FFmpeg: libavcodec/x86/vc1dsp

00001 /*
00002  * VC-1 and WMV3 - DSP functions MMX-optimized
00003  * Copyright (c) 2007 Christophe GISQUET <christophe.gisquet@free.fr>
00004  *
00005  * Permission is hereby granted, free of charge, to any person
00006  * obtaining a copy of this software and associated documentation
00007  * files (the "Software"), to deal in the Software without
00008  * restriction, including without limitation the rights to use,
00009  * copy, modify, merge, publish, distribute, sublicense, and/or sell
00010  * copies of the Software, and to permit persons to whom the
00011  * Software is furnished to do so, subject to the following
00012  * conditions:
00013  *
00014  * The above copyright notice and this permission notice shall be
00015  * included in all copies or substantial portions of the Software.
00016  *
00017  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
00018  * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES
00019  * OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
00020  * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT
00021  * HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY,
00022  * WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
00023  * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
00024  * OTHER DEALINGS IN THE SOFTWARE.
00025  */
00026 
00027 #include "libavutil/cpu.h"
00028 #include "libavutil/x86_cpu.h"
00029 #include "libavcodec/dsputil.h"
00030 #include "dsputil_mmx.h"
00031 #include "libavcodec/vc1dsp.h"
00032 
00033 #define OP_PUT(S,D)
00034 #define OP_AVG(S,D) "pavgb " #S ", " #D " \n\t"
00035 
00037 #define NORMALIZE_MMX(SHIFT)                                    \
00038      "paddw     %%mm7, %%mm3           \n\t" /* +bias-r */      \
00039      "paddw     %%mm7, %%mm4           \n\t" /* +bias-r */      \
00040      "psraw     "SHIFT", %%mm3         \n\t"                    \
00041      "psraw     "SHIFT", %%mm4         \n\t"
00042 
00043 #define TRANSFER_DO_PACK(OP)                    \
00044      "packuswb  %%mm4, %%mm3           \n\t"    \
00045      OP((%2), %%mm3)                            \
00046      "movq      %%mm3, (%2)            \n\t"
00047 
00048 #define TRANSFER_DONT_PACK(OP)                  \
00049      OP(0(%2), %%mm3)                           \
00050      OP(8(%2), %%mm4)                           \
00051      "movq      %%mm3, 0(%2)           \n\t"    \
00052      "movq      %%mm4, 8(%2)           \n\t"
00053 
00055 #define DO_UNPACK(reg)  "punpcklbw %%mm0, " reg "\n\t"
00056 #define DONT_UNPACK(reg)
00057 
00059 #define LOAD_ROUNDER_MMX(ROUND)                 \
00060      "movd      "ROUND", %%mm7         \n\t"    \
00061      "punpcklwd %%mm7, %%mm7           \n\t"    \
00062      "punpckldq %%mm7, %%mm7           \n\t"
00063 
00064 #define SHIFT2_LINE(OFF, R0,R1,R2,R3)           \
00065     "paddw     %%mm"#R2", %%mm"#R1"    \n\t"    \
00066     "movd      (%0,%3), %%mm"#R0"      \n\t"    \
00067     "pmullw    %%mm6, %%mm"#R1"        \n\t"    \
00068     "punpcklbw %%mm0, %%mm"#R0"        \n\t"    \
00069     "movd      (%0,%2), %%mm"#R3"      \n\t"    \
00070     "psubw     %%mm"#R0", %%mm"#R1"    \n\t"    \
00071     "punpcklbw %%mm0, %%mm"#R3"        \n\t"    \
00072     "paddw     %%mm7, %%mm"#R1"        \n\t"    \
00073     "psubw     %%mm"#R3", %%mm"#R1"    \n\t"    \
00074     "psraw     %4, %%mm"#R1"           \n\t"    \
00075     "movq      %%mm"#R1", "#OFF"(%1)   \n\t"    \
00076     "add       %2, %0                  \n\t"
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 
00227 #define MSPEL_FILTER13_CORE(UNPACK, MOVQ, A1, A2, A3, A4)       \
00228      MOVQ "*0+"A1", %%mm1       \n\t"                           \
00229      MOVQ "*4+"A1", %%mm2       \n\t"                           \
00230      UNPACK("%%mm1")                                            \
00231      UNPACK("%%mm2")                                            \
00232      "pmullw    "MANGLE(ff_pw_3)", %%mm1\n\t"                   \
00233      "pmullw    "MANGLE(ff_pw_3)", %%mm2\n\t"                   \
00234      MOVQ "*0+"A2", %%mm3       \n\t"                           \
