FFmpeg: libavcodec/x86/vc1dsp

00001 /*
00002  * VC-1 and WMV3 - DSP functions MMX-optimized
00003  * Copyright (c) 2007 Christophe GISQUET <christophe.gisquet@free.fr>
00004  *
00005  * Permission is hereby granted, free of charge, to any person
00006  * obtaining a copy of this software and associated documentation
00007  * files (the "Software"), to deal in the Software without
00008  * restriction, including without limitation the rights to use,
00009  * copy, modify, merge, publish, distribute, sublicense, and/or sell
00010  * copies of the Software, and to permit persons to whom the
00011  * Software is furnished to do so, subject to the following
00012  * conditions:
00013  *
00014  * The above copyright notice and this permission notice shall be
00015  * included in all copies or substantial portions of the Software.
00016  *
00017  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
00018  * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES
00019  * OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
00020  * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT
00021  * HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY,
00022  * WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
00023  * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
00024  * OTHER DEALINGS IN THE SOFTWARE.
00025  */
00026 
00027 #include "libavutil/cpu.h"
00028 #include "libavutil/x86_cpu.h"
00029 #include "libavcodec/dsputil.h"
00030 #include "dsputil_mmx.h"
00031 
00032 #define OP_PUT(S,D)
00033 #define OP_AVG(S,D) "pavgb " #S ", " #D " \n\t"
00034 
00036 #define NORMALIZE_MMX(SHIFT)                                    \
00037      "paddw     %%mm7, %%mm3           \n\t" /* +bias-r */      \
00038      "paddw     %%mm7, %%mm4           \n\t" /* +bias-r */      \
00039      "psraw     "SHIFT", %%mm3         \n\t"                    \
00040      "psraw     "SHIFT", %%mm4         \n\t"
00041 
00042 #define TRANSFER_DO_PACK(OP)                    \
00043      "packuswb  %%mm4, %%mm3           \n\t"    \
00044      OP((%2), %%mm3)                            \
00045      "movq      %%mm3, (%2)            \n\t"
00046 
00047 #define TRANSFER_DONT_PACK(OP)                  \
00048      OP(0(%2), %%mm3)                           \
00049      OP(8(%2), %%mm4)                           \
00050      "movq      %%mm3, 0(%2)           \n\t"    \
00051      "movq      %%mm4, 8(%2)           \n\t"
00052 
00054 #define DO_UNPACK(reg)  "punpcklbw %%mm0, " reg "\n\t"
00055 #define DONT_UNPACK(reg)
00056 
00058 #define LOAD_ROUNDER_MMX(ROUND)                 \
00059      "movd      "ROUND", %%mm7         \n\t"    \
00060      "punpcklwd %%mm7, %%mm7           \n\t"    \
00061      "punpckldq %%mm7, %%mm7           \n\t"
00062 
00063 #define SHIFT2_LINE(OFF, R0,R1,R2,R3)           \
00064     "paddw     %%mm"#R2", %%mm"#R1"    \n\t"    \
00065     "movd      (%0,%3), %%mm"#R0"      \n\t"    \
00066     "pmullw    %%mm6, %%mm"#R1"        \n\t"    \
00067     "punpcklbw %%mm0, %%mm"#R0"        \n\t"    \
00068     "movd      (%0,%2), %%mm"#R3"      \n\t"    \
00069     "psubw     %%mm"#R0", %%mm"#R1"    \n\t"    \
00070     "punpcklbw %%mm0, %%mm"#R3"        \n\t"    \
00071     "paddw     %%mm7, %%mm"#R1"        \n\t"    \
00072     "psubw     %%mm"#R3", %%mm"#R1"    \n\t"    \
00073     "psraw     %4, %%mm"#R1"           \n\t"    \
00074     "movq      %%mm"#R1", "#OFF"(%1)   \n\t"    \
00075     "add       %2, %0                  \n\t"
00076 
00078 static void vc1_put_ver_16b_shift2_mmx(int16_t *dst,
00079                                        const uint8_t *src, x86_reg stride,
00080                                        int rnd, int64_t shift)
00081 {
00082     __asm__ volatile(
00083         "mov       $3, %%"REG_c"           \n\t"
00084         LOAD_ROUNDER_MMX("%5")
00085         "movq      "MANGLE(ff_pw_9)", %%mm6 \n\t"
00086         "1:                                \n\t"
00087         "movd      (%0), %%mm2             \n\t"
00088         "add       %2, %0                  \n\t"
00089         "movd      (%0), %%mm3             \n\t"
00090         "punpcklbw %%mm0, %%mm2            \n\t"
00091         "punpcklbw %%mm0, %%mm3            \n\t"
00092         SHIFT2_LINE(  0, 1, 2, 3, 4)
00093         SHIFT2_LINE( 24, 2, 3, 4, 1)
00094         SHIFT2_LINE( 48, 3, 4, 1, 2)
00095         SHIFT2_LINE( 72, 4, 1, 2, 3)
00096         SHIFT2_LINE( 96, 1, 2, 3, 4)
00097         SHIFT2_LINE(120, 2, 3, 4, 1)
00098         SHIFT2_LINE(144, 3, 4, 1, 2)
00099         SHIFT2_LINE(168, 4, 1, 2, 3)
00100         "sub       %6, %0                  \n\t"
00101         "add       $8, %1                  \n\t"
00102         "dec       %%"REG_c"               \n\t"
