00001
00002
00003
00004
00005
00006
00007
00008
00009
00010
00011
00012
00013
00014
00015
00016
00017
00018
00019
00020
00021
00022
00023
00024 #include "dsputil.h"
00025 #include "x86_cpu.h"
00026
00027 DECLARE_ASM_CONST(8, uint64_t, round_tab[3])={
00028 0x0000000000000000ULL,
00029 0x0001000100010001ULL,
00030 0x0002000200020002ULL,
00031 };
00032
00033 DECLARE_ASM_CONST(8, uint64_t, bone)= 0x0101010101010101LL;
00034
00035 static inline void sad8_1_mmx(uint8_t *blk1, uint8_t *blk2, int stride, int h)
00036 {
00037 long len= -(stride*h);
00038 asm volatile(
00039 ASMALIGN(4)
00040 "1: \n\t"
00041 "movq (%1, %%"REG_a"), %%mm0 \n\t"
00042 "movq (%2, %%"REG_a"), %%mm2 \n\t"
00043 "movq (%2, %%"REG_a"), %%mm4 \n\t"
00044 "add %3, %%"REG_a" \n\t"
00045 "psubusb %%mm0, %%mm2 \n\t"
00046 "psubusb %%mm4, %%mm0 \n\t"
00047 "movq (%1, %%"REG_a"), %%mm1 \n\t"
00048 "movq (%2, %%"REG_a"), %%mm3 \n\t"
00049 "movq (%2, %%"REG_a"), %%mm5 \n\t"
00050 "psubusb %%mm1, %%mm3 \n\t"
00051 "psubusb %%mm5, %%mm1 \n\t"
00052 "por %%mm2, %%mm0 \n\t"
00053 "por %%mm1, %%mm3 \n\t"
00054 "movq %%mm0, %%mm1 \n\t"
00055 "movq %%mm3, %%mm2 \n\t"
00056 "punpcklbw %%mm7, %%mm0 \n\t"
00057 "punpckhbw %%mm7, %%mm1 \n\t"
00058 "punpcklbw %%mm7, %%mm3 \n\t"
00059 "punpckhbw %%mm7, %%mm2 \n\t"
00060 "paddw %%mm1, %%mm0 \n\t"
00061 "paddw %%mm3, %%mm2 \n\t"
00062 "paddw %%mm2, %%mm0 \n\t"
00063 "paddw %%mm0, %%mm6 \n\t"
00064 "add %3, %%"REG_a" \n\t"
00065 " js 1b \n\t"
00066 : "+a" (len)
00067 : "r" (blk1 - len), "r" (blk2 - len), "r" ((long)stride)
00068 );
00069 }
00070
00071 static inline void sad8_1_mmx2(uint8_t *blk1, uint8_t *blk2, int stride, int h)
00072 {
00073 asm volatile(
00074 ASMALIGN(4)
00075 "1: \n\t"
00076 "movq (%1), %%mm0 \n\t"
00077 "movq (%1, %3), %%mm1 \n\t"
00078 "psadbw (%2), %%mm0 \n\t"
00079 "psadbw (%2, %3), %%mm1 \n\t"
00080 "paddw %%mm0, %%mm6 \n\t"
00081 "paddw %%mm1, %%mm6 \n\t"
00082 "lea (%1,%3,2), %1 \n\t"
00083 "lea (%2,%3,2), %2 \n\t"
00084 "sub $2, %0 \n\t"
00085 " jg 1b \n\t"
00086 : "+r" (h), "+r" (blk1), "+r" (blk2)
00087 : "r" ((long)stride)
00088 );
00089 }
00090
00091 static int sad16_sse2(void *v, uint8_t *blk2, uint8_t *blk1, int stride, int h)
00092 {
00093 int ret;
00094 asm volatile(
00095 "pxor %%xmm6, %%xmm6 \n\t"
00096 ASMALIGN(4)
00097 "1: \n\t"
00098 "movdqu (%1), %%xmm0 \n\t"
00099 "movdqu (%1, %3), %%xmm1 \n\t"
00100 "psadbw (%2), %%xmm0 \n\t"
00101 "psadbw (%2, %3), %%xmm1 \n\t"
00102 "paddw %%xmm0, %%xmm6 \n\t"
00103 "paddw %%xmm1, %%xmm6 \n\t"
00104 "lea (%1,%3,2), %1 \n\t"
00105 "lea (%2,%3,2), %2 \n\t"
