From 4d62b8911eefba01417fdcf8d4599b4fe4f13da6 Mon Sep 17 00:00:00 2001 From: Stanca Serban Date: Fri, 19 Apr 2024 08:07:52 +0300 Subject: [PATCH] Backed out 2 changesets (bug 1890935, bug 1882209) for causing Android build bustages. CLOSED TREE Backed out changeset 05cb55554fc0 (bug 1882209) Backed out changeset 67b0625e2cfa (bug 1890935) --- Cargo.lock | 25 +- config/makefiles/rust.mk | 2 +- docs/writing-rust-code/update-policy.md | 4 +- python/mozboot/mozboot/util.py | 2 +- supply-chain/audits.toml | 6 - supply-chain/imports.lock | 7 - taskcluster/kinds/toolchain/rust.yml | 6 +- .../any_all_workaround/.cargo-checksum.json | 1 - .../rust/any_all_workaround/Cargo.toml | 25 - third_party/rust/any_all_workaround/README.md | 13 - .../rust/any_all_workaround/src/lib.rs | 99 -- .../rust/encoding_rs/.cargo-checksum.json | 2 +- third_party/rust/encoding_rs/Cargo.toml | 19 +- third_party/rust/encoding_rs/README.md | 34 +- third_party/rust/encoding_rs/src/ascii.rs | 388 +---- third_party/rust/encoding_rs/src/handles.rs | 36 +- third_party/rust/encoding_rs/src/lib.rs | 13 +- third_party/rust/encoding_rs/src/mem.rs | 18 +- .../rust/encoding_rs/src/simd_funcs.rs | 146 +- .../rust/encoding_rs/src/single_byte.rs | 64 +- .../rust/encoding_rs/src/x_user_defined.rs | 10 +- .../rust/packed_simd/.cargo-checksum.json | 1 + third_party/rust/packed_simd/Cargo.toml | 83 + .../LICENSE-APACHE | 0 .../LICENSE-MIT | 0 third_party/rust/packed_simd/README.md | 144 ++ third_party/rust/packed_simd/bors.toml | 3 + third_party/rust/packed_simd/build.rs | 6 + third_party/rust/packed_simd/ci/all.sh | 71 + .../packed_simd/ci/android-install-ndk.sh | 21 + .../packed_simd/ci/android-install-sdk.sh | 60 + .../rust/packed_simd/ci/android-sysimage.sh | 56 + third_party/rust/packed_simd/ci/benchmark.sh | 32 + .../ci/deploy_and_run_on_ios_simulator.rs | 176 +++ .../docker/aarch64-linux-android/Dockerfile | 47 + .../aarch64-unknown-linux-gnu/Dockerfile | 14 + .../arm-unknown-linux-gnueabi/Dockerfile | 15 + .../arm-unknown-linux-gnueabihf/Dockerfile | 13 + .../docker/armv7-linux-androideabi/Dockerfile | 47 + .../armv7-unknown-linux-gnueabihf/Dockerfile | 13 + .../docker/i586-unknown-linux-gnu/Dockerfile | 7 + .../docker/i686-unknown-linux-gnu/Dockerfile | 7 + .../docker/mips-unknown-linux-gnu/Dockerfile | 13 + .../mips64-unknown-linux-gnuabi64/Dockerfile | 10 + .../Dockerfile | 10 + .../mipsel-unknown-linux-musl/Dockerfile | 25 + .../powerpc-unknown-linux-gnu/Dockerfile | 13 + .../powerpc64-unknown-linux-gnu/Dockerfile | 17 + .../powerpc64le-unknown-linux-gnu/Dockerfile | 11 + .../docker/s390x-unknown-linux-gnu/Dockerfile | 20 + .../sparc64-unknown-linux-gnu/Dockerfile | 18 + .../thumbv7neon-linux-androideabi/Dockerfile | 47 + .../Dockerfile | 13 + .../docker/wasm32-unknown-unknown/Dockerfile | 39 + .../ci/docker/x86_64-linux-android/Dockerfile | 31 + .../Dockerfile | 16 + .../x86_64-unknown-linux-gnu/Dockerfile | 10 + third_party/rust/packed_simd/ci/dox.sh | 27 + .../rust/packed_simd/ci/linux-s390x.sh | 18 + .../rust/packed_simd/ci/linux-sparc64.sh | 17 + third_party/rust/packed_simd/ci/lld-shim.rs | 11 + .../rust/packed_simd/ci/max_line_width.sh | 17 + third_party/rust/packed_simd/ci/run-docker.sh | 38 + third_party/rust/packed_simd/ci/run.sh | 99 ++ .../rust/packed_simd/ci/run_examples.sh | 51 + .../rust/packed_simd/ci/runtest-android.rs | 45 + .../rust/packed_simd/ci/setup_benchmarks.sh | 7 + .../rust/packed_simd/ci/test-runner-linux | 24 + third_party/rust/packed_simd/contributing.md | 67 + .../rust/packed_simd/perf-guide/book.toml | 12 + .../packed_simd/perf-guide/src/SUMMARY.md | 21 + .../rust/packed_simd/perf-guide/src/ascii.css | 4 + .../perf-guide/src/bound_checks.md | 22 + .../perf-guide/src/float-math/approx.md | 8 + .../perf-guide/src/float-math/fma.md | 6 + .../perf-guide/src/float-math/fp.md | 3 + .../perf-guide/src/float-math/svml.md | 7 + .../perf-guide/src/introduction.md | 26 + .../packed_simd/perf-guide/src/prof/linux.md | 107 ++ .../packed_simd/perf-guide/src/prof/mca.md | 100 ++ .../perf-guide/src/prof/profiling.md | 14 + .../src/target-feature/attribute.md | 5 + .../perf-guide/src/target-feature/features.md | 13 + .../perf-guide/src/target-feature/inlining.md | 5 + .../perf-guide/src/target-feature/practice.md | 31 + .../perf-guide/src/target-feature/runtime.md | 5 + .../src/target-feature/rustflags.md | 77 + .../perf-guide/src/vert-hor-ops.md | 76 + third_party/rust/packed_simd/rust-toolchain | 1 + third_party/rust/packed_simd/rustfmt.toml | 5 + third_party/rust/packed_simd/src/api.rs | 309 ++++ .../rust/packed_simd/src/api/bit_manip.rs | 129 ++ .../rust/packed_simd/src/api/bitmask.rs | 79 + third_party/rust/packed_simd/src/api/cast.rs | 108 ++ .../rust/packed_simd/src/api/cast/macros.rs | 82 + .../rust/packed_simd/src/api/cast/v128.rs | 302 ++++ .../rust/packed_simd/src/api/cast/v16.rs | 68 + .../rust/packed_simd/src/api/cast/v256.rs | 298 ++++ .../rust/packed_simd/src/api/cast/v32.rs | 132 ++ .../rust/packed_simd/src/api/cast/v512.rs | 209 +++ .../rust/packed_simd/src/api/cast/v64.rs | 208 +++ third_party/rust/packed_simd/src/api/cmp.rs | 16 + .../rust/packed_simd/src/api/cmp/eq.rs | 27 + .../rust/packed_simd/src/api/cmp/ord.rs | 43 + .../packed_simd/src/api/cmp/partial_eq.rs | 65 + .../packed_simd/src/api/cmp/partial_ord.rs | 230 +++ .../rust/packed_simd/src/api/cmp/vertical.rs | 114 ++ .../rust/packed_simd/src/api/default.rs | 30 + third_party/rust/packed_simd/src/api/fmt.rs | 12 + .../rust/packed_simd/src/api/fmt/binary.rs | 54 + .../rust/packed_simd/src/api/fmt/debug.rs | 60 + .../rust/packed_simd/src/api/fmt/lower_hex.rs | 54 + .../rust/packed_simd/src/api/fmt/octal.rs | 54 + .../rust/packed_simd/src/api/fmt/upper_hex.rs | 54 + third_party/rust/packed_simd/src/api/from.rs | 7 + .../packed_simd/src/api/from/from_array.rs | 124 ++ .../packed_simd/src/api/from/from_vector.rs | 67 + third_party/rust/packed_simd/src/api/hash.rs | 49 + .../rust/packed_simd/src/api/into_bits.rs | 59 + .../src/api/into_bits/arch_specific.rs | 345 +++++ .../packed_simd/src/api/into_bits/macros.rs | 74 + .../packed_simd/src/api/into_bits/v128.rs | 232 +++ .../rust/packed_simd/src/api/into_bits/v16.rs | 9 + .../packed_simd/src/api/into_bits/v256.rs | 232 +++ .../rust/packed_simd/src/api/into_bits/v32.rs | 13 + .../packed_simd/src/api/into_bits/v512.rs | 232 +++ .../rust/packed_simd/src/api/into_bits/v64.rs | 18 + third_party/rust/packed_simd/src/api/math.rs | 4 + .../rust/packed_simd/src/api/math/float.rs | 64 + .../packed_simd/src/api/math/float/abs.rs | 31 + .../packed_simd/src/api/math/float/consts.rs | 74 + .../packed_simd/src/api/math/float/cos.rs | 44 + .../packed_simd/src/api/math/float/exp.rs | 33 + .../rust/packed_simd/src/api/math/float/ln.rs | 33 + .../packed_simd/src/api/math/float/mul_add.rs | 44 + .../src/api/math/float/mul_adde.rs | 48 + .../packed_simd/src/api/math/float/powf.rs | 36 + .../packed_simd/src/api/math/float/recpre.rs | 36 + .../packed_simd/src/api/math/float/rsqrte.rs | 40 + .../packed_simd/src/api/math/float/sin.rs | 50 + .../packed_simd/src/api/math/float/sqrt.rs | 35 + .../packed_simd/src/api/math/float/sqrte.rs | 44 + .../packed_simd/src/api/math/float/tanh.rs | 29 + .../rust/packed_simd/src/api/minimal.rs | 6 + .../rust/packed_simd/src/api/minimal/iuf.rs | 169 ++ .../rust/packed_simd/src/api/minimal/mask.rs | 176 +++ .../rust/packed_simd/src/api/minimal/ptr.rs | 1373 +++++++++++++++++ third_party/rust/packed_simd/src/api/ops.rs | 32 + .../src/api/ops/scalar_arithmetic.rs | 203 +++ .../packed_simd/src/api/ops/scalar_bitwise.rs | 162 ++ .../src/api/ops/scalar_mask_bitwise.rs | 140 ++ .../packed_simd/src/api/ops/scalar_shifts.rs | 106 ++ .../src/api/ops/vector_arithmetic.rs | 148 ++ .../packed_simd/src/api/ops/vector_bitwise.rs | 129 ++ .../src/api/ops/vector_float_min_max.rs | 74 + .../src/api/ops/vector_int_min_max.rs | 57 + .../src/api/ops/vector_mask_bitwise.rs | 116 ++ .../packed_simd/src/api/ops/vector_neg.rs | 43 + .../packed_simd/src/api/ops/vector_rotates.rs | 92 ++ .../packed_simd/src/api/ops/vector_shifts.rs | 106 ++ third_party/rust/packed_simd/src/api/ptr.rs | 4 + .../packed_simd/src/api/ptr/gather_scatter.rs | 216 +++ .../rust/packed_simd/src/api/reductions.rs | 12 + .../packed_simd/src/api/reductions/bitwise.rs | 151 ++ .../src/api/reductions/float_arithmetic.rs | 313 ++++ .../src/api/reductions/integer_arithmetic.rs | 193 +++ .../packed_simd/src/api/reductions/mask.rs | 89 ++ .../packed_simd/src/api/reductions/min_max.rs | 360 +++++ .../rust/packed_simd/src/api/select.rs | 73 + .../rust/packed_simd/src/api/shuffle.rs | 184 +++ .../rust/packed_simd/src/api/shuffle1_dyn.rs | 159 ++ third_party/rust/packed_simd/src/api/slice.rs | 7 + .../packed_simd/src/api/slice/from_slice.rs | 202 +++ .../src/api/slice/write_to_slice.rs | 196 +++ .../rust/packed_simd/src/api/swap_bytes.rs | 192 +++ third_party/rust/packed_simd/src/codegen.rs | 62 + .../rust/packed_simd/src/codegen/bit_manip.rs | 347 +++++ .../rust/packed_simd/src/codegen/llvm.rs | 122 ++ .../rust/packed_simd/src/codegen/math.rs | 3 + .../packed_simd/src/codegen/math/float.rs | 19 + .../packed_simd/src/codegen/math/float/abs.rs | 103 ++ .../packed_simd/src/codegen/math/float/cos.rs | 103 ++ .../src/codegen/math/float/cos_pi.rs | 87 ++ .../packed_simd/src/codegen/math/float/exp.rs | 112 ++ .../packed_simd/src/codegen/math/float/ln.rs | 112 ++ .../src/codegen/math/float/macros.rs | 470 ++++++ .../src/codegen/math/float/mul_add.rs | 109 ++ .../src/codegen/math/float/mul_adde.rs | 60 + .../src/codegen/math/float/powf.rs | 112 ++ .../packed_simd/src/codegen/math/float/sin.rs | 103 ++ .../src/codegen/math/float/sin_cos_pi.rs | 188 +++ .../src/codegen/math/float/sin_pi.rs | 87 ++ .../src/codegen/math/float/sqrt.rs | 103 ++ .../src/codegen/math/float/sqrte.rs | 67 + .../src/codegen/math/float/tanh.rs | 120 ++ .../src/codegen/pointer_sized_int.rs | 28 + .../packed_simd/src/codegen/reductions.rs | 1 + .../src/codegen/reductions/mask.rs | 69 + .../src/codegen/reductions/mask/aarch64.rs | 81 + .../src/codegen/reductions/mask/arm.rs | 56 + .../src/codegen/reductions/mask/fallback.rs | 8 + .../codegen/reductions/mask/fallback_impl.rs | 237 +++ .../src/codegen/reductions/mask/x86.rs | 216 +++ .../src/codegen/reductions/mask/x86/avx.rs | 95 ++ .../src/codegen/reductions/mask/x86/avx2.rs | 35 + .../src/codegen/reductions/mask/x86/sse.rs | 35 + .../src/codegen/reductions/mask/x86/sse2.rs | 68 + .../rust/packed_simd/src/codegen/shuffle.rs | 150 ++ .../packed_simd/src/codegen/shuffle1_dyn.rs | 408 +++++ .../packed_simd/src/codegen/swap_bytes.rs | 149 ++ .../rust/packed_simd/src/codegen/v128.rs | 46 + .../rust/packed_simd/src/codegen/v16.rs | 7 + .../rust/packed_simd/src/codegen/v256.rs | 78 + .../rust/packed_simd/src/codegen/v32.rs | 11 + .../rust/packed_simd/src/codegen/v512.rs | 145 ++ .../rust/packed_simd/src/codegen/v64.rs | 21 + .../rust/packed_simd/src/codegen/vPtr.rs | 35 + .../rust/packed_simd/src/codegen/vSize.rs | 16 + third_party/rust/packed_simd/src/lib.rs | 348 +++++ third_party/rust/packed_simd/src/masks.rs | 126 ++ third_party/rust/packed_simd/src/sealed.rs | 42 + third_party/rust/packed_simd/src/testing.rs | 8 + .../rust/packed_simd/src/testing/macros.rs | 44 + .../rust/packed_simd/src/testing/utils.rs | 130 ++ third_party/rust/packed_simd/src/v128.rs | 80 + third_party/rust/packed_simd/src/v16.rs | 16 + third_party/rust/packed_simd/src/v256.rs | 86 ++ third_party/rust/packed_simd/src/v32.rs | 29 + third_party/rust/packed_simd/src/v512.rs | 99 ++ third_party/rust/packed_simd/src/v64.rs | 66 + third_party/rust/packed_simd/src/vPtr.rs | 34 + third_party/rust/packed_simd/src/vSize.rs | 53 + .../rust/packed_simd/tests/endianness.rs | 268 ++++ 233 files changed, 18455 insertions(+), 782 deletions(-) delete mode 100644 third_party/rust/any_all_workaround/.cargo-checksum.json delete mode 100644 third_party/rust/any_all_workaround/Cargo.toml delete mode 100644 third_party/rust/any_all_workaround/README.md delete mode 100644 third_party/rust/any_all_workaround/src/lib.rs create mode 100644 third_party/rust/packed_simd/.cargo-checksum.json create mode 100644 third_party/rust/packed_simd/Cargo.toml rename third_party/rust/{any_all_workaround => packed_simd}/LICENSE-APACHE (100%) rename third_party/rust/{any_all_workaround => packed_simd}/LICENSE-MIT (100%) create mode 100644 third_party/rust/packed_simd/README.md create mode 100644 third_party/rust/packed_simd/bors.toml create mode 100644 third_party/rust/packed_simd/build.rs create mode 100755 third_party/rust/packed_simd/ci/all.sh create mode 100644 third_party/rust/packed_simd/ci/android-install-ndk.sh create mode 100644 third_party/rust/packed_simd/ci/android-install-sdk.sh create mode 100644 third_party/rust/packed_simd/ci/android-sysimage.sh create mode 100755 third_party/rust/packed_simd/ci/benchmark.sh create mode 100644 third_party/rust/packed_simd/ci/deploy_and_run_on_ios_simulator.rs create mode 100644 third_party/rust/packed_simd/ci/docker/aarch64-linux-android/Dockerfile create mode 100644 third_party/rust/packed_simd/ci/docker/aarch64-unknown-linux-gnu/Dockerfile create mode 100644 third_party/rust/packed_simd/ci/docker/arm-unknown-linux-gnueabi/Dockerfile create mode 100644 third_party/rust/packed_simd/ci/docker/arm-unknown-linux-gnueabihf/Dockerfile create mode 100644 third_party/rust/packed_simd/ci/docker/armv7-linux-androideabi/Dockerfile create mode 100644 third_party/rust/packed_simd/ci/docker/armv7-unknown-linux-gnueabihf/Dockerfile create mode 100644 third_party/rust/packed_simd/ci/docker/i586-unknown-linux-gnu/Dockerfile create mode 100644 third_party/rust/packed_simd/ci/docker/i686-unknown-linux-gnu/Dockerfile create mode 100644 third_party/rust/packed_simd/ci/docker/mips-unknown-linux-gnu/Dockerfile create mode 100644 third_party/rust/packed_simd/ci/docker/mips64-unknown-linux-gnuabi64/Dockerfile create mode 100644 third_party/rust/packed_simd/ci/docker/mips64el-unknown-linux-gnuabi64/Dockerfile create mode 100644 third_party/rust/packed_simd/ci/docker/mipsel-unknown-linux-musl/Dockerfile create mode 100644 third_party/rust/packed_simd/ci/docker/powerpc-unknown-linux-gnu/Dockerfile create mode 100644 third_party/rust/packed_simd/ci/docker/powerpc64-unknown-linux-gnu/Dockerfile create mode 100644 third_party/rust/packed_simd/ci/docker/powerpc64le-unknown-linux-gnu/Dockerfile create mode 100644 third_party/rust/packed_simd/ci/docker/s390x-unknown-linux-gnu/Dockerfile create mode 100644 third_party/rust/packed_simd/ci/docker/sparc64-unknown-linux-gnu/Dockerfile create mode 100644 third_party/rust/packed_simd/ci/docker/thumbv7neon-linux-androideabi/Dockerfile create mode 100644 third_party/rust/packed_simd/ci/docker/thumbv7neon-unknown-linux-gnueabihf/Dockerfile create mode 100644 third_party/rust/packed_simd/ci/docker/wasm32-unknown-unknown/Dockerfile create mode 100644 third_party/rust/packed_simd/ci/docker/x86_64-linux-android/Dockerfile create mode 100644 third_party/rust/packed_simd/ci/docker/x86_64-unknown-linux-gnu-emulated/Dockerfile create mode 100644 third_party/rust/packed_simd/ci/docker/x86_64-unknown-linux-gnu/Dockerfile create mode 100755 third_party/rust/packed_simd/ci/dox.sh create mode 100644 third_party/rust/packed_simd/ci/linux-s390x.sh create mode 100644 third_party/rust/packed_simd/ci/linux-sparc64.sh create mode 100644 third_party/rust/packed_simd/ci/lld-shim.rs create mode 100755 third_party/rust/packed_simd/ci/max_line_width.sh create mode 100755 third_party/rust/packed_simd/ci/run-docker.sh create mode 100755 third_party/rust/packed_simd/ci/run.sh create mode 100644 third_party/rust/packed_simd/ci/run_examples.sh create mode 100644 third_party/rust/packed_simd/ci/runtest-android.rs create mode 100755 third_party/rust/packed_simd/ci/setup_benchmarks.sh create mode 100755 third_party/rust/packed_simd/ci/test-runner-linux create mode 100644 third_party/rust/packed_simd/contributing.md create mode 100644 third_party/rust/packed_simd/perf-guide/book.toml create mode 100644 third_party/rust/packed_simd/perf-guide/src/SUMMARY.md create mode 100644 third_party/rust/packed_simd/perf-guide/src/ascii.css create mode 100644 third_party/rust/packed_simd/perf-guide/src/bound_checks.md create mode 100644 third_party/rust/packed_simd/perf-guide/src/float-math/approx.md create mode 100644 third_party/rust/packed_simd/perf-guide/src/float-math/fma.md create mode 100644 third_party/rust/packed_simd/perf-guide/src/float-math/fp.md create mode 100644 third_party/rust/packed_simd/perf-guide/src/float-math/svml.md create mode 100644 third_party/rust/packed_simd/perf-guide/src/introduction.md create mode 100644 third_party/rust/packed_simd/perf-guide/src/prof/linux.md create mode 100644 third_party/rust/packed_simd/perf-guide/src/prof/mca.md create mode 100644 third_party/rust/packed_simd/perf-guide/src/prof/profiling.md create mode 100644 third_party/rust/packed_simd/perf-guide/src/target-feature/attribute.md create mode 100644 third_party/rust/packed_simd/perf-guide/src/target-feature/features.md create mode 100644 third_party/rust/packed_simd/perf-guide/src/target-feature/inlining.md create mode 100644 third_party/rust/packed_simd/perf-guide/src/target-feature/practice.md create mode 100644 third_party/rust/packed_simd/perf-guide/src/target-feature/runtime.md create mode 100644 third_party/rust/packed_simd/perf-guide/src/target-feature/rustflags.md create mode 100644 third_party/rust/packed_simd/perf-guide/src/vert-hor-ops.md create mode 100644 third_party/rust/packed_simd/rust-toolchain create mode 100644 third_party/rust/packed_simd/rustfmt.toml create mode 100644 third_party/rust/packed_simd/src/api.rs create mode 100644 third_party/rust/packed_simd/src/api/bit_manip.rs create mode 100644 third_party/rust/packed_simd/src/api/bitmask.rs create mode 100644 third_party/rust/packed_simd/src/api/cast.rs create mode 100644 third_party/rust/packed_simd/src/api/cast/macros.rs create mode 100644 third_party/rust/packed_simd/src/api/cast/v128.rs create mode 100644 third_party/rust/packed_simd/src/api/cast/v16.rs create mode 100644 third_party/rust/packed_simd/src/api/cast/v256.rs create mode 100644 third_party/rust/packed_simd/src/api/cast/v32.rs create mode 100644 third_party/rust/packed_simd/src/api/cast/v512.rs create mode 100644 third_party/rust/packed_simd/src/api/cast/v64.rs create mode 100644 third_party/rust/packed_simd/src/api/cmp.rs create mode 100644 third_party/rust/packed_simd/src/api/cmp/eq.rs create mode 100644 third_party/rust/packed_simd/src/api/cmp/ord.rs create mode 100644 third_party/rust/packed_simd/src/api/cmp/partial_eq.rs create mode 100644 third_party/rust/packed_simd/src/api/cmp/partial_ord.rs create mode 100644 third_party/rust/packed_simd/src/api/cmp/vertical.rs create mode 100644 third_party/rust/packed_simd/src/api/default.rs create mode 100644 third_party/rust/packed_simd/src/api/fmt.rs create mode 100644 third_party/rust/packed_simd/src/api/fmt/binary.rs create mode 100644 third_party/rust/packed_simd/src/api/fmt/debug.rs create mode 100644 third_party/rust/packed_simd/src/api/fmt/lower_hex.rs create mode 100644 third_party/rust/packed_simd/src/api/fmt/octal.rs create mode 100644 third_party/rust/packed_simd/src/api/fmt/upper_hex.rs create mode 100644 third_party/rust/packed_simd/src/api/from.rs create mode 100644 third_party/rust/packed_simd/src/api/from/from_array.rs create mode 100644 third_party/rust/packed_simd/src/api/from/from_vector.rs create mode 100644 third_party/rust/packed_simd/src/api/hash.rs create mode 100644 third_party/rust/packed_simd/src/api/into_bits.rs create mode 100644 third_party/rust/packed_simd/src/api/into_bits/arch_specific.rs create mode 100644 third_party/rust/packed_simd/src/api/into_bits/macros.rs create mode 100644 third_party/rust/packed_simd/src/api/into_bits/v128.rs create mode 100644 third_party/rust/packed_simd/src/api/into_bits/v16.rs create mode 100644 third_party/rust/packed_simd/src/api/into_bits/v256.rs create mode 100644 third_party/rust/packed_simd/src/api/into_bits/v32.rs create mode 100644 third_party/rust/packed_simd/src/api/into_bits/v512.rs create mode 100644 third_party/rust/packed_simd/src/api/into_bits/v64.rs create mode 100644 third_party/rust/packed_simd/src/api/math.rs create mode 100644 third_party/rust/packed_simd/src/api/math/float.rs create mode 100644 third_party/rust/packed_simd/src/api/math/float/abs.rs create mode 100644 third_party/rust/packed_simd/src/api/math/float/consts.rs create mode 100644 third_party/rust/packed_simd/src/api/math/float/cos.rs create mode 100644 third_party/rust/packed_simd/src/api/math/float/exp.rs create mode 100644 third_party/rust/packed_simd/src/api/math/float/ln.rs create mode 100644 third_party/rust/packed_simd/src/api/math/float/mul_add.rs create mode 100644 third_party/rust/packed_simd/src/api/math/float/mul_adde.rs create mode 100644 third_party/rust/packed_simd/src/api/math/float/powf.rs create mode 100644 third_party/rust/packed_simd/src/api/math/float/recpre.rs create mode 100644 third_party/rust/packed_simd/src/api/math/float/rsqrte.rs create mode 100644 third_party/rust/packed_simd/src/api/math/float/sin.rs create mode 100644 third_party/rust/packed_simd/src/api/math/float/sqrt.rs create mode 100644 third_party/rust/packed_simd/src/api/math/float/sqrte.rs create mode 100644 third_party/rust/packed_simd/src/api/math/float/tanh.rs create mode 100644 third_party/rust/packed_simd/src/api/minimal.rs create mode 100644 third_party/rust/packed_simd/src/api/minimal/iuf.rs create mode 100644 third_party/rust/packed_simd/src/api/minimal/mask.rs create mode 100644 third_party/rust/packed_simd/src/api/minimal/ptr.rs create mode 100644 third_party/rust/packed_simd/src/api/ops.rs create mode 100644 third_party/rust/packed_simd/src/api/ops/scalar_arithmetic.rs create mode 100644 third_party/rust/packed_simd/src/api/ops/scalar_bitwise.rs create mode 100644 third_party/rust/packed_simd/src/api/ops/scalar_mask_bitwise.rs create mode 100644 third_party/rust/packed_simd/src/api/ops/scalar_shifts.rs create mode 100644 third_party/rust/packed_simd/src/api/ops/vector_arithmetic.rs create mode 100644 third_party/rust/packed_simd/src/api/ops/vector_bitwise.rs create mode 100644 third_party/rust/packed_simd/src/api/ops/vector_float_min_max.rs create mode 100644 third_party/rust/packed_simd/src/api/ops/vector_int_min_max.rs create mode 100644 third_party/rust/packed_simd/src/api/ops/vector_mask_bitwise.rs create mode 100644 third_party/rust/packed_simd/src/api/ops/vector_neg.rs create mode 100644 third_party/rust/packed_simd/src/api/ops/vector_rotates.rs create mode 100644 third_party/rust/packed_simd/src/api/ops/vector_shifts.rs create mode 100644 third_party/rust/packed_simd/src/api/ptr.rs create mode 100644 third_party/rust/packed_simd/src/api/ptr/gather_scatter.rs create mode 100644 third_party/rust/packed_simd/src/api/reductions.rs create mode 100644 third_party/rust/packed_simd/src/api/reductions/bitwise.rs create mode 100644 third_party/rust/packed_simd/src/api/reductions/float_arithmetic.rs create mode 100644 third_party/rust/packed_simd/src/api/reductions/integer_arithmetic.rs create mode 100644 third_party/rust/packed_simd/src/api/reductions/mask.rs create mode 100644 third_party/rust/packed_simd/src/api/reductions/min_max.rs create mode 100644 third_party/rust/packed_simd/src/api/select.rs create mode 100644 third_party/rust/packed_simd/src/api/shuffle.rs create mode 100644 third_party/rust/packed_simd/src/api/shuffle1_dyn.rs create mode 100644 third_party/rust/packed_simd/src/api/slice.rs create mode 100644 third_party/rust/packed_simd/src/api/slice/from_slice.rs create mode 100644 third_party/rust/packed_simd/src/api/slice/write_to_slice.rs create mode 100644 third_party/rust/packed_simd/src/api/swap_bytes.rs create mode 100644 third_party/rust/packed_simd/src/codegen.rs create mode 100644 third_party/rust/packed_simd/src/codegen/bit_manip.rs create mode 100644 third_party/rust/packed_simd/src/codegen/llvm.rs create mode 100644 third_party/rust/packed_simd/src/codegen/math.rs create mode 100644 third_party/rust/packed_simd/src/codegen/math/float.rs create mode 100644 third_party/rust/packed_simd/src/codegen/math/float/abs.rs create mode 100644 third_party/rust/packed_simd/src/codegen/math/float/cos.rs create mode 100644 third_party/rust/packed_simd/src/codegen/math/float/cos_pi.rs create mode 100644 third_party/rust/packed_simd/src/codegen/math/float/exp.rs create mode 100644 third_party/rust/packed_simd/src/codegen/math/float/ln.rs create mode 100644 third_party/rust/packed_simd/src/codegen/math/float/macros.rs create mode 100644 third_party/rust/packed_simd/src/codegen/math/float/mul_add.rs create mode 100644 third_party/rust/packed_simd/src/codegen/math/float/mul_adde.rs create mode 100644 third_party/rust/packed_simd/src/codegen/math/float/powf.rs create mode 100644 third_party/rust/packed_simd/src/codegen/math/float/sin.rs create mode 100644 third_party/rust/packed_simd/src/codegen/math/float/sin_cos_pi.rs create mode 100644 third_party/rust/packed_simd/src/codegen/math/float/sin_pi.rs create mode 100644 third_party/rust/packed_simd/src/codegen/math/float/sqrt.rs create mode 100644 third_party/rust/packed_simd/src/codegen/math/float/sqrte.rs create mode 100644 third_party/rust/packed_simd/src/codegen/math/float/tanh.rs create mode 100644 third_party/rust/packed_simd/src/codegen/pointer_sized_int.rs create mode 100644 third_party/rust/packed_simd/src/codegen/reductions.rs create mode 100644 third_party/rust/packed_simd/src/codegen/reductions/mask.rs create mode 100644 third_party/rust/packed_simd/src/codegen/reductions/mask/aarch64.rs create mode 100644 third_party/rust/packed_simd/src/codegen/reductions/mask/arm.rs create mode 100644 third_party/rust/packed_simd/src/codegen/reductions/mask/fallback.rs create mode 100644 third_party/rust/packed_simd/src/codegen/reductions/mask/fallback_impl.rs create mode 100644 third_party/rust/packed_simd/src/codegen/reductions/mask/x86.rs create mode 100644 third_party/rust/packed_simd/src/codegen/reductions/mask/x86/avx.rs create mode 100644 third_party/rust/packed_simd/src/codegen/reductions/mask/x86/avx2.rs create mode 100644 third_party/rust/packed_simd/src/codegen/reductions/mask/x86/sse.rs create mode 100644 third_party/rust/packed_simd/src/codegen/reductions/mask/x86/sse2.rs create mode 100644 third_party/rust/packed_simd/src/codegen/shuffle.rs create mode 100644 third_party/rust/packed_simd/src/codegen/shuffle1_dyn.rs create mode 100644 third_party/rust/packed_simd/src/codegen/swap_bytes.rs create mode 100644 third_party/rust/packed_simd/src/codegen/v128.rs create mode 100644 third_party/rust/packed_simd/src/codegen/v16.rs create mode 100644 third_party/rust/packed_simd/src/codegen/v256.rs create mode 100644 third_party/rust/packed_simd/src/codegen/v32.rs create mode 100644 third_party/rust/packed_simd/src/codegen/v512.rs create mode 100644 third_party/rust/packed_simd/src/codegen/v64.rs create mode 100644 third_party/rust/packed_simd/src/codegen/vPtr.rs create mode 100644 third_party/rust/packed_simd/src/codegen/vSize.rs create mode 100644 third_party/rust/packed_simd/src/lib.rs create mode 100644 third_party/rust/packed_simd/src/masks.rs create mode 100644 third_party/rust/packed_simd/src/sealed.rs create mode 100644 third_party/rust/packed_simd/src/testing.rs create mode 100644 third_party/rust/packed_simd/src/testing/macros.rs create mode 100644 third_party/rust/packed_simd/src/testing/utils.rs create mode 100644 third_party/rust/packed_simd/src/v128.rs create mode 100644 third_party/rust/packed_simd/src/v16.rs create mode 100644 third_party/rust/packed_simd/src/v256.rs create mode 100644 third_party/rust/packed_simd/src/v32.rs create mode 100644 third_party/rust/packed_simd/src/v512.rs create mode 100644 third_party/rust/packed_simd/src/v64.rs create mode 100644 third_party/rust/packed_simd/src/vPtr.rs create mode 100644 third_party/rust/packed_simd/src/vSize.rs create mode 100644 third_party/rust/packed_simd/tests/endianness.rs diff --git a/Cargo.lock b/Cargo.lock index e90fcfd85181..41811670d529 100644 --- a/Cargo.lock +++ b/Cargo.lock @@ -91,15 +91,6 @@ version = "1.0.3" source = "registry+https://github.com/rust-lang/crates.io-index" checksum = "b84bf0a05bbb2a83e5eb6fa36bb6e87baa08193c35ff52bbf6b38d8af2890e46" -[[package]] -name = "any_all_workaround" -version = "0.1.0" -source = "registry+https://github.com/rust-lang/crates.io-index" -checksum = "88fea40735f2cc320a5133ce772d39c571bd6c9b0d4c1a326926eecdd5af2e86" -dependencies = [ - "cfg-if 1.0.0", -] - [[package]] name = "anyhow" version = "1.0.69" @@ -1583,12 +1574,12 @@ dependencies = [ [[package]] name = "encoding_rs" -version = "0.8.34" +version = "0.8.33" source = "registry+https://github.com/rust-lang/crates.io-index" -checksum = "b45de904aa0b010bce2ab45264d0631681847fa7b6f2eaa7dab7619943bc4f59" +checksum = "7268b386296a025e474d5140678f75d6de9493ae55a5d709eeb9dd08149945e1" dependencies = [ - "any_all_workaround", "cfg-if 1.0.0", + "packed_simd", ] [[package]] @@ -4329,6 +4320,16 @@ dependencies = [ "oxilangtag", ] +[[package]] +name = "packed_simd" +version = "0.3.9" +source = "registry+https://github.com/rust-lang/crates.io-index" +checksum = "1f9f08af0c877571712e2e3e686ad79efad9657dbf0f7c3c8ba943ff6c38932d" +dependencies = [ + "cfg-if 1.0.0", + "num-traits", +] + [[package]] name = "parking_lot" version = "0.12.1" diff --git a/config/makefiles/rust.mk b/config/makefiles/rust.mk index 4b36aa99bc29..52c4e402f9d1 100644 --- a/config/makefiles/rust.mk +++ b/config/makefiles/rust.mk @@ -265,7 +265,7 @@ endif ifndef RUSTC_BOOTSTRAP RUSTC_BOOTSTRAP := mozglue_static,qcms ifdef MOZ_RUST_SIMD -RUSTC_BOOTSTRAP := $(RUSTC_BOOTSTRAP),encoding_rs,any_all_workaround +RUSTC_BOOTSTRAP := $(RUSTC_BOOTSTRAP),encoding_rs,packed_simd endif export RUSTC_BOOTSTRAP endif diff --git a/docs/writing-rust-code/update-policy.md b/docs/writing-rust-code/update-policy.md index 60eed74e944e..b23e7b6e22fa 100644 --- a/docs/writing-rust-code/update-policy.md +++ b/docs/writing-rust-code/update-policy.md @@ -152,9 +152,9 @@ Here are the Rust versions for each Firefox version. | Firefox 123 | Rust 1.75.0 | 1.70.0 | 2023 December 28 | 2024 January 11 | 2024 February 20 | Firefox 124 | Rust 1.76.0 | 1.70.0 | 2024 February 8 | 2024 February 15 | 2024 March 19 | Firefox 125 | Rust 1.76.0 | 1.74.0 | 2024 February 8 | 2024 March 14 | 2024 April 16 -| Firefox 126 | Rust 1.77.2 | 1.74.0 | 2024 March 28 | 2024 April 11 | 2024 May 14 | **Estimated** | -| Firefox 127 | Rust 1.78.0 | 1.76.0 | 2024 May 2 | 2024 May 9 | 2024 June 11 +| Firefox 126 | Rust 1.77.2 | ? | 2024 March 28 | 2024 April 11 | 2024 May 14 +| Firefox 127 | Rust 1.78.0 | ? | 2024 May 2 | 2024 May 9 | 2024 June 11 | Firefox 128 | Rust 1.78.0 | ? | 2024 May 2 | 2024 June 6 | 2024 July 9 | Firefox 129 | Rust 1.79.0 | ? | 2024 June 13 | 2024 July 4 | 2024 August 6 | Firefox 130 | Rust 1.80.0 | ? | 2024 July 25 | 2024 August 1 | 2024 September 3 diff --git a/python/mozboot/mozboot/util.py b/python/mozboot/mozboot/util.py index d523319d8bf7..2d26ffb93483 100644 --- a/python/mozboot/mozboot/util.py +++ b/python/mozboot/mozboot/util.py @@ -11,7 +11,7 @@ import certifi from mach.site import PythonVirtualenv from mach.util import get_state_dir -MINIMUM_RUST_VERSION = "1.76.0" +MINIMUM_RUST_VERSION = "1.74.0" def get_tools_dir(srcdir=False): diff --git a/supply-chain/audits.toml b/supply-chain/audits.toml index 1fa87868d8e2..1f3ca08be19f 100644 --- a/supply-chain/audits.toml +++ b/supply-chain/audits.toml @@ -588,12 +588,6 @@ who = "Mike Hommey " criteria = "safe-to-deploy" delta = "0.1.4 -> 0.1.5" -[[audits.any_all_workaround]] -who = "Henri Sivonen " -criteria = "safe-to-deploy" -version = "0.1.0" -notes = "The little code that is in this crate I reviewed and modified from packed_simd (which has previously been vendored in full instead of just this small part)." - [[audits.anyhow]] who = "Mike Hommey " criteria = "safe-to-deploy" diff --git a/supply-chain/imports.lock b/supply-chain/imports.lock index 018e601596ab..69a8160437c5 100644 --- a/supply-chain/imports.lock +++ b/supply-chain/imports.lock @@ -190,13 +190,6 @@ user-id = 4484 user-login = "hsivonen" user-name = "Henri Sivonen" -[[publisher.encoding_rs]] -version = "0.8.34" -when = "2024-04-10" -user-id = 4484 -user-login = "hsivonen" -user-name = "Henri Sivonen" - [[publisher.errno]] version = "0.3.8" when = "2023-11-28" diff --git a/taskcluster/kinds/toolchain/rust.yml b/taskcluster/kinds/toolchain/rust.yml index e1e2778f68a4..437c497d7147 100644 --- a/taskcluster/kinds/toolchain/rust.yml +++ b/taskcluster/kinds/toolchain/rust.yml @@ -11,12 +11,12 @@ job-defaults: script: repack_rust.py toolchain-artifact: public/build/rustc.tar.zst -linux64-rust-1.76: +linux64-rust-1.74: treeherder: - symbol: TL(rust-1.76) + symbol: TL(rust-1.74) run: arguments: [ - '--channel', '1.76.0', + '--channel', '1.74.0', '--host', 'x86_64-unknown-linux-gnu', '--target', 'x86_64-unknown-linux-gnu', '--target', 'i686-unknown-linux-gnu', diff --git a/third_party/rust/any_all_workaround/.cargo-checksum.json b/third_party/rust/any_all_workaround/.cargo-checksum.json deleted file mode 100644 index 1d6b87a1d8a3..000000000000 --- a/third_party/rust/any_all_workaround/.cargo-checksum.json +++ /dev/null @@ -1 +0,0 @@ -{"files":{"Cargo.toml":"c38be4bc8ef1c4df398b3eae589681d1bbb54a2577c71d592e12db0af757c472","LICENSE-APACHE":"a60eea817514531668d7e00765731449fe14d059d3249e0bc93b36de45f759f2","LICENSE-MIT":"6485b8ed310d3f0340bf1ad1f47645069ce4069dcc6bb46c7d5c6faf41de1fdb","README.md":"abebbd2620f915c70a873dd8221d99eadd8d017b7b194c22f3e0051f1fde193f","src/lib.rs":"e8a36b888f0f20accd4e7bfb2db9196e42b4be2d1014cb675981543d1372c610"},"package":"88fea40735f2cc320a5133ce772d39c571bd6c9b0d4c1a326926eecdd5af2e86"} \ No newline at end of file diff --git a/third_party/rust/any_all_workaround/Cargo.toml b/third_party/rust/any_all_workaround/Cargo.toml deleted file mode 100644 index ae749e1e5b2c..000000000000 --- a/third_party/rust/any_all_workaround/Cargo.toml +++ /dev/null @@ -1,25 +0,0 @@ -# THIS FILE IS AUTOMATICALLY GENERATED BY CARGO -# -# When uploading crates to the registry Cargo will automatically -# "normalize" Cargo.toml files for maximal compatibility -# with all versions of Cargo and also rewrite `path` dependencies -# to registry (e.g., crates.io) dependencies. -# -# If you are reading this file be aware that the original Cargo.toml -# will likely look very different (and much more reasonable). -# See Cargo.toml.orig for the original contents. - -[package] -edition = "2021" -name = "any_all_workaround" -version = "0.1.0" -authors = ["Henri Sivonen "] -description = "Workaround for bad LLVM codegen for boolean reductions on 32-bit ARM" -homepage = "https://docs.rs/any_all_workaround/" -documentation = "https://docs.rs/any_all_workaround/" -readme = "README.md" -license = "MIT OR Apache-2.0" -repository = "https://github.com/hsivonen/any_all_workaround" - -[dependencies.cfg-if] -version = "1.0" diff --git a/third_party/rust/any_all_workaround/README.md b/third_party/rust/any_all_workaround/README.md deleted file mode 100644 index 836e693f42de..000000000000 --- a/third_party/rust/any_all_workaround/README.md +++ /dev/null @@ -1,13 +0,0 @@ -# any_all_workaround - -This is a workaround for bad codegen ([Rust bug](https://github.com/rust-lang/portable-simd/issues/146), [LLVM bug](https://github.com/llvm/llvm-project/issues/50466)) for the `any()` and `all()` reductions for NEON-backed SIMD vectors on 32-bit ARM. On other platforms these delegate to `any()` and `all()` in `core::simd`. - -The plan is to abandon this crate once the LLVM bug is fixed or `core::simd` works around the LLVM bug. - -The code is forked from the [`packed_simd` crate](https://raw.githubusercontent.com/hsivonen/packed_simd/d938e39bee9bc5c222f5f2f2a0df9e53b5ce36ae/src/codegen/reductions/mask/arm.rs). - -This crate requires Nightly Rust as it depends on the `portable_simd` feature. - -# License - -`MIT OR Apache-2.0`, since that's how `packed_simd` is licensed. diff --git a/third_party/rust/any_all_workaround/src/lib.rs b/third_party/rust/any_all_workaround/src/lib.rs deleted file mode 100644 index 57bb81e14df2..000000000000 --- a/third_party/rust/any_all_workaround/src/lib.rs +++ /dev/null @@ -1,99 +0,0 @@ -// This code began as a fork of -// https://raw.githubusercontent.com/rust-lang/packed_simd/d938e39bee9bc5c222f5f2f2a0df9e53b5ce36ae/src/codegen/reductions/mask/arm.rs -// which didn't have a license header on the file, but Cargo.toml said "MIT OR Apache-2.0". -// See LICENSE-MIT and LICENSE-APACHE. - -#![no_std] -#![feature(portable_simd)] -#![cfg_attr( - all( - target_arch = "arm", - target_endian = "little", - target_feature = "neon", - target_feature = "v7" - ), - feature(stdarch_arm_neon_intrinsics) -)] - -use cfg_if::cfg_if; -use core::simd::mask16x8; -use core::simd::mask32x4; -use core::simd::mask8x16; - -cfg_if! { - if #[cfg(all(target_arch = "arm", target_endian = "little", target_feature = "neon", target_feature = "v7"))] { - use core::simd::mask8x8; - use core::simd::mask16x4; - use core::simd::mask32x2; - macro_rules! arm_128_v7_neon_impl { - ($all:ident, $any:ident, $id:ident, $half:ident, $vpmin:ident, $vpmax:ident) => { - #[inline] - pub fn $all(s: $id) -> bool { - use core::arch::arm::$vpmin; - use core::mem::transmute; - unsafe { - union U { - halves: ($half, $half), - vec: $id, - } - let halves = U { vec: s }.halves; - let h: $half = transmute($vpmin(transmute(halves.0), transmute(halves.1))); - h.all() - } - } - #[inline] - pub fn $any(s: $id) -> bool { - use core::arch::arm::$vpmax; - use core::mem::transmute; - unsafe { - union U { - halves: ($half, $half), - vec: $id, - } - let halves = U { vec: s }.halves; - let h: $half = transmute($vpmax(transmute(halves.0), transmute(halves.1))); - h.any() - } - } - } - } - } else { - macro_rules! arm_128_v7_neon_impl { - ($all:ident, $any:ident, $id:ident, $half:ident, $vpmin:ident, $vpmax:ident) => { - #[inline(always)] - pub fn $all(s: $id) -> bool { - s.all() - } - #[inline(always)] - pub fn $any(s: $id) -> bool { - s.any() - } - } - } - } -} - -arm_128_v7_neon_impl!( - all_mask8x16, - any_mask8x16, - mask8x16, - mask8x8, - vpmin_u8, - vpmax_u8 -); -arm_128_v7_neon_impl!( - all_mask16x8, - any_mask16x8, - mask16x8, - mask16x4, - vpmin_u16, - vpmax_u16 -); -arm_128_v7_neon_impl!( - all_mask32x4, - any_mask32x4, - mask32x4, - mask32x2, - vpmin_u32, - vpmax_u32 -); diff --git a/third_party/rust/encoding_rs/.cargo-checksum.json b/third_party/rust/encoding_rs/.cargo-checksum.json index d5e91dd7e2ac..a5e1f1ba2513 100644 --- a/third_party/rust/encoding_rs/.cargo-checksum.json +++ b/third_party/rust/encoding_rs/.cargo-checksum.json @@ -1 +1 @@ -{"files":{"CONTRIBUTING.md":"ca1901f3e8532fb4cec894fd3664f0eaa898c0c4b961d1b992d1ed54eacf362a","COPYRIGHT":"11789f45bb180841cd362a5eee6789c68ddb573a11105e30768c308a6add0190","Cargo.toml":"22a4d210c92dae9f32c6944ef340ee8fdd027f99c081577e8907123e2a93383e","Ideas.md":"b7452893f500163868d8de52c09addaf91e1632454ed02e892c467ed7ec39dbd","LICENSE-APACHE":"cfc7749b96f63bd31c3c42b5c471bf756814053e847c10f3eb003417bc523d30","LICENSE-MIT":"3fa4ca83dcc9237839b1bdeb2e6d16bdfb5ec0c5ce42b24694d8bbf0dcbef72c","LICENSE-WHATWG":"838118388fe5c2e7f1dbbaeed13e1c7f3ebf88be91319c7c1d77c18e987d1a50","README.md":"1d08aefcb92afa81b18154049c9abbcad4540a23f7172e9f9bbed5af33f1a087","ci/miri.sh":"43cb8d82f49e3bfe2d2274b6ccd6f0714a4188ccef0cecc040829883cfdbee25","doc/Big5.txt":"f73a2edc5cb6c2d140ba6e07f4542e1c4a234950378acde1df93480f0ca0be0b","doc/EUC-JP.txt":"ee2818b907d0137f40a9ab9fd525fc700a44dbdddb6cf0c157a656566bae4bf1","doc/EUC-KR.txt":"71d9e2ccf3b124e8bdfb433c8cf2773fd878077038d0cec3c7237a50f4a78a30","doc/GBK.txt":"c1b522b5a799884e5001da661f42c5a8f4d0acb9ef1d74b206f22b5f65365606","doc/IBM866.txt":"a5a433e804d0f83af785015179fbc1d9b0eaf1f7960efcd04093e136b51fbd0e","doc/ISO-2022-JP.txt":"af86684f5a8f0e2868d7b2c292860140c3d2e5527530ca091f1b28198e8e2fe6","doc/ISO-8859-10.txt":"6d3949ad7c81ca176895101ed81a1db7df1060d64e262880b94bd31bb344ab4d","doc/ISO-8859-13.txt":"3951dd89cf93f7729148091683cf8511f4529388b7dc8dcd0d62eaed55be93fa","doc/ISO-8859-14.txt":"3d330784a0374fd255a38b47949675cc7168c800530534b0a01cac6edc623adc","doc/ISO-8859-15.txt":"24b1084aab5127a85aab99153f86e24694d0a3615f53b5ce23683f97cf66c47a","doc/ISO-8859-16.txt":"ce0272559b92ba76d7a7e476f6424ae4a5cc72e75b183611b08392e44add4d25","doc/ISO-8859-2.txt":"18ceff88c13d1b5ba455a3919b1e3de489045c4c3d2dd7e8527c125c75d54aad","doc/ISO-8859-3.txt":"21798404c68f4f5db59223362f24999da96968c0628427321fccce7d2849a130","doc/ISO-8859-4.txt":"d27f6520c6c5bfbcc19176b71d081cdb3bccde1622bb3e420d5680e812632d53","doc/ISO-8859-5.txt":"a10ec8d6ea7a78ad15da7275f6cb1a3365118527e28f9af6d0d5830501303f3a","doc/ISO-8859-6.txt":"ccda8a2efc96115336bdd77776637b9712425e44fbcf745353b9057fbef144e7","doc/ISO-8859-7.txt":"17900fa1f27a445958f0a77d7d9056be375a6bd7ee4492aa680c7c1500bab85e","doc/ISO-8859-8-I.txt":"8357555646d54265a9b9ffa3e68b08d132312f1561c60108ff9b8b1167b6ecf2","doc/ISO-8859-8.txt":"72cd6f3afb7b4a9c16a66a362473315770b7755d72c86c870e52fc3eba86c8af","doc/KOI8-R.txt":"839cf19a38da994488004ed7814b1f6151640156a9a2af02bf2efca745fb5966","doc/KOI8-U.txt":"0cc76624ed1f024183e2298b7e019957da2c70c8ca06e0fc4e6f353f50a5054f","doc/Shift_JIS.txt":"34c49141818cb9ddbcf59cc858f78a79be8ad148d563f26415108ae1f148443f","doc/UTF-16BE.txt":"e2e280d8acbaa6d2a6b3569d60e17500a285f2baa0df3363dd85537cd5a1ef8f","doc/UTF-16LE.txt":"70bdc170e3fc5298ba68f10125fb5eeb8b077036cc96bb4416c4de396f6d76c1","doc/UTF-8.txt":"ea7bae742e613010ced002cf4b601a737d2203fad65e115611451bc4428f548a","doc/gb18030.txt":"dc71378a8f07a2d8659f69ee81fb8791fef56ba86f124b429978285237bb4a7b","doc/macintosh.txt":"57491e53866711b4672d9b9ff35380b9dac9e0d8e3d6c20bdd6140603687c023","doc/replacement.txt":"4b6c3bbd7999d9d4108a281594bd02d13607e334a95465afff8c2c08d395f0e4","doc/windows-1250.txt":"61296bb6a21cdab602300d32ecfba434cb82de5ac3bc88d58710d2f125e28d39","doc/windows-1251.txt":"7deea1c61dea1485c8ff02db2c7d578db7a9aab63ab1cfd02ec04b515864689e","doc/windows-1252.txt":"933ef3bdddfce5ee132b9f1a1aa8b47423d2587bbe475b19028d0a6d38e180b6","doc/windows-1253.txt":"1a38748b88e99071a5c7b3d5456ead4caedeabab50d50d658be105bc113714de","doc/windows-1254.txt":"f8372f86c6f8d642563cd6ddc025260553292a39423df1683a98670bd7bf2b47","doc/windows-1255.txt":"4e5852494730054e2da258a74e1b9d780abbcdd8ce22ebc218ca2efe9e90493d","doc/windows-1256.txt":"c0879c5172abedead302a406e8f60d9cd9598694a0ffa4fd288ffe4fef7b8ea1","doc/windows-1257.txt":"c28a0c9f964fcb2b46d21f537c402446501a2800670481d6abf9fd9e9018d523","doc/windows-1258.txt":"5019ae4d61805c79aacbf17c93793342dbb098d65a1837783bc3e2c6d6a23602","doc/windows-874.txt":"4ef0e4501c5feba8b17aee1818602ed44b36ca8475db771ce2fc16d392cabecc","doc/x-mac-cyrillic.txt":"58be154d8a888ca3d484b83b44f749823ef339ab27f14d90ca9a856f5050a8bd","doc/x-user-defined.txt":"f9cd07c4321bf5cfb0be4bdddd251072999b04a6cf7a6f5bc63709a84e2c1ffc","generate-encoding-data.py":"be989dd25c6b946e3e8745fdc8e8a80fcf24b3be99ad0b4b78153ba3f6ab6310","rustfmt.toml":"85c1a3b4382fd89e991cbb81b70fb52780472edc064c963943cdaaa56e0a2030","src/ascii.rs":"588e38b01e666d5e7462617ea7e90a108d608dec9e016f3d273ac0744af2e05d","src/big5.rs":"ec6e2913011a38e9a3e825a1731f139a7ca1d5b264fefae51a3cc1a68a57cef9","src/data.rs":"8a617cc57032092d65850eb27e00de687c80aea3299e839a1f58b42d0b35abf3","src/euc_jp.rs":"32047f5b540188c4cb19c07165f846b9786a09f18e315ed3e9bda1293dae52aa","src/euc_kr.rs":"9b25afc72d9378700eecfac58d55ad1c5946d6cd0ccde2c29c08200ef2de6bb9","src/gb18030.rs":"808587168d73f0c80f8520f0ca9b161866ed2efeb17a05e85fdf3b8efe7ba28a","src/handles.rs":"b08cef1f5785bb6a4822f2e844c6df1b046b737b7a075e4593eaa8c4208e9fe2","src/iso_2022_jp.rs":"9bb485e82574f4b7d4b2364f0ff276acb6a0bc111758420a3b0ec5e04c196652","src/lib.rs":"834f44b670ec48ee82c0e12223d1567313fdd9f88bca5f4b117c82c1828f559f","src/macros.rs":"200997f8870de8bfd8cdc475e92115df42108c0df661e49d3d1cbc32056e1d99","src/mem.rs":"948571137d3b151df8db4fb2c733e74ae595d055cdf0ad83abcab9341d6adabe","src/replacement.rs":"7660b34a53f8c1ca2bdfa0e51e843ec28326950952ad8bc96569feb93ac62308","src/shift_jis.rs":"6951ae67e36b1a12fa3a30734957f444d8b1b4ae0e2bde52060b29bd0f16d9d9","src/simd_funcs.rs":"05c6e77af74bfe73cd39a752067c11425d6b46e5da419910f54bf75a5c02a984","src/single_byte.rs":"3ad87116fb339434a4b58e8f2b15485f2b66b9f7814d708f16194ed08f6d6ccf","src/test_data/big5_in.txt":"4c5a8691f8dc717311889c63894026d2fb62725a86c4208ca274a9cc8d42a503","src/test_data/big5_in_ref.txt":"99d399e17750cf9c7cf30bb253dbfe35b81c4fcbdead93cfa48b1429213473c7","src/test_data/big5_out.txt":"6193ca97c297aa20e09396038d18e938bb7ea331c26f0f2454097296723a0b13","src/test_data/big5_out_ref.txt":"36567691f557df144f6cc520015a87038dfa156f296fcf103b56ae9a718be1fc","src/test_data/euc_kr_in.txt":"c86a7224f3215fa0d04e685622a752fdc72763e8ae076230c7fd62de57ec4074","src/test_data/euc_kr_in_ref.txt":"1f419f4ca47d708b54c73c461545a022ae2e20498fdbf8005a483d752a204883","src/test_data/euc_kr_out.txt":"e7f32e026f70be1e1b58e0047baf7d3d2c520269c4f9b9992e158b4decb0a1a3","src/test_data/euc_kr_out_ref.txt":"c9907857980b20b8e9e3b584482ed6567a2be6185d72237b6322f0404944924e","src/test_data/gb18030_in.txt":"ab7231b2d3e9afacdbd7d7f3b9e5361a7ff9f7e1cfdb4f3bd905b9362b309e53","src/test_data/gb18030_in_ref.txt":"dc5069421adca2043c55f5012b55a76fdff651d22e6e699fd0978f8d5706815c","src/test_data/gb18030_out.txt":"f0208d527f5ca63de7d9a0323be8d5cf12d8a104b2943d92c2701f0c3364dac1","src/test_data/gb18030_out_ref.txt":"6819fe47627e4ea01027003fc514b9f21a1322e732d7f1fb92cc6c5455bc6c07","src/test_data/iso_2022_jp_in.txt":"cd24bbdcb1834e25db54646fbf4c41560a13dc7540f6be3dba4f5d97d44513af","src/test_data/iso_2022_jp_in_ref.txt":"3dc4e6a5e06471942d086b16c9440945e78415f6f3f47e43717e4bc2eac2cdf5","src/test_data/iso_2022_jp_out.txt":"9b6f015329dda6c3f9ee5ce6dbd6fa9c89acc21283e886836c78b8d833480c21","src/test_data/iso_2022_jp_out_ref.txt":"78cb260093a20116ad9a42f43b05d1848c5ab100b6b9a850749809e943884b35","src/test_data/jis0208_in.txt":"6df3030553ffb0a6615bb33dc8ea9dca6d9623a9028e2ffec754ce3c3da824cc","src/test_data/jis0208_in_ref.txt":"3dc4e6a5e06471942d086b16c9440945e78415f6f3f47e43717e4bc2eac2cdf5","src/test_data/jis0208_out.txt":"4ec24477e1675ce750733bdc3c5add1cd27b6bd4ce1f09289564646e9654e857","src/test_data/jis0208_out_ref.txt":"c3e1cef5032b2b1d93a406f31ff940c4e2dfe8859b8b17ca2761fee7a75a0e48","src/test_data/jis0212_in.txt":"c011f0dd72bd7c8cd922df9374ef8d2769a77190514c77f6c62b415852eeb9fe","src/test_data/jis0212_in_ref.txt":"7d9458b3d2f73e7092a7f505c08ce1d233dde18aa679fbcf9889256239cc9e06","src/test_data/shift_jis_in.txt":"02e389ccef0dd2122e63f503899402cb7f797912c2444cc80ab93131116c5524","src/test_data/shift_jis_in_ref.txt":"512f985950ca902e643c88682dba9708b7c38d3c5ec2925168ab00ac94ab19f9","src/test_data/shift_jis_out.txt":"5fbc44da7bf639bf6cfe0fa1fd3eba7102b88f81919c9ea991302712f69426fb","src/test_data/shift_jis_out_ref.txt":"466322c6fed8286c64582731755290c2296508efdd258826e6279686649b481f","src/test_labels_names.rs":"23a2e11b02b3b8d15fb5613a625e3edb2c61e70e3c581abfd638719a4088200d","src/testing.rs":"f59e671e95a98a56f6b573e8c6be4d71e670bf52f7e20eb1605d990aafa1894e","src/utf_16.rs":"c071a147fad38d750c2c247e141b76b929a48007b99f26b2922b9caecdaf2f25","src/utf_8.rs":"7b7d887b347f1aefa03246b028a36a72758a4ce76c28f3b45c19467851aa7839","src/variant.rs":"1fab5363588a1554a7169de8731ea9cded7ac63ea35caabdd1c27a8dde68c27b","src/x_user_defined.rs":"9456ca46168ef86c98399a2536f577ef7be3cdde90c0c51392d8ac48519d3fae"},"package":"b45de904aa0b010bce2ab45264d0631681847fa7b6f2eaa7dab7619943bc4f59"} \ No newline at end of file +{"files":{"CONTRIBUTING.md":"ca1901f3e8532fb4cec894fd3664f0eaa898c0c4b961d1b992d1ed54eacf362a","COPYRIGHT":"11789f45bb180841cd362a5eee6789c68ddb573a11105e30768c308a6add0190","Cargo.toml":"42fa83322aa9fd6723b77d35d0cacb92cbb6e7f573ce11c55f5225292866f8f4","Ideas.md":"b7452893f500163868d8de52c09addaf91e1632454ed02e892c467ed7ec39dbd","LICENSE-APACHE":"cfc7749b96f63bd31c3c42b5c471bf756814053e847c10f3eb003417bc523d30","LICENSE-MIT":"3fa4ca83dcc9237839b1bdeb2e6d16bdfb5ec0c5ce42b24694d8bbf0dcbef72c","LICENSE-WHATWG":"838118388fe5c2e7f1dbbaeed13e1c7f3ebf88be91319c7c1d77c18e987d1a50","README.md":"d938e8ab0b9ab67e74a1a4f48f23fdce956d0ad3a3f6147ae7612a92763c88d5","ci/miri.sh":"43cb8d82f49e3bfe2d2274b6ccd6f0714a4188ccef0cecc040829883cfdbee25","doc/Big5.txt":"f73a2edc5cb6c2d140ba6e07f4542e1c4a234950378acde1df93480f0ca0be0b","doc/EUC-JP.txt":"ee2818b907d0137f40a9ab9fd525fc700a44dbdddb6cf0c157a656566bae4bf1","doc/EUC-KR.txt":"71d9e2ccf3b124e8bdfb433c8cf2773fd878077038d0cec3c7237a50f4a78a30","doc/GBK.txt":"c1b522b5a799884e5001da661f42c5a8f4d0acb9ef1d74b206f22b5f65365606","doc/IBM866.txt":"a5a433e804d0f83af785015179fbc1d9b0eaf1f7960efcd04093e136b51fbd0e","doc/ISO-2022-JP.txt":"af86684f5a8f0e2868d7b2c292860140c3d2e5527530ca091f1b28198e8e2fe6","doc/ISO-8859-10.txt":"6d3949ad7c81ca176895101ed81a1db7df1060d64e262880b94bd31bb344ab4d","doc/ISO-8859-13.txt":"3951dd89cf93f7729148091683cf8511f4529388b7dc8dcd0d62eaed55be93fa","doc/ISO-8859-14.txt":"3d330784a0374fd255a38b47949675cc7168c800530534b0a01cac6edc623adc","doc/ISO-8859-15.txt":"24b1084aab5127a85aab99153f86e24694d0a3615f53b5ce23683f97cf66c47a","doc/ISO-8859-16.txt":"ce0272559b92ba76d7a7e476f6424ae4a5cc72e75b183611b08392e44add4d25","doc/ISO-8859-2.txt":"18ceff88c13d1b5ba455a3919b1e3de489045c4c3d2dd7e8527c125c75d54aad","doc/ISO-8859-3.txt":"21798404c68f4f5db59223362f24999da96968c0628427321fccce7d2849a130","doc/ISO-8859-4.txt":"d27f6520c6c5bfbcc19176b71d081cdb3bccde1622bb3e420d5680e812632d53","doc/ISO-8859-5.txt":"a10ec8d6ea7a78ad15da7275f6cb1a3365118527e28f9af6d0d5830501303f3a","doc/ISO-8859-6.txt":"ccda8a2efc96115336bdd77776637b9712425e44fbcf745353b9057fbef144e7","doc/ISO-8859-7.txt":"17900fa1f27a445958f0a77d7d9056be375a6bd7ee4492aa680c7c1500bab85e","doc/ISO-8859-8-I.txt":"8357555646d54265a9b9ffa3e68b08d132312f1561c60108ff9b8b1167b6ecf2","doc/ISO-8859-8.txt":"72cd6f3afb7b4a9c16a66a362473315770b7755d72c86c870e52fc3eba86c8af","doc/KOI8-R.txt":"839cf19a38da994488004ed7814b1f6151640156a9a2af02bf2efca745fb5966","doc/KOI8-U.txt":"0cc76624ed1f024183e2298b7e019957da2c70c8ca06e0fc4e6f353f50a5054f","doc/Shift_JIS.txt":"34c49141818cb9ddbcf59cc858f78a79be8ad148d563f26415108ae1f148443f","doc/UTF-16BE.txt":"e2e280d8acbaa6d2a6b3569d60e17500a285f2baa0df3363dd85537cd5a1ef8f","doc/UTF-16LE.txt":"70bdc170e3fc5298ba68f10125fb5eeb8b077036cc96bb4416c4de396f6d76c1","doc/UTF-8.txt":"ea7bae742e613010ced002cf4b601a737d2203fad65e115611451bc4428f548a","doc/gb18030.txt":"dc71378a8f07a2d8659f69ee81fb8791fef56ba86f124b429978285237bb4a7b","doc/macintosh.txt":"57491e53866711b4672d9b9ff35380b9dac9e0d8e3d6c20bdd6140603687c023","doc/replacement.txt":"4b6c3bbd7999d9d4108a281594bd02d13607e334a95465afff8c2c08d395f0e4","doc/windows-1250.txt":"61296bb6a21cdab602300d32ecfba434cb82de5ac3bc88d58710d2f125e28d39","doc/windows-1251.txt":"7deea1c61dea1485c8ff02db2c7d578db7a9aab63ab1cfd02ec04b515864689e","doc/windows-1252.txt":"933ef3bdddfce5ee132b9f1a1aa8b47423d2587bbe475b19028d0a6d38e180b6","doc/windows-1253.txt":"1a38748b88e99071a5c7b3d5456ead4caedeabab50d50d658be105bc113714de","doc/windows-1254.txt":"f8372f86c6f8d642563cd6ddc025260553292a39423df1683a98670bd7bf2b47","doc/windows-1255.txt":"4e5852494730054e2da258a74e1b9d780abbcdd8ce22ebc218ca2efe9e90493d","doc/windows-1256.txt":"c0879c5172abedead302a406e8f60d9cd9598694a0ffa4fd288ffe4fef7b8ea1","doc/windows-1257.txt":"c28a0c9f964fcb2b46d21f537c402446501a2800670481d6abf9fd9e9018d523","doc/windows-1258.txt":"5019ae4d61805c79aacbf17c93793342dbb098d65a1837783bc3e2c6d6a23602","doc/windows-874.txt":"4ef0e4501c5feba8b17aee1818602ed44b36ca8475db771ce2fc16d392cabecc","doc/x-mac-cyrillic.txt":"58be154d8a888ca3d484b83b44f749823ef339ab27f14d90ca9a856f5050a8bd","doc/x-user-defined.txt":"f9cd07c4321bf5cfb0be4bdddd251072999b04a6cf7a6f5bc63709a84e2c1ffc","generate-encoding-data.py":"be989dd25c6b946e3e8745fdc8e8a80fcf24b3be99ad0b4b78153ba3f6ab6310","rustfmt.toml":"85c1a3b4382fd89e991cbb81b70fb52780472edc064c963943cdaaa56e0a2030","src/ascii.rs":"c44c002641adb5ebc4368707a8cc0a076d2f33e6a5c27b1b69988eb515f5653d","src/big5.rs":"ec6e2913011a38e9a3e825a1731f139a7ca1d5b264fefae51a3cc1a68a57cef9","src/data.rs":"8a617cc57032092d65850eb27e00de687c80aea3299e839a1f58b42d0b35abf3","src/euc_jp.rs":"32047f5b540188c4cb19c07165f846b9786a09f18e315ed3e9bda1293dae52aa","src/euc_kr.rs":"9b25afc72d9378700eecfac58d55ad1c5946d6cd0ccde2c29c08200ef2de6bb9","src/gb18030.rs":"808587168d73f0c80f8520f0ca9b161866ed2efeb17a05e85fdf3b8efe7ba28a","src/handles.rs":"cc83dc0754751d67f5688a65c5e0191cba02f6bacce81a0813a243cba55eef7a","src/iso_2022_jp.rs":"9bb485e82574f4b7d4b2364f0ff276acb6a0bc111758420a3b0ec5e04c196652","src/lib.rs":"1dc07b818e45846b16ddcaf0de46c8862dd7df8099123ec38b95c3f8ad9c91ec","src/macros.rs":"200997f8870de8bfd8cdc475e92115df42108c0df661e49d3d1cbc32056e1d99","src/mem.rs":"0bf34103e0ad1b842a13a082dee2b920b05cf4fb0f145c9ee7f608f4cb4a544f","src/replacement.rs":"7660b34a53f8c1ca2bdfa0e51e843ec28326950952ad8bc96569feb93ac62308","src/shift_jis.rs":"6951ae67e36b1a12fa3a30734957f444d8b1b4ae0e2bde52060b29bd0f16d9d9","src/simd_funcs.rs":"2612aba86e1d201096d7e47a859bc3444f85934cc82d8adc6d39a4304d9eecfc","src/single_byte.rs":"3c9e9c1f946ae622c725ba9421240c1faa9a05e95fa10dd4642a25cb276a1edc","src/test_data/big5_in.txt":"4c5a8691f8dc717311889c63894026d2fb62725a86c4208ca274a9cc8d42a503","src/test_data/big5_in_ref.txt":"99d399e17750cf9c7cf30bb253dbfe35b81c4fcbdead93cfa48b1429213473c7","src/test_data/big5_out.txt":"6193ca97c297aa20e09396038d18e938bb7ea331c26f0f2454097296723a0b13","src/test_data/big5_out_ref.txt":"36567691f557df144f6cc520015a87038dfa156f296fcf103b56ae9a718be1fc","src/test_data/euc_kr_in.txt":"c86a7224f3215fa0d04e685622a752fdc72763e8ae076230c7fd62de57ec4074","src/test_data/euc_kr_in_ref.txt":"1f419f4ca47d708b54c73c461545a022ae2e20498fdbf8005a483d752a204883","src/test_data/euc_kr_out.txt":"e7f32e026f70be1e1b58e0047baf7d3d2c520269c4f9b9992e158b4decb0a1a3","src/test_data/euc_kr_out_ref.txt":"c9907857980b20b8e9e3b584482ed6567a2be6185d72237b6322f0404944924e","src/test_data/gb18030_in.txt":"ab7231b2d3e9afacdbd7d7f3b9e5361a7ff9f7e1cfdb4f3bd905b9362b309e53","src/test_data/gb18030_in_ref.txt":"dc5069421adca2043c55f5012b55a76fdff651d22e6e699fd0978f8d5706815c","src/test_data/gb18030_out.txt":"f0208d527f5ca63de7d9a0323be8d5cf12d8a104b2943d92c2701f0c3364dac1","src/test_data/gb18030_out_ref.txt":"6819fe47627e4ea01027003fc514b9f21a1322e732d7f1fb92cc6c5455bc6c07","src/test_data/iso_2022_jp_in.txt":"cd24bbdcb1834e25db54646fbf4c41560a13dc7540f6be3dba4f5d97d44513af","src/test_data/iso_2022_jp_in_ref.txt":"3dc4e6a5e06471942d086b16c9440945e78415f6f3f47e43717e4bc2eac2cdf5","src/test_data/iso_2022_jp_out.txt":"9b6f015329dda6c3f9ee5ce6dbd6fa9c89acc21283e886836c78b8d833480c21","src/test_data/iso_2022_jp_out_ref.txt":"78cb260093a20116ad9a42f43b05d1848c5ab100b6b9a850749809e943884b35","src/test_data/jis0208_in.txt":"6df3030553ffb0a6615bb33dc8ea9dca6d9623a9028e2ffec754ce3c3da824cc","src/test_data/jis0208_in_ref.txt":"3dc4e6a5e06471942d086b16c9440945e78415f6f3f47e43717e4bc2eac2cdf5","src/test_data/jis0208_out.txt":"4ec24477e1675ce750733bdc3c5add1cd27b6bd4ce1f09289564646e9654e857","src/test_data/jis0208_out_ref.txt":"c3e1cef5032b2b1d93a406f31ff940c4e2dfe8859b8b17ca2761fee7a75a0e48","src/test_data/jis0212_in.txt":"c011f0dd72bd7c8cd922df9374ef8d2769a77190514c77f6c62b415852eeb9fe","src/test_data/jis0212_in_ref.txt":"7d9458b3d2f73e7092a7f505c08ce1d233dde18aa679fbcf9889256239cc9e06","src/test_data/shift_jis_in.txt":"02e389ccef0dd2122e63f503899402cb7f797912c2444cc80ab93131116c5524","src/test_data/shift_jis_in_ref.txt":"512f985950ca902e643c88682dba9708b7c38d3c5ec2925168ab00ac94ab19f9","src/test_data/shift_jis_out.txt":"5fbc44da7bf639bf6cfe0fa1fd3eba7102b88f81919c9ea991302712f69426fb","src/test_data/shift_jis_out_ref.txt":"466322c6fed8286c64582731755290c2296508efdd258826e6279686649b481f","src/test_labels_names.rs":"23a2e11b02b3b8d15fb5613a625e3edb2c61e70e3c581abfd638719a4088200d","src/testing.rs":"f59e671e95a98a56f6b573e8c6be4d71e670bf52f7e20eb1605d990aafa1894e","src/utf_16.rs":"c071a147fad38d750c2c247e141b76b929a48007b99f26b2922b9caecdaf2f25","src/utf_8.rs":"7b7d887b347f1aefa03246b028a36a72758a4ce76c28f3b45c19467851aa7839","src/variant.rs":"1fab5363588a1554a7169de8731ea9cded7ac63ea35caabdd1c27a8dde68c27b","src/x_user_defined.rs":"c9c010730dfb9f141d4fed19350c08a21af240913a54bb64f5ca89ff93b6b7d1"},"package":"7268b386296a025e474d5140678f75d6de9493ae55a5d709eeb9dd08149945e1"} \ No newline at end of file diff --git a/third_party/rust/encoding_rs/Cargo.toml b/third_party/rust/encoding_rs/Cargo.toml index 08bb61136b79..e9fd6c0af257 100644 --- a/third_party/rust/encoding_rs/Cargo.toml +++ b/third_party/rust/encoding_rs/Cargo.toml @@ -11,9 +11,8 @@ [package] edition = "2018" -rust-version = "1.36" name = "encoding_rs" -version = "0.8.34" +version = "0.8.33" authors = ["Henri Sivonen "] description = "A Gecko-oriented implementation of the Encoding Standard" homepage = "https://docs.rs/encoding_rs/" @@ -37,13 +36,13 @@ repository = "https://github.com/hsivonen/encoding_rs" [profile.release] lto = true -[dependencies.any_all_workaround] -version = "0.1.0" -optional = true - [dependencies.cfg-if] version = "1.0" +[dependencies.packed_simd] +version = "0.3.9" +optional = true + [dependencies.serde] version = "1.0" optional = true @@ -75,4 +74,10 @@ fast-legacy-encode = [ less-slow-big5-hanzi-encode = [] less-slow-gb-hanzi-encode = [] less-slow-kanji-encode = [] -simd-accel = ["any_all_workaround"] +simd-accel = [ + "packed_simd", + "packed_simd/into_bits", +] + +[badges.travis-ci] +repository = "hsivonen/encoding_rs" diff --git a/third_party/rust/encoding_rs/README.md b/third_party/rust/encoding_rs/README.md index ed0e954630f7..c0b596c95199 100644 --- a/third_party/rust/encoding_rs/README.md +++ b/third_party/rust/encoding_rs/README.md @@ -167,15 +167,13 @@ There are currently these optional cargo features: ### `simd-accel` -Enables SIMD acceleration using the nightly-dependent `portable_simd` standard -library feature. +Enables SIMD acceleration using the nightly-dependent `packed_simd` crate. This is an opt-in feature, because enabling this feature _opts out_ of Rust's guarantees of future compilers compiling old code (aka. "stability story"). Currently, this has not been tested to be an improvement except for these -targets and enabling the `simd-accel` feature is expected to break the build -on other targets: +targets: * x86_64 * i686 @@ -187,6 +185,22 @@ above, and you are prepared _to have to revise your configuration when updating Rust_, you should enable this feature. Otherwise, please _do not_ enable this feature. +_Note!_ If you are compiling for a target that does not have 128-bit SIMD +enabled as part of the target definition and you are enabling 128-bit SIMD +using `-C target_feature`, you need to enable the `core_arch` Cargo feature +for `packed_simd` to compile a crates.io snapshot of `core_arch` instead of +using the standard-library copy of `core::arch`, because the `core::arch` +module of the pre-compiled standard library has been compiled with the +assumption that the CPU doesn't have 128-bit SIMD. At present this applies +mainly to 32-bit ARM targets whose first component does not include the +substring `neon`. + +The encoding_rs side of things has not been properly set up for POWER, +PowerPC, MIPS, etc., SIMD at this time, so even if you were to follow +the advice from the previous paragraph, you probably shouldn't use +the `simd-accel` option on the less mainstream architectures at this +time. + Used by Firefox. ### `serde` @@ -367,9 +381,8 @@ as semver-breaking, because this crate depends on `cfg-if`, which doesn't appear to treat MSRV changes as semver-breaking, so it would be useless for this crate to treat MSRV changes as semver-breaking. -As of 2024-04-04, MSRV appears to be Rust 1.36.0 for using the crate and +As of 2021-02-04, MSRV appears to be Rust 1.36.0 for using the crate and 1.42.0 for doc tests to pass without errors about the global allocator. -With the `simd-accel` feature, the MSRV is even higher. ## Compatibility with rust-encoding @@ -433,17 +446,10 @@ To regenerate the generated code: - [x] Add actually fast CJK encode options. - [ ] ~Investigate [Bob Steagall's lookup table acceleration for UTF-8](https://github.com/BobSteagall/CppNow2018/blob/master/FastConversionFromUTF-8/Fast%20Conversion%20From%20UTF-8%20with%20C%2B%2B%2C%20DFAs%2C%20and%20SSE%20Intrinsics%20-%20Bob%20Steagall%20-%20C%2B%2BNow%202018.pdf).~ - [x] Provide a build mode that works without `alloc` (with lesser API surface). -- [x] Migrate to `std::simd` ~once it is stable and declare 1.0.~ -- [ ] Migrate `unsafe` slice access by larger types than `u8`/`u16` to `align_to`. +- [ ] Migrate to `std::simd` once it is stable and declare 1.0. ## Release Notes -### 0.8.34 - -* Use the `portable_simd` nightly feature of the standard library instead of the `packed_simd` crate. Only affects the `simd-accel` optional nightly feature. -* Internal documentation improvements and minor code improvements around `unsafe`. -* Added `rust-version` to `Cargo.toml`. - ### 0.8.33 * Use `packed_simd` instead of `packed_simd_2` again now that updates are back under the `packed_simd` name. Only affects the `simd-accel` optional nightly feature. diff --git a/third_party/rust/encoding_rs/src/ascii.rs b/third_party/rust/encoding_rs/src/ascii.rs index 80233f285e7a..90644de7a474 100644 --- a/third_party/rust/encoding_rs/src/ascii.rs +++ b/third_party/rust/encoding_rs/src/ascii.rs @@ -51,8 +51,6 @@ cfg_if! { } } -// Safety invariants for masks: data & mask = 0 for valid ASCII or basic latin utf-16 - // `as` truncates, so works on 32-bit, too. #[allow(dead_code)] pub const ASCII_MASK: usize = 0x8080_8080_8080_8080u64 as usize; @@ -64,9 +62,6 @@ pub const BASIC_LATIN_MASK: usize = 0xFF80_FF80_FF80_FF80u64 as usize; #[allow(unused_macros)] macro_rules! ascii_naive { ($name:ident, $src_unit:ty, $dst_unit:ty) => { - /// Safety: src and dst must have len_unit elements and be aligned - /// Safety-usable invariant: will return Some() when it fails - /// to convert. The first value will be a u8 that is > 127. #[inline(always)] pub unsafe fn $name( src: *const $src_unit, @@ -76,13 +71,10 @@ macro_rules! ascii_naive { // Yes, manually omitting the bound check here matters // a lot for perf. for i in 0..len { - // Safety: len invariant used here let code_unit = *(src.add(i)); - // Safety: Upholds safety-usable invariant here if code_unit > 127 { return Some((code_unit, i)); } - // Safety: len invariant used here *(dst.add(i)) = code_unit as $dst_unit; } return None; @@ -93,15 +85,9 @@ macro_rules! ascii_naive { #[allow(unused_macros)] macro_rules! ascii_alu { ($name:ident, - // safety invariant: src/dst MUST be u8 $src_unit:ty, $dst_unit:ty, - // Safety invariant: stride_fn must consume and produce two usizes, and return the index of the first non-ascii when it fails $stride_fn:ident) => { - /// Safety: src and dst must have len elements, src is valid for read, dst is valid for - /// write - /// Safety-usable invariant: will return Some() when it fails - /// to convert. The first value will be a u8 that is > 127. #[cfg_attr(feature = "cargo-clippy", allow(never_loop, cast_ptr_alignment))] #[inline(always)] pub unsafe fn $name( @@ -112,7 +98,6 @@ macro_rules! ascii_alu { let mut offset = 0usize; // This loop is only broken out of as a `goto` forward loop { - // Safety: until_alignment becomes the number of bytes we need to munch until we are aligned to usize let mut until_alignment = { // Check if the other unit aligns if we move the narrower unit // to alignment. @@ -121,7 +106,6 @@ macro_rules! ascii_alu { let src_alignment = (src as usize) & ALU_ALIGNMENT_MASK; let dst_alignment = (dst as usize) & ALU_ALIGNMENT_MASK; if src_alignment != dst_alignment { - // Safety: bails early and ends up in the naïve branch where usize-alignment doesn't matter break; } (ALU_ALIGNMENT - src_alignment) & ALU_ALIGNMENT_MASK @@ -150,40 +134,25 @@ macro_rules! ascii_alu { // x86_64 should be using SSE2 in due course, keeping the move // to alignment here. It would be good to test on more ARM CPUs // and on real MIPS and POWER hardware. - // - // Safety: This is the naïve code once again, for `until_alignment` bytes while until_alignment != 0 { let code_unit = *(src.add(offset)); if code_unit > 127 { - // Safety: Upholds safety-usable invariant here return Some((code_unit, offset)); } *(dst.add(offset)) = code_unit as $dst_unit; - // Safety: offset is the number of bytes copied so far offset += 1; until_alignment -= 1; } let len_minus_stride = len - ALU_STRIDE_SIZE; loop { - // Safety: num_ascii is known to be a byte index of a non-ascii byte due to stride_fn's invariant if let Some(num_ascii) = $stride_fn( - // Safety: These are known to be valid and aligned since we have at - // least ALU_STRIDE_SIZE data in these buffers, and offset is the - // number of elements copied so far, which according to the - // until_alignment calculation above will cause both src and dst to be - // aligned to usize after this add src.add(offset) as *const usize, dst.add(offset) as *mut usize, ) { offset += num_ascii; - // Safety: Upholds safety-usable invariant here by indexing into non-ascii byte return Some((*(src.add(offset)), offset)); } - // Safety: offset continues to be the number of bytes copied so far, and - // maintains usize alignment for the next loop iteration offset += ALU_STRIDE_SIZE; - // Safety: This is `offset > len - stride. This loop will continue as long as - // `offset <= len - stride`, which means there are `stride` bytes to still be read. if offset > len_minus_stride { break; } @@ -191,17 +160,11 @@ macro_rules! ascii_alu { } break; } - - // Safety: This is the naïve code, same as ascii_naive, and has no requirements - // other than src/dst being valid for the the right lens while offset < len { - // Safety: len invariant used here let code_unit = *(src.add(offset)); if code_unit > 127 { - // Safety: Upholds safety-usable invariant here return Some((code_unit, offset)); } - // Safety: len invariant used here *(dst.add(offset)) = code_unit as $dst_unit; offset += 1; } @@ -213,16 +176,9 @@ macro_rules! ascii_alu { #[allow(unused_macros)] macro_rules! basic_latin_alu { ($name:ident, - // safety invariant: use u8 for src/dest for ascii, and u16 for basic_latin $src_unit:ty, $dst_unit:ty, - // safety invariant: stride function must munch ALU_STRIDE_SIZE*size(src_unit) bytes off of src and - // write ALU_STRIDE_SIZE*size(dst_unit) bytes to dst $stride_fn:ident) => { - /// Safety: src and dst must have len elements, src is valid for read, dst is valid for - /// write - /// Safety-usable invariant: will return Some() when it fails - /// to convert. The first value will be a u8 that is > 127. #[cfg_attr( feature = "cargo-clippy", allow(never_loop, cast_ptr_alignment, cast_lossless) @@ -236,8 +192,6 @@ macro_rules! basic_latin_alu { let mut offset = 0usize; // This loop is only broken out of as a `goto` forward loop { - // Safety: until_alignment becomes the number of bytes we need to munch from src/dest until we are aligned to usize - // We ensure basic-latin has the same alignment as ascii, starting with ascii since it is smaller. let mut until_alignment = { // Check if the other unit aligns if we move the narrower unit // to alignment. @@ -283,37 +237,24 @@ macro_rules! basic_latin_alu { // x86_64 should be using SSE2 in due course, keeping the move // to alignment here. It would be good to test on more ARM CPUs // and on real MIPS and POWER hardware. - // - // Safety: This is the naïve code once again, for `until_alignment` bytes while until_alignment != 0 { let code_unit = *(src.add(offset)); if code_unit > 127 { - // Safety: Upholds safety-usable invariant here return Some((code_unit, offset)); } *(dst.add(offset)) = code_unit as $dst_unit; - // Safety: offset is the number of bytes copied so far offset += 1; until_alignment -= 1; } let len_minus_stride = len - ALU_STRIDE_SIZE; loop { if !$stride_fn( - // Safety: These are known to be valid and aligned since we have at - // least ALU_STRIDE_SIZE data in these buffers, and offset is the - // number of elements copied so far, which according to the - // until_alignment calculation above will cause both src and dst to be - // aligned to usize after this add src.add(offset) as *const usize, dst.add(offset) as *mut usize, ) { break; } - // Safety: offset continues to be the number of bytes copied so far, and - // maintains usize alignment for the next loop iteration offset += ALU_STRIDE_SIZE; - // Safety: This is `offset > len - stride. This loop will continue as long as - // `offset <= len - stride`, which means there are `stride` bytes to still be read. if offset > len_minus_stride { break; } @@ -321,15 +262,11 @@ macro_rules! basic_latin_alu { } break; } - // Safety: This is the naïve code once again, for leftover bytes while offset < len { - // Safety: len invariant used here let code_unit = *(src.add(offset)); if code_unit > 127 { - // Safety: Upholds safety-usable invariant here return Some((code_unit, offset)); } - // Safety: len invariant used here *(dst.add(offset)) = code_unit as $dst_unit; offset += 1; } @@ -340,11 +277,7 @@ macro_rules! basic_latin_alu { #[allow(unused_macros)] macro_rules! latin1_alu { - // safety invariant: stride function must munch ALU_STRIDE_SIZE*size(src_unit) bytes off of src and - // write ALU_STRIDE_SIZE*size(dst_unit) bytes to dst ($name:ident, $src_unit:ty, $dst_unit:ty, $stride_fn:ident) => { - /// Safety: src and dst must have len elements, src is valid for read, dst is valid for - /// write #[cfg_attr( feature = "cargo-clippy", allow(never_loop, cast_ptr_alignment, cast_lossless) @@ -354,8 +287,6 @@ macro_rules! latin1_alu { let mut offset = 0usize; // This loop is only broken out of as a `goto` forward loop { - // Safety: until_alignment becomes the number of bytes we need to munch from src/dest until we are aligned to usize - // We ensure the UTF-16 side has the same alignment as the Latin-1 side, starting with Latin-1 since it is smaller. let mut until_alignment = { if ::core::mem::size_of::<$src_unit>() < ::core::mem::size_of::<$dst_unit>() { // unpack @@ -382,30 +313,19 @@ macro_rules! latin1_alu { } }; if until_alignment + ALU_STRIDE_SIZE <= len { - // Safety: This is the naïve code once again, for `until_alignment` bytes while until_alignment != 0 { let code_unit = *(src.add(offset)); *(dst.add(offset)) = code_unit as $dst_unit; - // Safety: offset is the number of bytes copied so far offset += 1; until_alignment -= 1; } let len_minus_stride = len - ALU_STRIDE_SIZE; loop { $stride_fn( - // Safety: These are known to be valid and aligned since we have at - // least ALU_STRIDE_SIZE data in these buffers, and offset is the - // number of elements copied so far, which according to the - // until_alignment calculation above will cause both src and dst to be - // aligned to usize after this add src.add(offset) as *const usize, dst.add(offset) as *mut usize, ); - // Safety: offset continues to be the number of bytes copied so far, and - // maintains usize alignment for the next loop iteration offset += ALU_STRIDE_SIZE; - // Safety: This is `offset > len - stride. This loop will continue as long as - // `offset <= len - stride`, which means there are `stride` bytes to still be read. if offset > len_minus_stride { break; } @@ -413,9 +333,7 @@ macro_rules! latin1_alu { } break; } - // Safety: This is the naïve code once again, for leftover bytes while offset < len { - // Safety: len invariant used here let code_unit = *(src.add(offset)); *(dst.add(offset)) = code_unit as $dst_unit; offset += 1; @@ -430,19 +348,11 @@ macro_rules! ascii_simd_check_align { $name:ident, $src_unit:ty, $dst_unit:ty, - // Safety: This function must require aligned src/dest that are valid for reading/writing SIMD_STRIDE_SIZE src_unit/dst_unit $stride_both_aligned:ident, - // Safety: This function must require aligned/unaligned src/dest that are valid for reading/writing SIMD_STRIDE_SIZE src_unit/dst_unit $stride_src_aligned:ident, - // Safety: This function must require unaligned/aligned src/dest that are valid for reading/writing SIMD_STRIDE_SIZE src_unit/dst_unit $stride_dst_aligned:ident, - // Safety: This function must require unaligned src/dest that are valid for reading/writing SIMD_STRIDE_SIZE src_unit/dst_unit $stride_neither_aligned:ident ) => { - /// Safety: src/dst must be valid for reads/writes of `len` elements of their units. - /// - /// Safety-usable invariant: will return Some() when it encounters non-ASCII, with the first element in the Some being - /// guaranteed to be non-ASCII (> 127), and the second being the offset where it is found #[inline(always)] pub unsafe fn $name( src: *const $src_unit, @@ -450,7 +360,6 @@ macro_rules! ascii_simd_check_align { len: usize, ) -> Option<($src_unit, usize)> { let mut offset = 0usize; - // Safety: if this check succeeds we're valid for reading/writing at least `SIMD_STRIDE_SIZE` elements. if SIMD_STRIDE_SIZE <= len { let len_minus_stride = len - SIMD_STRIDE_SIZE; // XXX Should we first process one stride unconditionally as unaligned to @@ -459,29 +368,23 @@ macro_rules! ascii_simd_check_align { // on Haswell, it would make sense to just use unaligned and not bother // checking. Need to benchmark older architectures before deciding. let dst_masked = (dst as usize) & SIMD_ALIGNMENT_MASK; - // Safety: checking whether src is aligned if ((src as usize) & SIMD_ALIGNMENT_MASK) == 0 { - // Safety: Checking whether dst is aligned if dst_masked == 0 { loop { - // Safety: We're valid to read/write SIMD_STRIDE_SIZE elements and have the appropriate alignments if !$stride_both_aligned(src.add(offset), dst.add(offset)) { break; } offset += SIMD_STRIDE_SIZE; - // Safety: This is `offset > len - SIMD_STRIDE_SIZE` which means we always have at least `SIMD_STRIDE_SIZE` elements to munch next time. if offset > len_minus_stride { break; } } } else { loop { - // Safety: We're valid to read/write SIMD_STRIDE_SIZE elements and have the appropriate alignments if !$stride_src_aligned(src.add(offset), dst.add(offset)) { break; } offset += SIMD_STRIDE_SIZE; - // Safety: This is `offset > len - SIMD_STRIDE_SIZE` which means we always have at least `SIMD_STRIDE_SIZE` elements to munch next time. if offset > len_minus_stride { break; } @@ -490,24 +393,20 @@ macro_rules! ascii_simd_check_align { } else { if dst_masked == 0 { loop { - // Safety: We're valid to read/write SIMD_STRIDE_SIZE elements and have the appropriate alignments if !$stride_dst_aligned(src.add(offset), dst.add(offset)) { break; } offset += SIMD_STRIDE_SIZE; - // Safety: This is `offset > len - SIMD_STRIDE_SIZE` which means we always have at least `SIMD_STRIDE_SIZE` elements to munch next time. if offset > len_minus_stride { break; } } } else { loop { - // Safety: We're valid to read/write SIMD_STRIDE_SIZE elements and have the appropriate alignments if !$stride_neither_aligned(src.add(offset), dst.add(offset)) { break; } offset += SIMD_STRIDE_SIZE; - // Safety: This is `offset > len - SIMD_STRIDE_SIZE` which means we always have at least `SIMD_STRIDE_SIZE` elements to munch next time. if offset > len_minus_stride { break; } @@ -516,10 +415,8 @@ macro_rules! ascii_simd_check_align { } } while offset < len { - // Safety: uses len invariant here and below let code_unit = *(src.add(offset)); if code_unit > 127 { - // Safety: upholds safety-usable invariant return Some((code_unit, offset)); } *(dst.add(offset)) = code_unit as $dst_unit; @@ -536,21 +433,13 @@ macro_rules! ascii_simd_check_align_unrolled { $name:ident, $src_unit:ty, $dst_unit:ty, - // Safety: This function must require aligned src/dest that are valid for reading/writing SIMD_STRIDE_SIZE src_unit/dst_unit $stride_both_aligned:ident, - // Safety: This function must require aligned/unaligned src/dest that are valid for reading/writing SIMD_STRIDE_SIZE src_unit/dst_unit $stride_src_aligned:ident, - // Safety: This function must require unaligned src/dest that are valid for reading/writing SIMD_STRIDE_SIZE src_unit/dst_unit $stride_neither_aligned:ident, - // Safety: This function must require aligned src/dest that are valid for reading/writing 2*SIMD_STRIDE_SIZE src_unit/dst_unit $double_stride_both_aligned:ident, - // Safety: This function must require aligned/unaligned src/dest that are valid for reading/writing 2*SIMD_STRIDE_SIZE src_unit/dst_unit $double_stride_src_aligned:ident ) => { - /// Safety: src/dst must be valid for reads/writes of `len` elements of their units. - /// - /// Safety-usable invariant: will return Some() when it encounters non-ASCII, with the first element in the Some being - /// guaranteed to be non-ASCII (> 127), and the second being the offset where it is found #[inline(always)] + #[inline(always)] pub unsafe fn $name( src: *const $src_unit, dst: *mut $dst_unit, @@ -561,10 +450,8 @@ macro_rules! ascii_simd_check_align_unrolled { // This loop is only broken out of as a goto forward without // actually looping 'outer: loop { - // Safety: if this check succeeds we're valid for reading/writing at least `SIMD_STRIDE_SIZE` elements. if SIMD_STRIDE_SIZE <= len { // First, process one unaligned - // Safety: this is safe to call since we're valid for this read/write if !$stride_neither_aligned(src, dst) { break 'outer; } @@ -574,54 +461,37 @@ macro_rules! ascii_simd_check_align_unrolled { // there will be enough more to justify more expense // in the case of non-ASCII. // Use aligned reads for the sake of old microachitectures. - // - // Safety: this correctly calculates the number of src_units that need to be read before the remaining list is aligned. - // This is less that SIMD_ALIGNMENT, which is also SIMD_STRIDE_SIZE (as documented) let until_alignment = ((SIMD_ALIGNMENT - ((src.add(offset) as usize) & SIMD_ALIGNMENT_MASK)) & SIMD_ALIGNMENT_MASK) / unit_size; - // Safety: This addition won't overflow, because even in the 32-bit PAE case the + // This addition won't overflow, because even in the 32-bit PAE case the // address space holds enough code that the slice length can't be that // close to address space size. // offset now equals SIMD_STRIDE_SIZE, hence times 3 below. - // - // Safety: if this check succeeds we're valid for reading/writing at least `2 * SIMD_STRIDE_SIZE` elements plus `until_alignment`. - // The extra SIMD_STRIDE_SIZE in the condition is because `offset` is already `SIMD_STRIDE_SIZE`. if until_alignment + (SIMD_STRIDE_SIZE * 3) <= len { if until_alignment != 0 { - // Safety: this is safe to call since we're valid for this read/write (and more), and don't care about alignment - // This will copy over bytes that get decoded twice since it's not incrementing `offset` by SIMD_STRIDE_SIZE. This is fine. if !$stride_neither_aligned(src.add(offset), dst.add(offset)) { break; } offset += until_alignment; } - // Safety: At this point we're valid for reading/writing 2*SIMD_STRIDE_SIZE elements - // Safety: Now `offset` is aligned for `src` let len_minus_stride_times_two = len - (SIMD_STRIDE_SIZE * 2); - // Safety: This is whether dst is aligned let dst_masked = (dst.add(offset) as usize) & SIMD_ALIGNMENT_MASK; if dst_masked == 0 { loop { - // Safety: both are aligned, we can call the aligned function. We're valid for reading/writing double stride from the initial condition - // and the loop break condition below if let Some(advance) = $double_stride_both_aligned(src.add(offset), dst.add(offset)) { offset += advance; let code_unit = *(src.add(offset)); - // Safety: uses safety-usable invariant on ascii_to_ascii_simd_double_stride to return - // guaranteed non-ascii return Some((code_unit, offset)); } offset += SIMD_STRIDE_SIZE * 2; - // Safety: This is `offset > len - 2 * SIMD_STRIDE_SIZE` which means we always have at least `2 * SIMD_STRIDE_SIZE` elements to munch next time. if offset > len_minus_stride_times_two { break; } } - // Safety: We're valid for reading/writing one more, and can still assume alignment if offset + SIMD_STRIDE_SIZE <= len { if !$stride_both_aligned(src.add(offset), dst.add(offset)) { break 'outer; @@ -630,25 +500,18 @@ macro_rules! ascii_simd_check_align_unrolled { } } else { loop { - // Safety: only src is aligned here. We're valid for reading/writing double stride from the initial condition - // and the loop break condition below if let Some(advance) = $double_stride_src_aligned(src.add(offset), dst.add(offset)) { offset += advance; let code_unit = *(src.add(offset)); - // Safety: uses safety-usable invariant on ascii_to_ascii_simd_double_stride to return - // guaranteed non-ascii return Some((code_unit, offset)); } offset += SIMD_STRIDE_SIZE * 2; - // Safety: This is `offset > len - 2 * SIMD_STRIDE_SIZE` which means we always have at least `2 * SIMD_STRIDE_SIZE` elements to munch next time. - if offset > len_minus_stride_times_two { break; } } - // Safety: We're valid for reading/writing one more, and can still assume alignment if offset + SIMD_STRIDE_SIZE <= len { if !$stride_src_aligned(src.add(offset), dst.add(offset)) { break 'outer; @@ -659,13 +522,11 @@ macro_rules! ascii_simd_check_align_unrolled { } else { // At most two iterations, so unroll if offset + SIMD_STRIDE_SIZE <= len { - // Safety: The check above ensures we're allowed to read/write this, and we don't use alignment if !$stride_neither_aligned(src.add(offset), dst.add(offset)) { break; } offset += SIMD_STRIDE_SIZE; if offset + SIMD_STRIDE_SIZE <= len { - // Safety: The check above ensures we're allowed to read/write this, and we don't use alignment if !$stride_neither_aligned(src.add(offset), dst.add(offset)) { break; } @@ -677,10 +538,8 @@ macro_rules! ascii_simd_check_align_unrolled { break 'outer; } while offset < len { - // Safety: relies straightforwardly on the `len` invariant let code_unit = *(src.add(offset)); if code_unit > 127 { - // Safety-usable invariant upheld here return Some((code_unit, offset)); } *(dst.add(offset)) = code_unit as $dst_unit; @@ -697,45 +556,30 @@ macro_rules! latin1_simd_check_align { $name:ident, $src_unit:ty, $dst_unit:ty, - // Safety: This function must require aligned src/dest that are valid for reading/writing SIMD_STRIDE_SIZE src_unit/dst_unit $stride_both_aligned:ident, - // Safety: This function must require aligned/unaligned src/dest that are valid for reading/writing SIMD_STRIDE_SIZE src_unit/dst_unit $stride_src_aligned:ident, - // Safety: This function must require unaligned/aligned src/dest that are valid for reading/writing SIMD_STRIDE_SIZE src_unit/dst_unit $stride_dst_aligned:ident, - // Safety: This function must require unaligned src/dest that are valid for reading/writing SIMD_STRIDE_SIZE src_unit/dst_unit $stride_neither_aligned:ident - ) => { - /// Safety: src/dst must be valid for reads/writes of `len` elements of their units. #[inline(always)] pub unsafe fn $name(src: *const $src_unit, dst: *mut $dst_unit, len: usize) { let mut offset = 0usize; - // Safety: if this check succeeds we're valid for reading/writing at least `SIMD_STRIDE_SIZE` elements. if SIMD_STRIDE_SIZE <= len { let len_minus_stride = len - SIMD_STRIDE_SIZE; - // Whether dst is aligned let dst_masked = (dst as usize) & SIMD_ALIGNMENT_MASK; - // Whether src is aligned if ((src as usize) & SIMD_ALIGNMENT_MASK) == 0 { if dst_masked == 0 { loop { - // Safety: Both were aligned, we can use the aligned function $stride_both_aligned(src.add(offset), dst.add(offset)); offset += SIMD_STRIDE_SIZE; - // Safety: This is `offset > len - SIMD_STRIDE_SIZE`, which means in the next iteration we're valid for - // reading/writing at least SIMD_STRIDE_SIZE elements. if offset > len_minus_stride { break; } } } else { loop { - // Safety: src was aligned, dst was not $stride_src_aligned(src.add(offset), dst.add(offset)); offset += SIMD_STRIDE_SIZE; - // Safety: This is `offset > len - SIMD_STRIDE_SIZE`, which means in the next iteration we're valid for - // reading/writing at least SIMD_STRIDE_SIZE elements. if offset > len_minus_stride { break; } @@ -744,22 +588,16 @@ macro_rules! latin1_simd_check_align { } else { if dst_masked == 0 { loop { - // Safety: src was aligned, dst was not $stride_dst_aligned(src.add(offset), dst.add(offset)); offset += SIMD_STRIDE_SIZE; - // Safety: This is `offset > len - SIMD_STRIDE_SIZE`, which means in the next iteration we're valid for - // reading/writing at least SIMD_STRIDE_SIZE elements. if offset > len_minus_stride { break; } } } else { loop { - // Safety: Neither were aligned $stride_neither_aligned(src.add(offset), dst.add(offset)); offset += SIMD_STRIDE_SIZE; - // Safety: This is `offset > len - SIMD_STRIDE_SIZE`, which means in the next iteration we're valid for - // reading/writing at least SIMD_STRIDE_SIZE elements. if offset > len_minus_stride { break; } @@ -768,7 +606,6 @@ macro_rules! latin1_simd_check_align { } } while offset < len { - // Safety: relies straightforwardly on the `len` invariant let code_unit = *(src.add(offset)); *(dst.add(offset)) = code_unit as $dst_unit; offset += 1; @@ -783,74 +620,56 @@ macro_rules! latin1_simd_check_align_unrolled { $name:ident, $src_unit:ty, $dst_unit:ty, - // Safety: This function must require aligned src/dest that are valid for reading/writing SIMD_STRIDE_SIZE src_unit/dst_unit $stride_both_aligned:ident, - // Safety: This function must require aligned/unaligned src/dest that are valid for reading/writing SIMD_STRIDE_SIZE src_unit/dst_unit $stride_src_aligned:ident, - // Safety: This function must require unaligned/aligned src/dest that are valid for reading/writing SIMD_STRIDE_SIZE src_unit/dst_unit $stride_dst_aligned:ident, - // Safety: This function must require unaligned src/dest that are valid for reading/writing SIMD_STRIDE_SIZE src_unit/dst_unit $stride_neither_aligned:ident ) => { - /// Safety: src/dst must be valid for reads/writes of `len` elements of their units. #[inline(always)] pub unsafe fn $name(src: *const $src_unit, dst: *mut $dst_unit, len: usize) { let unit_size = ::core::mem::size_of::<$src_unit>(); let mut offset = 0usize; - // Safety: if this check succeeds we're valid for reading/writing at least `SIMD_STRIDE_SIZE` elements. if SIMD_STRIDE_SIZE <= len { - // Safety: this correctly calculates the number of src_units that need to be read before the remaining list is aligned. - // This is by definition less than SIMD_STRIDE_SIZE. let mut until_alignment = ((SIMD_STRIDE_SIZE - ((src as usize) & SIMD_ALIGNMENT_MASK)) & SIMD_ALIGNMENT_MASK) / unit_size; while until_alignment != 0 { - // Safety: This is a straightforward copy, since until_alignment is < SIMD_STRIDE_SIZE < len, this is in-bounds *(dst.add(offset)) = *(src.add(offset)) as $dst_unit; offset += 1; until_alignment -= 1; } - // Safety: here offset will be `until_alignment`, i.e. enough to align `src`. let len_minus_stride = len - SIMD_STRIDE_SIZE; - // Safety: if this check succeeds we're valid for reading/writing at least `2 * SIMD_STRIDE_SIZE` elements. if offset + SIMD_STRIDE_SIZE * 2 <= len { let len_minus_stride_times_two = len_minus_stride - SIMD_STRIDE_SIZE; - // Safety: at this point src is known to be aligned at offset, dst is not. if (dst.add(offset) as usize) & SIMD_ALIGNMENT_MASK == 0 { loop { - // Safety: We checked alignment of dst above, we can use the alignment functions. We're allowed to read/write 2*SIMD_STRIDE_SIZE elements, which we do. $stride_both_aligned(src.add(offset), dst.add(offset)); offset += SIMD_STRIDE_SIZE; $stride_both_aligned(src.add(offset), dst.add(offset)); offset += SIMD_STRIDE_SIZE; - // Safety: This is `offset > len - 2 * SIMD_STRIDE_SIZE` which means we always have at least `2 * SIMD_STRIDE_SIZE` elements to munch next time. if offset > len_minus_stride_times_two { break; } } } else { loop { - // Safety: we ensured alignment of src already. $stride_src_aligned(src.add(offset), dst.add(offset)); offset += SIMD_STRIDE_SIZE; $stride_src_aligned(src.add(offset), dst.add(offset)); offset += SIMD_STRIDE_SIZE; - // Safety: This is `offset > len - 2 * SIMD_STRIDE_SIZE` which means we always have at least `2 * SIMD_STRIDE_SIZE` elements to munch next time. if offset > len_minus_stride_times_two { break; } } } } - // Safety: This is `offset > len - SIMD_STRIDE_SIZE` which means we are valid to munch SIMD_STRIDE_SIZE more elements, which we do if offset < len_minus_stride { $stride_src_aligned(src.add(offset), dst.add(offset)); offset += SIMD_STRIDE_SIZE; } } while offset < len { - // Safety: uses len invariant here and below let code_unit = *(src.add(offset)); // On x86_64, this loop autovectorizes but in the pack // case there are instructions whose purpose is to make sure @@ -874,12 +693,7 @@ macro_rules! latin1_simd_check_align_unrolled { #[allow(unused_macros)] macro_rules! ascii_simd_unalign { - // Safety: stride_neither_aligned must be a function that requires src/dest be valid for unaligned reads/writes for SIMD_STRIDE_SIZE elements of type src_unit/dest_unit ($name:ident, $src_unit:ty, $dst_unit:ty, $stride_neither_aligned:ident) => { - /// Safety: src and dst must be valid for reads/writes of len elements of type src_unit/dst_unit - /// - /// Safety-usable invariant: will return Some() when it encounters non-ASCII, with the first element in the Some being - /// guaranteed to be non-ASCII (> 127), and the second being the offset where it is found #[inline(always)] pub unsafe fn $name( src: *const $src_unit, @@ -887,26 +701,21 @@ macro_rules! ascii_simd_unalign { len: usize, ) -> Option<($src_unit, usize)> { let mut offset = 0usize; - // Safety: if this check succeeds we're valid for reading/writing at least `stride` elements. if SIMD_STRIDE_SIZE <= len { let len_minus_stride = len - SIMD_STRIDE_SIZE; loop { - // Safety: We know we're valid for `stride` reads/writes, so we can call this function. We don't need alignment. if !