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/cpu.h"
00023 #include "libavutil/x86_cpu.h"
00024 #include "libavcodec/avcodec.h"
00025 #include "libavcodec/snow.h"
00026 #include "libavcodec/dwt.h"
00027 #include "dsputil_mmx.h"
00028 
00029 static void ff_snow_horizontal_compose97i_sse2(IDWTELEM *b, int width){
00030     const int w2= (width+1)>>1;
00031     DECLARE_ALIGNED(16, IDWTELEM, temp)[width>>1];
00032     const int w_l= (width>>1);
00033     const int w_r= w2 - 1;
00034     int i;
00035 
00036     { // Lift 0
00037         IDWTELEM * const ref = b + w2 - 1;
00038         IDWTELEM b_0 = b[0]; //By allowing the first entry in b[0] to be calculated twice
00039         // (the first time erroneously), we allow the SSE2 code to run an extra pass.
00040         // The savings in code and time are well worth having to store this value and
00041         // calculate b[0] correctly afterwards.
00042 
00043         i = 0;
00044         __asm__ volatile(
00045             "pcmpeqd   %%xmm7, %%xmm7         \n\t"
00046             "pcmpeqd   %%xmm3, %%xmm3         \n\t"
00047             "psllw         $1, %%xmm3         \n\t"
00048             "paddw     %%xmm7, %%xmm3         \n\t"
00049             "psllw        $13, %%xmm3         \n\t"
00050         ::);
00051         for(; i<w_l-15; i+=16){
00052             __asm__ volatile(
00053                 "movdqu   (%1), %%xmm1        \n\t"
00054                 "movdqu 16(%1), %%xmm5        \n\t"
00055                 "movdqu  2(%1), %%xmm2        \n\t"
00056                 "movdqu 18(%1), %%xmm6        \n\t"
00057                 "paddw  %%xmm1, %%xmm2        \n\t"
00058                 "paddw  %%xmm5, %%xmm6        \n\t"
00059                 "paddw  %%xmm7, %%xmm2        \n\t"
00060                 "paddw  %%xmm7, %%xmm6        \n\t"
00061                 "pmulhw %%xmm3, %%xmm2        \n\t"
00062                 "pmulhw %%xmm3, %%xmm6        \n\t"
00063                 "paddw    (%0), %%xmm2        \n\t"
00064                 "paddw  16(%0), %%xmm6        \n\t"
00065                 "movdqa %%xmm2, (%0)          \n\t"
00066                 "movdqa %%xmm6, 16(%0)        \n\t"
00067                 :: "r"(&b[i]), "r"(&ref[i])
00068                 : "memory"
00069             );
00070         }
00071         snow_horizontal_compose_lift_lead_out(i, b, b, ref, width, w_l, 0, W_DM, W_DO, W_DS);
00072         b[0] = b_0 - ((W_DM * 2 * ref[1]+W_DO)>>W_DS);
00073     }
00074 
00075     { // Lift 1
00076         IDWTELEM * const dst = b+w2;
00077 
00078         i = 0;
00079         for(; (((x86_reg)&dst[i]) & 0x1F) && i<w_r; i++){
00080             dst[i] = dst[i] - (b[i] + b[i + 1]);
00081         }
00082         for(; i<w_r-15; i+=16){
00083             __asm__ volatile(
00084                 "movdqu   (%1), %%xmm1        \n\t"
00085                 "movdqu 16(%1), %%xmm5        \n\t"
00086                 "movdqu  2(%1), %%xmm2        \n\t"
00087                 "movdqu 18(%1), %%xmm6        \n\t"
00088                 "paddw  %%xmm1, %%xmm2        \n\t"
00089                 "paddw  %%xmm5, %%xmm6        \n\t"
00090                 "movdqa   (%0), %%xmm0        \n\t"
00091                 "movdqa 16(%0), %%xmm4        \n\t"
00092                 "psubw  %%xmm2, %%xmm0        \n\t"
00093                 "psubw  %%xmm6, %%xmm4        \n\t"
00094                 "movdqa %%xmm0, (%0)          \n\t"
00095                 "movdqa %%xmm4, 16(%0)        \n\t"
00096                 :: "r"(&dst[i]), "r"(&b[i])
00097                 : "memory"
00098             );
00099         }
00100         snow_horizontal_compose_lift_lead_out(i, dst, dst, b, width, w_r, 1, W_CM, W_CO, W_CS);
00101     }
00102 
00103     { // Lift 2
00104         IDWTELEM * const ref = b+w2 - 1;
00105         IDWTELEM b_0 = b[0];
00106 
00107         i = 0;
00108         __asm__ volatile(
00109             "psllw         $15, %%xmm7        \n\t"
00110             "pcmpeqw    %%xmm6, %%xmm6        \n\t"
00111             "psrlw         $13, %%xmm6        \n\t"
00112             "paddw      %%xmm7, %%xmm6        \n\t"
00113         ::);
00114         for(; i<w_l-15; i+=16){
00115             __asm__ volatile(
00116                 "movdqu   (%1), %%xmm0        \n\t"
00117                 "movdqu 16(%1), %%xmm4        \n\t"
00118                 "movdqu  2(%1), %%xmm1        \n\t"
00119                 "movdqu 18(%1), %%xmm5        \n\t" //FIXME try aligned reads and shifts
00120                 "paddw  %%xmm6, %%xmm0        \n\t"
00121                 "paddw  %%xmm6, %%xmm4        \n\t"
00122                 "paddw  %%xmm7, %%xmm1        \n\t"
00123                 "paddw  %%xmm7, %%xmm5        \n\t"
00124                 "pavgw  %%xmm1, %%xmm0        \n\t"
00125                 "pavgw  %%xmm5, %%xmm4        \n\t"
00126                 "psubw  %%xmm7, %%xmm0        \n\t"
00127                 "psubw  %%xmm7, %%xmm4        \n\t"
00128                 "psraw      $1, %%xmm0        \n\t"
00129                 "psraw      $1, %%xmm4        \n\t"
00130                 "movdqa   (%0), %%xmm1        \n\t"
00131                 "movdqa 16(%0), %%xmm5        \n\t"
00132                 "paddw  %%xmm1, %%xmm0        \n\t"
00133                 "paddw  %%xmm5, %%xmm4        \n\t"
00134                 "psraw      $2, %%xmm0        \n\t"
00135                 "psraw      $2, %%xmm4        \n\t"
00136                 "paddw  %%xmm1, %%xmm0        \n\t"
00137                 "paddw  %%xmm5, %%xmm4        \n\t"
