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/x86_cpu.h"
00028 #include "libavcodec/dsputil.h"
00029 #include "dsputil_mmx.h"
00030 
00031 #define OP_PUT(S,D)
00032 #define OP_AVG(S,D) "pavgb " #S ", " #D " \n\t"
00033 
00035 #define NORMALIZE_MMX(SHIFT)                                    \
00036      "paddw     %%mm7, %%mm3           \n\t" /* +bias-r */      \
00037      "paddw     %%mm7, %%mm4           \n\t" /* +bias-r */      \
00038      "psraw     "SHIFT", %%mm3         \n\t"                    \
00039      "psraw     "SHIFT", %%mm4         \n\t"
00040 
00041 #define TRANSFER_DO_PACK(OP)                    \
00042      "packuswb  %%mm4, %%mm3           \n\t"    \
00043      OP((%2), %%mm3)                            \
00044      "movq      %%mm3, (%2)            \n\t"
00045 
00046 #define TRANSFER_DONT_PACK(OP)                  \
00047      OP(0(%2), %%mm3)                           \
00048      OP(8(%2), %%mm4)                           \
00049      "movq      %%mm3, 0(%2)           \n\t"    \
00050      "movq      %%mm4, 8(%2)           \n\t"
00051 
00053 #define DO_UNPACK(reg)  "punpcklbw %%mm0, " reg "\n\t"
00054 #define DONT_UNPACK(reg)
00055 
00057 #define LOAD_ROUNDER_MMX(ROUND)                 \
00058      "movd      "ROUND", %%mm7         \n\t"    \
00059      "punpcklwd %%mm7, %%mm7           \n\t"    \
00060      "punpckldq %%mm7, %%mm7           \n\t"
00061 
00062 #define SHIFT2_LINE(OFF, R0,R1,R2,R3)           \
00063     "paddw     %%mm"#R2", %%mm"#R1"    \n\t"    \
00064     "movd      (%0,%3), %%mm"#R0"      \n\t"    \
00065     "pmullw    %%mm6, %%mm"#R1"        \n\t"    \
00066     "punpcklbw %%mm0, %%mm"#R0"        \n\t"    \
00067     "movd      (%0,%2), %%mm"#R3"      \n\t"    \
00068     "psubw     %%mm"#R0", %%mm"#R1"    \n\t"    \
00069     "punpcklbw %%mm0, %%mm"#R3"        \n\t"    \
00070     "paddw     %%mm7, %%mm"#R1"        \n\t"    \
00071     "psubw     %%mm"#R3", %%mm"#R1"    \n\t"    \
00072     "psraw     %4, %%mm"#R1"           \n\t"    \
00073     "movq      %%mm"#R1", "#OFF"(%1)   \n\t"    \
00074     "add       %2, %0                  \n\t"
00075 
00076 DECLARE_ALIGNED(16, const uint64_t, ff_pw_9) = 0x0009000900090009ULL;
00077 
00079 static void vc1_put_ver_16b_shift2_mmx(int16_t *dst,
00080                                        const uint8_t *src, x86_reg stride,
00081                                        int rnd, int64_t shift)
00082 {
00083     __asm__ volatile(
00084         "mov       $3, %%"REG_c"           \n\t"
00085         LOAD_ROUNDER_MMX("%5")
00086         "movq      "MANGLE(ff_pw_9)", %%mm6 \n\t"
00087         "1:                                \n\t"
00088         "movd      (%0), %%mm2             \n\t"
00089         "add       %2, %0                  \n\t"
00090         "movd      (%0), %%mm3             \n\t"
00091         "punpcklbw %%mm0, %%mm2            \n\t"
00092         "punpcklbw %%mm0, %%mm3            \n\t"
00093         SHIFT2_LINE(  0, 1, 2, 3, 4)
00094         SHIFT2_LINE( 24, 2, 3, 4, 1)
00095         SHIFT2_LINE( 48, 3, 4, 1, 2)
00096         SHIFT2_LINE( 72, 4, 1, 2, 3)
00097         SHIFT2_LINE( 96, 1, 2, 3, 4)
00098         SHIFT2_LINE(120, 2, 3, 4, 1)
00099         SHIFT2_LINE(144, 3, 4, 1, 2)
00100         SHIFT2_LINE(168, 4, 1, 2, 3)
00101         "sub       %6, %0                  \n\t"
00102         "add       $8, %1                  \n\t"
00103         "dec       %%"REG_c"               \n\t"
00104         "jnz 1b                            \n\t"
00105         : "+r"(src), "+r"(dst)
00106         : "r"(stride), "r"(-2*stride),
00107           "m"(shift), "m"(rnd), "r"(9*stride-4)
00108         : "%"REG_c, "memory"
00109     );
00110 }
00111 
00116 #define VC1_HOR_16b_SHIFT2(OP, OPNAME)\
00117 static void OPNAME ## vc1_hor_16b_shift2_mmx(uint8_t *dst, x86_reg stride,\
00118                                              const int16_t *src, int rnd)\
00119 {\
00120     int h = 8;\
00121 \
00122     src -= 1;\
00123     rnd -= (-1+9+9-1)*1024; /* Add -1024 bias */\
00124     __asm__ volatile(\
00125         LOAD_ROUNDER_MMX("%4")\
00126         "movq      "MANGLE(ff_pw_128)", %%mm6\n\t"\
00127         "movq      "MANGLE(ff_pw_9)", %%mm5 \n\t"\
