Daniel Jahre
20 years ago
1 changed files with 206 additions and 0 deletions
Split View
Diff Options
@ -0,0 +1,206 @@ |
|||
diff -rduN SDL-old/src/audio/SDL_mixer_MMX.c SDL-new/src/audio/SDL_mixer_MMX.c
|
|||
--- SDL-old/src/audio/SDL_mixer_MMX.c 2002-11-09 07:13:28.000000000 +0100
|
|||
+++ SDL-new/src/audio/SDL_mixer_MMX.c 2004-05-18 15:28:23.000000000 +0200
|
|||
@@ -19,9 +19,9 @@
|
|||
" movl %1,%%esi\n" // esi = src |
|||
" movl %3,%%eax\n" // eax = volume |
|||
|
|||
-" movl %2,%%ebx\n" // ebx = size
|
|||
+" movl %2,%%edx\n" // edx = size
|
|||
|
|||
-" shrl $4,%%ebx\n" // process 16 bytes per iteration = 8 samples
|
|||
+" shrl $4,%%edx\n" // process 16 bytes per iteration = 8 samples
|
|||
|
|||
" jz .endS16\n" |
|||
|
|||
@@ -95,7 +95,7 @@
|
|||
|
|||
" addl $16,%%edi\n" |
|||
|
|||
-" dec %%ebx\n"
|
|||
+" dec %%edx\n"
|
|||
|
|||
" jnz .mixloopS16\n" |
|||
|
|||
@@ -105,7 +105,7 @@
|
|||
: |
|||
: "m" (dst), "m"(src),"m"(size), |
|||
"m"(volume) |
|||
- : "eax","ebx", "esi", "edi","memory"
|
|||
+ : "eax","edx", "esi", "edi","memory"
|
|||
); |
|||
} |
|||
|
|||
@@ -123,7 +123,7 @@
|
|||
" movl %1,%%esi\n" // esi = src |
|||
" movl %3,%%eax\n" // eax = volume |
|||
|
|||
-" movd %%ebx,%%mm0\n"
|
|||
+" movd %%edx,%%mm0\n"
|
|||
" movq %%mm0,%%mm1\n" |
|||
" psllq $16,%%mm0\n" |
|||
" por %%mm1,%%mm0\n" |
|||
@@ -132,10 +132,10 @@
|
|||
" psllq $16,%%mm0\n" |
|||
" por %%mm1,%%mm0\n" |
|||
|
|||
-" movl %2,%%ebx\n" // ebx = size
|
|||
-" shr $3,%%ebx\n" // process 8 bytes per iteration = 8 samples
|
|||
+" movl %2,%%edx\n" // edx = size
|
|||
+" shr $3,%%edx\n" // process 8 bytes per iteration = 8 samples
|
|||
|
|||
-" cmp $0,%%ebx\n"
|
|||
+" cmp $0,%%edx\n"
|
|||
" je .endS8\n" |
|||
|
|||
".align 16\n" |
|||
@@ -169,7 +169,7 @@
|
|||
" movq %%mm3,(%%edi)\n" // store back to ram |
|||
" addl $8,%%edi\n" |
|||
|
|||
-" dec %%ebx\n"
|
|||
+" dec %%edx\n"
|
|||
|
|||
" jnz .mixloopS8\n" |
|||
|
|||
@@ -178,7 +178,7 @@
|
|||
: |
|||
: "m" (dst), "m"(src),"m"(size), |
|||
"m"(volume) |
|||
- : "eax","ebx", "esi", "edi","memory"
|
|||
+ : "eax","edx", "esi", "edi","memory"
|
|||
); |
|||
} |
|||
#endif |
|||
diff -rduN SDL-old/src/cpuinfo/SDL_cpuinfo.c SDL-new/src/cpuinfo/SDL_cpuinfo.c
|
|||
--- SDL-old/src/cpuinfo/SDL_cpuinfo.c 2004-02-10 16:31:35.000000000 +0100
|
|||
+++ SDL-new/src/cpuinfo/SDL_cpuinfo.c 2004-05-18 15:29:31.000000000 +0200
|
|||
@@ -105,7 +105,7 @@
|
|||
int features = 0; |
|||