00103         "jnz 1b                            \n\t"
00104         : "+r"(src), "+r"(dst)
00105         : "r"(stride), "r"(-2*stride),
00106           "m"(shift), "m"(rnd), "r"(9*stride-4)
00107         : "%"REG_c, "memory"
00108     );
00109 }
00110 
00115 #define VC1_HOR_16b_SHIFT2(OP, OPNAME)\
00116 static void OPNAME ## vc1_hor_16b_shift2_mmx(uint8_t *dst, x86_reg stride,\
00117                                              const int16_t *src, int rnd)\
00118 {\
00119     int h = 8;\
00120 \
00121     src -= 1;\
00122     rnd -= (-1+9+9-1)*1024; /* Add -1024 bias */\
00123     __asm__ volatile(\
00124         LOAD_ROUNDER_MMX("%4")\
00125         "movq      "MANGLE(ff_pw_128)", %%mm6\n\t"\
00126         "movq      "MANGLE(ff_pw_9)", %%mm5 \n\t"\
00127         "1:                                \n\t"\
00128         "movq      2*0+0(%1), %%mm1        \n\t"\
00129         "movq      2*0+8(%1), %%mm2        \n\t"\
00130         "movq      2*1+0(%1), %%mm3        \n\t"\
00131         "movq      2*1+8(%1), %%mm4        \n\t"\
00132         "paddw     2*3+0(%1), %%mm1        \n\t"\
00133         "paddw     2*3+8(%1), %%mm2        \n\t"\
00134         "paddw     2*2+0(%1), %%mm3        \n\t"\
00135         "paddw     2*2+8(%1), %%mm4        \n\t"\
00136         "pmullw    %%mm5, %%mm3            \n\t"\
00137         "pmullw    %%mm5, %%mm4            \n\t"\
00138         "psubw     %%mm1, %%mm3            \n\t"\
00139         "psubw     %%mm2, %%mm4            \n\t"\
00140         NORMALIZE_MMX("$7")\
00141         /* Remove bias */\
00142         "paddw     %%mm6, %%mm3            \n\t"\
00143         "paddw     %%mm6, %%mm4            \n\t"\
00144         TRANSFER_DO_PACK(OP)\
00145         "add       $24, %1                 \n\t"\
00146         "add       %3, %2                  \n\t"\
00147         "decl      %0                      \n\t"\
00148         "jnz 1b                            \n\t"\
00149         : "+r"(h), "+r" (src),  "+r" (dst)\
00150         : "r"(stride), "m"(rnd)\
00151         : "memory"\
00152     );\
00153 }
00154 
00155 VC1_HOR_16b_SHIFT2(OP_PUT, put_)
00156 VC1_HOR_16b_SHIFT2(OP_AVG, avg_)
00157 
00158 
00163 #define VC1_SHIFT2(OP, OPNAME)\
00164 static void OPNAME ## vc1_shift2_mmx(uint8_t *dst, const uint8_t *src,\
00165                                      x86_reg stride, int rnd, x86_reg offset)\
00166 {\
00167     rnd = 8-rnd;\
00168     __asm__ volatile(\
00169         "mov       $8, %%"REG_c"           \n\t"\
00170         LOAD_ROUNDER_MMX("%5")\
00171         "movq      "MANGLE(ff_pw_9)", %%mm6\n\t"\
00172         "1:                                \n\t"\
00173         "movd      0(%0   ), %%mm3         \n\t"\
00174         "movd      4(%0   ), %%mm4         \n\t"\
00175         "movd      0(%0,%2), %%mm1         \n\t"\
00176         "movd      4(%0,%2), %%mm2         \n\t"\
00177         "add       %2, %0                  \n\t"\
00178         "punpcklbw %%mm0, %%mm3            \n\t"\
00179         "punpcklbw %%mm0, %%mm4            \n\t"\
00180         "punpcklbw %%mm0, %%mm1            \n\t"\
00181         "punpcklbw %%mm0, %%mm2            \n\t"\
00182         "paddw     %%mm1, %%mm3            \n\t"\
00183         "paddw     %%mm2, %%mm4            \n\t"\
00184         "movd      0(%0,%3), %%mm1         \n\t"\
00185         "movd      4(%0,%3), %%mm2         \n\t"\
00186         "pmullw    %%mm6, %%mm3            \n\t" /* 0,9,9,0*/\
00187         "pmullw    %%mm6, %%mm4            \n\t" /* 0,9,9,0*/\
00188         "punpcklbw %%mm0, %%mm1            \n\t"\
00189         "punpcklbw %%mm0, %%mm2            \n\t"\
00190         "psubw     %%mm1, %%mm3            \n\t" /*-1,9,9,0*/\
00191         "psubw     %%mm2, %%mm4            \n\t" /*-1,9,9,0*/\
00192         "movd      0(%0,%2), %%mm1         \n\t"\
00193         "movd      4(%0,%2), %%mm2         \n\t"\
00194         "punpcklbw %%mm0, %%mm1            \n\t"\
00195         "punpcklbw %%mm0, %%mm2            \n\t"\
00196         "psubw     %%mm1, %%mm3            \n\t" /*-1,9,9,-1*/\
00197         "psubw     %%mm2, %%mm4            \n\t" /*-1,9,9,-1*/\
00198         NORMALIZE_MMX("$4")\
00199         "packuswb  %%mm4, %%mm3            \n\t"\
00200         OP((%1), %%mm3)\
00201         "movq      %%mm3, (%1)             \n\t"\
00202         "add       %6, %0                  \n\t"\
00203         "add       %4, %1                  \n\t"\
00204         "dec       %%"REG_c"               \n\t"\
00205         "jnz 1b                            \n\t"\
00206         : "+r"(src),  "+r"(dst)\
00207         : "r"(offset), "r"(-2*offset), "g"(stride), "m"(rnd),\
00208           "g"(stride-offset)\
00209         : "%"REG_c, "memory"\
