2 * Copyright (c) 2000, 2001 Fabrice Bellard
3 * Copyright (c) 2002-2004 Michael Niedermayer <michaelni@gmx.at>
5 * This file is part of FFmpeg.
7 * FFmpeg is free software; you can redistribute it and/or
8 * modify it under the terms of the GNU Lesser General Public
9 * License as published by the Free Software Foundation; either
10 * version 2.1 of the License, or (at your option) any later version.
12 * FFmpeg is distributed in the hope that it will be useful,
13 * but WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 * Lesser General Public License for more details.
17 * You should have received a copy of the GNU Lesser General Public
18 * License along with FFmpeg; if not, write to the Free Software
19 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
23 #include "libavutil/attributes.h"
24 #include "libavutil/cpu.h"
25 #include "libavutil/internal.h"
26 #include "libavutil/x86/asm.h"
27 #include "libavutil/x86/cpu.h"
28 #include "libavcodec/avcodec.h"
29 #include "libavcodec/dsputil.h"
30 #include "libavcodec/pixels.h"
31 #include "libavcodec/simple_idct.h"
32 #include "libavcodec/version.h"
33 #include "dsputil_x86.h"
35 #include "idct_xvid.h"
37 void ff_put_pixels8_l2_mmxext(uint8_t *dst, uint8_t *src1, uint8_t *src2,
38 int dstStride, int src1Stride, int h);
39 void ff_put_no_rnd_pixels8_l2_mmxext(uint8_t *dst, uint8_t *src1,
40 uint8_t *src2, int dstStride,
41 int src1Stride, int h);
42 void ff_avg_pixels8_l2_mmxext(uint8_t *dst, uint8_t *src1, uint8_t *src2,
43 int dstStride, int src1Stride, int h);
44 void ff_put_pixels16_l2_mmxext(uint8_t *dst, uint8_t *src1, uint8_t *src2,
45 int dstStride, int src1Stride, int h);
46 void ff_avg_pixels16_l2_mmxext(uint8_t *dst, uint8_t *src1, uint8_t *src2,
47 int dstStride, int src1Stride, int h);
48 void ff_put_no_rnd_pixels16_l2_mmxext(uint8_t *dst, uint8_t *src1, uint8_t *src2,
49 int dstStride, int src1Stride, int h);
50 void ff_put_mpeg4_qpel16_h_lowpass_mmxext(uint8_t *dst, uint8_t *src,
51 int dstStride, int srcStride, int h);
52 void ff_avg_mpeg4_qpel16_h_lowpass_mmxext(uint8_t *dst, uint8_t *src,
53 int dstStride, int srcStride, int h);
54 void ff_put_no_rnd_mpeg4_qpel16_h_lowpass_mmxext(uint8_t *dst, uint8_t *src,
55 int dstStride, int srcStride,
57 void ff_put_mpeg4_qpel8_h_lowpass_mmxext(uint8_t *dst, uint8_t *src,
58 int dstStride, int srcStride, int h);
59 void ff_avg_mpeg4_qpel8_h_lowpass_mmxext(uint8_t *dst, uint8_t *src,
60 int dstStride, int srcStride, int h);
61 void ff_put_no_rnd_mpeg4_qpel8_h_lowpass_mmxext(uint8_t *dst, uint8_t *src,
62 int dstStride, int srcStride,
64 void ff_put_mpeg4_qpel16_v_lowpass_mmxext(uint8_t *dst, uint8_t *src,
65 int dstStride, int srcStride);
66 void ff_avg_mpeg4_qpel16_v_lowpass_mmxext(uint8_t *dst, uint8_t *src,
67 int dstStride, int srcStride);
