ryujinx/SDL
1
0
Fork 0
mirror of https://github.com/Ryujinx/SDL.git synced 2025-01-25 17:51:14 +00:00
Code Issues Packages Projects Releases Wiki Activity

Fixed bug #6990: fix computation of alpha in BlitRGBtoRGBPixelAlphaMMX

Browse source
This commit is contained in:
Sylvain 2023-01-06 13:27:53 +01:00 committed by Sam Lantinga
parent a099172754
commit 0ebda87425
1 changed files with 37 additions and 8 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

45
src/video/SDL_blit_A.c
View file

@ -332,9 +332,19 @@ static void BlitRGBtoRGBPixelAlphaMMX(SDL_BlitInfo *info)
Uint32 ashift = sf->Ashift; Uint32 ashift = sf->Ashift;
Uint64 multmask, multmask2; Uint64 multmask, multmask2;
__m64 src1, dst1, mm_alpha, mm_zero, mm_alpha2; __m64 src1, dst1, mm_alpha, mm_zero, mm_alpha2, mm_one_alpha;
mm_zero = _mm_setzero_si64(); /* 0 -> mm_zero */ mm_zero = _mm_setzero_si64(); /* 0 -> mm_zero */
if (amask == 0xFF000000) { /* 1 in the alpha channel -> mm_one_alpha */
mm_one_alpha = _mm_set_pi16(1, 0, 0, 0);
} else if (amask == 0x00FF0000) {
mm_one_alpha = _mm_set_pi16(0, 1, 0, 0);
} else if (amask == 0x0000FF00) {
mm_one_alpha = _mm_set_pi16(0, 0, 1, 0);
} else {
mm_one_alpha = _mm_set_pi16(0, 0, 0, 1);
}
multmask = 0x00FF; multmask = 0x00FF;
multmask <<= (ashift * 2); multmask <<= (ashift * 2);
multmask2 = 0x00FF00FF00FF00FFULL; multmask2 = 0x00FF00FF00FF00FFULL;
@ -361,14 +371,33 @@ static void BlitRGBtoRGBPixelAlphaMMX(SDL_BlitInfo *info)
mm_alpha = _mm_or_si64(mm_alpha2, *(__m64 *) & multmask); /* 0F0A0A0A -> mm_alpha */ mm_alpha = _mm_or_si64(mm_alpha2, *(__m64 *) & multmask); /* 0F0A0A0A -> mm_alpha */
mm_alpha2 = _mm_xor_si64(mm_alpha2, *(__m64 *) & multmask2); /* 255 - mm_alpha -> mm_alpha */ mm_alpha2 = _mm_xor_si64(mm_alpha2, *(__m64 *) & multmask2); /* 255 - mm_alpha -> mm_alpha */
/* blend */ /*
Alpha blending is:
dstRGB = (srcRGB * srcA) + (dstRGB * (1-srcA))
dstA = srcA + (dstA * (1-srcA)) *
Here, 'src1' is:
srcRGB * srcA
srcA
And 'dst1' is:
dstRGB * (1-srcA)
dstA * (1-srcA)
so that *dstp is 'src1 + dst1'
src1 is computed using mullo_pi16: (X * mask) >> 8, but is approximate for srcA ((srcA * 255) >> 8).
need to a 1 to get an exact result: (srcA * 256) >> 8 == srcA
*/
mm_alpha = _mm_add_pi16(mm_alpha, mm_one_alpha);
/* blend */
src1 = _mm_mullo_pi16(src1, mm_alpha); src1 = _mm_mullo_pi16(src1, mm_alpha);
src1 = _mm_srli_pi16(src1, 8); src1 = _mm_srli_pi16(src1, 8);
dst1 = _mm_mullo_pi16(dst1, mm_alpha2); dst1 = _mm_mullo_pi16(dst1, mm_alpha2);
dst1 = _mm_srli_pi16(dst1, 8); dst1 = _mm_srli_pi16(dst1, 8);
dst1 = _mm_add_pi16(src1, dst1); dst1 = _mm_add_pi16(src1, dst1);
dst1 = _mm_packs_pu16(dst1, mm_zero); dst1 = _mm_packs_pu16(dst1, mm_zero);
