NEON : Complex operations from Armv8.3-a #1077
Conversation
|
About the binary operations and 16-bit floating points: they might work when |
|
Thanks for your help. I have fixed it and pushed the code again! |
simde/arm/neon/cmla_lane.h
Outdated
| result = simde_float16x4_from_private(r_); | ||
| return result; |
simde/arm/neon/cmla_lane.h
Outdated
| #if defined(SIMDE_ARM_NEON_A32V8_ENABLE_NATIVE_ALIASES) | ||
| #undef vcmla_lane_f16 | ||
| #define vcmla_lane_f16(r, a, b, lane) simde_vcmla_lane_f16(r, a, b, lane) | ||
| #endif |
There was a problem hiding this comment.
| #if defined(SIMDE_ARM_NEON_A32V8_ENABLE_NATIVE_ALIASES) | |
| #undef vcmla_lane_f16 | |
| #define vcmla_lane_f16(r, a, b, lane) simde_vcmla_lane_f16(r, a, b, lane) | |
| #endif | |
| #if defined(SIMDE_ARM_NEON_A32V8_ENABLE_NATIVE_ALIASES) | |
| #undef vcmla_lane_f16 | |
| #define vcmla_lane_f16(r, a, b, lane) simde_vcmla_lane_f16(r, a, b, lane) | |
| #endif |
simde/arm/neon/cmla_rot90_lane.h
Outdated
| simde_float32x4_private r_ = | ||
| simde_float32x4_to_private(simde_vcvt_f32_f16(r)), |
There was a problem hiding this comment.
This style is more readable
| simde_float32x4_private r_ = | |
| simde_float32x4_to_private(simde_vcvt_f32_f16(r)), | |
| simde_float32x4_private r_ = simde_float32x4_to_private( | |
| simde_vcvt_f32_f16(r)), |
test/arm/neon/cadd_rot270.c
Outdated
|
|
||
|
|
There was a problem hiding this comment.
In general, there are several instances of too many blank lines
test/arm/neon/cmla_rot180_lane.c
Outdated
|
|
||
| // simde_float32x4_t r = simde_vcmlaq_rot180_laneq_f32(r_, a, b, | ||
| // test_vec[i].lane); simde_test_arm_neon_write_f32x4(2, r, | ||
| // SIMDE_TEST_VEC_POS_LAST); |
There was a problem hiding this comment.
Please tidy up your test code as well; thank you!
| // simde_float32x4_t r = simde_vcmlaq_rot180_laneq_f32(r_, a, b, | |
| // test_vec[i].lane); simde_test_arm_neon_write_f32x4(2, r, | |
| // SIMDE_TEST_VEC_POS_LAST); |
There was a problem hiding this comment.
Hello! I've cleaned up the code and reformatted it using Clang-Format with the LLVM style. Apologies for the coding style and redundant comments.
There was a problem hiding this comment.
Thanks for the clean up. SIMDe does not (yet) have an official clang-format style. I see that the LLVM style still uses ColumnLimit: 80, which I find to be too narrow.
I will add that guidance to https://github.com/simd-everywhere/simde/wiki/Coding-Style for future contributors
There was a problem hiding this comment.
It looks like all the compiler issues are fixed (I think the rpm-build:fedora-rawhide-i386 failure is not your fault). Please fix the formatting errors and merge/re-base on the latest commits in https://github.com/simd-everywhere/simde/tree/master ; I'll then merge this PR! Thank you @wewe5215 !
|
@mr-c Hello! I have reformatted the code and rebased it on the latest commit. I really appreciate your patience! |
…ne_f{16/32} and vcmlaq_laneq_f{16/32}
…nd vcmlaq_rot90_lane_f{16/32} and vcmlaq_rot90_laneq_f{16/32}
… and vcmlaq_rot180_lane_f{16/32} and vcmlaq_rot180_laneq_f{16/32}
…_f16 and vcmla{/q}_lane{/q}_f16
… vcmla{/q}_lane{/q}_f16
mr-c
force-pushed
the
complex_operation
branch
from
c51131a to
19ed113
Compare
This pull request includes initial implementations and corresponding test cases listed below
Sorry for the typo in commit fa9a14d.
It is [Neon] Add vcmla_rot270_lane_f{16/32} and vcmla_rot270_laneq_f{16/32} and vcmlaq_rot270_lane_f{16/32} and vcmlaq_rot270_laneq_f{16/32}