falkTX
/
FFmpeg
mirror of https://github.com/falkTX/FFmpeg.git
1
0
Fork 0
Code Issues 0 Releases 338 Wiki Activity
Browse Source

Make some assembler constants global instead of declaring them multiple times. 

Originally committed as revision 25910 to svn://svn.mplayerhq.hu/mplayer/trunk/libswscale
tags/v0.5
Reimar Döffinger 18 years ago
parent
d6773f563d
commit
5802683a97
6 changed files with 43 additions and 60 deletions
Split View
Diff Options
Show Stats Download Patch File Download Diff File
  1. +0
    -13
      libswscale/rgb2rgb.c
  2. +11
    -11
      libswscale/rgb2rgb_template.c
  3. +13
    -13
      libswscale/swscale.c
  4. +15
    -15
      libswscale/swscale_template.c
  5. +0
    -4
      libswscale/yuv2rgb.c
  6. +4
    -4
      libswscale/yuv2rgb_template.c

+ 0
- 13
libswscale/rgb2rgb.c View File

@@ -122,19 +122,6 @@ DECLARE_ASM_CONST(8, uint64_t, red_15mask) = 0x00007c0000007c00ULL;
DECLARE_ASM_CONST(8, uint64_t, green_15mask) = 0x000003e0000003e0ULL;
DECLARE_ASM_CONST(8, uint64_t, blue_15mask) = 0x0000001f0000001fULL;

#ifdef FAST_BGR2YV12
DECLARE_ASM_CONST(8, uint64_t, bgr2YCoeff) = 0x000000210041000DULL;
DECLARE_ASM_CONST(8, uint64_t, bgr2UCoeff) = 0x0000FFEEFFDC0038ULL;
DECLARE_ASM_CONST(8, uint64_t, bgr2VCoeff) = 0x00000038FFD2FFF8ULL;
#else
DECLARE_ASM_CONST(8, uint64_t, bgr2YCoeff) = 0x000020E540830C8BULL;
DECLARE_ASM_CONST(8, uint64_t, bgr2UCoeff) = 0x0000ED0FDAC23831ULL;
DECLARE_ASM_CONST(8, uint64_t, bgr2VCoeff) = 0x00003831D0E6F6EAULL;
#endif
DECLARE_ASM_CONST(8, uint64_t, bgr2YOffset) = 0x1010101010101010ULL;
DECLARE_ASM_CONST(8, uint64_t, bgr2UVOffset) = 0x8080808080808080ULL;
DECLARE_ASM_CONST(8, uint64_t, w1111) = 0x0001000100010001ULL;

#if 0
static volatile uint64_t __attribute__((aligned(8))) b5Dither;
static volatile uint64_t __attribute__((aligned(8))) g5Dither;


+ 11
- 11
libswscale/rgb2rgb_template.c View File

@@ -2147,8 +2147,8 @@ static inline void RENAME(rgb24toyv12)(const uint8_t *src, uint8_t *ydst, uint8_
{
asm volatile(
"mov %2, %%"REG_a" \n\t"
"movq "MANGLE(bgr2YCoeff)", %%mm6 \n\t"
"movq "MANGLE(w1111)", %%mm5 \n\t"
"movq "MANGLE(ff_bgr2YCoeff)", %%mm6 \n\t"
"movq "MANGLE(ff_w1111)", %%mm5 \n\t"
"pxor %%mm7, %%mm7 \n\t"
"lea (%%"REG_a", %%"REG_a", 2), %%"REG_d" \n\t"
ASMALIGN(4)
@@ -2206,7 +2206,7 @@ static inline void RENAME(rgb24toyv12)(const uint8_t *src, uint8_t *ydst, uint8_
"psraw $7, %%mm4 \n\t"

"packuswb %%mm4, %%mm0 \n\t"
"paddusb "MANGLE(bgr2YOffset)", %%mm0 \n\t"
"paddusb "MANGLE(ff_bgr2YOffset)", %%mm0 \n\t"

