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 "libavutil/x86/cpu.h"
00031 #include "libavcodec/dsputil.h"
00032 #include "dsputil_mmx.h"
00033 #include "libavcodec/vc1dsp.h"
00034 #include "vc1dsp.h"
00035 
00036 #if HAVE_INLINE_ASM
00037 
00038 #define OP_PUT(S,D)
00039 #define OP_AVG(S,D) "pavgb " #S ", " #D " \n\t"
00040 
00042 #define NORMALIZE_MMX(SHIFT)                                    \
00043      "paddw     %%mm7, %%mm3           \n\t" /* +bias-r */      \
00044      "paddw     %%mm7, %%mm4           \n\t" /* +bias-r */      \
00045      "psraw     "SHIFT", %%mm3         \n\t"                    \
00046      "psraw     "SHIFT", %%mm4         \n\t"
00047 
00048 #define TRANSFER_DO_PACK(OP)                    \
00049      "packuswb  %%mm4, %%mm3           \n\t"    \
00050      OP((%2), %%mm3)                            \
00051      "movq      %%mm3, (%2)            \n\t"
00052 
00053 #define TRANSFER_DONT_PACK(OP)                  \
00054      OP(0(%2), %%mm3)                           \
00055      OP(8(%2), %%mm4)                           \
00056      "movq      %%mm3, 0(%2)           \n\t"    \
00057      "movq      %%mm4, 8(%2)           \n\t"
00058 
00060 #define DO_UNPACK(reg)  "punpcklbw %%mm0, " reg "\n\t"
00061 #define DONT_UNPACK(reg)
00062 
00064 #define LOAD_ROUNDER_MMX(ROUND)                 \
00065      "movd      "ROUND", %%mm7         \n\t"    \
00066      "punpcklwd %%mm7, %%mm7           \n\t"    \
00067      "punpckldq %%mm7, %%mm7           \n\t"
00068 
00069 #define SHIFT2_LINE(OFF, R0,R1,R2,R3)           \
00070     "paddw     %%mm"#R2", %%mm"#R1"    \n\t"    \
00071     "movd      (%0,%3), %%mm"#R0"      \n\t"    \
00072     "pmullw    %%mm6, %%mm"#R1"        \n\t"    \
00073     "punpcklbw %%mm0, %%mm"#R0"        \n\t"    \
00074     "movd      (%0,%2), %%mm"#R3"      \n\t"    \
00075     "psubw     %%mm"#R0", %%mm"#R1"    \n\t"    \
00076     "punpcklbw %%mm0, %%mm"#R3"        \n\t"    \
00077     "paddw     %%mm7, %%mm"#R1"        \n\t"    \
00078     "psubw     %%mm"#R3", %%mm"#R1"    \n\t"    \
00079     "psraw     %4, %%mm"#R1"           \n\t"    \
00080     "movq      %%mm"#R1", "#OFF"(%1)   \n\t"    \
00081     "add       %2, %0                  \n\t"
00082 
00084 static void vc1_put_ver_16b_shift2_mmx(int16_t *dst,
00085                                        const uint8_t *src, x86_reg stride,
00086                                        int rnd, int64_t shift)
00087 {
00088     __asm__ volatile(
00089         "mov       $3, %%"REG_c"           \n\t"
00090         LOAD_ROUNDER_MMX("%5")
00091         "movq      "MANGLE(ff_pw_9)", %%mm6 \n\t"
00092         "1:                                \n\t"
00093         "movd      (%0), %%mm2             \n\t"
00094         "add       %2, %0                  \n\t"
00095         "movd      (%0), %%mm3             \n\t"
00096         "punpcklbw %%mm0, %%mm2            \n\t"
00097         "punpcklbw %%mm0, %%mm3            \n\t"
00098         SHIFT2_LINE(  0, 1, 2, 3, 4)
00099         SHIFT2_LINE( 24, 2, 3, 4, 1)
00100         SHIFT2_LINE( 48, 3, 4, 1, 2)
00101         SHIFT2_LINE( 72, 4, 1, 2, 3)
00102         SHIFT2_LINE( 96, 1, 2, 3, 4)
00103         SHIFT2_LINE(120, 2, 3, 4, 1)
00104         SHIFT2_LINE(144, 3, 4, 1, 2)
00105         SHIFT2_LINE(168, 4, 1, 2, 3)
00106         "sub       %6, %0                  \n\t"
00107         "add       $8, %1                  \n\t"
00108         "dec       %%"REG_c"               \n\t"
00109         "jnz 1b                            \n\t"
00110         : "+r"(src), "+r"(dst)
00111         : "r"(stride), "r"(-2*stride),
00112           "m"(shift), "m"(rnd), "r"(9*stride-4)
00113         : "%"REG_c, "memory"
00114     );
00115 }
00116 
00121 #define VC1_HOR_16b_SHIFT2(OP, OPNAME)\
00122 static void OPNAME ## vc1_hor_16b_shift2_mmx(uint8_t *dst, x86_reg stride,\
00123                                              const int16_t *src, int rnd)\
00124 {\
00125     int h = 8;\
00126 \
00127     src -= 1;\
00128     rnd -= (-1+9+9-1)*1024; /* Add -1024 bias */\
00129     __asm__ volatile(\
00130         LOAD_ROUNDER_MMX("%4")\
00131         "movq      "MANGLE(ff_pw_128)", %%mm6\n\t"\
00132         "movq      "MANGLE(ff_pw_9)", %%mm5 \n\t"\
00133         "1:                                \n\t"\