68 void ff_put_no_rnd_mpeg4_qpel16_v_lowpass_mmxext(uint8_t *dst, uint8_t *src,
69 int dstStride, int srcStride);
70 void ff_put_mpeg4_qpel8_v_lowpass_mmxext(uint8_t *dst, uint8_t *src,
71 int dstStride, int srcStride);
72 void ff_avg_mpeg4_qpel8_v_lowpass_mmxext(uint8_t *dst, uint8_t *src,
73 int dstStride, int srcStride);
74 void ff_put_no_rnd_mpeg4_qpel8_v_lowpass_mmxext(uint8_t *dst, uint8_t *src,
75 int dstStride, int srcStride);
76 #define ff_put_no_rnd_pixels16_mmxext ff_put_pixels16_mmx
77 #define ff_put_no_rnd_pixels8_mmxext ff_put_pixels8_mmx
79 int32_t ff_scalarproduct_int16_mmxext(const int16_t *v1, const int16_t *v2,
81 int32_t ff_scalarproduct_int16_sse2(const int16_t *v1, const int16_t *v2,
83 int32_t ff_scalarproduct_and_madd_int16_mmxext(int16_t *v1, const int16_t *v2,
86 int32_t ff_scalarproduct_and_madd_int16_sse2(int16_t *v1, const int16_t *v2,
89 int32_t ff_scalarproduct_and_madd_int16_ssse3(int16_t *v1, const int16_t *v2,
93 void ff_bswap32_buf_ssse3(uint32_t *dst, const uint32_t *src, int w);
94 void ff_bswap32_buf_sse2(uint32_t *dst, const uint32_t *src, int w);
96 void ff_add_hfyu_median_prediction_mmxext(uint8_t *dst, const uint8_t *top,
97 const uint8_t *diff, int w,
98 int *left, int *left_top);
99 int ff_add_hfyu_left_prediction_ssse3(uint8_t *dst, const uint8_t *src,
101 int ff_add_hfyu_left_prediction_sse4(uint8_t *dst, const uint8_t *src,
104 void ff_vector_clip_int32_mmx(int32_t *dst, const int32_t *src,
105 int32_t min, int32_t max, unsigned int len);
106 void ff_vector_clip_int32_sse2(int32_t *dst, const int32_t *src,
107 int32_t min, int32_t max, unsigned int len);
108 void ff_vector_clip_int32_int_sse2(int32_t *dst, const int32_t *src,
109 int32_t min, int32_t max, unsigned int len);
110 void ff_vector_clip_int32_sse4(int32_t *dst, const int32_t *src,
111 int32_t min, int32_t max, unsigned int len);
115 #define ff_put_pixels16_mmxext ff_put_pixels16_mmx
116 #define ff_put_pixels8_mmxext ff_put_pixels8_mmx
118 #define QPEL_OP(OPNAME, RND, MMX) \
119 static void OPNAME ## qpel8_mc00_ ## MMX(uint8_t *dst, uint8_t *src, \
122 ff_ ## OPNAME ## pixels8_ ## MMX(dst, src, stride, 8); \
125 static void OPNAME ## qpel8_mc10_ ## MMX(uint8_t *dst, uint8_t *src, \
129 uint8_t *const half = (uint8_t *) temp; \
130 ff_put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(half, src, 8, \
132 ff_ ## OPNAME ## pixels8_l2_ ## MMX(dst, src, half, \
133 stride, stride, 8); \
136 static void OPNAME ## qpel8_mc20_ ## MMX(uint8_t *dst, uint8_t *src, \
139 ff_ ## OPNAME ## mpeg4_qpel8_h_lowpass_ ## MMX(dst, src, stride, \
143 static void OPNAME ## qpel8_mc30_ ## MMX(uint8_t *dst, uint8_t *src, \
147 uint8_t *const half = (uint8_t *) temp; \
148 ff_put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(half, src, 8, \