MOVNTQ" %%mm0, (%1, %%"REG_a") \n\t"
"add $8, %%"REG_a" \n\t"
@@ -2220,8 +2220,8 @@ static inline void RENAME(rgb24toyv12)(const uint8_t *src, uint8_t *ydst, uint8_
src -= srcStride*2;
asm volatile(
"mov %4, %%"REG_a" \n\t"
"movq "MANGLE(w1111)", %%mm5 \n\t"
"movq "MANGLE(bgr2UCoeff)", %%mm6 \n\t"
"movq "MANGLE(ff_w1111)", %%mm5 \n\t"
"movq "MANGLE(ff_bgr2UCoeff)", %%mm6 \n\t"
"pxor %%mm7, %%mm7 \n\t"
"lea (%%"REG_a", %%"REG_a", 2), %%"REG_d" \n\t"
"add %%"REG_d", %%"REG_d" \n\t"
@@ -2270,8 +2270,8 @@ static inline void RENAME(rgb24toyv12)(const uint8_t *src, uint8_t *ydst, uint8_
"psrlw $2, %%mm0 \n\t"
"psrlw $2, %%mm2 \n\t"
#endif
"movq "MANGLE(bgr2VCoeff)", %%mm1 \n\t"
"movq "MANGLE(bgr2VCoeff)", %%mm3 \n\t"
"movq "MANGLE(ff_bgr2VCoeff)", %%mm1 \n\t"
"movq "MANGLE(ff_bgr2VCoeff)", %%mm3 \n\t"

"pmaddwd %%mm0, %%mm1 \n\t"
"pmaddwd %%mm2, %%mm3 \n\t"
@@ -2328,12 +2328,12 @@ static inline void RENAME(rgb24toyv12)(const uint8_t *src, uint8_t *ydst, uint8_
"paddw %%mm1, %%mm5 \n\t"
"paddw %%mm3, %%mm2 \n\t"
"paddw %%mm5, %%mm2 \n\t"
"movq "MANGLE(w1111)", %%mm5 \n\t"
"movq "MANGLE(ff_w1111)", %%mm5 \n\t"
"psrlw $2, %%mm4 \n\t"
"psrlw $2, %%mm2 \n\t"
#endif
"movq "MANGLE(bgr2VCoeff)", %%mm1 \n\t"
"movq "MANGLE(bgr2VCoeff)", %%mm3 \n\t"
"movq "MANGLE(ff_bgr2VCoeff)", %%mm1 \n\t"
"movq "MANGLE(ff_bgr2VCoeff)", %%mm3 \n\t"

"pmaddwd %%mm4, %%mm1 \n\t"
"pmaddwd %%mm2, %%mm3 \n\t"
@@ -2357,7 +2357,7 @@ static inline void RENAME(rgb24toyv12)(const uint8_t *src, uint8_t *ydst, uint8_
"punpckldq %%mm4, %%mm0 \n\t"
"punpckhdq %%mm4, %%mm1 \n\t"
"packsswb %%mm1, %%mm0 \n\t"
"paddb "MANGLE(bgr2UVOffset)", %%mm0 \n\t"
"paddb "MANGLE(ff_bgr2UVOffset)", %%mm0 \n\t"
"movd %%mm0, (%2, %%"REG_a") \n\t"
"punpckhdq %%mm0, %%mm0 \n\t"
"movd %%mm0, (%3, %%"REG_a") \n\t"


+ 13
- 13
libswscale/swscale.c View File

@@ -209,22 +209,22 @@ DECLARE_ASM_CONST(8, uint64_t, b15Mask)= 0x001F001F001F001FLL;
DECLARE_ASM_CONST(8, uint64_t, g15Mask)= 0x03E003E003E003E0LL;
DECLARE_ASM_CONST(8, uint64_t, r15Mask)= 0x7C007C007C007C00LL;

DECLARE_ASM_CONST(8, uint64_t, M24A)= 0x00FF0000FF0000FFLL;
DECLARE_ASM_CONST(8, uint64_t, M24B)= 0xFF0000FF0000FF00LL;
DECLARE_ASM_CONST(8, uint64_t, M24C)= 0x0000FF0000FF0000LL;
DECLARE_ALIGNED(8, const uint64_t, ff_M24A) = 0x00FF0000FF0000FFLL;
DECLARE_ALIGNED(8, const uint64_t, ff_M24B) = 0xFF0000FF0000FF00LL;
DECLARE_ALIGNED(8, const uint64_t, ff_M24C) = 0x0000FF0000FF0000LL;

