FFmpeg: libavcodec/x86/snowdsp

00001 /*
00002  * MMX and SSE2 optimized snow DSP utils
00003  * Copyright (c) 2005-2006 Robert Edele <yartrebo@earthlink.net>
00004  *
00005  * This file is part of FFmpeg.
00006  *
00007  * FFmpeg is free software; you can redistribute it and/or
00008  * modify it under the terms of the GNU Lesser General Public
00009  * License as published by the Free Software Foundation; either
00010  * version 2.1 of the License, or (at your option) any later version.
00011  *
00012  * FFmpeg is distributed in the hope that it will be useful,
00013  * but WITHOUT ANY WARRANTY; without even the implied warranty of
00014  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
00015  * Lesser General Public License for more details.
00016  *
00017  * You should have received a copy of the GNU Lesser General Public
00018  * License along with FFmpeg; if not, write to the Free Software
00019  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
00020  */
00021 
00022 #include "libavutil/x86_cpu.h"
00023 #include "libavcodec/avcodec.h"
00024 #include "libavcodec/snow.h"
00025 
00026 void ff_snow_horizontal_compose97i_sse2(IDWTELEM *b, int width){
00027     const int w2= (width+1)>>1;
00028     DECLARE_ALIGNED_16(IDWTELEM, temp[width>>1]);
00029     const int w_l= (width>>1);
00030     const int w_r= w2 - 1;
00031     int i;
00032 
00033     { // Lift 0
00034         IDWTELEM * const ref = b + w2 - 1;
00035         IDWTELEM b_0 = b[0]; //By allowing the first entry in b[0] to be calculated twice
00036         // (the first time erroneously), we allow the SSE2 code to run an extra pass.
00037         // The savings in code and time are well worth having to store this value and
00038         // calculate b[0] correctly afterwards.
00039 
00040         i = 0;
00041         __asm__ volatile(
00042             "pcmpeqd   %%xmm7, %%xmm7         \n\t"
00043             "pcmpeqd   %%xmm3, %%xmm3         \n\t"
00044             "psllw         $1, %%xmm3         \n\t"
00045             "paddw     %%xmm7, %%xmm3         \n\t"
00046             "psllw        $13, %%xmm3         \n\t"
00047         ::);
00048         for(; i<w_l-15; i+=16){
00049             __asm__ volatile(
00050                 "movdqu   (%1), %%xmm1        \n\t"
00051                 "movdqu 16(%1), %%xmm5        \n\t"
00052                 "movdqu  2(%1), %%xmm2        \n\t"
00053                 "movdqu 18(%1), %%xmm6        \n\t"
00054                 "paddw  %%xmm1, %%xmm2        \n\t"
00055                 "paddw  %%xmm5, %%xmm6        \n\t"
00056                 "paddw  %%xmm7, %%xmm2        \n\t"
00057                 "paddw  %%xmm7, %%xmm6        \n\t"
00058                 "pmulhw %%xmm3, %%xmm2        \n\t"
00059                 "pmulhw %%xmm3, %%xmm6        \n\t"
00060                 "paddw    (%0), %%xmm2        \n\t"
00061                 "paddw  16(%0), %%xmm6        \n\t"
00062                 "movdqa %%xmm2, (%0)          \n\t"
00063                 "movdqa %%xmm6, 16(%0)        \n\t"
00064                 :: "r"(&b[i]), "r"(&ref[i])
00065                 : "memory"
00066             );
00067         }
00068         snow_horizontal_compose_lift_lead_out(i, b, b, ref, width, w_l, 0, W_DM, W_DO, W_DS);
00069         b[0] = b_0 - ((W_DM * 2 * ref[1]+W_DO)>>W_DS);
00070     }
00071 
00072     { // Lift 1
00073         IDWTELEM * const dst = b+w2;
00074 
00075         i = 0;
00076         for(; (((x86_reg)&dst[i]) & 0x1F) && i<w_r; i++){
00077             dst[i] = dst[i] - (b[i] + b[i + 1]);
00078         }
00079         for(; i<w_r-15; i+=16){
00080             __asm__ volatile(
00081                 "movdqu   (%1), %%xmm1        \n\t"
00082                 "movdqu 16(%1), %%xmm5        \n\t"
00083                 "movdqu  2(%1), %%xmm2        \n\t"
00084                 "movdqu 18(%1), %%xmm6        \n\t"
00085                 "paddw  %%xmm1, %%xmm2        \n\t"
00086                 "paddw  %%xmm5, %%xmm6        \n\t"
00087                 "movdqa   (%0), %%xmm0        \n\t"
00088                 "movdqa 16(%0), %%xmm4        \n\t"
00089                 "psubw  %%xmm2, %%xmm0        \n\t"
00090                 "psubw  %%xmm6, %%xmm4        \n\t"
00091                 "movdqa %%xmm0, (%0)          \n\t"
00092                 "movdqa %%xmm4, 16(%0)        \n\t"
00093                 :: "r"(&dst[i]), "r"(&b[i])
00094                 : "memory"
00095             );
00096         }
00097         snow_horizontal_compose_lift_lead_out(i, dst, dst, b, width, w_r, 1, W_CM, W_CO, W_CS);
00098     }
00099 
00100     { // Lift 2
00101         IDWTELEM * const ref = b+w2 - 1;
00102         IDWTELEM b_0 = b[0];
00103 
00104         i = 0;
00105         __asm__ volatile(
00106             "psllw         $15, %%xmm7        \n\t"
00107             "pcmpeqw    %%xmm6, %%xmm6        \n\t"
00108             "psrlw         $13, %%xmm6        \n\t"
00109             "paddw      %%xmm7, %%xmm6        \n\t"
00110         ::);
00111         for(; i<w_l-15; i+=16){
00112             __asm__ volatile(
00113                 "movdqu   (%1), %%xmm0        \n\t"
00114                 "movdqu 16(%1), %%xmm4        \n\t"
00115                 "movdqu  2(%1), %%xmm1        \n\t"
00116                 "movdqu 18(%1), %%xmm5        \n\t" //FIXME try aligned reads and shifts
00117                 "paddw  %%xmm6, %%xmm0        \n\t"
00118                 "paddw  %%xmm6, %%xmm4        \n\t"
00119                 "paddw  %%xmm7, %%xmm1        \n\t"
00120                 "paddw  %%xmm7, %%xmm5        \n\t"
00121                 "pavgw  %%xmm1, %%xmm0        \n\t"