00210     );\
00211 }
00212 
00213 VC1_SHIFT2(OP_PUT, put_)
00214 VC1_SHIFT2(OP_AVG, avg_)
00215 
00226 #define MSPEL_FILTER13_CORE(UNPACK, MOVQ, A1, A2, A3, A4)       \
00227      MOVQ "*0+"A1", %%mm1       \n\t"                           \
00228      MOVQ "*4+"A1", %%mm2       \n\t"                           \
00229      UNPACK("%%mm1")                                            \
00230      UNPACK("%%mm2")                                            \
00231      "pmullw    "MANGLE(ff_pw_3)", %%mm1\n\t"                   \
00232      "pmullw    "MANGLE(ff_pw_3)", %%mm2\n\t"                   \
00233      MOVQ "*0+"A2", %%mm3       \n\t"                           \
00234      MOVQ "*4+"A2", %%mm4       \n\t"                           \
00235      UNPACK("%%mm3")                                            \
00236      UNPACK("%%mm4")                                            \
00237      "pmullw    %%mm6, %%mm3    \n\t" /* *18 */                 \
00238      "pmullw    %%mm6, %%mm4    \n\t" /* *18 */                 \
00239      "psubw     %%mm1, %%mm3    \n\t" /* 18,-3 */               \
00240      "psubw     %%mm2, %%mm4    \n\t" /* 18,-3 */               \
00241      MOVQ "*0+"A4", %%mm1       \n\t"                           \
00242      MOVQ "*4+"A4", %%mm2       \n\t"                           \
00243      UNPACK("%%mm1")                                            \
00244      UNPACK("%%mm2")                                            \
00245      "psllw     $2, %%mm1       \n\t" /* 4* */                  \
00246      "psllw     $2, %%mm2       \n\t" /* 4* */                  \
00247      "psubw     %%mm1, %%mm3    \n\t" /* -4,18,-3 */            \
00248      "psubw     %%mm2, %%mm4    \n\t" /* -4,18,-3 */            \
00249      MOVQ "*0+"A3", %%mm1       \n\t"                           \
00250      MOVQ "*4+"A3", %%mm2       \n\t"                           \
00251      UNPACK("%%mm1")                                            \
00252      UNPACK("%%mm2")                                            \
00253      "pmullw    %%mm5, %%mm1    \n\t" /* *53 */                 \
00254      "pmullw    %%mm5, %%mm2    \n\t" /* *53 */                 \
00255      "paddw     %%mm1, %%mm3    \n\t" /* 4,53,18,-3 */          \
00256      "paddw     %%mm2, %%mm4    \n\t" /* 4,53,18,-3 */
00257 
00266 #define MSPEL_FILTER13_VER_16B(NAME, A1, A2, A3, A4)                    \
00267 static void                                                             \
00268 vc1_put_ver_16b_ ## NAME ## _mmx(int16_t *dst, const uint8_t *src,      \
00269                                  x86_reg src_stride,                   \
00270                                  int rnd, int64_t shift)                \
00271 {                                                                       \
00272     int h = 8;                                                          \
00273     src -= src_stride;                                                  \
00274     __asm__ volatile(                                                       \
00275         LOAD_ROUNDER_MMX("%5")                                          \
00276         "movq      "MANGLE(ff_pw_53)", %%mm5\n\t"                       \
00277         "movq      "MANGLE(ff_pw_18)", %%mm6\n\t"                       \
00278         ASMALIGN(3)                                                     \
00279         "1:                        \n\t"                                \
00280         MSPEL_FILTER13_CORE(DO_UNPACK, "movd  1", A1, A2, A3, A4)       \
00281         NORMALIZE_MMX("%6")                                             \
00282         TRANSFER_DONT_PACK(OP_PUT)                                      \
00283         /* Last 3 (in fact 4) bytes on the line */                      \
00284         "movd      8+"A1", %%mm1   \n\t"                                \
00285         DO_UNPACK("%%mm1")                                              \
00286         "movq      %%mm1, %%mm3    \n\t"                                \
00287         "paddw     %%mm1, %%mm1    \n\t"                                \
00288         "paddw     %%mm3, %%mm1    \n\t" /* 3* */                       \
00289         "movd      8+"A2", %%mm3   \n\t"                                \
00290         DO_UNPACK("%%mm3")                                              \
00291         "pmullw    %%mm6, %%mm3    \n\t" /* *18 */                      \
00292         "psubw     %%mm1, %%mm3    \n\t" /*18,-3 */                     \
00293         "movd      8+"A3", %%mm1   \n\t"                                \
00294         DO_UNPACK("%%mm1")                                              \
