opt_impl - Branch - jvdd/argminmax

feat: optimize + add `f16` scalar implementations

#44Merged

Comparing

opt_impl

(

dcc1db9

) with

main

(

9c9f373

)

+10%

Improvement: 9

Untouched: 147

Benchmarks

156 total

scalar_f16_argmin_rn

benches/bench_f16_return_nan.rs::benches::argminmax_rn_f16_random_array_long

+10%

634 µs577.1 µs

scalar_f16_argmax_rn

benches/bench_f16_return_nan.rs::benches::argminmax_rn_f16_random_array_long

+10%

634 µs577.1 µs

sse_f64_argmax_rn

benches/bench_f64_return_nan.rs::benches::argminmax_rn_f64_random_array_long

+9%

701.1 µs644.1 µs

sse_f64_argmin_rn

benches/bench_f64_return_nan.rs::benches::argminmax_rn_f64_random_array_long

+9%

701.1 µs644.1 µs

sse_f64_argminmax_rn

benches/bench_f64_return_nan.rs::benches::argminmax_rn_f64_random_array_long

+9%

701 µs644.1 µs

sse_i64_argmin

benches/bench_i64.rs::benches::argminmax_i64_random_array_long

+8%

551.6 µs508.9 µs

sse_u64_argmin

benches/bench_u64.rs::benches::argminmax_u64_random_array_long

+8%

565.9 µs523.2 µs

sse_i64_argminmax

benches/bench_i64.rs::benches::argminmax_i64_random_array_long

+8%

608.6 µs566 µs

sse_u64_argminmax

benches/bench_u64.rs::benches::argminmax_u64_random_array_long

+7%

622.9 µs580.2 µs

avx2_u64_argmin

benches/bench_u64.rs::benches::argminmax_u64_random_array_long

+2%

437.9 µs430.8 µs

avx2_i64_argmin

benches/bench_i64.rs::benches::argminmax_i64_random_array_long

+2%

437.9 µs430.7 µs

avx2_i64_argminmax

benches/bench_i64.rs::benches::argminmax_i64_random_array_long

+2%

459.4 µs452.2 µs

avx2_u64_argminmax

benches/bench_u64.rs::benches::argminmax_u64_random_array_long

+2%

459.4 µs452.3 µs

impl_i64_argmin

benches/bench_i64.rs::benches::argminmax_i64_random_array_long

+2%

459.6 µs452.5 µs

impl_i64_argminmax

benches/bench_i64.rs::benches::argminmax_i64_random_array_long

+2%

459.6 µs452.5 µs

impl_i64_argmax

benches/bench_i64.rs::benches::argminmax_i64_random_array_long

+2%

459.6 µs452.5 µs

impl_u64_argmin

benches/bench_u64.rs::benches::argminmax_u64_random_array_long

+2%

459.6 µs452.5 µs

impl_u64_argminmax

benches/bench_u64.rs::benches::argminmax_u64_random_array_long

+2%

459.6 µs452.5 µs

impl_u64_argmax

benches/bench_u64.rs::benches::argminmax_u64_random_array_long

+2%

459.6 µs452.5 µs

avx2_f64_argmin_rn

benches/bench_f64_return_nan.rs::benches::argminmax_rn_f64_random_array_long

+1%

495 µs487.8 µs

avx2_f64_argmax_rn

benches/bench_f64_return_nan.rs::benches::argminmax_rn_f64_random_array_long

+1%

495 µs487.8 µs

avx2_f64_argminmax_rn

benches/bench_f64_return_nan.rs::benches::argminmax_rn_f64_random_array_long

+1%

494.9 µs487.7 µs

impl_f64_argminmax_rn

benches/bench_f64_return_nan.rs::benches::argminmax_rn_f64_random_array_long

+1%

495.1 µs488 µs

impl_f64_argmax_rn

benches/bench_f64_return_nan.rs::benches::argminmax_rn_f64_random_array_long

+1%

495.1 µs488 µs

impl_f64_argmin_rn

benches/bench_f64_return_nan.rs::benches::argminmax_rn_f64_random_array_long

+1%

495.1 µs488 µs

Commits

Click on a commit to change the comparison range

Base

main

9c9f373

-21.22%

perf: scalar first check >

c5f02ee

3 years ago

by jvdd

-9.8%

perf: let compiler use argminmax impl -> else if => if

391690d

3 years ago

by jvdd

:broom:

83ea5a8

3 years ago

by jvdd

-2.93%

:pray:

3d9b3d7

3 years ago

by jvdd

:pray:

43a74f4

3 years ago

by jvdd

+33.95%

:broom:

66ae9b0

3 years ago

by jvdd

-4.3%

:pray:

aef4ffe

3 years ago

by jvdd

×1,100

:rocket:

68645f1

3 years ago

by jvdd

:broom: add scalar ignore nan impl

f4feebd

3 years ago

by jvdd

feat: test & implement scalar f16 ignore nans

dcc1db9

3 years ago

by jvdd

Home Terms Privacy Docs