Mercurial > libavcodec.hg

comparison x86/dsputil_h264_template_mmx.c @ 8519:cc64e1343397 libavcodec

Find changesets by keywords (author, files, the commit message), revision number or hash, or revset expression.
Use H264 MMX chroma functions to accelerate RV40 decoding. Patch by Mathieu Velten (matmaul A gmail)
author cehoyos
date Sun, 04 Jan 2009 01:36:11 +0000
parents 7768bdfd4f7b
children 98970e51365a
comparison
equal deleted inserted replaced
8518:f2c406b05158 8519:cc64e1343397
23 * MMX optimized version of (put|avg)_h264_chroma_mc8. 23 * MMX optimized version of (put|avg)_h264_chroma_mc8.
24 * H264_CHROMA_MC8_TMPL must be defined to the desired function name 24 * H264_CHROMA_MC8_TMPL must be defined to the desired function name
25 * H264_CHROMA_OP must be defined to empty for put and pavgb/pavgusb for avg 25 * H264_CHROMA_OP must be defined to empty for put and pavgb/pavgusb for avg
26 * H264_CHROMA_MC8_MV0 must be defined to a (put|avg)_pixels8 function 26 * H264_CHROMA_MC8_MV0 must be defined to a (put|avg)_pixels8 function
27 */ 27 */
28 static void H264_CHROMA_MC8_TMPL(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int stride, int h, int x, int y, int rnd) 28 static void H264_CHROMA_MC8_TMPL(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int stride, int h, int x, int y, const uint64_t *rnd_reg)
29 { 29 {
30 const uint64_t *rnd_reg;
31 DECLARE_ALIGNED_8(uint64_t, AA); 30 DECLARE_ALIGNED_8(uint64_t, AA);
32 DECLARE_ALIGNED_8(uint64_t, DD); 31 DECLARE_ALIGNED_8(uint64_t, DD);
33 int i; 32 int i;
34 33
35 if(y==0 && x==0) { 34 if(y==0 && x==0) {
43 if(y==0 || x==0) 42 if(y==0 || x==0)
44 { 43 {
45 /* 1 dimensional filter only */ 44 /* 1 dimensional filter only */
46 const int dxy = x ? 1 : stride; 45 const int dxy = x ? 1 : stride;
47 46
48 rnd_reg = rnd ? &ff_pw_4 : &ff_pw_3;
49
50 __asm__ volatile( 47 __asm__ volatile(
51 "movd %0, %%mm5\n\t" 48 "movd %0, %%mm5\n\t"
52 "movq %1, %%mm4\n\t" 49 "movq %1, %%mm4\n\t"
53 "movq %2, %%mm6\n\t" /* mm6 = rnd */ 50 "movq %2, %%mm6\n\t" /* mm6 = rnd >> 3 */
54 "punpcklwd %%mm5, %%mm5\n\t" 51 "punpcklwd %%mm5, %%mm5\n\t"
55 "punpckldq %%mm5, %%mm5\n\t" /* mm5 = B = x */ 52 "punpckldq %%mm5, %%mm5\n\t" /* mm5 = B = x */
56 "pxor %%mm7, %%mm7\n\t" 53 "pxor %%mm7, %%mm7\n\t"
57 "psubw %%mm5, %%mm4\n\t" /* mm4 = A = 8-x */ 54 "psubw %%mm5, %%mm4\n\t" /* mm4 = A = 8-x */
58 :: "rm"(x+y), "m"(ff_pw_8), "m"(*rnd_reg)); 55 :: "rm"(x+y), "m"(ff_pw_8), "m"(*(rnd_reg+1)));
59 56
60 for(i=0; i<h; i++) { 57 for(i=0; i<h; i++) {
61 __asm__ volatile( 58 __asm__ volatile(
62 /* mm0 = src[0..7], mm1 = src[1..8] */ 59 /* mm0 = src[0..7], mm1 = src[1..8] */
63 "movq %0, %%mm0\n\t" 60 "movq %0, %%mm0\n\t"
76 "pmullw %%mm4, %%mm0\n\t" 73 "pmullw %%mm4, %%mm0\n\t"
77 "pmullw %%mm4, %%mm1\n\t" 74 "pmullw %%mm4, %%mm1\n\t"
78 "pmullw %%mm5, %%mm2\n\t" 75 "pmullw %%mm5, %%mm2\n\t"
79 "pmullw %%mm5, %%mm3\n\t" 76 "pmullw %%mm5, %%mm3\n\t"
