2 * MMX optimized DSP utils
3 * Copyright (c) 2000, 2001 Fabrice Bellard
4 * Copyright (c) 2002-2004 Michael Niedermayer <michaelni@gmx.at>
6 * This file is part of FFmpeg.
8 * FFmpeg is free software; you can redistribute it and/or
9 * modify it under the terms of the GNU Lesser General Public
10 * License as published by the Free Software Foundation; either
11 * version 2.1 of the License, or (at your option) any later version.
13 * FFmpeg is distributed in the hope that it will be useful,
14 * but WITHOUT ANY WARRANTY; without even the implied warranty of
15 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
16 * Lesser General Public License for more details.
18 * You should have received a copy of the GNU Lesser General Public
19 * License along with FFmpeg; if not, write to the Free Software
20 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
22 * MMX optimization by Nick Kurshev <nickols_k@mail.ru>
25 #include "libavutil/cpu.h"
26 #include "libavutil/x86_cpu.h"
27 #include "libavcodec/dsputil.h"
28 #include "libavcodec/h264dsp.h"
29 #include "libavcodec/mpegvideo.h"
30 #include "libavcodec/simple_idct.h"
31 #include "libavcodec/ac3dec.h"
32 #include "dsputil_mmx.h"
33 #include "idct_xvid.h"
38 /* pixel operations */
39 DECLARE_ALIGNED(8, const uint64_t, ff_bone) = 0x0101010101010101ULL;
40 DECLARE_ALIGNED(8, const uint64_t, ff_wtwo) = 0x0002000200020002ULL;
42 DECLARE_ALIGNED(16, const uint64_t, ff_pdw_80000000)[2] =
43 {0x8000000080000000ULL, 0x8000000080000000ULL};
45 DECLARE_ALIGNED(16, const xmm_reg, ff_pw_1 ) = {0x0001000100010001ULL, 0x0001000100010001ULL};
46 DECLARE_ALIGNED(16, const xmm_reg, ff_pw_2 ) = {0x0002000200020002ULL, 0x0002000200020002ULL};
47 DECLARE_ALIGNED(16, const xmm_reg, ff_pw_3 ) = {0x0003000300030003ULL, 0x0003000300030003ULL};
48 DECLARE_ALIGNED(16, const xmm_reg, ff_pw_4 ) = {0x0004000400040004ULL, 0x0004000400040004ULL};
49 DECLARE_ALIGNED(16, const xmm_reg, ff_pw_5 ) = {0x0005000500050005ULL, 0x0005000500050005ULL};
50 DECLARE_ALIGNED(16, const xmm_reg, ff_pw_8 ) = {0x0008000800080008ULL, 0x0008000800080008ULL};
51 DECLARE_ALIGNED(16, const xmm_reg, ff_pw_9 ) = {0x0009000900090009ULL, 0x0009000900090009ULL};
52 DECLARE_ALIGNED(8, const uint64_t, ff_pw_15 ) = 0x000F000F000F000FULL;
53 DECLARE_ALIGNED(16, const xmm_reg, ff_pw_16 ) = {0x0010001000100010ULL, 0x0010001000100010ULL};
54 DECLARE_ALIGNED(16, const xmm_reg, ff_pw_17 ) = {0x0011001100110011ULL, 0x0011001100110011ULL};
55 DECLARE_ALIGNED(16, const xmm_reg, ff_pw_18 ) = {0x0012001200120012ULL, 0x0012001200120012ULL};
56 DECLARE_ALIGNED(8, const uint64_t, ff_pw_20 ) = 0x0014001400140014ULL;
57 DECLARE_ALIGNED(16, const xmm_reg, ff_pw_27 ) = {0x001B001B001B001BULL, 0x001B001B001B001BULL};
58 DECLARE_ALIGNED(16, const xmm_reg, ff_pw_28 ) = {0x001C001C001C001CULL, 0x001C001C001C001CULL};
59 DECLARE_ALIGNED(16, const xmm_reg, ff_pw_32 ) = {0x0020002000200020ULL, 0x0020002000200020ULL};
60 DECLARE_ALIGNED(8, const uint64_t, ff_pw_42 ) = 0x002A002A002A002AULL;
61 DECLARE_ALIGNED(8, const uint64_t, ff_pw_53 ) = 0x0035003500350035ULL;
62 DECLARE_ALIGNED(16, const xmm_reg, ff_pw_63 ) = {0x003F003F003F003FULL, 0x003F003F003F003FULL};
63 DECLARE_ALIGNED(16, const xmm_reg, ff_pw_64 ) = {0x0040004000400040ULL, 0x0040004000400040ULL};
64 DECLARE_ALIGNED(8, const uint64_t, ff_pw_96 ) = 0x0060006000600060ULL;
65 DECLARE_ALIGNED(8, const uint64_t, ff_pw_128) = 0x0080008000800080ULL;
66 DECLARE_ALIGNED(8, const uint64_t, ff_pw_255) = 0x00ff00ff00ff00ffULL;
68 DECLARE_ALIGNED(16, const xmm_reg, ff_pb_0 ) = {0x0000000000000000ULL, 0x0000000000000000ULL};
69 DECLARE_ALIGNED(16, const xmm_reg, ff_pb_1 ) = {0x0101010101010101ULL, 0x0101010101010101ULL};
70 DECLARE_ALIGNED(16, const xmm_reg, ff_pb_3 ) = {0x0303030303030303ULL, 0x0303030303030303ULL};
71 DECLARE_ALIGNED(16, const xmm_reg, ff_pb_4 ) = {0x0404040404040404ULL, 0x0404040404040404ULL};
72 DECLARE_ALIGNED(8, const uint64_t, ff_pb_7 ) = 0x0707070707070707ULL;
73 DECLARE_ALIGNED(8, const uint64_t, ff_pb_1F ) = 0x1F1F1F1F1F1F1F1FULL;
74 DECLARE_ALIGNED(8, const uint64_t, ff_pb_3F ) = 0x3F3F3F3F3F3F3F3FULL;
75 DECLARE_ALIGNED(16, const xmm_reg, ff_pb_80 ) = {0x8080808080808080ULL, 0x8080808080808080ULL};
76 DECLARE_ALIGNED(8, const uint64_t, ff_pb_81 ) = 0x8181818181818181ULL;
77 DECLARE_ALIGNED(16, const xmm_reg, ff_pb_A1 ) = {0xA1A1A1A1A1A1A1A1ULL, 0xA1A1A1A1A1A1A1A1ULL};
78 DECLARE_ALIGNED(16, const xmm_reg, ff_pb_F8 ) = {0xF8F8F8F8F8F8F8F8ULL, 0xF8F8F8F8F8F8F8F8ULL};
79 DECLARE_ALIGNED(8, const uint64_t, ff_pb_FC ) = 0xFCFCFCFCFCFCFCFCULL;
80 DECLARE_ALIGNED(16, const xmm_reg, ff_pb_FE ) = {0xFEFEFEFEFEFEFEFEULL, 0xFEFEFEFEFEFEFEFEULL};
82 DECLARE_ALIGNED(16, const double, ff_pd_1)[2] = { 1.0, 1.0 };
83 DECLARE_ALIGNED(16, const double, ff_pd_2)[2] = { 2.0, 2.0 };
85 #define JUMPALIGN() __asm__ volatile (".p2align 3"::)
86 #define MOVQ_ZERO(regd) __asm__ volatile ("pxor %%" #regd ", %%" #regd ::)
88 #define MOVQ_BFE(regd) \
90 "pcmpeqd %%" #regd ", %%" #regd " \n\t"\
91 "paddb %%" #regd ", %%" #regd " \n\t" ::)
94 #define MOVQ_BONE(regd) __asm__ volatile ("movq %0, %%" #regd " \n\t" ::"m"(ff_bone))
95 #define MOVQ_WTWO(regd) __asm__ volatile ("movq %0, %%" #regd " \n\t" ::"m"(ff_wtwo))
97 // for shared library it's better to use this way for accessing constants
99 #define MOVQ_BONE(regd) \
101 "pcmpeqd %%" #regd ", %%" #regd " \n\t" \
102 "psrlw $15, %%" #regd " \n\t" \
103 "packuswb %%" #regd ", %%" #regd " \n\t" ::)
105 #define MOVQ_WTWO(regd) \
107 "pcmpeqd %%" #regd ", %%" #regd " \n\t" \
108 "psrlw $15, %%" #regd " \n\t" \
109 "psllw $1, %%" #regd " \n\t"::)
113 // using regr as temporary and for the output result
114 // first argument is unmodifed and second is trashed
115 // regfe is supposed to contain 0xfefefefefefefefe
116 #define PAVGB_MMX_NO_RND(rega, regb, regr, regfe) \
117 "movq " #rega ", " #regr " \n\t"\
118 "pand " #regb ", " #regr " \n\t"\
119 "pxor " #rega ", " #regb " \n\t"\
120 "pand " #regfe "," #regb " \n\t"\
121 "psrlq $1, " #regb " \n\t"\
122 "paddb " #regb ", " #regr " \n\t"
124 #define PAVGB_MMX(rega, regb, regr, regfe) \
125 "movq " #rega ", " #regr " \n\t"\
126 "por " #regb ", " #regr " \n\t"\
127 "pxor " #rega ", " #regb " \n\t"\
128 "pand " #regfe "," #regb " \n\t"\
129 "psrlq $1, " #regb " \n\t"\
130 "psubb " #regb ", " #regr " \n\t"
132 // mm6 is supposed to contain 0xfefefefefefefefe
133 #define PAVGBP_MMX_NO_RND(rega, regb, regr, regc, regd, regp) \
134 "movq " #rega ", " #regr " \n\t"\
135 "movq " #regc ", " #regp " \n\t"\
136 "pand " #regb ", " #regr " \n\t"\
137 "pand " #regd ", " #regp " \n\t"\
138 "pxor " #rega ", " #regb " \n\t"\
139 "pxor " #regc ", " #regd " \n\t"\
140 "pand %%mm6, " #regb " \n\t"\
141 "pand %%mm6, " #regd " \n\t"\
142 "psrlq $1, " #regb " \n\t"\
143 "psrlq $1, " #regd " \n\t"\
144 "paddb " #regb ", " #regr " \n\t"\
145 "paddb " #regd ", " #regp " \n\t"
147 #define PAVGBP_MMX(rega, regb, regr, regc, regd, regp) \
148 "movq " #rega ", " #regr " \n\t"\
149 "movq " #regc ", " #regp " \n\t"\
150 "por " #regb ", " #regr " \n\t"\
151 "por " #regd ", " #regp " \n\t"\
152 "pxor " #rega ", " #regb " \n\t"\
153 "pxor " #regc ", " #regd " \n\t"\
154 "pand %%mm6, " #regb " \n\t"\
155 "pand %%mm6, " #regd " \n\t"\
156 "psrlq $1, " #regd " \n\t"\
157 "psrlq $1, " #regb " \n\t"\
158 "psubb " #regb ", " #regr " \n\t"\
159 "psubb " #regd ", " #regp " \n\t"
161 /***********************************/
162 /* MMX no rounding */
163 #define DEF(x, y) x ## _no_rnd_ ## y ##_mmx
164 #define SET_RND MOVQ_WONE
165 #define PAVGBP(a, b, c, d, e, f) PAVGBP_MMX_NO_RND(a, b, c, d, e, f)
166 #define PAVGB(a, b, c, e) PAVGB_MMX_NO_RND(a, b, c, e)