$stride_neither_aligned(src.add(offset), dst.add(offset)) { break; } offset += SIMD_STRIDE_SIZE; - // This is `offset > len - stride` which means we always have at least `stride` elements to munch next time. if offset > len_minus_stride { break; } } } while offset < len { - // Safety: Uses len invariant here and below let code_unit = *(src.add(offset)); if code_unit > 127 { - // Safety-usable invariant upheld here return Some((code_unit, offset)); } *(dst.add(offset)) = code_unit as $dst_unit; @@ -919,27 +728,21 @@ macro_rules! ascii_simd_unalign { #[allow(unused_macros)] macro_rules! latin1_simd_unalign { - // Safety: stride_neither_aligned must be a function that requires src/dest be valid for unaligned reads/writes for SIMD_STRIDE_SIZE elements of type src_unit/dest_unit ($name:ident, $src_unit:ty, $dst_unit:ty, $stride_neither_aligned:ident) => { - /// Safety: src and dst must be valid for unaligned reads/writes of len elements of type src_unit/dst_unit #[inline(always)] pub unsafe fn $name(src: *const $src_unit, dst: *mut $dst_unit, len: usize) { let mut offset = 0usize; - // Safety: if this check succeeds we're valid for reading/writing at least `stride` elements. if SIMD_STRIDE_SIZE <= len { let len_minus_stride = len - SIMD_STRIDE_SIZE; loop { - // Safety: We know we're valid for `stride` reads/writes, so we can call this function. We don't need alignment. $stride_neither_aligned(src.add(offset), dst.add(offset)); offset += SIMD_STRIDE_SIZE; - // This is `offset > len - stride` which means we always have at least `stride` elements to munch next time. if offset > len_minus_stride { break; } } } while offset < len { - // Safety: Uses len invariant here let code_unit = *(src.add(offset)); *(dst.add(offset)) = code_unit as $dst_unit; offset += 1; @@ -950,11 +753,7 @@ macro_rules! latin1_simd_unalign { #[allow(unused_macros)] macro_rules! ascii_to_ascii_simd_stride { - // Safety: load/store must be valid for 16 bytes of read/write, which may be unaligned. (candidates: `(load|store)(16|8)_(unaligned|aligned)` functions) ($name:ident, $load:ident, $store:ident) => { - /// Safety: src and dst must be valid for 16 bytes of read/write according to - /// the $load/$store fn, which may allow for unaligned reads/writes or require - /// alignment to either 16x8 or u8x16. #[inline(always)] pub unsafe fn $name(src: *const u8, dst: *mut u8) -> bool { let simd = $load(src); @@ -969,32 +768,19 @@ macro_rules! ascii_to_ascii_simd_stride { #[allow(unused_macros)] macro_rules! ascii_to_ascii_simd_double_stride { - // Safety: store must be valid for 32 bytes of write, which may be unaligned (candidates: `store(8|16)_(aligned|unaligned)`) ($name:ident, $store:ident) => { - /// Safety: src must be valid for 32 bytes of aligned u8x16 read - /// dst must be valid for 32 bytes of unaligned write according to - /// the $store fn, which may allow for unaligned writes or require - /// alignment to either 16x8 or u8x16. - /// - /// Safety-usable invariant: Returns Some(index) if the element at `index` is invalid ASCII #[inline(always)] pub unsafe fn $name(src: *const u8, dst: *mut u8) -> Option { let first = load16_aligned(src); let second = load16_aligned(src.add(SIMD_STRIDE_SIZE)); $store(dst, first); if unlikely(!simd_is_ascii(first | second)) { - // Safety: mask_ascii produces a mask of all the high bits. let mask_first = mask_ascii(first); if mask_first != 0 { - // Safety: on little endian systems this will be the number of ascii bytes - // before the first non-ascii, i.e. valid for indexing src - // TODO SAFETY: What about big-endian systems? return Some(mask_first.trailing_zeros() as usize); } $store(dst.add(SIMD_STRIDE_SIZE), second); let mask_second = mask_ascii(second); - // Safety: on little endian systems this will be the number of ascii bytes - // before the first non-ascii, i.e. valid for indexing src return Some(SIMD_STRIDE_SIZE + mask_second.trailing_zeros() as usize); } $store(dst.add(SIMD_STRIDE_SIZE), second); @@ -1005,11 +791,7 @@ macro_rules! ascii_to_ascii_simd_double_stride { #[allow(unused_macros)] macro_rules! ascii_to_basic_latin_simd_stride { - // Safety: load/store must be valid for 16 bytes of read/write, which may be unaligned. (candidates: `(load|store)(16|8)_(unaligned|aligned)` functions) ($name:ident, $load:ident, $store:ident) => { - /// Safety: src and dst must be valid for 16/32 bytes of read/write according to - /// the $load/$store fn, which may allow for unaligned reads/writes or require - /// alignment to either 16x8 or u8x16. #[inline(always)] pub unsafe fn $name(src: *const u8, dst: *mut u16) -> bool { let simd = $load(src); @@ -1026,18 +808,13 @@ macro_rules! ascii_to_basic_latin_simd_stride { #[allow(unused_macros)] macro_rules! ascii_to_basic_latin_simd_double_stride { - // Safety: store must be valid for 16 bytes of write, which may be unaligned ($name:ident, $store:ident) => { - /// Safety: src must be valid for 2*SIMD_STRIDE_SIZE bytes of aligned reads, - /// aligned to either 16x8 or u8x16. - /// dst must be valid for 2*SIMD_STRIDE_SIZE bytes of aligned or unaligned reads #[inline(always)] pub unsafe fn $name(src: *const u8, dst: *mut u16) -> Option { let first = load16_aligned(src); let second = load16_aligned(src.add(SIMD_STRIDE_SIZE)); let (a, b) = simd_unpack(first); $store(dst, a); - // Safety: divide by 2 since it's a u16 pointer $store(dst.add(SIMD_STRIDE_SIZE / 2), b); if unlikely(!simd_is_ascii(first | second)) { let mask_first = mask_ascii(first); @@ -1060,11 +837,7 @@ macro_rules! ascii_to_basic_latin_simd_double_stride { #[allow(unused_macros)] macro_rules! unpack_simd_stride { - // Safety: load/store must be valid for 16 bytes of read/write, which may be unaligned. (candidates: `(load|store)(16|8)_(unaligned|aligned)` functions) ($name:ident, $load:ident, $store:ident) => { - /// Safety: src and dst must be valid for 16 bytes of read/write according to - /// the $load/$store fn, which may allow for unaligned reads/writes or require - /// alignment to either 16x8 or u8x16. #[inline(always)] pub unsafe fn $name(src: *const u8, dst: *mut u16) { let simd = $load(src); @@ -1077,11 +850,7 @@ macro_rules! unpack_simd_stride { #[allow(unused_macros)] macro_rules! basic_latin_to_ascii_simd_stride { - // Safety: load/store must be valid for 16 bytes of read/write, which may be unaligned. (candidates: `(load|store)(16|8)_(unaligned|aligned)` functions) ($name:ident, $load:ident, $store:ident) => { - /// Safety: src and dst must be valid for 32/16 bytes of read/write according to - /// the $load/$store fn, which may allow for unaligned reads/writes or require - /// alignment to either 16x8 or u8x16. #[inline(always)] pub unsafe fn $name(src: *const u16, dst: *mut u8) -> bool { let first = $load(src); @@ -1098,11 +867,7 @@ macro_rules! basic_latin_to_ascii_simd_stride { #[allow(unused_macros)] macro_rules! pack_simd_stride { - // Safety: load/store must be valid for 16 bytes of read/write, which may be unaligned. (candidates: `(load|store)(16|8)_(unaligned|aligned)` functions) ($name:ident, $load:ident, $store:ident) => { - /// Safety: src and dst must be valid for 32/16 bytes of read/write according to - /// the $load/$store fn, which may allow for unaligned reads/writes or require - /// alignment to either 16x8 or u8x16. #[inline(always)] pub unsafe fn $name(src: *const u16, dst: *mut u8) { let first = $load(src); @@ -1128,8 +893,6 @@ cfg_if! { pub const ALU_ALIGNMENT_MASK: usize = 7; - // Safety for stride macros: We stick to the load8_aligned/etc family of functions. We consistently produce - // neither_unaligned variants using only unaligned inputs. ascii_to_ascii_simd_stride!(ascii_to_ascii_stride_neither_aligned, load16_unaligned, store16_unaligned); ascii_to_basic_latin_simd_stride!(ascii_to_basic_latin_stride_neither_aligned, load16_unaligned, store8_unaligned); @@ -1138,8 +901,6 @@ cfg_if! { basic_latin_to_ascii_simd_stride!(basic_latin_to_ascii_stride_neither_aligned, load8_unaligned, store16_unaligned); pack_simd_stride!(pack_stride_neither_aligned, load8_unaligned, store16_unaligned); - // Safety for conversion macros: We use the unalign macro with unalign functions above. All stride functions were produced - // by stride macros that universally munch a single SIMD_STRIDE_SIZE worth of elements. ascii_simd_unalign!(ascii_to_ascii, u8, u8, ascii_to_ascii_stride_neither_aligned); ascii_simd_unalign!(ascii_to_basic_latin, u8, u16, ascii_to_basic_latin_stride_neither_aligned); ascii_simd_unalign!(basic_latin_to_ascii, u16, u8, basic_latin_to_ascii_stride_neither_aligned); @@ -1158,9 +919,6 @@ cfg_if! { pub const SIMD_ALIGNMENT_MASK: usize = 15; - // Safety for stride macros: We stick to the load8_aligned/etc family of functions. We consistently name - // aligned/unaligned functions according to src/dst being aligned/unaligned - ascii_to_ascii_simd_stride!(ascii_to_ascii_stride_both_aligned, load16_aligned, store16_aligned); ascii_to_ascii_simd_stride!(ascii_to_ascii_stride_src_aligned, load16_aligned, store16_unaligned); ascii_to_ascii_simd_stride!(ascii_to_ascii_stride_dst_aligned, load16_unaligned, store16_aligned); @@ -1186,9 +944,6 @@ cfg_if! { pack_simd_stride!(pack_stride_dst_aligned, load8_unaligned, store16_aligned); pack_simd_stride!(pack_stride_neither_aligned, load8_unaligned, store16_unaligned); - // Safety for conversion macros: We use the correct pattern of both/src/dst/neither here. All stride functions were produced - // by stride macros that universally munch a single SIMD_STRIDE_SIZE worth of elements. - ascii_simd_check_align!(ascii_to_ascii, u8, u8, ascii_to_ascii_stride_both_aligned, ascii_to_ascii_stride_src_aligned, ascii_to_ascii_stride_dst_aligned, ascii_to_ascii_stride_neither_aligned); ascii_simd_check_align!(ascii_to_basic_latin, u8, u16, ascii_to_basic_latin_stride_both_aligned, ascii_to_basic_latin_stride_src_aligned, ascii_to_basic_latin_stride_dst_aligned, ascii_to_basic_latin_stride_neither_aligned); ascii_simd_check_align!(basic_latin_to_ascii, u16, u8, basic_latin_to_ascii_stride_both_aligned, basic_latin_to_ascii_stride_src_aligned, basic_latin_to_ascii_stride_dst_aligned, basic_latin_to_ascii_stride_neither_aligned); @@ -1203,16 +958,12 @@ cfg_if! { pub const SIMD_STRIDE_SIZE: usize = 16; - /// Safety-usable invariant: This should be identical to SIMD_STRIDE_SIZE (used by ascii_simd_check_align_unrolled) pub const SIMD_ALIGNMENT: usize = 16; pub const MAX_STRIDE_SIZE: usize = 16; pub const SIMD_ALIGNMENT_MASK: usize = 15; - // Safety for stride macros: We stick to the load8_aligned/etc family of functions. We consistently name - // aligned/unaligned functions according to src/dst being aligned/unaligned - ascii_to_ascii_simd_double_stride!(ascii_to_ascii_simd_double_stride_both_aligned, store16_aligned); ascii_to_ascii_simd_double_stride!(ascii_to_ascii_simd_double_stride_src_aligned, store16_unaligned); @@ -1238,9 +989,6 @@ cfg_if! { pack_simd_stride!(pack_stride_both_aligned, load8_aligned, store16_aligned); pack_simd_stride!(pack_stride_src_aligned, load8_aligned, store16_unaligned); - // Safety for conversion macros: We use the correct pattern of both/src/dst/neither/double_both/double_src here. All stride functions were produced - // by stride macros that universally munch a single SIMD_STRIDE_SIZE worth of elements. - ascii_simd_check_align_unrolled!(ascii_to_ascii, u8, u8, ascii_to_ascii_stride_both_aligned, ascii_to_ascii_stride_src_aligned, ascii_to_ascii_stride_neither_aligned, ascii_to_ascii_simd_double_stride_both_aligned, ascii_to_ascii_simd_double_stride_src_aligned); ascii_simd_check_align_unrolled!(ascii_to_basic_latin, u8, u16, ascii_to_basic_latin_stride_both_aligned, ascii_to_basic_latin_stride_src_aligned, ascii_to_basic_latin_stride_neither_aligned, ascii_to_basic_latin_simd_double_stride_both_aligned, ascii_to_basic_latin_simd_double_stride_src_aligned); @@ -1250,21 +998,14 @@ cfg_if! { } else if #[cfg(all(target_endian = "little", target_pointer_width = "64"))] { // Aligned ALU word, little-endian, 64-bit - /// Safety invariant: this is the amount of bytes consumed by - /// unpack_alu. This will be twice the pointer width, as it consumes two usizes. - /// This is also the number of bytes produced by pack_alu. - /// This is also the number of u16 code units produced/consumed by unpack_alu/pack_alu respectively. pub const ALU_STRIDE_SIZE: usize = 16; pub const MAX_STRIDE_SIZE: usize = 16; - // Safety invariant: this is the pointer width in bytes pub const ALU_ALIGNMENT: usize = 8; - // Safety invariant: this is a mask for getting the bits of a pointer not aligned to ALU_ALIGNMENT pub const ALU_ALIGNMENT_MASK: usize = 7; - /// Safety: dst must point to valid space for writing four `usize`s #[inline(always)] unsafe fn unpack_alu(word: usize, second_word: usize, dst: *mut usize) { let first = ((0x0000_0000_FF00_0000usize & word) << 24) | @@ -1283,14 +1024,12 @@ cfg_if! { ((0x00FF_0000_0000_0000usize & second_word) >> 16) | ((0x0000_FF00_0000_0000usize & second_word) >> 24) | ((0x0000_00FF_0000_0000usize & second_word) >> 32); - // Safety: fn invariant used here *dst = first; *(dst.add(1)) = second; *(dst.add(2)) = third; *(dst.add(3)) = fourth; } - /// Safety: dst must point to valid space for writing two `usize`s #[inline(always)] unsafe fn pack_alu(first: usize, second: usize, third: usize, fourth: usize, dst: *mut usize) { let word = ((0x00FF_0000_0000_0000usize & second) << 8) | @@ -1309,28 +1048,20 @@ cfg_if! { ((0x0000_00FF_0000_0000usize & third) >> 16) | ((0x0000_0000_00FF_0000usize & third) >> 8) | (0x0000_0000_0000_00FFusize & third); - // Safety: fn invariant used here *dst = word; *(dst.add(1)) = second_word; } } else if #[cfg(all(target_endian = "little", target_pointer_width = "32"))] { // Aligned ALU word, little-endian, 32-bit - /// Safety invariant: this is the amount of bytes consumed by - /// unpack_alu. This will be twice the pointer width, as it consumes two usizes. - /// This is also the number of bytes produced by pack_alu. - /// This is also the number of u16 code units produced/consumed by unpack_alu/pack_alu respectively. pub const ALU_STRIDE_SIZE: usize = 8; pub const MAX_STRIDE_SIZE: usize = 8; - // Safety invariant: this is the pointer width in bytes pub const ALU_ALIGNMENT: usize = 4; - // Safety invariant: this is a mask for getting the bits of a pointer not aligned to ALU_ALIGNMENT pub const ALU_ALIGNMENT_MASK: usize = 3; - /// Safety: dst must point to valid space for writing four `usize`s #[inline(always)] unsafe fn unpack_alu(word: usize, second_word: usize, dst: *mut usize) { let first = ((0x0000_FF00usize & word) << 8) | @@ -1341,14 +1072,12 @@ cfg_if! { (0x0000_00FFusize & second_word); let fourth = ((0xFF00_0000usize & second_word) >> 8) | ((0x00FF_0000usize & second_word) >> 16); - // Safety: fn invariant used here *dst = first; *(dst.add(1)) = second; *(dst.add(2)) = third; *(dst.add(3)) = fourth; } - /// Safety: dst must point to valid space for writing two `usize`s #[inline(always)] unsafe fn pack_alu(first: usize, second: usize, third: usize, fourth: usize, dst: *mut usize) { let word = ((0x00FF_0000usize & second) << 8) | @@ -1359,28 +1088,20 @@ cfg_if! { ((0x0000_00FFusize & fourth) << 16) | ((0x00FF_0000usize & third) >> 8) | (0x0000_00FFusize & third); - // Safety: fn invariant used here *dst = word; *(dst.add(1)) = second_word; } } else if #[cfg(all(target_endian = "big", target_pointer_width = "64"))] { // Aligned ALU word, big-endian, 64-bit - /// Safety invariant: this is the amount of bytes consumed by - /// unpack_alu. This will be twice the pointer width, as it consumes two usizes. - /// This is also the number of bytes produced by pack_alu. - /// This is also the number of u16 code units produced/consumed by unpack_alu/pack_alu respectively. pub const ALU_STRIDE_SIZE: usize = 16; pub const MAX_STRIDE_SIZE: usize = 16; - // Safety invariant: this is the pointer width in bytes pub const ALU_ALIGNMENT: usize = 8; - // Safety invariant: this is a mask for getting the bits of a pointer not aligned to ALU_ALIGNMENT pub const ALU_ALIGNMENT_MASK: usize = 7; - /// Safety: dst must point to valid space for writing four `usize`s #[inline(always)] unsafe fn unpack_alu(word: usize, second_word: usize, dst: *mut usize) { let first = ((0xFF00_0000_0000_0000usize & word) >> 8) | @@ -1399,14 +1120,12 @@ cfg_if! { ((0x0000_0000_00FF_0000usize & second_word) << 16) | ((0x0000_0000_0000_FF00usize & second_word) << 8) | (0x0000_0000_0000_00FFusize & second_word); - // Safety: fn invariant used here *dst = first; *(dst.add(1)) = second; *(dst.add(2)) = third; *(dst.add(3)) = fourth; } - /// Safety: dst must point to valid space for writing two `usize`s #[inline(always)] unsafe fn pack_alu(first: usize, second: usize, third: usize, fourth: usize, dst: *mut usize) { let word = ((0x00FF0000_00000000usize & first) << 8) | @@ -1425,28 +1144,20 @@ cfg_if! { ((0x000000FF_00000000usize & fourth) >> 16) | ((0x00000000_00FF0000usize & fourth) >> 8) | (0x00000000_000000FFusize & fourth); - // Safety: fn invariant used here *dst = word; *(dst.add(1)) = second_word; } } else if #[cfg(all(target_endian = "big", target_pointer_width = "32"))] { // Aligned ALU word, big-endian, 32-bit - /// Safety invariant: this is the amount of bytes consumed by - /// unpack_alu. This will be twice the pointer width, as it consumes two usizes. - /// This is also the number of bytes produced by pack_alu. - /// This is also the number of u16 code units produced/consumed by unpack_alu/pack_alu respectively. pub const ALU_STRIDE_SIZE: usize = 8; pub const MAX_STRIDE_SIZE: usize = 8; - // Safety invariant: this is the pointer width in bytes pub const ALU_ALIGNMENT: usize = 4; - // Safety invariant: this is a mask for getting the bits of a pointer not aligned to ALU_ALIGNMENT pub const ALU_ALIGNMENT_MASK: usize = 3; - /// Safety: dst must point to valid space for writing four `usize`s #[inline(always)] unsafe fn unpack_alu(word: usize, second_word: usize, dst: *mut usize) { let first = ((0xFF00_0000usize & word) >> 8) | @@ -1457,14 +1168,12 @@ cfg_if! { ((0x00FF_0000usize & second_word) >> 16); let fourth = ((0x0000_FF00usize & second_word) << 8) | (0x0000_00FFusize & second_word); - // Safety: fn invariant used here *dst = first; *(dst.add(1)) = second; *(dst.add(2)) = third; *(dst.add(3)) = fourth; } - /// Safety: dst must point to valid space for writing two `usize`s #[inline(always)] unsafe fn pack_alu(first: usize, second: usize, third: usize, fourth: usize, dst: *mut usize) { let word = ((0x00FF_0000usize & first) << 8) | @@ -1475,7 +1184,6 @@ cfg_if! { ((0x0000_00FFusize & third) << 16) | ((0x00FF_0000usize & fourth) >> 8) | (0x0000_00FFusize & fourth); - // Safety: fn invariant used here *dst = word; *(dst.add(1)) = second_word; } @@ -1487,8 +1195,6 @@ cfg_if! { } cfg_if! { - // Safety-usable invariant: this counts the zeroes from the "first byte" of utf-8 data packed into a usize - // with the target endianness if #[cfg(target_endian = "little")] { #[allow(dead_code)] #[inline(always)] @@ -1506,24 +1212,19 @@ cfg_if! { cfg_if! { if #[cfg(all(feature = "simd-accel", target_endian = "little", target_arch = "disabled"))] { - /// Safety-usable invariant: Will return the value and position of the first non-ASCII byte in the slice in a Some if found. - /// In other words, the first element of the Some is always `> 127` #[inline(always)] pub fn validate_ascii(slice: &[u8]) -> Option<(u8, usize)> { let src = slice.as_ptr(); let len = slice.len(); let mut offset = 0usize; - // Safety: if this check succeeds we're valid for reading/writing at least `stride` elements. if SIMD_STRIDE_SIZE <= len { let len_minus_stride = len - SIMD_STRIDE_SIZE; loop { - // Safety: src at offset is valid for a `SIMD_STRIDE_SIZE` read let simd = unsafe { load16_unaligned(src.add(offset)) }; if !simd_is_ascii(simd) { break; } offset += SIMD_STRIDE_SIZE; - // This is `offset > len - SIMD_STRIDE_SIZE` which means we always have at least `SIMD_STRIDE_SIZE` elements to munch next time. if offset > len_minus_stride { break; } @@ -1532,7 +1233,6 @@ cfg_if! { while offset < len { let code_unit = slice[offset]; if code_unit > 127 { - // Safety: Safety-usable invariant upheld here return Some((code_unit, offset)); } offset += 1; @@ -1540,17 +1240,13 @@ cfg_if! { None } } else if #[cfg(all(feature = "simd-accel", target_feature = "sse2"))] { - /// Safety-usable invariant: will return Some() when it encounters non-ASCII, with the first element in the Some being - /// guaranteed to be non-ASCII (> 127), and the second being the offset where it is found #[inline(always)] pub fn validate_ascii(slice: &[u8]) -> Option<(u8, usize)> { let src = slice.as_ptr(); let len = slice.len(); let mut offset = 0usize; - // Safety: if this check succeeds we're valid for reading at least `stride` elements. if SIMD_STRIDE_SIZE <= len { // First, process one unaligned vector - // Safety: src is valid for a `SIMD_STRIDE_SIZE` read let simd = unsafe { load16_unaligned(src) }; let mask = mask_ascii(simd); if mask != 0 { @@ -1559,26 +1255,18 @@ cfg_if! { return Some((non_ascii, offset)); } offset = SIMD_STRIDE_SIZE; - // Safety: Now that offset has changed we don't yet know how much it is valid for // We have now seen 16 ASCII bytes. Let's guess that // there will be enough more to justify more expense // in the case of non-ASCII. // Use aligned reads for the sake of old microachitectures. - // Safety: this correctly calculates the number of src_units that need to be read before the remaining list is aligned. - // This is by definition less than SIMD_ALIGNMENT, which is defined to be equal to SIMD_STRIDE_SIZE. let until_alignment = unsafe { (SIMD_ALIGNMENT - ((src.add(offset) as usize) & SIMD_ALIGNMENT_MASK)) & SIMD_ALIGNMENT_MASK }; // This addition won't overflow, because even in the 32-bit PAE case the // address space holds enough code that the slice length can't be that // close to address space size. // offset now equals SIMD_STRIDE_SIZE, hence times 3 below. - // - // Safety: if this check succeeds we're valid for reading at least `2 * SIMD_STRIDE_SIZE` elements plus `until_alignment`. - // The extra SIMD_STRIDE_SIZE in the condition is because `offset` is already `SIMD_STRIDE_SIZE`. if until_alignment + (SIMD_STRIDE_SIZE * 3) <= len { if until_alignment != 0 { - // Safety: this is safe to call since we're valid for this read (and more), and don't care about alignment - // This will copy over bytes that get decoded twice since it's not incrementing `offset` by SIMD_STRIDE_SIZE. This is fine. let simd = unsafe { load16_unaligned(src.add(offset)) }; let mask = mask_ascii(simd); if mask != 0 { @@ -1588,78 +1276,53 @@ cfg_if! { } offset += until_alignment; } - // Safety: At this point we're valid for reading 2*SIMD_STRIDE_SIZE elements - // Safety: Now `offset` is aligned for `src` let len_minus_stride_times_two = len - (SIMD_STRIDE_SIZE * 2); loop { - // Safety: We were valid for this read, and were aligned. let first = unsafe { load16_aligned(src.add(offset)) }; let second = unsafe { load16_aligned(src.add(offset + SIMD_STRIDE_SIZE)) }; if !simd_is_ascii(first | second) { - // Safety: mask_ascii produces a mask of all the high bits. let mask_first = mask_ascii(first); if mask_first != 0 { - // Safety: on little endian systems this will be the number of ascii bytes - // before the first non-ascii, i.e. valid for indexing src - // TODO SAFETY: What about big-endian systems? offset += mask_first.trailing_zeros() as usize; } else { let mask_second = mask_ascii(second); - // Safety: on little endian systems this will be the number of ascii bytes - // before the first non-ascii, i.e. valid for indexing src offset += SIMD_STRIDE_SIZE + mask_second.trailing_zeros() as usize; } - // Safety: We know this is non-ASCII, and can uphold the safety-usable invariant here let non_ascii = unsafe { *src.add(offset) }; - return Some((non_ascii, offset)); } offset += SIMD_STRIDE_SIZE * 2; - // Safety: This is `offset > len - 2 * SIMD_STRIDE_SIZE` which means we always have at least `2 * SIMD_STRIDE_SIZE` elements to munch next time. if offset > len_minus_stride_times_two { break; } } - // Safety: if this check succeeds we're valid for reading at least `SIMD_STRIDE_SIZE` if offset + SIMD_STRIDE_SIZE <= len { - // Safety: We were valid for this read, and were aligned. - let simd = unsafe { load16_aligned(src.add(offset)) }; - // Safety: mask_ascii produces a mask of all the high bits. - let mask = mask_ascii(simd); + let simd = unsafe { load16_aligned(src.add(offset)) }; + let mask = mask_ascii(simd); if mask != 0 { - // Safety: on little endian systems this will be the number of ascii bytes - // before the first non-ascii, i.e. valid for indexing src offset += mask.trailing_zeros() as usize; let non_ascii = unsafe { *src.add(offset) }; - // Safety: We know this is non-ASCII, and can uphold the safety-usable invariant here return Some((non_ascii, offset)); } offset += SIMD_STRIDE_SIZE; } } else { - // Safety: this is the unaligned branch // At most two iterations, so unroll - // Safety: if this check succeeds we're valid for reading at least `SIMD_STRIDE_SIZE` if offset + SIMD_STRIDE_SIZE <= len { - // Safety: We're valid for this read but must use an unaligned read let simd = unsafe { load16_unaligned(src.add(offset)) }; let mask = mask_ascii(simd); if mask != 0 { offset += mask.trailing_zeros() as usize; let non_ascii = unsafe { *src.add(offset) }; - // Safety-usable invariant upheld here (same as above) return Some((non_ascii, offset)); } offset += SIMD_STRIDE_SIZE; - // Safety: if this check succeeds we're valid for reading at least `SIMD_STRIDE_SIZE` if offset + SIMD_STRIDE_SIZE <= len { - // Safety: We're valid for this read but must use an unaligned read let simd = unsafe { load16_unaligned(src.add(offset)) }; let mask = mask_ascii(simd); if mask != 0 { offset += mask.trailing_zeros() as usize; let non_ascii = unsafe { *src.add(offset) }; - // Safety-usable invariant upheld here (same as above) return Some((non_ascii, offset)); } offset += SIMD_STRIDE_SIZE; @@ -1668,10 +1331,8 @@ cfg_if! { } } while offset < len { - // Safety: relies straightforwardly on the `len` invariant let code_unit = unsafe { *(src.add(offset)) }; if code_unit > 127 { - // Safety-usable invariant upheld here return Some((code_unit, offset)); } offset += 1; @@ -1679,40 +1340,31 @@ cfg_if! { None } } else { - // Safety-usable invariant: returns byte index of first non-ascii byte #[inline(always)] fn find_non_ascii(word: usize, second_word: usize) -> Option { let word_masked = word & ASCII_MASK; let second_masked = second_word & ASCII_MASK; if (word_masked | second_masked) == 0 { - // Both are ascii, invariant upheld return None; } if word_masked != 0 { let zeros = count_zeros(word_masked); - // `zeros` now contains 0 to 7 (for the seven bits of masked ASCII in little endian, - // or up to 7 bits of non-ASCII in big endian if the first byte is non-ASCII) + // `zeros` now contains 7 (for the seven bits of non-ASCII) // plus 8 times the number of ASCII in text order before the // non-ASCII byte in the little-endian case or 8 times the number of ASCII in // text order before the non-ASCII byte in the big-endian case. let num_ascii = (zeros >> 3) as usize; - // Safety-usable invariant upheld here return Some(num_ascii); } let zeros = count_zeros(second_masked); - // `zeros` now contains 0 to 7 (for the seven bits of masked ASCII in little endian, - // or up to 7 bits of non-ASCII in big endian if the first byte is non-ASCII) + // `zeros` now contains 7 (for the seven bits of non-ASCII) // plus 8 times the number of ASCII in text order before the // non-ASCII byte in the little-endian case or 8 times the number of ASCII in // text order before the non-ASCII byte in the big-endian case. let num_ascii = (zeros >> 3) as usize; - // Safety-usable invariant upheld here Some(ALU_ALIGNMENT + num_ascii) } - /// Safety: `src` must be valid for the reads of two `usize`s - /// - /// Safety-usable invariant: will return byte index of first non-ascii byte #[inline(always)] unsafe fn validate_ascii_stride(src: *const usize) -> Option { let word = *src; @@ -1720,8 +1372,6 @@ cfg_if! { find_non_ascii(word, second_word) } - /// Safety-usable invariant: will return Some() when it encounters non-ASCII, with the first element in the Some being - /// guaranteed to be non-ASCII (> 127), and the second being the offset where it is found #[cfg_attr(feature = "cargo-clippy", allow(cast_ptr_alignment))] #[inline(always)] pub fn validate_ascii(slice: &[u8]) -> Option<(u8, usize)> { @@ -1729,30 +1379,23 @@ cfg_if! { let len = slice.len(); let mut offset = 0usize; let mut until_alignment = (ALU_ALIGNMENT - ((src as usize) & ALU_ALIGNMENT_MASK)) & ALU_ALIGNMENT_MASK; - // Safety: If this check fails we're valid to read `until_alignment + ALU_STRIDE_SIZE` elements if until_alignment + ALU_STRIDE_SIZE <= len { while until_alignment != 0 { let code_unit = slice[offset]; if code_unit > 127 { - // Safety-usable invairant upheld here return Some((code_unit, offset)); } offset += 1; until_alignment -= 1; } - // Safety: At this point we have read until_alignment elements and - // are valid for `ALU_STRIDE_SIZE` more. let len_minus_stride = len - ALU_STRIDE_SIZE; loop { - // Safety: we were valid for this read let ptr = unsafe { src.add(offset) as *const usize }; if let Some(num_ascii) = unsafe { validate_ascii_stride(ptr) } { offset += num_ascii; - // Safety-usable invairant upheld here using the invariant from validate_ascii_stride() return Some((unsafe { *(src.add(offset)) }, offset)); } offset += ALU_STRIDE_SIZE; - // Safety: This is `offset > ALU_STRIDE_SIZE` which means we always have at least `2 * ALU_STRIDE_SIZE` elements to munch next time. if offset > len_minus_stride { break; } @@ -1761,7 +1404,6 @@ cfg_if! { while offset < len { let code_unit = slice[offset]; if code_unit > 127 { - // Safety-usable invairant upheld here return Some((code_unit, offset)); } offset += 1; @@ -1786,47 +1428,36 @@ cfg_if! { pub const ALU_ALIGNMENT_MASK: usize = 3; } else { - // Safety: src points to two valid `usize`s, dst points to four valid `usize`s #[inline(always)] unsafe fn unpack_latin1_stride_alu(src: *const usize, dst: *mut usize) { - // Safety: src safety invariant used here let word = *src; let second_word = *(src.add(1)); - // Safety: dst safety invariant passed down unpack_alu(word, second_word, dst); } - // Safety: src points to four valid `usize`s, dst points to two valid `usize`s #[inline(always)] unsafe fn pack_latin1_stride_alu(src: *const usize, dst: *mut usize) { - // Safety: src safety invariant used here let first = *src; let second = *(src.add(1)); let third = *(src.add(2)); let fourth = *(src.add(3)); - // Safety: dst safety invariant passed down pack_alu(first, second, third, fourth, dst); } - // Safety: src points to two valid `usize`s, dst points to four valid `usize`s #[inline(always)] unsafe fn ascii_to_basic_latin_stride_alu(src: *const usize, dst: *mut usize) -> bool { - // Safety: src safety invariant used here let word = *src; let second_word = *(src.add(1)); // Check if the words contains non-ASCII if (word & ASCII_MASK) | (second_word & ASCII_MASK) != 0 { return false; } - // Safety: dst safety invariant passed down unpack_alu(word, second_word, dst); true } - // Safety: src points four valid `usize`s, dst points to two valid `usize`s #[inline(always)] unsafe fn basic_latin_to_ascii_stride_alu(src: *const usize, dst: *mut usize) -> bool { - // Safety: src safety invariant used here let first = *src; let second = *(src.add(1)); let third = *(src.add(2)); @@ -1834,22 +1465,16 @@ cfg_if! { if (first & BASIC_LATIN_MASK) | (second & BASIC_LATIN_MASK) | (third & BASIC_LATIN_MASK) | (fourth & BASIC_LATIN_MASK) != 0 { return false; } - // Safety: dst safety invariant passed down pack_alu(first, second, third, fourth, dst); true } - // Safety: src, dst both point to two valid `usize`s each - // Safety-usable invariant: Will return byte index of first non-ascii byte. #[inline(always)] unsafe fn ascii_to_ascii_stride(src: *const usize, dst: *mut usize) -> Option { - // Safety: src safety invariant used here let word = *src; let second_word = *(src.add(1)); - // Safety: src safety invariant used here *dst = word; *(dst.add(1)) = second_word; - // Relies on safety-usable invariant here find_non_ascii(word, second_word) } @@ -1857,7 +1482,6 @@ cfg_if! { basic_latin_alu!(basic_latin_to_ascii, u16, u8, basic_latin_to_ascii_stride_alu); latin1_alu!(unpack_latin1, u8, u16, unpack_latin1_stride_alu); latin1_alu!(pack_latin1, u16, u8, pack_latin1_stride_alu); - // Safety invariant upheld: ascii_to_ascii_stride will return byte index of first non-ascii if found ascii_alu!(ascii_to_ascii, u8, u8, ascii_to_ascii_stride); } } diff --git a/third_party/rust/encoding_rs/src/handles.rs b/third_party/rust/encoding_rs/src/handles.rs index f44a834672e9..b5404c01d9bd 100644 --- a/third_party/rust/encoding_rs/src/handles.rs +++ b/third_party/rust/encoding_rs/src/handles.rs @@ -34,7 +34,7 @@ use crate::simd_funcs::*; all(target_endian = "little", target_feature = "neon") ) ))] -use core::simd::u16x8; +use packed_simd::u16x8; use super::DecoderResult; use super::EncoderResult; @@ -90,23 +90,19 @@ impl Endian for LittleEndian { #[derive(Debug, Copy, Clone)] struct UnalignedU16Slice { - // Safety invariant: ptr must be valid for reading 2*len bytes ptr: *const u8, len: usize, } impl UnalignedU16Slice { - /// Safety: ptr must be valid for reading 2*len bytes #[inline(always)] pub unsafe fn new(ptr: *const u8, len: usize) -> UnalignedU16Slice { - // Safety: field invariant passed up to caller here UnalignedU16Slice { ptr, len } } #[inline(always)] pub fn trim_last(&mut self) { assert!(self.len > 0); - // Safety: invariant upheld here: a slice is still valid with a shorter len self.len -= 1; } @@ -117,9 +113,7 @@ impl UnalignedU16Slice { assert!(i < self.len); unsafe { let mut u: MaybeUninit = MaybeUninit::uninit(); - // Safety: i is at most len - 1, which works here ::core::ptr::copy_nonoverlapping(self.ptr.add(i * 2), u.as_mut_ptr() as *mut u8, 2); - // Safety: valid read above lets us do this u.assume_init() } } @@ -127,13 +121,8 @@ impl UnalignedU16Slice { #[cfg(feature = "simd-accel")] #[inline(always)] pub fn simd_at(&self, i: usize) -> u16x8 { - // Safety: i/len are on the scale of u16s, each one corresponds to 2 u8s assert!(i + SIMD_STRIDE_SIZE / 2 <= self.len); let byte_index = i * 2; - // Safety: load16_unaligned needs SIMD_STRIDE_SIZE=16 u8 elements to read, - // or 16/2 = 8 u16 elements to read. - // We have checked that we have at least that many above. - unsafe { to_u16_lanes(load16_unaligned(self.ptr.add(byte_index))) } } @@ -147,7 +136,6 @@ impl UnalignedU16Slice { // XXX the return value should be restricted not to // outlive self. assert!(from <= self.len); - // Safety: This upholds the same invariant: `from` is in bounds and we're returning a shorter slice unsafe { UnalignedU16Slice::new(self.ptr.add(from * 2), self.len - from) } } @@ -156,8 +144,6 @@ impl UnalignedU16Slice { pub fn copy_bmp_to(&self, other: &mut [u16]) -> Option<(u16, usize)> { assert!(self.len <= other.len()); let mut offset = 0; - // Safety: SIMD_STRIDE_SIZE is measured in bytes, whereas len is in u16s. We check we can - // munch SIMD_STRIDE_SIZE / 2 u16s which means we can write SIMD_STRIDE_SIZE u8s if SIMD_STRIDE_SIZE / 2 <= self.len { let len_minus_stride = self.len - SIMD_STRIDE_SIZE / 2; loop { @@ -165,7 +151,6 @@ impl UnalignedU16Slice { if E::OPPOSITE_ENDIAN { simd = simd_byte_swap(simd); } - // Safety: we have enough space on the other side to write this unsafe { store8_unaligned(other.as_mut_ptr().add(offset), simd); } @@ -173,7 +158,6 @@ impl UnalignedU16Slice { break; } offset += SIMD_STRIDE_SIZE / 2; - // Safety: This ensures we still have space for writing SIMD_STRIDE_SIZE u8s if offset > len_minus_stride { break; } @@ -252,7 +236,6 @@ fn copy_unaligned_basic_latin_to_ascii( ) -> CopyAsciiResult { let len = ::core::cmp::min(src.len(), dst.len()); let mut offset = 0; - // Safety: This check ensures we are able to read/write at least SIMD_STRIDE_SIZE elements if SIMD_STRIDE_SIZE <= len { let len_minus_stride = len - SIMD_STRIDE_SIZE; loop { @@ -266,13 +249,10 @@ fn copy_unaligned_basic_latin_to_ascii( break; } let packed = simd_pack(first, second); - // Safety: We are able to write SIMD_STRIDE_SIZE elements in this iteration unsafe { store16_unaligned(dst.as_mut_ptr().add(offset), packed); } offset += SIMD_STRIDE_SIZE; - // Safety: This is `offset > len - SIMD_STRIDE_SIZE`, which ensures that we can write at least SIMD_STRIDE_SIZE elements - // in the next iteration if offset > len_minus_stride { break; } @@ -657,7 +637,7 @@ impl<'a> Utf16Destination<'a> { self.write_code_unit((0xDC00 + (astral & 0x3FF)) as u16); } #[inline(always)] - fn write_surrogate_pair(&mut self, high: u16, low: u16) { + pub fn write_surrogate_pair(&mut self, high: u16, low: u16) { self.write_code_unit(high); self.write_code_unit(low); } @@ -666,7 +646,6 @@ impl<'a> Utf16Destination<'a> { self.write_bmp_excl_ascii(combined); self.write_bmp_excl_ascii(combining); } - // Safety-usable invariant: CopyAsciiResult::GoOn will only contain bytes >=0x80 #[inline(always)] pub fn copy_ascii_from_check_space_bmp<'b>( &'b mut self, @@ -680,8 +659,6 @@ impl<'a> Utf16Destination<'a> { } else { (DecoderResult::InputEmpty, src_remaining.len()) }; - // Safety: This function is documented as needing valid pointers for src/dest and len, which - // is true since we've passed the minumum length of the two match unsafe { ascii_to_basic_latin(src_remaining.as_ptr(), dst_remaining.as_mut_ptr(), length) } { @@ -690,20 +667,16 @@ impl<'a> Utf16Destination<'a> { self.pos += length; return CopyAsciiResult::Stop((pending, source.pos, self.pos)); } - // Safety: the function is documented as returning bytes >=0x80 in the Some Some((non_ascii, consumed)) => { source.pos += consumed; self.pos += consumed; source.pos += 1; // +1 for non_ascii - // Safety: non-ascii bubbled out here non_ascii } } }; - // Safety: non-ascii returned here CopyAsciiResult::GoOn((non_ascii_ret, Utf16BmpHandle::new(self))) } - // Safety-usable invariant: CopyAsciiResult::GoOn will only contain bytes >=0x80 #[inline(always)] pub fn copy_ascii_from_check_space_astral<'b>( &'b mut self, @@ -718,8 +691,6 @@ impl<'a> Utf16Destination<'a> { } else { (DecoderResult::InputEmpty, src_remaining.len()) }; - // Safety: This function is documented as needing valid pointers for src/dest and len, which - // is true since we've passed the minumum length of the two match unsafe { ascii_to_basic_latin(src_remaining.as_ptr(), dst_remaining.as_mut_ptr(), length) } { @@ -728,13 +699,11 @@ impl<'a> Utf16Destination<'a> { self.pos += length; return CopyAsciiResult::Stop((pending, source.pos, self.pos)); } - // Safety: the function is documented as returning bytes >=0x80 in the Some Some((non_ascii, consumed)) => { source.pos += consumed; self.pos += consumed; if self.pos + 1 < dst_len { source.pos += 1; // +1 for non_ascii - // Safety: non-ascii bubbled out here non_ascii } else { return CopyAsciiResult::Stop(( @@ -746,7 +715,6 @@ impl<'a> Utf16Destination<'a> { } } }; - // Safety: non-ascii returned here CopyAsciiResult::GoOn((non_ascii_ret, Utf16AstralHandle::new(self))) } #[inline(always)] diff --git a/third_party/rust/encoding_rs/src/lib.rs b/third_party/rust/encoding_rs/src/lib.rs index 1faf02e6bdf3..6cc920ef882d 100644 --- a/third_party/rust/encoding_rs/src/lib.rs +++ b/third_party/rust/encoding_rs/src/lib.rs @@ -689,7 +689,7 @@ //! for discussion about the UTF-16 family. #![no_std] -#![cfg_attr(feature = "simd-accel", feature(core_intrinsics, portable_simd))] +#![cfg_attr(feature = "simd-accel", feature(core_intrinsics))] #[cfg(feature = "alloc")] #[cfg_attr(test, macro_use)] @@ -699,6 +699,17 @@ extern crate core; #[macro_use] extern crate cfg_if; +#[cfg(all( + feature = "simd-accel", + any( + target_feature = "sse2", + all(target_endian = "little", target_arch = "aarch64"), + all(target_endian = "little", target_feature = "neon") + ) +))] +#[macro_use(shuffle)] +extern crate packed_simd; + #[cfg(feature = "serde")] extern crate serde; diff --git a/third_party/rust/encoding_rs/src/mem.rs b/third_party/rust/encoding_rs/src/mem.rs index 0f9f3c19772a..ba8d9e3f4ca7 100644 --- a/third_party/rust/encoding_rs/src/mem.rs +++ b/third_party/rust/encoding_rs/src/mem.rs @@ -116,11 +116,6 @@ macro_rules! by_unit_check_alu { } let len_minus_stride = len - ALU_ALIGNMENT / unit_size; if offset + (4 * (ALU_ALIGNMENT / unit_size)) <= len { - // Safety: the above check lets us perform 4 consecutive reads of - // length ALU_ALIGNMENT / unit_size. ALU_ALIGNMENT is the size of usize, and unit_size - // is the size of the `src` pointer, so this is equal to performing four usize reads. - // - // This invariant is upheld on all loop iterations let len_minus_unroll = len - (4 * (ALU_ALIGNMENT / unit_size)); loop { let unroll_accu = unsafe { *(src.add(offset) as *const usize) } @@ -139,14 +134,12 @@ macro_rules! by_unit_check_alu { return false; } offset += 4 * (ALU_ALIGNMENT / unit_size); - // Safety: this check lets us continue to perform the 4 reads earlier if offset > len_minus_unroll { break; } } } while offset <= len_minus_stride { - // Safety: the above check lets us perform one usize read. accu |= unsafe { *(src.add(offset) as *const usize) }; offset += ALU_ALIGNMENT / unit_size; } @@ -196,11 +189,6 @@ macro_rules! by_unit_check_simd { } let len_minus_stride = len - SIMD_STRIDE_SIZE / unit_size; if offset + (4 * (SIMD_STRIDE_SIZE / unit_size)) <= len { - // Safety: the above check lets us perform 4 consecutive reads of - // length SIMD_STRIDE_SIZE / unit_size. SIMD_STRIDE_SIZE is the size of $simd_ty, and unit_size - // is the size of the `src` pointer, so this is equal to performing four $simd_ty reads. - // - // This invariant is upheld on all loop iterations let len_minus_unroll = len - (4 * (SIMD_STRIDE_SIZE / unit_size)); loop { let unroll_accu = unsafe { *(src.add(offset) as *const $simd_ty) } @@ -220,7 +208,6 @@ macro_rules! by_unit_check_simd { return false; } offset += 4 * (SIMD_STRIDE_SIZE / unit_size); - // Safety: this check lets us continue to perform the 4 reads earlier if offset > len_minus_unroll { break; } @@ -228,7 +215,6 @@ macro_rules! by_unit_check_simd { } let mut simd_accu = $splat; while offset <= len_minus_stride { - // Safety: the above check lets us perform one $simd_ty read. simd_accu = simd_accu | unsafe { *(src.add(offset) as *const $simd_ty) }; offset += SIMD_STRIDE_SIZE / unit_size; } @@ -248,8 +234,8 @@ macro_rules! by_unit_check_simd { cfg_if! { if #[cfg(all(feature = "simd-accel", any(target_feature = "sse2", all(target_endian = "little", target_arch = "aarch64"), all(target_endian = "little", target_feature = "neon"))))] { use crate::simd_funcs::*; - use core::simd::u8x16; - use core::simd::u16x8; + use packed_simd::u8x16; + use packed_simd::u16x8; const SIMD_ALIGNMENT: usize = 16; diff --git a/third_party/rust/encoding_rs/src/simd_funcs.rs b/third_party/rust/encoding_rs/src/simd_funcs.rs index 5ae00e62e0f7..96feeab5a657 100644 --- a/third_party/rust/encoding_rs/src/simd_funcs.rs +++ b/third_party/rust/encoding_rs/src/simd_funcs.rs @@ -7,74 +7,55 @@ // option. This file may not be copied, modified, or distributed // except according to those terms. -use any_all_workaround::all_mask16x8; -use any_all_workaround::all_mask8x16; -use any_all_workaround::any_mask16x8; -use any_all_workaround::any_mask8x16; -use core::simd::cmp::SimdPartialEq; -use core::simd::cmp::SimdPartialOrd; -use core::simd::mask16x8; -use core::simd::mask8x16; -use core::simd::simd_swizzle; -use core::simd::u16x8; -use core::simd::u8x16; -use core::simd::ToBytes; +use packed_simd::u16x8; +use packed_simd::u8x16; +use packed_simd::IntoBits; // TODO: Migrate unaligned access to stdlib code if/when the RFC // https://github.com/rust-lang/rfcs/pull/1725 is implemented. -/// Safety invariant: ptr must be valid for an unaligned read of 16 bytes #[inline(always)] pub unsafe fn load16_unaligned(ptr: *const u8) -> u8x16 { - let mut simd = ::core::mem::MaybeUninit::::uninit(); - ::core::ptr::copy_nonoverlapping(ptr, simd.as_mut_ptr() as *mut u8, 16); - // Safety: copied 16 bytes of initialized memory into this, it is now initialized - simd.assume_init() + let mut simd = ::core::mem::uninitialized(); + ::core::ptr::copy_nonoverlapping(ptr, &mut simd as *mut u8x16 as *mut u8, 16); + simd } -/// Safety invariant: ptr must be valid for an aligned-for-u8x16 read of 16 bytes #[allow(dead_code)] #[inline(always)] pub unsafe fn load16_aligned(ptr: *const u8) -> u8x16 { *(ptr as *const u8x16) } -/// Safety invariant: ptr must be valid for an unaligned store of 16 bytes #[inline(always)] pub unsafe fn store16_unaligned(ptr: *mut u8, s: u8x16) { ::core::ptr::copy_nonoverlapping(&s as *const u8x16 as *const u8, ptr, 16); } -/// Safety invariant: ptr must be valid for an aligned-for-u8x16 store of 16 bytes #[allow(dead_code)] #[inline(always)] pub unsafe fn store16_aligned(ptr: *mut u8, s: u8x16) { *(ptr as *mut u8x16) = s; } -/// Safety invariant: ptr must be valid for an unaligned read of 16 bytes #[inline(always)] pub unsafe fn load8_unaligned(ptr: *const u16) -> u16x8 { - let mut simd = ::core::mem::MaybeUninit::::uninit(); - ::core::ptr::copy_nonoverlapping(ptr as *const u8, simd.as_mut_ptr() as *mut u8, 16); - // Safety: copied 16 bytes of initialized memory into this, it is now initialized - simd.assume_init() + let mut simd = ::core::mem::uninitialized(); + ::core::ptr::copy_nonoverlapping(ptr as *const u8, &mut simd as *mut u16x8 as *mut u8, 16); + simd } -/// Safety invariant: ptr must be valid for an aligned-for-u16x8 read of 16 bytes #[allow(dead_code)] #[inline(always)] pub unsafe fn load8_aligned(ptr: *const u16) -> u16x8 { *(ptr as *const u16x8) } -/// Safety invariant: ptr must be valid for an unaligned store of 16 bytes #[inline(always)] pub unsafe fn store8_unaligned(ptr: *mut u16, s: u16x8) { ::core::ptr::copy_nonoverlapping(&s as *const u16x8 as *const u8, ptr as *mut u8, 16); } -/// Safety invariant: ptr must be valid for an aligned-for-u16x8 store of 16 bytes #[allow(dead_code)] #[inline(always)] pub unsafe fn store8_aligned(ptr: *mut u16, s: u16x8) { @@ -119,7 +100,7 @@ pub fn simd_byte_swap(s: u16x8) -> u16x8 { #[inline(always)] pub fn to_u16_lanes(s: u8x16) -> u16x8 { - u16x8::from_ne_bytes(s) + s.into_bits() } cfg_if! { @@ -127,11 +108,10 @@ cfg_if! { // Expose low-level mask instead of higher-level conclusion, // because the non-ASCII case would perform less well otherwise. - // Safety-usable invariant: This returned value is whether each high bit is set #[inline(always)] pub fn mask_ascii(s: u8x16) -> i32 { unsafe { - _mm_movemask_epi8(s.into()) + _mm_movemask_epi8(s.into_bits()) } } @@ -145,16 +125,14 @@ cfg_if! { #[inline(always)] pub fn simd_is_ascii(s: u8x16) -> bool { unsafe { - // Safety: We have cfg()d the correct platform - _mm_movemask_epi8(s.into()) == 0 + _mm_movemask_epi8(s.into_bits()) == 0 } } } else if #[cfg(target_arch = "aarch64")]{ #[inline(always)] pub fn simd_is_ascii(s: u8x16) -> bool { unsafe { - // Safety: We have cfg()d the correct platform - vmaxvq_u8(s.into()) < 0x80 + vmaxvq_u8(s.into_bits()) < 0x80 } } } else { @@ -163,7 +141,7 @@ cfg_if! { // This optimizes better on ARM than // the lt formulation. let highest_ascii = u8x16::splat(0x7F); - !any_mask8x16(s.simd_gt(highest_ascii)) + !s.gt(highest_ascii).any() } } } @@ -176,21 +154,20 @@ cfg_if! { return true; } let above_str_latin1 = u8x16::splat(0xC4); - s.simd_lt(above_str_latin1).all() + s.lt(above_str_latin1).all() } } else if #[cfg(target_arch = "aarch64")]{ #[inline(always)] pub fn simd_is_str_latin1(s: u8x16) -> bool { unsafe { - // Safety: We have cfg()d the correct platform - vmaxvq_u8(s.into()) < 0xC4 + vmaxvq_u8(s.into_bits()) < 0xC4 } } } else { #[inline(always)] pub fn simd_is_str_latin1(s: u8x16) -> bool { let above_str_latin1 = u8x16::splat(0xC4); - all_mask8x16(s.simd_lt(above_str_latin1)) + s.lt(above_str_latin1).all() } } } @@ -200,23 +177,21 @@ cfg_if! { #[inline(always)] pub fn simd_is_basic_latin(s: u16x8) -> bool { unsafe { - // Safety: We have cfg()d the correct platform - vmaxvq_u16(s.into()) < 0x80 + vmaxvq_u16(s.into_bits()) < 0x80 } } #[inline(always)] pub fn simd_is_latin1(s: u16x8) -> bool { unsafe { - // Safety: We have cfg()d the correct platform - vmaxvq_u16(s.into()) < 0x100 + vmaxvq_u16(s.into_bits()) < 0x100 } } } else { #[inline(always)] pub fn simd_is_basic_latin(s: u16x8) -> bool { let above_ascii = u16x8::splat(0x80); - all_mask16x8(s.simd_lt(above_ascii)) + s.lt(above_ascii).all() } #[inline(always)] @@ -225,7 +200,7 @@ cfg_if! { // seems faster in this case while the above // function is better the other way round... let highest_latin1 = u16x8::splat(0xFF); - !any_mask16x8(s.simd_gt(highest_latin1)) + !s.gt(highest_latin1).any() } } } @@ -234,7 +209,7 @@ cfg_if! { pub fn contains_surrogates(s: u16x8) -> bool { let mask = u16x8::splat(0xF800); let surrogate_bits = u16x8::splat(0xD800); - any_mask16x8((s & mask).simd_eq(surrogate_bits)) + (s & mask).eq(surrogate_bits).any() } cfg_if! { @@ -242,8 +217,7 @@ cfg_if! { macro_rules! aarch64_return_false_if_below_hebrew { ($s:ident) => ({ unsafe { - // Safety: We have cfg()d the correct platform - if vmaxvq_u16($s.into()) < 0x0590 { + if vmaxvq_u16($s.into_bits()) < 0x0590 { return false; } } @@ -260,7 +234,7 @@ cfg_if! { macro_rules! non_aarch64_return_false_if_all { ($s:ident) => ({ - if all_mask16x8($s) { + if $s.all() { return false; } }) @@ -271,7 +245,7 @@ cfg_if! { macro_rules! in_range16x8 { ($s:ident, $start:expr, $end:expr) => {{ // SIMD sub is wrapping - ($s - u16x8::splat($start)).simd_lt(u16x8::splat($end - $start)) + ($s - u16x8::splat($start)).lt(u16x8::splat($end - $start)) }}; } @@ -285,44 +259,43 @@ pub fn is_u16x8_bidi(s: u16x8) -> bool { aarch64_return_false_if_below_hebrew!(s); - let below_hebrew = s.simd_lt(u16x8::splat(0x0590)); + let below_hebrew = s.lt(u16x8::splat(0x0590)); non_aarch64_return_false_if_all!(below_hebrew); - if all_mask16x8( - below_hebrew | in_range16x8!(s, 0x0900, 0x200F) | in_range16x8!(s, 0x2068, 0xD802), - ) { + if (below_hebrew | in_range16x8!(s, 0x0900, 0x200F) | in_range16x8!(s, 0x2068, 0xD802)).all() { return false; } // Quick refutation failed. Let's do the full check. - any_mask16x8( - (in_range16x8!(s, 0x0590, 0x0900) - | in_range16x8!(s, 0xFB1D, 0xFE00) - | in_range16x8!(s, 0xFE70, 0xFEFF) - | in_range16x8!(s, 0xD802, 0xD804) - | in_range16x8!(s, 0xD83A, 0xD83C) - | s.simd_eq(u16x8::splat(0x200F)) - | s.simd_eq(u16x8::splat(0x202B)) - | s.simd_eq(u16x8::splat(0x202E)) - | s.simd_eq(u16x8::splat(0x2067))), - ) + (in_range16x8!(s, 0x0590, 0x0900) + | in_range16x8!(s, 0xFB1D, 0xFE00) + | in_range16x8!(s, 0xFE70, 0xFEFF) + | in_range16x8!(s, 0xD802, 0xD804) + | in_range16x8!(s, 0xD83A, 0xD83C) + | s.eq(u16x8::splat(0x200F)) + | s.eq(u16x8::splat(0x202B)) + | s.eq(u16x8::splat(0x202E)) + | s.eq(u16x8::splat(0x2067))) + .any() } #[inline(always)] pub fn simd_unpack(s: u8x16) -> (u16x8, u16x8) { - let first: u8x16 = simd_swizzle!( - s, - u8x16::splat(0), - [0, 16, 1, 17, 2, 18, 3, 19, 4, 20, 5, 21, 6, 22, 7, 23] - ); - let second: u8x16 = simd_swizzle!( - s, - u8x16::splat(0), - [8, 24, 9, 25, 10, 26, 11, 27, 12, 28, 13, 29, 14, 30, 15, 31] - ); - (u16x8::from_ne_bytes(first), u16x8::from_ne_bytes(second)) + unsafe { + let first: u8x16 = shuffle!( + s, + u8x16::splat(0), + [0, 16, 1, 17, 2, 18, 3, 19, 4, 20, 5, 21, 6, 22, 7, 23] + ); + let second: u8x16 = shuffle!( + s, + u8x16::splat(0), + [8, 24, 9, 25, 10, 26, 11, 27, 12, 28, 13, 29, 14, 30, 15, 31] + ); + (first.into_bits(), second.into_bits()) + } } cfg_if! { @@ -330,20 +303,21 @@ cfg_if! { #[inline(always)] pub fn simd_pack(a: u16x8, b: u16x8) -> u8x16 { unsafe { - // Safety: We have cfg()d the correct platform - _mm_packus_epi16(a.into(), b.into()).into() + _mm_packus_epi16(a.into_bits(), b.into_bits()).into_bits() } } } else { #[inline(always)] pub fn simd_pack(a: u16x8, b: u16x8) -> u8x16 { - let first: u8x16 = a.to_ne_bytes(); - let second: u8x16 = b.to_ne_bytes(); - simd_swizzle!( - first, - second, - [0, 2, 4, 6, 8, 10, 12, 14, 16, 18, 20, 22, 24, 26, 28, 30] - ) + unsafe { + let first: u8x16 = a.into_bits(); + let second: u8x16 = b.into_bits(); + shuffle!( + first, + second, + [0, 2, 4, 6, 8, 10, 12, 14, 16, 18, 20, 22, 24, 26, 28, 30] + ) + } } } } diff --git a/third_party/rust/encoding_rs/src/single_byte.rs b/third_party/rust/encoding_rs/src/single_byte.rs index b7a4bf23da3a..b3b6089d318b 100644 --- a/third_party/rust/encoding_rs/src/single_byte.rs +++ b/third_party/rust/encoding_rs/src/single_byte.rs @@ -53,9 +53,6 @@ impl SingleByteDecoder { // statically omit the bound check when accessing // `[u16; 128]` with an index // `non_ascii as usize - 0x80usize`. - // - // Safety: `non_ascii` is a u8 byte >=0x80, from the invariants - // on Utf8Destination::copy_ascii_from_check_space_bmp() let mapped = unsafe { *(self.table.get_unchecked(non_ascii as usize - 0x80usize)) }; // let mapped = self.table[non_ascii as usize - 0x80usize]; @@ -154,12 +151,9 @@ impl SingleByteDecoder { } else { (DecoderResult::InputEmpty, src.len()) }; - // Safety invariant: converted <= length. Quite often we have `converted < length` - // which will be separately marked. let mut converted = 0usize; 'outermost: loop { match unsafe { - // Safety: length is the minimum length, `src/dst + x` will always be valid for reads/writes of `len - x` ascii_to_basic_latin( src.as_ptr().add(converted), dst.as_mut_ptr().add(converted), @@ -170,12 +164,6 @@ impl SingleByteDecoder { return (pending, length, length); } Some((mut non_ascii, consumed)) => { - // Safety invariant: `converted <= length` upheld, since this can only consume - // up to `length - converted` bytes. - // - // Furthermore, in this context, - // we can assume `converted < length` since this branch is only ever hit when - // ascii_to_basic_latin fails to consume the entire slice converted += consumed; 'middle: loop { // `converted` doesn't count the reading of `non_ascii` yet. @@ -184,9 +172,6 @@ impl SingleByteDecoder { // statically omit the bound check when accessing // `[u16; 128]` with an index // `non_ascii as usize - 0x80usize`. - // - // Safety: We can rely on `non_ascii` being between `0x80` and `0xFF` due to - // the invariants of `ascii_to_basic_latin()`, and our table has enough space for that. let mapped = unsafe { *(self.table.get_unchecked(non_ascii as usize - 0x80usize)) }; // let mapped = self.table[non_ascii as usize - 0x80usize]; @@ -198,10 +183,9 @@ impl SingleByteDecoder { ); } unsafe { - // Safety: As mentioned above, `converted < length` + // The bound check has already been performed *(dst.get_unchecked_mut(converted)) = mapped; } - // Safety: `converted <= length` upheld, since `converted < length` before this converted += 1; // Next, handle ASCII punctuation and non-ASCII without // going back to ASCII acceleration. Non-ASCII scripts @@ -214,10 +198,7 @@ impl SingleByteDecoder { if converted == length { return (pending, length, length); } - // Safety: We are back to `converted < length` because of the == above - // and can perform this check. let mut b = unsafe { *(src.get_unchecked(converted)) }; - // Safety: `converted < length` is upheld for this loop 'innermost: loop { if b > 127 { non_ascii = b; @@ -227,20 +208,15 @@ impl SingleByteDecoder { // byte unconditionally instead of trying to unread it // to make it part of the next SIMD stride. unsafe { - // Safety: `converted < length` is true for this loop *(dst.get_unchecked_mut(converted)) = u16::from(b); } - // Safety: We are now at `converted <= length`. We should *not* `continue` - // the loop without reverifying converted += 1; if b < 60 { // We've got punctuation if converted == length { return (pending, length, length); } - // Safety: we're back to `converted <= length` because of the == above b = unsafe { *(src.get_unchecked(converted)) }; - // Safety: The loop continues as `converted < length` continue 'innermost; } // We've got markup or ASCII text @@ -258,8 +234,6 @@ impl SingleByteDecoder { loop { if let Some((non_ascii, offset)) = validate_ascii(bytes) { total += offset; - // Safety: We can rely on `non_ascii` being between `0x80` and `0xFF` due to - // the invariants of `ascii_to_basic_latin()`, and our table has enough space for that. let mapped = unsafe { *(self.table.get_unchecked(non_ascii as usize - 0x80usize)) }; if mapped != u16::from(non_ascii) { return total; @@ -410,12 +384,9 @@ impl SingleByteEncoder { } else { (EncoderResult::InputEmpty, src.len()) }; - // Safety invariant: converted <= length. Quite often we have `converted < length` - // which will be separately marked. let mut converted = 0usize; 'outermost: loop { match unsafe { - // Safety: length is the minimum length, `src/dst + x` will always be valid for reads/writes of `len - x` basic_latin_to_ascii( src.as_ptr().add(converted), dst.as_mut_ptr().add(converted), @@ -426,23 +397,15 @@ impl SingleByteEncoder { return (pending, length, length); } Some((mut non_ascii, consumed)) => { - // Safety invariant: `converted <= length` upheld, since this can only consume - // up to `length - converted` bytes. - // - // Furthermore, in this context, - // we can assume `converted < length` since this branch is only ever hit when - // ascii_to_basic_latin fails to consume the entire slice converted += consumed; 'middle: loop { // `converted` doesn't count the reading of `non_ascii` yet. match self.encode_u16(non_ascii) { Some(byte) => { unsafe { - // Safety: we're allowed this access since `converted < length` *(dst.get_unchecked_mut(converted)) = byte; } converted += 1; - // `converted <= length` now } None => { // At this point, we need to know if we @@ -458,8 +421,6 @@ impl SingleByteEncoder { converted, ); } - // Safety: convered < length from outside the match, and `converted + 1 != length`, - // So `converted + 1 < length` as well. We're in bounds let second = u32::from(unsafe { *src.get_unchecked(converted + 1) }); if second & 0xFC00u32 != 0xDC00u32 { @@ -471,18 +432,6 @@ impl SingleByteEncoder { } // The next code unit is a low surrogate. let astral: char = unsafe { - // Safety: We can rely on non_ascii being 0xD800-0xDBFF since the high bits are 0xD800 - // Then, (non_ascii << 10 - 0xD800 << 10) becomes between (0 to 0x3FF) << 10, which is between - // 0x400 to 0xffc00. Adding the 0x10000 gives a range of 0x10400 to 0x10fc00. Subtracting the 0xDC00 - // gives 0x2800 to 0x102000 - // The second term is between 0xDC00 and 0xDFFF from the check above. This gives a maximum - // possible range of (0x10400 + 0xDC00) to (0x102000 + 0xDFFF) which is 0x1E000 to 0x10ffff. - // This is in range. - // - // From a Unicode principles perspective this can also be verified as we have checked that `non_ascii` is a high surrogate - // (0xD800..=0xDBFF), and that `second` is a low surrogate (`0xDC00..=0xDFFF`), and we are applying reverse of the UTC16 transformation - // algorithm , by applying the high surrogate - 0xD800 to the - // high ten bits, and the low surrogate - 0xDc00 to the low ten bits, and then adding 0x10000 ::core::char::from_u32_unchecked( (u32::from(non_ascii) << 10) + second - (((0xD800u32 << 10) - 0x1_0000u32) + 0xDC00u32), @@ -507,7 +456,6 @@ impl SingleByteEncoder { converted + 1, // +1 `for non_ascii` converted, ); - // Safety: This branch diverges, so no need to uphold invariants on `converted` } } // Next, handle ASCII punctuation and non-ASCII without @@ -521,12 +469,8 @@ impl SingleByteEncoder { if converted == length { return (pending, length, length); } - // Safety: we're back to `converted < length` due to the == above and can perform - // the unchecked read let mut unit = unsafe { *(src.get_unchecked(converted)) }; 'innermost: loop { - // Safety: This loop always begins with `converted < length`, see - // the invariant outside and the comment on the continue below if unit > 127 { non_ascii = unit; continue 'middle; @@ -535,25 +479,19 @@ impl SingleByteEncoder { // byte unconditionally instead of trying to unread it // to make it part of the next SIMD stride. unsafe { - // Safety: Can rely on converted < length *(dst.get_unchecked_mut(converted)) = unit as u8; } converted += 1; - // `converted <= length` here if unit < 60 { // We've got punctuation if converted == length { return (pending, length, length); } - // Safety: `converted < length` due to the == above. The read is safe. unit = unsafe { *(src.get_unchecked(converted)) }; - // Safety: This only happens if `converted < length`, maintaining it continue 'innermost; } // We've got markup or ASCII text continue 'outermost; - // Safety: All other routes to here diverge so the continue is the only - // way to run the innermost loop. } } } diff --git a/third_party/rust/encoding_rs/src/x_user_defined.rs b/third_party/rust/encoding_rs/src/x_user_defined.rs index 7af7d5e3d69d..103c9afba9eb 100644 --- a/third_party/rust/encoding_rs/src/x_user_defined.rs +++ b/third_party/rust/encoding_rs/src/x_user_defined.rs @@ -14,13 +14,12 @@ use crate::variant::*; cfg_if! { if #[cfg(feature = "simd-accel")] { use simd_funcs::*; - use core::simd::u16x8; - use core::simd::cmp::SimdPartialOrd; + use packed_simd::u16x8; #[inline(always)] fn shift_upper(unpacked: u16x8) -> u16x8 { let highest_ascii = u16x8::splat(0x7F); - unpacked + unpacked.simd_gt(highest_ascii).select(u16x8::splat(0xF700), u16x8::splat(0)) } + unpacked + unpacked.gt(highest_ascii).select(u16x8::splat(0xF700), u16x8::splat(0)) } } else { } } @@ -117,15 +116,10 @@ impl UserDefinedDecoder { let simd_iterations = length >> 4; let src_ptr = src.as_ptr(); let dst_ptr = dst.as_mut_ptr(); - // Safety: This is `for i in 0..length / 16` for i in 0..simd_iterations { - // Safety: This is in bounds: length is the minumum valid length for both src/dst - // and i ranges to length/16, so multiplying by 16 will always be `< length` and can do - // a 16 byte read let input = unsafe { load16_unaligned(src_ptr.add(i * 16)) }; let (first, second) = simd_unpack(input); unsafe { - // Safety: same as above, but this is two consecutive 8-byte reads store8_unaligned(dst_ptr.add(i * 16), shift_upper(first)); store8_unaligned(dst_ptr.add((i * 16) + 8), shift_upper(second)); } diff --git a/third_party/rust/packed_simd/.cargo-checksum.json b/third_party/rust/packed_simd/.cargo-checksum.json new file mode 100644 index 000000000000..0d75492f06bc --- /dev/null +++ b/third_party/rust/packed_simd/.cargo-checksum.json @@ -0,0 +1 @@ +{"files":{"Cargo.toml":"ae219b55eab1f8cd8c3497d7327ea0e5426dba23d2446f7981ca799ad80b6a52","LICENSE-APACHE":"a60eea817514531668d7e00765731449fe14d059d3249e0bc93b36de45f759f2","LICENSE-MIT":"6485b8ed310d3f0340bf1ad1f47645069ce4069dcc6bb46c7d5c6faf41de1fdb","README.md":"c4ac7027a9ab7d7858aa8957d7454dbfcdbb81e605b6a171f05310cc3cad3762","bors.toml":"dee881dc69b9b7834e4eba5d95c3ed5a416d4628815a167d6a22d4cb4fb064b8","build.rs":"019ed29c43989782d8eec3a961654cfc172d7a7898da4eca8f654700af7e1988","ci/all.sh":"2ae6b2445b4db83833e40b37efd0016c6b9879ee988b9b3ef94db5439a3e1606","ci/android-install-ndk.sh":"bdcf93ba9043ac1184e2c504a3d40c47c6c1601d882e0f0a27a8eb56fbabcb5f","ci/android-install-sdk.sh":"3490432022c5c8f5a115c084f7a9aca1626f96c0c87ffb62019228c4346b47e4","ci/android-sysimage.sh":"ebf4e5daa1f0fe1b2092b79f0f3f161c4c4275cb744e52352c4d81ab451e4c5a","ci/benchmark.sh":"b61d19ef6b90deba8fb79dee74c8b062d94844676293da346da87bb78a9a49a4","ci/deploy_and_run_on_ios_simulator.rs":"ec8ecf82d92072676aa47f0d1a3d021b60a7ae3531153ef12d2ff4541fc294dc","ci/docker/aarch64-linux-android/Dockerfile":"ace2e7d33c87bc0f6d3962a4a3408c04557646f7f51ab99cfbf574906796b016","ci/docker/aarch64-unknown-linux-gnu/Dockerfile":"da88c0d50f16dc08448c7fdf1fa5ed2cbe576acf9e7dd85b5b818621b2a8c702","ci/docker/arm-unknown-linux-gnueabi/Dockerfile":"bb5f8ae890707c128652290ffc544447643bf12037ddd73c6ad6989f848cb380","ci/docker/arm-unknown-linux-gnueabihf/Dockerfile":"1afaefcbc05b740859acd4e067bc92439be6bcbe8f2e9678474fb434bcd398d9","ci/docker/armv7-linux-androideabi/Dockerfile":"370e55d3330a413a3ccf677b3afb3e0ef9018a5fab263faa97ae8ac017fc2286","ci/docker/armv7-unknown-linux-gnueabihf/Dockerfile":"8282ea707a94109beed47a57574755e2d58401735904a03f85fb64c578c53b4f","ci/docker/i586-unknown-linux-gnu/Dockerfile":"49792922269f371bd29da4727e9085101b27be67a6b97755d0196c63317f7abb","ci/docker/i686-unknown-linux-gnu/Dockerfile":"49792922269f371bd29da4727e9085101b27be67a6b97755d0196c63317f7abb","ci/docker/mips-unknown-linux-gnu/Dockerfile":"b2ebc25797612c4f8395fe9d407725156044955bfbcf442036b7f55b43a5f9da","ci/docker/mips64-unknown-linux-gnuabi64/Dockerfile":"b0c1692ac65bc56dd30494b1993d8e929c48cc9c4b92029b7c7592af6d4f9220","ci/docker/mips64el-unknown-linux-gnuabi64/Dockerfile":"4e9249c179300138141d0b2b7401b11897f64aed69f541f078c1db4594df2827","ci/docker/mipsel-unknown-linux-musl/Dockerfile":"3164c52b0dcbb01afa78292b15b5c43503ccf0491cf6eb801ec2bf22ae274e52","ci/docker/powerpc-unknown-linux-gnu/Dockerfile":"ae8274309928620a5dd232a46264e05399bb746288ebee3843a71c4162208cc3","ci/docker/powerpc64-unknown-linux-gnu/Dockerfile":"ba5fbc4bf3bb91cd50b407248da31225681efc8f2be7618f4a0ab1219b389508","ci/docker/powerpc64le-unknown-linux-gnu/Dockerfile":"53f97f8b9b5aca7534b9bf9ea48f35175052cd2a560a107e01ad270731c032fc","ci/docker/s390x-unknown-linux-gnu/Dockerfile":"89f5421cf06d817ae94092987e914472ef384ad2d1fff2735be3d8786ba11214","ci/docker/sparc64-unknown-linux-gnu/Dockerfile":"83eba19576486f9d10d7c037d669d72b31a65565a479f30b22aab36aaa2ff8dc","ci/docker/thumbv7neon-linux-androideabi/Dockerfile":"c2decd5591bd7a09378901bef629cd944acf052eb55e4f35b79eb9cb4d62246a","ci/docker/thumbv7neon-unknown-linux-gnueabihf/Dockerfile":"51955a8bf3c4d440f47382af6f5426ebff94ab01a04da36175babda9a057740f","ci/docker/wasm32-unknown-unknown/Dockerfile":"b982b421c70db476900df5b60e19ef8815e6c7dae22687225002780cab7b0a76","ci/docker/x86_64-linux-android/Dockerfile":"a17ebdb186ce2dd6b62100b5a439e05a1ab9adab113e2508843e121aaea52992","ci/docker/x86_64-unknown-linux-gnu-emulated/Dockerfile":"44b6203d9290bfdc53d81219f0937e1110847a23dd982ec8c4de388354f01536","ci/docker/x86_64-unknown-linux-gnu/Dockerfile":"7f4e3ca5fa288ea70edb4d1f75309708cd30b192e2e4444e61c4d5b3b58f89cf","ci/dox.sh":"434e9611c52e389312d2b03564adf09429f10cc76fe66a8644adb104903b87b7","ci/linux-s390x.sh":"d6b732d7795b4ba131326aff893bca6228a7d2eb0e9402f135705413dbbe0dce","ci/linux-sparc64.sh":"c92966838b1ab7ad3b7a344833ee726aba6b647cf5952e56f0ad1ba420b13325","ci/lld-shim.rs":"3d7f71ec23a49e2b67f694a0168786f9a954dda15f5a138815d966643fd3fcc3","ci/max_line_width.sh":"0a1518bba4c9ecaa55694cb2e9930d0e19c265baabf73143f17f9cf285aaa5bb","ci/run-docker.sh":"92e036390ad9b0d16f109579df1b5ced2e72e9afea40c7d011400ebd3a2a90de","ci/run.sh":"9afabc961e0ee83b87201f3fd554c19e5b0c36f3a95d013595e276c9882dd0a4","ci/run_examples.sh":"d1a23c6c35374a0678ba5114b9b8fefd8be0a79e774872a8bf0898d1baca18d0","ci/runtest-android.rs":"145a8e9799a5223975061fe7e586ade5669ee4877a7d7a4cf6b4ab48e8e36c7c","ci/setup_benchmarks.sh":"fae3960023f6f3d1388cd2ad22fdbab4b075f1f29dd4292d7994a20783beb6cf","ci/test-runner-linux":"c8aa6025cff5306f4f31d0c61dc5f9d4dd5a1d189ab613ef8d4c367c694d9ccd","contributing.md":"2d2629310ad4d464c482bdbb5819f0d6ce223c576aeef2cdce6a1f6857085ea5","perf-guide/book.toml":"115a98284126c6b180178b44713314cc494f08a71662ee2ce15cf67f17a51064","perf-guide/src/SUMMARY.md":"3e03bffc991fdc2050f3d51842d72d9d21ea6abab56a3baf3b2d5973a78b89e1","perf-guide/src/ascii.css":"29afb08833b2fe2250f0412e1fa1161a2432a0820a14953c87124407417c741a","perf-guide/src/bound_checks.md":"5e4991ff58a183ef0cd9fdc1feb4cd12d083b44bdf87393bbb0927808ef3ce7d","perf-guide/src/float-math/approx.md":"8c09032fa2d795a0c5db1775826c850d28eb2627846d0965c60ee72de63735ad","perf-guide/src/float-math/fma.md":"311076ba4b741d604a82e74b83a8d7e8c318fcbd7f64c4392d1cf5af95c60243","perf-guide/src/float-math/fp.md":"04153e775ab6e4f0d7837bcc515230d327b04edfa34c84ce9c9e10ebaeef2be8","perf-guide/src/float-math/svml.md":"0798873b8eedaeda5fed62dc91645b57c20775a02d3cd74d8bd06958f1516506","perf-guide/src/introduction.md":"9f5a19e9e6751f25d2daad39891a0cc600974527ec4c8305843f9618910671bd","perf-guide/src/prof/linux.md":"447731eb5de7d69166728fdbc5ecb0c0c9db678ea493b45a592d67dd002184c0","perf-guide/src/prof/mca.md":"f56d54f3d20e7aa4d32052186e8237b03d65971eb5d112802b442570ff11d344","perf-guide/src/prof/profiling.md":"8a650c0fd6ede0964789bb6577557eeef1d8226a896788602ce61528e260e43c","perf-guide/src/target-feature/attribute.md":"615f88dca0a707b6c416fa605435dd6e1fb5361cc639429cbf68cd87624bd78b","perf-guide/src/target-feature/features.md":"17077760ff24c006b606dd21889c53d87228f4311f3ba3a574f9afdeacd86165","perf-guide/src/target-feature/inlining.md":"7ed1d7068d8173a00d84c16cfe5871cd68b9f04f8d0cca2d01ebc84957ebf2f6","perf-guide/src/target-feature/practice.md":"c4b371842e0086df178488fec97f20def8f0c62ee588bcd25fd948b9b1fa227e","perf-guide/src/target-feature/runtime.md":"835425f5ee597fb3e51d36e725a81ebee29f4561231d19563cd4da81dbb1cfcb","perf-guide/src/target-feature/rustflags.md":"01197acf6f0adec8db32b8591811f69cecb6555a2b05dc5d5ec27d0e3f7b065e","perf-guide/src/vert-hor-ops.md":"c6211c0ee91e60552ec592d89d9d957eedc21dee3cbd89e1ad6765ea06a27471","rust-toolchain":"58bea07cb6d97f9cfcd5c8f98b1feca0fb81cce5b0bf29a8e70ed2641956e9a6","rustfmt.toml":"d99a43f3f8ef9e425cf01c333fba9f0051f888f5d87ab4e8f63c2f7d0fe6620f","src/api.rs":"45508c6c0241519fc01a7f00c9105554c24c312c4e46900ef9c75139ea438305","src/api/bit_manip.rs":"27f3097fc0a11e3c4107049d9779e680dcd67407a066704008a6b9c4fd529e05","src/api/bitmask.rs":"058ebc38a2e0363f07a441d3e9a4775aaec57ccb170a0e5d5efa5dc4743ab07b","src/api/cast.rs":"03b94a3d316ac7b7be7068810044911e965e889a0ace7bae762749ca74a92747","src/api/cast/macros.rs":"b0a14d0c83ad2ebb7a275180f6d9e3f2bc312ba57a7d3d6c39fad4e0f20f9408","src/api/cast/v128.rs":"edd0994efac4379dff26e178423a52dbb3ffeb38b1fc97cae975d744c00b4fb6","src/api/cast/v16.rs":"96bd98c2d21b0663abe6c0ab33005b1fa693f3db7ee6795351391343863484da","src/api/cast/v256.rs":"8c31fe91f5e78ef737dfba6979cc1240210cb094a89d284fe459bf8a991ca24b","src/api/cast/v32.rs":"a99a79dd84d2a5e6adf9db98705675915bd03fd1287d489c7fe38e84d7e4a086","src/api/cast/v512.rs":"c0dd526f41ed7b8a71c3743d91267554ec0a0c75834ccc2e3ecb0ef3004af642","src/api/cast/v64.rs":"6572fdba2a1241a6cd666d3f0cce3306cd2cb7e5e236172e59d5d4351c8a88af","src/api/cmp.rs":"357c3a2a09c6d4611c32dd7fa95be2fae933d513e229026ec9b44451a77b884e","src/api/cmp/eq.rs":"60f70f355bae4cb5b17db53204cacc3890f70670611c17df638d4c04f7cc8075","src/api/cmp/ord.rs":"589f7234761c294fa5df8f525bc4acd5a47cdb602207d524a0d4e19804cd9695","src/api/cmp/partial_eq.rs":"902ccb8aa01fd5738b30ba0b712669c21d4801958907e03bad23432c7dba0198","src/api/cmp/partial_ord.rs":"9db0c37d7434cdfc62d8d66912e972fa3d8c115ab2af051a6f45e414bd3e4f1c","src/api/cmp/vertical.rs":"de3d62f38eba817299aa16f1e1939954c9a447e316509397465c2830852ba053","src/api/default.rs":"67bf21c134127d12a7028c8b88a57f0ceee8ccbd74976da8ca74eb9f16a174d5","src/api/fmt.rs":"67fb804bb86b6cd77cf8cd492b5733ce437071b66fe3297278b8a6552c325dda","src/api/fmt/binary.rs":"02b2b287f7404f8a983813cf70c87108c8da3835578b63ab303116885f609413","src/api/fmt/debug.rs":"56e1c3bdc092747344fffaafff9da7163ee7827857f6fb7cb1c9923eca4f6fa0","src/api/fmt/lower_hex.rs":"558fd592f7f485712fb051509cecc7174a21e6bf62e5ce64766e75afc97bb8e1","src/api/fmt/octal.rs":"3b2e70877a4f368c7704f8e254236c014c365c74d93371c1feb5f030e6c66422","src/api/fmt/upper_hex.rs":"2a442f666bc80e22d41f903f881238fe114dd49344c3ed69849250e853cafc5d","src/api/from.rs":"2e599d8329cb05eaf06224cc441355c4b7b51254fc19256619333be8c149d444","src/api/from/from_array.rs":"5d2cc700568376bf6ee1fe5e406da3bc2d488ff155644bf73d06a1349b73fc53","src/api/from/from_vector.rs":"9764371aa9e6005aace74dea14f59e5611a095b7cf42707940924749282c52f0","src/api/hash.rs":"5076ece87969592c876486f5b1ea8affbeaec379d1a14a30859e0aa5592019de","src/api/into_bits.rs":"8f8011627250e23e66b5c0ca641afb079d8232674bb1354140b536bdbea63e55","src/api/into_bits/arch_specific.rs":"e7445021f3908326bfee758835e5fc5ad56aa1baa77fc1c58abe4350c66c670a","src/api/into_bits/macros.rs":"bb4fe99be2af6a21d805efab44c8e4e61a7b2adb42a65504a0cf26d13efdadcd","src/api/into_bits/v128.rs":"145a44922b09a5ca5b62d88a461d327d399a997a15db4b11d7b17e554a9fa4c0","src/api/into_bits/v16.rs":"f4f4f61ba88aa51b158ec56ca3dce234349aea0daf2b3029a14ab5125d1e41e5","src/api/into_bits/v256.rs":"8cea9c5d9809f11323cb7cdc53b83df593fd17caf926251e412ae9777bed547f","src/api/into_bits/v32.rs":"905ba683d342fa32f4202b80bb46530807bd0a5b588f6c2e8c9f475223c47775","src/api/into_bits/v512.rs":"e25afa1fbf088a5d58e7d75d197b6cd4c56637ea28542ba18e46a451f29d04e7","src/api/into_bits/v64.rs":"d6238022ccff7b92e55b3f6017fc269acb6f36330a6d7e8fb389853a0f1b6478","src/api/math.rs":"8b2a2fc651917a850539f993aa0b9e5bf4da67b11685285b8de8cdca311719ec","src/api/math/float.rs":"969a75cdb3743c5ac7cde653d1a7f659ac65f2a5afb004c9928a7b34b79c3e39","src/api/math/float/abs.rs":"5b6b2701e2e11135b7ce58a05052ea8120e10e4702c95d046b9d21b827b26bf8","src/api/math/float/consts.rs":"6302c9261da4291d144d5bb53493cdd073498feb40955fb6860ea3c4d06c978a","src/api/math/float/cos.rs":"4c2dd7173728ef189314f1576c9486e03be21b7da98843b2f9011282a7979e31","src/api/math/float/exp.rs":"7c6d5f1e304f498a01cfa23b92380c815d7da0ad94eae3483783bc377d287eef","src/api/math/float/ln.rs":"54c7583f3df793b39ff57534fade27b41bb992439e5dc178252f5ca3190a3e54","src/api/math/float/mul_add.rs":"62cac77660d20159276d4c9ef066eb90c81cbddb808e8e157182c607625ad2eb","src/api/math/float/mul_adde.rs":"bae056ee9f3a70df39ec3c3b2f6437c65303888a7b843ef1a5bcf1f5aca0e602","src/api/math/float/powf.rs":"9ddb938984b36d39d82a82f862f80df8f7fb013f1d222d45698d41d88472f568","src/api/math/float/recpre.rs":"589225794ff1dbf31158dff660e6d4509ecc8befbb57c633900dea5ac0b840d6","src/api/math/float/rsqrte.rs":"a32abdcc318d7ccc8448231f54d75b884b7cbeb03a7d595713ab6243036f4dbf","src/api/math/float/sin.rs":"cbd3622b7df74f19691743001c8cf747a201f8977ad90542fee915f37dcd1e49","src/api/math/float/sqrt.rs":"0c66d5d63fb08e4d99c6b82a8828e41173aff1ac9fa1a2764a11fac217ccf2ac","src/api/math/float/sqrte.rs":"731e1c9f321b662accdd27dacb3aac2e8043b7aecb2f2161dde733bd9f025362","src/api/math/float/tanh.rs":"e57940434cc05981b086f0f3b92d32caceb38d67b90aebce5d3ed8e07c80538f","src/api/minimal.rs":"1f22bcc528555444e76de569ec0ae2029b9ae9d04805efeafa93369c8098036b","src/api/minimal/iuf.rs":"819cff26d3e196f807645bcc1d79eb27d9f175edb89910f2274d52a1e913cd11","src/api/minimal/mask.rs":"0cae10ae1fc65f5070e686c0c79bfba27b86b33d6c399367bd4848fb367dcec4","src/api/minimal/ptr.rs":"f74d7a4925d7209faebc26ea8315259cb2c08ec65789a70869e595649a9bc39a","src/api/ops.rs":"3e273b277a0f3019d42c3c59ca94a5afd4885d5ae6d2182e5089bbeec9de42ee","src/api/ops/scalar_arithmetic.rs":"d2d5ad897a59dd0787544f927e0e7ca4072c3e58b0f4a2324083312b0d5a21d7","src/api/ops/scalar_bitwise.rs":"482204e459ca6be79568e1c9f70adbe2d2151412ddf122fb2161be8ebb51c40c","src/api/ops/scalar_mask_bitwise.rs":"c250f52042e37b22d57256c80d4604104cfd2fbe2a2e127c676267270ca5d350","src/api/ops/scalar_shifts.rs":"c4773d435c3f9da4454327e6fbb2b5b41a1c0ebb1cca7372e69dc7a344a1b6e4","src/api/ops/vector_arithmetic.rs":"ddca15d09ddeef502c2ed66117a62300ca65d87e959e8b622d767bdf1c307910","src/api/ops/vector_bitwise.rs":"b3968f7005b649edcc22a54e2379b14d5ee19045f2e784029805781ae043b5ee","src/api/ops/vector_float_min_max.rs":"76bf8cb607e2c442923c1da1061a6b80d742d607408033c2a3761161114cf2a0","src/api/ops/vector_int_min_max.rs":"a378789c6ff9b32a51fbd0a97ffd36ed102cd1fe6a067d2b02017c1df342def6","src/api/ops/vector_mask_bitwise.rs":"5052d18517d765415d40327e6e8e55a312daaca0a5e2aec959bfa54b1675f9c8","src/api/ops/vector_neg.rs":"5c62f6b0221983cdbd23cd0a3af3672e6ba1255f0dfe8b19aae6fbd6503e231b","src/api/ops/vector_rotates.rs":"6c3f761d9d551f6365a8a95539ceace4b1a02e0b12d144f34ed68db94e88cff4","src/api/ops/vector_shifts.rs":"e510be14127c0ffd58a2573a39701da3557d66bedec09837ac8bbd44d579da00","src/api/ptr.rs":"8a793251bed6130dcfb2f1519ceaa18b751bbb15875928d0fb6deb5a5e07523a","src/api/ptr/gather_scatter.rs":"3d614f9d5b4ca201a9f7e46af4405e1d2c28ecee1620297c23b52e37b92cc0ea","src/api/reductions.rs":"ae5baca81352ecd44526d6c30c0a1feeda475ec73ddd3c3ec6b14e944e5448ee","src/api/reductions/bitwise.rs":"8bf910ae226188bd15fc7e125f058cd2566b6186fcd0cd8fd020f352c39ce139","src/api/reductions/float_arithmetic.rs":"47a5679896db2cbb56c31372fe42143da015b6beae7db5d2f3a0309ddf427ae1","src/api/reductions/integer_arithmetic.rs":"c2df3cf7493cca4174f2c65aea422a3d20d8a23af03f8d57cef72c19fee8f20d","src/api/reductions/mask.rs":"db83327a950e33a317f37fd33ca4e20c347fb415975ec024f3e23da8509425af","src/api/reductions/min_max.rs":"6af8c9aa45c69961b1b6fc205395f4767d4421869fb105fb3d563c5605fc13cd","src/api/select.rs":"6b07e7e8026df561f7307221a896f0fbb272536f41b9109040ac094c24c69331","src/api/shuffle.rs":"be7faff9b59654926df12897b2f98a4baa7d6acf2af1aaf93d388ba6e96f83ec","src/api/shuffle1_dyn.rs":"bfea5a91905b31444e9ef7ca6eddb7a9606b7e22d3f71bb842eb2795a0346620","src/api/slice.rs":"ee87484e8af329547b9a5d4f2a69e8bed6ea10bbd96270d706083843d4eea2ac","src/api/slice/from_slice.rs":"3735363000737104a8fc5f394ad8c31ec14e885952bd57647dd2a84001aee0a6","src/api/slice/write_to_slice.rs":"79d09c64d00724783c77c42e4583eeec97b18db94cf2ae146b564c3f85cfefd6","src/api/swap_bytes.rs":"05b4262eaade2f63e6cd3b780c19a03aecd2459d4cc4360051fc088887179a6e","src/codegen.rs":"db4f232fb9f5728db310b87dc8c4733be48afacab1053798c06106bef9a42b05","src/codegen/bit_manip.rs":"525ea6ff7ad1e043b6f6136992166f1803ed5563b7f6fc292c1c40257d20e264","src/codegen/llvm.rs":"12e748b4928c3be6cc12b4165c3041a3d0efccf6195338ecd3d88b8fdb0bbcc7","src/codegen/math.rs":"dfcf02ad34e2fdfe22c3f1cc2822001cc895e65031b4d06e585e5047839febb7","src/codegen/math/float.rs":"b2f31f479c5c70a6ff9ad33872c1e65506f72882b77a2e3f9e71c42e92af9355","src/codegen/math/float/abs.rs":"d5aaadcf540bdb9b4264dca6471a255fd7bf509e763bef0239c0144a68466fea","src/codegen/math/float/cos.rs":"17f28d2900c852dca221fa9c92a9cd5fe7fd2df8d427bbc60216c749b2be013d","src/codegen/math/float/cos_pi.rs":"dbaf9f443f9846a491d4ec52210a7b5835dd593b03366e3135b05c37d70f9d6c","src/codegen/math/float/exp.rs":"d300058a4bcc7ae7976f216f81902cd73a9e603ad63880dff3bbc866c27a9f37","src/codegen/math/float/ln.rs":"c851e211e43f8256093ba75b03ae0c307c9962ee66d94f09b4dd80068190cbdf","src/codegen/math/float/macros.rs":"fc9924869ed85e4795983af228cacf23158f4f35919adce16c920ad4a3f0a009","src/codegen/math/float/mul_add.rs":"041a5b69d5991d93ef795351b17560c10faf80b78fd26ad7df42a239b32cf9de","src/codegen/math/float/mul_adde.rs":"d71d5f0f3333b62a7439b823cb7adf5340ea1555ce820fb4a3f4cb922f73f5f5","src/codegen/math/float/powf.rs":"9742c3877f1a5509ca5c9492a40884b6579ba6dd11c26b7112e63f70666b395d","src/codegen/math/float/sin.rs":"0e9868d35531566509f3a01d85d5253045eb4afa8525d8407dcc1f5f33c56036","src/codegen/math/float/sin_cos_pi.rs":"8e6b6142d7dd240cdb36669722e82ab9810a2261e86e659f7d97a942ad8b1258","src/codegen/math/float/sin_pi.rs":"bb6d39db8f921e03a301fc5206ac1a61a97def8a2cb83b87ccf189f3fc48d548","src/codegen/math/float/sqrt.rs":"e6ebb0c5f428efad1f672b9a8fe4e58534dbf1ea5a8fe092ce5ce76b52fe89cb","src/codegen/math/float/sqrte.rs":"23acfaea38d0e081a6d9021c1094e813d0cfd12c58c1eca9662aade5e625d51c","src/codegen/math/float/tanh.rs":"816fd107f134920fb1a21cd792029d4b89306f6cf16d6f030cc1136823b033e7","src/codegen/pointer_sized_int.rs":"6ca13c214b6cf7e0929dbe18e96a16fc0bb7d8799608df29c4c8115490f99e01","src/codegen/reductions.rs":"8eb18ebac76985d2aa30262a2edd8cb004230b511a765d657525f677a585c12c","src/codegen/reductions/mask.rs":"e67f35a1f4d156a4894a2d6ea5a935b4d898cf70eefb2715f5c1cc165e776c11","src/codegen/reductions/mask/aarch64.rs":"84b101c17cad1ede4eb6d38cada0ac7da239dba8cea3badd3829b967e558431f","src/codegen/reductions/mask/arm.rs":"aaa07129bd078ae7e677cf8b8e67ec9f30536606a0c7ed1baaa18fd1793bb218","src/codegen/reductions/mask/fallback.rs":"3eb9319d2c7cf19216b607b8459612c4e027b643cf11b036937d36896bf76786","src/codegen/reductions/mask/fallback_impl.rs":"76547f396e55ef403327c77c314cf8db8c7a5c9b9819bfb925abeacf130249e5","src/codegen/reductions/mask/x86.rs":"36dcd8af4ab99730a078ed113d3955f74eb1a2876e2e6d9f224e0ff462c216d1","src/codegen/reductions/mask/x86/avx.rs":"3a40868b38c86e35aefb96d7578de6322efe89d8135e0366359b54ddd06f861a","src/codegen/reductions/mask/x86/avx2.rs":"677aed3f056285285daa3adff8bc65e739630b4424defa6d9665e160f027507e","src/codegen/reductions/mask/x86/sse.rs":"8522f6ed03f6c32dd577d4298df477c08aeaaa38563706f29096e1911ed731f2","src/codegen/reductions/mask/x86/sse2.rs":"54ec56e49b0c6841eccb719e4f310d65fe767c04136b2ec20bd8b9d7d9897b9e","src/codegen/shuffle.rs":"1ec2930f4e1acc43ac30b518af298d466a79e9e75734a51c380b7810efd1a27f","src/codegen/shuffle1_dyn.rs":"3f13ca1597378758d05106bf5ff3715eee531f3cb6d88f48b9182bd6c9386b51","src/codegen/swap_bytes.rs":"c67c86e91ca3fc77539e0efcea081a3c62548cccf503963ae408f2e86f4e6a21","src/codegen/v128.rs":"94226b31ec403d18d9d2fe06713f147c9c79e9b5f9105089088266313f843185","src/codegen/v16.rs":"ddec4ffb66b6f7aaffb9a1780c5ddba82557abd74f45073d335047e04cf74924","src/codegen/v256.rs":"6b63917f0444118d6b1595bff2045e59b97c4d24012bd575f69f1f0efc5a0241","src/codegen/v32.rs":"3477b3c5540aed86e61e2f5807dd31db947413cec9181c587d93ed6ec74f0eba","src/codegen/v512.rs":"5854f99d3aabc4cd42b28a20d9ce447756dc2ba024a409a69b6a8ae1f1842fc5","src/codegen/v64.rs":"e9e89caebfe63d10c0cbca61e4dfdba3b7e02ee0989170f80beed23237ddd950","src/codegen/vPtr.rs":"f0753b405cdc865bdf8e82c6505f299ea1f96136239ebbaf7f9ce93d310764b8","src/codegen/vSize.rs":"c89f5fdeb28ac4c8272ed1816fce03d9d95308cc32bb2533bd8b20cd5ac102ac","src/lib.rs":"05048c6a85ec65cf902d9dd8f757a3f76392b703a6794ea71f0d41500a89f78f","src/masks.rs":"70fc0abe4c2907ce2a491c574e1cfb9f3423385da2e1a923a48c9c13f8ba6ed8","src/sealed.rs":"ae7fdeaf5d84cd7710ed730ca72ca7eaba93df6cb0acb183e5c0a7327acf197f","src/testing.rs":"896669c08d8c801448a4d2fadc9d633eda0fbe879d229997e2a182e31278e469","src/testing/macros.rs":"403bbc5ecb7c786fe36156df302d0c07a8122408dbb15f7474d7682224ba1106","src/testing/utils.rs":"41912a92266dfe884647fc035e4242fd746100df8e839808ae0397af3759a3c8","src/v128.rs":"16cf9a8e7156b899ee9b9cd3f2dba9d13ec63289bea8c3ee9ae2e43ad9510288","src/v16.rs":"cb6465cf1e00bf530183af1819b9fe3d7eec978f8765d5e85d9b58a39a4b4045","src/v256.rs":"fe235017da18c7f3c361831c60e3173ad304d8ea1e95d64ebebc79da2d708511","src/v32.rs":"145d347855bac59b2de6508f9e594654e6c330423af9edc0e2ac8f4d1abdf45e","src/v512.rs":"f372f277f3e62eb5c945bb1c460333fdb17b6974fcc876633788ff53bded9599","src/v64.rs":"0b8079881b71575e3414be0b7f8f7eaba65281ba6732f2b2f61f73e95b6f48f7","src/vPtr.rs":"8b3e433d487180bb4304ff71245ecad90f0010f43e139a72027b672abe58facc","src/vSize.rs":"eda5aa020706cbf94d15bada41a0c2a35fc8f3f37cb7c2cd6f34d201399a495e","tests/endianness.rs":"5147f86d224c4c540b772033da2f994cad9bc9c035f38ec21e23bc4e55f8a759"},"package":"1f9f08af0c877571712e2e3e686ad79efad9657dbf0f7c3c8ba943ff6c38932d"} \ No newline at end of file diff --git a/third_party/rust/packed_simd/Cargo.toml b/third_party/rust/packed_simd/Cargo.toml new file mode 100644 index 000000000000..77fc096b0552 --- /dev/null +++ b/third_party/rust/packed_simd/Cargo.toml @@ -0,0 +1,83 @@ +# THIS FILE IS AUTOMATICALLY GENERATED BY CARGO +# +# When uploading crates to the registry Cargo will automatically +# "normalize" Cargo.toml files for maximal compatibility +# with all versions of Cargo and also rewrite `path` dependencies +# to registry (e.g., crates.io) dependencies. +# +# If you are reading this file be aware that the original Cargo.toml +# will likely look very different (and much more reasonable). +# See Cargo.toml.orig for the original contents. + +[package] +edition = "2018" +name = "packed_simd" +version = "0.3.9" +build = "build.rs" +description = "Portable Packed SIMD vectors" +homepage = "https://github.com/rust-lang/packed_simd" +documentation = "https://docs.rs/crate/packed_simd/" +readme = "README.md" +keywords = [ + "simd", + "vector", + "portability", +] +categories = [ + "hardware-support", + "concurrency", + "no-std", + "data-structures", +] +license = "MIT OR Apache-2.0" +repository = "https://github.com/rust-lang/packed_simd" + +[package.metadata.docs.rs] +features = ["into_bits"] +rustdoc-args = [ + "--cfg", + "doc_cfg", +] + +[dependencies.cfg-if] +version = "1.0.0" + +[dependencies.core_arch] +version = "0.1.5" +optional = true + +[dependencies.num-traits] +version = "0.2.14" +features = ["libm"] +default-features = false + +[dev-dependencies.arrayvec] +version = "^0.5" +default-features = false + +[dev-dependencies.paste] +version = "^1" + +[features] +default = [] +into_bits = [] +libcore_neon = [] + +[target."cfg(target_arch = \"x86_64\")".dependencies.sleef-sys] +version = "0.1.2" +optional = true + +[target.wasm32-unknown-unknown.dev-dependencies.wasm-bindgen] +version = "=0.2.87" + +[target.wasm32-unknown-unknown.dev-dependencies.wasm-bindgen-test] +version = "=0.3.37" + +[badges.is-it-maintained-issue-resolution] +repository = "rust-lang/packed_simd" + +[badges.is-it-maintained-open-issues] +repository = "rust-lang/packed_simd" + +[badges.maintenance] +status = "experimental" diff --git a/third_party/rust/any_all_workaround/LICENSE-APACHE b/third_party/rust/packed_simd/LICENSE-APACHE similarity index 100% rename from third_party/rust/any_all_workaround/LICENSE-APACHE rename to third_party/rust/packed_simd/LICENSE-APACHE diff --git a/third_party/rust/any_all_workaround/LICENSE-MIT b/third_party/rust/packed_simd/LICENSE-MIT similarity index 100% rename from third_party/rust/any_all_workaround/LICENSE-MIT rename to third_party/rust/packed_simd/LICENSE-MIT diff --git a/third_party/rust/packed_simd/README.md b/third_party/rust/packed_simd/README.md new file mode 100644 index 000000000000..59db13fe4f94 --- /dev/null +++ b/third_party/rust/packed_simd/README.md @@ -0,0 +1,144 @@ +# `Simd<[T; N]>` + +## Implementation of [Rust RFC #2366: `std::simd`][rfc2366] + +[![Latest Version]][crates.io] [![docs]][master_docs] + +**WARNING**: this crate only supports the most recent nightly Rust toolchain +and will be superseded by [`#![feature(portable_simd)]`](https://github.com/rust-lang/portable-simd). + +## Documentation + +* [API docs (`master` branch)][master_docs] +* [Performance guide][perf_guide] +* [API docs (`docs.rs`)][docs.rs] +* [RFC2366 `std::simd`][rfc2366]: - contains motivation, design rationale, + discussion, etc. + +## Examples + +Most of the examples come with both a scalar and a vectorized implementation. + +* [`aobench`](https://github.com/rust-lang-nursery/packed_simd/tree/master/examples/aobench) +* [`fannkuch_redux`](https://github.com/rust-lang-nursery/packed_simd/tree/master/examples/fannkuch_redux) +* [`matrix inverse`](https://github.com/rust-lang-nursery/packed_simd/tree/master/examples/matrix_inverse) +* [`mandelbrot`](https://github.com/rust-lang-nursery/packed_simd/tree/master/examples/mandelbrot) +* [`n-body`](https://github.com/rust-lang-nursery/packed_simd/tree/master/examples/nbody) +* [`options_pricing`](https://github.com/rust-lang-nursery/packed_simd/tree/master/examples/options_pricing) +* [`spectral_norm`](https://github.com/rust-lang-nursery/packed_simd/tree/master/examples/spectral_norm) +* [`triangle transform`](https://github.com/rust-lang-nursery/packed_simd/tree/master/examples/triangle_xform) +* [`stencil`](https://github.com/rust-lang-nursery/packed_simd/tree/master/examples/stencil) +* [`vector dot product`](https://github.com/rust-lang-nursery/packed_simd/tree/master/examples/dot_product) + +## Cargo features + +* `into_bits` (default: disabled): enables `FromBits`/`IntoBits` trait + implementations for the vector types. These allow reinterpreting the bits of a + vector type as those of another vector type safely by just using the + `.into_bits()` method. + +## Performance + +The following [ISPC] examples are also part of `packed_simd`'s +[`examples/`](https://github.com/rust-lang-nursery/packed_simd/tree/master/examples/) +directory, where `packed_simd`+[`rayon`][rayon] are used to emulate [ISPC]'s +Single-Program-Multiple-Data (SPMD) programming model. The performance results +on different hardware is shown in the `readme.md` of each example. The following +table summarizes the performance ranges, where `+` means speed-up and `-` +slowdown: + +* `aobench`: `[-1.02x, +1.53x]`, +* `stencil`: `[+1.06x, +1.72x]`, +* `mandelbrot`: `[-1.74x, +1.2x]`, +* `options_pricing`: + * `black_scholes`: `+1.0x` + * `binomial_put`: `+1.4x` + + While SPMD is not the intended use case for `packed_simd`, it is possible to + combine the library with [`rayon`][rayon] to poorly emulate [ISPC]'s SPMD programming + model in Rust. Writing performant code is not as straightforward as with + [ISPC], but with some care (e.g. see the [Performance Guide][perf_guide]) one + can easily match and often out-perform [ISPC]'s "default performance". + +## Platform support + +The following table describes the supported platforms: `build` shows whether +the library compiles without issues for a given target, while `run` shows +whether the test suite passes for a given target. + +| **Linux** | **build** | **run** | +|---------------------------------------|-----------|---------| +| `i586-unknown-linux-gnu` | ✓ | ✗ | +| `i686-unknown-linux-gnu` | ✓ | ✗ | +| `x86_64-unknown-linux-gnu` | ✓ | ✓ | +| `arm-unknown-linux-gnueabihf` | ✓ | ✓ | +| `armv7-unknown-linux-gnueabi` | ✓ | ✓ | +| `aarch64-unknown-linux-gnu` | ✓ | ✓ | +| `powerpc-unknown-linux-gnu` | ✓ | ✗ | +| `powerpc64-unknown-linux-gnu` | ✓ | ✗ | +| `powerpc64le-unknown-linux-gnu` | ✓ | ✓ | +| `s390x-unknown-linux-gnu` | ✓ | ✗ | +| `sparc64-unknown-linux-gnu` | ✓ | ✗ | +| `thumbv7neon-unknown-linux-gnueabihf` | ✓ | ✓ | +| **MacOSX** | **build** | **run** | +| `x86_64-apple-darwin` | ✓ | ✓ | +| **Android** | **build** | **run** | +| `x86_64-linux-android` | ✓ | ✓ | +| `armv7-linux-androideabi` | ✓ | ✗ | +| `aarch64-linux-android` | ✓ | ✗ | +| `thumbv7neon-linux-androideabi` | ✓ | ✗ | +| **iOS** | **build** | **run** | +| `x86_64-apple-ios` | ✗ | ✗ | +| `aarch64-apple-ios` | ✗ | ✗ | + + +## Machine code verification + +The +[`verify/`](https://github.com/rust-lang-nursery/packed_simd/tree/master/verify) +crate tests disassembles the portable packed vector APIs at run-time and +compares the generated machine code against the desired one to make sure that +this crate remains efficient. + +## License + +This project is licensed under either of + +* [Apache License, Version 2.0](http://www.apache.org/licenses/LICENSE-2.0) + ([LICENSE-APACHE](LICENSE-APACHE)) + +* [MIT License](http://opensource.org/licenses/MIT) + ([LICENSE-MIT](LICENSE-MIT)) + +at your option. + +## Contributing + +We welcome all people who want to contribute. +Please see the [contributing instructions] for more information. + +Contributions in any form (issues, pull requests, etc.) to this project +must adhere to Rust's [Code of Conduct]. + +Unless you explicitly state otherwise, any contribution intentionally submitted +for inclusion in `packed_simd` by you, as defined in the Apache-2.0 license, shall be +dual licensed as above, without any additional terms or conditions. + +[travis]: https://travis-ci.com/rust-lang/packed_simd +[Travis-CI Status]: https://travis-ci.com/rust-lang/packed_simd.svg?branch=master +[appveyor]: https://ci.appveyor.com/project/gnzlbg/packed-simd +[Appveyor Status]: https://ci.appveyor.com/api/projects/status/hd7v9dvr442hgdix?svg=true +[Latest Version]: https://img.shields.io/crates/v/packed_simd.svg +[crates.io]: https://crates.io/crates/packed_simd +[docs]: https://docs.rs/packed_simd/badge.svg +[docs.rs]: https://docs.rs/packed_simd +[master_docs]: https://rust-lang-nursery.github.io/packed_simd/packed_simd/ +[perf_guide]: https://rust-lang-nursery.github.io/packed_simd/perf-guide/ +[rfc2366]: https://github.com/rust-lang/rfcs/pull/2366 +[ISPC]: https://ispc.github.io/ +[rayon]: https://crates.io/crates/rayon +[boost_license]: https://www.boost.org/LICENSE_1_0.txt +[SLEEF]: https://sleef.org/ +[sleef_sys]: https://crates.io/crates/sleef-sys +[contributing instructions]: contributing.md +[Code of Conduct]: https://www.rust-lang.org/en-US/conduct.html diff --git a/third_party/rust/packed_simd/bors.toml b/third_party/rust/packed_simd/bors.toml new file mode 100644 index 000000000000..6d302dc85cf6 --- /dev/null +++ b/third_party/rust/packed_simd/bors.toml @@ -0,0 +1,3 @@ +status = [ + "continuous-integration/travis-ci/push" +] \ No newline at end of file diff --git a/third_party/rust/packed_simd/build.rs b/third_party/rust/packed_simd/build.rs new file mode 100644 index 000000000000..e87298a2de23 --- /dev/null +++ b/third_party/rust/packed_simd/build.rs @@ -0,0 +1,6 @@ +fn main() { + let target = std::env::var("TARGET").expect("TARGET environment variable not defined"); + if target.contains("neon") { + println!