00134         "movq      2*0+0(%1), %%mm1        \n\t"\
00135         "movq      2*0+8(%1), %%mm2        \n\t"\
00136         "movq      2*1+0(%1), %%mm3        \n\t"\
00137         "movq      2*1+8(%1), %%mm4        \n\t"\
00138         "paddw     2*3+0(%1), %%mm1        \n\t"\
00139         "paddw     2*3+8(%1), %%mm2        \n\t"\
00140         "paddw     2*2+0(%1), %%mm3        \n\t"\
00141         "paddw     2*2+8(%1), %%mm4        \n\t"\
00142         "pmullw    %%mm5, %%mm3            \n\t"\
00143         "pmullw    %%mm5, %%mm4            \n\t"\
00144         "psubw     %%mm1, %%mm3            \n\t"\
00145         "psubw     %%mm2, %%mm4            \n\t"\
00146         NORMALIZE_MMX("$7")\
00147         /* Remove bias */\
00148         "paddw     %%mm6, %%mm3            \n\t"\
00149         "paddw     %%mm6, %%mm4            \n\t"\
00150         TRANSFER_DO_PACK(OP)\
00151         "add       $24, %1                 \n\t"\
00152         "add       %3, %2                  \n\t"\
00153         "decl      %0                      \n\t"\
00154         "jnz 1b                            \n\t"\
00155         : "+r"(h), "+r" (src),  "+r" (dst)\
00156         : "r"(stride), "m"(rnd)\
00157         : "memory"\
00158     );\
00159 }
00160 
00161 VC1_HOR_16b_SHIFT2(OP_PUT, put_)
00162 VC1_HOR_16b_SHIFT2(OP_AVG, avg_)
00163 
00164 
00169 #define VC1_SHIFT2(OP, OPNAME)\
00170 static void OPNAME ## vc1_shift2_mmx(uint8_t *dst, const uint8_t *src,\
00171                                      x86_reg stride, int rnd, x86_reg offset)\
00172 {\
00173     rnd = 8-rnd;\
00174     __asm__ volatile(\
00175         "mov       $8, %%"REG_c"           \n\t"\
00176         LOAD_ROUNDER_MMX("%5")\
00177         "movq      "MANGLE(ff_pw_9)", %%mm6\n\t"\
00178         "1:                                \n\t"\
00179         "movd      0(%0   ), %%mm3         \n\t"\
00180         "movd      4(%0   ), %%mm4         \n\t"\
00181         "movd      0(%0,%2), %%mm1         \n\t"\
00182         "movd      4(%0,%2), %%mm2         \n\t"\
00183         "add       %2, %0                  \n\t"\
00184         "punpcklbw %%mm0, %%mm3            \n\t"\
00185         "punpcklbw %%mm0, %%mm4            \n\t"\
00186         "punpcklbw %%mm0, %%mm1            \n\t"\
00187         "punpcklbw %%mm0, %%mm2            \n\t"\
00188         "paddw     %%mm1, %%mm3            \n\t"\
00189         "paddw     %%mm2, %%mm4            \n\t"\
00190         "movd      0(%0,%3), %%mm1         \n\t"\
00191         "movd      4(%0,%3), %%mm2         \n\t"\
00192         "pmullw    %%mm6, %%mm3            \n\t" /* 0,9,9,0*/\
00193         "pmullw    %%mm6, %%mm4            \n\t" /* 0,9,9,0*/\
00194         "punpcklbw %%mm0, %%mm1            \n\t"\
00195         "punpcklbw %%mm0, %%mm2            \n\t"\
00196         "psubw     %%mm1, %%mm3            \n\t" /*-1,9,9,0*/\
00197         "psubw     %%mm2, %%mm4            \n\t" /*-1,9,9,0*/\
00198         "movd      0(%0,%2), %%mm1         \n\t"\
00199         "movd      4(%0,%2), %%mm2         \n\t"\
00200         "punpcklbw %%mm0, %%mm1            \n\t"\
00201         "punpcklbw %%mm0, %%mm2            \n\t"\
00202         "psubw     %%mm1, %%mm3            \n\t" /*-1,9,9,-1*/\
00203         "psubw     %%mm2, %%mm4            \n\t" /*-1,9,9,-1*/\
00204         NORMALIZE_MMX("$4")\
00205         "packuswb  %%mm4, %%mm3            \n\t"\
00206         OP((%1), %%mm3)\
00207         "movq      %%mm3, (%1)             \n\t"\
00208         "add       %6, %0                  \n\t"\
00209         "add       %4, %1                  \n\t"\
00210         "dec       %%"REG_c"               \n\t"\
00211         "jnz 1b                            \n\t"\
00212         : "+r"(src),  "+r"(dst)\
00213         : "r"(offset), "r"(-2*offset), "g"(stride), "m"(rnd),\
00214           "g"(stride-offset)\
00215         : "%"REG_c, "memory"\
00216     );\
00217 }
00218 
00219 VC1_SHIFT2(OP_PUT, put_)
00220 VC1_SHIFT2(OP_AVG, avg_)
00221 
00232 #define MSPEL_FILTER13_CORE(UNPACK, MOVQ, A1, A2, A3, A4)       \
00233      MOVQ "*0+"A1", %%mm1       \n\t"                           \
00234      MOVQ "*4+"A1", %%mm2       \n\t"                           \
00235      UNPACK("%%mm1")                                            \
