Chiebot-Mirror
/
FFmpeg
mirror of https://github.com/FFmpeg/FFmpeg.git
8
0
Fork 0
Code Issues Projects Releases Wiki Activity

optimize H264_DEBLOCK_P0_Q0

Browse Source 
2.5% faster filter_mb_fast() on P3

Originally committed as revision 6877 to svn://svn.ffmpeg.org/ffmpeg/trunk
pull/126/head
Michael Niedermayer 18 years ago
parent dd7e46e7c3
commit e9f1885c21
2 changed files with 23 additions and 40 deletions
Whitespace
Show all changes Ignore whitespace when comparing lines Ignore changes in amount of whitespace Ignore changes in whitespace at EOL
Split View
Diff Options
Show Stats Download Patch File Download Diff File
  1. 2
      libavcodec/i386/dsputil_mmx.c
  2. 61
      libavcodec/i386/h264dsp_mmx.c

2
libavcodec/i386/dsputil_mmx.c
Unescape View File

@ -58,6 +58,8 @@ static const uint64_t ff_pb_1 attribute_used __attribute__ ((aligned(8))) = 0x0
static const uint64_t ff_pb_3 attribute_used __attribute__ ((aligned(8))) = 0x0303030303030303ULL;
static const uint64_t ff_pb_7 attribute_used __attribute__ ((aligned(8))) = 0x0707070707070707ULL;
static const uint64_t ff_pb_3F attribute_used __attribute__ ((aligned(8))) = 0x3F3F3F3F3F3F3F3FULL;
static const uint64_t ff_pb_A1 attribute_used __attribute__ ((aligned(8))) = 0xA1A1A1A1A1A1A1A1ULL;
static const uint64_t ff_pb_5F attribute_used __attribute__ ((aligned(8))) = 0x5F5F5F5F5F5F5F5FULL;
static const uint64_t ff_pb_FC attribute_used __attribute__ ((aligned(8))) = 0xFCFCFCFCFCFCFCFCULL;
#define JUMPALIGN() __asm __volatile (ASMALIGN(3)::)

61
libavcodec/i386/h264dsp_mmx.c
Unescape View File

@ -337,46 +337,27 @@ static void ff_h264_idct8_dc_add_mmx2(uint8_t *dst, int16_t *block, int stride)
// out: mm1=p0' mm2=q0'
// clobbers: mm0,3-6
#define H264_DEBLOCK_P0_Q0(pb_01, pb_3f)\
/* a = q0^p0^((p1-q1)>>2) */\
"movq %%mm0, %%mm4 \n\t"\
"psubb %%mm3, %%mm4 \n\t"\
"psrlw $2, %%mm4 \n\t"\
"pxor %%mm1, %%mm4 \n\t"\
"pxor %%mm2, %%mm4 \n\t"\
/* b = p0^(q1>>2) */\
"psrlw $2, %%mm3 \n\t"\
"pand "#pb_3f", %%mm3 \n\t"\
"movq %%mm1, %%mm5 \n\t"\
"pxor %%mm3, %%mm5 \n\t"\
/* c = q0^(p1>>2) */\
"psrlw $2, %%mm0 \n\t"\
"pand "#pb_3f", %%mm0 \n\t"\
"movq %%mm2, %%mm6 \n\t"\
"pxor %%mm0, %%mm6 \n\t"\
/* d = (c^b) & ~(b^a) & 1 */\
"pxor %%mm5, %%mm6 \n\t"\
"pxor %%mm4, %%mm5 \n\t"\
"pandn %%mm6, %%mm5 \n\t"\
"pand "#pb_01", %%mm5 \n\t"\
/* delta = (avg(q0, p1>>2) + (d&a))
* - (avg(p0, q1>>2) + (d&~a)) */\
"pavgb %%mm2, %%mm0 \n\t"\
"pand %%mm5, %%mm4 \n\t"\
"paddusb %%mm4, %%mm0 \n\t"\
"pavgb %%mm1, %%mm3 \n\t"\
"pxor %%mm5, %%mm4 \n\t"\
"paddusb %%mm4, %%mm3 \n\t"\
/* p0 += clip(delta, -tc0, tc0)
* q0 -= clip(delta, -tc0, tc0) */\
"movq %%mm0, %%mm4 \n\t"\
"psubusb %%mm3, %%mm0 \n\t"\
"psubusb %%mm4, %%mm3 \n\t"\
"pminub %%mm7, %%mm0 \n\t"\
"pminub %%mm7, %%mm3 \n\t"\
"paddusb %%mm0, %%mm1 \n\t"\
"paddusb %%mm3, %%mm2 \n\t"\
"psubusb %%mm3, %%mm1 \n\t"\
"psubusb %%mm0, %%mm2 \n\t"
"movq %%mm1 , %%mm5 \n\t"\
"pxor %%mm2 , %%mm5 \n\t" /* p0^q0*/\
"pand "#pb_01" , %%mm5 \n\t" /* (p0^q0)&1*/\
"pcmpeqb %%mm4 , %%mm4 \n\t"\
"pxor %%mm4 , %%mm3 \n\t"\
"pavgb %%mm0 , %%mm3 \n\t" /* (p1 - q1 + 256)>>1*/\
"pavgb "MANGLE(ff_pb_3)" , %%mm3 \n\t" /*(((p1 - q1 + 256)>>1)+4)>>1 = 64+2+(p1-q1)>>2*/\
"pxor %%mm1 , %%mm4 \n\t"\
"pavgb %%mm2 , %%mm4 \n\t" /* (q0 - p0 + 256)>>1*/\
"pavgb %%mm5 , %%mm3 \n\t"\
"paddb %%mm4 , %%mm3 \n\t" /* d+128+33*/\
"pxor %%mm6 , %%mm6 \n\t" /* 0*/\
"psubb %%mm3 , %%mm6 \n\t" /* 128-33-d*/\
"psubusb "MANGLE(ff_pb_A1)" , %%mm3 \n\t"\
"psubusb "MANGLE(ff_pb_5F)" , %%mm6 \n\t"\
"pminub %%mm7 , %%mm3 \n\t"\
"pminub %%mm7 , %%mm6 \n\t"\
"paddusb %%mm3 , %%mm1 \n\t"\
"paddusb %%mm6 , %%mm2 \n\t"\
"psubusb %%mm6 , %%mm1 \n\t"\
"psubusb %%mm3 , %%mm2 \n\t"
// in: mm0=p1 mm1=p0 mm2=q0 mm3=q1 mm7=(tc&mask) %8=mm_bone
// out: (q1addr) = clip( (q2+((p0+q0+1)>>1))>>1, q1-tc0, q1+tc0 )

Write Preview
Loading…
Cancel
Save