167 #define OP_AVG(a, b, c, e) PAVGB_MMX(a, b, c, e)
169 #include "dsputil_mmx_rnd_template.c"
175 /***********************************/
178 #define DEF(x, y) x ## _ ## y ##_mmx
179 #define SET_RND MOVQ_WTWO
180 #define PAVGBP(a, b, c, d, e, f) PAVGBP_MMX(a, b, c, d, e, f)
181 #define PAVGB(a, b, c, e) PAVGB_MMX(a, b, c, e)
183 #include "dsputil_mmx_rnd_template.c"
191 /***********************************/
194 #define DEF(x) x ## _3dnow
195 #define PAVGB "pavgusb"
198 #include "dsputil_mmx_avg_template.c"
204 /***********************************/
207 #define DEF(x) x ## _mmx2
209 /* Introduced only in MMX2 set */
210 #define PAVGB "pavgb"
213 #include "dsputil_mmx_avg_template.c"
219 #define put_no_rnd_pixels16_mmx put_pixels16_mmx
220 #define put_no_rnd_pixels8_mmx put_pixels8_mmx
221 #define put_pixels16_mmx2 put_pixels16_mmx
222 #define put_pixels8_mmx2 put_pixels8_mmx
223 #define put_pixels4_mmx2 put_pixels4_mmx
224 #define put_no_rnd_pixels16_mmx2 put_no_rnd_pixels16_mmx
225 #define put_no_rnd_pixels8_mmx2 put_no_rnd_pixels8_mmx
226 #define put_pixels16_3dnow put_pixels16_mmx
227 #define put_pixels8_3dnow put_pixels8_mmx
228 #define put_pixels4_3dnow put_pixels4_mmx
229 #define put_no_rnd_pixels16_3dnow put_no_rnd_pixels16_mmx
230 #define put_no_rnd_pixels8_3dnow put_no_rnd_pixels8_mmx
232 /***********************************/
235 void ff_put_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels, int line_size)
240 /* read the pixels */
245 "movq %3, %%mm0 \n\t"
246 "movq 8%3, %%mm1 \n\t"
247 "movq 16%3, %%mm2 \n\t"
248 "movq 24%3, %%mm3 \n\t"
249 "movq 32%3, %%mm4 \n\t"
250 "movq 40%3, %%mm5 \n\t"
251 "movq 48%3, %%mm6 \n\t"
252 "movq 56%3, %%mm7 \n\t"
253 "packuswb %%mm1, %%mm0 \n\t"
254 "packuswb %%mm3, %%mm2 \n\t"
255 "packuswb %%mm5, %%mm4 \n\t"
256 "packuswb %%mm7, %%mm6 \n\t"
257 "movq %%mm0, (%0) \n\t"
258 "movq %%mm2, (%0, %1) \n\t"
259 "movq %%mm4, (%0, %1, 2) \n\t"
260 "movq %%mm6, (%0, %2) \n\t"
261 ::"r" (pix), "r" ((x86_reg)line_size), "r" ((x86_reg)line_size*3), "m"(*p)
266 // if here would be an exact copy of the code above
267 // compiler would generate some very strange code
270 "movq (%3), %%mm0 \n\t"
271 "movq 8(%3), %%mm1 \n\t"
272 "movq 16(%3), %%mm2 \n\t"
273 "movq 24(%3), %%mm3 \n\t"
274 "movq 32(%3), %%mm4 \n\t"
275 "movq 40(%3), %%mm5 \n\t"
276 "movq 48(%3), %%mm6 \n\t"
277 "movq 56(%3), %%mm7 \n\t"
278 "packuswb %%mm1, %%mm0 \n\t"
279 "packuswb %%mm3, %%mm2 \n\t"
280 "packuswb %%mm5, %%mm4 \n\t"
281 "packuswb %%mm7, %%mm6 \n\t"
282 "movq %%mm0, (%0) \n\t"
283 "movq %%mm2, (%0, %1) \n\t"
284 "movq %%mm4, (%0, %1, 2) \n\t"
285 "movq %%mm6, (%0, %2) \n\t"
286 ::"r" (pix), "r" ((x86_reg)line_size), "r" ((x86_reg)line_size*3), "r"(p)
290 #define put_signed_pixels_clamped_mmx_half(off) \
291 "movq "#off"(%2), %%mm1 \n\t"\
292 "movq 16+"#off"(%2), %%mm2 \n\t"\
293 "movq 32+"#off"(%2), %%mm3 \n\t"\
294 "movq 48+"#off"(%2), %%mm4 \n\t"\
295 "packsswb 8+"#off"(%2), %%mm1 \n\t"\
296 "packsswb 24+"#off"(%2), %%mm2 \n\t"\
297 "packsswb 40+"#off"(%2), %%mm3 \n\t"\
298 "packsswb 56+"#off"(%2), %%mm4 \n\t"\
299 "paddb %%mm0, %%mm1 \n\t"\
300 "paddb %%mm0, %%mm2 \n\t"\
301 "paddb %%mm0, %%mm3 \n\t"\
302 "paddb %%mm0, %%mm4 \n\t"\
303 "movq %%mm1, (%0) \n\t"\
304 "movq %%mm2, (%0, %3) \n\t"\
305 "movq %%mm3, (%0, %3, 2) \n\t"\
306 "movq %%mm4, (%0, %1) \n\t"
308 void ff_put_signed_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels, int line_size)
310 x86_reg line_skip = line_size;
314 "movq "MANGLE(ff_pb_80)", %%mm0 \n\t"
315 "lea (%3, %3, 2), %1 \n\t"
316 put_signed_pixels_clamped_mmx_half(0)
317 "lea (%0, %3, 4), %0 \n\t"
318 put_signed_pixels_clamped_mmx_half(64)
319 :"+&r" (pixels), "=&r" (line_skip3)
320 :"r" (block), "r"(line_skip)
324 void ff_add_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels, int line_size)
330 /* read the pixels */
337 "movq (%2), %%mm0 \n\t"
338 "movq 8(%2), %%mm1 \n\t"
339 "movq 16(%2), %%mm2 \n\t"
340 "movq 24(%2), %%mm3 \n\t"
341 "movq %0, %%mm4 \n\t"
342 "movq %1, %%mm6 \n\t"
343 "movq %%mm4, %%mm5 \n\t"
344 "punpcklbw %%mm7, %%mm4 \n\t"
345 "punpckhbw %%mm7, %%mm5 \n\t"
346 "paddsw %%mm4, %%mm0 \n\t"
347 "paddsw %%mm5, %%mm1 \n\t"
348 "movq %%mm6, %%mm5 \n\t"
349 "punpcklbw %%mm7, %%mm6 \n\t"
350 "punpckhbw %%mm7, %%mm5 \n\t"
351 "paddsw %%mm6, %%mm2 \n\t"
352 "paddsw %%mm5, %%mm3 \n\t"
353 "packuswb %%mm1, %%mm0 \n\t"
354 "packuswb %%mm3, %%mm2 \n\t"
355 "movq %%mm0, %0 \n\t"
356 "movq %%mm2, %1 \n\t"
357 :"+m"(*pix), "+m"(*(pix+line_size))
365 static void put_pixels4_mmx(uint8_t *block, const uint8_t *pixels, int line_size, int h)
368 "lea (%3, %3), %%"REG_a" \n\t"
371 "movd (%1), %%mm0 \n\t"
372 "movd (%1, %3), %%mm1 \n\t"
373 "movd %%mm0, (%2) \n\t"
374 "movd %%mm1, (%2, %3) \n\t"
375 "add %%"REG_a", %1 \n\t"
376 "add %%"REG_a", %2 \n\t"
377 "movd (%1), %%mm0 \n\t"
378 "movd (%1, %3), %%mm1 \n\t"
379 "movd %%mm0, (%2) \n\t"
380 "movd %%mm1, (%2, %3) \n\t"
381 "add %%"REG_a", %1 \n\t"
382 "add %%"REG_a", %2 \n\t"
385 : "+g"(h), "+r" (pixels), "+r" (block)
386 : "r"((x86_reg)line_size)
391 static void put_pixels8_mmx(uint8_t *block, const uint8_t *pixels, int line_size, int h)
394 "lea (%3, %3), %%"REG_a" \n\t"
397 "movq (%1), %%mm0 \n\t"
398 "movq (%1, %3), %%mm1 \n\t"
399 "movq %%mm0, (%2) \n\t"
400 "movq %%mm1, (%2, %3) \n\t"
401 "add %%"REG_a", %1 \n\t"
402 "add %%"REG_a", %2 \n\t"
403 "movq (%1), %%mm0 \n\t"
404 "movq (%1, %3), %%mm1 \n\t"
405 "movq %%mm0, (%2) \n\t"
406 "movq %%mm1, (%2, %3) \n\t"
407 "add %%"REG_a", %1 \n\t"
408 "add %%"REG_a", %2 \n\t"
411 : "+g"(h), "+r" (pixels), "+r" (block)
412 : "r"((x86_reg)line_size)
417 static void put_pixels16_mmx(uint8_t *block, const uint8_t *pixels, int line_size, int h)
420 "lea (%3, %3), %%"REG_a" \n\t"
423 "movq (%1), %%mm0 \n\t"
424 "movq 8(%1), %%mm4 \n\t"
425 "movq (%1, %3), %%mm1 \n\t"
426 "movq 8(%1, %3), %%mm5 \n\t"
427 "movq %%mm0, (%2) \n\t"
428 "movq %%mm4, 8(%2) \n\t"
429 "movq %%mm1, (%2, %3) \n\t"
430 "movq %%mm5, 8(%2, %3) \n\t"
431 "add %%"REG_a", %1 \n\t"
432 "add %%"REG_a", %2 \n\t"
433 "movq (%1), %%mm0 \n\t"
434 "movq 8(%1), %%mm4 \n\t"
435 "movq (%1, %3), %%mm1 \n\t"
436 "movq 8(%1, %3), %%mm5 \n\t"
437 "movq %%mm0, (%2) \n\t"
438 "movq %%mm4, 8(%2) \n\t"
439 "movq %%mm1, (%2, %3) \n\t"
440 "movq %%mm5, 8(%2, %3) \n\t"
441 "add %%"REG_a", %1 \n\t"
442 "add %%"REG_a", %2 \n\t"
445 : "+g"(h), "+r" (pixels), "+r" (block)
446 : "r"((x86_reg)line_size)
451 static void put_pixels16_sse2(uint8_t *block, const uint8_t *pixels, int line_size, int h)
455 "movdqu (%1), %%xmm0 \n\t"
456 "movdqu (%1,%3), %%xmm1 \n\t"
457 "movdqu (%1,%3,2), %%xmm2 \n\t"
458 "movdqu (%1,%4), %%xmm3 \n\t"
459 "movdqa %%xmm0, (%2) \n\t"
460 "movdqa %%xmm1, (%2,%3) \n\t"
461 "movdqa %%xmm2, (%2,%3,2) \n\t"
462 "movdqa %%xmm3, (%2,%4) \n\t"
464 "lea (%1,%3,4), %1 \n\t"
465 "lea (%2,%3,4), %2 \n\t"
467 : "+g"(h), "+r" (pixels), "+r" (block)
468 : "r"((x86_reg)line_size), "r"((x86_reg)3L*line_size)
473 static void avg_pixels16_sse2(uint8_t *block, const uint8_t *pixels, int line_size, int h)
477 "movdqu (%1), %%xmm0 \n\t"
478 "movdqu (%1,%3), %%xmm1 \n\t"
479 "movdqu (%1,%3,2), %%xmm2 \n\t"
480 "movdqu (%1,%4), %%xmm3 \n\t"
481 "pavgb (%2), %%xmm0 \n\t"
482 "pavgb (%2,%3), %%xmm1 \n\t"
483 "pavgb (%2,%3,2), %%xmm2 \n\t"
484 "pavgb (%2,%4), %%xmm3 \n\t"
485 "movdqa %%xmm0, (%2) \n\t"
486 "movdqa %%xmm1, (%2,%3) \n\t"
487 "movdqa %%xmm2, (%2,%3,2) \n\t"
488 "movdqa %%xmm3, (%2,%4) \n\t"
490 "lea (%1,%3,4), %1 \n\t"
491 "lea (%2,%3,4), %2 \n\t"
493 : "+g"(h), "+r" (pixels), "+r" (block)
494 : "r"((x86_reg)line_size), "r"((x86_reg)3L*line_size)
499 #define CLEAR_BLOCKS(name,n) \