00236      UNPACK("%%mm2")                                            \
00237      "pmullw    "MANGLE(ff_pw_3)", %%mm1\n\t"                   \
00238      "pmullw    "MANGLE(ff_pw_3)", %%mm2\n\t"                   \
00239      MOVQ "*0+"A2", %%mm3       \n\t"                           \
00240      MOVQ "*4+"A2", %%mm4       \n\t"                           \
00241      UNPACK("%%mm3")                                            \
00242      UNPACK("%%mm4")                                            \
00243      "pmullw    %%mm6, %%mm3    \n\t" /* *18 */                 \
00244      "pmullw    %%mm6, %%mm4    \n\t" /* *18 */                 \
00245      "psubw     %%mm1, %%mm3    \n\t" /* 18,-3 */               \
00246      "psubw     %%mm2, %%mm4    \n\t" /* 18,-3 */               \
00247      MOVQ "*0+"A4", %%mm1       \n\t"                           \
00248      MOVQ "*4+"A4", %%mm2       \n\t"                           \
00249      UNPACK("%%mm1")                                            \
00250      UNPACK("%%mm2")                                            \
00251      "psllw     $2, %%mm1       \n\t" /* 4* */                  \
00252      "psllw     $2, %%mm2       \n\t" /* 4* */                  \
00253      "psubw     %%mm1, %%mm3    \n\t" /* -4,18,-3 */            \
00254      "psubw     %%mm2, %%mm4    \n\t" /* -4,18,-3 */            \
00255      MOVQ "*0+"A3", %%mm1       \n\t"                           \
00256      MOVQ "*4+"A3", %%mm2       \n\t"                           \
00257      UNPACK("%%mm1")                                            \
00258      UNPACK("%%mm2")                                            \
00259      "pmullw    %%mm5, %%mm1    \n\t" /* *53 */                 \
00260      "pmullw    %%mm5, %%mm2    \n\t" /* *53 */                 \
00261      "paddw     %%mm1, %%mm3    \n\t" /* 4,53,18,-3 */          \
00262      "paddw     %%mm2, %%mm4    \n\t" /* 4,53,18,-3 */
00263 
00272 #define MSPEL_FILTER13_VER_16B(NAME, A1, A2, A3, A4)                    \
00273 static void                                                             \
00274 vc1_put_ver_16b_ ## NAME ## _mmx(int16_t *dst, const uint8_t *src,      \
00275                                  x86_reg src_stride,                   \
00276                                  int rnd, int64_t shift)                \
00277 {                                                                       \
00278     int h = 8;                                                          \
00279     src -= src_stride;                                                  \
00280     __asm__ volatile(                                                       \
00281         LOAD_ROUNDER_MMX("%5")                                          \
00282         "movq      "MANGLE(ff_pw_53)", %%mm5\n\t"                       \
00283         "movq      "MANGLE(ff_pw_18)", %%mm6\n\t"                       \
00284         ".p2align 3                \n\t"                                \
00285         "1:                        \n\t"                                \
00286         MSPEL_FILTER13_CORE(DO_UNPACK, "movd  1", A1, A2, A3, A4)       \
00287         NORMALIZE_MMX("%6")                                             \
00288         TRANSFER_DONT_PACK(OP_PUT)                                      \
00289         /* Last 3 (in fact 4) bytes on the line */                      \
00290         "movd      8+"A1", %%mm1   \n\t"                                \
00291         DO_UNPACK("%%mm1")                                              \
00292         "movq      %%mm1, %%mm3    \n\t"                                \
00293         "paddw     %%mm1, %%mm1    \n\t"                                \
00294         "paddw     %%mm3, %%mm1    \n\t" /* 3* */                       \
00295         "movd      8+"A2", %%mm3   \n\t"                                \
00296         DO_UNPACK("%%mm3")                                              \
00297         "pmullw    %%mm6, %%mm3    \n\t" /* *18 */                      \
00298         "psubw     %%mm1, %%mm3    \n\t" /*18,-3 */                     \
00299         "movd      8+"A3", %%mm1   \n\t"                                \
00300         DO_UNPACK("%%mm1")                                              \
