PageRenderTime 53ms CodeModel.GetById 14ms app.highlight 30ms RepoModel.GetById 1ms app.codeStats 1ms

/libavcodec/x86/vc1dsp_mmx.c

http://github.com/FFmpeg/FFmpeg
C | 486 lines | 373 code | 34 blank | 79 comment | 5 complexity | 2a17156c4a703b9c7b2825fd05c73597 MD5 | raw file
  1/*
  2 * VC-1 and WMV3 - DSP functions MMX-optimized
  3 * Copyright (c) 2007 Christophe GISQUET <christophe.gisquet@free.fr>
  4 *
  5 * Permission is hereby granted, free of charge, to any person
  6 * obtaining a copy of this software and associated documentation
  7 * files (the "Software"), to deal in the Software without
  8 * restriction, including without limitation the rights to use,
  9 * copy, modify, merge, publish, distribute, sublicense, and/or sell
 10 * copies of the Software, and to permit persons to whom the
 11 * Software is furnished to do so, subject to the following
 12 * conditions:
 13 *
 14 * The above copyright notice and this permission notice shall be
 15 * included in all copies or substantial portions of the Software.
 16 *
 17 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
 18 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES
 19 * OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
 20 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT
 21 * HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY,
 22 * WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
 23 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
 24 * OTHER DEALINGS IN THE SOFTWARE.
 25 */
 26
 27#include "libavutil/cpu.h"
 28#include "libavutil/mem.h"
 29#include "libavutil/x86/asm.h"
 30#include "libavutil/x86/cpu.h"
 31#include "libavcodec/vc1dsp.h"
 32#include "constants.h"
 33#include "fpel.h"
 34#include "vc1dsp.h"
 35
 36#if HAVE_6REGS && HAVE_INLINE_ASM && HAVE_MMX_EXTERNAL
 37
 38void ff_vc1_put_ver_16b_shift2_mmx(int16_t *dst,
 39                                   const uint8_t *src, x86_reg stride,
 40                                   int rnd, int64_t shift);
 41void ff_vc1_put_hor_16b_shift2_mmx(uint8_t *dst, x86_reg stride,
 42                                   const int16_t *src, int rnd);
 43void ff_vc1_avg_hor_16b_shift2_mmxext(uint8_t *dst, x86_reg stride,
 44                                      const int16_t *src, int rnd);
 45
 46#define OP_PUT(S,D)
 47#define OP_AVG(S,D) "pavgb " #S ", " #D " \n\t"
 48
 49/** Add rounder from mm7 to mm3 and pack result at destination */
 50#define NORMALIZE_MMX(SHIFT)                                    \
 51     "paddw     %%mm7, %%mm3           \n\t" /* +bias-r */      \
 52     "paddw     %%mm7, %%mm4           \n\t" /* +bias-r */      \
 53     "psraw     "SHIFT", %%mm3         \n\t"                    \
 54     "psraw     "SHIFT", %%mm4         \n\t"
 55
 56#define TRANSFER_DO_PACK(OP)                    \
 57     "packuswb  %%mm4, %%mm3           \n\t"    \
 58     OP((%2), %%mm3)                            \
 59     "movq      %%mm3, (%2)            \n\t"
 60
 61#define TRANSFER_DONT_PACK(OP)                  \
 62     OP(0(%2), %%mm3)                           \
 63     OP(8(%2), %%mm4)                           \
 64     "movq      %%mm3, 0(%2)           \n\t"    \
 65     "movq      %%mm4, 8(%2)           \n\t"
 66
 67/** @see MSPEL_FILTER13_CORE for use as UNPACK macro */
 68#define DO_UNPACK(reg)  "punpcklbw %%mm0, " reg "\n\t"
 69#define DONT_UNPACK(reg)
 70
 71/** Compute the rounder 32-r or 8-r and unpacks it to mm7 */
 72#define LOAD_ROUNDER_MMX(ROUND)                 \
 73     "movd      "ROUND", %%mm7         \n\t"    \
 74     "punpcklwd %%mm7, %%mm7           \n\t"    \
 75     "punpckldq %%mm7, %%mm7           \n\t"
 76
 77/**
 78 * Purely vertical or horizontal 1/2 shift interpolation.
