Mercurial > libavcodec.hg
changeset 132:c4caf29acc1a libavcodec
3dnow dering
| author | michael |
|---|---|
| date | Thu, 25 Oct 2001 22:50:03 +0000 |
| parents | f91bb7527628 |
| children | e0bf2e9ce24c |
| files | libpostproc/postprocess.c libpostproc/postprocess_template.c |
| diffstat | 2 files changed, 64 insertions(+), 20 deletions(-) [+] |
line wrap: on
line diff
--- a/libpostproc/postprocess.c Thu Oct 25 21:57:42 2001 +0000 +++ b/libpostproc/postprocess.c Thu Oct 25 22:50:03 2001 +0000 @@ -26,7 +26,7 @@ isHorizMinMaxOk a E doHorizLowPass E e e doHorizDefFilter Ec Ec Ec -deRing e +deRing e e* Vertical RKAlgo1 E a a Horizontal RKAlgo1 a a Vertical X1 a E E @@ -1808,7 +1808,7 @@ static inline void dering(uint8_t src[], int stride, int QP) { -#ifdef HAVE_MMX2 +#if defined (HAVE_MMX2) || defined (HAVE_3DNOW) asm volatile( "movq pQPb, %%mm0 \n\t" "paddusb %%mm0, %%mm0 \n\t" @@ -1821,11 +1821,20 @@ "pcmpeqb %%mm6, %%mm6 \n\t" "pxor %%mm7, %%mm7 \n\t" - +#ifdef HAVE_MMX2 #define FIND_MIN_MAX(addr)\ "movq " #addr ", %%mm0 \n\t"\ "pminub %%mm0, %%mm6 \n\t"\ "pmaxub %%mm0, %%mm7 \n\t" +#else +#define FIND_MIN_MAX(addr)\ + "movq " #addr ", %%mm0 \n\t"\ + "movq %%mm6, %%mm1 \n\t"\ + "psubusb %%mm0, %%mm7 \n\t"\ + "paddb %%mm0, %%mm7 \n\t"\ + "psubusb %%mm0, %%mm1 \n\t"\ + "psubb %%mm1, %%mm6 \n\t" +#endif FIND_MIN_MAX((%%eax)) FIND_MIN_MAX((%%eax, %1)) @@ -1838,36 +1847,49 @@ "movq %%mm6, %%mm4 \n\t" "psrlq $8, %%mm6 \n\t" +#ifdef HAVE_MMX2 "pminub %%mm4, %%mm6 \n\t" // min of pixels -#ifdef HAVE_MMX2 "pshufw $0xF9, %%mm6, %%mm4 \n\t" "pminub %%mm4, %%mm6 \n\t" // min of pixels "pshufw $0xFE, %%mm6, %%mm4 \n\t" + "pminub %%mm4, %%mm6 \n\t" #else + "movq %%mm6, %%mm1 \n\t" + "psubusb %%mm4, %%mm1 \n\t" + "psubb %%mm1, %%mm6 \n\t" "movq %%mm6, %%mm4 \n\t" "psrlq $16, %%mm6 \n\t" - "pminub %%mm4, %%mm6 \n\t" + "movq %%mm6, %%mm1 \n\t" + "psubusb %%mm4, %%mm1 \n\t" + "psubb %%mm1, %%mm6 \n\t" "movq %%mm6, %%mm4 \n\t" "psrlq $32, %%mm6 \n\t" + "movq %%mm6, %%mm1 \n\t" + "psubusb %%mm4, %%mm1 \n\t" + "psubb %%mm1, %%mm6 \n\t" #endif - "pminub %%mm4, %%mm6 \n\t" "movq %%mm7, %%mm4 \n\t" "psrlq $8, %%mm7 \n\t" +#ifdef HAVE_MMX2 "pmaxub %%mm4, %%mm7 \n\t" // max of pixels -#ifdef HAVE_MMX2 "pshufw $0xF9, %%mm7, %%mm4 \n\t" - "pmaxub %%mm4, %%mm7 \n\t" // min of pixels + "pmaxub %%mm4, %%mm7 \n\t" "pshufw $0xFE, %%mm7, %%mm4 \n\t" + "pmaxub %%mm4, %%mm7 \n\t" #else + "psubusb %%mm4, %%mm7 \n\t" + "paddb %%mm4, %%mm7 \n\t" "movq %%mm7, %%mm4 \n\t" "psrlq $16, %%mm7 \n\t" - "pmaxub %%mm4, %%mm7 \n\t" + "psubusb %%mm4, %%mm7 \n\t" + "paddb %%mm4, %%mm7 \n\t" "movq %%mm7, %%mm4 \n\t" "psrlq $32, %%mm7 \n\t" + "psubusb %%mm4, %%mm7 \n\t" + "paddb %%mm4, %%mm7 \n\t" #endif - "pmaxub %%mm4, %%mm7 \n\t" PAVGB(%%mm6, %%mm7) // a=(max + min)/2 "punpcklbw %%mm7, %%mm7 \n\t" "punpcklbw %%mm7, %%mm7 \n\t"