150 ff_ ## OPNAME ## pixels8_l2_ ## MMX(dst, src + 1, half, stride, \
154 static void OPNAME ## qpel8_mc01_ ## MMX(uint8_t *dst, uint8_t *src, \
158 uint8_t *const half = (uint8_t *) temp; \
159 ff_put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(half, src, \
161 ff_ ## OPNAME ## pixels8_l2_ ## MMX(dst, src, half, \
162 stride, stride, 8); \
165 static void OPNAME ## qpel8_mc02_ ## MMX(uint8_t *dst, uint8_t *src, \
168 ff_ ## OPNAME ## mpeg4_qpel8_v_lowpass_ ## MMX(dst, src, \
172 static void OPNAME ## qpel8_mc03_ ## MMX(uint8_t *dst, uint8_t *src, \
176 uint8_t *const half = (uint8_t *) temp; \
177 ff_put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(half, src, \
179 ff_ ## OPNAME ## pixels8_l2_ ## MMX(dst, src + stride, half, stride,\
183 static void OPNAME ## qpel8_mc11_ ## MMX(uint8_t *dst, uint8_t *src, \
186 uint64_t half[8 + 9]; \
187 uint8_t *const halfH = (uint8_t *) half + 64; \
188 uint8_t *const halfHV = (uint8_t *) half; \
189 ff_put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, \
191 ff_put ## RND ## pixels8_l2_ ## MMX(halfH, src, halfH, 8, \
193 ff_put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(halfHV, halfH, 8, 8);\
194 ff_ ## OPNAME ## pixels8_l2_ ## MMX(dst, halfH, halfHV, \
198 static void OPNAME ## qpel8_mc31_ ## MMX(uint8_t *dst, uint8_t *src, \
201 uint64_t half[8 + 9]; \
202 uint8_t *const halfH = (uint8_t *) half + 64; \
203 uint8_t *const halfHV = (uint8_t *) half; \
204 ff_put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, \
206 ff_put ## RND ## pixels8_l2_ ## MMX(halfH, src + 1, halfH, 8, \
208 ff_put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(halfHV, halfH, 8, 8);\
209 ff_ ## OPNAME ## pixels8_l2_ ## MMX(dst, halfH, halfHV, \
213 static void OPNAME ## qpel8_mc13_ ## MMX(uint8_t *dst, uint8_t *src, \
216 uint64_t half[8 + 9]; \
217 uint8_t *const halfH = (uint8_t *) half + 64; \
218 uint8_t *const halfHV = (uint8_t *) half; \
219 ff_put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, \
221 ff_put ## RND ## pixels8_l2_ ## MMX(halfH, src, halfH, 8, \
223 ff_put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(halfHV, halfH, 8, 8);\
224 ff_ ## OPNAME ## pixels8_l2_ ## MMX(dst, halfH + 8, halfHV, \
228 static void OPNAME ## qpel8_mc33_ ## MMX(uint8_t *dst, uint8_t *src, \
231 uint64_t half[8 + 9]; \
232 uint8_t *const halfH = (uint8_t *) half + 64; \
233 uint8_t *const halfHV = (uint8_t *) half; \
234 ff_put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, \
236 ff_put ## RND ## pixels8_l2_ ## MMX(halfH, src + 1, halfH, 8, \
238 ff_put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(halfHV, halfH, 8, 8);\
239 ff_ ## OPNAME ## pixels8_l2_ ## MMX(dst, halfH + 8, halfHV, \
243 static void OPNAME ## qpel8_mc21_ ## MMX(uint8_t *dst, uint8_t *src, \
246 uint64_t half[8 + 9]; \