00122                 "pavgw  %%xmm5, %%xmm4        \n\t"
00123                 "psubw  %%xmm7, %%xmm0        \n\t"
00124                 "psubw  %%xmm7, %%xmm4        \n\t"
00125                 "psraw      $1, %%xmm0        \n\t"
00126                 "psraw      $1, %%xmm4        \n\t"
00127                 "movdqa   (%0), %%xmm1        \n\t"
00128                 "movdqa 16(%0), %%xmm5        \n\t"
00129                 "paddw  %%xmm1, %%xmm0        \n\t"
00130                 "paddw  %%xmm5, %%xmm4        \n\t"
00131                 "psraw      $2, %%xmm0        \n\t"
00132                 "psraw      $2, %%xmm4        \n\t"
00133                 "paddw  %%xmm1, %%xmm0        \n\t"
00134                 "paddw  %%xmm5, %%xmm4        \n\t"
00135                 "movdqa %%xmm0, (%0)          \n\t"
00136                 "movdqa %%xmm4, 16(%0)        \n\t"
00137                 :: "r"(&b[i]), "r"(&ref[i])
00138                 : "memory"
00139             );
00140         }
00141         snow_horizontal_compose_liftS_lead_out(i, b, b, ref, width, w_l);
00142         b[0] = b_0 + ((2 * ref[1] + W_BO-1 + 4 * b_0) >> W_BS);
00143     }
00144 
00145     { // Lift 3
00146         IDWTELEM * const src = b+w2;
00147 
00148         i = 0;
00149         for(; (((x86_reg)&temp[i]) & 0x1F) && i<w_r; i++){
00150             temp[i] = src[i] - ((-W_AM*(b[i] + b[i+1]))>>W_AS);
00151         }
00152         for(; i<w_r-7; i+=8){
00153             __asm__ volatile(
00154                 "movdqu  2(%1), %%xmm2        \n\t"
00155                 "movdqu 18(%1), %%xmm6        \n\t"
00156                 "paddw    (%1), %%xmm2        \n\t"
00157                 "paddw  16(%1), %%xmm6        \n\t"
00158                 "movdqu   (%0), %%xmm0        \n\t"
00159                 "movdqu 16(%0), %%xmm4        \n\t"
00160                 "paddw  %%xmm2, %%xmm0        \n\t"
00161                 "paddw  %%xmm6, %%xmm4        \n\t"
00162                 "psraw      $1, %%xmm2        \n\t"
00163                 "psraw      $1, %%xmm6        \n\t"
00164                 "paddw  %%xmm0, %%xmm2        \n\t"
00165                 "paddw  %%xmm4, %%xmm6        \n\t"
00166                 "movdqa %%xmm2, (%2)          \n\t"
00167                 "movdqa %%xmm6, 16(%2)        \n\t"
00168                 :: "r"(&src[i]), "r"(&b[i]), "r"(&temp[i])
00169                  : "memory"
00170                );
00171         }
00172         snow_horizontal_compose_lift_lead_out(i, temp, src, b, width, w_r, 1, -W_AM, W_AO+1, W_AS);
00173     }
00174 
00175     {
00176         snow_interleave_line_header(&i, width, b, temp);
00177 
00178         for (; (i & 0x3E) != 0x3E; i-=2){
00179             b[i+1] = temp[i>>1];
00180             b[i] = b[i>>1];
00181         }
00182         for (i-=62; i>=0; i-=64){
00183             __asm__ volatile(
00184                 "movdqa      (%1), %%xmm0       \n\t"
00185                 "movdqa    16(%1), %%xmm2       \n\t"
00186                 "movdqa    32(%1), %%xmm4       \n\t"
00187                 "movdqa    48(%1), %%xmm6       \n\t"
00188                 "movdqa      (%1), %%xmm1       \n\t"
00189                 "movdqa    16(%1), %%xmm3       \n\t"
00190                 "movdqa    32(%1), %%xmm5       \n\t"
00191                 "movdqa    48(%1), %%xmm7       \n\t"
00192                 "punpcklwd   (%2), %%xmm0       \n\t"
00193                 "punpcklwd 16(%2), %%xmm2       \n\t"
00194                 "punpcklwd 32(%2), %%xmm4       \n\t"
00195                 "punpcklwd 48(%2), %%xmm6       \n\t"
00196                 "movdqa    %%xmm0, (%0)         \n\t"
00197                 "movdqa    %%xmm2, 32(%0)       \n\t"
00198                 "movdqa    %%xmm4, 64(%0)       \n\t"
00199                 "movdqa    %%xmm6, 96(%0)       \n\t"
00200                 "punpckhwd   (%2), %%xmm1       \n\t"
00201                 "punpckhwd 16(%2), %%xmm3       \n\t"
00202                 "punpckhwd 32(%2), %%xmm5       \n\t"
00203                 "punpckhwd 48(%2), %%xmm7       \n\t"
00204                 "movdqa    %%xmm1, 16(%0)       \n\t"
00205                 "movdqa    %%xmm3, 48(%0)       \n\t"
00206                 "movdqa    %%xmm5, 80(%0)       \n\t"
00207                 "movdqa    %%xmm7, 112(%0)      \n\t"
00208                 :: "r"(&(b)[i]), "r"(&(b)[i>>1]), "r"(&(temp)[i>>1])
00209                  : "memory"
00210                );
00211         }
00212     }
00213 }
00214 
00215 void ff_snow_horizontal_compose97i_mmx(IDWTELEM *b, int width){
00216     const int w2= (width+1)>>1;
00217     IDWTELEM temp[width >> 1];
00218     const int w_l= (width>>1);
00219     const int w_r= w2 - 1;
00220     int i;
00221 
00222     { // Lift 0
00223         IDWTELEM * const ref = b + w2 - 1;
00224 
00225         i = 1;
00226         b[0] = b[0] - ((W_DM * 2 * ref[1]+W_DO)>>W_DS);
00227         __asm__ volatile(
00228             "pcmpeqw    %%mm7, %%mm7         \n\t"
00229             "pcmpeqw    %%mm3, %%mm3         \n\t"
00230             "psllw         $1, %%mm3         \n\t"
00231             "paddw      %%mm7, %%mm3         \n\t"
00232             "psllw        $13, %%mm3         \n\t"
00233            ::);
00234         for(; i<w_l-7; i+=8){
00235             __asm__ volatile(
00236                 "movq     (%1), %%mm2        \n\t"
00237                 "movq    8(%1), %%mm6        \n\t"
00238                 "paddw   2(%1), %%mm2        \n\t"
00239                 "paddw  10(%1), %%mm6        \n\t"
00240                 "paddw   %%mm7, %%mm2        \n\t"
00241                 "paddw   %%mm7, %%mm6        \n\t"