00301         "pmullw    %%mm5, %%mm1    \n\t" /* *53 */                      \
00302         "paddw     %%mm1, %%mm3    \n\t" /*53,18,-3 */                  \
00303         "movd      8+"A4", %%mm1   \n\t"                                \
00304         DO_UNPACK("%%mm1")                                              \
00305         "psllw     $2, %%mm1       \n\t" /* 4* */                       \
00306         "psubw     %%mm1, %%mm3    \n\t"                                \
00307         "paddw     %%mm7, %%mm3    \n\t"                                \
00308         "psraw     %6, %%mm3       \n\t"                                \
00309         "movq      %%mm3, 16(%2)   \n\t"                                \
00310         "add       %3, %1          \n\t"                                \
00311         "add       $24, %2         \n\t"                                \
00312         "decl      %0              \n\t"                                \
00313         "jnz 1b                    \n\t"                                \
00314         : "+r"(h), "+r" (src),  "+r" (dst)                              \
00315         : "r"(src_stride), "r"(3*src_stride),                           \
00316           "m"(rnd), "m"(shift)                                          \
00317         : "memory"                                                      \
00318     );                                                                  \
00319 }
00320 
00328 #define MSPEL_FILTER13_HOR_16B(NAME, A1, A2, A3, A4, OP, OPNAME)        \
00329 static void                                                             \
00330 OPNAME ## vc1_hor_16b_ ## NAME ## _mmx(uint8_t *dst, x86_reg stride,    \
00331                                  const int16_t *src, int rnd)           \
00332 {                                                                       \
00333     int h = 8;                                                          \
00334     src -= 1;                                                           \
00335     rnd -= (-4+58+13-3)*256; /* Add -256 bias */                        \
00336     __asm__ volatile(                                                       \
00337         LOAD_ROUNDER_MMX("%4")                                          \
00338         "movq      "MANGLE(ff_pw_18)", %%mm6   \n\t"                    \
00339         "movq      "MANGLE(ff_pw_53)", %%mm5   \n\t"                    \
00340         ".p2align 3                \n\t"                                \
00341         "1:                        \n\t"                                \
00342         MSPEL_FILTER13_CORE(DONT_UNPACK, "movq 2", A1, A2, A3, A4)      \
00343         NORMALIZE_MMX("$7")                                             \
00344         /* Remove bias */                                               \
00345         "paddw     "MANGLE(ff_pw_128)", %%mm3  \n\t"                    \
00346         "paddw     "MANGLE(ff_pw_128)", %%mm4  \n\t"                    \
00347         TRANSFER_DO_PACK(OP)                                            \
00348         "add       $24, %1         \n\t"                                \
00349         "add       %3, %2          \n\t"                                \
00350         "decl      %0              \n\t"                                \
00351         "jnz 1b                    \n\t"                                \
00352         : "+r"(h), "+r" (src),  "+r" (dst)                              \
00353         : "r"(stride), "m"(rnd)                                         \
00354         : "memory"                                                      \
00355     );                                                                  \
00356 }
00357 
00366 #define MSPEL_FILTER13_8B(NAME, A1, A2, A3, A4, OP, OPNAME)             \
00367 static void                                                             \
00368 OPNAME ## vc1_## NAME ## _mmx(uint8_t *dst, const uint8_t *src,         \
00369                         x86_reg stride, int rnd, x86_reg offset)      \
00370 {                                                                       \
00371     int h = 8;                                                          \
00372     src -= offset;                                                      \
00373     rnd = 32-rnd;                                                       \