247 uint8_t *const halfH = (uint8_t *) half + 64; \
248 uint8_t *const halfHV = (uint8_t *) half; \
249 ff_put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, \
251 ff_put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(halfHV, halfH, 8, 8);\
252 ff_ ## OPNAME ## pixels8_l2_ ## MMX(dst, halfH, halfHV, \
256 static void OPNAME ## qpel8_mc23_ ## MMX(uint8_t *dst, uint8_t *src, \
259 uint64_t half[8 + 9]; \
260 uint8_t *const halfH = (uint8_t *) half + 64; \
261 uint8_t *const halfHV = (uint8_t *) half; \
262 ff_put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, \
264 ff_put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(halfHV, halfH, 8, 8);\
265 ff_ ## OPNAME ## pixels8_l2_ ## MMX(dst, halfH + 8, halfHV, \
269 static void OPNAME ## qpel8_mc12_ ## MMX(uint8_t *dst, uint8_t *src, \
272 uint64_t half[8 + 9]; \
273 uint8_t *const halfH = (uint8_t *) half; \
274 ff_put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, \
276 ff_put ## RND ## pixels8_l2_ ## MMX(halfH, src, halfH, \
278 ff_ ## OPNAME ## mpeg4_qpel8_v_lowpass_ ## MMX(dst, halfH, \
282 static void OPNAME ## qpel8_mc32_ ## MMX(uint8_t *dst, uint8_t *src, \
285 uint64_t half[8 + 9]; \
286 uint8_t *const halfH = (uint8_t *) half; \
287 ff_put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, \
289 ff_put ## RND ## pixels8_l2_ ## MMX(halfH, src + 1, halfH, 8, \
291 ff_ ## OPNAME ## mpeg4_qpel8_v_lowpass_ ## MMX(dst, halfH, \
295 static void OPNAME ## qpel8_mc22_ ## MMX(uint8_t *dst, uint8_t *src, \
299 uint8_t *const halfH = (uint8_t *) half; \
300 ff_put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, \
302 ff_ ## OPNAME ## mpeg4_qpel8_v_lowpass_ ## MMX(dst, halfH, \
306 static void OPNAME ## qpel16_mc00_ ## MMX(uint8_t *dst, uint8_t *src, \
309 ff_ ## OPNAME ## pixels16_ ## MMX(dst, src, stride, 16); \
312 static void OPNAME ## qpel16_mc10_ ## MMX(uint8_t *dst, uint8_t *src, \
316 uint8_t *const half = (uint8_t *) temp; \
317 ff_put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(half, src, 16, \
319 ff_ ## OPNAME ## pixels16_l2_ ## MMX(dst, src, half, stride, \
323 static void OPNAME ## qpel16_mc20_ ## MMX(uint8_t *dst, uint8_t *src, \
326 ff_ ## OPNAME ## mpeg4_qpel16_h_lowpass_ ## MMX(dst, src, \
327 stride, stride, 16);\
330 static void OPNAME ## qpel16_mc30_ ## MMX(uint8_t *dst, uint8_t *src, \
334 uint8_t *const half = (uint8_t*) temp; \
335 ff_put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(half, src, 16, \
337 ff_ ## OPNAME ## pixels16_l2_ ## MMX(dst, src + 1, half, \
338 stride, stride, 16); \
341 static void OPNAME ## qpel16_mc01_ ## MMX(uint8_t *dst, uint8_t *src, \
345 uint8_t *const half = (uint8_t *) temp; \
346 ff_put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(half, src, 16, \
348 ff_ ## OPNAME ## pixels16_l2_ ## MMX(dst, src, half, stride, \