("cargo:rustc-cfg=libcore_neon"); + } +} diff --git a/third_party/rust/packed_simd/ci/all.sh b/third_party/rust/packed_simd/ci/all.sh new file mode 100755 index 000000000000..55a1fa2efefe --- /dev/null +++ b/third_party/rust/packed_simd/ci/all.sh @@ -0,0 +1,71 @@ +#!/usr/bin/env bash +# +# Performs an operation on all targets + +set -ex + +: "${1?The all.sh script requires one argument.}" + +op=$1 + +cargo_clean() { + cargo clean +} + +cargo_check_fmt() { + cargo fmt --all -- --check +} + +cargo_fmt() { + cargo fmt --all +} + +cargo_clippy() { + cargo clippy --all -- -D clippy::perf +} + +CMD="-1" + +case $op in + clean*) + CMD=cargo_clean + ;; + check_fmt*) + CMD=cargo_check_fmt + ;; + fmt*) + CMD=cargo_fmt + ;; + clippy) + CMD=cargo_clippy + ;; + *) + echo "Unknown operation: \"${op}\"" + exit 1 + ;; +esac + +echo "Operation is: ${CMD}" + +# On src/ +$CMD + +# Check examples/ +for dir in examples/*/ +do + dir=${dir%*/} + ( + cd "${dir%*/}" + $CMD + ) +done + +( + cd verify/verify + $CMD +) + +( + cd micro_benchmarks + $CMD +) diff --git a/third_party/rust/packed_simd/ci/android-install-ndk.sh b/third_party/rust/packed_simd/ci/android-install-ndk.sh new file mode 100644 index 000000000000..5370853937c7 --- /dev/null +++ b/third_party/rust/packed_simd/ci/android-install-ndk.sh @@ -0,0 +1,21 @@ +#!/usr/bin/env sh +# Copyright 2016 The Rust Project Developers. See the COPYRIGHT +# file at the top-level directory of this distribution and at +# http://rust-lang.org/COPYRIGHT. +# +# Licensed under the Apache License, Version 2.0 or the MIT license +# , at your +# option. This file may not be copied, modified, or distributed +# except according to those terms. + +set -ex + +ANDROID_NDK_URL=https://dl.google.com/android/repository +ANDROID_NDK_ARCHIVE=android-ndk-r25b-linux.zip + +curl -fO "$ANDROID_NDK_URL/$ANDROID_NDK_ARCHIVE" +unzip -q $ANDROID_NDK_ARCHIVE +rm $ANDROID_NDK_ARCHIVE +mv android-ndk-* ndk +rm -rf android-ndk-* diff --git a/third_party/rust/packed_simd/ci/android-install-sdk.sh b/third_party/rust/packed_simd/ci/android-install-sdk.sh new file mode 100644 index 000000000000..6b5ac09ab04a --- /dev/null +++ b/third_party/rust/packed_simd/ci/android-install-sdk.sh @@ -0,0 +1,60 @@ +#!/usr/bin/env sh +# Copyright 2016 The Rust Project Developers. See the COPYRIGHT +# file at the top-level directory of this distribution and at +# http://rust-lang.org/COPYRIGHT. +# +# Licensed under the Apache License, Version 2.0 or the MIT license +# , at your +# option. This file may not be copied, modified, or distributed +# except according to those terms. + +set -ex + +# Prep the SDK and emulator +# +# Note that the update process requires that we accept a bunch of licenses, and +# we can't just pipe `yes` into it for some reason, so we take the same strategy +# located in https://github.com/appunite/docker by just wrapping it in a script +# which apparently magically accepts the licenses. + +mkdir sdk +curl --retry 5 https://dl.google.com/android/repository/sdk-tools-linux-3859397.zip -O +unzip -d sdk sdk-tools-linux-3859397.zip + +case "$1" in + arm | armv7) + abi=armeabi-v7a + ;; + + aarch64) + abi=arm64-v8a + ;; + + i686) + abi=x86 + ;; + + x86_64) + abi=x86_64 + ;; + + *) + echo "invalid arch: $1" + exit 1 + ;; +esac; + +# --no_https avoids + # javax.net.ssl.SSLHandshakeException: sun.security.validator.ValidatorException: No trusted certificate found +yes | ./sdk/tools/bin/sdkmanager --licenses --no_https +yes | ./sdk/tools/bin/sdkmanager --no_https \ + "emulator" \ + "platform-tools" \ + "platforms;android-24" \ + "system-images;android-24;default;$abi" + +echo "no" | + ./sdk/tools/bin/avdmanager create avd \ + --name "${1}" \ + --package "system-images;android-24;default;$abi" diff --git a/third_party/rust/packed_simd/ci/android-sysimage.sh b/third_party/rust/packed_simd/ci/android-sysimage.sh new file mode 100644 index 000000000000..9eabd7c8d94f --- /dev/null +++ b/third_party/rust/packed_simd/ci/android-sysimage.sh @@ -0,0 +1,56 @@ +#!/usr/bin/env bash + +# Copyright 2017 The Rust Project Developers. See the COPYRIGHT +# file at the top-level directory of this distribution and at +# http://rust-lang.org/COPYRIGHT. +# +# Licensed under the Apache License, Version 2.0 or the MIT license +# , at your +# option. This file may not be copied, modified, or distributed +# except according to those terms. + +set -ex + +URL=https://dl.google.com/android/repository/sys-img/android + +main() { + local arch="${1}" + local name="${2}" + local dest=/system + local td + td="$(mktemp -d)" + + apt-get install --no-install-recommends e2tools + + pushd "${td}" + curl --retry 5 -O "${URL}/${name}" + unzip -q "${name}" + + local system + system="$(find . -name system.img)" + mkdir -p ${dest}/{bin,lib,lib64} + + # Extract android linker and libraries to /system + # This allows android executables to be run directly (or with qemu) + if [ "${arch}" = "x86_64" ] || [ "${arch}" = "arm64" ]; then + e2cp -p "${system}:/bin/linker64" "${dest}/bin/" + e2cp -p "${system}:/lib64/libdl.so" "${dest}/lib64/" + e2cp -p "${system}:/lib64/libc.so" "${dest}/lib64/" + e2cp -p "${system}:/lib64/libm.so" "${dest}/lib64/" + else + e2cp -p "${system}:/bin/linker" "${dest}/bin/" + e2cp -p "${system}:/lib/libdl.so" "${dest}/lib/" + e2cp -p "${system}:/lib/libc.so" "${dest}/lib/" + e2cp -p "${system}:/lib/libm.so" "${dest}/lib/" + fi + + # clean up + apt-get purge --auto-remove -y e2tools + + popd + + rm -rf "${td}" +} + +main "${@}" diff --git a/third_party/rust/packed_simd/ci/benchmark.sh b/third_party/rust/packed_simd/ci/benchmark.sh new file mode 100755 index 000000000000..3635b9e371d1 --- /dev/null +++ b/third_party/rust/packed_simd/ci/benchmark.sh @@ -0,0 +1,32 @@ +#!/usr/bin/env bash +# +# Runs all benchmarks. Controlled by the following environment variables: +# +# FEATURES={} - cargo features to pass to all benchmarks (e.g. core_arch,sleef-sys,ispc) +# NORUN={1} - only builds the benchmarks + +set -ex + +if [[ ${NORUN} != 1 ]]; then + # Most benchmarks require hyperfine; require it upfront. + hash hyperfine 2>/dev/null || { echo >&2 "hyperfine is not in PATH."; exit 1; } +fi + + +# If the ispc benchmark feature is enabled, ispc must be in the path of the +# benchmarks. +if echo "$FEATURES" | grep -q "ispc"; then + hash ispc 2>/dev/null || { echo >&2 "ispc is not in PATH."; exit 1; } +fi + +# An example with a benchmark.sh is a benchmark: +for dir in examples/*/ +do + dir=${dir%*/} + cd ${dir%*/} + if [ -f "benchmark.sh" ]; then + ./benchmark.sh + fi + cd - +done + diff --git a/third_party/rust/packed_simd/ci/deploy_and_run_on_ios_simulator.rs b/third_party/rust/packed_simd/ci/deploy_and_run_on_ios_simulator.rs new file mode 100644 index 000000000000..c0fe52c35659 --- /dev/null +++ b/third_party/rust/packed_simd/ci/deploy_and_run_on_ios_simulator.rs @@ -0,0 +1,176 @@ +// Copyright 2017 The Rust Project Developers. See the COPYRIGHT +// file at the top-level directory of this distribution and at +// http://rust-lang.org/COPYRIGHT. +// +// Licensed under the Apache License, Version 2.0 or the MIT license +// , at your +// option. This file may not be copied, modified, or distributed +// except according to those terms. + +// This is a script to deploy and execute a binary on an iOS simulator. +// The primary use of this is to be able to run unit tests on the simulator and +// retrieve the results. +// +// To do this through Cargo instead, use Dinghy +// (https://github.com/snipsco/dinghy): cargo dinghy install, then cargo dinghy +// test. + +use std::env; +use std::fs::{self, File}; +use std::io::Write; +use std::path::Path; +use std::process; +use std::process::Command; + +macro_rules! t { + ($e:expr) => (match $e { + Ok(e) => e, + Err(e) => panic!("{} failed with: {}", stringify!($e), e), + }) +} + +// Step one: Wrap as an app +fn package_as_simulator_app(crate_name: &str, test_binary_path: &Path) { + println!("Packaging simulator app"); + drop(fs::remove_dir_all("ios_simulator_app")); + t!(fs::create_dir("ios_simulator_app")); + t!(fs::copy(test_binary_path, + Path::new("ios_simulator_app").join(crate_name))); + + let mut f = t!(File::create("ios_simulator_app/Info.plist")); + t!(f.write_all(format!(r#" + + + + + CFBundleExecutable + {} + CFBundleIdentifier + com.rust.unittests + + + "#, crate_name).as_bytes())); +} + +// Step two: Start the iOS simulator +fn start_simulator() { + println!("Looking for iOS simulator"); + let output = t!(Command::new("xcrun").arg("simctl").arg("list").output()); + assert!(output.status.success()); + let mut simulator_exists = false; + let mut simulator_booted = false; + let mut found_rust_sim = false; + let stdout = t!(String::from_utf8(output.stdout)); + for line in stdout.lines() { + if line.contains("rust_ios") { + if found_rust_sim { + panic!("Duplicate rust_ios simulators found. Please \ + double-check xcrun simctl list."); + } + simulator_exists = true; + simulator_booted = line.contains("(Booted)"); + found_rust_sim = true; + } + } + + if simulator_exists == false { + println!("Creating iOS simulator"); + Command::new("xcrun") + .arg("simctl") + .arg("create") + .arg("rust_ios") + .arg("com.apple.CoreSimulator.SimDeviceType.iPhone-SE") + .arg("com.apple.CoreSimulator.SimRuntime.iOS-10-2") + .check_status(); + } else if simulator_booted == true { + println!("Shutting down already-booted simulator"); + Command::new("xcrun") + .arg("simctl") + .arg("shutdown") + .arg("rust_ios") + .check_status(); + } + + println!("Starting iOS simulator"); + // We can't uninstall the app (if present) as that will hang if the + // simulator isn't completely booted; just erase the simulator instead. + Command::new("xcrun").arg("simctl").arg("erase").arg("rust_ios").check_status(); + Command::new("xcrun").arg("simctl").arg("boot").arg("rust_ios").check_status(); +} + +// Step three: Install the app +fn install_app_to_simulator() { + println!("Installing app to simulator"); + Command::new("xcrun") + .arg("simctl") + .arg("install") + .arg("booted") + .arg("ios_simulator_app/") + .check_status(); +} + +// Step four: Run the app +fn run_app_on_simulator() { + println!("Running app"); + let output = t!(Command::new("xcrun") + .arg("simctl") + .arg("launch") + .arg("--console") + .arg("booted") + .arg("com.rust.unittests") + .output()); + + println!("stdout --\n{}\n", String::from_utf8_lossy(&output.stdout)); + println!("stderr --\n{}\n", String::from_utf8_lossy(&output.stderr)); + + let stdout = String::from_utf8_lossy(&output.stdout); + let failed = stdout.lines() + .find(|l| l.contains("FAILED")) + .map(|l| l.contains("FAILED")) + .unwrap_or(false); + + let passed = stdout.lines() + .find(|l| l.contains("test result: ok")) + .map(|l| l.contains("test result: ok")) + .unwrap_or(false); + + println!("Shutting down simulator"); + Command::new("xcrun") + .arg("simctl") + .arg("shutdown") + .arg("rust_ios") + .check_status(); + if !(passed && !failed) { + panic!("tests didn't pass"); + } +} + +trait CheckStatus { + fn check_status(&mut self); +} + +impl CheckStatus for Command { + fn check_status(&mut self) { + println!("\trunning: {:?}", self); + assert!(t!(self.status()).success()); + } +} + +fn main() { + let args: Vec = env::args().collect(); + if args.len() != 2 { + println!("Usage: {} ", args[0]); + process::exit(-1); + } + + let test_binary_path = Path::new(&args[1]); + let crate_name = test_binary_path.file_name().unwrap(); + + package_as_simulator_app(crate_name.to_str().unwrap(), test_binary_path); + start_simulator(); + install_app_to_simulator(); + run_app_on_simulator(); +} diff --git a/third_party/rust/packed_simd/ci/docker/aarch64-linux-android/Dockerfile b/third_party/rust/packed_simd/ci/docker/aarch64-linux-android/Dockerfile new file mode 100644 index 000000000000..27bde89c5a8d --- /dev/null +++ b/third_party/rust/packed_simd/ci/docker/aarch64-linux-android/Dockerfile @@ -0,0 +1,47 @@ +FROM ubuntu:16.04 + +RUN dpkg --add-architecture i386 && \ + apt-get update && \ + apt-get install -y --no-install-recommends \ + file \ + make \ + curl \ + ca-certificates \ + python \ + unzip \ + expect \ + openjdk-9-jre \ + libstdc++6:i386 \ + libpulse0 \ + gcc \ + libc6-dev + +WORKDIR /android/ +COPY android* /android/ + +ENV ANDROID_ARCH=aarch64 +ENV PATH=$PATH:/android/ndk-$ANDROID_ARCH/bin:/android/sdk/tools:/android/sdk/platform-tools + +RUN sh /android/android-install-ndk.sh $ANDROID_ARCH +RUN sh /android/android-install-sdk.sh $ANDROID_ARCH +RUN mv /root/.android /tmp +RUN chmod 777 -R /tmp/.android +RUN chmod 755 /android/sdk/tools/* /android/sdk/emulator/qemu/linux-x86_64/* + +ENV PATH=$PATH:/rust/bin \ + CARGO_TARGET_AARCH64_LINUX_ANDROID_LINKER=aarch64-linux-android-gcc \ + CARGO_TARGET_AARCH64_LINUX_ANDROID_RUNNER=/tmp/runtest \ + OBJDUMP=aarch64-linux-android-objdump \ + HOME=/tmp + +ADD runtest-android.rs /tmp/runtest.rs +ENTRYPOINT [ \ + "bash", \ + "-c", \ + # set SHELL so android can detect a 64bits system, see + # http://stackoverflow.com/a/41789144 + "SHELL=/bin/dash /android/sdk/emulator/emulator @aarch64 -no-window & \ + rustc /tmp/runtest.rs -o /tmp/runtest && \ + exec \"$@\"", \ + "--" \ +] diff --git a/third_party/rust/packed_simd/ci/docker/aarch64-unknown-linux-gnu/Dockerfile b/third_party/rust/packed_simd/ci/docker/aarch64-unknown-linux-gnu/Dockerfile new file mode 100644 index 000000000000..41ff4729ac59 --- /dev/null +++ b/third_party/rust/packed_simd/ci/docker/aarch64-unknown-linux-gnu/Dockerfile @@ -0,0 +1,14 @@ +FROM ubuntu:18.04 +RUN apt-get update && apt-get install -y --no-install-recommends \ + gcc \ + ca-certificates \ + libc6-dev \ + gcc-aarch64-linux-gnu \ + libc6-dev-arm64-cross \ + qemu-user \ + make \ + file + +ENV CARGO_TARGET_AARCH64_UNKNOWN_LINUX_GNU_LINKER=aarch64-linux-gnu-gcc \ + CARGO_TARGET_AARCH64_UNKNOWN_LINUX_GNU_RUNNER="qemu-aarch64 -L /usr/aarch64-linux-gnu" \ + OBJDUMP=aarch64-linux-gnu-objdump diff --git a/third_party/rust/packed_simd/ci/docker/arm-unknown-linux-gnueabi/Dockerfile b/third_party/rust/packed_simd/ci/docker/arm-unknown-linux-gnueabi/Dockerfile new file mode 100644 index 000000000000..e1c591dd979a --- /dev/null +++ b/third_party/rust/packed_simd/ci/docker/arm-unknown-linux-gnueabi/Dockerfile @@ -0,0 +1,15 @@ +FROM ubuntu:18.04 +RUN apt-get update && apt-get install -y --no-install-recommends \ + gcc \ + ca-certificates \ + libc6-dev \ + libc6-armel-cross \ + libc6-dev-armel-cross \ + binutils-arm-linux-gnueabi \ + gcc-arm-linux-gnueabi \ + qemu-user \ + make \ + file +ENV CARGO_TARGET_ARM_UNKNOWN_LINUX_GNUEABI_LINKER=arm-linux-gnueabi-gcc \ + CARGO_TARGET_ARM_UNKNOWN_LINUX_GNUEABI_RUNNER="qemu-arm -L /usr/arm-linux-gnueabi" \ + OBJDUMP=arm-linux-gnueabi-objdump diff --git a/third_party/rust/packed_simd/ci/docker/arm-unknown-linux-gnueabihf/Dockerfile b/third_party/rust/packed_simd/ci/docker/arm-unknown-linux-gnueabihf/Dockerfile new file mode 100644 index 000000000000..757b79e7ecc1 --- /dev/null +++ b/third_party/rust/packed_simd/ci/docker/arm-unknown-linux-gnueabihf/Dockerfile @@ -0,0 +1,13 @@ +FROM ubuntu:18.04 +RUN apt-get update && apt-get install -y --no-install-recommends \ + gcc \ + ca-certificates \ + libc6-dev \ + gcc-arm-linux-gnueabihf \ + libc6-dev-armhf-cross \ + qemu-user \ + make \ + file +ENV CARGO_TARGET_ARM_UNKNOWN_LINUX_GNUEABIHF_LINKER=arm-linux-gnueabihf-gcc \ + CARGO_TARGET_ARM_UNKNOWN_LINUX_GNUEABIHF_RUNNER="qemu-arm -L /usr/arm-linux-gnueabihf" \ + OBJDUMP=arm-linux-gnueabihf-objdump diff --git a/third_party/rust/packed_simd/ci/docker/armv7-linux-androideabi/Dockerfile b/third_party/rust/packed_simd/ci/docker/armv7-linux-androideabi/Dockerfile new file mode 100644 index 000000000000..995a9e30e65e --- /dev/null +++ b/third_party/rust/packed_simd/ci/docker/armv7-linux-androideabi/Dockerfile @@ -0,0 +1,47 @@ +FROM ubuntu:16.04 + +RUN dpkg --add-architecture i386 && \ + apt-get update && \ + apt-get install -y --no-install-recommends \ + file \ + make \ + curl \ + ca-certificates \ + python \ + unzip \ + expect \ + openjdk-9-jre \ + libstdc++6:i386 \ + libpulse0 \ + gcc \ + libc6-dev + +WORKDIR /android/ +COPY android* /android/ + +ENV ANDROID_ARCH=arm +ENV PATH=$PATH:/android/ndk-$ANDROID_ARCH/bin:/android/sdk/tools:/android/sdk/platform-tools + +RUN sh /android/android-install-ndk.sh $ANDROID_ARCH +RUN sh /android/android-install-sdk.sh $ANDROID_ARCH +RUN mv /root/.android /tmp +RUN chmod 777 -R /tmp/.android +RUN chmod 755 /android/sdk/tools/* /android/sdk/emulator/qemu/linux-x86_64/* + +ENV PATH=$PATH:/rust/bin \ + CARGO_TARGET_ARM_LINUX_ANDROIDEABI_LINKER=arm-linux-androideabi-gcc \ + CARGO_TARGET_ARM_LINUX_ANDROIDEABI_RUNNER=/tmp/runtest \ + OBJDUMP=arm-linux-androideabi-objdump \ + HOME=/tmp + +ADD runtest-android.rs /tmp/runtest.rs +ENTRYPOINT [ \ + "bash", \ + "-c", \ + # set SHELL so android can detect a 64bits system, see + # http://stackoverflow.com/a/41789144 + "SHELL=/bin/dash /android/sdk/emulator/emulator @arm -no-window & \ + rustc /tmp/runtest.rs -o /tmp/runtest && \ + exec \"$@\"", \ + "--" \ +] diff --git a/third_party/rust/packed_simd/ci/docker/armv7-unknown-linux-gnueabihf/Dockerfile b/third_party/rust/packed_simd/ci/docker/armv7-unknown-linux-gnueabihf/Dockerfile new file mode 100644 index 000000000000..253906293374 --- /dev/null +++ b/third_party/rust/packed_simd/ci/docker/armv7-unknown-linux-gnueabihf/Dockerfile @@ -0,0 +1,13 @@ +FROM ubuntu:18.04 +RUN apt-get update && apt-get install -y --no-install-recommends \ + gcc \ + ca-certificates \ + libc6-dev \ + gcc-arm-linux-gnueabihf \ + libc6-dev-armhf-cross \ + qemu-user \ + make \ + file +ENV CARGO_TARGET_ARMV7_UNKNOWN_LINUX_GNUEABIHF_LINKER=arm-linux-gnueabihf-gcc \ + CARGO_TARGET_ARMV7_UNKNOWN_LINUX_GNUEABIHF_RUNNER="qemu-arm -L /usr/arm-linux-gnueabihf" \ + OBJDUMP=arm-linux-gnueabihf-objdump diff --git a/third_party/rust/packed_simd/ci/docker/i586-unknown-linux-gnu/Dockerfile b/third_party/rust/packed_simd/ci/docker/i586-unknown-linux-gnu/Dockerfile new file mode 100644 index 000000000000..01093698f679 --- /dev/null +++ b/third_party/rust/packed_simd/ci/docker/i586-unknown-linux-gnu/Dockerfile @@ -0,0 +1,7 @@ +FROM ubuntu:18.04 +RUN apt-get update && apt-get install -y --no-install-recommends \ + gcc-multilib \ + libc6-dev \ + file \ + make \ + ca-certificates diff --git a/third_party/rust/packed_simd/ci/docker/i686-unknown-linux-gnu/Dockerfile b/third_party/rust/packed_simd/ci/docker/i686-unknown-linux-gnu/Dockerfile new file mode 100644 index 000000000000..01093698f679 --- /dev/null +++ b/third_party/rust/packed_simd/ci/docker/i686-unknown-linux-gnu/Dockerfile @@ -0,0 +1,7 @@ +FROM ubuntu:18.04 +RUN apt-get update && apt-get install -y --no-install-recommends \ + gcc-multilib \ + libc6-dev \ + file \ + make \ + ca-certificates diff --git a/third_party/rust/packed_simd/ci/docker/mips-unknown-linux-gnu/Dockerfile b/third_party/rust/packed_simd/ci/docker/mips-unknown-linux-gnu/Dockerfile new file mode 100644 index 000000000000..3bd471e87d4d --- /dev/null +++ b/third_party/rust/packed_simd/ci/docker/mips-unknown-linux-gnu/Dockerfile @@ -0,0 +1,13 @@ +FROM ubuntu:18.04 + +RUN apt-get update && apt-get install -y --no-install-recommends \ + gcc libc6-dev qemu-user ca-certificates \ + gcc-mips-linux-gnu libc6-dev-mips-cross \ + qemu-system-mips \ + qemu-user \ + make \ + file + +ENV CARGO_TARGET_MIPS_UNKNOWN_LINUX_GNU_LINKER=mips-linux-gnu-gcc \ + CARGO_TARGET_MIPS_UNKNOWN_LINUX_GNU_RUNNER="qemu-mips -L /usr/mips-linux-gnu" \ + OBJDUMP=mips-linux-gnu-objdump \ No newline at end of file diff --git a/third_party/rust/packed_simd/ci/docker/mips64-unknown-linux-gnuabi64/Dockerfile b/third_party/rust/packed_simd/ci/docker/mips64-unknown-linux-gnuabi64/Dockerfile new file mode 100644 index 000000000000..f26f1f38eb22 --- /dev/null +++ b/third_party/rust/packed_simd/ci/docker/mips64-unknown-linux-gnuabi64/Dockerfile @@ -0,0 +1,10 @@ +FROM ubuntu:18.04 + +RUN apt-get update && apt-get install -y --no-install-recommends \ + gcc libc6-dev qemu-user ca-certificates \ + gcc-mips64-linux-gnuabi64 libc6-dev-mips64-cross \ + qemu-system-mips64 qemu-user + +ENV CARGO_TARGET_MIPS64_UNKNOWN_LINUX_GNUABI64_LINKER=mips64-linux-gnuabi64-gcc \ + CARGO_TARGET_MIPS64_UNKNOWN_LINUX_GNUABI64_RUNNER="qemu-mips64 -L /usr/mips64-linux-gnuabi64" \ + OBJDUMP=mips64-linux-gnuabi64-objdump \ No newline at end of file diff --git a/third_party/rust/packed_simd/ci/docker/mips64el-unknown-linux-gnuabi64/Dockerfile b/third_party/rust/packed_simd/ci/docker/mips64el-unknown-linux-gnuabi64/Dockerfile new file mode 100644 index 000000000000..7d9f0bd99250 --- /dev/null +++ b/third_party/rust/packed_simd/ci/docker/mips64el-unknown-linux-gnuabi64/Dockerfile @@ -0,0 +1,10 @@ +FROM ubuntu:18.04 + +RUN apt-get update && apt-get install -y --no-install-recommends \ + gcc libc6-dev qemu-user ca-certificates \ + gcc-mips64el-linux-gnuabi64 libc6-dev-mips64el-cross \ + qemu-system-mips64el + +ENV CARGO_TARGET_MIPS64EL_UNKNOWN_LINUX_GNUABI64_LINKER=mips64el-linux-gnuabi64-gcc \ + CARGO_TARGET_MIPS64EL_UNKNOWN_LINUX_GNUABI64_RUNNER="qemu-mips64el -L /usr/mips64el-linux-gnuabi64" \ + OBJDUMP=mips64el-linux-gnuabi64-objdump \ No newline at end of file diff --git a/third_party/rust/packed_simd/ci/docker/mipsel-unknown-linux-musl/Dockerfile b/third_party/rust/packed_simd/ci/docker/mipsel-unknown-linux-musl/Dockerfile new file mode 100644 index 000000000000..7488662ef281 --- /dev/null +++ b/third_party/rust/packed_simd/ci/docker/mipsel-unknown-linux-musl/Dockerfile @@ -0,0 +1,25 @@ +FROM ubuntu:18.10 + +RUN apt-get update && \ + apt-get install -y --no-install-recommends \ + ca-certificates \ + gcc \ + libc6-dev \ + make \ + qemu-user \ + qemu-system-mips \ + bzip2 \ + curl \ + file + +RUN mkdir /toolchain + +# Note that this originally came from: +# https://downloads.openwrt.org/snapshots/trunk/malta/generic/OpenWrt-Toolchain-malta-le_gcc-5.3.0_musl-1.1.15.Linux-x86_64.tar.bz2 +RUN curl -L https://ci-mirrors.rust-lang.org/libc/OpenWrt-Toolchain-malta-le_gcc-5.3.0_musl-1.1.15.Linux-x86_64.tar.bz2 | \ + tar xjf - -C /toolchain --strip-components=2 + +ENV PATH=$PATH:/rust/bin:/toolchain/bin \ + CC_mipsel_unknown_linux_musl=mipsel-openwrt-linux-gcc \ + CARGO_TARGET_MIPSEL_UNKNOWN_LINUX_MUSL_LINKER=mipsel-openwrt-linux-gcc \ + CARGO_TARGET_MIPSEL_UNKNOWN_LINUX_MUSL_RUNNER="qemu-mipsel -L /toolchain" diff --git a/third_party/rust/packed_simd/ci/docker/powerpc-unknown-linux-gnu/Dockerfile b/third_party/rust/packed_simd/ci/docker/powerpc-unknown-linux-gnu/Dockerfile new file mode 100644 index 000000000000..15ba58e60ce2 --- /dev/null +++ b/third_party/rust/packed_simd/ci/docker/powerpc-unknown-linux-gnu/Dockerfile @@ -0,0 +1,13 @@ +FROM ubuntu:22.04 + +RUN apt-get update && apt-get install -y --no-install-recommends \ + gcc libc6-dev qemu-user ca-certificates \ + gcc-powerpc-linux-gnu libc6-dev-powerpc-cross \ + qemu-system-ppc \ + make \ + file + +ENV CARGO_TARGET_POWERPC_UNKNOWN_LINUX_GNU_LINKER=powerpc-linux-gnu-gcc \ + CARGO_TARGET_POWERPC_UNKNOWN_LINUX_GNU_RUNNER="qemu-ppc -cpu Vger -L /usr/powerpc-linux-gnu" \ + CC=powerpc-linux-gnu-gcc \ + OBJDUMP=powerpc-linux-gnu-objdump diff --git a/third_party/rust/packed_simd/ci/docker/powerpc64-unknown-linux-gnu/Dockerfile b/third_party/rust/packed_simd/ci/docker/powerpc64-unknown-linux-gnu/Dockerfile new file mode 100644 index 000000000000..21c296dc442a --- /dev/null +++ b/third_party/rust/packed_simd/ci/docker/powerpc64-unknown-linux-gnu/Dockerfile @@ -0,0 +1,17 @@ +FROM ubuntu:22.04 + +RUN apt-get update && apt-get install -y --no-install-recommends \ + gcc \ + ca-certificates \ + libc6-dev \ + gcc-powerpc64-linux-gnu \ + libc6-dev-ppc64-cross \ + qemu-user \ + qemu-system-ppc \ + make \ + file + +ENV CARGO_TARGET_POWERPC64_UNKNOWN_LINUX_GNU_LINKER=powerpc64-linux-gnu-gcc \ + CARGO_TARGET_POWERPC64_UNKNOWN_LINUX_GNU_RUNNER="qemu-ppc64 -L /usr/powerpc64-linux-gnu" \ + CC=powerpc64-linux-gnu-gcc \ + OBJDUMP=powerpc64-linux-gnu-objdump diff --git a/third_party/rust/packed_simd/ci/docker/powerpc64le-unknown-linux-gnu/Dockerfile b/third_party/rust/packed_simd/ci/docker/powerpc64le-unknown-linux-gnu/Dockerfile new file mode 100644 index 000000000000..8034145fc06e --- /dev/null +++ b/third_party/rust/packed_simd/ci/docker/powerpc64le-unknown-linux-gnu/Dockerfile @@ -0,0 +1,11 @@ +FROM ubuntu:22.04 + +RUN apt-get update && apt-get install -y --no-install-recommends \ + gcc libc6-dev qemu-user ca-certificates \ + gcc-powerpc64le-linux-gnu libc6-dev-ppc64el-cross \ + qemu-system-ppc file make + +ENV CARGO_TARGET_POWERPC64LE_UNKNOWN_LINUX_GNU_LINKER=powerpc64le-linux-gnu-gcc \ + CARGO_TARGET_POWERPC64LE_UNKNOWN_LINUX_GNU_RUNNER="qemu-ppc64le -L /usr/powerpc64le-linux-gnu" \ + CC=powerpc64le-linux-gnu-gcc \ + OBJDUMP=powerpc64le-linux-gnu-objdump diff --git a/third_party/rust/packed_simd/ci/docker/s390x-unknown-linux-gnu/Dockerfile b/third_party/rust/packed_simd/ci/docker/s390x-unknown-linux-gnu/Dockerfile new file mode 100644 index 000000000000..e785ca370c4d --- /dev/null +++ b/third_party/rust/packed_simd/ci/docker/s390x-unknown-linux-gnu/Dockerfile @@ -0,0 +1,20 @@ +FROM ubuntu:22.04 + +RUN apt-get update && \ + apt-get install -y --no-install-recommends \ + ca-certificates \ + curl \ + cmake \ + gcc \ + libc6-dev \ + g++-s390x-linux-gnu \ + libc6-dev-s390x-cross \ + qemu-user \ + make \ + file + +ENV CARGO_TARGET_S390X_UNKNOWN_LINUX_GNU_LINKER=s390x-linux-gnu-gcc \ + CARGO_TARGET_S390X_UNKNOWN_LINUX_GNU_RUNNER="qemu-s390x -L /usr/s390x-linux-gnu" \ + CC_s390x_unknown_linux_gnu=s390x-linux-gnu-gcc \ + CXX_s390x_unknown_linux_gnu=s390x-linux-gnu-g++ \ + OBJDUMP=s390x-linux-gnu-objdump diff --git a/third_party/rust/packed_simd/ci/docker/sparc64-unknown-linux-gnu/Dockerfile b/third_party/rust/packed_simd/ci/docker/sparc64-unknown-linux-gnu/Dockerfile new file mode 100644 index 000000000000..c35f4d8f31a7 --- /dev/null +++ b/third_party/rust/packed_simd/ci/docker/sparc64-unknown-linux-gnu/Dockerfile @@ -0,0 +1,18 @@ +FROM debian:bookworm + +RUN apt-get update && apt-get install -y --no-install-recommends \ + curl ca-certificates \ + gcc libc6-dev \ + gcc-sparc64-linux-gnu libc6-dev-sparc64-cross \ + qemu-system-sparc64 openbios-sparc seabios ipxe-qemu \ + p7zip-full cpio + +COPY linux-sparc64.sh / +RUN bash /linux-sparc64.sh + +COPY test-runner-linux / + +ENV CARGO_TARGET_SPARC64_UNKNOWN_LINUX_GNU_LINKER=sparc64-linux-gnu-gcc \ + CARGO_TARGET_SPARC64_UNKNOWN_LINUX_GNU_RUNNER="/test-runner-linux sparc64" \ + CC_sparc64_unknown_linux_gnu=sparc64-linux-gnu-gcc \ + PATH=$PATH:/rust/bin diff --git a/third_party/rust/packed_simd/ci/docker/thumbv7neon-linux-androideabi/Dockerfile b/third_party/rust/packed_simd/ci/docker/thumbv7neon-linux-androideabi/Dockerfile new file mode 100644 index 000000000000..c1da77109c12 --- /dev/null +++ b/third_party/rust/packed_simd/ci/docker/thumbv7neon-linux-androideabi/Dockerfile @@ -0,0 +1,47 @@ +FROM ubuntu:16.04 + +RUN dpkg --add-architecture i386 && \ + apt-get update && \ + apt-get install -y --no-install-recommends \ + file \ + make \ + curl \ + ca-certificates \ + python \ + unzip \ + expect \ + openjdk-9-jre \ + libstdc++6:i386 \ + libpulse0 \ + gcc \ + libc6-dev + +WORKDIR /android/ +COPY android* /android/ + +ENV ANDROID_ARCH=arm +ENV PATH=$PATH:/android/ndk-$ANDROID_ARCH/bin:/android/sdk/tools:/android/sdk/platform-tools + +RUN sh /android/android-install-ndk.sh $ANDROID_ARCH +RUN sh /android/android-install-sdk.sh $ANDROID_ARCH +RUN mv /root/.android /tmp +RUN chmod 777 -R /tmp/.android +RUN chmod 755 /android/sdk/tools/* /android/sdk/emulator/qemu/linux-x86_64/* + +ENV PATH=$PATH:/rust/bin \ + CARGO_TARGET_THUMBV7NEON_LINUX_ANDROIDEABI_LINKER=arm-linux-androideabi-gcc \ + CARGO_TARGET_THUMBV7NEON_LINUX_ANDROIDEABI_RUNNER=/tmp/runtest \ + OBJDUMP=arm-linux-androideabi-objdump \ + HOME=/tmp + +ADD runtest-android.rs /tmp/runtest.rs +ENTRYPOINT [ \ + "bash", \ + "-c", \ + # set SHELL so android can detect a 64bits system, see + # http://stackoverflow.com/a/41789144 + "SHELL=/bin/dash /android/sdk/emulator/emulator @arm -no-window & \ + rustc /tmp/runtest.rs -o /tmp/runtest && \ + exec \"$@\"", \ + "--" \ +] diff --git a/third_party/rust/packed_simd/ci/docker/thumbv7neon-unknown-linux-gnueabihf/Dockerfile b/third_party/rust/packed_simd/ci/docker/thumbv7neon-unknown-linux-gnueabihf/Dockerfile new file mode 100644 index 000000000000..588d23c65ae5 --- /dev/null +++ b/third_party/rust/packed_simd/ci/docker/thumbv7neon-unknown-linux-gnueabihf/Dockerfile @@ -0,0 +1,13 @@ +FROM ubuntu:18.04 +RUN apt-get update && apt-get install -y --no-install-recommends \ + gcc \ + ca-certificates \ + libc6-dev \ + gcc-arm-linux-gnueabihf \ + libc6-dev-armhf-cross \ + qemu-user \ + make \ + file +ENV CARGO_TARGET_THUMBV7NEON_UNKNOWN_LINUX_GNUEABIHF_LINKER=arm-linux-gnueabihf-gcc \ + CARGO_TARGET_THUMBV7NEON_UNKNOWN_LINUX_GNUEABIHF_RUNNER="qemu-arm -L /usr/arm-linux-gnueabihf" \ + OBJDUMP=arm-linux-gnueabihf-objdump diff --git a/third_party/rust/packed_simd/ci/docker/wasm32-unknown-unknown/Dockerfile b/third_party/rust/packed_simd/ci/docker/wasm32-unknown-unknown/Dockerfile new file mode 100644 index 000000000000..51ee13e6c975 --- /dev/null +++ b/third_party/rust/packed_simd/ci/docker/wasm32-unknown-unknown/Dockerfile @@ -0,0 +1,39 @@ +FROM ubuntu:22.04 + +RUN apt-get update -y && apt-get install -y --no-install-recommends \ + ca-certificates \ + clang \ + cmake \ + curl \ + git \ + libc6-dev \ + make \ + ninja-build \ + python-is-python3 \ + xz-utils + +# Install `wasm2wat` +RUN git clone --recursive https://github.com/WebAssembly/wabt +RUN make -C wabt -j$(nproc) +ENV PATH=$PATH:/wabt/bin + +# Install `wasm-bindgen-test-runner` +RUN curl -L https://github.com/rustwasm/wasm-bindgen/releases/download/0.2.87/wasm-bindgen-0.2.87-x86_64-unknown-linux-musl.tar.gz \ + | tar xzf - +# Keep in sync with the version on Cargo.toml. +ENV PATH=$PATH:/wasm-bindgen-0.2.87-x86_64-unknown-linux-musl +ENV CARGO_TARGET_WASM32_UNKNOWN_UNKNOWN_RUNNER=wasm-bindgen-test-runner + +# Install `node` +RUN curl https://nodejs.org/dist/v14.16.0/node-v14.16.0-linux-x64.tar.xz | tar xJf - +ENV PATH=$PATH:/node-v14.16.0-linux-x64/bin + +# We use a shim linker that removes `--strip-debug` when passed to LLD. While +# this typically results in invalid debug information in release mode it doesn't +# result in an invalid names section which is what we're interested in. +COPY lld-shim.rs / +ENV CARGO_TARGET_WASM32_UNKNOWN_UNKNOWN_LINKER=/tmp/lld-shim + +# Rustc isn't available until this container starts, so defer compilation of the +# shim. +ENTRYPOINT /rust/bin/rustc /lld-shim.rs -o /tmp/lld-shim && exec bash "$@" diff --git a/third_party/rust/packed_simd/ci/docker/x86_64-linux-android/Dockerfile b/third_party/rust/packed_simd/ci/docker/x86_64-linux-android/Dockerfile new file mode 100644 index 000000000000..785936d347d2 --- /dev/null +++ b/third_party/rust/packed_simd/ci/docker/x86_64-linux-android/Dockerfile @@ -0,0 +1,31 @@ +FROM ubuntu:20.04 + +RUN apt-get update && \ + apt-get install -y --no-install-recommends \ + ca-certificates \ + curl \ + gcc \ + libc-dev \ + python \ + unzip \ + file \ + make + +WORKDIR /android/ +ENV ANDROID_ARCH=x86_64 +COPY android-install-ndk.sh /android/ +RUN sh /android/android-install-ndk.sh + +ENV STDARCH_ASSERT_INSTR_LIMIT=30 + +# We do not run x86_64-linux-android tests on an android emulator. +# See ci/android-sysimage.sh for informations about how tests are run. +COPY android-sysimage.sh /android/ +RUN bash /android/android-sysimage.sh x86_64 x86_64-24_r07.zip + +ENV PATH=$PATH:/rust/bin:/android/ndk/toolchains/llvm/prebuilt/linux-x86_64/bin \ + CARGO_TARGET_X86_64_LINUX_ANDROID_LINKER=x86_64-linux-android21-clang \ + CC_x86_64_linux_android=x86_64-linux-android21-clang \ + CXX_x86_64_linux_android=x86_64-linux-android21-clang++ \ + OBJDUMP=llvm-objdump \ + HOME=/tmp diff --git a/third_party/rust/packed_simd/ci/docker/x86_64-unknown-linux-gnu-emulated/Dockerfile b/third_party/rust/packed_simd/ci/docker/x86_64-unknown-linux-gnu-emulated/Dockerfile new file mode 100644 index 000000000000..a6bbe6653928 --- /dev/null +++ b/third_party/rust/packed_simd/ci/docker/x86_64-unknown-linux-gnu-emulated/Dockerfile @@ -0,0 +1,16 @@ +FROM ubuntu:18.04 +RUN apt-get update && apt-get install -y --no-install-recommends \ + gcc \ + libc6-dev \ + file \ + make \ + ca-certificates \ + wget \ + bzip2 \ + cmake \ + libclang-dev \ + clang + +RUN wget https://github.com/gnzlbg/intel_sde/raw/master/sde-external-8.16.0-2018-01-30-lin.tar.bz2 +RUN tar -xjf sde-external-8.16.0-2018-01-30-lin.tar.bz2 +ENV CARGO_TARGET_X86_64_UNKNOWN_LINUX_GNU_RUNNER="/sde-external-8.16.0-2018-01-30-lin/sde64 --" diff --git a/third_party/rust/packed_simd/ci/docker/x86_64-unknown-linux-gnu/Dockerfile b/third_party/rust/packed_simd/ci/docker/x86_64-unknown-linux-gnu/Dockerfile new file mode 100644 index 000000000000..ce5bb88e625d --- /dev/null +++ b/third_party/rust/packed_simd/ci/docker/x86_64-unknown-linux-gnu/Dockerfile @@ -0,0 +1,10 @@ +FROM ubuntu:18.04 +RUN apt-get update && apt-get install -y --no-install-recommends \ + gcc \ + libc6-dev \ + file \ + make \ + ca-certificates \ + cmake \ + libclang-dev \ + clang diff --git a/third_party/rust/packed_simd/ci/dox.sh b/third_party/rust/packed_simd/ci/dox.sh new file mode 100755 index 000000000000..560eaadcc880 --- /dev/null +++ b/third_party/rust/packed_simd/ci/dox.sh @@ -0,0 +1,27 @@ +#!/bin/sh + +set -ex + +rm -rf target/doc +mkdir -p target/doc + +# Build API documentation +cargo doc --features=into_bits + +# Build Performance Guide +# FIXME: https://github.com/rust-lang-nursery/mdBook/issues/780 +# mdbook build perf-guide -d target/doc/perf-guide +cd perf-guide +mdbook build +cd - +cp -r perf-guide/book target/doc/perf-guide + +# If we're on travis, not a PR, and on the right branch, publish! +if [ "$TRAVIS_PULL_REQUEST" = "false" ] && [ "$TRAVIS_BRANCH" = "master" ]; then + python3 -vV + pip -vV + python3.9 -vV + pip install ghp_import --user + ghp-import -n target/doc + git push -qf https://${GH_PAGES}@github.com/${TRAVIS_REPO_SLUG}.git gh-pages +fi diff --git a/third_party/rust/packed_simd/ci/linux-s390x.sh b/third_party/rust/packed_simd/ci/linux-s390x.sh new file mode 100644 index 000000000000..972abeec569e --- /dev/null +++ b/third_party/rust/packed_simd/ci/linux-s390x.sh @@ -0,0 +1,18 @@ +set -ex + +mkdir -m 777 /qemu +cd /qemu + +curl -LO https://github.com/qemu/qemu/raw/master/pc-bios/s390-ccw.img +curl -LO http://ftp.debian.org/debian/dists/testing/main/installer-s390x/20170828/images/generic/kernel.debian +curl -LO http://ftp.debian.org/debian/dists/testing/main/installer-s390x/20170828/images/generic/initrd.debian + +mv kernel.debian kernel +mv initrd.debian initrd.gz + +mkdir init +cd init +gunzip -c ../initrd.gz | cpio -id +rm ../initrd.gz +cp /usr/s390x-linux-gnu/lib/libgcc_s.so.1 usr/lib/ +chmod a+w . diff --git a/third_party/rust/packed_simd/ci/linux-sparc64.sh b/third_party/rust/packed_simd/ci/linux-sparc64.sh new file mode 100644 index 000000000000..4452b120e1b6 --- /dev/null +++ b/third_party/rust/packed_simd/ci/linux-sparc64.sh @@ -0,0 +1,17 @@ +set -ex + +mkdir -m 777 /qemu +cd /qemu + +curl -LO https://cdimage.debian.org/cdimage/ports/9.0/sparc64/iso-cd/debian-9.0-sparc64-NETINST-1.iso +7z e debian-9.0-sparc64-NETINST-1.iso boot/initrd.gz +7z e debian-9.0-sparc64-NETINST-1.iso boot/sparc64 +mv sparc64 kernel +rm debian-9.0-sparc64-NETINST-1.iso + +mkdir init +cd init +gunzip -c ../initrd.gz | cpio -id +rm ../initrd.gz +cp /usr/sparc64-linux-gnu/lib/libgcc_s.so.1 usr/lib/ +chmod a+w . diff --git a/third_party/rust/packed_simd/ci/lld-shim.rs b/third_party/rust/packed_simd/ci/lld-shim.rs new file mode 100644 index 000000000000..10263869e8dc --- /dev/null +++ b/third_party/rust/packed_simd/ci/lld-shim.rs @@ -0,0 +1,11 @@ +use std::os::unix::prelude::*; +use std::process::Command; +use std::env; + +fn main() { + let args = env::args() + .skip(1) + .filter(|s| s != "--strip-debug") + .collect::>(); + panic!("failed to exec: {}", Command::new("rust-lld").args(&args).exec()); +} diff --git a/third_party/rust/packed_simd/ci/max_line_width.sh b/third_party/rust/packed_simd/ci/max_line_width.sh new file mode 100755 index 000000000000..f70639b6f89b --- /dev/null +++ b/third_party/rust/packed_simd/ci/max_line_width.sh @@ -0,0 +1,17 @@ +#!/usr/bin/env sh + +set -x + +export success=true + +find . -iname '*.rs' | while read -r file; do + result=$(grep '.\{79\}' "${file}" | grep --invert 'http') + if [ "${result}" = "" ] + then + : + else + echo "file \"${file}\": $result" + exit 1 + fi +done + diff --git a/third_party/rust/packed_simd/ci/run-docker.sh b/third_party/rust/packed_simd/ci/run-docker.sh new file mode 100755 index 000000000000..abdd6852fc3a --- /dev/null +++ b/third_party/rust/packed_simd/ci/run-docker.sh @@ -0,0 +1,38 @@ +# Small script to run tests for a target (or all targets) inside all the +# respective docker images. + +set -ex + +run() { + echo "Building docker container for TARGET=${TARGET} RUSTFLAGS=${RUSTFLAGS}" + docker build -t packed_simd -f ci/docker/${TARGET}/Dockerfile ci/ + mkdir -p target + target=$(echo "${TARGET}" | sed 's/-emulated//') + echo "Running docker" + docker run \ + --user `id -u`:`id -g` \ + --rm \ + --init \ + --volume $HOME/.cargo:/cargo \ + --env CARGO_HOME=/cargo \ + --volume `rustc --print sysroot`:/rust:ro \ + --env TARGET=$target \ + --env NORUN \ + --env NOVERIFY \ + --env RUSTFLAGS \ + --volume `pwd`:/checkout:ro \ + --volume `pwd`/target:/checkout/target \ + --workdir /checkout \ + --privileged \ + packed_simd \ + bash \ + -c 'PATH=$PATH:/rust/bin exec ci/run.sh' +} + +if [ -z "${TARGET}" ]; then + for d in `ls ci/docker/`; do + run $d + done +else + run ${TARGET} +fi diff --git a/third_party/rust/packed_simd/ci/run.sh b/third_party/rust/packed_simd/ci/run.sh new file mode 100755 index 000000000000..b1b22ebd7c80 --- /dev/null +++ b/third_party/rust/packed_simd/ci/run.sh @@ -0,0 +1,99 @@ +#!/usr/bin/env bash + +set -ex + +: ${TARGET?"The TARGET environment variable must be set."} + +# Tests are all super fast anyway, and they fault often enough on travis that +# having only one thread increases debuggability to be worth it. +#export RUST_TEST_THREADS=1 +#export RUST_BACKTRACE=full +#export RUST_TEST_NOCAPTURE=1 + +# Some appveyor builds run out-of-memory; this attempts to mitigate that: +# https://github.com/rust-lang-nursery/packed_simd/issues/39 +# export RUSTFLAGS="${RUSTFLAGS} -C codegen-units=1" +# export CARGO_BUILD_JOBS=1 + +export CARGO_SUBCMD=test +if [[ "${NORUN}" == "1" ]]; then + export CARGO_SUBCMD=build +fi + +if [[ ${TARGET} == "x86_64-apple-ios" ]] || [[ ${TARGET} == "i386-apple-ios" ]]; then + export RUSTFLAGS="${RUSTFLAGS} -Clink-arg=-mios-simulator-version-min=7.0" + rustc ./ci/deploy_and_run_on_ios_simulator.rs -o $HOME/runtest + export CARGO_TARGET_X86_64_APPLE_IOS_RUNNER=$HOME/runtest + export CARGO_TARGET_I386_APPLE_IOS_RUNNER=$HOME/runtest +fi + +# The source directory is read-only. Need to copy internal crates to the target +# directory for their Cargo.lock to be properly written. +mkdir target || true + +rustc --version +cargo --version +echo "TARGET=${TARGET}" +echo "HOST=${HOST}" +echo "RUSTFLAGS=${RUSTFLAGS}" +echo "NORUN=${NORUN}" +echo "NOVERIFY=${NOVERIFY}" +echo "CARGO_SUBCMD=${CARGO_SUBCMD}" +echo "CARGO_BUILD_JOBS=${CARGO_BUILD_JOBS}" +echo "CARGO_INCREMENTAL=${CARGO_INCREMENTAL}" +echo "RUST_TEST_THREADS=${RUST_TEST_THREADS}" +echo "RUST_BACKTRACE=${RUST_BACKTRACE}" +echo "RUST_TEST_NOCAPTURE=${RUST_TEST_NOCAPTURE}" + +cargo_test() { + cmd="cargo ${CARGO_SUBCMD} --verbose --target=${TARGET} ${@}" + if [ "${NORUN}" != "1" ] + then + if [ "$TARGET" != "wasm32-unknown-unknown" ] + then + cmd="$cmd -- --quiet" + fi + fi + mkdir target || true + ${cmd} 2>&1 | tee > target/output + if [[ ${PIPESTATUS[0]} != 0 ]]; then + cat target/output + return 1 + fi +} + +cargo_test_impl() { + ORIGINAL_RUSTFLAGS=${RUSTFLAGS} + RUSTFLAGS="${ORIGINAL_RUSTFLAGS} --cfg test_v16 --cfg test_v32 --cfg test_v64" cargo_test ${@} + RUSTFLAGS="${ORIGINAL_RUSTFLAGS} --cfg test_v128 --cfg test_v256" cargo_test ${@} + RUSTFLAGS="${ORIGINAL_RUSTFLAGS} --cfg test_v512" cargo_test ${@} + RUSTFLAGS=${ORIGINAL_RUSTFLAGS} +} + +# Debug run: +if [[ "${TARGET}" != "wasm32-unknown-unknown" ]]; then + # Run wasm32-unknown-unknown in release mode only + cargo_test_impl +fi + +if [[ "${TARGET}" == "x86_64-unknown-linux-gnu" ]] || [[ "${TARGET}" == "x86_64-pc-windows-msvc" ]]; then + # use sleef on linux and windows x86_64 builds + # FIXME: Use `core_arch,sleef-sys` features once they works again + cargo_test_impl --release --features=into_bits +else + # FIXME: Use `core_arch` feature once it works again + cargo_test_impl --release --features=into_bits +fi + +# Verify code generation +if [[ "${NOVERIFY}" != "1" ]]; then + cp -r verify/verify target/verify + export STDSIMD_ASSERT_INSTR_LIMIT=30 + if [[ "${TARGET}" == "i586-unknown-linux-gnu" ]]; then + export STDSIMD_ASSERT_INSTR_LIMIT=50 + fi + cargo_test --release --manifest-path=target/verify/Cargo.toml +fi + +# FIXME: Figure out which examples take too long to run and ignore or adjust those +#. ci/run_examples.sh diff --git a/third_party/rust/packed_simd/ci/run_examples.sh b/third_party/rust/packed_simd/ci/run_examples.sh new file mode 100644 index 000000000000..5b26b18afb20 --- /dev/null +++ b/third_party/rust/packed_simd/ci/run_examples.sh @@ -0,0 +1,51 @@ +# Runs all examples. + +# FIXME: https://github.com/rust-lang-nursery/packed_simd/issues/55 +# All examples fail to build for `armv7-apple-ios`. +if [[ ${TARGET} == "armv7-apple-ios" ]]; then + exit 0 +fi + +# FIXME: travis exceeds 50 minutes on these targets +# Skipping the examples is an attempt at preventing travis from timing-out +if [[ ${TARGET} == "arm-linux-androidabi" ]] || [[ ${TARGET} == "aarch64-linux-androidabi" ]] \ + || [[ ${TARGET} == "sparc64-unknown-linux-gnu" ]]; then + exit 0 +fi + +if [[ ${TARGET} == "wasm32-unknown-unknown" ]]; then + exit 0 +fi + +cp -r examples/aobench target/aobench +cargo_test --manifest-path=target/aobench/Cargo.toml --release --no-default-features +cargo_test --manifest-path=target/aobench/Cargo.toml --release --features=256bit + +cp -r examples/dot_product target/dot_product +cargo_test --manifest-path=target/dot_product/Cargo.toml --release + +cp -r examples/fannkuch_redux target/fannkuch_redux +cargo_test --manifest-path=target/fannkuch_redux/Cargo.toml --release + +# FIXME: https://github.com/rust-lang-nursery/packed_simd/issues/56 +if [[ ${TARGET} != "i586-unknown-linux-gnu" ]]; then + cp -r examples/mandelbrot target/mandelbrot + cargo_test --manifest-path=target/mandelbrot/Cargo.toml --release +fi + +cp -r examples/matrix_inverse target/matrix_inverse +cargo_test --manifest-path=target/matrix_inverse/Cargo.toml --release + +cp -r examples/nbody target/nbody +cargo_test --manifest-path=target/nbody/Cargo.toml --release + +cp -r examples/spectral_norm target/spectral_norm +cargo_test --manifest-path=target/spectral_norm/Cargo.toml --release + +if [[ ${TARGET} != "i586-unknown-linux-gnu" ]]; then + cp -r examples/stencil target/stencil + cargo_test --manifest-path=target/stencil/Cargo.toml --release +fi + +cp -r examples/triangle_xform target/triangle_xform +cargo_test --manifest-path=target/triangle_xform/Cargo.toml --release diff --git a/third_party/rust/packed_simd/ci/runtest-android.rs b/third_party/rust/packed_simd/ci/runtest-android.rs new file mode 100644 index 000000000000..ed1cd80c834a --- /dev/null +++ b/third_party/rust/packed_simd/ci/runtest-android.rs @@ -0,0 +1,45 @@ +use std::env; +use std::process::Command; +use std::path::{Path, PathBuf}; + +fn main() { + let args = env::args_os() + .skip(1) + .filter(|arg| arg != "--quiet") + .collect::>(); + assert_eq!(args.len(), 1); + let test = PathBuf::from(&args[0]); + let dst = Path::new("/data/local/tmp").join(test.file_name().unwrap()); + + let status = Command::new("adb") + .arg("wait-for-device") + .status() + .expect("failed to run: adb wait-for-device"); + assert!(status.success()); + + let status = Command::new("adb") + .arg("push") + .arg(&test) + .arg(&dst) + .status() + .expect("failed to run: adb pushr"); + assert!(status.success()); + + let output = Command::new("adb") + .arg("shell") + .arg(&dst) + .output() + .expect("failed to run: adb shell"); + assert!(status.success()); + + println!("status: {}\nstdout ---\n{}\nstderr ---\n{}", + output.status, + String::from_utf8_lossy(&output.stdout), + String::from_utf8_lossy(&output.stderr)); + + let stdout = String::from_utf8_lossy(&output.stdout); + let mut lines = stdout.lines().filter(|l| l.starts_with("test result")); + if !lines.all(|l| l.contains("test result: ok") && l.contains("0 failed")) { + panic!("failed to find successful test run"); + } +} diff --git a/third_party/rust/packed_simd/ci/setup_benchmarks.sh b/third_party/rust/packed_simd/ci/setup_benchmarks.sh new file mode 100755 index 000000000000..cd41a7851303 --- /dev/null +++ b/third_party/rust/packed_simd/ci/setup_benchmarks.sh @@ -0,0 +1,7 @@ +#!/usr/bin/env bash + +set -ex + +# Get latest ISPC binary for the target and put it in the path +git clone https://github.com/gnzlbg/ispc-binaries +cp ispc-binaries/ispc-${TARGET} ispc diff --git a/third_party/rust/packed_simd/ci/test-runner-linux b/third_party/rust/packed_simd/ci/test-runner-linux new file mode 100755 index 000000000000..0654f63bfdb9 --- /dev/null +++ b/third_party/rust/packed_simd/ci/test-runner-linux @@ -0,0 +1,24 @@ +#!/bin/sh + +set -e + +arch=$1 +prog=$2 + +cd /qemu/init +cp -f $2 prog +find . | cpio --create --format='newc' --quiet | gzip > ../initrd.gz +cd .. + +timeout 30s qemu-system-$arch \ + -m 1024 \ + -nographic \ + -kernel kernel \ + -initrd initrd.gz \ + -append init=/prog > output || true + +# remove kernel messages +tr -d '\r' < output | egrep -v '^\[' + +# if the output contains a failure, return error +! grep FAILED output > /dev/null diff --git a/third_party/rust/packed_simd/contributing.md b/third_party/rust/packed_simd/contributing.md new file mode 100644 index 000000000000..79af8c199e8f --- /dev/null +++ b/third_party/rust/packed_simd/contributing.md @@ -0,0 +1,67 @@ +# Contributing to `packed_simd` + +Welcome! If you are reading this document, it means you are interested in contributing +to the `packed_simd` crate. + +## Reporting issues + +All issues with this crate are tracked using GitHub's [Issue Tracker]. + +You can use issues to bring bugs to the attention of the maintainers, to discuss +certain problems encountered with the crate, or to request new features (although +feature requests should be limited to things mentioned in the [RFC]). + +One thing to keep in mind is to always use the **latest** nightly toolchain when +working on this crate. Due to the nature of this project, we use a lot of unstable +features, meaning breakage happens often. + +[Issue Tracker]: https://github.com/rust-lang-nursery/packed_simd/issues +[RFC]: https://github.com/rust-lang/rfcs/pull/2366 + +### LLVM issues + +The Rust compiler relies on [LLVM](https://llvm.org/) for machine code generation, +and quite a few LLVM bugs have been discovered during the development of this project. + +If you encounter issues with incorrect/suboptimal codegen, which you do not encounter +when using the [SIMD vendor intrinsics](https://doc.rust-lang.org/nightly/std/arch/), +it is likely the issue is with LLVM, or this crate's interaction with it. + +You should first open an issue **in this repo** to help us track the problem, and we +will help determine what is the exact cause of the problem. +If LLVM is indeed the cause, the issue will be reported upstream to the +[LLVM bugtracker](https://bugs.llvm.org/). + +## Submitting Pull Requests + +New code is submitted to the crate using GitHub's [pull request] mechanism. +You should first fork this repository, make your changes (preferably in a new +branch), then use GitHub's web UI to create a new PR. + +[pull request]: https://help.github.com/articles/about-pull-requests/ + +### Examples + +The `examples` directory contains code showcasing SIMD code written with this crate, +usually in comparison to scalar or ISPC code. If you have a project / idea which +uses SIMD, we'd love to add it to the examples list. + +Every example should include a small `README`, describing the example code's purpose. +If your example could potentially work as a benchmark, then add a `benchmark.sh` +script to allow running the example benchmark code in CI. See an existing example's +[`benchmark.sh`](examples/aobench/benchmark.sh) for a sample. + +Don't forget to update the crate's top-level `README` with a link to your example. + +### Perf guide + +The objective of the [performance guide][perf-guide] is to be a comprehensive +resource detailing the process of optimizing Rust code with SIMD support. + +If you believe a certain section could be reworded, or if you have any tips & tricks +related to SIMD which you'd like to share, please open a PR. + +[mdBook] is used to manage the formatting of the guide as a book. + +[perf-guide]: https://rust-lang-nursery.github.io/packed_simd/perf-guide/ +[mdBook]: https://github.com/rust-lang-nursery/mdBook diff --git a/third_party/rust/packed_simd/perf-guide/book.toml b/third_party/rust/packed_simd/perf-guide/book.toml new file mode 100644 index 000000000000..69ba3053ca25 --- /dev/null +++ b/third_party/rust/packed_simd/perf-guide/book.toml @@ -0,0 +1,12 @@ +[book] +authors = ["Gonzalo Brito Gadeschi", "Gabriel Majeri"] +multilingual = false +src = "src" +title = "Rust SIMD Performance Guide" +description = "This book describes how to write performant SIMD code in Rust." + +[build] +create-missing = false + +[output.html] +additional-css = ["./src/ascii.css"] diff --git a/third_party/rust/packed_simd/perf-guide/src/SUMMARY.md b/third_party/rust/packed_simd/perf-guide/src/SUMMARY.md new file mode 100644 index 000000000000..1e76898865c5 --- /dev/null +++ b/third_party/rust/packed_simd/perf-guide/src/SUMMARY.md @@ -0,0 +1,21 @@ +# Summary + +[Introduction](./introduction.md) + +- [Floating-point Math](./float-math/fp.md) + - [Short-vector Math Library](./float-math/svml.md) + - [Approximate functions](./float-math/approx.md) + - [Fused multiply-accumulate](./float-math/fma.md) + +- [Target features](./target-feature/features.md) + - [Using `RUSTFLAGS`](./target-feature/rustflags.md) + - [Using the `target_feature` attribute](./target-feature/attribute.md) + - [Interaction with inlining](./target-feature/inlining.md) + - [Detecting features at runtime](./target-feature/runtime.md) + +- [Bounds checking](./bound_checks.md) +- [Vertical and horizontal operations](./vert-hor-ops.md) + +- [Performance profiling](./prof/profiling.md) + - [Profiling on Linux](./prof/linux.md) + - [Using machine code analyzers](./prof/mca.md) diff --git a/third_party/rust/packed_simd/perf-guide/src/ascii.css b/third_party/rust/packed_simd/perf-guide/src/ascii.css new file mode 100644 index 000000000000..4c02651195f9 --- /dev/null +++ b/third_party/rust/packed_simd/perf-guide/src/ascii.css @@ -0,0 +1,4 @@ +code { + /* "Source Code Pro" breaks ASCII art */ + font-family: Consolas, "Ubuntu Mono", Menlo, "DejaVu Sans Mono", monospace; +} diff --git a/third_party/rust/packed_simd/perf-guide/src/bound_checks.md b/third_party/rust/packed_simd/perf-guide/src/bound_checks.md new file mode 100644 index 000000000000..2eeedb5ac829 --- /dev/null +++ b/third_party/rust/packed_simd/perf-guide/src/bound_checks.md @@ -0,0 +1,22 @@ +# Bounds checking + +Reading and writing packed vectors to/from slices is checked by default. +Independently of the configuration options used, the safe functions: + +* `Simd<[T; N]>::from_slice_aligned(& s[..])` +* `Simd<[T; N]>::write_to_slice_aligned(&mut s[..])` + +always check that: + +* the slice is big enough to hold the vector +* the slice is suitably aligned to perform an aligned load/store for a `Simd<[T; + N]>` (this alignment is often much larger than that of `T`). + +There are `_unaligned` versions that use unaligned load and stores, as well as +`unsafe` `_unchecked` that do not perform any checks iff `debug-assertions = +false` / `debug = false`. That is, the `_unchecked` methods do still assert size +and alignment in debug builds and could also do so in release builds depending +on the configuration options. + +These assertions do often significantly impact performance and you should be +aware of them. diff --git a/third_party/rust/packed_simd/perf-guide/src/float-math/approx.md b/third_party/rust/packed_simd/perf-guide/src/float-math/approx.md new file mode 100644 index 000000000000..2237c67ec4b3 --- /dev/null +++ b/third_party/rust/packed_simd/perf-guide/src/float-math/approx.md @@ -0,0 +1,8 @@ +# Approximate functions + + diff --git a/third_party/rust/packed_simd/perf-guide/src/float-math/fma.md b/third_party/rust/packed_simd/perf-guide/src/float-math/fma.md new file mode 100644 index 000000000000..357748383d63 --- /dev/null +++ b/third_party/rust/packed_simd/perf-guide/src/float-math/fma.md @@ -0,0 +1,6 @@ +# Fused Multiply Add + + diff --git a/third_party/rust/packed_simd/perf-guide/src/float-math/fp.md b/third_party/rust/packed_simd/perf-guide/src/float-math/fp.md new file mode 100644 index 000000000000..711fcc4fd598 --- /dev/null +++ b/third_party/rust/packed_simd/perf-guide/src/float-math/fp.md @@ -0,0 +1,3 @@ +# Floating-point math + +This chapter contains information pertaining to working with floating-point numbers. diff --git a/third_party/rust/packed_simd/perf-guide/src/float-math/svml.md b/third_party/rust/packed_simd/perf-guide/src/float-math/svml.md new file mode 100644 index 000000000000..266c2531cc04 --- /dev/null +++ b/third_party/rust/packed_simd/perf-guide/src/float-math/svml.md @@ -0,0 +1,7 @@ +# Short Vector Math Library + + diff --git a/third_party/rust/packed_simd/perf-guide/src/introduction.md b/third_party/rust/packed_simd/perf-guide/src/introduction.md new file mode 100644 index 000000000000..7243e19c8a54 --- /dev/null +++ b/third_party/rust/packed_simd/perf-guide/src/introduction.md @@ -0,0 +1,26 @@ +# Introduction + +## What is SIMD + + + +## History of SIMD in Rust + + + +## Discover packed_simd + + + +Writing fast and portable SIMD algorithms using `packed_simd` is, unfortunately, +not trivial. There are many pitfals that one should be aware of, and some idioms +that help avoid those pitfalls. + +This book attempts to document these best practices and provides practical examples +on how to apply the tips to _your_ code. diff --git a/third_party/rust/packed_simd/perf-guide/src/prof/linux.md b/third_party/rust/packed_simd/perf-guide/src/prof/linux.md new file mode 100644 index 000000000000..96c7d67bc476 --- /dev/null +++ b/third_party/rust/packed_simd/perf-guide/src/prof/linux.md @@ -0,0 +1,107 @@ +# Performance profiling on Linux + +## Using `perf` + +[perf](https://perf.wiki.kernel.org/) is the most powerful performance profiler +for Linux, featuring support for various hardware Performance Monitoring Units, +as well as integration with the kernel's performance events framework. + +We will only look at how can the `perf` command can be used to profile SIMD code. +Full system profiling is outside of the scope of this book. + +### Recording + +The first step is to record a program's execution during an average workload. +It helps if you can isolate the parts of your program which have performance +issues, and set up a benchmark which can be easily (re)run. + +Build the benchmark binary in release mode, after having enabled debug info: + +```sh +$ cargo build --release +Finished release [optimized + debuginfo] target(s) in 0.02s +``` + +Then use the `perf record` subcommand: + +```sh +$ perf record --call-graph=dwarf ./target/release/my-program +[ perf record: Woken up 10 times to write data ] +[ perf record: Captured and wrote 2,356 MB perf.data (292 samples) ] +``` + +Instead of using `--call-graph=dwarf`, which can become pretty slow, you can use +`--call-graph=lbr` if you have a processor with support for Last Branch Record +(i.e. Intel Haswell and newer). + +`perf` will, by default, record the count of CPU cycles it takes to execute +various parts of your program. You can use the `-e` command line option +to enable other performance events, such as `cache-misses`. Use `perf list` +to get a list of all hardware counters supported by your CPU. + +### Viewing the report + +The next step is getting a bird's eye view of the program's execution. +`perf` provides a `ncurses`-based interface which will get you started. + +Use `perf report` to open a visualization of your program's performance: + +```sh +perf report --hierarchy -M intel +``` + +`--hierarchy` will display a tree-like structure of where your program spent +most of its time. `-M intel` enables disassembly output with Intel syntax, which +is subjectively more readable than the default AT&T syntax. + +Here is the output from profiling the `nbody` benchmark: + +``` +- 100,00% nbody + - 94,18% nbody + + 93,48% [.] nbody_lib::simd::advance + + 0,70% [.] nbody_lib::run + + 5,06% libc-2.28.so +``` + +If you move with the arrow keys to any node in the tree, you can the press `a` +to have `perf` _annotate_ that node. This means it will: + +- disassemble the function + +- associate every instruction with the percentage of time which was spent executing it + +- interleaves the disassembly with the source code, + assuming it found the debug symbols + (you can use `s` to toggle this behaviour) + +`perf` will, by default, open the instruction which it identified as being the +hottest spot in the function: + +``` +0,76 │ movapd xmm2,xmm0 +0,38 │ movhlps xmm2,xmm0 + │ addpd xmm2,xmm0 + │ unpcklpd xmm1,xmm2 +12,50 │ sqrtpd xmm0,xmm1 +1,52 │ mulpd xmm0,xmm1 +``` + +In this case, `sqrtpd` will be highlighted in red, since that's the instruction +which the CPU spends most of its time executing. + +## Using Valgrind + +Valgrind is a set of tools which initially helped C/C++ programmers find unsafe +memory accesses in their code. Nowadays the project also has + +- a heap profiler called `massif` + +- a cache utilization profiler called `cachegrind` + +- a call-graph performance profiler called `callgrind` + + diff --git a/third_party/rust/packed_simd/perf-guide/src/prof/mca.md b/third_party/rust/packed_simd/perf-guide/src/prof/mca.md new file mode 100644 index 000000000000..65ddf1a4eb3a --- /dev/null +++ b/third_party/rust/packed_simd/perf-guide/src/prof/mca.md @@ -0,0 +1,100 @@ +# Machine code analysis tools + +## The microarchitecture of modern CPUs + +While you might have heard of Instruction Set Architectures, such as `x86` or +`arm` or `mips`, the term _microarchitecture_ (also written here as _µ-arch_), +refers to the internal details of an actual family of CPUs, such as Intel's +_Haswell_ or AMD's _Jaguar_. + +Replacing scalar code with SIMD code will improve performance on all CPUs +supporting the required vector extensions. +However, due to microarchitectural differences, the actual speed-up at +runtime might vary. + +**Example**: a simple example arises when optimizing for AMD K8 CPUs. +The assembly generated for an empty function should look like this: + +```asm +nop +ret +``` + +The `nop` is used to align the `ret` instruction for better performance. +However, the compiler will actually generated the following code: + +```asm +repz ret +``` + +The `repz` instruction will repeat the following instruction until a certain +condition. Of course, in this situation, the function will simply immediately +return, and the `ret` instruction is still aligned. +However, AMD K8's branch predictor performs better with the latter code. + +For those looking to absolutely maximize performance for a certain target µ-arch, +you will have to read some CPU manuals, or ask the compiler to do it for you +with `-C target-cpu`. + +### Summary of CPU internals + +Modern processors are able to execute instructions out-of-order for better performance, +by utilizing tricks such as [branch prediction], [instruction pipelining], +or [superscalar execution]. + +[branch prediction]: https://en.wikipedia.org/wiki/Branch_predictor +[instruction pipelining]: https://en.wikipedia.org/wiki/Instruction_pipelining +[superscalar execution]: https://en.wikipedia.org/wiki/Superscalar_processor + +SIMD instructions are also subject to these optimizations, meaning it can get pretty +difficult to determine where the slowdown happens. +For example, if the profiler reports a store operation is slow, one of two things +could be happening: + +- the store is limited by the CPU's memory bandwidth, which is actually an ideal + scenario, all things considered; + +- memory bandwidth is nowhere near its peak, but the value to be stored is at the + end of a long chain of operations, and this store is where the profiler + encountered the pipeline stall; + +Since most profilers are simple tools which don't understand the subtleties of +instruction scheduling, you + +## Analyzing the machine code + +Certain tools have knowledge of internal CPU microarchitecture, i.e. they know + +- how many physical [register files] a CPU actually has + +- what is the latency / throughtput of an instruction + +- what [µ-ops] are generated for a set of instructions + +and many other architectural details. + +[register files]: https://en.wikipedia.org/wiki/Register_file +[µ-ops]: https://en.wikipedia.org/wiki/Micro-operation + +These tools are therefore able to provide accurate information as to why some +instructions are inefficient, and where the bottleneck is. + +The disadvantage is that the output of these tools requires advanced knowledge +of the target architecture to understand, i.e. they **cannot** point out what +the cause of the issue is explicitly. + +## Intel's Architecture Code Analyzer (IACA) + +[IACA] is a free tool offered by Intel for analyzing the performance of various +computational kernels. + +Being a proprietary, closed source tool, it _only_ supports Intel's µ-arches. + +[IACA]: https://software.intel.com/en-us/articles/intel-architecture-code-analyzer + +## llvm-mca + + diff --git a/third_party/rust/packed_simd/perf-guide/src/prof/profiling.md b/third_party/rust/packed_simd/perf-guide/src/prof/profiling.md new file mode 100644 index 000000000000..02ba78d2f22f --- /dev/null +++ b/third_party/rust/packed_simd/perf-guide/src/prof/profiling.md @@ -0,0 +1,14 @@ +# Performance profiling + +While the rest of the book provides practical advice on how to improve the performance +of SIMD code, this chapter is dedicated to [**performance profiling**][profiling]. +Profiling consists of recording a program's execution in order to identify program +hotspots. + +**Important**: most profilers require debug information in order to accurately +link the program hotspots back to the corresponding source code lines. Rust will +disable debug info generation by default for optimized builds, but you can change +that [in your `Cargo.toml`][cargo-ref]. + +[profiling]: https://en.wikipedia.org/wiki/Profiling_(computer_programming) +[cargo-ref]: https://doc.rust-lang.org/cargo/reference/manifest.html#the-profile-sections diff --git a/third_party/rust/packed_simd/perf-guide/src/target-feature/attribute.md b/third_party/rust/packed_simd/perf-guide/src/target-feature/attribute.md new file mode 100644 index 000000000000..ee670fea5bd8 --- /dev/null +++ b/third_party/rust/packed_simd/perf-guide/src/target-feature/attribute.md @@ -0,0 +1,5 @@ +# The `target_feature` attribute + + diff --git a/third_party/rust/packed_simd/perf-guide/src/target-feature/features.md b/third_party/rust/packed_simd/perf-guide/src/target-feature/features.md new file mode 100644 index 000000000000..b93030ca6708 --- /dev/null +++ b/third_party/rust/packed_simd/perf-guide/src/target-feature/features.md @@ -0,0 +1,13 @@ +# Enabling target features + +Not all processors of a certain architecture will have SIMD processing units, +and using a SIMD instruction which is not supported will trigger undefined behavior. + +To allow building safe, portable programs, the Rust compiler will **not**, by default, +generate any sort of vector instructions, unless it can statically determine +they are supported. For example, on AMD64, SSE2 support is architecturally guaranteed. +The `x86_64-apple-darwin` target enables up to SSSE3. The get a defintive list of +which features are enabled by default on various platforms, refer to the target +specifications [in the compiler's source code][targets]. + +[targets]: https://github.com/rust-lang/rust/tree/master/src/librustc_target/spec diff --git a/third_party/rust/packed_simd/perf-guide/src/target-feature/inlining.md b/third_party/rust/packed_simd/perf-guide/src/target-feature/inlining.md new file mode 100644 index 000000000000..86705102a74b --- /dev/null +++ b/third_party/rust/packed_simd/perf-guide/src/target-feature/inlining.md @@ -0,0 +1,5 @@ +# Inlining + + diff --git a/third_party/rust/packed_simd/perf-guide/src/target-feature/practice.md b/third_party/rust/packed_simd/perf-guide/src/target-feature/practice.md new file mode 100644 index 000000000000..5b55c61c268a --- /dev/null +++ b/third_party/rust/packed_simd/perf-guide/src/target-feature/practice.md @@ -0,0 +1,31 @@ +# Target features in practice + +Using `RUSTFLAGS` will allow the crate being compiled, as well as all its +transitive dependencies to use certain target features. + +A tehnique used to avoid undefined behavior at runtime is to compile and +ship multiple binaries, each compiled with a certain set of features. +This might not be feasible in some cases, and can quickly get out of hand +as more and more vector extensions are added to an architecture. + +Rust can be more flexible: you can build a single binary/library which automatically +picks the best supported vector instructions depending on the host machine. +The trick consists of monomorphizing parts of the code during building, and then +using run-time feature detection to select the right code path when running. + + + +**NOTE** (x86 specific): because the AVX (256-bit) registers extend the existing +SSE (128-bit) registers, mixing SSE and AVX instructions in a program can cause +performance issues. + +The solution is to compile all code, even the code written with 128-bit vectors, +with the AVX target feature enabled. This will cause the compiler to prefix the +generated instructions with the [VEX] prefix. + +[VEX]: https://en.wikipedia.org/wiki/VEX_prefix diff --git a/third_party/rust/packed_simd/perf-guide/src/target-feature/runtime.md b/third_party/rust/packed_simd/perf-guide/src/target-feature/runtime.md new file mode 100644 index 000000000000..47ddcc8660db --- /dev/null +++ b/third_party/rust/packed_simd/perf-guide/src/target-feature/runtime.md @@ -0,0 +1,5 @@ +# Detecting host features at runtime + + diff --git a/third_party/rust/packed_simd/perf-guide/src/target-feature/rustflags.md b/third_party/rust/packed_simd/perf-guide/src/target-feature/rustflags.md new file mode 100644 index 000000000000..f4c1d1304aaa --- /dev/null +++ b/third_party/rust/packed_simd/perf-guide/src/target-feature/rustflags.md @@ -0,0 +1,77 @@ +# Using RUSTFLAGS + +One of the easiest ways to benefit from SIMD is to allow the compiler +to generate code using certain vector instruction extensions. + +The environment variable `RUSTFLAGS` can be used to pass options for code +generation to the Rust compiler. These flags will affect **all** compiled crates. + +There are two flags which can be used to enable specific vector extensions: + +## target-feature + +- Syntax: `-C target-feature=` + +- Provides the compiler with a comma-separated set of instruction extensions + to enable. + + **Example**: Use `-C target-feature=+sse3,+avx` to enable generating instructions + for [Streaming SIMD Extensions 3](https://en.wikipedia.org/wiki/SSE3) and + [Advanced Vector Extensions](https://en.wikipedia.org/wiki/Advanced_Vector_Extensions). + +- To list target triples for all targets supported by Rust, use: + + ```sh + rustc --print target-list + ``` + +- To list all support target features for a certain target triple, use: + + ```sh + rustc --target=${TRIPLE} --print target-features + ``` + +- Note that all CPU features are independent, and will have to be enabled individually. + + **Example**: Setting `-C target-feature=+avx2` will _not_ enable `fma`, even though + all CPUs which support AVX2 also support FMA. To enable both, one has to use + `-C target-feature=+avx2,+fma` + +- Some features also depend on other features, which need to be enabled for the + target instructions to be generated. + + **Example**: Unless `v7` is specified as the target CPU (see below), to enable + NEON on ARM it is necessary to use `-C target-feature=+v7,+neon`. + +## target-cpu + +- Syntax: `-C target-cpu=` + +- Sets the identifier of a CPU family / model for which to build and optimize the code. + + **Example**: `RUSTFLAGS='-C target-cpu=cortex-a75'` + +- To list all supported target CPUs for a certain target triple, use: + + ```sh + rustc --target=${TRIPLE} --print target-cpus + ``` + + **Example**: + + ```sh + rustc --target=i686-pc-windows-msvc --print target-cpus + ``` + +- The compiler will translate this into a list of target features. Therefore, + individual feature checks (`#[cfg(target_feature = "...")]`) will still + work properly. + +- It will cause the code generator to optimize the generated code for that + specific CPU model. + +- Using `native` as the CPU model will cause Rust to generate and optimize code + for the CPU running the compiler. It is useful when building programs which you + plan to only use locally. This should never be used when the generated programs + are meant to be run on other computers, such as when packaging for distribution + or cross-compiling. diff --git a/third_party/rust/packed_simd/perf-guide/src/vert-hor-ops.md b/third_party/rust/packed_simd/perf-guide/src/vert-hor-ops.md new file mode 100644 index 000000000000..d0dd1be12a19 --- /dev/null +++ b/third_party/rust/packed_simd/perf-guide/src/vert-hor-ops.md @@ -0,0 +1,76 @@ +# Vertical and horizontal operations + +In SIMD terminology, each vector has a certain "width" (number of lanes). +A vector processor is able to perform two kinds of operations on a vector: + +- Vertical operations: + operate on two vectors of the same width, result has same width + +**Example**: vertical addition of two `f32x4` vectors + + %0 == | 2 | -3.5 | 0 | 7 | + + + + + + %1 == | 4 | 1.5 | -1 | 0 | + = = = = + %0 + %1 == | 6 | -2 | -1 | 7 | + +- Horizontal operations: + reduce the elements of two vectors in some way, + the result's elements combine information from the two original ones + +**Example**: horizontal addition of two `u64x2` vectors + + %0 == | 1 | 3 | + └─+───┘ + └───────┐ + │ + %1 == | 4 | -1 | │ + └─+──┘ │ + └───┐ │ + │ │ + ┌─────│───┘ + ▼ ▼ + %0 + %1 == | 4 | 3 | + +## Performance consideration of horizontal operations + +The result of vertical operations, like vector negation: `-a`, for a given lane, +does not depend on the result of the operation for the other lanes. The result +of horizontal operations, like the vector `sum` reduction: `a.sum()`, depends on +the value of all vector lanes. + +In virtually all architectures vertical operations are fast, while horizontal +operations are, by comparison, very slow. + +Consider the following two functions for computing the sum of all `f32` values +in a slice: + +```rust +fn fast_sum(x: &[f32]) -> f32 { + assert!(x.len() % 4 == 0); + let mut sum = f32x4::splat(0.); // [0., 0., 0., 0.] + for i in (0..x.len()).step_by(4) { + sum += f32x4::from_slice_unaligned(&x[i..]); + } + sum.sum() +} + +fn slow_sum(x: &[f32]) -> f32 { + assert!(x.len() % 4 == 0); + let mut sum: f32 = 0.; + for i in (0..x.len()).step_by(4) { + sum += f32x4::from_slice_unaligned(&x[i..]).sum(); + } + sum +} +``` + +The inner loop over the slice is where the bulk of the work actually happens. +There, the `fast_sum` function perform vertical operations into a vector, doing +a single horizontal reduction at the end, while the `slow_sum` function performs +horizontal vector operations inside of the loop. + +On all widely-used architectures, `fast_sum` is a large constant factor faster +than `slow_sum`. You can run the [slice_sum]() example and see for yourself. On +the particular machine tested there the algorithm using the horizontal vector +addition is 2.7x slower than the one using vertical vector operations! diff --git a/third_party/rust/packed_simd/rust-toolchain b/third_party/rust/packed_simd/rust-toolchain new file mode 100644 index 000000000000..bf867e0ae5b6 --- /dev/null +++ b/third_party/rust/packed_simd/rust-toolchain @@ -0,0 +1 @@ +nightly diff --git a/third_party/rust/packed_simd/rustfmt.toml b/third_party/rust/packed_simd/rustfmt.toml new file mode 100644 index 000000000000..7316518b9980 --- /dev/null +++ b/third_party/rust/packed_simd/rustfmt.toml @@ -0,0 +1,5 @@ +max_width = 110 +use_small_heuristics = "Max" +wrap_comments = true +edition = "2018" +error_on_line_overflow = true \ No newline at end of file diff --git a/third_party/rust/packed_simd/src/api.rs b/third_party/rust/packed_simd/src/api.rs new file mode 100644 index 000000000000..262fc4ee693f --- /dev/null +++ b/third_party/rust/packed_simd/src/api.rs @@ -0,0 +1,309 @@ +//! Implements the Simd<[T; N]> APIs + +#[macro_use] +mod bitmask; +pub(crate) mod cast; +#[macro_use] +mod cmp; +#[macro_use] +mod default; +#[macro_use] +mod fmt; +#[macro_use] +mod from; +#[macro_use] +mod hash; +#[macro_use] +mod math; +#[macro_use] +mod minimal; +#[macro_use] +mod ops; +#[macro_use] +mod ptr; +#[macro_use] +mod reductions; +#[macro_use] +mod select; +#[macro_use] +mod shuffle; +#[macro_use] +mod shuffle1_dyn; +#[macro_use] +mod slice; +#[macro_use] +mod swap_bytes; +#[macro_use] +mod bit_manip; + +#[cfg(feature = "into_bits")] +pub(crate) mod into_bits; + +macro_rules! impl_i { + ([$elem_ty:ident; $elem_n:expr]: $tuple_id:ident, $mask_ty:ident + | $ielem_ty:ident, $ibitmask_ty:ident | $test_tt:tt | $($elem_ids:ident),* + | From: $($from_vec_ty:ident),* | $(#[$doc:meta])*) => { + impl_minimal_iuf!([$elem_ty; $elem_n]: $tuple_id | $ielem_ty | $test_tt + | $($elem_ids),* | $(#[$doc])*); + impl_ops_vector_arithmetic!([$elem_ty; $elem_n]: $tuple_id | $test_tt); + impl_ops_scalar_arithmetic!([$elem_ty; $elem_n]: $tuple_id | $test_tt); + impl_ops_vector_bitwise!( + [$elem_ty; $elem_n]: $tuple_id | $test_tt | (!(0 as $elem_ty), 0) + ); + impl_ops_scalar_bitwise!( + [$elem_ty; $elem_n]: $tuple_id | $test_tt | (!(0 as $elem_ty), 0) + ); + impl_ops_vector_shifts!([$elem_ty; $elem_n]: $tuple_id | $test_tt); + impl_ops_scalar_shifts!([$elem_ty; $elem_n]: $tuple_id | $test_tt); + impl_ops_vector_rotates!([$elem_ty; $elem_n]: $tuple_id | $test_tt); + impl_ops_vector_neg!([$elem_ty; $elem_n]: $tuple_id | $test_tt); + impl_ops_vector_int_min_max!( + [$elem_ty; $elem_n]: $tuple_id | $test_tt + ); + impl_reduction_integer_arithmetic!( + [$elem_ty; $elem_n]: $tuple_id | $ielem_ty | $test_tt + ); + impl_reduction_min_max!( + [$elem_ty; $elem_n]: $tuple_id | $ielem_ty | $test_tt + ); + impl_reduction_bitwise!( + [$elem_ty; $elem_n]: $tuple_id | $ielem_ty | $test_tt + | (|x|{ x as $elem_ty }) | (!(0 as $elem_ty), 0) + ); + impl_fmt_debug!([$elem_ty; $elem_n]: $tuple_id | $test_tt); + impl_fmt_lower_hex!([$elem_ty; $elem_n]: $tuple_id | $test_tt); + impl_fmt_upper_hex!([$elem_ty; $elem_n]: $tuple_id | $test_tt); + impl_fmt_octal!([$elem_ty; $elem_n]: $tuple_id | $test_tt); + impl_fmt_binary!([$elem_ty; $elem_n]: $tuple_id | $test_tt); + impl_from_array!([$elem_ty; $elem_n]: $tuple_id | $test_tt | (1, 1)); + impl_from_vectors!( + [$elem_ty; $elem_n]: $tuple_id | $test_tt | $($from_vec_ty),* + ); + impl_default!([$elem_ty; $elem_n]: $tuple_id | $test_tt); + impl_hash!([$elem_ty; $elem_n]: $tuple_id | $test_tt); + impl_slice_from_slice!([$elem_ty; $elem_n]: $tuple_id | $test_tt); + impl_slice_write_to_slice!([$elem_ty; $elem_n]: $tuple_id | $test_tt); + impl_swap_bytes!([$elem_ty; $elem_n]: $tuple_id | $test_tt); + impl_bit_manip!([$elem_ty; $elem_n]: $tuple_id | $test_tt); + impl_shuffle1_dyn!([$elem_ty; $elem_n]: $tuple_id | $test_tt); + impl_cmp_partial_eq!( + [$elem_ty; $elem_n]: $tuple_id | $test_tt | (0, 1) + ); + impl_cmp_eq!([$elem_ty; $elem_n]: $tuple_id | $test_tt | (0, 1)); + impl_cmp_vertical!( + [$elem_ty; $elem_n]: $tuple_id, $mask_ty, false, (1, 0) | $test_tt + ); + impl_cmp_partial_ord!([$elem_ty; $elem_n]: $tuple_id | $test_tt); + impl_cmp_ord!([$elem_ty; $elem_n]: $tuple_id | $test_tt | (0, 1)); + impl_bitmask!($tuple_id | $ibitmask_ty | (-1, 0) | $test_tt); + + test_select!($elem_ty, $mask_ty, $tuple_id, (1, 2) | $test_tt); + test_cmp_partial_ord_int!([$elem_ty; $elem_n]: $tuple_id | $test_tt); + test_shuffle1_dyn!([$elem_ty; $elem_n]: $tuple_id | $test_tt); + } +} + +macro_rules! impl_u { + ([$elem_ty:ident; $elem_n:expr]: $tuple_id:ident, $mask_ty:ident + | $ielem_ty:ident, $ibitmask_ty:ident | $test_tt:tt | $($elem_ids:ident),* + | From: $($from_vec_ty:ident),* | $(#[$doc:meta])*) => { + impl_minimal_iuf!([$elem_ty; $elem_n]: $tuple_id | $ielem_ty | $test_tt + | $($elem_ids),* | $(#[$doc])*); + impl_ops_vector_arithmetic!([$elem_ty; $elem_n]: $tuple_id | $test_tt); + impl_ops_scalar_arithmetic!([$elem_ty; $elem_n]: $tuple_id | $test_tt); + impl_ops_vector_bitwise!( + [$elem_ty; $elem_n]: $tuple_id | $test_tt | (!(0 as $elem_ty), 0) + ); + impl_ops_scalar_bitwise!( + [$elem_ty; $elem_n]: $tuple_id | $test_tt | (!(0 as $elem_ty), 0) + ); + impl_ops_vector_shifts!([$elem_ty; $elem_n]: $tuple_id | $test_tt); + impl_ops_scalar_shifts!([$elem_ty; $elem_n]: $tuple_id | $test_tt); + impl_ops_vector_rotates!([$elem_ty; $elem_n]: $tuple_id | $test_tt); + impl_ops_vector_int_min_max!( + [$elem_ty; $elem_n]: $tuple_id | $test_tt + ); + impl_reduction_integer_arithmetic!( + [$elem_ty; $elem_n]: $tuple_id | $ielem_ty | $test_tt + ); + impl_reduction_min_max!( + [$elem_ty; $elem_n]: $tuple_id | $ielem_ty | $test_tt + ); + impl_reduction_bitwise!( + [$elem_ty; $elem_n]: $tuple_id | $ielem_ty | $test_tt + | (|x|{ x as $elem_ty }) | (!(0 as $elem_ty), 0) + ); + impl_fmt_debug!([$elem_ty; $elem_n]: $tuple_id | $test_tt); + impl_fmt_lower_hex!([$elem_ty; $elem_n]: $tuple_id | $test_tt); + impl_fmt_upper_hex!([$elem_ty; $elem_n]: $tuple_id | $test_tt); + impl_fmt_octal!([$elem_ty; $elem_n]: $tuple_id | $test_tt); + impl_fmt_binary!([$elem_ty; $elem_n]: $tuple_id | $test_tt); + impl_from_array!([$elem_ty; $elem_n]: $tuple_id | $test_tt | (1, 1)); + impl_from_vectors!( + [$elem_ty; $elem_n]: $tuple_id | $test_tt | $($from_vec_ty),* + ); + impl_default!