00128         "1:                                \n\t"\
00129         "movq      2*0+0(%1), %%mm1        \n\t"\
00130         "movq      2*0+8(%1), %%mm2        \n\t"\
00131         "movq      2*1+0(%1), %%mm3        \n\t"\
00132         "movq      2*1+8(%1), %%mm4        \n\t"\
00133         "paddw     2*3+0(%1), %%mm1        \n\t"\
00134         "paddw     2*3+8(%1), %%mm2        \n\t"\
00135         "paddw     2*2+0(%1), %%mm3        \n\t"\
00136         "paddw     2*2+8(%1), %%mm4        \n\t"\
00137         "pmullw    %%mm5, %%mm3            \n\t"\
00138         "pmullw    %%mm5, %%mm4            \n\t"\
00139         "psubw     %%mm1, %%mm3            \n\t"\
00140         "psubw     %%mm2, %%mm4            \n\t"\
00141         NORMALIZE_MMX("$7")\
00142         /* Remove bias */\
00143         "paddw     %%mm6, %%mm3            \n\t"\
00144         "paddw     %%mm6, %%mm4            \n\t"\
00145         TRANSFER_DO_PACK(OP)\
00146         "add       $24, %1                 \n\t"\
00147         "add       %3, %2                  \n\t"\
00148         "decl      %0                      \n\t"\
00149         "jnz 1b                            \n\t"\
00150         : "+r"(h), "+r" (src),  "+r" (dst)\
00151         : "r"(stride), "m"(rnd)\
00152         : "memory"\
00153     );\
00154 }
00155 
00156 VC1_HOR_16b_SHIFT2(OP_PUT, put_)
00157 VC1_HOR_16b_SHIFT2(OP_AVG, avg_)
00158 
00159 
00164 #define VC1_SHIFT2(OP, OPNAME)\
00165 static void OPNAME ## vc1_shift2_mmx(uint8_t *dst, const uint8_t *src,\
00166                                      x86_reg stride, int rnd, x86_reg offset)\
00167 {\
00168     rnd = 8-rnd;\
00169     __asm__ volatile(\
00170         "mov       $8, %%"REG_c"           \n\t"\
00171         LOAD_ROUNDER_MMX("%5")\
00172         "movq      "MANGLE(ff_pw_9)", %%mm6\n\t"\
00173         "1:                                \n\t"\
00174         "movd      0(%0   ), %%mm3         \n\t"\
00175         "movd      4(%0   ), %%mm4         \n\t"\
00176         "movd      0(%0,%2), %%mm1         \n\t"\
00177         "movd      4(%0,%2), %%mm2         \n\t"\
00178         "add       %2, %0                  \n\t"\
00179         "punpcklbw %%mm0, %%mm3            \n\t"\
00180         "punpcklbw %%mm0, %%mm4            \n\t"\
00181         "punpcklbw %%mm0, %%mm1            \n\t"\
00182         "punpcklbw %%mm0, %%mm2            \n\t"\
00183         "paddw     %%mm1, %%mm3            \n\t"\
00184         "paddw     %%mm2, %%mm4            \n\t"\
00185         "movd      0(%0,%3), %%mm1         \n\t"\
00186         "movd      4(%0,%3), %%mm2         \n\t"\
00187         "pmullw    %%mm6, %%mm3            \n\t" /* 0,9,9,0*/\
00188         "pmullw    %%mm6, %%mm4            \n\t" /* 0,9,9,0*/\
00189         "punpcklbw %%mm0, %%mm1            \n\t"\
00190         "punpcklbw %%mm0, %%mm2            \n\t"\
00191         "psubw     %%mm1, %%mm3            \n\t" /*-1,9,9,0*/\
00192         "psubw     %%mm2, %%mm4            \n\t" /*-1,9,9,0*/\
00193         "movd      0(%0,%2), %%mm1         \n\t"\
00194         "movd      4(%0,%2), %%mm2         \n\t"\
00195         "punpcklbw %%mm0, %%mm1            \n\t"\
00196         "punpcklbw %%mm0, %%mm2            \n\t"\
00197         "psubw     %%mm1, %%mm3            \n\t" /*-1,9,9,-1*/\
00198         "psubw     %%mm2, %%mm4            \n\t" /*-1,9,9,-1*/\
00199         NORMALIZE_MMX("$4")\
00200         "packuswb  %%mm4, %%mm3            \n\t"\
00201         OP((%1), %%mm3)\
00202         "movq      %%mm3, (%1)             \n\t"\
00203         "add       %6, %0                  \n\t"\
00204         "add       %4, %1                  \n\t"\
00205         "dec       %%"REG_c"               \n\t"\
00206         "jnz 1b                            \n\t"\
00207         : "+r"(src),  "+r"(dst)\
00208         : "r"(offset), "r"(-2*offset), "g"(stride), "m"(rnd),\
00209           "g"(stride-offset)\
00210         : "%"REG_c, "memory"\
00211     );\
00212 }
00213 
00214 VC1_SHIFT2(OP_PUT, put_)
00215 VC1_SHIFT2(OP_AVG, avg_)
00216 
00221 DECLARE_ASM_CONST(16, uint64_t, ff_pw_53) = 0x0035003500350035ULL;
00222 DECLARE_ASM_CONST(16, uint64_t, ff_pw_18) = 0x0012001200120012ULL;
00223 
00234 #define MSPEL_FILTER13_CORE(UNPACK, MOVQ, A1, A2, A3, A4)       \