00138                 "movdqa %%xmm0, (%0)          \n\t"
00139                 "movdqa %%xmm4, 16(%0)        \n\t"
00140                 :: "r"(&b[i]), "r"(&ref[i])
00141                 : "memory"
00142             );
00143         }
00144         snow_horizontal_compose_liftS_lead_out(i, b, b, ref, width, w_l);
00145         b[0] = b_0 + ((2 * ref[1] + W_BO-1 + 4 * b_0) >> W_BS);
00146     }
00147 
00148     { // Lift 3
00149         IDWTELEM * const src = b+w2;
00150 
00151         i = 0;
00152         for(; (((x86_reg)&temp[i]) & 0x1F) && i<w_r; i++){
00153             temp[i] = src[i] - ((-W_AM*(b[i] + b[i+1]))>>W_AS);
00154         }
00155         for(; i<w_r-7; i+=8){
00156             __asm__ volatile(
00157                 "movdqu  2(%1), %%xmm2        \n\t"
00158                 "movdqu 18(%1), %%xmm6        \n\t"
00159                 "paddw    (%1), %%xmm2        \n\t"
00160                 "paddw  16(%1), %%xmm6        \n\t"
00161                 "movdqu   (%0), %%xmm0        \n\t"
00162                 "movdqu 16(%0), %%xmm4        \n\t"
00163                 "paddw  %%xmm2, %%xmm0        \n\t"
00164                 "paddw  %%xmm6, %%xmm4        \n\t"
00165                 "psraw      $1, %%xmm2        \n\t"
00166                 "psraw      $1, %%xmm6        \n\t"
00167                 "paddw  %%xmm0, %%xmm2        \n\t"
00168                 "paddw  %%xmm4, %%xmm6        \n\t"
00169                 "movdqa %%xmm2, (%2)          \n\t"
00170                 "movdqa %%xmm6, 16(%2)        \n\t"
00171                 :: "r"(&src[i]), "r"(&b[i]), "r"(&temp[i])
00172                  : "memory"
00173                );
00174         }
00175         snow_horizontal_compose_lift_lead_out(i, temp, src, b, width, w_r, 1, -W_AM, W_AO+1, W_AS);
00176     }
00177 
00178     {
00179         snow_interleave_line_header(&i, width, b, temp);
00180 
00181         for (; (i & 0x3E) != 0x3E; i-=2){
00182             b[i+1] = temp[i>>1];
00183             b[i] = b[i>>1];
00184         }
00185         for (i-=62; i>=0; i-=64){
00186             __asm__ volatile(
00187                 "movdqa      (%1), %%xmm0       \n\t"
00188                 "movdqa    16(%1), %%xmm2       \n\t"
00189                 "movdqa    32(%1), %%xmm4       \n\t"
00190                 "movdqa    48(%1), %%xmm6       \n\t"
00191                 "movdqa      (%1), %%xmm1       \n\t"
00192                 "movdqa    16(%1), %%xmm3       \n\t"
00193                 "movdqa    32(%1), %%xmm5       \n\t"
00194                 "movdqa    48(%1), %%xmm7       \n\t"
00195                 "punpcklwd   (%2), %%xmm0       \n\t"
00196                 "punpcklwd 16(%2), %%xmm2       \n\t"
00197                 "punpcklwd 32(%2), %%xmm4       \n\t"
00198                 "punpcklwd 48(%2), %%xmm6       \n\t"
00199                 "movdqa    %%xmm0, (%0)         \n\t"
00200                 "movdqa    %%xmm2, 32(%0)       \n\t"
00201                 "movdqa    %%xmm4, 64(%0)       \n\t"
00202                 "movdqa    %%xmm6, 96(%0)       \n\t"
00203                 "punpckhwd   (%2), %%xmm1       \n\t"
00204                 "punpckhwd 16(%2), %%xmm3       \n\t"
00205                 "punpckhwd 32(%2), %%xmm5       \n\t"
00206                 "punpckhwd 48(%2), %%xmm7       \n\t"
00207                 "movdqa    %%xmm1, 16(%0)       \n\t"
00208                 "movdqa    %%xmm3, 48(%0)       \n\t"
00209                 "movdqa    %%xmm5, 80(%0)       \n\t"
00210                 "movdqa    %%xmm7, 112(%0)      \n\t"
00211                 :: "r"(&(b)[i]), "r"(&(b)[i>>1]), "r"(&(temp)[i>>1])
00212                  : "memory"
00213                );
00214         }
00215     }
00216 }
00217 
00218 static void ff_snow_horizontal_compose97i_mmx(IDWTELEM *b, int width){
00219     const int w2= (width+1)>>1;
00220     IDWTELEM temp[width >> 1];
00221     const int w_l= (width>>1);
00222     const int w_r= w2 - 1;
00223     int i;
00224 
00225     { // Lift 0
00226         IDWTELEM * const ref = b + w2 - 1;
00227 
00228         i = 1;
00229         b[0] = b[0] - ((W_DM * 2 * ref[1]+W_DO)>>W_DS);
00230         __asm__ volatile(
00231             "pcmpeqw    %%mm7, %%mm7         \n\t"
00232             "pcmpeqw    %%mm3, %%mm3         \n\t"
00233             "psllw         $1, %%mm3         \n\t"
00234             "paddw      %%mm7, %%mm3         \n\t"
00235             "psllw        $13, %%mm3         \n\t"
00236            ::);
00237         for(; i<w_l-7; i+=8){
00238             __asm__ volatile(
00239                 "movq     (%1), %%mm2        \n\t"
00240                 "movq    8(%1), %%mm6        \n\t"
00241                 "paddw   2(%1), %%mm2        \n\t"
00242                 "paddw  10(%1), %%mm6        \n\t"
00243                 "paddw   %%mm7, %%mm2        \n\t"
00244                 "paddw   %%mm7, %%mm6        \n\t"
00245                 "pmulhw  %%mm3, %%mm2        \n\t"
00246                 "pmulhw  %%mm3, %%mm6        \n\t"
00247                 "paddw    (%0), %%mm2        \n\t"
00248                 "paddw   8(%0), %%mm6        \n\t"
00249                 "movq    %%mm2, (%0)         \n\t"
00250                 "movq    %%mm6, 8(%0)        \n\t"
00251                 :: "r"(&b[i]), "r"(&ref[i])
00252                  : "memory"
00253                );
00254         }
00255         snow_horizontal_compose_lift_lead_out(i, b, b, ref, width, w_l, 0, W_DM, W_DO, W_DS);
00256     }
00257 
00258     { // Lift 1
00259         IDWTELEM * const dst = b+w2;
00260 
00261         i = 0;
