NEON: Implement some f16XN types and f16 related intrinsics. #1071

yyctw · 2023-10-03T08:28:05Z

Hi all, this is Eric from Andes Technology Corporation. This PR includes

Add the types simde_float16x4x2_t and simde_float16x8x2_t
Add 105 initial implementations and corresponding test cases in 35 families which are listed below:

mul, sub, ext, div, min, max, fma, abs, cgt, clt, clez, padd, pmax
get_lane, get_low, get_high, mul_lane, mul_n, set_lane
combine, cvt, ld2, ld1_dup, st2, zip1, zip2, zip, reinterpret
rsqrts, recps, recpe, rndn, rsqrte, sqrt, qrshrnh_n

All CI tests have passed on my fork! Thanks for reading and any recommendations are welcome!

mr-c · 2023-10-03T15:59:49Z

@yyctw Thank you very much for this contribution! Can you rebase your branch to bring it up to date?

yyctw · 2023-10-04T00:06:16Z

@yyctw Thank you very much for this contribution! Can you rebase your branch to bring it up to date?

Thank you for your suggestion! I've already rebased my branch to ensure it's up to date.

mr-c

Looking great! One request about the copyright lines

simde/arm/neon/div.h

yyctw added 29 commits October 4, 2023 07:59

[NEON] Add vmulq_f16 and vmul_f16.

61a0796

[NEON] Add vmulq_f16 and vmul_f16 test.

93c46e8

[NEON] Add vget_lane_f16 and vgetq_lane_f16.

6da9af3

[NEON] Add vsubh_f16 and vsub_f16.

b7f061e

[NEON] Add vextq_f16.

a47dcb1

[NEON] Add vget_low_f16.

5729c35

[NEON] Add vmulq_lane_f16.

70e5111

[NEON] Add vmul_n_f16.

216586a

[NEON] Add vget_high_f16.

5450c94

[NEON] Add vsetq_lane_f16.

58f0c43

[NEON] Add vcombine_f16.

b43ada1

[NEON] Add vcvtaq_s32_f32, vcvtas_s32_f32, and vcvta_s32_f32.

3d9934c

[NEON] Add vpadd_f16.

b7733b9

[NEON] Add vuzp1_f16.

e833ddc

[NEON] Add vuzp2_f16.

c60221d

[NEON] Add vmaxq_f16 and vmax_f16.

2a896e7

[NEON] Add vcvtas_u32_f32, vcvta_u32_f32, vcvtaq_u32_f32.

32cda10

[NEON] Add type simde_float16x8x2_t.

d519b56

[NEON] Add vld2q_f16.

478356e

[NEON] Add vld1q_dup_f16.

ab7f4fb

[NEON] Add vpmax_f16.

ca01aba

[NEON] Add vrsqrtsq_f16, vrsqrtsh_f16, vrsqrts_f16.

eb8ee00

[NEON] Add vcgtq_f16, vcgt_f16, vcgth_f16.

dd34fd2

[NEON] Add vdiv_f32 and vdivq_f32.

5c5a172

[NEON] Add vrecps_f16 and vrecpsq_f16.

106bfd5

[NEON] Add vset_lane_f16.

a131556

[NEON] Add vrecpe_f16, vrecpeq_f16.

70b035f

[NEON] Add vfmaq_f16.

3e712b7

[NEON] Add vabsq_f16 and vabs_f16.

9e77aa6

yyctw added 17 commits October 4, 2023 07:59

[NEON] Add vmin_f16 and vminq_f16.

e8d6248

[NEON] Add vclt_f16.

32cae83

[NEON] Add vclezq_f16, vclez_f16, and vclezh_f16.

50749e5

[NEON] Add vzip_f16 and vzipq_f16.

576a2ff

[NEON] Add vzip1_f16 and vzip1q_f16.

5e78de6

[NEON] Add vzip2_f16 and vzip2q_f16.

c8e87d6

[NEON] Add vst2_f16 and vst2q_f16.

873c97f

[NEON] Add type simde_float16x4x2.

4381cc6

[NEON] Add 16 intrinsics of vreinterpret series.

013bc55

[SIMDE] Add sqrtl() in simde_math_sqrtl.

f85ab38

[NEON] Add vrndnq_f16, vrndns_f16, and vrndn_f16.

11ea848

[NEON] Add sqrt in meson.build.

c9d8953

[NEON] Add 7 sqrt related intrinsics.

25c1052

[Fix] Add the judge whether define sqrt() or not.

edb38c3

[NEON] Add vrsqrteq_f16, vrsqrte_f16, and vrsqrteh_f16.

6fef767

[NEON] Add vqrshrnh_n_s16 and vqrshrnh_n_u16.

9addc89

[NEON] Add vqrshrunh_n_s16.

c44cfac

yyctw force-pushed the fp16 branch from c61fb07 to fb1e779 Compare October 4, 2023 00:00

yyctw closed this Oct 4, 2023

yyctw reopened this Oct 4, 2023

yyctw force-pushed the fp16 branch from fb1e779 to bfe4e00 Compare October 4, 2023 01:16

[Fix] Add new conditions for fp16 intrinsics.

329a841

yyctw force-pushed the fp16 branch from bfe4e00 to 329a841 Compare October 4, 2023 02:28

mr-c requested changes Oct 4, 2023

View reviewed changes

simde/arm/neon/div.h Outdated Show resolved Hide resolved

[License] Add Copyright.

bea3422

yyctw requested a review from mr-c October 6, 2023 08:03

mr-c enabled auto-merge (squash) October 6, 2023 08:05

mr-c approved these changes Oct 6, 2023

View reviewed changes

mr-c merged commit aae2245 into simd-everywhere:master Oct 6, 2023
79 of 80 checks passed

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

NEON: Implement some f16XN types and f16 related intrinsics. #1071

NEON: Implement some f16XN types and f16 related intrinsics. #1071

yyctw commented Oct 3, 2023

mr-c commented Oct 3, 2023

yyctw commented Oct 4, 2023 •

edited

mr-c left a comment

NEON: Implement some f16XN types and f16 related intrinsics. #1071

NEON: Implement some f16XN types and f16 related intrinsics. #1071

Conversation

yyctw commented Oct 3, 2023

mr-c commented Oct 3, 2023

yyctw commented Oct 4, 2023 • edited

mr-c left a comment

Choose a reason for hiding this comment

yyctw commented Oct 4, 2023 •

edited