500 static void name(DCTELEM *blocks)\
503 "pxor %%mm7, %%mm7 \n\t"\
504 "mov %1, %%"REG_a" \n\t"\
506 "movq %%mm7, (%0, %%"REG_a") \n\t"\
507 "movq %%mm7, 8(%0, %%"REG_a") \n\t"\
508 "movq %%mm7, 16(%0, %%"REG_a") \n\t"\
509 "movq %%mm7, 24(%0, %%"REG_a") \n\t"\
510 "add $32, %%"REG_a" \n\t"\
512 : : "r" (((uint8_t *)blocks)+128*n),\
517 CLEAR_BLOCKS(clear_blocks_mmx, 6)
518 CLEAR_BLOCKS(clear_block_mmx, 1)
520 static void clear_block_sse(DCTELEM *block)
523 "xorps %%xmm0, %%xmm0 \n"
524 "movaps %%xmm0, (%0) \n"
525 "movaps %%xmm0, 16(%0) \n"
526 "movaps %%xmm0, 32(%0) \n"
527 "movaps %%xmm0, 48(%0) \n"
528 "movaps %%xmm0, 64(%0) \n"
529 "movaps %%xmm0, 80(%0) \n"
530 "movaps %%xmm0, 96(%0) \n"
531 "movaps %%xmm0, 112(%0) \n"
537 static void clear_blocks_sse(DCTELEM *blocks)
540 "xorps %%xmm0, %%xmm0 \n"
541 "mov %1, %%"REG_a" \n"
543 "movaps %%xmm0, (%0, %%"REG_a") \n"
544 "movaps %%xmm0, 16(%0, %%"REG_a") \n"
545 "movaps %%xmm0, 32(%0, %%"REG_a") \n"
546 "movaps %%xmm0, 48(%0, %%"REG_a") \n"
547 "movaps %%xmm0, 64(%0, %%"REG_a") \n"
548 "movaps %%xmm0, 80(%0, %%"REG_a") \n"
549 "movaps %%xmm0, 96(%0, %%"REG_a") \n"
550 "movaps %%xmm0, 112(%0, %%"REG_a") \n"
551 "add $128, %%"REG_a" \n"
553 : : "r" (((uint8_t *)blocks)+128*6),
559 static void add_bytes_mmx(uint8_t *dst, uint8_t *src, int w){
564 "movq (%1, %0), %%mm0 \n\t"
565 "movq (%2, %0), %%mm1 \n\t"
566 "paddb %%mm0, %%mm1 \n\t"
567 "movq %%mm1, (%2, %0) \n\t"
568 "movq 8(%1, %0), %%mm0 \n\t"
569 "movq 8(%2, %0), %%mm1 \n\t"
570 "paddb %%mm0, %%mm1 \n\t"
571 "movq %%mm1, 8(%2, %0) \n\t"
577 : "r"(src), "r"(dst), "r"((x86_reg)w-15)
580 dst[i+0] += src[i+0];
584 static void add_hfyu_median_prediction_cmov(uint8_t *dst, const uint8_t *top, const uint8_t *diff, int w, int *left, int *left_top) {
587 int l = *left & 0xff;
588 int tl = *left_top & 0xff;
593 "movzbl (%3,%4), %2 \n"
606 "add (%6,%4), %b0 \n"
607 "mov %b0, (%5,%4) \n"
610 :"+&q"(l), "+&q"(tl), "=&r"(t), "=&q"(x), "+&r"(w2)
611 :"r"(dst+w), "r"(diff+w), "rm"(top+w)
618 #define H263_LOOP_FILTER \
619 "pxor %%mm7, %%mm7 \n\t"\
620 "movq %0, %%mm0 \n\t"\
621 "movq %0, %%mm1 \n\t"\
622 "movq %3, %%mm2 \n\t"\
623 "movq %3, %%mm3 \n\t"\
624 "punpcklbw %%mm7, %%mm0 \n\t"\
625 "punpckhbw %%mm7, %%mm1 \n\t"\
626 "punpcklbw %%mm7, %%mm2 \n\t"\
627 "punpckhbw %%mm7, %%mm3 \n\t"\
628 "psubw %%mm2, %%mm0 \n\t"\
629 "psubw %%mm3, %%mm1 \n\t"\
630 "movq %1, %%mm2 \n\t"\
631 "movq %1, %%mm3 \n\t"\
632 "movq %2, %%mm4 \n\t"\
633 "movq %2, %%mm5 \n\t"\
634 "punpcklbw %%mm7, %%mm2 \n\t"\
635 "punpckhbw %%mm7, %%mm3 \n\t"\
636 "punpcklbw %%mm7, %%mm4 \n\t"\
637 "punpckhbw %%mm7, %%mm5 \n\t"\
638 "psubw %%mm2, %%mm4 \n\t"\
639 "psubw %%mm3, %%mm5 \n\t"\
640 "psllw $2, %%mm4 \n\t"\
641 "psllw $2, %%mm5 \n\t"\
642 "paddw %%mm0, %%mm4 \n\t"\
643 "paddw %%mm1, %%mm5 \n\t"\
644 "pxor %%mm6, %%mm6 \n\t"\
645 "pcmpgtw %%mm4, %%mm6 \n\t"\
646 "pcmpgtw %%mm5, %%mm7 \n\t"\
647 "pxor %%mm6, %%mm4 \n\t"\
648 "pxor %%mm7, %%mm5 \n\t"\
649 "psubw %%mm6, %%mm4 \n\t"\
650 "psubw %%mm7, %%mm5 \n\t"\
651 "psrlw $3, %%mm4 \n\t"\
652 "psrlw $3, %%mm5 \n\t"\
653 "packuswb %%mm5, %%mm4 \n\t"\
654 "packsswb %%mm7, %%mm6 \n\t"\
655 "pxor %%mm7, %%mm7 \n\t"\
656 "movd %4, %%mm2 \n\t"\
657 "punpcklbw %%mm2, %%mm2 \n\t"\
658 "punpcklbw %%mm2, %%mm2 \n\t"\
659 "punpcklbw %%mm2, %%mm2 \n\t"\
660 "psubusb %%mm4, %%mm2 \n\t"\
661 "movq %%mm2, %%mm3 \n\t"\
662 "psubusb %%mm4, %%mm3 \n\t"\
663 "psubb %%mm3, %%mm2 \n\t"\
664 "movq %1, %%mm3 \n\t"\
665 "movq %2, %%mm4 \n\t"\
666 "pxor %%mm6, %%mm3 \n\t"\
667 "pxor %%mm6, %%mm4 \n\t"\
668 "paddusb %%mm2, %%mm3 \n\t"\
669 "psubusb %%mm2, %%mm4 \n\t"\
670 "pxor %%mm6, %%mm3 \n\t"\
671 "pxor %%mm6, %%mm4 \n\t"\
672 "paddusb %%mm2, %%mm2 \n\t"\
673 "packsswb %%mm1, %%mm0 \n\t"\
674 "pcmpgtb %%mm0, %%mm7 \n\t"\
675 "pxor %%mm7, %%mm0 \n\t"\
676 "psubb %%mm7, %%mm0 \n\t"\
677 "movq %%mm0, %%mm1 \n\t"\
678 "psubusb %%mm2, %%mm0 \n\t"\
679 "psubb %%mm0, %%mm1 \n\t"\
680 "pand %5, %%mm1 \n\t"\
681 "psrlw $2, %%mm1 \n\t"\
682 "pxor %%mm7, %%mm1 \n\t"\
683 "psubb %%mm7, %%mm1 \n\t"\
684 "movq %0, %%mm5 \n\t"\
685 "movq %3, %%mm6 \n\t"\
686 "psubb %%mm1, %%mm5 \n\t"\
687 "paddb %%mm1, %%mm6 \n\t"
689 static void h263_v_loop_filter_mmx(uint8_t *src, int stride, int qscale){
690 if(CONFIG_H263_DECODER || CONFIG_H263_ENCODER) {
691 const int strength= ff_h263_loop_filter_strength[qscale];
697 "movq %%mm3, %1 \n\t"
698 "movq %%mm4, %2 \n\t"
699 "movq %%mm5, %0 \n\t"
700 "movq %%mm6, %3 \n\t"
701 : "+m" (*(uint64_t*)(src - 2*stride)),
702 "+m" (*(uint64_t*)(src - 1*stride)),
703 "+m" (*(uint64_t*)(src + 0*stride)),
704 "+m" (*(uint64_t*)(src + 1*stride))
705 : "g" (2*strength), "m"(ff_pb_FC)
710 static void h263_h_loop_filter_mmx(uint8_t *src, int stride, int qscale){
711 if(CONFIG_H263_DECODER || CONFIG_H263_ENCODER) {
712 const int strength= ff_h263_loop_filter_strength[qscale];
713 DECLARE_ALIGNED(8, uint64_t, temp)[4];
714 uint8_t *btemp= (uint8_t*)temp;
718 transpose4x4(btemp , src , 8, stride);
719 transpose4x4(btemp+4, src + 4*stride, 8, stride);
721 H263_LOOP_FILTER // 5 3 4 6
727 : "g" (2*strength), "m"(ff_pb_FC)
731 "movq %%mm5, %%mm1 \n\t"
732 "movq %%mm4, %%mm0 \n\t"
733 "punpcklbw %%mm3, %%mm5 \n\t"
734 "punpcklbw %%mm6, %%mm4 \n\t"
735 "punpckhbw %%mm3, %%mm1 \n\t"
736 "punpckhbw %%mm6, %%mm0 \n\t"
737 "movq %%mm5, %%mm3 \n\t"
738 "movq %%mm1, %%mm6 \n\t"
739 "punpcklwd %%mm4, %%mm5 \n\t"
740 "punpcklwd %%mm0, %%mm1 \n\t"
741 "punpckhwd %%mm4, %%mm3 \n\t"
742 "punpckhwd %%mm0, %%mm6 \n\t"
743 "movd %%mm5, (%0) \n\t"
744 "punpckhdq %%mm5, %%mm5 \n\t"
745 "movd %%mm5, (%0,%2) \n\t"
746 "movd %%mm3, (%0,%2,2) \n\t"
747 "punpckhdq %%mm3, %%mm3 \n\t"
748 "movd %%mm3, (%0,%3) \n\t"
749 "movd %%mm1, (%1) \n\t"
750 "punpckhdq %%mm1, %%mm1 \n\t"
751 "movd %%mm1, (%1,%2) \n\t"
752 "movd %%mm6, (%1,%2,2) \n\t"
753 "punpckhdq %%mm6, %%mm6 \n\t"
754 "movd %%mm6, (%1,%3) \n\t"
756 "r" (src + 4*stride),
757 "r" ((x86_reg) stride ),
758 "r" ((x86_reg)(3*stride))
763 /* draw the edges of width 'w' of an image of size width, height
764 this mmx version can only handle w==8 || w==16 */
765 static void draw_edges_mmx(uint8_t *buf, int wrap, int width, int height, int w, int h, int sides)
767 uint8_t *ptr, *last_line;
770 last_line = buf + (height - 1) * wrap;
777 "movd (%0), %%mm0 \n\t"
778 "punpcklbw %%mm0, %%mm0 \n\t"
779 "punpcklwd %%mm0, %%mm0 \n\t"
780 "punpckldq %%mm0, %%mm0 \n\t"
781 "movq %%mm0, -8(%0) \n\t"
782 "movq -8(%0, %2), %%mm1 \n\t"
783 "punpckhbw %%mm1, %%mm1 \n\t"
784 "punpckhwd %%mm1, %%mm1 \n\t"
785 "punpckhdq %%mm1, %%mm1 \n\t"
786 "movq %%mm1, (%0, %2) \n\t"
791 : "r" ((x86_reg)wrap), "r" ((x86_reg)width), "r" (ptr + wrap*height)
798 "movd (%0), %%mm0 \n\t"
799 "punpcklbw %%mm0, %%mm0 \n\t"
800 "punpcklwd %%mm0, %%mm0 \n\t"
801 "punpckldq %%mm0, %%mm0 \n\t"
802 "movq %%mm0, -8(%0) \n\t"
803 "movq %%mm0, -16(%0) \n\t"
804 "movq -8(%0, %2), %%mm1 \n\t"
805 "punpckhbw %%mm1, %%mm1 \n\t"
806 "punpckhwd %%mm1, %%mm1 \n\t"
807 "punpckhdq %%mm1, %%mm1 \n\t"
808 "movq %%mm1, (%0, %2) \n\t"
809 "movq %%mm1, 8(%0, %2) \n\t"
814 : "r" ((x86_reg)wrap), "r" ((x86_reg)width), "r" (ptr + wrap*height)
818 /* top and bottom (and hopefully also the corners) */
819 if (sides&EDGE_TOP) {
820 for(i = 0; i < h; i += 4) {
821 ptr= buf - (i + 1) * wrap - w;
824 "movq (%1, %0), %%mm0 \n\t"
825 "movq %%mm0, (%0) \n\t"
826 "movq %%mm0, (%0, %2) \n\t"
827 "movq %%mm0, (%0, %2, 2) \n\t"
828 "movq %%mm0, (%0, %3) \n\t"
833 : "r" ((x86_reg)buf - (x86_reg)ptr - w), "r" ((x86_reg)-wrap), "r" ((x86_reg)-wrap*3), "r" (ptr+width+2*w)
838 if (sides&EDGE_BOTTOM) {
839 for(i = 0; i < w; i += 4) {
840 ptr= last_line + (i + 1) * wrap - w;
