drewcassidy/quicktex
1
0
Fork 0
mirror of https://github.com/drewcassidy/quicktex.git synced 2024-09-13 06:37:34 +00:00
Code Issues Projects Releases Activity

Add emulated 128-bit support and fix x86

Browse Source
This commit is contained in:
Andrew Cassidy 2022-05-15 20:03:05 -07:00
parent f7b0cbe76b
commit 8168d6e249
1 changed files with 7 additions and 4 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

11
quicktex/VecUtil.h
View File

@ -18,9 +18,7 @@
*/
#pragma once
#include <hwy/highway.h>
namespace hn = hwy::HWY_NAMESPACE;
#if HWY_TARGET == HWY_NEON
@ -52,6 +50,11 @@ inline int32_t WideningSumS16(const Vec_s16 v) {
#if HWY_TARGET == HWY_SCALAR
// In Scalar mode this is a no-op, since there's only one lane
return (int32_t)v.raw;
#elif HWY_TARGET == HWY_EMU128
// In emulated 128-bit mode, do the addition serially
int acc = 0;
for (unsigned i = 0; i < hn::MaxLanes(TagS16); i++) { acc += v.raw[i]; }
return acc;
#elif HWY_TARGET == HWY_NEON
static_assert(hn::MaxLanes(TagS16) == 8);
static_assert(hn::MaxLanes(TagS32) == 4);
@ -66,7 +69,7 @@ inline int32_t WideningSumS16(const Vec_s16 v) {
static_assert(hn::MaxLanes(TagS32) == 8);
// Pairwise widening sum with multiply by 1, then sum all N/2 widened lanes
auto paired = Vec_s32(_mm256_madd_epi16(v.raw, __mm256_set1_epi16(1)));
auto paired = Vec_s32{_mm256_madd_epi16(v.raw, __mm256_set1_epi16(1))};
auto sums = SumOfLanes(paired);
return hn::GetLane(sums);
#else
@ -74,7 +77,7 @@ inline int32_t WideningSumS16(const Vec_s16 v) {
static_assert(hn::MaxLanes(TagS32) == 4);
// Pairwise widening sum with multiply by 1, then sum all N/2 widened lanes
auto paired = Vec_s32(_mm_madd_epi16(v.raw, _mm_set1_epi16(1)));
auto paired = Vec_s32{_mm_madd_epi16(v.raw, _mm_set1_epi16(1))};
auto sums = SumOfLanes(paired);
return hn::GetLane(sums);
#endif