([$elem_ty; $elem_n]: $tuple_id | $test_tt); + impl_hash!([$elem_ty; $elem_n]: $tuple_id | $test_tt); + impl_slice_from_slice!([$elem_ty; $elem_n]: $tuple_id | $test_tt); + impl_slice_write_to_slice!([$elem_ty; $elem_n]: $tuple_id | $test_tt); + impl_swap_bytes!([$elem_ty; $elem_n]: $tuple_id | $test_tt); + impl_bit_manip!([$elem_ty; $elem_n]: $tuple_id | $test_tt); + impl_shuffle1_dyn!([$elem_ty; $elem_n]: $tuple_id | $test_tt); + impl_cmp_partial_eq!( + [$elem_ty; $elem_n]: $tuple_id | $test_tt | (1, 0) + ); + impl_cmp_eq!([$elem_ty; $elem_n]: $tuple_id | $test_tt | (0, 1)); + impl_cmp_vertical!( + [$elem_ty; $elem_n]: $tuple_id, $mask_ty, false, (1, 0) | $test_tt + ); + impl_cmp_partial_ord!([$elem_ty; $elem_n]: $tuple_id | $test_tt); + impl_cmp_ord!([$elem_ty; $elem_n]: $tuple_id | $test_tt | (0, 1)); + impl_bitmask!($tuple_id | $ibitmask_ty | ($ielem_ty::max_value(), 0) | + $test_tt); + + test_select!($elem_ty, $mask_ty, $tuple_id, (1, 2) | $test_tt); + test_cmp_partial_ord_int!([$elem_ty; $elem_n]: $tuple_id | $test_tt); + test_shuffle1_dyn!([$elem_ty; $elem_n]: $tuple_id | $test_tt); + } +} + +macro_rules! impl_f { + ([$elem_ty:ident; $elem_n:expr]: $tuple_id:ident, $mask_ty:ident + | $ielem_ty:ident | $test_tt:tt | $($elem_ids:ident),* + | From: $($from_vec_ty:ident),* | $(#[$doc:meta])*) => { + impl_minimal_iuf!([$elem_ty; $elem_n]: $tuple_id | $ielem_ty | $test_tt + | $($elem_ids),* | $(#[$doc])*); + impl_ops_vector_arithmetic!([$elem_ty; $elem_n]: $tuple_id | $test_tt); + impl_ops_scalar_arithmetic!([$elem_ty; $elem_n]: $tuple_id | $test_tt); + impl_ops_vector_neg!([$elem_ty; $elem_n]: $tuple_id | $test_tt); + impl_ops_vector_float_min_max!( + [$elem_ty; $elem_n]: $tuple_id | $test_tt + ); + impl_reduction_float_arithmetic!( + [$elem_ty; $elem_n]: $tuple_id | $test_tt); + impl_reduction_min_max!( + [$elem_ty; $elem_n]: $tuple_id | $ielem_ty | $test_tt + ); + impl_fmt_debug!([$elem_ty; $elem_n]: $tuple_id | $test_tt); + impl_from_array!([$elem_ty; $elem_n]: $tuple_id | $test_tt | (1., 1.)); + impl_from_vectors!( + [$elem_ty; $elem_n]: $tuple_id | $test_tt | $($from_vec_ty),* + ); + impl_default!([$elem_ty; $elem_n]: $tuple_id | $test_tt); + impl_cmp_partial_eq!( + [$elem_ty; $elem_n]: $tuple_id | $test_tt | (1., 0.) + ); + impl_slice_from_slice!([$elem_ty; $elem_n]: $tuple_id | $test_tt); + impl_slice_write_to_slice!([$elem_ty; $elem_n]: $tuple_id | $test_tt); + impl_shuffle1_dyn!([$elem_ty; $elem_n]: $tuple_id | $test_tt); + + impl_float_consts!([$elem_ty; $elem_n]: $tuple_id); + impl_float_category!([$elem_ty; $elem_n]: $tuple_id, $mask_ty); + + // floating-point math + impl_math_float_abs!([$elem_ty; $elem_n]: $tuple_id | $test_tt); + impl_math_float_cos!([$elem_ty; $elem_n]: $tuple_id | $test_tt); + impl_math_float_exp!([$elem_ty; $elem_n]: $tuple_id | $test_tt); + impl_math_float_ln!([$elem_ty; $elem_n]: $tuple_id | $test_tt); + impl_math_float_mul_add!([$elem_ty; $elem_n]: $tuple_id | $test_tt); + impl_math_float_mul_adde!([$elem_ty; $elem_n]: $tuple_id | $test_tt); + impl_math_float_powf!([$elem_ty; $elem_n]: $tuple_id | $test_tt); + impl_math_float_recpre!([$elem_ty; $elem_n]: $tuple_id | $test_tt); + impl_math_float_rsqrte!([$elem_ty; $elem_n]: $tuple_id | $test_tt); + impl_math_float_sin!([$elem_ty; $elem_n]: $tuple_id | $test_tt); + impl_math_float_sqrt!([$elem_ty; $elem_n]: $tuple_id | $test_tt); + impl_math_float_sqrte!([$elem_ty; $elem_n]: $tuple_id | $test_tt); + impl_math_float_tanh!([$elem_ty; $elem_n]: $tuple_id | $test_tt); + impl_cmp_vertical!( + [$elem_ty; $elem_n]: $tuple_id, $mask_ty, false, (1., 0.) + | $test_tt + ); + + test_select!($elem_ty, $mask_ty, $tuple_id, (1., 2.) | $test_tt); + test_reduction_float_min_max!( + [$elem_ty; $elem_n]: $tuple_id | $test_tt + ); + test_shuffle1_dyn!([$elem_ty; $elem_n]: $tuple_id | $test_tt); + } +} + +macro_rules! impl_m { + ([$elem_ty:ident; $elem_n:expr]: $tuple_id:ident + | $ielem_ty:ident, $ibitmask_ty:ident + | $test_tt:tt | $($elem_ids:ident),* | From: $($from_vec_ty:ident),* + | $(#[$doc:meta])*) => { + impl_minimal_mask!( + [$elem_ty; $elem_n]: $tuple_id | $ielem_ty | $test_tt + | $($elem_ids),* | $(#[$doc])* + ); + impl_ops_vector_mask_bitwise!( + [$elem_ty; $elem_n]: $tuple_id | $test_tt | (true, false) + ); + impl_ops_scalar_mask_bitwise!( + [$elem_ty; $elem_n]: $tuple_id | $test_tt | (true, false) + ); + impl_reduction_bitwise!( + [bool; $elem_n]: $tuple_id | $ielem_ty | $test_tt + | (|x|{ x != 0 }) | (true, false) + ); + impl_reduction_mask!([$elem_ty; $elem_n]: $tuple_id | $test_tt); + impl_fmt_debug!([bool; $elem_n]: $tuple_id | $test_tt); + impl_from_array!( + [$elem_ty; $elem_n]: $tuple_id | $test_tt + | (crate::$elem_ty::new(true), true) + ); + impl_from_vectors!( + [$elem_ty; $elem_n]: $tuple_id | $test_tt | $($from_vec_ty),* + ); + impl_default!([bool; $elem_n]: $tuple_id | $test_tt); + impl_cmp_partial_eq!( + [$elem_ty; $elem_n]: $tuple_id | $test_tt | (true, false) + ); + impl_cmp_eq!( + [$elem_ty; $elem_n]: $tuple_id | $test_tt | (true, false) + ); + impl_cmp_vertical!( + [$elem_ty; $elem_n]: $tuple_id, $tuple_id, true, (true, false) + | $test_tt + ); + impl_select!([$elem_ty; $elem_n]: $tuple_id | $test_tt); + impl_cmp_partial_ord!([$elem_ty; $elem_n]: $tuple_id | $test_tt); + impl_cmp_ord!( + [$elem_ty; $elem_n]: $tuple_id | $test_tt | (false, true) + ); + impl_shuffle1_dyn!([$elem_ty; $elem_n]: $tuple_id | $test_tt); + impl_bitmask!($tuple_id | $ibitmask_ty | (true, false) | $test_tt); + + test_cmp_partial_ord_mask!([$elem_ty; $elem_n]: $tuple_id | $test_tt); + test_shuffle1_dyn_mask!([$elem_ty; $elem_n]: $tuple_id | $test_tt); + } +} + +macro_rules! impl_const_p { + ([$elem_ty:ty; $elem_n:expr]: $tuple_id:ident, $mask_ty:ident, + $usize_ty:ident, $isize_ty:ident + | $test_tt:tt | $($elem_ids:ident),* + | From: $($from_vec_ty:ident),* | $(#[$doc:meta])*) => { + impl_minimal_p!( + [$elem_ty; $elem_n]: $tuple_id, $mask_ty, $usize_ty, $isize_ty + | ref_ | $test_tt | $($elem_ids),* + | (1 as $elem_ty, 0 as $elem_ty) | $(#[$doc])* + ); + impl_ptr_read!([$elem_ty; $elem_n]: $tuple_id, $mask_ty | $test_tt); + } +} + +macro_rules! impl_mut_p { + ([$elem_ty:ty; $elem_n:expr]: $tuple_id:ident, $mask_ty:ident, + $usize_ty:ident, $isize_ty:ident + | $test_tt:tt | $($elem_ids:ident),* + | From: $($from_vec_ty:ident),* | $(#[$doc:meta])*) => { + impl_minimal_p!( + [$elem_ty; $elem_n]: $tuple_id, $mask_ty, $usize_ty, $isize_ty + | ref_mut_ | $test_tt | $($elem_ids),* + | (1 as $elem_ty, 0 as $elem_ty) | $(#[$doc])* + ); + impl_ptr_read!([$elem_ty; $elem_n]: $tuple_id, $mask_ty | $test_tt); + impl_ptr_write!([$elem_ty; $elem_n]: $tuple_id, $mask_ty | $test_tt); + } +} diff --git a/third_party/rust/packed_simd/src/api/bit_manip.rs b/third_party/rust/packed_simd/src/api/bit_manip.rs new file mode 100644 index 000000000000..c1e90bb0fb85 --- /dev/null +++ b/third_party/rust/packed_simd/src/api/bit_manip.rs @@ -0,0 +1,129 @@ +//! Bit manipulations. + +macro_rules! impl_bit_manip { + ([$elem_ty:ident; $elem_count:expr]: $id:ident | $test_tt:tt) => { + impl $id { + /// Returns the number of ones in the binary representation of + /// the lanes of `self`. + #[inline] + pub fn count_ones(self) -> Self { + super::codegen::bit_manip::BitManip::ctpop(self) + } + + /// Returns the number of zeros in the binary representation of + /// the lanes of `self`. + #[inline] + pub fn count_zeros(self) -> Self { + super::codegen::bit_manip::BitManip::ctpop(!self) + } + + /// Returns the number of leading zeros in the binary + /// representation of the lanes of `self`. + #[inline] + pub fn leading_zeros(self) -> Self { + super::codegen::bit_manip::BitManip::ctlz(self) + } + + /// Returns the number of trailing zeros in the binary + /// representation of the lanes of `self`. + #[inline] + pub fn trailing_zeros(self) -> Self { + super::codegen::bit_manip::BitManip::cttz(self) + } + } + + test_if! { + $test_tt: + paste::item! { + #[allow(overflowing_literals)] + pub mod [<$id _bit_manip>] { + #![allow(const_item_mutation)] + use super::*; + + const LANE_WIDTH: usize = mem::size_of::<$elem_ty>() * 8; + + macro_rules! test_func { + ($x:expr, $func:ident) => {{ + let mut actual = $x; + for i in 0..$id::lanes() { + actual = actual.replace( + i, + $x.extract(i).$func() as $elem_ty + ); + } + let expected = $x.$func(); + assert_eq!(actual, expected); + }}; + } + + const BYTES: [u8; 64] = [ + 0, 1, 2, 3, 4, 5, 6, 7, + 8, 9, 10, 11, 12, 13, 14, 15, + 16, 17, 18, 19, 20, 21, 22, 23, + 24, 25, 26, 27, 28, 29, 30, 31, + 32, 33, 34, 35, 36, 37, 38, 39, + 40, 41, 42, 43, 44, 45, 46, 47, + 48, 49, 50, 51, 52, 53, 54, 55, + 56, 57, 58, 59, 60, 61, 62, 63, + ]; + + fn load_bytes() -> $id { + let elems: &mut [$elem_ty] = unsafe { + slice::from_raw_parts_mut( + BYTES.as_mut_ptr() as *mut $elem_ty, + $id::lanes(), + ) + }; + $id::from_slice_unaligned(elems) + } + + #[cfg_attr(not(target_arch = "wasm32"), test)] + #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] + fn count_ones() { + test_func!($id::splat(0), count_ones); + test_func!($id::splat(!0), count_ones); + test_func!(load_bytes(), count_ones); + } + + #[cfg_attr(not(target_arch = "wasm32"), test)] + #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] + fn count_zeros() { + test_func!($id::splat(0), count_zeros); + test_func!($id::splat(!0), count_zeros); + test_func!(load_bytes(), count_zeros); + } + + #[cfg_attr(not(target_arch = "wasm32"), test)] + #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] + fn leading_zeros() { + test_func!($id::splat(0), leading_zeros); + test_func!($id::splat(1), leading_zeros); + // some implementations use `pshufb` which has unique + // behavior when the 8th bit is set. + test_func!($id::splat(0b1000_0010), leading_zeros); + test_func!($id::splat(!0), leading_zeros); + test_func!( + $id::splat(1 << (LANE_WIDTH - 1)), + leading_zeros + ); + test_func!(load_bytes(), leading_zeros); + } + + #[cfg_attr(not(target_arch = "wasm32"), test)] + #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] + fn trailing_zeros() { + test_func!($id::splat(0), trailing_zeros); + test_func!($id::splat(1), trailing_zeros); + test_func!($id::splat(0b1000_0010), trailing_zeros); + test_func!($id::splat(!0), trailing_zeros); + test_func!( + $id::splat(1 << (LANE_WIDTH - 1)), + trailing_zeros + ); + test_func!(load_bytes(), trailing_zeros); + } + } + } + } + }; +} diff --git a/third_party/rust/packed_simd/src/api/bitmask.rs b/third_party/rust/packed_simd/src/api/bitmask.rs new file mode 100644 index 000000000000..8f4868f32844 --- /dev/null +++ b/third_party/rust/packed_simd/src/api/bitmask.rs @@ -0,0 +1,79 @@ +//! Bitmask API + +macro_rules! impl_bitmask { + ($id:ident | $ibitmask_ty:ident | ($set:expr, $clear:expr) + | $test_tt:tt) => { + impl $id { + /// Creates a bitmask with the MSB of each vector lane. + /// + /// If the vector has less than 8 lanes, the bits that do not + /// correspond to any vector lanes are cleared. + #[inline] + pub fn bitmask(self) -> $ibitmask_ty { + unsafe { codegen::llvm::simd_bitmask(self.0) } + } + } + + test_if! { + $test_tt: + paste::item! { + #[cfg(not( + // FIXME: https://github.com/rust-lang-nursery/packed_simd/issues/210 + target_endian = "big" + ))] + pub mod [<$id _bitmask>] { + use super::*; + #[cfg_attr(not(target_arch = "wasm32"), test)] + #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] + fn bitmask() { + // clear all lanes + let vec = $id::splat($clear as _); + let bitmask: $ibitmask_ty = 0; + assert_eq!(vec.bitmask(), bitmask); + + // set even lanes + let mut vec = $id::splat($clear as _); + for i in 0..$id::lanes() { + if i % 2 == 0 { + vec = vec.replace(i, $set as _); + } + } + // create bitmask with even lanes set: + let mut bitmask: $ibitmask_ty = 0; + for i in 0..$id::lanes() { + if i % 2 == 0 { + bitmask |= 1 << i; + } + } + assert_eq!(vec.bitmask(), bitmask); + + + // set odd lanes + let mut vec = $id::splat($clear as _); + for i in 0..$id::lanes() { + if i % 2 != 0 { + vec = vec.replace(i, $set as _); + } + } + // create bitmask with odd lanes set: + let mut bitmask: $ibitmask_ty = 0; + for i in 0..$id::lanes() { + if i % 2 != 0 { + bitmask |= 1 << i; + } + } + assert_eq!(vec.bitmask(), bitmask); + + // set all lanes + let vec = $id::splat($set as _); + let mut bitmask: $ibitmask_ty = 0; + for i in 0..$id::lanes() { + bitmask |= 1 << i; + } + assert_eq!(vec.bitmask(), bitmask); + } + } + } + } + }; +} diff --git a/third_party/rust/packed_simd/src/api/cast.rs b/third_party/rust/packed_simd/src/api/cast.rs new file mode 100644 index 000000000000..f1c32ca1a38b --- /dev/null +++ b/third_party/rust/packed_simd/src/api/cast.rs @@ -0,0 +1,108 @@ +//! Implementation of `FromCast` and `IntoCast`. +#![allow(clippy::module_name_repetitions)] + +/// Numeric cast from `T` to `Self`. +/// +/// > Note: This is a temporary workaround until the conversion traits +/// specified > in [RFC2484] are implemented. +/// +/// Numeric cast between vectors with the same number of lanes, such that: +/// +/// * casting integer vectors whose lane types have the same size (e.g. `i32xN` +/// -> `u32xN`) is a **no-op**, +/// +/// * casting from a larger integer to a smaller integer (e.g. `u32xN` -> +/// `u8xN`) will **truncate**, +/// +/// * casting from a smaller integer to a larger integer (e.g. `u8xN` -> +/// `u32xN`) will: +/// * **zero-extend** if the source is unsigned, or +/// * **sign-extend** if the source is signed, +/// +/// * casting from a float to an integer will **round the float towards zero**, +/// +/// * casting from an integer to float will produce the floating point +/// representation of the integer, **rounding to nearest, ties to even**, +/// +/// * casting from an `f32` to an `f64` is perfect and lossless, +/// +/// * casting from an `f64` to an `f32` **rounds to nearest, ties to even**. +/// +/// [RFC2484]: https://github.com/rust-lang/rfcs/pull/2484 +pub trait FromCast: crate::marker::Sized { + /// Numeric cast from `T` to `Self`. + fn from_cast(_: T) -> Self; +} + +/// Numeric cast from `Self` to `T`. +/// +/// > Note: This is a temporary workaround until the conversion traits +/// specified > in [RFC2484] are implemented. +/// +/// Numeric cast between vectors with the same number of lanes, such that: +/// +/// * casting integer vectors whose lane types have the same size (e.g. `i32xN` +/// -> `u32xN`) is a **no-op**, +/// +/// * casting from a larger integer to a smaller integer (e.g. `u32xN` -> +/// `u8xN`) will **truncate**, +/// +/// * casting from a smaller integer to a larger integer (e.g. `u8xN` -> +/// `u32xN`) will: +/// * **zero-extend** if the source is unsigned, or +/// * **sign-extend** if the source is signed, +/// +/// * casting from a float to an integer will **round the float towards zero**, +/// +/// * casting from an integer to float will produce the floating point +/// representation of the integer, **rounding to nearest, ties to even**, +/// +/// * casting from an `f32` to an `f64` is perfect and lossless, +/// +/// * casting from an `f64` to an `f32` **rounds to nearest, ties to even**. +/// +/// [RFC2484]: https://github.com/rust-lang/rfcs/pull/2484 +pub trait Cast: crate::marker::Sized { + /// Numeric cast from `self` to `T`. + fn cast(self) -> T; +} + +/// `FromCast` implies `Cast`. +impl Cast for T +where + U: FromCast, +{ + #[inline] + fn cast(self) -> U { + U::from_cast(self) + } +} + +/// `FromCast` and `Cast` are reflexive +impl FromCast for T { + #[inline] + fn from_cast(t: Self) -> Self { + t + } +} + +#[macro_use] +mod macros; + +mod v16; +pub use self::v16::*; + +mod v32; +pub use self::v32::*; + +mod v64; +pub use self::v64::*; + +mod v128; +pub use self::v128::*; + +mod v256; +pub use self::v256::*; + +mod v512; +pub use self::v512::*; diff --git a/third_party/rust/packed_simd/src/api/cast/macros.rs b/third_party/rust/packed_simd/src/api/cast/macros.rs new file mode 100644 index 000000000000..3bb29f0b80b7 --- /dev/null +++ b/third_party/rust/packed_simd/src/api/cast/macros.rs @@ -0,0 +1,82 @@ +//! Macros implementing `FromCast` + +macro_rules! impl_from_cast_ { + ($id:ident[$test_tt:tt]: $from_ty:ident) => { + impl crate::api::cast::FromCast<$from_ty> for $id { + #[inline] + fn from_cast(x: $from_ty) -> Self { + use crate::llvm::simd_cast; + debug_assert_eq!($from_ty::lanes(), $id::lanes()); + Simd(unsafe { simd_cast(x.0) }) + } + } + + test_if!{ + $test_tt: + paste::item! { + pub mod [<$id _from_cast_ $from_ty>] { + use super::*; + #[cfg_attr(not(target_arch = "wasm32"), test)] #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] + fn test() { + assert_eq!($id::lanes(), $from_ty::lanes()); + } + } + } + } + }; +} + +macro_rules! impl_from_cast { + ($id:ident[$test_tt:tt]: $($from_ty:ident),*) => { + $( + impl_from_cast_!($id[$test_tt]: $from_ty); + )* + } +} + +macro_rules! impl_from_cast_mask_ { + ($id:ident[$test_tt:tt]: $from_ty:ident) => { + impl crate::api::cast::FromCast<$from_ty> for $id { + #[inline] + fn from_cast(x: $from_ty) -> Self { + debug_assert_eq!($from_ty::lanes(), $id::lanes()); + x.ne($from_ty::default()) + .select($id::splat(true), $id::splat(false)) + } + } + + test_if!{ + $test_tt: + paste::item! { + pub mod [<$id _from_cast_ $from_ty>] { + use super::*; + #[cfg_attr(not(target_arch = "wasm32"), test)] #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] + fn test() { + assert_eq!($id::lanes(), $from_ty::lanes()); + + let x = $from_ty::default(); + let m: $id = x.cast(); + assert!(m.none()); + } + } + } + } + }; +} + +macro_rules! impl_from_cast_mask { + ($id:ident[$test_tt:tt]: $($from_ty:ident),*) => { + $( + impl_from_cast_mask_!($id[$test_tt]: $from_ty); + )* + } +} + +#[allow(unused)] +macro_rules! impl_into_cast { + ($id:ident[$test_tt:tt]: $($from_ty:ident),*) => { + $( + impl_from_cast_!($from_ty[$test_tt]: $id); + )* + } +} diff --git a/third_party/rust/packed_simd/src/api/cast/v128.rs b/third_party/rust/packed_simd/src/api/cast/v128.rs new file mode 100644 index 000000000000..2e10b97b7737 --- /dev/null +++ b/third_party/rust/packed_simd/src/api/cast/v128.rs @@ -0,0 +1,302 @@ +//! `FromCast` and `IntoCast` implementations for portable 128-bit wide vectors +#[rustfmt::skip] + +use crate::*; + +impl_from_cast!(i8x16[test_v128]: u8x16, m8x16, i16x16, u16x16, m16x16, i32x16, u32x16, f32x16, m32x16); +impl_from_cast!(u8x16[test_v128]: i8x16, m8x16, i16x16, u16x16, m16x16, i32x16, u32x16, f32x16, m32x16); +impl_from_cast_mask!(m8x16[test_v128]: i8x16, u8x16, i16x16, u16x16, m16x16, i32x16, u32x16, f32x16, m32x16); + +impl_from_cast!( + i16x8[test_v128]: i8x8, + u8x8, + m8x8, + u16x8, + m16x8, + i32x8, + u32x8, + f32x8, + m32x8, + i64x8, + u64x8, + f64x8, + m64x8, + isizex8, + usizex8, + msizex8 +); +impl_from_cast!( + u16x8[test_v128]: i8x8, + u8x8, + m8x8, + i16x8, + m16x8, + i32x8, + u32x8, + f32x8, + m32x8, + i64x8, + u64x8, + f64x8, + m64x8, + isizex8, + usizex8, + msizex8 +); +impl_from_cast_mask!( + m16x8[test_v128]: i8x8, + u8x8, + m8x8, + i16x8, + u16x8, + i32x8, + u32x8, + f32x8, + m32x8, + i64x8, + u64x8, + f64x8, + m64x8, + isizex8, + usizex8, + msizex8 +); + +impl_from_cast!( + i32x4[test_v128]: i8x4, + u8x4, + m8x4, + i16x4, + u16x4, + m16x4, + u32x4, + f32x4, + m32x4, + i64x4, + u64x4, + f64x4, + m64x4, + i128x4, + u128x4, + m128x4, + isizex4, + usizex4, + msizex4 +); +impl_from_cast!( + u32x4[test_v128]: i8x4, + u8x4, + m8x4, + i16x4, + u16x4, + m16x4, + i32x4, + f32x4, + m32x4, + i64x4, + u64x4, + f64x4, + m64x4, + i128x4, + u128x4, + m128x4, + isizex4, + usizex4, + msizex4 +); +impl_from_cast!( + f32x4[test_v128]: i8x4, + u8x4, + m8x4, + i16x4, + u16x4, + m16x4, + i32x4, + u32x4, + m32x4, + i64x4, + u64x4, + f64x4, + m64x4, + i128x4, + u128x4, + m128x4, + isizex4, + usizex4, + msizex4 +); +impl_from_cast_mask!( + m32x4[test_v128]: i8x4, + u8x4, + m8x4, + i16x4, + u16x4, + m16x4, + i32x4, + u32x4, + f32x4, + i64x4, + u64x4, + f64x4, + m64x4, + i128x4, + u128x4, + m128x4, + isizex4, + usizex4, + msizex4 +); + +impl_from_cast!( + i64x2[test_v128]: i8x2, + u8x2, + m8x2, + i16x2, + u16x2, + m16x2, + i32x2, + u32x2, + f32x2, + m32x2, + u64x2, + f64x2, + m64x2, + i128x2, + u128x2, + m128x2, + isizex2, + usizex2, + msizex2 +); +impl_from_cast!( + u64x2[test_v128]: i8x2, + u8x2, + m8x2, + i16x2, + u16x2, + m16x2, + i32x2, + u32x2, + f32x2, + m32x2, + i64x2, + f64x2, + m64x2, + i128x2, + u128x2, + m128x2, + isizex2, + usizex2, + msizex2 +); +impl_from_cast!( + f64x2[test_v128]: i8x2, + u8x2, + m8x2, + i16x2, + u16x2, + m16x2, + i32x2, + u32x2, + f32x2, + m32x2, + i64x2, + u64x2, + m64x2, + i128x2, + u128x2, + m128x2, + isizex2, + usizex2, + msizex2 +); +impl_from_cast_mask!( + m64x2[test_v128]: i8x2, + u8x2, + m8x2, + i16x2, + u16x2, + m16x2, + i32x2, + u32x2, + f32x2, + m32x2, + i64x2, + u64x2, + f64x2, + i128x2, + u128x2, + m128x2, + isizex2, + usizex2, + msizex2 +); + +impl_from_cast!( + isizex2[test_v128]: i8x2, + u8x2, + m8x2, + i16x2, + u16x2, + m16x2, + i32x2, + u32x2, + f32x2, + m32x2, + i64x2, + u64x2, + f64x2, + m64x2, + i128x2, + u128x2, + m128x2, + usizex2, + msizex2 +); +impl_from_cast!( + usizex2[test_v128]: i8x2, + u8x2, + m8x2, + i16x2, + u16x2, + m16x2, + i32x2, + u32x2, + f32x2, + m32x2, + i64x2, + u64x2, + f64x2, + m64x2, + i128x2, + u128x2, + m128x2, + isizex2, + msizex2 +); +impl_from_cast_mask!( + msizex2[test_v128]: i8x2, + u8x2, + m8x2, + i16x2, + u16x2, + m16x2, + i32x2, + u32x2, + f32x2, + m32x2, + i64x2, + u64x2, + f64x2, + m64x2, + i128x2, + u128x2, + m128x2, + isizex2, + usizex2 +); + +// FIXME[test_v128]: 64-bit single element vectors into_cast impls +impl_from_cast!(i128x1[test_v128]: u128x1, m128x1); +impl_from_cast!(u128x1[test_v128]: i128x1, m128x1); +impl_from_cast!(m128x1[test_v128]: i128x1, u128x1); diff --git a/third_party/rust/packed_simd/src/api/cast/v16.rs b/third_party/rust/packed_simd/src/api/cast/v16.rs new file mode 100644 index 000000000000..896febacbb42 --- /dev/null +++ b/third_party/rust/packed_simd/src/api/cast/v16.rs @@ -0,0 +1,68 @@ +//! `FromCast` and `IntoCast` implementations for portable 16-bit wide vectors +#[rustfmt::skip] + +use crate::*; + +impl_from_cast!( + i8x2[test_v16]: u8x2, + m8x2, + i16x2, + u16x2, + m16x2, + i32x2, + u32x2, + f32x2, + m32x2, + i64x2, + u64x2, + f64x2, + m64x2, + i128x2, + u128x2, + m128x2, + isizex2, + usizex2, + msizex2 +); +impl_from_cast!( + u8x2[test_v16]: i8x2, + m8x2, + i16x2, + u16x2, + m16x2, + i32x2, + u32x2, + f32x2, + m32x2, + i64x2, + u64x2, + f64x2, + m64x2, + i128x2, + u128x2, + m128x2, + isizex2, + usizex2, + msizex2 +); +impl_from_cast_mask!( + m8x2[test_v16]: i8x2, + u8x2, + i16x2, + u16x2, + m16x2, + i32x2, + u32x2, + f32x2, + m32x2, + i64x2, + u64x2, + f64x2, + m64x2, + i128x2, + u128x2, + m128x2, + isizex2, + usizex2, + msizex2 +); diff --git a/third_party/rust/packed_simd/src/api/cast/v256.rs b/third_party/rust/packed_simd/src/api/cast/v256.rs new file mode 100644 index 000000000000..fe0c835e3c90 --- /dev/null +++ b/third_party/rust/packed_simd/src/api/cast/v256.rs @@ -0,0 +1,298 @@ +//! `FromCast` and `IntoCast` implementations for portable 256-bit wide vectors +#[rustfmt::skip] + +use crate::*; + +impl_from_cast!(i8x32[test_v256]: u8x32, m8x32, i16x32, u16x32, m16x32); +impl_from_cast!(u8x32[test_v256]: i8x32, m8x32, i16x32, u16x32, m16x32); +impl_from_cast_mask!(m8x32[test_v256]: i8x32, u8x32, i16x32, u16x32, m16x32); + +impl_from_cast!(i16x16[test_v256]: i8x16, u8x16, m8x16, u16x16, m16x16, i32x16, u32x16, f32x16, m32x16); +impl_from_cast!(u16x16[test_v256]: i8x16, u8x16, m8x16, i16x16, m16x16, i32x16, u32x16, f32x16, m32x16); +impl_from_cast_mask!(m16x16[test_v256]: i8x16, u8x16, m8x16, i16x16, u16x16, i32x16, u32x16, f32x16, m32x16); + +impl_from_cast!( + i32x8[test_v256]: i8x8, + u8x8, + m8x8, + i16x8, + u16x8, + m16x8, + u32x8, + f32x8, + m32x8, + i64x8, + u64x8, + f64x8, + m64x8, + isizex8, + usizex8, + msizex8 +); +impl_from_cast!( + u32x8[test_v256]: i8x8, + u8x8, + m8x8, + i16x8, + u16x8, + m16x8, + i32x8, + f32x8, + m32x8, + i64x8, + u64x8, + f64x8, + m64x8, + isizex8, + usizex8, + msizex8 +); +impl_from_cast!( + f32x8[test_v256]: i8x8, + u8x8, + m8x8, + i16x8, + u16x8, + m16x8, + i32x8, + u32x8, + m32x8, + i64x8, + u64x8, + f64x8, + m64x8, + isizex8, + usizex8, + msizex8 +); +impl_from_cast_mask!( + m32x8[test_v256]: i8x8, + u8x8, + m8x8, + i16x8, + u16x8, + m16x8, + i32x8, + u32x8, + f32x8, + i64x8, + u64x8, + f64x8, + m64x8, + isizex8, + usizex8, + msizex8 +); + +impl_from_cast!( + i64x4[test_v256]: i8x4, + u8x4, + m8x4, + i16x4, + u16x4, + m16x4, + i32x4, + u32x4, + f32x4, + m32x4, + u64x4, + f64x4, + m64x4, + i128x4, + u128x4, + m128x4, + isizex4, + usizex4, + msizex4 +); +impl_from_cast!( + u64x4[test_v256]: i8x4, + u8x4, + m8x4, + i16x4, + u16x4, + m16x4, + i32x4, + u32x4, + f32x4, + m32x4, + i64x4, + f64x4, + m64x4, + i128x4, + u128x4, + m128x4, + isizex4, + usizex4, + msizex4 +); +impl_from_cast!( + f64x4[test_v256]: i8x4, + u8x4, + m8x4, + i16x4, + u16x4, + m16x4, + i32x4, + u32x4, + f32x4, + m32x4, + i64x4, + u64x4, + m64x4, + i128x4, + u128x4, + m128x4, + isizex4, + usizex4, + msizex4 +); +impl_from_cast_mask!( + m64x4[test_v256]: i8x4, + u8x4, + m8x4, + i16x4, + u16x4, + m16x4, + i32x4, + u32x4, + f32x4, + m32x4, + i64x4, + u64x4, + f64x4, + i128x4, + u128x4, + m128x4, + isizex4, + usizex4, + msizex4 +); + +impl_from_cast!( + i128x2[test_v256]: i8x2, + u8x2, + m8x2, + i16x2, + u16x2, + m16x2, + i32x2, + u32x2, + f32x2, + m32x2, + i64x2, + u64x2, + f64x2, + m64x2, + u128x2, + m128x2, + isizex2, + usizex2, + msizex2 +); +impl_from_cast!( + u128x2[test_v256]: i8x2, + u8x2, + m8x2, + i16x2, + u16x2, + m16x2, + i32x2, + u32x2, + f32x2, + m32x2, + i64x2, + u64x2, + f64x2, + m64x2, + i128x2, + m128x2, + isizex2, + usizex2, + msizex2 +); +impl_from_cast_mask!( + m128x2[test_v256]: i8x2, + u8x2, + m8x2, + i16x2, + u16x2, + m16x2, + i32x2, + u32x2, + f32x2, + m32x2, + i64x2, + u64x2, + m64x2, + f64x2, + i128x2, + u128x2, + isizex2, + usizex2, + msizex2 +); + +impl_from_cast!( + isizex4[test_v256]: i8x4, + u8x4, + m8x4, + i16x4, + u16x4, + m16x4, + i32x4, + u32x4, + f32x4, + m32x4, + i64x4, + u64x4, + f64x4, + m64x4, + i128x4, + u128x4, + m128x4, + usizex4, + msizex4 +); +impl_from_cast!( + usizex4[test_v256]: i8x4, + u8x4, + m8x4, + i16x4, + u16x4, + m16x4, + i32x4, + u32x4, + f32x4, + m32x4, + i64x4, + u64x4, + f64x4, + m64x4, + i128x4, + u128x4, + m128x4, + isizex4, + msizex4 +); +impl_from_cast_mask!( + msizex4[test_v256]: i8x4, + u8x4, + m8x4, + i16x4, + u16x4, + m16x4, + i32x4, + u32x4, + f32x4, + m32x4, + i64x4, + u64x4, + f64x4, + m64x4, + i128x4, + u128x4, + m128x4, + isizex4, + usizex4 +); diff --git a/third_party/rust/packed_simd/src/api/cast/v32.rs b/third_party/rust/packed_simd/src/api/cast/v32.rs new file mode 100644 index 000000000000..4ad1cbf74d2e --- /dev/null +++ b/third_party/rust/packed_simd/src/api/cast/v32.rs @@ -0,0 +1,132 @@ +//! `FromCast` and `IntoCast` implementations for portable 32-bit wide vectors +#[rustfmt::skip] + +use crate::*; + +impl_from_cast!( + i8x4[test_v32]: u8x4, + m8x4, + i16x4, + u16x4, + m16x4, + i32x4, + u32x4, + f32x4, + m32x4, + i64x4, + u64x4, + f64x4, + m64x4, + i128x4, + u128x4, + m128x4, + isizex4, + usizex4, + msizex4 +); +impl_from_cast!( + u8x4[test_v32]: i8x4, + m8x4, + i16x4, + u16x4, + m16x4, + i32x4, + u32x4, + f32x4, + m32x4, + i64x4, + u64x4, + f64x4, + m64x4, + i128x4, + u128x4, + m128x4, + isizex4, + usizex4, + msizex4 +); +impl_from_cast_mask!( + m8x4[test_v32]: i8x4, + u8x4, + i16x4, + u16x4, + m16x4, + i32x4, + u32x4, + f32x4, + m32x4, + i64x4, + u64x4, + f64x4, + m64x4, + i128x4, + u128x4, + m128x4, + isizex4, + usizex4, + msizex4 +); + +impl_from_cast!( + i16x2[test_v32]: i8x2, + u8x2, + m8x2, + u16x2, + m16x2, + i32x2, + u32x2, + f32x2, + m32x2, + i64x2, + u64x2, + f64x2, + m64x2, + i128x2, + u128x2, + m128x2, + isizex2, + usizex2, + msizex2 +); +impl_from_cast!( + u16x2[test_v32]: i8x2, + u8x2, + m8x2, + i16x2, + m16x2, + i32x2, + u32x2, + f32x2, + m32x2, + i64x2, + u64x2, + f64x2, + m64x2, + i128x2, + u128x2, + m128x2, + isizex2, + usizex2, + msizex2 +); +impl_from_cast_mask!( + m16x2[test_v32]: i8x2, + u8x2, + m8x2, + i16x2, + u16x2, + i32x2, + u32x2, + f32x2, + m32x2, + i64x2, + u64x2, + f64x2, + m64x2, + i128x2, + u128x2, + m128x2, + isizex2, + usizex2, + msizex2 +); diff --git a/third_party/rust/packed_simd/src/api/cast/v512.rs b/third_party/rust/packed_simd/src/api/cast/v512.rs new file mode 100644 index 000000000000..b64605045e33 --- /dev/null +++ b/third_party/rust/packed_simd/src/api/cast/v512.rs @@ -0,0 +1,209 @@ +//! `FromCast` and `IntoCast` implementations for portable 512-bit wide vectors +#[rustfmt::skip] + +use crate::*; + +impl_from_cast!(i8x64[test_v512]: u8x64, m8x64); +impl_from_cast!(u8x64[test_v512]: i8x64, m8x64); +impl_from_cast_mask!(m8x64[test_v512]: i8x64, u8x64); + +impl_from_cast!(i16x32[test_v512]: i8x32, u8x32, m8x32, u16x32, m16x32); +impl_from_cast!(u16x32[test_v512]: i8x32, u8x32, m8x32, i16x32, m16x32); +impl_from_cast_mask!(m16x32[test_v512]: i8x32, u8x32, m8x32, i16x32, u16x32); + +impl_from_cast!(i32x16[test_v512]: i8x16, u8x16, m8x16, i16x16, u16x16, m16x16, u32x16, f32x16, m32x16); +impl_from_cast!(u32x16[test_v512]: i8x16, u8x16, m8x16, i16x16, u16x16, m16x16, i32x16, f32x16, m32x16); +impl_from_cast!(f32x16[test_v512]: i8x16, u8x16, m8x16, i16x16, u16x16, m16x16, i32x16, u32x16, m32x16); +impl_from_cast_mask!(m32x16[test_v512]: i8x16, u8x16, m8x16, i16x16, u16x16, m16x16, i32x16, u32x16, f32x16); + +impl_from_cast!( + i64x8[test_v512]: i8x8, + u8x8, + m8x8, + i16x8, + u16x8, + m16x8, + i32x8, + u32x8, + f32x8, + m32x8, + u64x8, + f64x8, + m64x8, + isizex8, + usizex8, + msizex8 +); +impl_from_cast!( + u64x8[test_v512]: i8x8, + u8x8, + m8x8, + i16x8, + u16x8, + m16x8, + i32x8, + u32x8, + f32x8, + m32x8, + i64x8, + f64x8, + m64x8, + isizex8, + usizex8, + msizex8 +); +impl_from_cast!( + f64x8[test_v512]: i8x8, + u8x8, + m8x8, + i16x8, + u16x8, + m16x8, + i32x8, + u32x8, + f32x8, + m32x8, + i64x8, + u64x8, + m64x8, + isizex8, + usizex8, + msizex8 +); +impl_from_cast_mask!( + m64x8[test_v512]: i8x8, + u8x8, + m8x8, + i16x8, + u16x8, + m16x8, + i32x8, + u32x8, + f32x8, + m32x8, + i64x8, + u64x8, + f64x8, + isizex8, + usizex8, + msizex8 +); + +impl_from_cast!( + i128x4[test_v512]: i8x4, + u8x4, + m8x4, + i16x4, + u16x4, + m16x4, + i32x4, + u32x4, + f32x4, + m32x4, + i64x4, + u64x4, + f64x4, + m64x4, + u128x4, + m128x4, + isizex4, + usizex4, + msizex4 +); +impl_from_cast!( + u128x4[test_v512]: i8x4, + u8x4, + m8x4, + i16x4, + u16x4, + m16x4, + i32x4, + u32x4, + f32x4, + m32x4, + i64x4, + u64x4, + f64x4, + m64x4, + i128x4, + m128x4, + isizex4, + usizex4, + msizex4 +); +impl_from_cast_mask!( + m128x4[test_v512]: i8x4, + u8x4, + m8x4, + i16x4, + u16x4, + m16x4, + i32x4, + u32x4, + f32x4, + m32x4, + i64x4, + u64x4, + m64x4, + f64x4, + i128x4, + u128x4, + isizex4, + usizex4, + msizex4 +); + +impl_from_cast!( + isizex8[test_v512]: i8x8, + u8x8, + m8x8, + i16x8, + u16x8, + m16x8, + i32x8, + u32x8, + f32x8, + m32x8, + i64x8, + u64x8, + f64x8, + m64x8, + usizex8, + msizex8 +); +impl_from_cast!( + usizex8[test_v512]: i8x8, + u8x8, + m8x8, + i16x8, + u16x8, + m16x8, + i32x8, + u32x8, + f32x8, + m32x8, + i64x8, + u64x8, + f64x8, + m64x8, + isizex8, + msizex8 +); +impl_from_cast_mask!( + msizex8[test_v512]: i8x8, + u8x8, + m8x8, + i16x8, + u16x8, + m16x8, + i32x8, + u32x8, + f32x8, + m32x8, + i64x8, + u64x8, + f64x8, + m64x8, + isizex8, + usizex8 +); diff --git a/third_party/rust/packed_simd/src/api/cast/v64.rs b/third_party/rust/packed_simd/src/api/cast/v64.rs new file mode 100644 index 000000000000..b23d1a49171f --- /dev/null +++ b/third_party/rust/packed_simd/src/api/cast/v64.rs @@ -0,0 +1,208 @@ +//! `FromCast` and `IntoCast` implementations for portable 64-bit wide vectors +#[rustfmt::skip] + +use crate::*; + +impl_from_cast!( + i8x8[test_v64]: u8x8, + m8x8, + i16x8, + u16x8, + m16x8, + i32x8, + u32x8, + f32x8, + m32x8, + i64x8, + u64x8, + f64x8, + m64x8, + isizex8, + usizex8, + msizex8 +); +impl_from_cast!( + u8x8[test_v64]: i8x8, + m8x8, + i16x8, + u16x8, + m16x8, + i32x8, + u32x8, + f32x8, + m32x8, + i64x8, + u64x8, + f64x8, + m64x8, + isizex8, + usizex8, + msizex8 +); +impl_from_cast_mask!( + m8x8[test_v64]: i8x8, + u8x8, + i16x8, + u16x8, + m16x8, + i32x8, + u32x8, + f32x8, + m32x8, + i64x8, + u64x8, + f64x8, + m64x8, + isizex8, + usizex8, + msizex8 +); + +impl_from_cast!( + i16x4[test_v64]: i8x4, + u8x4, + m8x4, + u16x4, + m16x4, + i32x4, + u32x4, + f32x4, + m32x4, + i64x4, + u64x4, + f64x4, + m64x4, + i128x4, + u128x4, + m128x4, + isizex4, + usizex4, + msizex4 +); +impl_from_cast!( + u16x4[test_v64]: i8x4, + u8x4, + m8x4, + i16x4, + m16x4, + i32x4, + u32x4, + f32x4, + m32x4, + i64x4, + u64x4, + f64x4, + m64x4, + i128x4, + u128x4, + m128x4, + isizex4, + usizex4, + msizex4 +); +impl_from_cast_mask!( + m16x4[test_v64]: i8x4, + u8x4, + m8x4, + i16x4, + u16x4, + i32x4, + u32x4, + f32x4, + m32x4, + i64x4, + u64x4, + f64x4, + m64x4, + i128x4, + u128x4, + m128x4, + isizex4, + usizex4, + msizex4 +); + +impl_from_cast!( + i32x2[test_v64]: i8x2, + u8x2, + m8x2, + i16x2, + u16x2, + m16x2, + u32x2, + f32x2, + m32x2, + i64x2, + u64x2, + f64x2, + m64x2, + i128x2, + u128x2, + m128x2, + isizex2, + usizex2, + msizex2 +); +impl_from_cast!( + u32x2[test_v64]: i8x2, + u8x2, + m8x2, + i16x2, + u16x2, + m16x2, + i32x2, + f32x2, + m32x2, + i64x2, + u64x2, + f64x2, + m64x2, + i128x2, + u128x2, + m128x2, + isizex2, + usizex2, + msizex2 +); +impl_from_cast!( + f32x2[test_v64]: i8x2, + u8x2, + m8x2, + i16x2, + u16x2, + m16x2, + i32x2, + u32x2, + m32x2, + i64x2, + u64x2, + f64x2, + m64x2, + i128x2, + u128x2, + m128x2, + isizex2, + usizex2, + msizex2 +); +impl_from_cast_mask!( + m32x2[test_v64]: i8x2, + u8x2, + m8x2, + i16x2, + u16x2, + m16x2, + i32x2, + u32x2, + f32x2, + i64x2, + u64x2, + f64x2, + m64x2, + i128x2, + u128x2, + m128x2, + isizex2, + usizex2, + msizex2 +); diff --git a/third_party/rust/packed_simd/src/api/cmp.rs b/third_party/rust/packed_simd/src/api/cmp.rs new file mode 100644 index 000000000000..6d5301ddddbd --- /dev/null +++ b/third_party/rust/packed_simd/src/api/cmp.rs @@ -0,0 +1,16 @@ +//! Implement cmp traits for vector types + +#[macro_use] +mod partial_eq; + +#[macro_use] +mod eq; + +#[macro_use] +mod partial_ord; + +#[macro_use] +mod ord; + +#[macro_use] +mod vertical; diff --git a/third_party/rust/packed_simd/src/api/cmp/eq.rs b/third_party/rust/packed_simd/src/api/cmp/eq.rs new file mode 100644 index 000000000000..3c55d0dce57e --- /dev/null +++ b/third_party/rust/packed_simd/src/api/cmp/eq.rs @@ -0,0 +1,27 @@ +//! Implements `Eq` for vector types. + +macro_rules! impl_cmp_eq { + ( + [$elem_ty:ident; $elem_count:expr]: + $id:ident | $test_tt:tt | + ($true:expr, $false:expr) + ) => { + impl crate::cmp::Eq for $id {} + impl crate::cmp::Eq for LexicographicallyOrdered<$id> {} + + test_if!{ + $test_tt: + paste::item! { + pub mod [<$id _cmp_eq>] { + use super::*; + #[cfg_attr(not(target_arch = "wasm32"), test)] #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] + fn eq() { + fn foo(_: E) {} + let a = $id::splat($false); + foo(a); + } + } + } + } + }; +} diff --git a/third_party/rust/packed_simd/src/api/cmp/ord.rs b/third_party/rust/packed_simd/src/api/cmp/ord.rs new file mode 100644 index 000000000000..e54ba3bfde9a --- /dev/null +++ b/third_party/rust/packed_simd/src/api/cmp/ord.rs @@ -0,0 +1,43 @@ +//! Implements `Ord` for vector types. + +macro_rules! impl_cmp_ord { + ( + [$elem_ty:ident; $elem_count:expr]: + $id:ident | $test_tt:tt | + ($true:expr, $false:expr) + ) => { + impl $id { + /// Returns a wrapper that implements `Ord`. + #[inline] + pub fn lex_ord(&self) -> LexicographicallyOrdered<$id> { + LexicographicallyOrdered(*self) + } + } + + impl crate::cmp::Ord for LexicographicallyOrdered<$id> { + #[inline] + fn cmp(&self, other: &Self) -> crate::cmp::Ordering { + match self.partial_cmp(other) { + Some(x) => x, + None => unsafe { crate::hint::unreachable_unchecked() }, + } + } + } + + test_if!{ + $test_tt: + paste::item! { + pub mod [<$id _cmp_ord>] { + use super::*; + #[cfg_attr(not(target_arch = "wasm32"), test)] #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] + fn eq() { + fn foo(_: E) {} + let a = $id::splat($false); + foo(a.partial_lex_ord()); + foo(a.lex_ord()); + } + } + } + } + }; +} diff --git a/third_party/rust/packed_simd/src/api/cmp/partial_eq.rs b/third_party/rust/packed_simd/src/api/cmp/partial_eq.rs new file mode 100644 index 000000000000..d69dd47425f7 --- /dev/null +++ b/third_party/rust/packed_simd/src/api/cmp/partial_eq.rs @@ -0,0 +1,65 @@ +//! Implements `PartialEq` for vector types. + +macro_rules! impl_cmp_partial_eq { + ( + [$elem_ty:ident; $elem_count:expr]: + $id:ident | $test_tt:tt | + ($true:expr, $false:expr) + ) => { + // FIXME: https://github.com/rust-lang-nursery/rust-clippy/issues/2892 + #[allow(clippy::partialeq_ne_impl)] + impl crate::cmp::PartialEq<$id> for $id { + #[inline] + fn eq(&self, other: &Self) -> bool { + $id::eq(*self, *other).all() + } + #[inline] + fn ne(&self, other: &Self) -> bool { + $id::ne(*self, *other).any() + } + } + + // FIXME: https://github.com/rust-lang-nursery/rust-clippy/issues/2892 + #[allow(clippy::partialeq_ne_impl)] + impl crate::cmp::PartialEq> for LexicographicallyOrdered<$id> { + #[inline] + fn eq(&self, other: &Self) -> bool { + self.0 == other.0 + } + #[inline] + fn ne(&self, other: &Self) -> bool { + self.0 != other.0 + } + } + + test_if! { + $test_tt: + paste::item! { + pub mod [<$id _cmp_PartialEq>] { + use super::*; + #[cfg_attr(not(target_arch = "wasm32"), test)] + #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] + fn partial_eq() { + let a = $id::splat($false); + let b = $id::splat($true); + + assert!(a != b); + assert!(!(a == b)); + assert!(a == a); + assert!(!(a != a)); + + if $id::lanes() > 1 { + let a = $id::splat($false).replace(0, $true); + let b = $id::splat($true); + + assert!(a != b); + assert!(!(a == b)); + assert!(a == a); + assert!(!(a != a)); + } + } + } + } + } + }; +} diff --git a/third_party/rust/packed_simd/src/api/cmp/partial_ord.rs b/third_party/rust/packed_simd/src/api/cmp/partial_ord.rs new file mode 100644 index 000000000000..76ed9ebe4e5a --- /dev/null +++ b/third_party/rust/packed_simd/src/api/cmp/partial_ord.rs @@ -0,0 +1,230 @@ +//! Implements `PartialOrd` for vector types. +//! +//! This implements a lexicographical order. + +macro_rules! impl_cmp_partial_ord { + ([$elem_ty:ident; $elem_count:expr]: $id:ident | $test_tt:tt) => { + impl $id { + /// Returns a wrapper that implements `PartialOrd`. + #[inline] + pub fn partial_lex_ord(&self) -> LexicographicallyOrdered<$id> { + LexicographicallyOrdered(*self) + } + } + + impl crate::cmp::PartialOrd> for LexicographicallyOrdered<$id> { + #[inline] + fn partial_cmp(&self, other: &Self) -> Option { + if PartialEq::eq(self, other) { + Some(crate::cmp::Ordering::Equal) + } else if PartialOrd::lt(self, other) { + Some(crate::cmp::Ordering::Less) + } else if PartialOrd::gt(self, other) { + Some(crate::cmp::Ordering::Greater) + } else { + None + } + } + #[inline] + fn lt(&self, other: &Self) -> bool { + let m_lt = self.0.lt(other.0); + let m_eq = self.0.eq(other.0); + for i in 0..$id::lanes() { + if m_eq.extract(i) { + continue; + } + return m_lt.extract(i); + } + false + } + #[inline] + fn le(&self, other: &Self) -> bool { + self.lt(other) | PartialEq::eq(self, other) + } + #[inline] + fn ge(&self, other: &Self) -> bool { + self.gt(other) | PartialEq::eq(self, other) + } + #[inline] + fn gt(&self, other: &Self) -> bool { + let m_gt = self.0.gt(other.0); + let m_eq = self.0.eq(other.0); + for i in 0..$id::lanes() { + if m_eq.extract(i) { + continue; + } + return m_gt.extract(i); + } + false + } + } + }; +} + +macro_rules! test_cmp_partial_ord_int { + ([$elem_ty:ident; $elem_count:expr]: $id:ident | $test_tt:tt) => { + test_if!{ + $test_tt: + paste::item! { + pub mod [<$id _cmp_PartialOrd>] { + use super::*; + #[cfg_attr(not(target_arch = "wasm32"), test)] #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] + fn partial_lex_ord() { + use crate::testing::utils::{test_cmp}; + // constant values + let a = $id::splat(0); + let b = $id::splat(1); + + test_cmp(a.partial_lex_ord(), b.partial_lex_ord(), + Some(crate::cmp::Ordering::Less)); + test_cmp(b.partial_lex_ord(), a.partial_lex_ord(), + Some(crate::cmp::Ordering::Greater)); + test_cmp(a.partial_lex_ord(), a.partial_lex_ord(), + Some(crate::cmp::Ordering::Equal)); + test_cmp(b.partial_lex_ord(), b.partial_lex_ord(), + Some(crate::cmp::Ordering::Equal)); + + // variable values: a = [0, 1, 2, 3]; b = [3, 2, 1, 0] + let mut a = $id::splat(0); + let mut b = $id::splat(0); + for i in 0..$id::lanes() { + a = a.replace(i, i as $elem_ty); + b = b.replace(i, ($id::lanes() - i) as $elem_ty); + } + test_cmp(a.partial_lex_ord(), b.partial_lex_ord(), + Some(crate::cmp::Ordering::Less)); + test_cmp(b.partial_lex_ord(), a.partial_lex_ord(), + Some(crate::cmp::Ordering::Greater)); + test_cmp(a.partial_lex_ord(), a.partial_lex_ord(), + Some(crate::cmp::Ordering::Equal)); + test_cmp(b.partial_lex_ord(), b.partial_lex_ord(), + Some(crate::cmp::Ordering::Equal)); + + // variable values: a = [0, 1, 2, 3]; b = [0, 1, 2, 4] + let mut b = a; + b = b.replace( + $id::lanes() - 1, + a.extract($id::lanes() - 1) + 1 as $elem_ty + ); + test_cmp(a.partial_lex_ord(), b.partial_lex_ord(), + Some(crate::cmp::Ordering::Less)); + test_cmp(b.partial_lex_ord(), a.partial_lex_ord(), + Some(crate::cmp::Ordering::Greater)); + test_cmp(a.partial_lex_ord(), a.partial_lex_ord(), + Some(crate::cmp::Ordering::Equal)); + test_cmp(b.partial_lex_ord(), b.partial_lex_ord(), + Some(crate::cmp::Ordering::Equal)); + + if $id::lanes() > 2 { + // variable values a = [0, 1, 0, 0]; b = [0, 1, 2, 3] + let b = a; + let mut a = $id::splat(0); + a = a.replace(1, 1 as $elem_ty); + test_cmp(a.partial_lex_ord(), b.partial_lex_ord(), + Some(crate::cmp::Ordering::Less)); + test_cmp(b.partial_lex_ord(), a.partial_lex_ord(), + Some(crate::cmp::Ordering::Greater)); + test_cmp(a.partial_lex_ord(), a.partial_lex_ord(), + Some(crate::cmp::Ordering::Equal)); + test_cmp(b.partial_lex_ord(), b.partial_lex_ord(), + Some(crate::cmp::Ordering::Equal)); + + // variable values: a = [0, 1, 2, 3]; b = [0, 1, 3, 2] + let mut b = a; + b = b.replace( + 2, a.extract($id::lanes() - 1) + 1 as $elem_ty + ); + test_cmp(a.partial_lex_ord(), b.partial_lex_ord(), + Some(crate::cmp::Ordering::Less)); + test_cmp(b.partial_lex_ord(), a.partial_lex_ord(), + Some(crate::cmp::Ordering::Greater)); + test_cmp(a.partial_lex_ord(), a.partial_lex_ord(), + Some(crate::cmp::Ordering::Equal)); + test_cmp(b.partial_lex_ord(), b.partial_lex_ord(), + Some(crate::cmp::Ordering::Equal)); + } + } + } + } + } + }; +} + +macro_rules! test_cmp_partial_ord_mask { + ([$elem_ty:ident; $elem_count:expr]: $id:ident | $test_tt:tt) => { + test_if!{ + $test_tt: + paste::item! { + pub mod [<$id _cmp_PartialOrd>] { + use super::*; + #[cfg_attr(not(target_arch = "wasm32"), test)] #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] + fn partial_lex_ord() { + use crate::testing::utils::{test_cmp}; + use crate::cmp::Ordering; + + // constant values + let a = $id::splat(false); + let b = $id::splat(true); + + test_cmp(a.partial_lex_ord(), b.partial_lex_ord(), + Some(Ordering::Less)); + test_cmp(b.partial_lex_ord(), a.partial_lex_ord(), + Some(Ordering::Greater)); + test_cmp(a.partial_lex_ord(), a.partial_lex_ord(), + Some(Ordering::Equal)); + test_cmp(b.partial_lex_ord(), b.partial_lex_ord(), + Some(Ordering::Equal)); + + // variable values: + // a = [false, false, false, false]; + // b = [false, false, false, true] + let a = $id::splat(false); + let mut b = $id::splat(false); + b = b.replace($id::lanes() - 1, true); + test_cmp(a.partial_lex_ord(), b.partial_lex_ord(), + Some(Ordering::Less)); + test_cmp(b.partial_lex_ord(), a.partial_lex_ord(), + Some(Ordering::Greater)); + test_cmp(a.partial_lex_ord(), a.partial_lex_ord(), + Some(Ordering::Equal)); + test_cmp(b.partial_lex_ord(), b.partial_lex_ord(), + Some(Ordering::Equal)); + + // variable values: + // a = [true, true, true, false]; + // b = [true, true, true, true] + let mut a = $id::splat(true); + let b = $id::splat(true); + a = a.replace($id::lanes() - 1, false); + test_cmp(a.partial_lex_ord(), b.partial_lex_ord(), + Some(Ordering::Less)); + test_cmp(b.partial_lex_ord(), a.partial_lex_ord(), + Some(Ordering::Greater)); + test_cmp(a.partial_lex_ord(), a.partial_lex_ord(), + Some(Ordering::Equal)); + test_cmp(b.partial_lex_ord(), b.partial_lex_ord(), + Some(Ordering::Equal)); + + if $id::lanes() > 2 { + // variable values + // a = [false, true, false, false]; + // b = [false, true, true, true] + let mut a = $id::splat(false); + let mut b = $id::splat(true); + a = a.replace(1, true); + b = b.replace(0, false); + test_cmp(a.partial_lex_ord(), b.partial_lex_ord(), + Some(Ordering::Less)); + test_cmp(b.partial_lex_ord(), a.partial_lex_ord(), + Some(Ordering::Greater)); + test_cmp(a.partial_lex_ord(), a.partial_lex_ord(), + Some(Ordering::Equal)); + test_cmp(b.partial_lex_ord(), b.partial_lex_ord(), + Some(Ordering::Equal)); + } + } + } + } + } + }; +} diff --git a/third_party/rust/packed_simd/src/api/cmp/vertical.rs b/third_party/rust/packed_simd/src/api/cmp/vertical.rs new file mode 100644 index 000000000000..ea4a0d1a3467 --- /dev/null +++ b/third_party/rust/packed_simd/src/api/cmp/vertical.rs @@ -0,0 +1,114 @@ +//! Vertical (lane-wise) vector comparisons returning vector masks. + +macro_rules! impl_cmp_vertical { + ( + [$elem_ty:ident; $elem_count:expr]: + $id:ident, + $mask_ty:ident, + $is_mask:expr,($true:expr, $false:expr) | $test_tt:tt + ) => { + impl $id { + /// Lane-wise equality comparison. + #[inline] + pub fn eq(self, other: Self) -> $mask_ty { + use crate::llvm::simd_eq; + Simd(unsafe { simd_eq(self.0, other.0) }) + } + + /// Lane-wise inequality comparison. + #[inline] + pub fn ne(self, other: Self) -> $mask_ty { + use crate::llvm::simd_ne; + Simd(unsafe { simd_ne(self.0, other.0) }) + } + + /// Lane-wise less-than comparison. + #[inline] + pub fn lt(self, other: Self) -> $mask_ty { + use crate::llvm::{simd_gt, simd_lt}; + if $is_mask { + Simd(unsafe { simd_gt(self.0, other.0) }) + } else { + Simd(unsafe { simd_lt(self.0, other.0) }) + } + } + + /// Lane-wise less-than-or-equals comparison. + #[inline] + pub fn le(self, other: Self) -> $mask_ty { + use crate::llvm::{simd_ge, simd_le}; + if $is_mask { + Simd(unsafe { simd_ge(self.0, other.0) }) + } else { + Simd(unsafe { simd_le(self.0, other.0) }) + } + } + + /// Lane-wise greater-than comparison. + #[inline] + pub fn gt(self, other: Self) -> $mask_ty { + use crate::llvm::{simd_gt, simd_lt}; + if $is_mask { + Simd(unsafe { simd_lt(self.0, other.0) }) + } else { + Simd(unsafe { simd_gt(self.0, other.0) }) + } + } + + /// Lane-wise greater-than-or-equals comparison. + #[inline] + pub fn ge(self, other: Self) -> $mask_ty { + use crate::llvm::{simd_ge, simd_le}; + if $is_mask { + Simd(unsafe { simd_le(self.0, other.0) }) + } else { + Simd(unsafe { simd_ge(self.0, other.0) }) + } + } + } + test_if!{ + $test_tt: + paste::item! { + pub mod [<$id _cmp_vertical>] { + use super::*; + #[cfg_attr(not(target_arch = "wasm32"), test)] #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] + fn cmp() { + let a = $id::splat($false); + let b = $id::splat($true); + + let r = a.lt(b); + let e = $mask_ty::splat(true); + assert!(r == e); + let r = a.le(b); + assert!(r == e); + + let e = $mask_ty::splat(false); + let r = a.gt(b); + assert!(r == e); + let r = a.ge(b); + assert!(r == e); + let r = a.eq(b); + assert!(r == e); + + let mut a = a; + let mut b = b; + let mut e = e; + for i in 0..$id::lanes() { + if i % 2 == 0 { + a = a.replace(i, $false); + b = b.replace(i, $true); + e = e.replace(i, true); + } else { + a = a.replace(i, $true); + b = b.replace(i, $false); + e = e.replace(i, false); + } + } + let r = a.lt(b); + assert!(r == e); + } + } + } + } + }; +} diff --git a/third_party/rust/packed_simd/src/api/default.rs b/third_party/rust/packed_simd/src/api/default.rs new file mode 100644 index 000000000000..7af55ea77a85 --- /dev/null +++ b/third_party/rust/packed_simd/src/api/default.rs @@ -0,0 +1,30 @@ +//! Implements `Default` for vector types. + +macro_rules! impl_default { + ([$elem_ty:ident; $elem_count:expr]: $id:ident | $test_tt:tt) => { + impl Default for $id { + #[inline] + fn default() -> Self { + Self::splat($elem_ty::default()) + } + } + + test_if!{ + $test_tt: + paste::item! { + // Comparisons use integer casts within mantissa^1 range. + #[allow(clippy::float_cmp)] + pub mod [<$id _default>] { + use super::*; + #[cfg_attr(not(target_arch = "wasm32"), test)] #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] + fn default() { + let a = $id::default(); + for i in 0..$id::lanes() { + assert_eq!(a.extract(i), $elem_ty::default()); + } + } + } + } + } + }; +} diff --git a/third_party/rust/packed_simd/src/api/fmt.rs b/third_party/rust/packed_simd/src/api/fmt.rs new file mode 100644 index 000000000000..f3f55c401548 --- /dev/null +++ b/third_party/rust/packed_simd/src/api/fmt.rs @@ -0,0 +1,12 @@ +//! Implements formatting APIs + +#[macro_use] +mod debug; +#[macro_use] +mod lower_hex; +#[macro_use] +mod upper_hex; +#[macro_use] +mod octal; +#[macro_use] +mod binary; diff --git a/third_party/rust/packed_simd/src/api/fmt/binary.rs b/third_party/rust/packed_simd/src/api/fmt/binary.rs new file mode 100644 index 000000000000..91c082555908 --- /dev/null +++ b/third_party/rust/packed_simd/src/api/fmt/binary.rs @@ -0,0 +1,54 @@ +//! Implement Octal formatting + +macro_rules! impl_fmt_binary { + ([$elem_ty:ident; $elem_count:expr]: $id:ident | $test_tt:tt) => { + impl crate::fmt::Binary for $id { + #[allow(clippy::missing_inline_in_public_items)] + fn fmt(&self, f: &mut crate::fmt::Formatter<'_>) -> crate::fmt::Result { + write!(f, "{}(", stringify!($id))?; + for i in 0..$elem_count { + if i > 0 { + write!(f, ", ")?; + } + self.extract(i).fmt(f)?; + } + write!(f, ")") + } + } + test_if! { + $test_tt: + paste::item! { + pub mod [<$id _fmt_binary>] { + use super::*; + #[cfg_attr(not(target_arch = "wasm32"), test)] + #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] + fn binary() { + use arrayvec::{ArrayString,ArrayVec}; + type TinyString = ArrayString<[u8; 512]>; + + use crate::fmt::Write; + let v = $id::splat($elem_ty::default()); + let mut s = TinyString::new(); + write!(&mut s, "{:#b}", v).unwrap(); + + let mut beg = TinyString::new(); + write!(&mut beg, "{}(", stringify!($id)).unwrap(); + assert!(s.starts_with(beg.as_str())); + assert!(s.ends_with(")")); + let s: ArrayVec<[TinyString; 64]> + = s.replace(beg.as_str(), "") + .replace(")", "").split(",") + .map(|v| TinyString::from(v.trim()).unwrap()) + .collect(); + assert_eq!(s.len(), $id::lanes()); + for (index, ss) in s.into_iter().enumerate() { + let mut e = TinyString::new(); + write!(&mut e, "{:#b}", v.extract(index)).unwrap(); + assert_eq!(ss, e); + } + } + } + } + } + }; +} diff --git a/third_party/rust/packed_simd/src/api/fmt/debug.rs b/third_party/rust/packed_simd/src/api/fmt/debug.rs new file mode 100644 index 000000000000..1e209b3bff77 --- /dev/null +++ b/third_party/rust/packed_simd/src/api/fmt/debug.rs @@ -0,0 +1,60 @@ +//! Implement debug formatting + +macro_rules! impl_fmt_debug_tests { + ([$elem_ty:ty; $elem_count:expr]: $id:ident | $test_tt:tt) => { + test_if! { + $test_tt: + paste::item! { + pub mod [<$id _fmt_debug>] { + use super::*; + #[cfg_attr(not(target_arch = "wasm32"), test)] + #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] + fn debug() { + use arrayvec::{ArrayString,ArrayVec}; + type TinyString = ArrayString<[u8; 512]>; + + use crate::fmt::Write; + let v = $id::default(); + let mut s = TinyString::new(); + write!(&mut s, "{:?}", v).unwrap(); + + let mut beg = TinyString::new(); + write!(&mut beg, "{}(", stringify!($id)).unwrap(); + assert!(s.starts_with(beg.as_str())); + assert!(s.ends_with(")")); + let s: ArrayVec<[TinyString; 64]> + = s.replace(beg.as_str(), "") + .replace(")", "").split(",") + .map(|v| TinyString::from(v.trim()).unwrap()) + .collect(); + assert_eq!(s.len(), $id::lanes()); + for (index, ss) in s.into_iter().enumerate() { + let mut e = TinyString::new(); + write!(&mut e, "{:?}", v.extract(index)).unwrap(); + assert_eq!(ss, e); + } + } + } + } + } + }; +} + +macro_rules! impl_fmt_debug { + ([$elem_ty:ty; $elem_count:expr]: $id:ident | $test_tt:tt) => { + impl crate::fmt::Debug for $id { + #[allow(clippy::missing_inline_in_public_items)] + fn fmt(&self, f: &mut crate::fmt::Formatter<'_>) -> crate::fmt::Result { + write!(f, "{}(", stringify!($id))?; + for i in 0..$elem_count { + if i > 0 { + write!(f, ", ")?; + } + self.extract(i).fmt(f)?; + } + write!(f, ")") + } + } + impl_fmt_debug_tests!([$elem_ty; $elem_count]: $id | $test_tt); + }; +} diff --git a/third_party/rust/packed_simd/src/api/fmt/lower_hex.rs b/third_party/rust/packed_simd/src/api/fmt/lower_hex.rs new file mode 100644 index 000000000000..8f11d3119b70 --- /dev/null +++ b/third_party/rust/packed_simd/src/api/fmt/lower_hex.rs @@ -0,0 +1,54 @@ +//! Implement `LowerHex` formatting + +macro_rules! impl_fmt_lower_hex { + ([$elem_ty:ident; $elem_count:expr]: $id:ident | $test_tt:tt) => { + impl crate::fmt::LowerHex for $id { + #[allow(clippy::missing_inline_in_public_items)] + fn fmt(&self, f: &mut crate::fmt::Formatter<'_>) -> crate::fmt::Result { + write!(f, "{}(", stringify!($id))?; + for i in 0..$elem_count { + if i > 0 { + write!(f, ", ")?; + } + self.extract(i).fmt(f)?; + } + write!(f, ")") + } + } + test_if! { + $test_tt: + paste::item! { + pub mod [<$id _fmt_lower_hex>] { + use super::*; + #[cfg_attr(not(target_arch = "wasm32"), test)] + #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] + fn lower_hex() { + use arrayvec::{ArrayString,ArrayVec}; + type TinyString = ArrayString<[u8; 512]>; + + use crate::fmt::Write; + let v = $id::splat($elem_ty::default()); + let mut s = TinyString::new(); + write!(&mut s, "{:#x}", v).unwrap(); + + let mut beg = TinyString::new(); + write!(&mut beg, "{}(", stringify!($id)).unwrap(); + assert!(s.starts_with(beg.as_str())); + assert!(s.ends_with(")")); + let s: ArrayVec<[TinyString; 64]> + = s.replace(beg.as_str(), "").replace(")", "") + .split(",") + .map(|v| TinyString::from(v.trim()).unwrap()) + .collect(); + assert_eq!(s.len(), $id::lanes()); + for (index, ss) in s.into_iter().enumerate() { + let mut e = TinyString::new(); + write!(&mut e, "{:#x}", v.extract(index)).unwrap(); + assert_eq!(ss, e); + } + } + } + } + } + }; +} diff --git a/third_party/rust/packed_simd/src/api/fmt/octal.rs b/third_party/rust/packed_simd/src/api/fmt/octal.rs new file mode 100644 index 000000000000..e708e094cee0 --- /dev/null +++ b/third_party/rust/packed_simd/src/api/fmt/octal.rs @@ -0,0 +1,54 @@ +//! Implement Octal formatting + +macro_rules! impl_fmt_octal { + ([$elem_ty:ident; $elem_count:expr]: $id:ident | $test_tt:tt) => { + impl crate::fmt::Octal for $id { + #[allow(clippy::missing_inline_in_public_items)] + fn fmt(&self, f: &mut crate::fmt::Formatter<'_>) -> crate::fmt::Result { + write!(f, "{}(", stringify!($id))?; + for i in 0..$elem_count { + if i > 0 { + write!(f, ", ")?; + } + self.extract(i).fmt(f)?; + } + write!(f, ")") + } + } + test_if! { + $test_tt: + paste::item! { + pub mod [<$id _fmt_octal>] { + use super::*; + #[cfg_attr(not(target_arch = "wasm32"), test)] + #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] + fn octal_hex() { + use arrayvec::{ArrayString,ArrayVec}; + type TinyString = ArrayString<[u8; 512]>; + + use crate::fmt::Write; + let v = $id::splat($elem_ty::default()); + let mut s = TinyString::new(); + write!(&mut s, "{:#o}", v).unwrap(); + + let mut beg = TinyString::new(); + write!(&mut beg, "{}(", stringify!($id)).unwrap(); + assert!(s.starts_with(beg.as_str())); + assert!(s.ends_with(")")); + let s: ArrayVec<[TinyString; 64]> + = s.replace(beg.as_str(), "").replace(")", "") + .split(",") + .map(|v| TinyString::from(v.trim()).unwrap()) + .collect(); + assert_eq!(s.len(), $id::lanes()); + for (index, ss) in s.into_iter().enumerate() { + let mut e = TinyString::new(); + write!(&mut e, "{:#o}", v.extract(index)).unwrap(); + assert_eq!(ss, e); + } + } + } + } + } + }; +} diff --git a/third_party/rust/packed_simd/src/api/fmt/upper_hex.rs b/third_party/rust/packed_simd/src/api/fmt/upper_hex.rs new file mode 100644 index 000000000000..5ad455706baa --- /dev/null +++ b/third_party/rust/packed_simd/src/api/fmt/upper_hex.rs @@ -0,0 +1,54 @@ +//! Implement `UpperHex` formatting + +macro_rules! impl_fmt_upper_hex { + ([$elem_ty:ident; $elem_count:expr]: $id:ident | $test_tt:tt) => { + impl crate::fmt::UpperHex for $id { + #[allow(clippy::missing_inline_in_public_items)] + fn fmt(&self, f: &mut crate::fmt::Formatter<'_>) -> crate::fmt::Result { + write!(f, "{}(", stringify!($id))?; + for i in 0..$elem_count { + if i > 0 { + write!(f, ", ")?; + } + self.extract(i).fmt(f)?; + } + write!(f, ")") + } + } + test_if! { + $test_tt: + paste::item! { + pub mod [<$id _fmt_upper_hex>] { + use super::*; + #[cfg_attr(not(target_arch = "wasm32"), test)] + #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] + fn upper_hex() { + use arrayvec::{ArrayString,ArrayVec}; + type TinyString = ArrayString<[u8; 512]>; + + use crate::fmt::Write; + let v = $id::splat($elem_ty::default()); + let mut s = TinyString::new(); + write!(&mut s, "{:#X}", v).unwrap(); + + let mut beg = TinyString::new(); + write!(&mut beg, "{}(", stringify!($id)).unwrap(); + assert!(s.starts_with(beg.as_str())); + assert!(s.ends_with(")")); + let s: ArrayVec<[TinyString; 64]> + = s.replace(beg.as_str(), "").replace(")", "") + .split(",") + .map(|v| TinyString::from(v.trim()).unwrap()) + .collect(); + assert_eq!(s.len(), $id::lanes()); + for (index, ss) in s.into_iter().enumerate() { + let mut e = TinyString::new(); + write!(&mut e, "{:#X}", v.extract(index)).unwrap(); + assert_eq!(ss, e); + } + } + } + } + } + }; +} diff --git a/third_party/rust/packed_simd/src/api/from.rs b/third_party/rust/packed_simd/src/api/from.rs new file mode 100644 index 000000000000..c30c4d6e216d --- /dev/null +++ b/third_party/rust/packed_simd/src/api/from.rs @@ -0,0 +1,7 @@ +//! Implementations of the `From` and `Into` traits + +#[macro_use] +mod from_array; + +#[macro_use] +mod from_vector; diff --git a/third_party/rust/packed_simd/src/api/from/from_array.rs b/third_party/rust/packed_simd/src/api/from/from_array.rs new file mode 100644 index 000000000000..5c7801ddaf09 --- /dev/null +++ b/third_party/rust/packed_simd/src/api/from/from_array.rs @@ -0,0 +1,124 @@ +//! Implements `From<[T; N]>` and `Into<[T; N]>` for vector types. + +macro_rules! impl_from_array { + ([$elem_ty:ident; $elem_count:expr]: $id:ident | $test_tt:tt + | ($non_default_array:expr, $non_default_vec:expr)) => { + impl From<[$elem_ty; $elem_count]> for $id { + #[inline] + fn from(array: [$elem_ty; $elem_count]) -> Self { + union U { + array: [$elem_ty; $elem_count], + vec: $id, + } + unsafe { U { array }.vec } + } + } + + impl From<$id> for [$elem_ty; $elem_count] { + #[inline] + fn from(vec: $id) -> Self { + union U { + array: [$elem_ty; $elem_count], + vec: $id, + } + unsafe { U { vec }.array } + } + } + + // FIXME: `Into::into` is not inline, but due to + // the blanket impl in `std`, which is not + // marked `default`, we cannot override it here with + // specialization. + /* + impl Into<[$elem_ty; $elem_count]> for $id { + #[inline] + fn into(self) -> [$elem_ty; $elem_count] { + union U { + array: [$elem_ty; $elem_count], + vec: $id, + } + unsafe { U { vec: self }.array } + } + } + + impl Into<$id> for [$elem_ty; $elem_count] { + #[inline] + fn into(self) -> $id { + union U { + array: [$elem_ty; $elem_count], + vec: $id, + } + unsafe { U { array: self }.vec } + } + } + */ + + test_if! { + $test_tt: + paste::item! { + // Comparisons use integer casts within mantissa^1 range. + #[allow(clippy::float_cmp)] + mod [<$id _from>] { + use super::*; + #[test] + #[cfg_attr(miri, ignore)] + fn array() { + let vec: $id = Default::default(); + + // FIXME: Workaround for arrays with more than 32 + // elements. + // + // Safe because we never take a reference to any + // uninitialized element. + union W { + array: [$elem_ty; $elem_count], + other: () + } + let mut array = W { other: () }; + for i in 0..$elem_count { + let default: $elem_ty = Default::default(); + // note: array.other is the active member and + // initialized so we can take a reference to it: + let p = unsafe { + &mut array.other as *mut () as *mut $elem_ty + }; + // note: default is a valid bit-pattern for + // $elem_ty: + unsafe { + crate::ptr::write(p.wrapping_add(i), default) + }; + } + // note: the array variant of the union is properly + // initialized: + let mut array = unsafe { + array.array + }; + + array[0] = $non_default_array; + let vec = vec.replace(0, $non_default_vec); + + let vec_from_array = $id::from(array); + assert_eq!(vec_from_array, vec); + let array_from_vec + = <[$elem_ty; $elem_count]>::from(vec); + // FIXME: Workaround for arrays with more than 32 + // elements. + for i in 0..$elem_count { + assert_eq!(array_from_vec[i], array[i]); + } + + let vec_from_into_array: $id = array.into(); + assert_eq!(vec_from_into_array, vec); + let array_from_into_vec: [$elem_ty; $elem_count] + = vec.into(); + // FIXME: Workaround for arrays with more than 32 + // elements. + for i in 0..$elem_count { + assert_eq!(array_from_into_vec[i], array[i]); + } + } + } + } + } + }; +} diff --git a/third_party/rust/packed_simd/src/api/from/from_vector.rs b/third_party/rust/packed_simd/src/api/from/from_vector.rs new file mode 100644 index 000000000000..55f70016d51d --- /dev/null +++ b/third_party/rust/packed_simd/src/api/from/from_vector.rs @@ -0,0 +1,67 @@ +//! Implements `From` and `Into` for vector types. + +macro_rules! impl_from_vector { + ([$elem_ty:ident; $elem_count:expr]: $id:ident | $test_tt:tt + | $source:ident) => { + impl From<$source> for $id { + #[inline] + fn from(source: $source) -> Self { + fn static_assert_same_number_of_lanes() + where + T: crate::sealed::Simd, + U: crate::sealed::Simd, + { + } + use crate::llvm::simd_cast; + static_assert_same_number_of_lanes::<$id, $source>(); + Simd(unsafe { simd_cast(source.0) }) + } + } + + // FIXME: `Into::into` is not inline, but due to the blanket impl in + // `std`, which is not marked `default`, we cannot override it here + // with specialization. + + /* + impl Into<$id> for $source { + #[inline] + fn into(self) -> $id { + unsafe { simd_cast(self) } + } + } + */ + + test_if! { + $test_tt: + paste::item! { + pub mod [<$id _from_ $source>] { + use super::*; + #[cfg_attr(not(target_arch = "wasm32"), test)] + #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] + fn from() { + assert_eq!($id::lanes(), $source::lanes()); + let source: $source = Default::default(); + let vec: $id = Default::default(); + + let e = $id::from(source); + assert_eq!(e, vec); + + let e: $id = source.into(); + assert_eq!(e, vec); + } + } + } + } + }; +} + +macro_rules! impl_from_vectors { + ([$elem_ty:ident; $elem_count:expr]: $id:ident | $test_tt:tt + | $($source:ident),*) => { + $( + impl_from_vector!( + [$elem_ty; $elem_count]: $id | $test_tt | $source + ); + )* + } +} diff --git a/third_party/rust/packed_simd/src/api/hash.rs b/third_party/rust/packed_simd/src/api/hash.rs new file mode 100644 index 000000000000..ee80eff939c3 --- /dev/null +++ b/third_party/rust/packed_simd/src/api/hash.rs @@ -0,0 +1,49 @@ +//! Implements `Hash` for vector types. + +macro_rules! impl_hash { + ([$elem_ty:ident; $elem_count:expr]: $id:ident | $test_tt:tt) => { + impl crate::hash::Hash for $id { + #[inline] + fn hash(&self, state: &mut H) { + unsafe { + union A { + data: [$elem_ty; $id::lanes()], + vec: $id, + } + A { vec: *self }.data.hash(state) + } + } + } + + test_if! { + $test_tt: + paste::item! { + pub mod [<$id _hash>] { + use super::*; + #[cfg_attr(not(target_arch = "wasm32"), test)] #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] + fn hash() { + use crate::hash::{Hash, Hasher}; + #[allow(deprecated)] + use crate::hash::{SipHasher13}; + type A = [$elem_ty; $id::lanes()]; + let a: A = [42 as $elem_ty; $id::lanes()]; + assert_eq!( + crate::mem::size_of::(), + crate::mem::size_of::<$id>() + ); + #[allow(deprecated)] + let mut a_hash = SipHasher13::new(); + let mut v_hash = a_hash.clone(); + a.hash(&mut a_hash); + + // Integer within mantissa^1 range. + #[allow(clippy::float_cmp)] + let v = $id::splat(42 as $elem_ty); + v.hash(&mut v_hash); + assert_eq!(a_hash.finish(), v_hash.finish()); + } + } + } + } + }; +} diff --git a/third_party/rust/packed_simd/src/api/into_bits.rs b/third_party/rust/packed_simd/src/api/into_bits.rs new file mode 100644 index 000000000000..03fbe4bff78f --- /dev/null +++ b/third_party/rust/packed_simd/src/api/into_bits.rs @@ -0,0 +1,59 @@ +//! Implementation of `FromBits` and `IntoBits`. + +/// Safe lossless bitwise conversion from `T` to `Self`. +#[cfg_attr(doc_cfg, doc(cfg(feature = "into_bits")))] +pub trait FromBits: crate::marker::Sized { + /// Safe lossless bitwise transmute from `T` to `Self`. + fn from_bits(t: T) -> Self; +} + +/// Safe lossless bitwise conversion from `Self` to `T`. +#[cfg_attr(doc_cfg, doc(cfg(feature = "into_bits")))] +pub trait IntoBits: crate::marker::Sized { + /// Safe lossless bitwise transmute from `self` to `T`. + fn into_bits(self) -> T; +} + +/// `FromBits` implies `IntoBits`. +impl IntoBits for T +where + U: FromBits, +{ + #[inline] + fn into_bits(self) -> U { + debug_assert!(crate::mem::size_of::() == crate::mem::size_of::()); + U::from_bits(self) + } +} + +/// `FromBits` and `IntoBits` are reflexive +impl FromBits for T { + #[inline] + fn from_bits(t: Self) -> Self { + t + } +} + +#[macro_use] +mod macros; + +mod v16; +pub use self::v16::*; + +mod v32; +pub use self::v32::*; + +mod v64; +pub use self::v64::*; + +mod v128; +pub use self::v128::*; + +mod v256; +pub use self::v256::*; + +mod v512; +pub use self::v512::*; + +mod arch_specific; +pub use self::arch_specific::*; diff --git a/third_party/rust/packed_simd/src/api/into_bits/arch_specific.rs b/third_party/rust/packed_simd/src/api/into_bits/arch_specific.rs new file mode 100644 index 000000000000..bfac915576c3 --- /dev/null +++ b/third_party/rust/packed_simd/src/api/into_bits/arch_specific.rs @@ -0,0 +1,345 @@ +//! `FromBits` and `IntoBits` between portable vector types and the +//! architecture-specific vector types. +#[rustfmt::skip] + +// FIXME: MIPS FromBits/IntoBits + +#[allow(unused)] +use crate::*; + +/// This macro implements FromBits for the portable and the architecture +/// specific vector types. +/// +/// The "leaf" case is at the bottom, and the most generic case is at the top. +/// The generic case is split into smaller cases recursively. +macro_rules! impl_arch { + ([$arch_head_i:ident[$arch_head_tt:tt]: $($arch_head_ty:ident),*], + $([$arch_tail_i:ident[$arch_tail_tt:tt]: $($arch_tail_ty:ident),*]),* | + from: $($from_ty:ident),* | into: $($into_ty:ident),* | + test: $test_tt:tt) => { + impl_arch!( + [$arch_head_i[$arch_head_tt]: $($arch_head_ty),*] | + from: $($from_ty),* | + into: $($into_ty),* | + test: $test_tt + ); + impl_arch!( + $([$arch_tail_i[$arch_tail_tt]: $($arch_tail_ty),*]),* | + from: $($from_ty),* | + into: $($into_ty),* | + test: $test_tt + ); + }; + ([$arch:ident[$arch_tt:tt]: $($arch_ty:ident),*] | + from: $($from_ty:ident),* | into: $($into_ty:ident),* | + test: $test_tt:tt) => { + // note: if target is "arm", "+v7,+neon" must be enabled + // and the std library must be recompiled with them + #[cfg(any( + not(target_arch = "arm"), + all(target_feature = "v7", target_feature = "neon", + any(feature = "core_arch", libcore_neon))) + )] + // note: if target is "powerpc", "altivec" must be enabled + // and the std library must be recompiled with it + #[cfg(any( + not(target_arch = "powerpc"), + all(target_feature = "altivec", feature = "core_arch"), + ))] + #[cfg(target_arch = $arch_tt)] + use crate::arch::$arch::{ + $($arch_ty),* + }; + + #[cfg(any( + not(target_arch = "arm"), + all(target_feature = "v7", target_feature = "neon", + any(feature = "core_arch", libcore_neon))) + )] + #[cfg(any( + not(target_arch = "powerpc"), + all(target_feature = "altivec", feature = "core_arch"), + ))] + #[cfg(target_arch = $arch_tt)] + impl_arch!($($arch_ty),* | $($from_ty),* | $($into_ty),* | + test: $test_tt); + }; + ($arch_head:ident, $($arch_tail:ident),* | $($from_ty:ident),* + | $($into_ty:ident),* | test: $test_tt:tt) => { + impl_arch!($arch_head | $($from_ty),* | $($into_ty),* | + test: $test_tt); + impl_arch!($($arch_tail),* | $($from_ty),* | $($into_ty),* | + test: $test_tt); + }; + ($arch_head:ident | $($from_ty:ident),* | $($into_ty:ident),* | + test: $test_tt:tt) => { + impl_from_bits!($arch_head[$test_tt]: $($from_ty),*); + impl_into_bits!($arch_head[$test_tt]: $($into_ty),*); + }; +} + +//////////////////////////////////////////////////////////////////////////////// +// Implementations for the 64-bit wide vector types: + +// FIXME: 64-bit single element types +// FIXME: arm/aarch float16x4_t missing +impl_arch!( + [ + arm["arm"]: int8x8_t, + uint8x8_t, + poly8x8_t, + int16x4_t, + uint16x4_t, + poly16x4_t, + int32x2_t, + uint32x2_t, + float32x2_t, + int64x1_t, + uint64x1_t + ], + [ + aarch64["aarch64"]: int8x8_t, + uint8x8_t, + poly8x8_t, + int16x4_t, + uint16x4_t, + poly16x4_t, + int32x2_t, + uint32x2_t, + float32x2_t, + int64x1_t, + uint64x1_t, + float64x1_t + ] | from: i8x8, + u8x8, + m8x8, + i16x4, + u16x4, + m16x4, + i32x2, + u32x2, + f32x2, + m32x2 | into: i8x8, + u8x8, + i16x4, + u16x4, + i32x2, + u32x2, + f32x2 | test: test_v64 +); + +//////////////////////////////////////////////////////////////////////////////// +// Implementations for the 128-bit wide vector types: + +// FIXME: arm/aarch float16x8_t missing +// FIXME: ppc vector_pixel missing +// FIXME: ppc64 vector_Float16 missing +// FIXME: ppc64 vector_signed_long_long missing +// FIXME: ppc64 vector_unsigned_long_long missing +// FIXME: ppc64 vector_bool_long_long missing +// FIXME: ppc64 vector_signed___int128 missing +// FIXME: ppc64 vector_unsigned___int128 missing +impl_arch!( + [x86["x86"]: __m128, __m128i, __m128d], + [x86_64["x86_64"]: __m128, __m128i, __m128d], + [ + arm["arm"]: int8x16_t, + uint8x16_t, + poly8x16_t, + int16x8_t, + uint16x8_t, + poly16x8_t, + int32x4_t, + uint32x4_t, + float32x4_t, + int64x2_t, + uint64x2_t + ], + [ + aarch64["aarch64"]: int8x16_t, + uint8x16_t, + poly8x16_t, + int16x8_t, + uint16x8_t, + poly16x8_t, + int32x4_t, + uint32x4_t, + float32x4_t, + int64x2_t, + uint64x2_t, + float64x2_t + ], + [ + powerpc["powerpc"]: vector_signed_char, + vector_unsigned_char, + vector_signed_short, + vector_unsigned_short, + vector_signed_int, + vector_unsigned_int, + vector_float + ], + [ + powerpc64["powerpc64"]: vector_signed_char, + vector_unsigned_char, + vector_signed_short, + vector_unsigned_short, + vector_signed_int, + vector_unsigned_int, + vector_float, + vector_signed_long, + vector_unsigned_long, + vector_double + ] | from: i8x16, + u8x16, + m8x16, + i16x8, + u16x8, + m16x8, + i32x4, + u32x4, + f32x4, + m32x4, + i64x2, + u64x2, + f64x2, + m64x2, + i128x1, + u128x1, + m128x1 | into: i8x16, + u8x16, + i16x8, + u16x8, + i32x4, + u32x4, + f32x4, + i64x2, + u64x2, + f64x2, + i128x1, + u128x1 | test: test_v128 +); + +impl_arch!( + [powerpc["powerpc"]: vector_bool_char], + [powerpc64["powerpc64"]: vector_bool_char] | from: m8x16, + m16x8, + m32x4, + m64x2, + m128x1 | into: i8x16, + u8x16, + i16x8, + u16x8, + i32x4, + u32x4, + f32x4, + i64x2, + u64x2, + f64x2, + i128x1, + u128x1, + // Masks: + m8x16 | test: test_v128 +); + +impl_arch!( + [powerpc["powerpc"]: vector_bool_short], + [powerpc64["powerpc64"]: vector_bool_short] | from: m16x8, + m32x4, + m64x2, + m128x1 | into: i8x16, + u8x16, + i16x8, + u16x8, + i32x4, + u32x4, + f32x4, + i64x2, + u64x2, + f64x2, + i128x1, + u128x1, + // Masks: + m8x16, + m16x8 | test: test_v128 +); + +impl_arch!( + [powerpc["powerpc"]: vector_bool_int], + [powerpc64["powerpc64"]: vector_bool_int] | from: m32x4, + m64x2, + m128x1 | into: i8x16, + u8x16, + i16x8, + u16x8, + i32x4, + u32x4, + f32x4, + i64x2, + u64x2, + f64x2, + i128x1, + u128x1, + // Masks: + m8x16, + m16x8, + m32x4 | test: test_v128 +); + +impl_arch!( + [powerpc64["powerpc64"]: vector_bool_long] | from: m64x2, + m128x1 | into: i8x16, + u8x16, + i16x8, + u16x8, + i32x4, + u32x4, + f32x4, + i64x2, + u64x2, + f64x2, + i128x1, + u128x1, + // Masks: + m8x16, + m16x8, + m32x4, + m64x2 | test: test_v128 +); + +//////////////////////////////////////////////////////////////////////////////// +// Implementations for the 256-bit wide vector types + +impl_arch!( + [x86["x86"]: __m256, __m256i, __m256d], + [x86_64["x86_64"]: __m256, __m256i, __m256d] | from: i8x32, + u8x32, + m8x32, + i16x16, + u16x16, + m16x16, + i32x8, + u32x8, + f32x8, + m32x8, + i64x4, + u64x4, + f64x4, + m64x4, + i128x2, + u128x2, + m128x2 | into: i8x32, + u8x32, + i16x16, + u16x16, + i32x8, + u32x8, + f32x8, + i64x4, + u64x4, + f64x4, + i128x2, + u128x2 | test: test_v256 +); + +//////////////////////////////////////////////////////////////////////////////// +// FIXME: Implementations for the 512-bit wide vector types diff --git a/third_party/rust/packed_simd/src/api/into_bits/macros.rs b/third_party/rust/packed_simd/src/api/into_bits/macros.rs new file mode 100644 index 000000000000..265ab34ae0f3 --- /dev/null +++ b/third_party/rust/packed_simd/src/api/into_bits/macros.rs @@ -0,0 +1,74 @@ +//! Macros implementing `FromBits` + +macro_rules! impl_from_bits_ { + ($id:ident[$test_tt:tt]: $from_ty:ident) => { + impl crate::api::into_bits::FromBits<$from_ty> for $id { + #[inline] + fn from_bits(x: $from_ty) -> Self { + unsafe { crate::mem::transmute(x) } + } + } + + test_if! { + $test_tt: + paste::item! { + pub mod [<$id _from_bits_ $from_ty>] { + use super::*; + #[cfg_attr(not(target_arch = "wasm32"), test)] + #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] + fn test() { + use crate::{ + ptr::{read_unaligned}, + mem::{size_of, zeroed} + }; + use crate::IntoBits; + assert_eq!(size_of::<$id>(), + size_of::<$from_ty>()); + // This is safe because we never create a reference to + // uninitialized memory: + let a: $from_ty = unsafe { zeroed() }; + + let b_0: $id = crate::FromBits::from_bits(a); + let b_1: $id = a.into_bits(); + + // Check that these are byte-wise equal, that is, + // that the bit patterns are identical: + for i in 0..size_of::<$id>() { + // This is safe because we only read initialized + // memory in bounds. Also, taking a reference to + // `b_i` is ok because the fields are initialized. + unsafe { + let b_0_v: u8 = read_unaligned( + (&b_0 as *const $id as *const u8) + .wrapping_add(i) + ); + let b_1_v: u8 = read_unaligned( + (&b_1 as *const $id as *const u8) + .wrapping_add(i) + ); + assert_eq!(b_0_v, b_1_v); + } + } + } + } + } + } + }; +} + +macro_rules! impl_from_bits { + ($id:ident[$test_tt:tt]: $($from_ty:ident),*) => { + $( + impl_from_bits_!($id[$test_tt]: $from_ty); + )* + } +} + +#[allow(unused)] +macro_rules! impl_into_bits { + ($id:ident[$test_tt:tt]: $($from_ty:ident),*) => { + $( + impl_from_bits_!($from_ty[$test_tt]: $id); + )* + } +} diff --git a/third_party/rust/packed_simd/src/api/into_bits/v128.rs b/third_party/rust/packed_simd/src/api/into_bits/v128.rs new file mode 100644 index 000000000000..639c09c2c4bd --- /dev/null +++ b/third_party/rust/packed_simd/src/api/into_bits/v128.rs @@ -0,0 +1,232 @@ +//! `FromBits` and `IntoBits` implementations for portable 128-bit wide vectors +#[rustfmt::skip] + +#[allow(unused)] // wasm_bindgen_test +use crate::*; + +impl_from_bits!( + i8x16[test_v128]: u8x16, + m8x16, + i16x8, + u16x8, + m16x8, + i32x4, + u32x4, + f32x4, + m32x4, + i64x2, + u64x2, + f64x2, + m64x2, + i128x1, + u128x1, + m128x1 +); +impl_from_bits!( + u8x16[test_v128]: i8x16, + m8x16, + i16x8, + u16x8, + m16x8, + i32x4, + u32x4, + f32x4, + m32x4, + i64x2, + u64x2, + f64x2, + m64x2, + i128x1, + u128x1, + m128x1 +); +impl_from_bits!(m8x16[test_v128]: m16x8, m32x4, m64x2, m128x1); + +impl_from_bits!( + i16x8[test_v128]: i8x16, + u8x16, + m8x16, + u16x8, + m16x8, + i32x4, + u32x4, + f32x4, + m32x4, + i64x2, + u64x2, + f64x2, + m64x2, + i128x1, + u128x1, + m128x1 +); +impl_from_bits!( + u16x8[test_v128]: i8x16, + u8x16, + m8x16, + i16x8, + m16x8, + i32x4, + u32x4, + f32x4, + m32x4, + i64x2, + u64x2, + f64x2, + m64x2, + i128x1, + u128x1, + m128x1 +); +impl_from_bits!(m16x8[test_v128]: m32x4, m64x2, m128x1); + +impl_from_bits!( + i32x4[test_v128]: i8x16, + u8x16, + m8x16, + i16x8, + u16x8, + m16x8, + u32x4, + f32x4, + m32x4, + i64x2, + u64x2, + f64x2, + m64x2, + i128x1, + u128x1, + m128x1 +); +impl_from_bits!( + u32x4[test_v128]: i8x16, + u8x16, + m8x16, + i16x8, + u16x8, + m16x8, + i32x4, + f32x4, + m32x4, + i64x2, + u64x2, + f64x2, + m64x2, + i128x1, + u128x1, + m128x1 +); +impl_from_bits!( + f32x4[test_v128]: i8x16, + u8x16, + m8x16, + i16x8, + u16x8, + m16x8, + i32x4, + u32x4, + m32x4, + i64x2, + u64x2, + f64x2, + m64x2, + i128x1, + u128x1, + m128x1 +); +impl_from_bits!(m32x4[test_v128]: m64x2, m128x1); + +impl_from_bits!( + i64x2[test_v128]: i8x16, + u8x16, + m8x16, + i16x8, + u16x8, + m16x8, + i32x4, + u32x4, + f32x4, + m32x4, + u64x2, + f64x2, + m64x2, + i128x1, + u128x1, + m128x1 +); +impl_from_bits!( + u64x2[test_v128]: i8x16, + u8x16, + m8x16, + i16x8, + u16x8, + m16x8, + i32x4, + u32x4, + f32x4, + m32x4, + i64x2, + f64x2, + m64x2, + i128x1, + u128x1, + m128x1 +); +impl_from_bits!( + f64x2[test_v128]: i8x16, + u8x16, + m8x16, + i16x8, + u16x8, + m16x8, + i32x4, + u32x4, + f32x4, + m32x4, + i64x2, + u64x2, + m64x2, + i128x1, + u128x1, + m128x1 +); +impl_from_bits!(m64x2[test_v128]: m128x1); + +impl_from_bits!( + i128x1[test_v128]: i8x16, + u8x16, + m8x16, + i16x8, + u16x8, + m16x8, + i32x4, + u32x4, + f32x4, + m32x4, + i64x2, + u64x2, + f64x2, + m64x2, + u128x1, + m128x1 +); +impl_from_bits!( + u128x1[test_v128]: i8x16, + u8x16, + m8x16, + i16x8, + u16x8, + m16x8, + i32x4, + u32x4, + f32x4, + m32x4, + i64x2, + u64x2, + f64x2, + m64x2, + i128x1, + m128x1 +); +// note: m128x1 cannot be constructed from all the other masks bit patterns in +// here diff --git a/third_party/rust/packed_simd/src/api/into_bits/v16.rs b/third_party/rust/packed_simd/src/api/into_bits/v16.rs new file mode 100644 index 000000000000..e44d0e7f9a18 --- /dev/null +++ b/third_party/rust/packed_simd/src/api/into_bits/v16.rs @@ -0,0 +1,9 @@ +//! `FromBits` and `IntoBits` implementations for portable 16-bit wide vectors +#[rustfmt::skip] + +#[allow(unused)] // wasm_bindgen_test +use crate::*; + +impl_from_bits!(i8x2[test_v16]: u8x2, m8x2); +impl_from_bits!(u8x2[test_v16]: i8x2, m8x2); +// note: m8x2 cannot be constructed from all i8x2 or u8x2 bit patterns diff --git a/third_party/rust/packed_simd/src/api/into_bits/v256.rs b/third_party/rust/packed_simd/src/api/into_bits/v256.rs new file mode 100644 index 000000000000..e432bbbc9f05 --- /dev/null +++ b/third_party/rust/packed_simd/src/api/into_bits/v256.rs @@ -0,0 +1,232 @@ +//! `FromBits` and `IntoBits` implementations for portable 256-bit wide vectors +#[rustfmt::skip] + +#[allow(unused)] // wasm_bindgen_test +use crate::*; + +impl_from_bits!( + i8x32[test_v256]: u8x32, + m8x32, + i16x16, + u16x16, + m16x16, + i32x8, + u32x8, + f32x8, + m32x8, + i64x4, + u64x4, + f64x4, + m64x4, + i128x2, + u128x2, + m128x2 +); +impl_from_bits!( + u8x32[test_v256]: i8x32, + m8x32, + i16x16, + u16x16, + m16x16, + i32x8, + u32x8, + f32x8, + m32x8, + i64x4, + u64x4, + f64x4, + m64x4, + i128x2, + u128x2, + m128x2 +); +impl_from_bits!(m8x32[test_v256]: m16x16, m32x8, m64x4, m128x2); + +impl_from_bits!( + i16x16[test_v256]: i8x32, + u8x32, + m8x32, + u16x16, + m16x16, + i32x8, + u32x8, + f32x8, + m32x8, + i64x4, + u64x4, + f64x4, + m64x4, + i128x2, + u128x2, + m128x2 +); +impl_from_bits!( + u16x16[test_v256]: i8x32, + u8x32, + m8x32, + i16x16, + m16x16, + i32x8, + u32x8, + f32x8, + m32x8, + i64x4, + u64x4, + f64x4, + m64x4, + i128x2, + u128x2, + m128x2 +); +impl_from_bits!(m16x16[test_v256]: m32x8, m64x4, m128x2); + +impl_from_bits!( + i32x8[test_v256]: i8x32, + u8x32, + m8x32, + i16x16, + u16x16, + m16x16, + u32x8, + f32x8, + m32x8, + i64x4, + u64x4, + f64x4, + m64x4, + i128x2, + u128x2, + m128x2 +); +impl_from_bits!( + u32x8[test_v256]: i8x32, + u8x32, + m8x32, + i16x16, + u16x16, + m16x16, + i32x8, + f32x8, + m32x8, + i64x4, + u64x4, + f64x4, + m64x4, + i128x2, + u128x2, + m128x2 +); +impl_from_bits!( + f32x8[test_v256]: i8x32, + u8x32, + m8x32, + i16x16, + u16x16, + m16x16, + i32x8, + u32x8, + m32x8, + i64x4, + u64x4, + f64x4, + m64x4, + i128x2, + u128x2, + m128x2 +); +impl_from_bits!(m32x8[test_v256]: m64x4, m128x2); + +impl_from_bits!( + i64x4[test_v256]: i8x32, + u8x32, + m8x32, + i16x16, + u16x16, + m16x16, + i32x8, + u32x8, + f32x8, + m32x8, + u64x4, + f64x4, + m64x4, + i128x2, + u128x2, + m128x2 +); +impl_from_bits!( + u64x4[test_v256]: i8x32, + u8x32, + m8x32, + i16x16, + u16x16, + m16x16, + i32x8, + u32x8, + f32x8, + m32x8, + i64x4, + f64x4, + m64x4, + i128x2, + u128x2, + m128x2 +); +impl_from_bits!( + f64x4[test_v256]: i8x32, + u8x32, + m8x32, + i16x16, + u16x16, + m16x16, + i32x8, + u32x8, + f32x8, + m32x8, + i64x4, + u64x4, + m64x4, + i128x2, + u128x2, + m128x2 +); +impl_from_bits!(m64x4[test_v256]: m128x2); + +impl_from_bits!( + i128x2[test_v256]: i8x32, + u8x32, + m8x32, + i16x16, + u16x16, + m16x16, + i32x8, + u32x8, + f32x8, + m32x8, + i64x4, + u64x4, + f64x4, + m64x4, + u128x2, + m128x2 +); +impl_from_bits!( + u128x2[test_v256]: i8x32, + u8x32, + m8x32, + i16x16, + u16x16, + m16x16, + i32x8, + u32x8, + f32x8, + m32x8, + i64x4, + u64x4, + f64x4, + m64x4, + i128x2, + m128x2 +); +// note: m128x2 cannot be constructed from all the other masks bit patterns in +// here diff --git a/third_party/rust/packed_simd/src/api/into_bits/v32.rs b/third_party/rust/packed_simd/src/api/into_bits/v32.rs new file mode 100644 index 000000000000..5dba38a17976 --- /dev/null +++ b/third_party/rust/packed_simd/src/api/into_bits/v32.rs @@ -0,0 +1,13 @@ +//! `FromBits` and `IntoBits` implementations for portable 32-bit wide vectors +#[rustfmt::skip] + +#[allow(unused)] // wasm_bindgen_test +use crate::*; + +impl_from_bits!(i8x4[test_v32]: u8x4, m8x4, i16x2, u16x2, m16x2); +impl_from_bits!(u8x4[test_v32]: i8x4, m8x4, i16x2, u16x2, m16x2); +impl_from_bits!(m8x4[test_v32]: m16x2); + +impl_from_bits!(i16x2[test_v32]: i8x4, u8x4, m8x4, u16x2, m16x2); +impl_from_bits!(u16x2[test_v32]: i8x4, u8x4, m8x4, i16x2, m16x2); +// note: m16x2 cannot be constructed from all m8x4 bit patterns diff --git a/third_party/rust/packed_simd/src/api/into_bits/v512.rs b/third_party/rust/packed_simd/src/api/into_bits/v512.rs new file mode 100644 index 000000000000..f6e9bb8bf7a3 --- /dev/null +++ b/third_party/rust/packed_simd/src/api/into_bits/v512.rs @@ -0,0 +1,232 @@ +//! `FromBits` and `IntoBits` implementations for portable 512-bit wide vectors +#[rustfmt::skip] + +#[allow(unused)] // wasm_bindgen_test +use crate::*; + +impl_from_bits!( + i8x64[test_v512]: u8x64, + m8x64, + i16x32, + u16x32, + m16x32, + i32x16, + u32x16, + f32x16, + m32x16, + i64x8, + u64x8, + f64x8, + m64x8, + i128x4, + u128x4, + m128x4 +); +impl_from_bits!( + u8x64[test_v512]: i8x64, + m8x64, + i16x32, + u16x32, + m16x32, + i32x16, + u32x16, + f32x16, + m32x16, + i64x8, + u64x8, + f64x8, + m64x8, + i128x4, + u128x4, + m128x4 +); +impl_from_bits!(m8x64[test_v512]: m16x32, m32x16, m64x8, m128x4); + +impl_from_bits!( + i16x32[test_v512]: i8x64, + u8x64, + m8x64, + u16x32, + m16x32, + i32x16, + u32x16, + f32x16, + m32x16, + i64x8, + u64x8, + f64x8, + m64x8, + i128x4, + u128x4, + m128x4 +); +impl_from_bits!( + u16x32[test_v512]: i8x64, + u8x64, + m8x64, + i16x32, + m16x32, + i32x16, + u32x16, + f32x16, + m32x16, + i64x8, + u64x8, + f64x8, + m64x8, + i128x4, + u128x4, + m128x4 +); +impl_from_bits!(m16x32[test_v512]: m32x16, m64x8, m128x4); + +impl_from_bits!( + i32x16[test_v512]: i8x64, + u8x64, + m8x64, + i16x32, + u16x32, + m16x32, + u32x16, + f32x16, + m32x16, + i64x8, + u64x8, + f64x8, + m64x8, + i128x4, + u128x4, + m128x4 +); +impl_from_bits!( + u32x16[test_v512]: i8x64, + u8x64, + m8x64, + i16x32, + u16x32, + m16x32, + i32x16, + f32x16, + m32x16, + i64x8, + u64x8, + f64x8, + m64x8, + i128x4, + u128x4, + m128x4 +); +impl_from_bits!( + f32x16[test_v512]: i8x64, + u8x64, + m8x64, + i16x32, + u16x32, + m16x32, + i32x16, + u32x16, + m32x16, + i64x8, + u64x8, + f64x8, + m64x8, + i128x4, + u128x4, + m128x4 +); +impl_from_bits!(m32x16[test_v512]: m64x8, m128x4); + +impl_from_bits!( + i64x8[test_v512]: i8x64, + u8x64, + m8x64, + i16x32, + u16x32, + m16x32, + i32x16, + u32x16, + f32x16, + m32x16, + u64x8, + f64x8, + m64x8, + i128x4, + u128x4, + m128x4 +); +impl_from_bits!( + u64x8[test_v512]: i8x64, + u8x64, + m8x64, + i16x32, + u16x32, + m16x32, + i32x16, + u32x16, + f32x16, + m32x16, + i64x8, + f64x8, + m64x8, + i128x4, + u128x4, + m128x4 +); +impl_from_bits!( + f64x8[test_v512]: i8x64, + u8x64, + m8x64, + i16x32, + u16x32, + m16x32, + i32x16, + u32x16, + f32x16, + m32x16, + i64x8, + u64x8, + m64x8, + i128x4, + u128x4, + m128x4 +); +impl_from_bits!(m64x8[test_v512]: m128x4); + +impl_from_bits!( + i128x4[test_v512]: i8x64, + u8x64, + m8x64, + i16x32, + u16x32, + m16x32, + i32x16, + u32x16, + f32x16, + m32x16, + i64x8, + u64x8, + f64x8, + m64x8, + u128x4, + m128x4 +); +impl_from_bits!( + u128x4[test_v512]: i8x64, + u8x64, + m8x64, + i16x32, + u16x32, + m16x32, + i32x16, + u32x16, + f32x16, + m32x16, + i64x8, + u64x8, + f64x8, + m64x8, + i128x4, + m128x4 +); +// note: m128x4 cannot be constructed from all the other masks bit patterns in +// here diff --git a/third_party/rust/packed_simd/src/api/into_bits/v64.rs b/third_party/rust/packed_simd/src/api/into_bits/v64.rs new file mode 100644 index 000000000000..5b065f1bd5f7 --- /dev/null +++ b/third_party/rust/packed_simd/src/api/into_bits/v64.rs @@ -0,0 +1,18 @@ +//! `FromBits` and `IntoBits` implementations for portable 64-bit wide vectors +#[rustfmt::skip] + +#[allow(unused)] // wasm_bindgen_test +use crate::*; + +impl_from_bits!(i8x8[test_v64]: u8x8, m8x8, i16x4, u16x4, m16x4, i32x2, u32x2, f32x2, m32x2); +impl_from_bits!(u8x8[test_v64]: i8x8, m8x8, i16x4, u16x4, m16x4, i32x2, u32x2, f32x2, m32x2); +impl_from_bits!(m8x8[test_v64]: m16x4, m32x2); + +impl_from_bits!(i16x4[test_v64]: i8x8, u8x8, m8x8, u16x4, m16x4, i32x2, u32x2, f32x2, m32x2); +impl_from_bits!(u16x4[test_v64]: i8x8, u8x8, m8x8, i16x4, m16x4, i32x2, u32x2, f32x2, m32x2); +impl_from_bits!(m16x4[test_v64]: m32x2); + +impl_from_bits!(i32x2[test_v64]: i8x8, u8x8, m8x8, i16x4, u16x4, m16x4, u32x2, f32x2, m32x2); +impl_from_bits!(u32x2[test_v64]: i8x8, u8x8, m8x8, i16x4, u16x4, m16x4, i32x2, f32x2, m32x2); +impl_from_bits!