Skip to content

vectorize GetBestLengths2() hotspot loop. #146

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
yumeyao wants to merge 1 commit into
base: master
Choose a base branch
from
Open

vectorize GetBestLengths2() hotspot loop. #146

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
40 changes: 40 additions & 0 deletions src/zopfli/squeeze.c
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -38,6 +38,10 @@ Author: jyrki.alakuijala@gmail.com (Jyrki Alakuijala)
#include "../LzFind.h"
#include "../threadLocal.h"

#ifdef __SSE2__
#include <immintrin.h>
#endif

static void CopyStats(const SymbolStats* source, SymbolStats* dest) {
memcpy(dest->litlens, source->litlens, 288 * sizeof(dest->litlens[0]));
memcpy(dest->dists, source->dists, 32 * sizeof(dest->dists[0]));
Expand Down Expand Up @@ -370,6 +374,42 @@ static void GetBestLengths2(const unsigned char* in, size_t instart, size_t inen
unsigned dist = *mp++;
float price2 = price + disttable[dist];
dist <<=9;
#if defined(__AVX__)
for (; curr + 8 < len; curr+=8) {
__m256 x8 = _mm256_add_ps(_mm256_set1_ps(price2), _mm256_loadu_ps(&litlentable[curr]));
__m256 vcost = _mm256_loadu_ps(&costs[j + curr]);
_mm256_storeu_ps(&costs[j + curr], _mm256_min_ps(x8, vcost));
#if defined(__AVX2__)
__m256i vlength = _mm256_add_epi32(_mm256_set1_epi32(curr + dist), _mm256_setr_epi32(0,1,2,3,4,5,6,7));
vlength = _mm256_blendv_epi8(_mm256_loadu_si256((__m256i*)&length_array[j + curr]), vlength, cmp_mask);
#else
__m128i vlength_low = _mm_add_epi32(_mm_set1_epi32(curr + dist), _mm_setr_epi32(0, 1, 2, 3));
__m128i vlength_high = _mm_add_epi32(_mm_set1_epi32(curr + dist), _mm_setr_epi32(4, 5, 6, 7));
__m256i vlength = _mm256_set_m128i(vlength_high, vlength_low);
vlength = _mm256_castps_si256(_mm256_blendv_ps(_mm256_castsi256_ps(_mm256_loadu_si256((__m256i*)&length_array[j + curr])),
_mm256_castsi256_ps(vlength), _mm256_cmp_ps(x8, vcost, _CMP_LT_OQ)));
#endif
_mm256_storeu_si256((__m256i*)&length_array[j + curr], vlength);
}
#endif
#if defined(__SSE2__)
for (; curr + 4 < len; curr += 4) {
__m128 x4 = _mm_add_ps(_mm_set1_ps(price2), _mm_loadu_ps(&litlentable[curr]));
__m128 vcost = _mm_loadu_ps(&costs[j + curr]);
_mm_storeu_ps(&costs[j + curr], _mm_min_ps(x4, vcost));
__m128i vlength = _mm_add_epi32(_mm_set1_epi32(curr + dist), _mm_setr_epi32(0, 1, 2, 3));
__m128i cmp_mask = _mm_castps_si128(_mm_cmplt_ps(x4, vcost));
#if defined(__SSE4_1__)
vlength = _mm_blendv_epi8(_mm_loadu_si128((__m128i*)&length_array[j + curr]), vlength, cmp_mask);
#else
vlength = _mm_or_si128(_mm_and_si128(vlength, cmp_mask), _mm_andnot_si128(cmp_mask, old_vlength));
#endif
_mm_storeu_si128((__m128i*)&length_array[j + curr], vlength);
}
#ifdef __GNUC__
#pragma GCC unroll 1 /* no unroll */
#endif
#endif
for (; curr <= len; curr++) {
float x = price2 + litlentable[curr];
if (x < costs[j + curr]){
Expand Down