python-pillow/Pillow
1
1
Fork 0
mirror of https://github.com/python-pillow/Pillow.git synced 2025-08-20 20:24:45 +03:00
Code Issues Packages Projects Releases Wiki Activity

SIMD ColorLUT. improve performance by preliminary index calculation

Browse Source
This commit is contained in:
Alexander 2018-03-27 09:50:38 +03:00 committed by Alexander Karpinsky
parent f50dba4aad
commit c7e3b0ead9
1 changed files with 14 additions and 8 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

18
src/libImaging/ColorLUT.c
View File

@ -104,17 +104,23 @@ ImagingColorLUT3D_linear(
ImagingSectionEnter(&cookie); ImagingSectionEnter(&cookie);
for (y = 0; y < imOut->ysize; y++) { for (y = 0; y < imOut->ysize; y++) {
UINT8 *rowIn = (UINT8 *)imIn->image[y]; UINT8* rowIn = (UINT8 *)imIn->image[y];
char *rowOut = (char *)imOut->image[y]; UINT32* rowOut = (UINT32 *)imOut->image[y];
for (x = 0; x < imOut->xsize; x++) {
__m128i index = _mm_mullo_epi32(scale, __m128i index = _mm_mullo_epi32(scale,
_mm_cvtepu8_epi32(*(__m128i *) &rowIn[x*4])); _mm_cvtepu8_epi32(*(__m128i *) &rowIn[0]));
__m128i shift = _mm_srli_epi32(
_mm_and_si128(scale_mask, index), (SCALE_BITS - SHIFT_BITS));
int idx = table_channels * _mm_extract_epi32( int idx = table_channels * _mm_extract_epi32(
_mm_hadd_epi32(_mm_hadd_epi32( _mm_hadd_epi32(_mm_hadd_epi32(
_mm_madd_epi16(index_mul, _mm_srli_epi32(index, SCALE_BITS)), _mm_madd_epi16(index_mul, _mm_srli_epi32(index, SCALE_BITS)),
_mm_setzero_si128()), _mm_setzero_si128()), 0); _mm_setzero_si128()), _mm_setzero_si128()), 0);
for (x = 0; x < imOut->xsize; x++) {
__m128i next_index = _mm_mullo_epi32(scale,
_mm_cvtepu8_epi32(*(__m128i *) &rowIn[x*4 + 4]));
int next_idx = table_channels * _mm_extract_epi32(
_mm_hadd_epi32(_mm_hadd_epi32(
_mm_madd_epi16(index_mul, _mm_srli_epi32(next_index, SCALE_BITS)),
_mm_setzero_si128()), _mm_setzero_si128()), 0);
__m128i shift = _mm_srli_epi32(
_mm_and_si128(scale_mask, index), (SCALE_BITS - SHIFT_BITS));
__m128i shift1D, shift2D, shift3D; __m128i shift1D, shift2D, shift3D;
__m128i source, left, right, result; __m128i source, left, right, result;
__m128i leftleft, leftright, rightleft, rightright; __m128i leftleft, leftright, rightleft, rightright;