--- a/libpostproc/postprocess_template.c Thu Oct 25 21:57:42 2001 +0000 +++ b/libpostproc/postprocess_template.c Thu Oct 25 22:50:03 2001 +0000 @@ -26,7 +26,7 @@ isHorizMinMaxOk a E doHorizLowPass E e e doHorizDefFilter Ec Ec Ec -deRing e +deRing e e* Vertical RKAlgo1 E a a Horizontal RKAlgo1 a a Vertical X1 a E E @@ -1808,7 +1808,7 @@ static inline void dering(uint8_t src[], int stride, int QP) { -#ifdef HAVE_MMX2 +#if defined (HAVE_MMX2) || defined (HAVE_3DNOW) asm volatile( "movq pQPb, %%mm0 \n\t" "paddusb %%mm0, %%mm0 \n\t" @@ -1821,11 +1821,20 @@ "pcmpeqb %%mm6, %%mm6 \n\t" "pxor %%mm7, %%mm7 \n\t" - +#ifdef HAVE_MMX2 #define FIND_MIN_MAX(addr)\ "movq " #addr ", %%mm0 \n\t"\ "pminub %%mm0, %%mm6 \n\t"\ "pmaxub %%mm0, %%mm7 \n\t" +#else +#define FIND_MIN_MAX(addr)\ + "movq " #addr ", %%mm0 \n\t"\ + "movq %%mm6, %%mm1 \n\t"\ + "psubusb %%mm0, %%mm7 \n\t"\ + "paddb %%mm0, %%mm7 \n\t"\ + "psubusb %%mm0, %%mm1 \n\t"\ + "psubb %%mm1, %%mm6 \n\t" +#endif FIND_MIN_MAX((%%eax)) FIND_MIN_MAX((%%eax, %1)) @@ -1838,36 +1847,49 @@ "movq %%mm6, %%mm4 \n\t" "psrlq $8, %%mm6 \n\t" +#ifdef HAVE_MMX2 "pminub %%mm4, %%mm6 \n\t" // min of pixels -#ifdef HAVE_MMX2 "pshufw $0xF9, %%mm6, %%mm4 \n\t" "pminub %%mm4, %%mm6 \n\t" // min of pixels "pshufw $0xFE, %%mm6, %%mm4 \n\t" + "pminub %%mm4, %%mm6 \n\t" #else + "movq %%mm6, %%mm1 \n\t" + "psubusb %%mm4, %%mm1 \n\t" + "psubb %%mm1, %%mm6 \n\t" "movq %%mm6, %%mm4 \n\t" "psrlq $16, %%mm6 \n\t" - "pminub %%mm4, %%mm6 \n\t" + "movq %%mm6, %%mm1 \n\t" + "psubusb %%mm4, %%mm1 \n\t" + "psubb %%mm1, %%mm6 \n\t" "movq %%mm6, %%mm4 \n\t" "psrlq $32, %%mm6 \n\t" + "movq %%mm6, %%mm1 \n\t" + "psubusb %%mm4, %%mm1 \n\t" + "psubb %%mm1, %%mm6 \n\t" #endif - "pminub %%mm4, %%mm6 \n\t" "movq %%mm7, %%mm4 \n\t" "psrlq $8, %%mm7 \n\t" +#ifdef HAVE_MMX2 "pmaxub %%mm4, %%mm7 \n\t" // max of pixels -#ifdef HAVE_MMX2 "pshufw $0xF9, %%mm7, %%mm4 \n\t" - "pmaxub %%mm4, %%mm7 \n\t" // min of pixels + "pmaxub %%mm4, %%mm7 \n\t" "pshufw $0xFE, %%mm7, %%mm4 \n\t" + "pmaxub %%mm4, %%mm7 \n\t" #else + "psubusb %%mm4, %%mm7 \n\t" + "paddb %%mm4, %%mm7 \n\t" "movq %%mm7, %%mm4 \n\t" "psrlq $16, %%mm7 \n\t" - "pmaxub %%mm4, %%mm7 \n\t" + "psubusb %%mm4, %%mm7 \n\t" + "paddb %%mm4, %%mm7 \n\t" "movq %%mm7, %%mm4 \n\t" "psrlq $32, %%mm7 \n\t" + "psubusb %%mm4, %%mm7 \n\t" + "paddb %%mm4, %%mm7 \n\t" #endif - "pmaxub %%mm4, %%mm7 \n\t" PAVGB(%%mm6, %%mm7) // a=(max + min)/2 "punpcklbw %%mm7, %%mm7 \n\t" "punpcklbw %%mm7, %%mm7 \n\t"