00242                 "pmulhw  %%mm3, %%mm2        \n\t"
00243                 "pmulhw  %%mm3, %%mm6        \n\t"
00244                 "paddw    (%0), %%mm2        \n\t"
00245                 "paddw   8(%0), %%mm6        \n\t"
00246                 "movq    %%mm2, (%0)         \n\t"
00247                 "movq    %%mm6, 8(%0)        \n\t"
00248                 :: "r"(&b[i]), "r"(&ref[i])
00249                  : "memory"
00250                );
00251         }
00252         snow_horizontal_compose_lift_lead_out(i, b, b, ref, width, w_l, 0, W_DM, W_DO, W_DS);
00253     }
00254 
00255     { // Lift 1
00256         IDWTELEM * const dst = b+w2;
00257 
00258         i = 0;
00259         for(; i<w_r-7; i+=8){
00260             __asm__ volatile(
00261                 "movq     (%1), %%mm2        \n\t"
00262                 "movq    8(%1), %%mm6        \n\t"
00263                 "paddw   2(%1), %%mm2        \n\t"
00264                 "paddw  10(%1), %%mm6        \n\t"
00265                 "movq     (%0), %%mm0        \n\t"
00266                 "movq    8(%0), %%mm4        \n\t"
00267                 "psubw   %%mm2, %%mm0        \n\t"
00268                 "psubw   %%mm6, %%mm4        \n\t"
00269                 "movq    %%mm0, (%0)         \n\t"
00270                 "movq    %%mm4, 8(%0)        \n\t"
00271                 :: "r"(&dst[i]), "r"(&b[i])
00272                  : "memory"
00273                );
00274         }
00275         snow_horizontal_compose_lift_lead_out(i, dst, dst, b, width, w_r, 1, W_CM, W_CO, W_CS);
00276     }
00277 
00278     { // Lift 2
00279         IDWTELEM * const ref = b+w2 - 1;
00280 
00281         i = 1;
00282         b[0] = b[0] + (((2 * ref[1] + W_BO) + 4 * b[0]) >> W_BS);
00283         __asm__ volatile(
00284             "psllw         $15, %%mm7        \n\t"
00285             "pcmpeqw     %%mm6, %%mm6        \n\t"
00286             "psrlw         $13, %%mm6        \n\t"
00287             "paddw       %%mm7, %%mm6        \n\t"
00288            ::);
00289         for(; i<w_l-7; i+=8){
00290             __asm__ volatile(
00291                 "movq     (%1), %%mm0        \n\t"
00292                 "movq    8(%1), %%mm4        \n\t"
00293                 "movq    2(%1), %%mm1        \n\t"
00294                 "movq   10(%1), %%mm5        \n\t"
00295                 "paddw   %%mm6, %%mm0        \n\t"
00296                 "paddw   %%mm6, %%mm4        \n\t"
00297                 "paddw   %%mm7, %%mm1        \n\t"
00298                 "paddw   %%mm7, %%mm5        \n\t"
00299                 "pavgw   %%mm1, %%mm0        \n\t"
00300                 "pavgw   %%mm5, %%mm4        \n\t"
00301                 "psubw   %%mm7, %%mm0        \n\t"
00302                 "psubw   %%mm7, %%mm4        \n\t"
00303                 "psraw      $1, %%mm0        \n\t"
00304                 "psraw      $1, %%mm4        \n\t"
00305                 "movq     (%0), %%mm1        \n\t"
00306                 "movq    8(%0), %%mm5        \n\t"
00307                 "paddw   %%mm1, %%mm0        \n\t"
00308                 "paddw   %%mm5, %%mm4        \n\t"
00309                 "psraw      $2, %%mm0        \n\t"
00310                 "psraw      $2, %%mm4        \n\t"
00311                 "paddw   %%mm1, %%mm0        \n\t"
00312                 "paddw   %%mm5, %%mm4        \n\t"
00313                 "movq    %%mm0, (%0)         \n\t"
00314                 "movq    %%mm4, 8(%0)        \n\t"
00315                 :: "r"(&b[i]), "r"(&ref[i])
00316                  : "memory"
00317                );
00318         }
00319         snow_horizontal_compose_liftS_lead_out(i, b, b, ref, width, w_l);
00320     }
00321 
00322     { // Lift 3
00323         IDWTELEM * const src = b+w2;
00324         i = 0;
00325 
00326         for(; i<w_r-7; i+=8){
00327             __asm__ volatile(
00328                 "movq    2(%1), %%mm2        \n\t"
00329                 "movq   10(%1), %%mm6        \n\t"
00330                 "paddw    (%1), %%mm2        \n\t"
00331                 "paddw   8(%1), %%mm6        \n\t"
00332                 "movq     (%0), %%mm0        \n\t"
00333                 "movq    8(%0), %%mm4        \n\t"
00334                 "paddw   %%mm2, %%mm0        \n\t"
00335                 "paddw   %%mm6, %%mm4        \n\t"
00336                 "psraw      $1, %%mm2        \n\t"
00337                 "psraw      $1, %%mm6        \n\t"
00338                 "paddw   %%mm0, %%mm2        \n\t"
00339                 "paddw   %%mm4, %%mm6        \n\t"
00340                 "movq    %%mm2, (%2)         \n\t"
00341                 "movq    %%mm6, 8(%2)        \n\t"
00342                 :: "r"(&src[i]), "r"(&b[i]), "r"(&temp[i])
00343                  : "memory"
00344                );
00345         }
00346         snow_horizontal_compose_lift_lead_out(i, temp, src, b, width, w_r, 1, -W_AM, W_AO+1, W_AS);
00347     }
00348 
00349     {
00350         snow_interleave_line_header(&i, width, b, temp);
00351 
00352         for (; (i & 0x1E) != 0x1E; i-=2){
00353             b[i+1] = temp[i>>1];
00354             b[i] = b[i>>1];
00355         }
00356         for (i-=30; i>=0; i-=32){
00357             __asm__ volatile(
00358                 "movq        (%1), %%mm0       \n\t"
00359                 "movq       8(%1), %%mm2       \n\t"
00360                 "movq      16(%1), %%mm4       \n\t"
00361                 "movq      24(%1), %%mm6       \n\t"
00362                 "movq        (%1), %%mm1       \n\t"
00363                 "movq       8(%1), %%mm3       \n\t"