352 static void OPNAME ## qpel16_mc02_ ## MMX(uint8_t *dst, uint8_t *src, \
355 ff_ ## OPNAME ## mpeg4_qpel16_v_lowpass_ ## MMX(dst, src, \
359 static void OPNAME ## qpel16_mc03_ ## MMX(uint8_t *dst, uint8_t *src, \
363 uint8_t *const half = (uint8_t *) temp; \
364 ff_put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(half, src, 16, \
366 ff_ ## OPNAME ## pixels16_l2_ ## MMX(dst, src+stride, half, \
367 stride, stride, 16); \
370 static void OPNAME ## qpel16_mc11_ ## MMX(uint8_t *dst, uint8_t *src, \
373 uint64_t half[16 * 2 + 17 * 2]; \
374 uint8_t *const halfH = (uint8_t *) half + 256; \
375 uint8_t *const halfHV = (uint8_t *) half; \
376 ff_put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16, \
378 ff_put ## RND ## pixels16_l2_ ## MMX(halfH, src, halfH, 16, \
380 ff_put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(halfHV, halfH, \
382 ff_ ## OPNAME ## pixels16_l2_ ## MMX(dst, halfH, halfHV, \
386 static void OPNAME ## qpel16_mc31_ ## MMX(uint8_t *dst, uint8_t *src, \
389 uint64_t half[16 * 2 + 17 * 2]; \
390 uint8_t *const halfH = (uint8_t *) half + 256; \
391 uint8_t *const halfHV = (uint8_t *) half; \
392 ff_put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16, \
394 ff_put ## RND ## pixels16_l2_ ## MMX(halfH, src + 1, halfH, 16, \
396 ff_put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(halfHV, halfH, \
398 ff_ ## OPNAME ## pixels16_l2_ ## MMX(dst, halfH, halfHV, \
402 static void OPNAME ## qpel16_mc13_ ## MMX(uint8_t *dst, uint8_t *src, \
405 uint64_t half[16 * 2 + 17 * 2]; \
406 uint8_t *const halfH = (uint8_t *) half + 256; \
407 uint8_t *const halfHV = (uint8_t *) half; \
408 ff_put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16, \
410 ff_put ## RND ## pixels16_l2_ ## MMX(halfH, src, halfH, 16, \
412 ff_put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(halfHV, halfH, \
414 ff_ ## OPNAME ## pixels16_l2_ ## MMX(dst, halfH + 16, halfHV, \
418 static void OPNAME ## qpel16_mc33_ ## MMX(uint8_t *dst, uint8_t *src, \
421 uint64_t half[16 * 2 + 17 * 2]; \
422 uint8_t *const halfH = (uint8_t *) half + 256; \
423 uint8_t *const halfHV = (uint8_t *) half; \
424 ff_put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16, \
426 ff_put ## RND ## pixels16_l2_ ## MMX(halfH, src + 1, halfH, 16, \
428 ff_put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(halfHV, halfH, \
430 ff_ ## OPNAME ## pixels16_l2_ ## MMX(dst, halfH + 16, halfHV, \
434 static void OPNAME ## qpel16_mc21_ ## MMX(uint8_t *dst, uint8_t *src, \
437 uint64_t half[16 * 2 + 17 * 2]; \
438 uint8_t *const halfH = (uint8_t *) half + 256; \
439 uint8_t *const halfHV = (uint8_t *) half; \
440 ff_put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16, \
442 ff_put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(halfHV, halfH, \