00235      MOVQ "*4+"A2", %%mm4       \n\t"                           \
00236      UNPACK("%%mm3")                                            \
00237      UNPACK("%%mm4")                                            \
00238      "pmullw    %%mm6, %%mm3    \n\t" /* *18 */                 \
00239      "pmullw    %%mm6, %%mm4    \n\t" /* *18 */                 \
00240      "psubw     %%mm1, %%mm3    \n\t" /* 18,-3 */               \
00241      "psubw     %%mm2, %%mm4    \n\t" /* 18,-3 */               \
00242      MOVQ "*0+"A4", %%mm1       \n\t"                           \
00243      MOVQ "*4+"A4", %%mm2       \n\t"                           \
00244      UNPACK("%%mm1")                                            \
00245      UNPACK("%%mm2")                                            \
00246      "psllw     $2, %%mm1       \n\t" /* 4* */                  \
00247      "psllw     $2, %%mm2       \n\t" /* 4* */                  \
00248      "psubw     %%mm1, %%mm3    \n\t" /* -4,18,-3 */            \
00249      "psubw     %%mm2, %%mm4    \n\t" /* -4,18,-3 */            \
00250      MOVQ "*0+"A3", %%mm1       \n\t"                           \
00251      MOVQ "*4+"A3", %%mm2       \n\t"                           \
00252      UNPACK("%%mm1")                                            \
00253      UNPACK("%%mm2")                                            \
00254      "pmullw    %%mm5, %%mm1    \n\t" /* *53 */                 \
00255      "pmullw    %%mm5, %%mm2    \n\t" /* *53 */                 \
00256      "paddw     %%mm1, %%mm3    \n\t" /* 4,53,18,-3 */          \
00257      "paddw     %%mm2, %%mm4    \n\t" /* 4,53,18,-3 */
00258 
00267 #define MSPEL_FILTER13_VER_16B(NAME, A1, A2, A3, A4)                    \
00268 static void                                                             \
00269 vc1_put_ver_16b_ ## NAME ## _mmx(int16_t *dst, const uint8_t *src,      \
00270                                  x86_reg src_stride,                   \
00271                                  int rnd, int64_t shift)                \
00272 {                                                                       \
00273     int h = 8;                                                          \
00274     src -= src_stride;                                                  \
00275     __asm__ volatile(                                                       \
00276         LOAD_ROUNDER_MMX("%5")                                          \
00277         "movq      "MANGLE(ff_pw_53)", %%mm5\n\t"                       \
00278         "movq      "MANGLE(ff_pw_18)", %%mm6\n\t"                       \
00279         ".p2align 3                \n\t"                                \
00280         "1:                        \n\t"                                \
00281         MSPEL_FILTER13_CORE(DO_UNPACK, "movd  1", A1, A2, A3, A4)       \
00282         NORMALIZE_MMX("%6")                                             \
00283         TRANSFER_DONT_PACK(OP_PUT)                                      \
00284         /* Last 3 (in fact 4) bytes on the line */                      \
00285         "movd      8+"A1", %%mm1   \n\t"                                \
00286         DO_UNPACK("%%mm1")                                              \
00287         "movq      %%mm1, %%mm3    \n\t"                                \
00288         "paddw     %%mm1, %%mm1    \n\t"                                \
00289         "paddw     %%mm3, %%mm1    \n\t" /* 3* */                       \
00290         "movd      8+"A2", %%mm3   \n\t"                                \
00291         DO_UNPACK("%%mm3")                                              \
00292         "pmullw    %%mm6, %%mm3    \n\t" /* *18 */                      \
00293         "psubw     %%mm1, %%mm3    \n\t" /*18,-3 */                     \
00294         "movd      8+"A3", %%mm1   \n\t"                                \
00295         DO_UNPACK("%%mm1")                                              \
00296         "pmullw    %%mm5, %%mm1    \n\t" /* *53 */                      \
00297         "paddw     %%mm1, %%mm3    \n\t" /*53,18,-3 */                  \