843 "movq (%1, %0), %%mm0 \n\t"
844 "movq %%mm0, (%0) \n\t"
845 "movq %%mm0, (%0, %2) \n\t"
846 "movq %%mm0, (%0, %2, 2) \n\t"
847 "movq %%mm0, (%0, %3) \n\t"
852 : "r" ((x86_reg)last_line - (x86_reg)ptr - w), "r" ((x86_reg)wrap), "r" ((x86_reg)wrap*3), "r" (ptr+width+2*w)
858 #define QPEL_V_LOW(m3,m4,m5,m6, pw_20, pw_3, rnd, in0, in1, in2, in7, out, OP)\
859 "paddw " #m4 ", " #m3 " \n\t" /* x1 */\
860 "movq "MANGLE(ff_pw_20)", %%mm4 \n\t" /* 20 */\
861 "pmullw " #m3 ", %%mm4 \n\t" /* 20x1 */\
862 "movq "#in7", " #m3 " \n\t" /* d */\
863 "movq "#in0", %%mm5 \n\t" /* D */\
864 "paddw " #m3 ", %%mm5 \n\t" /* x4 */\
865 "psubw %%mm5, %%mm4 \n\t" /* 20x1 - x4 */\
866 "movq "#in1", %%mm5 \n\t" /* C */\
867 "movq "#in2", %%mm6 \n\t" /* B */\
868 "paddw " #m6 ", %%mm5 \n\t" /* x3 */\
869 "paddw " #m5 ", %%mm6 \n\t" /* x2 */\
870 "paddw %%mm6, %%mm6 \n\t" /* 2x2 */\
871 "psubw %%mm6, %%mm5 \n\t" /* -2x2 + x3 */\
872 "pmullw "MANGLE(ff_pw_3)", %%mm5 \n\t" /* -6x2 + 3x3 */\
873 "paddw " #rnd ", %%mm4 \n\t" /* x2 */\
874 "paddw %%mm4, %%mm5 \n\t" /* 20x1 - 6x2 + 3x3 - x4 */\
875 "psraw $5, %%mm5 \n\t"\
876 "packuswb %%mm5, %%mm5 \n\t"\
877 OP(%%mm5, out, %%mm7, d)
879 #define QPEL_BASE(OPNAME, ROUNDER, RND, OP_MMX2, OP_3DNOW)\
880 static void OPNAME ## mpeg4_qpel16_h_lowpass_mmx2(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h){\
884 "pxor %%mm7, %%mm7 \n\t"\
886 "movq (%0), %%mm0 \n\t" /* ABCDEFGH */\
887 "movq %%mm0, %%mm1 \n\t" /* ABCDEFGH */\
888 "movq %%mm0, %%mm2 \n\t" /* ABCDEFGH */\
889 "punpcklbw %%mm7, %%mm0 \n\t" /* 0A0B0C0D */\
890 "punpckhbw %%mm7, %%mm1 \n\t" /* 0E0F0G0H */\
891 "pshufw $0x90, %%mm0, %%mm5 \n\t" /* 0A0A0B0C */\
892 "pshufw $0x41, %%mm0, %%mm6 \n\t" /* 0B0A0A0B */\
893 "movq %%mm2, %%mm3 \n\t" /* ABCDEFGH */\
894 "movq %%mm2, %%mm4 \n\t" /* ABCDEFGH */\
895 "psllq $8, %%mm2 \n\t" /* 0ABCDEFG */\
896 "psllq $16, %%mm3 \n\t" /* 00ABCDEF */\
897 "psllq $24, %%mm4 \n\t" /* 000ABCDE */\
898 "punpckhbw %%mm7, %%mm2 \n\t" /* 0D0E0F0G */\
899 "punpckhbw %%mm7, %%mm3 \n\t" /* 0C0D0E0F */\
900 "punpckhbw %%mm7, %%mm4 \n\t" /* 0B0C0D0E */\
901 "paddw %%mm3, %%mm5 \n\t" /* b */\
902 "paddw %%mm2, %%mm6 \n\t" /* c */\
903 "paddw %%mm5, %%mm5 \n\t" /* 2b */\
904 "psubw %%mm5, %%mm6 \n\t" /* c - 2b */\
905 "pshufw $0x06, %%mm0, %%mm5 \n\t" /* 0C0B0A0A */\
906 "pmullw "MANGLE(ff_pw_3)", %%mm6 \n\t" /* 3c - 6b */\
907 "paddw %%mm4, %%mm0 \n\t" /* a */\
908 "paddw %%mm1, %%mm5 \n\t" /* d */\
909 "pmullw "MANGLE(ff_pw_20)", %%mm0 \n\t" /* 20a */\
910 "psubw %%mm5, %%mm0 \n\t" /* 20a - d */\
911 "paddw %6, %%mm6 \n\t"\
912 "paddw %%mm6, %%mm0 \n\t" /* 20a - 6b + 3c - d */\
913 "psraw $5, %%mm0 \n\t"\
914 "movq %%mm0, %5 \n\t"\
915 /* mm1=EFGH, mm2=DEFG, mm3=CDEF, mm4=BCDE, mm7=0 */\
917 "movq 5(%0), %%mm0 \n\t" /* FGHIJKLM */\
918 "movq %%mm0, %%mm5 \n\t" /* FGHIJKLM */\
919 "movq %%mm0, %%mm6 \n\t" /* FGHIJKLM */\
920 "psrlq $8, %%mm0 \n\t" /* GHIJKLM0 */\
921 "psrlq $16, %%mm5 \n\t" /* HIJKLM00 */\
922 "punpcklbw %%mm7, %%mm0 \n\t" /* 0G0H0I0J */\
923 "punpcklbw %%mm7, %%mm5 \n\t" /* 0H0I0J0K */\
924 "paddw %%mm0, %%mm2 \n\t" /* b */\
925 "paddw %%mm5, %%mm3 \n\t" /* c */\
926 "paddw %%mm2, %%mm2 \n\t" /* 2b */\
927 "psubw %%mm2, %%mm3 \n\t" /* c - 2b */\
928 "movq %%mm6, %%mm2 \n\t" /* FGHIJKLM */\
929 "psrlq $24, %%mm6 \n\t" /* IJKLM000 */\
930 "punpcklbw %%mm7, %%mm2 \n\t" /* 0F0G0H0I */\
931 "punpcklbw %%mm7, %%mm6 \n\t" /* 0I0J0K0L */\
932 "pmullw "MANGLE(ff_pw_3)", %%mm3 \n\t" /* 3c - 6b */\
933 "paddw %%mm2, %%mm1 \n\t" /* a */\
934 "paddw %%mm6, %%mm4 \n\t" /* d */\
935 "pmullw "MANGLE(ff_pw_20)", %%mm1 \n\t" /* 20a */\
936 "psubw %%mm4, %%mm3 \n\t" /* - 6b +3c - d */\
937 "paddw %6, %%mm1 \n\t"\
938 "paddw %%mm1, %%mm3 \n\t" /* 20a - 6b +3c - d */\
939 "psraw $5, %%mm3 \n\t"\
940 "movq %5, %%mm1 \n\t"\
941 "packuswb %%mm3, %%mm1 \n\t"\
942 OP_MMX2(%%mm1, (%1),%%mm4, q)\
943 /* mm0= GHIJ, mm2=FGHI, mm5=HIJK, mm6=IJKL, mm7=0 */\
945 "movq 9(%0), %%mm1 \n\t" /* JKLMNOPQ */\
946 "movq %%mm1, %%mm4 \n\t" /* JKLMNOPQ */\
947 "movq %%mm1, %%mm3 \n\t" /* JKLMNOPQ */\
948 "psrlq $8, %%mm1 \n\t" /* KLMNOPQ0 */\
949 "psrlq $16, %%mm4 \n\t" /* LMNOPQ00 */\
950 "punpcklbw %%mm7, %%mm1 \n\t" /* 0K0L0M0N */\
951 "punpcklbw %%mm7, %%mm4 \n\t" /* 0L0M0N0O */\
952 "paddw %%mm1, %%mm5 \n\t" /* b */\
953 "paddw %%mm4, %%mm0 \n\t" /* c */\
954 "paddw %%mm5, %%mm5 \n\t" /* 2b */\
955 "psubw %%mm5, %%mm0 \n\t" /* c - 2b */\
956 "movq %%mm3, %%mm5 \n\t" /* JKLMNOPQ */\
957 "psrlq $24, %%mm3 \n\t" /* MNOPQ000 */\
958 "pmullw "MANGLE(ff_pw_3)", %%mm0 \n\t" /* 3c - 6b */\
959 "punpcklbw %%mm7, %%mm3 \n\t" /* 0M0N0O0P */\
960 "paddw %%mm3, %%mm2 \n\t" /* d */\
961 "psubw %%mm2, %%mm0 \n\t" /* -6b + 3c - d */\
962 "movq %%mm5, %%mm2 \n\t" /* JKLMNOPQ */\
963 "punpcklbw %%mm7, %%mm2 \n\t" /* 0J0K0L0M */\
964 "punpckhbw %%mm7, %%mm5 \n\t" /* 0N0O0P0Q */\
965 "paddw %%mm2, %%mm6 \n\t" /* a */\
966 "pmullw "MANGLE(ff_pw_20)", %%mm6 \n\t" /* 20a */\
967 "paddw %6, %%mm0 \n\t"\
968 "paddw %%mm6, %%mm0 \n\t" /* 20a - 6b + 3c - d */\
969 "psraw $5, %%mm0 \n\t"\
970 /* mm1=KLMN, mm2=JKLM, mm3=MNOP, mm4=LMNO, mm5=NOPQ mm7=0 */\
972 "paddw %%mm5, %%mm3 \n\t" /* a */\
973 "pshufw $0xF9, %%mm5, %%mm6 \n\t" /* 0O0P0Q0Q */\
974 "paddw %%mm4, %%mm6 \n\t" /* b */\
975 "pshufw $0xBE, %%mm5, %%mm4 \n\t" /* 0P0Q0Q0P */\
976 "pshufw $0x6F, %%mm5, %%mm5 \n\t" /* 0Q0Q0P0O */\
977 "paddw %%mm1, %%mm4 \n\t" /* c */\
978 "paddw %%mm2, %%mm5 \n\t" /* d */\
979 "paddw %%mm6, %%mm6 \n\t" /* 2b */\
980 "psubw %%mm6, %%mm4 \n\t" /* c - 2b */\
981 "pmullw "MANGLE(ff_pw_20)", %%mm3 \n\t" /* 20a */\
982 "pmullw "MANGLE(ff_pw_3)", %%mm4 \n\t" /* 3c - 6b */\
983 "psubw %%mm5, %%mm3 \n\t" /* -6b + 3c - d */\
984 "paddw %6, %%mm4 \n\t"\
985 "paddw %%mm3, %%mm4 \n\t" /* 20a - 6b + 3c - d */\
986 "psraw $5, %%mm4 \n\t"\
987 "packuswb %%mm4, %%mm0 \n\t"\
988 OP_MMX2(%%mm0, 8(%1), %%mm4, q)\
994 : "+a"(src), "+c"(dst), "+D"(h)\
995 : "d"((x86_reg)srcStride), "S"((x86_reg)dstStride), /*"m"(ff_pw_20), "m"(ff_pw_3),*/ "m"(temp), "m"(ROUNDER)\
1000 static void OPNAME ## mpeg4_qpel16_h_lowpass_3dnow(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h){\
1003 /* quick HACK, XXX FIXME MUST be optimized */\
1006 temp[ 0]= (src[ 0]+src[ 1])*20 - (src[ 0]+src[ 2])*6 + (src[ 1]+src[ 3])*3 - (src[ 2]+src[ 4]);\
1007 temp[ 1]= (src[ 1]+src[ 2])*20 - (src[ 0]+src[ 3])*6 + (src[ 0]+src[ 4])*3 - (src[ 1]+src[ 5]);\
1008 temp[ 2]= (src[ 2]+src[ 3])*20 - (src[ 1]+src[ 4])*6 + (src[ 0]+src[ 5])*3 - (src[ 0]+src[ 6]);\
1009 temp[ 3]= (src[ 3]+src[ 4])*20 - (src[ 2]+src[ 5])*6 + (src[ 1]+src[ 6])*3 - (src[ 0]+src[ 7]);\
1010 temp[ 4]= (src[ 4]+src[ 5])*20 - (src[ 3]+src[ 6])*6 + (src[ 2]+src[ 7])*3 - (src[ 1]+src[ 8]);\
1011 temp[ 5]= (src[ 5]+src[ 6])*20 - (src[ 4]+src[ 7])*6 + (src[ 3]+src[ 8])*3 - (src[ 2]+src[ 9]);\
1012 temp[ 6]= (src[ 6]+src[ 7])*20 - (src[ 5]+src[ 8])*6 + (src[ 4]+src[ 9])*3 - (src[ 3]+src[10]);\
1013 temp[ 7]= (src[ 7]+src[ 8])*20 - (src[ 6]+src[ 9])*6 + (src[ 5]+src[10])*3 - (src[ 4]+src[11]);\
1014 temp[ 8]= (src[ 8]+src[ 9])*20 - (src[ 7]+src[10])*6 + (src[ 6]+src[11])*3 - (src[ 5]+src[12]);\
1015 temp[ 9]= (src[ 9]+src[10])*20 - (src[ 8]+src[11])*6 + (src[ 7]+src[12])*3 - (src[ 6]+src[13]);\
1016 temp[10]= (src[10]+src[11])*20 - (src[ 9]+src[12])*6 + (src[ 8]+src[13])*3 - (src[ 7]+src[14]);\
1017 temp[11]= (src[11]+src[12])*20 - (src[10]+src[13])*6 + (src[ 9]+src[14])*3 - (src[ 8]+src[15]);\
1018 temp[12]= (src[12]+src[13])*20 - (src[11]+src[14])*6 + (src[10]+src[15])*3 - (src[ 9]+src[16]);\
1019 temp[13]= (src[13]+src[14])*20 - (src[12]+src[15])*6 + (src[11]+src[16])*3 - (src[10]+src[16]);\