(f32x2[test_v64]: i8x8, u8x8, m8x8, i16x4, u16x4, m16x4, i32x2, u32x2, m32x2); +// note: m32x2 cannot be constructed from all m16x4 or m8x8 bit patterns diff --git a/third_party/rust/packed_simd/src/api/math.rs b/third_party/rust/packed_simd/src/api/math.rs new file mode 100644 index 000000000000..e7a8d256baf5 --- /dev/null +++ b/third_party/rust/packed_simd/src/api/math.rs @@ -0,0 +1,4 @@ +//! Implements vertical math operations + +#[macro_use] +mod float; diff --git a/third_party/rust/packed_simd/src/api/math/float.rs b/third_party/rust/packed_simd/src/api/math/float.rs new file mode 100644 index 000000000000..d5d2bee2eb88 --- /dev/null +++ b/third_party/rust/packed_simd/src/api/math/float.rs @@ -0,0 +1,64 @@ +//! Implements vertical floating-point math operations. + +#[macro_use] +mod abs; + +#[macro_use] +mod consts; + +#[macro_use] +mod cos; + +#[macro_use] +mod exp; + +#[macro_use] +mod powf; + +#[macro_use] +mod ln; + +#[macro_use] +mod mul_add; + +#[macro_use] +mod mul_adde; + +#[macro_use] +mod recpre; + +#[macro_use] +mod rsqrte; + +#[macro_use] +mod sin; + +#[macro_use] +mod sqrt; + +#[macro_use] +mod sqrte; + +#[macro_use] +mod tanh; + +macro_rules! impl_float_category { + ([$elem_ty:ident; $elem_count:expr]: $id:ident, $mask_ty:ident) => { + impl $id { + #[inline] + pub fn is_nan(self) -> $mask_ty { + self.ne(self) + } + + #[inline] + pub fn is_infinite(self) -> $mask_ty { + self.eq(Self::INFINITY) | self.eq(Self::NEG_INFINITY) + } + + #[inline] + pub fn is_finite(self) -> $mask_ty { + !(self.is_nan() | self.is_infinite()) + } + } + }; +} diff --git a/third_party/rust/packed_simd/src/api/math/float/abs.rs b/third_party/rust/packed_simd/src/api/math/float/abs.rs new file mode 100644 index 000000000000..1865bdb68ec6 --- /dev/null +++ b/third_party/rust/packed_simd/src/api/math/float/abs.rs @@ -0,0 +1,31 @@ +//! Implements vertical (lane-wise) floating-point `abs`. + +macro_rules! impl_math_float_abs { + ([$elem_ty:ident; $elem_count:expr]: $id:ident | $test_tt:tt) => { + impl $id { + /// Absolute value. + #[inline] + pub fn abs(self) -> Self { + use crate::codegen::math::float::abs::Abs; + Abs::abs(self) + } + } + + test_if!{ + $test_tt: + paste::item! { + pub mod [<$id _math_abs>] { + use super::*; + #[cfg_attr(not(target_arch = "wasm32"), test)] #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] + fn abs() { + let o = $id::splat(1 as $elem_ty); + assert_eq!(o, o.abs()); + + let mo = $id::splat(-1 as $elem_ty); + assert_eq!(o, mo.abs()); + } + } + } + } + }; +} diff --git a/third_party/rust/packed_simd/src/api/math/float/consts.rs b/third_party/rust/packed_simd/src/api/math/float/consts.rs new file mode 100644 index 000000000000..7f41acbf1b51 --- /dev/null +++ b/third_party/rust/packed_simd/src/api/math/float/consts.rs @@ -0,0 +1,74 @@ +macro_rules! impl_float_consts { + ([$elem_ty:ident; $elem_count:expr]: $id:ident) => { + impl $id { + /// Machine epsilon value. + pub const EPSILON: $id = $id::splat(core::$elem_ty::EPSILON); + + /// Smallest finite value. + pub const MIN: $id = $id::splat(core::$elem_ty::MIN); + + /// Smallest positive normal value. + pub const MIN_POSITIVE: $id = $id::splat(core::$elem_ty::MIN_POSITIVE); + + /// Largest finite value. + pub const MAX: $id = $id::splat(core::$elem_ty::MAX); + + /// Not a Number (NaN). + pub const NAN: $id = $id::splat(core::$elem_ty::NAN); + + /// Infinity (∞). + pub const INFINITY: $id = $id::splat(core::$elem_ty::INFINITY); + + /// Negative infinity (-∞). + pub const NEG_INFINITY: $id = $id::splat(core::$elem_ty::NEG_INFINITY); + + /// Archimedes' constant (π) + pub const PI: $id = $id::splat(core::$elem_ty::consts::PI); + + /// π/2 + pub const FRAC_PI_2: $id = $id::splat(core::$elem_ty::consts::FRAC_PI_2); + + /// π/3 + pub const FRAC_PI_3: $id = $id::splat(core::$elem_ty::consts::FRAC_PI_3); + + /// π/4 + pub const FRAC_PI_4: $id = $id::splat(core::$elem_ty::consts::FRAC_PI_4); + + /// π/6 + pub const FRAC_PI_6: $id = $id::splat(core::$elem_ty::consts::FRAC_PI_6); + + /// π/8 + pub const FRAC_PI_8: $id = $id::splat(core::$elem_ty::consts::FRAC_PI_8); + + /// 1/π + pub const FRAC_1_PI: $id = $id::splat(core::$elem_ty::consts::FRAC_1_PI); + + /// 2/π + pub const FRAC_2_PI: $id = $id::splat(core::$elem_ty::consts::FRAC_2_PI); + + /// 2/sqrt(π) + pub const FRAC_2_SQRT_PI: $id = $id::splat(core::$elem_ty::consts::FRAC_2_SQRT_PI); + + /// sqrt(2) + pub const SQRT_2: $id = $id::splat(core::$elem_ty::consts::SQRT_2); + + /// 1/sqrt(2) + pub const FRAC_1_SQRT_2: $id = $id::splat(core::$elem_ty::consts::FRAC_1_SQRT_2); + + /// Euler's number (e) + pub const E: $id = $id::splat(core::$elem_ty::consts::E); + + /// log2(e) + pub const LOG2_E: $id = $id::splat(core::$elem_ty::consts::LOG2_E); + + /// log10(e) + pub const LOG10_E: $id = $id::splat(core::$elem_ty::consts::LOG10_E); + + /// ln(2) + pub const LN_2: $id = $id::splat(core::$elem_ty::consts::LN_2); + + /// ln(10) + pub const LN_10: $id = $id::splat(core::$elem_ty::consts::LN_10); + } + }; +} diff --git a/third_party/rust/packed_simd/src/api/math/float/cos.rs b/third_party/rust/packed_simd/src/api/math/float/cos.rs new file mode 100644 index 000000000000..e5b8f46036c7 --- /dev/null +++ b/third_party/rust/packed_simd/src/api/math/float/cos.rs @@ -0,0 +1,44 @@ +//! Implements vertical (lane-wise) floating-point `cos`. + +macro_rules! impl_math_float_cos { + ([$elem_ty:ident; $elem_count:expr]: $id:ident | $test_tt:tt) => { + impl $id { + /// Cosine. + #[inline] + pub fn cos(self) -> Self { + use crate::codegen::math::float::cos::Cos; + Cos::cos(self) + } + + /// Cosine of `self * PI`. + #[inline] + pub fn cos_pi(self) -> Self { + use crate::codegen::math::float::cos_pi::CosPi; + CosPi::cos_pi(self) + } + } + + test_if!{ + $test_tt: + paste::item! { + pub mod [<$id _math_cos>] { + use super::*; + #[cfg_attr(not(target_arch = "wasm32"), test)] #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] + fn cos() { + use crate::$elem_ty::consts::PI; + let z = $id::splat(0 as $elem_ty); + let o = $id::splat(1 as $elem_ty); + let p = $id::splat(PI as $elem_ty); + let ph = $id::splat(PI as $elem_ty / 2.); + let z_r = $id::splat((PI as $elem_ty / 2.).cos()); + let o_r = $id::splat((PI as $elem_ty).cos()); + + assert_eq!(o, z.cos()); + assert_eq!(z_r, ph.cos()); + assert_eq!(o_r, p.cos()); + } + } + } + } + }; +} diff --git a/third_party/rust/packed_simd/src/api/math/float/exp.rs b/third_party/rust/packed_simd/src/api/math/float/exp.rs new file mode 100644 index 000000000000..e3356d853a83 --- /dev/null +++ b/third_party/rust/packed_simd/src/api/math/float/exp.rs @@ -0,0 +1,33 @@ +//! Implements vertical (lane-wise) floating-point `exp`. + +macro_rules! impl_math_float_exp { + ([$elem_ty:ident; $elem_count:expr]: $id:ident | $test_tt:tt) => { + impl $id { + /// Returns the exponential function of `self`: `e^(self)`. + #[inline] + pub fn exp(self) -> Self { + use crate::codegen::math::float::exp::Exp; + Exp::exp(self) + } + } + + test_if!{ + $test_tt: + paste::item! { + pub mod [<$id _math_exp>] { + use super::*; + #[cfg_attr(not(target_arch = "wasm32"), test)] #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] + fn exp() { + let z = $id::splat(0 as $elem_ty); + let o = $id::splat(1 as $elem_ty); + assert_eq!(o, z.exp()); + + let e = $id::splat(crate::f64::consts::E as $elem_ty); + let tol = $id::splat(2.4e-4 as $elem_ty); + assert!((e - o.exp()).abs().le(tol).all()); + } + } + } + } + }; +} diff --git a/third_party/rust/packed_simd/src/api/math/float/ln.rs b/third_party/rust/packed_simd/src/api/math/float/ln.rs new file mode 100644 index 000000000000..5ceb9173ae05 --- /dev/null +++ b/third_party/rust/packed_simd/src/api/math/float/ln.rs @@ -0,0 +1,33 @@ +//! Implements vertical (lane-wise) floating-point `ln`. + +macro_rules! impl_math_float_ln { + ([$elem_ty:ident; $elem_count:expr]: $id:ident | $test_tt:tt) => { + impl $id { + /// Returns the natural logarithm of `self`. + #[inline] + pub fn ln(self) -> Self { + use crate::codegen::math::float::ln::Ln; + Ln::ln(self) + } + } + + test_if!{ + $test_tt: + paste::item! { + pub mod [<$id _math_ln>] { + use super::*; + #[cfg_attr(not(target_arch = "wasm32"), test)] #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] + fn ln() { + let z = $id::splat(0 as $elem_ty); + let o = $id::splat(1 as $elem_ty); + assert_eq!(z, o.ln()); + + let e = $id::splat(crate::f64::consts::E as $elem_ty); + let tol = $id::splat(2.4e-4 as $elem_ty); + assert!((o - e.ln()).abs().le(tol).all()); + } + } + } + } + }; +} diff --git a/third_party/rust/packed_simd/src/api/math/float/mul_add.rs b/third_party/rust/packed_simd/src/api/math/float/mul_add.rs new file mode 100644 index 000000000000..4b170ee2b755 --- /dev/null +++ b/third_party/rust/packed_simd/src/api/math/float/mul_add.rs @@ -0,0 +1,44 @@ +//! Implements vertical (lane-wise) floating-point `mul_add`. + +macro_rules! impl_math_float_mul_add { + ([$elem_ty:ident; $elem_count:expr]: $id:ident | $test_tt:tt) => { + impl $id { + /// Fused multiply add: `self * y + z` + #[inline] + pub fn mul_add(self, y: Self, z: Self) -> Self { + use crate::codegen::math::float::mul_add::MulAdd; + MulAdd::mul_add(self, y, z) + } + } + + test_if!{ + $test_tt: + paste::item! { + pub mod [<$id _math_mul_add>] { + use super::*; + #[cfg_attr(not(target_arch = "wasm32"), test)] #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] + fn mul_add() { + let z = $id::splat(0 as $elem_ty); + let o = $id::splat(1 as $elem_ty); + let t = $id::splat(2 as $elem_ty); + let t3 = $id::splat(3 as $elem_ty); + let f = $id::splat(4 as $elem_ty); + + assert_eq!(z, z.mul_add(z, z)); + assert_eq!(o, o.mul_add(o, z)); + assert_eq!(o, o.mul_add(z, o)); + assert_eq!(o, z.mul_add(o, o)); + + assert_eq!(t, o.mul_add(o, o)); + assert_eq!(t, o.mul_add(t, z)); + assert_eq!(t, t.mul_add(o, z)); + + assert_eq!(f, t.mul_add(t, z)); + assert_eq!(f, t.mul_add(o, t)); + assert_eq!(t3, t.mul_add(o, o)); + } + } + } + } + }; +} diff --git a/third_party/rust/packed_simd/src/api/math/float/mul_adde.rs b/third_party/rust/packed_simd/src/api/math/float/mul_adde.rs new file mode 100644 index 000000000000..c5b27110f2d7 --- /dev/null +++ b/third_party/rust/packed_simd/src/api/math/float/mul_adde.rs @@ -0,0 +1,48 @@ +//! Implements vertical (lane-wise) floating-point `mul_adde`. + +macro_rules! impl_math_float_mul_adde { + ([$elem_ty:ident; $elem_count:expr]: $id:ident | $test_tt:tt) => { + impl $id { + /// Fused multiply add estimate: ~= `self * y + z` + /// + /// While fused multiply-add (`fma`) has infinite precision, + /// `mul_adde` has _at worst_ the same precision of a multiply followed by an add. + /// This might be more efficient on architectures that do not have an `fma` instruction. + #[inline] + pub fn mul_adde(self, y: Self, z: Self) -> Self { + use crate::codegen::math::float::mul_adde::MulAddE; + MulAddE::mul_adde(self, y, z) + } + } + + test_if!{ + $test_tt: + paste::item! { + pub mod [<$id _math_mul_adde>] { + use super::*; + #[cfg_attr(not(target_arch = "wasm32"), test)] #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] + fn mul_adde() { + let z = $id::splat(0 as $elem_ty); + let o = $id::splat(1 as $elem_ty); + let t = $id::splat(2 as $elem_ty); + let t3 = $id::splat(3 as $elem_ty); + let f = $id::splat(4 as $elem_ty); + + assert_eq!(z, z.mul_adde(z, z)); + assert_eq!(o, o.mul_adde(o, z)); + assert_eq!(o, o.mul_adde(z, o)); + assert_eq!(o, z.mul_adde(o, o)); + + assert_eq!(t, o.mul_adde(o, o)); + assert_eq!(t, o.mul_adde(t, z)); + assert_eq!(t, t.mul_adde(o, z)); + + assert_eq!(f, t.mul_adde(t, z)); + assert_eq!(f, t.mul_adde(o, t)); + assert_eq!(t3, t.mul_adde(o, o)); + } + } + } + } + }; +} diff --git a/third_party/rust/packed_simd/src/api/math/float/powf.rs b/third_party/rust/packed_simd/src/api/math/float/powf.rs new file mode 100644 index 000000000000..83dc9ff9c05e --- /dev/null +++ b/third_party/rust/packed_simd/src/api/math/float/powf.rs @@ -0,0 +1,36 @@ +//! Implements vertical (lane-wise) floating-point `powf`. + +macro_rules! impl_math_float_powf { + ([$elem_ty:ident; $elem_count:expr]: $id:ident | $test_tt:tt) => { + impl $id { + /// Raises `self` number to the floating point power of `x`. + #[inline] + pub fn powf(self, x: Self) -> Self { + use crate::codegen::math::float::powf::Powf; + Powf::powf(self, x) + } + } + + test_if!{ + $test_tt: + paste::item! { + pub mod [<$id _math_powf>] { + use super::*; + #[cfg_attr(not(target_arch = "wasm32"), test)] #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] + fn powf() { + let z = $id::splat(0 as $elem_ty); + let o = $id::splat(1 as $elem_ty); + let t = $id::splat(2 as $elem_ty); + assert_eq!(o, o.powf(z)); + assert_eq!(o, t.powf(z)); + assert_eq!(o, o.powf(o)); + assert_eq!(t, t.powf(o)); + + let f = $id::splat(4 as $elem_ty); + assert_eq!(f, t.powf(t)); + } + } + } + } + }; +} diff --git a/third_party/rust/packed_simd/src/api/math/float/recpre.rs b/third_party/rust/packed_simd/src/api/math/float/recpre.rs new file mode 100644 index 000000000000..127f0b2ff674 --- /dev/null +++ b/third_party/rust/packed_simd/src/api/math/float/recpre.rs @@ -0,0 +1,36 @@ +//! Implements vertical (lane-wise) floating-point `recpre`. + +macro_rules! impl_math_float_recpre { + ([$elem_ty:ident; $elem_count:expr]: $id:ident | $test_tt:tt) => { + impl $id { + /// Reciprocal estimate: `~= 1. / self`. + /// + /// FIXME: The precision of the estimate is currently unspecified. + #[inline] + pub fn recpre(self) -> Self { + $id::splat(1.) / self + } + } + + test_if!{ + $test_tt: + paste::item! { + pub mod [<$id _math_recpre>] { + use super::*; + #[cfg_attr(not(target_arch = "wasm32"), test)] #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] + fn recpre() { + let tol = $id::splat(2.4e-4 as $elem_ty); + let o = $id::splat(1 as $elem_ty); + let error = (o - o.recpre()).abs(); + assert!(error.le(tol).all()); + + let t = $id::splat(2 as $elem_ty); + let e = 0.5; + let error = (e - t.recpre()).abs(); + assert!(error.le(tol).all()); + } + } + } + } + }; +} diff --git a/third_party/rust/packed_simd/src/api/math/float/rsqrte.rs b/third_party/rust/packed_simd/src/api/math/float/rsqrte.rs new file mode 100644 index 000000000000..c77977f7b1cd --- /dev/null +++ b/third_party/rust/packed_simd/src/api/math/float/rsqrte.rs @@ -0,0 +1,40 @@ +//! Implements vertical (lane-wise) floating-point `rsqrte`. + +macro_rules! impl_math_float_rsqrte { + ([$elem_ty:ident; $elem_count:expr]: $id:ident | $test_tt:tt) => { + impl $id { + /// Reciprocal square-root estimate: `~= 1. / self.sqrt()`. + /// + /// FIXME: The precision of the estimate is currently unspecified. + #[inline] + pub fn rsqrte(self) -> Self { + unsafe { + use crate::llvm::simd_fsqrt; + $id::splat(1.) / Simd(simd_fsqrt(self.0)) + } + } + } + + test_if!{ + $test_tt: + paste::item! { + pub mod [<$id _math_rsqrte>] { + use super::*; + #[cfg_attr(not(target_arch = "wasm32"), test)] #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] + fn rsqrte() { + use crate::$elem_ty::consts::SQRT_2; + let tol = $id::splat(2.4e-4 as $elem_ty); + let o = $id::splat(1 as $elem_ty); + let error = (o - o.rsqrte()).abs(); + assert!(error.le(tol).all()); + + let t = $id::splat(2 as $elem_ty); + let e = 1. / SQRT_2; + let error = (e - t.rsqrte()).abs(); + assert!(error.le(tol).all()); + } + } + } + } + }; +} diff --git a/third_party/rust/packed_simd/src/api/math/float/sin.rs b/third_party/rust/packed_simd/src/api/math/float/sin.rs new file mode 100644 index 000000000000..49908319b126 --- /dev/null +++ b/third_party/rust/packed_simd/src/api/math/float/sin.rs @@ -0,0 +1,50 @@ +//! Implements vertical (lane-wise) floating-point `sin`. + +macro_rules! impl_math_float_sin { + ([$elem_ty:ident; $elem_count:expr]: $id:ident | $test_tt:tt) => { + impl $id { + /// Sine. + #[inline] + pub fn sin(self) -> Self { + use crate::codegen::math::float::sin::Sin; + Sin::sin(self) + } + + /// Sine of `self * PI`. + #[inline] + pub fn sin_pi(self) -> Self { + use crate::codegen::math::float::sin_pi::SinPi; + SinPi::sin_pi(self) + } + + /// Sine and cosine of `self * PI`. + #[inline] + pub fn sin_cos_pi(self) -> (Self, Self) { + use crate::codegen::math::float::sin_cos_pi::SinCosPi; + SinCosPi::sin_cos_pi(self) + } + } + + test_if!{ + $test_tt: + paste::item! { + pub mod [<$id _math_sin>] { + use super::*; + #[cfg_attr(not(target_arch = "wasm32"), test)] #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] + fn sin() { + use crate::$elem_ty::consts::PI; + let z = $id::splat(0 as $elem_ty); + let p = $id::splat(PI as $elem_ty); + let ph = $id::splat(PI as $elem_ty / 2.); + let o_r = $id::splat((PI as $elem_ty / 2.).sin()); + let z_r = $id::splat((PI as $elem_ty).sin()); + + assert_eq!(z, z.sin()); + assert_eq!(o_r, ph.sin()); + assert_eq!(z_r, p.sin()); + } + } + } + } + }; +} diff --git a/third_party/rust/packed_simd/src/api/math/float/sqrt.rs b/third_party/rust/packed_simd/src/api/math/float/sqrt.rs new file mode 100644 index 000000000000..ae624122d0e2 --- /dev/null +++ b/third_party/rust/packed_simd/src/api/math/float/sqrt.rs @@ -0,0 +1,35 @@ +//! Implements vertical (lane-wise) floating-point `sqrt`. + +macro_rules! impl_math_float_sqrt { + ([$elem_ty:ident; $elem_count:expr]: $id:ident | $test_tt:tt) => { + impl $id { + #[inline] + pub fn sqrt(self) -> Self { + use crate::codegen::math::float::sqrt::Sqrt; + Sqrt::sqrt(self) + } + } + + test_if!{ + $test_tt: + paste::item! { + pub mod [<$id _math_sqrt>] { + use super::*; + #[cfg_attr(not(target_arch = "wasm32"), test)] #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] + fn sqrt() { + use crate::$elem_ty::consts::SQRT_2; + let z = $id::splat(0 as $elem_ty); + let o = $id::splat(1 as $elem_ty); + assert_eq!(z, z.sqrt()); + assert_eq!(o, o.sqrt()); + + let t = $id::splat(2 as $elem_ty); + let e = $id::splat(SQRT_2); + assert_eq!(e, t.sqrt()); + + } + } + } + } + }; +} diff --git a/third_party/rust/packed_simd/src/api/math/float/sqrte.rs b/third_party/rust/packed_simd/src/api/math/float/sqrte.rs new file mode 100644 index 000000000000..f7ffad748d9c --- /dev/null +++ b/third_party/rust/packed_simd/src/api/math/float/sqrte.rs @@ -0,0 +1,44 @@ +//! Implements vertical (lane-wise) floating-point `sqrte`. + +macro_rules! impl_math_float_sqrte { + ([$elem_ty:ident; $elem_count:expr]: $id:ident | $test_tt:tt) => { + impl $id { + /// Square-root estimate. + /// + /// FIXME: The precision of the estimate is currently unspecified. + #[inline] + pub fn sqrte(self) -> Self { + use crate::codegen::math::float::sqrte::Sqrte; + Sqrte::sqrte(self) + } + } + + test_if!{ + $test_tt: + paste::item! { + pub mod [<$id _math_sqrte>] { + use super::*; + #[cfg_attr(not(target_arch = "wasm32"), test)] #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] + fn sqrte() { + use crate::$elem_ty::consts::SQRT_2; + let tol = $id::splat(2.4e-4 as $elem_ty); + + let z = $id::splat(0 as $elem_ty); + let error = (z - z.sqrte()).abs(); + assert!(error.le(tol).all()); + + let o = $id::splat(1 as $elem_ty); + let error = (o - o.sqrte()).abs(); + assert!(error.le(tol).all()); + + let t = $id::splat(2 as $elem_ty); + let e = $id::splat(SQRT_2 as $elem_ty); + let error = (e - t.sqrte()).abs(); + + assert!(error.le(tol).all()); + } + } + } + } + }; +} diff --git a/third_party/rust/packed_simd/src/api/math/float/tanh.rs b/third_party/rust/packed_simd/src/api/math/float/tanh.rs new file mode 100644 index 000000000000..acfd93caaa5d --- /dev/null +++ b/third_party/rust/packed_simd/src/api/math/float/tanh.rs @@ -0,0 +1,29 @@ +//! Implements vertical (lane-wise) floating-point `tanh`. + +macro_rules! impl_math_float_tanh { + ([$elem_ty:ident; $elem_count:expr]: $id:ident | $test_tt:tt) => { + impl $id { + /// Tanh. + #[inline] + pub fn tanh(self) -> Self { + use crate::codegen::math::float::tanh::Tanh; + Tanh::tanh(self) + } + } + + test_if!{ + $test_tt: + paste::item! { + pub mod [<$id _math_tanh>] { + use super::*; + #[cfg_attr(not(target_arch = "wasm32"), test)] #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] + fn tanh() { + let z = $id::splat(0 as $elem_ty); + + assert_eq!(z, z.tanh()); + } + } + } + } + }; +} diff --git a/third_party/rust/packed_simd/src/api/minimal.rs b/third_party/rust/packed_simd/src/api/minimal.rs new file mode 100644 index 000000000000..840d9e32585d --- /dev/null +++ b/third_party/rust/packed_simd/src/api/minimal.rs @@ -0,0 +1,6 @@ +#[macro_use] +mod iuf; +#[macro_use] +mod mask; +#[macro_use] +mod ptr; diff --git a/third_party/rust/packed_simd/src/api/minimal/iuf.rs b/third_party/rust/packed_simd/src/api/minimal/iuf.rs new file mode 100644 index 000000000000..a155ac178a26 --- /dev/null +++ b/third_party/rust/packed_simd/src/api/minimal/iuf.rs @@ -0,0 +1,169 @@ +//! Minimal API of signed integer, unsigned integer, and floating-point +//! vectors. + +macro_rules! impl_minimal_iuf { + ([$elem_ty:ident; $elem_count:expr]: $id:ident | $ielem_ty:ident | + $test_tt:tt | $($elem_name:ident),+ | $(#[$doc:meta])*) => { + + $(#[$doc])* + pub type $id = Simd<[$elem_ty; $elem_count]>; + + impl sealed::Simd for $id { + type Element = $elem_ty; + const LANES: usize = $elem_count; + type LanesType = [u32; $elem_count]; + } + + impl $id { + /// Creates a new instance with each vector elements initialized + /// with the provided values. + #[inline] + #[allow(clippy::too_many_arguments)] + pub const fn new($($elem_name: $elem_ty),*) -> Self { + Simd(codegen::$id($($elem_name as $ielem_ty),*)) + } + + /// Returns the number of vector lanes. + #[inline] + pub const fn lanes() -> usize { + $elem_count + } + + /// Constructs a new instance with each element initialized to + /// `value`. + #[inline] + pub const fn splat(value: $elem_ty) -> Self { + Simd(codegen::$id($({ + #[allow(non_camel_case_types, dead_code)] + struct $elem_name; + value as $ielem_ty + }),*)) + } + + /// Extracts the value at `index`. + /// + /// # Panics + /// + /// If `index >= Self::lanes()`. + #[inline] + pub fn extract(self, index: usize) -> $elem_ty { + assert!(index < $elem_count); + unsafe { self.extract_unchecked(index) } + } + + /// Extracts the value at `index`. + /// + /// # Safety + /// + /// If `index >= Self::lanes()` the behavior is undefined. + #[inline] + pub unsafe fn extract_unchecked(self, index: usize) -> $elem_ty { + use crate::llvm::simd_extract; + let e: $ielem_ty = simd_extract(self.0, index as u32); + e as $elem_ty + } + + /// Returns a new vector where the value at `index` is replaced by `new_value`. + /// + /// # Panics + /// + /// If `index >= Self::lanes()`. + #[inline] + #[must_use = "replace does not modify the original value - \ + it returns a new vector with the value at `index` \ + replaced by `new_value`d" + ] + pub fn replace(self, index: usize, new_value: $elem_ty) -> Self { + assert!(index < $elem_count); + unsafe { self.replace_unchecked(index, new_value) } + } + + /// Returns a new vector where the value at `index` is replaced by `new_value`. + /// + /// # Safety + /// + /// If `index >= Self::lanes()` the behavior is undefined. + #[inline] + #[must_use = "replace_unchecked does not modify the original value - \ + it returns a new vector with the value at `index` \ + replaced by `new_value`d" + ] + pub unsafe fn replace_unchecked( + self, + index: usize, + new_value: $elem_ty, + ) -> Self { + use crate::llvm::simd_insert; + Simd(simd_insert(self.0, index as u32, new_value as $ielem_ty)) + } + } + + test_if!{ + $test_tt: + paste::item! { + // Comparisons use integer casts within mantissa^1 range. + #[allow(clippy::float_cmp)] + pub mod [<$id _minimal>] { + use super::*; + #[cfg_attr(not(target_arch = "wasm32"), test)] + #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] + fn minimal() { + // lanes: + assert_eq!($elem_count, $id::lanes()); + + // splat and extract / extract_unchecked: + const VAL: $elem_ty = 7 as $elem_ty; + const VEC: $id = $id::splat(VAL); + for i in 0..$id::lanes() { + assert_eq!(VAL, VEC.extract(i)); + assert_eq!( + VAL, unsafe { VEC.extract_unchecked(i) } + ); + } + + // replace / replace_unchecked + let new_vec = VEC.replace(0, 42 as $elem_ty); + for i in 0..$id::lanes() { + if i == 0 { + assert_eq!(42 as $elem_ty, new_vec.extract(i)); + } else { + assert_eq!(VAL, new_vec.extract(i)); + } + } + let new_vec = unsafe { + VEC.replace_unchecked(0, 42 as $elem_ty) + }; + for i in 0..$id::lanes() { + if i == 0 { + assert_eq!(42 as $elem_ty, new_vec.extract(i)); + } else { + assert_eq!(VAL, new_vec.extract(i)); + } + } + } + + // FIXME: wasm-bindgen-test does not support #[should_panic] + // #[cfg_attr(not(target_arch = "wasm32"), test)] #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] + #[cfg(not(target_arch = "wasm32"))] + #[test] + #[should_panic] + fn extract_panic_oob() { + const VAL: $elem_ty = 7 as $elem_ty; + const VEC: $id = $id::splat(VAL); + let _ = VEC.extract($id::lanes()); + } + // FIXME: wasm-bindgen-test does not support #[should_panic] + // #[cfg_attr(not(target_arch = "wasm32"), test)] #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] + #[cfg(not(target_arch = "wasm32"))] + #[test] + #[should_panic] + fn replace_panic_oob() { + const VAL: $elem_ty = 7 as $elem_ty; + const VEC: $id = $id::splat(VAL); + let _ = VEC.replace($id::lanes(), 42 as $elem_ty); + } + } + } + } + } +} diff --git a/third_party/rust/packed_simd/src/api/minimal/mask.rs b/third_party/rust/packed_simd/src/api/minimal/mask.rs new file mode 100644 index 000000000000..a420060b423d --- /dev/null +++ b/third_party/rust/packed_simd/src/api/minimal/mask.rs @@ -0,0 +1,176 @@ +//! Minimal API of mask vectors. + +macro_rules! impl_minimal_mask { + ([$elem_ty:ident; $elem_count:expr]: $id:ident | $ielem_ty:ident + | $test_tt:tt | $($elem_name:ident),+ | $(#[$doc:meta])*) => { + $(#[$doc])* + pub type $id = Simd<[$elem_ty; $elem_count]>; + + impl sealed::Simd for $id { + type Element = $elem_ty; + const LANES: usize = $elem_count; + type LanesType = [u32; $elem_count]; + } + + impl $id { + /// Creates a new instance with each vector elements initialized + /// with the provided values. + #[inline] + #[allow(clippy::too_many_arguments)] + pub const fn new($($elem_name: bool),*) -> Self { + Simd(codegen::$id($(Self::bool_to_internal($elem_name)),*)) + } + + /// Converts a boolean type into the type of the vector lanes. + #[inline] + #[allow(clippy::indexing_slicing)] + const fn bool_to_internal(x: bool) -> $ielem_ty { + [0 as $ielem_ty, !(0 as $ielem_ty)][x as usize] + } + + /// Returns the number of vector lanes. + #[inline] + pub const fn lanes() -> usize { + $elem_count + } + + /// Constructs a new instance with each element initialized to + /// `value`. + #[inline] + pub const fn splat(value: bool) -> Self { + Simd(codegen::$id($({ + #[allow(non_camel_case_types, dead_code)] + struct $elem_name; + Self::bool_to_internal(value) + }),*)) + } + + /// Extracts the value at `index`. + /// + /// # Panics + /// + /// If `index >= Self::lanes()`. + #[inline] + pub fn extract(self, index: usize) -> bool { + assert!(index < $elem_count); + unsafe { self.extract_unchecked(index) } + } + + /// Extracts the value at `index`. + /// + /// # Safety + /// + /// If `index >= Self::lanes()` the behavior is undefined. + #[inline] + pub unsafe fn extract_unchecked(self, index: usize) -> bool { + use crate::llvm::simd_extract; + let x: $ielem_ty = simd_extract(self.0, index as u32); + x != 0 + } + + /// Returns a new vector where the value at `index` is replaced by + /// `new_value`. + /// + /// # Panics + /// + /// If `index >= Self::lanes()`. + #[inline] + #[must_use = "replace does not modify the original value - \ + it returns a new vector with the value at `index` \ + replaced by `new_value`d" + ] + pub fn replace(self, index: usize, new_value: bool) -> Self { + assert!(index < $elem_count); + unsafe { self.replace_unchecked(index, new_value) } + } + + /// Returns a new vector where the value at `index` is replaced by + /// `new_value`. + /// + /// # Safety + /// + /// If `index >= Self::lanes()` the behavior is undefined. + #[inline] + #[must_use = "replace_unchecked does not modify the original value - \ + it returns a new vector with the value at `index` \ + replaced by `new_value`d" + ] + pub unsafe fn replace_unchecked( + self, + index: usize, + new_value: bool, + ) -> Self { + use crate::llvm::simd_insert; + Simd(simd_insert(self.0, index as u32, + Self::bool_to_internal(new_value))) + } + } + + test_if!{ + $test_tt: + paste::item! { + pub mod [<$id _minimal>] { + use super::*; + #[cfg_attr(not(target_arch = "wasm32"), test)] + #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] + fn minimal() { + // TODO: test new + + // lanes: + assert_eq!($elem_count, $id::lanes()); + + // splat and extract / extract_unchecked: + let vec = $id::splat(true); + for i in 0..$id::lanes() { + assert_eq!(true, vec.extract(i)); + assert_eq!(true, + unsafe { vec.extract_unchecked(i) } + ); + } + + // replace / replace_unchecked + let new_vec = vec.replace(0, false); + for i in 0..$id::lanes() { + if i == 0 { + assert_eq!(false, new_vec.extract(i)); + } else { + assert_eq!(true, new_vec.extract(i)); + } + } + let new_vec = unsafe { + vec.replace_unchecked(0, false) + }; + for i in 0..$id::lanes() { + if i == 0 { + assert_eq!(false, new_vec.extract(i)); + } else { + assert_eq!(true, new_vec.extract(i)); + } + } + } + + // FIXME: wasm-bindgen-test does not support #[should_panic] + // #[cfg_attr(not(target_arch = "wasm32"), test)] + // #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] + #[cfg(not(target_arch = "wasm32"))] + #[test] + #[should_panic] + fn extract_panic_oob() { + let vec = $id::splat(false); + let _ = vec.extract($id::lanes()); + } + // FIXME: wasm-bindgen-test does not support #[should_panic] + // #[cfg_attr(not(target_arch = "wasm32"), test)] + // #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] + #[cfg(not(target_arch = "wasm32"))] + #[test] + #[should_panic] + fn replace_panic_oob() { + let vec = $id::splat(false); + let _ = vec.replace($id::lanes(), true); + } + } + } + } + } +} diff --git a/third_party/rust/packed_simd/src/api/minimal/ptr.rs b/third_party/rust/packed_simd/src/api/minimal/ptr.rs new file mode 100644 index 000000000000..d9e47c9ccbfb --- /dev/null +++ b/third_party/rust/packed_simd/src/api/minimal/ptr.rs @@ -0,0 +1,1373 @@ +//! Minimal API of pointer vectors. + +macro_rules! impl_minimal_p { + ([$elem_ty:ty; $elem_count:expr]: $id:ident, $mask_ty:ident, + $usize_ty:ident, $isize_ty:ident | $ref:ident | $test_tt:tt + | $($elem_name:ident),+ | ($true:expr, $false:expr) | + $(#[$doc:meta])*) => { + + $(#[$doc])* + pub type $id = Simd<[$elem_ty; $elem_count]>; + + impl sealed::Simd for $id { + type Element = $elem_ty; + const LANES: usize = $elem_count; + type LanesType = [u32; $elem_count]; + } + + impl $id { + /// Creates a new instance with each vector elements initialized + /// with the provided values. + #[inline] + #[allow(clippy::too_many_arguments)] + pub const fn new($($elem_name: $elem_ty),*) -> Self { + Simd(codegen::$id($($elem_name),*)) + } + + /// Returns the number of vector lanes. + #[inline] + pub const fn lanes() -> usize { + $elem_count + } + + /// Constructs a new instance with each element initialized to + /// `value`. + #[inline] + pub const fn splat(value: $elem_ty) -> Self { + Simd(codegen::$id($({ + #[allow(non_camel_case_types, dead_code)] + struct $elem_name; + value + }),*)) + } + + /// Constructs a new instance with each element initialized to + /// `null`. + #[inline] + pub const fn null() -> Self { + Self::splat(crate::ptr::null_mut() as $elem_ty) + } + + /// Returns a mask that selects those lanes that contain `null` + /// pointers. + #[inline] + pub fn is_null(self) -> $mask_ty { + self.eq(Self::null()) + } + + /// Extracts the value at `index`. + /// + /// # Panics + /// + /// If `index >= Self::lanes()`. + #[inline] + pub fn extract(self, index: usize) -> $elem_ty { + assert!(index < $elem_count); + unsafe { self.extract_unchecked(index) } + } + + /// Extracts the value at `index`. + /// + /// # Safety + /// + /// If `index >= Self::lanes()` the behavior is undefined. + #[inline] + pub unsafe fn extract_unchecked(self, index: usize) -> $elem_ty { + use crate::llvm::simd_extract; + simd_extract(self.0, index as u32) + } + + /// Returns a new vector where the value at `index` is replaced by + /// `new_value`. + /// + /// # Panics + /// + /// If `index >= Self::lanes()`. + #[inline] + #[must_use = "replace does not modify the original value - \ + it returns a new vector with the value at `index` \ + replaced by `new_value`d" + ] + #[allow(clippy::not_unsafe_ptr_arg_deref)] + pub fn replace(self, index: usize, new_value: $elem_ty) -> Self { + assert!(index < $elem_count); + unsafe { self.replace_unchecked(index, new_value) } + } + + /// Returns a new vector where the value at `index` is replaced by `new_value`. + /// + /// # Safety + /// + /// If `index >= Self::lanes()` the behavior is undefined. + #[inline] + #[must_use = "replace_unchecked does not modify the original value - \ + it returns a new vector with the value at `index` \ + replaced by `new_value`d" + ] + pub unsafe fn replace_unchecked( + self, + index: usize, + new_value: $elem_ty, + ) -> Self { + use crate::llvm::simd_insert; + Simd(simd_insert(self.0, index as u32, new_value)) + } + } + + + test_if!{ + $test_tt: + paste::item! { + pub mod [<$id _minimal>] { + use super::*; + #[cfg_attr(not(target_arch = "wasm32"), test)] + #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] + fn minimal() { + // lanes: + assert_eq!($elem_count, $id::::lanes()); + + // splat and extract / extract_unchecked: + let VAL7: <$id as sealed::Simd>::Element + = $ref!(7); + let VAL42: <$id as sealed::Simd>::Element + = $ref!(42); + let VEC: $id = $id::splat(VAL7); + for i in 0..$id::::lanes() { + assert_eq!(VAL7, VEC.extract(i)); + assert_eq!( + VAL7, unsafe { VEC.extract_unchecked(i) } + ); + } + + // replace / replace_unchecked + let new_vec = VEC.replace(0, VAL42); + for i in 0..$id::::lanes() { + if i == 0 { + assert_eq!(VAL42, new_vec.extract(i)); + } else { + assert_eq!(VAL7, new_vec.extract(i)); + } + } + let new_vec = unsafe { + VEC.replace_unchecked(0, VAL42) + }; + for i in 0..$id::::lanes() { + if i == 0 { + assert_eq!(VAL42, new_vec.extract(i)); + } else { + assert_eq!(VAL7, new_vec.extract(i)); + } + } + + let mut n = $id::::null(); + assert_eq!( + n, + $id::::splat(unsafe { crate::mem::zeroed() }) + ); + assert!(n.is_null().all()); + n = n.replace( + 0, unsafe { crate::mem::transmute(1_isize) } + ); + assert!(!n.is_null().all()); + if $id::::lanes() > 1 { + assert!(n.is_null().any()); + } else { + assert!(!n.is_null().any()); + } + } + + // FIXME: wasm-bindgen-test does not support #[should_panic] + // #[cfg_attr(not(target_arch = "wasm32"), test)] + // #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] + #[cfg(not(target_arch = "wasm32"))] + #[test] + #[should_panic] + fn extract_panic_oob() { + let VAL: <$id as sealed::Simd>::Element + = $ref!(7); + let VEC: $id = $id::splat(VAL); + let _ = VEC.extract($id::::lanes()); + } + + // FIXME: wasm-bindgen-test does not support #[should_panic] + // #[cfg_attr(not(target_arch = "wasm32"), test)] + // #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] + #[cfg(not(target_arch = "wasm32"))] + #[test] + #[should_panic] + fn replace_panic_oob() { + let VAL: <$id as sealed::Simd>::Element + = $ref!(7); + let VAL42: <$id as sealed::Simd>::Element + = $ref!(42); + let VEC: $id = $id::splat(VAL); + let _ = VEC.replace($id::::lanes(), VAL42); + } + } + } + } + + impl crate::fmt::Debug for $id { + #[allow(clippy::missing_inline_in_public_items)] + fn fmt(&self, f: &mut crate::fmt::Formatter<'_>) + -> crate::fmt::Result { + write!( + f, + "{}<{}>(", + stringify!($id), + crate::intrinsics::type_name::() + )?; + for i in 0..$elem_count { + if i > 0 { + write!(f, ", ")?; + } + self.extract(i).fmt(f)?; + } + write!(f, ")") + } + } + + test_if!{ + $test_tt: + paste::item! { + pub mod [<$id _fmt_debug>] { + use super::*; + #[cfg_attr(not(target_arch = "wasm32"), test)] + #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] + fn debug() { + use arrayvec::{ArrayString,ArrayVec}; + type TinyString = ArrayString<[u8; 512]>; + + use crate::fmt::Write; + let v = $id::::default(); + let mut s = TinyString::new(); + write!(&mut s, "{:?}", v).unwrap(); + + let mut beg = TinyString::new(); + write!(&mut beg, "{}(", stringify!($id)).unwrap(); + assert!( + s.starts_with(beg.as_str()), + "s = {} (should start with = {})", s, beg + ); + assert!(s.ends_with(")")); + let s: ArrayVec<[TinyString; 64]> + = s.replace(beg.as_str(), "") + .replace(")", "").split(",") + .map(|v| TinyString::from(v.trim()).unwrap()) + .collect(); + assert_eq!(s.len(), $id::::lanes()); + for (index, ss) in s.into_iter().enumerate() { + let mut e = TinyString::new(); + write!(&mut e, "{:?}", v.extract(index)).unwrap(); + assert_eq!(ss, e); + } + } + } + } + } + + impl Default for $id { + #[inline] + fn default() -> Self { + // FIXME: ptrs do not implement default + Self::null() + } + } + + test_if!{ + $test_tt: + paste::item! { + pub mod [<$id _default>] { + use super::*; + #[cfg_attr(not(target_arch = "wasm32"), test)] + #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] + fn default() { + let a = $id::::default(); + for i in 0..$id::::lanes() { + assert_eq!( + a.extract(i), unsafe { crate::mem::zeroed() } + ); + } + } + } + } + } + + impl $id { + /// Lane-wise equality comparison. + #[inline] + pub fn eq(self, other: Self) -> $mask_ty { + unsafe { + use crate::llvm::simd_eq; + let a: $usize_ty = crate::mem::transmute(self); + let b: $usize_ty = crate::mem::transmute(other); + Simd(simd_eq(a.0, b.0)) + } + } + + /// Lane-wise inequality comparison. + #[inline] + pub fn ne(self, other: Self) -> $mask_ty { + unsafe { + use crate::llvm::simd_ne; + let a: $usize_ty = crate::mem::transmute(self); + let b: $usize_ty = crate::mem::transmute(other); + Simd(simd_ne(a.0, b.0)) + } + } + + /// Lane-wise less-than comparison. + #[inline] + pub fn lt(self, other: Self) -> $mask_ty { + unsafe { + use crate::llvm::simd_lt; + let a: $usize_ty = crate::mem::transmute(self); + let b: $usize_ty = crate::mem::transmute(other); + Simd(simd_lt(a.0, b.0)) + } + } + + /// Lane-wise less-than-or-equals comparison. + #[inline] + pub fn le(self, other: Self) -> $mask_ty { + unsafe { + use crate::llvm::simd_le; + let a: $usize_ty = crate::mem::transmute(self); + let b: $usize_ty = crate::mem::transmute(other); + Simd(simd_le(a.0, b.0)) + } + } + + /// Lane-wise greater-than comparison. + #[inline] + pub fn gt(self, other: Self) -> $mask_ty { + unsafe { + use crate::llvm::simd_gt; + let a: $usize_ty = crate::mem::transmute(self); + let b: $usize_ty = crate::mem::transmute(other); + Simd(simd_gt(a.0, b.0)) + } + } + + /// Lane-wise greater-than-or-equals comparison. + #[inline] + pub fn ge(self, other: Self) -> $mask_ty { + unsafe { + use crate::llvm::simd_ge; + let a: $usize_ty = crate::mem::transmute(self); + let b: $usize_ty = crate::mem::transmute(other); + Simd(simd_ge(a.0, b.0)) + } + } + } + + test_if!{ + $test_tt: + paste::item! { + pub mod [<$id _cmp_vertical>] { + use super::*; + #[cfg_attr(not(target_arch = "wasm32"), test)] + #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] + fn cmp() { + let a = $id::::null(); + let b = $id::::splat(unsafe { + crate::mem::transmute(1_isize) + }); + + let r = a.lt(b); + let e = $mask_ty::splat(true); + assert!(r == e); + let r = a.le(b); + assert!(r == e); + + let e = $mask_ty::splat(false); + let r = a.gt(b); + assert!(r == e); + let r = a.ge(b); + assert!(r == e); + let r = a.eq(b); + assert!(r == e); + + let mut a = a; + let mut b = b; + let mut e = e; + for i in 0..$id::::lanes() { + if i % 2 == 0 { + a = a.replace( + i, + unsafe { crate::mem::transmute(0_isize) } + ); + b = b.replace( + i, + unsafe { crate::mem::transmute(1_isize) } + ); + e = e.replace(i, true); + } else { + a = a.replace( + i, + unsafe { crate::mem::transmute(1_isize) } + ); + b = b.replace( + i, + unsafe { crate::mem::transmute(0_isize) } + ); + e = e.replace(i, false); + } + } + let r = a.lt(b); + assert!(r == e); + } + } + } + } + + #[allow(clippy::partialeq_ne_impl)] + impl crate::cmp::PartialEq<$id> for $id { + #[inline] + fn eq(&self, other: &Self) -> bool { + $id::::eq(*self, *other).all() + } + #[inline] + fn ne(&self, other: &Self) -> bool { + $id::::ne(*self, *other).any() + } + } + + // FIXME: https://github.com/rust-lang-nursery/rust-clippy/issues/2892 + #[allow(clippy::partialeq_ne_impl)] + impl crate::cmp::PartialEq>> + for LexicographicallyOrdered<$id> + { + #[inline] + fn eq(&self, other: &Self) -> bool { + self.0 == other.0 + } + #[inline] + fn ne(&self, other: &Self) -> bool { + self.0 != other.0 + } + } + + test_if!{ + $test_tt: + paste::item! { + pub mod [<$id _cmp_PartialEq>] { + use super::*; + #[cfg_attr(not(target_arch = "wasm32"), test)] + #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] + fn partial_eq() { + let a = $id::::null(); + let b = $id::::splat(unsafe { + crate::mem::transmute(1_isize) + }); + + assert!(a != b); + assert!(!(a == b)); + assert!(a == a); + assert!(!(a != a)); + + if $id::::lanes() > 1 { + let a = $id::::null().replace(0, unsafe { + crate::mem::transmute(1_isize) + }); + let b = $id::::splat(unsafe { + crate::mem::transmute(1_isize) + }); + + assert!(a != b); + assert!(!(a == b)); + assert!(a == a); + assert!(!(a != a)); + } + } + } + } + } + + impl crate::cmp::Eq for $id {} + impl crate::cmp::Eq for LexicographicallyOrdered<$id> {} + + test_if!{ + $test_tt: + paste::item! { + pub mod [<$id _cmp_eq>] { + use super::*; + #[cfg_attr(not(target_arch = "wasm32"), test)] + #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] + fn eq() { + fn foo(_: E) {} + let a = $id::::null(); + foo(a); + } + } + } + } + + impl From<[$elem_ty; $elem_count]> for $id { + #[inline] + fn from(array: [$elem_ty; $elem_count]) -> Self { + unsafe { + // FIXME: unnecessary zeroing; better than UB. + let mut u: Self = crate::mem::zeroed(); + crate::ptr::copy_nonoverlapping( + &array as *const [$elem_ty; $elem_count] as *const u8, + &mut u as *mut Self as *mut u8, + crate::mem::size_of::() + ); + u + } + } + } + impl Into<[$elem_ty; $elem_count]> for $id { + #[inline] + fn into(self) -> [$elem_ty; $elem_count] { + unsafe { + // FIXME: unnecessary zeroing; better than UB. + let mut u: [$elem_ty; $elem_count] = crate::mem::zeroed(); + crate::ptr::copy_nonoverlapping( + &self as *const $id as *const u8, + &mut u as *mut [$elem_ty; $elem_count] as *mut u8, + crate::mem::size_of::() + ); + u + } + } + } + + test_if!{ + $test_tt: + paste::item! { + pub mod [<$id _from>] { + use super::*; + #[cfg_attr(not(target_arch = "wasm32"), test)] + #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] + fn array() { + let values = [1_i32; $elem_count]; + + let mut vec: $id = Default::default(); + let mut array = [ + $id::::null().extract(0); $elem_count + ]; + + for i in 0..$elem_count { + let ptr = &values[i] as *const i32 as *mut i32; + vec = vec.replace(i, ptr); + array[i] = ptr; + } + + // FIXME: there is no impl of From<$id> for [$elem_ty; N] + // let a0 = From::from(vec); + // assert_eq!(a0, array); + #[allow(unused_assignments)] + let mut a1 = array; + a1 = vec.into(); + assert_eq!(a1, array); + + let v0: $id = From::from(array); + assert_eq!(v0, vec); + let v1: $id = array.into(); + assert_eq!(v1, vec); + } + } + } + } + + impl $id { + /// Instantiates a new vector with the values of the `slice`. + /// + /// # Panics + /// + /// If `slice.len() < Self::lanes()` or `&slice[0]` is not aligned + /// to an `align_of::()` boundary. + #[inline] + pub fn from_slice_aligned(slice: &[$elem_ty]) -> Self { + unsafe { + assert!(slice.len() >= $elem_count); + let target_ptr = slice.as_ptr(); + assert!( + target_ptr.align_offset(crate::mem::align_of::()) + == 0 + ); + Self::from_slice_aligned_unchecked(slice) + } + } + + /// Instantiates a new vector with the values of the `slice`. + /// + /// # Panics + /// + /// If `slice.len() < Self::lanes()`. + #[inline] + pub fn from_slice_unaligned(slice: &[$elem_ty]) -> Self { + unsafe { + assert!(slice.len() >= $elem_count); + Self::from_slice_unaligned_unchecked(slice) + } + } + + /// Instantiates a new vector with the values of the `slice`. + /// + /// # Safety + /// + /// If `slice.len() < Self::lanes()` or `&slice[0]` is not aligned + /// to an `align_of::()` boundary, the behavior is undefined. + #[inline] + pub unsafe fn from_slice_aligned_unchecked(slice: &[$elem_ty]) + -> Self { + #[allow(clippy::cast_ptr_alignment)] + *(slice.as_ptr().cast()) + } + + /// Instantiates a new vector with the values of the `slice`. + /// + /// # Safety + /// + /// If `slice.len() < Self::lanes()` the behavior is undefined. + #[inline] + pub unsafe fn from_slice_unaligned_unchecked( + slice: &[$elem_ty], + ) -> Self { + use crate::mem::size_of; + let target_ptr = slice.as_ptr().cast(); + let mut x = Self::splat(crate::ptr::null_mut() as $elem_ty); + let self_ptr = &mut x as *mut Self as *mut u8; + crate::ptr::copy_nonoverlapping( + target_ptr, + self_ptr, + size_of::(), + ); + x + } + } + + test_if!{ + $test_tt: + paste::item! { + pub mod [<$id _slice_from_slice>] { + use super::*; + use crate::iter::Iterator; + + #[cfg_attr(not(target_arch = "wasm32"), test)] + #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] + fn from_slice_unaligned() { + let (null, non_null) = ptr_vals!($id); + + let mut unaligned = [ + non_null; $id::::lanes() + 1 + ]; + unaligned[0] = null; + let vec = $id::::from_slice_unaligned( + &unaligned[1..] + ); + for (index, &b) in unaligned.iter().enumerate() { + if index == 0 { + assert_eq!(b, null); + } else { + assert_eq!(b, non_null); + assert_eq!(b, vec.extract(index - 1)); + } + } + } + + // FIXME: wasm-bindgen-test does not support #[should_panic] + // #[cfg_attr(not(target_arch = "wasm32"), test)] + // #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] + #[cfg(not(target_arch = "wasm32"))] + #[test] + #[should_panic] + fn from_slice_unaligned_fail() { + let (_null, non_null) = ptr_vals!($id); + let unaligned = [non_null; $id::::lanes() + 1]; + // the slice is not large enough => panic + let _vec = $id::::from_slice_unaligned( + &unaligned[2..] + ); + } + + union A { + data: [<$id as sealed::Simd>::Element; + 2 * $id::::lanes()], + _vec: $id, + } + + #[cfg_attr(not(target_arch = "wasm32"), test)] + #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] + fn from_slice_aligned() { + let (null, non_null) = ptr_vals!($id); + let mut aligned = A { + data: [null; 2 * $id::::lanes()], + }; + for i in + $id::::lanes()..(2 * $id::::lanes()) { + unsafe { + aligned.data[i] = non_null; + } + } + + let vec = unsafe { + $id::::from_slice_aligned( + &aligned.data[$id::::lanes()..] + ) + }; + for (index, &b) in unsafe { + aligned.data.iter().enumerate() + } { + if index < $id::::lanes() { + assert_eq!(b, null); + } else { + assert_eq!(b, non_null); + assert_eq!( + b, vec.extract(index - $id::::lanes()) + ); + } + } + } + + // FIXME: wasm-bindgen-test does not support #[should_panic] + // #[cfg_attr(not(target_arch = "wasm32"), test)] + // #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] + #[cfg(not(target_arch = "wasm32"))] + #[test] + #[should_panic] + fn from_slice_aligned_fail_lanes() { + let (_null, non_null) = ptr_vals!($id); + let aligned = A { + data: [non_null; 2 * $id::::lanes()], + }; + // the slice is not large enough => panic + let _vec = unsafe { + $id::::from_slice_aligned( + &aligned.data[2 * $id::::lanes()..] + ) + }; + } + + // FIXME: wasm-bindgen-test does not support #[should_panic] + // #[cfg_attr(not(target_arch = "wasm32"), test)] + // #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] + #[cfg(not(target_arch = "wasm32"))] + #[test] + #[should_panic] + fn from_slice_aligned_fail_align() { + unsafe { + let (null, _non_null) = ptr_vals!($id); + let aligned = A { + data: [null; 2 * $id::::lanes()], + }; + + // get a pointer to the front of data + let ptr = aligned.data.as_ptr(); + // offset pointer by one element + let ptr = ptr.wrapping_add(1); + + if ptr.align_offset( + crate::mem::align_of::<$id>() + ) == 0 { + // the pointer is properly aligned, so + // from_slice_aligned won't fail here (e.g. this + // can happen for i128x1). So we panic to make + // the "should_fail" test pass: + panic!("ok"); + } + + // create a slice - this is safe, because the + // elements of the slice exist, are properly + // initialized, and properly aligned: + let s = slice::from_raw_parts( + ptr, $id::::lanes() + ); + // this should always panic because the slice + // alignment does not match the alignment + // requirements for the vector type: + let _vec = $id::::from_slice_aligned(s); + } + } + } + } + } + + impl $id { + /// Writes the values of the vector to the `slice`. + /// + /// # Panics + /// + /// If `slice.len() < Self::lanes()` or `&slice[0]` is not + /// aligned to an `align_of::()` boundary. + #[inline] + pub fn write_to_slice_aligned(self, slice: &mut [$elem_ty]) { + unsafe { + assert!(slice.len() >= $elem_count); + let target_ptr = slice.as_mut_ptr(); + assert!( + target_ptr.align_offset(crate::mem::align_of::()) + == 0 + ); + self.write_to_slice_aligned_unchecked(slice); + } + } + + /// Writes the values of the vector to the `slice`. + /// + /// # Panics + /// + /// If `slice.len() < Self::lanes()`. + #[inline] + pub fn write_to_slice_unaligned(self, slice: &mut [$elem_ty]) { + unsafe { + assert!(slice.len() >= $elem_count); + self.write_to_slice_unaligned_unchecked(slice); + } + } + + /// Writes the values of the vector to the `slice`. + /// + /// # Safety + /// + /// If `slice.len() < Self::lanes()` or `&slice[0]` is not + /// aligned to an `align_of::()` boundary, the behavior is + /// undefined. + #[inline] + pub unsafe fn write_to_slice_aligned_unchecked( + self, slice: &mut [$elem_ty], + ) { + #[allow(clippy::cast_ptr_alignment)] + *(slice.as_mut_ptr().cast()) = self; + } + + /// Writes the values of the vector to the `slice`. + /// + /// # Safety + /// + /// If `slice.len() < Self::lanes()` the behavior is undefined. + #[inline] + pub unsafe fn write_to_slice_unaligned_unchecked( + self, slice: &mut [$elem_ty], + ) { + let target_ptr = slice.as_mut_ptr().cast(); + let self_ptr = &self as *const Self as *const u8; + crate::ptr::copy_nonoverlapping( + self_ptr, + target_ptr, + crate::mem::size_of::(), + ); + } + } + + test_if!{ + $test_tt: + paste::item! { + pub mod [<$id _slice_write_to_slice>] { + use super::*; + use crate::iter::Iterator; + + #[cfg_attr(not(target_arch = "wasm32"), test)] + #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] + fn write_to_slice_unaligned() { + let (null, non_null) = ptr_vals!($id); + let mut unaligned = [null; $id::::lanes() + 1]; + let vec = $id::::splat(non_null); + vec.write_to_slice_unaligned(&mut unaligned[1..]); + for (index, &b) in unaligned.iter().enumerate() { + if index == 0 { + assert_eq!(b, null); + } else { + assert_eq!(b, non_null); + assert_eq!(b, vec.extract(index - 1)); + } + } + } + + // FIXME: wasm-bindgen-test does not support #[should_panic] + // #[cfg_attr(not(target_arch = "wasm32"), test)] + // #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] + #[cfg(not(target_arch = "wasm32"))] + #[test] + #[should_panic] + fn write_to_slice_unaligned_fail() { + let (null, non_null) = ptr_vals!($id); + let mut unaligned = [null; $id::::lanes() + 1]; + let vec = $id::::splat(non_null); + // the slice is not large enough => panic + vec.write_to_slice_unaligned(&mut unaligned[2..]); + } + + union A { + data: [<$id as sealed::Simd>::Element; + 2 * $id::::lanes()], + _vec: $id, + } + + #[cfg_attr(not(target_arch = "wasm32"), test)] + #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] + fn write_to_slice_aligned() { + let (null, non_null) = ptr_vals!($id); + let mut aligned = A { + data: [null; 2 * $id::::lanes()], + }; + let vec = $id::::splat(non_null); + unsafe { + vec.write_to_slice_aligned( + &mut aligned.data[$id::::lanes()..] + ) + }; + for (index, &b) in + unsafe { aligned.data.iter().enumerate() } { + if index < $id::::lanes() { + assert_eq!(b, null); + } else { + assert_eq!(b, non_null); + assert_eq!( + b, vec.extract(index - $id::::lanes()) + ); + } + } + } + + // FIXME: wasm-bindgen-test does not support #[should_panic] + // #[cfg_attr(not(target_arch = "wasm32"), test)] + // #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] + #[cfg(not(target_arch = "wasm32"))] + #[test] + #[should_panic] + fn write_to_slice_aligned_fail_lanes() { + let (null, non_null) = ptr_vals!($id); + let mut aligned = A { + data: [null; 2 * $id::::lanes()], + }; + let vec = $id::::splat(non_null); + // the slice is not large enough => panic + unsafe { + vec.write_to_slice_aligned( + &mut aligned.data[2 * $id::::lanes()..] + ) + }; + } + + // FIXME: wasm-bindgen-test does not support #[should_panic] + // #[cfg_attr(not(target_arch = "wasm32"), test)] + // #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] + #[cfg(not(target_arch = "wasm32"))] + #[test] + #[should_panic] + fn write_to_slice_aligned_fail_align() { + let (null, non_null) = ptr_vals!($id); + unsafe { + let mut aligned = A { + data: [null; 2 * $id::::lanes()], + }; + + // get a pointer to the front of data + let ptr = aligned.data.as_mut_ptr(); + // offset pointer by one element + let ptr = ptr.wrapping_add(1); + + if ptr.align_offset( + crate::mem::align_of::<$id>() + ) == 0 { + // the pointer is properly aligned, so + // write_to_slice_aligned won't fail here (e.g. + // this can happen for i128x1). So we panic to + // make the "should_fail" test pass: + panic!("ok"); + } + + // create a slice - this is safe, because the + // elements of the slice exist, are properly + // initialized, and properly aligned: + let s = slice::from_raw_parts_mut( + ptr, $id::::lanes() + ); + // this should always panic because the slice + // alignment does not match the alignment + // requirements for the vector type: + let vec = $id::::splat(non_null); + vec.write_to_slice_aligned(s); + } + } + } + } + } + + impl crate::hash::Hash for $id { + #[inline] + fn hash(&self, state: &mut H) { + let s: $usize_ty = unsafe { crate::mem::transmute(*self) }; + s.hash(state) + } + } + + test_if! { + $test_tt: + paste::item! { + pub mod [<$id _hash>] { + use super::*; + #[cfg_attr(not(target_arch = "wasm32"), test)] + #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] + fn hash() { + use crate::hash::{Hash, Hasher}; + #[allow(deprecated)] + use crate::hash::{SipHasher13}; + + let values = [1_i32; $elem_count]; + + let mut vec: $id = Default::default(); + let mut array = [ + $id::::null().extract(0); + $elem_count + ]; + + for i in 0..$elem_count { + let ptr = &values[i] as *const i32 as *mut i32; + vec = vec.replace(i, ptr); + array[i] = ptr; + } + + #[allow(deprecated)] + let mut a_hash = SipHasher13::new(); + let mut v_hash = a_hash.clone(); + array.hash(&mut a_hash); + vec.hash(&mut v_hash); + assert_eq!(a_hash.finish(), v_hash.finish()); + } + } + } + } + + impl $id { + /// Calculates the offset from a pointer. + /// + /// `count` is in units of `T`; e.g. a count of `3` represents a + /// pointer offset of `3 * size_of::()` bytes. + /// + /// # Safety + /// + /// If any of the following conditions are violated, the result is + /// Undefined Behavior: + /// + /// * Both the starting and resulting pointer must be either in + /// bounds or one byte past the end of an allocated object. + /// + /// * The computed offset, in bytes, cannot overflow an `isize`. + /// + /// * The offset being in bounds cannot rely on "wrapping around" + /// the address space. That is, the infinite-precision sum, in bytes + /// must fit in a `usize`. + /// + /// The compiler and standard library generally tries to ensure + /// allocations never reach a size where an offset is a concern. For + /// instance, `Vec` and `Box` ensure they never allocate more than + /// `isize::MAX` bytes, so `vec.as_ptr().offset(vec.len() as isize)` + /// is always safe. + /// + /// Most platforms fundamentally can't even construct such an + /// allocation. For instance, no known 64-bit platform can ever + /// serve a request for 263 bytes due to page-table limitations or + /// splitting the address space. However, some 32-bit and 16-bit + /// platforms may successfully serve a request for more than + /// `isize::MAX` bytes with things like Physical Address Extension. + /// As such, memory acquired directly from allocators or memory + /// mapped files may be too large to handle with this function. + /// + /// Consider using `wrapping_offset` instead if these constraints + /// are difficult to satisfy. The only advantage of this method is + /// that it enables more aggressive compiler optimizations. + #[inline] + pub unsafe fn offset(self, count: $isize_ty) -> Self { + // FIXME: should use LLVM's `add nsw nuw` + self.wrapping_offset(count) + } + + /// Calculates the offset from a pointer using wrapping arithmetic. + /// + /// `count` is in units of `T`; e.g. a count of `3` represents a + /// pointer offset of `3 * size_of::()` bytes. + /// + /// # Safety + /// + /// The resulting pointer does not need to be in bounds, but it is + /// potentially hazardous to dereference (which requires unsafe). + /// + /// Always use `.offset(count)` instead when possible, because + /// offset allows the compiler to optimize better. + #[inline] + pub fn wrapping_offset(self, count: $isize_ty) -> Self { + unsafe { + let x: $isize_ty = crate::mem::transmute(self); + // note: {+,*} currently performs a `wrapping_{add, mul}` + crate::mem::transmute( + x + (count * crate::mem::size_of::() as isize) + ) + } + } + + /// Calculates the distance between two pointers. + /// + /// The returned value is in units of `T`: the distance in bytes is + /// divided by `mem::size_of::()`. + /// + /// This function is the inverse of offset. + /// + /// # Safety + /// + /// If any of the following conditions are violated, the result is + /// Undefined Behavior: + /// + /// * Both the starting and other pointer must be either in bounds + /// or one byte past the end of the same allocated object. + /// + /// * The distance between the pointers, in bytes, cannot overflow + /// an `isize`. + /// + /// * The distance between the pointers, in bytes, must be an exact + /// multiple of the size of `T`. + /// + /// * The distance being in bounds cannot rely on "wrapping around" + /// the address space. + /// + /// The compiler and standard library generally try to ensure + /// allocations never reach a size where an offset is a concern. For + /// instance, `Vec` and `Box` ensure they never allocate more than + /// `isize::MAX` bytes, so `ptr_into_vec.offset_from(vec.as_ptr())` + /// is always safe. + /// + /// Most platforms fundamentally can't even construct such an + /// allocation. For instance, no known 64-bit platform can ever + /// serve a request for 263 bytes due to page-table limitations or + /// splitting the address space. However, some 32-bit and 16-bit + /// platforms may successfully serve a request for more than + /// `isize::MAX` bytes with things like Physical Address Extension. + /// As such, memory acquired directly from allocators or memory + /// mapped files may be too large to handle with this function. + /// + /// Consider using `wrapping_offset_from` instead if these constraints + /// are difficult to satisfy. The only advantage of this method is + /// that it enables more aggressive compiler optimizations. + #[inline] + pub unsafe fn offset_from(self, origin: Self) -> $isize_ty { + // FIXME: should use LLVM's `sub nsw nuw`. + self.wrapping_offset_from(origin) + } + + /// Calculates the distance between two pointers. + /// + /// The returned value is in units of `T`: the distance in bytes is + /// divided by `mem::size_of::()`. + /// + /// If the address different between the two pointers is not a + /// multiple of `mem::size_of::()` then the result of the + /// division is rounded towards zero. + /// + /// Though this method is safe for any two pointers, note that its + /// result will be mostly useless if the two pointers aren't into + /// the same allocated object, for example if they point to two + /// different local variables. + #[inline] + pub fn wrapping_offset_from(self, origin: Self) -> $isize_ty { + let x: $isize_ty = unsafe { crate::mem::transmute(self) }; + let y: $isize_ty = unsafe { crate::mem::transmute(origin) }; + // note: {-,/} currently perform wrapping_{sub, div} + (y - x) / (crate::mem::size_of::() as isize) + } + + /// Calculates the offset from a pointer (convenience for + /// `.offset(count as isize)`). + /// + /// `count` is in units of `T`; e.g. a count of 3 represents a + /// pointer offset of `3 * size_of::()` bytes. + /// + /// # Safety + /// + /// If any of the following conditions are violated, the result is + /// Undefined Behavior: + /// + /// * Both the starting and resulting pointer must be either in + /// bounds or one byte past the end of an allocated object. + /// + /// * The computed offset, in bytes, cannot overflow an `isize`. + /// + /// * The offset being in bounds cannot rely on "wrapping around" + /// the address space. That is, the infinite-precision sum must fit + /// in a `usize`. + /// + /// The compiler and standard library generally tries to ensure + /// allocations never reach a size where an offset is a concern. For + /// instance, `Vec` and `Box` ensure they never allocate more than + /// `isize::MAX` bytes, so `vec.as_ptr().add(vec.len())` is always + /// safe. + /// + /// Most platforms fundamentally can't even construct such an + /// allocation. For instance, no known 64-bit platform can ever + /// serve a request for 263 bytes due to page-table limitations or + /// splitting the address space. However, some 32-bit and 16-bit + /// platforms may successfully serve a request for more than + /// `isize::MAX` bytes with things like Physical Address Extension. + /// As such, memory acquired directly from allocators or memory + /// mapped files may be too large to handle with this function. + /// + /// Consider using `wrapping_offset` instead if these constraints + /// are difficult to satisfy. The only advantage of this method is + /// that it enables more aggressive compiler optimizations. + #[inline] + #[allow(clippy::should_implement_trait)] + pub unsafe fn add(self, count: $usize_ty) -> Self { + self.offset(count.cast()) + } + + /// Calculates the offset from a pointer (convenience for + /// `.offset((count as isize).wrapping_neg())`). + /// + /// `count` is in units of T; e.g. a `count` of 3 represents a + /// pointer offset of `3 * size_of::()` bytes. + /// + /// # Safety + /// + /// If any of the following conditions are violated, the result is + /// Undefined Behavior: + /// + /// * Both the starting and resulting pointer must be either in + /// bounds or one byte past the end of an allocated object. + /// + /// * The computed offset cannot exceed `isize::MAX` **bytes**. + /// + /// * The offset being in bounds cannot rely on "wrapping around" + /// the address space. That is, the infinite-precision sum must fit + /// in a usize. + /// + /// The compiler and standard library generally tries to ensure + /// allocations never reach a size where an offset is a concern. For + /// instance, `Vec` and `Box` ensure they never allocate more than + /// `isize::MAX` bytes, so + /// `vec.as_ptr().add(vec.len()).sub(vec.len())` is always safe. + /// + /// Most platforms fundamentally can't even construct such an + /// allocation. For instance, no known 64-bit platform can ever + /// serve a request for 263 bytes due to page-table + /// limitations or splitting the address space. However, some 32-bit + /// and 16-bit platforms may successfully serve a request for more + /// than `isize::MAX` bytes with things like Physical Address + /// Extension. As such, memory acquired directly from allocators or + /// memory mapped files *may* be too large to handle with this + /// function. + /// + /// Consider using `wrapping_offset` instead if these constraints + /// are difficult to satisfy. The only advantage of this method is + /// that it enables more aggressive compiler optimizations. + #[inline] + #[allow(clippy::should_implement_trait)] + pub unsafe fn sub(self, count: $usize_ty) -> Self { + let x: $isize_ty = count.cast(); + // note: - is currently wrapping_neg + self.offset(-x) + } + + /// Calculates the offset from a pointer using wrapping arithmetic. + /// (convenience for `.wrapping_offset(count as isize)`) + /// + /// `count` is in units of T; e.g. a `count` of 3 represents a + /// pointer offset of `3 * size_of::()` bytes. + /// + /// # Safety + /// + /// The resulting pointer does not need to be in bounds, but it is + /// potentially hazardous to dereference (which requires `unsafe`). + /// + /// Always use `.add(count)` instead when possible, because `add` + /// allows the compiler to optimize better. + #[inline] + pub fn wrapping_add(self, count: $usize_ty) -> Self { + self.wrapping_offset(count.cast()) + } + + /// Calculates the offset from a pointer using wrapping arithmetic. + /// (convenience for `.wrapping_offset((count as + /// isize).wrapping_sub())`) + /// + /// `count` is in units of T; e.g. a `count` of 3 represents a + /// pointer offset of `3 * size_of::()` bytes. + /// + /// # Safety + /// + /// The resulting pointer does not need to be in bounds, but it is + /// potentially hazardous to dereference (which requires `unsafe`). + /// + /// Always use `.sub(count)` instead when possible, because `sub` + /// allows the compiler to optimize better. + #[inline] + pub fn wrapping_sub(self, count: $usize_ty) -> Self { + let x: $isize_ty = count.cast(); + self.wrapping_offset(-1 * x) + } + } + + impl $id { + /// Shuffle vector elements according to `indices`. + #[inline] + pub fn shuffle1_dyn(self, indices: I) -> Self + where + Self: codegen::shuffle1_dyn::Shuffle1Dyn, + { + codegen::shuffle1_dyn::Shuffle1Dyn::shuffle1_dyn(self, indices) + } + } + + test_if! { + $test_tt: + paste::item! { + pub mod [<$id _shuffle1_dyn>] { + use super::*; + #[cfg_attr(not(target_arch = "wasm32"), test)] + #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] + fn shuffle1_dyn() { + let (null, non_null) = ptr_vals!($id); + + // alternating = [non_null, null, non_null, null, ...] + let mut alternating = $id::::splat(null); + for i in 0..$id::::lanes() { + if i % 2 == 0 { + alternating = alternating.replace(i, non_null); + } + } + + type Indices = <$id + as codegen::shuffle1_dyn::Shuffle1Dyn>::Indices; + // even = [0, 0, 2, 2, 4, 4, ..] + let even = { + let mut v = Indices::splat(0); + for i in 0..$id::::lanes() { + if i % 2 == 0 { + v = v.replace(i, (i as u8).into()); + } else { + v = v.replace(i, (i as u8 - 1).into()); + } + } + v + }; + // odd = [1, 1, 3, 3, 5, 5, ...] + let odd = { + let mut v = Indices::splat(0); + for i in 0..$id::::lanes() { + if i % 2 != 0 { + v = v.replace(i, (i as u8).into()); + } else { + v = v.replace(i, (i as u8 + 1).into()); + } + } + v + }; + + assert_eq!( + alternating.shuffle1_dyn(even), + $id::::splat(non_null) + ); + if $id::::lanes() > 1 { + assert_eq!( + alternating.shuffle1_dyn(odd), + $id::::splat(null) + ); + } + } + } + } + } + }; +} diff --git a/third_party/rust/packed_simd/src/api/ops.rs b/third_party/rust/packed_simd/src/api/ops.rs new file mode 100644 index 000000000000..f71c98795da3 --- /dev/null +++ b/third_party/rust/packed_simd/src/api/ops.rs @@ -0,0 +1,32 @@ +//! Implementation of the `ops` traits +#[macro_use] +mod vector_mask_bitwise; +#[macro_use] +mod scalar_mask_bitwise; + +#[macro_use] +mod vector_arithmetic; +#[macro_use] +mod scalar_arithmetic; + +#[macro_use] +mod vector_bitwise; +#[macro_use] +mod scalar_bitwise; + +#[macro_use] +mod vector_shifts; +#[macro_use] +mod scalar_shifts; + +#[macro_use] +mod vector_rotates; + +#[macro_use] +mod vector_neg; + +#[macro_use] +mod vector_int_min_max; + +#[macro_use] +mod vector_float_min_max; diff --git a/third_party/rust/packed_simd/src/api/ops/scalar_arithmetic.rs b/third_party/rust/packed_simd/src/api/ops/scalar_arithmetic.rs new file mode 100644 index 000000000000..da1a2037eaaf --- /dev/null +++ b/third_party/rust/packed_simd/src/api/ops/scalar_arithmetic.rs @@ -0,0 +1,203 @@ +//! Vertical (lane-wise) vector-scalar / scalar-vector arithmetic operations. + +macro_rules! impl_ops_scalar_arithmetic { + ([$elem_ty:ident; $elem_count:expr]: $id:ident | $test_tt:tt) => { + impl crate::ops::Add<$elem_ty> for $id { + type Output = Self; + #[inline] + fn add(self, other: $elem_ty) -> Self { + self + $id::splat(other) + } + } + impl crate::ops::Add<$id> for $elem_ty { + type Output = $id; + #[inline] + fn add(self, other: $id) -> $id { + $id::splat(self) + other + } + } + + impl crate::ops::Sub<$elem_ty> for $id { + type Output = Self; + #[inline] + fn sub(self, other: $elem_ty) -> Self { + self - $id::splat(other) + } + } + impl crate::ops::Sub<$id> for $elem_ty { + type Output = $id; + #[inline] + fn sub(self, other: $id) -> $id { + $id::splat(self) - other + } + } + + impl crate::ops::Mul<$elem_ty> for $id { + type Output = Self; + #[inline] + fn mul(self, other: $elem_ty) -> Self { + self * $id::splat(other) + } + } + impl crate::ops::Mul<$id> for $elem_ty { + type Output = $id; + #[inline] + fn mul(self, other: $id) -> $id { + $id::splat(self) * other + } + } + + impl crate::ops::Div<$elem_ty> for $id { + type Output = Self; + #[inline] + fn div(self, other: $elem_ty) -> Self { + self / $id::splat(other) + } + } + impl crate::ops::Div<$id> for $elem_ty { + type Output = $id; + #[inline] + fn div(self, other: $id) -> $id { + $id::splat(self) / other + } + } + + impl crate::ops::Rem<$elem_ty> for $id { + type Output = Self; + #[inline] + fn rem(self, other: $elem_ty) -> Self { + self % $id::splat(other) + } + } + impl crate::ops::Rem<$id> for $elem_ty { + type Output = $id; + #[inline] + fn rem(self, other: $id) -> $id { + $id::splat(self) % other + } + } + + impl crate::ops::AddAssign<$elem_ty> for $id { + #[inline] + fn add_assign(&mut self, other: $elem_ty) { + *self = *self + other; + } + } + + impl crate::ops::SubAssign<$elem_ty> for $id { + #[inline] + fn sub_assign(&mut self, other: $elem_ty) { + *self = *self - other; + } + } + + impl crate::ops::MulAssign<$elem_ty> for $id { + #[inline] + fn mul_assign(&mut self, other: $elem_ty) { + *self = *self * other; + } + } + + impl crate::ops::DivAssign<$elem_ty> for $id { + #[inline] + fn div_assign(&mut self, other: $elem_ty) { + *self = *self / other; + } + } + + impl crate::ops::RemAssign<$elem_ty> for $id { + #[inline] + fn rem_assign(&mut self, other: $elem_ty) { + *self = *self % other; + } + } + + test_if!{ + $test_tt: + paste::item! { + pub mod [<$id _ops_scalar_arith>] { + use super::*; + #[cfg_attr(not(target_arch = "wasm32"), test)] #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] + fn ops_scalar_arithmetic() { + let zi = 0 as $elem_ty; + let oi = 1 as $elem_ty; + let ti = 2 as $elem_ty; + let fi = 4 as $elem_ty; + let z = $id::splat(zi); + let o = $id::splat(oi); + let t = $id::splat(ti); + let f = $id::splat(fi); + + // add + assert_eq!(zi + z, z); + assert_eq!(z + zi, z); + assert_eq!(oi + z, o); + assert_eq!(o + zi, o); + assert_eq!(ti + z, t); + assert_eq!(t + zi, t); + assert_eq!(ti + t, f); + assert_eq!(t + ti, f); + // sub + assert_eq!(zi - z, z); + assert_eq!(z - zi, z); + assert_eq!(oi - z, o); + assert_eq!(o - zi, o); + assert_eq!(ti - z, t); + assert_eq!(t - zi, t); + assert_eq!(fi - t, t); + assert_eq!(f - ti, t); + assert_eq!(f - o - o, t); + assert_eq!(f - oi - oi, t); + // mul + assert_eq!(zi * z, z); + assert_eq!(z * zi, z); + assert_eq!(zi * o, z); + assert_eq!(z * oi, z); + assert_eq!(zi * t, z); + assert_eq!(z * ti, z); + assert_eq!(oi * t, t); + assert_eq!(o * ti, t); + assert_eq!(ti * t, f); + assert_eq!(t * ti, f); + // div + assert_eq!(zi / o, z); + assert_eq!(z / oi, z); + assert_eq!(ti / o, t); + assert_eq!(t / oi, t); + assert_eq!(fi / o, f); + assert_eq!(f / oi, f); + assert_eq!(ti / t, o); + assert_eq!(t / ti, o); + assert_eq!(fi / t, t); + assert_eq!(f / ti, t); + // rem + assert_eq!(oi % o, z); + assert_eq!(o % oi, z); + assert_eq!(fi % t, z); + assert_eq!(f % ti, z); + + { + let mut v = z; + assert_eq!(v, z); + v += oi; // add_assign + assert_eq!(v, o); + v -= oi; // sub_assign + assert_eq!(v, z); + v = t; + v *= oi; // mul_assign + assert_eq!(v, t); + v *= ti; + assert_eq!(v, f); + v /= oi; // div_assign + assert_eq!(v, f); + v /= ti; + assert_eq!(v, t); + v %= ti; // rem_assign + assert_eq!(v, z); + } + } + } + } + } + }; +} diff --git a/third_party/rust/packed_simd/src/api/ops/scalar_bitwise.rs b/third_party/rust/packed_simd/src/api/ops/scalar_bitwise.rs new file mode 100644 index 000000000000..88216769aec4 --- /dev/null +++ b/third_party/rust/packed_simd/src/api/ops/scalar_bitwise.rs @@ -0,0 +1,162 @@ +//! Vertical (lane-wise) vector-scalar / scalar-vector bitwise operations. + +macro_rules! impl_ops_scalar_bitwise { + ( + [$elem_ty:ident; $elem_count:expr]: + $id:ident | $test_tt:tt | + ($true:expr, $false:expr) + ) => { + impl crate::ops::BitXor<$elem_ty> for $id { + type Output = Self; + #[inline] + fn bitxor(self, other: $elem_ty) -> Self { + self ^ $id::splat(other) + } + } + impl crate::ops::BitXor<$id> for $elem_ty { + type Output = $id; + #[inline] + fn bitxor(self, other: $id) -> $id { + $id::splat(self) ^ other + } + } + + impl crate::ops::BitAnd<$elem_ty> for $id { + type Output = Self; + #[inline] + fn bitand(self, other: $elem_ty) -> Self { + self & $id::splat(other) + } + } + impl crate::ops::BitAnd<$id> for $elem_ty { + type Output = $id; + #[inline] + fn bitand(self, other: $id) -> $id { + $id::splat(self) & other + } + } + + impl crate::ops::BitOr<$elem_ty> for $id { + type Output = Self; + #[inline] + fn bitor(self, other: $elem_ty) -> Self { + self | $id::splat(other) + } + } + impl crate::ops::BitOr<$id> for $elem_ty { + type Output = $id; + #[inline] + fn bitor(self, other: $id) -> $id { + $id::splat(self) | other + } + } + + impl crate::ops::BitAndAssign<$elem_ty> for $id { + #[inline] + fn bitand_assign(&mut self, other: $elem_ty) { + *self = *self & other; + } + } + impl crate::ops::BitOrAssign<$elem_ty> for $id { + #[inline] + fn bitor_assign(&mut self, other: $elem_ty) { + *self = *self | other; + } + } + impl crate::ops::BitXorAssign<$elem_ty> for $id { + #[inline] + fn bitxor_assign(&mut self, other: $elem_ty) { + *self = *self ^ other; + } + } + + test_if!{ + $test_tt: + paste::item! { + pub mod [<$id _ops_scalar_bitwise>] { + use super::*; + + #[cfg_attr(not(target_arch = "wasm32"), test)] #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] + fn ops_scalar_bitwise() { + let zi = 0 as $elem_ty; + let oi = 1 as $elem_ty; + let ti = 2 as $elem_ty; + let z = $id::splat(zi); + let o = $id::splat(oi); + let t = $id::splat(ti); + + // BitAnd: + assert_eq!(oi & o, o); + assert_eq!(o & oi, o); + assert_eq!(oi & z, z); + assert_eq!(o & zi, z); + assert_eq!(zi & o, z); + assert_eq!(z & oi, z); + assert_eq!(zi & z, z); + assert_eq!(z & zi, z); + + assert_eq!(ti & t, t); + assert_eq!(t & ti, t); + assert_eq!(ti & o, z); + assert_eq!(t & oi, z); + assert_eq!(oi & t, z); + assert_eq!(o & ti, z); + + // BitOr: + assert_eq!(oi | o, o); + assert_eq!(o | oi, o); + assert_eq!(oi | z, o); + assert_eq!(o | zi, o); + assert_eq!(zi | o, o); + assert_eq!(z | oi, o); + assert_eq!(zi | z, z); + assert_eq!(z | zi, z); + + assert_eq!(ti | t, t); + assert_eq!(t | ti, t); + assert_eq!(zi | t, t); + assert_eq!(z | ti, t); + assert_eq!(ti | z, t); + assert_eq!(t | zi, t); + + // BitXOR: + assert_eq!(oi ^ o, z); + assert_eq!(o ^ oi, z); + assert_eq!(zi ^ z, z); + assert_eq!(z ^ zi, z); + assert_eq!(zi ^ o, o); + assert_eq!(z ^ oi, o); + assert_eq!(oi ^ z, o); + assert_eq!(o ^ zi, o); + + assert_eq!(ti ^ t, z); + assert_eq!(t ^ ti, z); + assert_eq!(ti ^ z, t); + assert_eq!(t ^ zi, t); + assert_eq!(zi ^ t, t); + assert_eq!(z ^ ti, t); + + { + // AndAssign: + let mut v = o; + v &= ti; + assert_eq!(v, z); + } + { + // OrAssign: + let mut v = z; + v |= oi; + assert_eq!(v, o); + } + { + // XORAssign: + let mut v = z; + v ^= oi; + assert_eq!(v, o); + } + } + } + } + } + }; +} diff --git a/third_party/rust/packed_simd/src/api/ops/scalar_mask_bitwise.rs b/third_party/rust/packed_simd/src/api/ops/scalar_mask_bitwise.rs new file mode 100644 index 000000000000..523a85207b6b --- /dev/null +++ b/third_party/rust/packed_simd/src/api/ops/scalar_mask_bitwise.rs @@ -0,0 +1,140 @@ +//! Vertical (lane-wise) vector-vector bitwise operations. + +macro_rules! impl_ops_scalar_mask_bitwise { + ( + [$elem_ty:ident; $elem_count:expr]: + $id:ident | $test_tt:tt | + ($true:expr, $false:expr) + ) => { + impl crate::ops::BitXor for $id { + type Output = Self; + #[inline] + fn bitxor(self, other: bool) -> Self { + self ^ $id::splat(other) + } + } + impl crate::ops::BitXor<$id> for bool { + type Output = $id; + #[inline] + fn bitxor(self, other: $id) -> $id { + $id::splat(self) ^ other + } + } + + impl crate::ops::BitAnd for $id { + type Output = Self; + #[inline] + fn bitand(self, other: bool) -> Self { + self & $id::splat(other) + } + } + impl crate::ops::BitAnd<$id> for bool { + type Output = $id; + #[inline] + fn bitand(self, other: $id) -> $id { + $id::splat(self) & other + } + } + + impl crate::ops::BitOr for $id { + type Output = Self; + #[inline] + fn bitor(self, other: bool) -> Self { + self | $id::splat(other) + } + } + impl crate::ops::BitOr<$id> for bool { + type Output = $id; + #[inline] + fn bitor(self, other: $id) -> $id { + $id::splat(self) | other + } + } + + impl crate::ops::BitAndAssign for $id { + #[inline] + fn bitand_assign(&mut self, other: bool) { + *self = *self & other; + } + } + impl crate::ops::BitOrAssign for $id { + #[inline] + fn bitor_assign(&mut self, other: bool) { + *self = *self | other; + } + } + impl crate::ops::BitXorAssign for $id { + #[inline] + fn bitxor_assign(&mut self, other: bool) { + *self = *self ^ other; + } + } + + test_if!{ + $test_tt: + paste::item! { + pub mod [<$id _ops_scalar_mask_bitwise>] { + use super::*; + #[cfg_attr(not(target_arch = "wasm32"), test)] #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] + fn ops_scalar_mask_bitwise() { + let ti = true; + let fi = false; + let t = $id::splat(ti); + let f = $id::splat(fi); + assert!(t != f); + assert!(!(t == f)); + + // BitAnd: + assert_eq!(ti & f, f); + assert_eq!(t & fi, f); + assert_eq!(fi & t, f); + assert_eq!(f & ti, f); + assert_eq!(ti & t, t); + assert_eq!(t & ti, t); + assert_eq!(fi & f, f); + assert_eq!(f & fi, f); + + // BitOr: + assert_eq!(ti | f, t); + assert_eq!(t | fi, t); + assert_eq!(fi | t, t); + assert_eq!(f | ti, t); + assert_eq!(ti | t, t); + assert_eq!(t | ti, t); + assert_eq!(fi | f, f); + assert_eq!(f | fi, f); + + // BitXOR: + assert_eq!(ti ^ f, t); + assert_eq!(t ^ fi, t); + assert_eq!(fi ^ t, t); + assert_eq!(f ^ ti, t); + assert_eq!(ti ^ t, f); + assert_eq!(t ^ ti, f); + assert_eq!(fi ^ f, f); + assert_eq!(f ^ fi, f); + + { + // AndAssign: + let mut v = f; + v &= ti; + assert_eq!(v, f); + } + { + // OrAssign: + let mut v = f; + v |= ti; + assert_eq!(v, t); + } + { + // XORAssign: + let mut v = f; + v ^= ti; + assert_eq!(v, t); + } + } + } + } + } + }; +} diff --git a/third_party/rust/packed_simd/src/api/ops/scalar_shifts.rs b/third_party/rust/packed_simd/src/api/ops/scalar_shifts.rs new file mode 100644 index 000000000000..4a7a096263d0 --- /dev/null +++ b/third_party/rust/packed_simd/src/api/ops/scalar_shifts.rs @@ -0,0 +1,106 @@ +//! Vertical (lane-wise) vector-scalar shifts operations. + +macro_rules! impl_ops_scalar_shifts { + ([$elem_ty:ident; $elem_count:expr]: $id:ident | $test_tt:tt) => { + impl crate::ops::Shl for $id { + type Output = Self; + #[inline] + fn shl(self, other: u32) -> Self { + self << $id::splat(other as $elem_ty) + } + } + impl crate::ops::Shr for $id { + type Output = Self; + #[inline] + fn shr(self, other: u32) -> Self { + self >> $id::splat(other as $elem_ty) + } + } + + impl crate::ops::ShlAssign for $id { + #[inline] + fn shl_assign(&mut self, other: u32) { + *self = *self << other; + } + } + impl crate::ops::ShrAssign for $id { + #[inline] + fn shr_assign(&mut self, other: u32) { + *self = *self >> other; + } + } + test_if!{ + $test_tt: + paste::item! { + pub mod [<$id _ops_scalar_shifts>] { + use super::*; + #[cfg_attr(not(target_arch = "wasm32"), test)] #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] + #[cfg_attr(any(target_arch = "s390x", target_arch = "sparc64"), + allow(unreachable_code, unused_variables) + )] + #[cfg(not(target_arch = "aarch64"))] + //~^ FIXME: https://github.com/rust-lang/packed_simd/issues/317 + fn ops_scalar_shifts() { + let z = $id::splat(0 as $elem_ty); + let o = $id::splat(1 as $elem_ty); + let t = $id::splat(2 as $elem_ty); + let f = $id::splat(4 as $elem_ty); + + { + let zi = 0 as u32; + let oi = 1 as u32; + let ti = 2 as u32; + let maxi + = (mem::size_of::<$elem_ty>() * 8 - 1) as u32; + + // shr + assert_eq!(z >> zi, z); + assert_eq!(z >> oi, z); + assert_eq!(z >> ti, z); + assert_eq!(z >> ti, z); + + #[cfg(any(target_arch = "s390x", target_arch = "sparc64"))] { + // FIXME: https://github.com/rust-lang-nursery/packed_simd/issues/13 + return; + } + + assert_eq!(o >> zi, o); + assert_eq!(t >> zi, t); + assert_eq!(f >> zi, f); + assert_eq!(f >> maxi, z); + + assert_eq!(o >> oi, z); + assert_eq!(t >> oi, o); + assert_eq!(t >> ti, z); + assert_eq!(f >> oi, t); + assert_eq!(f >> ti, o); + assert_eq!(f >> maxi, z); + + // shl + assert_eq!(z << zi, z); + assert_eq!(o << zi, o); + assert_eq!(t << zi, t); + assert_eq!(f << zi, f); + assert_eq!(f << maxi, z); + + assert_eq!(o << oi, t); + assert_eq!(o << ti, f); + assert_eq!(t << oi, f); + + { // shr_assign + let mut v = o; + v >>= oi; + assert_eq!(v, z); + } + { // shl_assign + let mut v = o; + v <<= oi; + assert_eq!(v, t); + } + } + } + } + } + } + }; +} diff --git a/third_party/rust/packed_simd/src/api/ops/vector_arithmetic.rs b/third_party/rust/packed_simd/src/api/ops/vector_arithmetic.rs new file mode 100644 index 000000000000..7057f52d0317 --- /dev/null +++ b/third_party/rust/packed_simd/src/api/ops/vector_arithmetic.rs @@ -0,0 +1,148 @@ +//! Vertical (lane-wise) vector-vector arithmetic operations. + +macro_rules! impl_ops_vector_arithmetic { + ([$elem_ty:ident; $elem_count:expr]: $id:ident | $test_tt:tt) => { + impl crate::ops::Add for $id { + type Output = Self; + #[inline] + fn add(self, other: Self) -> Self { + use crate::llvm::simd_add; + unsafe { Simd(simd_add(self.0, other.0)) } + } + } + + impl crate::ops::Sub for $id { + type Output = Self; + #[inline] + fn sub(self, other: Self) -> Self { + use crate::llvm::simd_sub; + unsafe { Simd(simd_sub(self.0, other.0)) } + } + } + + impl crate::ops::Mul for $id { + type Output = Self; + #[inline] + fn mul(self, other: Self) -> Self { + use crate::llvm::simd_mul; + unsafe { Simd(simd_mul(self.0, other.0)) } + } + } + + impl crate::ops::Div for $id { + type Output = Self; + #[inline] + fn div(self, other: Self) -> Self { + use crate::llvm::simd_div; + unsafe { Simd(simd_div(self.0, other.0)) } + } + } + + impl crate::ops::Rem for $id { + type Output = Self; + #[inline] + fn rem(self, other: Self) -> Self { + use crate::llvm::simd_rem; + unsafe { Simd(simd_rem(self.0, other.0)) } + } + } + + impl crate::ops::AddAssign for $id { + #[inline] + fn add_assign(&mut self, other: Self) { + *self = *self + other; + } + } + + impl crate::ops::SubAssign for $id { + #[inline] + fn sub_assign(&mut self, other: Self) { + *self = *self - other; + } + } + + impl crate::ops::MulAssign for $id { + #[inline] + fn mul_assign(&mut self, other: Self) { + *self = *self * other; + } + } + + impl crate::ops::DivAssign for $id { + #[inline] + fn div_assign(&mut self, other: Self) { + *self = *self / other; + } + } + + impl crate::ops::RemAssign for $id { + #[inline] + fn rem_assign(&mut self, other: Self) { + *self = *self % other; + } + } + + test_if!{ + $test_tt: + paste::item! { + pub mod [<$id _ops_vector_arith>] { + use super::*; + #[cfg_attr(not(target_arch = "wasm32"), test)] #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] + fn ops_vector_arithmetic() { + let z = $id::splat(0 as $elem_ty); + let o = $id::splat(1 as $elem_ty); + let t = $id::splat(2 as $elem_ty); + let f = $id::splat(4 as $elem_ty); + + // add + assert_eq!(z + z, z); + assert_eq!(o + z, o); + assert_eq!(t + z, t); + assert_eq!(t + t, f); + // sub + assert_eq!(z - z, z); + assert_eq!(o - z, o); + assert_eq!(t - z, t); + assert_eq!(f - t, t); + assert_eq!(f - o - o, t); + // mul + assert_eq!(z * z, z); + assert_eq!(z * o, z); + assert_eq!(z * t, z); + assert_eq!(o * t, t); + assert_eq!(t * t, f); + // div + assert_eq!(z / o, z); + assert_eq!(t / o, t); + assert_eq!(f / o, f); + assert_eq!(t / t, o); + assert_eq!(f / t, t); + // rem + assert_eq!(o % o, z); + assert_eq!(f % t, z); + + { + let mut v = z; + assert_eq!(v, z); + v += o; // add_assign + assert_eq!(v, o); + v -= o; // sub_assign + assert_eq!(v, z); + v = t; + v *= o; // mul_assign + assert_eq!(v, t); + v *= t; + assert_eq!(v, f); + v /= o; // div_assign + assert_eq!(v, f); + v /= t; + assert_eq!(v, t); + v %= t; // rem_assign + assert_eq!(v, z); + } + } + } + } + } + }; +} diff --git a/third_party/rust/packed_simd/src/api/ops/vector_bitwise.rs b/third_party/rust/packed_simd/src/api/ops/vector_bitwise.rs new file mode 100644 index 000000000000..7be9603fa261 --- /dev/null +++ b/third_party/rust/packed_simd/src/api/ops/vector_bitwise.rs @@ -0,0 +1,129 @@ +//! Vertical (lane-wise) vector-vector bitwise operations. + +macro_rules! impl_ops_vector_bitwise { + ( + [$elem_ty:ident; $elem_count:expr]: + $id:ident | $test_tt:tt | + ($true:expr, $false:expr) + ) => { + impl crate::ops::Not for $id { + type Output = Self; + #[inline] + fn not(self) -> Self { + Self::splat($true) ^ self + } + } + impl crate::ops::BitXor for $id { + type Output = Self; + #[inline] + fn bitxor(self, other: Self) -> Self { + use crate::llvm::simd_xor; + unsafe { Simd(simd_xor(self.0, other.0)) } + } + } + impl crate::ops::BitAnd for $id { + type Output = Self; + #[inline] + fn bitand(self, other: Self) -> Self { + use crate::llvm::simd_and; + unsafe { Simd(simd_and(self.0, other.0)) } + } + } + impl crate::ops::BitOr for $id { + type Output = Self; + #[inline] + fn bitor(self, other: Self) -> Self { + use crate::llvm::simd_or; + unsafe { Simd(simd_or(self.0, other.0)) } + } + } + impl crate::ops::BitAndAssign for $id { + #[inline] + fn bitand_assign(&mut self, other: Self) { + *self = *self & other; + } + } + impl crate::ops::BitOrAssign for $id { + #[inline] + fn bitor_assign(&mut self, other: Self) { + *self = *self | other; + } + } + impl crate::ops::BitXorAssign for $id { + #[inline] + fn bitxor_assign(&mut self, other: Self) { + *self = *self ^ other; + } + } + + test_if!{ + $test_tt: + paste::item! { + pub mod [<$id _ops_vector_bitwise>] { + use super::*; + #[cfg_attr(not(target_arch = "wasm32"), test)] #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] + fn ops_vector_bitwise() { + + let z = $id::splat(0 as $elem_ty); + let o = $id::splat(1 as $elem_ty); + let t = $id::splat(2 as $elem_ty); + let m = $id::splat(!z.extract(0)); + + // Not: + assert_eq!(!z, m); + assert_eq!(!m, z); + + // BitAnd: + assert_eq!(o & o, o); + assert_eq!(o & z, z); + assert_eq!(z & o, z); + assert_eq!(z & z, z); + + assert_eq!(t & t, t); + assert_eq!(t & o, z); + assert_eq!(o & t, z); + + // BitOr: + assert_eq!(o | o, o); + assert_eq!(o | z, o); + assert_eq!(z | o, o); + assert_eq!(z | z, z); + + assert_eq!(t | t, t); + assert_eq!(z | t, t); + assert_eq!(t | z, t); + + // BitXOR: + assert_eq!(o ^ o, z); + assert_eq!