00235      MOVQ "*0+"A1", %%mm1       \n\t"                           \
00236      MOVQ "*4+"A1", %%mm2       \n\t"                           \
00237      UNPACK("%%mm1")                                            \
00238      UNPACK("%%mm2")                                            \
00239      "pmullw    "MANGLE(ff_pw_3)", %%mm1\n\t"                   \
00240      "pmullw    "MANGLE(ff_pw_3)", %%mm2\n\t"                   \
00241      MOVQ "*0+"A2", %%mm3       \n\t"                           \
00242      MOVQ "*4+"A2", %%mm4       \n\t"                           \
00243      UNPACK("%%mm3")                                            \
00244      UNPACK("%%mm4")                                            \
00245      "pmullw    %%mm6, %%mm3    \n\t" /* *18 */                 \
00246      "pmullw    %%mm6, %%mm4    \n\t" /* *18 */                 \
00247      "psubw     %%mm1, %%mm3    \n\t" /* 18,-3 */               \
00248      "psubw     %%mm2, %%mm4    \n\t" /* 18,-3 */               \
00249      MOVQ "*0+"A4", %%mm1       \n\t"                           \
00250      MOVQ "*4+"A4", %%mm2       \n\t"                           \
00251      UNPACK("%%mm1")                                            \
00252      UNPACK("%%mm2")                                            \
00253      "psllw     $2, %%mm1       \n\t" /* 4* */                  \
00254      "psllw     $2, %%mm2       \n\t" /* 4* */                  \
00255      "psubw     %%mm1, %%mm3    \n\t" /* -4,18,-3 */            \
00256      "psubw     %%mm2, %%mm4    \n\t" /* -4,18,-3 */            \
00257      MOVQ "*0+"A3", %%mm1       \n\t"                           \
00258      MOVQ "*4+"A3", %%mm2       \n\t"                           \
00259      UNPACK("%%mm1")                                            \
00260      UNPACK("%%mm2")                                            \
00261      "pmullw    %%mm5, %%mm1    \n\t" /* *53 */                 \
00262      "pmullw    %%mm5, %%mm2    \n\t" /* *53 */                 \
00263      "paddw     %%mm1, %%mm3    \n\t" /* 4,53,18,-3 */          \
00264      "paddw     %%mm2, %%mm4    \n\t" /* 4,53,18,-3 */
00265 
00274 #define MSPEL_FILTER13_VER_16B(NAME, A1, A2, A3, A4)                    \
00275 static void                                                             \
00276 vc1_put_ver_16b_ ## NAME ## _mmx(int16_t *dst, const uint8_t *src,      \
00277                                  x86_reg src_stride,                   \
00278                                  int rnd, int64_t shift)                \
00279 {                                                                       \
00280     int h = 8;                                                          \
00281     src -= src_stride;                                                  \
00282     __asm__ volatile(                                                       \
00283         LOAD_ROUNDER_MMX("%5")                                          \
00284         "movq      "MANGLE(ff_pw_53)", %%mm5\n\t"                       \
00285         "movq      "MANGLE(ff_pw_18)", %%mm6\n\t"                       \
00286         ASMALIGN(3)                                                     \
00287         "1:                        \n\t"                                \
00288         MSPEL_FILTER13_CORE(DO_UNPACK, "movd  1", A1, A2, A3, A4)       \
00289         NORMALIZE_MMX("%6")                                             \
00290         TRANSFER_DONT_PACK(OP_PUT)                                      \
00291         /* Last 3 (in fact 4) bytes on the line */                      \
00292         "movd      8+"A1", %%mm1   \n\t"                                \
00293         DO_UNPACK("%%mm1")                                              \
00294         "movq      %%mm1, %%mm3    \n\t"                                \
00295         "paddw     %%mm1, %%mm1    \n\t"                                \
00296         "paddw     %%mm3, %%mm1    \n\t" /* 3* */                       \
00297         "movd      8+"A2", %%mm3   \n\t"                                \
00298         DO_UNPACK("%%mm3")                                              \
00299         "pmullw    %%mm6, %%mm3    \n\t" /* *18 */                      \