1020 temp[14]= (src[14]+src[15])*20 - (src[13]+src[16])*6 + (src[12]+src[16])*3 - (src[11]+src[15]);\
1021 temp[15]= (src[15]+src[16])*20 - (src[14]+src[16])*6 + (src[13]+src[15])*3 - (src[12]+src[14]);\
1023 "movq (%0), %%mm0 \n\t"\
1024 "movq 8(%0), %%mm1 \n\t"\
1025 "paddw %2, %%mm0 \n\t"\
1026 "paddw %2, %%mm1 \n\t"\
1027 "psraw $5, %%mm0 \n\t"\
1028 "psraw $5, %%mm1 \n\t"\
1029 "packuswb %%mm1, %%mm0 \n\t"\
1030 OP_3DNOW(%%mm0, (%1), %%mm1, q)\
1031 "movq 16(%0), %%mm0 \n\t"\
1032 "movq 24(%0), %%mm1 \n\t"\
1033 "paddw %2, %%mm0 \n\t"\
1034 "paddw %2, %%mm1 \n\t"\
1035 "psraw $5, %%mm0 \n\t"\
1036 "psraw $5, %%mm1 \n\t"\
1037 "packuswb %%mm1, %%mm0 \n\t"\
1038 OP_3DNOW(%%mm0, 8(%1), %%mm1, q)\
1039 :: "r"(temp), "r"(dst), "m"(ROUNDER)\
1047 static void OPNAME ## mpeg4_qpel8_h_lowpass_mmx2(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h){\
1049 "pxor %%mm7, %%mm7 \n\t"\
1051 "movq (%0), %%mm0 \n\t" /* ABCDEFGH */\
1052 "movq %%mm0, %%mm1 \n\t" /* ABCDEFGH */\
1053 "movq %%mm0, %%mm2 \n\t" /* ABCDEFGH */\
1054 "punpcklbw %%mm7, %%mm0 \n\t" /* 0A0B0C0D */\
1055 "punpckhbw %%mm7, %%mm1 \n\t" /* 0E0F0G0H */\
1056 "pshufw $0x90, %%mm0, %%mm5 \n\t" /* 0A0A0B0C */\
1057 "pshufw $0x41, %%mm0, %%mm6 \n\t" /* 0B0A0A0B */\
1058 "movq %%mm2, %%mm3 \n\t" /* ABCDEFGH */\
1059 "movq %%mm2, %%mm4 \n\t" /* ABCDEFGH */\
1060 "psllq $8, %%mm2 \n\t" /* 0ABCDEFG */\
1061 "psllq $16, %%mm3 \n\t" /* 00ABCDEF */\
1062 "psllq $24, %%mm4 \n\t" /* 000ABCDE */\
1063 "punpckhbw %%mm7, %%mm2 \n\t" /* 0D0E0F0G */\
1064 "punpckhbw %%mm7, %%mm3 \n\t" /* 0C0D0E0F */\
1065 "punpckhbw %%mm7, %%mm4 \n\t" /* 0B0C0D0E */\
1066 "paddw %%mm3, %%mm5 \n\t" /* b */\
1067 "paddw %%mm2, %%mm6 \n\t" /* c */\
1068 "paddw %%mm5, %%mm5 \n\t" /* 2b */\
1069 "psubw %%mm5, %%mm6 \n\t" /* c - 2b */\
1070 "pshufw $0x06, %%mm0, %%mm5 \n\t" /* 0C0B0A0A */\
1071 "pmullw "MANGLE(ff_pw_3)", %%mm6 \n\t" /* 3c - 6b */\
1072 "paddw %%mm4, %%mm0 \n\t" /* a */\
1073 "paddw %%mm1, %%mm5 \n\t" /* d */\
1074 "pmullw "MANGLE(ff_pw_20)", %%mm0 \n\t" /* 20a */\
1075 "psubw %%mm5, %%mm0 \n\t" /* 20a - d */\
1076 "paddw %5, %%mm6 \n\t"\
1077 "paddw %%mm6, %%mm0 \n\t" /* 20a - 6b + 3c - d */\
1078 "psraw $5, %%mm0 \n\t"\
1079 /* mm1=EFGH, mm2=DEFG, mm3=CDEF, mm4=BCDE, mm7=0 */\
1081 "movd 5(%0), %%mm5 \n\t" /* FGHI */\
1082 "punpcklbw %%mm7, %%mm5 \n\t" /* 0F0G0H0I */\
1083 "pshufw $0xF9, %%mm5, %%mm6 \n\t" /* 0G0H0I0I */\
1084 "paddw %%mm5, %%mm1 \n\t" /* a */\
1085 "paddw %%mm6, %%mm2 \n\t" /* b */\
1086 "pshufw $0xBE, %%mm5, %%mm6 \n\t" /* 0H0I0I0H */\
1087 "pshufw $0x6F, %%mm5, %%mm5 \n\t" /* 0I0I0H0G */\
1088 "paddw %%mm6, %%mm3 \n\t" /* c */\
1089 "paddw %%mm5, %%mm4 \n\t" /* d */\
1090 "paddw %%mm2, %%mm2 \n\t" /* 2b */\
1091 "psubw %%mm2, %%mm3 \n\t" /* c - 2b */\
1092 "pmullw "MANGLE(ff_pw_20)", %%mm1 \n\t" /* 20a */\
1093 "pmullw "MANGLE(ff_pw_3)", %%mm3 \n\t" /* 3c - 6b */\
1094 "psubw %%mm4, %%mm3 \n\t" /* -6b + 3c - d */\
1095 "paddw %5, %%mm1 \n\t"\
1096 "paddw %%mm1, %%mm3 \n\t" /* 20a - 6b + 3c - d */\
1097 "psraw $5, %%mm3 \n\t"\
1098 "packuswb %%mm3, %%mm0 \n\t"\
1099 OP_MMX2(%%mm0, (%1), %%mm4, q)\
1105 : "+a"(src), "+c"(dst), "+d"(h)\
1106 : "S"((x86_reg)srcStride), "D"((x86_reg)dstStride), /*"m"(ff_pw_20), "m"(ff_pw_3),*/ "m"(ROUNDER)\
1111 static void OPNAME ## mpeg4_qpel8_h_lowpass_3dnow(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h){\
1114 /* quick HACK, XXX FIXME MUST be optimized */\
1117 temp[ 0]= (src[ 0]+src[ 1])*20 - (src[ 0]+src[ 2])*6 + (src[ 1]+src[ 3])*3 - (src[ 2]+src[ 4]);\
1118 temp[ 1]= (src[ 1]+src[ 2])*20 - (src[ 0]+src[ 3])*6 + (src[ 0]+src[ 4])*3 - (src[ 1]+src[ 5]);\
1119 temp[ 2]= (src[ 2]+src[ 3])*20 - (src[ 1]+src[ 4])*6 + (src[ 0]+src[ 5])*3 - (src[ 0]+src[ 6]);\
1120 temp[ 3]= (src[ 3]+src[ 4])*20 - (src[ 2]+src[ 5])*6 + (src[ 1]+src[ 6])*3 - (src[ 0]+src[ 7]);\
1121 temp[ 4]= (src[ 4]+src[ 5])*20 - (src[ 3]+src[ 6])*6 + (src[ 2]+src[ 7])*3 - (src[ 1]+src[ 8]);\
1122 temp[ 5]= (src[ 5]+src[ 6])*20 - (src[ 4]+src[ 7])*6 + (src[ 3]+src[ 8])*3 - (src[ 2]+src[ 8]);\
1123 temp[ 6]= (src[ 6]+src[ 7])*20 - (src[ 5]+src[ 8])*6 + (src[ 4]+src[ 8])*3 - (src[ 3]+src[ 7]);\
1124 temp[ 7]= (src[ 7]+src[ 8])*20 - (src[ 6]+src[ 8])*6 + (src[ 5]+src[ 7])*3 - (src[ 4]+src[ 6]);\
1126 "movq (%0), %%mm0 \n\t"\
1127 "movq 8(%0), %%mm1 \n\t"\
1128 "paddw %2, %%mm0 \n\t"\
1129 "paddw %2, %%mm1 \n\t"\
1130 "psraw $5, %%mm0 \n\t"\
1131 "psraw $5, %%mm1 \n\t"\
1132 "packuswb %%mm1, %%mm0 \n\t"\
1133 OP_3DNOW(%%mm0, (%1), %%mm1, q)\
1134 :: "r"(temp), "r"(dst), "m"(ROUNDER)\
1142 #define QPEL_OP(OPNAME, ROUNDER, RND, OP, MMX)\
1144 static void OPNAME ## mpeg4_qpel16_v_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
1145 uint64_t temp[17*4];\
1146 uint64_t *temp_ptr= temp;\
1151 "pxor %%mm7, %%mm7 \n\t"\
1153 "movq (%0), %%mm0 \n\t"\
1154 "movq (%0), %%mm1 \n\t"\
1155 "movq 8(%0), %%mm2 \n\t"\
1156 "movq 8(%0), %%mm3 \n\t"\
1157 "punpcklbw %%mm7, %%mm0 \n\t"\
1158 "punpckhbw %%mm7, %%mm1 \n\t"\
1159 "punpcklbw %%mm7, %%mm2 \n\t"\
1160 "punpckhbw %%mm7, %%mm3 \n\t"\
1161 "movq %%mm0, (%1) \n\t"\
1162 "movq %%mm1, 17*8(%1) \n\t"\
1163 "movq %%mm2, 2*17*8(%1) \n\t"\
1164 "movq %%mm3, 3*17*8(%1) \n\t"\
1169 : "+r" (src), "+r" (temp_ptr), "+r"(count)\
1170 : "r" ((x86_reg)srcStride)\
1177 /*FIXME reorder for speed */\
1179 /*"pxor %%mm7, %%mm7 \n\t"*/\
1181 "movq (%0), %%mm0 \n\t"\
1182 "movq 8(%0), %%mm1 \n\t"\
1183 "movq 16(%0), %%mm2 \n\t"\
1184 "movq 24(%0), %%mm3 \n\t"\
1185 QPEL_V_LOW(%%mm0, %%mm1, %%mm2, %%mm3, %5, %6, %5, 16(%0), 8(%0), (%0), 32(%0), (%1), OP)\
1186 QPEL_V_LOW(%%mm1, %%mm2, %%mm3, %%mm0, %5, %6, %5, 8(%0), (%0), (%0), 40(%0), (%1, %3), OP)\
1188 QPEL_V_LOW(%%mm2, %%mm3, %%mm0, %%mm1, %5, %6, %5, (%0), (%0), 8(%0), 48(%0), (%1), OP)\
1190 QPEL_V_LOW(%%mm3, %%mm0, %%mm1, %%mm2, %5, %6, %5, (%0), 8(%0), 16(%0), 56(%0), (%1, %3), OP)\
1192 QPEL_V_LOW(%%mm0, %%mm1, %%mm2, %%mm3, %5, %6, %5, 8(%0), 16(%0), 24(%0), 64(%0), (%1), OP)\
1193 QPEL_V_LOW(%%mm1, %%mm2, %%mm3, %%mm0, %5, %6, %5, 16(%0), 24(%0), 32(%0), 72(%0), (%1, %3), OP)\
1195 QPEL_V_LOW(%%mm2, %%mm3, %%mm0, %%mm1, %5, %6, %5, 24(%0), 32(%0), 40(%0), 80(%0), (%1), OP)\
1196 QPEL_V_LOW(%%mm3, %%mm0, %%mm1, %%mm2, %5, %6, %5, 32(%0), 40(%0), 48(%0), 88(%0), (%1, %3), OP)\
1198 QPEL_V_LOW(%%mm0, %%mm1, %%mm2, %%mm3, %5, %6, %5, 40(%0), 48(%0), 56(%0), 96(%0), (%1), OP)\
1199 QPEL_V_LOW(%%mm1, %%mm2, %%mm3, %%mm0, %5, %6, %5, 48(%0), 56(%0), 64(%0),104(%0), (%1, %3), OP)\
1201 QPEL_V_LOW(%%mm2, %%mm3, %%mm0, %%mm1, %5, %6, %5, 56(%0), 64(%0), 72(%0),112(%0), (%1), OP)\
1202 QPEL_V_LOW(%%mm3, %%mm0, %%mm1, %%mm2, %5, %6, %5, 64(%0), 72(%0), 80(%0),120(%0), (%1, %3), OP)\
1204 QPEL_V_LOW(%%mm0, %%mm1, %%mm2, %%mm3, %5, %6, %5, 72(%0), 80(%0), 88(%0),128(%0), (%1), OP)\
1206 QPEL_V_LOW(%%mm1, %%mm2, %%mm3, %%mm0, %5, %6, %5, 80(%0), 88(%0), 96(%0),128(%0), (%1, %3), OP)\
1208 QPEL_V_LOW(%%mm2, %%mm3, %%mm0, %%mm1, %5, %6, %5, 88(%0), 96(%0),104(%0),120(%0), (%1), OP)\
1209 QPEL_V_LOW(%%mm3, %%mm0, %%mm1, %%mm2, %5, %6, %5, 96(%0),104(%0),112(%0),112(%0), (%1, %3), OP)\
1211 "add $136, %0 \n\t"\
1216 : "+r"(temp_ptr), "+r"(dst), "+g"(count)\
1217 : "r"((x86_reg)dstStride), "r"(2*(x86_reg)dstStride), /*"m"(ff_pw_20), "m"(ff_pw_3),*/ "m"(ROUNDER), "g"(4-14*(x86_reg)dstStride)\
1222 static void OPNAME ## mpeg4_qpel8_v_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
1223 uint64_t temp[9*2];\
1224 uint64_t *temp_ptr= temp;\
1229 "pxor %%mm7, %%mm7 \n\t"\
1231 "movq (%0), %%mm0 \n\t"\
1232 "movq (%0), %%mm1 \n\t"\
1233 "punpcklbw %%mm7, %%mm0 \n\t"\
1234 "punpckhbw %%mm7, %%mm1 \n\t"\
1235 "movq %%mm0, (%1) \n\t"\
1236 "movq %%mm1, 9*8(%1) \n\t"\
1241 : "+r" (src), "+r" (temp_ptr), "+r"(count)\