00295         "pmullw    %%mm5, %%mm1    \n\t" /* *53 */                      \
00296         "paddw     %%mm1, %%mm3    \n\t" /*53,18,-3 */                  \
00297         "movd      8+"A4", %%mm1   \n\t"                                \
00298         DO_UNPACK("%%mm1")                                              \
00299         "psllw     $2, %%mm1       \n\t" /* 4* */                       \
00300         "psubw     %%mm1, %%mm3    \n\t"                                \
00301         "paddw     %%mm7, %%mm3    \n\t"                                \
00302         "psraw     %6, %%mm3       \n\t"                                \
00303         "movq      %%mm3, 16(%2)   \n\t"                                \
00304         "add       %3, %1          \n\t"                                \
00305         "add       $24, %2         \n\t"                                \
00306         "decl      %0              \n\t"                                \
00307         "jnz 1b                    \n\t"                                \
00308         : "+r"(h), "+r" (src),  "+r" (dst)                              \
00309         : "r"(src_stride), "r"(3*src_stride),                           \
00310           "m"(rnd), "m"(shift)                                          \
00311         : "memory"                                                      \
00312     );                                                                  \
00313 }
00314 
00322 #define MSPEL_FILTER13_HOR_16B(NAME, A1, A2, A3, A4, OP, OPNAME)        \
00323 static void                                                             \
00324 OPNAME ## vc1_hor_16b_ ## NAME ## _mmx(uint8_t *dst, x86_reg stride,    \
00325                                  const int16_t *src, int rnd)           \
00326 {                                                                       \
00327     int h = 8;                                                          \
00328     src -= 1;                                                           \
00329     rnd -= (-4+58+13-3)*256; /* Add -256 bias */                        \
00330     __asm__ volatile(                                                       \
00331         LOAD_ROUNDER_MMX("%4")                                          \
00332         "movq      "MANGLE(ff_pw_18)", %%mm6   \n\t"                    \
00333         "movq      "MANGLE(ff_pw_53)", %%mm5   \n\t"                    \
00334         ASMALIGN(3)                                                     \
00335         "1:                        \n\t"                                \
00336         MSPEL_FILTER13_CORE(DONT_UNPACK, "movq 2", A1, A2, A3, A4)      \
00337         NORMALIZE_MMX("$7")                                             \
00338         /* Remove bias */                                               \
00339         "paddw     "MANGLE(ff_pw_128)", %%mm3  \n\t"                    \
00340         "paddw     "MANGLE(ff_pw_128)", %%mm4  \n\t"                    \
00341         TRANSFER_DO_PACK(OP)                                            \
00342         "add       $24, %1         \n\t"                                \
00343         "add       %3, %2          \n\t"                                \
00344         "decl      %0              \n\t"                                \
00345         "jnz 1b                    \n\t"                                \
00346         : "+r"(h), "+r" (src),  "+r" (dst)                              \
00347         : "r"(stride), "m"(rnd)                                         \
00348         : "memory"                                                      \
00349     );                                                                  \
00350 }
00351 
00360 #define MSPEL_FILTER13_8B(NAME, A1, A2, A3, A4, OP, OPNAME)             \
00361 static void                                                             \
00362 OPNAME ## vc1_## NAME ## _mmx(uint8_t *dst, const uint8_t *src,         \
00363                         x86_reg stride, int rnd, x86_reg offset)      \
00364 {                                                                       \
00365     int h = 8;                                                          \
00366     src -= offset;                                                      \
00367     rnd = 32-rnd;                                                       \
00368     __asm__ volatile (                                                      \
00369         LOAD_ROUNDER_MMX("%6")                                          \
00370         "movq      "MANGLE(ff_pw_53)", %%mm5       \n\t"                \
00371         "movq      "MANGLE(ff_pw_18)", %%mm6       \n\t"                \
00372         ASMALIGN(3)                                                     \
