python-pillow/Pillow
1
1
Fork 0
mirror of https://github.com/python-pillow/Pillow.git synced 2025-06-07 22:53:24 +03:00
Code Issues Packages Projects Releases Wiki Activity

Add SIMD example

Browse Source
This commit is contained in:
Aleksandr Karpinskii 2024-07-06 15:30:22 +04:00
parent e0649632ff
commit 73f76e55c3
1 changed files with 41 additions and 0 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

41
src/libImaging/Bands.c
View File

@ -21,6 +21,9 @@ Imaging
ImagingGetBand(Imaging imIn, int band) { ImagingGetBand(Imaging imIn, int band) {
Imaging imOut; Imaging imOut;
int x, y; int x, y;
#ifdef __SSE4__
__m128i shuffle_mask;
#endif
/* Check arguments */ /* Check arguments */
if (!imIn || imIn->type != IMAGING_TYPE_UINT8) { if (!imIn || imIn->type != IMAGING_TYPE_UINT8) {
@ -46,14 +49,25 @@ ImagingGetBand(Imaging imIn, int band) {
return NULL; return NULL;
} }
#ifdef __SSE4__
shuffle_mask = _mm_set_epi8(
-1,-1,-1,-1, -1,-1,-1,-1, -1,-1,-1,-1, 12+band,8+band,4+band,0+band);
#endif
/* Extract band from image */ /* Extract band from image */
for (y = 0; y < imIn->ysize; y++) { for (y = 0; y < imIn->ysize; y++) {
UINT8 *in = (UINT8 *)imIn->image[y] + band; UINT8 *in = (UINT8 *)imIn->image[y] + band;
UINT8 *out = imOut->image8[y]; UINT8 *out = imOut->image8[y];
x = 0; x = 0;
for (; x < imIn->xsize - 3; x += 4) { for (; x < imIn->xsize - 3; x += 4) {
#ifdef __SSE4__
__m128i source = _mm_loadu_si128((__m128i *)(in - band));
*((UINT32 *)(out + x)) = _mm_cvtsi128_si32(
_mm_shuffle_epi8(source, shuffle_mask));
#else
UINT32 v = MAKE_UINT32(in[0], in[4], in[8], in[12]); UINT32 v = MAKE_UINT32(in[0], in[4], in[8], in[12]);
memcpy(out + x, &v, sizeof(v)); memcpy(out + x, &v, sizeof(v));
#endif
in += 16; in += 16;
} }
for (; x < imIn->xsize; x++) { for (; x < imIn->xsize; x++) {
@ -99,10 +113,18 @@ ImagingSplit(Imaging imIn, Imaging bands[4]) {
UINT8 *out1 = bands[1]->image8[y]; UINT8 *out1 = bands[1]->image8[y];
x = 0; x = 0;
for (; x < imIn->xsize - 3; x += 4) { for (; x < imIn->xsize - 3; x += 4) {
#ifdef __SSE4__
__m128i source = _mm_loadu_si128((__m128i *)in);
source = _mm_shuffle_epi8(source, _mm_set_epi8(
15,11,7,3, 14,10,6,2, 13,9,5,1, 12,8,4,0));
*((UINT32 *)(out0 + x)) = _mm_cvtsi128_si32(source);
*((UINT32 *)(out1 + x)) = _mm_cvtsi128_si32(_mm_srli_si128(source, 12));
#else
UINT32 v = MAKE_UINT32(in[0], in[4], in[8], in[12]); UINT32 v = MAKE_UINT32(in[0], in[4], in[8], in[12]);
memcpy(out0 + x, &v, sizeof(v)); memcpy(out0 + x, &v, sizeof(v));
v = MAKE_UINT32(in[0 + 3], in[4 + 3], in[8 + 3], in[12 + 3]); v = MAKE_UINT32(in[0 + 3], in[4 + 3], in[8 + 3], in[12 + 3]);
memcpy(out1 + x, &v, sizeof(v)); memcpy(out1 + x, &v, sizeof(v));
#endif