(z ^ z, z); + assert_eq!(z ^ o, o); + assert_eq!(o ^ z, o); + + assert_eq!(t ^ t, z); + assert_eq!(t ^ z, t); + assert_eq!(z ^ t, t); + + { + // AndAssign: + let mut v = o; + v &= t; + assert_eq!(v, z); + } + { + // OrAssign: + let mut v = z; + v |= o; + assert_eq!(v, o); + } + { + // XORAssign: + let mut v = z; + v ^= o; + assert_eq!(v, o); + } + } + } + } + } + }; +} diff --git a/third_party/rust/packed_simd/src/api/ops/vector_float_min_max.rs b/third_party/rust/packed_simd/src/api/ops/vector_float_min_max.rs new file mode 100644 index 000000000000..8310667b7a8d --- /dev/null +++ b/third_party/rust/packed_simd/src/api/ops/vector_float_min_max.rs @@ -0,0 +1,74 @@ +//! Vertical (lane-wise) vector `min` and `max` for floating-point vectors. + +macro_rules! impl_ops_vector_float_min_max { + ([$elem_ty:ident; $elem_count:expr]: $id:ident | $test_tt:tt) => { + impl $id { + /// Minimum of two vectors. + /// + /// Returns a new vector containing the minimum value of each of + /// the input vector lanes. + #[inline] + pub fn min(self, x: Self) -> Self { + use crate::llvm::simd_fmin; + unsafe { Simd(simd_fmin(self.0, x.0)) } + } + + /// Maximum of two vectors. + /// + /// Returns a new vector containing the maximum value of each of + /// the input vector lanes. + #[inline] + pub fn max(self, x: Self) -> Self { + use crate::llvm::simd_fmax; + unsafe { Simd(simd_fmax(self.0, x.0)) } + } + } + test_if!{ + $test_tt: + paste::item! { + #[cfg(not(any( + // FIXME: https://github.com/rust-lang-nursery/packed_simd/issues/223 + all(target_arch = "mips", target_endian = "big"), + target_arch = "mips64", + )))] + pub mod [<$id _ops_vector_min_max>] { + use super::*; + #[cfg_attr(not(target_arch = "wasm32"), test)] #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] + fn min_max() { + let n = crate::$elem_ty::NAN; + let o = $id::splat(1. as $elem_ty); + let t = $id::splat(2. as $elem_ty); + + let mut m = o; // [1., 2., 1., 2., ...] + let mut on = o; + for i in 0..$id::lanes() { + if i % 2 == 0 { + m = m.replace(i, 2. as $elem_ty); + on = on.replace(i, n); + } + } + + assert_eq!(o.min(t), o); + assert_eq!(t.min(o), o); + assert_eq!(m.min(o), o); + assert_eq!(o.min(m), o); + assert_eq!(m.min(t), m); + assert_eq!(t.min(m), m); + + assert_eq!(o.max(t), t); + assert_eq!(t.max(o), t); + assert_eq!(m.max(o), m); + assert_eq!(o.max(m), m); + assert_eq!(m.max(t), t); + assert_eq!(t.max(m), t); + + assert_eq!(on.min(o), o); + assert_eq!(o.min(on), o); + assert_eq!(on.max(o), o); + assert_eq!(o.max(on), o); + } + } + } + } + }; +} diff --git a/third_party/rust/packed_simd/src/api/ops/vector_int_min_max.rs b/third_party/rust/packed_simd/src/api/ops/vector_int_min_max.rs new file mode 100644 index 000000000000..36ea98e6bf32 --- /dev/null +++ b/third_party/rust/packed_simd/src/api/ops/vector_int_min_max.rs @@ -0,0 +1,57 @@ +//! Vertical (lane-wise) vector `min` and `max` for integer vectors. + +macro_rules! impl_ops_vector_int_min_max { + ([$elem_ty:ident; $elem_count:expr]: $id:ident | $test_tt:tt) => { + impl $id { + /// Minimum of two vectors. + /// + /// Returns a new vector containing the minimum value of each of + /// the input vector lanes. + #[inline] + pub fn min(self, x: Self) -> Self { + self.lt(x).select(self, x) + } + + /// Maximum of two vectors. + /// + /// Returns a new vector containing the maximum value of each of + /// the input vector lanes. + #[inline] + pub fn max(self, x: Self) -> Self { + self.gt(x).select(self, x) + } + } + test_if!{$test_tt: + paste::item! { + pub mod [<$id _ops_vector_min_max>] { + use super::*; + #[cfg_attr(not(target_arch = "wasm32"), test)] #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] + fn min_max() { + let o = $id::splat(1 as $elem_ty); + let t = $id::splat(2 as $elem_ty); + + let mut m = o; + for i in 0..$id::lanes() { + if i % 2 == 0 { + m = m.replace(i, 2 as $elem_ty); + } + } + assert_eq!(o.min(t), o); + assert_eq!(t.min(o), o); + assert_eq!(m.min(o), o); + assert_eq!(o.min(m), o); + assert_eq!(m.min(t), m); + assert_eq!(t.min(m), m); + + assert_eq!(o.max(t), t); + assert_eq!(t.max(o), t); + assert_eq!(m.max(o), m); + assert_eq!(o.max(m), m); + assert_eq!(m.max(t), t); + assert_eq!(t.max(m), t); + } + } + } + } + }; +} diff --git a/third_party/rust/packed_simd/src/api/ops/vector_mask_bitwise.rs b/third_party/rust/packed_simd/src/api/ops/vector_mask_bitwise.rs new file mode 100644 index 000000000000..295fc1ca81c9 --- /dev/null +++ b/third_party/rust/packed_simd/src/api/ops/vector_mask_bitwise.rs @@ -0,0 +1,116 @@ +//! Vertical (lane-wise) vector-vector bitwise operations. + +macro_rules! impl_ops_vector_mask_bitwise { + ( + [$elem_ty:ident; $elem_count:expr]: + $id:ident | $test_tt:tt | + ($true:expr, $false:expr) + ) => { + impl crate::ops::Not for $id { + type Output = Self; + #[inline] + fn not(self) -> Self { + Self::splat($true) ^ self + } + } + impl crate::ops::BitXor for $id { + type Output = Self; + #[inline] + fn bitxor(self, other: Self) -> Self { + use crate::llvm::simd_xor; + unsafe { Simd(simd_xor(self.0, other.0)) } + } + } + impl crate::ops::BitAnd for $id { + type Output = Self; + #[inline] + fn bitand(self, other: Self) -> Self { + use crate::llvm::simd_and; + unsafe { Simd(simd_and(self.0, other.0)) } + } + } + impl crate::ops::BitOr for $id { + type Output = Self; + #[inline] + fn bitor(self, other: Self) -> Self { + use crate::llvm::simd_or; + unsafe { Simd(simd_or(self.0, other.0)) } + } + } + impl crate::ops::BitAndAssign for $id { + #[inline] + fn bitand_assign(&mut self, other: Self) { + *self = *self & other; + } + } + impl crate::ops::BitOrAssign for $id { + #[inline] + fn bitor_assign(&mut self, other: Self) { + *self = *self | other; + } + } + impl crate::ops::BitXorAssign for $id { + #[inline] + fn bitxor_assign(&mut self, other: Self) { + *self = *self ^ other; + } + } + + test_if!{ + $test_tt: + paste::item! { + pub mod [<$id _ops_vector_mask_bitwise>] { + use super::*; + #[cfg_attr(not(target_arch = "wasm32"), test)] #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] + fn ops_vector_mask_bitwise() { + let t = $id::splat(true); + let f = $id::splat(false); + assert!(t != f); + assert!(!(t == f)); + + // Not: + assert_eq!(!t, f); + assert_eq!(t, !f); + + // BitAnd: + assert_eq!(t & f, f); + assert_eq!(f & t, f); + assert_eq!(t & t, t); + assert_eq!(f & f, f); + + // BitOr: + assert_eq!(t | f, t); + assert_eq!(f | t, t); + assert_eq!(t | t, t); + assert_eq!(f | f, f); + + // BitXOR: + assert_eq!(t ^ f, t); + assert_eq!(f ^ t, t); + assert_eq!(t ^ t, f); + assert_eq!(f ^ f, f); + + { + // AndAssign: + let mut v = f; + v &= t; + assert_eq!(v, f); + } + { + // OrAssign: + let mut v = f; + v |= t; + assert_eq!(v, t); + } + { + // XORAssign: + let mut v = f; + v ^= t; + assert_eq!(v, t); + } + } + } + } + } + }; +} diff --git a/third_party/rust/packed_simd/src/api/ops/vector_neg.rs b/third_party/rust/packed_simd/src/api/ops/vector_neg.rs new file mode 100644 index 000000000000..e2d91fd2fed6 --- /dev/null +++ b/third_party/rust/packed_simd/src/api/ops/vector_neg.rs @@ -0,0 +1,43 @@ +//! Vertical (lane-wise) vector `Neg`. + +macro_rules! impl_ops_vector_neg { + ([$elem_ty:ident; $elem_count:expr]: $id:ident | $test_tt:tt) => { + impl crate::ops::Neg for $id { + type Output = Self; + #[inline] + fn neg(self) -> Self { + Self::splat(-1 as $elem_ty) * self + } + } + test_if!{ + $test_tt: + paste::item! { + pub mod [<$id _ops_vector_neg>] { + use super::*; + #[cfg_attr(not(target_arch = "wasm32"), test)] #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] + fn neg() { + let z = $id::splat(0 as $elem_ty); + let o = $id::splat(1 as $elem_ty); + let t = $id::splat(2 as $elem_ty); + let f = $id::splat(4 as $elem_ty); + + let nz = $id::splat(-(0 as $elem_ty)); + let no = $id::splat(-(1 as $elem_ty)); + let nt = $id::splat(-(2 as $elem_ty)); + let nf = $id::splat(-(4 as $elem_ty)); + + assert_eq!(-z, nz); + assert_eq!(-o, no); + assert_eq!(-t, nt); + assert_eq!(-f, nf); + + assert_eq!(z, -nz); + assert_eq!(o, -no); + assert_eq!(t, -nt); + assert_eq!(f, -nf); + } + } + } + } + }; +} diff --git a/third_party/rust/packed_simd/src/api/ops/vector_rotates.rs b/third_party/rust/packed_simd/src/api/ops/vector_rotates.rs new file mode 100644 index 000000000000..6c4bed72a2a8 --- /dev/null +++ b/third_party/rust/packed_simd/src/api/ops/vector_rotates.rs @@ -0,0 +1,92 @@ +//! Vertical (lane-wise) vector rotates operations. +#![allow(unused)] + +macro_rules! impl_ops_vector_rotates { + ([$elem_ty:ident; $elem_count:expr]: $id:ident | $test_tt:tt) => { + impl $id { + /// Shifts the bits of each lane to the left by the specified + /// amount in the corresponding lane of `n`, wrapping the + /// truncated bits to the end of the resulting integer. + /// + /// Note: this is neither the same operation as `<<` nor equivalent + /// to `slice::rotate_left`. + #[inline] + pub fn rotate_left(self, n: $id) -> $id { + const LANE_WIDTH: $elem_ty = + crate::mem::size_of::<$elem_ty>() as $elem_ty * 8; + // Protect against undefined behavior for over-long bit shifts + let n = n % LANE_WIDTH; + (self << n) | (self >> ((LANE_WIDTH - n) % LANE_WIDTH)) + } + + /// Shifts the bits of each lane to the right by the specified + /// amount in the corresponding lane of `n`, wrapping the + /// truncated bits to the beginning of the resulting integer. + /// + /// Note: this is neither the same operation as `>>` nor equivalent + /// to `slice::rotate_right`. + #[inline] + pub fn rotate_right(self, n: $id) -> $id { + const LANE_WIDTH: $elem_ty = + crate::mem::size_of::<$elem_ty>() as $elem_ty * 8; + // Protect against undefined behavior for over-long bit shifts + let n = n % LANE_WIDTH; + (self >> n) | (self << ((LANE_WIDTH - n) % LANE_WIDTH)) + } + } + + test_if!{ + $test_tt: + paste::item! { + // FIXME: + // https://github.com/rust-lang-nursery/packed_simd/issues/75 + #[cfg(not(any( + target_arch = "s390x", + target_arch = "sparc64", + )))] + pub mod [<$id _ops_vector_rotate>] { + use super::*; + #[cfg_attr(not(target_arch = "wasm32"), test)] #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] + #[cfg(not(target_arch = "aarch64"))] + //~^ FIXME: https://github.com/rust-lang/packed_simd/issues/317 + fn rotate_ops() { + let z = $id::splat(0 as $elem_ty); + let o = $id::splat(1 as $elem_ty); + let t = $id::splat(2 as $elem_ty); + let f = $id::splat(4 as $elem_ty); + + let max = $id::splat( + (mem::size_of::<$elem_ty>() * 8 - 1) as $elem_ty); + + // rotate_right + assert_eq!(z.rotate_right(z), z); + assert_eq!(z.rotate_right(o), z); + assert_eq!(z.rotate_right(t), z); + + assert_eq!(o.rotate_right(z), o); + assert_eq!(t.rotate_right(z), t); + assert_eq!(f.rotate_right(z), f); + assert_eq!(f.rotate_right(max), f << 1); + + assert_eq!(o.rotate_right(o), o << max); + assert_eq!(t.rotate_right(o), o); + assert_eq!(t.rotate_right(t), o << max); + assert_eq!(f.rotate_right(o), t); + assert_eq!(f.rotate_right(t), o); + + // rotate_left + assert_eq!(z.rotate_left(z), z); + assert_eq!(o.rotate_left(z), o); + assert_eq!(t.rotate_left(z), t); + assert_eq!(f.rotate_left(z), f); + assert_eq!(f.rotate_left(max), t); + + assert_eq!(o.rotate_left(o), t); + assert_eq!(o.rotate_left(t), f); + assert_eq!(t.rotate_left(o), f); + } + } + } + } + }; +} diff --git a/third_party/rust/packed_simd/src/api/ops/vector_shifts.rs b/third_party/rust/packed_simd/src/api/ops/vector_shifts.rs new file mode 100644 index 000000000000..8bb5ac2fc03a --- /dev/null +++ b/third_party/rust/packed_simd/src/api/ops/vector_shifts.rs @@ -0,0 +1,106 @@ +//! Vertical (lane-wise) vector-vector shifts operations. + +macro_rules! impl_ops_vector_shifts { + ([$elem_ty:ident; $elem_count:expr]: $id:ident | $test_tt:tt) => { + impl crate::ops::Shl<$id> for $id { + type Output = Self; + #[inline] + fn shl(self, other: Self) -> Self { + use crate::llvm::simd_shl; + unsafe { Simd(simd_shl(self.0, other.0)) } + } + } + impl crate::ops::Shr<$id> for $id { + type Output = Self; + #[inline] + fn shr(self, other: Self) -> Self { + use crate::llvm::simd_shr; + unsafe { Simd(simd_shr(self.0, other.0)) } + } + } + impl crate::ops::ShlAssign<$id> for $id { + #[inline] + fn shl_assign(&mut self, other: Self) { + *self = *self << other; + } + } + impl crate::ops::ShrAssign<$id> for $id { + #[inline] + fn shr_assign(&mut self, other: Self) { + *self = *self >> other; + } + } + test_if!{ + $test_tt: + paste::item! { + pub mod [<$id _ops_vector_shifts>] { + use super::*; + #[cfg_attr(not(target_arch = "wasm32"), test)] #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] + #[cfg_attr(any(target_arch = "s390x", target_arch = "sparc64"), + allow(unreachable_code, unused_variables) + )] + #[cfg(not(target_arch = "aarch64"))] + //~^ FIXME: https://github.com/rust-lang/packed_simd/issues/317 + fn ops_vector_shifts() { + let z = $id::splat(0 as $elem_ty); + let o = $id::splat(1 as $elem_ty); + let t = $id::splat(2 as $elem_ty); + let f = $id::splat(4 as $elem_ty); + + let max =$id::splat( + (mem::size_of::<$elem_ty>() * 8 - 1) as $elem_ty + ); + + // shr + assert_eq!(z >> z, z); + assert_eq!(z >> o, z); + assert_eq!(z >> t, z); + assert_eq!(z >> t, z); + + #[cfg(any(target_arch = "s390x", target_arch = "sparc64"))] { + // FIXME: rust produces bad codegen for shifts: + // https://github.com/rust-lang-nursery/packed_simd/issues/13 + return; + } + + assert_eq!(o >> z, o); + assert_eq!(t >> z, t); + assert_eq!(f >> z, f); + assert_eq!(f >> max, z); + + assert_eq!(o >> o, z); + assert_eq!(t >> o, o); + assert_eq!(t >> t, z); + assert_eq!(f >> o, t); + assert_eq!(f >> t, o); + assert_eq!(f >> max, z); + + // shl + assert_eq!(z << z, z); + assert_eq!(o << z, o); + assert_eq!(t << z, t); + assert_eq!(f << z, f); + assert_eq!(f << max, z); + + assert_eq!(o << o, t); + assert_eq!(o << t, f); + assert_eq!(t << o, f); + + { + // shr_assign + let mut v = o; + v >>= o; + assert_eq!(v, z); + } + { + // shl_assign + let mut v = o; + v <<= o; + assert_eq!(v, t); + } + } + } + } + } + }; +} diff --git a/third_party/rust/packed_simd/src/api/ptr.rs b/third_party/rust/packed_simd/src/api/ptr.rs new file mode 100644 index 000000000000..d2e523a49faf --- /dev/null +++ b/third_party/rust/packed_simd/src/api/ptr.rs @@ -0,0 +1,4 @@ +//! Vector of pointers + +#[macro_use] +mod gather_scatter; diff --git a/third_party/rust/packed_simd/src/api/ptr/gather_scatter.rs b/third_party/rust/packed_simd/src/api/ptr/gather_scatter.rs new file mode 100644 index 000000000000..374482ac31c6 --- /dev/null +++ b/third_party/rust/packed_simd/src/api/ptr/gather_scatter.rs @@ -0,0 +1,216 @@ +//! Implements masked gather and scatters for vectors of pointers + +macro_rules! impl_ptr_read { + ([$elem_ty:ty; $elem_count:expr]: $id:ident, $mask_ty:ident + | $test_tt:tt) => { + impl $id + where + [T; $elem_count]: sealed::SimdArray, + { + /// Reads selected vector elements from memory. + /// + /// Instantiates a new vector by reading the values from `self` for + /// those lanes whose `mask` is `true`, and using the elements of + /// `value` otherwise. + /// + /// No memory is accessed for those lanes of `self` whose `mask` is + /// `false`. + /// + /// # Safety + /// + /// This method is unsafe because it dereferences raw pointers. The + /// pointers must be aligned to `mem::align_of::()`. + #[inline] + pub unsafe fn read( + self, + mask: Simd<[M; $elem_count]>, + value: Simd<[T; $elem_count]>, + ) -> Simd<[T; $elem_count]> + where + M: sealed::Mask, + [M; $elem_count]: sealed::SimdArray, + { + use crate::llvm::simd_gather; + Simd(simd_gather(value.0, self.0, mask.0)) + } + } + + test_if! { + $test_tt: + paste::item! { + mod [<$id _read>] { + use super::*; + #[test] + fn read() { + let mut v = [0_i32; $elem_count]; + for i in 0..$elem_count { + v[i] = i as i32; + } + + let mut ptr = $id::::null(); + + for i in 0..$elem_count { + ptr = ptr.replace(i, + &v[i] as *const i32 as *mut i32 + ); + } + + // all mask elements are true: + let mask = $mask_ty::splat(true); + let def = Simd::<[i32; $elem_count]>::splat(42_i32); + let r: Simd<[i32; $elem_count]> = unsafe { + ptr.read(mask, def) + }; + assert_eq!( + r, + Simd::<[i32; $elem_count]>::from_slice_unaligned( + &v + ) + ); + + let mut mask = mask; + for i in 0..$elem_count { + if i % 2 != 0 { + mask = mask.replace(i, false); + } + } + + // even mask elements are true, odd ones are false: + let r: Simd<[i32; $elem_count]> = unsafe { + ptr.read(mask, def) + }; + let mut e = v; + for i in 0..$elem_count { + if i % 2 != 0 { + e[i] = 42; + } + } + assert_eq!( + r, + Simd::<[i32; $elem_count]>::from_slice_unaligned( + &e + ) + ); + + // all mask elements are false: + let mask = $mask_ty::splat(false); + let def = Simd::<[i32; $elem_count]>::splat(42_i32); + let r: Simd<[i32; $elem_count]> = unsafe { + ptr.read(mask, def) } + ; + assert_eq!(r, def); + } + } + } + } + }; +} + +macro_rules! impl_ptr_write { + ([$elem_ty:ty; $elem_count:expr]: $id:ident, $mask_ty:ident + | $test_tt:tt) => { + impl $id + where + [T; $elem_count]: sealed::SimdArray, + { + /// Writes selected vector elements to memory. + /// + /// Writes the lanes of `values` for which the mask is `true` to + /// their corresponding memory addresses in `self`. + /// + /// No memory is accessed for those lanes of `self` whose `mask` is + /// `false`. + /// + /// Overlapping memory addresses of `self` are written to in order + /// from the lest-significant to the most-significant element. + /// + /// # Safety + /// + /// This method is unsafe because it dereferences raw pointers. The + /// pointers must be aligned to `mem::align_of::()`. + #[inline] + pub unsafe fn write(self, mask: Simd<[M; $elem_count]>, value: Simd<[T; $elem_count]>) + where + M: sealed::Mask, + [M; $elem_count]: sealed::SimdArray, + { + use crate::llvm::simd_scatter; + simd_scatter(value.0, self.0, mask.0) + } + } + + test_if! { + $test_tt: + paste::item! { + mod [<$id _write>] { + use super::*; + #[test] + fn write() { + // forty_two = [42, 42, 42, ...] + let forty_two + = Simd::<[i32; $elem_count]>::splat(42_i32); + + // This test will write to this array + let mut arr = [0_i32; $elem_count]; + for i in 0..$elem_count { + arr[i] = i as i32; + } + // arr = [0, 1, 2, ...] + + let mut ptr = $id::::null(); + for i in 0..$elem_count { + ptr = ptr.replace(i, unsafe { + arr.as_ptr().add(i) as *mut i32 + }); + } + // ptr = [&arr[0], &arr[1], ...] + + // write `forty_two` to all elements of `v` + { + let backup = arr; + unsafe { + ptr.write($mask_ty::splat(true), forty_two) + }; + assert_eq!(arr, [42_i32; $elem_count]); + arr = backup; // arr = [0, 1, 2, ...] + } + + // write 42 to even elements of arr: + { + // set odd elements of the mask to false + let mut mask = $mask_ty::splat(true); + for i in 0..$elem_count { + if i % 2 != 0 { + mask = mask.replace(i, false); + } + } + // mask = [true, false, true, false, ...] + + // expected result r = [42, 1, 42, 3, 42, 5, ...] + let mut r = arr; + for i in 0..$elem_count { + if i % 2 == 0 { + r[i] = 42; + } + } + + let backup = arr; + unsafe { ptr.write(mask, forty_two) }; + assert_eq!(arr, r); + arr = backup; // arr = [0, 1, 2, 3, ...] + } + + // write 42 to no elements of arr + { + let backup = arr; + unsafe { + ptr.write($mask_ty::splat(false), forty_two) + }; + assert_eq!(arr, backup); + } + } + } + } + } + }; +} diff --git a/third_party/rust/packed_simd/src/api/reductions.rs b/third_party/rust/packed_simd/src/api/reductions.rs new file mode 100644 index 000000000000..54d2f0cc7f08 --- /dev/null +++ b/third_party/rust/packed_simd/src/api/reductions.rs @@ -0,0 +1,12 @@ +//! Reductions + +#[macro_use] +mod float_arithmetic; +#[macro_use] +mod integer_arithmetic; +#[macro_use] +mod bitwise; +#[macro_use] +mod mask; +#[macro_use] +mod min_max; diff --git a/third_party/rust/packed_simd/src/api/reductions/bitwise.rs b/third_party/rust/packed_simd/src/api/reductions/bitwise.rs new file mode 100644 index 000000000000..5bad4f474b16 --- /dev/null +++ b/third_party/rust/packed_simd/src/api/reductions/bitwise.rs @@ -0,0 +1,151 @@ +//! Implements portable horizontal bitwise vector reductions. +#![allow(unused)] + +macro_rules! impl_reduction_bitwise { + ( + [$elem_ty:ident; $elem_count:expr]: + $id:ident | $ielem_ty:ident | $test_tt:tt | + ($convert:expr) | + ($true:expr, $false:expr) + ) => { + impl $id { + /// Lane-wise bitwise `and` of the vector elements. + /// + /// Note: if the vector has one lane, the first element of the + /// vector is returned. + #[inline] + pub fn and(self) -> $elem_ty { + #[cfg(not(target_arch = "aarch64"))] + { + use crate::llvm::simd_reduce_and; + let r: $ielem_ty = unsafe { simd_reduce_and(self.0) }; + $convert(r) + } + #[cfg(target_arch = "aarch64")] + { + // FIXME: broken on aarch64 + // https://github.com/rust-lang-nursery/packed_simd/issues/15 + let mut x = self.extract(0) as $elem_ty; + for i in 1..$id::lanes() { + x &= self.extract(i) as $elem_ty; + } + x + } + } + + /// Lane-wise bitwise `or` of the vector elements. + /// + /// Note: if the vector has one lane, the first element of the + /// vector is returned. + #[inline] + pub fn or(self) -> $elem_ty { + #[cfg(not(target_arch = "aarch64"))] + { + use crate::llvm::simd_reduce_or; + let r: $ielem_ty = unsafe { simd_reduce_or(self.0) }; + $convert(r) + } + #[cfg(target_arch = "aarch64")] + { + // FIXME: broken on aarch64 + // https://github.com/rust-lang-nursery/packed_simd/issues/15 + let mut x = self.extract(0) as $elem_ty; + for i in 1..$id::lanes() { + x |= self.extract(i) as $elem_ty; + } + x + } + } + + /// Lane-wise bitwise `xor` of the vector elements. + /// + /// Note: if the vector has one lane, the first element of the + /// vector is returned. + #[inline] + pub fn xor(self) -> $elem_ty { + #[cfg(not(target_arch = "aarch64"))] + { + use crate::llvm::simd_reduce_xor; + let r: $ielem_ty = unsafe { simd_reduce_xor(self.0) }; + $convert(r) + } + #[cfg(target_arch = "aarch64")] + { + // FIXME: broken on aarch64 + // https://github.com/rust-lang-nursery/packed_simd/issues/15 + let mut x = self.extract(0) as $elem_ty; + for i in 1..$id::lanes() { + x ^= self.extract(i) as $elem_ty; + } + x + } + } + } + + test_if!{ + $test_tt: + paste::item! { + pub mod [<$id _reduction_bitwise>] { + use super::*; + + #[cfg_attr(not(target_arch = "wasm32"), test)] #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] + fn and() { + let v = $id::splat($false); + assert_eq!(v.and(), $false); + let v = $id::splat($true); + assert_eq!(v.and(), $true); + let v = $id::splat($false); + let v = v.replace(0, $true); + if $id::lanes() > 1 { + assert_eq!(v.and(), $false); + } else { + assert_eq!(v.and(), $true); + } + let v = $id::splat($true); + let v = v.replace(0, $false); + assert_eq!(v.and(), $false); + + } + #[cfg_attr(not(target_arch = "wasm32"), test)] #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] + fn or() { + let v = $id::splat($false); + assert_eq!(v.or(), $false); + let v = $id::splat($true); + assert_eq!(v.or(), $true); + let v = $id::splat($false); + let v = v.replace(0, $true); + assert_eq!(v.or(), $true); + let v = $id::splat($true); + let v = v.replace(0, $false); + if $id::lanes() > 1 { + assert_eq!(v.or(), $true); + } else { + assert_eq!(v.or(), $false); + } + } + #[cfg_attr(not(target_arch = "wasm32"), test)] #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] + fn xor() { + let v = $id::splat($false); + assert_eq!(v.xor(), $false); + let v = $id::splat($true); + if $id::lanes() > 1 { + assert_eq!(v.xor(), $false); + } else { + assert_eq!(v.xor(), $true); + } + let v = $id::splat($false); + let v = v.replace(0, $true); + assert_eq!(v.xor(), $true); + let v = $id::splat($true); + let v = v.replace(0, $false); + if $id::lanes() > 1 { + assert_eq!(v.xor(), $true); + } else { + assert_eq!(v.xor(), $false); + } + } + } + } + } + }; +} diff --git a/third_party/rust/packed_simd/src/api/reductions/float_arithmetic.rs b/third_party/rust/packed_simd/src/api/reductions/float_arithmetic.rs new file mode 100644 index 000000000000..9dc8783dbbf6 --- /dev/null +++ b/third_party/rust/packed_simd/src/api/reductions/float_arithmetic.rs @@ -0,0 +1,313 @@ +//! Implements portable horizontal float vector arithmetic reductions. + +macro_rules! impl_reduction_float_arithmetic { + ([$elem_ty:ident; $elem_count:expr]: $id:ident | $test_tt:tt) => { + impl $id { + /// Horizontal sum of the vector elements. + /// + /// The intrinsic performs a tree-reduction of the vector elements. + /// That is, for an 8 element vector: + /// + /// > ((x0 + x1) + (x2 + x3)) + ((x4 + x5) + (x6 + x7)) + /// + /// If one of the vector element is `NaN` the reduction returns + /// `NaN`. The resulting `NaN` is not required to be equal to any + /// of the `NaN`s in the vector. + #[inline] + pub fn sum(self) -> $elem_ty { + #[cfg(not(target_arch = "aarch64"))] + { + use crate::llvm::simd_reduce_add_ordered; + unsafe { simd_reduce_add_ordered(self.0, 0 as $elem_ty) } + } + #[cfg(target_arch = "aarch64")] + { + // FIXME: broken on AArch64 + // https://github.com/rust-lang-nursery/packed_simd/issues/15 + let mut x = self.extract(0) as $elem_ty; + for i in 1..$id::lanes() { + x += self.extract(i) as $elem_ty; + } + x + } + } + + /// Horizontal product of the vector elements. + /// + /// The intrinsic performs a tree-reduction of the vector elements. + /// That is, for an 8 element vector: + /// + /// > ((x0 * x1) * (x2 * x3)) * ((x4 * x5) * (x6 * x7)) + /// + /// If one of the vector element is `NaN` the reduction returns + /// `NaN`. The resulting `NaN` is not required to be equal to any + /// of the `NaN`s in the vector. + #[inline] + pub fn product(self) -> $elem_ty { + #[cfg(not(target_arch = "aarch64"))] + { + use crate::llvm::simd_reduce_mul_ordered; + unsafe { simd_reduce_mul_ordered(self.0, 1 as $elem_ty) } + } + #[cfg(target_arch = "aarch64")] + { + // FIXME: broken on AArch64 + // https://github.com/rust-lang-nursery/packed_simd/issues/15 + let mut x = self.extract(0) as $elem_ty; + for i in 1..$id::lanes() { + x *= self.extract(i) as $elem_ty; + } + x + } + } + } + + impl crate::iter::Sum for $id { + #[inline] + fn sum>(iter: I) -> $id { + iter.fold($id::splat(0.), crate::ops::Add::add) + } + } + + impl crate::iter::Product for $id { + #[inline] + fn product>(iter: I) -> $id { + iter.fold($id::splat(1.), crate::ops::Mul::mul) + } + } + + impl<'a> crate::iter::Sum<&'a $id> for $id { + #[inline] + fn sum>(iter: I) -> $id { + iter.fold($id::splat(0.), |a, b| crate::ops::Add::add(a, *b)) + } + } + + impl<'a> crate::iter::Product<&'a $id> for $id { + #[inline] + fn product>(iter: I) -> $id { + iter.fold($id::splat(1.), |a, b| crate::ops::Mul::mul(a, *b)) + } + } + + test_if! { + $test_tt: + paste::item! { + // Comparisons use integer casts within mantissa^1 range. + #[allow(clippy::float_cmp)] + pub mod [<$id _reduction_float_arith>] { + use super::*; + fn alternating(x: usize) -> $id { + let mut v = $id::splat(1 as $elem_ty); + for i in 0..$id::lanes() { + if i % x == 0 { + v = v.replace(i, 2 as $elem_ty); + } + } + v + } + + #[cfg_attr(not(target_arch = "wasm32"), test)] + #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] + fn sum() { + let v = $id::splat(0 as $elem_ty); + assert_eq!(v.sum(), 0 as $elem_ty); + let v = $id::splat(1 as $elem_ty); + assert_eq!(v.sum(), $id::lanes() as $elem_ty); + let v = alternating(2); + assert_eq!( + v.sum(), + ($id::lanes() / 2 + $id::lanes()) as $elem_ty + ); + } + #[cfg_attr(not(target_arch = "wasm32"), test)] + #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] + fn product() { + let v = $id::splat(0 as $elem_ty); + assert_eq!(v.product(), 0 as $elem_ty); + let v = $id::splat(1 as $elem_ty); + assert_eq!(v.product(), 1 as $elem_ty); + let f = match $id::lanes() { + 64 => 16, + 32 => 8, + 16 => 4, + _ => 2, + }; + let v = alternating(f); + assert_eq!( + v.product(), + (2_usize.pow(($id::lanes() / f) as u32) + as $elem_ty) + ); + } + + #[cfg_attr(not(target_arch = "wasm32"), test)] + #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] + #[allow(unreachable_code)] + fn sum_nan() { + // FIXME: https://bugs.llvm.org/show_bug.cgi?id=36732 + // https://github.com/rust-lang-nursery/packed_simd/issues/6 + return; + + let n0 = crate::$elem_ty::NAN; + let v0 = $id::splat(-3.0); + for i in 0..$id::lanes() { + let mut v = v0.replace(i, n0); + // If the vector contains a NaN the result is NaN: + assert!( + v.sum().is_nan(), + "nan at {} => {} | {:?}", + i, + v.sum(), + v + ); + for j in 0..i { + v = v.replace(j, n0); + assert!(v.sum().is_nan()); + } + } + let v = $id::splat(n0); + assert!(v.sum().is_nan(), "all nans | {:?}", v); + } + + #[cfg_attr(not(target_arch = "wasm32"), test)] + #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] + #[allow(unreachable_code)] + fn product_nan() { + // FIXME: https://bugs.llvm.org/show_bug.cgi?id=36732 + // https://github.com/rust-lang-nursery/packed_simd/issues/6 + return; + + let n0 = crate::$elem_ty::NAN; + let v0 = $id::splat(-3.0); + for i in 0..$id::lanes() { + let mut v = v0.replace(i, n0); + // If the vector contains a NaN the result is NaN: + assert!( + v.product().is_nan(), + "nan at {} => {} | {:?}", + i, + v.product(), + v + ); + for j in 0..i { + v = v.replace(j, n0); + assert!(v.product().is_nan()); + } + } + let v = $id::splat(n0); + assert!(v.product().is_nan(), "all nans | {:?}", v); + } + + #[cfg_attr(not(target_arch = "wasm32"), test)] + #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] + #[allow(unused, dead_code)] + fn sum_roundoff() { + // Performs a tree-reduction + fn tree_reduce_sum(a: &[$elem_ty]) -> $elem_ty { + assert!(!a.is_empty()); + if a.len() == 1 { + a[0] + } else if a.len() == 2 { + a[0] + a[1] + } else { + let mid = a.len() / 2; + let (left, right) = a.split_at(mid); + tree_reduce_sum(left) + tree_reduce_sum(right) + } + } + + let mut start = crate::$elem_ty::EPSILON; + let mut scalar_reduction = 0. as $elem_ty; + + let mut v = $id::splat(0. as $elem_ty); + for i in 0..$id::lanes() { + let c = if i % 2 == 0 { 1e3 } else { -1. }; + start *= ::core::$elem_ty::consts::PI * c; + scalar_reduction += start; + v = v.replace(i, start); + } + let simd_reduction = v.sum(); + + let mut a = [0. as $elem_ty; $id::lanes()]; + v.write_to_slice_unaligned(&mut a); + let tree_reduction = tree_reduce_sum(&a); + + // tolerate 1 ULP difference: + let red_bits = simd_reduction.to_bits(); + let tree_bits = tree_reduction.to_bits(); + assert!( + if red_bits > tree_bits { + red_bits - tree_bits + } else { + tree_bits - red_bits + } < 2, + "vector: {:?} | simd_reduction: {:?} | \ +tree_reduction: {} | scalar_reduction: {}", + v, + simd_reduction, + tree_reduction, + scalar_reduction + ); + } + + #[cfg_attr(not(target_arch = "wasm32"), test)] + #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] + #[allow(unused, dead_code)] + fn product_roundoff() { + use ::core::convert::TryInto; + // Performs a tree-reduction + fn tree_reduce_product(a: &[$elem_ty]) -> $elem_ty { + assert!(!a.is_empty()); + if a.len() == 1 { + a[0] + } else if a.len() == 2 { + a[0] * a[1] + } else { + let mid = a.len() / 2; + let (left, right) = a.split_at(mid); + tree_reduce_product(left) + * tree_reduce_product(right) + } + } + + let mut start = crate::$elem_ty::EPSILON; + let mut scalar_reduction = 1. as $elem_ty; + + let mut v = $id::splat(0. as $elem_ty); + for i in 0..$id::lanes() { + let c = if i % 2 == 0 { 1e3 } else { -1. }; + start *= ::core::$elem_ty::consts::PI * c; + scalar_reduction *= start; + v = v.replace(i, start); + } + let simd_reduction = v.product(); + + let mut a = [0. as $elem_ty; $id::lanes()]; + v.write_to_slice_unaligned(&mut a); + let tree_reduction = tree_reduce_product(&a); + + // FIXME: Too imprecise, even only for product(f32x8). + // Figure out how to narrow this down. + let ulp_limit = $id::lanes() / 2; + let red_bits = simd_reduction.to_bits(); + let tree_bits = tree_reduction.to_bits(); + assert!( + if red_bits > tree_bits { + red_bits - tree_bits + } else { + tree_bits - red_bits + } < ulp_limit.try_into().unwrap(), + "vector: {:?} | simd_reduction: {:?} | \ +tree_reduction: {} | scalar_reduction: {}", + v, + simd_reduction, + tree_reduction, + scalar_reduction + ); + } + } + } + } + }; +} diff --git a/third_party/rust/packed_simd/src/api/reductions/integer_arithmetic.rs b/third_party/rust/packed_simd/src/api/reductions/integer_arithmetic.rs new file mode 100644 index 000000000000..e99e6cb5d796 --- /dev/null +++ b/third_party/rust/packed_simd/src/api/reductions/integer_arithmetic.rs @@ -0,0 +1,193 @@ +//! Implements portable horizontal integer vector arithmetic reductions. + +macro_rules! impl_reduction_integer_arithmetic { + ([$elem_ty:ident; $elem_count:expr]: $id:ident | $ielem_ty:ident + | $test_tt:tt) => { + impl $id { + /// Horizontal wrapping sum of the vector elements. + /// + /// The intrinsic performs a tree-reduction of the vector elements. + /// That is, for an 8 element vector: + /// + /// > ((x0 + x1) + (x2 + x3)) + ((x4 + x5) + (x6 + x7)) + /// + /// If an operation overflows it returns the mathematical result + /// modulo `2^n` where `n` is the number of times it overflows. + #[inline] + pub fn wrapping_sum(self) -> $elem_ty { + #[cfg(not(target_arch = "aarch64"))] + { + use crate::llvm::simd_reduce_add_ordered; + let v: $ielem_ty = unsafe { simd_reduce_add_ordered(self.0, 0 as $ielem_ty) }; + v as $elem_ty + } + #[cfg(target_arch = "aarch64")] + { + // FIXME: broken on AArch64 + // https://github.com/rust-lang-nursery/packed_simd/issues/15 + let mut x = self.extract(0) as $elem_ty; + for i in 1..$id::lanes() { + x = x.wrapping_add(self.extract(i) as $elem_ty); + } + x + } + } + + /// Horizontal wrapping product of the vector elements. + /// + /// The intrinsic performs a tree-reduction of the vector elements. + /// That is, for an 8 element vector: + /// + /// > ((x0 * x1) * (x2 * x3)) * ((x4 * x5) * (x6 * x7)) + /// + /// If an operation overflows it returns the mathematical result + /// modulo `2^n` where `n` is the number of times it overflows. + #[inline] + pub fn wrapping_product(self) -> $elem_ty { + #[cfg(not(target_arch = "aarch64"))] + { + use crate::llvm::simd_reduce_mul_ordered; + let v: $ielem_ty = unsafe { simd_reduce_mul_ordered(self.0, 1 as $ielem_ty) }; + v as $elem_ty + } + #[cfg(target_arch = "aarch64")] + { + // FIXME: broken on AArch64 + // https://github.com/rust-lang-nursery/packed_simd/issues/15 + let mut x = self.extract(0) as $elem_ty; + for i in 1..$id::lanes() { + x = x.wrapping_mul(self.extract(i) as $elem_ty); + } + x + } + } + } + + impl crate::iter::Sum for $id { + #[inline] + fn sum>(iter: I) -> $id { + iter.fold($id::splat(0), crate::ops::Add::add) + } + } + + impl crate::iter::Product for $id { + #[inline] + fn product>(iter: I) -> $id { + iter.fold($id::splat(1), crate::ops::Mul::mul) + } + } + + impl<'a> crate::iter::Sum<&'a $id> for $id { + #[inline] + fn sum>(iter: I) -> $id { + iter.fold($id::splat(0), |a, b| crate::ops::Add::add(a, *b)) + } + } + + impl<'a> crate::iter::Product<&'a $id> for $id { + #[inline] + fn product>(iter: I) -> $id { + iter.fold($id::splat(1), |a, b| crate::ops::Mul::mul(a, *b)) + } + } + + test_if! { + $test_tt: + paste::item! { + pub mod [<$id _reduction_int_arith>] { + use super::*; + + fn alternating(x: usize) -> $id { + let mut v = $id::splat(1 as $elem_ty); + for i in 0..$id::lanes() { + if i % x == 0 { + v = v.replace(i, 2 as $elem_ty); + } + } + v + } + + #[cfg_attr(not(target_arch = "wasm32"), test)] + #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] + fn wrapping_sum() { + let v = $id::splat(0 as $elem_ty); + assert_eq!(v.wrapping_sum(), 0 as $elem_ty); + let v = $id::splat(1 as $elem_ty); + assert_eq!(v.wrapping_sum(), $id::lanes() as $elem_ty); + let v = alternating(2); + if $id::lanes() > 1 { + assert_eq!( + v.wrapping_sum(), + ($id::lanes() / 2 + $id::lanes()) as $elem_ty + ); + } else { + assert_eq!( + v.wrapping_sum(), + 2 as $elem_ty + ); + } + } + #[cfg_attr(not(target_arch = "wasm32"), test)] + #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] + fn wrapping_sum_overflow() { + let start = $elem_ty::max_value() + - ($id::lanes() as $elem_ty / 2); + + let v = $id::splat(start as $elem_ty); + let vwrapping_sum = v.wrapping_sum(); + + let mut wrapping_sum = start; + for _ in 1..$id::lanes() { + wrapping_sum = wrapping_sum.wrapping_add(start); + } + assert_eq!(wrapping_sum, vwrapping_sum, "v = {:?}", v); + } + + #[cfg_attr(not(target_arch = "wasm32"), test)] + #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] + fn wrapping_product() { + let v = $id::splat(0 as $elem_ty); + assert_eq!(v.wrapping_product(), 0 as $elem_ty); + let v = $id::splat(1 as $elem_ty); + assert_eq!(v.wrapping_product(), 1 as $elem_ty); + let f = match $id::lanes() { + 64 => 16, + 32 => 8, + 16 => 4, + _ => 2, + }; + let v = alternating(f); + if $id::lanes() > 1 { + assert_eq!( + v.wrapping_product(), + (2_usize.pow(($id::lanes() / f) as u32) + as $elem_ty) + ); + } else { + assert_eq!( + v.wrapping_product(), + 2 as $elem_ty + ); + } + } + + #[cfg_attr(not(target_arch = "wasm32"), test)] + #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] + fn wrapping_product_overflow() { + let start = $elem_ty::max_value() + - ($id::lanes() as $elem_ty / 2); + + let v = $id::splat(start as $elem_ty); + let vmul = v.wrapping_product(); + + let mut mul = start; + for _ in 1..$id::lanes() { + mul = mul.wrapping_mul(start); + } + assert_eq!(mul, vmul, "v = {:?}", v); + } + } + } + } + }; +} diff --git a/third_party/rust/packed_simd/src/api/reductions/mask.rs b/third_party/rust/packed_simd/src/api/reductions/mask.rs new file mode 100644 index 000000000000..0dd6a84e7e8d --- /dev/null +++ b/third_party/rust/packed_simd/src/api/reductions/mask.rs @@ -0,0 +1,89 @@ +//! Implements portable horizontal mask reductions. + +macro_rules! impl_reduction_mask { + ([$elem_ty:ident; $elem_count:expr]: $id:ident | $test_tt:tt) => { + impl $id { + /// Are `all` vector lanes `true`? + #[inline] + pub fn all(self) -> bool { + unsafe { crate::codegen::reductions::mask::All::all(self) } + } + /// Is `any` vector lane `true`? + #[inline] + pub fn any(self) -> bool { + unsafe { crate::codegen::reductions::mask::Any::any(self) } + } + /// Are `all` vector lanes `false`? + #[inline] + pub fn none(self) -> bool { + !self.any() + } + } + + test_if! { + $test_tt: + paste::item! { + pub mod [<$id _reduction>] { + use super::*; + #[cfg_attr(not(target_arch = "wasm32"), test)] + #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] + fn all() { + let a = $id::splat(true); + assert!(a.all()); + let a = $id::splat(false); + assert!(!a.all()); + + if $id::lanes() > 1 { + for i in 0..$id::lanes() { + let mut a = $id::splat(true); + a = a.replace(i, false); + assert!(!a.all()); + let mut a = $id::splat(false); + a = a.replace(i, true); + assert!(!a.all()); + } + } + } + #[cfg_attr(not(target_arch = "wasm32"), test)] + #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] + fn any() { + let a = $id::splat(true); + assert!(a.any()); + let a = $id::splat(false); + assert!(!a.any()); + + if $id::lanes() > 1 { + for i in 0..$id::lanes() { + let mut a = $id::splat(true); + a = a.replace(i, false); + assert!(a.any()); + let mut a = $id::splat(false); + a = a.replace(i, true); + assert!(a.any()); + } + } + } + #[cfg_attr(not(target_arch = "wasm32"), test)] + #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] + fn none() { + let a = $id::splat(true); + assert!(!a.none()); + let a = $id::splat(false); + assert!(a.none()); + + if $id::lanes() > 1 { + for i in 0..$id::lanes() { + let mut a = $id::splat(true); + a = a.replace(i, false); + assert!(!a.none()); + let mut a = $id::splat(false); + a = a.replace(i, true); + assert!(!a.none()); + } + } + } + } + } + } + }; +} diff --git a/third_party/rust/packed_simd/src/api/reductions/min_max.rs b/third_party/rust/packed_simd/src/api/reductions/min_max.rs new file mode 100644 index 000000000000..a3ce13a45123 --- /dev/null +++ b/third_party/rust/packed_simd/src/api/reductions/min_max.rs @@ -0,0 +1,360 @@ +//! Implements portable horizontal vector min/max reductions. + +macro_rules! impl_reduction_min_max { + ([$elem_ty:ident; $elem_count:expr]: $id:ident + | $ielem_ty:ident | $test_tt:tt) => { + impl $id { + /// Largest vector element value. + #[inline] + pub fn max_element(self) -> $elem_ty { + #[cfg(not(any( + target_arch = "aarch64", + target_arch = "arm", + target_arch = "powerpc64", + target_arch = "wasm32", + )))] + { + use crate::llvm::simd_reduce_max; + let v: $ielem_ty = unsafe { simd_reduce_max(self.0) }; + v as $elem_ty + } + #[cfg(any( + target_arch = "aarch64", + target_arch = "arm", + target_arch = "powerpc64", + target_arch = "wasm32", + ))] + { + // FIXME: broken on AArch64 + // https://github.com/rust-lang-nursery/packed_simd/issues/15 + // FIXME: broken on WASM32 + // https://github.com/rust-lang-nursery/packed_simd/issues/91 + let mut x = self.extract(0); + for i in 1..$id::lanes() { + x = x.max(self.extract(i)); + } + x + } + } + + /// Smallest vector element value. + #[inline] + pub fn min_element(self) -> $elem_ty { + #[cfg(not(any( + target_arch = "aarch64", + target_arch = "arm", + all(target_arch = "x86", not(target_feature = "sse2")), + target_arch = "powerpc64", + target_arch = "wasm32", + ),))] + { + use crate::llvm::simd_reduce_min; + let v: $ielem_ty = unsafe { simd_reduce_min(self.0) }; + v as $elem_ty + } + #[cfg(any( + target_arch = "aarch64", + target_arch = "arm", + all(target_arch = "x86", not(target_feature = "sse2")), + target_arch = "powerpc64", + target_arch = "wasm32", + ))] + { + // FIXME: broken on AArch64 + // https://github.com/rust-lang-nursery/packed_simd/issues/15 + // FIXME: broken on i586-unknown-linux-gnu + // https://github.com/rust-lang-nursery/packed_simd/issues/22 + // FIXME: broken on WASM32 + // https://github.com/rust-lang-nursery/packed_simd/issues/91 + let mut x = self.extract(0); + for i in 1..$id::lanes() { + x = x.min(self.extract(i)); + } + x + } + } + } + test_if! {$test_tt: + paste::item! { + // Comparisons use integer casts within mantissa^1 range. + #[allow(clippy::float_cmp)] + pub mod [<$id _reduction_min_max>] { + use super::*; + #[cfg_attr(not(target_arch = "wasm32"), test)] + #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] + pub fn max_element() { + let v = $id::splat(0 as $elem_ty); + assert_eq!(v.max_element(), 0 as $elem_ty); + if $id::lanes() > 1 { + let v = v.replace(1, 1 as $elem_ty); + assert_eq!(v.max_element(), 1 as $elem_ty); + } + let v = v.replace(0, 2 as $elem_ty); + assert_eq!(v.max_element(), 2 as $elem_ty); + } + + #[cfg_attr(not(target_arch = "wasm32"), test)] + #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] + pub fn min_element() { + let v = $id::splat(0 as $elem_ty); + assert_eq!(v.min_element(), 0 as $elem_ty); + if $id::lanes() > 1 { + let v = v.replace(1, 1 as $elem_ty); + assert_eq!(v.min_element(), 0 as $elem_ty); + } + let v = $id::splat(1 as $elem_ty); + let v = v.replace(0, 2 as $elem_ty); + if $id::lanes() > 1 { + assert_eq!(v.min_element(), 1 as $elem_ty); + } else { + assert_eq!(v.min_element(), 2 as $elem_ty); + } + if $id::lanes() > 1 { + let v = $id::splat(2 as $elem_ty); + let v = v.replace(1, 1 as $elem_ty); + assert_eq!(v.min_element(), 1 as $elem_ty); + } + } + } + } + } + }; +} + +macro_rules! test_reduction_float_min_max { + ([$elem_ty:ident; $elem_count:expr]: $id:ident | $test_tt:tt) => { + test_if! { + $test_tt: + paste::item! { + // Comparisons use integer casts within mantissa^1 range. + #[allow(clippy::float_cmp)] + pub mod [<$id _reduction_min_max_nan>] { + use super::*; + #[cfg_attr(not(target_arch = "wasm32"), test)] + #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] + fn min_element_test() { + let n = crate::$elem_ty::NAN; + + assert_eq!(n.min(-3.), -3.); + assert_eq!((-3. as $elem_ty).min(n), -3.); + + let v0 = $id::splat(-3.); + + let target_with_broken_last_lane_nan = !cfg!(any( + target_arch = "arm", target_arch = "aarch64", + all(target_arch = "x86", + not(target_feature = "sse2") + ), + target_arch = "powerpc64", + target_arch = "wasm32", + )); + + // The vector is initialized to `-3.`s: [-3, -3, -3, -3] + for i in 0..$id::lanes() { + // We replace the i-th element of the vector with + // `NaN`: [-3, -3, -3, NaN] + let mut v = v0.replace(i, n); + + // If the NaN is in the last place, the LLVM + // implementation of these methods is broken on some + // targets: + if i == $id::lanes() - 1 && + target_with_broken_last_lane_nan { + assert_eq!(v.min_element(), -3., + "[A]: nan at {} => {} | {:?}", + i, v.min_element(), v); + + // If we replace all the elements in the vector + // up-to the `i-th` lane with `NaN`s, the result + // is still always `-3.` unless all elements of + // the vector are `NaN`s: + for j in 0..i { + v = v.replace(j, n); + if j == i-1 { + assert!(v.min_element().is_nan(), + "[B]: nan at {} => {} | {:?}", + i, v.min_element(), v); + } else { + assert_eq!(v.min_element(), -3., + "[B]: nan at {} => {} | {:?}", + i, v.min_element(), v); + } + } + + // We are done here, since we were in the last + // lane which is the last iteration of the loop. + break + } + + // We are not in the last lane, and there is only + // one `NaN` in the vector. + + // If the vector has one lane, the result is `NaN`: + if $id::lanes() == 1 { + assert!(v.min_element().is_nan(), + "[C]: all nans | v={:?} | min={} | \ +is_nan: {}", + v, v.min_element(), + v.min_element().is_nan() + ); + + // And we are done, since the vector only has + // one lane anyways. + break; + } + + // The vector has more than one lane, since there is + // only one `NaN` in the vector, the result is + // always `-3`. + assert_eq!(v.min_element(), -3., + "[D]: nan at {} => {} | {:?}", + i, v.min_element(), v); + + // If we replace all the elements in the vector + // up-to the `i-th` lane with `NaN`s, the result is + // still always `-3.` unless all elements of the + // vector are `NaN`s: + for j in 0..i { + v = v.replace(j, n); + + if i == $id::lanes() - 1 && j == i - 1 { + // All elements of the vector are `NaN`s, + // therefore the result is NaN as well. + // + // Note: the #lanes of the vector is > 1, so + // "i - 1" does not overflow. + assert!(v.min_element().is_nan(), + "[E]: all nans | v={:?} | min={} | \ +is_nan: {}", + v, v.min_element(), + v.min_element().is_nan()); + } else { + // There are non-`NaN` elements in the + // vector, therefore the result is `-3.`: + assert_eq!(v.min_element(), -3., + "[F]: nan at {} => {} | {:?}", + i, v.min_element(), v); + } + } + } + + // If the vector contains all NaNs the result is NaN: + assert!($id::splat(n).min_element().is_nan(), + "all nans | v={:?} | min={} | is_nan: {}", + $id::splat(n), $id::splat(n).min_element(), + $id::splat(n).min_element().is_nan()); + } + #[cfg_attr(not(target_arch = "wasm32"), test)] + #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] + fn max_element_test() { + let n = crate::$elem_ty::NAN; + + assert_eq!(n.max(-3.), -3.); + assert_eq!((-3. as $elem_ty).max(n), -3.); + + let v0 = $id::splat(-3.); + + let target_with_broken_last_lane_nan = !cfg!(any( + target_arch = "arm", target_arch = "aarch64", + target_arch = "powerpc64", target_arch = "wasm32", + )); + + // The vector is initialized to `-3.`s: [-3, -3, -3, -3] + for i in 0..$id::lanes() { + // We replace the i-th element of the vector with + // `NaN`: [-3, -3, -3, NaN] + let mut v = v0.replace(i, n); + + // If the NaN is in the last place, the LLVM + // implementation of these methods is broken on some + // targets: + if i == $id::lanes() - 1 && + target_with_broken_last_lane_nan { + assert_eq!(v.max_element(), -3., + "[A]: nan at {} => {} | {:?}", + i, v.max_element(), v); + + // If we replace all the elements in the vector + // up-to the `i-th` lane with `NaN`s, the result + // is still always `-3.` unless all elements of + // the vector are `NaN`s: + for j in 0..i { + v = v.replace(j, n); + if j == i-1 { + assert!(v.min_element().is_nan(), + "[B]: nan at {} => {} | {:?}", + i, v.min_element(), v); + } else { + assert_eq!(v.max_element(), -3., + "[B]: nan at {} => {} | {:?}", + i, v.max_element(), v); + } + } + + // We are done here, since we were in the last + // lane which is the last iteration of the loop. + break + } + + // We are not in the last lane, and there is only + // one `NaN` in the vector. + + // If the vector has one lane, the result is `NaN`: + if $id::lanes() == 1 { + assert!(v.max_element().is_nan(), + "[C]: all nans | v={:?} | min={} | \ +is_nan: {}", + v, v.max_element(), + v.max_element().is_nan()); + + // And we are done, since the vector only has + // one lane anyways. + break; + } + + // The vector has more than one lane, since there is + // only one `NaN` in the vector, the result is + // always `-3`. + assert_eq!(v.max_element(), -3., + "[D]: nan at {} => {} | {:?}", + i, v.max_element(), v); + + // If we replace all the elements in the vector + // up-to the `i-th` lane with `NaN`s, the result is + // still always `-3.` unless all elements of the + // vector are `NaN`s: + for j in 0..i { + v = v.replace(j, n); + + if i == $id::lanes() - 1 && j == i - 1 { + // All elements of the vector are `NaN`s, + // therefore the result is NaN as well. + // + // Note: the #lanes of the vector is > 1, so + // "i - 1" does not overflow. + assert!(v.max_element().is_nan(), + "[E]: all nans | v={:?} | max={} | \ +is_nan: {}", + v, v.max_element(), + v.max_element().is_nan()); + } else { + // There are non-`NaN` elements in the + // vector, therefore the result is `-3.`: + assert_eq!(v.max_element(), -3., + "[F]: nan at {} => {} | {:?}", + i, v.max_element(), v); + } + } + } + + // If the vector contains all NaNs the result is NaN: + assert!($id::splat(n).max_element().is_nan(), + "all nans | v={:?} | max={} | is_nan: {}", + $id::splat(n), $id::splat(n).max_element(), + $id::splat(n).max_element().is_nan()); + } + } + } + } + }; +} diff --git a/third_party/rust/packed_simd/src/api/select.rs b/third_party/rust/packed_simd/src/api/select.rs new file mode 100644 index 000000000000..daf62947219e --- /dev/null +++ b/third_party/rust/packed_simd/src/api/select.rs @@ -0,0 +1,73 @@ +//! Implements mask's `select`. + +/// Implements mask select method +macro_rules! impl_select { + ([$elem_ty:ident; $elem_count:expr]: $id:ident | $test_tt:tt) => { + impl $id { + /// Selects elements of `a` and `b` using mask. + /// + /// The lanes of the result for which the mask is `true` contain + /// the values of `a`. The remaining lanes contain the values of + /// `b`. + #[inline] + pub fn select(self, a: Simd, b: Simd) -> Simd + where + T: sealed::SimdArray::NT>, + { + use crate::llvm::simd_select; + Simd(unsafe { simd_select(self.0, a.0, b.0) }) + } + } + + test_select!(bool, $id, $id, (false, true) | $test_tt); + }; +} + +macro_rules! test_select { + ( + $elem_ty:ident, + $mask_ty:ident, + $vec_ty:ident,($small:expr, $large:expr) | + $test_tt:tt + ) => { + test_if! { + $test_tt: + paste::item! { + pub mod [<$vec_ty _select>] { + use super::*; + + #[cfg_attr(not(target_arch = "wasm32"), test)] #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] + fn select() { + let o = $small as $elem_ty; + let t = $large as $elem_ty; + + let a = $vec_ty::splat(o); + let b = $vec_ty::splat(t); + let m = a.lt(b); + assert_eq!(m.select(a, b), a); + + let m = b.lt(a); + assert_eq!(m.select(b, a), a); + + let mut c = a; + let mut d = b; + let mut m_e = $mask_ty::splat(false); + for i in 0..$vec_ty::lanes() { + if i % 2 == 0 { + let c_tmp = c.extract(i); + c = c.replace(i, d.extract(i)); + d = d.replace(i, c_tmp); + } else { + m_e = m_e.replace(i, true); + } + } + + let m = c.lt(d); + assert_eq!(m_e, m); + assert_eq!(m.select(c, d), a); + } + } + } + } + }; +} diff --git a/third_party/rust/packed_simd/src/api/shuffle.rs b/third_party/rust/packed_simd/src/api/shuffle.rs new file mode 100644 index 000000000000..1c17bd766e6d --- /dev/null +++ b/third_party/rust/packed_simd/src/api/shuffle.rs @@ -0,0 +1,184 @@ +//! Implements portable vector shuffles with immediate indices. + +// FIXME: comprehensive tests +// https://github.com/rust-lang-nursery/packed_simd/issues/20 + +/// Shuffles vector elements. +/// +/// This macro returns a new vector that contains a shuffle of the elements in +/// one (`shuffle!(vec, [indices...])`) or two (`shuffle!(vec0, vec1, +/// [indices...])`) input vectors. +/// +/// The type of `vec0` and `vec1` must be equal, and the element type of the +/// resulting vector is the element type of the input vector. +/// +/// The number of `indices` must be a power-of-two in range `[0, 64)`, since +/// currently, the largest vector supported by the library has 64 lanes. The +/// length of the resulting vector equals the number of indices provided. +/// +/// The indices must be in range `[0, M * N)` where `M` is the number of input +/// vectors (`1` or `2`) and `N` is the number of lanes of the input vectors. +/// The indices `i` in range `[0, N)` refer to the `i`-th element of `vec0`, +/// while the indices in range `[N, 2*N)` refer to the `i - N`-th element of +/// `vec1`. +/// +/// # Examples +/// +/// Shuffling elements of two vectors: +/// +/// ``` +/// # use packed_simd::*; +/// # fn main() { +/// // Shuffle allows reordering the elements: +/// let x = i32x4::new(1, 2, 3, 4); +/// let y = i32x4::new(5, 6, 7, 8); +/// let r = shuffle!(x, y, [4, 0, 5, 1]); +/// assert_eq!(r, i32x4::new(5, 1, 6, 2)); +/// +/// // The resulting vector can als be smaller than the input: +/// let r = shuffle!(x, y, [1, 6]); +/// assert_eq!(r, i32x2::new(2, 7)); +/// +/// // Or larger: +/// let r = shuffle!(x, y, [1, 3, 4, 2, 1, 7, 2, 2]); +/// assert_eq!(r, i32x8::new(2, 4, 5, 3, 2, 8, 3, 3)); +/// // At most 2 * the number of lanes in the input vector. +/// # } +/// ``` +/// +/// Shuffling elements of one vector: +/// +/// ``` +/// # use packed_simd::*; +/// # fn main() { +/// // Shuffle allows reordering the elements of a vector: +/// let x = i32x4::new(1, 2, 3, 4); +/// let r = shuffle!(x, [2, 1, 3, 0]); +/// assert_eq!(r, i32x4::new(3, 2, 4, 1)); +/// +/// // The resulting vector can be smaller than the input: +/// let r = shuffle!(x, [1, 3]); +/// assert_eq!(r, i32x2::new(2, 4)); +/// +/// // Equal: +/// let r = shuffle!(x, [1, 3, 2, 0]); +/// assert_eq!(r, i32x4::new(2, 4, 3, 1)); +/// +/// // Or larger: +/// let r = shuffle!(x, [1, 3, 2, 2, 1, 3, 2, 2]); +/// assert_eq!(r, i32x8::new(2, 4, 3, 3, 2, 4, 3, 3)); +/// // At most 2 * the number of lanes in the input vector. +/// # } +/// ``` +#[macro_export] +macro_rules! shuffle { + ($vec0:expr, $vec1:expr, [$l0:expr, $l1:expr]) => {{ + #[allow(unused_unsafe)] + unsafe { + $crate::Simd($crate::__shuffle_vector2::<{[$l0, $l1]}, _, _>( + $vec0.0, + $vec1.0, + )) + } + }}; + ($vec0:expr, $vec1:expr, [$l0:expr, $l1:expr, $l2:expr, $l3:expr]) => {{ + #[allow(unused_unsafe)] + unsafe { + $crate::Simd($crate::__shuffle_vector4::<{[$l0, $l1, $l2, $l3]}, _, _>( + $vec0.0, + $vec1.0, + )) + } + }}; + ($vec0:expr, $vec1:expr, + [$l0:expr, $l1:expr, $l2:expr, $l3:expr, + $l4:expr, $l5:expr, $l6:expr, $l7:expr]) => {{ + #[allow(unused_unsafe)] + unsafe { + $crate::Simd($crate::__shuffle_vector8::<{[$l0, $l1, $l2, $l3, $l4, $l5, $l6, $l7]}, _, _>( + $vec0.0, + $vec1.0, + )) + } + }}; + ($vec0:expr, $vec1:expr, + [$l0:expr, $l1:expr, $l2:expr, $l3:expr, + $l4:expr, $l5:expr, $l6:expr, $l7:expr, + $l8:expr, $l9:expr, $l10:expr, $l11:expr, + $l12:expr, $l13:expr, $l14:expr, $l15:expr]) => {{ + #[allow(unused_unsafe)] + unsafe { + $crate::Simd($crate::__shuffle_vector16::<{ + [ + $l0, $l1, $l2, $l3, $l4, $l5, $l6, $l7, $l8, $l9, $l10, + $l11, $l12, $l13, $l14, $l15, + ] + }, _, _>( + $vec0.0, + $vec1.0, + )) + } + }}; + ($vec0:expr, $vec1:expr, + [$l0:expr, $l1:expr, $l2:expr, $l3:expr, + $l4:expr, $l5:expr, $l6:expr, $l7:expr, + $l8:expr, $l9:expr, $l10:expr, $l11:expr, + $l12:expr, $l13:expr, $l14:expr, $l15:expr, + $l16:expr, $l17:expr, $l18:expr, $l19:expr, + $l20:expr, $l21:expr, $l22:expr, $l23:expr, + $l24:expr, $l25:expr, $l26:expr, $l27:expr, + $l28:expr, $l29:expr, $l30:expr, $l31:expr]) => {{ + #[allow(unused_unsafe)] + unsafe { + $crate::Simd($crate::__shuffle_vector32::<{ + [ + $l0, $l1, $l2, $l3, $l4, $l5, $l6, $l7, $l8, $l9, $l10, + $l11, $l12, $l13, $l14, $l15, $l16, $l17, $l18, $l19, + $l20, $l21, $l22, $l23, $l24, $l25, $l26, $l27, $l28, + $l29, $l30, $l31, + ] + }, _, _>( + $vec0.0, + $vec1.0, + )) + } + }}; + ($vec0:expr, $vec1:expr, + [$l0:expr, $l1:expr, $l2:expr, $l3:expr, + $l4:expr, $l5:expr, $l6:expr, $l7:expr, + $l8:expr, $l9:expr, $l10:expr, $l11:expr, + $l12:expr, $l13:expr, $l14:expr, $l15:expr, + $l16:expr, $l17:expr, $l18:expr, $l19:expr, + $l20:expr, $l21:expr, $l22:expr, $l23:expr, + $l24:expr, $l25:expr, $l26:expr, $l27:expr, + $l28:expr, $l29:expr, $l30:expr, $l31:expr, + $l32:expr, $l33:expr, $l34:expr, $l35:expr, + $l36:expr, $l37:expr, $l38:expr, $l39:expr, + $l40:expr, $l41:expr, $l42:expr, $l43:expr, + $l44:expr, $l45:expr, $l46:expr, $l47:expr, + $l48:expr, $l49:expr, $l50:expr, $l51:expr, + $l52:expr, $l53:expr, $l54:expr, $l55:expr, + $l56:expr, $l57:expr, $l58:expr, $l59:expr, + $l60:expr, $l61:expr, $l62:expr, $l63:expr]) => {{ + #[allow(unused_unsafe)] + unsafe { + $crate::Simd($crate::__shuffle_vector64::<{[ + $l0, $l1, $l2, $l3, $l4, $l5, $l6, $l7, $l8, $l9, $l10, + $l11, $l12, $l13, $l14, $l15, $l16, $l17, $l18, $l19, + $l20, $l21, $l22, $l23, $l24, $l25, $l26, $l27, $l28, + $l29, $l30, $l31, $l32, $l33, $l34, $l35, $l36, $l37, + $l38, $l39, $l40, $l41, $l42, $l43, $l44, $l45, $l46, + $l47, $l48, $l49, $l50, $l51, $l52, $l53, $l54, $l55, + $l56, $l57, $l58, $l59, $l60, $l61, $l62, $l63, + ]}, _, _>( + $vec0.0, + $vec1.0, + )) + } + }}; + ($vec:expr, [$($l:expr),*]) => { + match $vec { + v => shuffle!(v, v, [$($l),*]) + } + }; +} diff --git a/third_party/rust/packed_simd/src/api/shuffle1_dyn.rs b/third_party/rust/packed_simd/src/api/shuffle1_dyn.rs new file mode 100644 index 000000000000..64536be6cba1 --- /dev/null +++ b/third_party/rust/packed_simd/src/api/shuffle1_dyn.rs @@ -0,0 +1,159 @@ +//! Shuffle vector elements according to a dynamic vector of indices. + +macro_rules! impl_shuffle1_dyn { + ([$elem_ty:ident; $elem_count:expr]: $id:ident | $test_tt:tt) => { + impl $id { + /// Shuffle vector elements according to `indices`. + #[inline] + pub fn shuffle1_dyn(self, indices: I) -> Self + where + Self: codegen::shuffle1_dyn::Shuffle1Dyn, + { + codegen::shuffle1_dyn::Shuffle1Dyn::shuffle1_dyn(self, indices) + } + } + }; +} + +macro_rules! test_shuffle1_dyn { + ([$elem_ty:ident; $elem_count:expr]: $id:ident | $test_tt:tt) => { + test_if! { + $test_tt: + paste::item! { + pub mod [<$id _shuffle1_dyn>] { + use super::*; + #[cfg_attr(not(target_arch = "wasm32"), test)] #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] + fn shuffle1_dyn() { + let increasing = { + let mut v = $id::splat(0 as $elem_ty); + for i in 0..$id::lanes() { + v = v.replace(i, i as $elem_ty); + } + v + }; + let decreasing = { + let mut v = $id::splat(0 as $elem_ty); + for i in 0..$id::lanes() { + v = v.replace( + i, + ($id::lanes() - 1 - i) as $elem_ty + ); + } + v + }; + + type Indices = < + $id as codegen::shuffle1_dyn::Shuffle1Dyn + >::Indices; + let increasing_ids: Indices = increasing.cast(); + let decreasing_ids: Indices = decreasing.cast(); + + assert_eq!( + increasing.shuffle1_dyn(increasing_ids), + increasing, + "(i,i)=>i" + ); + assert_eq!( + decreasing.shuffle1_dyn(increasing_ids), + decreasing, + "(d,i)=>d" + ); + assert_eq!( + increasing.shuffle1_dyn(decreasing_ids), + decreasing, + "(i,d)=>d" + ); + assert_eq!( + decreasing.shuffle1_dyn(decreasing_ids), + increasing, + "(d,d)=>i" + ); + + for i in 0..$id::lanes() { + let v_ids: Indices + = $id::splat(i as $elem_ty).cast(); + assert_eq!(increasing.shuffle1_dyn(v_ids), + $id::splat(increasing.extract(i)) + ); + assert_eq!(decreasing.shuffle1_dyn(v_ids), + $id::splat(decreasing.extract(i)) + ); + assert_eq!( + $id::splat(i as $elem_ty) + .shuffle1_dyn(increasing_ids), + $id::splat(i as $elem_ty) + ); + assert_eq!( + $id::splat(i as $elem_ty) + .shuffle1_dyn(decreasing_ids), + $id::splat(i as $elem_ty) + ); + } + } + } + } + } + }; +} + +macro_rules! test_shuffle1_dyn_mask { + ([$elem_ty:ident; $elem_count:expr]: $id:ident | $test_tt:tt) => { + test_if! { + $test_tt: + paste::item! { + pub mod [<$id _shuffle1_dyn>] { + use super::*; + #[cfg_attr(not(target_arch = "wasm32"), test)] #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] + fn shuffle1_dyn() { + // alternating = [true, false, true, false, ...] + let mut alternating = $id::splat(false); + for i in 0..$id::lanes() { + if i % 2 == 0 { + alternating = alternating.replace(i, true); + } + } + + type Indices = < + $id as codegen::shuffle1_dyn::Shuffle1Dyn + >::Indices; + // even = [0, 0, 2, 2, 4, 4, ..] + let even = { + let mut v = Indices::splat(0); + for i in 0..$id::lanes() { + if i % 2 == 0 { + v = v.replace(i, (i as u8).into()); + } else { + v = v.replace(i, (i as u8 - 1).into()); + } + } + v + }; + // odd = [1, 1, 3, 3, 5, 5, ...] + let odd = { + let mut v = Indices::splat(0); + for i in 0..$id::lanes() { + if i % 2 != 0 { + v = v.replace(i, (i as u8).into()); + } else { + v = v.replace(i, (i as u8 + 1).into()); + } + } + v + }; + + assert_eq!( + alternating.shuffle1_dyn(even), + $id::splat(true) + ); + if $id::lanes() > 1 { + assert_eq!( + alternating.shuffle1_dyn(odd), + $id::splat(false) + ); + } + } + } + } + } + }; +} diff --git a/third_party/rust/packed_simd/src/api/slice.rs b/third_party/rust/packed_simd/src/api/slice.rs new file mode 100644 index 000000000000..526b848b5c06 --- /dev/null +++ b/third_party/rust/packed_simd/src/api/slice.rs @@ -0,0 +1,7 @@ +//! Slice from/to methods + +#[macro_use] +mod from_slice; + +#[macro_use] +mod write_to_slice; diff --git a/third_party/rust/packed_simd/src/api/slice/from_slice.rs b/third_party/rust/packed_simd/src/api/slice/from_slice.rs new file mode 100644 index 000000000000..cafd6f8213b1 --- /dev/null +++ b/third_party/rust/packed_simd/src/api/slice/from_slice.rs @@ -0,0 +1,202 @@ +//! Implements methods to read a vector type from a slice. + +macro_rules! impl_slice_from_slice { + ([$elem_ty:ident; $elem_count:expr]: $id:ident | $test_tt:tt) => { + impl $id { + /// Instantiates a new vector with the values of the `slice`. + /// + /// # Panics + /// + /// If `slice.len() < Self::lanes()` or `&slice[0]` is not aligned + /// to an `align_of::()` boundary. + #[inline] + pub fn from_slice_aligned(slice: &[$elem_ty]) -> Self { + unsafe { + assert!(slice.len() >= $elem_count); + let target_ptr = slice.as_ptr(); + assert_eq!(target_ptr.align_offset(crate::mem::align_of::()), 0); + Self::from_slice_aligned_unchecked(slice) + } + } + + /// Instantiates a new vector with the values of the `slice`. + /// + /// # Panics + /// + /// If `slice.len() < Self::lanes()`. + #[inline] + pub fn from_slice_unaligned(slice: &[$elem_ty]) -> Self { + unsafe { + assert!(slice.len() >= $elem_count); + Self::from_slice_unaligned_unchecked(slice) + } + } + + /// Instantiates a new vector with the values of the `slice`. + /// + /// # Safety + /// + /// If `slice.len() < Self::lanes()` or `&slice[0]` is not aligned + /// to an `align_of::()` boundary, the behavior is undefined. + #[inline] + pub unsafe fn from_slice_aligned_unchecked(slice: &[$elem_ty]) -> Self { + debug_assert!(slice.len() >= $elem_count); + let target_ptr = slice.as_ptr(); + debug_assert_eq!(target_ptr.align_offset(crate::mem::align_of::()), 0); + + #[allow(clippy::cast_ptr_alignment)] + *(target_ptr as *const Self) + } + + /// Instantiates a new vector with the values of the `slice`. + /// + /// # Safety + /// + /// If `slice.len() < Self::lanes()` the behavior is undefined. + #[inline] + pub unsafe fn from_slice_unaligned_unchecked(slice: &[$elem_ty]) -> Self { + use crate::mem::size_of; + debug_assert!(slice.len() >= $elem_count); + let target_ptr = slice.as_ptr().cast(); + let mut x = Self::splat(0 as $elem_ty); + let self_ptr = &mut x as *mut Self as *mut u8; + crate::ptr::copy_nonoverlapping(target_ptr, self_ptr, size_of::()); + x + } + } + + test_if! { + $test_tt: + paste::item! { + // Comparisons use integer casts within mantissa^1 range. + #[allow(clippy::float_cmp)] + pub mod [<$id _slice_from_slice>] { + use super::*; + use crate::iter::Iterator; + + #[cfg_attr(not(target_arch = "wasm32"), test)] + #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] + fn from_slice_unaligned() { + let mut unaligned = [42 as $elem_ty; $id::lanes() + 1]; + unaligned[0] = 0 as $elem_ty; + let vec = $id::from_slice_unaligned(&unaligned[1..]); + for (index, &b) in unaligned.iter().enumerate() { + if index == 0 { + assert_eq!(b, 0 as $elem_ty); + } else { + assert_eq!(b, 42 as $elem_ty); + assert_eq!(b, vec.extract(index - 1)); + } + } + } + + // FIXME: wasm-bindgen-test does not support #[should_panic] + // #[cfg_attr(not(target_arch = "wasm32"), test)] + // #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] + #[cfg(not(target_arch = "wasm32"))] + #[test] + #[should_panic] + fn from_slice_unaligned_fail() { + let mut unaligned = [42 as $elem_ty; $id::lanes() + 1]; + unaligned[0] = 0 as $elem_ty; + // the slice is not large enough => panic + let _vec = $id::from_slice_unaligned(&unaligned[2..]); + } + + union A { + data: [$elem_ty; 2 * $id::lanes()], + _vec: $id, + } + + #[cfg_attr(not(target_arch = "wasm32"), test)] + #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] + fn from_slice_aligned() { + let mut aligned = A { + data: [0 as $elem_ty; 2 * $id::lanes()], + }; + for i in $id::lanes()..(2 * $id::lanes()) { + unsafe { + aligned.data[i] = 42 as $elem_ty; + } + } + + let vec = unsafe { + $id::from_slice_aligned( + &aligned.data[$id::lanes()..] + ) + }; + for (index, &b) in + unsafe { aligned.data.iter().enumerate() } { + if index < $id::lanes() { + assert_eq!(b, 0 as $elem_ty); + } else { + assert_eq!(b, 42 as $elem_ty); + assert_eq!( + b, vec.extract(index - $id::lanes()) + ); + } + } + } + + // FIXME: wasm-bindgen-test does not support #[should_panic] + // #[cfg_attr(not(target_arch = "wasm32"), test)] + // #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] + #[cfg(not(target_arch = "wasm32"))] + #[test] + #[should_panic] + fn from_slice_aligned_fail_lanes() { + let aligned = A { + data: [0 as $elem_ty; 2 * $id::lanes()], + }; + let _vec = unsafe { + $id::from_slice_aligned( + &aligned.data[2 * $id::lanes()..] + ) + }; + } + + // FIXME: wasm-bindgen-test does not support #[should_panic] + // #[cfg_attr(not(target_arch = "wasm32"), test)] + // #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] + #[cfg(not(target_arch = "wasm32"))] + #[test] + #[should_panic] + fn from_slice_aligned_fail_align() { + unsafe { + let aligned = A { + data: [0 as $elem_ty; 2 * $id::lanes()], + }; + + // get a pointer to the front of data + let ptr: *const $elem_ty = aligned.data.as_ptr() + as *const $elem_ty; + // offset pointer by one element + let ptr = ptr.wrapping_add(1); + + if ptr.align_offset( + crate::mem::align_of::<$id>() + ) == 0 { + // the pointer is properly aligned, so + // from_slice_aligned won't fail here (e.g. this + // can happen for i128x1). So we panic to make + // the "should_fail" test pass: + panic!("ok"); + } + + // create a slice - this is safe, because the + // elements of the slice exist, are properly + // initialized, and properly aligned: + let s: &[$elem_ty] = slice::from_raw_parts( + ptr, $id::lanes() + ); + // this should always panic because the slice + // alignment does not match the alignment + // requirements for the vector type: + let _vec = $id::from_slice_aligned(s); + } + } + } + } + } + }; +} diff --git a/third_party/rust/packed_simd/src/api/slice/write_to_slice.rs b/third_party/rust/packed_simd/src/api/slice/write_to_slice.rs new file mode 100644 index 000000000000..5abd4916e026 --- /dev/null +++ b/third_party/rust/packed_simd/src/api/slice/write_to_slice.rs @@ -0,0 +1,196 @@ +//! Implements methods to write a vector type to a slice. + +macro_rules! impl_slice_write_to_slice { + ([$elem_ty:ident; $elem_count:expr]: $id:ident | $test_tt:tt) => { + impl $id { + /// Writes the values of the vector to the `slice`. + /// + /// # Panics + /// + /// If `slice.len() < Self::lanes()` or `&slice[0]` is not + /// aligned to an `align_of::()` boundary. + #[inline] + pub fn write_to_slice_aligned(self, slice: &mut [$elem_ty]) { + unsafe { + assert!(slice.len() >= $elem_count); + let target_ptr = slice.as_mut_ptr(); + assert_eq!(target_ptr.align_offset(crate::mem::align_of::()), 0); + self.write_to_slice_aligned_unchecked(slice); + } + } + + /// Writes the values of the vector to the `slice`. + /// + /// # Panics + /// + /// If `slice.len() < Self::lanes()`. + #[inline] + pub fn write_to_slice_unaligned(self, slice: &mut [$elem_ty]) { + unsafe { + assert!(slice.len() >= $elem_count); + self.write_to_slice_unaligned_unchecked(slice); + } + } + + /// Writes the values of the vector to the `slice`. + /// + /// # Safety + /// + /// If `slice.len() < Self::lanes()` or `&slice[0]` is not + /// aligned to an `align_of::()` boundary, the behavior is + /// undefined. + #[inline] + pub unsafe fn write_to_slice_aligned_unchecked(self, slice: &mut [$elem_ty]) { + debug_assert!(slice.len() >= $elem_count); + let target_ptr = slice.as_mut_ptr(); + debug_assert_eq!(target_ptr.align_offset(crate::mem::align_of::()), 0); + + #[allow(clippy::cast_ptr_alignment)] + #[allow(clippy::cast_ptr_alignment)] + #[allow(clippy::cast_ptr_alignment)] + #[allow(clippy::cast_ptr_alignment)] + *(target_ptr as *mut Self) = self; + } + + /// Writes the values of the vector to the `slice`. + /// + /// # Safety + /// + /// If `slice.len() < Self::lanes()` the behavior is undefined. + #[inline] + pub unsafe fn write_to_slice_unaligned_unchecked(self, slice: &mut [$elem_ty]) { + debug_assert!(slice.len() >= $elem_count); + let target_ptr = slice.as_mut_ptr().cast(); + let self_ptr = &self as *const Self as *const u8; + crate::ptr::copy_nonoverlapping(self_ptr, target_ptr, crate::mem::size_of::()); + } + } + + test_if! { + $test_tt: + paste::item! { + // Comparisons use integer casts within mantissa^1 range. + #[allow(clippy::float_cmp)] + pub mod [<$id _slice_write_to_slice>] { + use super::*; + use crate::iter::Iterator; + + #[cfg_attr(not(target_arch = "wasm32"), test)] + #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] + fn write_to_slice_unaligned() { + let mut unaligned = [0 as $elem_ty; $id::lanes() + 1]; + let vec = $id::splat(42 as $elem_ty); + vec.write_to_slice_unaligned(&mut unaligned[1..]); + for (index, &b) in unaligned.iter().enumerate() { + if index == 0 { + assert_eq!(b, 0 as $elem_ty); + } else { + assert_eq!(b, 42 as $elem_ty); + assert_eq!(b, vec.extract(index - 1)); + } + } + } + + // FIXME: wasm-bindgen-test does not support #[should_panic] + // #[cfg_attr(not(target_arch = "wasm32"), test)] + // #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] + #[cfg(not(target_arch = "wasm32"))] + #[test] + #[should_panic] + fn write_to_slice_unaligned_fail() { + let mut unaligned = [0 as $elem_ty; $id::lanes() + 1]; + let vec = $id::splat(42 as $elem_ty); + vec.write_to_slice_unaligned(&mut unaligned[2..]); + } + + union A { + data: [$elem_ty; 2 * $id::lanes()], + _vec: $id, + } + + #[cfg_attr(not(target_arch = "wasm32"), test)] + #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] + fn write_to_slice_aligned() { + let mut aligned = A { + data: [0 as $elem_ty; 2 * $id::lanes()], + }; + let vec = $id::splat(42 as $elem_ty); + unsafe { + vec.write_to_slice_aligned( + &mut aligned.data[$id::lanes()..] + ); + for (idx, &b) in aligned.data.iter().enumerate() { + if idx < $id::lanes() { + assert_eq!(b, 0 as $elem_ty); + } else { + assert_eq!(b, 42 as $elem_ty); + assert_eq!( + b, vec.extract(idx - $id::lanes()) + ); + } + } + } + } + + // FIXME: wasm-bindgen-test does not support #[should_panic] + // #[cfg_attr(not(target_arch = "wasm32"), test)] + // #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] + #[cfg(not(target_arch = "wasm32"))] + #[test] + #[should_panic] + fn write_to_slice_aligned_fail_lanes() { + let mut aligned = A { + data: [0 as $elem_ty; 2 * $id::lanes()], + }; + let vec = $id::splat(42 as $elem_ty); + unsafe { + vec.write_to_slice_aligned( + &mut aligned.data[2 * $id::lanes()..] + ) + }; + } + + // FIXME: wasm-bindgen-test does not support #[should_panic] + // #[cfg_attr(not(target_arch = "wasm32"), test)] + // #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] + #[cfg(not(target_arch = "wasm32"))] + #[test] + #[should_panic] + fn write_to_slice_aligned_fail_align() { + unsafe { + let mut aligned = A { + data: [0 as $elem_ty; 2 * $id::lanes()], + }; + + // get a pointer to the front of data + let ptr: *mut $elem_ty + = aligned.data.as_mut_ptr() as *mut $elem_ty; + // offset pointer by one element + let ptr = ptr.wrapping_add(1); + + if ptr.align_offset(crate::mem::align_of::<$id>()) + == 0 { + // the pointer is properly aligned, so + // write_to_slice_aligned won't fail here (e.g. + // this can happen for i128x1). So we panic to + // make the "should_fail" test pass: + panic!("ok"); + } + + // create a slice - this is safe, because the + // elements of the slice exist, are properly + // initialized, and properly aligned: + let s: &mut [$elem_ty] + = slice::from_raw_parts_mut(ptr, $id::lanes()); + // this should always panic because the slice + // alignment does not match the alignment + // requirements for the vector type: + let vec = $id::splat(42 as $elem_ty); + vec.write_to_slice_aligned(s); + } + } + } + } + } + }; +} diff --git a/third_party/rust/packed_simd/src/api/swap_bytes.rs b/third_party/rust/packed_simd/src/api/swap_bytes.rs new file mode 100644 index 000000000000..4649ed679b9d --- /dev/null +++ b/third_party/rust/packed_simd/src/api/swap_bytes.rs @@ -0,0 +1,192 @@ +//! Horizontal swap bytes + +macro_rules! impl_swap_bytes { + ([$elem_ty:ident; $elem_count:expr]: $id:ident | $test_tt:tt) => { + impl $id { + /// Reverses the byte order of the vector. + #[inline] + pub fn swap_bytes(self) -> Self { + super::codegen::swap_bytes::SwapBytes::swap_bytes(self) + } + + /// Converts self to little endian from the target's endianness. + /// + /// On little endian this is a no-op. On big endian the bytes are + /// swapped. + #[inline] + pub fn to_le(self) -> Self { + #[cfg(target_endian = "little")] + { + self + } + #[cfg(not(target_endian = "little"))] + { + self.swap_bytes() + } + } + + /// Converts self to big endian from the target's endianness. + /// + /// On big endian this is a no-op. On little endian the bytes are + /// swapped. + #[inline] + pub fn to_be(self) -> Self { + #[cfg(target_endian = "big")] + { + self + } + #[cfg(not(target_endian = "big"))] + { + self.swap_bytes() + } + } + + /// Converts a vector from little endian to the target's endianness. + /// + /// On little endian this is a no-op. On big endian the bytes are + /// swapped. + #[inline] + pub fn from_le(x: Self) -> Self { + #[cfg(target_endian = "little")] + { + x + } + #[cfg(not(target_endian = "little"))] + { + x.swap_bytes() + } + } + + /// Converts a vector from big endian to the target's endianness. + /// + /// On big endian this is a no-op. On little endian the bytes are + /// swapped. + #[inline] + pub fn from_be(x: Self) -> Self { + #[cfg(target_endian = "big")] + { + x + } + #[cfg(not(target_endian = "big"))] + { + x.swap_bytes() + } + } + } + + test_if! { + $test_tt: + paste::item! { + pub mod [<$id _swap_bytes>] { + use super::*; + + const BYTES: [u8; 64] = [ + 0, 1, 2, 3, 4, 5, 6, 7, + 8, 9, 10, 11, 12, 13, 14, 15, + 16, 17, 18, 19, 20, 21, 22, 23, + 24, 25, 26, 27, 28, 29, 30, 31, + 32, 33, 34, 35, 36, 37, 38, 39, + 40, 41, 42, 43, 44, 45, 46, 47, + 48, 49, 50, 51, 52, 53, 54, 55, + 56, 57, 58, 59, 60, 61, 62, 63, + ]; + + macro_rules! swap { + ($func: ident) => {{ + // catch possible future >512 vectors + assert!(mem::size_of::<$id>() <= 64); + + let mut actual = BYTES; + let elems: &mut [$elem_ty] = unsafe { + slice::from_raw_parts_mut( + actual.as_mut_ptr() as *mut $elem_ty, + $id::lanes(), + ) + }; + + let vec = $id::from_slice_unaligned(elems); + $id::$func(vec).write_to_slice_unaligned(elems); + + actual + }}; + } + + macro_rules! test_swap { + ($func: ident) => {{ + let actual = swap!($func); + let expected = + BYTES.iter().rev() + .skip(64 - crate::mem::size_of::<$id>()); + assert!(actual.iter().zip(expected) + .all(|(x, y)| x == y)); + }}; + } + + macro_rules! test_no_swap { + ($func: ident) => {{ + let actual = swap!($func); + let expected = BYTES.iter() + .take(mem::size_of::<$id>()); + + assert!(actual.iter().zip(expected) + .all(|(x, y)| x == y)); + }}; + } + + #[cfg_attr(not(target_arch = "wasm32"), test)] #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] + fn swap_bytes() { + test_swap!(swap_bytes); + } + + #[cfg_attr(not(target_arch = "wasm32"), test)] #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] + fn to_le() { + #[cfg(target_endian = "little")] + { + test_no_swap!(to_le); + } + #[cfg(not(target_endian = "little"))] + { + test_swap!(to_le); + } + } + + #[cfg_attr(not(target_arch = "wasm32"), test)] #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] + fn to_be() { + #[cfg(target_endian = "big")] + { + test_no_swap!(to_be); + } + #[cfg(not(target_endian = "big"))] + { + test_swap!(to_be); + } + } + + #[cfg_attr(not(target_arch = "wasm32"), test)] #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] + fn from_le() { + #[cfg(target_endian = "little")] + { + test_no_swap!(from_le); + } + #[cfg(not(target_endian = "little"))] + { + test_swap!(from_le); + } + } + + #[cfg_attr(not(target_arch = "wasm32"), test)] #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] + fn from_be() { + #[cfg(target_endian = "big")] + { + test_no_swap!(from_be); + } + #[cfg(not(target_endian = "big"))] + { + test_swap!(from_be); + } + } + } + } + } + }; +} diff --git a/third_party/rust/packed_simd/src/codegen.rs b/third_party/rust/packed_simd/src/codegen.rs new file mode 100644 index 000000000000..8a9e971486d7 --- /dev/null +++ b/third_party/rust/packed_simd/src/codegen.rs @@ -0,0 +1,62 @@ +//! Code-generation utilities + +pub(crate) mod bit_manip; +pub(crate) mod llvm; +pub(crate) mod math; +pub(crate) mod reductions; +pub(crate) mod shuffle; +pub(crate) mod shuffle1_dyn; +pub(crate) mod swap_bytes; + +macro_rules! impl_simd_array { + ([$elem_ty:ident; $elem_count:expr]: + $tuple_id:ident | $($elem_tys:ident),*) => { + #[derive(Copy, Clone)] + #[repr(simd)] + pub struct $tuple_id($(pub(crate) $elem_tys),*); + //^^^^^^^ leaked through SimdArray + + impl crate::sealed::Seal for [$elem_ty; $elem_count] {} + + impl crate::sealed::SimdArray for [$elem_ty; $elem_count] { + type Tuple = $tuple_id; + type T = $elem_ty; + const N: usize = $elem_count; + type NT = [u32; $elem_count]; + } + + impl crate::sealed::Seal for $tuple_id {} + impl crate::sealed::Simd for $tuple_id { + type Element = $elem_ty; + const LANES: usize = $elem_count; + type LanesType = [u32; $elem_count]; + } + + } +} + +pub(crate) mod pointer_sized_int; + +pub(crate) mod v16; +pub(crate) use self::v16::*; + +pub(crate) mod v32; +pub(crate) use self::v32::*; + +pub(crate) mod v64; +pub(crate) use self::v64::*; + +pub(crate) mod v128; +pub(crate) use self::v128::*; + +pub(crate) mod v256; +pub(crate) use self::v256::*; + +pub(crate) mod v512; +pub(crate) use self::v512::*; + +pub(crate) mod vSize; +pub(crate) use self::vSize::*; + +pub(crate) mod vPtr; +pub(crate) use self::vPtr::*; diff --git a/third_party/rust/packed_simd/src/codegen/bit_manip.rs b/third_party/rust/packed_simd/src/codegen/bit_manip.rs new file mode 100644 index 000000000000..32d8d717a076 --- /dev/null +++ b/third_party/rust/packed_simd/src/codegen/bit_manip.rs @@ -0,0 +1,347 @@ +//! LLVM bit manipulation intrinsics. +#[rustfmt::skip] + +pub(crate) use crate::*; + +#[allow(improper_ctypes, dead_code)] +extern "C" { + #[link_name = "llvm.ctlz.v2i8"] + fn ctlz_u8x2(x: u8x2, is_zero_undef: bool) -> u8x2; + #[link_name = "llvm.ctlz.v4i8"] + fn ctlz_u8x4(x: u8x4, is_zero_undef: bool) -> u8x4; + #[link_name = "llvm.ctlz.v8i8"] + fn ctlz_u8x8(x: u8x8, is_zero_undef: bool) -> u8x8; + #[link_name = "llvm.ctlz.v16i8"] + fn ctlz_u8x16(x: u8x16, is_zero_undef: bool) -> u8x16; + #[link_name = "llvm.ctlz.v32i8"] + fn ctlz_u8x32(x: u8x32, is_zero_undef: bool) -> u8x32; + #[link_name = "llvm.ctlz.v64i8"] + fn ctlz_u8x64(x: u8x64, is_zero_undef: bool) -> u8x64; + + #[link_name = "llvm.ctlz.v2i16"] + fn ctlz_u16x2(x: u16x2, is_zero_undef: bool) -> u16x2; + #[link_name = "llvm.ctlz.v4i16"] + fn ctlz_u16x4(x: u16x4, is_zero_undef: bool) -> u16x4; + #[link_name = "llvm.ctlz.v8i16"] + fn ctlz_u16x8(x: u16x8, is_zero_undef: bool) -> u16x8; + #[link_name = "llvm.ctlz.v16i16"] + fn ctlz_u16x16(x: u16x16, is_zero_undef: bool) -> u16x16; + #[link_name = "llvm.ctlz.v32i16"] + fn ctlz_u16x32(x: u16x32, is_zero_undef: bool) -> u16x32; + + #[link_name = "llvm.ctlz.v2i32"] + fn ctlz_u32x2(x: u32x2, is_zero_undef: bool) -> u32x2; + #[link_name = "llvm.ctlz.v4i32"] + fn ctlz_u32x4(x: u32x4, is_zero_undef: bool) -> u32x4; + #[link_name = "llvm.ctlz.v8i32"] + fn ctlz_u32x8(x: u32x8, is_zero_undef: bool) -> u32x8; + #[link_name = "llvm.ctlz.v16i32"] + fn ctlz_u32x16(x: u32x16, is_zero_undef: bool) -> u32x16; + + #[link_name = "llvm.ctlz.v2i64"] + fn ctlz_u64x2(x: u64x2, is_zero_undef: bool) -> u64x2; + #[link_name = "llvm.ctlz.v4i64"] + fn ctlz_u64x4(x: u64x4, is_zero_undef: bool) -> u64x4; + #[link_name = "llvm.ctlz.v8i64"] + fn ctlz_u64x8(x: u64x8, is_zero_undef: bool) -> u64x8; + + #[link_name = "llvm.ctlz.v1i128"] + fn ctlz_u128x1(x: u128x1, is_zero_undef: bool) -> u128x1; + #[link_name = "llvm.ctlz.v2i128"] + fn ctlz_u128x2(x: u128x2, is_zero_undef: bool) -> u128x2; + #[link_name = "llvm.ctlz.v4i128"] + fn ctlz_u128x4(x: u128x4, is_zero_undef: bool) -> u128x4; + + #[link_name = "llvm.cttz.v2i8"] + fn cttz_u8x2(x: u8x2, is_zero_undef: bool) -> u8x2; + #[link_name = "llvm.cttz.v4i8"] + fn cttz_u8x4(x: u8x4, is_zero_undef: bool) -> u8x4; + #[link_name = "llvm.cttz.v8i8"] + fn cttz_u8x8(x: u8x8, is_zero_undef: bool) -> u8x8; + #[link_name = "llvm.cttz.v16i8"] + fn cttz_u8x16(x: u8x16, is_zero_undef: bool) -> u8x16; + #[link_name = "llvm.cttz.v32i8"] + fn cttz_u8x32(x: u8x32, is_zero_undef: bool) -> u8x32; + #[link_name = "llvm.cttz.v64i8"] + fn cttz_u8x64(x: u8x64, is_zero_undef: bool) -> u8x64; + + #[link_name = "llvm.cttz.v2i16"] + fn cttz_u16x2(x: u16x2, is_zero_undef: bool) -> u16x2; + #[link_name = "llvm.cttz.v4i16"] + fn cttz_u16x4(x: u16x4, is_zero_undef: bool) -> u16x4; + #[link_name = "llvm.cttz.v8i16"] + fn cttz_u16x8(x: u16x8, is_zero_undef: bool) -> u16x8; + #[link_name = "llvm.cttz.v16i16"] + fn cttz_u16x16(x: u16x16, is_zero_undef: bool) -> u16x16; + #[link_name = "llvm.cttz.v32i16"] + fn cttz_u16x32(x: u16x32, is_zero_undef: bool) -> u16x32; + + #[link_name = "llvm.cttz.v2i32"] + fn cttz_u32x2(x: u32x2, is_zero_undef: bool) -> u32x2; + #[link_name = "llvm.cttz.v4i32"] + fn cttz_u32x4(x: u32x4, is_zero_undef: bool) -> u32x4; + #[link_name = "llvm.cttz.v8i32"] + fn cttz_u32x8(x: u32x8, is_zero_undef: bool) -> u32x8; + #[link_name = "llvm.cttz.v16i32"] + fn cttz_u32x16(x: u32x16, is_zero_undef: bool) -> u32x16; + + #[link_name = "llvm.cttz.v2i64"] + fn cttz_u64x2(x: u64x2, is_zero_undef: bool) -> u64x2; + #[link_name = "llvm.cttz.v4i64"] + fn cttz_u64x4(x: u64x4, is_zero_undef: bool) -> u64x4; + #[link_name = "llvm.cttz.v8i64"] + fn cttz_u64x8(x: u64x8, is_zero_undef: bool) -> u64x8; + + #[link_name = "llvm.cttz.v1i128"] + fn cttz_u128x1(x: u128x1, is_zero_undef: bool) -> u128x1; + #[link_name = "llvm.cttz.v2i128"] + fn cttz_u128x2(x: u128x2, is_zero_undef: bool) -> u128x2; + #[link_name = "llvm.cttz.v4i128"] + fn cttz_u128x4(x: u128x4, is_zero_undef: bool) -> u128x4; + + #[link_name = "llvm.ctpop.v2i8"] + fn ctpop_u8x2(x: u8x2) -> u8x2; + #[link_name = "llvm.ctpop.v4i8"] + fn ctpop_u8x4(x: u8x4) -> u8x4; + #[link_name = "llvm.ctpop.v8i8"] + fn ctpop_u8x8(x: u8x8) -> u8x8; + #[link_name = "llvm.ctpop.v16i8"] + fn ctpop_u8x16(x: u8x16) -> u8x16; + #[link_name = "llvm.ctpop.v32i8"] + fn ctpop_u8x32(x: u8x32) -> u8x32; + #[link_name = "llvm.ctpop.v64i8"] + fn ctpop_u8x64(x: u8x64) -> u8x64; + + #[link_name = "llvm.ctpop.v2i16"] + fn ctpop_u16x2(x: u16x2) -> u16x2; + #[link_name = "llvm.ctpop.v4i16"] + fn ctpop_u16x4(x: u16x4) -> u16x4; + #[link_name = "llvm.ctpop.v8i16"] + fn ctpop_u16x8(x: u16x8) -> u16x8; + #[link_name = "llvm.ctpop.v16i16"] + fn ctpop_u16x16(x: u16x16) -> u16x16; + #[link_name = "llvm.ctpop.v32i16"] + fn ctpop_u16x32(x: u16x32) -> u16x32; + + #[link_name = "llvm.ctpop.v2i32"] + fn ctpop_u32x2(x: u32x2) -> u32x2; + #[link_name = "llvm.ctpop.v4i32"] + fn ctpop_u32x4(x: u32x4) -> u32x4; + #[link_name = "llvm.ctpop.v8i32"] + fn ctpop_u32x8(x: u32x8) -> u32x8; + #[link_name = "llvm.ctpop.v16i32"] + fn ctpop_u32x16(x: u32x16) -> u32x16; + + #[link_name = "llvm.ctpop.v2i64"] + fn ctpop_u64x2(x: u64x2) -> u64x2; + #[link_name = "llvm.ctpop.v4i64"] + fn ctpop_u64x4(x: u64x4) -> u64x4; + #[link_name = "llvm.ctpop.v8i64"] + fn ctpop_u64x8(x: u64x8) -> u64x8; + + #[link_name = "llvm.ctpop.v1i128"] + fn ctpop_u128x1(x: u128x1) -> u128x1; + #[link_name = "llvm.ctpop.v2i128"] + fn ctpop_u128x2(x: u128x2) -> u128x2; + #[link_name = "llvm.ctpop.v4i128"] + fn ctpop_u128x4(x: u128x4) -> u128x4; +} + +pub(crate) trait BitManip { + fn ctpop(self) -> Self; + fn ctlz(self) -> Self; + fn cttz(self) -> Self; +} + +macro_rules! impl_bit_manip { + (inner: $ty:ident, $scalar:ty, $uty:ident, + $ctpop:ident, $ctlz:ident, $cttz:ident) => { + // FIXME: several LLVM intrinsics break on s390x https://github.com/rust-lang-nursery/packed_simd/issues/192 + #[cfg(target_arch = "s390x")] + impl_bit_manip! { scalar: $ty, $scalar } + #[cfg(not(target_arch = "s390x"))] + impl BitManip for $ty { + #[inline] + fn ctpop(self) -> Self { + let y: $uty = self.cast(); + unsafe { $ctpop(y).cast() } + } + + #[inline] + fn ctlz(self) -> Self { + let y: $uty = self.cast(); + // the ctxx intrinsics need compile-time constant + // `is_zero_undef` + unsafe { $ctlz(y, false).cast() } + } + + #[inline] + fn cttz(self) -> Self { + let y: $uty = self.cast(); + unsafe { $cttz(y, false).cast() } + } + } + }; + (sized_inner: $ty:ident, $scalar:ty, $uty:ident) => { + #[cfg(target_arch = "s390x")] + impl_bit_manip! { scalar: $ty, $scalar } + #[cfg(not(target_arch = "s390x"))] + impl BitManip for $ty { + #[inline] + fn ctpop(self) -> Self { + let y: $uty = self.cast(); + $uty::ctpop(y).cast() + } + + #[inline] + fn ctlz(self) -> Self { + let y: $uty = self.cast(); + $uty::ctlz(y).cast() + } + + #[inline] + fn cttz(self) -> Self { + let y: $uty = self.cast(); + $uty::cttz(y).cast() + } + } + }; + (scalar: $ty:ident, $scalar:ty) => { + impl BitManip for $ty { + #[inline] + fn ctpop(self) -> Self { + let mut ones = self; + for i in 0..Self::lanes() { + ones = ones.replace(i, self.extract(i).count_ones() as $scalar); + } + ones + } + + #[inline] + fn ctlz(self) -> Self { + let mut lz = self; + for i in 0..Self::lanes() { + lz = lz.replace(i, self.extract(i).leading_zeros() as $scalar); + } + lz + } + + #[inline] + fn cttz(self) -> Self { + let mut tz = self; + for i in 0..Self::lanes() { + tz = tz.replace(i, self.extract(i).trailing_zeros() as $scalar); + } + tz + } + } + }; + ($uty:ident, $uscalar:ty, $ity:ident, $iscalar:ty, + $ctpop:ident, $ctlz:ident, $cttz:ident) => { + impl_bit_manip! { inner: $uty, $uscalar, $uty, $ctpop, $ctlz, $cttz } + impl_bit_manip! { inner: $ity, $iscalar, $uty, $ctpop, $ctlz, $cttz } + }; + (sized: $usize:ident, $uscalar:ty, $isize:ident, + $iscalar:ty, $ty:ident) => { + impl_bit_manip! { sized_inner: $usize, $uscalar, $ty } + impl_bit_manip! { sized_inner: $isize, $iscalar, $ty } + }; +} + +impl_bit_manip! { u8x2 , u8, i8x2, i8, ctpop_u8x2, ctlz_u8x2, cttz_u8x2 } +impl_bit_manip! { u8x4 , u8, i8x4, i8, ctpop_u8x4, ctlz_u8x4, cttz_u8x4 } +#[cfg(not(target_arch = "aarch64"))] // see below +impl_bit_manip! { u8x8 , u8, i8x8, i8, ctpop_u8x8, ctlz_u8x8, cttz_u8x8 } +impl_bit_manip! { u8x16 , u8, i8x16, i8, ctpop_u8x16, ctlz_u8x16, cttz_u8x16 } +impl_bit_manip! { u8x32 , u8, i8x32, i8, ctpop_u8x32, ctlz_u8x32, cttz_u8x32 } +impl_bit_manip! { u8x64 , u8, i8x64, i8, ctpop_u8x64, ctlz_u8x64, cttz_u8x64 } +impl_bit_manip! { u16x2 , u16, i16x2, i16, ctpop_u16x2, ctlz_u16x2, cttz_u16x2 } +impl_bit_manip! { u16x4 , u16, i16x4, i16, ctpop_u16x4, ctlz_u16x4, cttz_u16x4 } +impl_bit_manip! { u16x8 , u16, i16x8, i16, ctpop_u16x8, ctlz_u16x8, cttz_u16x8 } +impl_bit_manip! { u16x16 , u16, i16x16, i16, ctpop_u16x16, ctlz_u16x16, cttz_u16x16 } +impl_bit_manip! { u16x32 , u16, i16x32, i16, ctpop_u16x32, ctlz_u16x32, cttz_u16x32 } +impl_bit_manip! { u32x2 , u32, i32x2, i32, ctpop_u32x2, ctlz_u32x2, cttz_u32x2 } +impl_bit_manip! { u32x4 , u32, i32x4, i32, ctpop_u32x4, ctlz_u32x4, cttz_u32x4 } +impl_bit_manip! { u32x8 , u32, i32x8, i32, ctpop_u32x8, ctlz_u32x8, cttz_u32x8 } +impl_bit_manip! { u32x16 , u32, i32x16, i32, ctpop_u32x16, ctlz_u32x16, cttz_u32x16 } +impl_bit_manip! { u64x2 , u64, i64x2, i64, ctpop_u64x2, ctlz_u64x2, cttz_u64x2 } +impl_bit_manip! { u64x4 , u64, i64x4, i64, ctpop_u64x4, ctlz_u64x4, cttz_u64x4 } +impl_bit_manip! { u64x8 , u64, i64x8, i64, ctpop_u64x8, ctlz_u64x8, cttz_u64x8 } +impl_bit_manip! { u128x1 , u128, i128x1, i128, ctpop_u128x1, ctlz_u128x1, cttz_u128x1 } +impl_bit_manip! { u128x2 , u128, i128x2, i128, ctpop_u128x2, ctlz_u128x2, cttz_u128x2 } +impl_bit_manip! { u128x4 , u128, i128x4, i128, ctpop_u128x4, ctlz_u128x4, cttz_u128x4 } + +#[cfg(target_arch = "aarch64")] +impl BitManip for u8x8 { + #[inline] + fn ctpop(self) -> Self { + let y: u8x8 = self.cast(); + unsafe { ctpop_u8x8(y).cast() } + } + + #[inline] + fn ctlz(self) -> Self { + let y: u8x8 = self.cast(); + unsafe { ctlz_u8x8(y, false).cast() } + } + + #[inline] + fn cttz(self) -> Self { + // FIXME: LLVM cttz.v8i8 broken on aarch64 https://github.com/rust-lang-nursery/packed_simd/issues/191 + // OPTIMIZE: adapt the algorithm used for v8i16/etc to Rust's aarch64 + // intrinsics + let mut tz = self; + for i in 0..Self::lanes() { + tz = tz.replace(i, self.extract(i).trailing_zeros() as u8); + } + tz + } +} +#[cfg(target_arch = "aarch64")] +impl BitManip for i8x8 { + #[inline] + fn ctpop(self) -> Self { + let y: u8x8 = self.cast(); + unsafe { ctpop_u8x8(y).cast() } + } + + #[inline] + fn ctlz(self) -> Self { + let y: u8x8 = self.cast(); + unsafe { ctlz_u8x8(y, false).cast() } + } + + #[inline] + fn cttz(self) -> Self { + // FIXME: LLVM cttz.v8i8 broken on aarch64 https://github.com/rust-lang-nursery/packed_simd/issues/191 + // OPTIMIZE: adapt the algorithm used for v8i16/etc to Rust's aarch64 + // intrinsics + let mut tz = self; + for i in 0..Self::lanes() { + tz = tz.replace(i, self.extract(i).trailing_zeros() as i8); + } + tz + } +} + +cfg_if! { + if #[cfg(target_pointer_width = "8")] { + impl_bit_manip! { sized: usizex2, usize, isizex2, isize, u8x2 } + impl_bit_manip! { sized: usizex4, usize, isizex4, isize, u8x4 } + impl_bit_manip! { sized: usizex8, usize, isizex8, isize, u8x8 } + } else if #[cfg(target_pointer_width = "16")] { + impl_bit_manip! { sized: usizex2, usize, isizex2, isize, u16x2 } + impl_bit_manip! { sized: usizex4, usize, isizex4, isize, u16x4 } + impl_bit_manip! { sized: usizex8, usize, isizex8, isize, u16x8 } + } else if #[cfg(target_pointer_width = "32")] { + impl_bit_manip! { sized: usizex2, usize, isizex2, isize, u32x2 } + impl_bit_manip! { sized: usizex4, usize, isizex4, isize, u32x4 } + impl_bit_manip! { sized: usizex8, usize, isizex8, isize, u32x8 } + } else if #[cfg(target_pointer_width = "64")] { + impl_bit_manip! { sized: usizex2, usize, isizex2, isize, u64x2 } + impl_bit_manip! { sized: usizex4, usize, isizex4, isize, u64x4 } + impl_bit_manip! { sized: usizex8, usize, isizex8, isize, u64x8 } + } else { + compile_error!