Mercurial > libavcodec.hg

comparison i386/h264dsp_mmx.c @ 6320:ffb2a7b80d6d libavcodec

Find changesets by keywords (author, files, the commit message), revision number or hash, or revset expression.
ff_h264_idct8_add_sse2. compared to mmx, 217->126 cycles on core2, 262->220 on k8.
author lorenm
date Sun, 03 Feb 2008 07:05:11 +0000
parents 4089a1ae6558
children 57bd93f81a14
comparison
equal deleted inserted replaced
6319:4089a1ae6558 6320:ffb2a7b80d6d
73 73
74 /* mm2=s02+s13 mm3=s02-s13 mm4=d02+d13 mm1=d02-d13 */ 74 /* mm2=s02+s13 mm3=s02-s13 mm4=d02+d13 mm1=d02-d13 */
75 IDCT4_1D( %%mm4, %%mm2, %%mm3, %%mm0, %%mm1 ) 75 IDCT4_1D( %%mm4, %%mm2, %%mm3, %%mm0, %%mm1 )
76 76
77 "pxor %%mm7, %%mm7 \n\t" 77 "pxor %%mm7, %%mm7 \n\t"
78 :: "m"(ff_pw_32)); 78 :: "m"(*ff_pw_32));
79 79
80 asm volatile( 80 asm volatile(
81 STORE_DIFF_4P( %%mm0, %%mm1, %%mm7) 81 STORE_DIFF_4P( %%mm0, %%mm1, %%mm7)
82 "add %1, %0 \n\t" 82 "add %1, %0 \n\t"
83 STORE_DIFF_4P( %%mm2, %%mm1, %%mm7) 83 STORE_DIFF_4P( %%mm2, %%mm1, %%mm7)
207 : "memory" 207 : "memory"
208 ); 208 );
209 } 209 }
210 210
211 add_pixels_clamped_mmx(b2, dst, stride); 211 add_pixels_clamped_mmx(b2, dst, stride);
212 }
213
214 #define STORE_DIFF_8P( p, d, t, z )\
215 "movq "#d", "#t" \n"\
216 "psraw $6, "#p" \n"\
217 "punpcklbw "#z", "#t" \n"\
218 "paddsw "#t", "#p" \n"\
219 "packuswb "#p", "#p" \n"\
220 "movq "#p", "#d" \n"
221
222 #define H264_IDCT8_1D_SSE2(a,b,c,d,e,f,g,h)\
223 "movdqa "#c", "#a" \n"\
224 "movdqa "#g", "#e" \n"\
225 "psraw $1, "#c" \n"\
226 "psraw $1, "#g" \n"\
227 "psubw "#e", "#c" \n"\
228 "paddw "#a", "#g" \n"\
229 "movdqa "#b", "#e" \n"\
230 "psraw $1, "#e" \n"\
231 "paddw "#b", "#e" \n"\
232 "paddw "#d", "#e" \n"\
233 "paddw "#f", "#e" \n"\
234 "movdqa "#f", "#a" \n"\
235 "psraw $1, "#a" \n"\
236 "paddw "#f", "#a" \n"\
237 "paddw "#h", "#a" \n"\
238 "psubw "#b", "#a" \n"\
239 "psubw "#d", "#b" \n"\
240 "psubw "#d", "#f" \n"\
241 "paddw "#h", "#b" \n"\
242 "psubw "#h", "#f" \n"\
243 "psraw $1, "#d" \n"\
244 "psraw $1, "#h" \n"\
245 "psubw "#d", "#b" \n"\
246 "psubw "#h", "#f" \n"\
247 "movdqa "#e", "#d" \n"\
248 "movdqa "#a", "#h" \n"\
249 "psraw $2, "#d" \n"\
250 "psraw $2, "#h" \n"\
251 "paddw "#f", "#d" \n"\
252 "paddw "#b", "#h" \n"\
253 "psraw $2, "#f" \n"\
254 "psraw $2, "#b" \n"\
255 "psubw "#f", "#e" \n"\
256 "psubw "#a", "#b" \n"\
257 "movdqa 0x00(%1), "#a" \n"\
258 "movdqa 0x40(%1), "#f" \n"\
259 SUMSUB_BA(f, a)\
260 SUMSUB_BA(g, f)\
261 SUMSUB_BA(c, a)\
262 SUMSUB_BA(e, g)\
263 SUMSUB_BA(b, c)\
264 SUMSUB_BA(h, a)\
265 SUMSUB_BA(d, f)
266
267 static void ff_h264_idct8_add_sse2(uint8_t *dst, int16_t *block, int stride)
268 {
269 asm volatile(