00262         for(; i<w_r-7; i+=8){
00263             __asm__ volatile(
00264                 "movq     (%1), %%mm2        \n\t"
00265                 "movq    8(%1), %%mm6        \n\t"
00266                 "paddw   2(%1), %%mm2        \n\t"
00267                 "paddw  10(%1), %%mm6        \n\t"
00268                 "movq     (%0), %%mm0        \n\t"
00269                 "movq    8(%0), %%mm4        \n\t"
00270                 "psubw   %%mm2, %%mm0        \n\t"
00271                 "psubw   %%mm6, %%mm4        \n\t"
00272                 "movq    %%mm0, (%0)         \n\t"
00273                 "movq    %%mm4, 8(%0)        \n\t"
00274                 :: "r"(&dst[i]), "r"(&b[i])
00275                  : "memory"
00276                );
00277         }
00278         snow_horizontal_compose_lift_lead_out(i, dst, dst, b, width, w_r, 1, W_CM, W_CO, W_CS);
00279     }
00280 
00281     { // Lift 2
00282         IDWTELEM * const ref = b+w2 - 1;
00283 
00284         i = 1;
00285         b[0] = b[0] + (((2 * ref[1] + W_BO) + 4 * b[0]) >> W_BS);
00286         __asm__ volatile(
00287             "psllw         $15, %%mm7        \n\t"
00288             "pcmpeqw     %%mm6, %%mm6        \n\t"
00289             "psrlw         $13, %%mm6        \n\t"
00290             "paddw       %%mm7, %%mm6        \n\t"
00291            ::);
00292         for(; i<w_l-7; i+=8){
00293             __asm__ volatile(
00294                 "movq     (%1), %%mm0        \n\t"
00295                 "movq    8(%1), %%mm4        \n\t"
00296                 "movq    2(%1), %%mm1        \n\t"
00297                 "movq   10(%1), %%mm5        \n\t"
00298                 "paddw   %%mm6, %%mm0        \n\t"
00299                 "paddw   %%mm6, %%mm4        \n\t"
00300                 "paddw   %%mm7, %%mm1        \n\t"
00301                 "paddw   %%mm7, %%mm5        \n\t"
00302                 "pavgw   %%mm1, %%mm0        \n\t"
00303                 "pavgw   %%mm5, %%mm4        \n\t"
00304                 "psubw   %%mm7, %%mm0        \n\t"
00305                 "psubw   %%mm7, %%mm4        \n\t"
00306                 "psraw      $1, %%mm0        \n\t"
00307                 "psraw      $1, %%mm4        \n\t"
00308                 "movq     (%0), %%mm1        \n\t"
00309                 "movq    8(%0), %%mm5        \n\t"
00310                 "paddw   %%mm1, %%mm0        \n\t"
00311                 "paddw   %%mm5, %%mm4        \n\t"
00312                 "psraw      $2, %%mm0        \n\t"
00313                 "psraw      $2, %%mm4        \n\t"
00314                 "paddw   %%mm1, %%mm0        \n\t"
00315                 "paddw   %%mm5, %%mm4        \n\t"
00316                 "movq    %%mm0, (%0)         \n\t"
00317                 "movq    %%mm4, 8(%0)        \n\t"
00318                 :: "r"(&b[i]), "r"(&ref[i])
00319                  : "memory"
00320                );
00321         }
00322         snow_horizontal_compose_liftS_lead_out(i, b, b, ref, width, w_l);
00323     }
00324 
00325     { // Lift 3
00326         IDWTELEM * const src = b+w2;
00327         i = 0;
00328 
00329         for(; i<w_r-7; i+=8){
00330             __asm__ volatile(
00331                 "movq    2(%1), %%mm2        \n\t"
00332                 "movq   10(%1), %%mm6        \n\t"
00333                 "paddw    (%1), %%mm2        \n\t"
00334                 "paddw   8(%1), %%mm6        \n\t"
00335                 "movq     (%0), %%mm0        \n\t"
00336                 "movq    8(%0), %%mm4        \n\t"
00337                 "paddw   %%mm2, %%mm0        \n\t"
00338                 "paddw   %%mm6, %%mm4        \n\t"
00339                 "psraw      $1, %%mm2        \n\t"
00340                 "psraw      $1, %%mm6        \n\t"
00341                 "paddw   %%mm0, %%mm2        \n\t"
00342                 "paddw   %%mm4, %%mm6        \n\t"
00343                 "movq    %%mm2, (%2)         \n\t"
00344                 "movq    %%mm6, 8(%2)        \n\t"
00345                 :: "r"(&src[i]), "r"(&b[i]), "r"(&temp[i])
00346                  : "memory"
00347                );
00348         }
00349         snow_horizontal_compose_lift_lead_out(i, temp, src, b, width, w_r, 1, -W_AM, W_AO+1, W_AS);
00350     }
00351 
00352     {
00353         snow_interleave_line_header(&i, width, b, temp);
00354 
00355         for (; (i & 0x1E) != 0x1E; i-=2){
00356             b[i+1] = temp[i>>1];
00357             b[i] = b[i>>1];
00358         }
00359         for (i-=30; i>=0; i-=32){
00360             __asm__ volatile(
00361                 "movq        (%1), %%mm0       \n\t"
00362                 "movq       8(%1), %%mm2       \n\t"
00363                 "movq      16(%1), %%mm4       \n\t"
00364                 "movq      24(%1), %%mm6       \n\t"
00365                 "movq        (%1), %%mm1       \n\t"
00366                 "movq       8(%1), %%mm3       \n\t"
00367                 "movq      16(%1), %%mm5       \n\t"
00368                 "movq      24(%1), %%mm7       \n\t"
00369                 "punpcklwd   (%2), %%mm0       \n\t"
00370                 "punpcklwd  8(%2), %%mm2       \n\t"
00371                 "punpcklwd 16(%2), %%mm4       \n\t"
00372                 "punpcklwd 24(%2), %%mm6       \n\t"
00373                 "movq       %%mm0, (%0)        \n\t"
00374                 "movq       %%mm2, 16(%0)      \n\t"
00375                 "movq       %%mm4, 32(%0)      \n\t"
00376                 "movq       %%mm6, 48(%0)      \n\t"
00377                 "punpckhwd   (%2), %%mm1       \n\t"