00106 "sub $2, %0 \n\t"
00107 " jg 1b \n\t"
00108 : "+r" (h), "+r" (blk1), "+r" (blk2)
00109 : "r" ((long)stride)
00110 );
00111 asm volatile(
00112 "movhlps %%xmm6, %%xmm0 \n\t"
00113 "paddw %%xmm0, %%xmm6 \n\t"
00114 "movd %%xmm6, %0 \n\t"
00115 : "=r"(ret)
00116 );
00117 return ret;
00118 }
00119
00120 static inline void sad8_x2a_mmx2(uint8_t *blk1, uint8_t *blk2, int stride, int h)
00121 {
00122 asm volatile(
00123 ASMALIGN(4)
00124 "1: \n\t"
00125 "movq (%1), %%mm0 \n\t"
00126 "movq (%1, %3), %%mm1 \n\t"
00127 "pavgb 1(%1), %%mm0 \n\t"
00128 "pavgb 1(%1, %3), %%mm1 \n\t"
00129 "psadbw (%2), %%mm0 \n\t"
00130 "psadbw (%2, %3), %%mm1 \n\t"
00131 "paddw %%mm0, %%mm6 \n\t"
00132 "paddw %%mm1, %%mm6 \n\t"
00133 "lea (%1,%3,2), %1 \n\t"
00134 "lea (%2,%3,2), %2 \n\t"
00135 "sub $2, %0 \n\t"
00136 " jg 1b \n\t"
00137 : "+r" (h), "+r" (blk1), "+r" (blk2)
00138 : "r" ((long)stride)
00139 );
00140 }
00141
00142 static inline void sad8_y2a_mmx2(uint8_t *blk1, uint8_t *blk2, int stride, int h)
00143 {
00144 asm volatile(
00145 "movq (%1), %%mm0 \n\t"
00146 "add %3, %1 \n\t"
00147 ASMALIGN(4)
00148 "1: \n\t"
00149 "movq (%1), %%mm1 \n\t"
00150 "movq (%1, %3), %%mm2 \n\t"
00151 "pavgb %%mm1, %%mm0 \n\t"
00152 "pavgb %%mm2, %%mm1 \n\t"
00153 "psadbw (%2), %%mm0 \n\t"
00154 "psadbw (%2, %3), %%mm1 \n\t"
00155 "paddw %%mm0, %%mm6 \n\t"
00156 "paddw %%mm1, %%mm6 \n\t"
00157 "movq %%mm2, %%mm0 \n\t"
00158 "lea (%1,%3,2), %1 \n\t"
00159 "lea (%2,%3,2), %2 \n\t"
00160 "sub $2, %0 \n\t"
00161 " jg 1b \n\t"
00162 : "+r" (h), "+r" (blk1), "+r" (blk2)
00163 : "r" ((long)stride)
00164 );
00165 }
00166
00167 static inline void sad8_4_mmx2(uint8_t *blk1, uint8_t *blk2, int stride, int h)
00168 {
00169 asm volatile(
00170 "movq "MANGLE(bone)", %%mm5 \n\t"
00171 "movq (%1), %%mm0 \n\t"
00172 "pavgb 1(%1), %%mm0 \n\t"
00173 "add %3, %1 \n\t"
00174 ASMALIGN(4)
00175 "1: \n\t"
00176 "movq (%1), %%mm1 \n\t"
00177 "movq (%1,%3), %%mm2 \n\t"
00178 "pavgb 1(%1), %%mm1 \n\t"
00179 "pavgb 1(%1,%3), %%mm2 \n\t"
00180 "psubusb %%mm5, %%mm1 \n\t"
00181 "pavgb %%mm1, %%mm0 \n\t"
00182 "pavgb %%mm2, %%mm1 \n\t"
00183 "psadbw (%2), %%mm0 \n\t"
00184 "psadbw (%2,%3), %%mm1 \n\t"
00185 "paddw %%mm0, %%mm6 \n\t"
00186 "paddw %%mm1, %%mm6 \n\t"
00187 "movq %%mm2, %%mm0 \n\t"
00188 "lea (%1,%3,2), %1 \n\t"
00189 "lea (%2,%3,2), %2 \n\t"
00190 "sub $2, %0 \n\t"
00191 " jg 1b \n\t"
00192 : "+r" (h), "+r" (blk1), "+r" (blk2)
00193 : "r" ((long)stride)
00194 );
00195 }
00196
00197 static inline void sad8_2_mmx(uint8_t *blk1a, uint8_t *blk1b, uint8_t *blk2, int stride, int h)
00198 {
