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/mem.h"
00029 #include "libavutil/x86/asm.h"
00030 #include "libavcodec/dsputil.h"
00031 #include "dsputil_mmx.h"
00032 #include "libavcodec/vc1dsp.h"
00033 
00034 #if HAVE_INLINE_ASM
00035 
00036 #define OP_PUT(S,D)
00037 #define OP_AVG(S,D) "pavgb " #S ", " #D " \n\t"
00038 
00040 #define NORMALIZE_MMX(SHIFT)                                    \
00041      "paddw     %%mm7, %%mm3           \n\t" /* +bias-r */      \
00042      "paddw     %%mm7, %%mm4           \n\t" /* +bias-r */      \
00043      "psraw     "SHIFT", %%mm3         \n\t"                    \
00044      "psraw     "SHIFT", %%mm4         \n\t"
00045 
00046 #define TRANSFER_DO_PACK(OP)                    \
00047      "packuswb  %%mm4, %%mm3           \n\t"    \
00048      OP((%2), %%mm3)                            \
00049      "movq      %%mm3, (%2)            \n\t"
00050 
00051 #define TRANSFER_DONT_PACK(OP)                  \
00052      OP(0(%2), %%mm3)                           \
00053      OP(8(%2), %%mm4)                           \
00054      "movq      %%mm3, 0(%2)           \n\t"    \
00055      "movq      %%mm4, 8(%2)           \n\t"
00056 
00058 #define DO_UNPACK(reg)  "punpcklbw %%mm0, " reg "\n\t"
00059 #define DONT_UNPACK(reg)
00060 
00062 #define LOAD_ROUNDER_MMX(ROUND)                 \
00063      "movd      "ROUND", %%mm7         \n\t"    \
00064      "punpcklwd %%mm7, %%mm7           \n\t"    \
00065      "punpckldq %%mm7, %%mm7           \n\t"
00066 
00067 #define SHIFT2_LINE(OFF, R0,R1,R2,R3)           \
00068     "paddw     %%mm"#R2", %%mm"#R1"    \n\t"    \
00069     "movd      (%0,%3), %%mm"#R0"      \n\t"    \
00070     "pmullw    %%mm6, %%mm"#R1"        \n\t"    \
00071     "punpcklbw %%mm0, %%mm"#R0"        \n\t"    \
00072     "movd      (%0,%2), %%mm"#R3"      \n\t"    \
00073     "psubw     %%mm"#R0", %%mm"#R1"    \n\t"    \
00074     "punpcklbw %%mm0, %%mm"#R3"        \n\t"    \
00075     "paddw     %%mm7, %%mm"#R1"        \n\t"    \
00076     "psubw     %%mm"#R3", %%mm"#R1"    \n\t"    \
00077     "psraw     %4, %%mm"#R1"           \n\t"    \
00078     "movq      %%mm"#R1", "#OFF"(%1)   \n\t"    \
00079     "add       %2, %0                  \n\t"
00080 
00082 static void vc1_put_ver_16b_shift2_mmx(int16_t *dst,
00083                                        const uint8_t *src, x86_reg stride,
00084                                        int rnd, int64_t shift)
00085 {
00086     __asm__ volatile(
00087         "mov       $3, %%"REG_c"           \n\t"
00088         LOAD_ROUNDER_MMX("%5")
00089         "movq      "MANGLE(ff_pw_9)", %%mm6 \n\t"
00090         "1:                                \n\t"
00091         "movd      (%0), %%mm2             \n\t"
00092         "add       %2, %0                  \n\t"
00093         "movd      (%0), %%mm3             \n\t"
00094         "punpcklbw %%mm0, %%mm2            \n\t"
00095         "punpcklbw %%mm0, %%mm3            \n\t"
00096         SHIFT2_LINE(  0, 1, 2, 3, 4)
00097         SHIFT2_LINE( 24, 2, 3, 4, 1)
00098         SHIFT2_LINE( 48, 3, 4, 1, 2)
00099         SHIFT2_LINE( 72, 4, 1, 2, 3)
00100         SHIFT2_LINE( 96, 1, 2, 3, 4)
00101         SHIFT2_LINE(120, 2, 3, 4, 1)
00102         SHIFT2_LINE(144, 3, 4, 1, 2)
00103         SHIFT2_LINE(168, 4, 1, 2, 3)
00104         "sub       %6, %0                  \n\t"
00105         "add       $8, %1                  \n\t"
00106         "dec       %%"REG_c"               \n\t"
00107         "jnz 1b                            \n\t"
00108         : "+r"(src), "+r"(dst)
00109         : "r"(stride), "r"(-2*stride),
00110           "m"(shift), "m"(rnd), "r"(9*stride-4)
00111         : "%"REG_c, "memory"
00112     );
00113 }
00114 
00119 #define VC1_HOR_16b_SHIFT2(OP, OPNAME)\
00120 static void OPNAME ## vc1_hor_16b_shift2_mmx(uint8_t *dst, x86_reg stride,\
00121                                              const int16_t *src, int rnd)\
00122 {\
00123     int h = 8;\
00124 \
00125     src -= 1;\
00126     rnd -= (-1+9+9-1)*1024; /* Add -1024 bias */\
00127     __asm__ volatile(\
00128         LOAD_ROUNDER_MMX("%4")\
00129         "movq      "MANGLE(ff_pw_128)", %%mm6\n\t"\
00130         "movq      "MANGLE(ff_pw_9)", %%mm5 \n\t"\
00131         "1:                                \n\t"\
00132         "movq      2*0+0(%1), %%mm1        \n\t"\
00133         "movq      2*0+8(%1), %%mm2        \n\t"\
00134         "movq      2*1+0(%1), %%mm3        \n\t"\
00135         "movq      2*1+8(%1), %%mm4        \n\t"\
00136         "paddw     2*3+0(%1), %%mm1        \n\t"\
00137         "paddw     2*3+8(%1), %%mm2        \n\t"\
00138         "paddw     2*2+0(%1), %%mm3        \n\t"\
00139         "paddw     2*2+8(%1), %%mm4        \n\t"\