1242 : "r" ((x86_reg)srcStride)\
1249 /*FIXME reorder for speed */\
1251 /*"pxor %%mm7, %%mm7 \n\t"*/\
1253 "movq (%0), %%mm0 \n\t"\
1254 "movq 8(%0), %%mm1 \n\t"\
1255 "movq 16(%0), %%mm2 \n\t"\
1256 "movq 24(%0), %%mm3 \n\t"\
1257 QPEL_V_LOW(%%mm0, %%mm1, %%mm2, %%mm3, %5, %6, %5, 16(%0), 8(%0), (%0), 32(%0), (%1), OP)\
1258 QPEL_V_LOW(%%mm1, %%mm2, %%mm3, %%mm0, %5, %6, %5, 8(%0), (%0), (%0), 40(%0), (%1, %3), OP)\
1260 QPEL_V_LOW(%%mm2, %%mm3, %%mm0, %%mm1, %5, %6, %5, (%0), (%0), 8(%0), 48(%0), (%1), OP)\
1262 QPEL_V_LOW(%%mm3, %%mm0, %%mm1, %%mm2, %5, %6, %5, (%0), 8(%0), 16(%0), 56(%0), (%1, %3), OP)\
1264 QPEL_V_LOW(%%mm0, %%mm1, %%mm2, %%mm3, %5, %6, %5, 8(%0), 16(%0), 24(%0), 64(%0), (%1), OP)\
1266 QPEL_V_LOW(%%mm1, %%mm2, %%mm3, %%mm0, %5, %6, %5, 16(%0), 24(%0), 32(%0), 64(%0), (%1, %3), OP)\
1268 QPEL_V_LOW(%%mm2, %%mm3, %%mm0, %%mm1, %5, %6, %5, 24(%0), 32(%0), 40(%0), 56(%0), (%1), OP)\
1269 QPEL_V_LOW(%%mm3, %%mm0, %%mm1, %%mm2, %5, %6, %5, 32(%0), 40(%0), 48(%0), 48(%0), (%1, %3), OP)\
1276 : "+r"(temp_ptr), "+r"(dst), "+g"(count)\
1277 : "r"((x86_reg)dstStride), "r"(2*(x86_reg)dstStride), /*"m"(ff_pw_20), "m"(ff_pw_3),*/ "m"(ROUNDER), "g"(4-6*(x86_reg)dstStride)\
1282 static void OPNAME ## qpel8_mc00_ ## MMX (uint8_t *dst, uint8_t *src, int stride){\
1283 OPNAME ## pixels8_ ## MMX(dst, src, stride, 8);\
1286 static void OPNAME ## qpel8_mc10_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1288 uint8_t * const half= (uint8_t*)temp;\
1289 put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(half, src, 8, stride, 8);\
1290 OPNAME ## pixels8_l2_ ## MMX(dst, src, half, stride, stride, 8);\
1293 static void OPNAME ## qpel8_mc20_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1294 OPNAME ## mpeg4_qpel8_h_lowpass_ ## MMX(dst, src, stride, stride, 8);\
1297 static void OPNAME ## qpel8_mc30_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1299 uint8_t * const half= (uint8_t*)temp;\
1300 put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(half, src, 8, stride, 8);\
1301 OPNAME ## pixels8_l2_ ## MMX(dst, src+1, half, stride, stride, 8);\
1304 static void OPNAME ## qpel8_mc01_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1306 uint8_t * const half= (uint8_t*)temp;\
1307 put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(half, src, 8, stride);\
1308 OPNAME ## pixels8_l2_ ## MMX(dst, src, half, stride, stride, 8);\
1311 static void OPNAME ## qpel8_mc02_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1312 OPNAME ## mpeg4_qpel8_v_lowpass_ ## MMX(dst, src, stride, stride);\
1315 static void OPNAME ## qpel8_mc03_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1317 uint8_t * const half= (uint8_t*)temp;\
1318 put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(half, src, 8, stride);\
1319 OPNAME ## pixels8_l2_ ## MMX(dst, src+stride, half, stride, stride, 8);\
1321 static void OPNAME ## qpel8_mc11_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1322 uint64_t half[8 + 9];\
1323 uint8_t * const halfH= ((uint8_t*)half) + 64;\
1324 uint8_t * const halfHV= ((uint8_t*)half);\
1325 put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, stride, 9);\
1326 put ## RND ## pixels8_l2_ ## MMX(halfH, src, halfH, 8, stride, 9);\
1327 put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(halfHV, halfH, 8, 8);\
1328 OPNAME ## pixels8_l2_ ## MMX(dst, halfH, halfHV, stride, 8, 8);\
1330 static void OPNAME ## qpel8_mc31_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1331 uint64_t half[8 + 9];\
1332 uint8_t * const halfH= ((uint8_t*)half) + 64;\
1333 uint8_t * const halfHV= ((uint8_t*)half);\
1334 put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, stride, 9);\
1335 put ## RND ## pixels8_l2_ ## MMX(halfH, src+1, halfH, 8, stride, 9);\
1336 put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(halfHV, halfH, 8, 8);\
1337 OPNAME ## pixels8_l2_ ## MMX(dst, halfH, halfHV, stride, 8, 8);\
1339 static void OPNAME ## qpel8_mc13_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1340 uint64_t half[8 + 9];\
1341 uint8_t * const halfH= ((uint8_t*)half) + 64;\
1342 uint8_t * const halfHV= ((uint8_t*)half);\
1343 put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, stride, 9);\
1344 put ## RND ## pixels8_l2_ ## MMX(halfH, src, halfH, 8, stride, 9);\
1345 put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(halfHV, halfH, 8, 8);\
1346 OPNAME ## pixels8_l2_ ## MMX(dst, halfH+8, halfHV, stride, 8, 8);\
1348 static void OPNAME ## qpel8_mc33_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1349 uint64_t half[8 + 9];\
1350 uint8_t * const halfH= ((uint8_t*)half) + 64;\
1351 uint8_t * const halfHV= ((uint8_t*)half);\
1352 put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, stride, 9);\
1353 put ## RND ## pixels8_l2_ ## MMX(halfH, src+1, halfH, 8, stride, 9);\
1354 put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(halfHV, halfH, 8, 8);\
1355 OPNAME ## pixels8_l2_ ## MMX(dst, halfH+8, halfHV, stride, 8, 8);\
1357 static void OPNAME ## qpel8_mc21_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1358 uint64_t half[8 + 9];\
1359 uint8_t * const halfH= ((uint8_t*)half) + 64;\
1360 uint8_t * const halfHV= ((uint8_t*)half);\
1361 put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, stride, 9);\
1362 put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(halfHV, halfH, 8, 8);\
1363 OPNAME ## pixels8_l2_ ## MMX(dst, halfH, halfHV, stride, 8, 8);\
1365 static void OPNAME ## qpel8_mc23_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1366 uint64_t half[8 + 9];\
1367 uint8_t * const halfH= ((uint8_t*)half) + 64;\
1368 uint8_t * const halfHV= ((uint8_t*)half);\
1369 put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, stride, 9);\
1370 put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(halfHV, halfH, 8, 8);\
1371 OPNAME ## pixels8_l2_ ## MMX(dst, halfH+8, halfHV, stride, 8, 8);\
1373 static void OPNAME ## qpel8_mc12_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1374 uint64_t half[8 + 9];\
1375 uint8_t * const halfH= ((uint8_t*)half);\
1376 put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, stride, 9);\
1377 put ## RND ## pixels8_l2_ ## MMX(halfH, src, halfH, 8, stride, 9);\
1378 OPNAME ## mpeg4_qpel8_v_lowpass_ ## MMX(dst, halfH, stride, 8);\
1380 static void OPNAME ## qpel8_mc32_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1381 uint64_t half[8 + 9];\
1382 uint8_t * const halfH= ((uint8_t*)half);\
1383 put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, stride, 9);\
1384 put ## RND ## pixels8_l2_ ## MMX(halfH, src+1, halfH, 8, stride, 9);\
1385 OPNAME ## mpeg4_qpel8_v_lowpass_ ## MMX(dst, halfH, stride, 8);\
1387 static void OPNAME ## qpel8_mc22_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1389 uint8_t * const halfH= ((uint8_t*)half);\
1390 put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, stride, 9);\
1391 OPNAME ## mpeg4_qpel8_v_lowpass_ ## MMX(dst, halfH, stride, 8);\
1393 static void OPNAME ## qpel16_mc00_ ## MMX (uint8_t *dst, uint8_t *src, int stride){\
1394 OPNAME ## pixels16_ ## MMX(dst, src, stride, 16);\
1397 static void OPNAME ## qpel16_mc10_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1399 uint8_t * const half= (uint8_t*)temp;\
1400 put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(half, src, 16, stride, 16);\
1401 OPNAME ## pixels16_l2_ ## MMX(dst, src, half, stride, stride, 16);\
1404 static void OPNAME ## qpel16_mc20_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1405 OPNAME ## mpeg4_qpel16_h_lowpass_ ## MMX(dst, src, stride, stride, 16);\
1408 static void OPNAME ## qpel16_mc30_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1410 uint8_t * const half= (uint8_t*)temp;\