00300         "psubw     %%mm1, %%mm3    \n\t" /*18,-3 */                     \
00301         "movd      8+"A3", %%mm1   \n\t"                                \
00302         DO_UNPACK("%%mm1")                                              \
00303         "pmullw    %%mm5, %%mm1    \n\t" /* *53 */                      \
00304         "paddw     %%mm1, %%mm3    \n\t" /*53,18,-3 */                  \
00305         "movd      8+"A4", %%mm1   \n\t"                                \
00306         DO_UNPACK("%%mm1")                                              \
00307         "psllw     $2, %%mm1       \n\t" /* 4* */                       \
00308         "psubw     %%mm1, %%mm3    \n\t"                                \
00309         "paddw     %%mm7, %%mm3    \n\t"                                \
00310         "psraw     %6, %%mm3       \n\t"                                \
00311         "movq      %%mm3, 16(%2)   \n\t"                                \
00312         "add       %3, %1          \n\t"                                \
00313         "add       $24, %2         \n\t"                                \
00314         "decl      %0              \n\t"                                \
00315         "jnz 1b                    \n\t"                                \
00316         : "+r"(h), "+r" (src),  "+r" (dst)                              \
00317         : "r"(src_stride), "r"(3*src_stride),                           \
00318           "m"(rnd), "m"(shift)                                          \
00319         : "memory"                                                      \
00320     );                                                                  \
00321 }
00322 
00330 #define MSPEL_FILTER13_HOR_16B(NAME, A1, A2, A3, A4, OP, OPNAME)        \
00331 static void                                                             \
00332 OPNAME ## vc1_hor_16b_ ## NAME ## _mmx(uint8_t *dst, x86_reg stride,    \
00333                                  const int16_t *src, int rnd)           \
00334 {                                                                       \
00335     int h = 8;                                                          \
00336     src -= 1;                                                           \
00337     rnd -= (-4+58+13-3)*256; /* Add -256 bias */                        \
00338     __asm__ volatile(                                                       \
00339         LOAD_ROUNDER_MMX("%4")                                          \
00340         "movq      "MANGLE(ff_pw_18)", %%mm6   \n\t"                    \
00341         "movq      "MANGLE(ff_pw_53)", %%mm5   \n\t"                    \
00342         ASMALIGN(3)                                                     \
00343         "1:                        \n\t"                                \
00344         MSPEL_FILTER13_CORE(DONT_UNPACK, "movq 2", A1, A2, A3, A4)      \
00345         NORMALIZE_MMX("$7")                                             \
00346         /* Remove bias */                                               \
00347         "paddw     "MANGLE(ff_pw_128)", %%mm3  \n\t"                    \
00348         "paddw     "MANGLE(ff_pw_128)", %%mm4  \n\t"                    \
00349         TRANSFER_DO_PACK(OP)                                            \
00350         "add       $24, %1         \n\t"                                \
00351         "add       %3, %2          \n\t"                                \
00352         "decl      %0              \n\t"                                \
00353         "jnz 1b                    \n\t"                                \
00354         : "+r"(h), "+r" (src),  "+r" (dst)                              \
00355         : "r"(stride), "m"(rnd)                                         \
00356         : "memory"                                                      \
00357     );                                                                  \
00358 }
00359 
00368 #define MSPEL_FILTER13_8B(NAME, A1, A2, A3, A4, OP, OPNAME)             \
00369 static void                                                             \
00370 OPNAME ## vc1_## NAME ## _mmx(uint8_t *dst, const uint8_t *src,         \
00371                         x86_reg stride, int rnd, x86_reg offset)      \