00140         "pmullw    %%mm5, %%mm3            \n\t"\
00141         "pmullw    %%mm5, %%mm4            \n\t"\
00142         "psubw     %%mm1, %%mm3            \n\t"\
00143         "psubw     %%mm2, %%mm4            \n\t"\
00144         NORMALIZE_MMX("$7")\
00145         /* Remove bias */\
00146         "paddw     %%mm6, %%mm3            \n\t"\
00147         "paddw     %%mm6, %%mm4            \n\t"\
00148         TRANSFER_DO_PACK(OP)\
00149         "add       $24, %1                 \n\t"\
00150         "add       %3, %2                  \n\t"\
00151         "decl      %0                      \n\t"\
00152         "jnz 1b                            \n\t"\
00153         : "+r"(h), "+r" (src),  "+r" (dst)\
00154         : "r"(stride), "m"(rnd)\
00155         : "memory"\
00156     );\
00157 }
00158 
00159 VC1_HOR_16b_SHIFT2(OP_PUT, put_)
00160 VC1_HOR_16b_SHIFT2(OP_AVG, avg_)
00161 
00162 
00167 #define VC1_SHIFT2(OP, OPNAME)\
00168 static void OPNAME ## vc1_shift2_mmx(uint8_t *dst, const uint8_t *src,\
00169                                      x86_reg stride, int rnd, x86_reg offset)\
00170 {\
00171     rnd = 8-rnd;\
00172     __asm__ volatile(\
00173         "mov       $8, %%"REG_c"           \n\t"\
00174         LOAD_ROUNDER_MMX("%5")\
00175         "movq      "MANGLE(ff_pw_9)", %%mm6\n\t"\
00176         "1:                                \n\t"\
00177         "movd      0(%0   ), %%mm3         \n\t"\
00178         "movd      4(%0   ), %%mm4         \n\t"\
00179         "movd      0(%0,%2), %%mm1         \n\t"\
00180         "movd      4(%0,%2), %%mm2         \n\t"\
00181         "add       %2, %0                  \n\t"\
00182         "punpcklbw %%mm0, %%mm3            \n\t"\
00183         "punpcklbw %%mm0, %%mm4            \n\t"\
00184         "punpcklbw %%mm0, %%mm1            \n\t"\
00185         "punpcklbw %%mm0, %%mm2            \n\t"\
00186         "paddw     %%mm1, %%mm3            \n\t"\
00187         "paddw     %%mm2, %%mm4            \n\t"\
00188         "movd      0(%0,%3), %%mm1         \n\t"\
00189         "movd      4(%0,%3), %%mm2         \n\t"\
00190         "pmullw    %%mm6, %%mm3            \n\t" /* 0,9,9,0*/\
00191         "pmullw    %%mm6, %%mm4            \n\t" /* 0,9,9,0*/\
00192         "punpcklbw %%mm0, %%mm1            \n\t"\
00193         "punpcklbw %%mm0, %%mm2            \n\t"\
00194         "psubw     %%mm1, %%mm3            \n\t" /*-1,9,9,0*/\
00195         "psubw     %%mm2, %%mm4            \n\t" /*-1,9,9,0*/\
00196         "movd      0(%0,%2), %%mm1         \n\t"\
00197         "movd      4(%0,%2), %%mm2         \n\t"\
00198         "punpcklbw %%mm0, %%mm1            \n\t"\
00199         "punpcklbw %%mm0, %%mm2            \n\t"\
00200         "psubw     %%mm1, %%mm3            \n\t" /*-1,9,9,-1*/\
00201         "psubw     %%mm2, %%mm4            \n\t" /*-1,9,9,-1*/\
00202         NORMALIZE_MMX("$4")\
00203         "packuswb  %%mm4, %%mm3            \n\t"\
00204         OP((%1), %%mm3)\
00205         "movq      %%mm3, (%1)             \n\t"\
00206         "add       %6, %0                  \n\t"\
00207         "add       %4, %1                  \n\t"\
00208         "dec       %%"REG_c"               \n\t"\
00209         "jnz 1b                            \n\t"\
00210         : "+r"(src),  "+r"(dst)\
00211         : "r"(offset), "r"(-2*offset), "g"(stride), "m"(rnd),\
00212           "g"(stride-offset)\
00213         : "%"REG_c, "memory"\
00214     );\
00215 }
00216 
00217 VC1_SHIFT2(OP_PUT, put_)
00218 VC1_SHIFT2(OP_AVG, avg_)
00219 
00230 #define MSPEL_FILTER13_CORE(UNPACK, MOVQ, A1, A2, A3, A4)       \
00231      MOVQ "*0+"A1", %%mm1       \n\t"                           \
00232      MOVQ "*4+"A1", %%mm2       \n\t"                           \
00233      UNPACK("%%mm1")                                            \
00234      UNPACK("%%mm2")                                            \
00235      "pmullw    "MANGLE(ff_pw_3)", %%mm1\n\t"                   \
00236      "pmullw    "MANGLE(ff_pw_3)", %%mm2\n\t"                   \
00237      MOVQ "*0+"A2", %%mm3       \n\t"                           \
00238      MOVQ "*4+"A2", %%mm4       \n\t"                           \
00239      UNPACK("%%mm3")                                            \
00240      UNPACK("%%mm4")                                            \
00241      "pmullw    %%mm6, %%mm3    \n\t" /* *18 */                 \
00242      "pmullw    %%mm6, %%mm4    \n\t" /* *18 */                 \
00243      "psubw     %%mm1, %%mm3    \n\t" /* 18,-3 */               \
00244      "psubw     %%mm2, %%mm4    \n\t" /* 18,-3 */               \
00245      MOVQ "*0+"A4", %%mm1       \n\t"                           \
00246      MOVQ "*4+"A4", %%mm2       \n\t"                           \