1411 put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(half, src, 16, stride, 16);\
1412 OPNAME ## pixels16_l2_ ## MMX(dst, src+1, half, stride, stride, 16);\
1415 static void OPNAME ## qpel16_mc01_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1417 uint8_t * const half= (uint8_t*)temp;\
1418 put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(half, src, 16, stride);\
1419 OPNAME ## pixels16_l2_ ## MMX(dst, src, half, stride, stride, 16);\
1422 static void OPNAME ## qpel16_mc02_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1423 OPNAME ## mpeg4_qpel16_v_lowpass_ ## MMX(dst, src, stride, stride);\
1426 static void OPNAME ## qpel16_mc03_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1428 uint8_t * const half= (uint8_t*)temp;\
1429 put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(half, src, 16, stride);\
1430 OPNAME ## pixels16_l2_ ## MMX(dst, src+stride, half, stride, stride, 16);\
1432 static void OPNAME ## qpel16_mc11_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1433 uint64_t half[16*2 + 17*2];\
1434 uint8_t * const halfH= ((uint8_t*)half) + 256;\
1435 uint8_t * const halfHV= ((uint8_t*)half);\
1436 put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16, stride, 17);\
1437 put ## RND ## pixels16_l2_ ## MMX(halfH, src, halfH, 16, stride, 17);\
1438 put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(halfHV, halfH, 16, 16);\
1439 OPNAME ## pixels16_l2_ ## MMX(dst, halfH, halfHV, stride, 16, 16);\
1441 static void OPNAME ## qpel16_mc31_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1442 uint64_t half[16*2 + 17*2];\
1443 uint8_t * const halfH= ((uint8_t*)half) + 256;\
1444 uint8_t * const halfHV= ((uint8_t*)half);\
1445 put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16, stride, 17);\
1446 put ## RND ## pixels16_l2_ ## MMX(halfH, src+1, halfH, 16, stride, 17);\
1447 put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(halfHV, halfH, 16, 16);\
1448 OPNAME ## pixels16_l2_ ## MMX(dst, halfH, halfHV, stride, 16, 16);\
1450 static void OPNAME ## qpel16_mc13_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1451 uint64_t half[16*2 + 17*2];\
1452 uint8_t * const halfH= ((uint8_t*)half) + 256;\
1453 uint8_t * const halfHV= ((uint8_t*)half);\
1454 put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16, stride, 17);\
1455 put ## RND ## pixels16_l2_ ## MMX(halfH, src, halfH, 16, stride, 17);\
1456 put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(halfHV, halfH, 16, 16);\
1457 OPNAME ## pixels16_l2_ ## MMX(dst, halfH+16, halfHV, stride, 16, 16);\
1459 static void OPNAME ## qpel16_mc33_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1460 uint64_t half[16*2 + 17*2];\
1461 uint8_t * const halfH= ((uint8_t*)half) + 256;\
1462 uint8_t * const halfHV= ((uint8_t*)half);\
1463 put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16, stride, 17);\
1464 put ## RND ## pixels16_l2_ ## MMX(halfH, src+1, halfH, 16, stride, 17);\
1465 put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(halfHV, halfH, 16, 16);\
1466 OPNAME ## pixels16_l2_ ## MMX(dst, halfH+16, halfHV, stride, 16, 16);\
1468 static void OPNAME ## qpel16_mc21_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1469 uint64_t half[16*2 + 17*2];\
1470 uint8_t * const halfH= ((uint8_t*)half) + 256;\
1471 uint8_t * const halfHV= ((uint8_t*)half);\
1472 put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16, stride, 17);\
1473 put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(halfHV, halfH, 16, 16);\
1474 OPNAME ## pixels16_l2_ ## MMX(dst, halfH, halfHV, stride, 16, 16);\
1476 static void OPNAME ## qpel16_mc23_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1477 uint64_t half[16*2 + 17*2];\
1478 uint8_t * const halfH= ((uint8_t*)half) + 256;\
1479 uint8_t * const halfHV= ((uint8_t*)half);\
1480 put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16, stride, 17);\
1481 put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(halfHV, halfH, 16, 16);\
1482 OPNAME ## pixels16_l2_ ## MMX(dst, halfH+16, halfHV, stride, 16, 16);\
1484 static void OPNAME ## qpel16_mc12_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1485 uint64_t half[17*2];\
1486 uint8_t * const halfH= ((uint8_t*)half);\
1487 put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16, stride, 17);\
1488 put ## RND ## pixels16_l2_ ## MMX(halfH, src, halfH, 16, stride, 17);\
1489 OPNAME ## mpeg4_qpel16_v_lowpass_ ## MMX(dst, halfH, stride, 16);\
1491 static void OPNAME ## qpel16_mc32_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1492 uint64_t half[17*2];\
1493 uint8_t * const halfH= ((uint8_t*)half);\
1494 put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16, stride, 17);\
1495 put ## RND ## pixels16_l2_ ## MMX(halfH, src+1, halfH, 16, stride, 17);\
1496 OPNAME ## mpeg4_qpel16_v_lowpass_ ## MMX(dst, halfH, stride, 16);\
1498 static void OPNAME ## qpel16_mc22_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1499 uint64_t half[17*2];\
1500 uint8_t * const halfH= ((uint8_t*)half);\
1501 put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16, stride, 17);\
1502 OPNAME ## mpeg4_qpel16_v_lowpass_ ## MMX(dst, halfH, stride, 16);\
1505 #define PUT_OP(a,b,temp, size) "mov" #size " " #a ", " #b " \n\t"
1506 #define AVG_3DNOW_OP(a,b,temp, size) \
1507 "mov" #size " " #b ", " #temp " \n\t"\
1508 "pavgusb " #temp ", " #a " \n\t"\
1509 "mov" #size " " #a ", " #b " \n\t"
1510 #define AVG_MMX2_OP(a,b,temp, size) \
1511 "mov" #size " " #b ", " #temp " \n\t"\
1512 "pavgb " #temp ", " #a " \n\t"\
1513 "mov" #size " " #a ", " #b " \n\t"
1515 QPEL_BASE(put_ , ff_pw_16, _ , PUT_OP, PUT_OP)
1516 QPEL_BASE(avg_ , ff_pw_16, _ , AVG_MMX2_OP, AVG_3DNOW_OP)
1517 QPEL_BASE(put_no_rnd_, ff_pw_15, _no_rnd_, PUT_OP, PUT_OP)
1518 QPEL_OP(put_ , ff_pw_16, _ , PUT_OP, 3dnow)
1519 QPEL_OP(avg_ , ff_pw_16, _ , AVG_3DNOW_OP, 3dnow)
1520 QPEL_OP(put_no_rnd_, ff_pw_15, _no_rnd_, PUT_OP, 3dnow)
1521 QPEL_OP(put_ , ff_pw_16, _ , PUT_OP, mmx2)
1522 QPEL_OP(avg_ , ff_pw_16, _ , AVG_MMX2_OP, mmx2)
1523 QPEL_OP(put_no_rnd_, ff_pw_15, _no_rnd_, PUT_OP, mmx2)
1525 /***********************************/
1526 /* bilinear qpel: not compliant to any spec, only for -lavdopts fast */
1528 #define QPEL_2TAP_XY(OPNAME, SIZE, MMX, XY, HPEL)\
1529 static void OPNAME ## 2tap_qpel ## SIZE ## _mc ## XY ## _ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1530 OPNAME ## pixels ## SIZE ## HPEL(dst, src, stride, SIZE);\
1532 #define QPEL_2TAP_L3(OPNAME, SIZE, MMX, XY, S0, S1, S2)\
1533 static void OPNAME ## 2tap_qpel ## SIZE ## _mc ## XY ## _ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1534 OPNAME ## 2tap_qpel ## SIZE ## _l3_ ## MMX(dst, src+S0, stride, SIZE, S1, S2);\
1537 #define QPEL_2TAP(OPNAME, SIZE, MMX)\
1538 QPEL_2TAP_XY(OPNAME, SIZE, MMX, 20, _x2_ ## MMX)\
1539 QPEL_2TAP_XY(OPNAME, SIZE, MMX, 02, _y2_ ## MMX)\
1540 QPEL_2TAP_XY(OPNAME, SIZE, MMX, 22, _xy2_mmx)\
1541 static const qpel_mc_func OPNAME ## 2tap_qpel ## SIZE ## _mc00_ ## MMX =\
1542 OPNAME ## qpel ## SIZE ## _mc00_ ## MMX;\
1543 static const qpel_mc_func OPNAME ## 2tap_qpel ## SIZE ## _mc21_ ## MMX =\
1544 OPNAME ## 2tap_qpel ## SIZE ## _mc20_ ## MMX;\
1545 static const qpel_mc_func OPNAME ## 2tap_qpel ## SIZE ## _mc12_ ## MMX =\
1546 OPNAME ## 2tap_qpel ## SIZE ## _mc02_ ## MMX;\
1547 static void OPNAME ## 2tap_qpel ## SIZE ## _mc32_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1548 OPNAME ## pixels ## SIZE ## _y2_ ## MMX(dst, src+1, stride, SIZE);\
1550 static void OPNAME ## 2tap_qpel ## SIZE ## _mc23_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1551 OPNAME ## pixels ## SIZE ## _x2_ ## MMX(dst, src+stride, stride, SIZE);\