00378                 "punpckhwd  8(%2), %%mm3       \n\t"
00379                 "punpckhwd 16(%2), %%mm5       \n\t"
00380                 "punpckhwd 24(%2), %%mm7       \n\t"
00381                 "movq       %%mm1, 8(%0)       \n\t"
00382                 "movq       %%mm3, 24(%0)      \n\t"
00383                 "movq       %%mm5, 40(%0)      \n\t"
00384                 "movq       %%mm7, 56(%0)      \n\t"
00385                 :: "r"(&b[i]), "r"(&b[i>>1]), "r"(&temp[i>>1])
00386                  : "memory"
00387                );
00388         }
00389     }
00390 }
00391 
00392 #if HAVE_7REGS
00393 #define snow_vertical_compose_sse2_load_add(op,r,t0,t1,t2,t3)\
00394         ""op" ("r",%%"REG_d"), %%"t0"      \n\t"\
00395         ""op" 16("r",%%"REG_d"), %%"t1"    \n\t"\
00396         ""op" 32("r",%%"REG_d"), %%"t2"    \n\t"\
00397         ""op" 48("r",%%"REG_d"), %%"t3"    \n\t"
00398 
00399 #define snow_vertical_compose_sse2_load(r,t0,t1,t2,t3)\
00400         snow_vertical_compose_sse2_load_add("movdqa",r,t0,t1,t2,t3)
00401 
00402 #define snow_vertical_compose_sse2_add(r,t0,t1,t2,t3)\
00403         snow_vertical_compose_sse2_load_add("paddw",r,t0,t1,t2,t3)
00404 
00405 #define snow_vertical_compose_r2r_sub(s0,s1,s2,s3,t0,t1,t2,t3)\
00406         "psubw %%"s0", %%"t0" \n\t"\
00407         "psubw %%"s1", %%"t1" \n\t"\
00408         "psubw %%"s2", %%"t2" \n\t"\
00409         "psubw %%"s3", %%"t3" \n\t"
00410 
00411 #define snow_vertical_compose_sse2_store(w,s0,s1,s2,s3)\
00412         "movdqa %%"s0", ("w",%%"REG_d")      \n\t"\
00413         "movdqa %%"s1", 16("w",%%"REG_d")    \n\t"\
00414         "movdqa %%"s2", 32("w",%%"REG_d")    \n\t"\
00415         "movdqa %%"s3", 48("w",%%"REG_d")    \n\t"
00416 
00417 #define snow_vertical_compose_sra(n,t0,t1,t2,t3)\
00418         "psraw $"n", %%"t0" \n\t"\
00419         "psraw $"n", %%"t1" \n\t"\
00420         "psraw $"n", %%"t2" \n\t"\
00421         "psraw $"n", %%"t3" \n\t"
00422 
00423 #define snow_vertical_compose_r2r_add(s0,s1,s2,s3,t0,t1,t2,t3)\
00424         "paddw %%"s0", %%"t0" \n\t"\
00425         "paddw %%"s1", %%"t1" \n\t"\
00426         "paddw %%"s2", %%"t2" \n\t"\
00427         "paddw %%"s3", %%"t3" \n\t"
00428 
00429 #define snow_vertical_compose_r2r_pmulhw(s0,s1,s2,s3,t0,t1,t2,t3)\
00430         "pmulhw %%"s0", %%"t0" \n\t"\
00431         "pmulhw %%"s1", %%"t1" \n\t"\
00432         "pmulhw %%"s2", %%"t2" \n\t"\
00433         "pmulhw %%"s3", %%"t3" \n\t"
00434 
00435 #define snow_vertical_compose_sse2_move(s0,s1,s2,s3,t0,t1,t2,t3)\
00436         "movdqa %%"s0", %%"t0" \n\t"\
00437         "movdqa %%"s1", %%"t1" \n\t"\
00438         "movdqa %%"s2", %%"t2" \n\t"\
00439         "movdqa %%"s3", %%"t3" \n\t"
00440 
00441 static void ff_snow_vertical_compose97i_sse2(IDWTELEM *b0, IDWTELEM *b1, IDWTELEM *b2, IDWTELEM *b3, IDWTELEM *b4, IDWTELEM *b5, int width){
00442     x86_reg i = width;
00443 
00444     while(i & 0x1F)
00445     {
00446         i--;
00447         b4[i] -= (W_DM*(b3[i] + b5[i])+W_DO)>>W_DS;
00448         b3[i] -= (W_CM*(b2[i] + b4[i])+W_CO)>>W_CS;
00449         b2[i] += (W_BM*(b1[i] + b3[i])+4*b2[i]+W_BO)>>W_BS;
00450         b1[i] += (W_AM*(b0[i] + b2[i])+W_AO)>>W_AS;
00451     }
00452     i+=i;
00453 
00454          __asm__ volatile (
00455         "jmp 2f                                      \n\t"
00456         "1:                                          \n\t"
00457         snow_vertical_compose_sse2_load("%4","xmm0","xmm2","xmm4","xmm6")
00458         snow_vertical_compose_sse2_add("%6","xmm0","xmm2","xmm4","xmm6")
00459 
00460 
00461         "pcmpeqw    %%xmm0, %%xmm0                   \n\t"
00462         "pcmpeqw    %%xmm2, %%xmm2                   \n\t"
00463         "paddw      %%xmm2, %%xmm2                   \n\t"
00464         "paddw      %%xmm0, %%xmm2                   \n\t"
00465         "psllw         $13, %%xmm2                   \n\t"
00466         snow_vertical_compose_r2r_add("xmm0","xmm0","xmm0","xmm0","xmm1","xmm3","xmm5","xmm7")
00467         snow_vertical_compose_r2r_pmulhw("xmm2","xmm2","xmm2","xmm2","xmm1","xmm3","xmm5","xmm7")
00468         snow_vertical_compose_sse2_add("%5","xmm1","xmm3","xmm5","xmm7")
00469         snow_vertical_compose_sse2_store("%5","xmm1","xmm3","xmm5","xmm7")
00470         snow_vertical_compose_sse2_load("%4","xmm0","xmm2","xmm4","xmm6")
00471         snow_vertical_compose_sse2_add("%3","xmm1","xmm3","xmm5","xmm7")
00472         snow_vertical_compose_r2r_sub("xmm1","xmm3","xmm5","xmm7","xmm0","xmm2","xmm4","xmm6")
00473         snow_vertical_compose_sse2_store("%4","xmm0","xmm2","xmm4","xmm6")
00474 
00475         "pcmpeqw %%xmm7, %%xmm7                      \n\t"
00476         "pcmpeqw %%xmm5, %%xmm5                      \n\t"
00477         "psllw $15, %%xmm7                           \n\t"
00478         "psrlw $13, %%xmm5                           \n\t"
00479         "paddw %%xmm7, %%xmm5                        \n\t"
00480         snow_vertical_compose_r2r_add("xmm5","xmm5","xmm5","xmm5","xmm0","xmm2","xmm4","xmm6")
00481         "movq   (%2,%%"REG_d"), %%xmm1        \n\t"
00482         "movq  8(%2,%%"REG_d"), %%xmm3        \n\t"
00483         "paddw %%xmm7, %%xmm1                        \n\t"
00484         "paddw %%xmm7, %%xmm3                        \n\t"
00485         "pavgw %%xmm1, %%xmm0                        \n\t"
00486         "pavgw %%xmm3, %%xmm2                        \n\t"
00487         "movq 16(%2,%%"REG_d"), %%xmm1        \n\t"
00488         "movq 24(%2,%%"REG_d"), %%xmm3        \n\t"