#ifdef FAST_BGR2YV12
DECLARE_ASM_CONST(8, uint64_t, bgr2YCoeff) = 0x000000210041000DULL;
DECLARE_ASM_CONST(8, uint64_t, bgr2UCoeff) = 0x0000FFEEFFDC0038ULL;
DECLARE_ASM_CONST(8, uint64_t, bgr2VCoeff) = 0x00000038FFD2FFF8ULL;
DECLARE_ALIGNED(8, const uint64_t, ff_bgr2YCoeff) = 0x000000210041000DULL;
DECLARE_ALIGNED(8, const uint64_t, ff_bgr2UCoeff) = 0x0000FFEEFFDC0038ULL;
DECLARE_ALIGNED(8, const uint64_t, ff_bgr2VCoeff) = 0x00000038FFD2FFF8ULL;
#else
DECLARE_ASM_CONST(8, uint64_t, bgr2YCoeff) = 0x000020E540830C8BULL;
DECLARE_ASM_CONST(8, uint64_t, bgr2UCoeff) = 0x0000ED0FDAC23831ULL;
DECLARE_ASM_CONST(8, uint64_t, bgr2VCoeff) = 0x00003831D0E6F6EAULL;
DECLARE_ALIGNED(8, const uint64_t, ff_bgr2YCoeff) = 0x000020E540830C8BULL;
DECLARE_ALIGNED(8, const uint64_t, ff_bgr2UCoeff) = 0x0000ED0FDAC23831ULL;
DECLARE_ALIGNED(8, const uint64_t, ff_bgr2VCoeff) = 0x00003831D0E6F6EAULL;
#endif /* FAST_BGR2YV12 */
DECLARE_ASM_CONST(8, uint64_t, bgr2YOffset) = 0x1010101010101010ULL;
DECLARE_ASM_CONST(8, uint64_t, bgr2UVOffset) = 0x8080808080808080ULL;
DECLARE_ASM_CONST(8, uint64_t, w1111) = 0x0001000100010001ULL;
DECLARE_ALIGNED(8, const uint64_t, ff_bgr2YOffset) = 0x1010101010101010ULL;
DECLARE_ALIGNED(8, const uint64_t, ff_bgr2UVOffset) = 0x8080808080808080ULL;
DECLARE_ALIGNED(8, const uint64_t, ff_w1111) = 0x0001000100010001ULL;
#endif /* defined(ARCH_X86) */

// clipping helper table for C implementations:
@@ -367,7 +367,7 @@ void in_asm_used_var_warning_killer()
{
volatile int i= bF8+bFC+w10+
bm00001111+bm00000111+bm11111000+b16Mask+g16Mask+r16Mask+b15Mask+g15Mask+r15Mask+
M24A+M24B+M24C+w02 + b5Dither+g5Dither+r5Dither+g6Dither+dither4[0]+dither8[0]+bm01010101;
ff_M24A+ff_M24B+ff_M24C+w02 + b5Dither+g5Dither+r5Dither+g6Dither+dither4[0]+dither8[0]+bm01010101;
if (i) i=0;
}
#endif