00374     __asm__ volatile (                                                      \
00375         LOAD_ROUNDER_MMX("%6")                                          \
00376         "movq      "MANGLE(ff_pw_53)", %%mm5       \n\t"                \
00377         "movq      "MANGLE(ff_pw_18)", %%mm6       \n\t"                \
00378         ".p2align 3                \n\t"                                \
00379         "1:                        \n\t"                                \
00380         MSPEL_FILTER13_CORE(DO_UNPACK, "movd   1", A1, A2, A3, A4)      \
00381         NORMALIZE_MMX("$6")                                             \
00382         TRANSFER_DO_PACK(OP)                                            \
00383         "add       %5, %1          \n\t"                                \
00384         "add       %5, %2          \n\t"                                \
00385         "decl      %0              \n\t"                                \
00386         "jnz 1b                    \n\t"                                \
00387         : "+r"(h), "+r" (src),  "+r" (dst)                              \
00388         : "r"(offset), "r"(3*offset), "g"(stride), "m"(rnd)             \
00389         : "memory"                                                      \
00390     );                                                                  \
00391 }
00392 
00394 MSPEL_FILTER13_8B     (shift1, "0(%1,%4  )", "0(%1,%3,2)", "0(%1,%3  )", "0(%1     )", OP_PUT, put_)
00395 MSPEL_FILTER13_8B     (shift1, "0(%1,%4  )", "0(%1,%3,2)", "0(%1,%3  )", "0(%1     )", OP_AVG, avg_)
00396 MSPEL_FILTER13_VER_16B(shift1, "0(%1,%4  )", "0(%1,%3,2)", "0(%1,%3  )", "0(%1     )")
00397 MSPEL_FILTER13_HOR_16B(shift1, "2*3(%1)", "2*2(%1)", "2*1(%1)", "2*0(%1)", OP_PUT, put_)
00398 MSPEL_FILTER13_HOR_16B(shift1, "2*3(%1)", "2*2(%1)", "2*1(%1)", "2*0(%1)", OP_AVG, avg_)
00399 
00401 MSPEL_FILTER13_8B     (shift3, "0(%1     )", "0(%1,%3  )", "0(%1,%3,2)", "0(%1,%4  )", OP_PUT, put_)
00402 MSPEL_FILTER13_8B     (shift3, "0(%1     )", "0(%1,%3  )", "0(%1,%3,2)", "0(%1,%4  )", OP_AVG, avg_)
00403 MSPEL_FILTER13_VER_16B(shift3, "0(%1     )", "0(%1,%3  )", "0(%1,%3,2)", "0(%1,%4  )")
00404 MSPEL_FILTER13_HOR_16B(shift3, "2*0(%1)", "2*1(%1)", "2*2(%1)", "2*3(%1)", OP_PUT, put_)
00405 MSPEL_FILTER13_HOR_16B(shift3, "2*0(%1)", "2*1(%1)", "2*2(%1)", "2*3(%1)", OP_AVG, avg_)
00406 
00407 typedef void (*vc1_mspel_mc_filter_ver_16bits)(int16_t *dst, const uint8_t *src, x86_reg src_stride, int rnd, int64_t shift);
00408 typedef void (*vc1_mspel_mc_filter_hor_16bits)(uint8_t *dst, x86_reg dst_stride, const int16_t *src, int rnd);
00409 typedef void (*vc1_mspel_mc_filter_8bits)(uint8_t *dst, const uint8_t *src, x86_reg stride, int rnd, x86_reg offset);
00410 
00422 #define VC1_MSPEL_MC(OP)\
00423 static void OP ## vc1_mspel_mc(uint8_t *dst, const uint8_t *src, int stride,\
00424                                int hmode, int vmode, int rnd)\
00425 {\
00426     static const vc1_mspel_mc_filter_ver_16bits vc1_put_shift_ver_16bits[] =\
00427          { NULL, vc1_put_ver_16b_shift1_mmx, vc1_put_ver_16b_shift2_mmx, vc1_put_ver_16b_shift3_mmx };\
00428     static const vc1_mspel_mc_filter_hor_16bits vc1_put_shift_hor_16bits[] =\
00429          { NULL, OP ## vc1_hor_16b_shift1_mmx, OP ## vc1_hor_16b_shift2_mmx, OP ## vc1_hor_16b_shift3_mmx };\
00430     static const vc1_mspel_mc_filter_8bits vc1_put_shift_8bits[] =\
00431          { NULL, OP ## vc1_shift1_mmx, OP ## vc1_shift2_mmx, OP ## vc1_shift3_mmx };\
00432 \
00433     __asm__ volatile(\
00434         "pxor %%mm0, %%mm0         \n\t"\
00435         ::: "memory"\
00436     );\
00437 \
00438     if (vmode) { /* Vertical filter to apply */\
00439         if (hmode) { /* Horizontal filter to apply, output to tmp */\
00440             static const int shift_value[] = { 0, 5, 1, 5 };\
00441             int              shift = (shift_value[hmode]+shift_value[vmode])>>1;\
00442             int              r;\
00443             DECLARE_ALIGNED(16, int16_t, tmp)[12*8];\
00444 \
00445             r = (1<<(shift-1)) + rnd-1;\
00446             vc1_put_shift_ver_16bits[vmode](tmp, src-1, stride, r, shift);\
00447 \
00448             vc1_put_shift_hor_16bits[hmode](dst, stride, tmp+1, 64-rnd);\