#if defined(__GNUC__) && defined(i386) |
|||
__asm__ ( |
|||
-" movl %%ebx,%%edi\n"
|
|||
+" movl %%edx,%%edi\n"
|
|||
" xorl %%eax,%%eax # Set up for CPUID instruction \n" |
|||
" cpuid # Get and save vendor ID \n" |
|||
" cmpl $1,%%eax # Make sure 1 is valid input for CPUID\n" |
|||
@@ -115,10 +115,10 @@
|
|||
" cpuid # Get family/model/stepping/features\n" |
|||
" movl %%edx,%0 \n" |
|||
"1: \n" |
|||
-" movl %%edi,%%ebx\n"
|
|||
+" movl %%edi,%%edx\n"
|
|||
: "=m" (features) |
|||
: |
|||
- : "%eax", "%ebx", "%ecx", "%edx", "%edi"
|
|||
+ : "%eax", "%edx", "%ecx", "%edx", "%edi"
|
|||
); |
|||
#elif defined(_MSC_VER) |
|||
__asm { |
|||
@@ -141,7 +141,7 @@
|
|||
int features = 0; |
|||
#if defined(__GNUC__) && defined(i386) |
|||
__asm__ ( |
|||
-" movl %%ebx,%%edi\n"
|
|||
+" movl %%edx,%%edi\n"
|
|||
" movl $0x80000000,%%eax # Query for extended functions \n" |
|||
" cpuid # Get extended function limit \n" |
|||
" cmpl $0x80000001,%%eax \n" |
|||
@@ -150,10 +150,10 @@
|
|||
" cpuid # and get the information \n" |
|||
" movl %%edx,%0 \n" |
|||
"1: \n" |
|||
-" movl %%edi,%%ebx\n"
|
|||
+" movl %%edi,%%edx\n"
|
|||
: "=m" (features) |
|||
: |
|||
- : "%eax", "%ebx", "%ecx", "%edx", "%edi"
|
|||
+ : "%eax", "%edx", "%ecx", "%edx", "%edi"
|
|||
); |
|||
#elif defined(_MSC_VER) |
|||
__asm { |
|||
diff -rduN SDL-old/src/video/SDL_yuv_mmx.c SDL-new/src/video/SDL_yuv_mmx.c
|
|||
--- SDL-old/src/video/SDL_yuv_mmx.c 2004-02-18 18:22:04.000000000 +0100
|
|||
+++ SDL-new/src/video/SDL_yuv_mmx.c 2004-05-17 22:53:31.000000000 +0200
|
|||
@@ -120,12 +120,12 @@
|
|||
"movd (%2), %%mm2\n" // 0 0 0 0 l3 l2 l1 l0 |
|||
"punpcklbw %%mm7,%%mm1\n" // 0 v3 0 v2 00 v1 00 v0 |
|||
"punpckldq %%mm1,%%mm1\n" // 00 v1 00 v0 00 v1 00 v0 |
|||
- "psubw _MMX_0080w,%%mm1\n" // mm1-128:r1 r1 r0 r0 r1 r1 r0 r0
|
|||
+ "psubw %[_MMX_0080w],%%mm1\n" // mm1-128:r1 r1 r0 r0 r1 r1 r0 r0
|
|||
|
|||
// create Cr_g (result in mm0) |
|||
"movq %%mm1,%%mm0\n" // r1 r1 r0 r0 r1 r1 r0 r0 |
|||
- "pmullw _MMX_VgrnRGB,%%mm0\n"// red*-46dec=0.7136*64
|
|||
- "pmullw _MMX_VredRGB,%%mm1\n"// red*89dec=1.4013*64
|
|||
+ "pmullw %[_MMX_VgrnRGB],%%mm0\n"// red*-46dec=0.7136*64
|
|||
+ "pmullw %[_MMX_VredRGB],%%mm1\n"// red*89dec=1.4013*64
|
|||
"psraw $6, %%mm0\n" // red=red/64 |
|||
"psraw $6, %%mm1\n" // red=red/64 |
|||
|
|||
@@ -134,8 +134,8 @@
|
|||
"movq (%2,%4),%%mm3\n" // 0 0 0 0 L3 L2 L1 L0 |
|||
"punpckldq %%mm3,%%mm2\n" // L3 L2 L1 L0 l3 l2 l1 l0 |
|||