00373         "1:                        \n\t"                                \
00374         MSPEL_FILTER13_CORE(DO_UNPACK, "movd   1", A1, A2, A3, A4)      \
00375         NORMALIZE_MMX("$6")                                             \
00376         TRANSFER_DO_PACK(OP)                                            \
00377         "add       %5, %1          \n\t"                                \
00378         "add       %5, %2          \n\t"                                \
00379         "decl      %0              \n\t"                                \
00380         "jnz 1b                    \n\t"                                \
00381         : "+r"(h), "+r" (src),  "+r" (dst)                              \
00382         : "r"(offset), "r"(3*offset), "g"(stride), "m"(rnd)             \
00383         : "memory"                                                      \
00384     );                                                                  \
00385 }
00386 
00388 MSPEL_FILTER13_8B     (shift1, "0(%1,%4  )", "0(%1,%3,2)", "0(%1,%3  )", "0(%1     )", OP_PUT, put_)
00389 MSPEL_FILTER13_8B     (shift1, "0(%1,%4  )", "0(%1,%3,2)", "0(%1,%3  )", "0(%1     )", OP_AVG, avg_)
00390 MSPEL_FILTER13_VER_16B(shift1, "0(%1,%4  )", "0(%1,%3,2)", "0(%1,%3  )", "0(%1     )")
00391 MSPEL_FILTER13_HOR_16B(shift1, "2*3(%1)", "2*2(%1)", "2*1(%1)", "2*0(%1)", OP_PUT, put_)
00392 MSPEL_FILTER13_HOR_16B(shift1, "2*3(%1)", "2*2(%1)", "2*1(%1)", "2*0(%1)", OP_AVG, avg_)
00393 
00395 MSPEL_FILTER13_8B     (shift3, "0(%1     )", "0(%1,%3  )", "0(%1,%3,2)", "0(%1,%4  )", OP_PUT, put_)
00396 MSPEL_FILTER13_8B     (shift3, "0(%1     )", "0(%1,%3  )", "0(%1,%3,2)", "0(%1,%4  )", OP_AVG, avg_)
00397 MSPEL_FILTER13_VER_16B(shift3, "0(%1     )", "0(%1,%3  )", "0(%1,%3,2)", "0(%1,%4  )")
00398 MSPEL_FILTER13_HOR_16B(shift3, "2*0(%1)", "2*1(%1)", "2*2(%1)", "2*3(%1)", OP_PUT, put_)
00399 MSPEL_FILTER13_HOR_16B(shift3, "2*0(%1)", "2*1(%1)", "2*2(%1)", "2*3(%1)", OP_AVG, avg_)
00400 
00401 typedef void (*vc1_mspel_mc_filter_ver_16bits)(int16_t *dst, const uint8_t *src, x86_reg src_stride, int rnd, int64_t shift);
00402 typedef void (*vc1_mspel_mc_filter_hor_16bits)(uint8_t *dst, x86_reg dst_stride, const int16_t *src, int rnd);
00403 typedef void (*vc1_mspel_mc_filter_8bits)(uint8_t *dst, const uint8_t *src, x86_reg stride, int rnd, x86_reg offset);
00404 
00416 #define VC1_MSPEL_MC(OP)\
00417 static void OP ## vc1_mspel_mc(uint8_t *dst, const uint8_t *src, int stride,\
00418                                int hmode, int vmode, int rnd)\
00419 {\
00420     static const vc1_mspel_mc_filter_ver_16bits vc1_put_shift_ver_16bits[] =\
00421          { NULL, vc1_put_ver_16b_shift1_mmx, vc1_put_ver_16b_shift2_mmx, vc1_put_ver_16b_shift3_mmx };\
00422     static const vc1_mspel_mc_filter_hor_16bits vc1_put_shift_hor_16bits[] =\
00423          { NULL, OP ## vc1_hor_16b_shift1_mmx, OP ## vc1_hor_16b_shift2_mmx, OP ## vc1_hor_16b_shift3_mmx };\
00424     static const vc1_mspel_mc_filter_8bits vc1_put_shift_8bits[] =\
00425          { NULL, OP ## vc1_shift1_mmx, OP ## vc1_shift2_mmx, OP ## vc1_shift3_mmx };\
00426 \
00427     __asm__ volatile(\
00428         "pxor %%mm0, %%mm0         \n\t"\
00429         ::: "memory"\
00430     );\
00431 \
00432     if (vmode) { /* Vertical filter to apply */\
00433         if (hmode) { /* Horizontal filter to apply, output to tmp */\
00434             static const int shift_value[] = { 0, 5, 1, 5 };\
00435             int              shift = (shift_value[hmode]+shift_value[vmode])>>1;\
00436             int              r;\
00437             DECLARE_ALIGNED(16, int16_t, tmp)[12*8];\
00438 \
00439             r = (1<<(shift-1)) + rnd-1;\
00440             vc1_put_shift_ver_16bits[vmode](tmp, src-1, stride, r, shift);\
00441 \
00442             vc1_put_shift_hor_16bits[hmode](dst, stride, tmp+1, 64-rnd);\
00443             return;\
00444         }\
00445         else { /* No horizontal filter, output 8 lines to dst */\
00446             vc1_put_shift_8bits[vmode](dst, src, stride, 1-rnd, stride);\
00447             return;\
00448         }\
00449     }\
00450 \
00451     /* Horizontal mode with no vertical mode */\
00452     vc1_put_shift_8bits[hmode](dst, src, stride, rnd, 1);\
00453 }
00454 
00455 VC1_MSPEL_MC(put_)
00456 VC1_MSPEL_MC(avg_)
00457 
00459 #define DECLARE_FUNCTION(a, b)                                          \
00460 static void put_vc1_mspel_mc ## a ## b ## _mmx(uint8_t *dst, const uint8_t *src, int stride, int rnd) { \
00461      put_vc1_mspel_mc(dst, src, stride, a, b, rnd);                     \
00462 }\
00463 static void avg_vc1_mspel_mc ## a ## b ## _mmx2(uint8_t *dst, const uint8_t *src, int stride, int rnd) { \