00199 long len= -(stride*h);
00200 asm volatile(
00201 ASMALIGN(4)
00202 "1: \n\t"
00203 "movq (%1, %%"REG_a"), %%mm0 \n\t"
00204 "movq (%2, %%"REG_a"), %%mm1 \n\t"
00205 "movq (%1, %%"REG_a"), %%mm2 \n\t"
00206 "movq (%2, %%"REG_a"), %%mm3 \n\t"
00207 "punpcklbw %%mm7, %%mm0 \n\t"
00208 "punpcklbw %%mm7, %%mm1 \n\t"
00209 "punpckhbw %%mm7, %%mm2 \n\t"
00210 "punpckhbw %%mm7, %%mm3 \n\t"
00211 "paddw %%mm0, %%mm1 \n\t"
00212 "paddw %%mm2, %%mm3 \n\t"
00213 "movq (%3, %%"REG_a"), %%mm4 \n\t"
00214 "movq (%3, %%"REG_a"), %%mm2 \n\t"
00215 "paddw %%mm5, %%mm1 \n\t"
00216 "paddw %%mm5, %%mm3 \n\t"
00217 "psrlw $1, %%mm1 \n\t"
00218 "psrlw $1, %%mm3 \n\t"
00219 "packuswb %%mm3, %%mm1 \n\t"
00220 "psubusb %%mm1, %%mm4 \n\t"
00221 "psubusb %%mm2, %%mm1 \n\t"
00222 "por %%mm4, %%mm1 \n\t"
00223 "movq %%mm1, %%mm0 \n\t"
00224 "punpcklbw %%mm7, %%mm0 \n\t"
00225 "punpckhbw %%mm7, %%mm1 \n\t"
00226 "paddw %%mm1, %%mm0 \n\t"
00227 "paddw %%mm0, %%mm6 \n\t"
00228 "add %4, %%"REG_a" \n\t"
00229 " js 1b \n\t"
00230 : "+a" (len)
00231 : "r" (blk1a - len), "r" (blk1b -len), "r" (blk2 - len), "r" ((long)stride)
00232 );
00233 }
00234
00235 static inline void sad8_4_mmx(uint8_t *blk1, uint8_t *blk2, int stride, int h)
00236 {
00237 long len= -(stride*h);
00238 asm volatile(
00239 "movq (%1, %%"REG_a"), %%mm0 \n\t"
00240 "movq 1(%1, %%"REG_a"), %%mm2 \n\t"
00241 "movq %%mm0, %%mm1 \n\t"
00242 "movq %%mm2, %%mm3 \n\t"
00243 "punpcklbw %%mm7, %%mm0 \n\t"
00244 "punpckhbw %%mm7, %%mm1 \n\t"
00245 "punpcklbw %%mm7, %%mm2 \n\t"
00246 "punpckhbw %%mm7, %%mm3 \n\t"
00247 "paddw %%mm2, %%mm0 \n\t"
00248 "paddw %%mm3, %%mm1 \n\t"
00249 ASMALIGN(4)
00250 "1: \n\t"
00251 "movq (%2, %%"REG_a"), %%mm2 \n\t"
00252 "movq 1(%2, %%"REG_a"), %%mm4 \n\t"
00253 "movq %%mm2, %%mm3 \n\t"
00254 "movq %%mm4, %%mm5 \n\t"
00255 "punpcklbw %%mm7, %%mm2 \n\t"
00256 "punpckhbw %%mm7, %%mm3 \n\t"
00257 "punpcklbw %%mm7, %%mm4 \n\t"
00258 "punpckhbw %%mm7, %%mm5 \n\t"
00259 "paddw %%mm4, %%mm2 \n\t"
00260 "paddw %%mm5, %%mm3 \n\t"
00261 "movq 16+"MANGLE(round_tab)", %%mm5 \n\t"
00262 "paddw %%mm2, %%mm0 \n\t"
00263 "paddw %%mm3, %%mm1 \n\t"
00264 "paddw %%mm5, %%mm0 \n\t"
00265 "paddw %%mm5, %%mm1 \n\t"
00266 "movq (%3, %%"REG_a"), %%mm4 \n\t"
00267 "movq (%3, %%"REG_a"), %%mm5 \n\t"
00268 "psrlw $2, %%mm0 \n\t"
00269 "psrlw $2, %%mm1 \n\t"
00270 "packuswb %%mm1, %%mm0 \n\t"
00271 "psubusb %%mm0, %%mm4 \n\t"
00272 "psubusb %%mm5, %%mm0 \n\t"
00273 "por %%mm4, %%mm0 \n\t"
00274 "movq %%mm0, %%mm4 \n\t"
00275 "punpcklbw %%mm7, %%mm0 \n\t"
00276 "punpckhbw %%mm7, %%mm4 \n\t"
