mirror of
https://gitlab.winehq.org/wine/wine-gecko.git
synced 2024-09-13 09:24:08 -07:00
98 lines
3.6 KiB
Diff
98 lines
3.6 KiB
Diff
|
Index: lib/dec/x86_vc/mmxfrag.c
|
||
|
|
===================================================================
|
||
|
|
--- lib/dec/x86_vc/mmxfrag.c (revision 16142)
|
||
|
|
+++ lib/dec/x86_vc/mmxfrag.c (working copy)
|
||
|
|
@@ -27,12 +27,14 @@
|
||
|
|
|
||
|
|
void oc_frag_recon_intra_mmx(unsigned char *_dst,int _dst_ystride,
|
||
|
|
const ogg_int16_t *_residue){
|
||
|
|
+ int _save_ebx;
|
||
|
|
/* ---------------------------------------------------------------------
|
||
|
|
This function does the inter reconstruction step with 8 iterations
|
||
|
|
unrolled. The iteration for each instruction is noted by the #id in the
|
||
|
|
comments (in case you want to reconstruct it)
|
||
|
|
--------------------------------------------------------------------- */
|
||
|
|
_asm{
|
||
|
|
+ mov [_save_ebx], ebx
|
||
|
|
mov edi, [_residue] /* load residue ptr */
|
||
|
|
mov eax, 0x00800080 /* generate constant */
|
||
|
|
mov ebx, [_dst_ystride] /* load dst-stride */
|
||
|
|
@@ -93,6 +95,7 @@
|
||
|
|
packuswb mm3, mm4 /* #8 pack to byte */
|
||
|
|
movq [edx + ecx*2], mm1 /* #7 write row */
|
||
|
|
movq [edx + eax], mm3 /* #8 write row */
|
||
|
|
+ mov ebx, [_save_ebx]
|
||
|
|
}
|
||
|
|
}
|
||
|
|
|
||
|
|
@@ -100,6 +103,7 @@
|
||
|
|
|
||
|
|
void oc_frag_recon_inter_mmx (unsigned char *_dst, int _dst_ystride,
|
||
|
|
const unsigned char *_src, int _src_ystride, const ogg_int16_t *_residue){
|
||
|
|
+ int _save_ebx;
|
||
|
|
/* ---------------------------------------------------------------------
|
||
|
|
This function does the inter reconstruction step with two iterations
|
||
|
|
running in parallel to hide some load-latencies and break the dependency
|
||
|
|
@@ -107,6 +111,7 @@
|
||
|
|
comments (in case you want to reconstruct it)
|
||
|
|
--------------------------------------------------------------------- */
|
||
|
|
_asm{
|
||
|
|
+ mov [_save_ebx], ebx
|
||
|
|
pxor mm0, mm0 /* generate constant 0 */
|
||
|
|
mov esi, [_src]
|
||
|
|
mov edi, [_residue]
|
||
|
|
@@ -143,6 +148,7 @@
|
||
|
|
movq [edx + ebx], mm7 /* #2 write row */
|
||
|
|
lea edx, [edx+ebx*2] /* dst += stride * 2 */
|
||
|
|
jne nextchunk
|
||
|
|
+ mov ebx, [_save_ebx]
|
||
|
|
}
|
||
|
|
}
|
||
|
|
|
||
|
|
@@ -150,6 +156,7 @@
|
||
|
|
void oc_frag_recon_inter2_mmx(unsigned char *_dst, int _dst_ystride,
|
||
|
|
const unsigned char *_src1, int _src1_ystride, const unsigned char *_src2,
|
||
|
|
int _src2_ystride,const ogg_int16_t *_residue){
|
||
|
|
+ int _save_ebx;
|
||
|
|
/* ---------------------------------------------------------------------
|
||
|
|
This function does the inter2 reconstruction step.The building of the
|
||
|
|
average is done with a bit-twiddeling trick to avoid excessive register
|
||
|
|
@@ -166,6 +173,7 @@
|
||
|
|
using the pavgb instruction let me know and I'll do the 3dnow codepath.
|
||
|
|
--------------------------------------------------------------------- */
|
||
|
|
_asm{
|
||
|
|
+ mov [_save_ebx], ebx
|
||
|
|
mov eax, 0xfefefefe
|
||
|
|
mov esi, [_src1]
|
||
|
|
mov edi, [_src2]
|
||
|
|
@@ -204,6 +212,7 @@
|
||
|
|
packuswb mm2, mm3 /* pack and saturate */
|
||
|
|
movq [edx], mm2 /* write row */
|
||
|
|
jne nextrow
|
||
|
|
+ mov ebx, [_save_ebx]
|
||
|
|
}
|
||
|
|
}
|
||
|
|
|
||
|
|
Index: lib/dec/x86_vc/mmxloopfilter.c
|
||
|
|
===================================================================
|
||
|
|
--- lib/dec/x86_vc/mmxloopfilter.c (revision 16142)
|
||
|
|
+++ lib/dec/x86_vc/mmxloopfilter.c (working copy)
|
||
|
|
@@ -38,7 +38,7 @@
|
||
|
|
_asm {
|
||
|
|
mov eax, [_pix]
|
||
|
|
mov edx, [_ystride]
|
||
|
|
- mov ebx, [_ll]
|
||
|
|
+ mov ecx, [_ll]
|
||
|
|
|
||
|
|
/* _pix -= ystride */
|
||
|
|
sub eax, edx
|
||
|
|
@@ -104,7 +104,7 @@
|
||
|
|
/*Free up mm5.*/
|
||
|
|
packuswb mm4, mm5
|
||
|
|
/*mm0=L L L L*/
|
||
|
|
- movq mm0, [ebx]
|
||
|
|
+ movq mm0, [ecx]
|
||
|
|
/*if(R_i<-2L||R_i>2L)R_i=0:*/
|
||
|
|
movq mm5, mm2
|
||
|
|
pxor mm6, mm6
|