00489         "paddw %%xmm7, %%xmm1                        \n\t"
00490         "paddw %%xmm7, %%xmm3                        \n\t"
00491         "pavgw %%xmm1, %%xmm4                        \n\t"
00492         "pavgw %%xmm3, %%xmm6                        \n\t"
00493         snow_vertical_compose_r2r_sub("xmm7","xmm7","xmm7","xmm7","xmm0","xmm2","xmm4","xmm6")
00494         snow_vertical_compose_sra("1","xmm0","xmm2","xmm4","xmm6")
00495         snow_vertical_compose_sse2_add("%3","xmm0","xmm2","xmm4","xmm6")
00496 
00497         snow_vertical_compose_sra("2","xmm0","xmm2","xmm4","xmm6")
00498         snow_vertical_compose_sse2_add("%3","xmm0","xmm2","xmm4","xmm6")
00499         snow_vertical_compose_sse2_store("%3","xmm0","xmm2","xmm4","xmm6")
00500         snow_vertical_compose_sse2_add("%1","xmm0","xmm2","xmm4","xmm6")
00501         snow_vertical_compose_sse2_move("xmm0","xmm2","xmm4","xmm6","xmm1","xmm3","xmm5","xmm7")
00502         snow_vertical_compose_sra("1","xmm0","xmm2","xmm4","xmm6")
00503         snow_vertical_compose_r2r_add("xmm1","xmm3","xmm5","xmm7","xmm0","xmm2","xmm4","xmm6")
00504         snow_vertical_compose_sse2_add("%2","xmm0","xmm2","xmm4","xmm6")
00505         snow_vertical_compose_sse2_store("%2","xmm0","xmm2","xmm4","xmm6")
00506 
00507         "2:                                          \n\t"
00508         "sub $64, %%"REG_d"                          \n\t"
00509         "jge 1b                                      \n\t"
00510         :"+d"(i)
00511         :"r"(b0),"r"(b1),"r"(b2),"r"(b3),"r"(b4),"r"(b5));
00512 }
00513 
00514 #define snow_vertical_compose_mmx_load_add(op,r,t0,t1,t2,t3)\
00515         ""op" ("r",%%"REG_d"), %%"t0"   \n\t"\
00516         ""op" 8("r",%%"REG_d"), %%"t1"  \n\t"\
00517         ""op" 16("r",%%"REG_d"), %%"t2" \n\t"\
00518         ""op" 24("r",%%"REG_d"), %%"t3" \n\t"
00519 
00520 #define snow_vertical_compose_mmx_load(r,t0,t1,t2,t3)\
00521         snow_vertical_compose_mmx_load_add("movq",r,t0,t1,t2,t3)
00522 
00523 #define snow_vertical_compose_mmx_add(r,t0,t1,t2,t3)\
00524         snow_vertical_compose_mmx_load_add("paddw",r,t0,t1,t2,t3)
00525 
00526 #define snow_vertical_compose_mmx_store(w,s0,s1,s2,s3)\
00527         "movq %%"s0", ("w",%%"REG_d")   \n\t"\
00528         "movq %%"s1", 8("w",%%"REG_d")  \n\t"\
00529         "movq %%"s2", 16("w",%%"REG_d") \n\t"\
00530         "movq %%"s3", 24("w",%%"REG_d") \n\t"
00531 
00532 #define snow_vertical_compose_mmx_move(s0,s1,s2,s3,t0,t1,t2,t3)\
00533         "movq %%"s0", %%"t0" \n\t"\
00534         "movq %%"s1", %%"t1" \n\t"\
00535         "movq %%"s2", %%"t2" \n\t"\
00536         "movq %%"s3", %%"t3" \n\t"
00537 
00538 
00539 static void ff_snow_vertical_compose97i_mmx(IDWTELEM *b0, IDWTELEM *b1, IDWTELEM *b2, IDWTELEM *b3, IDWTELEM *b4, IDWTELEM *b5, int width){
00540     x86_reg i = width;
00541     while(i & 15)
00542     {
00543         i--;
00544         b4[i] -= (W_DM*(b3[i] + b5[i])+W_DO)>>W_DS;
00545         b3[i] -= (W_CM*(b2[i] + b4[i])+W_CO)>>W_CS;
00546         b2[i] += (W_BM*(b1[i] + b3[i])+4*b2[i]+W_BO)>>W_BS;
00547         b1[i] += (W_AM*(b0[i] + b2[i])+W_AO)>>W_AS;
00548     }
00549     i+=i;
00550     __asm__ volatile(
00551         "jmp 2f                                      \n\t"
00552         "1:                                          \n\t"
00553 
00554         snow_vertical_compose_mmx_load("%4","mm1","mm3","mm5","mm7")
00555         snow_vertical_compose_mmx_add("%6","mm1","mm3","mm5","mm7")
00556         "pcmpeqw    %%mm0, %%mm0                     \n\t"
00557         "pcmpeqw    %%mm2, %%mm2                     \n\t"
00558         "paddw      %%mm2, %%mm2                     \n\t"
00559         "paddw      %%mm0, %%mm2                     \n\t"
00560         "psllw        $13, %%mm2                     \n\t"
00561         snow_vertical_compose_r2r_add("mm0","mm0","mm0","mm0","mm1","mm3","mm5","mm7")
00562         snow_vertical_compose_r2r_pmulhw("mm2","mm2","mm2","mm2","mm1","mm3","mm5","mm7")
00563         snow_vertical_compose_mmx_add("%5","mm1","mm3","mm5","mm7")
00564         snow_vertical_compose_mmx_store("%5","mm1","mm3","mm5","mm7")
00565         snow_vertical_compose_mmx_load("%4","mm0","mm2","mm4","mm6")
00566         snow_vertical_compose_mmx_add("%3","mm1","mm3","mm5","mm7")
00567         snow_vertical_compose_r2r_sub("mm1","mm3","mm5","mm7","mm0","mm2","mm4","mm6")
00568         snow_vertical_compose_mmx_store("%4","mm0","mm2","mm4","mm6")
00569         "pcmpeqw %%mm7, %%mm7                        \n\t"
00570         "pcmpeqw %%mm5, %%mm5                        \n\t"
00571         "psllw $15, %%mm7                            \n\t"
00572         "psrlw $13, %%mm5                            \n\t"
00573         "paddw %%mm7, %%mm5                          \n\t"
00574         snow_vertical_compose_r2r_add("mm5","mm5","mm5","mm5","mm0","mm2","mm4","mm6")
00575         "movq   (%2,%%"REG_d"), %%mm1         \n\t"
00576         "movq  8(%2,%%"REG_d"), %%mm3         \n\t"
00577         "paddw %%mm7, %%mm1                          \n\t"
00578         "paddw %%mm7, %%mm3                          \n\t"
00579         "pavgw %%mm1, %%mm0                          \n\t"
00580         "pavgw %%mm3, %%mm2                          \n\t"
00581         "movq 16(%2,%%"REG_d"), %%mm1         \n\t"
00582         "movq 24(%2,%%"REG_d"), %%mm3         \n\t"
00583         "paddw %%mm7, %%mm1                          \n\t"
00584         "paddw %%mm7, %%mm3                          \n\t"