00277 "paddw %%mm0, %%mm6 \n\t"
00278 "paddw %%mm4, %%mm6 \n\t"
00279 "movq %%mm2, %%mm0 \n\t"
00280 "movq %%mm3, %%mm1 \n\t"
00281 "add %4, %%"REG_a" \n\t"
00282 " js 1b \n\t"
00283 : "+a" (len)
00284 : "r" (blk1 - len), "r" (blk1 -len + stride), "r" (blk2 - len), "r" ((long)stride)
00285 );
00286 }
00287
00288 static inline int sum_mmx(void)
00289 {
00290 int ret;
00291 asm volatile(
00292 "movq %%mm6, %%mm0 \n\t"
00293 "psrlq $32, %%mm6 \n\t"
00294 "paddw %%mm0, %%mm6 \n\t"
00295 "movq %%mm6, %%mm0 \n\t"
00296 "psrlq $16, %%mm6 \n\t"
00297 "paddw %%mm0, %%mm6 \n\t"
00298 "movd %%mm6, %0 \n\t"
00299 : "=r" (ret)
00300 );
00301 return ret&0xFFFF;
00302 }
00303
00304 static inline int sum_mmx2(void)
00305 {
00306 int ret;
00307 asm volatile(
00308 "movd %%mm6, %0 \n\t"
00309 : "=r" (ret)
00310 );
00311 return ret;
00312 }
00313
00314 static inline void sad8_x2a_mmx(uint8_t *blk1, uint8_t *blk2, int stride, int h)
00315 {
00316 sad8_2_mmx(blk1, blk1+1, blk2, stride, h);
00317 }
00318 static inline void sad8_y2a_mmx(uint8_t *blk1, uint8_t *blk2, int stride, int h)
00319 {
00320 sad8_2_mmx(blk1, blk1+stride, blk2, stride, h);
00321 }
00322
00323
00324 #define PIX_SAD(suf)\
00325 static int sad8_ ## suf(void *v, uint8_t *blk2, uint8_t *blk1, int stride, int h)\
00326 {\
00327 assert(h==8);\
00328 asm volatile("pxor %%mm7, %%mm7 \n\t"\
00329 "pxor %%mm6, %%mm6 \n\t":);\
00330 \
00331 sad8_1_ ## suf(blk1, blk2, stride, 8);\
00332 \
00333 return sum_ ## suf();\
00334 }\
00335 static int sad8_x2_ ## suf(void *v, uint8_t *blk2, uint8_t *blk1, int stride, int h)\
00336 {\
00337 assert(h==8);\
00338 asm volatile("pxor %%mm7, %%mm7 \n\t"\
00339 "pxor %%mm6, %%mm6 \n\t"\
00340 "movq %0, %%mm5 \n\t"\
00341 :: "m"(round_tab[1]) \
00342 );\
00343 \
00344 sad8_x2a_ ## suf(blk1, blk2, stride, 8);\
00345 \
00346 return sum_ ## suf();\
00347 }\
00348 \
00349 static int sad8_y2_ ## suf(void *v, uint8_t *blk2, uint8_t *blk1, int stride, int h)\
00350 {\
00351 assert(h==8);\
00352 asm volatile("pxor %%mm7, %%mm7 \n\t"\
00353 "pxor %%mm6, %%mm6 \n\t"\
00354 "movq %0, %%mm5 \n\t"\
00355 :: "m"(round_tab[1]) \
00356 );\
00357 \
00358 sad8_y2a_ ## suf(blk1, blk2, stride, 8);\
00359 \
00360 return sum_ ## suf();\
00361 }\
00362 \
00363 static int sad8_xy2_ ## suf(void *v, uint8_t *blk2, uint8_t *blk1, int stride, int h)\
00364 {\
00365 assert(h==8);\
00366 asm volatile("pxor %%mm7, %%mm7 \n\t"\
00367 "pxor %%mm6, %%mm6 \n\t"\
00368 ::);\
00369 \
00370 sad8_4_ ## suf(blk1, blk2, stride, 8);\
00371 \
00372 return sum_ ## suf();\
00373 }\
00374 \
