desktop/avidemux/2.5.3_field_asm_fix.diff
2010-07-19 11:35:51 +00:00

136 lines
4.3 KiB
Diff

diff --git a/avidemux/ADM_coreImage/src/ADM_interlaceUtil.cpp b/avidemux/ADM_coreImage/src/ADM_interlaceUtil.cpp
index 33b2a7f..99f0664 100644
--- a/avidemux/ADM_coreImage/src/ADM_interlaceUtil.cpp
+++ b/avidemux/ADM_coreImage/src/ADM_interlaceUtil.cpp
@@ -30,19 +30,8 @@ Skip factor=2
***************************************************************************/
#include "ADM_default.h"
-
-//#include "ADM_editor/ADM_edit.hxx"
#include "ADM_videoFilter.h"
-//#define MMX_TRACE
-#warning remove mmxmacro and debug asm
-//#define ASM_ILACING
-
-
-#include "ADM_mmxMacros.h"
-
-
-
#define SKIP_FACTOR 2 // 2^SKIPFACTOR=SKIP_LINEAR+1
#define SKIP_LINEAR 3
@@ -62,7 +51,7 @@ Skip factor=2
*/
#if defined(ADM_CPU_X86) && defined(ASM_ILACING)
-
+#warning REWRITE IN PLAIN ASM
static uint32_t ADMVideo_interlaceCount_MMX( uint8_t *src ,uint32_t w, uint32_t h);
static uint8_t * FUNNY_MANGLE(_l_p) =NULL;
static uint8_t * FUNNY_MANGLE(_l_c) =NULL;
diff --git a/avidemux/ADM_coreImage/src/ADM_vidFieldASM.cpp b/avidemux/ADM_coreImage/src/ADM_vidFieldASM.cpp
index 61d11d4..479731c 100644
--- a/avidemux/ADM_coreImage/src/ADM_vidFieldASM.cpp
+++ b/avidemux/ADM_coreImage/src/ADM_vidFieldASM.cpp
@@ -20,15 +20,13 @@
***************************************************************************/
#include "ADM_default.h"
-
#include "ADM_videoFilter.h"
-
#include"ADM_vidField.h"
#ifdef ADM_CPU_X86
// #define DEBUG_DEINT 1
// #define MMX_TRACE 1
- #include "ADM_mmxMacros.h"
+// #include "ADM_mmxMacros.h"
void myDeintASM(void);
@@ -36,25 +34,27 @@
static uint8_t * FUNNY_MANGLE(_l_p) , * FUNNY_MANGLE(_l_c) ,* FUNNY_MANGLE(_l_n);
static uint8_t * FUNNY_MANGLE(_l_e) , * FUNNY_MANGLE(_l_e2);
#define EXPAND(x) (x)+((x)<<16)+((x)<<32) +((x)<<48)
-static mmx_t _mmTHRESH1;
-static mmx_t _mmTHRESH2;
+static uint64_t __attribute__((used)) __attribute__ ((__aligned__ (8))) FUNNY_MANGLE(_mmTHRESH1) ;
+static uint64_t __attribute__((used)) __attribute__ ((__aligned__ (8))) FUNNY_MANGLE(_mmTHRESH2) ;
-#define COMPUTE_MMX \
-punpcklbw_r2r(mm5,mm0); /*c expand 4 bytes -> 4 word */ \
-punpcklbw_r2r(mm5,mm1); /*p*/ \
-punpcklbw_r2r(mm5,mm2); /* n*/ \
-movq_r2r(mm0,mm3); /* mm3 also c*/ \
-psubw_r2r(mm1,mm0) ; /* mm0=mm0-mm1 = c-p*/ \
-psubw_r2r(mm2,mm3) ; /* mm3=mm3-mm2 = c-n*/ \
-psraw_i2r(1,mm0); /* to protect from overflow*/ \
-psraw_i2r(1,mm3);\
-pmullw_r2r(mm0,mm3); /* mm3=(c-p)*(c-n) / 4;*/ \
-movq_r2r(mm3,mm0) ; /* mm0 also c-p*c-n */ \
-pcmpgtw_r2r(mm4,mm3); /* keep only > size*/ \
-pcmpgtw_r2r(mm6,mm0); /* keep only > size*/ \
-packsswb_r2r(mm5,mm0); \
-packsswb_r2r(mm5,mm3);
+#define COMPUTE_MMX \
+__asm__ __volatile__(\
+"punpcklbw %%mm5,%%mm0\n\t" \
+"punpcklbw %%mm5,%%mm1\n\t" \
+"punpcklbw %%mm5,%%mm2\n\t" \
+"movq %%mm0,%%mm3\n\t" \
+"psubw %%mm1,%%mm0\n\t" \
+"psubw %%mm2,%%mm3\n\t" \
+"psraw $1,%%mm0\n\t" \
+"psraw $1,%%mm3\n\t" \
+"pmullw %%mm0,%%mm3\n\t" \
+"movq %%mm3,%%mm0\n\t" \
+"pcmpgtw %%mm4,%%mm3\n\t" \
+"pcmpgtw %%mm6,%%mm0\n\t" \
+"packsswb %%mm5,%%mm0\n\t" \
+"packsswb %%mm5,%%mm3\n\t" \
+::)
#endif
void ADMVideoFields::hasMotion_C(uint8_t *p,uint8_t *c,
@@ -90,8 +90,8 @@ void ADMVideoFields::hasMotion_MMX(uint8_t *p,uint8_t *c,
{
- _mmTHRESH1.uq=EXPAND((uint64_t ) ((_param->motion_trigger*_param->motion_trigger)>>2) );
- _mmTHRESH2.uq=EXPAND((uint64_t ) ((_param->blend_trigger*_param->blend_trigger)>>2) );
+ _mmTHRESH1=EXPAND((uint64_t ) ((_param->motion_trigger*_param->motion_trigger)>>2) );
+ _mmTHRESH2=EXPAND((uint64_t ) ((_param->blend_trigger*_param->blend_trigger)>>2) );
_l_h=_info.height-2;
_l_w=_info.width>>2;
@@ -102,10 +102,11 @@ void ADMVideoFields::hasMotion_MMX(uint8_t *p,uint8_t *c,
_l_e=e;
_l_e2=e2;
// printf("\n MMX \n");
-
- pxor_r2r(mm5,mm5);
- movq_m2r(_mmTHRESH1,mm4);
- movq_m2r(_mmTHRESH2,mm6);
+ __asm__ __volatile__ (
+ "pxor %%mm5,%%mm5\n\t"
+ "movq "Mangle(_mmTHRESH1)",%%mm4\n\t"
+ "movq "Mangle(_mmTHRESH2)",%%mm6\n\t"
+ ::);
myDeintASM();
}
#if !defined(DEBUG_DEINT)
@@ -149,7 +150,7 @@ void myDeintASM(void)
:
: "eax", "ecx","edx","esi"
);
- emms();
+ __asm__ __volatile__ ("emms");
}
#else