00585         "pavgw %%mm1, %%mm4                          \n\t"
00586         "pavgw %%mm3, %%mm6                          \n\t"
00587         snow_vertical_compose_r2r_sub("mm7","mm7","mm7","mm7","mm0","mm2","mm4","mm6")
00588         snow_vertical_compose_sra("1","mm0","mm2","mm4","mm6")
00589         snow_vertical_compose_mmx_add("%3","mm0","mm2","mm4","mm6")
00590 
00591         snow_vertical_compose_sra("2","mm0","mm2","mm4","mm6")
00592         snow_vertical_compose_mmx_add("%3","mm0","mm2","mm4","mm6")
00593         snow_vertical_compose_mmx_store("%3","mm0","mm2","mm4","mm6")
00594         snow_vertical_compose_mmx_add("%1","mm0","mm2","mm4","mm6")
00595         snow_vertical_compose_mmx_move("mm0","mm2","mm4","mm6","mm1","mm3","mm5","mm7")
00596         snow_vertical_compose_sra("1","mm0","mm2","mm4","mm6")
00597         snow_vertical_compose_r2r_add("mm1","mm3","mm5","mm7","mm0","mm2","mm4","mm6")
00598         snow_vertical_compose_mmx_add("%2","mm0","mm2","mm4","mm6")
00599         snow_vertical_compose_mmx_store("%2","mm0","mm2","mm4","mm6")
00600 
00601         "2:                                          \n\t"
00602         "sub $32, %%"REG_d"                          \n\t"
00603         "jge 1b                                      \n\t"
00604         :"+d"(i)
00605         :"r"(b0),"r"(b1),"r"(b2),"r"(b3),"r"(b4),"r"(b5));
00606 }
00607 #endif //HAVE_7REGS
00608 
00609 #define snow_inner_add_yblock_sse2_header \
00610     IDWTELEM * * dst_array = sb->line + src_y;\
00611     x86_reg tmp;\
00612     __asm__ volatile(\
00613              "mov  %7, %%"REG_c"             \n\t"\
00614              "mov  %6, %2                    \n\t"\
00615              "mov  %4, %%"REG_S"             \n\t"\
00616              "pxor %%xmm7, %%xmm7            \n\t" /* 0 */\
00617              "pcmpeqd %%xmm3, %%xmm3         \n\t"\
00618              "psllw $15, %%xmm3              \n\t"\
00619              "psrlw $12, %%xmm3              \n\t" /* FRAC_BITS >> 1 */\
00620              "1:                             \n\t"\
00621              "mov %1, %%"REG_D"              \n\t"\
00622              "mov (%%"REG_D"), %%"REG_D"     \n\t"\
00623              "add %3, %%"REG_D"              \n\t"
00624 
00625 #define snow_inner_add_yblock_sse2_start_8(out_reg1, out_reg2, ptr_offset, s_offset)\
00626              "mov "PTR_SIZE"*"ptr_offset"(%%"REG_a"), %%"REG_d"; \n\t"\
00627              "movq (%%"REG_d"), %%"out_reg1" \n\t"\
00628              "movq (%%"REG_d", %%"REG_c"), %%"out_reg2" \n\t"\
00629              "punpcklbw %%xmm7, %%"out_reg1" \n\t"\
00630              "punpcklbw %%xmm7, %%"out_reg2" \n\t"\
00631              "movq "s_offset"(%%"REG_S"), %%xmm0 \n\t"\
00632              "movq "s_offset"+16(%%"REG_S"), %%xmm4 \n\t"\
00633              "punpcklbw %%xmm7, %%xmm0       \n\t"\
00634              "punpcklbw %%xmm7, %%xmm4       \n\t"\
00635              "pmullw %%xmm0, %%"out_reg1"    \n\t"\
00636              "pmullw %%xmm4, %%"out_reg2"    \n\t"
00637 
00638 #define snow_inner_add_yblock_sse2_start_16(out_reg1, out_reg2, ptr_offset, s_offset)\
00639              "mov "PTR_SIZE"*"ptr_offset"(%%"REG_a"), %%"REG_d"; \n\t"\
00640              "movq (%%"REG_d"), %%"out_reg1" \n\t"\
00641              "movq 8(%%"REG_d"), %%"out_reg2" \n\t"\
00642              "punpcklbw %%xmm7, %%"out_reg1" \n\t"\
00643              "punpcklbw %%xmm7, %%"out_reg2" \n\t"\
00644              "movq "s_offset"(%%"REG_S"), %%xmm0 \n\t"\
00645              "movq "s_offset"+8(%%"REG_S"), %%xmm4 \n\t"\
00646              "punpcklbw %%xmm7, %%xmm0       \n\t"\
00647              "punpcklbw %%xmm7, %%xmm4       \n\t"\
00648              "pmullw %%xmm0, %%"out_reg1"    \n\t"\
00649              "pmullw %%xmm4, %%"out_reg2"    \n\t"
00650 
00651 #define snow_inner_add_yblock_sse2_accum_8(ptr_offset, s_offset) \
00652              snow_inner_add_yblock_sse2_start_8("xmm2", "xmm6", ptr_offset, s_offset)\
00653              "paddusw %%xmm2, %%xmm1         \n\t"\
00654              "paddusw %%xmm6, %%xmm5         \n\t"
00655 
00656 #define snow_inner_add_yblock_sse2_accum_16(ptr_offset, s_offset) \
00657              snow_inner_add_yblock_sse2_start_16("xmm2", "xmm6", ptr_offset, s_offset)\
00658              "paddusw %%xmm2, %%xmm1         \n\t"\
00659              "paddusw %%xmm6, %%xmm5         \n\t"
00660 
00661 #define snow_inner_add_yblock_sse2_end_common1\
00662              "add $32, %%"REG_S"             \n\t"\
00663              "add %%"REG_c", %0              \n\t"\
00664              "add %%"REG_c", "PTR_SIZE"*3(%%"REG_a");\n\t"\
00665              "add %%"REG_c", "PTR_SIZE"*2(%%"REG_a");\n\t"\
00666              "add %%"REG_c", "PTR_SIZE"*1(%%"REG_a");\n\t"\
00667              "add %%"REG_c", (%%"REG_a")     \n\t"
00668 
00669 #define snow_inner_add_yblock_sse2_end_common2\
00670              "jnz 1b                         \n\t"\
00671              :"+m"(dst8),"+m"(dst_array),"=&r"(tmp)\
00672              :\
00673              "rm"((x86_reg)(src_x<<1)),"m"(obmc),"a"(block),"m"(b_h),"m"(src_stride):\
00674              "%"REG_c"","%"REG_S"","%"REG_D"","%"REG_d"");
00675 
00676 #define snow_inner_add_yblock_sse2_end_8\
00677              "sal $1, %%"REG_c"              \n\t"\
00678              "add"OPSIZE" $"PTR_SIZE"*2, %1  \n\t"\
00679              snow_inner_add_yblock_sse2_end_common1\
00680              "sar $1, %%"REG_c"              \n\t"\
00681              "sub $2, %2                     \n\t"\
00682              snow_inner_add_yblock_sse2_end_common2