00298         "movd      8+"A4", %%mm1   \n\t"                                \
00299         DO_UNPACK("%%mm1")                                              \
00300         "psllw     $2, %%mm1       \n\t" /* 4* */                       \
00301         "psubw     %%mm1, %%mm3    \n\t"                                \
00302         "paddw     %%mm7, %%mm3    \n\t"                                \
00303         "psraw     %6, %%mm3       \n\t"                                \
00304         "movq      %%mm3, 16(%2)   \n\t"                                \
00305         "add       %3, %1          \n\t"                                \
00306         "add       $24, %2         \n\t"                                \
00307         "decl      %0              \n\t"                                \
00308         "jnz 1b                    \n\t"                                \
00309         : "+r"(h), "+r" (src),  "+r" (dst)                              \
00310         : "r"(src_stride), "r"(3*src_stride),                           \
00311           "m"(rnd), "m"(shift)                                          \
00312         : "memory"                                                      \
00313     );                                                                  \
00314 }
00315 
00323 #define MSPEL_FILTER13_HOR_16B(NAME, A1, A2, A3, A4, OP, OPNAME)        \
00324 static void                                                             \
00325 OPNAME ## vc1_hor_16b_ ## NAME ## _mmx(uint8_t *dst, x86_reg stride,    \
00326                                  const int16_t *src, int rnd)           \
00327 {                                                                       \
00328     int h = 8;                                                          \
00329     src -= 1;                                                           \
00330     rnd -= (-4+58+13-3)*256; /* Add -256 bias */                        \
00331     __asm__ volatile(                                                       \
00332         LOAD_ROUNDER_MMX("%4")                                          \
00333         "movq      "MANGLE(ff_pw_18)", %%mm6   \n\t"                    \
00334         "movq      "MANGLE(ff_pw_53)", %%mm5   \n\t"                    \
00335         ".p2align 3                \n\t"                                \
00336         "1:                        \n\t"                                \
00337         MSPEL_FILTER13_CORE(DONT_UNPACK, "movq 2", A1, A2, A3, A4)      \
00338         NORMALIZE_MMX("$7")                                             \
00339         /* Remove bias */                                               \
00340         "paddw     "MANGLE(ff_pw_128)", %%mm3  \n\t"                    \
00341         "paddw     "MANGLE(ff_pw_128)", %%mm4  \n\t"                    \
00342         TRANSFER_DO_PACK(OP)                                            \
00343         "add       $24, %1         \n\t"                                \
00344         "add       %3, %2          \n\t"                                \
00345         "decl      %0              \n\t"                                \
00346         "jnz 1b                    \n\t"                                \
00347         : "+r"(h), "+r" (src),  "+r" (dst)                              \
00348         : "r"(stride), "m"(rnd)                                         \
00349         : "memory"                                                      \
00350     );                                                                  \
00351 }
00352 
00361 #define MSPEL_FILTER13_8B(NAME, A1, A2, A3, A4, OP, OPNAME)             \
00362 static void                                                             \
00363 OPNAME ## vc1_## NAME ## _mmx(uint8_t *dst, const uint8_t *src,         \
00364                         x86_reg stride, int rnd, x86_reg offset)      \
00365 {                                                                       \
00366     int h = 8;                                                          \
00367     src -= offset;                                                      \
00368     rnd = 32-rnd;                                                       \