00247      UNPACK("%%mm1")                                            \
00248      UNPACK("%%mm2")                                            \
00249      "psllw     $2, %%mm1       \n\t" /* 4* */                  \
00250      "psllw     $2, %%mm2       \n\t" /* 4* */                  \
00251      "psubw     %%mm1, %%mm3    \n\t" /* -4,18,-3 */            \
00252      "psubw     %%mm2, %%mm4    \n\t" /* -4,18,-3 */            \
00253      MOVQ "*0+"A3", %%mm1       \n\t"                           \
00254      MOVQ "*4+"A3", %%mm2       \n\t"                           \
00255      UNPACK("%%mm1")                                            \
00256      UNPACK("%%mm2")                                            \
00257      "pmullw    %%mm5, %%mm1    \n\t" /* *53 */                 \
00258      "pmullw    %%mm5, %%mm2    \n\t" /* *53 */                 \
00259      "paddw     %%mm1, %%mm3    \n\t" /* 4,53,18,-3 */          \
00260      "paddw     %%mm2, %%mm4    \n\t" /* 4,53,18,-3 */
00261 
00270 #define MSPEL_FILTER13_VER_16B(NAME, A1, A2, A3, A4)                    \
00271 static void                                                             \
00272 vc1_put_ver_16b_ ## NAME ## _mmx(int16_t *dst, const uint8_t *src,      \
00273                                  x86_reg src_stride,                   \
00274                                  int rnd, int64_t shift)                \
00275 {                                                                       \
00276     int h = 8;                                                          \
00277     src -= src_stride;                                                  \
00278     __asm__ volatile(                                                       \
00279         LOAD_ROUNDER_MMX("%5")                                          \
00280         "movq      "MANGLE(ff_pw_53)", %%mm5\n\t"                       \
00281         "movq      "MANGLE(ff_pw_18)", %%mm6\n\t"                       \
00282         ".p2align 3                \n\t"                                \
00283         "1:                        \n\t"                                \
00284         MSPEL_FILTER13_CORE(DO_UNPACK, "movd  1", A1, A2, A3, A4)       \
00285         NORMALIZE_MMX("%6")                                             \
00286         TRANSFER_DONT_PACK(OP_PUT)                                      \
00287         /* Last 3 (in fact 4) bytes on the line */                      \
00288         "movd      8+"A1", %%mm1   \n\t"                                \
00289         DO_UNPACK("%%mm1")                                              \
00290         "movq      %%mm1, %%mm3    \n\t"                                \
00291         "paddw     %%mm1, %%mm1    \n\t"                                \
00292         "paddw     %%mm3, %%mm1    \n\t" /* 3* */                       \
00293         "movd      8+"A2", %%mm3   \n\t"                                \
00294         DO_UNPACK("%%mm3")                                              \
00295         "pmullw    %%mm6, %%mm3    \n\t" /* *18 */                      \
00296         "psubw     %%mm1, %%mm3    \n\t" /*18,-3 */                     \
00297         "movd      8+"A3", %%mm1   \n\t"                                \
00298         DO_UNPACK("%%mm1")                                              \
00299         "pmullw    %%mm5, %%mm1    \n\t" /* *53 */                      \
00300         "paddw     %%mm1, %%mm3    \n\t" /*53,18,-3 */                  \
00301         "movd      8+"A4", %%mm1   \n\t"                                \
00302         DO_UNPACK("%%mm1")                                              \
00303         "psllw     $2, %%mm1       \n\t" /* 4* */                       \
00304         "psubw     %%mm1, %%mm3    \n\t"                                \
00305         "paddw     %%mm7, %%mm3    \n\t"                                \
00306         "psraw     %6, %%mm3       \n\t"                                \
00307         "movq      %%mm3, 16(%2)   \n\t"                                \
00308         "add       %3, %1          \n\t"                                \
00309         "add       $24, %2         \n\t"                                \
00310         "decl      %0              \n\t"                                \
00311         "jnz 1b                    \n\t"                                \
00312         : "+r"(h), "+r" (src),  "+r" (dst)                              \
00313         : "r"(src_stride), "r"(3*src_stride),                           \
00314           "m"(rnd), "m"(shift)                                          \
00315         : "memory"                                                      \
00316     );                                                                  \