80 77
81 /* dst[0..7] = (A * src[0..7] + B * src[1..8] + 4) >> 3 */ 78 /* dst[0..7] = (A * src[0..7] + B * src[1..8] + (rnd >> 3)) >> 3 */
82 "paddw %%mm6, %%mm0\n\t" 79 "paddw %%mm6, %%mm0\n\t"
83 "paddw %%mm6, %%mm1\n\t" 80 "paddw %%mm6, %%mm1\n\t"
84 "paddw %%mm2, %%mm0\n\t" 81 "paddw %%mm2, %%mm0\n\t"
85 "paddw %%mm3, %%mm1\n\t" 82 "paddw %%mm3, %%mm1\n\t"
86 "psrlw $3, %%mm0\n\t" 83 "psrlw $3, %%mm0\n\t"
95 } 92 }
96 return; 93 return;
97 } 94 }
98 95
99 /* general case, bilinear */ 96 /* general case, bilinear */
100 rnd_reg = rnd ? &ff_pw_32.a : &ff_pw_28.a;
101 __asm__ volatile("movd %2, %%mm4\n\t" 97 __asm__ volatile("movd %2, %%mm4\n\t"
102 "movd %3, %%mm6\n\t" 98 "movd %3, %%mm6\n\t"
103 "punpcklwd %%mm4, %%mm4\n\t" 99 "punpcklwd %%mm4, %%mm4\n\t"
104 "punpcklwd %%mm6, %%mm6\n\t" 100 "punpcklwd %%mm6, %%mm6\n\t"
105 "punpckldq %%mm4, %%mm4\n\t" /* mm4 = x words */ 101 "punpckldq %%mm4, %%mm4\n\t" /* mm4 = x words */
170 "paddw %%mm4, %%mm3\n\t" 166 "paddw %%mm4, %%mm3\n\t"
171 "movq %0, %%mm0\n\t" 167 "movq %0, %%mm0\n\t"
172 : : "m" (src[0]), "m" (src[1]), "m" (DD)); 168 : : "m" (src[0]), "m" (src[1]), "m" (DD));
173 169
174 __asm__ volatile( 170 __asm__ volatile(
175 /* dst[0..7] = ([mm2,mm3] + 32) >> 6 */ 171 /* dst[0..7] = ([mm2,mm3] + rnd) >> 6 */
176 "paddw %1, %%mm2\n\t" 172 "paddw %1, %%mm2\n\t"
177 "paddw %1, %%mm3\n\t" 173 "paddw %1, %%mm3\n\t"
178 "psrlw $6, %%mm2\n\t" 174 "psrlw $6, %%mm2\n\t"
179 "psrlw $6, %%mm3\n\t" 175 "psrlw $6, %%mm3\n\t"
180 "packuswb %%mm3, %%mm2\n\t" 176 "packuswb %%mm3, %%mm2\n\t"
183 : "=m" (dst[0]) : "m" (*rnd_reg)); 179 : "=m" (dst[0]) : "m" (*rnd_reg));
184 dst+= stride; 180 dst+= stride;
185 } 181 }
186 } 182 }
187 183
188 static void H264_CHROMA_MC4_TMPL(uint8_t *dst/*align 4*/, uint8_t *src/*align 1*/, int stride, int h, int x, int y) 184 static void H264_CHROMA_MC4_TMPL(uint8_t *dst/*align 4*/, uint8_t *src/*align 1*/, int stride, int h, int x, int y, const uint64_t *rnd_reg)
189 { 185 {
190 __asm__ volatile( 186 __asm__ volatile(
191 "pxor %%mm7, %%mm7 \n\t" 187 "pxor %%mm7, %%mm7 \n\t"
192 "movd %5, %%mm2 \n\t" 188 "movd %5, %%mm2 \n\t"
193 "movd %6, %%mm3 \n\t" 189 "movd %6, %%mm3 \n\t"
247 "movd %%mm1, (%0) \n\t" 243 "movd %%mm1, (%0) \n\t"
248 "add %3, %0 \n\t" 244 "add %3, %0 \n\t"
249 "sub $2, %2 \n\t" 245 "sub $2, %2 \n\t"
250 "jnz 1b \n\t" 246 "jnz 1b \n\t"
251 : "+r"(dst), "+r"(src), "+r"(h) 247 : "+r"(dst), "+r"(src), "+r"(h)
252 : "r"((x86_reg)stride), "m"(ff_pw_32), "m"(x), "m"(y) 248 : "r"((x86_reg)stride), "m"(*rnd_reg), "m"(x), "m"(y)
253 ); 249 );
254 } 250 }
255 251
256 #ifdef H264_CHROMA_MC2_TMPL 252 #ifdef H264_CHROMA_MC2_TMPL
257 static void H264_CHROMA_MC2_TMPL(uint8_t *dst/*align 2*/, uint8_t *src/*align 1*/, int stride, int h, int x, int y) 253 static void H264_CHROMA_MC2_TMPL(uint8_t *dst/*align 2*/, uint8_t *src/*align 1*/, int stride, int h, int x, int y)