00364                 "movq      16(%1), %%mm5       \n\t"
00365                 "movq      24(%1), %%mm7       \n\t"
00366                 "punpcklwd   (%2), %%mm0       \n\t"
00367                 "punpcklwd  8(%2), %%mm2       \n\t"
00368                 "punpcklwd 16(%2), %%mm4       \n\t"
00369                 "punpcklwd 24(%2), %%mm6       \n\t"
00370                 "movq       %%mm0, (%0)        \n\t"
00371                 "movq       %%mm2, 16(%0)      \n\t"
00372                 "movq       %%mm4, 32(%0)      \n\t"
00373                 "movq       %%mm6, 48(%0)      \n\t"
00374                 "punpckhwd   (%2), %%mm1       \n\t"
00375                 "punpckhwd  8(%2), %%mm3       \n\t"
00376                 "punpckhwd 16(%2), %%mm5       \n\t"
00377                 "punpckhwd 24(%2), %%mm7       \n\t"
00378                 "movq       %%mm1, 8(%0)       \n\t"
00379                 "movq       %%mm3, 24(%0)      \n\t"
00380                 "movq       %%mm5, 40(%0)      \n\t"
00381                 "movq       %%mm7, 56(%0)      \n\t"
00382                 :: "r"(&b[i]), "r"(&b[i>>1]), "r"(&temp[i>>1])
00383                  : "memory"
00384                );
00385         }
00386     }
00387 }
00388 
00389 #if HAVE_7REGS
00390 #define snow_vertical_compose_sse2_load_add(op,r,t0,t1,t2,t3)\
00391         ""op" ("r",%%"REG_d"), %%"t0"      \n\t"\
00392         ""op" 16("r",%%"REG_d"), %%"t1"    \n\t"\
00393         ""op" 32("r",%%"REG_d"), %%"t2"    \n\t"\
00394         ""op" 48("r",%%"REG_d"), %%"t3"    \n\t"
00395 
00396 #define snow_vertical_compose_sse2_load(r,t0,t1,t2,t3)\
00397         snow_vertical_compose_sse2_load_add("movdqa",r,t0,t1,t2,t3)
00398 
00399 #define snow_vertical_compose_sse2_add(r,t0,t1,t2,t3)\
00400         snow_vertical_compose_sse2_load_add("paddw",r,t0,t1,t2,t3)
00401 
00402 #define snow_vertical_compose_r2r_sub(s0,s1,s2,s3,t0,t1,t2,t3)\
00403         "psubw %%"s0", %%"t0" \n\t"\
00404         "psubw %%"s1", %%"t1" \n\t"\
00405         "psubw %%"s2", %%"t2" \n\t"\
00406         "psubw %%"s3", %%"t3" \n\t"
00407 
00408 #define snow_vertical_compose_sse2_store(w,s0,s1,s2,s3)\
00409         "movdqa %%"s0", ("w",%%"REG_d")      \n\t"\
00410         "movdqa %%"s1", 16("w",%%"REG_d")    \n\t"\
00411         "movdqa %%"s2", 32("w",%%"REG_d")    \n\t"\
00412         "movdqa %%"s3", 48("w",%%"REG_d")    \n\t"
00413 
00414 #define snow_vertical_compose_sra(n,t0,t1,t2,t3)\
00415         "psraw $"n", %%"t0" \n\t"\
00416         "psraw $"n", %%"t1" \n\t"\
00417         "psraw $"n", %%"t2" \n\t"\
00418         "psraw $"n", %%"t3" \n\t"
00419 
00420 #define snow_vertical_compose_r2r_add(s0,s1,s2,s3,t0,t1,t2,t3)\
00421         "paddw %%"s0", %%"t0" \n\t"\
00422         "paddw %%"s1", %%"t1" \n\t"\
00423         "paddw %%"s2", %%"t2" \n\t"\
00424         "paddw %%"s3", %%"t3" \n\t"
00425 
00426 #define snow_vertical_compose_r2r_pmulhw(s0,s1,s2,s3,t0,t1,t2,t3)\
00427         "pmulhw %%"s0", %%"t0" \n\t"\
00428         "pmulhw %%"s1", %%"t1" \n\t"\
00429         "pmulhw %%"s2", %%"t2" \n\t"\
00430         "pmulhw %%"s3", %%"t3" \n\t"
00431 
00432 #define snow_vertical_compose_sse2_move(s0,s1,s2,s3,t0,t1,t2,t3)\
00433         "movdqa %%"s0", %%"t0" \n\t"\
00434         "movdqa %%"s1", %%"t1" \n\t"\
00435         "movdqa %%"s2", %%"t2" \n\t"\
00436         "movdqa %%"s3", %%"t3" \n\t"
00437 
00438 void ff_snow_vertical_compose97i_sse2(IDWTELEM *b0, IDWTELEM *b1, IDWTELEM *b2, IDWTELEM *b3, IDWTELEM *b4, IDWTELEM *b5, int width){
00439     x86_reg i = width;
00440 
00441     while(i & 0x1F)
00442     {
00443         i--;
00444         b4[i] -= (W_DM*(b3[i] + b5[i])+W_DO)>>W_DS;
00445         b3[i] -= (W_CM*(b2[i] + b4[i])+W_CO)>>W_CS;
00446         b2[i] += (W_BM*(b1[i] + b3[i])+4*b2[i]+W_BO)>>W_BS;
00447         b1[i] += (W_AM*(b0[i] + b2[i])+W_AO)>>W_AS;
00448     }
00449     i+=i;
00450 
00451          __asm__ volatile (
00452         "jmp 2f                                      \n\t"
00453         "1:                                          \n\t"
00454         snow_vertical_compose_sse2_load("%4","xmm0","xmm2","xmm4","xmm6")
00455         snow_vertical_compose_sse2_add("%6","xmm0","xmm2","xmm4","xmm6")
00456 
00457 
00458         "pcmpeqw    %%xmm0, %%xmm0                   \n\t"
00459         "pcmpeqw    %%xmm2, %%xmm2                   \n\t"
00460         "paddw      %%xmm2, %%xmm2                   \n\t"
00461         "paddw      %%xmm0, %%xmm2                   \n\t"
00462         "psllw         $13, %%xmm2                   \n\t"
00463         snow_vertical_compose_r2r_add("xmm0","xmm0","xmm0","xmm0","xmm1","xmm3","xmm5","xmm7")
00464         snow_vertical_compose_r2r_pmulhw("xmm2","xmm2","xmm2","xmm2","xmm1","xmm3","xmm5","xmm7")
00465         snow_vertical_compose_sse2_add("%5","xmm1","xmm3","xmm5","xmm7")
00466         snow_vertical_compose_sse2_store("%5","xmm1","xmm3","xmm5","xmm7")
00467         snow_vertical_compose_sse2_load("%4","xmm0","xmm2","xmm4","xmm6")
00468         snow_vertical_compose_sse2_add("%3","xmm1","xmm3","xmm5","xmm7")
00469         snow_vertical_compose_r2r_sub("xmm1","xmm3","xmm5","xmm7","xmm0","xmm2","xmm4","xmm6")
00470         snow_vertical_compose_sse2_store("%4","xmm0","xmm2","xmm4","xmm6")
00471 
00472         "pcmpeqw %%xmm7, %%xmm7                      \n\t"
00473         "pcmpeqw %%xmm5, %%xmm5                      \n\t"
00474         "psllw $15, %%xmm7                           \n\t"
00475         "psrlw $13, %%xmm5                           \n\t"
00476         "paddw %%xmm7, %%xmm5                        \n\t"