00369     __asm__ volatile (                                                      \
00370         LOAD_ROUNDER_MMX("%6")                                          \
00371         "movq      "MANGLE(ff_pw_53)", %%mm5       \n\t"                \
00372         "movq      "MANGLE(ff_pw_18)", %%mm6       \n\t"                \
00373         ".p2align 3                \n\t"                                \
00374         "1:                        \n\t"                                \
00375         MSPEL_FILTER13_CORE(DO_UNPACK, "movd   1", A1, A2, A3, A4)      \
00376         NORMALIZE_MMX("$6")                                             \
00377         TRANSFER_DO_PACK(OP)                                            \
00378         "add       %5, %1          \n\t"                                \
00379         "add       %5, %2          \n\t"                                \
00380         "decl      %0              \n\t"                                \
00381         "jnz 1b                    \n\t"                                \
00382         : "+r"(h), "+r" (src),  "+r" (dst)                              \
00383         : "r"(offset), "r"(3*offset), "g"(stride), "m"(rnd)             \
00384         : "memory"                                                      \
00385     );                                                                  \
00386 }
00387 
00389 MSPEL_FILTER13_8B     (shift1, "0(%1,%4  )", "0(%1,%3,2)", "0(%1,%3  )", "0(%1     )", OP_PUT, put_)
00390 MSPEL_FILTER13_8B     (shift1, "0(%1,%4  )", "0(%1,%3,2)", "0(%1,%3  )", "0(%1     )", OP_AVG, avg_)
00391 MSPEL_FILTER13_VER_16B(shift1, "0(%1,%4  )", "0(%1,%3,2)", "0(%1,%3  )", "0(%1     )")
00392 MSPEL_FILTER13_HOR_16B(shift1, "2*3(%1)", "2*2(%1)", "2*1(%1)", "2*0(%1)", OP_PUT, put_)
00393 MSPEL_FILTER13_HOR_16B(shift1, "2*3(%1)", "2*2(%1)", "2*1(%1)", "2*0(%1)", OP_AVG, avg_)
00394 
00396 MSPEL_FILTER13_8B     (shift3, "0(%1     )", "0(%1,%3  )", "0(%1,%3,2)", "0(%1,%4  )", OP_PUT, put_)
00397 MSPEL_FILTER13_8B     (shift3, "0(%1     )", "0(%1,%3  )", "0(%1,%3,2)", "0(%1,%4  )", OP_AVG, avg_)
00398 MSPEL_FILTER13_VER_16B(shift3, "0(%1     )", "0(%1,%3  )", "0(%1,%3,2)", "0(%1,%4  )")
00399 MSPEL_FILTER13_HOR_16B(shift3, "2*0(%1)", "2*1(%1)", "2*2(%1)", "2*3(%1)", OP_PUT, put_)
00400 MSPEL_FILTER13_HOR_16B(shift3, "2*0(%1)", "2*1(%1)", "2*2(%1)", "2*3(%1)", OP_AVG, avg_)
00401 
00402 typedef void (*vc1_mspel_mc_filter_ver_16bits)(int16_t *dst, const uint8_t *src, x86_reg src_stride, int rnd, int64_t shift);
00403 typedef void (*vc1_mspel_mc_filter_hor_16bits)(uint8_t *dst, x86_reg dst_stride, const int16_t *src, int rnd);
00404 typedef void (*vc1_mspel_mc_filter_8bits)(uint8_t *dst, const uint8_t *src, x86_reg stride, int rnd, x86_reg offset);
00405 
00417 #define VC1_MSPEL_MC(OP)\
00418 static void OP ## vc1_mspel_mc(uint8_t *dst, const uint8_t *src, int stride,\
00419                                int hmode, int vmode, int rnd)\
00420 {\
00421     static const vc1_mspel_mc_filter_ver_16bits vc1_put_shift_ver_16bits[] =\
00422          { NULL, vc1_put_ver_16b_shift1_mmx, vc1_put_ver_16b_shift2_mmx, vc1_put_ver_16b_shift3_mmx };\
00423     static const vc1_mspel_mc_filter_hor_16bits vc1_put_shift_hor_16bits[] =\
00424          { NULL, OP ## vc1_hor_16b_shift1_mmx, OP ## vc1_hor_16b_shift2_mmx, OP ## vc1_hor_16b_shift3_mmx };\
00425     static const vc1_mspel_mc_filter_8bits vc1_put_shift_8bits[] =\
00426          { NULL, OP ## vc1_shift1_mmx, OP ## vc1_shift2_mmx, OP ## vc1_shift3_mmx };\
00427 \
00428     __asm__ volatile(\
00429         "pxor %%mm0, %%mm0         \n\t"\
00430         ::: "memory"\
00431     );\
00432 \
00433     if (vmode) { /* Vertical filter to apply */\
00434         if (hmode) { /* Horizontal filter to apply, output to tmp */\
00435             static const int shift_value[] = { 0, 5, 1, 5 };\
00436             int              shift = (shift_value[hmode]+shift_value[vmode])>>1;\
00437             int              r;\
00438             DECLARE_ALIGNED(16, int16_t, tmp)[12*8];\
00439 \
00440             r = (1<<(shift-1)) + rnd-1;\
