python-pillow/Pillow
1
1
Fork 0
mirror of https://github.com/python-pillow/Pillow.git synced 2025-08-21 04:34:47 +03:00
Code Issues Packages Projects Releases Wiki Activity

SIMD AlphaComposite. fast div aproximation

Browse Source
This commit is contained in:
Alexander 2017-01-23 04:56:37 +03:00 committed by Alexander Karpinsky
parent a48b385426
commit 5a939e33f7
1 changed files with 8 additions and 4 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

12
src/libImaging/AlphaComposite.c
View File

@ -106,8 +106,10 @@ ImagingAlphaComposite(Imaging imDst, Imaging imSrc) {
mm_out_a = _mm256_add_epi32(_mm256_mullo_epi16(mm_src_a, vmm_max_alpha), mm_blend); mm_out_a = _mm256_add_epi32(_mm256_mullo_epi16(mm_src_a, vmm_max_alpha), mm_blend);
mm_coef1 = _mm256_mullo_epi32(mm_src_a, vmm_max_alpha2); mm_coef1 = _mm256_mullo_epi32(mm_src_a, vmm_max_alpha2);
// 8 bits // 8 bits
mm_coef1 = _mm256_cvtps_epi32(_mm256_div_ps(_mm256_cvtepi32_ps(mm_coef1), mm_coef1 = _mm256_cvtps_epi32(
_mm256_cvtepi32_ps(mm_out_a))); _mm256_mul_ps(_mm256_cvtepi32_ps(mm_coef1),
_mm256_rcp_ps(_mm256_cvtepi32_ps(mm_out_a)))
);
// 8 bits // 8 bits
mm_coef2 = _mm256_sub_epi32(vmm_max_alpha, mm_coef1); mm_coef2 = _mm256_sub_epi32(vmm_max_alpha, mm_coef1);
@ -169,8 +171,10 @@ ImagingAlphaComposite(Imaging imDst, Imaging imSrc) {
// coef1 = src->a * 255 * 255 / outa // coef1 = src->a * 255 * 255 / outa
mm_coef1 = _mm_mullo_epi32(mm_src_a, mm_max_alpha2); mm_coef1 = _mm_mullo_epi32(mm_src_a, mm_max_alpha2);
// [8] xx xx xx c3 xx xx xx c2 xx xx xx c1 xx xx xx c0 // [8] xx xx xx c3 xx xx xx c2 xx xx xx c1 xx xx xx c0
mm_coef1 = _mm_cvtps_epi32(_mm_div_ps(_mm_cvtepi32_ps(mm_coef1), mm_coef1 = _mm_cvtps_epi32(
_mm_cvtepi32_ps(mm_out_a))); _mm_mul_ps(_mm_cvtepi32_ps(mm_coef1),
_mm_rcp_ps(_mm_cvtepi32_ps(mm_out_a)))
);
// [8] xx xx xx c3 xx xx xx c2 xx xx xx c1 xx xx xx c0 // [8] xx xx xx c3 xx xx xx c2 xx xx xx c1 xx xx xx c0
mm_coef2 = _mm_sub_epi32(mm_max_alpha, mm_coef1); mm_coef2 = _mm_sub_epi32(mm_max_alpha, mm_coef1);