00477         snow_vertical_compose_r2r_add("xmm5","xmm5","xmm5","xmm5","xmm0","xmm2","xmm4","xmm6")
00478         "movq   (%2,%%"REG_d"), %%xmm1        \n\t"
00479         "movq  8(%2,%%"REG_d"), %%xmm3        \n\t"
00480         "paddw %%xmm7, %%xmm1                        \n\t"
00481         "paddw %%xmm7, %%xmm3                        \n\t"
00482         "pavgw %%xmm1, %%xmm0                        \n\t"
00483         "pavgw %%xmm3, %%xmm2                        \n\t"
00484         "movq 16(%2,%%"REG_d"), %%xmm1        \n\t"
00485         "movq 24(%2,%%"REG_d"), %%xmm3        \n\t"
00486         "paddw %%xmm7, %%xmm1                        \n\t"
00487         "paddw %%xmm7, %%xmm3                        \n\t"
00488         "pavgw %%xmm1, %%xmm4                        \n\t"
00489         "pavgw %%xmm3, %%xmm6                        \n\t"
00490         snow_vertical_compose_r2r_sub("xmm7","xmm7","xmm7","xmm7","xmm0","xmm2","xmm4","xmm6")
00491         snow_vertical_compose_sra("1","xmm0","xmm2","xmm4","xmm6")
00492         snow_vertical_compose_sse2_add("%3","xmm0","xmm2","xmm4","xmm6")
00493 
00494         snow_vertical_compose_sra("2","xmm0","xmm2","xmm4","xmm6")
00495         snow_vertical_compose_sse2_add("%3","xmm0","xmm2","xmm4","xmm6")
00496         snow_vertical_compose_sse2_store("%3","xmm0","xmm2","xmm4","xmm6")
00497         snow_vertical_compose_sse2_add("%1","xmm0","xmm2","xmm4","xmm6")
00498         snow_vertical_compose_sse2_move("xmm0","xmm2","xmm4","xmm6","xmm1","xmm3","xmm5","xmm7")
00499         snow_vertical_compose_sra("1","xmm0","xmm2","xmm4","xmm6")
00500         snow_vertical_compose_r2r_add("xmm1","xmm3","xmm5","xmm7","xmm0","xmm2","xmm4","xmm6")
00501         snow_vertical_compose_sse2_add("%2","xmm0","xmm2","xmm4","xmm6")
00502         snow_vertical_compose_sse2_store("%2","xmm0","xmm2","xmm4","xmm6")
00503 
00504         "2:                                          \n\t"
00505         "sub $64, %%"REG_d"                          \n\t"
00506         "jge 1b                                      \n\t"
00507         :"+d"(i)
00508         :"r"(b0),"r"(b1),"r"(b2),"r"(b3),"r"(b4),"r"(b5));
00509 }
00510 
00511 #define snow_vertical_compose_mmx_load_add(op,r,t0,t1,t2,t3)\
00512         ""op" ("r",%%"REG_d"), %%"t0"   \n\t"\
00513         ""op" 8("r",%%"REG_d"), %%"t1"  \n\t"\
00514         ""op" 16("r",%%"REG_d"), %%"t2" \n\t"\
00515         ""op" 24("r",%%"REG_d"), %%"t3" \n\t"
00516 
00517 #define snow_vertical_compose_mmx_load(r,t0,t1,t2,t3)\
00518         snow_vertical_compose_mmx_load_add("movq",r,t0,t1,t2,t3)
00519 
00520 #define snow_vertical_compose_mmx_add(r,t0,t1,t2,t3)\
00521         snow_vertical_compose_mmx_load_add("paddw",r,t0,t1,t2,t3)
00522 
00523 #define snow_vertical_compose_mmx_store(w,s0,s1,s2,s3)\
00524         "movq %%"s0", ("w",%%"REG_d")   \n\t"\
00525         "movq %%"s1", 8("w",%%"REG_d")  \n\t"\
00526         "movq %%"s2", 16("w",%%"REG_d") \n\t"\
00527         "movq %%"s3", 24("w",%%"REG_d") \n\t"
00528 
00529 #define snow_vertical_compose_mmx_move(s0,s1,s2,s3,t0,t1,t2,t3)\
00530         "movq %%"s0", %%"t0" \n\t"\
00531         "movq %%"s1", %%"t1" \n\t"\
00532         "movq %%"s2", %%"t2" \n\t"\
00533         "movq %%"s3", %%"t3" \n\t"
00534 
00535 
00536 void ff_snow_vertical_compose97i_mmx(IDWTELEM *b0, IDWTELEM *b1, IDWTELEM *b2, IDWTELEM *b3, IDWTELEM *b4, IDWTELEM *b5, int width){
00537     x86_reg i = width;
00538     while(i & 15)
00539     {
00540         i--;
00541         b4[i] -= (W_DM*(b3[i] + b5[i])+W_DO)>>W_DS;
00542         b3[i] -= (W_CM*(b2[i] + b4[i])+W_CO)>>W_CS;
00543         b2[i] += (W_BM*(b1[i] + b3[i])+4*b2[i]+W_BO)>>W_BS;
00544         b1[i] += (W_AM*(b0[i] + b2[i])+W_AO)>>W_AS;
00545     }
00546     i+=i;
00547     __asm__ volatile(
00548         "jmp 2f                                      \n\t"
00549         "1:                                          \n\t"
00550 
00551         snow_vertical_compose_mmx_load("%4","mm1","mm3","mm5","mm7")
00552         snow_vertical_compose_mmx_add("%6","mm1","mm3","mm5","mm7")
00553         "pcmpeqw    %%mm0, %%mm0                     \n\t"
00554         "pcmpeqw    %%mm2, %%mm2                     \n\t"
00555         "paddw      %%mm2, %%mm2                     \n\t"
00556         "paddw      %%mm0, %%mm2                     \n\t"
00557         "psllw        $13, %%mm2                     \n\t"
00558         snow_vertical_compose_r2r_add("mm0","mm0","mm0","mm0","mm1","mm3","mm5","mm7")
00559         snow_vertical_compose_r2r_pmulhw("mm2","mm2","mm2","mm2","mm1","mm3","mm5","mm7")
00560         snow_vertical_compose_mmx_add("%5","mm1","mm3","mm5","mm7")
00561         snow_vertical_compose_mmx_store("%5","mm1","mm3","mm5","mm7")
00562         snow_vertical_compose_mmx_load("%4","mm0","mm2","mm4","mm6")
00563         snow_vertical_compose_mmx_add("%3","mm1","mm3","mm5","mm7")
00564         snow_vertical_compose_r2r_sub("mm1","mm3","mm5","mm7","mm0","mm2","mm4","mm6")
00565         snow_vertical_compose_mmx_store("%4","mm0","mm2","mm4","mm6")
00566         "pcmpeqw %%mm7, %%mm7                        \n\t"
00567         "pcmpeqw %%mm5, %%mm5                        \n\t"
00568         "psllw $15, %%mm7                            \n\t"
00569         "psrlw $13, %%mm5                            \n\t"
00570         "paddw %%mm7, %%mm5                          \n\t"
00571         snow_vertical_compose_r2r_add("mm5","mm5","mm5","mm5","mm0","mm2","mm4","mm6")
00572         "movq   (%2,%%"REG_d"), %%mm1         \n\t"
00573         "movq  8(%2,%%"REG_d"), %%mm3         \n\t"
00574         "paddw %%mm7, %%mm1                          \n\t"