00441             vc1_put_shift_ver_16bits[vmode](tmp, src-1, stride, r, shift);\
00442 \
00443             vc1_put_shift_hor_16bits[hmode](dst, stride, tmp+1, 64-rnd);\
00444             return;\
00445         }\
00446         else { /* No horizontal filter, output 8 lines to dst */\
00447             vc1_put_shift_8bits[vmode](dst, src, stride, 1-rnd, stride);\
00448             return;\
00449         }\
00450     }\
00451 \
00452     /* Horizontal mode with no vertical mode */\
00453     vc1_put_shift_8bits[hmode](dst, src, stride, rnd, 1);\
00454 }
00455 
00456 VC1_MSPEL_MC(put_)
00457 VC1_MSPEL_MC(avg_)
00458 
00460 #define DECLARE_FUNCTION(a, b)                                          \
00461 static void put_vc1_mspel_mc ## a ## b ## _mmx(uint8_t *dst, const uint8_t *src, int stride, int rnd) { \
00462      put_vc1_mspel_mc(dst, src, stride, a, b, rnd);                     \
00463 }\
00464 static void avg_vc1_mspel_mc ## a ## b ## _mmx2(uint8_t *dst, const uint8_t *src, int stride, int rnd) { \
00465      avg_vc1_mspel_mc(dst, src, stride, a, b, rnd);                     \
00466 }
00467 
00468 DECLARE_FUNCTION(0, 1)
00469 DECLARE_FUNCTION(0, 2)
00470 DECLARE_FUNCTION(0, 3)
00471 
00472 DECLARE_FUNCTION(1, 0)
00473 DECLARE_FUNCTION(1, 1)
00474 DECLARE_FUNCTION(1, 2)
00475 DECLARE_FUNCTION(1, 3)
00476 
00477 DECLARE_FUNCTION(2, 0)
00478 DECLARE_FUNCTION(2, 1)
00479 DECLARE_FUNCTION(2, 2)
00480 DECLARE_FUNCTION(2, 3)
00481 
00482 DECLARE_FUNCTION(3, 0)
00483 DECLARE_FUNCTION(3, 1)
00484 DECLARE_FUNCTION(3, 2)
00485 DECLARE_FUNCTION(3, 3)
00486 
00487 static void vc1_inv_trans_4x4_dc_mmx2(uint8_t *dest, int linesize, DCTELEM *block)
00488 {
00489     int dc = block[0];
00490     dc = (17 * dc +  4) >> 3;
00491     dc = (17 * dc + 64) >> 7;
00492     __asm__ volatile(
00493         "movd          %0, %%mm0 \n\t"
00494         "pshufw $0, %%mm0, %%mm0 \n\t"
00495         "pxor       %%mm1, %%mm1 \n\t"
00496         "psubw      %%mm0, %%mm1 \n\t"
00497         "packuswb   %%mm0, %%mm0 \n\t"
00498         "packuswb   %%mm1, %%mm1 \n\t"
00499         ::"r"(dc)
00500     );
00501     __asm__ volatile(
00502         "movd          %0, %%mm2 \n\t"
00503         "movd          %1, %%mm3 \n\t"
00504         "movd          %2, %%mm4 \n\t"
00505         "movd          %3, %%mm5 \n\t"
00506         "paddusb    %%mm0, %%mm2 \n\t"
00507         "paddusb    %%mm0, %%mm3 \n\t"
00508         "paddusb    %%mm0, %%mm4 \n\t"
00509         "paddusb    %%mm0, %%mm5 \n\t"
00510         "psubusb    %%mm1, %%mm2 \n\t"
00511         "psubusb    %%mm1, %%mm3 \n\t"
00512         "psubusb    %%mm1, %%mm4 \n\t"
00513         "psubusb    %%mm1, %%mm5 \n\t"
00514         "movd       %%mm2, %0    \n\t"
00515         "movd       %%mm3, %1    \n\t"
00516         "movd       %%mm4, %2    \n\t"
00517         "movd       %%mm5, %3    \n\t"
00518         :"+m"(*(uint32_t*)(dest+0*linesize)),
00519          "+m"(*(uint32_t*)(dest+1*linesize)),
00520          "+m"(*(uint32_t*)(dest+2*linesize)),
00521          "+m"(*(uint32_t*)(dest+3*linesize))
00522     );
00523 }
00524 
00525 static void vc1_inv_trans_4x8_dc_mmx2(uint8_t *dest, int linesize, DCTELEM *block)
00526 {
00527     int dc = block[0];
00528     dc = (17 * dc +  4) >> 3;
00529     dc = (12 * dc + 64) >> 7;
00530     __asm__ volatile(
00531         "movd          %0, %%mm0 \n\t"
00532         "pshufw $0, %%mm0, %%mm0 \n\t"
00533         "pxor       %%mm1, %%mm1 \n\t"
00534         "psubw      %%mm0, %%mm1 \n\t"
00535         "packuswb   %%mm0, %%mm0 \n\t"
00536         "packuswb   %%mm1, %%mm1 \n\t"
00537         ::"r"(dc)
00538     );
00539     __asm__ volatile(
00540         "movd          %0, %%mm2 \n\t"
00541         "movd          %1, %%mm3 \n\t"
00542         "movd          %2, %%mm4 \n\t"
00543         "movd          %3, %%mm5 \n\t"
00544         "paddusb    %%mm0, %%mm2 \n\t"
00545         "paddusb    %%mm0, %%mm3 \n\t"
00546         "paddusb    %%mm0, %%mm4 \n\t"
00547         "paddusb    %%mm0, %%mm5 \n\t"
00548         "psubusb    %%mm1, %%mm2 \n\t"
00549         "psubusb    %%mm1, %%mm3 \n\t"
00550         "psubusb    %%mm1, %%mm4 \n\t"