"movq %%mm2,%%mm4\n" // L3 L2 L1 L0 l3 l2 l1 l0 |
|||
- "pand _MMX_FF00w,%%mm2\n" // L3 0 L1 0 l3 0 l1 0
|
|||
- "pand _MMX_00FFw,%%mm4\n" // 0 L2 0 L0 0 l2 0 l0
|
|||
+ "pand %[_MMX_FF00w],%%mm2\n" // L3 0 L1 0 l3 0 l1 0
|
|||
+ "pand %[_MMX_00FFw],%%mm4\n" // 0 L2 0 L0 0 l2 0 l0
|
|||
"psrlw $8,%%mm2\n" // 0 L3 0 L1 0 l3 0 l1 |
|||
|
|||
// create R (result in mm6) |
|||
@@ -152,11 +152,11 @@
|
|||
"movd (%1), %%mm1\n" // 0 0 0 0 u3 u2 u1 u0 |
|||
"punpcklbw %%mm7,%%mm1\n" // 0 u3 0 u2 00 u1 00 u0 |
|||
"punpckldq %%mm1,%%mm1\n" // 00 u1 00 u0 00 u1 00 u0 |
|||
- "psubw _MMX_0080w,%%mm1\n" // mm1-128:u1 u1 u0 u0 u1 u1 u0 u0
|
|||
+ "psubw %[_MMX_0080w],%%mm1\n" // mm1-128:u1 u1 u0 u0 u1 u1 u0 u0
|
|||
// create Cb_g (result in mm5) |
|||
"movq %%mm1,%%mm5\n" // u1 u1 u0 u0 u1 u1 u0 u0 |
|||
- "pmullw _MMX_UgrnRGB,%%mm5\n" // blue*-109dec=1.7129*64
|
|||
- "pmullw _MMX_UbluRGB,%%mm1\n" // blue*114dec=1.78125*64
|
|||
+ "pmullw %[_MMX_UgrnRGB],%%mm5\n" // blue*-109dec=1.7129*64
|
|||
+ "pmullw %[_MMX_UbluRGB],%%mm1\n" // blue*114dec=1.78125*64
|
|||
"psraw $6, %%mm5\n" // blue=red/64 |
|||
"psraw $6, %%mm1\n" // blue=blue/64 |
|||
|
|||
@@ -238,9 +238,15 @@
|
|||
"popl %%ebx\n" |
|||
: |
|||
: "m" (cr), "r"(cb),"r"(lum), |
|||
- "r"(row1),"r"(cols),"r"(row2),"m"(x),"m"(y),"m"(mod)
|
|||
- : "%ebx"
|
|||
- );
|
|||
+ "r"(row1),"r"(cols),"r"(row2),"m"(x),"m"(y),"m"(mod),
|
|||
+ [_MMX_0080w] "m" (*_MMX_0080w),
|
|||
+ [_MMX_00FFw] "m" (*_MMX_00FFw),
|
|||
+ [_MMX_FF00w] "m" (*_MMX_FF00w),
|
|||
+ [_MMX_VgrnRGB] "m" (*_MMX_VgrnRGB),
|
|||
+ [_MMX_VredRGB] "m" (*_MMX_VredRGB),
|
|||
+ [_MMX_UgrnRGB] "m" (*_MMX_UgrnRGB),
|
|||
+ [_MMX_UbluRGB] "m" (*_MMX_UbluRGB)
|
|||
+ );
|
|||
} |
|||
|
|||
void Color565DitherYV12MMX1X( int *colortab, Uint32 *rgb_2_pix, |
|||
@@ -413,9 +419,17 @@
|
|||
"popl %%ebx\n" |
|||
: |
|||
:"m" (cr), "r"(cb),"r"(lum), |
|||
- "r"(row1),"r"(cols),"r"(row2),"m"(x),"m"(y),"m"(mod)
|
|||
- : "%ebx"
|
|||
- );
|
|||
+ "r"(row1),"r"(cols),"r"(row2),"m"(x),"m"(y),"m"(mod),
|
|||
+ [_MMX_0080w] "m" (*_MMX_0080w),
|
|||
+ [_MMX_Ugrn565] "m" (*_MMX_Ugrn565),
|
|||
+ [_MMX_Ublu5x5] "m" (*_MMX_Ublu5x5),
|
|||
+ [_MMX_00FFw] "m" (*_MMX_00FFw),
|
|||
+ [_MMX_Vgrn565] "m" (*_MMX_Vgrn565),
|
|||
+ [_MMX_Vred5x5] "m" (*_MMX_Vred5x5),
|
|||
+ [_MMX_Ycoeff] "m" (*_MMX_Ycoeff),
|
|||
+ [_MMX_red565] "m" (*_MMX_red565),
|
|||
+ [_MMX_grn565] "m" (*_MMX_grn565)
|
|||
+ );
|
|||
} |
|||
|
|||
#endif /* GCC i386 inline assembly */ |
|||