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 Filter. faster 3x3 singleband SSE4

Browse Source
This commit is contained in:
Alexander 2017-08-13 22:34:34 +03:00 committed by Alexander Karpinsky
parent 64ab983164
commit 05a5ff706a
1 changed files with 40 additions and 6 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

44
src/libImaging/Filter.c
View File

@ -107,6 +107,12 @@ void
ImagingFilter3x3(Imaging imOut, Imaging im, const float* kernel,
float offset)
{
#define MM_KERNEL1x3_SUM1(ss, row) \
ss = _mm_set1_ps(offset); \
ss = _mm_add_ps(ss, _mm_mul_ps(pix0##row, kernel0)); \
ss = _mm_add_ps(ss, _mm_mul_ps(pix1##row, kernel1)); \
ss = _mm_add_ps(ss, _mm_mul_ps(pix2##row, kernel2));
#define MM_KERNEL1x3_LOAD(row, x) \
pix0##row = _mm_cvtepi32_ps(_mm_cvtepu8_epi32(*(__m128i*) &in1[x])); \
pix1##row = _mm_cvtepi32_ps(_mm_cvtepu8_epi32(*(__m128i*) &in0[x])); \
@ -121,8 +127,6 @@ ImagingFilter3x3(Imaging imOut, Imaging im, const float* kernel,
memcpy(imOut->image[0], im->image[0], im->linesize);
if (im->bands == 1) {
// Add one time for rounding
offset += 0.5 / 4;
for (y = 1; y < im->ysize-1; y++) {
UINT8* in_1 = (UINT8*) im->image[y-1];
UINT8* in0 = (UINT8*) im->image[y];
@ -133,16 +137,46 @@ ImagingFilter3x3(Imaging imOut, Imaging im, const float* kernel,
__m128 kernel2 = _mm_set_ps(0, kernel[8], kernel[7], kernel[6]);
out[0] = in0[0];
for (x = 1; x < im->xsize-1; x++) {
__m128 ss = _mm_set1_ps(offset);;
x = 1;
for (; x < im->xsize-1-3; x += 4) {
__m128 ss0, ss1, ss2, ss3;
__m128 pix00, pix10, pix20;
__m128i ssi0;
MM_KERNEL1x3_LOAD(0, x-1);
MM_KERNEL1x3_SUM1(ss0, 0);
MM_KERNEL1x3_LOAD(0, x+0);
MM_KERNEL1x3_SUM1(ss1, 0);
MM_KERNEL1x3_LOAD(0, x+1);
MM_KERNEL1x3_SUM1(ss2, 0);
MM_KERNEL1x3_LOAD(0, x+2);
MM_KERNEL1x3_SUM1(ss3, 0);
ss0 = _mm_hadd_ps(ss0, ss1);
ss1 = _mm_hadd_ps(ss2, ss3);
ss0 = _mm_hadd_ps(ss0, ss1);
ssi0 = _mm_cvtps_epi32(ss0);
ssi0 = _mm_packs_epi32(ssi0, ssi0);
ssi0 = _mm_packus_epi16(ssi0, ssi0);
*((UINT32*) &out[x]) = _mm_cvtsi128_si32(ssi0);
}
for (; x < im->xsize-1; x++) {
__m128 ss;
__m128 pix00, pix10, pix20;
__m128i ssi0;
MM_KERNEL1x3_LOAD(0, x-1);
ss = _mm_set1_ps(offset);
ss = _mm_add_ps(ss, _mm_mul_ps(pix00, kernel0));
ss = _mm_add_ps(ss, _mm_mul_ps(pix10, kernel1));
ss = _mm_add_ps(ss, _mm_mul_ps(pix20, kernel2));
ss = _mm_hadd_ps(ss, ss);
ss = _mm_hadd_ps(ss, ss);
out[x] = clip8(_mm_cvtss_f32(ss));
ssi0 = _mm_cvtps_epi32(ss);
ssi0 = _mm_packs_epi32(ssi0, ssi0);
ssi0 = _mm_packus_epi16(ssi0, ssi0);
out[x] = _mm_cvtsi128_si32(ssi0);
}
out[x] = in0[x];
}