00449             return;\
00450         }\
00451         else { /* No horizontal filter, output 8 lines to dst */\
00452             vc1_put_shift_8bits[vmode](dst, src, stride, 1-rnd, stride);\
00453             return;\
00454         }\
00455     }\
00456 \
00457     /* Horizontal mode with no vertical mode */\
00458     vc1_put_shift_8bits[hmode](dst, src, stride, rnd, 1);\
00459 }
00460 
00461 VC1_MSPEL_MC(put_)
00462 VC1_MSPEL_MC(avg_)
00463 
00465 #define DECLARE_FUNCTION(a, b)                                          \
00466 static void put_vc1_mspel_mc ## a ## b ## _mmx(uint8_t *dst, const uint8_t *src, int stride, int rnd) { \
00467      put_vc1_mspel_mc(dst, src, stride, a, b, rnd);                     \
00468 }\
00469 static void avg_vc1_mspel_mc ## a ## b ## _mmxext(uint8_t *dst,         \
00470                                                   const uint8_t *src,   \
00471                                                   int stride, int rnd)  \
00472 {                                                                       \
00473      avg_vc1_mspel_mc(dst, src, stride, a, b, rnd);                     \
00474 }
00475 
00476 DECLARE_FUNCTION(0, 1)
00477 DECLARE_FUNCTION(0, 2)
00478 DECLARE_FUNCTION(0, 3)
00479 
00480 DECLARE_FUNCTION(1, 0)
00481 DECLARE_FUNCTION(1, 1)
00482 DECLARE_FUNCTION(1, 2)
00483 DECLARE_FUNCTION(1, 3)
00484 
00485 DECLARE_FUNCTION(2, 0)
00486 DECLARE_FUNCTION(2, 1)
00487 DECLARE_FUNCTION(2, 2)
00488 DECLARE_FUNCTION(2, 3)
00489 
00490 DECLARE_FUNCTION(3, 0)
00491 DECLARE_FUNCTION(3, 1)
00492 DECLARE_FUNCTION(3, 2)
00493 DECLARE_FUNCTION(3, 3)
00494 
00495 static void vc1_inv_trans_4x4_dc_mmxext(uint8_t *dest, int linesize,
00496                                         DCTELEM *block)
00497 {
00498     int dc = block[0];
00499     dc = (17 * dc +  4) >> 3;
00500     dc = (17 * dc + 64) >> 7;
00501     __asm__ volatile(
00502         "movd          %0, %%mm0 \n\t"
00503         "pshufw $0, %%mm0, %%mm0 \n\t"
00504         "pxor       %%mm1, %%mm1 \n\t"
00505         "psubw      %%mm0, %%mm1 \n\t"
00506         "packuswb   %%mm0, %%mm0 \n\t"
00507         "packuswb   %%mm1, %%mm1 \n\t"
00508         ::"r"(dc)
00509     );
00510     __asm__ volatile(
00511         "movd          %0, %%mm2 \n\t"
00512         "movd          %1, %%mm3 \n\t"
00513         "movd          %2, %%mm4 \n\t"
00514         "movd          %3, %%mm5 \n\t"
00515         "paddusb    %%mm0, %%mm2 \n\t"
00516         "paddusb    %%mm0, %%mm3 \n\t"
00517         "paddusb    %%mm0, %%mm4 \n\t"
00518         "paddusb    %%mm0, %%mm5 \n\t"
00519         "psubusb    %%mm1, %%mm2 \n\t"
00520         "psubusb    %%mm1, %%mm3 \n\t"
00521         "psubusb    %%mm1, %%mm4 \n\t"
00522         "psubusb    %%mm1, %%mm5 \n\t"
00523         "movd       %%mm2, %0    \n\t"
00524         "movd       %%mm3, %1    \n\t"
00525         "movd       %%mm4, %2    \n\t"
00526         "movd       %%mm5, %3    \n\t"
00527         :"+m"(*(uint32_t*)(dest+0*linesize)),
00528          "+m"(*(uint32_t*)(dest+1*linesize)),
00529          "+m"(*(uint32_t*)(dest+2*linesize)),
00530          "+m"(*(uint32_t*)(dest+3*linesize))
00531     );
00532 }
00533 
00534 static void vc1_inv_trans_4x8_dc_mmxext(uint8_t *dest, int linesize,
00535                                         DCTELEM *block)
00536 {
00537     int dc = block[0];
00538     dc = (17 * dc +  4) >> 3;
00539     dc = (12 * dc + 64) >> 7;
00540     __asm__ volatile(
00541         "movd          %0, %%mm0 \n\t"
00542         "pshufw $0, %%mm0, %%mm0 \n\t"
00543         "pxor       %%mm1, %%mm1 \n\t"
00544         "psubw      %%mm0, %%mm1 \n\t"
00545         "packuswb   %%mm0, %%mm0 \n\t"
00546         "packuswb   %%mm1, %%mm1 \n\t"
00547         ::"r"(dc)
00548     );
00549     __asm__ volatile(
00550         "movd          %0, %%mm2 \n\t"
00551         "movd          %1, %%mm3 \n\t"
00552         "movd          %2, %%mm4 \n\t"
00553         "movd          %3, %%mm5 \n\t"
00554         "paddusb    %%mm0, %%mm2 \n\t"
00555         "paddusb    %%mm0, %%mm3 \n\t"
00556         "paddusb    %%mm0, %%mm4 \n\t"
00557         "paddusb    %%mm0, %%mm5 \n\t"
00558         "psubusb    %%mm1, %%mm2 \n\t"
00559         "psubusb    %%mm1, %%mm3 \n\t"
00560         "psubusb    %%mm1, %%mm4 \n\t"