("unsupported target_pointer_width"); + } +} diff --git a/third_party/rust/packed_simd/src/codegen/llvm.rs b/third_party/rust/packed_simd/src/codegen/llvm.rs new file mode 100644 index 000000000000..bb482fac6625 --- /dev/null +++ b/third_party/rust/packed_simd/src/codegen/llvm.rs @@ -0,0 +1,122 @@ +//! LLVM's platform intrinsics +#![allow(dead_code)] + +use crate::sealed::Shuffle; +#[allow(unused_imports)] // FIXME: spurious warning? +use crate::sealed::Simd; + +extern "platform-intrinsic" { + fn simd_shuffle(x: T, y: T, idx: I) -> U; +} + +#[allow(clippy::missing_safety_doc)] +#[inline] +pub unsafe fn __shuffle_vector2(x: T, y: T) -> U +where + T: Simd, + ::Element: Shuffle<[u32; 2], Output = U>, +{ + simd_shuffle(x, y, IDX) +} + +#[allow(clippy::missing_safety_doc)] +#[inline] +pub unsafe fn __shuffle_vector4(x: T, y: T) -> U +where + T: Simd, + ::Element: Shuffle<[u32; 4], Output = U>, +{ + simd_shuffle(x, y, IDX) +} + +#[allow(clippy::missing_safety_doc)] +#[inline] +pub unsafe fn __shuffle_vector8(x: T, y: T) -> U +where + T: Simd, + ::Element: Shuffle<[u32; 8], Output = U>, +{ + simd_shuffle(x, y, IDX) +} + +#[allow(clippy::missing_safety_doc)] +#[inline] +pub unsafe fn __shuffle_vector16(x: T, y: T) -> U +where + T: Simd, + ::Element: Shuffle<[u32; 16], Output = U>, +{ + simd_shuffle(x, y, IDX) +} + +#[allow(clippy::missing_safety_doc)] +#[inline] +pub unsafe fn __shuffle_vector32(x: T, y: T) -> U +where + T: Simd, + ::Element: Shuffle<[u32; 32], Output = U>, +{ + simd_shuffle(x, y, IDX) +} + +#[allow(clippy::missing_safety_doc)] +#[inline] +pub unsafe fn __shuffle_vector64(x: T, y: T) -> U +where + T: Simd, + ::Element: Shuffle<[u32; 64], Output = U>, +{ + simd_shuffle(x, y, IDX) +} + +extern "platform-intrinsic" { + pub(crate) fn simd_eq(x: T, y: T) -> U; + pub(crate) fn simd_ne(x: T, y: T) -> U; + pub(crate) fn simd_lt(x: T, y: T) -> U; + pub(crate) fn simd_le(x: T, y: T) -> U; + pub(crate) fn simd_gt(x: T, y: T) -> U; + pub(crate) fn simd_ge(x: T, y: T) -> U; + + pub(crate) fn simd_insert(x: T, idx: u32, val: U) -> T; + pub(crate) fn simd_extract(x: T, idx: u32) -> U; + + pub(crate) fn simd_cast(x: T) -> U; + + pub(crate) fn simd_add(x: T, y: T) -> T; + pub(crate) fn simd_sub(x: T, y: T) -> T; + pub(crate) fn simd_mul(x: T, y: T) -> T; + pub(crate) fn simd_div(x: T, y: T) -> T; + pub(crate) fn simd_rem(x: T, y: T) -> T; + pub(crate) fn simd_shl(x: T, y: T) -> T; + pub(crate) fn simd_shr(x: T, y: T) -> T; + pub(crate) fn simd_and(x: T, y: T) -> T; + pub(crate) fn simd_or(x: T, y: T) -> T; + pub(crate) fn simd_xor(x: T, y: T) -> T; + + pub(crate) fn simd_reduce_add_unordered(x: T) -> U; + pub(crate) fn simd_reduce_mul_unordered(x: T) -> U; + pub(crate) fn simd_reduce_add_ordered(x: T, acc: U) -> U; + pub(crate) fn simd_reduce_mul_ordered(x: T, acc: U) -> U; + pub(crate) fn simd_reduce_min(x: T) -> U; + pub(crate) fn simd_reduce_max(x: T) -> U; + pub(crate) fn simd_reduce_min_nanless(x: T) -> U; + pub(crate) fn simd_reduce_max_nanless(x: T) -> U; + pub(crate) fn simd_reduce_and(x: T) -> U; + pub(crate) fn simd_reduce_or(x: T) -> U; + pub(crate) fn simd_reduce_xor(x: T) -> U; + pub(crate) fn simd_reduce_all(x: T) -> bool; + pub(crate) fn simd_reduce_any(x: T) -> bool; + + pub(crate) fn simd_select(m: M, a: T, b: T) -> T; + + pub(crate) fn simd_fmin(a: T, b: T) -> T; + pub(crate) fn simd_fmax(a: T, b: T) -> T; + + pub(crate) fn simd_fsqrt(a: T) -> T; + pub(crate) fn simd_fma(a: T, b: T, c: T) -> T; + + pub(crate) fn simd_gather(value: T, pointers: P, mask: M) -> T; + pub(crate) fn simd_scatter(value: T, pointers: P, mask: M); + + pub(crate) fn simd_bitmask(value: T) -> U; +} diff --git a/third_party/rust/packed_simd/src/codegen/math.rs b/third_party/rust/packed_simd/src/codegen/math.rs new file mode 100644 index 000000000000..9a0ea7a4e2d2 --- /dev/null +++ b/third_party/rust/packed_simd/src/codegen/math.rs @@ -0,0 +1,3 @@ +//! Vertical math operations + +pub(crate) mod float; diff --git a/third_party/rust/packed_simd/src/codegen/math/float.rs b/third_party/rust/packed_simd/src/codegen/math/float.rs new file mode 100644 index 000000000000..10d21831f157 --- /dev/null +++ b/third_party/rust/packed_simd/src/codegen/math/float.rs @@ -0,0 +1,19 @@ +//! Vertical floating-point math operations. +#![allow(clippy::useless_transmute)] + +#[macro_use] +pub(crate) mod macros; +pub(crate) mod abs; +pub(crate) mod cos; +pub(crate) mod cos_pi; +pub(crate) mod exp; +pub(crate) mod ln; +pub(crate) mod mul_add; +pub(crate) mod mul_adde; +pub(crate) mod powf; +pub(crate) mod sin; +pub(crate) mod sin_cos_pi; +pub(crate) mod sin_pi; +pub(crate) mod sqrt; +pub(crate) mod sqrte; +pub(crate) mod tanh; diff --git a/third_party/rust/packed_simd/src/codegen/math/float/abs.rs b/third_party/rust/packed_simd/src/codegen/math/float/abs.rs new file mode 100644 index 000000000000..34aacc25be75 --- /dev/null +++ b/third_party/rust/packed_simd/src/codegen/math/float/abs.rs @@ -0,0 +1,103 @@ +//! Vertical floating-point `fabs` +#![allow(unused)] + +// FIXME 64-bit 1 elem vectors fabs + +use crate::*; + +pub(crate) trait Abs { + fn abs(self) -> Self; +} + +#[allow(improper_ctypes)] +extern "C" { + #[link_name = "llvm.fabs.v2f32"] + fn fabs_v2f32(x: f32x2) -> f32x2; + #[link_name = "llvm.fabs.v4f32"] + fn fabs_v4f32(x: f32x4) -> f32x4; + #[link_name = "llvm.fabs.v8f32"] + fn fabs_v8f32(x: f32x8) -> f32x8; + #[link_name = "llvm.fabs.v16f32"] + fn fabs_v16f32(x: f32x16) -> f32x16; + /* FIXME 64-bit fabsgle elem vectors + #[link_name = "llvm.fabs.v1f64"] + fn fabs_v1f64(x: f64x1) -> f64x1; + */ + #[link_name = "llvm.fabs.v2f64"] + fn fabs_v2f64(x: f64x2) -> f64x2; + #[link_name = "llvm.fabs.v4f64"] + fn fabs_v4f64(x: f64x4) -> f64x4; + #[link_name = "llvm.fabs.v8f64"] + fn fabs_v8f64(x: f64x8) -> f64x8; + + #[link_name = "llvm.fabs.f32"] + fn fabs_f32(x: f32) -> f32; + #[link_name = "llvm.fabs.f64"] + fn fabs_f64(x: f64) -> f64; +} + +gen_unary_impl_table!(Abs, abs); + +cfg_if! { + if #[cfg(target_arch = "s390x")] { + // FIXME: https://github.com/rust-lang-nursery/packed_simd/issues/14 + impl_unary!(f32x2[f32; 2]: fabs_f32); + impl_unary!(f32x4[f32; 4]: fabs_f32); + impl_unary!(f32x8[f32; 8]: fabs_f32); + impl_unary!(f32x16[f32; 16]: fabs_f32); + + impl_unary!(f64x2[f64; 2]: fabs_f64); + impl_unary!(f64x4[f64; 4]: fabs_f64); + impl_unary!(f64x8[f64; 8]: fabs_f64); + } else if #[cfg(all(target_arch = "x86_64", feature = "sleef-sys"))] { + use sleef_sys::*; + cfg_if! { + if #[cfg(target_feature = "avx2")] { + impl_unary!(f32x2[t => f32x4]: Sleef_fabsf4_avx2128); + impl_unary!(f32x16[h => f32x8]: Sleef_fabsf8_avx2); + impl_unary!(f64x8[h => f64x4]: Sleef_fabsd4_avx2); + + impl_unary!(f32x4: Sleef_fabsf4_avx2128); + impl_unary!(f32x8: Sleef_fabsf8_avx2); + impl_unary!(f64x2: Sleef_fabsd2_avx2128); + impl_unary!(f64x4: Sleef_fabsd4_avx2); + } else if #[cfg(target_feature = "avx")] { + impl_unary!(f32x2[t => f32x4]: Sleef_fabsf4_sse4); + impl_unary!(f32x16[h => f32x8]: Sleef_fabsf8_avx); + impl_unary!(f64x8[h => f64x4]: Sleef_fabsd4_avx); + + impl_unary!(f32x4: Sleef_fabsf4_sse4); + impl_unary!(f32x8: Sleef_fabsf8_avx); + impl_unary!(f64x2: Sleef_fabsd2_sse4); + impl_unary!(f64x4: Sleef_fabsd4_avx); + } else if #[cfg(target_feature = "sse4.2")] { + impl_unary!(f32x2[t => f32x4]: Sleef_fabsf4_sse4); + impl_unary!(f32x16[q => f32x4]: Sleef_fabsf4_sse4); + impl_unary!(f64x8[q => f64x2]: Sleef_fabsd2_sse4); + + impl_unary!(f32x4: Sleef_fabsf4_sse4); + impl_unary!(f32x8[h => f32x4]: Sleef_fabsf4_sse4); + impl_unary!(f64x2: Sleef_fabsd2_sse4); + impl_unary!(f64x4[h => f64x2]: Sleef_fabsd2_sse4); + } else { + impl_unary!(f32x2[f32; 2]: fabs_f32); + impl_unary!(f32x16: fabs_v16f32); + impl_unary!(f64x8: fabs_v8f64); + + impl_unary!(f32x4: fabs_v4f32); + impl_unary!(f32x8: fabs_v8f32); + impl_unary!(f64x2: fabs_v2f64); + impl_unary!(f64x4: fabs_v4f64); + } + } + } else { + impl_unary!(f32x2[f32; 2]: fabs_f32); + impl_unary!(f32x4: fabs_v4f32); + impl_unary!(f32x8: fabs_v8f32); + impl_unary!(f32x16: fabs_v16f32); + + impl_unary!(f64x2: fabs_v2f64); + impl_unary!(f64x4: fabs_v4f64); + impl_unary!(f64x8: fabs_v8f64); + } +} diff --git a/third_party/rust/packed_simd/src/codegen/math/float/cos.rs b/third_party/rust/packed_simd/src/codegen/math/float/cos.rs new file mode 100644 index 000000000000..dec390cb74d4 --- /dev/null +++ b/third_party/rust/packed_simd/src/codegen/math/float/cos.rs @@ -0,0 +1,103 @@ +//! Vertical floating-point `cos` +#![allow(unused)] + +// FIXME 64-bit 1 elem vector cos + +use crate::*; + +pub(crate) trait Cos { + fn cos(self) -> Self; +} + +#[allow(improper_ctypes)] +extern "C" { + #[link_name = "llvm.cos.v2f32"] + fn cos_v2f32(x: f32x2) -> f32x2; + #[link_name = "llvm.cos.v4f32"] + fn cos_v4f32(x: f32x4) -> f32x4; + #[link_name = "llvm.cos.v8f32"] + fn cos_v8f32(x: f32x8) -> f32x8; + #[link_name = "llvm.cos.v16f32"] + fn cos_v16f32(x: f32x16) -> f32x16; + /* FIXME 64-bit cosgle elem vectors + #[link_name = "llvm.cos.v1f64"] + fn cos_v1f64(x: f64x1) -> f64x1; + */ + #[link_name = "llvm.cos.v2f64"] + fn cos_v2f64(x: f64x2) -> f64x2; + #[link_name = "llvm.cos.v4f64"] + fn cos_v4f64(x: f64x4) -> f64x4; + #[link_name = "llvm.cos.v8f64"] + fn cos_v8f64(x: f64x8) -> f64x8; + + #[link_name = "llvm.cos.f32"] + fn cos_f32(x: f32) -> f32; + #[link_name = "llvm.cos.f64"] + fn cos_f64(x: f64) -> f64; +} + +gen_unary_impl_table!(Cos, cos); + +cfg_if! { + if #[cfg(target_arch = "s390x")] { + // FIXME: https://github.com/rust-lang-nursery/packed_simd/issues/14 + impl_unary!(f32x2[f32; 2]: cos_f32); + impl_unary!(f32x4[f32; 4]: cos_f32); + impl_unary!(f32x8[f32; 8]: cos_f32); + impl_unary!(f32x16[f32; 16]: cos_f32); + + impl_unary!(f64x2[f64; 2]: cos_f64); + impl_unary!(f64x4[f64; 4]: cos_f64); + impl_unary!(f64x8[f64; 8]: cos_f64); + } else if #[cfg(all(target_arch = "x86_64", feature = "sleef-sys"))] { + use sleef_sys::*; + cfg_if! { + if #[cfg(target_feature = "avx2")] { + impl_unary!(f32x2[t => f32x4]: Sleef_cosf4_u10avx2128); + impl_unary!(f32x16[h => f32x8]: Sleef_cosf8_u10avx2); + impl_unary!(f64x8[h => f64x4]: Sleef_cosd4_u10avx2); + + impl_unary!(f32x4: Sleef_cosf4_u10avx2128); + impl_unary!(f32x8: Sleef_cosf8_u10avx2); + impl_unary!(f64x2: Sleef_cosd2_u10avx2128); + impl_unary!(f64x4: Sleef_cosd4_u10avx2); + } else if #[cfg(target_feature = "avx")] { + impl_unary!(f32x2[t => f32x4]: Sleef_cosf4_u10sse4); + impl_unary!(f32x16[h => f32x8]: Sleef_cosf8_u10avx); + impl_unary!(f64x8[h => f64x4]: Sleef_cosd4_u10avx); + + impl_unary!(f32x4: Sleef_cosf4_u10sse4); + impl_unary!(f32x8: Sleef_cosf8_u10avx); + impl_unary!(f64x2: Sleef_cosd2_u10sse4); + impl_unary!(f64x4: Sleef_cosd4_u10avx); + } else if #[cfg(target_feature = "sse4.2")] { + impl_unary!(f32x2[t => f32x4]: Sleef_cosf4_u10sse4); + impl_unary!(f32x16[q => f32x4]: Sleef_cosf4_u10sse4); + impl_unary!(f64x8[q => f64x2]: Sleef_cosd2_u10sse4); + + impl_unary!(f32x4: Sleef_cosf4_u10sse4); + impl_unary!(f32x8[h => f32x4]: Sleef_cosf4_u10sse4); + impl_unary!(f64x2: Sleef_cosd2_u10sse4); + impl_unary!(f64x4[h => f64x2]: Sleef_cosd2_u10sse4); + } else { + impl_unary!(f32x2[f32; 2]: cos_f32); + impl_unary!(f32x16: cos_v16f32); + impl_unary!(f64x8: cos_v8f64); + + impl_unary!(f32x4: cos_v4f32); + impl_unary!(f32x8: cos_v8f32); + impl_unary!(f64x2: cos_v2f64); + impl_unary!(f64x4: cos_v4f64); + } + } + } else { + impl_unary!(f32x2[f32; 2]: cos_f32); + impl_unary!(f32x4: cos_v4f32); + impl_unary!(f32x8: cos_v8f32); + impl_unary!(f32x16: cos_v16f32); + + impl_unary!(f64x2: cos_v2f64); + impl_unary!(f64x4: cos_v4f64); + impl_unary!(f64x8: cos_v8f64); + } +} diff --git a/third_party/rust/packed_simd/src/codegen/math/float/cos_pi.rs b/third_party/rust/packed_simd/src/codegen/math/float/cos_pi.rs new file mode 100644 index 000000000000..e283280ee44b --- /dev/null +++ b/third_party/rust/packed_simd/src/codegen/math/float/cos_pi.rs @@ -0,0 +1,87 @@ +//! Vertical floating-point `cos` +#![allow(unused)] + +// FIXME 64-bit 1 elem vectors cos_pi + +use crate::*; + +pub(crate) trait CosPi { + fn cos_pi(self) -> Self; +} + +gen_unary_impl_table!(CosPi, cos_pi); + +macro_rules! impl_def { + ($vid:ident, $PI:path) => { + impl CosPi for $vid { + #[inline] + fn cos_pi(self) -> Self { + (self * Self::splat($PI)).cos() + } + } + }; +} +macro_rules! impl_def32 { + ($vid:ident) => { + impl_def!($vid, crate::f32::consts::PI); + }; +} +macro_rules! impl_def64 { + ($vid:ident) => { + impl_def!($vid, crate::f64::consts::PI); + }; +} + +cfg_if! { + if #[cfg(all(target_arch = "x86_64", feature = "sleef-sys"))] { + use sleef_sys::*; + cfg_if! { + if #[cfg(target_feature = "avx2")] { + impl_unary!(f32x2[t => f32x4]: Sleef_cospif4_u05avx2128); + impl_unary!(f32x16[h => f32x8]: Sleef_cospif8_u05avx2); + impl_unary!(f64x8[h => f64x4]: Sleef_cospid4_u05avx2); + + impl_unary!(f32x4: Sleef_cospif4_u05avx2128); + impl_unary!(f32x8: Sleef_cospif8_u05avx2); + impl_unary!(f64x2: Sleef_cospid2_u05avx2128); + impl_unary!(f64x4: Sleef_cospid4_u05avx2); + } else if #[cfg(target_feature = "avx")] { + impl_unary!(f32x2[t => f32x4]: Sleef_cospif4_u05sse4); + impl_unary!(f32x16[h => f32x8]: Sleef_cospif8_u05avx); + impl_unary!(f64x8[h => f64x4]: Sleef_cospid4_u05avx); + + impl_unary!(f32x4: Sleef_cospif4_u05sse4); + impl_unary!(f32x8: Sleef_cospif8_u05avx); + impl_unary!(f64x2: Sleef_cospid2_u05sse4); + impl_unary!(f64x4: Sleef_cospid4_u05avx); + } else if #[cfg(target_feature = "sse4.2")] { + impl_unary!(f32x2[t => f32x4]: Sleef_cospif4_u05sse4); + impl_unary!(f32x16[q => f32x4]: Sleef_cospif4_u05sse4); + impl_unary!(f64x8[q => f64x2]: Sleef_cospid2_u05sse4); + + impl_unary!(f32x4: Sleef_cospif4_u05sse4); + impl_unary!(f32x8[h => f32x4]: Sleef_cospif4_u05sse4); + impl_unary!(f64x2: Sleef_cospid2_u05sse4); + impl_unary!(f64x4[h => f64x2]: Sleef_cospid2_u05sse4); + } else { + impl_def32!(f32x2); + impl_def32!(f32x4); + impl_def32!(f32x8); + impl_def32!(f32x16); + + impl_def64!(f64x2); + impl_def64!(f64x4); + impl_def64!(f64x8); + } + } + } else { + impl_def32!(f32x2); + impl_def32!(f32x4); + impl_def32!(f32x8); + impl_def32!(f32x16); + + impl_def64!(f64x2); + impl_def64!(f64x4); + impl_def64!(f64x8); + } +} diff --git a/third_party/rust/packed_simd/src/codegen/math/float/exp.rs b/third_party/rust/packed_simd/src/codegen/math/float/exp.rs new file mode 100644 index 000000000000..a7b20580e3f1 --- /dev/null +++ b/third_party/rust/packed_simd/src/codegen/math/float/exp.rs @@ -0,0 +1,112 @@ +//! Vertical floating-point `exp` +#![allow(unused)] + +// FIXME 64-bit expgle elem vectors misexpg + +use crate::*; + +pub(crate) trait Exp { + fn exp(self) -> Self; +} + +#[allow(improper_ctypes)] +extern "C" { + #[link_name = "llvm.exp.v2f32"] + fn exp_v2f32(x: f32x2) -> f32x2; + #[link_name = "llvm.exp.v4f32"] + fn exp_v4f32(x: f32x4) -> f32x4; + #[link_name = "llvm.exp.v8f32"] + fn exp_v8f32(x: f32x8) -> f32x8; + #[link_name = "llvm.exp.v16f32"] + fn exp_v16f32(x: f32x16) -> f32x16; + /* FIXME 64-bit expgle elem vectors + #[link_name = "llvm.exp.v1f64"] + fn exp_v1f64(x: f64x1) -> f64x1; + */ + #[link_name = "llvm.exp.v2f64"] + fn exp_v2f64(x: f64x2) -> f64x2; + #[link_name = "llvm.exp.v4f64"] + fn exp_v4f64(x: f64x4) -> f64x4; + #[link_name = "llvm.exp.v8f64"] + fn exp_v8f64(x: f64x8) -> f64x8; + + #[link_name = "llvm.exp.f32"] + fn exp_f32(x: f32) -> f32; + #[link_name = "llvm.exp.f64"] + fn exp_f64(x: f64) -> f64; +} + +gen_unary_impl_table!(Exp, exp); + +cfg_if! { + if #[cfg(target_arch = "s390x")] { + // FIXME: https://github.com/rust-lang-nursery/packed_simd/issues/14 + impl_unary!(f32x2[f32; 2]: exp_f32); + impl_unary!(f32x4[f32; 4]: exp_f32); + impl_unary!(f32x8[f32; 8]: exp_f32); + impl_unary!(f32x16[f32; 16]: exp_f32); + + impl_unary!(f64x2[f64; 2]: exp_f64); + impl_unary!(f64x4[f64; 4]: exp_f64); + impl_unary!(f64x8[f64; 8]: exp_f64); + } else if #[cfg(all(target_arch = "x86_64", feature = "sleef-sys"))] { + use sleef_sys::*; + cfg_if! { + if #[cfg(target_feature = "avx2")] { + impl_unary!(f32x2[t => f32x4]: Sleef_expf4_u10avx2128); + impl_unary!(f32x16[h => f32x8]: Sleef_expf8_u10avx2); + impl_unary!(f64x8[h => f64x4]: Sleef_expd4_u10avx2); + + impl_unary!(f32x4: Sleef_expf4_u10avx2128); + impl_unary!(f32x8: Sleef_expf8_u10avx2); + impl_unary!(f64x2: Sleef_expd2_u10avx2128); + impl_unary!(f64x4: Sleef_expd4_u10avx2); + } else if #[cfg(target_feature = "avx")] { + impl_unary!(f32x2[t => f32x4]: Sleef_expf4_u10sse4); + impl_unary!(f32x16[h => f32x8]: Sleef_expf8_u10avx); + impl_unary!(f64x8[h => f64x4]: Sleef_expd4_u10avx); + + impl_unary!(f32x4: Sleef_expf4_u10sse4); + impl_unary!(f32x8: Sleef_expf8_u10avx); + impl_unary!(f64x2: Sleef_expd2_u10sse4); + impl_unary!(f64x4: Sleef_expd4_u10avx); + } else if #[cfg(target_feature = "sse4.2")] { + impl_unary!(f32x2[t => f32x4]: Sleef_expf4_u10sse4); + impl_unary!(f32x16[q => f32x4]: Sleef_expf4_u10sse4); + impl_unary!(f64x8[q => f64x2]: Sleef_expd2_u10sse4); + + impl_unary!(f32x4: Sleef_expf4_u10sse4); + impl_unary!(f32x8[h => f32x4]: Sleef_expf4_u10sse4); + impl_unary!(f64x2: Sleef_expd2_u10sse4); + impl_unary!(f64x4[h => f64x2]: Sleef_expd2_u10sse4); + } else if #[cfg(target_feature = "sse2")] { + impl_unary!(f32x2[t => f32x4]: Sleef_expf4_u10sse2); + impl_unary!(f32x16[q => f32x4]: Sleef_expf4_u10sse2); + impl_unary!(f64x8[q => f64x2]: Sleef_expd2_u10sse2); + + impl_unary!(f32x4: Sleef_expf4_u10sse2); + impl_unary!(f32x8[h => f32x4]: Sleef_expf4_u10sse2); + impl_unary!(f64x2: Sleef_expd2_u10sse2); + impl_unary!(f64x4[h => f64x2]: Sleef_expd2_u10sse2); + } else { + impl_unary!(f32x2[f32; 2]: exp_f32); + impl_unary!(f32x16: exp_v16f32); + impl_unary!(f64x8: exp_v8f64); + + impl_unary!(f32x4: exp_v4f32); + impl_unary!(f32x8: exp_v8f32); + impl_unary!(f64x2: exp_v2f64); + impl_unary!(f64x4: exp_v4f64); + } + } + } else { + impl_unary!(f32x2[f32; 2]: exp_f32); + impl_unary!(f32x4: exp_v4f32); + impl_unary!(f32x8: exp_v8f32); + impl_unary!(f32x16: exp_v16f32); + + impl_unary!(f64x2: exp_v2f64); + impl_unary!(f64x4: exp_v4f64); + impl_unary!(f64x8: exp_v8f64); + } +} diff --git a/third_party/rust/packed_simd/src/codegen/math/float/ln.rs b/third_party/rust/packed_simd/src/codegen/math/float/ln.rs new file mode 100644 index 000000000000..a5e38cb40d1e --- /dev/null +++ b/third_party/rust/packed_simd/src/codegen/math/float/ln.rs @@ -0,0 +1,112 @@ +//! Vertical floating-point `ln` +#![allow(unused)] + +// FIXME 64-bit lngle elem vectors mislng + +use crate::*; + +pub(crate) trait Ln { + fn ln(self) -> Self; +} + +#[allow(improper_ctypes)] +extern "C" { + #[link_name = "llvm.log.v2f32"] + fn ln_v2f32(x: f32x2) -> f32x2; + #[link_name = "llvm.log.v4f32"] + fn ln_v4f32(x: f32x4) -> f32x4; + #[link_name = "llvm.log.v8f32"] + fn ln_v8f32(x: f32x8) -> f32x8; + #[link_name = "llvm.log.v16f32"] + fn ln_v16f32(x: f32x16) -> f32x16; + /* FIXME 64-bit lngle elem vectors + #[link_name = "llvm.log.v1f64"] + fn ln_v1f64(x: f64x1) -> f64x1; + */ + #[link_name = "llvm.log.v2f64"] + fn ln_v2f64(x: f64x2) -> f64x2; + #[link_name = "llvm.log.v4f64"] + fn ln_v4f64(x: f64x4) -> f64x4; + #[link_name = "llvm.log.v8f64"] + fn ln_v8f64(x: f64x8) -> f64x8; + + #[link_name = "llvm.log.f32"] + fn ln_f32(x: f32) -> f32; + #[link_name = "llvm.log.f64"] + fn ln_f64(x: f64) -> f64; +} + +gen_unary_impl_table!(Ln, ln); + +cfg_if! { + if #[cfg(target_arch = "s390x")] { + // FIXME: https://github.com/rust-lang-nursery/packed_simd/issues/14 + impl_unary!(f32x2[f32; 2]: ln_f32); + impl_unary!(f32x4[f32; 4]: ln_f32); + impl_unary!(f32x8[f32; 8]: ln_f32); + impl_unary!(f32x16[f32; 16]: ln_f32); + + impl_unary!(f64x2[f64; 2]: ln_f64); + impl_unary!(f64x4[f64; 4]: ln_f64); + impl_unary!(f64x8[f64; 8]: ln_f64); + } else if #[cfg(all(target_arch = "x86_64", feature = "sleef-sys"))] { + use sleef_sys::*; + cfg_if! { + if #[cfg(target_feature = "avx2")] { + impl_unary!(f32x2[t => f32x4]: Sleef_logf4_u10avx2128); + impl_unary!(f32x16[h => f32x8]: Sleef_logf8_u10avx2); + impl_unary!(f64x8[h => f64x4]: Sleef_logd4_u10avx2); + + impl_unary!(f32x4: Sleef_logf4_u10avx2128); + impl_unary!(f32x8: Sleef_logf8_u10avx2); + impl_unary!(f64x2: Sleef_logd2_u10avx2128); + impl_unary!(f64x4: Sleef_logd4_u10avx2); + } else if #[cfg(target_feature = "avx")] { + impl_unary!(f32x2[t => f32x4]: Sleef_logf4_u10sse4); + impl_unary!(f32x16[h => f32x8]: Sleef_logf8_u10avx); + impl_unary!(f64x8[h => f64x4]: Sleef_logd4_u10avx); + + impl_unary!(f32x4: Sleef_logf4_u10sse4); + impl_unary!(f32x8: Sleef_logf8_u10avx); + impl_unary!(f64x2: Sleef_logd2_u10sse4); + impl_unary!(f64x4: Sleef_logd4_u10avx); + } else if #[cfg(target_feature = "sse4.2")] { + impl_unary!(f32x2[t => f32x4]: Sleef_logf4_u10sse4); + impl_unary!(f32x16[q => f32x4]: Sleef_logf4_u10sse4); + impl_unary!(f64x8[q => f64x2]: Sleef_logd2_u10sse4); + + impl_unary!(f32x4: Sleef_logf4_u10sse4); + impl_unary!(f32x8[h => f32x4]: Sleef_logf4_u10sse4); + impl_unary!(f64x2: Sleef_logd2_u10sse4); + impl_unary!(f64x4[h => f64x2]: Sleef_logd2_u10sse4); + } else if #[cfg(target_feature = "sse2")] { + impl_unary!(f32x2[t => f32x4]: Sleef_logf4_u10sse2); + impl_unary!(f32x16[q => f32x4]: Sleef_logf4_u10sse2); + impl_unary!(f64x8[q => f64x2]: Sleef_logd2_u10sse2); + + impl_unary!(f32x4: Sleef_logf4_u10sse2); + impl_unary!(f32x8[h => f32x4]: Sleef_logf4_u10sse2); + impl_unary!(f64x2: Sleef_logd2_u10sse2); + impl_unary!(f64x4[h => f64x2]: Sleef_logd2_u10sse2); + } else { + impl_unary!(f32x2[f32; 2]: ln_f32); + impl_unary!(f32x16: ln_v16f32); + impl_unary!(f64x8: ln_v8f64); + + impl_unary!(f32x4: ln_v4f32); + impl_unary!(f32x8: ln_v8f32); + impl_unary!(f64x2: ln_v2f64); + impl_unary!(f64x4: ln_v4f64); + } + } + } else { + impl_unary!(f32x2[f32; 2]: ln_f32); + impl_unary!(f32x4: ln_v4f32); + impl_unary!(f32x8: ln_v8f32); + impl_unary!(f32x16: ln_v16f32); + + impl_unary!(f64x2: ln_v2f64); + impl_unary!(f64x4: ln_v4f64); + impl_unary!(f64x8: ln_v8f64); + } +} diff --git a/third_party/rust/packed_simd/src/codegen/math/float/macros.rs b/third_party/rust/packed_simd/src/codegen/math/float/macros.rs new file mode 100644 index 000000000000..8daee1afe205 --- /dev/null +++ b/third_party/rust/packed_simd/src/codegen/math/float/macros.rs @@ -0,0 +1,470 @@ +//! Utility macros +#![allow(unused)] + +macro_rules! impl_unary_ { + // implementation mapping 1:1 + (vec | $trait_id:ident, $trait_method:ident, $vec_id:ident, + $fun:ident) => { + impl $trait_id for $vec_id { + #[inline] + fn $trait_method(self) -> Self { + unsafe { + use crate::mem::transmute; + transmute($fun(transmute(self))) + } + } + } + }; + // implementation mapping 1:1 for when `$fun` is a generic function + // like some of the fp math rustc intrinsics (e.g. `fn fun(x: T) -> T`). + (gen | $trait_id:ident, $trait_method:ident, $vec_id:ident, + $fun:ident) => { + impl $trait_id for $vec_id { + #[inline] + fn $trait_method(self) -> Self { + unsafe { + use crate::mem::transmute; + transmute($fun(self.0)) + } + } + } + }; + (scalar | $trait_id:ident, $trait_method:ident, + $vec_id:ident, [$sid:ident; $scount:expr], $fun:ident) => { + impl $trait_id for $vec_id { + #[inline] + fn $trait_method(self) -> Self { + unsafe { + union U { + vec: $vec_id, + scalars: [$sid; $scount], + } + let mut scalars = U { vec: self }.scalars; + for i in &mut scalars { + *i = $fun(*i); + } + U { scalars }.vec + } + } + } + }; + // implementation calling fun twice on each of the vector halves: + (halves | $trait_id:ident, $trait_method:ident, $vec_id:ident, + $vech_id:ident, $fun:ident) => { + impl $trait_id for $vec_id { + #[inline] + fn $trait_method(self) -> Self { + unsafe { + use crate::mem::transmute; + union U { + vec: $vec_id, + halves: [$vech_id; 2], + } + + let mut halves = U { vec: self }.halves; + + *halves.get_unchecked_mut(0) = transmute($fun(transmute(*halves.get_unchecked(0)))); + *halves.get_unchecked_mut(1) = transmute($fun(transmute(*halves.get_unchecked(1)))); + + U { halves }.vec + } + } + } + }; + // implementation calling fun four times on each of the vector quarters: + (quarter | $trait_id:ident, $trait_method:ident, $vec_id:ident, + $vecq_id:ident, $fun:ident) => { + impl $trait_id for $vec_id { + #[inline] + fn $trait_method(self) -> Self { + unsafe { + use crate::mem::transmute; + union U { + vec: $vec_id, + quarters: [$vecq_id; 4], + } + + let mut quarters = U { vec: self }.quarters; + + *quarters.get_unchecked_mut(0) = transmute($fun(transmute(*quarters.get_unchecked(0)))); + *quarters.get_unchecked_mut(1) = transmute($fun(transmute(*quarters.get_unchecked(1)))); + *quarters.get_unchecked_mut(2) = transmute($fun(transmute(*quarters.get_unchecked(2)))); + *quarters.get_unchecked_mut(3) = transmute($fun(transmute(*quarters.get_unchecked(3)))); + + U { quarters }.vec + } + } + } + }; + // implementation calling fun once on a vector twice as large: + (twice | $trait_id:ident, $trait_method:ident, $vec_id:ident, + $vect_id:ident, $fun:ident) => { + impl $trait_id for $vec_id { + #[inline] + fn $trait_method(self) -> Self { + unsafe { + use crate::mem::{transmute, uninitialized}; + + union U { + vec: [$vec_id; 2], + twice: $vect_id, + } + + let twice = U { vec: [self, uninitialized()] }.twice; + let twice = transmute($fun(transmute(twice))); + + *(U { twice }.vec.get_unchecked(0)) + } + } + } + }; +} + +macro_rules! gen_unary_impl_table { + ($trait_id:ident, $trait_method:ident) => { + macro_rules! impl_unary { + ($vid:ident: $fun:ident) => { + impl_unary_!(vec | $trait_id, $trait_method, $vid, $fun); + }; + ($vid:ident[g]: $fun:ident) => { + impl_unary_!(gen | $trait_id, $trait_method, $vid, $fun); + }; + ($vid:ident[$sid:ident; $sc:expr]: $fun:ident) => { + impl_unary_!(scalar | $trait_id, $trait_method, $vid, [$sid; $sc], $fun); + }; + ($vid:ident[s]: $fun:ident) => { + impl_unary_!(scalar | $trait_id, $trait_method, $vid, $fun); + }; + ($vid:ident[h => $vid_h:ident]: $fun:ident) => { + impl_unary_!(halves | $trait_id, $trait_method, $vid, $vid_h, $fun); + }; + ($vid:ident[q => $vid_q:ident]: $fun:ident) => { + impl_unary_!(quarter | $trait_id, $trait_method, $vid, $vid_q, $fun); + }; + ($vid:ident[t => $vid_t:ident]: $fun:ident) => { + impl_unary_!(twice | $trait_id, $trait_method, $vid, $vid_t, $fun); + }; + } + }; +} + +macro_rules! impl_tertiary_ { + // implementation mapping 1:1 + (vec | $trait_id:ident, $trait_method:ident, $vec_id:ident, + $fun:ident) => { + impl $trait_id for $vec_id { + #[inline] + fn $trait_method(self, y: Self, z: Self) -> Self { + unsafe { + use crate::mem::transmute; + transmute($fun(transmute(self), transmute(y), transmute(z))) + } + } + } + }; + (scalar | $trait_id:ident, $trait_method:ident, + $vec_id:ident, [$sid:ident; $scount:expr], $fun:ident) => { + impl $trait_id for $vec_id { + #[inline] + fn $trait_method(self, y: Self, z: Self) -> Self { + unsafe { + union U { + vec: $vec_id, + scalars: [$sid; $scount], + } + let mut x = U { vec: self }.scalars; + let y = U { vec: y }.scalars; + let z = U { vec: z }.scalars; + for (x, (y, z)) in (&mut scalars).zip(&y).zip(&z) { + *i = $fun(*i, *y, *z); + } + U { vec: x }.vec + } + } + } + }; + // implementation calling fun twice on each of the vector halves: + (halves | $trait_id:ident, $trait_method:ident, $vec_id:ident, + $vech_id:ident, $fun:ident) => { + impl $trait_id for $vec_id { + #[inline] + fn $trait_method(self, y: Self, z: Self) -> Self { + unsafe { + use crate::mem::transmute; + union U { + vec: $vec_id, + halves: [$vech_id; 2], + } + + let mut x_halves = U { vec: self }.halves; + let y_halves = U { vec: y }.halves; + let z_halves = U { vec: z }.halves; + + *x_halves.get_unchecked_mut(0) = transmute($fun( + transmute(*x_halves.get_unchecked(0)), + transmute(*y_halves.get_unchecked(0)), + transmute(*z_halves.get_unchecked(0)), + )); + *x_halves.get_unchecked_mut(1) = transmute($fun( + transmute(*x_halves.get_unchecked(1)), + transmute(*y_halves.get_unchecked(1)), + transmute(*z_halves.get_unchecked(1)), + )); + + U { halves: x_halves }.vec + } + } + } + }; + // implementation calling fun four times on each of the vector quarters: + (quarter | $trait_id:ident, $trait_method:ident, $vec_id:ident, + $vecq_id:ident, $fun:ident) => { + impl $trait_id for $vec_id { + #[inline] + fn $trait_method(self, y: Self, z: Self) -> Self { + unsafe { + use crate::mem::transmute; + union U { + vec: $vec_id, + quarters: [$vecq_id; 4], + } + + let mut x_quarters = U { vec: self }.quarters; + let y_quarters = U { vec: y }.quarters; + let z_quarters = U { vec: z }.quarters; + + *x_quarters.get_unchecked_mut(0) = transmute($fun( + transmute(*x_quarters.get_unchecked(0)), + transmute(*y_quarters.get_unchecked(0)), + transmute(*z_quarters.get_unchecked(0)), + )); + + *x_quarters.get_unchecked_mut(1) = transmute($fun( + transmute(*x_quarters.get_unchecked(1)), + transmute(*y_quarters.get_unchecked(1)), + transmute(*z_quarters.get_unchecked(1)), + )); + + *x_quarters.get_unchecked_mut(2) = transmute($fun( + transmute(*x_quarters.get_unchecked(2)), + transmute(*y_quarters.get_unchecked(2)), + transmute(*z_quarters.get_unchecked(2)), + )); + + *x_quarters.get_unchecked_mut(3) = transmute($fun( + transmute(*x_quarters.get_unchecked(3)), + transmute(*y_quarters.get_unchecked(3)), + transmute(*z_quarters.get_unchecked(3)), + )); + + U { quarters: x_quarters }.vec + } + } + } + }; + // implementation calling fun once on a vector twice as large: + (twice | $trait_id:ident, $trait_method:ident, $vec_id:ident, + $vect_id:ident, $fun:ident) => { + impl $trait_id for $vec_id { + #[inline] + fn $trait_method(self, y: Self, z: Self) -> Self { + unsafe { + use crate::mem::{transmute, uninitialized}; + + union U { + vec: [$vec_id; 2], + twice: $vect_id, + } + + let x_twice = U { vec: [self, uninitialized()] }.twice; + let y_twice = U { vec: [y, uninitialized()] }.twice; + let z_twice = U { vec: [z, uninitialized()] }.twice; + let twice: $vect_id = + transmute($fun(transmute(x_twice), transmute(y_twice), transmute(z_twice))); + + *(U { twice }.vec.get_unchecked(0)) + } + } + } + }; +} + +macro_rules! gen_tertiary_impl_table { + ($trait_id:ident, $trait_method:ident) => { + macro_rules! impl_tertiary { + ($vid:ident: $fun:ident) => { + impl_tertiary_!(vec | $trait_id, $trait_method, $vid, $fun); + }; + ($vid:ident[$sid:ident; $sc:expr]: $fun:ident) => { + impl_tertiary_!(scalar | $trait_id, $trait_method, $vid, [$sid; $sc], $fun); + }; + ($vid:ident[s]: $fun:ident) => { + impl_tertiary_!(scalar | $trait_id, $trait_method, $vid, $fun); + }; + ($vid:ident[h => $vid_h:ident]: $fun:ident) => { + impl_tertiary_!(halves | $trait_id, $trait_method, $vid, $vid_h, $fun); + }; + ($vid:ident[q => $vid_q:ident]: $fun:ident) => { + impl_tertiary_!(quarter | $trait_id, $trait_method, $vid, $vid_q, $fun); + }; + ($vid:ident[t => $vid_t:ident]: $fun:ident) => { + impl_tertiary_!(twice | $trait_id, $trait_method, $vid, $vid_t, $fun); + }; + } + }; +} + +macro_rules! impl_binary_ { + // implementation mapping 1:1 + (vec | $trait_id:ident, $trait_method:ident, $vec_id:ident, + $fun:ident) => { + impl $trait_id for $vec_id { + #[inline] + fn $trait_method(self, y: Self) -> Self { + unsafe { + use crate::mem::transmute; + transmute($fun(transmute(self), transmute(y))) + } + } + } + }; + (scalar | $trait_id:ident, $trait_method:ident, + $vec_id:ident, [$sid:ident; $scount:expr], $fun:ident) => { + impl $trait_id for $vec_id { + #[inline] + fn $trait_method(self, y: Self) -> Self { + unsafe { + union U { + vec: $vec_id, + scalars: [$sid; $scount], + } + let mut x = U { vec: self }.scalars; + let y = U { vec: y }.scalars; + for (x, y) in x.iter_mut().zip(&y) { + *x = $fun(*x, *y); + } + U { scalars: x }.vec + } + } + } + }; + // implementation calling fun twice on each of the vector halves: + (halves | $trait_id:ident, $trait_method:ident, $vec_id:ident, + $vech_id:ident, $fun:ident) => { + impl $trait_id for $vec_id { + #[inline] + fn $trait_method(self, y: Self) -> Self { + unsafe { + use crate::mem::transmute; + union U { + vec: $vec_id, + halves: [$vech_id; 2], + } + + let mut x_halves = U { vec: self }.halves; + let y_halves = U { vec: y }.halves; + + *x_halves.get_unchecked_mut(0) = transmute($fun( + transmute(*x_halves.get_unchecked(0)), + transmute(*y_halves.get_unchecked(0)), + )); + *x_halves.get_unchecked_mut(1) = transmute($fun( + transmute(*x_halves.get_unchecked(1)), + transmute(*y_halves.get_unchecked(1)), + )); + + U { halves: x_halves }.vec + } + } + } + }; + // implementation calling fun four times on each of the vector quarters: + (quarter | $trait_id:ident, $trait_method:ident, $vec_id:ident, + $vecq_id:ident, $fun:ident) => { + impl $trait_id for $vec_id { + #[inline] + fn $trait_method(self, y: Self) -> Self { + unsafe { + use crate::mem::transmute; + union U { + vec: $vec_id, + quarters: [$vecq_id; 4], + } + + let mut x_quarters = U { vec: self }.quarters; + let y_quarters = U { vec: y }.quarters; + + *x_quarters.get_unchecked_mut(0) = transmute($fun( + transmute(*x_quarters.get_unchecked(0)), + transmute(*y_quarters.get_unchecked(0)), + )); + + *x_quarters.get_unchecked_mut(1) = transmute($fun( + transmute(*x_quarters.get_unchecked(1)), + transmute(*y_quarters.get_unchecked(1)), + )); + + *x_quarters.get_unchecked_mut(2) = transmute($fun( + transmute(*x_quarters.get_unchecked(2)), + transmute(*y_quarters.get_unchecked(2)), + )); + + *x_quarters.get_unchecked_mut(3) = transmute($fun( + transmute(*x_quarters.get_unchecked(3)), + transmute(*y_quarters.get_unchecked(3)), + )); + + U { quarters: x_quarters }.vec + } + } + } + }; + // implementation calling fun once on a vector twice as large: + (twice | $trait_id:ident, $trait_method:ident, $vec_id:ident, + $vect_id:ident, $fun:ident) => { + impl $trait_id for $vec_id { + #[inline] + fn $trait_method(self, y: Self) -> Self { + unsafe { + use crate::mem::{transmute, uninitialized}; + + union U { + vec: [$vec_id; 2], + twice: $vect_id, + } + + let x_twice = U { vec: [self, uninitialized()] }.twice; + let y_twice = U { vec: [y, uninitialized()] }.twice; + let twice: $vect_id = transmute($fun(transmute(x_twice), transmute(y_twice))); + + *(U { twice }.vec.get_unchecked(0)) + } + } + } + }; +} + +macro_rules! gen_binary_impl_table { + ($trait_id:ident, $trait_method:ident) => { + macro_rules! impl_binary { + ($vid:ident: $fun:ident) => { + impl_binary_!(vec | $trait_id, $trait_method, $vid, $fun); + }; + ($vid:ident[$sid:ident; $sc:expr]: $fun:ident) => { + impl_binary_!(scalar | $trait_id, $trait_method, $vid, [$sid; $sc], $fun); + }; + ($vid:ident[s]: $fun:ident) => { + impl_binary_!(scalar | $trait_id, $trait_method, $vid, $fun); + }; + ($vid:ident[h => $vid_h:ident]: $fun:ident) => { + impl_binary_!(halves | $trait_id, $trait_method, $vid, $vid_h, $fun); + }; + ($vid:ident[q => $vid_q:ident]: $fun:ident) => { + impl_binary_!(quarter | $trait_id, $trait_method, $vid, $vid_q, $fun); + }; + ($vid:ident[t => $vid_t:ident]: $fun:ident) => { + impl_binary_!(twice | $trait_id, $trait_method, $vid, $vid_t, $fun); + }; + } + }; +} diff --git a/third_party/rust/packed_simd/src/codegen/math/float/mul_add.rs b/third_party/rust/packed_simd/src/codegen/math/float/mul_add.rs new file mode 100644 index 000000000000..d37f30fa8614 --- /dev/null +++ b/third_party/rust/packed_simd/src/codegen/math/float/mul_add.rs @@ -0,0 +1,109 @@ +//! Vertical floating-point `mul_add` +#![allow(unused)] +use crate::*; + +// FIXME: 64-bit 1 element mul_add + +pub(crate) trait MulAdd { + fn mul_add(self, y: Self, z: Self) -> Self; +} + +#[cfg(not(target_arch = "s390x"))] +#[allow(improper_ctypes)] +extern "C" { + #[link_name = "llvm.fma.v2f32"] + fn fma_v2f32(x: f32x2, y: f32x2, z: f32x2) -> f32x2; + #[link_name = "llvm.fma.v4f32"] + fn fma_v4f32(x: f32x4, y: f32x4, z: f32x4) -> f32x4; + #[link_name = "llvm.fma.v8f32"] + fn fma_v8f32(x: f32x8, y: f32x8, z: f32x8) -> f32x8; + #[link_name = "llvm.fma.v16f32"] + fn fma_v16f32(x: f32x16, y: f32x16, z: f32x16) -> f32x16; + /* FIXME 64-bit single elem vectors + #[link_name = "llvm.fma.v1f64"] + fn fma_v1f64(x: f64x1, y: f64x1, z: f64x1) -> f64x1; + */ + #[link_name = "llvm.fma.v2f64"] + fn fma_v2f64(x: f64x2, y: f64x2, z: f64x2) -> f64x2; + #[link_name = "llvm.fma.v4f64"] + fn fma_v4f64(x: f64x4, y: f64x4, z: f64x4) -> f64x4; + #[link_name = "llvm.fma.v8f64"] + fn fma_v8f64(x: f64x8, y: f64x8, z: f64x8) -> f64x8; +} + +gen_tertiary_impl_table!(MulAdd, mul_add); + +cfg_if! { + if #[cfg(target_arch = "s390x")] { + // FIXME: https://github.com/rust-lang-nursery/packed_simd/issues/14 + macro_rules! impl_broken { + ($id:ident) => { + impl MulAdd for $id { + #[inline] + fn mul_add(self, y: Self, z: Self) -> Self { + self * y + z + } + } + }; + } + + impl_broken!(f32x2); + impl_broken!(f32x4); + impl_broken!(f32x8); + impl_broken!(f32x16); + + impl_broken!(f64x2); + impl_broken!(f64x4); + impl_broken!(f64x8); + } else if #[cfg(all(target_arch = "x86_64", feature = "sleef-sys"))] { + use sleef_sys::*; + cfg_if! { + if #[cfg(target_feature = "avx2")] { + impl_tertiary!(f32x2[t => f32x4]: Sleef_fmaf4_avx2128); + impl_tertiary!(f32x16[h => f32x8]: Sleef_fmaf8_avx2); + impl_tertiary!(f64x8[h => f64x4]: Sleef_fmad4_avx2); + + impl_tertiary!(f32x4: Sleef_fmaf4_avx2128); + impl_tertiary!(f32x8: Sleef_fmaf8_avx2); + impl_tertiary!(f64x2: Sleef_fmad2_avx2128); + impl_tertiary!(f64x4: Sleef_fmad4_avx2); + } else if #[cfg(target_feature = "avx")] { + impl_tertiary!(f32x2[t => f32x4]: Sleef_fmaf4_sse4); + impl_tertiary!(f32x16[h => f32x8]: Sleef_fmaf8_avx); + impl_tertiary!(f64x8[h => f64x4]: Sleef_fmad4_avx); + + impl_tertiary!(f32x4: Sleef_fmaf4_sse4); + impl_tertiary!(f32x8: Sleef_fmaf8_avx); + impl_tertiary!(f64x2: Sleef_fmad2_sse4); + impl_tertiary!(f64x4: Sleef_fmad4_avx); + } else if #[cfg(target_feature = "sse4.2")] { + impl_tertiary!(f32x2[t => f32x4]: Sleef_fmaf4_sse4); + impl_tertiary!(f32x16[q => f32x4]: Sleef_fmaf4_sse4); + impl_tertiary!(f64x8[q => f64x2]: Sleef_fmad2_sse4); + + impl_tertiary!(f32x4: Sleef_fmaf4_sse4); + impl_tertiary!(f32x8[h => f32x4]: Sleef_fmaf4_sse4); + impl_tertiary!(f64x2: Sleef_fmad2_sse4); + impl_tertiary!(f64x4[h => f64x2]: Sleef_fmad2_sse4); + } else { + impl_tertiary!(f32x2: fma_v2f32); + impl_tertiary!(f32x16: fma_v16f32); + impl_tertiary!(f64x8: fma_v8f64); + + impl_tertiary!(f32x4: fma_v4f32); + impl_tertiary!(f32x8: fma_v8f32); + impl_tertiary!(f64x2: fma_v2f64); + impl_tertiary!(f64x4: fma_v4f64); + } + } + } else { + impl_tertiary!(f32x2: fma_v2f32); + impl_tertiary!(f32x4: fma_v4f32); + impl_tertiary!(f32x8: fma_v8f32); + impl_tertiary!(f32x16: fma_v16f32); + // impl_tertiary!(f64x1: fma_v1f64); // FIXME 64-bit fmagle elem vectors + impl_tertiary!(f64x2: fma_v2f64); + impl_tertiary!(f64x4: fma_v4f64); + impl_tertiary!(f64x8: fma_v8f64); + } +} diff --git a/third_party/rust/packed_simd/src/codegen/math/float/mul_adde.rs b/third_party/rust/packed_simd/src/codegen/math/float/mul_adde.rs new file mode 100644 index 000000000000..c0baeacec20b --- /dev/null +++ b/third_party/rust/packed_simd/src/codegen/math/float/mul_adde.rs @@ -0,0 +1,60 @@ +//! Approximation for floating-point `mul_add` +use crate::*; + +// FIXME: 64-bit 1 element mul_adde + +pub(crate) trait MulAddE { + fn mul_adde(self, y: Self, z: Self) -> Self; +} + +#[cfg(not(target_arch = "s390x"))] +#[allow(improper_ctypes)] +extern "C" { + #[link_name = "llvm.fmuladd.v2f32"] + fn fmuladd_v2f32(x: f32x2, y: f32x2, z: f32x2) -> f32x2; + #[link_name = "llvm.fmuladd.v4f32"] + fn fmuladd_v4f32(x: f32x4, y: f32x4, z: f32x4) -> f32x4; + #[link_name = "llvm.fmuladd.v8f32"] + fn fmuladd_v8f32(x: f32x8, y: f32x8, z: f32x8) -> f32x8; + #[link_name = "llvm.fmuladd.v16f32"] + fn fmuladd_v16f32(x: f32x16, y: f32x16, z: f32x16) -> f32x16; + /* FIXME 64-bit single elem vectors + #[link_name = "llvm.fmuladd.v1f64"] + fn fmuladd_v1f64(x: f64x1, y: f64x1, z: f64x1) -> f64x1; + */ + #[link_name = "llvm.fmuladd.v2f64"] + fn fmuladd_v2f64(x: f64x2, y: f64x2, z: f64x2) -> f64x2; + #[link_name = "llvm.fmuladd.v4f64"] + fn fmuladd_v4f64(x: f64x4, y: f64x4, z: f64x4) -> f64x4; + #[link_name = "llvm.fmuladd.v8f64"] + fn fmuladd_v8f64(x: f64x8, y: f64x8, z: f64x8) -> f64x8; +} + +macro_rules! impl_mul_adde { + ($id:ident : $fn:ident) => { + impl MulAddE for $id { + #[inline] + fn mul_adde(self, y: Self, z: Self) -> Self { + #[cfg(not(target_arch = "s390x"))] + { + use crate::mem::transmute; + unsafe { transmute($fn(transmute(self), transmute(y), transmute(z))) } + } + #[cfg(target_arch = "s390x")] + { + // FIXME: https://github.com/rust-lang-nursery/packed_simd/issues/14 + self * y + z + } + } + } + }; +} + +impl_mul_adde!(f32x2: fmuladd_v2f32); +impl_mul_adde!(f32x4: fmuladd_v4f32); +impl_mul_adde!(f32x8: fmuladd_v8f32); +impl_mul_adde!(f32x16: fmuladd_v16f32); +// impl_mul_adde!(f64x1: fma_v1f64); // FIXME 64-bit fmagle elem vectors +impl_mul_adde!(f64x2: fmuladd_v2f64); +impl_mul_adde!(f64x4: fmuladd_v4f64); +impl_mul_adde!(f64x8: fmuladd_v8f64); diff --git a/third_party/rust/packed_simd/src/codegen/math/float/powf.rs b/third_party/rust/packed_simd/src/codegen/math/float/powf.rs new file mode 100644 index 000000000000..89ca52e96d81 --- /dev/null +++ b/third_party/rust/packed_simd/src/codegen/math/float/powf.rs @@ -0,0 +1,112 @@ +//! Vertical floating-point `powf` +#![allow(unused)] + +// FIXME 64-bit powfgle elem vectors mispowfg + +use crate::*; + +pub(crate) trait Powf { + fn powf(self, x: Self) -> Self; +} + +#[allow(improper_ctypes)] +extern "C" { + #[link_name = "llvm.pow.v2f32"] + fn powf_v2f32(x: f32x2, y: f32x2) -> f32x2; + #[link_name = "llvm.pow.v4f32"] + fn powf_v4f32(x: f32x4, y: f32x4) -> f32x4; + #[link_name = "llvm.pow.v8f32"] + fn powf_v8f32(x: f32x8, y: f32x8) -> f32x8; + #[link_name = "llvm.pow.v16f32"] + fn powf_v16f32(x: f32x16, y: f32x16) -> f32x16; + /* FIXME 64-bit powfgle elem vectors + #[link_name = "llvm.pow.v1f64"] + fn powf_v1f64(x: f64x1, y: f64x1) -> f64x1; + */ + #[link_name = "llvm.pow.v2f64"] + fn powf_v2f64(x: f64x2, y: f64x2) -> f64x2; + #[link_name = "llvm.pow.v4f64"] + fn powf_v4f64(x: f64x4, y: f64x4) -> f64x4; + #[link_name = "llvm.pow.v8f64"] + fn powf_v8f64(x: f64x8, y: f64x8) -> f64x8; + + #[link_name = "llvm.pow.f32"] + fn powf_f32(x: f32, y: f32) -> f32; + #[link_name = "llvm.pow.f64"] + fn powf_f64(x: f64, y: f64) -> f64; +} + +gen_binary_impl_table!(Powf, powf); + +cfg_if! { + if #[cfg(target_arch = "s390x")] { + // FIXME: https://github.com/rust-lang-nursery/packed_simd/issues/14 + impl_binary!(f32x2[f32; 2]: powf_f32); + impl_binary!(f32x4[f32; 4]: powf_f32); + impl_binary!(f32x8[f32; 8]: powf_f32); + impl_binary!(f32x16[f32; 16]: powf_f32); + + impl_binary!(f64x2[f64; 2]: powf_f64); + impl_binary!(f64x4[f64; 4]: powf_f64); + impl_binary!(f64x8[f64; 8]: powf_f64); + } else if #[cfg(all(target_arch = "x86_64", feature = "sleef-sys"))] { + use sleef_sys::*; + cfg_if! { + if #[cfg(target_feature = "avx2")] { + impl_binary!(f32x2[t => f32x4]: Sleef_powf4_u10avx2128); + impl_binary!(f32x16[h => f32x8]: Sleef_powf8_u10avx2); + impl_binary!(f64x8[h => f64x4]: Sleef_powd4_u10avx2); + + impl_binary!(f32x4: Sleef_powf4_u10avx2128); + impl_binary!(f32x8: Sleef_powf8_u10avx2); + impl_binary!(f64x2: Sleef_powd2_u10avx2128); + impl_binary!(f64x4: Sleef_powd4_u10avx2); + } else if #[cfg(target_feature = "avx")] { + impl_binary!(f32x2[t => f32x4]: Sleef_powf4_u10sse4); + impl_binary!(f32x16[h => f32x8]: Sleef_powf8_u10avx); + impl_binary!(f64x8[h => f64x4]: Sleef_powd4_u10avx); + + impl_binary!(f32x4: Sleef_powf4_u10sse4); + impl_binary!(f32x8: Sleef_powf8_u10avx); + impl_binary!(f64x2: Sleef_powd2_u10sse4); + impl_binary!(f64x4: Sleef_powd4_u10avx); + } else if #[cfg(target_feature = "sse4.2")] { + impl_binary!(f32x2[t => f32x4]: Sleef_powf4_u10sse4); + impl_binary!(f32x16[q => f32x4]: Sleef_powf4_u10sse4); + impl_binary!(f64x8[q => f64x2]: Sleef_powd2_u10sse4); + + impl_binary!(f32x4: Sleef_powf4_u10sse4); + impl_binary!(f32x8[h => f32x4]: Sleef_powf4_u10sse4); + impl_binary!(f64x2: Sleef_powd2_u10sse4); + impl_binary!(f64x4[h => f64x2]: Sleef_powd2_u10sse4); + } else if #[cfg(target_feature = "sse2")] { + impl_binary!(f32x2[t => f32x4]: Sleef_powf4_u10sse2); + impl_binary!(f32x16[q => f32x4]: Sleef_powf4_u10sse2); + impl_binary!(f64x8[q => f64x2]: Sleef_powd2_u10sse2); + + impl_binary!(f32x4: Sleef_powf4_u10sse2); + impl_binary!(f32x8[h => f32x4]: Sleef_powf4_u10sse2); + impl_binary!(f64x2: Sleef_powd2_u10sse2); + impl_binary!(f64x4[h => f64x2]: Sleef_powd2_u10sse2); + } else { + impl_binary!(f32x2[f32; 2]: powf_f32); + impl_binary!(f32x4: powf_v4f32); + impl_binary!(f32x8: powf_v8f32); + impl_binary!(f32x16: powf_v16f32); + + impl_binary!(f64x2: powf_v2f64); + impl_binary!(f64x4: powf_v4f64); + impl_binary!(f64x8: powf_v8f64); + } + } + } else { + impl_binary!(f32x2[f32; 2]: powf_f32); + impl_binary!(f32x4: powf_v4f32); + impl_binary!(f32x8: powf_v8f32); + impl_binary!(f32x16: powf_v16f32); + + impl_binary!(f64x2: powf_v2f64); + impl_binary!(f64x4: powf_v4f64); + impl_binary!(f64x8: powf_v8f64); + } +} diff --git a/third_party/rust/packed_simd/src/codegen/math/float/sin.rs b/third_party/rust/packed_simd/src/codegen/math/float/sin.rs new file mode 100644 index 000000000000..d881415909af --- /dev/null +++ b/third_party/rust/packed_simd/src/codegen/math/float/sin.rs @@ -0,0 +1,103 @@ +//! Vertical floating-point `sin` +#![allow(unused)] + +// FIXME 64-bit 1 elem vectors sin + +use crate::*; + +pub(crate) trait Sin { + fn sin(self) -> Self; +} + +#[allow(improper_ctypes)] +extern "C" { + #[link_name = "llvm.sin.v2f32"] + fn sin_v2f32(x: f32x2) -> f32x2; + #[link_name = "llvm.sin.v4f32"] + fn sin_v4f32(x: f32x4) -> f32x4; + #[link_name = "llvm.sin.v8f32"] + fn sin_v8f32(x: f32x8) -> f32x8; + #[link_name = "llvm.sin.v16f32"] + fn sin_v16f32(x: f32x16) -> f32x16; + /* FIXME 64-bit single elem vectors + #[link_name = "llvm.sin.v1f64"] + fn sin_v1f64(x: f64x1) -> f64x1; + */ + #[link_name = "llvm.sin.v2f64"] + fn sin_v2f64(x: f64x2) -> f64x2; + #[link_name = "llvm.sin.v4f64"] + fn sin_v4f64(x: f64x4) -> f64x4; + #[link_name = "llvm.sin.v8f64"] + fn sin_v8f64(x: f64x8) -> f64x8; + + #[link_name = "llvm.sin.f32"] + fn sin_f32(x: f32) -> f32; + #[link_name = "llvm.sin.f64"] + fn sin_f64(x: f64) -> f64; +} + +gen_unary_impl_table!(Sin, sin); + +cfg_if! { + if #[cfg(target_arch = "s390x")] { + // FIXME: https://github.com/rust-lang-nursery/packed_simd/issues/14 + impl_unary!(f32x2[f32; 2]: sin_f32); + impl_unary!(f32x4[f32; 4]: sin_f32); + impl_unary!(f32x8[f32; 8]: sin_f32); + impl_unary!(f32x16[f32; 16]: sin_f32); + + impl_unary!(f64x2[f64; 2]: sin_f64); + impl_unary!(f64x4[f64; 4]: sin_f64); + impl_unary!(f64x8[f64; 8]: sin_f64); + } else if #[cfg(all(target_arch = "x86_64", feature = "sleef-sys"))] { + use sleef_sys::*; + cfg_if! { + if #[cfg(target_feature = "avx2")] { + impl_unary!(f32x2[t => f32x4]: Sleef_sinf4_u10avx2128); + impl_unary!(f32x16[h => f32x8]: Sleef_sinf8_u10avx2); + impl_unary!(f64x8[h => f64x4]: Sleef_sind4_u10avx2); + + impl_unary!(f32x4: Sleef_sinf4_u10avx2128); + impl_unary!(f32x8: Sleef_sinf8_u10avx2); + impl_unary!(f64x2: Sleef_sind2_u10avx2128); + impl_unary!(f64x4: Sleef_sind4_u10avx2); + } else if #[cfg(target_feature = "avx")] { + impl_unary!(f32x2[t => f32x4]: Sleef_sinf4_u10sse4); + impl_unary!(f32x16[h => f32x8]: Sleef_sinf8_u10avx); + impl_unary!(f64x8[h => f64x4]: Sleef_sind4_u10avx); + + impl_unary!(f32x4: Sleef_sinf4_u10sse4); + impl_unary!(f32x8: Sleef_sinf8_u10avx); + impl_unary!(f64x2: Sleef_sind2_u10sse4); + impl_unary!(f64x4: Sleef_sind4_u10avx); + } else if #[cfg(target_feature = "sse4.2")] { + impl_unary!(f32x2[t => f32x4]: Sleef_sinf4_u10sse4); + impl_unary!(f32x16[q => f32x4]: Sleef_sinf4_u10sse4); + impl_unary!(f64x8[q => f64x2]: Sleef_sind2_u10sse4); + + impl_unary!(f32x4: Sleef_sinf4_u10sse4); + impl_unary!(f32x8[h => f32x4]: Sleef_sinf4_u10sse4); + impl_unary!(f64x2: Sleef_sind2_u10sse4); + impl_unary!(f64x4[h => f64x2]: Sleef_sind2_u10sse4); + } else { + impl_unary!(f32x2[f32; 2]: sin_f32); + impl_unary!(f32x16: sin_v16f32); + impl_unary!(f64x8: sin_v8f64); + + impl_unary!(f32x4: sin_v4f32); + impl_unary!(f32x8: sin_v8f32); + impl_unary!(f64x2: sin_v2f64); + impl_unary!(f64x4: sin_v4f64); + } + } + } else { + impl_unary!(f32x2[f32; 2]: sin_f32); + impl_unary!(f32x4: sin_v4f32); + impl_unary!(f32x8: sin_v8f32); + impl_unary!(f32x16: sin_v16f32); + + impl_unary!(f64x2: sin_v2f64); + impl_unary!(f64x4: sin_v4f64); + impl_unary!(f64x8: sin_v8f64); + } +} diff --git a/third_party/rust/packed_simd/src/codegen/math/float/sin_cos_pi.rs b/third_party/rust/packed_simd/src/codegen/math/float/sin_cos_pi.rs new file mode 100644 index 000000000000..b283d11111fd --- /dev/null +++ b/third_party/rust/packed_simd/src/codegen/math/float/sin_cos_pi.rs @@ -0,0 +1,188 @@ +//! Vertical floating-point `sin_cos` +#![allow(unused)] + +// FIXME 64-bit 1 elem vectors sin_cos + +use crate::*; + +pub(crate) trait SinCosPi: Sized { + type Output; + fn sin_cos_pi(self) -> Self::Output; +} + +macro_rules! impl_def { + ($vid:ident, $PI:path) => { + impl SinCosPi for $vid { + type Output = (Self, Self); + #[inline] + fn sin_cos_pi(self) -> Self::Output { + let v = self * Self::splat($PI); + (v.sin(), v.cos()) + } + } + }; +} + +macro_rules! impl_def32 { + ($vid:ident) => { + impl_def!($vid, crate::f32::consts::PI); + }; +} +macro_rules! impl_def64 { + ($vid:ident) => { + impl_def!($vid, crate::f64::consts::PI); + }; +} + +macro_rules! impl_unary_t { + ($vid:ident: $fun:ident) => { + impl SinCosPi for $vid { + type Output = (Self, Self); + fn sin_cos_pi(self) -> Self::Output { + unsafe { + use crate::mem::transmute; + transmute($fun(transmute(self))) + } + } + } + }; + ($vid:ident[t => $vid_t:ident]: $fun:ident) => { + impl SinCosPi for $vid { + type Output = (Self, Self); + fn sin_cos_pi(self) -> Self::Output { + unsafe { + use crate::mem::{transmute, uninitialized}; + + union U { + vec: [$vid; 2], + twice: $vid_t, + } + + let twice = U { vec: [self, uninitialized()] }.twice; + let twice = transmute($fun(transmute(twice))); + + union R { + twice: ($vid_t, $vid_t), + vecs: ([$vid; 2], [$vid; 2]), + } + let r = R { twice }.vecs; + (*r.0.get_unchecked(0), *r.0.get_unchecked(1)) + } + } + } + }; + ($vid:ident[h => $vid_h:ident]: $fun:ident) => { + impl SinCosPi for $vid { + type Output = (Self, Self); + fn sin_cos_pi(self) -> Self::Output { + unsafe { + use crate::mem::transmute; + + union U { + vec: $vid, + halves: [$vid_h; 2], + } + + let halves = U { vec: self }.halves; + + let res_0: ($vid_h, $vid_h) = transmute($fun(transmute(*halves.get_unchecked(0)))); + let res_1: ($vid_h, $vid_h) = transmute($fun(transmute(*halves.get_unchecked(1)))); + + union R { + result: ($vid, $vid), + halves: ([$vid_h; 2], [$vid_h; 2]), + } + R { halves: ([res_0.0, res_1.0], [res_0.1, res_1.1]) }.result + } + } + } + }; + ($vid:ident[q => $vid_q:ident]: $fun:ident) => { + impl SinCosPi for $vid { + type Output = (Self, Self); + fn sin_cos_pi(self) -> Self::Output { + unsafe { + use crate::mem::transmute; + + union U { + vec: $vid, + quarters: [$vid_q; 4], + } + + let quarters = U { vec: self }.quarters; + + let res_0: ($vid_q, $vid_q) = transmute($fun(transmute(*quarters.get_unchecked(0)))); + let res_1: ($vid_q, $vid_q) = transmute($fun(transmute(*quarters.get_unchecked(1)))); + let res_2: ($vid_q, $vid_q) = transmute($fun(transmute(*quarters.get_unchecked(2)))); + let res_3: ($vid_q, $vid_q) = transmute($fun(transmute(*quarters.get_unchecked(3)))); + + union R { + result: ($vid, $vid), + quarters: ([$vid_q; 4], [$vid_q; 4]), + } + R { + quarters: ( + [res_0.0, res_1.0, res_2.0, res_3.0], + [res_0.1, res_1.1, res_2.1, res_3.1], + ), + } + .result + } + } + } + }; +} + +cfg_if! { + if #[cfg(all(target_arch = "x86_64", feature = "sleef-sys"))] { + use sleef_sys::*; + cfg_if! { + if #[cfg(target_feature = "avx2")] { + impl_unary_t!(f32x2[t => f32x4]: Sleef_sincospif4_u05avx2128); + impl_unary_t!(f32x16[h => f32x8]: Sleef_sincospif8_u05avx2); + impl_unary_t!(f64x8[h => f64x4]: Sleef_sincospid4_u05avx2); + + impl_unary_t!(f32x4: Sleef_sincospif4_u05avx2128); + impl_unary_t!(f32x8: Sleef_sincospif8_u05avx2); + impl_unary_t!(f64x2: Sleef_sincospid2_u05avx2128); + impl_unary_t!(f64x4: Sleef_sincospid4_u05avx2); + } else if #[cfg(target_feature = "avx")] { + impl_unary_t!(f32x2[t => f32x4]: Sleef_sincospif4_u05sse4); + impl_unary_t!(f32x16[h => f32x8]: Sleef_sincospif8_u05avx); + impl_unary_t!(f64x8[h => f64x4]: Sleef_sincospid4_u05avx); + + impl_unary_t!(f32x4: Sleef_sincospif4_u05sse4); + impl_unary_t!(f32x8: Sleef_sincospif8_u05avx); + impl_unary_t!(f64x2: Sleef_sincospid2_u05sse4); + impl_unary_t!(f64x4: Sleef_sincospid4_u05avx); + } else if #[cfg(target_feature = "sse4.2")] { + impl_unary_t!(f32x2[t => f32x4]: Sleef_sincospif4_u05sse4); + impl_unary_t!(f32x16[q => f32x4]: Sleef_sincospif4_u05sse4); + impl_unary_t!(f64x8[q => f64x2]: Sleef_sincospid2_u05sse4); + + impl_unary_t!(f32x4: Sleef_sincospif4_u05sse4); + impl_unary_t!(f32x8[h => f32x4]: Sleef_sincospif4_u05sse4); + impl_unary_t!(f64x2: Sleef_sincospid2_u05sse4); + impl_unary_t!(f64x4[h => f64x2]: Sleef_sincospid2_u05sse4); + } else { + impl_def32!(f32x2); + impl_def32!(f32x4); + impl_def32!(f32x8); + impl_def32!(f32x16); + + impl_def64!(f64x2); + impl_def64!(f64x4); + impl_def64!(f64x8); + } + } + } else { + impl_def32!(f32x2); + impl_def32!(f32x4); + impl_def32!(f32x8); + impl_def32!(f32x16); + + impl_def64!(f64x2); + impl_def64!(f64x4); + impl_def64!(f64x8); + } +} diff --git a/third_party/rust/packed_simd/src/codegen/math/float/sin_pi.rs b/third_party/rust/packed_simd/src/codegen/math/float/sin_pi.rs new file mode 100644 index 000000000000..0c8f6bb12050 --- /dev/null +++ b/third_party/rust/packed_simd/src/codegen/math/float/sin_pi.rs @@ -0,0 +1,87 @@ +//! Vertical floating-point `sin_pi` +#![allow(unused)] + +// FIXME 64-bit 1 elem vectors sin_pi + +use crate::*; + +pub(crate) trait SinPi { + fn sin_pi(self) -> Self; +} + +gen_unary_impl_table!(SinPi, sin_pi); + +macro_rules! impl_def { + ($vid:ident, $PI:path) => { + impl SinPi for $vid { + #[inline] + fn sin_pi(self) -> Self { + (self * Self::splat($PI)).sin() + } + } + }; +} +macro_rules! impl_def32 { + ($vid:ident) => { + impl_def!($vid, crate::f32::consts::PI); + }; +} +macro_rules! impl_def64 { + ($vid:ident) => { + impl_def!($vid, crate::f64::consts::PI); + }; +} + +cfg_if! { + if #[cfg(all(target_arch = "x86_64", feature = "sleef-sys"))] { + use sleef_sys::*; + cfg_if! { + if #[cfg(target_feature = "avx2")] { + impl_unary!(f32x2[t => f32x4]: Sleef_sinpif4_u05avx2128); + impl_unary!(f32x16[h => f32x8]: Sleef_sinpif8_u05avx2); + impl_unary!(f64x8[h => f64x4]: Sleef_sinpid4_u05avx2); + + impl_unary!(f32x4: Sleef_sinpif4_u05avx2128); + impl_unary!(f32x8: Sleef_sinpif8_u05avx2); + impl_unary!(f64x2: Sleef_sinpid2_u05avx2128); + impl_unary!(f64x4: Sleef_sinpid4_u05avx2); + } else if #[cfg(target_feature = "avx")] { + impl_unary!(f32x2[t => f32x4]: Sleef_sinpif4_u05sse4); + impl_unary!(f32x16[h => f32x8]: Sleef_sinpif8_u05avx); + impl_unary!(f64x8[h => f64x4]: Sleef_sinpid4_u05avx); + + impl_unary!(f32x4: Sleef_sinpif4_u05sse4); + impl_unary!(f32x8: Sleef_sinpif8_u05avx); + impl_unary!(f64x2: Sleef_sinpid2_u05sse4); + impl_unary!(f64x4: Sleef_sinpid4_u05avx); + } else if #[cfg(target_feature = "sse4.2")] { + impl_unary!(f32x2[t => f32x4]: Sleef_sinpif4_u05sse4); + impl_unary!(f32x16[q => f32x4]: Sleef_sinpif4_u05sse4); + impl_unary!(f64x8[q => f64x2]: Sleef_sinpid2_u05sse4); + + impl_unary!(f32x4: Sleef_sinpif4_u05sse4); + impl_unary!(f32x8[h => f32x4]: Sleef_sinpif4_u05sse4); + impl_unary!(f64x2: Sleef_sinpid2_u05sse4); + impl_unary!(f64x4[h => f64x2]: Sleef_sinpid2_u05sse4); + } else { + impl_def32!(f32x2); + impl_def32!(f32x4); + impl_def32!(f32x8); + impl_def32!(f32x16); + + impl_def64!(f64x2); + impl_def64!(f64x4); + impl_def64!(f64x8); + } + } + } else { + impl_def32!(f32x2); + impl_def32!(f32x4); + impl_def32!(f32x8); + impl_def32!(f32x16); + + impl_def64!(f64x2); + impl_def64!(f64x4); + impl_def64!(f64x8); + } +} diff --git a/third_party/rust/packed_simd/src/codegen/math/float/sqrt.rs b/third_party/rust/packed_simd/src/codegen/math/float/sqrt.rs new file mode 100644 index 000000000000..67bb0a2a9c59 --- /dev/null +++ b/third_party/rust/packed_simd/src/codegen/math/float/sqrt.rs @@ -0,0 +1,103 @@ +//! Vertical floating-point `sqrt` +#![allow(unused)] + +// FIXME 64-bit 1 elem vectors sqrt + +use crate::*; + +pub(crate) trait Sqrt { + fn sqrt(self) -> Self; +} + +#[allow(improper_ctypes)] +extern "C" { + #[link_name = "llvm.sqrt.v2f32"] + fn sqrt_v2f32(x: f32x2) -> f32x2; + #[link_name = "llvm.sqrt.v4f32"] + fn sqrt_v4f32(x: f32x4) -> f32x4; + #[link_name = "llvm.sqrt.v8f32"] + fn sqrt_v8f32(x: f32x8) -> f32x8; + #[link_name = "llvm.sqrt.v16f32"] + fn sqrt_v16f32(x: f32x16) -> f32x16; + /* FIXME 64-bit sqrtgle elem vectors + #[link_name = "llvm.sqrt.v1f64"] + fn sqrt_v1f64(x: f64x1) -> f64x1; + */ + #[link_name = "llvm.sqrt.v2f64"] + fn sqrt_v2f64(x: f64x2) -> f64x2; + #[link_name = "llvm.sqrt.v4f64"] + fn sqrt_v4f64(x: f64x4) -> f64x4; + #[link_name = "llvm.sqrt.v8f64"] + fn sqrt_v8f64(x: f64x8) -> f64x8; + + #[link_name = "llvm.sqrt.f32"] + fn sqrt_f32(x: f32) -> f32; + #[link_name = "llvm.sqrt.f64"] + fn sqrt_f64(x: f64) -> f64; +} + +gen_unary_impl_table!(Sqrt, sqrt); + +cfg_if! { + if #[cfg(target_arch = "s390x")] { + // FIXME: https://github.com/rust-lang-nursery/packed_simd/issues/14 + impl_unary!(f32x2[f32; 2]: sqrt_f32); + impl_unary!(f32x4[f32; 4]: sqrt_f32); + impl_unary!(f32x8[f32; 8]: sqrt_f32); + impl_unary!(f32x16[f32; 16]: sqrt_f32); + + impl_unary!(f64x2[f64; 2]: sqrt_f64); + impl_unary!(f64x4[f64; 4]: sqrt_f64); + impl_unary!(f64x8[f64; 8]: sqrt_f64); + } else if #[cfg(all(target_arch = "x86_64", feature = "sleef-sys"))] { + use sleef_sys::*; + cfg_if! { + if #[cfg(target_feature = "avx2")] { + impl_unary!(f32x2[t => f32x4]: Sleef_sqrtf4_avx2128); + impl_unary!(f32x16[h => f32x8]: Sleef_sqrtf8_avx2); + impl_unary!(f64x8[h => f64x4]: Sleef_sqrtd4_avx2); + + impl_unary!(f32x4: Sleef_sqrtf4_avx2128); + impl_unary!(f32x8: Sleef_sqrtf8_avx2); + impl_unary!(f64x2: Sleef_sqrtd2_avx2128); + impl_unary!(f64x4: Sleef_sqrtd4_avx2); + } else if #[cfg(target_feature = "avx")] { + impl_unary!(f32x2[t => f32x4]: Sleef_sqrtf4_sse4); + impl_unary!(f32x16[h => f32x8]: Sleef_sqrtf8_avx); + impl_unary!(f64x8[h => f64x4]: Sleef_sqrtd4_avx); + + impl_unary!(f32x4: Sleef_sqrtf4_sse4); + impl_unary!(f32x8: Sleef_sqrtf8_avx); + impl_unary!(f64x2: Sleef_sqrtd2_sse4); + impl_unary!(f64x4: Sleef_sqrtd4_avx); + } else if #[cfg(target_feature = "sse4.2")] { + impl_unary!(f32x2[t => f32x4]: Sleef_sqrtf4_sse4); + impl_unary!(f32x16[q => f32x4]: Sleef_sqrtf4_sse4); + impl_unary!(f64x8[q => f64x2]: Sleef_sqrtd2_sse4); + + impl_unary!(f32x4: Sleef_sqrtf4_sse4); + impl_unary!(f32x8[h => f32x4]: Sleef_sqrtf4_sse4); + impl_unary!(f64x2: Sleef_sqrtd2_sse4); + impl_unary!(f64x4[h => f64x2]: Sleef_sqrtd2_sse4); + } else { + impl_unary!(f32x2[f32; 2]: sqrt_f32); + impl_unary!(f32x16: sqrt_v16f32); + impl_unary!(f64x8: sqrt_v8f64); + + impl_unary!(f32x4: sqrt_v4f32); + impl_unary!(f32x8: sqrt_v8f32); + impl_unary!(f64x2: sqrt_v2f64); + impl_unary!(f64x4: sqrt_v4f64); + } + } + } else { + impl_unary!(f32x2[f32; 2]: sqrt_f32); + impl_unary!(f32x4: sqrt_v4f32); + impl_unary!(f32x8: sqrt_v8f32); + impl_unary!(f32x16: sqrt_v16f32); + + impl_unary!(f64x2: sqrt_v2f64); + impl_unary!(f64x4: sqrt_v4f64); + impl_unary!(f64x8: sqrt_v8f64); + } +} diff --git a/third_party/rust/packed_simd/src/codegen/math/float/sqrte.rs b/third_party/rust/packed_simd/src/codegen/math/float/sqrte.rs new file mode 100644 index 000000000000..58a1de1f400f --- /dev/null +++ b/third_party/rust/packed_simd/src/codegen/math/float/sqrte.rs @@ -0,0 +1,67 @@ +//! Vertical floating-point `sqrt` +#![allow(unused)] + +// FIXME 64-bit 1 elem vectors sqrte + +use crate::llvm::simd_fsqrt; +use crate::*; + +pub(crate) trait Sqrte { + fn sqrte(self) -> Self; +} + +gen_unary_impl_table!(Sqrte, sqrte); + +cfg_if! { + if #[cfg(all(target_arch = "x86_64", feature = "sleef-sys"))] { + use sleef_sys::*; + cfg_if! { + if #[cfg(target_feature = "avx2")] { + impl_unary!(f32x2[t => f32x4]: Sleef_sqrtf4_u35avx2128); + impl_unary!(f32x16[h => f32x8]: Sleef_sqrtf8_u35avx2); + impl_unary!(f64x8[h => f64x4]: Sleef_sqrtd4_u35avx2); + + impl_unary!(f32x4: Sleef_sqrtf4_u35avx2128); + impl_unary!(f32x8: Sleef_sqrtf8_u35avx2); + impl_unary!(f64x2: Sleef_sqrtd2_u35avx2128); + impl_unary!(f64x4: Sleef_sqrtd4_u35avx2); + } else if #[cfg(target_feature = "avx")] { + impl_unary!(f32x2[t => f32x4]: Sleef_sqrtf4_u35sse4); + impl_unary!(f32x16[h => f32x8]: Sleef_sqrtf8_u35avx); + impl_unary!(f64x8[h => f64x4]: Sleef_sqrtd4_u35avx); + + impl_unary!(f32x4: Sleef_sqrtf4_u35sse4); + impl_unary!(f32x8: Sleef_sqrtf8_u35avx); + impl_unary!(f64x2: Sleef_sqrtd2_u35sse4); + impl_unary!(f64x4: Sleef_sqrtd4_u35avx); + } else if #[cfg(target_feature = "sse4.2")] { + impl_unary!(f32x2[t => f32x4]: Sleef_sqrtf4_u35sse4); + impl_unary!(f32x16[q => f32x4]: Sleef_sqrtf4_u35sse4); + impl_unary!(f64x8[q => f64x2]: Sleef_sqrtd2_u35sse4); + + impl_unary!(f32x4: Sleef_sqrtf4_u35sse4); + impl_unary!(f32x8[h => f32x4]: Sleef_sqrtf4_u35sse4); + impl_unary!(f64x2: Sleef_sqrtd2_u35sse4); + impl_unary!(f64x4[h => f64x2]: Sleef_sqrtd2_u35sse4); + } else { + impl_unary!(f32x2[g]: simd_fsqrt); + impl_unary!(f32x16[g]: simd_fsqrt); + impl_unary!(f64x8[g]: simd_fsqrt); + + impl_unary!(f32x4[g]: simd_fsqrt); + impl_unary!(f32x8[g]: simd_fsqrt); + impl_unary!(f64x2[g]: simd_fsqrt); + impl_unary!(f64x4[g]: simd_fsqrt); + } + } + } else { + impl_unary!(f32x2[g]: simd_fsqrt); + impl_unary!(f32x4[g]: simd_fsqrt); + impl_unary!(f32x8[g]: simd_fsqrt); + impl_unary!(f32x16[g]: simd_fsqrt); + + impl_unary!(f64x2[g]: simd_fsqrt); + impl_unary!(f64x4[g]: simd_fsqrt); + impl_unary!(f64x8[g]: simd_fsqrt); + } +} diff --git a/third_party/rust/packed_simd/src/codegen/math/float/tanh.rs b/third_party/rust/packed_simd/src/codegen/math/float/tanh.rs new file mode 100644 index 000000000000..4243b0d886c5 --- /dev/null +++ b/third_party/rust/packed_simd/src/codegen/math/float/tanh.rs @@ -0,0 +1,120 @@ +//! Vertical floating-point `tanh` +#![allow(unused)] + +// FIXME 64-bit 1 elem vectors tanh + +#[cfg(not(feature = "std"))] +use num_traits::Float; + +use crate::*; + +pub(crate) trait Tanh { + fn tanh(self) -> Self; +} + +macro_rules! define_tanh { + ($name:ident, $basetype:ty, $simdtype:ty, $lanes:expr, $trait:path) => { + fn $name(x: $simdtype) -> $simdtype { + use core::intrinsics::transmute; + let mut buf: [$basetype; $lanes] = unsafe { transmute(x) }; + for elem in &mut buf { + *elem = <$basetype as $trait>::tanh(*elem); + } + unsafe { transmute(buf) } + } + }; + + (f32 => $name:ident, $type:ty, $lanes:expr) => { + define_tanh!($name, f32, $type, $lanes, Float); + }; + + (f64 => $name:ident, $type:ty, $lanes:expr) => { + define_tanh!($name, f64, $type, $lanes, Float); + }; +} + +// llvm does not seem to expose the hyperbolic versions of trigonometric +// functions; we thus call the classical rust versions on all of them (which +// stem from cmath). +define_tanh!(f32 => tanh_v2f32, f32x2, 2); +define_tanh!(f32 => tanh_v4f32, f32x4, 4); +define_tanh!(f32 => tanh_v8f32, f32x8, 8); +define_tanh!(f32 => tanh_v16f32, f32x16, 16); + +define_tanh!(f64 => tanh_v2f64, f64x2, 2); +define_tanh!(f64 => tanh_v4f64, f64x4, 4); +define_tanh!(f64 => tanh_v8f64, f64x8, 8); + +fn tanh_f32(x: f32) -> f32 { + Float::tanh(x) +} + +fn tanh_f64(x: f64) -> f64 { + Float::tanh(x) +} + +gen_unary_impl_table!(Tanh, tanh); + +cfg_if! { + if #[cfg(target_arch = "s390x")] { + // FIXME: https://github.com/rust-lang-nursery/packed_simd/issues/14 + impl_unary!(f32x2[f32; 2]: tanh_f32); + impl_unary!(f32x4[f32; 4]: tanh_f32); + impl_unary!(f32x8[f32; 8]: tanh_f32); + impl_unary!(f32x16[f32; 16]: tanh_f32); + + impl_unary!(f64x2[f64; 2]: tanh_f64); + impl_unary!(f64x4[f64; 4]: tanh_f64); + impl_unary!(f64x8[f64; 8]: tanh_f64); + } else if #[cfg(all(target_arch = "x86_64", feature = "sleef-sys"))] { + use sleef_sys::*; + cfg_if! { + if #[cfg(target_feature = "avx2")] { + impl_unary!(f32x2[t => f32x4]: Sleef_tanhf4_u10avx2128); + impl_unary!(f32x16[h => f32x8]: Sleef_tanhf8_u10avx2); + impl_unary!(f64x8[h => f64x4]: Sleef_tanhd4_u10avx2); + + impl_unary!(f32x4: Sleef_tanhf4_u10avx2128); + impl_unary!(f32x8: Sleef_tanhf8_u10avx2); + impl_unary!(f64x2: Sleef_tanhd2_u10avx2128); + impl_unary!(f64x4: Sleef_tanhd4_u10avx2); + } else if #[cfg(target_feature = "avx")] { + impl_unary!(f32x2[t => f32x4]: Sleef_tanhf4_u10sse4); + impl_unary!(f32x16[h => f32x8]: Sleef_tanhf8_u10avx); + impl_unary!(f64x8[h => f64x4]: Sleef_tanhd4_u10avx); + + impl_unary!(f32x4: Sleef_tanhf4_u10sse4); + impl_unary!(f32x8: Sleef_tanhf8_u10avx); + impl_unary!(f64x2: Sleef_tanhd2_u10sse4); + impl_unary!(f64x4: Sleef_tanhd4_u10avx); + } else if #[cfg(target_feature = "sse4.2")] { + impl_unary!(f32x2[t => f32x4]: Sleef_tanhf4_u10sse4); + impl_unary!(f32x16[q => f32x4]: Sleef_tanhf4_u10sse4); + impl_unary!(f64x8[q => f64x2]: Sleef_tanhd2_u10sse4); + + impl_unary!(f32x4: Sleef_tanhf4_u10sse4); + impl_unary!(f32x8[h => f32x4]: Sleef_tanhf4_u10sse4); + impl_unary!(f64x2: Sleef_tanhd2_u10sse4); + impl_unary!(f64x4[h => f64x2]: Sleef_tanhd2_u10sse4); + } else { + impl_unary!(f32x2[f32; 2]: tanh_f32); + impl_unary!(f32x16: tanh_v16f32); + impl_unary!(f64x8: tanh_v8f64); + + impl_unary!(f32x4: tanh_v4f32); + impl_unary!(f32x8: tanh_v8f32); + impl_unary!(f64x2: tanh_v2f64); + impl_unary!(f64x4: tanh_v4f64); + } + } + } else { + impl_unary!(f32x2[f32; 2]: tanh_f32); + impl_unary!(f32x4: tanh_v4f32); + impl_unary!(f32x8: tanh_v8f32); + impl_unary!(f32x16: tanh_v16f32); + + impl_unary!(f64x2: tanh_v2f64); + impl_unary!(f64x4: tanh_v4f64); + impl_unary!(f64x8: tanh_v8f64); + } +} diff --git a/third_party/rust/packed_simd/src/codegen/pointer_sized_int.rs b/third_party/rust/packed_simd/src/codegen/pointer_sized_int.rs new file mode 100644 index 000000000000..55cbc297aaf5 --- /dev/null +++ b/third_party/rust/packed_simd/src/codegen/pointer_sized_int.rs @@ -0,0 +1,28 @@ +//! Provides `isize` and `usize` + +use cfg_if::cfg_if; + +cfg_if! { + if #[cfg(target_pointer_width = "8")] { + pub(crate) type isize_ = i8; + pub(crate) type usize_ = u8; + } else if #[cfg(target_pointer_width = "16")] { + pub(crate) type isize_ = i16; + pub(crate) type usize_ = u16; + } else if #[cfg(target_pointer_width = "32")] { + pub(crate) type isize_ = i32; + pub(crate) type usize_ = u32; + + } else if #[cfg(target_pointer_width = "64")] { + pub(crate) type isize_ = i64; + pub(crate) type usize_ = u64; + } else if #[cfg(target_pointer_width = "64")] { + pub(crate) type isize_ = i64; + pub(crate) type usize_ = u64; + } else if #[cfg(target_pointer_width = "128")] { + pub(crate) type isize_ = i128; + pub(crate) type usize_ = u128; + } else { + compile_error!("unsupported target_pointer_width"); + } +} diff --git a/third_party/rust/packed_simd/src/codegen/reductions.rs b/third_party/rust/packed_simd/src/codegen/reductions.rs new file mode 100644 index 000000000000..302ca6d88f33 --- /dev/null +++ b/third_party/rust/packed_simd/src/codegen/reductions.rs @@ -0,0 +1 @@ +pub(crate) mod mask; diff --git a/third_party/rust/packed_simd/src/codegen/reductions/mask.rs b/third_party/rust/packed_simd/src/codegen/reductions/mask.rs new file mode 100644 index 000000000000..a78bcc563267 --- /dev/null +++ b/third_party/rust/packed_simd/src/codegen/reductions/mask.rs @@ -0,0 +1,69 @@ +//! Code generation workaround for `all()` mask horizontal reduction. +//! +//! Works around [LLVM bug 36702]. +//! +//! [LLVM bug 36702]: https://bugs.llvm.org/show_bug.cgi?id=36702 +#![allow(unused_macros)] + +use crate::*; + +pub(crate) trait All: crate::marker::Sized { + unsafe fn all(self) -> bool; +} + +pub(crate) trait Any: crate::marker::Sized { + unsafe fn any(self) -> bool; +} + +#[macro_use] +mod fallback_impl; + +cfg_if! { + if #[cfg(any(target_arch = "x86", target_arch = "x86_64"))] { + #[macro_use] + mod x86; + } else if #[cfg(all(target_arch = "arm", target_feature = "v7", + target_feature = "neon", + any(feature = "core_arch", libcore_neon)))] { + #[macro_use] + mod arm; + } else if #[cfg(all(target_arch = "aarch64", target_feature = "neon"))] { + #[macro_use] + mod aarch64; + } else { + #[macro_use] + mod fallback; + } +} + +impl_mask_reductions!(m8x2); +impl_mask_reductions!(m8x4); +impl_mask_reductions!(m8x8); +impl_mask_reductions!(m8x16); +impl_mask_reductions!(m8x32); +impl_mask_reductions!(m8x64); + +impl_mask_reductions!(m16x2); +impl_mask_reductions!(m16x4); +impl_mask_reductions!(m16x8); +impl_mask_reductions!(m16x16); +impl_mask_reductions!(m16x32); + +impl_mask_reductions!(m32x2); +impl_mask_reductions!(m32x4); +impl_mask_reductions!(m32x8); +impl_mask_reductions!(m32x16); + +// FIXME: 64-bit single element vector +// impl_mask_reductions!(m64x1); +impl_mask_reductions!(m64x2); +impl_mask_reductions!(m64x4); +impl_mask_reductions!(m64x8); + +impl_mask_reductions!(m128x1); +impl_mask_reductions!(m128x2); +impl_mask_reductions!(m128x4); + +impl_mask_reductions!(msizex2); +impl_mask_reductions!(msizex4); +impl_mask_reductions!(msizex8); diff --git a/third_party/rust/packed_simd/src/codegen/reductions/mask/aarch64.rs b/third_party/rust/packed_simd/src/codegen/reductions/mask/aarch64.rs new file mode 100644 index 000000000000..b2db52c891c1 --- /dev/null +++ b/third_party/rust/packed_simd/src/codegen/reductions/mask/aarch64.rs @@ -0,0 +1,81 @@ +//! Mask reductions implementation for `aarch64` targets + +/// 128-bit wide vectors +macro_rules! aarch64_128_neon_impl { + ($id:ident, $vmin:ident, $vmax:ident) => { + impl All for $id { + #[inline] + #[target_feature(enable = "neon")] + unsafe fn all(self) -> bool { + use crate::arch::aarch64::$vmin; + $vmin(crate::mem::transmute(self)) != 0 + } + } + impl Any for $id { + #[inline] + #[target_feature(enable = "neon")] + unsafe fn any(self) -> bool { + use crate::arch::aarch64::$vmax; + $vmax(crate::mem::transmute(self)) != 0 + } + } + }; +} + +/// 64-bit wide vectors +macro_rules! aarch64_64_neon_impl { + ($id:ident, $vec128:ident) => { + impl All for $id { + #[inline] + #[target_feature(enable = "neon")] + unsafe fn all(self) -> bool { + // Duplicates the 64-bit vector into a 128-bit one and + // calls all on that. + union U { + halves: ($id, $id), + vec: $vec128, + } + U { halves: (self, self) }.vec.all() + } + } + impl Any for $id { + #[inline] + #[target_feature(enable = "neon")] + unsafe fn any(self) -> bool { + union U { + halves: ($id, $id), + vec: $vec128, + } + U { halves: (self, self) }.vec.any() + } + } + }; +} + +/// Mask reduction implementation for `aarch64` targets +macro_rules! impl_mask_reductions { + // 64-bit wide masks + (m8x8) => { + aarch64_64_neon_impl!(m8x8, m8x16); + }; + (m16x4) => { + aarch64_64_neon_impl!(m16x4, m16x8); + }; + (m32x2) => { + aarch64_64_neon_impl!(m32x2, m32x4); + }; + // 128-bit wide masks + (m8x16) => { + aarch64_128_neon_impl!(m8x16, vminvq_u8, vmaxvq_u8); + }; + (m16x8) => { + aarch64_128_neon_impl!(m16x8, vminvq_u16, vmaxvq_u16); + }; + (m32x4) => { + aarch64_128_neon_impl!(m32x4, vminvq_u32, vmaxvq_u32); + }; + // Fallback to LLVM's default code-generation: + ($id:ident) => { + fallback_impl!($id); + }; +} diff --git a/third_party/rust/packed_simd/src/codegen/reductions/mask/arm.rs b/third_party/rust/packed_simd/src/codegen/reductions/mask/arm.rs new file mode 100644 index 000000000000..41c3cbc58afa --- /dev/null +++ b/third_party/rust/packed_simd/src/codegen/reductions/mask/arm.rs @@ -0,0 +1,56 @@ +//! Mask reductions implementation for `arm` targets + +/// Implementation for ARM + v7 + NEON for 64-bit or 128-bit wide vectors with +/// more than two elements. +macro_rules! arm_128_v7_neon_impl { + ($id:ident, $half:ident, $vpmin:ident, $vpmax:ident) => { + impl All for $id { + #[inline] + #[target_feature(enable = "v7,neon")] + unsafe fn all(self) -> bool { + use crate::arch::arm::$vpmin; + use crate::mem::transmute; + union U { + halves: ($half, $half), + vec: $id, + } + let halves = U { vec: self }.halves; + let h: $half = transmute($vpmin(transmute(halves.0), transmute(halves.1))); + h.all() + } + } + impl Any for $id { + #[inline] + #[target_feature(enable = "v7,neon")] + unsafe fn any(self) -> bool { + use crate::arch::arm::$vpmax; + use crate::mem::transmute; + union U { + halves: ($half, $half), + vec: $id, + } + let halves = U { vec: self }.halves; + let h: $half = transmute($vpmax(transmute(halves.0), transmute(halves.1))); + h.any() + } + } + }; +} + +/// Mask reduction implementation for `arm` targets +macro_rules! impl_mask_reductions { + // 128-bit wide masks + (m8x16) => { + arm_128_v7_neon_impl!(m8x16, m8x8, vpmin_u8, vpmax_u8); + }; + (m16x8) => { + arm_128_v7_neon_impl!(m16x8, m16x4, vpmin_u16, vpmax_u16); + }; + (m32x4) => { + arm_128_v7_neon_impl!(m32x4, m32x2, vpmin_u32, vpmax_u32); + }; + // Fallback to LLVM's default code-generation: + ($id:ident) => { + fallback_impl!($id); + }; +} diff --git a/third_party/rust/packed_simd/src/codegen/reductions/mask/fallback.rs b/third_party/rust/packed_simd/src/codegen/reductions/mask/fallback.rs new file mode 100644 index 000000000000..4c377a68784e --- /dev/null +++ b/third_party/rust/packed_simd/src/codegen/reductions/mask/fallback.rs @@ -0,0 +1,8 @@ +//! Default mask reduction implementations. + +/// Default mask reduction implementation +macro_rules! impl_mask_reductions { + ($id:ident) => { + fallback_impl!($id); + }; +} diff --git a/third_party/rust/packed_simd/src/codegen/reductions/mask/fallback_impl.rs b/third_party/rust/packed_simd/src/codegen/reductions/mask/fallback_impl.rs new file mode 100644 index 000000000000..0d246e2fdab6 --- /dev/null +++ b/third_party/rust/packed_simd/src/codegen/reductions/mask/fallback_impl.rs @@ -0,0 +1,237 @@ +//! Default implementation of a mask reduction for any target. + +macro_rules! fallback_to_other_impl { + ($id:ident, $other:ident) => { + impl All for $id { + #[inline] + unsafe fn all(self) -> bool { + let m: $other = crate::mem::transmute(self); + m.all() + } + } + impl Any for $id { + #[inline] + unsafe fn any(self) -> bool { + let m: $other = crate::mem::transmute(self); + m.any() + } + } + }; +} + +/// Fallback implementation. +macro_rules! fallback_impl { + // 16-bit wide masks: + (m8x2) => { + impl All for m8x2 { + #[inline] + unsafe fn all(self) -> bool { + let i: u16 = crate::mem::transmute(self); + i == u16::max_value() + } + } + impl Any for m8x2 { + #[inline] + unsafe fn any(self) -> bool { + let i: u16 = crate::mem::transmute(self); + i != 0 + } + } + }; + // 32-bit wide masks + (m8x4) => { + impl All for m8x4 { + #[inline] + unsafe fn all(self) -> bool { + let i: u32 = crate::mem::transmute(self); + i == u32::max_value() + } + } + impl Any for m8x4 { + #[inline] + unsafe fn any(self) -> bool { + let i: u32 = crate::mem::transmute(self); + i != 0 + } + } + }; + (m16x2) => { + fallback_to_other_impl!(m16x2, m8x4); + }; + // 64-bit wide masks: + (m8x8) => { + impl All for m8x8 { + #[inline] + unsafe fn all(self) -> bool { + let i: u64 = crate::mem::transmute(self); + i == u64::max_value() + } + } + impl Any for m8x8 { + #[inline] + unsafe fn any(self) -> bool { + let i: u64 = crate::mem::transmute(self); + i != 0 + } + } + }; + (m16x4) => { + fallback_to_other_impl!(m16x4, m8x8); + }; + (m32x2) => { + fallback_to_other_impl!(m32x2, m16x4); + }; + // FIXME: 64x1 maxk + // 128-bit wide masks: + (m8x16) => { + impl All for m8x16 { + #[inline] + unsafe fn all(self) -> bool { + let i: u128 = crate::mem::transmute(self); + i == u128::max_value() + } + } + impl Any for m8x16 { + #[inline] + unsafe fn any(self) -> bool { + let i: u128 = crate::mem::transmute(self); + i != 0 + } + } + }; + (m16x8) => { + fallback_to_other_impl!(m16x8, m8x16); + }; + (m32x4) => { + fallback_to_other_impl!(m32x4, m16x8); + }; + (m64x2) => { + fallback_to_other_impl!(m64x2, m32x4); + }; + (m128x1) => { + fallback_to_other_impl!(m128x1, m64x2); + }; + // 256-bit wide masks + (m8x32) => { + impl All for m8x32 { + #[inline] + unsafe fn all(self) -> bool { + let i: [u128; 2] = crate::mem::transmute(self); + let o: [u128; 2] = [u128::max_value(); 2]; + i == o + } + } + impl Any for m8x32 { + #[inline] + unsafe fn any(self) -> bool { + let i: [u128; 2] = crate::mem::transmute(self); + let o: [u128; 2] = [0; 2]; + i != o + } + } + }; + (m16x16) => { + fallback_to_other_impl!(m16x16, m8x32); + }; + (m32x8) => { + fallback_to_other_impl!(m32x8, m16x16); + }; + (m64x4) => { + fallback_to_other_impl!(m64x4, m32x8); + }; + (m128x2) => { + fallback_to_other_impl!(m128x2, m64x4); + }; + // 512-bit wide masks + (m8x64) => { + impl All for m8x64 { + #[inline] + unsafe fn all(self) -> bool { + let i: [u128; 4] = crate::mem::transmute(self); + let o: [u128; 4] = [u128::max_value(); 4]; + i == o + } + } + impl Any for m8x64 { + #[inline] + unsafe fn any(self) -> bool { + let i: [u128; 4] = crate::mem::transmute(self); + let o: [u128; 4] = [0; 4]; + i != o + } + } + }; + (m16x32) => { + fallback_to_other_impl!(m16x32, m8x64); + }; + (m32x16) => { + fallback_to_other_impl!(m32x16, m16x32); + }; + (m64x8) => { + fallback_to_other_impl!(m64x8, m32x16); + }; + (m128x4) => { + fallback_to_other_impl!(m128x4, m64x8); + }; + // Masks with pointer-sized elements64 + (msizex2) => { + cfg_if! { + if #[cfg(target_pointer_width = "64")] { + fallback_to_other_impl!(msizex2, m64x2); + } else if #[cfg(target_pointer_width = "32")] { + fallback_to_other_impl!(msizex2, m32x2); + } else { + compile_error!("unsupported target_pointer_width"); + } + } + }; + (msizex4) => { + cfg_if! { + if #[cfg(target_pointer_width = "64")] { + fallback_to_other_impl!(msizex4, m64x4); + } else if #[cfg(target_pointer_width = "32")] { + fallback_to_other_impl!(msizex4, m32x4); + } else { + compile_error!("unsupported target_pointer_width"); + } + } + }; + (msizex8) => { + cfg_if! { + if #[cfg(target_pointer_width = "64")] { + fallback_to_other_impl!(msizex8, m64x8); + } else if #[cfg(target_pointer_width = "32")] { + fallback_to_other_impl!(msizex8, m32x8); + } else { + compile_error!("unsupported target_pointer_width"); + } + } + }; +} + +macro_rules! recurse_half { + ($vid:ident, $vid_h:ident) => { + impl All for $vid { + #[inline] + unsafe fn all(self) -> bool { + union U { + halves: ($vid_h, $vid_h), + vec: $vid, + } + let halves = U { vec: self }.halves; + halves.0.all() && halves.1.all() + } + } + impl Any for $vid { + #[inline] + unsafe fn any(self) -> bool { + union U { + halves: ($vid_h, $vid_h), + vec: $vid, + } + let halves = U { vec: self }.halves; + halves.0.any() || halves.1.any() + } + } + }; +} diff --git a/third_party/rust/packed_simd/src/codegen/reductions/mask/x86.rs b/third_party/rust/packed_simd/src/codegen/reductions/mask/x86.rs new file mode 100644 index 000000000000..4bf509806545 --- /dev/null +++ b/third_party/rust/packed_simd/src/codegen/reductions/mask/x86.rs @@ -0,0 +1,216 @@ +//! Mask reductions implementation for `x86` and `x86_64` targets + +#[cfg(target_feature = "sse")] +#[macro_use] +mod sse; + +#[cfg(target_feature = "sse2")] +#[macro_use] +mod sse2; + +#[cfg(target_feature = "avx")] +#[macro_use] +mod avx; + +#[cfg(target_feature = "avx2")] +#[macro_use] +mod avx2; + +/// x86 64-bit m8x8 implementation +macro_rules! x86_m8x8_impl { + ($id:ident) => { + fallback_impl!($id); + }; +} + +/// x86 128-bit m8x16 implementation +macro_rules! x86_m8x16_impl { + ($id:ident) => { + cfg_if! { + if #[cfg(target_feature = "sse2")] { + x86_m8x16_sse2_impl!($id); + } else { + fallback_impl!($id); + } + } + }; +} + +/// x86 128-bit m32x4 implementation +macro_rules! x86_m32x4_impl { + ($id:ident) => { + cfg_if! { + if #[cfg(target_feature = "sse")] { + x86_m32x4_sse_impl!($id); + } else { + fallback_impl!($id); + } + } + }; +} + +/// x86 128-bit m64x2 implementation +macro_rules! x86_m64x2_impl { + ($id:ident) => { + cfg_if! { + if #[cfg(target_feature = "sse2")] { + x86_m64x2_sse2_impl!($id); + } else if #[cfg(target_feature = "sse")] { + x86_m32x4_sse_impl!($id); + } else { + fallback_impl!($id); + } + } + }; +} + +/// x86 256-bit m8x32 implementation +macro_rules! x86_m8x32_impl { + ($id:ident, $half_id:ident) => { + cfg_if! { + if #[cfg(target_feature = "avx2")] { + x86_m8x32_avx2_impl!($id); + } else if #[cfg(target_feature = "avx")] { + x86_m8x32_avx_impl!($id); + } else if #[cfg(target_feature = "sse2")] { + recurse_half!($id, $half_id); + } else { + fallback_impl!($id); + } + } + }; +} + +/// x86 256-bit m32x8 implementation +macro_rules! x86_m32x8_impl { + ($id:ident, $half_id:ident) => { + cfg_if! { + if #[cfg(target_feature = "avx")] { + x86_m32x8_avx_impl!($id); + } else if #[cfg(target_feature = "sse")] { + recurse_half!($id, $half_id); + } else { + fallback_impl!($id); + } + } + }; +} + +/// x86 256-bit m64x4 implementation +macro_rules! x86_m64x4_impl { + ($id:ident, $half_id:ident) => { + cfg_if! { + if #[cfg(target_feature = "avx")] { + x86_m64x4_avx_impl!($id); + } else if #[cfg(target_feature = "sse")] { + recurse_half!($id, $half_id); + } else { + fallback_impl!($id); + } + } + }; +} + +/// Fallback implementation. +macro_rules! x86_intr_impl { + ($id:ident) => { + impl All for $id { + #[inline] + unsafe fn all(self) -> bool { + use crate::llvm::simd_reduce_all; + simd_reduce_all(self.0) + } + } + impl Any for $id { + #[inline] + unsafe fn any(self) -> bool { + use crate::llvm::simd_reduce_any; + simd_reduce_any(self.0) + } + } + }; +} + +/// Mask reduction implementation for `x86` and `x86_64` targets +macro_rules! impl_mask_reductions { + // 64-bit wide masks + (m8x8) => { + x86_m8x8_impl!(m8x8); + }; + (m16x4) => { + x86_m8x8_impl!(m16x4); + }; + (m32x2) => { + x86_m8x8_impl!(m32x2); + }; + // 128-bit wide masks + (m8x16) => { + x86_m8x16_impl!(m8x16); + }; + (m16x8) => { + x86_m8x16_impl!(m16x8); + }; + (m32x4) => { + x86_m32x4_impl!(m32x4); + }; + (m64x2) => { + x86_m64x2_impl!(m64x2); + }; + (m128x1) => { + x86_intr_impl!(m128x1); + }; + // 256-bit wide masks: + (m8x32) => { + x86_m8x32_impl!(m8x32, m8x16); + }; + (m16x16) => { + x86_m8x32_impl!(m16x16, m16x8); + }; + (m32x8) => { + x86_m32x8_impl!(m32x8, m32x4); + }; + (m64x4) => { + x86_m64x4_impl!(m64x4, m64x2); + }; + (m128x2) => { + x86_intr_impl!(m128x2); + }; + (msizex2) => { + cfg_if! { + if #[cfg(target_pointer_width = "64")] { + fallback_to_other_impl!(msizex2, m64x2); + } else if #[cfg(target_pointer_width = "32")] { + fallback_to_other_impl!(msizex2, m32x2); + } else { + compile_error!("unsupported target_pointer_width"); + } + } + }; + (msizex4) => { + cfg_if! { + if #[cfg(target_pointer_width = "64")] { + fallback_to_other_impl!