00683 
00684 #define snow_inner_add_yblock_sse2_end_16\
00685              "add"OPSIZE" $"PTR_SIZE"*1, %1  \n\t"\
00686              snow_inner_add_yblock_sse2_end_common1\
00687              "dec %2                         \n\t"\
00688              snow_inner_add_yblock_sse2_end_common2
00689 
00690 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,
00691                       int src_x, int src_y, x86_reg src_stride, slice_buffer * sb, int add, uint8_t * dst8){
00692 snow_inner_add_yblock_sse2_header
00693 snow_inner_add_yblock_sse2_start_8("xmm1", "xmm5", "3", "0")
00694 snow_inner_add_yblock_sse2_accum_8("2", "8")
00695 snow_inner_add_yblock_sse2_accum_8("1", "128")
00696 snow_inner_add_yblock_sse2_accum_8("0", "136")
00697 
00698              "mov %0, %%"REG_d"              \n\t"
00699              "movdqa (%%"REG_D"), %%xmm0     \n\t"
00700              "movdqa %%xmm1, %%xmm2          \n\t"
00701 
00702              "punpckhwd %%xmm7, %%xmm1       \n\t"
00703              "punpcklwd %%xmm7, %%xmm2       \n\t"
00704              "paddd %%xmm2, %%xmm0           \n\t"
00705              "movdqa 16(%%"REG_D"), %%xmm2   \n\t"
00706              "paddd %%xmm1, %%xmm2           \n\t"
00707              "paddd %%xmm3, %%xmm0           \n\t"
00708              "paddd %%xmm3, %%xmm2           \n\t"
00709 
00710              "mov %1, %%"REG_D"              \n\t"
00711              "mov "PTR_SIZE"(%%"REG_D"), %%"REG_D";\n\t"
00712              "add %3, %%"REG_D"              \n\t"
00713 
00714              "movdqa (%%"REG_D"), %%xmm4     \n\t"
00715              "movdqa %%xmm5, %%xmm6          \n\t"
00716              "punpckhwd %%xmm7, %%xmm5       \n\t"
00717              "punpcklwd %%xmm7, %%xmm6       \n\t"
00718              "paddd %%xmm6, %%xmm4           \n\t"
00719              "movdqa 16(%%"REG_D"), %%xmm6   \n\t"
00720              "paddd %%xmm5, %%xmm6           \n\t"
00721              "paddd %%xmm3, %%xmm4           \n\t"
00722              "paddd %%xmm3, %%xmm6           \n\t"
00723 
00724              "psrad $8, %%xmm0               \n\t" /* FRAC_BITS. */
00725              "psrad $8, %%xmm2               \n\t" /* FRAC_BITS. */
00726              "packssdw %%xmm2, %%xmm0        \n\t"
00727              "packuswb %%xmm7, %%xmm0        \n\t"
00728              "movq %%xmm0, (%%"REG_d")       \n\t"
00729 
00730              "psrad $8, %%xmm4               \n\t" /* FRAC_BITS. */
00731              "psrad $8, %%xmm6               \n\t" /* FRAC_BITS. */
00732              "packssdw %%xmm6, %%xmm4        \n\t"
00733              "packuswb %%xmm7, %%xmm4        \n\t"
00734              "movq %%xmm4, (%%"REG_d",%%"REG_c");\n\t"
00735 snow_inner_add_yblock_sse2_end_8
00736 }
00737 
00738 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,
00739                       int src_x, int src_y, x86_reg src_stride, slice_buffer * sb, int add, uint8_t * dst8){
00740 snow_inner_add_yblock_sse2_header
00741 snow_inner_add_yblock_sse2_start_16("xmm1", "xmm5", "3", "0")
00742 snow_inner_add_yblock_sse2_accum_16("2", "16")
00743 snow_inner_add_yblock_sse2_accum_16("1", "512")
00744 snow_inner_add_yblock_sse2_accum_16("0", "528")
00745 
00746              "mov %0, %%"REG_d"              \n\t"
00747              "psrlw $4, %%xmm1               \n\t"
00748              "psrlw $4, %%xmm5               \n\t"
00749              "paddw   (%%"REG_D"), %%xmm1    \n\t"
00750              "paddw 16(%%"REG_D"), %%xmm5    \n\t"
00751              "paddw %%xmm3, %%xmm1           \n\t"
00752              "paddw %%xmm3, %%xmm5           \n\t"
00753              "psraw $4, %%xmm1               \n\t" /* FRAC_BITS. */
00754              "psraw $4, %%xmm5               \n\t" /* FRAC_BITS. */
00755              "packuswb %%xmm5, %%xmm1        \n\t"
00756 
00757              "movdqu %%xmm1, (%%"REG_d")       \n\t"
00758 
00759 snow_inner_add_yblock_sse2_end_16
00760 }
00761 
00762 #define snow_inner_add_yblock_mmx_header \
00763     IDWTELEM * * dst_array = sb->line + src_y;\
00764     x86_reg tmp;\
00765     __asm__ volatile(\
00766              "mov  %7, %%"REG_c"             \n\t"\
00767              "mov  %6, %2                    \n\t"\
00768              "mov  %4, %%"REG_S"             \n\t"\
00769              "pxor %%mm7, %%mm7              \n\t" /* 0 */\
00770              "pcmpeqd %%mm3, %%mm3           \n\t"\
00771              "psllw $15, %%mm3               \n\t"\
00772              "psrlw $12, %%mm3               \n\t" /* FRAC_BITS >> 1 */\
00773              "1:                             \n\t"\
00774              "mov %1, %%"REG_D"              \n\t"\
00775              "mov (%%"REG_D"), %%"REG_D"     \n\t"\
00776              "add %3, %%"REG_D"              \n\t"
00777 
00778 #define snow_inner_add_yblock_mmx_start(out_reg1, out_reg2, ptr_offset, s_offset, d_offset)\
00779              "mov "PTR_SIZE"*"ptr_offset"(%%"REG_a"), %%"REG_d"; \n\t"\
00780              "movd "d_offset"(%%"REG_d"), %%"out_reg1" \n\t"\
00781              "movd "d_offset"+4(%%"REG_d"), %%"out_reg2" \n\t"\
00782              "punpcklbw %%mm7, %%"out_reg1" \n\t"\
00783              "punpcklbw %%mm7, %%"out_reg2" \n\t"\
00784              "movd "s_offset"(%%"REG_S"), %%mm0 \n\t"\
00785              "movd "s_offset"+4(%%"REG_S"), %%mm4 \n\t"\
00786              "punpcklbw %%mm7, %%mm0       \n\t"\
00787              "punpcklbw %%mm7, %%mm4       \n\t"\
00788              "pmullw %%mm0, %%"out_reg1"    \n\t"\
00789              "pmullw %%mm4, %%"out_reg2"    \n\t"