00551         "psubusb    %%mm1, %%mm5 \n\t"
00552         "movd       %%mm2, %0    \n\t"
00553         "movd       %%mm3, %1    \n\t"
00554         "movd       %%mm4, %2    \n\t"
00555         "movd       %%mm5, %3    \n\t"
00556         :"+m"(*(uint32_t*)(dest+0*linesize)),
00557          "+m"(*(uint32_t*)(dest+1*linesize)),
00558          "+m"(*(uint32_t*)(dest+2*linesize)),
00559          "+m"(*(uint32_t*)(dest+3*linesize))
00560     );
00561     dest += 4*linesize;
00562     __asm__ volatile(
00563         "movd          %0, %%mm2 \n\t"
00564         "movd          %1, %%mm3 \n\t"
00565         "movd          %2, %%mm4 \n\t"
00566         "movd          %3, %%mm5 \n\t"
00567         "paddusb    %%mm0, %%mm2 \n\t"
00568         "paddusb    %%mm0, %%mm3 \n\t"
00569         "paddusb    %%mm0, %%mm4 \n\t"
00570         "paddusb    %%mm0, %%mm5 \n\t"
00571         "psubusb    %%mm1, %%mm2 \n\t"
00572         "psubusb    %%mm1, %%mm3 \n\t"
00573         "psubusb    %%mm1, %%mm4 \n\t"
00574         "psubusb    %%mm1, %%mm5 \n\t"
00575         "movd       %%mm2, %0    \n\t"
00576         "movd       %%mm3, %1    \n\t"
00577         "movd       %%mm4, %2    \n\t"
00578         "movd       %%mm5, %3    \n\t"
00579         :"+m"(*(uint32_t*)(dest+0*linesize)),
00580          "+m"(*(uint32_t*)(dest+1*linesize)),
00581          "+m"(*(uint32_t*)(dest+2*linesize)),
00582          "+m"(*(uint32_t*)(dest+3*linesize))
00583     );
00584 }
00585 
00586 static void vc1_inv_trans_8x4_dc_mmx2(uint8_t *dest, int linesize, DCTELEM *block)
00587 {
00588     int dc = block[0];
00589     dc = ( 3 * dc +  1) >> 1;
00590     dc = (17 * dc + 64) >> 7;
00591     __asm__ volatile(
00592         "movd          %0, %%mm0 \n\t"
00593         "pshufw $0, %%mm0, %%mm0 \n\t"
00594         "pxor       %%mm1, %%mm1 \n\t"
00595         "psubw      %%mm0, %%mm1 \n\t"
00596         "packuswb   %%mm0, %%mm0 \n\t"
00597         "packuswb   %%mm1, %%mm1 \n\t"
00598         ::"r"(dc)
00599     );
00600     __asm__ volatile(
00601         "movq          %0, %%mm2 \n\t"
00602         "movq          %1, %%mm3 \n\t"
00603         "movq          %2, %%mm4 \n\t"
00604         "movq          %3, %%mm5 \n\t"
00605         "paddusb    %%mm0, %%mm2 \n\t"
00606         "paddusb    %%mm0, %%mm3 \n\t"
00607         "paddusb    %%mm0, %%mm4 \n\t"
00608         "paddusb    %%mm0, %%mm5 \n\t"
00609         "psubusb    %%mm1, %%mm2 \n\t"
00610         "psubusb    %%mm1, %%mm3 \n\t"
00611         "psubusb    %%mm1, %%mm4 \n\t"
00612         "psubusb    %%mm1, %%mm5 \n\t"
00613         "movq       %%mm2, %0    \n\t"
00614         "movq       %%mm3, %1    \n\t"
00615         "movq       %%mm4, %2    \n\t"
00616         "movq       %%mm5, %3    \n\t"
00617         :"+m"(*(uint32_t*)(dest+0*linesize)),
00618          "+m"(*(uint32_t*)(dest+1*linesize)),
00619          "+m"(*(uint32_t*)(dest+2*linesize)),
00620          "+m"(*(uint32_t*)(dest+3*linesize))
00621     );
00622 }
00623 
00624 static void vc1_inv_trans_8x8_dc_mmx2(uint8_t *dest, int linesize, DCTELEM *block)
00625 {
00626     int dc = block[0];
00627     dc = (3 * dc +  1) >> 1;
00628     dc = (3 * dc + 16) >> 5;
00629     __asm__ volatile(
00630         "movd          %0, %%mm0 \n\t"
00631         "pshufw $0, %%mm0, %%mm0 \n\t"
00632         "pxor       %%mm1, %%mm1 \n\t"
00633         "psubw      %%mm0, %%mm1 \n\t"
00634         "packuswb   %%mm0, %%mm0 \n\t"
00635         "packuswb   %%mm1, %%mm1 \n\t"
00636         ::"r"(dc)
00637     );
00638     __asm__ volatile(
00639         "movq          %0, %%mm2 \n\t"
00640         "movq          %1, %%mm3 \n\t"
00641         "movq          %2, %%mm4 \n\t"
00642         "movq          %3, %%mm5 \n\t"
00643         "paddusb    %%mm0, %%mm2 \n\t"
00644         "paddusb    %%mm0, %%mm3 \n\t"
00645         "paddusb    %%mm0, %%mm4 \n\t"
00646         "paddusb    %%mm0, %%mm5 \n\t"
00647         "psubusb    %%mm1, %%mm2 \n\t"
00648         "psubusb    %%mm1, %%mm3 \n\t"
00649         "psubusb    %%mm1, %%mm4 \n\t"
00650         "psubusb    %%mm1, %%mm5 \n\t"
00651         "movq       %%mm2, %0    \n\t"
00652         "movq       %%mm3, %1    \n\t"