 79 * Sacrifice mm6 for *9 factor.
 80 */
 81#define VC1_SHIFT2(OP, OPNAME)\
 82static void OPNAME ## vc1_shift2_mmx(uint8_t *dst, const uint8_t *src,\
 83                                     x86_reg stride, int rnd, x86_reg offset)\
 84{\
 85    rnd = 8-rnd;\
 86    __asm__ volatile(\
 87        "mov       $8, %%"FF_REG_c"        \n\t"\
 88        LOAD_ROUNDER_MMX("%5")\
 89        "movq      "MANGLE(ff_pw_9)", %%mm6\n\t"\
 90        "1:                                \n\t"\
 91        "movd      0(%0   ), %%mm3         \n\t"\
 92        "movd      4(%0   ), %%mm4         \n\t"\
 93        "movd      0(%0,%2), %%mm1         \n\t"\
 94        "movd      4(%0,%2), %%mm2         \n\t"\
 95        "add       %2, %0                  \n\t"\
 96        "punpcklbw %%mm0, %%mm3            \n\t"\
 97        "punpcklbw %%mm0, %%mm4            \n\t"\
 98        "punpcklbw %%mm0, %%mm1            \n\t"\
 99        "punpcklbw %%mm0, %%mm2            \n\t"\
100        "paddw     %%mm1, %%mm3            \n\t"\
101        "paddw     %%mm2, %%mm4            \n\t"\
102        "movd      0(%0,%3), %%mm1         \n\t"\
103        "movd      4(%0,%3), %%mm2         \n\t"\
104        "pmullw    %%mm6, %%mm3            \n\t" /* 0,9,9,0*/\
105        "pmullw    %%mm6, %%mm4            \n\t" /* 0,9,9,0*/\
106        "punpcklbw %%mm0, %%mm1            \n\t"\
107        "punpcklbw %%mm0, %%mm2            \n\t"\
108        "psubw     %%mm1, %%mm3            \n\t" /*-1,9,9,0*/\
109        "psubw     %%mm2, %%mm4            \n\t" /*-1,9,9,0*/\
110        "movd      0(%0,%2), %%mm1         \n\t"\
111        "movd      4(%0,%2), %%mm2         \n\t"\
112        "punpcklbw %%mm0, %%mm1            \n\t"\
113        "punpcklbw %%mm0, %%mm2            \n\t"\
114        "psubw     %%mm1, %%mm3            \n\t" /*-1,9,9,-1*/\
115        "psubw     %%mm2, %%mm4            \n\t" /*-1,9,9,-1*/\
116        NORMALIZE_MMX("$4")\
117        "packuswb  %%mm4, %%mm3            \n\t"\
118        OP((%1), %%mm3)\
119        "movq      %%mm3, (%1)             \n\t"\
120        "add       %6, %0                  \n\t"\
121        "add       %4, %1                  \n\t"\
122        "dec       %%"FF_REG_c"            \n\t"\
123        "jnz 1b                            \n\t"\
124        : "+r"(src),  "+r"(dst)\
125        : "r"(offset), "r"(-2*offset), "g"(stride), "m"(rnd),\
126          "g"(stride-offset)\
127          NAMED_CONSTRAINTS_ADD(ff_pw_9)\
128        : "%"FF_REG_c, "memory"\
129    );\
130}
131
132VC1_SHIFT2(OP_PUT, put_)
133VC1_SHIFT2(OP_AVG, avg_)
134
135/**
136 * Core of the 1/4 and 3/4 shift bicubic interpolation.
137 *
138 * @param UNPACK  Macro unpacking arguments from 8 to 16 bits (can be empty).
139 * @param MOVQ    "movd 1" or "movq 2", if data read is already unpacked.
140 * @param A1      Address of 1st tap (beware of unpacked/packed).