00372 {                                                                       \
00373     int h = 8;                                                          \
00374     src -= offset;                                                      \
00375     rnd = 32-rnd;                                                       \
00376     __asm__ volatile (                                                      \
00377         LOAD_ROUNDER_MMX("%6")                                          \
00378         "movq      "MANGLE(ff_pw_53)", %%mm5       \n\t"                \
00379         "movq      "MANGLE(ff_pw_18)", %%mm6       \n\t"                \
00380         ASMALIGN(3)                                                     \
00381         "1:                        \n\t"                                \
00382         MSPEL_FILTER13_CORE(DO_UNPACK, "movd   1", A1, A2, A3, A4)      \
00383         NORMALIZE_MMX("$6")                                             \
00384         TRANSFER_DO_PACK(OP)                                            \
00385         "add       %5, %1          \n\t"                                \
00386         "add       %5, %2          \n\t"                                \
00387         "decl      %0              \n\t"                                \
00388         "jnz 1b                    \n\t"                                \
00389         : "+r"(h), "+r" (src),  "+r" (dst)                              \
00390         : "r"(offset), "r"(3*offset), "g"(stride), "m"(rnd)             \
00391         : "memory"                                                      \
00392     );                                                                  \
00393 }
00394 
00396 MSPEL_FILTER13_8B     (shift1, "0(%1,%4  )", "0(%1,%3,2)", "0(%1,%3  )", "0(%1     )", OP_PUT, put_)
00397 MSPEL_FILTER13_8B     (shift1, "0(%1,%4  )", "0(%1,%3,2)", "0(%1,%3  )", "0(%1     )", OP_AVG, avg_)
00398 MSPEL_FILTER13_VER_16B(shift1, "0(%1,%4  )", "0(%1,%3,2)", "0(%1,%3  )", "0(%1     )")
00399 MSPEL_FILTER13_HOR_16B(shift1, "2*3(%1)", "2*2(%1)", "2*1(%1)", "2*0(%1)", OP_PUT, put_)
00400 MSPEL_FILTER13_HOR_16B(shift1, "2*3(%1)", "2*2(%1)", "2*1(%1)", "2*0(%1)", OP_AVG, avg_)
00401 
00403 MSPEL_FILTER13_8B     (shift3, "0(%1     )", "0(%1,%3  )", "0(%1,%3,2)", "0(%1,%4  )", OP_PUT, put_)
00404 MSPEL_FILTER13_8B     (shift3, "0(%1     )", "0(%1,%3  )", "0(%1,%3,2)", "0(%1,%4  )", OP_AVG, avg_)
00405 MSPEL_FILTER13_VER_16B(shift3, "0(%1     )", "0(%1,%3  )", "0(%1,%3,2)", "0(%1,%4  )")
00406 MSPEL_FILTER13_HOR_16B(shift3, "2*0(%1)", "2*1(%1)", "2*2(%1)", "2*3(%1)", OP_PUT, put_)
00407 MSPEL_FILTER13_HOR_16B(shift3, "2*0(%1)", "2*1(%1)", "2*2(%1)", "2*3(%1)", OP_AVG, avg_)
00408 
00409 typedef void (*vc1_mspel_mc_filter_ver_16bits)(int16_t *dst, const uint8_t *src, x86_reg src_stride, int rnd, int64_t shift);
00410 typedef void (*vc1_mspel_mc_filter_hor_16bits)(uint8_t *dst, x86_reg dst_stride, const int16_t *src, int rnd);
00411 typedef void (*vc1_mspel_mc_filter_8bits)(uint8_t *dst, const uint8_t *src, x86_reg stride, int rnd, x86_reg offset);
00412 
00424 #define VC1_MSPEL_MC(OP)\
00425 static void OP ## vc1_mspel_mc(uint8_t *dst, const uint8_t *src, int stride,\
00426                                int hmode, int vmode, int rnd)\
00427 {\
00428     static const vc1_mspel_mc_filter_ver_16bits vc1_put_shift_ver_16bits[] =\
00429          { NULL, vc1_put_ver_16b_shift1_mmx, vc1_put_ver_16b_shift2_mmx, vc1_put_ver_16b_shift3_mmx };\
00430     static const vc1_mspel_mc_filter_hor_16bits vc1_put_shift_hor_16bits[] =\
00431          { NULL, OP ## vc1_hor_16b_shift1_mmx, OP ## vc1_hor_16b_shift2_mmx, OP ## vc1_hor_16b_shift3_mmx };\
00432     static const vc1_mspel_mc_filter_8bits vc1_put_shift_8bits[] =\
00433          { NULL, OP ## vc1_shift1_mmx, OP ## vc1_shift2_mmx, OP ## vc1_shift3_mmx };\
00434 \
00435     __asm__ volatile(\
00436         "pxor %%mm0, %%mm0         \n\t"\
00437         ::: "memory"\
00438     );\
00439 \
00440     if (vmode) { /* Vertical filter to apply */\
00441         if (hmode) { /* Horizontal filter to apply, output to tmp */\
00442             static const int shift_value[] = { 0, 5, 1, 5 };\
