Benchmark against simd math functions from core::intrinsics::simd

[NamorNiradnug]

No description provided.

[NamorNiradnug]

На моем ноутбуке выдает такое:

test llvm_vec_exp      ... bench:     944,499 ns/iter (+/- 7,318)
test llvm_vec_sin      ... bench:     961,830 ns/iter (+/- 10,572)
test scalar_sin_bench  ... bench:     985,468 ns/iter (+/- 7,642)
test vec_sin_bench     ... bench:     133,412 ns/iter (+/- 976)
test vec_sin_cos_bench ... bench:     146,397 ns/iter (+/- 1,339)

Ура, оно работает сильно хуже, чем то, что у меня VCL! Юхуу

[NamorNiradnug]

test bench_cos_f32_core_simd ... bench:     961,397 ns/iter (+/- 25,370)
test bench_cos_f32_scalar    ... bench:     970,129 ns/iter (+/- 17,319)
test bench_cos_f32_vec       ... bench:     131,046 ns/iter (+/- 2,032)
test bench_cos_f64_core_simd ... bench:   1,973,315 ns/iter (+/- 18,569)
test bench_cos_f64_scalar    ... bench:   1,899,518 ns/iter (+/- 11,374)
test bench_cos_f64_vec       ... bench:     500,811 ns/iter (+/- 5,891)
test bench_sin_f32_core_simd ... bench:     968,157 ns/iter (+/- 8,476)
test bench_sin_f32_scalar    ... bench:   1,010,247 ns/iter (+/- 19,708)
test bench_sin_f32_vec       ... bench:     136,389 ns/iter (+/- 1,070)
test bench_sin_f64_core_simd ... bench:   1,935,117 ns/iter (+/- 7,962)
test bench_sin_f64_scalar    ... bench:   1,867,211 ns/iter (+/- 18,490)
test bench_sin_f64_vec       ... bench:     490,637 ns/iter (+/- 5,658)

даже хуже, чем скалярное для f64...

... что-то мне это не нравится

[NamorNiradnug]

UPD: ну, просто в Rust эти интринсики на работают: rust-lang/portable-simd#76.

[asokol123]

Огонь