141 * @param A2      Address of 2nd tap
142 * @param A3      Address of 3rd tap
143 * @param A4      Address of 4th tap
144 */
145#define MSPEL_FILTER13_CORE(UNPACK, MOVQ, A1, A2, A3, A4)       \
146     MOVQ "*0+"A1", %%mm1       \n\t"                           \
147     MOVQ "*4+"A1", %%mm2       \n\t"                           \
148     UNPACK("%%mm1")                                            \
149     UNPACK("%%mm2")                                            \
150     "pmullw    "MANGLE(ff_pw_3)", %%mm1\n\t"                   \
151     "pmullw    "MANGLE(ff_pw_3)", %%mm2\n\t"                   \
152     MOVQ "*0+"A2", %%mm3       \n\t"                           \
153     MOVQ "*4+"A2", %%mm4       \n\t"                           \
154     UNPACK("%%mm3")                                            \
155     UNPACK("%%mm4")                                            \
156     "pmullw    %%mm6, %%mm3    \n\t" /* *18 */                 \
157     "pmullw    %%mm6, %%mm4    \n\t" /* *18 */                 \
158     "psubw     %%mm1, %%mm3    \n\t" /* 18,-3 */               \
159     "psubw     %%mm2, %%mm4    \n\t" /* 18,-3 */               \
160     MOVQ "*0+"A4", %%mm1       \n\t"                           \
161     MOVQ "*4+"A4", %%mm2       \n\t"                           \
162     UNPACK("%%mm1")                                            \
163     UNPACK("%%mm2")                                            \
164     "psllw     $2, %%mm1       \n\t" /* 4* */                  \
165     "psllw     $2, %%mm2       \n\t" /* 4* */                  \
166     "psubw     %%mm1, %%mm3    \n\t" /* -4,18,-3 */            \
167     "psubw     %%mm2, %%mm4    \n\t" /* -4,18,-3 */            \
168     MOVQ "*0+"A3", %%mm1       \n\t"                           \
169     MOVQ "*4+"A3", %%mm2       \n\t"                           \
170     UNPACK("%%mm1")                                            \
171     UNPACK("%%mm2")                                            \
172     "pmullw    %%mm5, %%mm1    \n\t" /* *53 */                 \
173     "pmullw    %%mm5, %%mm2    \n\t" /* *53 */                 \
174     "paddw     %%mm1, %%mm3    \n\t" /* 4,53,18,-3 */          \
175     "paddw     %%mm2, %%mm4    \n\t" /* 4,53,18,-3 */
176
177/**
178 * Macro to build the vertical 16 bits version of vc1_put_shift[13].
179 * Here, offset=src_stride. Parameters passed A1 to A4 must use
180 * %3 (src_stride) and %4 (3*src_stride).
181 *
182 * @param  NAME   Either 1 or 3
183 * @see MSPEL_FILTER13_CORE for information on A1->A4
184 */
185#define MSPEL_FILTER13_VER_16B(NAME, A1, A2, A3, A4)                    \
186static void                                                             \
187vc1_put_ver_16b_ ## NAME ## _mmx(int16_t *dst, const uint8_t *src,      \
188                                 x86_reg src_stride,                   \
189                                 int rnd, int64_t shift)                \
190{                                                                       \
191    int h = 8;                                                          \
192    src -= src_stride;                                                  \
193    __asm__ volatile(                                                       \
194        LOAD_ROUNDER_MMX("%5")                                          \
195        "movq      "MANGLE(ff_pw_53)", %%mm5\n\t"                       \
196        "movq      "MANGLE(ff_pw_18)", %%mm6\n\t"                       \
197        ".p2align 3                \n\t"                                \
198        "1:                        \n\t"                                \
199        MSPEL_FILTER13_CORE(DO_UNPACK, "movd  1", A1, A2, A3, A4)       \
200        NORMALIZE_MMX("%6")                                             \
201        TRANSFER_DONT_PACK(OP_PUT)                                      \
202        /* Last 3 (in fact 4) bytes on the line */                      \
203        "movd      8+"A1", %%mm1   \n\t"                                \
204        DO_UNPACK("%%mm1")                                              \
205        "movq      %%mm1, %%mm3    \n\t"                                \
206        "paddw     %%mm1, %%mm1    \n\t"                                \
207        "paddw     %%mm3, %%mm1    \n\t" /* 3* */                       \
208        "movd      8+"A2", %%mm3   \n\t"                                \
209        DO_UNPACK("%%mm3")                                              \