270 "movdqa 0x10(%1), %%xmm1 \n"
271 "movdqa 0x20(%1), %%xmm2 \n"
272 "movdqa 0x30(%1), %%xmm3 \n"
273 "movdqa 0x50(%1), %%xmm5 \n"
274 "movdqa 0x60(%1), %%xmm6 \n"
275 "movdqa 0x70(%1), %%xmm7 \n"
276 H264_IDCT8_1D_SSE2(%%xmm0, %%xmm1, %%xmm2, %%xmm3, %%xmm4, %%xmm5, %%xmm6, %%xmm7)
277 TRANSPOSE8(%%xmm4, %%xmm1, %%xmm7, %%xmm3, %%xmm5, %%xmm0, %%xmm2, %%xmm6, (%1))
278 "paddw %4, %%xmm4 \n"
279 "movdqa %%xmm4, 0x00(%1) \n"
280 "movdqa %%xmm2, 0x40(%1) \n"
281 H264_IDCT8_1D_SSE2(%%xmm4, %%xmm0, %%xmm6, %%xmm3, %%xmm2, %%xmm5, %%xmm7, %%xmm1)
282 "movdqa %%xmm6, 0x60(%1) \n"
283 "movdqa %%xmm7, 0x70(%1) \n"
284 "pxor %%xmm7, %%xmm7 \n"
285 STORE_DIFF_8P(%%xmm2, (%0), %%xmm6, %%xmm7)
286 STORE_DIFF_8P(%%xmm0, (%0,%2), %%xmm6, %%xmm7)
287 STORE_DIFF_8P(%%xmm1, (%0,%2,2), %%xmm6, %%xmm7)
288 STORE_DIFF_8P(%%xmm3, (%0,%3), %%xmm6, %%xmm7)
289 "lea (%0,%2,4), %0 \n"
290 STORE_DIFF_8P(%%xmm5, (%0), %%xmm6, %%xmm7)
291 STORE_DIFF_8P(%%xmm4, (%0,%2), %%xmm6, %%xmm7)
292 "movdqa 0x60(%1), %%xmm0 \n"
293 "movdqa 0x70(%1), %%xmm1 \n"
294 STORE_DIFF_8P(%%xmm0, (%0,%2,2), %%xmm6, %%xmm7)
295 STORE_DIFF_8P(%%xmm1, (%0,%3), %%xmm6, %%xmm7)
296 :"+r"(dst)
297 :"r"(block), "r"((long)stride), "r"(3L*stride), "m"(*ff_pw_32)
298 );
212 } 299 }
213 300
214 static void ff_h264_idct_dc_add_mmx2(uint8_t *dst, int16_t *block, int stride) 301 static void ff_h264_idct_dc_add_mmx2(uint8_t *dst, int16_t *block, int stride)
215 { 302 {
216 int dc = (block[0] + 32) >> 6; 303 int dc = (block[0] + 32) >> 6;
837 "add $24, %0 \n\t"\ 924 "add $24, %0 \n\t"\
838 "add %3, %1 \n\t"\ 925 "add %3, %1 \n\t"\
839 "decl %2 \n\t"\ 926 "decl %2 \n\t"\
840 " jnz 1b \n\t"\ 927 " jnz 1b \n\t"\
841 : "+a"(tmp), "+c"(dst), "+m"(h)\ 928 : "+a"(tmp), "+c"(dst), "+m"(h)\
842 : "S"((long)dstStride), "m"(ff_pw_32)\ 929 : "S"((long)dstStride), "m"(*ff_pw_32)\
843 : "memory"\ 930 : "memory"\
844 );\ 931 );\
845 }\ 932 }\
846 \ 933 \
847 static av_noinline void OPNAME ## h264_qpel8_h_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\ 934 static av_noinline void OPNAME ## h264_qpel8_h_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
1111 "add $48, %0 \n\t"\ 1198 "add $48, %0 \n\t"\
1112 "add %3, %1 \n\t"\ 1199 "add %3, %1 \n\t"\
1113 "decl %2 \n\t"\ 1200 "decl %2 \n\t"\
1114 " jnz 1b \n\t"\ 1201 " jnz 1b \n\t"\
1115 : "+a"(tmp), "+c"(dst), "+m"(h)\ 1202 : "+a"(tmp), "+c"(dst), "+m"(h)\
1116 : "S"((long)dstStride), "m"(ff_pw_32)\ 1203 : "S"((long)dstStride), "m"(*ff_pw_32)\
1117 : "memory"\ 1204 : "memory"\
1118 );\ 1205 );\
1119 tmp += 8 - size*24;\ 1206 tmp += 8 - size*24;\
1120 dst += 8 - size*dstStride;\ 1207 dst += 8 - size*dstStride;\
1121 }while(w--);\ 1208 }while(w--);\