00443             int              shift = (shift_value[hmode]+shift_value[vmode])>>1;\
00444             int              r;\
00445             DECLARE_ALIGNED(16, int16_t, tmp)[12*8];\
00446 \
00447             r = (1<<(shift-1)) + rnd-1;\
00448             vc1_put_shift_ver_16bits[vmode](tmp, src-1, stride, r, shift);\
00449 \
00450             vc1_put_shift_hor_16bits[hmode](dst, stride, tmp+1, 64-rnd);\
00451             return;\
00452         }\
00453         else { /* No horizontal filter, output 8 lines to dst */\
00454             vc1_put_shift_8bits[vmode](dst, src, stride, 1-rnd, stride);\
00455             return;\
00456         }\
00457     }\
00458 \
00459     /* Horizontal mode with no vertical mode */\
00460     vc1_put_shift_8bits[hmode](dst, src, stride, rnd, 1);\
00461 }
00462 
00463 VC1_MSPEL_MC(put_)
00464 VC1_MSPEL_MC(avg_)
00465 
00467 #define DECLARE_FUNCTION(a, b)                                          \
00468 static void put_vc1_mspel_mc ## a ## b ## _mmx(uint8_t *dst, const uint8_t *src, int stride, int rnd) { \
00469      put_vc1_mspel_mc(dst, src, stride, a, b, rnd);                     \
00470 }\
00471 static void avg_vc1_mspel_mc ## a ## b ## _mmx2(uint8_t *dst, const uint8_t *src, int stride, int rnd) { \
00472      avg_vc1_mspel_mc(dst, src, stride, a, b, rnd);                     \
00473 }
00474 
00475 DECLARE_FUNCTION(0, 1)
00476 DECLARE_FUNCTION(0, 2)
00477 DECLARE_FUNCTION(0, 3)
00478 
00479 DECLARE_FUNCTION(1, 0)
00480 DECLARE_FUNCTION(1, 1)
00481 DECLARE_FUNCTION(1, 2)
00482 DECLARE_FUNCTION(1, 3)
00483 
00484 DECLARE_FUNCTION(2, 0)
00485 DECLARE_FUNCTION(2, 1)
00486 DECLARE_FUNCTION(2, 2)
00487 DECLARE_FUNCTION(2, 3)
00488 
00489 DECLARE_FUNCTION(3, 0)
00490 DECLARE_FUNCTION(3, 1)
00491 DECLARE_FUNCTION(3, 2)
00492 DECLARE_FUNCTION(3, 3)
00493 
00494 static void vc1_inv_trans_4x4_dc_mmx2(uint8_t *dest, int linesize, DCTELEM *block)
00495 {
00496     int dc = block[0];
00497     dc = (17 * dc +  4) >> 3;
00498     dc = (17 * dc + 64) >> 7;
00499     __asm__ volatile(
00500         "movd          %0, %%mm0 \n\t"
00501         "pshufw $0, %%mm0, %%mm0 \n\t"
00502         "pxor       %%mm1, %%mm1 \n\t"
00503         "psubw      %%mm0, %%mm1 \n\t"
00504         "packuswb   %%mm0, %%mm0 \n\t"
00505         "packuswb   %%mm1, %%mm1 \n\t"
00506         ::"r"(dc)
00507     );
00508     __asm__ volatile(
00509         "movd          %0, %%mm2 \n\t"
00510         "movd          %1, %%mm3 \n\t"
00511         "movd          %2, %%mm4 \n\t"
00512         "movd          %3, %%mm5 \n\t"
00513         "paddusb    %%mm0, %%mm2 \n\t"
00514         "paddusb    %%mm0, %%mm3 \n\t"
00515         "paddusb    %%mm0, %%mm4 \n\t"
00516         "paddusb    %%mm0, %%mm5 \n\t"
00517         "psubusb    %%mm1, %%mm2 \n\t"
00518         "psubusb    %%mm1, %%mm3 \n\t"
00519         "psubusb    %%mm1, %%mm4 \n\t"
00520         "psubusb    %%mm1, %%mm5 \n\t"
00521         "movd       %%mm2, %0    \n\t"
00522         "movd       %%mm3, %1    \n\t"
00523         "movd       %%mm4, %2    \n\t"
00524         "movd       %%mm5, %3    \n\t"
00525         :"+m"(*(uint32_t*)(dest+0*linesize)),
00526          "+m"(*(uint32_t*)(dest+1*linesize)),
00527          "+m"(*(uint32_t*)(dest+2*linesize)),
00528          "+m"(*(uint32_t*)(dest+3*linesize))
00529     );
00530 }
00531 
00532 static void vc1_inv_trans_4x8_dc_mmx2(uint8_t *dest, int linesize, DCTELEM *block)
00533 {
00534     int dc = block[0];
00535     dc = (17 * dc +  4) >> 3;
00536     dc = (12 * dc + 64) >> 7;
00537     __asm__ volatile(
00538         "movd          %0, %%mm0 \n\t"
00539         "pshufw $0, %%mm0, %%mm0 \n\t"
00540         "pxor       %%mm1, %%mm1 \n\t"
00541         "psubw      %%mm0, %%mm1 \n\t"
00542         "packuswb   %%mm0, %%mm0 \n\t"
00543         "packuswb   %%mm1, %%mm1 \n\t"
00544         ::"r"(dc)
00545     );
00546     __asm__ volatile(
00547         "movd          %0, %%mm2 \n\t"
00548         "movd          %1, %%mm3 \n\t"
00549         "movd          %2, %%mm4 \n\t"
00550         "movd          %3, %%mm5 \n\t"
00551         "paddusb    %%mm0, %%mm2 \n\t"
00552         "paddusb    %%mm0, %%mm3 \n\t"
00553         "paddusb    %%mm0, %%mm4 \n\t"
00554         "paddusb    %%mm0, %%mm5 \n\t"