210        "pmullw    %%mm6, %%mm3    \n\t" /* *18 */                      \
211        "psubw     %%mm1, %%mm3    \n\t" /*18,-3 */                     \
212        "movd      8+"A3", %%mm1   \n\t"                                \
213        DO_UNPACK("%%mm1")                                              \
214        "pmullw    %%mm5, %%mm1    \n\t" /* *53 */                      \
215        "paddw     %%mm1, %%mm3    \n\t" /*53,18,-3 */                  \
216        "movd      8+"A4", %%mm1   \n\t"                                \
217        DO_UNPACK("%%mm1")                                              \
218        "psllw     $2, %%mm1       \n\t" /* 4* */                       \
219        "psubw     %%mm1, %%mm3    \n\t"                                \
220        "paddw     %%mm7, %%mm3    \n\t"                                \
221        "psraw     %6, %%mm3       \n\t"                                \
222        "movq      %%mm3, 16(%2)   \n\t"                                \
223        "add       %3, %1          \n\t"                                \
224        "add       $24, %2         \n\t"                                \
225        "decl      %0              \n\t"                                \
226        "jnz 1b                    \n\t"                                \
227        : "+r"(h), "+r" (src),  "+r" (dst)                              \
228        : "r"(src_stride), "r"(3*src_stride),                           \
229          "m"(rnd), "m"(shift)                                          \
230          NAMED_CONSTRAINTS_ADD(ff_pw_3,ff_pw_53,ff_pw_18)              \
231        : "memory"                                                      \
232    );                                                                  \
233}
234
235/**
236 * Macro to build the horizontal 16 bits version of vc1_put_shift[13].
237 * Here, offset=16 bits, so parameters passed A1 to A4 should be simple.
238 *
239 * @param  NAME   Either 1 or 3
240 * @see MSPEL_FILTER13_CORE for information on A1->A4
241 */
242#define MSPEL_FILTER13_HOR_16B(NAME, A1, A2, A3, A4, OP, OPNAME)        \
243static void                                                             \
244OPNAME ## vc1_hor_16b_ ## NAME ## _mmx(uint8_t *dst, x86_reg stride,    \
245                                 const int16_t *src, int rnd)           \
246{                                                                       \
247    int h = 8;                                                          \
248    src -= 1;                                                           \
249    rnd -= (-4+58+13-3)*256; /* Add -256 bias */                        \
250    __asm__ volatile(                                                       \
251        LOAD_ROUNDER_MMX("%4")                                          \
252        "movq      "MANGLE(ff_pw_18)", %%mm6   \n\t"                    \
253        "movq      "MANGLE(ff_pw_53)", %%mm5   \n\t"                    \
254        ".p2align 3                \n\t"                                \
255        "1:                        \n\t"                                \
256        MSPEL_FILTER13_CORE(DONT_UNPACK, "movq 2", A1, A2, A3, A4)      \
257        NORMALIZE_MMX("$7")                                             \
258        /* Remove bias */                                               \
259        "paddw     "MANGLE(ff_pw_128)", %%mm3  \n\t"                    \
260        "paddw     "MANGLE(ff_pw_128)", %%mm4  \n\t"                    \
261        TRANSFER_DO_PACK(OP)                                            \
262        "add       $24, %1         \n\t"                                \
263        "add       %3, %2          \n\t"                                \
264        "decl      %0              \n\t"                                \
265        "jnz 1b                    \n\t"                                \
266        : "+r"(h), "+r" (src),  "+r" (dst)                              \
267        : "r"(stride), "m"(rnd)                                         \
268          NAMED_CONSTRAINTS_ADD(ff_pw_3,ff_pw_18,ff_pw_53,ff_pw_128)    \
269        : "memory"                                                      \
270    );                                                                  \
271}
272
273/**
274 * Macro to build the 8 bits, any direction, version of vc1_put_shift[13].
275 * Here, offset=src_stride. Parameters passed A1 to A4 must use
276 * %3 (offset) and %4 (3*offset).