00561         "psubusb    %%mm1, %%mm5 \n\t"
00562         "movd       %%mm2, %0    \n\t"
00563         "movd       %%mm3, %1    \n\t"
00564         "movd       %%mm4, %2    \n\t"
00565         "movd       %%mm5, %3    \n\t"
00566         :"+m"(*(uint32_t*)(dest+0*linesize)),
00567          "+m"(*(uint32_t*)(dest+1*linesize)),
00568          "+m"(*(uint32_t*)(dest+2*linesize)),
00569          "+m"(*(uint32_t*)(dest+3*linesize))
00570     );
00571     dest += 4*linesize;
00572     __asm__ volatile(
00573         "movd          %0, %%mm2 \n\t"
00574         "movd          %1, %%mm3 \n\t"
00575         "movd          %2, %%mm4 \n\t"
00576         "movd          %3, %%mm5 \n\t"
00577         "paddusb    %%mm0, %%mm2 \n\t"
00578         "paddusb    %%mm0, %%mm3 \n\t"
00579         "paddusb    %%mm0, %%mm4 \n\t"
00580         "paddusb    %%mm0, %%mm5 \n\t"
00581         "psubusb    %%mm1, %%mm2 \n\t"
00582         "psubusb    %%mm1, %%mm3 \n\t"
00583         "psubusb    %%mm1, %%mm4 \n\t"
00584         "psubusb    %%mm1, %%mm5 \n\t"
00585         "movd       %%mm2, %0    \n\t"
00586         "movd       %%mm3, %1    \n\t"
00587         "movd       %%mm4, %2    \n\t"
00588         "movd       %%mm5, %3    \n\t"
00589         :"+m"(*(uint32_t*)(dest+0*linesize)),
00590          "+m"(*(uint32_t*)(dest+1*linesize)),
00591          "+m"(*(uint32_t*)(dest+2*linesize)),
00592          "+m"(*(uint32_t*)(dest+3*linesize))
00593     );
00594 }
00595 
00596 static void vc1_inv_trans_8x4_dc_mmxext(uint8_t *dest, int linesize,
00597                                         DCTELEM *block)
00598 {
00599     int dc = block[0];
00600     dc = ( 3 * dc +  1) >> 1;
00601     dc = (17 * dc + 64) >> 7;
00602     __asm__ volatile(
00603         "movd          %0, %%mm0 \n\t"
00604         "pshufw $0, %%mm0, %%mm0 \n\t"
00605         "pxor       %%mm1, %%mm1 \n\t"
00606         "psubw      %%mm0, %%mm1 \n\t"
00607         "packuswb   %%mm0, %%mm0 \n\t"
00608         "packuswb   %%mm1, %%mm1 \n\t"
00609         ::"r"(dc)
00610     );
00611     __asm__ volatile(
00612         "movq          %0, %%mm2 \n\t"
00613         "movq          %1, %%mm3 \n\t"
00614         "movq          %2, %%mm4 \n\t"
00615         "movq          %3, %%mm5 \n\t"
00616         "paddusb    %%mm0, %%mm2 \n\t"
00617         "paddusb    %%mm0, %%mm3 \n\t"
00618         "paddusb    %%mm0, %%mm4 \n\t"
00619         "paddusb    %%mm0, %%mm5 \n\t"
00620         "psubusb    %%mm1, %%mm2 \n\t"
00621         "psubusb    %%mm1, %%mm3 \n\t"
00622         "psubusb    %%mm1, %%mm4 \n\t"
00623         "psubusb    %%mm1, %%mm5 \n\t"
00624         "movq       %%mm2, %0    \n\t"
00625         "movq       %%mm3, %1    \n\t"
00626         "movq       %%mm4, %2    \n\t"
00627         "movq       %%mm5, %3    \n\t"
00628         :"+m"(*(uint32_t*)(dest+0*linesize)),
00629          "+m"(*(uint32_t*)(dest+1*linesize)),
00630          "+m"(*(uint32_t*)(dest+2*linesize)),
00631          "+m"(*(uint32_t*)(dest+3*linesize))
00632     );
00633 }
00634 
00635 static void vc1_inv_trans_8x8_dc_mmxext(uint8_t *dest, int linesize,
00636                                         DCTELEM *block)
00637 {
00638     int dc = block[0];
00639     dc = (3 * dc +  1) >> 1;
00640     dc = (3 * dc + 16) >> 5;
00641     __asm__ volatile(
00642         "movd          %0, %%mm0 \n\t"
00643         "pshufw $0, %%mm0, %%mm0 \n\t"
00644         "pxor       %%mm1, %%mm1 \n\t"
00645         "psubw      %%mm0, %%mm1 \n\t"
00646         "packuswb   %%mm0, %%mm0 \n\t"
00647         "packuswb   %%mm1, %%mm1 \n\t"
00648         ::"r"(dc)
00649     );
00650     __asm__ volatile(
00651         "movq          %0, %%mm2 \n\t"
00652         "movq          %1, %%mm3 \n\t"
00653         "movq          %2, %%mm4 \n\t"
00654         "movq          %3, %%mm5 \n\t"
00655         "paddusb    %%mm0, %%mm2 \n\t"
00656         "paddusb    %%mm0, %%mm3 \n\t"
00657         "paddusb    %%mm0, %%mm4 \n\t"
00658         "paddusb    %%mm0, %%mm5 \n\t"
00659         "psubusb    %%mm1, %%mm2 \n\t"
00660         "psubusb    %%mm1, %%mm3 \n\t"
00661         "psubusb    %%mm1, %%mm4 \n\t"
00662         "psubusb    %%mm1, %%mm5 \n\t"
00663         "movq       %%mm2, %0    \n\t"
00664         "movq       %%mm3, %1    \n\t"