00653         "movq       %%mm4, %2    \n\t"
00654         "movq       %%mm5, %3    \n\t"
00655         :"+m"(*(uint32_t*)(dest+0*linesize)),
00656          "+m"(*(uint32_t*)(dest+1*linesize)),
00657          "+m"(*(uint32_t*)(dest+2*linesize)),
00658          "+m"(*(uint32_t*)(dest+3*linesize))
00659     );
00660     dest += 4*linesize;
00661     __asm__ volatile(
00662         "movq          %0, %%mm2 \n\t"
00663         "movq          %1, %%mm3 \n\t"
00664         "movq          %2, %%mm4 \n\t"
00665         "movq          %3, %%mm5 \n\t"
00666         "paddusb    %%mm0, %%mm2 \n\t"
00667         "paddusb    %%mm0, %%mm3 \n\t"
00668         "paddusb    %%mm0, %%mm4 \n\t"
00669         "paddusb    %%mm0, %%mm5 \n\t"
00670         "psubusb    %%mm1, %%mm2 \n\t"
00671         "psubusb    %%mm1, %%mm3 \n\t"
00672         "psubusb    %%mm1, %%mm4 \n\t"
00673         "psubusb    %%mm1, %%mm5 \n\t"
00674         "movq       %%mm2, %0    \n\t"
00675         "movq       %%mm3, %1    \n\t"
00676         "movq       %%mm4, %2    \n\t"
00677         "movq       %%mm5, %3    \n\t"
00678         :"+m"(*(uint32_t*)(dest+0*linesize)),
00679          "+m"(*(uint32_t*)(dest+1*linesize)),
00680          "+m"(*(uint32_t*)(dest+2*linesize)),
00681          "+m"(*(uint32_t*)(dest+3*linesize))
00682     );
00683 }
00684 
00685 #define LOOP_FILTER(EXT) \
00686 void ff_vc1_v_loop_filter4_ ## EXT(uint8_t *src, int stride, int pq); \
00687 void ff_vc1_h_loop_filter4_ ## EXT(uint8_t *src, int stride, int pq); \
00688 void ff_vc1_v_loop_filter8_ ## EXT(uint8_t *src, int stride, int pq); \
00689 void ff_vc1_h_loop_filter8_ ## EXT(uint8_t *src, int stride, int pq); \
00690 \
00691 static void vc1_v_loop_filter16_ ## EXT(uint8_t *src, int stride, int pq) \
00692 { \
00693     ff_vc1_v_loop_filter8_ ## EXT(src,   stride, pq); \
00694     ff_vc1_v_loop_filter8_ ## EXT(src+8, stride, pq); \
00695 } \
00696 \
00697 static void vc1_h_loop_filter16_ ## EXT(uint8_t *src, int stride, int pq) \
00698 { \
00699     ff_vc1_h_loop_filter8_ ## EXT(src,          stride, pq); \
00700     ff_vc1_h_loop_filter8_ ## EXT(src+8*stride, stride, pq); \
00701 }
00702 
00703 #if HAVE_YASM
00704 LOOP_FILTER(mmx2)
00705 LOOP_FILTER(sse2)
00706 LOOP_FILTER(ssse3)
00707 
00708 void ff_vc1_h_loop_filter8_sse4(uint8_t *src, int stride, int pq);
00709 
00710 static void vc1_h_loop_filter16_sse4(uint8_t *src, int stride, int pq)
00711 {
00712     ff_vc1_h_loop_filter8_sse4(src,          stride, pq);
00713     ff_vc1_h_loop_filter8_sse4(src+8*stride, stride, pq);
00714 }
00715 
00716 #endif
00717 
00718 void ff_put_vc1_chroma_mc8_mmx_nornd  (uint8_t *dst, uint8_t *src,
00719                                        int stride, int h, int x, int y);
00720 void ff_avg_vc1_chroma_mc8_mmx2_nornd (uint8_t *dst, uint8_t *src,
00721                                        int stride, int h, int x, int y);
00722 void ff_avg_vc1_chroma_mc8_3dnow_nornd(uint8_t *dst, uint8_t *src,
00723                                        int stride, int h, int x, int y);
00724 void ff_put_vc1_chroma_mc8_ssse3_nornd(uint8_t *dst, uint8_t *src,
00725                                        int stride, int h, int x, int y);
00726 void ff_avg_vc1_chroma_mc8_ssse3_nornd(uint8_t *dst, uint8_t *src,
00727                                        int stride, int h, int x, int y);
00728 
00729 void ff_vc1dsp_init_mmx(VC1DSPContext *dsp)
00730 {
00731     int mm_flags = av_get_cpu_flags();
00732 
00733     if (mm_flags & AV_CPU_FLAG_MMX) {
00734         dsp->put_vc1_mspel_pixels_tab[ 0] = ff_put_vc1_mspel_mc00_mmx;
00735         dsp->put_vc1_mspel_pixels_tab[ 4] = put_vc1_mspel_mc01_mmx;
00736         dsp->put_vc1_mspel_pixels_tab[ 8] = put_vc1_mspel_mc02_mmx;
00737         dsp->put_vc1_mspel_pixels_tab[12] = put_vc1_mspel_mc03_mmx;
00738 
00739         dsp->put_vc1_mspel_pixels_tab[ 1] = put_vc1_mspel_mc10_mmx;
00740         dsp->put_vc1_mspel_pixels_tab[ 5] = put_vc1_mspel_mc11_mmx;
00741         dsp->put_vc1_mspel_pixels_tab[ 9] = put_vc1_mspel_mc12_mmx;
00742         dsp->put_vc1_mspel_pixels_tab[13] = put_vc1_mspel_mc13_mmx;
00743 
00744         dsp->put_vc1_mspel_pixels_tab[ 2] = put_vc1_mspel_mc20_mmx;