277 *
278 * @param  NAME   Either 1 or 3
279 * @see MSPEL_FILTER13_CORE for information on A1->A4
280 */
281#define MSPEL_FILTER13_8B(NAME, A1, A2, A3, A4, OP, OPNAME)             \
282static void                                                             \
283OPNAME ## vc1_## NAME ## _mmx(uint8_t *dst, const uint8_t *src,         \
284                        x86_reg stride, int rnd, x86_reg offset)      \
285{                                                                       \
286    int h = 8;                                                          \
287    src -= offset;                                                      \
288    rnd = 32-rnd;                                                       \
289    __asm__ volatile (                                                      \
290        LOAD_ROUNDER_MMX("%6")                                          \
291        "movq      "MANGLE(ff_pw_53)", %%mm5       \n\t"                \
292        "movq      "MANGLE(ff_pw_18)", %%mm6       \n\t"                \
293        ".p2align 3                \n\t"                                \
294        "1:                        \n\t"                                \
295        MSPEL_FILTER13_CORE(DO_UNPACK, "movd   1", A1, A2, A3, A4)      \
296        NORMALIZE_MMX("$6")                                             \
297        TRANSFER_DO_PACK(OP)                                            \
298        "add       %5, %1          \n\t"                                \
299        "add       %5, %2          \n\t"                                \
300        "decl      %0              \n\t"                                \
301        "jnz 1b                    \n\t"                                \
302        : "+r"(h), "+r" (src),  "+r" (dst)                              \
303        : "r"(offset), "r"(3*offset), "g"(stride), "m"(rnd)             \
304          NAMED_CONSTRAINTS_ADD(ff_pw_53,ff_pw_18,ff_pw_3)              \
305        : "memory"                                                      \
306    );                                                                  \
307}
308
309/** 1/4 shift bicubic interpolation */
310MSPEL_FILTER13_8B     (shift1, "0(%1,%4  )", "0(%1,%3,2)", "0(%1,%3  )", "0(%1     )", OP_PUT, put_)
311MSPEL_FILTER13_8B     (shift1, "0(%1,%4  )", "0(%1,%3,2)", "0(%1,%3  )", "0(%1     )", OP_AVG, avg_)
312MSPEL_FILTER13_VER_16B(shift1, "0(%1,%4  )", "0(%1,%3,2)", "0(%1,%3  )", "0(%1     )")
313MSPEL_FILTER13_HOR_16B(shift1, "2*3(%1)", "2*2(%1)", "2*1(%1)", "2*0(%1)", OP_PUT, put_)
314MSPEL_FILTER13_HOR_16B(shift1, "2*3(%1)", "2*2(%1)", "2*1(%1)", "2*0(%1)", OP_AVG, avg_)
315
316/** 3/4 shift bicubic interpolation */
317MSPEL_FILTER13_8B     (shift3, "0(%1     )", "0(%1,%3  )", "0(%1,%3,2)", "0(%1,%4  )", OP_PUT, put_)
318MSPEL_FILTER13_8B     (shift3, "0(%1     )", "0(%1,%3  )", "0(%1,%3,2)", "0(%1,%4  )", OP_AVG, avg_)
319MSPEL_FILTER13_VER_16B(shift3, "0(%1     )", "0(%1,%3  )", "0(%1,%3,2)", "0(%1,%4  )")
320MSPEL_FILTER13_HOR_16B(shift3, "2*0(%1)", "2*1(%1)", "2*2(%1)", "2*3(%1)", OP_PUT, put_)
321MSPEL_FILTER13_HOR_16B(shift3, "2*0(%1)", "2*1(%1)", "2*2(%1)", "2*3(%1)", OP_AVG, avg_)
322
323typedef void (*vc1_mspel_mc_filter_ver_16bits)(int16_t *dst, const uint8_t *src, x86_reg src_stride, int rnd, int64_t shift);
324typedef void (*vc1_mspel_mc_filter_hor_16bits)(uint8_t *dst, x86_reg dst_stride, const int16_t *src, int rnd);
325typedef void (*vc1_mspel_mc_filter_8bits)(uint8_t *dst, const uint8_t *src, x86_reg stride, int rnd, x86_reg offset);
326
327/**
328 * Interpolate fractional pel values by applying proper vertical then
329 * horizontal filter.
330 *
331 * @param  dst     Destination buffer for interpolated pels.
332 * @param  src     Source buffer.
333 * @param  stride  Stride for both src and dst buffers.
334 * @param  hmode   Horizontal filter (expressed in quarter pixels shift).