444 ff_ ## OPNAME ## pixels16_l2_ ## MMX(dst, halfH, halfHV, \
448 static void OPNAME ## qpel16_mc23_ ## MMX(uint8_t *dst, uint8_t *src, \
451 uint64_t half[16 * 2 + 17 * 2]; \
452 uint8_t *const halfH = (uint8_t *) half + 256; \
453 uint8_t *const halfHV = (uint8_t *) half; \
454 ff_put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16, \
456 ff_put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(halfHV, halfH, \
458 ff_ ## OPNAME ## pixels16_l2_ ## MMX(dst, halfH + 16, halfHV, \
462 static void OPNAME ## qpel16_mc12_ ## MMX(uint8_t *dst, uint8_t *src, \
465 uint64_t half[17 * 2]; \
466 uint8_t *const halfH = (uint8_t *) half; \
467 ff_put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16, \
469 ff_put ## RND ## pixels16_l2_ ## MMX(halfH, src, halfH, 16, \
471 ff_ ## OPNAME ## mpeg4_qpel16_v_lowpass_ ## MMX(dst, halfH, \
475 static void OPNAME ## qpel16_mc32_ ## MMX(uint8_t *dst, uint8_t *src, \
478 uint64_t half[17 * 2]; \
479 uint8_t *const halfH = (uint8_t *) half; \
480 ff_put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16, \
482 ff_put ## RND ## pixels16_l2_ ## MMX(halfH, src + 1, halfH, 16, \
484 ff_ ## OPNAME ## mpeg4_qpel16_v_lowpass_ ## MMX(dst, halfH, \
488 static void OPNAME ## qpel16_mc22_ ## MMX(uint8_t *dst, uint8_t *src, \
491 uint64_t half[17 * 2]; \
492 uint8_t *const halfH = (uint8_t *) half; \
493 ff_put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16, \
495 ff_ ## OPNAME ## mpeg4_qpel16_v_lowpass_ ## MMX(dst, halfH, \
499 QPEL_OP(put_, _, mmxext)
500 QPEL_OP(avg_, _, mmxext)
501 QPEL_OP(put_no_rnd_, _no_rnd_, mmxext)
503 #endif /* HAVE_YASM */
505 #define SET_QPEL_FUNCS(PFX, IDX, SIZE, CPU, PREFIX) \
507 c->PFX ## _pixels_tab[IDX][ 0] = PREFIX ## PFX ## SIZE ## _mc00_ ## CPU; \
508 c->PFX ## _pixels_tab[IDX][ 1] = PREFIX ## PFX ## SIZE ## _mc10_ ## CPU; \
509 c->PFX ## _pixels_tab[IDX][ 2] = PREFIX ## PFX ## SIZE ## _mc20_ ## CPU; \
510 c->PFX ## _pixels_tab[IDX][ 3] = PREFIX ## PFX ## SIZE ## _mc30_ ## CPU; \
511 c->PFX ## _pixels_tab[IDX][ 4] = PREFIX ## PFX ## SIZE ## _mc01_ ## CPU; \
512 c->PFX ## _pixels_tab[IDX][ 5] = PREFIX ## PFX ## SIZE ## _mc11_ ## CPU; \
513 c->PFX ## _pixels_tab[IDX][ 6] = PREFIX ## PFX ## SIZE ## _mc21_ ## CPU; \
514 c->PFX ## _pixels_tab[IDX][ 7] = PREFIX ## PFX ## SIZE ## _mc31_ ## CPU; \
515 c->PFX ## _pixels_tab[IDX][ 8] = PREFIX ## PFX ## SIZE ## _mc02_ ## CPU; \
516 c->PFX ## _pixels_tab[IDX][ 9] = PREFIX ## PFX ## SIZE ## _mc12_ ## CPU; \
517 c->PFX ## _pixels_tab[IDX][10] = PREFIX ## PFX ## SIZE ## _mc22_ ## CPU; \
518 c->PFX ## _pixels_tab[IDX][11] = PREFIX ## PFX ## SIZE ## _mc32_ ## CPU; \
519 c->PFX ## _pixels_tab[IDX][12] = PREFIX ## PFX ## SIZE ## _mc03_ ## CPU; \
520 c->PFX ## _pixels_tab[IDX][13] = PREFIX ## PFX ## SIZE ## _mc13_ ## CPU; \