(msizex4, m64x4); + } else if #[cfg(target_pointer_width = "32")] { + fallback_to_other_impl!(msizex4, m32x4); + } else { + compile_error!("unsupported target_pointer_width"); + } + } + }; + (msizex8) => { + cfg_if! { + if #[cfg(target_pointer_width = "64")] { + fallback_to_other_impl!(msizex8, m64x8); + } else if #[cfg(target_pointer_width = "32")] { + fallback_to_other_impl!(msizex8, m32x8); + } else { + compile_error!("unsupported target_pointer_width"); + } + } + }; + + // Fallback to LLVM's default code-generation: + ($id:ident) => { + fallback_impl!($id); + }; +} diff --git a/third_party/rust/packed_simd/src/codegen/reductions/mask/x86/avx.rs b/third_party/rust/packed_simd/src/codegen/reductions/mask/x86/avx.rs new file mode 100644 index 000000000000..61f352d22839 --- /dev/null +++ b/third_party/rust/packed_simd/src/codegen/reductions/mask/x86/avx.rs @@ -0,0 +1,95 @@ +//! Mask reductions implementation for `x86` and `x86_64` targets with `AVX` + +/// `x86`/`x86_64` 256-bit `AVX` implementation +/// FIXME: it might be faster here to do two `_mm_movmask_epi8` +#[cfg(target_feature = "avx")] +macro_rules! x86_m8x32_avx_impl { + ($id:ident) => { + impl All for $id { + #[inline] + #[target_feature(enable = "avx")] + unsafe fn all(self) -> bool { + #[cfg(target_arch = "x86")] + use crate::arch::x86::_mm256_testc_si256; + #[cfg(target_arch = "x86_64")] + use crate::arch::x86_64::_mm256_testc_si256; + _mm256_testc_si256(crate::mem::transmute(self), crate::mem::transmute($id::splat(true))) != 0 + } + } + impl Any for $id { + #[inline] + #[target_feature(enable = "avx")] + unsafe fn any(self) -> bool { + #[cfg(target_arch = "x86")] + use crate::arch::x86::_mm256_testz_si256; + #[cfg(target_arch = "x86_64")] + use crate::arch::x86_64::_mm256_testz_si256; + _mm256_testz_si256(crate::mem::transmute(self), crate::mem::transmute(self)) == 0 + } + } + }; +} + +/// `x86`/`x86_64` 256-bit m32x8 `AVX` implementation +macro_rules! x86_m32x8_avx_impl { + ($id:ident) => { + impl All for $id { + #[inline] + #[target_feature(enable = "sse")] + unsafe fn all(self) -> bool { + #[cfg(target_arch = "x86")] + use crate::arch::x86::_mm256_movemask_ps; + #[cfg(target_arch = "x86_64")] + use crate::arch::x86_64::_mm256_movemask_ps; + // _mm256_movemask_ps(a) creates a 8bit mask containing the + // most significant bit of each lane of `a`. If all bits are + // set, then all 8 lanes of the mask are true. + _mm256_movemask_ps(crate::mem::transmute(self)) == 0b_1111_1111_i32 + } + } + impl Any for $id { + #[inline] + #[target_feature(enable = "sse")] + unsafe fn any(self) -> bool { + #[cfg(target_arch = "x86")] + use crate::arch::x86::_mm256_movemask_ps; + #[cfg(target_arch = "x86_64")] + use crate::arch::x86_64::_mm256_movemask_ps; + + _mm256_movemask_ps(crate::mem::transmute(self)) != 0 + } + } + }; +} + +/// `x86`/`x86_64` 256-bit m64x4 `AVX` implementation +macro_rules! x86_m64x4_avx_impl { + ($id:ident) => { + impl All for $id { + #[inline] + #[target_feature(enable = "sse")] + unsafe fn all(self) -> bool { + #[cfg(target_arch = "x86")] + use crate::arch::x86::_mm256_movemask_pd; + #[cfg(target_arch = "x86_64")] + use crate::arch::x86_64::_mm256_movemask_pd; + // _mm256_movemask_pd(a) creates a 4bit mask containing the + // most significant bit of each lane of `a`. If all bits are + // set, then all 4 lanes of the mask are true. + _mm256_movemask_pd(crate::mem::transmute(self)) == 0b_1111_i32 + } + } + impl Any for $id { + #[inline] + #[target_feature(enable = "sse")] + unsafe fn any(self) -> bool { + #[cfg(target_arch = "x86")] + use crate::arch::x86::_mm256_movemask_pd; + #[cfg(target_arch = "x86_64")] + use crate::arch::x86_64::_mm256_movemask_pd; + + _mm256_movemask_pd(crate::mem::transmute(self)) != 0 + } + } + }; +} diff --git a/third_party/rust/packed_simd/src/codegen/reductions/mask/x86/avx2.rs b/third_party/rust/packed_simd/src/codegen/reductions/mask/x86/avx2.rs new file mode 100644 index 000000000000..d37d02342092 --- /dev/null +++ b/third_party/rust/packed_simd/src/codegen/reductions/mask/x86/avx2.rs @@ -0,0 +1,35 @@ +//! Mask reductions implementation for `x86` and `x86_64` targets with `AVX2`. +#![allow(unused)] + +/// x86/x86_64 256-bit m8x32 AVX2 implementation +macro_rules! x86_m8x32_avx2_impl { + ($id:ident) => { + impl All for $id { + #[inline] + #[target_feature(enable = "sse2")] + unsafe fn all(self) -> bool { + #[cfg(target_arch = "x86")] + use crate::arch::x86::_mm256_movemask_epi8; + #[cfg(target_arch = "x86_64")] + use crate::arch::x86_64::_mm256_movemask_epi8; + // _mm256_movemask_epi8(a) creates a 32bit mask containing the + // most significant bit of each byte of `a`. If all + // bits are set, then all 32 lanes of the mask are + // true. + _mm256_movemask_epi8(crate::mem::transmute(self)) == -1_i32 + } + } + impl Any for $id { + #[inline] + #[target_feature(enable = "sse2")] + unsafe fn any(self) -> bool { + #[cfg(target_arch = "x86")] + use crate::arch::x86::_mm256_movemask_epi8; + #[cfg(target_arch = "x86_64")] + use crate::arch::x86_64::_mm256_movemask_epi8; + + _mm256_movemask_epi8(crate::mem::transmute(self)) != 0 + } + } + }; +} diff --git a/third_party/rust/packed_simd/src/codegen/reductions/mask/x86/sse.rs b/third_party/rust/packed_simd/src/codegen/reductions/mask/x86/sse.rs new file mode 100644 index 000000000000..e0c9aee92bd9 --- /dev/null +++ b/third_party/rust/packed_simd/src/codegen/reductions/mask/x86/sse.rs @@ -0,0 +1,35 @@ +//! Mask reductions implementation for `x86` and `x86_64` targets with `SSE`. +#![allow(unused)] + +/// `x86`/`x86_64` 128-bit `m32x4` `SSE` implementation +macro_rules! x86_m32x4_sse_impl { + ($id:ident) => { + impl All for $id { + #[inline] + #[target_feature(enable = "sse")] + unsafe fn all(self) -> bool { + #[cfg(target_arch = "x86")] + use crate::arch::x86::_mm_movemask_ps; + #[cfg(target_arch = "x86_64")] + use crate::arch::x86_64::_mm_movemask_ps; + // _mm_movemask_ps(a) creates a 4bit mask containing the + // most significant bit of each lane of `a`. If all + // bits are set, then all 4 lanes of the mask are + // true. + _mm_movemask_ps(crate::mem::transmute(self)) == 0b_1111_i32 + } + } + impl Any for $id { + #[inline] + #[target_feature(enable = "sse")] + unsafe fn any(self) -> bool { + #[cfg(target_arch = "x86")] + use crate::arch::x86::_mm_movemask_ps; + #[cfg(target_arch = "x86_64")] + use crate::arch::x86_64::_mm_movemask_ps; + + _mm_movemask_ps(crate::mem::transmute(self)) != 0 + } + } + }; +} diff --git a/third_party/rust/packed_simd/src/codegen/reductions/mask/x86/sse2.rs b/third_party/rust/packed_simd/src/codegen/reductions/mask/x86/sse2.rs new file mode 100644 index 000000000000..bbb52fa47e02 --- /dev/null +++ b/third_party/rust/packed_simd/src/codegen/reductions/mask/x86/sse2.rs @@ -0,0 +1,68 @@ +//! Mask reductions implementation for `x86` and `x86_64` targets with `SSE2`. +#![allow(unused)] + +/// `x86`/`x86_64` 128-bit m64x2 `SSE2` implementation +macro_rules! x86_m64x2_sse2_impl { + ($id:ident) => { + impl All for $id { + #[inline] + #[target_feature(enable = "sse")] + unsafe fn all(self) -> bool { + #[cfg(target_arch = "x86")] + use crate::arch::x86::_mm_movemask_pd; + #[cfg(target_arch = "x86_64")] + use crate::arch::x86_64::_mm_movemask_pd; + // _mm_movemask_pd(a) creates a 2bit mask containing the + // most significant bit of each lane of `a`. If all + // bits are set, then all 2 lanes of the mask are + // true. + _mm_movemask_pd(crate::mem::transmute(self)) == 0b_11_i32 + } + } + impl Any for $id { + #[inline] + #[target_feature(enable = "sse")] + unsafe fn any(self) -> bool { + #[cfg(target_arch = "x86")] + use crate::arch::x86::_mm_movemask_pd; + #[cfg(target_arch = "x86_64")] + use crate::arch::x86_64::_mm_movemask_pd; + + _mm_movemask_pd(crate::mem::transmute(self)) != 0 + } + } + }; +} + +/// `x86`/`x86_64` 128-bit m8x16 `SSE2` implementation +macro_rules! x86_m8x16_sse2_impl { + ($id:ident) => { + impl All for $id { + #[inline] + #[target_feature(enable = "sse2")] + unsafe fn all(self) -> bool { + #[cfg(target_arch = "x86")] + use crate::arch::x86::_mm_movemask_epi8; + #[cfg(target_arch = "x86_64")] + use crate::arch::x86_64::_mm_movemask_epi8; + // _mm_movemask_epi8(a) creates a 16bit mask containing the + // most significant bit of each byte of `a`. If all + // bits are set, then all 16 lanes of the mask are + // true. + _mm_movemask_epi8(crate::mem::transmute(self)) == i32::from(u16::max_value()) + } + } + impl Any for $id { + #[inline] + #[target_feature(enable = "sse2")] + unsafe fn any(self) -> bool { + #[cfg(target_arch = "x86")] + use crate::arch::x86::_mm_movemask_epi8; + #[cfg(target_arch = "x86_64")] + use crate::arch::x86_64::_mm_movemask_epi8; + + _mm_movemask_epi8(crate::mem::transmute(self)) != 0 + } + } + }; +} diff --git a/third_party/rust/packed_simd/src/codegen/shuffle.rs b/third_party/rust/packed_simd/src/codegen/shuffle.rs new file mode 100644 index 000000000000..d3acd48f5bed --- /dev/null +++ b/third_party/rust/packed_simd/src/codegen/shuffle.rs @@ -0,0 +1,150 @@ +//! Implementations of the `ShuffleResult` trait for the different numbers of +//! lanes and vector element types. + +use crate::masks::*; +use crate::sealed::{Seal, Shuffle}; + +macro_rules! impl_shuffle { + ($array:ty, $base:ty, $out:ty) => { + impl Seal<$array> for $base {} + impl Shuffle<$array> for $base { + type Output = $out; + } + }; +} + +impl_shuffle! { [u32; 2], i8, crate::codegen::i8x2 } +impl_shuffle! { [u32; 4], i8, crate::codegen::i8x4 } +impl_shuffle! { [u32; 8], i8, crate::codegen::i8x8 } +impl_shuffle! { [u32; 16], i8, crate::codegen::i8x16 } +impl_shuffle! { [u32; 32], i8, crate::codegen::i8x32 } +impl_shuffle! { [u32; 64], i8, crate::codegen::i8x64 } + +impl_shuffle! { [u32; 2], u8, crate::codegen::u8x2 } +impl_shuffle! { [u32; 4], u8, crate::codegen::u8x4 } +impl_shuffle! { [u32; 8], u8, crate::codegen::u8x8 } +impl_shuffle! { [u32; 16], u8, crate::codegen::u8x16 } +impl_shuffle! { [u32; 32], u8, crate::codegen::u8x32 } +impl_shuffle! { [u32; 64], u8, crate::codegen::u8x64 } + +impl_shuffle! { [u32; 2], m8, crate::codegen::m8x2 } +impl_shuffle! { [u32; 4], m8, crate::codegen::m8x4 } +impl_shuffle! { [u32; 8], m8, crate::codegen::m8x8 } +impl_shuffle! { [u32; 16], m8, crate::codegen::m8x16 } +impl_shuffle! { [u32; 32], m8, crate::codegen::m8x32 } +impl_shuffle! { [u32; 64], m8, crate::codegen::m8x64 } + +impl_shuffle! { [u32; 2], i16, crate::codegen::i16x2 } +impl_shuffle! { [u32; 4], i16, crate::codegen::i16x4 } +impl_shuffle! { [u32; 8], i16, crate::codegen::i16x8 } +impl_shuffle! { [u32; 16], i16, crate::codegen::i16x16 } +impl_shuffle! { [u32; 32], i16, crate::codegen::i16x32 } + +impl_shuffle! { [u32; 2], u16, crate::codegen::u16x2 } +impl_shuffle! { [u32; 4], u16, crate::codegen::u16x4 } +impl_shuffle! { [u32; 8], u16, crate::codegen::u16x8 } +impl_shuffle! { [u32; 16], u16, crate::codegen::u16x16 } +impl_shuffle! { [u32; 32], u16, crate::codegen::u16x32 } + +impl_shuffle! { [u32; 2], m16, crate::codegen::m16x2 } +impl_shuffle! { [u32; 4], m16, crate::codegen::m16x4 } +impl_shuffle! { [u32; 8], m16, crate::codegen::m16x8 } +impl_shuffle! { [u32; 16], m16, crate::codegen::m16x16 } + +impl_shuffle! { [u32; 2], i32, crate::codegen::i32x2 } +impl_shuffle! { [u32; 4], i32, crate::codegen::i32x4 } +impl_shuffle! { [u32; 8], i32, crate::codegen::i32x8 } +impl_shuffle! { [u32; 16], i32, crate::codegen::i32x16 } + +impl_shuffle! { [u32; 2], u32, crate::codegen::u32x2 } +impl_shuffle! { [u32; 4], u32, crate::codegen::u32x4 } +impl_shuffle! { [u32; 8], u32, crate::codegen::u32x8 } +impl_shuffle! { [u32; 16], u32, crate::codegen::u32x16 } + +impl_shuffle! { [u32; 2], f32, crate::codegen::f32x2 } +impl_shuffle! { [u32; 4], f32, crate::codegen::f32x4 } +impl_shuffle! { [u32; 8], f32, crate::codegen::f32x8 } +impl_shuffle! { [u32; 16], f32, crate::codegen::f32x16 } + +impl_shuffle! { [u32; 2], m32, crate::codegen::m32x2 } +impl_shuffle! { [u32; 4], m32, crate::codegen::m32x4 } +impl_shuffle! { [u32; 8], m32, crate::codegen::m32x8 } +impl_shuffle! { [u32; 16], m32, crate::codegen::m32x16 } + +/* FIXME: 64-bit single element vector +impl_shuffle! { [u32; 1], i64, crate::codegen::i64x1 } +*/ +impl_shuffle! { [u32; 2], i64, crate::codegen::i64x2 } +impl_shuffle! { [u32; 4], i64, crate::codegen::i64x4 } +impl_shuffle! { [u32; 8], i64, crate::codegen::i64x8 } + +/* FIXME: 64-bit single element vector +impl_shuffle! { [u32; 1], i64, crate::codegen::i64x1 } +*/ +impl_shuffle! { [u32; 2], u64, crate::codegen::u64x2 } +impl_shuffle! { [u32; 4], u64, crate::codegen::u64x4 } +impl_shuffle! { [u32; 8], u64, crate::codegen::u64x8 } + +/* FIXME: 64-bit single element vector +impl_shuffle! { [u32; 1], i64, crate::codegen::i64x1 } +*/ +impl_shuffle! { [u32; 2], f64, crate::codegen::f64x2 } +impl_shuffle! { [u32; 4], f64, crate::codegen::f64x4 } +impl_shuffle! { [u32; 8], f64, crate::codegen::f64x8 } + +/* FIXME: 64-bit single element vector +impl_shuffle! { [u32; 1], i64, crate::codegen::i64x1 } +*/ +impl_shuffle! { [u32; 2], m64, crate::codegen::m64x2 } +impl_shuffle! { [u32; 4], m64, crate::codegen::m64x4 } +impl_shuffle! { [u32; 8], m64, crate::codegen::m64x8 } + +impl_shuffle! { [u32; 2], isize, crate::codegen::isizex2 } +impl_shuffle! { [u32; 4], isize, crate::codegen::isizex4 } +impl_shuffle! { [u32; 8], isize, crate::codegen::isizex8 } + +impl_shuffle! { [u32; 2], usize, crate::codegen::usizex2 } +impl_shuffle! { [u32; 4], usize, crate::codegen::usizex4 } +impl_shuffle! { [u32; 8], usize, crate::codegen::usizex8 } + +impl_shuffle! { [u32; 2], msize, crate::codegen::msizex2 } +impl_shuffle! { [u32; 4], msize, crate::codegen::msizex4 } +impl_shuffle! { [u32; 8], msize, crate::codegen::msizex8 } + +impl Seal<[u32; 2]> for *const T {} +impl Shuffle<[u32; 2]> for *const T { + type Output = crate::codegen::cptrx2; +} +impl Seal<[u32; 4]> for *const T {} +impl Shuffle<[u32; 4]> for *const T { + type Output = crate::codegen::cptrx4; +} +impl Seal<[u32; 8]> for *const T {} +impl Shuffle<[u32; 8]> for *const T { + type Output = crate::codegen::cptrx8; +} + +impl Seal<[u32; 2]> for *mut T {} +impl Shuffle<[u32; 2]> for *mut T { + type Output = crate::codegen::mptrx2; +} +impl Seal<[u32; 4]> for *mut T {} +impl Shuffle<[u32; 4]> for *mut T { + type Output = crate::codegen::mptrx4; +} +impl Seal<[u32; 8]> for *mut T {} +impl Shuffle<[u32; 8]> for *mut T { + type Output = crate::codegen::mptrx8; +} + +impl_shuffle! { [u32; 1], i128, crate::codegen::i128x1 } +impl_shuffle! { [u32; 2], i128, crate::codegen::i128x2 } +impl_shuffle! { [u32; 4], i128, crate::codegen::i128x4 } + +impl_shuffle! { [u32; 1], u128, crate::codegen::u128x1 } +impl_shuffle! { [u32; 2], u128, crate::codegen::u128x2 } +impl_shuffle! { [u32; 4], u128, crate::codegen::u128x4 } + +impl_shuffle! { [u32; 1], m128, crate::codegen::m128x1 } +impl_shuffle! { [u32; 2], m128, crate::codegen::m128x2 } +impl_shuffle! { [u32; 4], m128, crate::codegen::m128x4 } diff --git a/third_party/rust/packed_simd/src/codegen/shuffle1_dyn.rs b/third_party/rust/packed_simd/src/codegen/shuffle1_dyn.rs new file mode 100644 index 000000000000..19d457a45bfc --- /dev/null +++ b/third_party/rust/packed_simd/src/codegen/shuffle1_dyn.rs @@ -0,0 +1,408 @@ +//! Shuffle vector lanes with run-time indices. + +use crate::*; + +pub trait Shuffle1Dyn { + type Indices; + fn shuffle1_dyn(self, _: Self::Indices) -> Self; +} + +// Fallback implementation +macro_rules! impl_fallback { + ($id:ident) => { + impl Shuffle1Dyn for $id { + type Indices = Self; + #[inline] + fn shuffle1_dyn(self, indices: Self::Indices) -> Self { + let mut result = Self::splat(0); + for i in 0..$id::lanes() { + result = result.replace(i, self.extract(indices.extract(i) as usize)); + } + result + } + } + }; +} + +macro_rules! impl_shuffle1_dyn { + (u8x8) => { + cfg_if! { + if #[cfg(all( + any( + all(target_arch = "aarch64", target_feature = "neon"), + all(target_arch = "doesnotexist", target_feature = "v7", + target_feature = "neon") + ), + any(feature = "core_arch", libcore_neon) + ) + )] { + impl Shuffle1Dyn for u8x8 { + type Indices = Self; + #[inline] + fn shuffle1_dyn(self, indices: Self::Indices) -> Self { + #[cfg(target_arch = "aarch64")] + use crate::arch::aarch64::vtbl1_u8; + #[cfg(target_arch = "doesnotexist")] + use crate::arch::arm::vtbl1_u8; + + // This is safe because the binary is compiled with + // neon enabled at compile-time and can therefore only + // run on CPUs that have it enabled. + unsafe { + Simd(mem::transmute( + vtbl1_u8(mem::transmute(self.0), + crate::mem::transmute(indices.0)) + )) + } + } + } + } else { + impl_fallback!(u8x8); + } + } + }; + (u8x16) => { + cfg_if! { + if #[cfg(all(any(target_arch = "x86", target_arch = "x86_64"), + target_feature = "ssse3"))] { + impl Shuffle1Dyn for u8x16 { + type Indices = Self; + #[inline] + fn shuffle1_dyn(self, indices: Self::Indices) -> Self { + #[cfg(target_arch = "x86")] + use crate::arch::x86::_mm_shuffle_epi8; + #[cfg(target_arch = "x86_64")] + use crate::arch::x86_64::_mm_shuffle_epi8; + // This is safe because the binary is compiled with + // ssse3 enabled at compile-time and can therefore only + // run on CPUs that have it enabled. + unsafe { + Simd(mem::transmute( + _mm_shuffle_epi8(mem::transmute(self.0), + crate::mem::transmute(indices)) + )) + } + } + } + } else if #[cfg(all(target_arch = "aarch64", target_feature = "neon", + any(feature = "core_arch", libcore_neon)))] { + impl Shuffle1Dyn for u8x16 { + type Indices = Self; + #[inline] + fn shuffle1_dyn(self, indices: Self::Indices) -> Self { + use crate::arch::aarch64::vqtbl1q_u8; + + // This is safe because the binary is compiled with + // neon enabled at compile-time and can therefore only + // run on CPUs that have it enabled. + unsafe { + Simd(mem::transmute( + vqtbl1q_u8(mem::transmute(self.0), + crate::mem::transmute(indices.0)) + )) + } + } + } + } else if #[cfg(all(target_arch = "doesnotexist", target_feature = "v7", + target_feature = "neon", + any(feature = "core_arch", libcore_neon)))] { + impl Shuffle1Dyn for u8x16 { + type Indices = Self; + #[inline] + fn shuffle1_dyn(self, indices: Self::Indices) -> Self { + use crate::arch::arm::vtbl2_u8; + + // This is safe because the binary is compiled with + // neon enabled at compile-time and can therefore only + // run on CPUs that have it enabled. + unsafe { + union U { + j: u8x16, + s: (u8x8, u8x8), + } + + let (i0, i1) = U { j: y }.s; + + let r0 = vtbl2_u8( + mem::transmute(x), + crate::mem::transmute(i0) + ); + let r1 = vtbl2_u8( + mem::transmute(x), + crate::mem::transmute(i1) + ); + + let r = U { s: (r0, r1) }.j; + + Simd(mem::transmute(r)) + } + } + } + } else { + impl_fallback!(u8x16); + } + } + }; + (u16x8) => { + impl Shuffle1Dyn for u16x8 { + type Indices = Self; + #[inline] + fn shuffle1_dyn(self, indices: Self::Indices) -> Self { + let indices: u8x8 = (indices * 2).cast(); + let indices: u8x16 = shuffle!(indices, [0, 0, 1, 1, 2, 2, 3, 3, 4, 4, 5, 5, 6, 6, 7, 7]); + let v = u8x16::new(0, 1, 0, 1, 0, 1, 0, 1, 0, 1, 0, 1, 0, 1, 0, 1); + let indices = indices + v; + unsafe { + let s: u8x16 = crate::mem::transmute(self); + crate::mem::transmute(s.shuffle1_dyn(indices)) + } + } + } + }; + (u32x4) => { + cfg_if! { + if #[cfg(all(any(target_arch = "x86", target_arch = "x86_64"), + target_feature = "avx"))] { + impl Shuffle1Dyn for u32x4 { + type Indices = Self; + #[inline] + fn shuffle1_dyn(self, indices: Self::Indices) -> Self { + #[cfg(target_arch = "x86")] + use crate::arch::x86::{_mm_permutevar_ps}; + #[cfg(target_arch = "x86_64")] + use crate::arch::x86_64::{_mm_permutevar_ps}; + + unsafe { + crate::mem::transmute( + _mm_permutevar_ps( + crate::mem::transmute(self.0), + crate::mem::transmute(indices.0) + ) + ) + } + } + } + } else { + impl Shuffle1Dyn for u32x4 { + type Indices = Self; + #[inline] + fn shuffle1_dyn(self, indices: Self::Indices) -> Self { + let indices: u8x4 = (indices * 4).cast(); + let indices: u8x16 = shuffle!( + indices, + [0, 0, 0, 0, 1, 1, 1, 1, 2, 2, 2, 2, 3, 3, 3, 3] + ); + let v = u8x16::new( + 0, 1, 2, 3, 0, 1, 2, 3, 0, 1, 2, 3, 0, 1, 2, 3 + ); + let indices = indices + v; + unsafe { + let s: u8x16 =crate::mem::transmute(self); + crate::mem::transmute(s.shuffle1_dyn(indices)) + } + } + } + } + } + }; + (u64x2) => { + cfg_if! { + if #[cfg(all(any(target_arch = "x86", target_arch = "x86_64"), + target_feature = "avx"))] { + impl Shuffle1Dyn for u64x2 { + type Indices = Self; + #[inline] + fn shuffle1_dyn(self, indices: Self::Indices) -> Self { + #[cfg(target_arch = "x86")] + use crate::arch::x86::{_mm_permutevar_pd}; + #[cfg(target_arch = "x86_64")] + use crate::arch::x86_64::{_mm_permutevar_pd}; + // _mm_permutevar_pd uses the _second_ bit of each + // element to perform the selection, that is: 0b00 => 0, + // 0b10 => 1: + let indices = indices << 1; + unsafe { + crate::mem::transmute( + _mm_permutevar_pd( + crate::mem::transmute(self), + crate::mem::transmute(indices) + ) + ) + } + } + } + } else { + impl Shuffle1Dyn for u64x2 { + type Indices = Self; + #[inline] + fn shuffle1_dyn(self, indices: Self::Indices) -> Self { + let indices: u8x2 = (indices * 8).cast(); + let indices: u8x16 = shuffle!( + indices, + [0, 0, 0, 0, 0, 0, 0, 0, 1, 1, 1, 1, 1, 1, 1, 1] + ); + let v = u8x16::new( + 0, 1, 2, 3, 4, 5, 6, 7, 0, 1, 2, 3, 4, 5, 6, 7 + ); + let indices = indices + v; + unsafe { + let s: u8x16 =crate::mem::transmute(self); + crate::mem::transmute(s.shuffle1_dyn(indices)) + } + } + } + } + } + }; + (u128x1) => { + impl Shuffle1Dyn for u128x1 { + type Indices = Self; + #[inline] + fn shuffle1_dyn(self, _indices: Self::Indices) -> Self { + self + } + } + }; + ($id:ident) => { + impl_fallback!($id); + }; +} + +impl_shuffle1_dyn!(u8x2); +impl_shuffle1_dyn!(u8x4); +impl_shuffle1_dyn!(u8x8); +impl_shuffle1_dyn!(u8x16); +impl_shuffle1_dyn!(u8x32); +impl_shuffle1_dyn!(u8x64); + +impl_shuffle1_dyn!(u16x2); +impl_shuffle1_dyn!(u16x4); +impl_shuffle1_dyn!(u16x8); +impl_shuffle1_dyn!(u16x16); +impl_shuffle1_dyn!(u16x32); + +impl_shuffle1_dyn!(u32x2); +impl_shuffle1_dyn!(u32x4); +impl_shuffle1_dyn!(u32x8); +impl_shuffle1_dyn!(u32x16); + +impl_shuffle1_dyn!(u64x2); +impl_shuffle1_dyn!(u64x4); +impl_shuffle1_dyn!(u64x8); + +impl_shuffle1_dyn!(usizex2); +impl_shuffle1_dyn!(usizex4); +impl_shuffle1_dyn!(usizex8); + +impl_shuffle1_dyn!(u128x1); +impl_shuffle1_dyn!(u128x2); +impl_shuffle1_dyn!(u128x4); + +// Implementation for non-unsigned vector types +macro_rules! impl_shuffle1_dyn_non_u { + ($id:ident, $uid:ident) => { + impl Shuffle1Dyn for $id { + type Indices = $uid; + #[inline] + fn shuffle1_dyn(self, indices: Self::Indices) -> Self { + unsafe { + let u: $uid = crate::mem::transmute(self); + crate::mem::transmute(u.shuffle1_dyn(indices)) + } + } + } + }; +} + +impl_shuffle1_dyn_non_u!(i8x2, u8x2); +impl_shuffle1_dyn_non_u!(i8x4, u8x4); +impl_shuffle1_dyn_non_u!(i8x8, u8x8); +impl_shuffle1_dyn_non_u!(i8x16, u8x16); +impl_shuffle1_dyn_non_u!(i8x32, u8x32); +impl_shuffle1_dyn_non_u!(i8x64, u8x64); + +impl_shuffle1_dyn_non_u!(i16x2, u16x2); +impl_shuffle1_dyn_non_u!(i16x4, u16x4); +impl_shuffle1_dyn_non_u!(i16x8, u16x8); +impl_shuffle1_dyn_non_u!(i16x16, u16x16); +impl_shuffle1_dyn_non_u!(i16x32, u16x32); + +impl_shuffle1_dyn_non_u!(i32x2, u32x2); +impl_shuffle1_dyn_non_u!(i32x4, u32x4); +impl_shuffle1_dyn_non_u!(i32x8, u32x8); +impl_shuffle1_dyn_non_u!(i32x16, u32x16); + +impl_shuffle1_dyn_non_u!(i64x2, u64x2); +impl_shuffle1_dyn_non_u!(i64x4, u64x4); +impl_shuffle1_dyn_non_u!(i64x8, u64x8); + +impl_shuffle1_dyn_non_u!(isizex2, usizex2); +impl_shuffle1_dyn_non_u!(isizex4, usizex4); +impl_shuffle1_dyn_non_u!(isizex8, usizex8); + +impl_shuffle1_dyn_non_u!(i128x1, u128x1); +impl_shuffle1_dyn_non_u!(i128x2, u128x2); +impl_shuffle1_dyn_non_u!(i128x4, u128x4); + +impl_shuffle1_dyn_non_u!(m8x2, u8x2); +impl_shuffle1_dyn_non_u!(m8x4, u8x4); +impl_shuffle1_dyn_non_u!(m8x8, u8x8); +impl_shuffle1_dyn_non_u!(m8x16, u8x16); +impl_shuffle1_dyn_non_u!(m8x32, u8x32); +impl_shuffle1_dyn_non_u!(m8x64, u8x64); + +impl_shuffle1_dyn_non_u!(m16x2, u16x2); +impl_shuffle1_dyn_non_u!(m16x4, u16x4); +impl_shuffle1_dyn_non_u!(m16x8, u16x8); +impl_shuffle1_dyn_non_u!(m16x16, u16x16); +impl_shuffle1_dyn_non_u!(m16x32, u16x32); + +impl_shuffle1_dyn_non_u!(m32x2, u32x2); +impl_shuffle1_dyn_non_u!(m32x4, u32x4); +impl_shuffle1_dyn_non_u!(m32x8, u32x8); +impl_shuffle1_dyn_non_u!(m32x16, u32x16); + +impl_shuffle1_dyn_non_u!(m64x2, u64x2); +impl_shuffle1_dyn_non_u!(m64x4, u64x4); +impl_shuffle1_dyn_non_u!(m64x8, u64x8); + +impl_shuffle1_dyn_non_u!(msizex2, usizex2); +impl_shuffle1_dyn_non_u!(msizex4, usizex4); +impl_shuffle1_dyn_non_u!(msizex8, usizex8); + +impl_shuffle1_dyn_non_u!(m128x1, u128x1); +impl_shuffle1_dyn_non_u!(m128x2, u128x2); +impl_shuffle1_dyn_non_u!(m128x4, u128x4); + +impl_shuffle1_dyn_non_u!(f32x2, u32x2); +impl_shuffle1_dyn_non_u!(f32x4, u32x4); +impl_shuffle1_dyn_non_u!(f32x8, u32x8); +impl_shuffle1_dyn_non_u!(f32x16, u32x16); + +impl_shuffle1_dyn_non_u!(f64x2, u64x2); +impl_shuffle1_dyn_non_u!(f64x4, u64x4); +impl_shuffle1_dyn_non_u!(f64x8, u64x8); + +// Implementation for non-unsigned vector types +macro_rules! impl_shuffle1_dyn_ptr { + ($id:ident, $uid:ident) => { + impl Shuffle1Dyn for $id { + type Indices = $uid; + #[inline] + fn shuffle1_dyn(self, indices: Self::Indices) -> Self { + unsafe { + let u: $uid = crate::mem::transmute(self); + crate::mem::transmute(u.shuffle1_dyn(indices)) + } + } + } + }; +} + +impl_shuffle1_dyn_ptr!(cptrx2, usizex2); +impl_shuffle1_dyn_ptr!(cptrx4, usizex4); +impl_shuffle1_dyn_ptr!(cptrx8, usizex8); + +impl_shuffle1_dyn_ptr!(mptrx2, usizex2); +impl_shuffle1_dyn_ptr!(mptrx4, usizex4); +impl_shuffle1_dyn_ptr!(mptrx8, usizex8); diff --git a/third_party/rust/packed_simd/src/codegen/swap_bytes.rs b/third_party/rust/packed_simd/src/codegen/swap_bytes.rs new file mode 100644 index 000000000000..9cf34a3e0401 --- /dev/null +++ b/third_party/rust/packed_simd/src/codegen/swap_bytes.rs @@ -0,0 +1,149 @@ +//! Horizontal swap bytes reductions. + +// FIXME: investigate using `llvm.bswap` +// https://github.com/rust-lang-nursery/packed_simd/issues/19 + +use crate::*; + +pub(crate) trait SwapBytes { + fn swap_bytes(self) -> Self; +} + +macro_rules! impl_swap_bytes { + (v16: $($id:ident,)+) => { + $( + impl SwapBytes for $id { + #[inline] + fn swap_bytes(self) -> Self { + shuffle!(self, [1, 0]) + } + } + )+ + }; + (v32: $($id:ident,)+) => { + $( + impl SwapBytes for $id { + #[inline] + #[allow(clippy::useless_transmute)] + fn swap_bytes(self) -> Self { + unsafe { + let bytes: u8x4 = crate::mem::transmute(self); + let result: u8x4 = shuffle!(bytes, [3, 2, 1, 0]); + crate::mem::transmute(result) + } + } + } + )+ + }; + (v64: $($id:ident,)+) => { + $( + impl SwapBytes for $id { + #[inline] + #[allow(clippy::useless_transmute)] + fn swap_bytes(self) -> Self { + unsafe { + let bytes: u8x8 = crate::mem::transmute(self); + let result: u8x8 = shuffle!( + bytes, [7, 6, 5, 4, 3, 2, 1, 0] + ); + crate::mem::transmute(result) + } + } + } + )+ + }; + (v128: $($id:ident,)+) => { + $( + impl SwapBytes for $id { + #[inline] + #[allow(clippy::useless_transmute)] + fn swap_bytes(self) -> Self { + unsafe { + let bytes: u8x16 = crate::mem::transmute(self); + let result: u8x16 = shuffle!(bytes, [ + 15, 14, 13, 12, 11, 10, 9, 8, + 7, 6, 5, 4, 3, 2, 1, 0 + ]); + crate::mem::transmute(result) + } + } + } + )+ + }; + (v256: $($id:ident,)+) => { + $( + impl SwapBytes for $id { + #[inline] + #[allow(clippy::useless_transmute)] + fn swap_bytes(self) -> Self { + unsafe { + let bytes: u8x32 = crate::mem::transmute(self); + let result: u8x32 = shuffle!(bytes, [ + 31, 30, 29, 28, 27, 26, 25, 24, + 23, 22, 21, 20, 19, 18, 17, 16, + 15, 14, 13, 12, 11, 10, 9, 8, + 7, 6, 5, 4, 3, 2, 1, 0 + ]); + crate::mem::transmute(result) + } + } + } + )+ + }; + (v512: $($id:ident,)+) => { + $( + impl SwapBytes for $id { + #[inline] + #[allow(clippy::useless_transmute)] + fn swap_bytes(self) -> Self { + unsafe { + let bytes: u8x64 = crate::mem::transmute(self); + let result: u8x64 = shuffle!(bytes, [ + 63, 62, 61, 60, 59, 58, 57, 56, + 55, 54, 53, 52, 51, 50, 49, 48, + 47, 46, 45, 44, 43, 42, 41, 40, + 39, 38, 37, 36, 35, 34, 33, 32, + 31, 30, 29, 28, 27, 26, 25, 24, + 23, 22, 21, 20, 19, 18, 17, 16, + 15, 14, 13, 12, 11, 10, 9, 8, + 7, 6, 5, 4, 3, 2, 1, 0 + ]); + crate::mem::transmute(result) + } + } + } + )+ + }; +} + +impl_swap_bytes!(v16: u8x2, i8x2,); +impl_swap_bytes!(v32: u8x4, i8x4, u16x2, i16x2,); +// FIXME: 64-bit single element vector +impl_swap_bytes!(v64: u8x8, i8x8, u16x4, i16x4, u32x2, i32x2 /* u64x1, i64x1, */,); + +impl_swap_bytes!(v128: u8x16, i8x16, u16x8, i16x8, u32x4, i32x4, u64x2, i64x2, u128x1, i128x1,); +impl_swap_bytes!(v256: u8x32, i8x32, u16x16, i16x16, u32x8, i32x8, u64x4, i64x4, u128x2, i128x2,); + +impl_swap_bytes!(v512: u8x64, i8x64, u16x32, i16x32, u32x16, i32x16, u64x8, i64x8, u128x4, i128x4,); + +cfg_if! { + if #[cfg(target_pointer_width = "8")] { + impl_swap_bytes!(v16: isizex2, usizex2,); + impl_swap_bytes!(v32: isizex4, usizex4,); + impl_swap_bytes!(v64: isizex8, usizex8,); + } else if #[cfg(target_pointer_width = "16")] { + impl_swap_bytes!(v32: isizex2, usizex2,); + impl_swap_bytes!(v64: isizex4, usizex4,); + impl_swap_bytes!(v128: isizex8, usizex8,); + } else if #[cfg(target_pointer_width = "32")] { + impl_swap_bytes!(v64: isizex2, usizex2,); + impl_swap_bytes!(v128: isizex4, usizex4,); + impl_swap_bytes!(v256: isizex8, usizex8,); + } else if #[cfg(target_pointer_width = "64")] { + impl_swap_bytes!(v128: isizex2, usizex2,); + impl_swap_bytes!(v256: isizex4, usizex4,); + impl_swap_bytes!(v512: isizex8, usizex8,); + } else { + compile_error!("unsupported target_pointer_width"); + } +} diff --git a/third_party/rust/packed_simd/src/codegen/v128.rs b/third_party/rust/packed_simd/src/codegen/v128.rs new file mode 100644 index 000000000000..9506424fadad --- /dev/null +++ b/third_party/rust/packed_simd/src/codegen/v128.rs @@ -0,0 +1,46 @@ +//! Internal 128-bit wide vector types + +use crate::masks::*; + +#[rustfmt::skip] +impl_simd_array!( + [i8; 16]: i8x16 | + i8, i8, i8, i8, + i8, i8, i8, i8, + i8, i8, i8, i8, + i8, i8, i8, i8 +); +#[rustfmt::skip] +impl_simd_array!( + [u8; 16]: u8x16 | + u8, u8, u8, u8, + u8, u8, u8, u8, + u8, u8, u8, u8, + u8, u8, u8, u8 +); +#[rustfmt::skip] +impl_simd_array!( + [m8; 16]: m8x16 | + i8, i8, i8, i8, + i8, i8, i8, i8, + i8, i8, i8, i8, + i8, i8, i8, i8 +); + +impl_simd_array!([i16; 8]: i16x8 | i16, i16, i16, i16, i16, i16, i16, i16); +impl_simd_array!([u16; 8]: u16x8 | u16, u16, u16, u16, u16, u16, u16, u16); +impl_simd_array!([m16; 8]: m16x8 | i16, i16, i16, i16, i16, i16, i16, i16); + +impl_simd_array!([i32; 4]: i32x4 | i32, i32, i32, i32); +impl_simd_array!([u32; 4]: u32x4 | u32, u32, u32, u32); +impl_simd_array!([f32; 4]: f32x4 | f32, f32, f32, f32); +impl_simd_array!([m32; 4]: m32x4 | i32, i32, i32, i32); + +impl_simd_array!([i64; 2]: i64x2 | i64, i64); +impl_simd_array!([u64; 2]: u64x2 | u64, u64); +impl_simd_array!([f64; 2]: f64x2 | f64, f64); +impl_simd_array!([m64; 2]: m64x2 | i64, i64); + +impl_simd_array!([i128; 1]: i128x1 | i128); +impl_simd_array!([u128; 1]: u128x1 | u128); +impl_simd_array!([m128; 1]: m128x1 | i128); diff --git a/third_party/rust/packed_simd/src/codegen/v16.rs b/third_party/rust/packed_simd/src/codegen/v16.rs new file mode 100644 index 000000000000..4d55a6d8998e --- /dev/null +++ b/third_party/rust/packed_simd/src/codegen/v16.rs @@ -0,0 +1,7 @@ +//! Internal 16-bit wide vector types + +use crate::masks::*; + +impl_simd_array!([i8; 2]: i8x2 | i8, i8); +impl_simd_array!([u8; 2]: u8x2 | u8, u8); +impl_simd_array!([m8; 2]: m8x2 | i8, i8); diff --git a/third_party/rust/packed_simd/src/codegen/v256.rs b/third_party/rust/packed_simd/src/codegen/v256.rs new file mode 100644 index 000000000000..5ca4759f0c0a --- /dev/null +++ b/third_party/rust/packed_simd/src/codegen/v256.rs @@ -0,0 +1,78 @@ +//! Internal 256-bit wide vector types + +use crate::masks::*; + +#[rustfmt::skip] +impl_simd_array!( + [i8; 32]: i8x32 | + i8, i8, i8, i8, + i8, i8, i8, i8, + i8, i8, i8, i8, + i8, i8, i8, i8, + i8, i8, i8, i8, + i8, i8, i8, i8, + i8, i8, i8, i8, + i8, i8, i8, i8 +); +#[rustfmt::skip] +impl_simd_array!( + [u8; 32]: u8x32 | + u8, u8, u8, u8, + u8, u8, u8, u8, + u8, u8, u8, u8, + u8, u8, u8, u8, + u8, u8, u8, u8, + u8, u8, u8, u8, + u8, u8, u8, u8, + u8, u8, u8, u8 +); +#[rustfmt::skip] +impl_simd_array!( + [m8; 32]: m8x32 | + i8, i8, i8, i8, + i8, i8, i8, i8, + i8, i8, i8, i8, + i8, i8, i8, i8, + i8, i8, i8, i8, + i8, i8, i8, i8, + i8, i8, i8, i8, + i8, i8, i8, i8 +); +#[rustfmt::skip] +impl_simd_array!( + [i16; 16]: i16x16 | + i16, i16, i16, i16, + i16, i16, i16, i16, + i16, i16, i16, i16, + i16, i16, i16, i16 +); +#[rustfmt::skip] +impl_simd_array!( + [u16; 16]: u16x16 | + u16, u16, u16, u16, + u16, u16, u16, u16, + u16, u16, u16, u16, + u16, u16, u16, u16 +); +#[rustfmt::skip] +impl_simd_array!( + [m16; 16]: m16x16 | + i16, i16, i16, i16, + i16, i16, i16, i16, + i16, i16, i16, i16, + i16, i16, i16, i16 +); + +impl_simd_array!([i32; 8]: i32x8 | i32, i32, i32, i32, i32, i32, i32, i32); +impl_simd_array!([u32; 8]: u32x8 | u32, u32, u32, u32, u32, u32, u32, u32); +impl_simd_array!([f32; 8]: f32x8 | f32, f32, f32, f32, f32, f32, f32, f32); +impl_simd_array!([m32; 8]: m32x8 | i32, i32, i32, i32, i32, i32, i32, i32); + +impl_simd_array!([i64; 4]: i64x4 | i64, i64, i64, i64); +impl_simd_array!([u64; 4]: u64x4 | u64, u64, u64, u64); +impl_simd_array!([f64; 4]: f64x4 | f64, f64, f64, f64); +impl_simd_array!([m64; 4]: m64x4 | i64, i64, i64, i64); + +impl_simd_array!([i128; 2]: i128x2 | i128, i128); +impl_simd_array!([u128; 2]: u128x2 | u128, u128); +impl_simd_array!([m128; 2]: m128x2 | i128, i128); diff --git a/third_party/rust/packed_simd/src/codegen/v32.rs b/third_party/rust/packed_simd/src/codegen/v32.rs new file mode 100644 index 000000000000..ae1dabd00c22 --- /dev/null +++ b/third_party/rust/packed_simd/src/codegen/v32.rs @@ -0,0 +1,11 @@ +//! Internal 32-bit wide vector types + +use crate::masks::*; + +impl_simd_array!([i8; 4]: i8x4 | i8, i8, i8, i8); +impl_simd_array!([u8; 4]: u8x4 | u8, u8, u8, u8); +impl_simd_array!([m8; 4]: m8x4 | i8, i8, i8, i8); + +impl_simd_array!([i16; 2]: i16x2 | i16, i16); +impl_simd_array!([u16; 2]: u16x2 | u16, u16); +impl_simd_array!([m16; 2]: m16x2 | i16, i16); diff --git a/third_party/rust/packed_simd/src/codegen/v512.rs b/third_party/rust/packed_simd/src/codegen/v512.rs new file mode 100644 index 000000000000..bf95110340d6 --- /dev/null +++ b/third_party/rust/packed_simd/src/codegen/v512.rs @@ -0,0 +1,145 @@ +//! Internal 512-bit wide vector types + +use crate::masks::*; + +#[rustfmt::skip] +impl_simd_array!( + [i8; 64]: i8x64 | + i8, i8, i8, i8, + i8, i8, i8, i8, + i8, i8, i8, i8, + i8, i8, i8, i8, + i8, i8, i8, i8, + i8, i8, i8, i8, + i8, i8, i8, i8, + i8, i8, i8, i8, + + i8, i8, i8, i8, + i8, i8, i8, i8, + i8, i8, i8, i8, + i8, i8, i8, i8, + i8, i8, i8, i8, + i8, i8, i8, i8, + i8, i8, i8, i8, + i8, i8, i8, i8 +); +#[rustfmt::skip] +impl_simd_array!( + [u8; 64]: u8x64 | + u8, u8, u8, u8, + u8, u8, u8, u8, + u8, u8, u8, u8, + u8, u8, u8, u8, + u8, u8, u8, u8, + u8, u8, u8, u8, + u8, u8, u8, u8, + u8, u8, u8, u8, + + u8, u8, u8, u8, + u8, u8, u8, u8, + u8, u8, u8, u8, + u8, u8, u8, u8, + u8, u8, u8, u8, + u8, u8, u8, u8, + u8, u8, u8, u8, + u8, u8, u8, u8 +); +#[rustfmt::skip] +impl_simd_array!( + [m8; 64]: m8x64 | + i8, i8, i8, i8, + i8, i8, i8, i8, + i8, i8, i8, i8, + i8, i8, i8, i8, + i8, i8, i8, i8, + i8, i8, i8, i8, + i8, i8, i8, i8, + i8, i8, i8, i8, + + i8, i8, i8, i8, + i8, i8, i8, i8, + i8, i8, i8, i8, + i8, i8, i8, i8, + i8, i8, i8, i8, + i8, i8, i8, i8, + i8, i8, i8, i8, + i8, i8, i8, i8 +); +#[rustfmt::skip] +impl_simd_array!( + [i16; 32]: i16x32 | + i16, i16, i16, i16, + i16, i16, i16, i16, + i16, i16, i16, i16, + i16, i16, i16, i16, + i16, i16, i16, i16, + i16, i16, i16, i16, + i16, i16, i16, i16, + i16, i16, i16, i16 +); +#[rustfmt::skip] +impl_simd_array!( + [u16; 32]: u16x32 | + u16, u16, u16, u16, + u16, u16, u16, u16, + u16, u16, u16, u16, + u16, u16, u16, u16, + u16, u16, u16, u16, + u16, u16, u16, u16, + u16, u16, u16, u16, + u16, u16, u16, u16 +); +#[rustfmt::skip] +impl_simd_array!( + [m16; 32]: m16x32 | + i16, i16, i16, i16, + i16, i16, i16, i16, + i16, i16, i16, i16, + i16, i16, i16, i16, + i16, i16, i16, i16, + i16, i16, i16, i16, + i16, i16, i16, i16, + i16, i16, i16, i16 +); + +#[rustfmt::skip] +impl_simd_array!( + [i32; 16]: i32x16 | + i32, i32, i32, i32, + i32, i32, i32, i32, + i32, i32, i32, i32, + i32, i32, i32, i32 +); +#[rustfmt::skip] +impl_simd_array!( + [u32; 16]: u32x16 | + u32, u32, u32, u32, + u32, u32, u32, u32, + u32, u32, u32, u32, + u32, u32, u32, u32 +); +#[rustfmt::skip] +impl_simd_array!( + [f32; 16]: f32x16 | + f32, f32, f32, f32, + f32, f32, f32, f32, + f32, f32, f32, f32, + f32, f32, f32, f32 +); +#[rustfmt::skip] +impl_simd_array!( + [m32; 16]: m32x16 | + i32, i32, i32, i32, + i32, i32, i32, i32, + i32, i32, i32, i32, + i32, i32, i32, i32 +); + +impl_simd_array!([i64; 8]: i64x8 | i64, i64, i64, i64, i64, i64, i64, i64); +impl_simd_array!([u64; 8]: u64x8 | u64, u64, u64, u64, u64, u64, u64, u64); +impl_simd_array!([f64; 8]: f64x8 | f64, f64, f64, f64, f64, f64, f64, f64); +impl_simd_array!([m64; 8]: m64x8 | i64, i64, i64, i64, i64, i64, i64, i64); + +impl_simd_array!([i128; 4]: i128x4 | i128, i128, i128, i128); +impl_simd_array!([u128; 4]: u128x4 | u128, u128, u128, u128); +impl_simd_array!([m128; 4]: m128x4 | i128, i128, i128, i128); diff --git a/third_party/rust/packed_simd/src/codegen/v64.rs b/third_party/rust/packed_simd/src/codegen/v64.rs new file mode 100644 index 000000000000..3cfb67c1a013 --- /dev/null +++ b/third_party/rust/packed_simd/src/codegen/v64.rs @@ -0,0 +1,21 @@ +//! Internal 64-bit wide vector types + +use crate::masks::*; + +impl_simd_array!([i8; 8]: i8x8 | i8, i8, i8, i8, i8, i8, i8, i8); +impl_simd_array!([u8; 8]: u8x8 | u8, u8, u8, u8, u8, u8, u8, u8); +impl_simd_array!([m8; 8]: m8x8 | i8, i8, i8, i8, i8, i8, i8, i8); + +impl_simd_array!([i16; 4]: i16x4 | i16, i16, i16, i16); +impl_simd_array!([u16; 4]: u16x4 | u16, u16, u16, u16); +impl_simd_array!([m16; 4]: m16x4 | i16, i16, i16, i16); + +impl_simd_array!([i32; 2]: i32x2 | i32, i32); +impl_simd_array!([u32; 2]: u32x2 | u32, u32); +impl_simd_array!([f32; 2]: f32x2 | f32, f32); +impl_simd_array!([m32; 2]: m32x2 | i32, i32); + +impl_simd_array!([i64; 1]: i64x1 | i64); +impl_simd_array!([u64; 1]: u64x1 | u64); +impl_simd_array!([f64; 1]: f64x1 | f64); +impl_simd_array!([m64; 1]: m64x1 | i64); diff --git a/third_party/rust/packed_simd/src/codegen/vPtr.rs b/third_party/rust/packed_simd/src/codegen/vPtr.rs new file mode 100644 index 000000000000..abd3aa877920 --- /dev/null +++ b/third_party/rust/packed_simd/src/codegen/vPtr.rs @@ -0,0 +1,35 @@ +//! Pointer vector types + +macro_rules! impl_simd_ptr { + ([$ptr_ty:ty; $elem_count:expr]: $tuple_id:ident | $ty:ident + | $($tys:ty),*) => { + #[derive(Copy, Clone)] + #[repr(simd)] + pub struct $tuple_id<$ty>($(pub(crate) $tys),*); + //^^^^^^^ leaked through SimdArray + + impl<$ty> crate::sealed::Seal for [$ptr_ty; $elem_count] {} + impl<$ty> crate::sealed::SimdArray for [$ptr_ty; $elem_count] { + type Tuple = $tuple_id<$ptr_ty>; + type T = $ptr_ty; + const N: usize = $elem_count; + type NT = [u32; $elem_count]; + } + + impl<$ty> crate::sealed::Seal for $tuple_id<$ptr_ty> {} + impl<$ty> crate::sealed::Simd for $tuple_id<$ptr_ty> { + type Element = $ptr_ty; + const LANES: usize = $elem_count; + type LanesType = [u32; $elem_count]; + } + + } +} + +impl_simd_ptr!([*const T; 2]: cptrx2 | T | T, T); +impl_simd_ptr!([*const T; 4]: cptrx4 | T | T, T, T, T); +impl_simd_ptr!([*const T; 8]: cptrx8 | T | T, T, T, T, T, T, T, T); + +impl_simd_ptr!([*mut T; 2]: mptrx2 | T | T, T); +impl_simd_ptr!([*mut T; 4]: mptrx4 | T | T, T, T, T); +impl_simd_ptr!([*mut T; 8]: mptrx8 | T | T, T, T, T, T, T, T, T); diff --git a/third_party/rust/packed_simd/src/codegen/vSize.rs b/third_party/rust/packed_simd/src/codegen/vSize.rs new file mode 100644 index 000000000000..d5db03991d61 --- /dev/null +++ b/third_party/rust/packed_simd/src/codegen/vSize.rs @@ -0,0 +1,16 @@ +//! Vector types with pointer-sized elements + +use crate::codegen::pointer_sized_int::{isize_, usize_}; +use crate::masks::*; + +impl_simd_array!([isize; 2]: isizex2 | isize_, isize_); +impl_simd_array!([usize; 2]: usizex2 | usize_, usize_); +impl_simd_array!([msize; 2]: msizex2 | isize_, isize_); + +impl_simd_array!([isize; 4]: isizex4 | isize_, isize_, isize_, isize_); +impl_simd_array!([usize; 4]: usizex4 | usize_, usize_, usize_, usize_); +impl_simd_array!([msize; 4]: msizex4 | isize_, isize_, isize_, isize_); + +impl_simd_array!([isize; 8]: isizex8 | isize_, isize_, isize_, isize_, isize_, isize_, isize_, isize_); +impl_simd_array!([usize; 8]: usizex8 | usize_, usize_, usize_, usize_, usize_, usize_, usize_, usize_); +impl_simd_array!([msize; 8]: msizex8 | isize_, isize_, isize_, isize_, isize_, isize_, isize_, isize_); diff --git a/third_party/rust/packed_simd/src/lib.rs b/third_party/rust/packed_simd/src/lib.rs new file mode 100644 index 000000000000..867cc10e9eb6 --- /dev/null +++ b/third_party/rust/packed_simd/src/lib.rs @@ -0,0 +1,348 @@ +//! # Portable packed SIMD vectors +//! +//! This crate is proposed for stabilization as `std::packed_simd` in [RFC2366: +//! `std::simd`](https://github.com/rust-lang/rfcs/pull/2366) . +//! +//! The examples available in the +//! [`examples/`](https://github.com/rust-lang-nursery/packed_simd/tree/master/examples) +//! sub-directory of the crate showcase how to use the library in practice. +//! +//! ## Table of contents +//! +//! - [Introduction](#introduction) +//! - [Vector types](#vector-types) +//! - [Conditional operations](#conditional-operations) +//! - [Conversions](#conversions) +//! - [Hardware Features](#hardware-features) +//! - [Performance guide](https://rust-lang-nursery.github.io/packed_simd/perf-guide/) +//! +//! ## Introduction +//! +//! This crate exports [`Simd<[T; N]>`][`Simd`]: a packed vector of `N` +//! elements of type `T` as well as many type aliases for this type: for +//! example, [`f32x4`], which is just an alias for `Simd<[f32; 4]>`. +//! +//! The operations on packed vectors are, by default, "vertical", that is, they +//! are applied to each vector lane in isolation of the others: +//! +//! ``` +//! # use packed_simd::*; +//! let a = i32x4::new(1, 2, 3, 4); +//! let b = i32x4::new(5, 6, 7, 8); +//! assert_eq!(a + b, i32x4::new(6, 8, 10, 12)); +//! ``` +//! +//! Many "horizontal" operations are also provided: +//! +//! ``` +//! # use packed_simd::*; +//! # let a = i32x4::new(1, 2, 3, 4); +//! assert_eq!(a.wrapping_sum(), 10); +//! ``` +//! +//! In virtually all architectures vertical operations are fast, while +//! horizontal operations are, by comparison, much slower. That is, the +//! most portably-efficient way of performing a reduction over a slice +//! is to collect the results into a vector using vertical operations, +//! and performing a single horizontal operation at the end: +//! +//! ``` +//! # use packed_simd::*; +//! fn reduce(x: &[i32]) -> i32 { +//! assert_eq!(x.len() % 4, 0); +//! let mut sum = i32x4::splat(0); // [0, 0, 0, 0] +//! for i in (0..x.len()).step_by(4) { +//! sum += i32x4::from_slice_unaligned(&x[i..]); +//! } +//! sum.wrapping_sum() +//! } +//! +//! let x = [0, 1, 2, 3, 4, 5, 6, 7]; +//! assert_eq!(reduce(&x), 28); +//! ``` +//! +//! ## Vector types +//! +//! The vector type aliases are named according to the following scheme: +//! +//! > `{element_type}x{number_of_lanes} == Simd<[element_type; +//! number_of_lanes]>` +//! +//! where the following element types are supported: +//! +//! * `i{element_width}`: signed integer +//! * `u{element_width}`: unsigned integer +//! * `f{element_width}`: float +//! * `m{element_width}`: mask (see below) +//! * `*{const,mut} T`: `const` and `mut` pointers +//! +//! ## Basic operations +//! +//! ``` +//! # use packed_simd::*; +//! // Sets all elements to `0`: +//! let a = i32x4::splat(0); +//! +//! // Reads a vector from a slice: +//! let mut arr = [0, 0, 0, 1, 2, 3, 4, 5]; +//! let b = i32x4::from_slice_unaligned(&arr); +//! +//! // Reads the 4-th element of a vector: +//! assert_eq!(b.extract(3), 1); +//! +//! // Returns a new vector where the 4-th element is replaced with `1`: +//! let a = a.replace(3, 1); +//! assert_eq!(a, b); +//! +//! // Writes a vector to a slice: +//! let a = a.replace(2, 1); +//! a.write_to_slice_unaligned(&mut arr[4..]); +//! assert_eq!(arr, [0, 0, 0, 1, 0, 0, 1, 1]); +//! ``` +//! +//! ## Conditional operations +//! +//! One often needs to perform an operation on some lanes of the vector. Vector +//! masks, like `m32x4`, allow selecting on which vector lanes an operation is +//! to be performed: +//! +//! ``` +//! # use packed_simd::*; +//! let a = i32x4::new(1, 1, 2, 2); +//! +//! // Add `1` to the first two lanes of the vector. +//! let m = m16x4::new(true, true, false, false); +//! let a = m.select(a + 1, a); +//! assert_eq!(a, i32x4::splat(2)); +//! ``` +//! +//! The elements of a vector mask are either `true` or `false`. Here `true` +//! means that a lane is "selected", while `false` means that a lane is not +//! selected. +//! +//! All vector masks implement a `mask.select(a: T, b: T) -> T` method that +//! works on all vectors that have the same number of lanes as the mask. The +//! resulting vector contains the elements of `a` for those lanes for which the +//! mask is `true`, and the elements of `b` otherwise. +//! +//! The example constructs a mask with the first two lanes set to `true` and +//! the last two lanes set to `false`. This selects the first two lanes of `a + +//! 1` and the last two lanes of `a`, producing a vector where the first two +//! lanes have been incremented by `1`. +//! +//! > note: mask `select` can be used on vector types that have the same number +//! > of lanes as the mask. The example shows this by using [`m16x4`] instead +//! > of [`m32x4`]. It is _typically_ more performant to use a mask element +//! > width equal to the element width of the vectors being operated upon. +//! > This is, however, not true for 512-bit wide vectors when targeting +//! > AVX-512, where the most efficient masks use only 1-bit per element. +//! +//! All vertical comparison operations returns masks: +//! +//! ``` +//! # use packed_simd::*; +//! let a = i32x4::new(1, 1, 3, 3); +//! let b = i32x4::new(2, 2, 0, 0); +//! +//! // ge: >= (Greater Eequal; see also lt, le, gt, eq, ne). +//! let m = a.ge(i32x4::splat(2)); +//! +//! if m.any() { +//! // all / any / none allow coherent control flow +//! let d = m.select(a, b); +//! assert_eq!(d, i32x4::new(2, 2, 3, 3)); +//! } +//! ``` +//! +//! ## Conversions +//! +//! * **lossless widening conversions**: [`From`]/[`Into`] are implemented for +//! vectors with the same number of lanes when the conversion is value +//! preserving (same as in `std`). +//! +//! * **safe bitwise conversions**: The cargo feature `into_bits` provides the +//! `IntoBits/FromBits` traits (`x.into_bits()`). These perform safe bitwise +//! `transmute`s when all bit patterns of the source type are valid bit +//! patterns of the target type and are also implemented for the +//! architecture-specific vector types of `std::arch`. For example, `let x: +//! u8x8 = m8x8::splat(true).into_bits();` is provided because all `m8x8` bit +//! patterns are valid `u8x8` bit patterns. However, the opposite is not +//! true, not all `u8x8` bit patterns are valid `m8x8` bit-patterns, so this +//! operation cannot be performed safely using `x.into_bits()`; one needs to +//! use `unsafe { crate::mem::transmute(x) }` for that, making sure that the +//! value in the `u8x8` is a valid bit-pattern of `m8x8`. +//! +//! * **numeric casts** (`as`): are performed using [`FromCast`]/[`Cast`] +//! (`x.cast()`), just like `as`: +//! +//! * casting integer vectors whose lane types have the same size (e.g. +//! `i32xN` -> `u32xN`) is a **no-op**, +//! +//! * casting from a larger integer to a smaller integer (e.g. `u32xN` -> +//! `u8xN`) will **truncate**, +//! +//! * casting from a smaller integer to a larger integer (e.g. `u8xN` -> +//! `u32xN`) will: +//! * **zero-extend** if the source is unsigned, or +//! * **sign-extend** if the source is signed, +//! +//! * casting from a float to an integer will **round the float towards +//! zero**, +//! +//! * casting from an integer to float will produce the floating point +//! representation of the integer, **rounding to nearest, ties to even**, +//! +//! * casting from an `f32` to an `f64` is perfect and lossless, +//! +//! * casting from an `f64` to an `f32` **rounds to nearest, ties to even**. +//! +//! Numeric casts are not very "precise": sometimes lossy, sometimes value +//! preserving, etc. +//! +//! ## Hardware Features +//! +//! This crate can use different hardware features based on your configured +//! `RUSTFLAGS`. For example, with no configured `RUSTFLAGS`, `u64x8` on +//! x86_64 will use SSE2 operations like `PCMPEQD`. If you configure +//! `RUSTFLAGS='-C target-feature=+avx2,+avx'` on supported x86_64 hardware +//! the same `u64x8` may use wider AVX2 operations like `VPCMPEQQ`. It is +//! important for performance and for hardware support requirements that +//! you choose an appropriate set of `target-feature` and `target-cpu` +//! options during builds. For more information, see the [Performance +//! guide](https://rust-lang-nursery.github.io/packed_simd/perf-guide/) + +#![feature( + adt_const_params, + repr_simd, + rustc_attrs, + platform_intrinsics, + stdsimd, + arm_target_feature, + link_llvm_intrinsics, + core_intrinsics, + stmt_expr_attributes, + custom_inner_attributes, +)] +#![allow(non_camel_case_types, non_snake_case, + // FIXME: these types are unsound in C FFI already + // See https://github.com/rust-lang/rust/issues/53346 + improper_ctypes_definitions, + incomplete_features, + clippy::cast_possible_truncation, + clippy::cast_lossless, + clippy::cast_possible_wrap, + clippy::cast_precision_loss, + // TODO: manually add the `#[must_use]` attribute where appropriate + clippy::must_use_candidate, + // This lint is currently broken for generic code + // See https://github.com/rust-lang/rust-clippy/issues/3410 + clippy::use_self, + clippy::wrong_self_convention, + clippy::from_over_into, +)] +#![cfg_attr(test, feature(hashmap_internals))] +#![cfg_attr(doc_cfg, feature(doc_cfg))] +#![deny(rust_2018_idioms, clippy::missing_inline_in_public_items)] +#![no_std] + +use cfg_if::cfg_if; + +cfg_if! { + if #[cfg(feature = "core_arch")] { + #[allow(unused_imports)] + use core_arch as arch; + } else { + #[allow(unused_imports)] + use core::arch; + } +} + +#[cfg(all(target_arch = "wasm32", test))] +use wasm_bindgen_test::*; + +#[allow(unused_imports)] +use core::{ + /* arch (handled above), */ cmp, f32, f64, fmt, hash, hint, i128, i16, i32, i64, i8, intrinsics, + isize, iter, marker, mem, ops, ptr, slice, u128, u16, u32, u64, u8, usize, +}; + +#[macro_use] +mod testing; +#[macro_use] +mod api; +mod codegen; +mod sealed; + +pub use crate::sealed::{Mask, Shuffle, Simd as SimdVector, SimdArray}; + +/// Packed SIMD vector type. +/// +/// # Examples +/// +/// ``` +/// # use packed_simd::Simd; +/// let v = Simd::<[i32; 4]>::new(0, 1, 2, 3); +/// assert_eq!(v.extract(2), 2); +/// ``` +#[repr(transparent)] +#[derive(Copy, Clone)] +pub struct Simd( + // FIXME: this type should be private, + // but it currently must be public for the + // `shuffle!` macro to work: it needs to + // access the internal `repr(simd)` type + // to call the shuffle intrinsics. + #[doc(hidden)] pub ::Tuple, +); + +impl sealed::Seal for Simd {} + +/// Wrapper over `T` implementing a lexicoraphical order via the `PartialOrd` +/// and/or `Ord` traits. +#[repr(transparent)] +#[derive(Copy, Clone, Debug)] +#[allow(clippy::missing_inline_in_public_items)] +pub struct LexicographicallyOrdered(T); + +mod masks; +pub use self::masks::*; + +mod v16; +pub use self::v16::*; + +mod v32; +pub use self::v32::*; + +mod v64; +pub use self::v64::*; + +mod v128; +pub use self::v128::*; + +mod v256; +pub use self::v256::*; + +mod v512; +pub use self::v512::*; + +mod vSize; +pub use self::vSize::*; + +mod vPtr; +pub use self::vPtr::*; + +pub use self::api::cast::*; + +#[cfg(feature = "into_bits")] +pub use self::api::into_bits::*; + +// Re-export the shuffle intrinsics required by the `shuffle!` macro. +#[doc(hidden)] +pub use self::codegen::llvm::{ + __shuffle_vector16, __shuffle_vector2, __shuffle_vector32, __shuffle_vector4, __shuffle_vector64, + __shuffle_vector8, +}; + +pub(crate) mod llvm { + pub(crate) use crate::codegen::llvm::*; +} diff --git a/third_party/rust/packed_simd/src/masks.rs b/third_party/rust/packed_simd/src/masks.rs new file mode 100644 index 000000000000..04534eab2ba0 --- /dev/null +++ b/third_party/rust/packed_simd/src/masks.rs @@ -0,0 +1,126 @@ +//! Mask types + +macro_rules! impl_mask_ty { + ($id:ident : $elem_ty:ident | #[$doc:meta]) => { + #[$doc] + #[derive(Copy, Clone)] + pub struct $id($elem_ty); + + impl crate::sealed::Seal for $id {} + impl crate::sealed::Mask for $id { + #[inline] + fn test(&self) -> bool { + $id::test(self) + } + } + + impl $id { + /// Instantiate a mask with `value` + #[inline] + pub fn new(x: bool) -> Self { + if x { + $id(!0) + } else { + $id(0) + } + } + /// Test if the mask is set + #[inline] + pub fn test(&self) -> bool { + self.0 != 0 + } + } + + impl Default for $id { + #[inline] + fn default() -> Self { + $id(0) + } + } + + #[allow(clippy::partialeq_ne_impl)] + impl PartialEq<$id> for $id { + #[inline] + fn eq(&self, other: &Self) -> bool { + self.0 == other.0 + } + #[inline] + fn ne(&self, other: &Self) -> bool { + self.0 != other.0 + } + } + + impl Eq for $id {} + + impl PartialOrd<$id> for $id { + #[inline] + fn partial_cmp(&self, other: &Self) -> Option { + use crate::cmp::Ordering; + if self == other { + Some(Ordering::Equal) + } else if self.0 > other.0 { + // Note: + // * false = 0_i + // * true == !0_i == -1_i + Some(Ordering::Less) + } else { + Some(Ordering::Greater) + } + } + + #[inline] + fn lt(&self, other: &Self) -> bool { + self.0 > other.0 + } + #[inline] + fn gt(&self, other: &Self) -> bool { + self.0 < other.0 + } + #[inline] + fn le(&self, other: &Self) -> bool { + self.0 >= other.0 + } + #[inline] + fn ge(&self, other: &Self) -> bool { + self.0 <= other.0 + } + } + + impl Ord for $id { + #[inline] + fn cmp(&self, other: &Self) -> crate::cmp::Ordering { + match self.partial_cmp(other) { + Some(x) => x, + None => unsafe { crate::hint::unreachable_unchecked() }, + } + } + } + + impl crate::hash::Hash for $id { + #[inline] + fn hash(&self, state: &mut H) { + (self.0 != 0).hash(state); + } + } + + impl crate::fmt::Debug for $id { + #[inline] + fn fmt(&self, fmtter: &mut crate::fmt::Formatter<'_>) -> Result<(), crate::fmt::Error> { + write!(fmtter, "{}({})", stringify!($id), self.0 != 0) + } + } + }; +} + +impl_mask_ty!(m8: i8 | /// 8-bit wide mask. +); +impl_mask_ty!(m16: i16 | /// 16-bit wide mask. +); +impl_mask_ty!(m32: i32 | /// 32-bit wide mask. +); +impl_mask_ty!(m64: i64 | /// 64-bit wide mask. +); +impl_mask_ty!(m128: i128 | /// 128-bit wide mask. +); +impl_mask_ty!(msize: isize | /// isize-wide mask. +); diff --git a/third_party/rust/packed_simd/src/sealed.rs b/third_party/rust/packed_simd/src/sealed.rs new file mode 100644 index 000000000000..0ec20300fdd4 --- /dev/null +++ b/third_party/rust/packed_simd/src/sealed.rs @@ -0,0 +1,42 @@ +//! Sealed traits + +/// A sealed trait, this is logically private to the crate +/// and will prevent implementations from outside the crate +pub trait Seal {} + +/// Trait implemented by arrays that can be SIMD types. +pub trait SimdArray: Seal { + /// The type of the #[repr(simd)] type. + type Tuple: Copy + Clone; + /// The element type of the vector. + type T; + /// The number of elements in the array. + const N: usize; + /// The type: `[u32; Self::N]`. + type NT; +} + +/// This traits is used to constraint the arguments +/// and result type of the portable shuffles. +#[doc(hidden)] +pub trait Shuffle: Seal { + // Lanes is a `[u32; N]` where `N` is the number of vector lanes + + /// The result type of the shuffle. + type Output; +} + +/// This trait is implemented by all SIMD vector types. +pub trait Simd: Seal { + /// Element type of the SIMD vector + type Element; + /// The number of elements in the SIMD vector. + const LANES: usize; + /// The type: `[u32; Self::N]`. + type LanesType; +} + +/// This trait is implemented by all mask types +pub trait Mask: Seal { + fn test(&self) -> bool; +} diff --git a/third_party/rust/packed_simd/src/testing.rs b/third_party/rust/packed_simd/src/testing.rs new file mode 100644 index 000000000000..6320b2805556 --- /dev/null +++ b/third_party/rust/packed_simd/src/testing.rs @@ -0,0 +1,8 @@ +//! Testing macros and other utilities. + +#[macro_use] +mod macros; + +#[cfg(test)] +#[macro_use] +pub(crate) mod utils; diff --git a/third_party/rust/packed_simd/src/testing/macros.rs b/third_party/rust/packed_simd/src/testing/macros.rs new file mode 100644 index 000000000000..7bc4268b9002 --- /dev/null +++ b/third_party/rust/packed_simd/src/testing/macros.rs @@ -0,0 +1,44 @@ +//! Testing macros + +macro_rules! test_if { + ($cfg_tt:tt: $it:item) => { + #[cfg(any( + // Test everything if: + // + // * tests are enabled, + // * no features about exclusively testing + // specific vector classes are enabled + all(test, not(any( + test_v16, + test_v32, + test_v64, + test_v128, + test_v256, + test_v512, + test_none, // disables all tests + ))), + // Test if: + // + // * tests are enabled + // * a particular cfg token tree returns true + all(test, $cfg_tt), + ))] + $it + }; +} + +#[cfg(test)] +#[allow(unused)] +macro_rules! ref_ { + ($anything:tt) => { + &$anything + }; +} + +#[cfg(test)] +#[allow(unused)] +macro_rules! ref_mut_ { + ($anything:tt) => { + &mut $anything + }; +} diff --git a/third_party/rust/packed_simd/src/testing/utils.rs b/third_party/rust/packed_simd/src/testing/utils.rs new file mode 100644 index 000000000000..7d8f39573964 --- /dev/null +++ b/third_party/rust/packed_simd/src/testing/utils.rs @@ -0,0 +1,130 @@ +//! Testing utilities + +#![allow(dead_code)] +// FIXME: Or don't. But it's true this is a problematic comparison. +#![allow(clippy::neg_cmp_op_on_partial_ord)] + +use crate::{cmp::PartialOrd, fmt::Debug, LexicographicallyOrdered}; + +/// Tests PartialOrd for `a` and `b` where `a < b` is true. +pub fn test_lt(a: LexicographicallyOrdered, b: LexicographicallyOrdered) +where + LexicographicallyOrdered: Debug + PartialOrd, +{ + assert!(a < b, "{:?}, {:?}", a, b); + assert!(b > a, "{:?}, {:?}", a, b); + + assert!(!(a == b), "{:?}, {:?}", a, b); + assert_ne!(a, b, "{:?}, {:?}", a, b); + + assert!(a <= b, "{:?}, {:?}", a, b); + assert!(b >= a, "{:?}, {:?}", a, b); + + // The elegance of the mathematical expression of irreflexivity is more + // than clippy can handle. + #[allow(clippy::eq_op)] + { + // Irreflexivity + assert!(!(a < a), "{:?}, {:?}", a, b); + assert!(!(b < b), "{:?}, {:?}", a, b); + assert!(!(a > a), "{:?}, {:?}", a, b); + assert!(!(b > b), "{:?}, {:?}", a, b); + + assert!(a <= a, "{:?}, {:?}", a, b); + assert!(b <= b, "{:?}, {:?}", a, b); + } +} + +/// Tests PartialOrd for `a` and `b` where `a <= b` is true. +pub fn test_le(a: LexicographicallyOrdered, b: LexicographicallyOrdered) +where + LexicographicallyOrdered: Debug + PartialOrd, +{ + assert!(a <= b, "{:?}, {:?}", a, b); + assert!(b >= a, "{:?}, {:?}", a, b); + + assert!(a <= b, "{:?}, {:?}", a, b); + assert!(b >= a, "{:?}, {:?}", a, b); + + if a == b { + assert!(!(a < b), "{:?}, {:?}", a, b); + assert!(!(b > a), "{:?}, {:?}", a, b); + + assert!(!(a != b), "{:?}, {:?}", a, b); + } else { + assert_ne!(a, b, "{:?}, {:?}", a, b); + test_lt(a, b); + } +} + +/// Test PartialOrd::partial_cmp for `a` and `b` returning `Ordering` +pub fn test_cmp( + a: LexicographicallyOrdered, + b: LexicographicallyOrdered, + o: Option, +) where + LexicographicallyOrdered: PartialOrd + Debug, + T: Debug + crate::sealed::Simd + Copy + Clone, + ::Element: Default + Copy + Clone + PartialOrd, +{ + assert!(T::LANES <= 64, "array length in these two arrays needs updating"); + let mut arr_a: [T::Element; 64] = [Default::default(); 64]; + let mut arr_b: [T::Element; 64] = [Default::default(); 64]; + + unsafe { crate::ptr::write_unaligned(arr_a.as_mut_ptr() as *mut LexicographicallyOrdered, a) } + unsafe { crate::ptr::write_unaligned(arr_b.as_mut_ptr() as *mut LexicographicallyOrdered, b) } + let expected = arr_a[0..T::LANES].partial_cmp(&arr_b[0..T::LANES]); + let result = a.partial_cmp(&b); + assert_eq!(expected, result, "{:?}, {:?}", a, b); + assert_eq!(o, result, "{:?}, {:?}", a, b); + match o { + Some(crate::cmp::Ordering::Less) => { + test_lt(a, b); + test_le(a, b); + } + Some(crate::cmp::Ordering::Greater) => { + test_lt(b, a); + test_le(b, a); + } + Some(crate::cmp::Ordering::Equal) => { + assert!(a == b, "{:?}, {:?}", a, b); + assert!(!(a != b), "{:?}, {:?}", a, b); + assert!(!(a < b), "{:?}, {:?}", a, b); + assert!(!(b < a), "{:?}, {:?}", a, b); + assert!(!(a > b), "{:?}, {:?}", a, b); + assert!(!(b > a), "{:?}, {:?}", a, b); + + test_le(a, b); + test_le(b, a); + } + None => { + assert!(!(a == b), "{:?}, {:?}", a, b); + assert!(!(a != b), "{:?}, {:?}", a, b); + assert!(!(a < b), "{:?}, {:?}", a, b); + assert!(!(a > b), "{:?}, {:?}", a, b); + assert!(!(b < a), "{:?}, {:?}", a, b); + assert!(!(b > a), "{:?}, {:?}", a, b); + assert!(!(a <= b), "{:?}, {:?}", a, b); + assert!(!(b <= a), "{:?}, {:?}", a, b); + assert!(!(a >= b), "{:?}, {:?}", a, b); + assert!(!(b >= a), "{:?}, {:?}", a, b); + } + } +} + +// Returns a tuple containing two distinct pointer values of the same type as +// the element type of the Simd vector `$id`. +#[allow(unused)] +macro_rules! ptr_vals { + ($id:ty) => { + // expands to an expression + #[allow(unused_unsafe)] + unsafe { + // all bits cleared + let clear: <$id as sealed::Simd>::Element = crate::mem::zeroed(); + // all bits set + let set: <$id as sealed::Simd>::Element = crate::mem::transmute(-1_isize); + (clear, set) + } + }; +} diff --git a/third_party/rust/packed_simd/src/v128.rs b/third_party/rust/packed_simd/src/v128.rs new file mode 100644 index 000000000000..7949f6619a4f --- /dev/null +++ b/third_party/rust/packed_simd/src/v128.rs @@ -0,0 +1,80 @@ +//! 128-bit wide vector types +#[rustfmt::skip] + +use crate::*; + +impl_i!([i8; 16]: i8x16, m8x16 | i8, u16 | test_v128 | + x0, x1, x2, x3, x4, x5, x6, x7, x8, x9, x10, x11, x12, x13, x14, x15 | + From: | + /// A 128-bit vector with 16 `i8` lanes. +); +impl_u!([u8; 16]: u8x16, m8x16 | u8, u16 | test_v128 | + x0, x1, x2, x3, x4, x5, x6, x7, x8, x9, x10, x11, x12, x13, x14, x15 | + From: | + /// A 128-bit vector with 16 `u8` lanes. +); +impl_m!([m8; 16]: m8x16 | i8, u16 | test_v128 | + x0, x1, x2, x3, x4, x5, x6, x7, x8, x9, x10, x11, x12, x13, x14, x15 | + From: m16x16 | + /// A 128-bit vector mask with 16 `m8` lanes. +); + +impl_i!([i16; 8]: i16x8, m16x8 | i16, u8 | test_v128 | x0, x1, x2, x3, x4, x5, x6, x7 | + From: i8x8, u8x8 | + /// A 128-bit vector with 8 `i16` lanes. +); +impl_u!([u16; 8]: u16x8, m16x8 | u16, u8 | test_v128 | x0, x1, x2, x3, x4, x5, x6, x7 | + From: u8x8 | + /// A 128-bit vector with 8 `u16` lanes. +); +impl_m!([m16; 8]: m16x8 | i16, u8 | test_v128 | x0, x1, x2, x3, x4, x5, x6, x7 | + From: m8x8, m32x8 | + /// A 128-bit vector mask with 8 `m16` lanes. +); + +impl_i!([i32; 4]: i32x4, m32x4 | i32, u8 | test_v128 | x0, x1, x2, x3 | + From: i8x4, u8x4, i16x4, u16x4 | + /// A 128-bit vector with 4 `i32` lanes. +); +impl_u!([u32; 4]: u32x4, m32x4 | u32, u8 | test_v128 | x0, x1, x2, x3 | + From: u8x4, u16x4 | + /// A 128-bit vector with 4 `u32` lanes. +); +impl_f!([f32; 4]: f32x4, m32x4 | f32 | test_v128 | x0, x1, x2, x3 | + From: i8x4, u8x4, i16x4, u16x4 | + /// A 128-bit vector with 4 `f32` lanes. +); +impl_m!([m32; 4]: m32x4 | i32, u8 | test_v128 | x0, x1, x2, x3 | + From: m8x4, m16x4, m64x4 | + /// A 128-bit vector mask with 4 `m32` lanes. +); + +impl_i!([i64; 2]: i64x2, m64x2 | i64, u8 | test_v128 | x0, x1 | + From: i8x2, u8x2, i16x2, u16x2, i32x2, u32x2 | + /// A 128-bit vector with 2 `i64` lanes. +); +impl_u!([u64; 2]: u64x2, m64x2 | u64, u8 | test_v128 | x0, x1 | + From: u8x2, u16x2, u32x2 | + /// A 128-bit vector with 2 `u64` lanes. +); +impl_f!([f64; 2]: f64x2, m64x2 | f64 | test_v128 | x0, x1 | + From: i8x2, u8x2, i16x2, u16x2, i32x2, u32x2, f32x2 | + /// A 128-bit vector with 2 `f64` lanes. +); +impl_m!([m64; 2]: m64x2 | i64, u8 | test_v128 | x0, x1 | + From: m8x2, m16x2, m32x2, m128x2 | + /// A 128-bit vector mask with 2 `m64` lanes. +); + +impl_i!([i128; 1]: i128x1, m128x1 | i128, u8 | test_v128 | x0 | + From: /*i8x1, u8x1, i16x1, u16x1, i32x1, u32x1, i64x1, u64x1 */ | // FIXME: unary small vector types + /// A 128-bit vector with 1 `i128` lane. +); +impl_u!([u128; 1]: u128x1, m128x1 | u128, u8 | test_v128 | x0 | + From: /*u8x1, u16x1, u32x1, u64x1 */ | // FIXME: unary small vector types + /// A 128-bit vector with 1 `u128` lane. +); +impl_m!([m128; 1]: m128x1 | i128, u8 | test_v128 | x0 | + From: /*m8x1, m16x1, m32x1, m64x1 */ | // FIXME: unary small vector types + /// A 128-bit vector mask with 1 `m128` lane. +); diff --git a/third_party/rust/packed_simd/src/v16.rs b/third_party/rust/packed_simd/src/v16.rs new file mode 100644 index 000000000000..4ca5afb2a7d5 --- /dev/null +++ b/third_party/rust/packed_simd/src/v16.rs @@ -0,0 +1,16 @@ +//! 16-bit wide vector types + +use crate::*; + +impl_i!([i8; 2]: i8x2, m8x2 | i8, u8 | test_v16 | x0, x1 | + From: | + /// A 16-bit vector with 2 `i8` lanes. +); +impl_u!([u8; 2]: u8x2, m8x2 | u8, u8 | test_v16 | x0, x1 | + From: | + /// A 16-bit vector with 2 `u8` lanes. +); +impl_m!([m8; 2]: m8x2 | i8, u8 | test_v16 | x0, x1 | + From: m16x2, m32x2, m64x2, m128x2 | + /// A 16-bit vector mask with 2 `m8` lanes. +); diff --git a/third_party/rust/packed_simd/src/v256.rs b/third_party/rust/packed_simd/src/v256.rs new file mode 100644 index 000000000000..f0c3bc281383 --- /dev/null +++ b/third_party/rust/packed_simd/src/v256.rs @@ -0,0 +1,86 @@ +//! 256-bit wide vector types +#[rustfmt::skip] + +use crate::*; + +impl_i!([i8; 32]: i8x32, m8x32 | i8, u32 | test_v256 | + x0, x1, x2, x3, x4, x5, x6, x7, x8, x9, x10, x11, x12, x13, x14, x15, + x16, x17, x18, x19, x20, x21, x22, x23, x24, x25, x26, x27, x28, x29, x30, x31 | + From: | + /// A 256-bit vector with 32 `i8` lanes. +); +impl_u!([u8; 32]: u8x32, m8x32 | u8, u32 | test_v256 | + x0, x1, x2, x3, x4, x5, x6, x7, x8, x9, x10, x11, x12, x13, x14, x15, + x16, x17, x18, x19, x20, x21, x22, x23, x24, x25, x26, x27, x28, x29, x30, x31 | + From: | + /// A 256-bit vector with 32 `u8` lanes. +); +impl_m!([m8; 32]: m8x32 | i8, u32 | test_v256 | + x0, x1, x2, x3, x4, x5, x6, x7, x8, x9, x10, x11, x12, x13, x14, x15, + x16, x17, x18, x19, x20, x21, x22, x23, x24, x25, x26, x27, x28, x29, x30, x31 | + From: | + /// A 256-bit vector mask with 32 `m8` lanes. +); + +impl_i!([i16; 16]: i16x16, m16x16 | i16, u16 | test_v256 | + x0, x1, x2, x3, x4, x5, x6, x7, x8, x9, x10, x11, x12, x13, x14, x15 | + From: i8x16, u8x16 | + /// A 256-bit vector with 16 `i16` lanes. +); +impl_u!([u16; 16]: u16x16, m16x16 | u16, u16 | test_v256 | + x0, x1, x2, x3, x4, x5, x6, x7, x8, x9, x10, x11, x12, x13, x14, x15 | + From: u8x16 | + /// A 256-bit vector with 16 `u16` lanes. +); +impl_m!([m16; 16]: m16x16 | i16, u16 | test_v256 | + x0, x1, x2, x3, x4, x5, x6, x7, x8, x9, x10, x11, x12, x13, x14, x15 | + From: m8x16 | + /// A 256-bit vector mask with 16 `m16` lanes. +); + +impl_i!([i32; 8]: i32x8, m32x8 | i32, u8 | test_v256 | x0, x1, x2, x3, x4, x5, x6, x7 | + From: i8x8, u8x8, i16x8, u16x8 | + /// A 256-bit vector with 8 `i32` lanes. +); +impl_u!([u32; 8]: u32x8, m32x8 | u32, u8 | test_v256 | x0, x1, x2, x3, x4, x5, x6, x7 | + From: u8x8, u16x8 | + /// A 256-bit vector with 8 `u32` lanes. +); +impl_f!([f32; 8]: f32x8, m32x8 | f32 | test_v256 | x0, x1, x2, x3, x4, x5, x6, x7 | + From: i8x8, u8x8, i16x8, u16x8 | + /// A 256-bit vector with 8 `f32` lanes. +); +impl_m!([m32; 8]: m32x8 | i32, u8 | test_v256 | x0, x1, x2, x3, x4, x5, x6, x7 | + From: m8x8, m16x8 | + /// A 256-bit vector mask with 8 `m32` lanes. +); + +impl_i!([i64; 4]: i64x4, m64x4 | i64, u8 | test_v256 | x0, x1, x2, x3 | + From: i8x4, u8x4, i16x4, u16x4, i32x4, u32x4 | + /// A 256-bit vector with 4 `i64` lanes. +); +impl_u!([u64; 4]: u64x4, m64x4 | u64, u8 | test_v256 | x0, x1, x2, x3 | + From: u8x4, u16x4, u32x4 | + /// A 256-bit vector with 4 `u64` lanes. +); +impl_f!([f64; 4]: f64x4, m64x4 | f64 | test_v256 | x0, x1, x2, x3 | + From: i8x4, u8x4, i16x4, u16x4, i32x4, u32x4, f32x4 | + /// A 256-bit vector with 4 `f64` lanes. +); +impl_m!([m64; 4]: m64x4 | i64, u8 | test_v256 | x0, x1, x2, x3 | + From: m8x4, m16x4, m32x4 | + /// A 256-bit vector mask with 4 `m64` lanes. +); + +impl_i!([i128; 2]: i128x2, m128x2 | i128, u8 | test_v256 | x0, x1 | + From: i8x2, u8x2, i16x2, u16x2, i32x2, u32x2, i64x2, u64x2 | + /// A 256-bit vector with 2 `i128` lanes. +); +impl_u!([u128; 2]: u128x2, m128x2 | u128, u8 | test_v256 | x0, x1 | + From: u8x2, u16x2, u32x2, u64x2 | + /// A 256-bit vector with 2 `u128` lanes. +); +impl_m!([m128; 2]: m128x2 | i128, u8 | test_v256 | x0, x1 | + From: m8x2, m16x2, m32x2, m64x2 | + /// A 256-bit vector mask with 2 `m128` lanes. +); diff --git a/third_party/rust/packed_simd/src/v32.rs b/third_party/rust/packed_simd/src/v32.rs new file mode 100644 index 000000000000..75a1838e5e0e --- /dev/null +++ b/third_party/rust/packed_simd/src/v32.rs @@ -0,0 +1,29 @@ +//! 32-bit wide vector types + +use crate::*; + +impl_i!([i8; 4]: i8x4, m8x4 | i8, u8 | test_v32 | x0, x1, x2, x3 | + From: | + /// A 32-bit vector with 4 `i8` lanes. +); +impl_u!([u8; 4]: u8x4, m8x4 | u8, u8 | test_v32 | x0, x1, x2, x3 | + From: | + /// A 32-bit vector with 4 `u8` lanes. +); +impl_m!([m8; 4]: m8x4 | i8, u8 | test_v32 | x0, x1, x2, x3 | + From: m16x4, m32x4, m64x4 | + /// A 32-bit vector mask with 4 `m8` lanes. +); + +impl_i!([i16; 2]: i16x2, m16x2 | i16, u8 | test_v32 | x0, x1 | + From: i8x2, u8x2 | + /// A 32-bit vector with 2 `i16` lanes. +); +impl_u!([u16; 2]: u16x2, m16x2 | u16, u8 | test_v32 | x0, x1 | + From: u8x2 | + /// A 32-bit vector with 2 `u16` lanes. +); +impl_m!([m16; 2]: m16x2 | i16, u8 | test_v32 | x0, x1 | + From: m8x2, m32x2, m64x2, m128x2 | + /// A 32-bit vector mask with 2 `m16` lanes. +); diff --git a/third_party/rust/packed_simd/src/v512.rs b/third_party/rust/packed_simd/src/v512.rs new file mode 100644 index 000000000000..4c8c71338aca --- /dev/null +++ b/third_party/rust/packed_simd/src/v512.rs @@ -0,0 +1,99 @@ +//! 512-bit wide vector types +#[rustfmt::skip] + +use crate::*; + +impl_i!([i8; 64]: i8x64, m8x64 | i8, u64 | test_v512 | + x0, x1, x2, x3, x4, x5, x6, x7, x8, x9, x10, x11, x12, x13, x14, x15, + x16, x17, x18, x19, x20, x21, x22, x23, x24, x25, x26, x27, x28, x29, x30, x31, + x32, x33, x34, x35, x36, x37, x38, x39, x40, x41, x42, x43, x44, x45, x46, x47, + x48, x49, x50, x51, x52, x53, x54, x55, x56, x57, x58, x59, x60, x61, x62, x63 | + From: | + /// A 512-bit vector with 64 `i8` lanes. +); +impl_u!([u8; 64]: u8x64, m8x64 | u8, u64 | test_v512 | + x0, x1, x2, x3, x4, x5, x6, x7, x8, x9, x10, x11, x12, x13, x14, x15, + x16, x17, x18, x19, x20, x21, x22, x23, x24, x25, x26, x27, x28, x29, x30, x31, + x32, x33, x34, x35, x36, x37, x38, x39, x40, x41, x42, x43, x44, x45, x46, x47, + x48, x49, x50, x51, x52, x53, x54, x55, x56, x57, x58, x59, x60, x61, x62, x63 | + From: | + /// A 512-bit vector with 64 `u8` lanes. +); +impl_m!([m8; 64]: m8x64 | i8, u64 | test_v512 | + x0, x1, x2, x3, x4, x5, x6, x7, x8, x9, x10, x11, x12, x13, x14, x15, + x16, x17, x18, x19, x20, x21, x22, x23, x24, x25, x26, x27, x28, x29, x30, x31, + x32, x33, x34, x35, x36, x37, x38, x39, x40, x41, x42, x43, x44, x45, x46, x47, + x48, x49, x50, x51, x52, x53, x54, x55, x56, x57, x58, x59, x60, x61, x62, x63 | + From: | + /// A 512-bit vector mask with 64 `m8` lanes. +); + +impl_i!([i16; 32]: i16x32, m16x32 | i16, u32 | test_v512 | + x0, x1, x2, x3, x4, x5, x6, x7, x8, x9, x10, x11, x12, x13, x14, x15, + x16, x17, x18, x19, x20, x21, x22, x23, x24, x25, x26, x27, x28, x29, x30, x31 | + From: i8x32, u8x32 | + /// A 512-bit vector with 32 `i16` lanes. +); +impl_u!([u16; 32]: u16x32, m16x32 | u16, u32 | test_v512 | + x0, x1, x2, x3, x4, x5, x6, x7, x8, x9, x10, x11, x12, x13, x14, x15, + x16, x17, x18, x19, x20, x21, x22, x23, x24, x25, x26, x27, x28, x29, x30, x31 | + From: u8x32 | + /// A 512-bit vector with 32 `u16` lanes. +); +impl_m!([m16; 32]: m16x32 | i16, u32 | test_v512 | + x0, x1, x2, x3, x4, x5, x6, x7, x8, x9, x10, x11, x12, x13, x14, x15, + x16, x17, x18, x19, x20, x21, x22, x23, x24, x25, x26, x27, x28, x29, x30, x31 | + From: m8x32 | + /// A 512-bit vector mask with 32 `m16` lanes. +); + +impl_i!([i32; 16]: i32x16, m32x16 | i32, u16 | test_v512 | + x0, x1, x2, x3, x4, x5, x6, x7, x8, x9, x10, x11, x12, x13, x14, x15 | + From: i8x16, u8x16, i16x16, u16x16 | + /// A 512-bit vector with 16 `i32` lanes. +); +impl_u!([u32; 16]: u32x16, m32x16 | u32, u16 | test_v512 | + x0, x1, x2, x3, x4, x5, x6, x7, x8, x9, x10, x11, x12, x13, x14, x15 | + From: u8x16, u16x16 | + /// A 512-bit vector with 16 `u32` lanes. +); +impl_f!([f32; 16]: f32x16, m32x16 | f32 | test_v512 | + x0, x1, x2, x3, x4, x5, x6, x7, x8, x9, x10, x11, x12, x13, x14, x15 | + From: i8x16, u8x16, i16x16, u16x16 | + /// A 512-bit vector with 16 `f32` lanes. +); +impl_m!([m32; 16]: m32x16 | i32, u16 | test_v512 | + x0, x1, x2, x3, x4, x5, x6, x7, x8, x9, x10, x11, x12, x13, x14, x15 | + From: m8x16, m16x16 | + /// A 512-bit vector mask with 16 `m32` lanes. +); + +impl_i!([i64; 8]: i64x8, m64x8 | i64, u8 | test_v512 | x0, x1, x2, x3, x4, x5, x6, x7 | + From: i8x8, u8x8, i16x8, u16x8, i32x8, u32x8 | + /// A 512-bit vector with 8 `i64` lanes. +); +impl_u!([u64; 8]: u64x8, m64x8 | u64, u8 | test_v512 | x0, x1, x2, x3, x4, x5, x6, x7 | + From: u8x8, u16x8, u32x8 | + /// A 512-bit vector with 8 `u64` lanes. +); +impl_f!([f64; 8]: f64x8, m64x8 | f64 | test_v512 | x0, x1, x2, x3, x4, x5, x6, x7 | + From: i8x8, u8x8, i16x8, u16x8, i32x8, u32x8, f32x8 | + /// A 512-bit vector with 8 `f64` lanes. +); +impl_m!([m64; 8]: m64x8 | i64, u8 | test_v512 | x0, x1, x2, x3, x4, x5, x6, x7 | + From: m8x8, m16x8, m32x8 | + /// A 512-bit vector mask with 8 `m64` lanes. +); + +impl_i!([i128; 4]: i128x4, m128x4 | i128, u8 | test_v512 | x0, x1, x2, x3 | + From: i8x4, u8x4, i16x4, u16x4, i32x4, u32x4, i64x4, u64x4 | + /// A 512-bit vector with 4 `i128` lanes. +); +impl_u!([u128; 4]: u128x4, m128x4 | u128, u8 | test_v512 | x0, x1, x2, x3 | + From: u8x4, u16x4, u32x4, u64x4 | + /// A 512-bit vector with 4 `u128` lanes. +); +impl_m!([m128; 4]: m128x4 | i128, u8 | test_v512 | x0, x1, x2, x3 | + From: m8x4, m16x4, m32x4, m64x4 | + /// A 512-bit vector mask with 4 `m128` lanes. +); diff --git a/third_party/rust/packed_simd/src/v64.rs b/third_party/rust/packed_simd/src/v64.rs new file mode 100644 index 000000000000..bf6b9de61005 --- /dev/null +++ b/third_party/rust/packed_simd/src/v64.rs @@ -0,0 +1,66 @@ +//! 64-bit wide vector types +#[rustfmt::skip] + +use super::*; + +impl_i!([i8; 8]: i8x8, m8x8 | i8, u8 | test_v64 | x0, x1, x2, x3, x4, x5, x6, x7 | + From: | + /// A 64-bit vector with 8 `i8` lanes. +); +impl_u!([u8; 8]: u8x8, m8x8 | u8, u8 | test_v64 | x0, x1, x2, x3, x4, x5, x6, x7 | + From: | + /// A 64-bit vector with 8 `u8` lanes. +); +impl_m!([m8; 8]: m8x8 | i8, u8 | test_v64 | x0, x1, x2, x3, x4, x5, x6, x7 | + From: m16x8, m32x8 | + /// A 64-bit vector mask with 8 `m8` lanes. +); + +impl_i!([i16; 4]: i16x4, m16x4 | i16, u8 | test_v64 | x0, x1, x2, x3 | + From: i8x4, u8x4 | + /// A 64-bit vector with 4 `i16` lanes. +); +impl_u!([u16; 4]: u16x4, m16x4 | u16, u8 | test_v64 | x0, x1, x2, x3 | + From: u8x4 | + /// A 64-bit vector with 4 `u16` lanes. +); +impl_m!([m16; 4]: m16x4 | i16, u8 | test_v64 | x0, x1, x2, x3 | + From: m8x4, m32x4, m64x4 | + /// A 64-bit vector mask with 4 `m16` lanes. +); + +impl_i!([i32; 2]: i32x2, m32x2 | i32, u8 | test_v64 | x0, x1 | + From: i8x2, u8x2, i16x2, u16x2 | + /// A 64-bit vector with 2 `i32` lanes. +); +impl_u!([u32; 2]: u32x2, m32x2 | u32, u8 | test_v64 | x0, x1 | + From: u8x2, u16x2 | + /// A 64-bit vector with 2 `u32` lanes. +); +impl_m!([m32; 2]: m32x2 | i32, u8 | test_v64 | x0, x1 | + From: m8x2, m16x2, m64x2, m128x2 | + /// A 64-bit vector mask with 2 `m32` lanes. +); +impl_f!([f32; 2]: f32x2, m32x2 | f32 | test_v64 | x0, x1 | + From: i8x2, u8x2, i16x2, u16x2 | + /// A 64-bit vector with 2 `f32` lanes. +); + +/* +impl_i!([i64; 1]: i64x1, m64x1 | i64, u8 | test_v64 | x0 | + From: /*i8x1, u8x1, i16x1, u16x1, i32x1, u32x1*/ | // FIXME: primitive to vector conversion + /// A 64-bit vector with 1 `i64` lanes. +); +impl_u!([u64; 1]: u64x1, m64x1 | u64, u8 | test_v64 | x0 | + From: /*u8x1, u16x1, u32x1*/ | // FIXME: primitive to vector conversion + /// A 64-bit vector with 1 `u64` lanes. +); +impl_m!([m64; 1]: m64x1 | i64, u8 | test_v64 | x0 | + From: /*m8x1, m16x1, m32x1, */ m128x1 | // FIXME: unary small vector types + /// A 64-bit vector mask with 1 `m64` lanes. +); +impl_f!([f64; 1]: f64x1, m64x1 | f64 | test_v64 | x0 | + From: /*i8x1, u8x1, i16x1, u16x1, i32x1, u32x1, f32x1*/ | // FIXME: unary small vector types + /// A 64-bit vector with 1 `f64` lanes. +); +*/ diff --git a/third_party/rust/packed_simd/src/vPtr.rs b/third_party/rust/packed_simd/src/vPtr.rs new file mode 100644 index 000000000000..e34cb170eb1c --- /dev/null +++ b/third_party/rust/packed_simd/src/vPtr.rs @@ -0,0 +1,34 @@ +//! Vectors of pointers +#[rustfmt::skip] + +use crate::*; + +impl_const_p!( + [*const T; 2]: cptrx2, msizex2, usizex2, isizex2 | test_v128 | x0, x1 | From: | + /// A vector with 2 `*const T` lanes +); + +impl_mut_p!( + [*mut T; 2]: mptrx2, msizex2, usizex2, isizex2 | test_v128 | x0, x1 | From: | + /// A vector with 2 `*mut T` lanes +); + +impl_const_p!( + [*const T; 4]: cptrx4, msizex4, usizex4, isizex4 | test_v256 | x0, x1, x2, x3 | From: | + /// A vector with 4 `*const T` lanes +); + +impl_mut_p!( + [*mut T; 4]: mptrx4, msizex4, usizex4, isizex4 | test_v256 | x0, x1, x2, x3 | From: | + /// A vector with 4 `*mut T` lanes +); + +impl_const_p!( + [*const T; 8]: cptrx8, msizex8, usizex8, isizex8 | test_v512 | x0, x1, x2, x3, x4, x5, x6, x7 | From: | + /// A vector with 8 `*const T` lanes +); + +impl_mut_p!( + [*mut T; 8]: mptrx8, msizex8, usizex8, isizex8 | test_v512 | x0, x1, x2, x3, x4, x5, x6, x7 | From: | + /// A vector with 8 `*mut T` lanes +); diff --git a/third_party/rust/packed_simd/src/vSize.rs b/third_party/rust/packed_simd/src/vSize.rs new file mode 100644 index 000000000000..b5d891006138 --- /dev/null +++ b/third_party/rust/packed_simd/src/vSize.rs @@ -0,0 +1,53 @@ +//! Vectors with pointer-sized elements + +use crate::codegen::pointer_sized_int::{isize_, usize_}; +use crate::*; + +impl_i!([isize; 2]: isizex2, msizex2 | isize_, u8 | test_v128 | + x0, x1| + From: | + /// A vector with 2 `isize` lanes. +); + +impl_u!([usize; 2]: usizex2, msizex2 | usize_, u8 | test_v128 | + x0, x1| + From: | + /// A vector with 2 `usize` lanes. +); +impl_m!([msize; 2]: msizex2 | isize_, u8 | test_v128 | + x0, x1 | + From: | + /// A vector mask with 2 `msize` lanes. +); + +impl_i!([isize; 4]: isizex4, msizex4 | isize_, u8 | test_v256 | + x0, x1, x2, x3 | + From: | + /// A vector with 4 `isize` lanes. +); +impl_u!([usize; 4]: usizex4, msizex4 | usize_, u8 | test_v256 | + x0, x1, x2, x3| + From: | + /// A vector with 4 `usize` lanes. +); +impl_m!([msize; 4]: msizex4 | isize_, u8 | test_v256 | + x0, x1, x2, x3 | + From: | + /// A vector mask with 4 `msize` lanes. +); + +impl_i!([isize; 8]: isizex8, msizex8 | isize_, u8 | test_v512 | + x0, x1, x2, x3, x4, x5, x6, x7 | + From: | + /// A vector with 8 `isize` lanes. +); +impl_u!([usize; 8]: usizex8, msizex8 | usize_, u8 | test_v512 | + x0, x1, x2, x3, x4, x5, x6, x7 | + From: | + /// A vector with 8 `usize` lanes. +); +impl_m!([msize; 8]: msizex8 | isize_, u8 | test_v512 | + x0, x1, x2, x3, x4, x5, x6, x7 | + From: | + /// A vector mask with 8 `msize` lanes. +); diff --git a/third_party/rust/packed_simd/tests/endianness.rs b/third_party/rust/packed_simd/tests/endianness.rs new file mode 100644 index 000000000000..17a7796b1bf8 --- /dev/null +++ b/third_party/rust/packed_simd/tests/endianness.rs @@ -0,0 +1,268 @@ +#[cfg(target_arch = "wasm32")] +use wasm_bindgen_test::*; + +use packed_simd::*; +use std::{mem, slice}; + +#[cfg_attr(not(target_arch = "wasm32"), test)] +#[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] +fn endian_indexing() { + let v = i32x4::new(0, 1, 2, 3); + assert_eq!(v.extract(0), 0); + assert_eq!(v.extract(1), 1); + assert_eq!(v.extract(2), 2); + assert_eq!(v.extract(3), 3); +} + +#[cfg_attr(not(target_arch = "wasm32"), test)] +#[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] +fn endian_bitcasts() { + #[rustfmt::skip] + let x = i8x16::new( + 0, 1, 2, 3, 4, 5, 6, 7, + 8, 9, 10, 11, 12, 13, 14, 15, + ); + let t: i16x8 = unsafe { mem::transmute(x) }; + let e: i16x8 = if cfg!(target_endian = "little") { + i16x8::new(256, 770, 1284, 1798, 2312, 2826, 3340, 3854) + } else { + i16x8::new(1, 515, 1029, 1543, 2057, 2571, 3085, 3599) + }; + assert_eq!(t, e); +} + +#[cfg_attr(not(target_arch = "wasm32"), test)] +#[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] +fn endian_casts() { + #[rustfmt::skip] + let x = i8x16::new( + 0, 1, 2, 3, 4, 5, 6, 7, + 8, 9, 10, 11, 12, 13, 14, 15, + ); + let t: i16x16 = x.into(); // simd_cast + #[rustfmt::skip] + let e = i16x16::new( + 0, 1, 2, 3, 4, 5, 6, 7, + 8, 9, 10, 11, 12, 13, 14, 15, + ); + assert_eq!(t, e); +} + +#[cfg_attr(not(target_arch = "wasm32"), test)] +#[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] +fn endian_load_and_stores() { + #[rustfmt::skip] + let x = i8x16::new( + 0, 1, 2, 3, 4, 5, 6, 7, + 8, 9, 10, 11, 12, 13, 14, 15, + ); + let mut y: [i16; 8] = [0; 8]; + x.write_to_slice_unaligned(unsafe { slice::from_raw_parts_mut(&mut y as *mut _ as *mut i8, 16) }); + + let e: [i16; 8] = if cfg!(target_endian = "little") { + [256, 770, 1284, 1798, 2312, 2826, 3340, 3854] + } else { + [1, 515, 1029, 1543, 2057, 2571, 3085, 3599] + }; + assert_eq!(y, e); + + let z = i8x16::from_slice_unaligned(unsafe { slice::from_raw_parts(&y as *const _ as *const i8, 16) }); + assert_eq!(z, x); +} + +#[cfg_attr(not(target_arch = "wasm32"), test)] +#[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] +fn endian_array_union() { + union A { + data: [f32; 4], + vec: f32x4, + } + let x: [f32; 4] = unsafe { A { vec: f32x4::new(0., 1., 2., 3.) }.data }; + // As all of these are integer values within the mantissa^1 range, it + // would be very unusual for them to actually fail to compare. + #[allow(clippy::float_cmp)] + { + assert_eq!(x[0], 0_f32); + assert_eq!(x[1], 1_f32); + assert_eq!(x[2], 2_f32); + assert_eq!(x[3], 3_f32); + } + let y: f32x4 = unsafe { A { data: [3., 2., 1., 0.] }.vec }; + assert_eq!(y, f32x4::new(3., 2., 1., 0.)); + + union B { + data: [i8; 16], + vec: i8x16, + } + #[rustfmt::skip] + let x = i8x16::new( + 0, 1, 2, 3, 4, 5, 6, 7, + 8, 9, 10, 11, 12, 13, 14, 15, + ); + let x: [i8; 16] = unsafe { B { vec: x }.data }; + + for (i, v) in x.iter().enumerate() { + assert_eq!(i as i8, *v); + } + + #[rustfmt::skip] + let y = [ + 15, 14, 13, 12, 11, 19, 9, 8, + 7, 6, 5, 4, 3, 2, 1, 0 + ]; + #[rustfmt::skip] + let e = i8x16::new( + 15, 14, 13, 12, 11, 19, 9, 8, + 7, 6, 5, 4, 3, 2, 1, 0 + ); + let z = unsafe { B { data: y }.vec }; + assert_eq!(z, e); + + union C { + data: [i16; 8], + vec: i8x16, + } + #[rustfmt::skip] + let x = i8x16::new( + 0, 1, 2, 3, 4, 5, 6, 7, + 8, 9, 10, 11, 12, 13, 14, 15, + ); + let x: [i16; 8] = unsafe { C { vec: x }.data }; + + let e: [i16; 8] = if cfg!(target_endian = "little") { + [256, 770, 1284, 1798, 2312, 2826, 3340, 3854] + } else { + [1, 515, 1029, 1543, 2057, 2571, 3085, 3599] + }; + assert_eq!(x, e); +} + +#[cfg_attr(not(target_arch = "wasm32"), test)] +#[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)] +fn endian_tuple_access() { + type F32x4T = (f32, f32, f32, f32); + union A { + data: F32x4T, + vec: f32x4, + } + let x: F32x4T = unsafe { A { vec: f32x4::new(0., 1., 2., 3.) }.data }; + // As all of these are integer values within the mantissa^1 range, it + // would be very unusual for them to actually fail to compare. + #[allow(clippy::float_cmp)] + { + assert_eq!(x.0, 0_f32); + assert_eq!(x.1, 1_f32); + assert_eq!(x.2, 2_f32); + assert_eq!(x.3, 3_f32); + } + let y: f32x4 = unsafe { A { data: (3., 2., 1., 0.) }.vec }; + assert_eq!(y, f32x4::new(3., 2., 1., 0.)); + + #[rustfmt::skip] + type I8x16T = (i8, i8, i8, i8, i8, i8, i8, i8, i8, i8, i8, i8, i8, i8, i8, i8); + union B { + data: I8x16T, + vec: i8x16, + } + + #[rustfmt::skip] + let x = i8x16::new( + 0, 1, 2, 3, 4, 5, 6, 7, + 8, 9, 10, 11, 12, 13, 14, 15, + ); + let x: I8x16T = unsafe { B { vec: x }.data }; + + assert_eq!(x.0, 0); + assert_eq!(x.1, 1); + assert_eq!(x.2, 2); + assert_eq!(x.3, 3); + assert_eq!(x.4, 4); + assert_eq!(x.5, 5); + assert_eq!(x.6, 6); + assert_eq!(x.7, 7); + assert_eq!(x.8, 8); + assert_eq!(x.9, 9); + assert_eq!(x.10, 10); + assert_eq!(x.11, 11); + assert_eq!(x.12, 12); + assert_eq!(x.13, 13); + assert_eq!(x.14, 14); + assert_eq!(x.15, 15); + + #[rustfmt::skip] + let y = ( + 15, 14, 13, 12, 11, 10, 9, 8, + 7, 6, 5, 4, 3, 2, 1, 0 + ); + let z: i8x16 = unsafe { B { data: y }.vec }; + #[rustfmt::skip] + let e = i8x16::new( + 15, 14, 13, 12, 11, 10, 9, 8, + 7, 6, 5, 4, 3, 2, 1, 0 + ); + assert_eq!(e, z); + + #[rustfmt::skip] + type I16x8T = (i16, i16, i16, i16, i16, i16, i16, i16); + union C { + data: I16x8T, + vec: i8x16, + } + + #[rustfmt::skip] + let x = i8x16::new( + 0, 1, 2, 3, 4, 5, 6, 7, + 8, 9, 10, 11, 12, 13, 14, 15, + ); + let x: I16x8T = unsafe { C { vec: x }.data }; + + let e: [i16; 8] = if cfg!(target_endian = "little") { + [256, 770, 1284, 1798, 2312, 2826, 3340, 3854] + } else { + [1, 515, 1029, 1543, 2057, 2571, 3085, 3599] + }; + assert_eq!(x.0, e[0]); + assert_eq!(x.1, e[1]); + assert_eq!(x.2, e[2]); + assert_eq!(x.3, e[3]); + assert_eq!(x.4, e[4]); + assert_eq!(x.5, e[5]); + assert_eq!(x.6, e[6]); + assert_eq!(x.7, e[7]); + + #[rustfmt::skip] + #[repr(C)] + #[derive(Copy ,Clone)] + pub struct Tup(pub i8, pub i8, pub i16, pub i8, pub i8, pub i16, + pub i8, pub i8, pub i16, pub i8, pub i8, pub i16); + + union D { + data: Tup, + vec: i8x16, + } + + #[rustfmt::skip] + let x = i8x16::new( + 0, 1, 2, 3, 4, 5, 6, 7, + 8, 9, 10, 11, 12, 13, 14, 15, + ); + let x: Tup = unsafe { D { vec: x }.data }; + + let e: [i16; 12] = if cfg!(target_endian = "little") { + [0, 1, 770, 4, 5, 1798, 8, 9, 2826, 12, 13, 3854] + } else { + [0, 1, 515, 4, 5, 1543, 8, 9, 2571, 12, 13, 3599] + }; + assert_eq!(x.0 as i16, e[0]); + assert_eq!(x.1 as i16, e[1]); + assert_eq!(x.2 as i16, e[2]); + assert_eq!(x.3 as i16, e[3]); + assert_eq!(x.4 as i16, e[4]); + assert_eq!(x.5 as i16, e[5]); + assert_eq!(x.6 as i16, e[6]); + assert_eq!(x.7 as i16, e[7]); + assert_eq!(x.8 as i16, e[8]); + assert_eq!(x.9 as i16, e[9]); + assert_eq!(x.10 as i16, e[10]); + assert_eq!(x.11 as i16, e[11]); +}