335 * @param  hmode   Vertical filter.
336 * @param  rnd     Rounding bias.
337 */
338#define VC1_MSPEL_MC(OP, INSTR)\
339static void OP ## vc1_mspel_mc(uint8_t *dst, const uint8_t *src, int stride,\
340                               int hmode, int vmode, int rnd)\
341{\
342    static const vc1_mspel_mc_filter_ver_16bits vc1_put_shift_ver_16bits[] =\
343         { NULL, vc1_put_ver_16b_shift1_mmx, ff_vc1_put_ver_16b_shift2_mmx, vc1_put_ver_16b_shift3_mmx };\
344    static const vc1_mspel_mc_filter_hor_16bits vc1_put_shift_hor_16bits[] =\
345         { NULL, OP ## vc1_hor_16b_shift1_mmx, ff_vc1_ ## OP ## hor_16b_shift2_ ## INSTR, OP ## vc1_hor_16b_shift3_mmx };\
346    static const vc1_mspel_mc_filter_8bits vc1_put_shift_8bits[] =\
347         { NULL, OP ## vc1_shift1_mmx, OP ## vc1_shift2_mmx, OP ## vc1_shift3_mmx };\
348\
349    __asm__ volatile(\
350        "pxor %%mm0, %%mm0         \n\t"\
351        ::: "memory"\
352    );\
353\
354    if (vmode) { /* Vertical filter to apply */\
355        if (hmode) { /* Horizontal filter to apply, output to tmp */\
356            static const int shift_value[] = { 0, 5, 1, 5 };\
357            int              shift = (shift_value[hmode]+shift_value[vmode])>>1;\
358            int              r;\
359            LOCAL_ALIGNED(16, int16_t, tmp, [12*8]);\
360\
361            r = (1<<(shift-1)) + rnd-1;\
362            vc1_put_shift_ver_16bits[vmode](tmp, src-1, stride, r, shift);\
363\
364            vc1_put_shift_hor_16bits[hmode](dst, stride, tmp+1, 64-rnd);\
365            return;\
366        }\
367        else { /* No horizontal filter, output 8 lines to dst */\
368            vc1_put_shift_8bits[vmode](dst, src, stride, 1-rnd, stride);\
369            return;\
370        }\
371    }\
372\
373    /* Horizontal mode with no vertical mode */\
374    vc1_put_shift_8bits[hmode](dst, src, stride, rnd, 1);\
375} \
376static void OP ## vc1_mspel_mc_16(uint8_t *dst, const uint8_t *src, \
377                                  int stride, int hmode, int vmode, int rnd)\
378{ \
379    OP ## vc1_mspel_mc(dst + 0, src + 0, stride, hmode, vmode, rnd); \
380    OP ## vc1_mspel_mc(dst + 8, src + 8, stride, hmode, vmode, rnd); \
381    dst += 8*stride; src += 8*stride; \
382    OP ## vc1_mspel_mc(dst + 0, src + 0, stride, hmode, vmode, rnd); \
383    OP ## vc1_mspel_mc(dst + 8, src + 8, stride, hmode, vmode, rnd); \
384}
385
386VC1_MSPEL_MC(put_, mmx)
387VC1_MSPEL_MC(avg_, mmxext)
388
389/** Macro to ease bicubic filter interpolation functions declarations */
390#define DECLARE_FUNCTION(a, b)                                          \
391static void put_vc1_mspel_mc ## a ## b ## _mmx(uint8_t *dst,            \
392                                               const uint8_t *src,      \
393                                               ptrdiff_t stride,        \
394                                               int rnd)                 \
395{                                                                       \
396     put_vc1_mspel_mc(dst, src, stride, a, b, rnd);                     \
397}\
398static void avg_vc1_mspel_mc ## a ## b ## _mmxext(uint8_t *dst,         \
399                                                  const uint8_t *src,   \
