guijan
/
legacy-imlib2
forked from old/legacy-imlib2
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

optimize mmx blending more.. uswed to do 15 million pixels/sec... now does 25 

million per sec.. compared to the C (9million per sec) thats pretty good now


SVN revision: 2454
This commit is contained in:
Carsten Haitzler 2000-04-10 18:12:40 +00:00
parent 5a8634780e
commit 3def8d48d3
2 changed files with 19 additions and 27 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

34
src/asm_blend.S
View File

@ -36,7 +36,7 @@ ret
__imlib_asm_blend_rgba_to_rgb(void *src, void *dst, int w, void *mmx_data)
******************************************************************************/
#define zero (%ebx)
#define components (%ebx)
#define mask_red 8(%ebx)
#define mask_green 16(%ebx)
#define mask_blue 24(%ebx)
@ -62,41 +62,33 @@ __imlib_asm_blend_rgba_to_rgb:
imul $4, %edx /* w *= 4; */
movl $0, %ecx /* int i = 0; */
pxor %mm1, %mm1 /* mm1 = 0; */
.for_loop__imlib_asm_blend_rgba_to_rgb:
movd (%esi,%ecx), %mm2 /* mm2 = src[i]; */
movd (%eax,%ecx), %mm3 /* mm3 = dst[i]; */
movq %mm2, %mm4 /* mm4 = mm2; */
pand mask_alpha, %mm4 /* mm4 &= mask_alpha */
psrlq $24, %mm4 /* mm4 >>= 24; */
psrlq $24, %mm4 /* mm4 >>= 24; (mm4 = oooooooA) */
movq zero, %mm1 /* mm1 = 0; */
punpcklbw %mm1, %mm2 /* mm2 = punpcklbw(mm1); */
punpcklbw %mm1, %mm3 /* mm3 = punpcklbw(mm1); */
punpcklbw %mm1, %mm2 /* mm2 = mm2(oAoRoGoB) */
punpcklbw %mm1, %mm3 /* mm3 = mm3(oAoRoGoB) */
psubw %mm3, %mm2 /* mm2 -= mm3; */
punpcklwd %mm4, %mm4 /* mm4 = punpcklwd(mm4); */
punpckldq %mm4, %mm4 /* mm4 = punpckldq(mm4); */
punpcklwd %mm4, %mm4 /* mm4 = mm4(oooooAoA) */
punpckldq %mm4, %mm4 /* mm4 = mm4(oAoAoAoA) */
pmullw %mm4, %mm2 /* mm2 *= mm4; */
psraw $8, %mm2 /* mm2 >>= 8; */
paddw %mm3, %mm2 /* mm2 += mm3; */
movq %mm2, %mm3 /* mm3 = mm2; */
pand mask_red, %mm3 /* mm3 &= mask_red; */
pand components, %mm2 /* mm2 &= components */
packuswb %mm1, %mm2 /* mm2 = mm2(ooooARGB) */
#if 1 /* need this to be correct - but normally can be left out */
por mask_alpha, %mm2 /* mm2 |= mask_alpha */
#endif
movq %mm2, %mm4 /* mm4 = mm2; */
pand mask_green, %mm4 /* mm4 &= mask_green; */
psrlq $8, %mm4 /* mm4 >>= 8; */
por %mm4, %mm3 /* mm3 |= mm4; */
pand mask_blue, %mm2 /* mm2 &= mask_blue; */
psrlq $16, %mm2 /* mm2 >>= 16; */
por %mm2, %mm3 /* mm3 |= mm2; */
movd %mm3, (%eax,%ecx) /* dst[i] = mm3; */
movd %mm2, (%eax,%ecx) /* dst[i] = mm3; */
add $4, %ecx /* i += 4; */
cmp %edx, %ecx /* if (i != w) */

12
src/blend.c
View File

@ -65,12 +65,12 @@
#ifdef DO_MMX_ASM
static DATA8 mmx_data[] =
{
0, 0, 0, 0, 0, 0, 0, 0, /* zero */
255,0, 0, 0, 0, 0, 0, 0, /* mask_red */
0, 0,255, 0, 0, 0, 0, 0, /* mask_green */
0, 0, 0, 0, 255,0, 0, 0, /* mask_blue */
0, 0, 0, 255,0, 0, 0, 0, /* mask_alpha */
255,255,255,255,255,255,0, 0 /* mask */
255, 0,255, 0,255, 0,255, 0, /* zero */
255, 0, 0, 0, 0, 0, 0, 0, /* mask_red */
0, 0,255, 0, 0, 0, 0, 0, /* mask_green */
0, 0, 0, 0, 255,0, 0, 0, /* mask_blue */
0, 0, 0,255, 0, 0, 0, 0, /* mask_alpha */
255,255,255,255,255,255, 0, 0 /* mask */
};
#endif