diff options
| author | Loren Merritt <lorenm@u.washington.edu> | 2010-05-19 13:31:11 +0000 |
|---|---|---|
| committer | Loren Merritt <lorenm@u.washington.edu> | 2010-05-19 13:31:11 +0000 |
| commit | 8a26b2c665672a5883b85283c3e2ec5bbb60acec (patch) | |
| tree | 79a75ce19c874c86b40f6f9567e569c4a4fa6b7c | |
| parent | 8d168a9207f231c22a04a5a2b252d0ab89477b02 (diff) | |
| download | ffmpeg-8a26b2c665672a5883b85283c3e2ec5bbb60acec.tar.gz | |
13% faster yuv420 to rgb15 mmx.
It is now faster than the old gpl version on conroe.
Originally committed as revision 31181 to svn://svn.mplayerhq.hu/mplayer/trunk/libswscale
| -rw-r--r-- | libswscale/x86/yuv2rgb_mmx.c | 3 | ||||
| -rw-r--r-- | libswscale/x86/yuv2rgb_template2.c | 37 |
2 files changed, 19 insertions, 21 deletions
diff --git a/libswscale/x86/yuv2rgb_mmx.c b/libswscale/x86/yuv2rgb_mmx.c index d042421ff7..7d3ab7150b 100644 --- a/libswscale/x86/yuv2rgb_mmx.c +++ b/libswscale/x86/yuv2rgb_mmx.c @@ -41,6 +41,9 @@ 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, pb_e0) = 0xe0e0e0e0e0e0e0e0ULL; +DECLARE_ASM_CONST(8, uint64_t, pb_03) = 0x0303030303030303ULL; +DECLARE_ASM_CONST(8, uint64_t, pb_07) = 0x0707070707070707ULL; //MMX versions #undef RENAME diff --git a/libswscale/x86/yuv2rgb_template2.c b/libswscale/x86/yuv2rgb_template2.c index 9f757e6344..5c062c10d4 100644 --- a/libswscale/x86/yuv2rgb_template2.c +++ b/libswscale/x86/yuv2rgb_template2.c @@ -162,31 +162,26 @@ __asm__ volatile (SFENCE"\n\t"EMMS); \ return srcSliceH; \ +#define IF0(x) +#define IF1(x) x -#define RGB_PACK16(gmask, gshift, rshift) \ +#define RGB_PACK16(gmask, is15) \ "pand "MANGLE(mmx_redmask)", %%mm0\n\t" \ "pand "MANGLE(mmx_redmask)", %%mm1\n\t" \ + "movq %%mm2, %%mm3\n\t" \ + "psllw $"AV_STRINGIFY(3-is15)", %%mm2\n\t" \ + "psrlw $"AV_STRINGIFY(5+is15)", %%mm3\n\t" \ "psrlw $3, %%mm0\n\t" \ - "pand "MANGLE(gmask)", %%mm2\n\t" \ - "movq %%mm0, %%mm5\n\t" \ - "movq %%mm1, %%mm6\n\t" \ - "movq %%mm2, %%mm7\n\t" \ - "punpcklbw %%mm4, %%mm0\n\t" \ - "punpcklbw %%mm4, %%mm1\n\t" \ - "punpcklbw %%mm4, %%mm2\n\t" \ - "punpckhbw %%mm4, %%mm5\n\t" \ - "punpckhbw %%mm4, %%mm6\n\t" \ - "punpckhbw %%mm4, %%mm7\n\t" \ - "psllw $"rshift", %%mm1\n\t" \ - "psllw $"rshift", %%mm6\n\t" \ - "psllw $"gshift", %%mm2\n\t" \ - "psllw $"gshift", %%mm7\n\t" \ - "por %%mm1, %%mm0\n\t" \ - "por %%mm6, %%mm5\n\t" \ + IF##is15("psrlw $1, %%mm1\n\t") \ + "pand "MANGLE(pb_e0)", %%mm2\n\t" \ + "pand "MANGLE(gmask)", %%mm3\n\t" \ "por %%mm2, %%mm0\n\t" \ - "por %%mm7, %%mm5\n\t" \ + "por %%mm3, %%mm1\n\t" \ + "movq %%mm0, %%mm2\n\t" \ + "punpcklbw %%mm1, %%mm0\n\t" \ + "punpckhbw %%mm1, %%mm2\n\t" \ MOVNTQ " %%mm0, (%1)\n\t" \ - MOVNTQ " %%mm5, 8(%1)\n\t" \ + MOVNTQ " %%mm2, 8(%1)\n\t" \ #define DITHER_RGB \ "paddusb "BLUE_DITHER"(%4), %%mm0\n\t" \ @@ -214,7 +209,7 @@ static inline int RENAME(yuv420_rgb15)(SwsContext *c, const uint8_t *src[], #ifdef DITHER1XBPP DITHER_RGB #endif - RGB_PACK16(mmx_redmask, "2", "7") + RGB_PACK16(pb_03, 1) YUV2RGB_ENDLOOP(2) YUV2RGB_OPERANDS @@ -242,7 +237,7 @@ static inline int RENAME(yuv420_rgb16)(SwsContext *c, const uint8_t *src[], #ifdef DITHER1XBPP DITHER_RGB #endif - RGB_PACK16(mmx_grnmask, "3", "8") + RGB_PACK16(pb_07, 0) YUV2RGB_ENDLOOP(2) YUV2RGB_OPERANDS |