00464      avg_vc1_mspel_mc(dst, src, stride, a, b, rnd);                     \
00465 }
00466 
00467 DECLARE_FUNCTION(0, 1)
00468 DECLARE_FUNCTION(0, 2)
00469 DECLARE_FUNCTION(0, 3)
00470 
00471 DECLARE_FUNCTION(1, 0)
00472 DECLARE_FUNCTION(1, 1)
00473 DECLARE_FUNCTION(1, 2)
00474 DECLARE_FUNCTION(1, 3)
00475 
00476 DECLARE_FUNCTION(2, 0)
00477 DECLARE_FUNCTION(2, 1)
00478 DECLARE_FUNCTION(2, 2)
00479 DECLARE_FUNCTION(2, 3)
00480 
00481 DECLARE_FUNCTION(3, 0)
00482 DECLARE_FUNCTION(3, 1)
00483 DECLARE_FUNCTION(3, 2)
00484 DECLARE_FUNCTION(3, 3)
00485 
00486 static void vc1_inv_trans_4x4_dc_mmx2(uint8_t *dest, int linesize, DCTELEM *block)
00487 {
00488     int dc = block[0];
00489     dc = (17 * dc +  4) >> 3;
00490     dc = (17 * dc + 64) >> 7;
00491     __asm__ volatile(
00492         "movd          %0, %%mm0 \n\t"
00493         "pshufw $0, %%mm0, %%mm0 \n\t"
00494         "pxor       %%mm1, %%mm1 \n\t"
00495         "psubw      %%mm0, %%mm1 \n\t"
00496         "packuswb   %%mm0, %%mm0 \n\t"
00497         "packuswb   %%mm1, %%mm1 \n\t"
00498         ::"r"(dc)
00499     );
00500     __asm__ volatile(
00501         "movd          %0, %%mm2 \n\t"
00502         "movd          %1, %%mm3 \n\t"
00503         "movd          %2, %%mm4 \n\t"
00504         "movd          %3, %%mm5 \n\t"
00505         "paddusb    %%mm0, %%mm2 \n\t"
00506         "paddusb    %%mm0, %%mm3 \n\t"
00507         "paddusb    %%mm0, %%mm4 \n\t"
00508         "paddusb    %%mm0, %%mm5 \n\t"
00509         "psubusb    %%mm1, %%mm2 \n\t"
00510         "psubusb    %%mm1, %%mm3 \n\t"
00511         "psubusb    %%mm1, %%mm4 \n\t"
00512         "psubusb    %%mm1, %%mm5 \n\t"
00513         "movd       %%mm2, %0    \n\t"
00514         "movd       %%mm3, %1    \n\t"
00515         "movd       %%mm4, %2    \n\t"
00516         "movd       %%mm5, %3    \n\t"
00517         :"+m"(*(uint32_t*)(dest+0*linesize)),
00518          "+m"(*(uint32_t*)(dest+1*linesize)),
00519          "+m"(*(uint32_t*)(dest+2*linesize)),
00520          "+m"(*(uint32_t*)(dest+3*linesize))
00521     );
00522 }
00523 
00524 static void vc1_inv_trans_4x8_dc_mmx2(uint8_t *dest, int linesize, DCTELEM *block)
00525 {
00526     int dc = block[0];
00527     dc = (17 * dc +  4) >> 3;
00528     dc = (12 * dc + 64) >> 7;
00529     __asm__ volatile(
00530         "movd          %0, %%mm0 \n\t"
00531         "pshufw $0, %%mm0, %%mm0 \n\t"
00532         "pxor       %%mm1, %%mm1 \n\t"
00533         "psubw      %%mm0, %%mm1 \n\t"
00534         "packuswb   %%mm0, %%mm0 \n\t"
00535         "packuswb   %%mm1, %%mm1 \n\t"
00536         ::"r"(dc)
00537     );
00538     __asm__ volatile(
00539         "movd          %0, %%mm2 \n\t"
00540         "movd          %1, %%mm3 \n\t"
00541         "movd          %2, %%mm4 \n\t"
00542         "movd          %3, %%mm5 \n\t"
00543         "paddusb    %%mm0, %%mm2 \n\t"
00544         "paddusb    %%mm0, %%mm3 \n\t"
00545         "paddusb    %%mm0, %%mm4 \n\t"
00546         "paddusb    %%mm0, %%mm5 \n\t"
00547         "psubusb    %%mm1, %%mm2 \n\t"
00548         "psubusb    %%mm1, %%mm3 \n\t"
00549         "psubusb    %%mm1, %%mm4 \n\t"
00550         "psubusb    %%mm1, %%mm5 \n\t"
00551         "movd       %%mm2, %0    \n\t"
00552         "movd       %%mm3, %1    \n\t"
00553         "movd       %%mm4, %2    \n\t"
00554         "movd       %%mm5, %3    \n\t"
00555         :"+m"(*(uint32_t*)(dest+0*linesize)),
00556          "+m"(*(uint32_t*)(dest+1*linesize)),
00557          "+m"(*(uint32_t*)(dest+2*linesize)),
00558          "+m"(*(uint32_t*)(dest+3*linesize))
00559     );
00560     dest += 4*linesize;
00561     __asm__ volatile(
00562         "movd          %0, %%mm2 \n\t"
00563         "movd          %1, %%mm3 \n\t"
00564         "movd          %2, %%mm4 \n\t"
00565         "movd          %3, %%mm5 \n\t"
00566         "paddusb    %%mm0, %%mm2 \n\t"
00567         "paddusb    %%mm0, %%mm3 \n\t"
00568         "paddusb    %%mm0, %%mm4 \n\t"
00569         "paddusb    %%mm0, %%mm5 \n\t"
00570         "psubusb    %%mm1, %%mm2 \n\t"
00571         "psubusb    %%mm1, %%mm3 \n\t"
00572         "psubusb    %%mm1, %%mm4 \n\t"
00573         "psubusb    %%mm1, %%mm5 \n\t"
00574         "movd       %%mm2, %0    \n\t"
00575         "movd       %%mm3, %1    \n\t"
00576         "movd       %%mm4, %2    \n\t"
00577         "movd       %%mm5, %3    \n\t"
00578         :"+m"(*(uint32_t*)(dest+0*linesize)),
00579          "+m"(*(uint32_t*)(dest+1*linesize)),
00580          "+m"(*(uint32_t*)(dest+2*linesize)),
00581          "+m"(*(uint32_t*)(dest+3*linesize))
00582     );
00583 }
00584 
00585 static void vc1_inv_trans_8x4_dc_mmx2(uint8_t *dest, int linesize, DCTELEM *block)