00790 
00791 #define snow_inner_add_yblock_mmx_accum(ptr_offset, s_offset, d_offset) \
00792              snow_inner_add_yblock_mmx_start("mm2", "mm6", ptr_offset, s_offset, d_offset)\
00793              "paddusw %%mm2, %%mm1         \n\t"\
00794              "paddusw %%mm6, %%mm5         \n\t"
00795 
00796 #define snow_inner_add_yblock_mmx_mix(read_offset, write_offset)\
00797              "mov %0, %%"REG_d"              \n\t"\
00798              "psrlw $4, %%mm1                \n\t"\
00799              "psrlw $4, %%mm5                \n\t"\
00800              "paddw "read_offset"(%%"REG_D"), %%mm1 \n\t"\
00801              "paddw "read_offset"+8(%%"REG_D"), %%mm5 \n\t"\
00802              "paddw %%mm3, %%mm1             \n\t"\
00803              "paddw %%mm3, %%mm5             \n\t"\
00804              "psraw $4, %%mm1                \n\t"\
00805              "psraw $4, %%mm5                \n\t"\
00806              "packuswb %%mm5, %%mm1          \n\t"\
00807              "movq %%mm1, "write_offset"(%%"REG_d") \n\t"
00808 
00809 #define snow_inner_add_yblock_mmx_end(s_step)\
00810              "add $"s_step", %%"REG_S"             \n\t"\
00811              "add %%"REG_c", "PTR_SIZE"*3(%%"REG_a");\n\t"\
00812              "add %%"REG_c", "PTR_SIZE"*2(%%"REG_a");\n\t"\
00813              "add %%"REG_c", "PTR_SIZE"*1(%%"REG_a");\n\t"\
00814              "add %%"REG_c", (%%"REG_a")     \n\t"\
00815              "add"OPSIZE " $"PTR_SIZE"*1, %1 \n\t"\
00816              "add %%"REG_c", %0              \n\t"\
00817              "dec %2                         \n\t"\
00818              "jnz 1b                         \n\t"\
00819              :"+m"(dst8),"+m"(dst_array),"=&r"(tmp)\
00820              :\
00821              "rm"((x86_reg)(src_x<<1)),"m"(obmc),"a"(block),"m"(b_h),"m"(src_stride):\
00822              "%"REG_c"","%"REG_S"","%"REG_D"","%"REG_d"");
00823 
00824 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,
00825                       int src_x, int src_y, x86_reg src_stride, slice_buffer * sb, int add, uint8_t * dst8){
00826 snow_inner_add_yblock_mmx_header
00827 snow_inner_add_yblock_mmx_start("mm1", "mm5", "3", "0", "0")
00828 snow_inner_add_yblock_mmx_accum("2", "8", "0")
00829 snow_inner_add_yblock_mmx_accum("1", "128", "0")
00830 snow_inner_add_yblock_mmx_accum("0", "136", "0")
00831 snow_inner_add_yblock_mmx_mix("0", "0")
00832 snow_inner_add_yblock_mmx_end("16")
00833 }
00834 
00835 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,
00836                       int src_x, int src_y, x86_reg src_stride, slice_buffer * sb, int add, uint8_t * dst8){
00837 snow_inner_add_yblock_mmx_header
00838 snow_inner_add_yblock_mmx_start("mm1", "mm5", "3", "0", "0")
00839 snow_inner_add_yblock_mmx_accum("2", "16", "0")
00840 snow_inner_add_yblock_mmx_accum("1", "512", "0")
00841 snow_inner_add_yblock_mmx_accum("0", "528", "0")
00842 snow_inner_add_yblock_mmx_mix("0", "0")
00843 
00844 snow_inner_add_yblock_mmx_start("mm1", "mm5", "3", "8", "8")
00845 snow_inner_add_yblock_mmx_accum("2", "24", "8")
00846 snow_inner_add_yblock_mmx_accum("1", "520", "8")
00847 snow_inner_add_yblock_mmx_accum("0", "536", "8")
00848 snow_inner_add_yblock_mmx_mix("16", "8")
00849 snow_inner_add_yblock_mmx_end("32")
00850 }
00851 
00852 static void ff_snow_inner_add_yblock_sse2(const uint8_t *obmc, const int obmc_stride, uint8_t * * block, int b_w, int b_h,
00853                            int src_x, int src_y, int src_stride, slice_buffer * sb, int add, uint8_t * dst8){
00854 
00855     if (b_w == 16)
00856         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);
00857     else if (b_w == 8 && obmc_stride == 16) {
00858         if (!(b_h & 1))
00859             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);
00860         else
00861             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);
00862     } else
00863          ff_snow_inner_add_yblock(obmc, obmc_stride, block, b_w, b_h, src_x,src_y, src_stride, sb, add, dst8);
00864 }
00865 
00866 static void ff_snow_inner_add_yblock_mmx(const uint8_t *obmc, const int obmc_stride, uint8_t * * block, int b_w, int b_h,
00867                           int src_x, int src_y, int src_stride, slice_buffer * sb, int add, uint8_t * dst8){
00868     if (b_w == 16)
00869         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);
00870     else if (b_w == 8 && obmc_stride == 16)
00871         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);
00872     else
00873         ff_snow_inner_add_yblock(obmc, obmc_stride, block, b_w, b_h, src_x,src_y, src_stride, sb, add, dst8);
00874 }
00875 
00876 void ff_dwt_init_x86(DWTContext *c)
00877 {
00878     int mm_flags = av_get_cpu_flags();
00879 
00880     if (mm_flags & AV_CPU_FLAG_MMX) {
00881         if(mm_flags & AV_CPU_FLAG_SSE2 & 0){
00882             c->horizontal_compose97i = ff_snow_horizontal_compose97i_sse2;
00883 #if HAVE_7REGS
00884             c->vertical_compose97i = ff_snow_vertical_compose97i_sse2;
00885 #endif
00886             c->inner_add_yblock = ff_snow_inner_add_yblock_sse2;
00887         }
00888         else{
00889             if(mm_flags & AV_CPU_FLAG_MMX2){
00890             c->horizontal_compose97i = ff_snow_horizontal_compose97i_mmx;
00891 #if HAVE_7REGS
00892             c->vertical_compose97i = ff_snow_vertical_compose97i_mmx;
00893 #endif
00894             }
00895             c->inner_add_yblock = ff_snow_inner_add_yblock_mmx;
00896         }
00897     }
00898 }
libavcodec/x86/snowdsp_mmx.c