*dstp = _mm_cvtsi64_si32(dst1); /* dst1 -> pixel */ *dstp = _mm_cvtsi64_si32(dst1); /* dst1 -> pixel */
} }
++srcp; ++srcp;
@ -664,14 +693,14 @@ static void BlitRGBtoRGBPixelAlphaMMX3DNOW(SDL_BlitInfo *info)
mm_alpha2 = _mm_xor_si64(mm_alpha2, *(__m64 *) & multmask2); /* 255 - mm_alpha -> mm_alpha */ mm_alpha2 = _mm_xor_si64(mm_alpha2, *(__m64 *) & multmask2); /* 255 - mm_alpha -> mm_alpha */
/* blend */ /* blend */
src1 = _mm_mullo_pi16(src1, mm_alpha); src1 = _mm_mullo_pi16(src1, mm_alpha);
src1 = _mm_srli_pi16(src1, 8); src1 = _mm_srli_pi16(src1, 8);
dst1 = _mm_mullo_pi16(dst1, mm_alpha2); dst1 = _mm_mullo_pi16(dst1, mm_alpha2);
dst1 = _mm_srli_pi16(dst1, 8); dst1 = _mm_srli_pi16(dst1, 8);
dst1 = _mm_add_pi16(src1, dst1); dst1 = _mm_add_pi16(src1, dst1);
dst1 = _mm_packs_pu16(dst1, mm_zero); dst1 = _mm_packs_pu16(dst1, mm_zero);
*dstp = _mm_cvtsi64_si32(dst1); /* dst1 -> pixel */ *dstp = _mm_cvtsi64_si32(dst1); /* dst1 -> pixel */
} }
++srcp; ++srcp;
@ -1024,7 +1053,7 @@ static void Blit555to555SurfaceAlphaMMX(SDL_BlitInfo *info)
dst2 = _mm_and_si64(dst2, rmask); /* dst2 & MASKRED -> dst2 */ dst2 = _mm_and_si64(dst2, rmask); /* dst2 & MASKRED -> dst2 */
mm_res = dst2; /* RED -> mm_res */ mm_res = dst2; /* RED -> mm_res */
/* green -- process the bits in place */ /* green -- process the bits in place */
src2 = src1; src2 = src1;
src2 = _mm_and_si64(src2, gmask); /* src & MASKGREEN -> src2 */ src2 = _mm_and_si64(src2, gmask); /* src & MASKGREEN -> src2 */
@ -1166,7 +1195,7 @@ static void BlitARGBto565PixelAlpha(SDL_BlitInfo *info)
compositioning used (>>8 instead of /255) doesn't handle compositioning used (>>8 instead of /255) doesn't handle
it correctly. Also special-case alpha=0 for speed? it correctly. Also special-case alpha=0 for speed?
Benchmark this! */ Benchmark this! */
if (alpha) { if (alpha) {
if (alpha == (SDL_ALPHA_OPAQUE >> 3)) { if (alpha == (SDL_ALPHA_OPAQUE >> 3)) {
*dstp = (Uint16)((s >> 8 & 0xf800) + (s >> 5 & 0x7e0) + (s >> 3 & 0x1f)); *dstp = (Uint16)((s >> 8 & 0xf800) + (s >> 5 & 0x7e0) + (s >> 3 & 0x1f));
} else { } else {
@ -1212,7 +1241,7 @@ static void BlitARGBto555PixelAlpha(SDL_BlitInfo *info)
compositioning used (>>8 instead of /255) doesn't handle compositioning used (>>8 instead of /255) doesn't handle
it correctly. Also special-case alpha=0 for speed? it correctly. Also special-case alpha=0 for speed?
Benchmark this! */ Benchmark this! */
if (alpha) { if (alpha) {
if (alpha == (SDL_ALPHA_OPAQUE >> 3)) { if (alpha == (SDL_ALPHA_OPAQUE >> 3)) {
*dstp = (Uint16)((s >> 9 & 0x7c00) + (s >> 6 & 0x3e0) + (s >> 3 & 0x1f)); *dstp = (Uint16)((s >> 9 & 0x7c00) + (s >> 6 & 0x3e0) + (s >> 3 & 0x1f));
} else { } else {