00586 {
00587     int dc = block[0];
00588     dc = ( 3 * dc +  1) >> 1;
00589     dc = (17 * dc + 64) >> 7;
00590     __asm__ volatile(
00591         "movd          %0, %%mm0 \n\t"
00592         "pshufw $0, %%mm0, %%mm0 \n\t"
00593         "pxor       %%mm1, %%mm1 \n\t"
00594         "psubw      %%mm0, %%mm1 \n\t"
00595         "packuswb   %%mm0, %%mm0 \n\t"
00596         "packuswb   %%mm1, %%mm1 \n\t"
00597         ::"r"(dc)
00598     );
00599     __asm__ volatile(
00600         "movq          %0, %%mm2 \n\t"
00601         "movq          %1, %%mm3 \n\t"
00602         "movq          %2, %%mm4 \n\t"
00603         "movq          %3, %%mm5 \n\t"
00604         "paddusb    %%mm0, %%mm2 \n\t"
00605         "paddusb    %%mm0, %%mm3 \n\t"
00606         "paddusb    %%mm0, %%mm4 \n\t"
00607         "paddusb    %%mm0, %%mm5 \n\t"
00608         "psubusb    %%mm1, %%mm2 \n\t"
00609         "psubusb    %%mm1, %%mm3 \n\t"
00610         "psubusb    %%mm1, %%mm4 \n\t"
00611         "psubusb    %%mm1, %%mm5 \n\t"
00612         "movq       %%mm2, %0    \n\t"
00613         "movq       %%mm3, %1    \n\t"
00614         "movq       %%mm4, %2    \n\t"
00615         "movq       %%mm5, %3    \n\t"
00616         :"+m"(*(uint32_t*)(dest+0*linesize)),
00617          "+m"(*(uint32_t*)(dest+1*linesize)),
00618          "+m"(*(uint32_t*)(dest+2*linesize)),
00619          "+m"(*(uint32_t*)(dest+3*linesize))
00620     );
00621 }
00622 
00623 static void vc1_inv_trans_8x8_dc_mmx2(uint8_t *dest, int linesize, DCTELEM *block)
00624 {
00625     int dc = block[0];
00626     dc = (3 * dc +  1) >> 1;
00627     dc = (3 * dc + 16) >> 5;
00628     __asm__ volatile(
00629         "movd          %0, %%mm0 \n\t"
00630         "pshufw $0, %%mm0, %%mm0 \n\t"
00631         "pxor       %%mm1, %%mm1 \n\t"
00632         "psubw      %%mm0, %%mm1 \n\t"
00633         "packuswb   %%mm0, %%mm0 \n\t"
00634         "packuswb   %%mm1, %%mm1 \n\t"
00635         ::"r"(dc)
00636     );
00637     __asm__ volatile(
00638         "movq          %0, %%mm2 \n\t"
00639         "movq          %1, %%mm3 \n\t"
00640         "movq          %2, %%mm4 \n\t"
00641         "movq          %3, %%mm5 \n\t"
00642         "paddusb    %%mm0, %%mm2 \n\t"
00643         "paddusb    %%mm0, %%mm3 \n\t"
00644         "paddusb    %%mm0, %%mm4 \n\t"
00645         "paddusb    %%mm0, %%mm5 \n\t"
00646         "psubusb    %%mm1, %%mm2 \n\t"
00647         "psubusb    %%mm1, %%mm3 \n\t"
00648         "psubusb    %%mm1, %%mm4 \n\t"
00649         "psubusb    %%mm1, %%mm5 \n\t"
00650         "movq       %%mm2, %0    \n\t"
00651         "movq       %%mm3, %1    \n\t"
00652         "movq       %%mm4, %2    \n\t"
00653         "movq       %%mm5, %3    \n\t"
00654         :"+m"(*(uint32_t*)(dest+0*linesize)),
00655          "+m"(*(uint32_t*)(dest+1*linesize)),
00656          "+m"(*(uint32_t*)(dest+2*linesize)),
00657          "+m"(*(uint32_t*)(dest+3*linesize))
00658     );
00659     dest += 4*linesize;
00660     __asm__ volatile(
00661         "movq          %0, %%mm2 \n\t"
00662         "movq          %1, %%mm3 \n\t"
00663         "movq          %2, %%mm4 \n\t"
00664         "movq          %3, %%mm5 \n\t"
00665         "paddusb    %%mm0, %%mm2 \n\t"
00666         "paddusb    %%mm0, %%mm3 \n\t"
00667         "paddusb    %%mm0, %%mm4 \n\t"
00668         "paddusb    %%mm0, %%mm5 \n\t"
00669         "psubusb    %%mm1, %%mm2 \n\t"
00670         "psubusb    %%mm1, %%mm3 \n\t"
00671         "psubusb    %%mm1, %%mm4 \n\t"
00672         "psubusb    %%mm1, %%mm5 \n\t"
00673         "movq       %%mm2, %0    \n\t"
00674         "movq       %%mm3, %1    \n\t"
00675         "movq       %%mm4, %2    \n\t"
00676         "movq       %%mm5, %3    \n\t"
00677         :"+m"(*(uint32_t*)(dest+0*linesize)),
00678          "+m"(*(uint32_t*)(dest+1*linesize)),
00679          "+m"(*(uint32_t*)(dest+2*linesize)),
00680          "+m"(*(uint32_t*)(dest+3*linesize))
00681     );
00682 }
00683 
00684 #define LOOP_FILTER(EXT) \
00685 void ff_vc1_v_loop_filter4_ ## EXT(uint8_t *src, int stride, int pq); \
00686 void ff_vc1_h_loop_filter4_ ## EXT(uint8_t *src, int stride, int pq); \
00687 void ff_vc1_v_loop_filter8_ ## EXT(uint8_t *src, int stride, int pq); \
00688 void ff_vc1_h_loop_filter8_ ## EXT(uint8_t *src, int stride, int pq); \
00689 \
00690 static void vc1_v_loop_filter16_ ## EXT(uint8_t *src, int stride, int pq) \
00691 { \
00692     ff_vc1_v_loop_filter8_ ## EXT(src,   stride, pq); \
00693     ff_vc1_v_loop_filter8_ ## EXT(src+8, stride, pq); \
00694 } \
00695 \
00696 static void vc1_h_loop_filter16_ ## EXT(uint8_t *src, int stride, int pq) \
00697 { \
00698     ff_vc1_h_loop_filter8_ ## EXT(src,          stride, pq); \