in += 16; in += 16;
} }
for (; x < imIn->xsize; x++) { for (; x < imIn->xsize; x++) {
@ -119,12 +141,21 @@ ImagingSplit(Imaging imIn, Imaging bands[4]) {
UINT8 *out2 = bands[2]->image8[y]; UINT8 *out2 = bands[2]->image8[y];
x = 0; x = 0;
for (; x < imIn->xsize - 3; x += 4) { for (; x < imIn->xsize - 3; x += 4) {
#ifdef __SSE4__
__m128i source = _mm_loadu_si128((__m128i *)in);
source = _mm_shuffle_epi8(source, _mm_set_epi8(
15,11,7,3, 14,10,6,2, 13,9,5,1, 12,8,4,0));
*((UINT32 *)(out0 + x)) = _mm_cvtsi128_si32(source);
*((UINT32 *)(out1 + x)) = _mm_cvtsi128_si32(_mm_srli_si128(source, 4));
*((UINT32 *)(out2 + x)) = _mm_cvtsi128_si32(_mm_srli_si128(source, 8));
#else
UINT32 v = MAKE_UINT32(in[0], in[4], in[8], in[12]); UINT32 v = MAKE_UINT32(in[0], in[4], in[8], in[12]);
memcpy(out0 + x, &v, sizeof(v)); memcpy(out0 + x, &v, sizeof(v));
v = MAKE_UINT32(in[0 + 1], in[4 + 1], in[8 + 1], in[12 + 1]); v = MAKE_UINT32(in[0 + 1], in[4 + 1], in[8 + 1], in[12 + 1]);
memcpy(out1 + x, &v, sizeof(v)); memcpy(out1 + x, &v, sizeof(v));
v = MAKE_UINT32(in[0 + 2], in[4 + 2], in[8 + 2], in[12 + 2]); v = MAKE_UINT32(in[0 + 2], in[4 + 2], in[8 + 2], in[12 + 2]);
memcpy(out2 + x, &v, sizeof(v)); memcpy(out2 + x, &v, sizeof(v));
#endif
in += 16; in += 16;
} }
for (; x < imIn->xsize; x++) { for (; x < imIn->xsize; x++) {
@ -143,6 +174,15 @@ ImagingSplit(Imaging imIn, Imaging bands[4]) {
UINT8 *out3 = bands[3]->image8[y]; UINT8 *out3 = bands[3]->image8[y];
x = 0; x = 0;
for (; x < imIn->xsize - 3; x += 4) { for (; x < imIn->xsize - 3; x += 4) {
#ifdef __SSE4__
__m128i source = _mm_loadu_si128((__m128i *)in);
source = _mm_shuffle_epi8(source, _mm_set_epi8(
15,11,7,3, 14,10,6,2, 13,9,5,1, 12,8,4,0));
*((UINT32 *)(out0 + x)) = _mm_cvtsi128_si32(source);
*((UINT32 *)(out1 + x)) = _mm_cvtsi128_si32(_mm_srli_si128(source, 4));
*((UINT32 *)(out2 + x)) = _mm_cvtsi128_si32(_mm_srli_si128(source, 8));
*((UINT32 *)(out3 + x)) = _mm_cvtsi128_si32(_mm_srli_si128(source, 12));
#else
UINT32 v = MAKE_UINT32(in[0], in[4], in[8], in[12]); UINT32 v = MAKE_UINT32(in[0], in[4], in[8], in[12]);
memcpy(out0 + x, &v, sizeof(v)); memcpy(out0 + x, &v, sizeof(v));
v = MAKE_UINT32(in[0 + 1], in[4 + 1], in[8 + 1], in[12 + 1]); v = MAKE_UINT32(in[0 + 1], in[4 + 1], in[8 + 1], in[12 + 1]);
@ -151,6 +191,7 @@ ImagingSplit(Imaging imIn, Imaging bands[4]) {
memcpy(out2 + x, &v, sizeof(v)); memcpy(out2 + x, &v, sizeof(v));
v = MAKE_UINT32(in[0 + 3], in[4 + 3], in[8 + 3], in[12 + 3]); v = MAKE_UINT32(in[0 + 3], in[4 + 3], in[8 + 3], in[12 + 3]);
memcpy(out3 + x, &v, sizeof(v)); memcpy(out3 + x, &v, sizeof(v));
#endif
in += 16; in += 16;
} }
for (; x < imIn->xsize; x++) { for (; x < imIn->xsize; x++) {