00575         "paddw %%mm7, %%mm3                          \n\t"
00576         "pavgw %%mm1, %%mm0                          \n\t"
00577         "pavgw %%mm3, %%mm2                          \n\t"
00578         "movq 16(%2,%%"REG_d"), %%mm1         \n\t"
00579         "movq 24(%2,%%"REG_d"), %%mm3         \n\t"
00580         "paddw %%mm7, %%mm1                          \n\t"
00581         "paddw %%mm7, %%mm3                          \n\t"
00582         "pavgw %%mm1, %%mm4                          \n\t"
00583         "pavgw %%mm3, %%mm6                          \n\t"
00584         snow_vertical_compose_r2r_sub("mm7","mm7","mm7","mm7","mm0","mm2","mm4","mm6")
00585         snow_vertical_compose_sra("1","mm0","mm2","mm4","mm6")
00586         snow_vertical_compose_mmx_add("%3","mm0","mm2","mm4","mm6")
00587 
00588         snow_vertical_compose_sra("2","mm0","mm2","mm4","mm6")
00589         snow_vertical_compose_mmx_add("%3","mm0","mm2","mm4","mm6")
00590         snow_vertical_compose_mmx_store("%3","mm0","mm2","mm4","mm6")
00591         snow_vertical_compose_mmx_add("%1","mm0","mm2","mm4","mm6")
00592         snow_vertical_compose_mmx_move("mm0","mm2","mm4","mm6","mm1","mm3","mm5","mm7")
00593         snow_vertical_compose_sra("1","mm0","mm2","mm4","mm6")
00594         snow_vertical_compose_r2r_add("mm1","mm3","mm5","mm7","mm0","mm2","mm4","mm6")
00595         snow_vertical_compose_mmx_add("%2","mm0","mm2","mm4","mm6")
00596         snow_vertical_compose_mmx_store("%2","mm0","mm2","mm4","mm6")
00597 
00598         "2:                                          \n\t"
00599         "sub $32, %%"REG_d"                          \n\t"
00600         "jge 1b                                      \n\t"
00601         :"+d"(i)
00602         :"r"(b0),"r"(b1),"r"(b2),"r"(b3),"r"(b4),"r"(b5));
00603 }
00604 #endif //HAVE_7REGS
00605 
00606 #define snow_inner_add_yblock_sse2_header \
00607     IDWTELEM * * dst_array = sb->line + src_y;\
00608     x86_reg tmp;\
00609     __asm__ volatile(\
00610              "mov  %7, %%"REG_c"             \n\t"\
00611              "mov  %6, %2                    \n\t"\
00612              "mov  %4, %%"REG_S"             \n\t"\
00613              "pxor %%xmm7, %%xmm7            \n\t" /* 0 */\
00614              "pcmpeqd %%xmm3, %%xmm3         \n\t"\
00615              "psllw $15, %%xmm3              \n\t"\
00616              "psrlw $12, %%xmm3              \n\t" /* FRAC_BITS >> 1 */\
00617              "1:                             \n\t"\
00618              "mov %1, %%"REG_D"              \n\t"\
00619              "mov (%%"REG_D"), %%"REG_D"     \n\t"\
00620              "add %3, %%"REG_D"              \n\t"
00621 
00622 #define snow_inner_add_yblock_sse2_start_8(out_reg1, out_reg2, ptr_offset, s_offset)\
00623              "mov "PTR_SIZE"*"ptr_offset"(%%"REG_a"), %%"REG_d"; \n\t"\
00624              "movq (%%"REG_d"), %%"out_reg1" \n\t"\
00625              "movq (%%"REG_d", %%"REG_c"), %%"out_reg2" \n\t"\
00626              "punpcklbw %%xmm7, %%"out_reg1" \n\t"\
00627              "punpcklbw %%xmm7, %%"out_reg2" \n\t"\
00628              "movq "s_offset"(%%"REG_S"), %%xmm0 \n\t"\
00629              "movq "s_offset"+16(%%"REG_S"), %%xmm4 \n\t"\
00630              "punpcklbw %%xmm7, %%xmm0       \n\t"\
00631              "punpcklbw %%xmm7, %%xmm4       \n\t"\
00632              "pmullw %%xmm0, %%"out_reg1"    \n\t"\
00633              "pmullw %%xmm4, %%"out_reg2"    \n\t"
00634 
00635 #define snow_inner_add_yblock_sse2_start_16(out_reg1, out_reg2, ptr_offset, s_offset)\
00636              "mov "PTR_SIZE"*"ptr_offset"(%%"REG_a"), %%"REG_d"; \n\t"\
00637              "movq (%%"REG_d"), %%"out_reg1" \n\t"\
00638              "movq 8(%%"REG_d"), %%"out_reg2" \n\t"\
00639              "punpcklbw %%xmm7, %%"out_reg1" \n\t"\
00640              "punpcklbw %%xmm7, %%"out_reg2" \n\t"\
00641              "movq "s_offset"(%%"REG_S"), %%xmm0 \n\t"\
00642              "movq "s_offset"+8(%%"REG_S"), %%xmm4 \n\t"\
00643              "punpcklbw %%xmm7, %%xmm0       \n\t"\
00644              "punpcklbw %%xmm7, %%xmm4       \n\t"\
00645              "pmullw %%xmm0, %%"out_reg1"    \n\t"\
00646              "pmullw %%xmm4, %%"out_reg2"    \n\t"
00647 
00648 #define snow_inner_add_yblock_sse2_accum_8(ptr_offset, s_offset) \
00649              snow_inner_add_yblock_sse2_start_8("xmm2", "xmm6", ptr_offset, s_offset)\
00650              "paddusw %%xmm2, %%xmm1         \n\t"\
00651              "paddusw %%xmm6, %%xmm5         \n\t"
00652 
00653 #define snow_inner_add_yblock_sse2_accum_16(ptr_offset, s_offset) \
00654              snow_inner_add_yblock_sse2_start_16("xmm2", "xmm6", ptr_offset, s_offset)\
00655              "paddusw %%xmm2, %%xmm1         \n\t"\
00656              "paddusw %%xmm6, %%xmm5         \n\t"
00657 
00658 #define snow_inner_add_yblock_sse2_end_common1\
00659              "add $32, %%"REG_S"             \n\t"\
00660              "add %%"REG_c", %0              \n\t"\
00661              "add %%"REG_c", "PTR_SIZE"*3(%%"REG_a");\n\t"\
00662              "add %%"REG_c", "PTR_SIZE"*2(%%"REG_a");\n\t"\
00663              "add %%"REG_c", "PTR_SIZE"*1(%%"REG_a");\n\t"\
00664              "add %%"REG_c", (%%"REG_a")     \n\t"
00665 
00666 #define snow_inner_add_yblock_sse2_end_common2\
00667              "jnz 1b                         \n\t"\
00668              :"+m"(dst8),"+m"(dst_array),"=&r"(tmp)\
00669              :\
00670              "rm"((x86_reg)(src_x<<1)),"m"(obmc),"a"(block),"m"((x86_reg)b_h),"m"((x86_reg)src_stride):\
00671              "%"REG_c"","%"REG_S"","%"REG_D"","%"REG_d"");