521 c->PFX ## _pixels_tab[IDX][14] = PREFIX ## PFX ## SIZE ## _mc23_ ## CPU; \
522 c->PFX ## _pixels_tab[IDX][15] = PREFIX ## PFX ## SIZE ## _mc33_ ## CPU; \
525 static av_cold void dsputil_init_mmx(DSPContext *c, AVCodecContext *avctx,
526 int cpu_flags, unsigned high_bit_depth)
529 c->put_pixels_clamped = ff_put_pixels_clamped_mmx;
530 c->add_pixels_clamped = ff_add_pixels_clamped_mmx;
532 if (!high_bit_depth) {
533 c->draw_edges = ff_draw_edges_mmx;
536 #if CONFIG_VIDEODSP && (ARCH_X86_32 || !HAVE_YASM)
540 c->add_bytes = ff_add_bytes_mmx;
541 #endif /* HAVE_MMX_INLINE */
543 #if HAVE_MMX_EXTERNAL
544 if (!high_bit_depth) {
545 c->clear_block = ff_clear_block_mmx;
546 c->clear_blocks = ff_clear_blocks_mmx;
548 c->vector_clip_int32 = ff_vector_clip_int32_mmx;
549 c->put_signed_pixels_clamped = ff_put_signed_pixels_clamped_mmx;
550 #endif /* HAVE_MMX_EXTERNAL */
553 static av_cold void dsputil_init_mmxext(DSPContext *c, AVCodecContext *avctx,
554 int cpu_flags, unsigned high_bit_depth)
556 #if HAVE_MMXEXT_INLINE
557 if (!high_bit_depth && avctx->idct_algo == FF_IDCT_XVIDMMX && avctx->lowres == 0) {
558 c->idct_put = ff_idct_xvid_mmxext_put;
559 c->idct_add = ff_idct_xvid_mmxext_add;
560 c->idct = ff_idct_xvid_mmxext;
562 #endif /* HAVE_MMXEXT_INLINE */
564 #if HAVE_MMXEXT_EXTERNAL
565 SET_QPEL_FUNCS(avg_qpel, 0, 16, mmxext, );
566 SET_QPEL_FUNCS(avg_qpel, 1, 8, mmxext, );
568 SET_QPEL_FUNCS(put_qpel, 0, 16, mmxext, );
569 SET_QPEL_FUNCS(put_qpel, 1, 8, mmxext, );
570 SET_QPEL_FUNCS(put_no_rnd_qpel, 0, 16, mmxext, );
571 SET_QPEL_FUNCS(put_no_rnd_qpel, 1, 8, mmxext, );
573 /* slower than cmov version on AMD */
574 if (!(cpu_flags & AV_CPU_FLAG_3DNOW))
575 c->add_hfyu_median_prediction = ff_add_hfyu_median_prediction_mmxext;
577 c->scalarproduct_int16 = ff_scalarproduct_int16_mmxext;
578 c->scalarproduct_and_madd_int16 = ff_scalarproduct_and_madd_int16_mmxext;
579 #endif /* HAVE_MMXEXT_EXTERNAL */
582 static av_cold void dsputil_init_sse(DSPContext *c, AVCodecContext *avctx,
583 int cpu_flags, unsigned high_bit_depth)
586 #if HAVE_SSE_EXTERNAL
587 c->vector_clipf = ff_vector_clipf_sse;
589 /* XvMCCreateBlocks() may not allocate 16-byte aligned blocks */
590 if (CONFIG_XVMC && avctx->hwaccel && avctx->hwaccel->decode_mb)
593 if (!high_bit_depth) {
594 c->clear_block = ff_clear_block_sse;
595 c->clear_blocks = ff_clear_blocks_sse;
598 #if HAVE_INLINE_ASM && CONFIG_VIDEODSP
601 #endif /* HAVE_YASM */
604 static av_cold void dsputil_init_sse2(DSPContext *c, AVCodecContext *avctx,
605 int cpu_flags, unsigned high_bit_depth)
608 if (!high_bit_depth && avctx->idct_algo == FF_IDCT_XVIDMMX && avctx->lowres == 0) {
609 c->idct_put = ff_idct_xvid_sse2_put;
610 c->idct_add = ff_idct_xvid_sse2_add;
611 c->idct = ff_idct_xvid_sse2;