00317 }
00318 
00326 #define MSPEL_FILTER13_HOR_16B(NAME, A1, A2, A3, A4, OP, OPNAME)        \
00327 static void                                                             \
00328 OPNAME ## vc1_hor_16b_ ## NAME ## _mmx(uint8_t *dst, x86_reg stride,    \
00329                                  const int16_t *src, int rnd)           \
00330 {                                                                       \
00331     int h = 8;                                                          \
00332     src -= 1;                                                           \
00333     rnd -= (-4+58+13-3)*256; /* Add -256 bias */                        \
00334     __asm__ volatile(                                                       \
00335         LOAD_ROUNDER_MMX("%4")                                          \
00336         "movq      "MANGLE(ff_pw_18)", %%mm6   \n\t"                    \
00337         "movq      "MANGLE(ff_pw_53)", %%mm5   \n\t"                    \
00338         ".p2align 3                \n\t"                                \
00339         "1:                        \n\t"                                \
00340         MSPEL_FILTER13_CORE(DONT_UNPACK, "movq 2", A1, A2, A3, A4)      \
00341         NORMALIZE_MMX("$7")                                             \
00342         /* Remove bias */                                               \
00343         "paddw     "MANGLE(ff_pw_128)", %%mm3  \n\t"                    \
00344         "paddw     "MANGLE(ff_pw_128)", %%mm4  \n\t"                    \
00345         TRANSFER_DO_PACK(OP)                                            \
00346         "add       $24, %1         \n\t"                                \
00347         "add       %3, %2          \n\t"                                \
00348         "decl      %0              \n\t"                                \
00349         "jnz 1b                    \n\t"                                \
00350         : "+r"(h), "+r" (src),  "+r" (dst)                              \
00351         : "r"(stride), "m"(rnd)                                         \
00352         : "memory"                                                      \
00353     );                                                                  \
00354 }
00355 
00364 #define MSPEL_FILTER13_8B(NAME, A1, A2, A3, A4, OP, OPNAME)             \
00365 static void                                                             \
00366 OPNAME ## vc1_## NAME ## _mmx(uint8_t *dst, const uint8_t *src,         \
00367                         x86_reg stride, int rnd, x86_reg offset)      \
00368 {                                                                       \
00369     int h = 8;                                                          \
00370     src -= offset;                                                      \
00371     rnd = 32-rnd;                                                       \
00372     __asm__ volatile (                                                      \
00373         LOAD_ROUNDER_MMX("%6")                                          \
00374         "movq      "MANGLE(ff_pw_53)", %%mm5       \n\t"                \
00375         "movq      "MANGLE(ff_pw_18)", %%mm6       \n\t"                \
00376         ".p2align 3                \n\t"                                \
00377         "1:                        \n\t"                                \
00378         MSPEL_FILTER13_CORE(DO_UNPACK, "movd   1", A1, A2, A3, A4)      \
00379         NORMALIZE_MMX("$6")                                             \
00380         TRANSFER_DO_PACK(OP)                                            \
00381         "add       %5, %1          \n\t"                                \
00382         "add       %5, %2          \n\t"                                \
00383         "decl      %0              \n\t"                                \
00384         "jnz 1b                    \n\t"                                \
00385         : "+r"(h), "+r" (src),  "+r" (dst)                              \
00386         : "r"(offset), "r"(3*offset), "g"(stride), "m"(rnd)             \
00387         : "memory"                                                      \
00388     );                                                                  \
00389 }
00390 
00392 MSPEL_FILTER13_8B     (shift1, "0(%1,%4  )", "0(%1,%3,2)", "0(%1,%3  )", "0(%1     )", OP_PUT, put_)
00393 MSPEL_FILTER13_8B     (shift1, "0(%1,%4  )", "0(%1,%3,2)", "0(%1,%3  )", "0(%1     )", OP_AVG, avg_)
00394 MSPEL_FILTER13_VER_16B(shift1, "0(%1,%4  )", "0(%1,%3,2)", "0(%1,%3  )", "0(%1     )")
00395 MSPEL_FILTER13_HOR_16B(shift1, "2*3(%1)", "2*2(%1)", "2*1(%1)", "2*0(%1)", OP_PUT, put_)
00396 MSPEL_FILTER13_HOR_16B(shift1, "2*3(%1)", "2*2(%1)", "2*1(%1)", "2*0(%1)", OP_AVG, avg_)