00745         dsp->put_vc1_mspel_pixels_tab[ 6] = put_vc1_mspel_mc21_mmx;
00746         dsp->put_vc1_mspel_pixels_tab[10] = put_vc1_mspel_mc22_mmx;
00747         dsp->put_vc1_mspel_pixels_tab[14] = put_vc1_mspel_mc23_mmx;
00748 
00749         dsp->put_vc1_mspel_pixels_tab[ 3] = put_vc1_mspel_mc30_mmx;
00750         dsp->put_vc1_mspel_pixels_tab[ 7] = put_vc1_mspel_mc31_mmx;
00751         dsp->put_vc1_mspel_pixels_tab[11] = put_vc1_mspel_mc32_mmx;
00752         dsp->put_vc1_mspel_pixels_tab[15] = put_vc1_mspel_mc33_mmx;
00753 
00754         if (HAVE_YASM)
00755             dsp->put_no_rnd_vc1_chroma_pixels_tab[0]= ff_put_vc1_chroma_mc8_mmx_nornd;
00756     }
00757 
00758     if (mm_flags & AV_CPU_FLAG_MMX2){
00759         dsp->avg_vc1_mspel_pixels_tab[ 0] = ff_avg_vc1_mspel_mc00_mmx2;
00760         dsp->avg_vc1_mspel_pixels_tab[ 4] = avg_vc1_mspel_mc01_mmx2;
00761         dsp->avg_vc1_mspel_pixels_tab[ 8] = avg_vc1_mspel_mc02_mmx2;
00762         dsp->avg_vc1_mspel_pixels_tab[12] = avg_vc1_mspel_mc03_mmx2;
00763 
00764         dsp->avg_vc1_mspel_pixels_tab[ 1] = avg_vc1_mspel_mc10_mmx2;
00765         dsp->avg_vc1_mspel_pixels_tab[ 5] = avg_vc1_mspel_mc11_mmx2;
00766         dsp->avg_vc1_mspel_pixels_tab[ 9] = avg_vc1_mspel_mc12_mmx2;
00767         dsp->avg_vc1_mspel_pixels_tab[13] = avg_vc1_mspel_mc13_mmx2;
00768 
00769         dsp->avg_vc1_mspel_pixels_tab[ 2] = avg_vc1_mspel_mc20_mmx2;
00770         dsp->avg_vc1_mspel_pixels_tab[ 6] = avg_vc1_mspel_mc21_mmx2;
00771         dsp->avg_vc1_mspel_pixels_tab[10] = avg_vc1_mspel_mc22_mmx2;
00772         dsp->avg_vc1_mspel_pixels_tab[14] = avg_vc1_mspel_mc23_mmx2;
00773 
00774         dsp->avg_vc1_mspel_pixels_tab[ 3] = avg_vc1_mspel_mc30_mmx2;
00775         dsp->avg_vc1_mspel_pixels_tab[ 7] = avg_vc1_mspel_mc31_mmx2;
00776         dsp->avg_vc1_mspel_pixels_tab[11] = avg_vc1_mspel_mc32_mmx2;
00777         dsp->avg_vc1_mspel_pixels_tab[15] = avg_vc1_mspel_mc33_mmx2;
00778 
00779         dsp->vc1_inv_trans_8x8_dc = vc1_inv_trans_8x8_dc_mmx2;
00780         dsp->vc1_inv_trans_4x8_dc = vc1_inv_trans_4x8_dc_mmx2;
00781         dsp->vc1_inv_trans_8x4_dc = vc1_inv_trans_8x4_dc_mmx2;
00782         dsp->vc1_inv_trans_4x4_dc = vc1_inv_trans_4x4_dc_mmx2;
00783 
00784         if (HAVE_YASM)
00785             dsp->avg_no_rnd_vc1_chroma_pixels_tab[0]= ff_avg_vc1_chroma_mc8_mmx2_nornd;
00786     } else if (HAVE_YASM && mm_flags & AV_CPU_FLAG_3DNOW) {
00787         dsp->avg_no_rnd_vc1_chroma_pixels_tab[0]= ff_avg_vc1_chroma_mc8_3dnow_nornd;
00788     }
00789 
00790     if (HAVE_YASM && mm_flags & AV_CPU_FLAG_SSSE3) {
00791         dsp->put_no_rnd_vc1_chroma_pixels_tab[0]= ff_put_vc1_chroma_mc8_ssse3_nornd;
00792         dsp->avg_no_rnd_vc1_chroma_pixels_tab[0]= ff_avg_vc1_chroma_mc8_ssse3_nornd;
00793     }
00794 
00795 #define ASSIGN_LF(EXT) \
00796         dsp->vc1_v_loop_filter4  = ff_vc1_v_loop_filter4_ ## EXT; \
00797         dsp->vc1_h_loop_filter4  = ff_vc1_h_loop_filter4_ ## EXT; \
00798         dsp->vc1_v_loop_filter8  = ff_vc1_v_loop_filter8_ ## EXT; \
00799         dsp->vc1_h_loop_filter8  = ff_vc1_h_loop_filter8_ ## EXT; \
00800         dsp->vc1_v_loop_filter16 = vc1_v_loop_filter16_ ## EXT; \
00801         dsp->vc1_h_loop_filter16 = vc1_h_loop_filter16_ ## EXT
00802 
00803 #if HAVE_YASM
00804     if (mm_flags & AV_CPU_FLAG_MMX) {
00805     }
00806     return;
00807     if (mm_flags & AV_CPU_FLAG_MMX2) {
00808         ASSIGN_LF(mmx2);
00809     }
00810 
00811     if (mm_flags & AV_CPU_FLAG_SSE2) {
00812         dsp->vc1_v_loop_filter8  = ff_vc1_v_loop_filter8_sse2;
00813         dsp->vc1_h_loop_filter8  = ff_vc1_h_loop_filter8_sse2;
00814         dsp->vc1_v_loop_filter16 = vc1_v_loop_filter16_sse2;
00815         dsp->vc1_h_loop_filter16 = vc1_h_loop_filter16_sse2;
00816     }
00817     if (mm_flags & AV_CPU_FLAG_SSSE3) {
00818         ASSIGN_LF(ssse3);
00819     }
00820     if (mm_flags & AV_CPU_FLAG_SSE4) {
00821         dsp->vc1_h_loop_filter8  = ff_vc1_h_loop_filter8_sse4;
00822         dsp->vc1_h_loop_filter16 = vc1_h_loop_filter16_sse4;
00823     }
00824 #endif
00825 }
libavcodec/x86/vc1dsp_mmx.c