00375 static int sad16_ ## suf(void *v, uint8_t *blk2, uint8_t *blk1, int stride, int h)\
00376 {\
00377 asm volatile("pxor %%mm7, %%mm7 \n\t"\
00378 "pxor %%mm6, %%mm6 \n\t":);\
00379 \
00380 sad8_1_ ## suf(blk1 , blk2 , stride, h);\
00381 sad8_1_ ## suf(blk1+8, blk2+8, stride, h);\
00382 \
00383 return sum_ ## suf();\
00384 }\
00385 static int sad16_x2_ ## suf(void *v, uint8_t *blk2, uint8_t *blk1, int stride, int h)\
00386 {\
00387 asm volatile("pxor %%mm7, %%mm7 \n\t"\
00388 "pxor %%mm6, %%mm6 \n\t"\
00389 "movq %0, %%mm5 \n\t"\
00390 :: "m"(round_tab[1]) \
00391 );\
00392 \
00393 sad8_x2a_ ## suf(blk1 , blk2 , stride, h);\
00394 sad8_x2a_ ## suf(blk1+8, blk2+8, stride, h);\
00395 \
00396 return sum_ ## suf();\
00397 }\
00398 static int sad16_y2_ ## suf(void *v, uint8_t *blk2, uint8_t *blk1, int stride, int h)\
00399 {\
00400 asm volatile("pxor %%mm7, %%mm7 \n\t"\
00401 "pxor %%mm6, %%mm6 \n\t"\
00402 "movq %0, %%mm5 \n\t"\
00403 :: "m"(round_tab[1]) \
00404 );\
00405 \
00406 sad8_y2a_ ## suf(blk1 , blk2 , stride, h);\
00407 sad8_y2a_ ## suf(blk1+8, blk2+8, stride, h);\
00408 \
00409 return sum_ ## suf();\
00410 }\
00411 static int sad16_xy2_ ## suf(void *v, uint8_t *blk2, uint8_t *blk1, int stride, int h)\
00412 {\
00413 asm volatile("pxor %%mm7, %%mm7 \n\t"\
00414 "pxor %%mm6, %%mm6 \n\t"\
00415 ::);\
00416 \
00417 sad8_4_ ## suf(blk1 , blk2 , stride, h);\
00418 sad8_4_ ## suf(blk1+8, blk2+8, stride, h);\
00419 \
00420 return sum_ ## suf();\
00421 }\
00422
00423 PIX_SAD(mmx)
00424 PIX_SAD(mmx2)
00425
00426 void dsputil_init_pix_mmx(DSPContext* c, AVCodecContext *avctx)
00427 {
00428 if (mm_flags & MM_MMX) {
00429 c->pix_abs[0][0] = sad16_mmx;
00430 c->pix_abs[0][1] = sad16_x2_mmx;
00431 c->pix_abs[0][2] = sad16_y2_mmx;
00432 c->pix_abs[0][3] = sad16_xy2_mmx;
00433 c->pix_abs[1][0] = sad8_mmx;
00434 c->pix_abs[1][1] = sad8_x2_mmx;
00435 c->pix_abs[1][2] = sad8_y2_mmx;
00436 c->pix_abs[1][3] = sad8_xy2_mmx;
00437
00438 c->sad[0]= sad16_mmx;
00439 c->sad[1]= sad8_mmx;
00440 }
00441 if (mm_flags & MM_MMXEXT) {
00442 c->pix_abs[0][0] = sad16_mmx2;
00443 c->pix_abs[1][0] = sad8_mmx2;
00444
00445 c->sad[0]= sad16_mmx2;
00446 c->sad[1]= sad8_mmx2;
00447
00448 if(!(avctx->flags & CODEC_FLAG_BITEXACT)){
00449 c->pix_abs[0][1] = sad16_x2_mmx2;
00450 c->pix_abs[0][2] = sad16_y2_mmx2;
00451 c->pix_abs[0][3] = sad16_xy2_mmx2;
00452 c->pix_abs[1][1] = sad8_x2_mmx2;
00453 c->pix_abs[1][2] = sad8_y2_mmx2;
00454 c->pix_abs[1][3] = sad8_xy2_mmx2;
00455 }
00456 }
00457 if ((mm_flags & MM_SSE2) && !(mm_flags & MM_3DNOW)) {
00458 c->sad[0]= sad16_sse2;
00459 }
00460 }