00397 
00399 MSPEL_FILTER13_8B     (shift3, "0(%1     )", "0(%1,%3  )", "0(%1,%3,2)", "0(%1,%4  )", OP_PUT, put_)
00400 MSPEL_FILTER13_8B     (shift3, "0(%1     )", "0(%1,%3  )", "0(%1,%3,2)", "0(%1,%4  )", OP_AVG, avg_)
00401 MSPEL_FILTER13_VER_16B(shift3, "0(%1     )", "0(%1,%3  )", "0(%1,%3,2)", "0(%1,%4  )")
00402 MSPEL_FILTER13_HOR_16B(shift3, "2*0(%1)", "2*1(%1)", "2*2(%1)", "2*3(%1)", OP_PUT, put_)
00403 MSPEL_FILTER13_HOR_16B(shift3, "2*0(%1)", "2*1(%1)", "2*2(%1)", "2*3(%1)", OP_AVG, avg_)
00404 
00405 typedef void (*vc1_mspel_mc_filter_ver_16bits)(int16_t *dst, const uint8_t *src, x86_reg src_stride, int rnd, int64_t shift);
00406 typedef void (*vc1_mspel_mc_filter_hor_16bits)(uint8_t *dst, x86_reg dst_stride, const int16_t *src, int rnd);
00407 typedef void (*vc1_mspel_mc_filter_8bits)(uint8_t *dst, const uint8_t *src, x86_reg stride, int rnd, x86_reg offset);
00408 
00420 #define VC1_MSPEL_MC(OP)\
00421 static void OP ## vc1_mspel_mc(uint8_t *dst, const uint8_t *src, int stride,\
00422                                int hmode, int vmode, int rnd)\
00423 {\
00424     static const vc1_mspel_mc_filter_ver_16bits vc1_put_shift_ver_16bits[] =\
00425          { NULL, vc1_put_ver_16b_shift1_mmx, vc1_put_ver_16b_shift2_mmx, vc1_put_ver_16b_shift3_mmx };\
00426     static const vc1_mspel_mc_filter_hor_16bits vc1_put_shift_hor_16bits[] =\
00427          { NULL, OP ## vc1_hor_16b_shift1_mmx, OP ## vc1_hor_16b_shift2_mmx, OP ## vc1_hor_16b_shift3_mmx };\
00428     static const vc1_mspel_mc_filter_8bits vc1_put_shift_8bits[] =\
00429          { NULL, OP ## vc1_shift1_mmx, OP ## vc1_shift2_mmx, OP ## vc1_shift3_mmx };\
00430 \
00431     __asm__ volatile(\
00432         "pxor %%mm0, %%mm0         \n\t"\
00433         ::: "memory"\
00434     );\
00435 \
00436     if (vmode) { /* Vertical filter to apply */\
00437         if (hmode) { /* Horizontal filter to apply, output to tmp */\
00438             static const int shift_value[] = { 0, 5, 1, 5 };\
00439             int              shift = (shift_value[hmode]+shift_value[vmode])>>1;\
00440             int              r;\
00441             DECLARE_ALIGNED(16, int16_t, tmp)[12*8];\
00442 \
00443             r = (1<<(shift-1)) + rnd-1;\
00444             vc1_put_shift_ver_16bits[vmode](tmp, src-1, stride, r, shift);\
00445 \
00446             vc1_put_shift_hor_16bits[hmode](dst, stride, tmp+1, 64-rnd);\
00447             return;\
00448         }\
00449         else { /* No horizontal filter, output 8 lines to dst */\
00450             vc1_put_shift_8bits[vmode](dst, src, stride, 1-rnd, stride);\
00451             return;\
00452         }\
00453     }\
00454 \
00455     /* Horizontal mode with no vertical mode */\
00456     vc1_put_shift_8bits[hmode](dst, src, stride, rnd, 1);\
00457 }
00458 
00459 VC1_MSPEL_MC(put_)
00460 VC1_MSPEL_MC(avg_)
00461 
00463 #define DECLARE_FUNCTION(a, b)                                          \
00464 static void put_vc1_mspel_mc ## a ## b ## _mmx(uint8_t *dst, const uint8_t *src, int stride, int rnd) { \
00465      put_vc1_mspel_mc(dst, src, stride, a, b, rnd);                     \
00466 }\
00467 static void avg_vc1_mspel_mc ## a ## b ## _mmx2(uint8_t *dst, const uint8_t *src, int stride, int rnd) { \
00468      avg_vc1_mspel_mc(dst, src, stride, a, b, rnd);                     \
00469 }
00470 
00471 DECLARE_FUNCTION(0, 1)
00472 DECLARE_FUNCTION(0, 2)
00473 DECLARE_FUNCTION(0, 3)
00474 
00475 DECLARE_FUNCTION(1, 0)
00476 DECLARE_FUNCTION(1, 1)
00477 DECLARE_FUNCTION(1, 2)
00478 DECLARE_FUNCTION(1, 3)
00479 
00480 DECLARE_FUNCTION(2, 0)
00481 DECLARE_FUNCTION(2, 1)
00482 DECLARE_FUNCTION(2, 2)
00483 DECLARE_FUNCTION(2, 3)
00484 
00485 DECLARE_FUNCTION(3, 0)
00486 DECLARE_FUNCTION(3, 1)
00487 DECLARE_FUNCTION(3, 2)
00488 DECLARE_FUNCTION(3, 3)
00489 
00490 static void vc1_inv_trans_4x4_dc_mmx2(uint8_t *dest, int linesize, DCTELEM *block)
00491 {
00492     int dc = block[0];
00493     dc = (17 * dc +  4) >> 3;
00494     dc = (17 * dc + 64) >> 7;
00495     __asm__ volatile(
00496         "movd          %0, %%mm0 \n\t"
00497         "pshufw $0, %%mm0, %%mm0 \n\t"
00498         "pxor       %%mm1, %%mm1 \n\t"
00499         "psubw      %%mm0, %%mm1 \n\t"
00500         "packuswb   %%mm0, %%mm0 \n\t"
00501         "packuswb   %%mm1, %%mm1 \n\t"
00502         ::"r"(dc)
00503     );
00504     __asm__ volatile(
00505         "movd          %0, %%mm2 \n\t"
00506         "movd          %1, %%mm3 \n\t"
00507         "movd          %2, %%mm4 \n\t"
00508         "movd          %3, %%mm5 \n\t"
00509         "paddusb    %%mm0, %%mm2 \n\t"
00510         "paddusb    %%mm0, %%mm3 \n\t"
00511         "paddusb    %%mm0, %%mm4 \n\t"
00512         "paddusb    %%mm0, %%mm5 \n\t"