00665         "movq       %%mm4, %2    \n\t"
00666         "movq       %%mm5, %3    \n\t"
00667         :"+m"(*(uint32_t*)(dest+0*linesize)),
00668          "+m"(*(uint32_t*)(dest+1*linesize)),
00669          "+m"(*(uint32_t*)(dest+2*linesize)),
00670          "+m"(*(uint32_t*)(dest+3*linesize))
00671     );
00672     dest += 4*linesize;
00673     __asm__ volatile(
00674         "movq          %0, %%mm2 \n\t"
00675         "movq          %1, %%mm3 \n\t"
00676         "movq          %2, %%mm4 \n\t"
00677         "movq          %3, %%mm5 \n\t"
00678         "paddusb    %%mm0, %%mm2 \n\t"
00679         "paddusb    %%mm0, %%mm3 \n\t"
00680         "paddusb    %%mm0, %%mm4 \n\t"
00681         "paddusb    %%mm0, %%mm5 \n\t"
00682         "psubusb    %%mm1, %%mm2 \n\t"
00683         "psubusb    %%mm1, %%mm3 \n\t"
00684         "psubusb    %%mm1, %%mm4 \n\t"
00685         "psubusb    %%mm1, %%mm5 \n\t"
00686         "movq       %%mm2, %0    \n\t"
00687         "movq       %%mm3, %1    \n\t"
00688         "movq       %%mm4, %2    \n\t"
00689         "movq       %%mm5, %3    \n\t"
00690         :"+m"(*(uint32_t*)(dest+0*linesize)),
00691          "+m"(*(uint32_t*)(dest+1*linesize)),
00692          "+m"(*(uint32_t*)(dest+2*linesize)),
00693          "+m"(*(uint32_t*)(dest+3*linesize))
00694     );
00695 }
00696 
00697 av_cold void ff_vc1dsp_init_mmx(VC1DSPContext *dsp)
00698 {
00699         dsp->put_vc1_mspel_pixels_tab[ 0] = ff_put_vc1_mspel_mc00_mmx;
00700         dsp->put_vc1_mspel_pixels_tab[ 4] = put_vc1_mspel_mc01_mmx;
00701         dsp->put_vc1_mspel_pixels_tab[ 8] = put_vc1_mspel_mc02_mmx;
00702         dsp->put_vc1_mspel_pixels_tab[12] = put_vc1_mspel_mc03_mmx;
00703 
00704         dsp->put_vc1_mspel_pixels_tab[ 1] = put_vc1_mspel_mc10_mmx;
00705         dsp->put_vc1_mspel_pixels_tab[ 5] = put_vc1_mspel_mc11_mmx;
00706         dsp->put_vc1_mspel_pixels_tab[ 9] = put_vc1_mspel_mc12_mmx;
00707         dsp->put_vc1_mspel_pixels_tab[13] = put_vc1_mspel_mc13_mmx;
00708 
00709         dsp->put_vc1_mspel_pixels_tab[ 2] = put_vc1_mspel_mc20_mmx;
00710         dsp->put_vc1_mspel_pixels_tab[ 6] = put_vc1_mspel_mc21_mmx;
00711         dsp->put_vc1_mspel_pixels_tab[10] = put_vc1_mspel_mc22_mmx;
00712         dsp->put_vc1_mspel_pixels_tab[14] = put_vc1_mspel_mc23_mmx;
00713 
00714         dsp->put_vc1_mspel_pixels_tab[ 3] = put_vc1_mspel_mc30_mmx;
00715         dsp->put_vc1_mspel_pixels_tab[ 7] = put_vc1_mspel_mc31_mmx;
00716         dsp->put_vc1_mspel_pixels_tab[11] = put_vc1_mspel_mc32_mmx;
00717         dsp->put_vc1_mspel_pixels_tab[15] = put_vc1_mspel_mc33_mmx;
00718 }
00719 
00720 av_cold void ff_vc1dsp_init_mmxext(VC1DSPContext *dsp)
00721 {
00722         dsp->avg_vc1_mspel_pixels_tab[ 0] = ff_avg_vc1_mspel_mc00_mmxext;
00723         dsp->avg_vc1_mspel_pixels_tab[ 4] = avg_vc1_mspel_mc01_mmxext;
00724         dsp->avg_vc1_mspel_pixels_tab[ 8] = avg_vc1_mspel_mc02_mmxext;
00725         dsp->avg_vc1_mspel_pixels_tab[12] = avg_vc1_mspel_mc03_mmxext;
00726 
00727         dsp->avg_vc1_mspel_pixels_tab[ 1] = avg_vc1_mspel_mc10_mmxext;
00728         dsp->avg_vc1_mspel_pixels_tab[ 5] = avg_vc1_mspel_mc11_mmxext;
00729         dsp->avg_vc1_mspel_pixels_tab[ 9] = avg_vc1_mspel_mc12_mmxext;
00730         dsp->avg_vc1_mspel_pixels_tab[13] = avg_vc1_mspel_mc13_mmxext;
00731 
00732         dsp->avg_vc1_mspel_pixels_tab[ 2] = avg_vc1_mspel_mc20_mmxext;
00733         dsp->avg_vc1_mspel_pixels_tab[ 6] = avg_vc1_mspel_mc21_mmxext;
00734         dsp->avg_vc1_mspel_pixels_tab[10] = avg_vc1_mspel_mc22_mmxext;
00735         dsp->avg_vc1_mspel_pixels_tab[14] = avg_vc1_mspel_mc23_mmxext;
00736 
00737         dsp->avg_vc1_mspel_pixels_tab[ 3] = avg_vc1_mspel_mc30_mmxext;
00738         dsp->avg_vc1_mspel_pixels_tab[ 7] = avg_vc1_mspel_mc31_mmxext;
00739         dsp->avg_vc1_mspel_pixels_tab[11] = avg_vc1_mspel_mc32_mmxext;
00740         dsp->avg_vc1_mspel_pixels_tab[15] = avg_vc1_mspel_mc33_mmxext;
00741 
00742         dsp->vc1_inv_trans_8x8_dc = vc1_inv_trans_8x8_dc_mmxext;
00743         dsp->vc1_inv_trans_4x8_dc = vc1_inv_trans_4x8_dc_mmxext;
00744         dsp->vc1_inv_trans_8x4_dc = vc1_inv_trans_8x4_dc_mmxext;
00745         dsp->vc1_inv_trans_4x4_dc = vc1_inv_trans_4x4_dc_mmxext;
00746 }
00747 #endif /* HAVE_INLINE_ASM */
libavcodec/x86/vc1dsp_mmx.c