00555         "psubusb    %%mm1, %%mm2 \n\t"
00556         "psubusb    %%mm1, %%mm3 \n\t"
00557         "psubusb    %%mm1, %%mm4 \n\t"
00558         "psubusb    %%mm1, %%mm5 \n\t"
00559         "movd       %%mm2, %0    \n\t"
00560         "movd       %%mm3, %1    \n\t"
00561         "movd       %%mm4, %2    \n\t"
00562         "movd       %%mm5, %3    \n\t"
00563         :"+m"(*(uint32_t*)(dest+0*linesize)),
00564          "+m"(*(uint32_t*)(dest+1*linesize)),
00565          "+m"(*(uint32_t*)(dest+2*linesize)),
00566          "+m"(*(uint32_t*)(dest+3*linesize))
00567     );
00568     dest += 4*linesize;
00569     __asm__ volatile(
00570         "movd          %0, %%mm2 \n\t"
00571         "movd          %1, %%mm3 \n\t"
00572         "movd          %2, %%mm4 \n\t"
00573         "movd          %3, %%mm5 \n\t"
00574         "paddusb    %%mm0, %%mm2 \n\t"
00575         "paddusb    %%mm0, %%mm3 \n\t"
00576         "paddusb    %%mm0, %%mm4 \n\t"
00577         "paddusb    %%mm0, %%mm5 \n\t"
00578         "psubusb    %%mm1, %%mm2 \n\t"
00579         "psubusb    %%mm1, %%mm3 \n\t"
00580         "psubusb    %%mm1, %%mm4 \n\t"
00581         "psubusb    %%mm1, %%mm5 \n\t"
00582         "movd       %%mm2, %0    \n\t"
00583         "movd       %%mm3, %1    \n\t"
00584         "movd       %%mm4, %2    \n\t"
00585         "movd       %%mm5, %3    \n\t"
00586         :"+m"(*(uint32_t*)(dest+0*linesize)),
00587          "+m"(*(uint32_t*)(dest+1*linesize)),
00588          "+m"(*(uint32_t*)(dest+2*linesize)),
00589          "+m"(*(uint32_t*)(dest+3*linesize))
00590     );
00591 }
00592 
00593 static void vc1_inv_trans_8x4_dc_mmx2(uint8_t *dest, int linesize, DCTELEM *block)
00594 {
00595     int dc = block[0];
00596     dc = ( 3 * dc +  1) >> 1;
00597     dc = (17 * dc + 64) >> 7;
00598     __asm__ volatile(
00599         "movd          %0, %%mm0 \n\t"
00600         "pshufw $0, %%mm0, %%mm0 \n\t"
00601         "pxor       %%mm1, %%mm1 \n\t"
00602         "psubw      %%mm0, %%mm1 \n\t"
00603         "packuswb   %%mm0, %%mm0 \n\t"
00604         "packuswb   %%mm1, %%mm1 \n\t"
00605         ::"r"(dc)
00606     );
00607     __asm__ volatile(
00608         "movq          %0, %%mm2 \n\t"
00609         "movq          %1, %%mm3 \n\t"
00610         "movq          %2, %%mm4 \n\t"
00611         "movq          %3, %%mm5 \n\t"
00612         "paddusb    %%mm0, %%mm2 \n\t"
00613         "paddusb    %%mm0, %%mm3 \n\t"
00614         "paddusb    %%mm0, %%mm4 \n\t"
00615         "paddusb    %%mm0, %%mm5 \n\t"
00616         "psubusb    %%mm1, %%mm2 \n\t"
00617         "psubusb    %%mm1, %%mm3 \n\t"
00618         "psubusb    %%mm1, %%mm4 \n\t"
00619         "psubusb    %%mm1, %%mm5 \n\t"
00620         "movq       %%mm2, %0    \n\t"
00621         "movq       %%mm3, %1    \n\t"
00622         "movq       %%mm4, %2    \n\t"
00623         "movq       %%mm5, %3    \n\t"
00624         :"+m"(*(uint32_t*)(dest+0*linesize)),
00625          "+m"(*(uint32_t*)(dest+1*linesize)),
00626          "+m"(*(uint32_t*)(dest+2*linesize)),
00627          "+m"(*(uint32_t*)(dest+3*linesize))
00628     );
00629 }
00630 
00631 static void vc1_inv_trans_8x8_dc_mmx2(uint8_t *dest, int linesize, DCTELEM *block)
00632 {
00633     int dc = block[0];
00634     dc = (3 * dc +  1) >> 1;
00635     dc = (3 * dc + 16) >> 5;
00636     __asm__ volatile(
00637         "movd          %0, %%mm0 \n\t"
00638         "pshufw $0, %%mm0, %%mm0 \n\t"
00639         "pxor       %%mm1, %%mm1 \n\t"
00640         "psubw      %%mm0, %%mm1 \n\t"
00641         "packuswb   %%mm0, %%mm0 \n\t"
00642         "packuswb   %%mm1, %%mm1 \n\t"
00643         ::"r"(dc)
00644     );
00645     __asm__ volatile(
00646         "movq          %0, %%mm2 \n\t"
00647         "movq          %1, %%mm3 \n\t"
00648         "movq          %2, %%mm4 \n\t"
00649         "movq          %3, %%mm5 \n\t"
00650         "paddusb    %%mm0, %%mm2 \n\t"
00651         "paddusb    %%mm0, %%mm3 \n\t"
00652         "paddusb    %%mm0, %%mm4 \n\t"
00653         "paddusb    %%mm0, %%mm5 \n\t"
00654         "psubusb    %%mm1, %%mm2 \n\t"
00655         "psubusb    %%mm1, %%mm3 \n\t"
00656         "psubusb    %%mm1, %%mm4 \n\t"
00657         "psubusb    %%mm1, %%mm5 \n\t"
00658         "movq       %%mm2, %0    \n\t"
