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 Filter. Macros for AVX

Browse Source
This commit is contained in:
Alexander 2017-09-04 16:20:49 +03:00 committed by Alexander Karpinsky
parent 2329ee5605
commit efe722fce3
1 changed files with 38 additions and 48 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

86
src/libImaging/Filter.c
View File

@ -245,6 +245,31 @@ ImagingFilter3x3(Imaging imOut, Imaging im, const float* kernel,
UINT32* in1 = (UINT32*) im->image[y+1]; UINT32* in1 = (UINT32*) im->image[y+1];
UINT32* out = (UINT32*) imOut->image[y]; UINT32* out = (UINT32*) imOut->image[y];
#if defined(__AVX2__) #if defined(__AVX2__)
#define MM256_LOAD(row, x) \
pix0##row = _mm256_cvtepi32_ps(_mm256_cvtepu8_epi32(*(__m128i*) &in1[x])); \
pix1##row = _mm256_cvtepi32_ps(_mm256_cvtepu8_epi32(*(__m128i*) &in0[x])); \
pix2##row = _mm256_cvtepi32_ps(_mm256_cvtepu8_epi32(*(__m128i*) &in_1[x]));
#define MM256_SUM(pixrow, kernelrow) \
ss = _mm256_add_ps(ss, _mm256_mul_ps(pix0##pixrow, kernel0##kernelrow)); \
ss = _mm256_add_ps(ss, _mm256_mul_ps(pix1##pixrow, kernel1##kernelrow)); \
ss = _mm256_add_ps(ss, _mm256_mul_ps(pix2##pixrow, kernel2##kernelrow));
#define MM256_PERMUTE(row, from0, from1, control) \
pix0##row = _mm256_permute2f128_ps(pix0##from0, pix0##from1, control); \
pix1##row = _mm256_permute2f128_ps(pix1##from0, pix1##from1, control); \
pix2##row = _mm256_permute2f128_ps(pix2##from0, pix2##from1, control);
#define MM256_OUT(x) \
ssi = _mm_cvtps_epi32(_mm_add_ps( \
_mm256_extractf128_ps(ss, 0), \
_mm256_extractf128_ps(ss, 1) \
)); \
ssi = _mm_packs_epi32(ssi, ssi); \
ssi = _mm_packus_epi16(ssi, ssi); \
out[x] = _mm_cvtsi128_si32(ssi);
__m256 kernel00 = _mm256_insertf128_ps( __m256 kernel00 = _mm256_insertf128_ps(
_mm256_set1_ps(kernel[0+0]), _mm256_set1_ps(kernel[0+0]),
_mm_set1_ps(kernel[0+1]), 1); _mm_set1_ps(kernel[0+1]), 1);
@ -262,60 +287,25 @@ ImagingFilter3x3(Imaging imOut, Imaging im, const float* kernel,
out[0] = in0[0]; out[0] = in0[0];
x = 1; x = 1;
pix00 = _mm256_cvtepi32_ps(_mm256_cvtepu8_epi32(*(__m128i*) &in1[0])); MM256_LOAD(0, 0);
pix10 = _mm256_cvtepi32_ps(_mm256_cvtepu8_epi32(*(__m128i*) &in0[0]));
pix20 = _mm256_cvtepi32_ps(_mm256_cvtepu8_epi32(*(__m128i*) &in_1[0]));
for (; x < im->xsize-1-1; x += 2) { for (; x < im->xsize-1-1; x += 2) {
__m256 ss; __m256 ss;
__m128i ssi; __m128i ssi;
pix01 = _mm256_cvtepi32_ps(_mm256_cvtepu8_epi32(*(__m128i*) &in1[x+1])); ss = _mm256_castps128_ps256(_mm_set1_ps(offset));
pix11 = _mm256_cvtepi32_ps(_mm256_cvtepu8_epi32(*(__m128i*) &in0[x+1])); MM256_SUM(0, 0);
pix21 = _mm256_cvtepi32_ps(_mm256_cvtepu8_epi32(*(__m128i*) &in_1[x+1])); MM256_LOAD(1, x+1);
MM256_SUM(1, 1);
MM256_OUT(x);
ss = _mm256_set1_ps(offset); ss = _mm256_castps128_ps256(_mm_set1_ps(offset));
ss = _mm256_add_ps(ss, _mm256_mul_ps(pix00, kernel00)); MM256_PERMUTE(0, 0, 1, 0x21);
ss = _mm256_add_ps(ss, _mm256_mul_ps(pix01, kernel01)); MM256_SUM(0, 0);
ss = _mm256_add_ps(ss, _mm256_mul_ps(pix10, kernel10)); MM256_PERMUTE(1, 0, 1, 0xf3);
ss = _mm256_add_ps(ss, _mm256_mul_ps(pix11, kernel11)); MM256_SUM(1, 1);
ss = _mm256_add_ps(ss, _mm256_mul_ps(pix20, kernel20)); MM256_OUT(x+1);
ss = _mm256_add_ps(ss, _mm256_mul_ps(pix21, kernel21));
ssi = _mm_cvtps_epi32(_mm_add_ps( MM256_PERMUTE(0, 0, 1, 0x21);
_mm256_extractf128_ps(ss, 0),
_mm256_extractf128_ps(ss, 1)
));
ssi = _mm_packs_epi32(ssi, ssi);
ssi = _mm_packus_epi16(ssi, ssi);
out[x] = _mm_cvtsi128_si32(ssi);
pix00 = _mm256_permute2f128_ps(pix00, pix01, 0x21);
pix10 = _mm256_permute2f128_ps(pix10, pix11, 0x21);
pix20 = _mm256_permute2f128_ps(pix20, pix21, 0x21);
pix01 = _mm256_permute2f128_ps(pix01, pix01, 0xf1);
pix11 = _mm256_permute2f128_ps(pix11, pix11, 0xf1);
pix21 = _mm256_permute2f128_ps(pix21, pix21, 0xf1);
ss = _mm256_set1_ps(offset);
ss = _mm256_add_ps(ss, _mm256_mul_ps(pix00, kernel00));
ss = _mm256_add_ps(ss, _mm256_mul_ps(pix01, kernel01));
ss = _mm256_add_ps(ss, _mm256_mul_ps(pix10, kernel10));
ss = _mm256_add_ps(ss, _mm256_mul_ps(pix11, kernel11));
ss = _mm256_add_ps(ss, _mm256_mul_ps(pix20, kernel20));
ss = _mm256_add_ps(ss, _mm256_mul_ps(pix21, kernel21));
ssi = _mm_cvtps_epi32(_mm_add_ps(
_mm256_extractf128_ps(ss, 0),
_mm256_extractf128_ps(ss, 1)
));
ssi = _mm_packs_epi32(ssi, ssi);
ssi = _mm_packus_epi16(ssi, ssi);
out[x+1] = _mm_cvtsi128_si32(ssi);
pix00 = _mm256_permute2f128_ps(pix00, pix01, 0x21);
pix10 = _mm256_permute2f128_ps(pix10, pix11, 0x21);
pix20 = _mm256_permute2f128_ps(pix20, pix21, 0x21);
} }
out[x] = in0[x]; out[x] = in0[x];
#else #else