00699     ff_vc1_h_loop_filter8_ ## EXT(src+8*stride, stride, pq); \
00700 }
00701 
00702 #if HAVE_YASM
00703 LOOP_FILTER(mmx)
00704 LOOP_FILTER(mmx2)
00705 LOOP_FILTER(sse2)
00706 LOOP_FILTER(ssse3)
00707 
00708 void ff_vc1_h_loop_filter8_sse4(uint8_t *src, int stride, int pq);
00709 
00710 static void vc1_h_loop_filter16_sse4(uint8_t *src, int stride, int pq)
00711 {
00712     ff_vc1_h_loop_filter8_sse4(src,          stride, pq);
00713     ff_vc1_h_loop_filter8_sse4(src+8*stride, stride, pq);
00714 }
00715 #endif
00716 
00717 void ff_vc1dsp_init_mmx(DSPContext* dsp, AVCodecContext *avctx) {
00718     int mm_flags = av_get_cpu_flags();
00719 
00720     dsp->put_vc1_mspel_pixels_tab[ 0] = ff_put_vc1_mspel_mc00_mmx;
00721     dsp->put_vc1_mspel_pixels_tab[ 4] = put_vc1_mspel_mc01_mmx;
00722     dsp->put_vc1_mspel_pixels_tab[ 8] = put_vc1_mspel_mc02_mmx;
00723     dsp->put_vc1_mspel_pixels_tab[12] = put_vc1_mspel_mc03_mmx;
00724 
00725     dsp->put_vc1_mspel_pixels_tab[ 1] = put_vc1_mspel_mc10_mmx;
00726     dsp->put_vc1_mspel_pixels_tab[ 5] = put_vc1_mspel_mc11_mmx;
00727     dsp->put_vc1_mspel_pixels_tab[ 9] = put_vc1_mspel_mc12_mmx;
00728     dsp->put_vc1_mspel_pixels_tab[13] = put_vc1_mspel_mc13_mmx;
00729 
00730     dsp->put_vc1_mspel_pixels_tab[ 2] = put_vc1_mspel_mc20_mmx;
00731     dsp->put_vc1_mspel_pixels_tab[ 6] = put_vc1_mspel_mc21_mmx;
00732     dsp->put_vc1_mspel_pixels_tab[10] = put_vc1_mspel_mc22_mmx;
00733     dsp->put_vc1_mspel_pixels_tab[14] = put_vc1_mspel_mc23_mmx;
00734 
00735     dsp->put_vc1_mspel_pixels_tab[ 3] = put_vc1_mspel_mc30_mmx;
00736     dsp->put_vc1_mspel_pixels_tab[ 7] = put_vc1_mspel_mc31_mmx;
00737     dsp->put_vc1_mspel_pixels_tab[11] = put_vc1_mspel_mc32_mmx;
00738     dsp->put_vc1_mspel_pixels_tab[15] = put_vc1_mspel_mc33_mmx;
00739 
00740     if (mm_flags & AV_CPU_FLAG_MMX2){
00741         dsp->avg_vc1_mspel_pixels_tab[ 0] = ff_avg_vc1_mspel_mc00_mmx2;
00742         dsp->avg_vc1_mspel_pixels_tab[ 4] = avg_vc1_mspel_mc01_mmx2;
00743         dsp->avg_vc1_mspel_pixels_tab[ 8] = avg_vc1_mspel_mc02_mmx2;
00744         dsp->avg_vc1_mspel_pixels_tab[12] = avg_vc1_mspel_mc03_mmx2;
00745 
00746         dsp->avg_vc1_mspel_pixels_tab[ 1] = avg_vc1_mspel_mc10_mmx2;
00747         dsp->avg_vc1_mspel_pixels_tab[ 5] = avg_vc1_mspel_mc11_mmx2;
00748         dsp->avg_vc1_mspel_pixels_tab[ 9] = avg_vc1_mspel_mc12_mmx2;
00749         dsp->avg_vc1_mspel_pixels_tab[13] = avg_vc1_mspel_mc13_mmx2;
00750 
00751         dsp->avg_vc1_mspel_pixels_tab[ 2] = avg_vc1_mspel_mc20_mmx2;
00752         dsp->avg_vc1_mspel_pixels_tab[ 6] = avg_vc1_mspel_mc21_mmx2;
00753         dsp->avg_vc1_mspel_pixels_tab[10] = avg_vc1_mspel_mc22_mmx2;
00754         dsp->avg_vc1_mspel_pixels_tab[14] = avg_vc1_mspel_mc23_mmx2;
00755 
00756         dsp->avg_vc1_mspel_pixels_tab[ 3] = avg_vc1_mspel_mc30_mmx2;
00757         dsp->avg_vc1_mspel_pixels_tab[ 7] = avg_vc1_mspel_mc31_mmx2;
00758         dsp->avg_vc1_mspel_pixels_tab[11] = avg_vc1_mspel_mc32_mmx2;
00759         dsp->avg_vc1_mspel_pixels_tab[15] = avg_vc1_mspel_mc33_mmx2;
00760 
00761         dsp->vc1_inv_trans_8x8_dc = vc1_inv_trans_8x8_dc_mmx2;
00762         dsp->vc1_inv_trans_4x8_dc = vc1_inv_trans_4x8_dc_mmx2;
00763         dsp->vc1_inv_trans_8x4_dc = vc1_inv_trans_8x4_dc_mmx2;
00764         dsp->vc1_inv_trans_4x4_dc = vc1_inv_trans_4x4_dc_mmx2;
00765     }
00766 
00767 #define ASSIGN_LF(EXT) \
00768         dsp->vc1_v_loop_filter4  = ff_vc1_v_loop_filter4_ ## EXT; \
00769         dsp->vc1_h_loop_filter4  = ff_vc1_h_loop_filter4_ ## EXT; \
00770         dsp->vc1_v_loop_filter8  = ff_vc1_v_loop_filter8_ ## EXT; \
00771         dsp->vc1_h_loop_filter8  = ff_vc1_h_loop_filter8_ ## EXT; \
00772         dsp->vc1_v_loop_filter16 = vc1_v_loop_filter16_ ## EXT; \
00773         dsp->vc1_h_loop_filter16 = vc1_h_loop_filter16_ ## EXT
00774 
00775 #if HAVE_YASM
00776     if (mm_flags & AV_CPU_FLAG_MMX) {
00777         ASSIGN_LF(mmx);
00778     }
00779     return;
00780     if (mm_flags & AV_CPU_FLAG_MMX2) {
00781         ASSIGN_LF(mmx2);
00782     }
00783     if (mm_flags & AV_CPU_FLAG_SSE2) {
00784         dsp->vc1_v_loop_filter8  = ff_vc1_v_loop_filter8_sse2;
00785         dsp->vc1_h_loop_filter8  = ff_vc1_h_loop_filter8_sse2;
00786         dsp->vc1_v_loop_filter16 = vc1_v_loop_filter16_sse2;
00787         dsp->vc1_h_loop_filter16 = vc1_h_loop_filter16_sse2;
00788     }
00789     if (mm_flags & AV_CPU_FLAG_SSSE3) {
00790         ASSIGN_LF(ssse3);
00791     }
00792     if (mm_flags & AV_CPU_FLAG_SSE4) {
00793         dsp->vc1_h_loop_filter8  = ff_vc1_h_loop_filter8_sse4;
00794         dsp->vc1_h_loop_filter16 = vc1_h_loop_filter16_sse4;
00795     }
00796 #endif
00797 }
libavcodec/x86/vc1dsp_mmx.c