00672 
00673 #define snow_inner_add_yblock_sse2_end_8\
00674              "sal $1, %%"REG_c"              \n\t"\
00675              "add $"PTR_SIZE"*2, %1          \n\t"\
00676              snow_inner_add_yblock_sse2_end_common1\
00677              "sar $1, %%"REG_c"              \n\t"\
00678              "sub $2, %2                     \n\t"\
00679              snow_inner_add_yblock_sse2_end_common2
00680 
00681 #define snow_inner_add_yblock_sse2_end_16\
00682              "add $"PTR_SIZE"*1, %1          \n\t"\
00683              snow_inner_add_yblock_sse2_end_common1\
00684              "dec %2                         \n\t"\
00685              snow_inner_add_yblock_sse2_end_common2
00686 
00687 static void inner_add_yblock_bw_8_obmc_16_bh_even_sse2(const uint8_t *obmc, const x86_reg obmc_stride, uint8_t * * block, int b_w, x86_reg b_h,
00688                       int src_x, int src_y, x86_reg src_stride, slice_buffer * sb, int add, uint8_t * dst8){
00689 snow_inner_add_yblock_sse2_header
00690 snow_inner_add_yblock_sse2_start_8("xmm1", "xmm5", "3", "0")
00691 snow_inner_add_yblock_sse2_accum_8("2", "8")
00692 snow_inner_add_yblock_sse2_accum_8("1", "128")
00693 snow_inner_add_yblock_sse2_accum_8("0", "136")
00694 
00695              "mov %0, %%"REG_d"              \n\t"
00696              "movdqa (%%"REG_D"), %%xmm0     \n\t"
00697              "movdqa %%xmm1, %%xmm2          \n\t"
00698 
00699              "punpckhwd %%xmm7, %%xmm1       \n\t"
00700              "punpcklwd %%xmm7, %%xmm2       \n\t"
00701              "paddd %%xmm2, %%xmm0           \n\t"
00702              "movdqa 16(%%"REG_D"), %%xmm2   \n\t"
00703              "paddd %%xmm1, %%xmm2           \n\t"
00704              "paddd %%xmm3, %%xmm0           \n\t"
00705              "paddd %%xmm3, %%xmm2           \n\t"
00706 
00707              "mov %1, %%"REG_D"              \n\t"
00708              "mov "PTR_SIZE"(%%"REG_D"), %%"REG_D";\n\t"
00709              "add %3, %%"REG_D"              \n\t"
00710 
00711              "movdqa (%%"REG_D"), %%xmm4     \n\t"
00712              "movdqa %%xmm5, %%xmm6          \n\t"
00713              "punpckhwd %%xmm7, %%xmm5       \n\t"
00714              "punpcklwd %%xmm7, %%xmm6       \n\t"
00715              "paddd %%xmm6, %%xmm4           \n\t"
00716              "movdqa 16(%%"REG_D"), %%xmm6   \n\t"
00717              "paddd %%xmm5, %%xmm6           \n\t"
00718              "paddd %%xmm3, %%xmm4           \n\t"
00719              "paddd %%xmm3, %%xmm6           \n\t"
00720 
00721              "psrad $8, %%xmm0               \n\t" /* FRAC_BITS. */
00722              "psrad $8, %%xmm2               \n\t" /* FRAC_BITS. */
00723              "packssdw %%xmm2, %%xmm0        \n\t"
00724              "packuswb %%xmm7, %%xmm0        \n\t"
00725              "movq %%xmm0, (%%"REG_d")       \n\t"
00726 
00727              "psrad $8, %%xmm4               \n\t" /* FRAC_BITS. */
00728              "psrad $8, %%xmm6               \n\t" /* FRAC_BITS. */
00729              "packssdw %%xmm6, %%xmm4        \n\t"
00730              "packuswb %%xmm7, %%xmm4        \n\t"
00731              "movq %%xmm4, (%%"REG_d",%%"REG_c");\n\t"
00732 snow_inner_add_yblock_sse2_end_8
00733 }
00734 
00735 static void inner_add_yblock_bw_16_obmc_32_sse2(const uint8_t *obmc, const x86_reg obmc_stride, uint8_t * * block, int b_w, x86_reg b_h,
00736                       int src_x, int src_y, x86_reg src_stride, slice_buffer * sb, int add, uint8_t * dst8){
00737 snow_inner_add_yblock_sse2_header
00738 snow_inner_add_yblock_sse2_start_16("xmm1", "xmm5", "3", "0")
00739 snow_inner_add_yblock_sse2_accum_16("2", "16")
00740 snow_inner_add_yblock_sse2_accum_16("1", "512")
00741 snow_inner_add_yblock_sse2_accum_16("0", "528")
00742 
00743              "mov %0, %%"REG_d"              \n\t"
00744              "psrlw $4, %%xmm1               \n\t"
00745              "psrlw $4, %%xmm5               \n\t"
00746              "paddw   (%%"REG_D"), %%xmm1    \n\t"
00747              "paddw 16(%%"REG_D"), %%xmm5    \n\t"
00748              "paddw %%xmm3, %%xmm1           \n\t"
00749              "paddw %%xmm3, %%xmm5           \n\t"
00750              "psraw $4, %%xmm1               \n\t" /* FRAC_BITS. */
00751              "psraw $4, %%xmm5               \n\t" /* FRAC_BITS. */
00752              "packuswb %%xmm5, %%xmm1        \n\t"
00753 
00754              "movdqu %%xmm1, (%%"REG_d")       \n\t"
00755 
00756 snow_inner_add_yblock_sse2_end_16
00757 }
00758 
00759 #define snow_inner_add_yblock_mmx_header \
00760     IDWTELEM * * dst_array = sb->line + src_y;\
00761     x86_reg tmp;\
00762     __asm__ volatile(\
00763              "mov  %7, %%"REG_c"             \n\t"\
00764              "mov  %6, %2                    \n\t"\
00765              "mov  %4, %%"REG_S"             \n\t"\
00766              "pxor %%mm7, %%mm7              \n\t" /* 0 */\
00767              "pcmpeqd %%mm3, %%mm3           \n\t"\
00768              "psllw $15, %%mm3               \n\t"\
00769              "psrlw $12, %%mm3               \n\t" /* FRAC_BITS >> 1 */\
00770              "1:                             \n\t"\
00771              "mov %1, %%"REG_D"              \n\t"\
00772              "mov (%%"REG_D"), %%"REG_D"     \n\t"\
00773              "add %3, %%"REG_D"              \n\t"
00774 
00775 #define snow_inner_add_yblock_mmx_start(out_reg1, out_reg2, ptr_offset, s_offset, d_offset)\
00776              "mov "PTR_SIZE"*"ptr_offset"(%%"REG_a"), %%"REG_d"; \n\t"\
00777              "movd "d_offset"(%%"REG_d"), %%"out_reg1" \n\t"\