612 c->idct_permutation_type = FF_SSE2_IDCT_PERM;
614 #endif /* HAVE_SSE2_INLINE */
616 #if HAVE_SSE2_EXTERNAL
617 c->scalarproduct_int16 = ff_scalarproduct_int16_sse2;
618 c->scalarproduct_and_madd_int16 = ff_scalarproduct_and_madd_int16_sse2;
619 if (cpu_flags & AV_CPU_FLAG_ATOM) {
620 c->vector_clip_int32 = ff_vector_clip_int32_int_sse2;
622 c->vector_clip_int32 = ff_vector_clip_int32_sse2;
624 c->bswap_buf = ff_bswap32_buf_sse2;
625 c->put_signed_pixels_clamped = ff_put_signed_pixels_clamped_sse2;
626 #endif /* HAVE_SSE2_EXTERNAL */
629 static av_cold void dsputil_init_ssse3(DSPContext *c, AVCodecContext *avctx,
630 int cpu_flags, unsigned high_bit_depth)
632 #if HAVE_SSSE3_EXTERNAL
633 c->add_hfyu_left_prediction = ff_add_hfyu_left_prediction_ssse3;
634 if (cpu_flags & AV_CPU_FLAG_SSE4) // not really SSE4, just slow on Conroe
635 c->add_hfyu_left_prediction = ff_add_hfyu_left_prediction_sse4;
637 if (!(cpu_flags & (AV_CPU_FLAG_SSE42 | AV_CPU_FLAG_3DNOW))) // cachesplit
638 c->scalarproduct_and_madd_int16 = ff_scalarproduct_and_madd_int16_ssse3;
639 c->bswap_buf = ff_bswap32_buf_ssse3;
640 #endif /* HAVE_SSSE3_EXTERNAL */
643 static av_cold void dsputil_init_sse4(DSPContext *c, AVCodecContext *avctx,
644 int cpu_flags, unsigned high_bit_depth)
646 #if HAVE_SSE4_EXTERNAL
647 c->vector_clip_int32 = ff_vector_clip_int32_sse4;
648 #endif /* HAVE_SSE4_EXTERNAL */
651 av_cold void ff_dsputil_init_x86(DSPContext *c, AVCodecContext *avctx,
652 unsigned high_bit_depth)
654 int cpu_flags = av_get_cpu_flags();
656 #if HAVE_7REGS && HAVE_INLINE_ASM
657 if (HAVE_MMX && cpu_flags & AV_CPU_FLAG_CMOV)
658 c->add_hfyu_median_prediction = ff_add_hfyu_median_prediction_cmov;
661 if (X86_MMX(cpu_flags)) {
663 const int idct_algo = avctx->idct_algo;
665 if (avctx->lowres == 0 && !high_bit_depth) {
666 if (idct_algo == FF_IDCT_AUTO || idct_algo == FF_IDCT_SIMPLEMMX) {
667 c->idct_put = ff_simple_idct_put_mmx;
668 c->idct_add = ff_simple_idct_add_mmx;
669 c->idct = ff_simple_idct_mmx;
670 c->idct_permutation_type = FF_SIMPLE_IDCT_PERM;
671 } else if (idct_algo == FF_IDCT_XVIDMMX) {
672 c->idct_put = ff_idct_xvid_mmx_put;
673 c->idct_add = ff_idct_xvid_mmx_add;
674 c->idct = ff_idct_xvid_mmx;
677 #endif /* HAVE_INLINE_ASM */
679 dsputil_init_mmx(c, avctx, cpu_flags, high_bit_depth);
682 if (X86_MMXEXT(cpu_flags))
683 dsputil_init_mmxext(c, avctx, cpu_flags, high_bit_depth);
685 if (X86_SSE(cpu_flags))
686 dsputil_init_sse(c, avctx, cpu_flags, high_bit_depth);
688 if (X86_SSE2(cpu_flags))
689 dsputil_init_sse2(c, avctx, cpu_flags, high_bit_depth);
691 if (EXTERNAL_SSSE3(cpu_flags))
692 dsputil_init_ssse3(c, avctx, cpu_flags, high_bit_depth);
694 if (EXTERNAL_SSE4(cpu_flags))
695 dsputil_init_sse4(c, avctx, cpu_flags, high_bit_depth);
698 ff_dsputilenc_init_mmx(c, avctx, high_bit_depth);