00513         "psubusb    %%mm1, %%mm2 \n\t"
00514         "psubusb    %%mm1, %%mm3 \n\t"
00515         "psubusb    %%mm1, %%mm4 \n\t"
00516         "psubusb    %%mm1, %%mm5 \n\t"
00517         "movd       %%mm2, %0    \n\t"
00518         "movd       %%mm3, %1    \n\t"
00519         "movd       %%mm4, %2    \n\t"
00520         "movd       %%mm5, %3    \n\t"
00521         :"+m"(*(uint32_t*)(dest+0*linesize)),
00522          "+m"(*(uint32_t*)(dest+1*linesize)),
00523          "+m"(*(uint32_t*)(dest+2*linesize)),
00524          "+m"(*(uint32_t*)(dest+3*linesize))
00525     );
00526 }
00527 
00528 static void vc1_inv_trans_4x8_dc_mmx2(uint8_t *dest, int linesize, DCTELEM *block)
00529 {
00530     int dc = block[0];
00531     dc = (17 * dc +  4) >> 3;
00532     dc = (12 * dc + 64) >> 7;
00533     __asm__ volatile(
00534         "movd          %0, %%mm0 \n\t"
00535         "pshufw $0, %%mm0, %%mm0 \n\t"
00536         "pxor       %%mm1, %%mm1 \n\t"
00537         "psubw      %%mm0, %%mm1 \n\t"
00538         "packuswb   %%mm0, %%mm0 \n\t"
00539         "packuswb   %%mm1, %%mm1 \n\t"
00540         ::"r"(dc)
00541     );
00542     __asm__ volatile(
00543         "movd          %0, %%mm2 \n\t"
00544         "movd          %1, %%mm3 \n\t"
00545         "movd          %2, %%mm4 \n\t"
00546         "movd          %3, %%mm5 \n\t"
00547         "paddusb    %%mm0, %%mm2 \n\t"
00548         "paddusb    %%mm0, %%mm3 \n\t"
00549         "paddusb    %%mm0, %%mm4 \n\t"
00550         "paddusb    %%mm0, %%mm5 \n\t"
00551         "psubusb    %%mm1, %%mm2 \n\t"
00552         "psubusb    %%mm1, %%mm3 \n\t"
00553         "psubusb    %%mm1, %%mm4 \n\t"
00554         "psubusb    %%mm1, %%mm5 \n\t"
00555         "movd       %%mm2, %0    \n\t"
00556         "movd       %%mm3, %1    \n\t"
00557         "movd       %%mm4, %2    \n\t"
00558         "movd       %%mm5, %3    \n\t"
00559         :"+m"(*(uint32_t*)(dest+0*linesize)),
00560          "+m"(*(uint32_t*)(dest+1*linesize)),
00561          "+m"(*(uint32_t*)(dest+2*linesize)),
00562          "+m"(*(uint32_t*)(dest+3*linesize))
00563     );
00564     dest += 4*linesize;
00565     __asm__ volatile(
00566         "movd          %0, %%mm2 \n\t"
00567         "movd          %1, %%mm3 \n\t"
00568         "movd          %2, %%mm4 \n\t"
00569         "movd          %3, %%mm5 \n\t"
00570         "paddusb    %%mm0, %%mm2 \n\t"
00571         "paddusb    %%mm0, %%mm3 \n\t"
00572         "paddusb    %%mm0, %%mm4 \n\t"
00573         "paddusb    %%mm0, %%mm5 \n\t"
00574         "psubusb    %%mm1, %%mm2 \n\t"
00575         "psubusb    %%mm1, %%mm3 \n\t"
00576         "psubusb    %%mm1, %%mm4 \n\t"
00577         "psubusb    %%mm1, %%mm5 \n\t"
00578         "movd       %%mm2, %0    \n\t"
00579         "movd       %%mm3, %1    \n\t"
00580         "movd       %%mm4, %2    \n\t"
00581         "movd       %%mm5, %3    \n\t"
00582         :"+m"(*(uint32_t*)(dest+0*linesize)),
00583          "+m"(*(uint32_t*)(dest+1*linesize)),
00584          "+m"(*(uint32_t*)(dest+2*linesize)),
00585          "+m"(*(uint32_t*)(dest+3*linesize))
00586     );
00587 }
00588 
00589 static void vc1_inv_trans_8x4_dc_mmx2(uint8_t *dest, int linesize, DCTELEM *block)
00590 {
00591     int dc = block[0];
00592     dc = ( 3 * dc +  1) >> 1;
00593     dc = (17 * dc + 64) >> 7;
00594     __asm__ volatile(
00595         "movd          %0, %%mm0 \n\t"
00596         "pshufw $0, %%mm0, %%mm0 \n\t"
00597         "pxor       %%mm1, %%mm1 \n\t"
00598         "psubw      %%mm0, %%mm1 \n\t"
00599         "packuswb   %%mm0, %%mm0 \n\t"
00600         "packuswb   %%mm1, %%mm1 \n\t"
00601         ::"r"(dc)
00602     );
00603     __asm__ volatile(
00604         "movq          %0, %%mm2 \n\t"
00605         "movq          %1, %%mm3 \n\t"
00606         "movq          %2, %%mm4 \n\t"
00607         "movq          %3, %%mm5 \n\t"
00608         "paddusb    %%mm0, %%mm2 \n\t"
00609         "paddusb    %%mm0, %%mm3 \n\t"
00610         "paddusb    %%mm0, %%mm4 \n\t"
00611         "paddusb    %%mm0, %%mm5 \n\t"
00612         "psubusb    %%mm1, %%mm2 \n\t"
00613         "psubusb    %%mm1, %%mm3 \n\t"
00614         "psubusb    %%mm1, %%mm4 \n\t"
00615         "psubusb    %%mm1, %%mm5 \n\t"
00616         "movq       %%mm2, %0    \n\t"
00617         "movq       %%mm3, %1    \n\t"
00618         "movq       %%mm4, %2    \n\t"
00619         "movq       %%mm5, %3    \n\t"
00620         :"+m"(*(uint32_t*)(dest+0*linesize)),
00621          "+m"(*(uint32_t*)(dest+1*linesize)),
00622          "+m"(*(uint32_t*)(dest+2*linesize)),
00623          "+m"(*(uint32_t*)(dest+3*linesize))
00624     );
00625 }
00626 
00627 static void vc1_inv_trans_8x8_dc_mmx2(uint8_t *dest, int linesize, DCTELEM *block)
00628 {
00629     int dc = block[0];