00778              "movd "d_offset"+4(%%"REG_d"), %%"out_reg2" \n\t"\
00779              "punpcklbw %%mm7, %%"out_reg1" \n\t"\
00780              "punpcklbw %%mm7, %%"out_reg2" \n\t"\
00781              "movd "s_offset"(%%"REG_S"), %%mm0 \n\t"\
00782              "movd "s_offset"+4(%%"REG_S"), %%mm4 \n\t"\
00783              "punpcklbw %%mm7, %%mm0       \n\t"\
00784              "punpcklbw %%mm7, %%mm4       \n\t"\
00785              "pmullw %%mm0, %%"out_reg1"    \n\t"\
00786              "pmullw %%mm4, %%"out_reg2"    \n\t"
00787 
00788 #define snow_inner_add_yblock_mmx_accum(ptr_offset, s_offset, d_offset) \
00789              snow_inner_add_yblock_mmx_start("mm2", "mm6", ptr_offset, s_offset, d_offset)\
00790              "paddusw %%mm2, %%mm1         \n\t"\
00791              "paddusw %%mm6, %%mm5         \n\t"
00792 
00793 #define snow_inner_add_yblock_mmx_mix(read_offset, write_offset)\
00794              "mov %0, %%"REG_d"              \n\t"\
00795              "psrlw $4, %%mm1                \n\t"\
00796              "psrlw $4, %%mm5                \n\t"\
00797              "paddw "read_offset"(%%"REG_D"), %%mm1 \n\t"\
00798              "paddw "read_offset"+8(%%"REG_D"), %%mm5 \n\t"\
00799              "paddw %%mm3, %%mm1             \n\t"\
00800              "paddw %%mm3, %%mm5             \n\t"\
00801              "psraw $4, %%mm1                \n\t"\
00802              "psraw $4, %%mm5                \n\t"\
00803              "packuswb %%mm5, %%mm1          \n\t"\
00804              "movq %%mm1, "write_offset"(%%"REG_d") \n\t"
00805 
00806 #define snow_inner_add_yblock_mmx_end(s_step)\
00807              "add $"s_step", %%"REG_S"             \n\t"\
00808              "add %%"REG_c", "PTR_SIZE"*3(%%"REG_a");\n\t"\
00809              "add %%"REG_c", "PTR_SIZE"*2(%%"REG_a");\n\t"\
00810              "add %%"REG_c", "PTR_SIZE"*1(%%"REG_a");\n\t"\
00811              "add %%"REG_c", (%%"REG_a")     \n\t"\
00812              "add $"PTR_SIZE"*1, %1          \n\t"\
00813              "add %%"REG_c", %0              \n\t"\
00814              "dec %2                         \n\t"\
00815              "jnz 1b                         \n\t"\
00816              :"+m"(dst8),"+m"(dst_array),"=&r"(tmp)\
00817              :\
00818              "rm"((x86_reg)(src_x<<1)),"m"(obmc),"a"(block),"m"((x86_reg)b_h),"m"((x86_reg)src_stride):\
00819              "%"REG_c"","%"REG_S"","%"REG_D"","%"REG_d"");
00820 
00821 static void inner_add_yblock_bw_8_obmc_16_mmx(const uint8_t *obmc, const x86_reg obmc_stride, uint8_t * * block, int b_w, x86_reg b_h,
00822                       int src_x, int src_y, x86_reg src_stride, slice_buffer * sb, int add, uint8_t * dst8){
00823 snow_inner_add_yblock_mmx_header
00824 snow_inner_add_yblock_mmx_start("mm1", "mm5", "3", "0", "0")
00825 snow_inner_add_yblock_mmx_accum("2", "8", "0")
00826 snow_inner_add_yblock_mmx_accum("1", "128", "0")
00827 snow_inner_add_yblock_mmx_accum("0", "136", "0")
00828 snow_inner_add_yblock_mmx_mix("0", "0")
00829 snow_inner_add_yblock_mmx_end("16")
00830 }
00831 
00832 static void inner_add_yblock_bw_16_obmc_32_mmx(const uint8_t *obmc, const x86_reg obmc_stride, uint8_t * * block, int b_w, x86_reg b_h,
00833                       int src_x, int src_y, x86_reg src_stride, slice_buffer * sb, int add, uint8_t * dst8){
00834 snow_inner_add_yblock_mmx_header
00835 snow_inner_add_yblock_mmx_start("mm1", "mm5", "3", "0", "0")
00836 snow_inner_add_yblock_mmx_accum("2", "16", "0")
00837 snow_inner_add_yblock_mmx_accum("1", "512", "0")
00838 snow_inner_add_yblock_mmx_accum("0", "528", "0")
00839 snow_inner_add_yblock_mmx_mix("0", "0")
00840 
00841 snow_inner_add_yblock_mmx_start("mm1", "mm5", "3", "8", "8")
00842 snow_inner_add_yblock_mmx_accum("2", "24", "8")
00843 snow_inner_add_yblock_mmx_accum("1", "520", "8")
00844 snow_inner_add_yblock_mmx_accum("0", "536", "8")
00845 snow_inner_add_yblock_mmx_mix("16", "8")
00846 snow_inner_add_yblock_mmx_end("32")
00847 }
00848 
00849 void ff_snow_inner_add_yblock_sse2(const uint8_t *obmc, const int obmc_stride, uint8_t * * block, int b_w, int b_h,
00850                            int src_x, int src_y, int src_stride, slice_buffer * sb, int add, uint8_t * dst8){
00851 
00852     if (b_w == 16)
00853         inner_add_yblock_bw_16_obmc_32_sse2(obmc, obmc_stride, block, b_w, b_h, src_x,src_y, src_stride, sb, add, dst8);
00854     else if (b_w == 8 && obmc_stride == 16) {
00855         if (!(b_h & 1))
00856             inner_add_yblock_bw_8_obmc_16_bh_even_sse2(obmc, obmc_stride, block, b_w, b_h, src_x,src_y, src_stride, sb, add, dst8);
00857         else
00858             inner_add_yblock_bw_8_obmc_16_mmx(obmc, obmc_stride, block, b_w, b_h, src_x,src_y, src_stride, sb, add, dst8);
00859     } else
00860          ff_snow_inner_add_yblock(obmc, obmc_stride, block, b_w, b_h, src_x,src_y, src_stride, sb, add, dst8);
00861 }
00862 
00863 void ff_snow_inner_add_yblock_mmx(const uint8_t *obmc, const int obmc_stride, uint8_t * * block, int b_w, int b_h,
00864                           int src_x, int src_y, int src_stride, slice_buffer * sb, int add, uint8_t * dst8){
00865     if (b_w == 16)
00866         inner_add_yblock_bw_16_obmc_32_mmx(obmc, obmc_stride, block, b_w, b_h, src_x,src_y, src_stride, sb, add, dst8);
00867     else if (b_w == 8 && obmc_stride == 16)
00868         inner_add_yblock_bw_8_obmc_16_mmx(obmc, obmc_stride, block, b_w, b_h, src_x,src_y, src_stride, sb, add, dst8);
00869     else
00870         ff_snow_inner_add_yblock(obmc, obmc_stride, block, b_w, b_h, src_x,src_y, src_stride, sb, add, dst8);
00871 }
libavcodec/x86/snowdsp_mmx.c