+ 15
- 15
libswscale/swscale_template.c View File

@@ -855,8 +855,8 @@

#define WRITEBGR24MMX2(dst, dstw, index) \
/* mm2=B, %%mm4=G, %%mm5=R, %%mm7=0 */\
"movq "MANGLE(M24A)", %%mm0 \n\t"\
"movq "MANGLE(M24C)", %%mm7 \n\t"\
"movq "MANGLE(ff_M24A)", %%mm0 \n\t"\
"movq "MANGLE(ff_M24C)", %%mm7 \n\t"\
"pshufw $0x50, %%mm2, %%mm1 \n\t" /* B3 B2 B3 B2 B1 B0 B1 B0 */\
"pshufw $0x50, %%mm4, %%mm3 \n\t" /* G3 G2 G3 G2 G1 G0 G1 G0 */\
"pshufw $0x00, %%mm5, %%mm6 \n\t" /* R1 R0 R1 R0 R1 R0 R1 R0 */\
@@ -875,7 +875,7 @@
"pshufw $0x55, %%mm4, %%mm3 \n\t" /* G4 G3 G4 G3 G4 G3 G4 G3 */\
"pshufw $0xA5, %%mm5, %%mm6 \n\t" /* R5 R4 R5 R4 R3 R2 R3 R2 */\
\
"pand "MANGLE(M24B)", %%mm1 \n\t" /* B5 B4 B3 */\
"pand "MANGLE(ff_M24B)", %%mm1 \n\t" /* B5 B4 B3 */\
"pand %%mm7, %%mm3 \n\t" /* G4 G3 */\
"pand %%mm0, %%mm6 \n\t" /* R4 R3 R2 */\
\
@@ -889,7 +889,7 @@
\
"pand %%mm7, %%mm1 \n\t" /* B7 B6 */\
"pand %%mm0, %%mm3 \n\t" /* G7 G6 G5 */\
"pand "MANGLE(M24B)", %%mm6 \n\t" /* R7 R6 R5 */\
"pand "MANGLE(ff_M24B)", %%mm6 \n\t" /* R7 R6 R5 */\
\
"por %%mm1, %%mm3 \n\t"\
"por %%mm3, %%mm6 \n\t"\
@@ -1859,8 +1859,8 @@ static inline void RENAME(bgr24ToY)(uint8_t *dst, uint8_t *src, long width)
#ifdef HAVE_MMX
asm volatile(
"mov %2, %%"REG_a" \n\t"
"movq "MANGLE(bgr2YCoeff)", %%mm6 \n\t"
"movq "MANGLE(w1111)", %%mm5 \n\t"
"movq "MANGLE(ff_bgr2YCoeff)", %%mm6 \n\t"
"movq "MANGLE(ff_w1111)", %%mm5 \n\t"
"pxor %%mm7, %%mm7 \n\t"
"lea (%%"REG_a", %%"REG_a", 2), %%"REG_d" \n\t"
ASMALIGN(4)
@@ -1918,7 +1918,7 @@ static inline void RENAME(bgr24ToY)(uint8_t *dst, uint8_t *src, long width)
"psraw $7, %%mm4 \n\t"

"packuswb %%mm4, %%mm0 \n\t"
"paddusb "MANGLE(bgr2YOffset)", %%mm0 \n\t"
"paddusb "MANGLE(ff_bgr2YOffset)", %%mm0 \n\t"

"movq %%mm0, (%1, %%"REG_a") \n\t"
"add $8, %%"REG_a" \n\t"
@@ -1944,8 +1944,8 @@ static inline void RENAME(bgr24ToUV)(uint8_t *dstU, uint8_t *dstV, uint8_t *src1
#ifdef HAVE_MMX
asm volatile(
"mov %3, %%"REG_a" \n\t"
"movq "MANGLE(w1111)", %%mm5 \n\t"
"movq "MANGLE(bgr2UCoeff)", %%mm6 \n\t"
"movq "MANGLE(ff_w1111)", %%mm5 \n\t"
"movq "MANGLE(ff_bgr2UCoeff)", %%mm6 \n\t"
"pxor %%mm7, %%mm7 \n\t"
"lea (%%"REG_a", %%"REG_a", 2), %%"REG_d" \n\t"
"add %%"REG_d", %%"REG_d" \n\t"
@@ -1977,8 +1977,8 @@ static inline void RENAME(bgr24ToUV)(uint8_t *dstU, uint8_t *dstV, uint8_t *src1
"psrlw $1, %%mm0 \n\t"
"psrlw $1, %%mm2 \n\t"
#endif
"movq "MANGLE(bgr2VCoeff)", %%mm1 \n\t"
"movq "MANGLE(bgr2VCoeff)", %%mm3 \n\t"
"movq "MANGLE(ff_bgr2VCoeff)", %%mm1 \n\t"
"movq "MANGLE(ff_bgr2VCoeff)", %%mm3 \n\t"

"pmaddwd %%mm0, %%mm1 \n\t"
"pmaddwd %%mm2, %%mm3 \n\t"
@@ -2019,12 +2019,12 @@ static inline void RENAME(bgr24ToUV)(uint8_t *dstU, uint8_t *dstV, uint8_t *src1
"punpcklbw %%mm7, %%mm5 \n\t"
"punpcklbw %%mm7, %%mm2 \n\t"
"paddw %%mm5, %%mm2 \n\t"
"movq "MANGLE(w1111)", %%mm5 \n\t"
"movq "MANGLE(ff_w1111)", %%mm5 \n\t"
"psrlw $2, %%mm4 \n\t"
"psrlw $2, %%mm2 \n\t"
#endif
"movq "MANGLE(bgr2VCoeff)", %%mm1 \n\t"
"movq "MANGLE(bgr2VCoeff)", %%mm3 \n\t"
"movq "MANGLE(ff_bgr2VCoeff)", %%mm1 \n\t"
"movq "MANGLE(ff_bgr2VCoeff)", %%mm3 \n\t"