00630     dc = (3 * dc +  1) >> 1;
00631     dc = (3 * dc + 16) >> 5;
00632     __asm__ volatile(
00633         "movd          %0, %%mm0 \n\t"
00634         "pshufw $0, %%mm0, %%mm0 \n\t"
00635         "pxor       %%mm1, %%mm1 \n\t"
00636         "psubw      %%mm0, %%mm1 \n\t"
00637         "packuswb   %%mm0, %%mm0 \n\t"
00638         "packuswb   %%mm1, %%mm1 \n\t"
00639         ::"r"(dc)
00640     );
00641     __asm__ volatile(
00642         "movq          %0, %%mm2 \n\t"
00643         "movq          %1, %%mm3 \n\t"
00644         "movq          %2, %%mm4 \n\t"
00645         "movq          %3, %%mm5 \n\t"
00646         "paddusb    %%mm0, %%mm2 \n\t"
00647         "paddusb    %%mm0, %%mm3 \n\t"
00648         "paddusb    %%mm0, %%mm4 \n\t"
00649         "paddusb    %%mm0, %%mm5 \n\t"
00650         "psubusb    %%mm1, %%mm2 \n\t"
00651         "psubusb    %%mm1, %%mm3 \n\t"
00652         "psubusb    %%mm1, %%mm4 \n\t"
00653         "psubusb    %%mm1, %%mm5 \n\t"
00654         "movq       %%mm2, %0    \n\t"
00655         "movq       %%mm3, %1    \n\t"
00656         "movq       %%mm4, %2    \n\t"
00657         "movq       %%mm5, %3    \n\t"
00658         :"+m"(*(uint32_t*)(dest+0*linesize)),
00659          "+m"(*(uint32_t*)(dest+1*linesize)),
00660          "+m"(*(uint32_t*)(dest+2*linesize)),
00661          "+m"(*(uint32_t*)(dest+3*linesize))
00662     );
00663     dest += 4*linesize;
00664     __asm__ volatile(
00665         "movq          %0, %%mm2 \n\t"
00666         "movq          %1, %%mm3 \n\t"
00667         "movq          %2, %%mm4 \n\t"
00668         "movq          %3, %%mm5 \n\t"
00669         "paddusb    %%mm0, %%mm2 \n\t"
00670         "paddusb    %%mm0, %%mm3 \n\t"
00671         "paddusb    %%mm0, %%mm4 \n\t"
00672         "paddusb    %%mm0, %%mm5 \n\t"
00673         "psubusb    %%mm1, %%mm2 \n\t"
00674         "psubusb    %%mm1, %%mm3 \n\t"
00675         "psubusb    %%mm1, %%mm4 \n\t"
00676         "psubusb    %%mm1, %%mm5 \n\t"
00677         "movq       %%mm2, %0    \n\t"
00678         "movq       %%mm3, %1    \n\t"
00679         "movq       %%mm4, %2    \n\t"
00680         "movq       %%mm5, %3    \n\t"
00681         :"+m"(*(uint32_t*)(dest+0*linesize)),
00682          "+m"(*(uint32_t*)(dest+1*linesize)),
00683          "+m"(*(uint32_t*)(dest+2*linesize)),
00684          "+m"(*(uint32_t*)(dest+3*linesize))
00685     );
00686 }
00687 
00688 #endif /* HAVE_INLINE_ASM */
00689 
00690 #define LOOP_FILTER(EXT) \
00691 void ff_vc1_v_loop_filter4_ ## EXT(uint8_t *src, int stride, int pq); \
00692 void ff_vc1_h_loop_filter4_ ## EXT(uint8_t *src, int stride, int pq); \
00693 void ff_vc1_v_loop_filter8_ ## EXT(uint8_t *src, int stride, int pq); \
00694 void ff_vc1_h_loop_filter8_ ## EXT(uint8_t *src, int stride, int pq); \
00695 \
00696 static void vc1_v_loop_filter16_ ## EXT(uint8_t *src, int stride, int pq) \
00697 { \
00698     ff_vc1_v_loop_filter8_ ## EXT(src,   stride, pq); \
00699     ff_vc1_v_loop_filter8_ ## EXT(src+8, stride, pq); \
00700 } \
00701 \
00702 static void vc1_h_loop_filter16_ ## EXT(uint8_t *src, int stride, int pq) \
00703 { \
00704     ff_vc1_h_loop_filter8_ ## EXT(src,          stride, pq); \
00705     ff_vc1_h_loop_filter8_ ## EXT(src+8*stride, stride, pq); \
00706 }
00707 
00708 #if HAVE_YASM
00709 LOOP_FILTER(mmx2)
00710 LOOP_FILTER(sse2)
00711 LOOP_FILTER(ssse3)
00712 
00713 void ff_vc1_h_loop_filter8_sse4(uint8_t *src, int stride, int pq);
00714 
00715 static void vc1_h_loop_filter16_sse4(uint8_t *src, int stride, int pq)
00716 {
00717     ff_vc1_h_loop_filter8_sse4(src,          stride, pq);
00718     ff_vc1_h_loop_filter8_sse4(src+8*stride, stride, pq);
00719 }
00720 #endif /* HAVE_YASM */
00721 
00722 void ff_put_vc1_chroma_mc8_mmx_nornd  (uint8_t *dst, uint8_t *src,
00723                                        int stride, int h, int x, int y);
00724 void ff_avg_vc1_chroma_mc8_mmx2_nornd (uint8_t *dst, uint8_t *src,
00725                                        int stride, int h, int x, int y);
00726 void ff_avg_vc1_chroma_mc8_3dnow_nornd(uint8_t *dst, uint8_t *src,
00727                                        int stride, int h, int x, int y);
00728 void ff_put_vc1_chroma_mc8_ssse3_nornd(uint8_t *dst, uint8_t *src,
00729                                        int stride, int h, int x, int y);
00730 void ff_avg_vc1_chroma_mc8_ssse3_nornd(uint8_t *dst, uint8_t *src,
00731                                        int stride, int h, int x, int y);
00732 
00733 void ff_vc1dsp_init_mmx(VC1DSPContext *dsp)
00734 {
00735     int mm_flags = av_get_cpu_flags();
00736 
00737 #if HAVE_INLINE_ASM
00738     if (mm_flags & AV_CPU_FLAG_MMX) {
00739         dsp->put_vc1_mspel_pixels_tab[ 0] = ff_put_vc1_mspel_mc00_mmx;