400                                                  ptrdiff_t stride,     \
401                                                  int rnd)              \
402{                                                                       \
403     avg_vc1_mspel_mc(dst, src, stride, a, b, rnd);                     \
404}\
405static void put_vc1_mspel_mc ## a ## b ## _16_mmx(uint8_t *dst,         \
406                                                  const uint8_t *src,   \
407                                                  ptrdiff_t stride,     \
408                                                  int rnd)              \
409{                                                                       \
410     put_vc1_mspel_mc_16(dst, src, stride, a, b, rnd);                  \
411}\
412static void avg_vc1_mspel_mc ## a ## b ## _16_mmxext(uint8_t *dst,      \
413                                                     const uint8_t *src,\
414                                                     ptrdiff_t stride,  \
415                                                     int rnd)           \
416{                                                                       \
417     avg_vc1_mspel_mc_16(dst, src, stride, a, b, rnd);                  \
418}
419
420DECLARE_FUNCTION(0, 1)
421DECLARE_FUNCTION(0, 2)
422DECLARE_FUNCTION(0, 3)
423
424DECLARE_FUNCTION(1, 0)
425DECLARE_FUNCTION(1, 1)
426DECLARE_FUNCTION(1, 2)
427DECLARE_FUNCTION(1, 3)
428
429DECLARE_FUNCTION(2, 0)
430DECLARE_FUNCTION(2, 1)
431DECLARE_FUNCTION(2, 2)
432DECLARE_FUNCTION(2, 3)
433
434DECLARE_FUNCTION(3, 0)
435DECLARE_FUNCTION(3, 1)
436DECLARE_FUNCTION(3, 2)
437DECLARE_FUNCTION(3, 3)
438
439#define FN_ASSIGN(OP, X, Y, INSN) \
440    dsp->OP##vc1_mspel_pixels_tab[1][X+4*Y] = OP##vc1_mspel_mc##X##Y##INSN; \
441    dsp->OP##vc1_mspel_pixels_tab[0][X+4*Y] = OP##vc1_mspel_mc##X##Y##_16##INSN
442
443av_cold void ff_vc1dsp_init_mmx(VC1DSPContext *dsp)
444{
445    FN_ASSIGN(put_, 0, 1, _mmx);
446    FN_ASSIGN(put_, 0, 2, _mmx);
447    FN_ASSIGN(put_, 0, 3, _mmx);
448
449    FN_ASSIGN(put_, 1, 0, _mmx);
450    FN_ASSIGN(put_, 1, 1, _mmx);
451    FN_ASSIGN(put_, 1, 2, _mmx);
452    FN_ASSIGN(put_, 1, 3, _mmx);
453
454    FN_ASSIGN(put_, 2, 0, _mmx);
455    FN_ASSIGN(put_, 2, 1, _mmx);
456    FN_ASSIGN(put_, 2, 2, _mmx);
457    FN_ASSIGN(put_, 2, 3, _mmx);
458
459    FN_ASSIGN(put_, 3, 0, _mmx);
460    FN_ASSIGN(put_, 3, 1, _mmx);
461    FN_ASSIGN(put_, 3, 2, _mmx);
462    FN_ASSIGN(put_, 3, 3, _mmx);
463}
464
465av_cold void ff_vc1dsp_init_mmxext(VC1DSPContext *dsp)
466{
467    FN_ASSIGN(avg_, 0, 1, _mmxext);
468    FN_ASSIGN(avg_, 0, 2, _mmxext);
469    FN_ASSIGN(avg_, 0, 3, _mmxext);
470
471    FN_ASSIGN(avg_, 1, 0, _mmxext);
472    FN_ASSIGN(avg_, 1, 1, _mmxext);
473    FN_ASSIGN(avg_, 1, 2, _mmxext);
474    FN_ASSIGN(avg_, 1, 3, _mmxext);
475
476    FN_ASSIGN(avg_, 2, 0, _mmxext);
477    FN_ASSIGN(avg_, 2, 1, _mmxext);
478    FN_ASSIGN(avg_, 2, 2, _mmxext);
479    FN_ASSIGN(avg_, 2, 3, _mmxext);
480
481    FN_ASSIGN(avg_, 3, 0, _mmxext);
482    FN_ASSIGN(avg_, 3, 1, _mmxext);
483    FN_ASSIGN(avg_, 3, 2, _mmxext);
484    FN_ASSIGN(avg_, 3, 3, _mmxext);
485}
486#endif /* HAVE_6REGS && HAVE_INLINE_ASM && HAVE_MMX_EXTERNAL */