"pmaddwd %%mm4, %%mm1 \n\t"
"pmaddwd %%mm2, %%mm3 \n\t"
@@ -2048,7 +2048,7 @@ static inline void RENAME(bgr24ToUV)(uint8_t *dstU, uint8_t *dstV, uint8_t *src1
"punpckldq %%mm4, %%mm0 \n\t"
"punpckhdq %%mm4, %%mm1 \n\t"
"packsswb %%mm1, %%mm0 \n\t"
"paddb "MANGLE(bgr2UVOffset)", %%mm0 \n\t"
"paddb "MANGLE(ff_bgr2UVOffset)", %%mm0 \n\t"

"movd %%mm0, (%1, %%"REG_a") \n\t"
"punpckhdq %%mm0, %%mm0 \n\t"


+ 0
- 4
libswscale/yuv2rgb.c View File

@@ -163,10 +163,6 @@ DECLARE_ASM_CONST(8, uint64_t, mmx_00ffw) = 0x00ff00ff00ff00ffULL;
DECLARE_ASM_CONST(8, uint64_t, mmx_redmask) = 0xf8f8f8f8f8f8f8f8ULL;
DECLARE_ASM_CONST(8, uint64_t, mmx_grnmask) = 0xfcfcfcfcfcfcfcfcULL;

DECLARE_ASM_CONST(8, uint64_t, M24A)= 0x00FF0000FF0000FFULL;
DECLARE_ASM_CONST(8, uint64_t, M24B)= 0xFF0000FF0000FF00ULL;
DECLARE_ASM_CONST(8, uint64_t, M24C)= 0x0000FF0000FF0000ULL;

// the volatile is required because gcc otherwise optimizes some writes away not knowing that these
// are read in the asm block
static volatile uint64_t attribute_used __attribute__((aligned(8))) b5Dither;


+ 4
- 4
libswscale/yuv2rgb_template.c View File

@@ -338,8 +338,8 @@ static inline int RENAME(yuv420_rgb24)(SwsContext *c, uint8_t* src[], int srcStr
YUV2RGB
/* mm0=B, %%mm2=G, %%mm1=R */
#ifdef HAVE_MMX2
"movq "MANGLE(M24A)", %%mm4 \n\t"
"movq "MANGLE(M24C)", %%mm7 \n\t"
"movq "MANGLE(ff_M24A)", %%mm4 \n\t"
"movq "MANGLE(ff_M24C)", %%mm7 \n\t"
"pshufw $0x50, %%mm0, %%mm5 \n\t" /* B3 B2 B3 B2 B1 B0 B1 B0 */
"pshufw $0x50, %%mm2, %%mm3 \n\t" /* G3 G2 G3 G2 G1 G0 G1 G0 */
"pshufw $0x00, %%mm1, %%mm6 \n\t" /* R1 R0 R1 R0 R1 R0 R1 R0 */
@@ -358,7 +358,7 @@ YUV2RGB
"pshufw $0x55, %%mm2, %%mm3 \n\t" /* G4 G3 G4 G3 G4 G3 G4 G3 */
"pshufw $0xA5, %%mm1, %%mm6 \n\t" /* R5 R4 R5 R4 R3 R2 R3 R2 */

"pand "MANGLE(M24B)", %%mm5 \n\t" /* B5 B4 B3 */
"pand "MANGLE(ff_M24B)", %%mm5 \n\t" /* B5 B4 B3 */
"pand %%mm7, %%mm3 \n\t" /* G4 G3 */
"pand %%mm4, %%mm6 \n\t" /* R4 R3 R2 */

@@ -373,7 +373,7 @@ YUV2RGB

"pand %%mm7, %%mm5 \n\t" /* B7 B6 */
"pand %%mm4, %%mm3 \n\t" /* G7 G6 G5 */
"pand "MANGLE(M24B)", %%mm6 \n\t" /* R7 R6 R5 */
"pand "MANGLE(ff_M24B)", %%mm6 \n\t" /* R7 R6 R5 */
"movd 4 (%3, %0), %%mm1;" /* Load 4 Cr 00 00 00 00 v3 v2 v1 v0 */
\
"por %%mm5, %%mm3 \n\t"


Write Preview
Loading…
Cancel
Save