00740         dsp->put_vc1_mspel_pixels_tab[ 4] = put_vc1_mspel_mc01_mmx;
00741         dsp->put_vc1_mspel_pixels_tab[ 8] = put_vc1_mspel_mc02_mmx;
00742         dsp->put_vc1_mspel_pixels_tab[12] = put_vc1_mspel_mc03_mmx;
00743 
00744         dsp->put_vc1_mspel_pixels_tab[ 1] = put_vc1_mspel_mc10_mmx;
00745         dsp->put_vc1_mspel_pixels_tab[ 5] = put_vc1_mspel_mc11_mmx;
00746         dsp->put_vc1_mspel_pixels_tab[ 9] = put_vc1_mspel_mc12_mmx;
00747         dsp->put_vc1_mspel_pixels_tab[13] = put_vc1_mspel_mc13_mmx;
00748 
00749         dsp->put_vc1_mspel_pixels_tab[ 2] = put_vc1_mspel_mc20_mmx;
00750         dsp->put_vc1_mspel_pixels_tab[ 6] = put_vc1_mspel_mc21_mmx;
00751         dsp->put_vc1_mspel_pixels_tab[10] = put_vc1_mspel_mc22_mmx;
00752         dsp->put_vc1_mspel_pixels_tab[14] = put_vc1_mspel_mc23_mmx;
00753 
00754         dsp->put_vc1_mspel_pixels_tab[ 3] = put_vc1_mspel_mc30_mmx;
00755         dsp->put_vc1_mspel_pixels_tab[ 7] = put_vc1_mspel_mc31_mmx;
00756         dsp->put_vc1_mspel_pixels_tab[11] = put_vc1_mspel_mc32_mmx;
00757         dsp->put_vc1_mspel_pixels_tab[15] = put_vc1_mspel_mc33_mmx;
00758 
00759         if (HAVE_YASM)
00760             dsp->put_no_rnd_vc1_chroma_pixels_tab[0]= ff_put_vc1_chroma_mc8_mmx_nornd;
00761     }
00762 
00763     if (mm_flags & AV_CPU_FLAG_MMXEXT) {
00764         dsp->avg_vc1_mspel_pixels_tab[ 0] = ff_avg_vc1_mspel_mc00_mmx2;
00765         dsp->avg_vc1_mspel_pixels_tab[ 4] = avg_vc1_mspel_mc01_mmx2;
00766         dsp->avg_vc1_mspel_pixels_tab[ 8] = avg_vc1_mspel_mc02_mmx2;
00767         dsp->avg_vc1_mspel_pixels_tab[12] = avg_vc1_mspel_mc03_mmx2;
00768 
00769         dsp->avg_vc1_mspel_pixels_tab[ 1] = avg_vc1_mspel_mc10_mmx2;
00770         dsp->avg_vc1_mspel_pixels_tab[ 5] = avg_vc1_mspel_mc11_mmx2;
00771         dsp->avg_vc1_mspel_pixels_tab[ 9] = avg_vc1_mspel_mc12_mmx2;
00772         dsp->avg_vc1_mspel_pixels_tab[13] = avg_vc1_mspel_mc13_mmx2;
00773 
00774         dsp->avg_vc1_mspel_pixels_tab[ 2] = avg_vc1_mspel_mc20_mmx2;
00775         dsp->avg_vc1_mspel_pixels_tab[ 6] = avg_vc1_mspel_mc21_mmx2;
00776         dsp->avg_vc1_mspel_pixels_tab[10] = avg_vc1_mspel_mc22_mmx2;
00777         dsp->avg_vc1_mspel_pixels_tab[14] = avg_vc1_mspel_mc23_mmx2;
00778 
00779         dsp->avg_vc1_mspel_pixels_tab[ 3] = avg_vc1_mspel_mc30_mmx2;
00780         dsp->avg_vc1_mspel_pixels_tab[ 7] = avg_vc1_mspel_mc31_mmx2;
00781         dsp->avg_vc1_mspel_pixels_tab[11] = avg_vc1_mspel_mc32_mmx2;
00782         dsp->avg_vc1_mspel_pixels_tab[15] = avg_vc1_mspel_mc33_mmx2;
00783 
00784         dsp->vc1_inv_trans_8x8_dc = vc1_inv_trans_8x8_dc_mmx2;
00785         dsp->vc1_inv_trans_4x8_dc = vc1_inv_trans_4x8_dc_mmx2;
00786         dsp->vc1_inv_trans_8x4_dc = vc1_inv_trans_8x4_dc_mmx2;
00787         dsp->vc1_inv_trans_4x4_dc = vc1_inv_trans_4x4_dc_mmx2;
00788 
00789         if (HAVE_YASM)
00790             dsp->avg_no_rnd_vc1_chroma_pixels_tab[0]= ff_avg_vc1_chroma_mc8_mmx2_nornd;
00791     } else if (HAVE_YASM && mm_flags & AV_CPU_FLAG_3DNOW) {
00792         dsp->avg_no_rnd_vc1_chroma_pixels_tab[0]= ff_avg_vc1_chroma_mc8_3dnow_nornd;
00793     }
00794 
00795     if (HAVE_YASM && mm_flags & AV_CPU_FLAG_SSSE3) {
00796         dsp->put_no_rnd_vc1_chroma_pixels_tab[0]= ff_put_vc1_chroma_mc8_ssse3_nornd;
00797         dsp->avg_no_rnd_vc1_chroma_pixels_tab[0]= ff_avg_vc1_chroma_mc8_ssse3_nornd;
00798     }
00799 #endif /* HAVE_INLINE_ASM */
00800 
00801 #define ASSIGN_LF(EXT) \
00802         dsp->vc1_v_loop_filter4  = ff_vc1_v_loop_filter4_ ## EXT; \
00803         dsp->vc1_h_loop_filter4  = ff_vc1_h_loop_filter4_ ## EXT; \
00804         dsp->vc1_v_loop_filter8  = ff_vc1_v_loop_filter8_ ## EXT; \
00805         dsp->vc1_h_loop_filter8  = ff_vc1_h_loop_filter8_ ## EXT; \
00806         dsp->vc1_v_loop_filter16 = vc1_v_loop_filter16_ ## EXT; \
00807         dsp->vc1_h_loop_filter16 = vc1_h_loop_filter16_ ## EXT
00808 
00809 #if HAVE_YASM
00810     if (mm_flags & AV_CPU_FLAG_MMX) {
00811     }
00812 
00813     if (mm_flags & AV_CPU_FLAG_MMXEXT) {
00814         ASSIGN_LF(mmx2);
00815     }
00816 
00817     if (mm_flags & AV_CPU_FLAG_SSE2) {
00818         dsp->vc1_v_loop_filter8  = ff_vc1_v_loop_filter8_sse2;
00819         dsp->vc1_h_loop_filter8  = ff_vc1_h_loop_filter8_sse2;
00820         dsp->vc1_v_loop_filter16 = vc1_v_loop_filter16_sse2;
00821         dsp->vc1_h_loop_filter16 = vc1_h_loop_filter16_sse2;
00822     }
00823     if (mm_flags & AV_CPU_FLAG_SSSE3) {
00824         ASSIGN_LF(ssse3);
00825     }
00826     if (mm_flags & AV_CPU_FLAG_SSE4) {
00827         dsp->vc1_h_loop_filter8  = ff_vc1_h_loop_filter8_sse4;
00828         dsp->vc1_h_loop_filter16 = vc1_h_loop_filter16_sse4;
00829     }
00830 #endif /* HAVE_YASM */
00831 }
libavcodec/x86/vc1dsp_mmx.c