00659         "movq       %%mm3, %1    \n\t"
00660         "movq       %%mm4, %2    \n\t"
00661         "movq       %%mm5, %3    \n\t"
00662         :"+m"(*(uint32_t*)(dest+0*linesize)),
00663          "+m"(*(uint32_t*)(dest+1*linesize)),
00664          "+m"(*(uint32_t*)(dest+2*linesize)),
00665          "+m"(*(uint32_t*)(dest+3*linesize))
00666     );
00667     dest += 4*linesize;
00668     __asm__ volatile(
00669         "movq          %0, %%mm2 \n\t"
00670         "movq          %1, %%mm3 \n\t"
00671         "movq          %2, %%mm4 \n\t"
00672         "movq          %3, %%mm5 \n\t"
00673         "paddusb    %%mm0, %%mm2 \n\t"
00674         "paddusb    %%mm0, %%mm3 \n\t"
00675         "paddusb    %%mm0, %%mm4 \n\t"
00676         "paddusb    %%mm0, %%mm5 \n\t"
00677         "psubusb    %%mm1, %%mm2 \n\t"
00678         "psubusb    %%mm1, %%mm3 \n\t"
00679         "psubusb    %%mm1, %%mm4 \n\t"
00680         "psubusb    %%mm1, %%mm5 \n\t"
00681         "movq       %%mm2, %0    \n\t"
00682         "movq       %%mm3, %1    \n\t"
00683         "movq       %%mm4, %2    \n\t"
00684         "movq       %%mm5, %3    \n\t"
00685         :"+m"(*(uint32_t*)(dest+0*linesize)),
00686          "+m"(*(uint32_t*)(dest+1*linesize)),
00687          "+m"(*(uint32_t*)(dest+2*linesize)),
00688          "+m"(*(uint32_t*)(dest+3*linesize))
00689     );
00690 }
00691 
00692 void ff_vc1dsp_init_mmx(DSPContext* dsp, AVCodecContext *avctx) {
00693     mm_flags = mm_support();
00694 
00695     dsp->put_vc1_mspel_pixels_tab[ 0] = ff_put_vc1_mspel_mc00_mmx;
00696     dsp->put_vc1_mspel_pixels_tab[ 4] = put_vc1_mspel_mc01_mmx;
00697     dsp->put_vc1_mspel_pixels_tab[ 8] = put_vc1_mspel_mc02_mmx;
00698     dsp->put_vc1_mspel_pixels_tab[12] = put_vc1_mspel_mc03_mmx;
00699 
00700     dsp->put_vc1_mspel_pixels_tab[ 1] = put_vc1_mspel_mc10_mmx;
00701     dsp->put_vc1_mspel_pixels_tab[ 5] = put_vc1_mspel_mc11_mmx;
00702     dsp->put_vc1_mspel_pixels_tab[ 9] = put_vc1_mspel_mc12_mmx;
00703     dsp->put_vc1_mspel_pixels_tab[13] = put_vc1_mspel_mc13_mmx;
00704 
00705     dsp->put_vc1_mspel_pixels_tab[ 2] = put_vc1_mspel_mc20_mmx;
00706     dsp->put_vc1_mspel_pixels_tab[ 6] = put_vc1_mspel_mc21_mmx;
00707     dsp->put_vc1_mspel_pixels_tab[10] = put_vc1_mspel_mc22_mmx;
00708     dsp->put_vc1_mspel_pixels_tab[14] = put_vc1_mspel_mc23_mmx;
00709 
00710     dsp->put_vc1_mspel_pixels_tab[ 3] = put_vc1_mspel_mc30_mmx;
00711     dsp->put_vc1_mspel_pixels_tab[ 7] = put_vc1_mspel_mc31_mmx;
00712     dsp->put_vc1_mspel_pixels_tab[11] = put_vc1_mspel_mc32_mmx;
00713     dsp->put_vc1_mspel_pixels_tab[15] = put_vc1_mspel_mc33_mmx;
00714 
00715     if (mm_flags & FF_MM_MMX2){
00716         dsp->avg_vc1_mspel_pixels_tab[ 0] = ff_avg_vc1_mspel_mc00_mmx2;
00717         dsp->avg_vc1_mspel_pixels_tab[ 4] = avg_vc1_mspel_mc01_mmx2;
00718         dsp->avg_vc1_mspel_pixels_tab[ 8] = avg_vc1_mspel_mc02_mmx2;
00719         dsp->avg_vc1_mspel_pixels_tab[12] = avg_vc1_mspel_mc03_mmx2;
00720 
00721         dsp->avg_vc1_mspel_pixels_tab[ 1] = avg_vc1_mspel_mc10_mmx2;
00722         dsp->avg_vc1_mspel_pixels_tab[ 5] = avg_vc1_mspel_mc11_mmx2;
00723         dsp->avg_vc1_mspel_pixels_tab[ 9] = avg_vc1_mspel_mc12_mmx2;
00724         dsp->avg_vc1_mspel_pixels_tab[13] = avg_vc1_mspel_mc13_mmx2;
00725 
00726         dsp->avg_vc1_mspel_pixels_tab[ 2] = avg_vc1_mspel_mc20_mmx2;
00727         dsp->avg_vc1_mspel_pixels_tab[ 6] = avg_vc1_mspel_mc21_mmx2;
00728         dsp->avg_vc1_mspel_pixels_tab[10] = avg_vc1_mspel_mc22_mmx2;
00729         dsp->avg_vc1_mspel_pixels_tab[14] = avg_vc1_mspel_mc23_mmx2;
00730 
00731         dsp->avg_vc1_mspel_pixels_tab[ 3] = avg_vc1_mspel_mc30_mmx2;
00732         dsp->avg_vc1_mspel_pixels_tab[ 7] = avg_vc1_mspel_mc31_mmx2;
00733         dsp->avg_vc1_mspel_pixels_tab[11] = avg_vc1_mspel_mc32_mmx2;
00734         dsp->avg_vc1_mspel_pixels_tab[15] = avg_vc1_mspel_mc33_mmx2;
00735 
00736         dsp->vc1_inv_trans_8x8_dc = vc1_inv_trans_8x8_dc_mmx2;
00737         dsp->vc1_inv_trans_4x8_dc = vc1_inv_trans_4x8_dc_mmx2;
00738         dsp->vc1_inv_trans_8x4_dc = vc1_inv_trans_8x4_dc_mmx2;
00739         dsp->vc1_inv_trans_4x4_dc = vc1_inv_trans_4x4_dc_mmx2;
00740     }
00741 }
libavcodec/x86/vc1dsp_mmx.c