1553 QPEL_2TAP_L3(OPNAME, SIZE, MMX, 10, 0, 1, 0)\
1554 QPEL_2TAP_L3(OPNAME, SIZE, MMX, 30, 1, -1, 0)\
1555 QPEL_2TAP_L3(OPNAME, SIZE, MMX, 01, 0, stride, 0)\
1556 QPEL_2TAP_L3(OPNAME, SIZE, MMX, 03, stride, -stride, 0)\
1557 QPEL_2TAP_L3(OPNAME, SIZE, MMX, 11, 0, stride, 1)\
1558 QPEL_2TAP_L3(OPNAME, SIZE, MMX, 31, 1, stride, -1)\
1559 QPEL_2TAP_L3(OPNAME, SIZE, MMX, 13, stride, -stride, 1)\
1560 QPEL_2TAP_L3(OPNAME, SIZE, MMX, 33, stride+1, -stride, -1)\
1562 QPEL_2TAP(put_, 16, mmx2)
1563 QPEL_2TAP(avg_, 16, mmx2)
1564 QPEL_2TAP(put_, 8, mmx2)
1565 QPEL_2TAP(avg_, 8, mmx2)
1566 QPEL_2TAP(put_, 16, 3dnow)
1567 QPEL_2TAP(avg_, 16, 3dnow)
1568 QPEL_2TAP(put_, 8, 3dnow)
1569 QPEL_2TAP(avg_, 8, 3dnow)
1573 typedef void emu_edge_core_func (uint8_t *buf, const uint8_t *src,
1574 x86_reg linesize, x86_reg start_y,
1575 x86_reg end_y, x86_reg block_h,
1576 x86_reg start_x, x86_reg end_x,
1578 extern emu_edge_core_func ff_emu_edge_core_mmx;
1579 extern emu_edge_core_func ff_emu_edge_core_sse;
1581 static av_always_inline
1582 void emulated_edge_mc(uint8_t *buf, const uint8_t *src, int linesize,
1583 int block_w, int block_h,
1584 int src_x, int src_y, int w, int h,
1585 emu_edge_core_func *core_fn)
1587 int start_y, start_x, end_y, end_x, src_y_add=0;
1590 src_y_add = h-1-src_y;
1592 }else if(src_y<=-block_h){
1593 src_y_add = 1-block_h-src_y;
1599 }else if(src_x<=-block_w){
1600 src+= (1-block_w-src_x);
1604 start_y= FFMAX(0, -src_y);
1605 start_x= FFMAX(0, -src_x);
1606 end_y= FFMIN(block_h, h-src_y);
1607 end_x= FFMIN(block_w, w-src_x);
1608 assert(start_x < end_x && block_w > 0);
1609 assert(start_y < end_y && block_h > 0);
1611 // fill in the to-be-copied part plus all above/below
1612 src += (src_y_add+start_y)*linesize + start_x;
1614 core_fn(buf, src, linesize, start_y, end_y, block_h, start_x, end_x, block_w);
1619 void emulated_edge_mc_mmx(uint8_t *buf, const uint8_t *src, int linesize,
1620 int block_w, int block_h,
1621 int src_x, int src_y, int w, int h)
1623 emulated_edge_mc(buf, src, linesize, block_w, block_h, src_x, src_y,
1624 w, h, &ff_emu_edge_core_mmx);
1628 void emulated_edge_mc_sse(uint8_t *buf, const uint8_t *src, int linesize,
1629 int block_w, int block_h,
1630 int src_x, int src_y, int w, int h)
1632 emulated_edge_mc(buf, src, linesize, block_w, block_h, src_x, src_y,
1633 w, h, &ff_emu_edge_core_sse);
1635 #endif /* HAVE_YASM */
1637 typedef void emulated_edge_mc_func (uint8_t *dst, const uint8_t *src,
1638 int linesize, int block_w, int block_h,
1639 int src_x, int src_y, int w, int h);
1641 static av_always_inline
1642 void gmc(uint8_t *dst, uint8_t *src, int stride, int h, int ox, int oy,
1643 int dxx, int dxy, int dyx, int dyy, int shift, int r, int width, int height,
1644 emulated_edge_mc_func *emu_edge_fn)
1647 const int ix = ox>>(16+shift);
1648 const int iy = oy>>(16+shift);
1649 const int oxs = ox>>4;
1650 const int oys = oy>>4;
1651 const int dxxs = dxx>>4;
1652 const int dxys = dxy>>4;
1653 const int dyxs = dyx>>4;
1654 const int dyys = dyy>>4;
1655 const uint16_t r4[4] = {r,r,r,r};
1656 const uint16_t dxy4[4] = {dxys,dxys,dxys,dxys};
1657 const uint16_t dyy4[4] = {dyys,dyys,dyys,dyys};
1658 const uint64_t shift2 = 2*shift;
1659 uint8_t edge_buf[(h+1)*stride];
1662 const int dxw = (dxx-(1<<(16+shift)))*(w-1);
1663 const int dyh = (dyy-(1<<(16+shift)))*(h-1);
1664 const int dxh = dxy*(h-1);
1665 const int dyw = dyx*(w-1);
1666 if( // non-constant fullpel offset (3% of blocks)
1667 ((ox^(ox+dxw)) | (ox^(ox+dxh)) | (ox^(ox+dxw+dxh)) |
1668 (oy^(oy+dyw)) | (oy^(oy+dyh)) | (oy^(oy+dyw+dyh))) >> (16+shift)
1669 // uses more than 16 bits of subpel mv (only at huge resolution)
1670 || (dxx|dxy|dyx|dyy)&15 )
1672 //FIXME could still use mmx for some of the rows
1673 ff_gmc_c(dst, src, stride, h, ox, oy, dxx, dxy, dyx, dyy, shift, r, width, height);
1677 src += ix + iy*stride;
1678 if( (unsigned)ix >= width-w ||
1679 (unsigned)iy >= height-h )
1681 emu_edge_fn(edge_buf, src, stride, w+1, h+1, ix, iy, width, height);
1686 "movd %0, %%mm6 \n\t"
1687 "pxor %%mm7, %%mm7 \n\t"
1688 "punpcklwd %%mm6, %%mm6 \n\t"
1689 "punpcklwd %%mm6, %%mm6 \n\t"
1693 for(x=0; x<w; x+=4){
1694 uint16_t dx4[4] = { oxs - dxys + dxxs*(x+0),
1695 oxs - dxys + dxxs*(x+1),
1696 oxs - dxys + dxxs*(x+2),
1697 oxs - dxys + dxxs*(x+3) };
1698 uint16_t dy4[4] = { oys - dyys + dyxs*(x+0),
1699 oys - dyys + dyxs*(x+1),
1700 oys - dyys + dyxs*(x+2),
1701 oys - dyys + dyxs*(x+3) };
1705 "movq %0, %%mm4 \n\t"
1706 "movq %1, %%mm5 \n\t"
1707 "paddw %2, %%mm4 \n\t"
1708 "paddw %3, %%mm5 \n\t"
1709 "movq %%mm4, %0 \n\t"
1710 "movq %%mm5, %1 \n\t"
1711 "psrlw $12, %%mm4 \n\t"
1712 "psrlw $12, %%mm5 \n\t"
1713 : "+m"(*dx4), "+m"(*dy4)
1714 : "m"(*dxy4), "m"(*dyy4)
1718 "movq %%mm6, %%mm2 \n\t"
1719 "movq %%mm6, %%mm1 \n\t"
1720 "psubw %%mm4, %%mm2 \n\t"
1721 "psubw %%mm5, %%mm1 \n\t"
1722 "movq %%mm2, %%mm0 \n\t"
1723 "movq %%mm4, %%mm3 \n\t"
1724 "pmullw %%mm1, %%mm0 \n\t" // (s-dx)*(s-dy)
1725 "pmullw %%mm5, %%mm3 \n\t" // dx*dy
1726 "pmullw %%mm5, %%mm2 \n\t" // (s-dx)*dy
1727 "pmullw %%mm4, %%mm1 \n\t" // dx*(s-dy)
1729 "movd %4, %%mm5 \n\t"
1730 "movd %3, %%mm4 \n\t"
1731 "punpcklbw %%mm7, %%mm5 \n\t"
1732 "punpcklbw %%mm7, %%mm4 \n\t"
1733 "pmullw %%mm5, %%mm3 \n\t" // src[1,1] * dx*dy
1734 "pmullw %%mm4, %%mm2 \n\t" // src[0,1] * (s-dx)*dy
1736 "movd %2, %%mm5 \n\t"
1737 "movd %1, %%mm4 \n\t"
1738 "punpcklbw %%mm7, %%mm5 \n\t"
1739 "punpcklbw %%mm7, %%mm4 \n\t"
1740 "pmullw %%mm5, %%mm1 \n\t" // src[1,0] * dx*(s-dy)
1741 "pmullw %%mm4, %%mm0 \n\t" // src[0,0] * (s-dx)*(s-dy)
1742 "paddw %5, %%mm1 \n\t"
1743 "paddw %%mm3, %%mm2 \n\t"
1744 "paddw %%mm1, %%mm0 \n\t"
1745 "paddw %%mm2, %%mm0 \n\t"
1747 "psrlw %6, %%mm0 \n\t"
1748 "packuswb %%mm0, %%mm0 \n\t"
1749 "movd %%mm0, %0 \n\t"
1751 : "=m"(dst[x+y*stride])
1752 : "m"(src[0]), "m"(src[1]),
1753 "m"(src[stride]), "m"(src[stride+1]),
1754 "m"(*r4), "m"(shift2)
1764 static void gmc_mmx(uint8_t *dst, uint8_t *src, int stride, int h, int ox, int oy,
1765 int dxx, int dxy, int dyx, int dyy, int shift, int r, int width, int height)
1767 gmc(dst, src, stride, h, ox, oy, dxx, dxy, dyx, dyy, shift, r,
1768 width, height, &emulated_edge_mc_mmx);
1771 static void gmc_sse(uint8_t *dst, uint8_t *src, int stride, int h, int ox, int oy,
1772 int dxx, int dxy, int dyx, int dyy, int shift, int r, int width, int height)
1774 gmc(dst, src, stride, h, ox, oy, dxx, dxy, dyx, dyy, shift, r,
1775 width, height, &emulated_edge_mc_sse);
1778 static void gmc_mmx(uint8_t *dst, uint8_t *src, int stride, int h, int ox, int oy,
1779 int dxx, int dxy, int dyx, int dyy, int shift, int r, int width, int height)
1781 gmc(dst, src, stride, h, ox, oy, dxx, dxy, dyx, dyy, shift, r,
1782 width, height, &ff_emulated_edge_mc_8);
1786 #define PREFETCH(name, op) \
1787 static void name(void *mem, int stride, int h){\
1788 const uint8_t *p= mem;\
1790 __asm__ volatile(#op" %0" :: "m"(*p));\
1794 PREFETCH(prefetch_mmx2, prefetcht0)
1795 PREFETCH(prefetch_3dnow, prefetch)
1798 #include "h264_qpel_mmx.c"
1800 void ff_put_h264_chroma_mc8_mmx_rnd (uint8_t *dst, uint8_t *src,
1801 int stride, int h, int x, int y);
1802 void ff_put_rv40_chroma_mc8_mmx (uint8_t *dst, uint8_t *src,
1803 int stride, int h, int x, int y);
1804 void ff_avg_h264_chroma_mc8_mmx2_rnd (uint8_t *dst, uint8_t *src,
1805 int stride, int h, int x, int y);
1806 void ff_avg_rv40_chroma_mc8_mmx2 (uint8_t *dst, uint8_t *src,
1807 int stride, int h, int x, int y);
1808 void ff_avg_h264_chroma_mc8_3dnow_rnd (uint8_t *dst, uint8_t *src,
1809 int stride, int h, int x, int y);
1810 void ff_avg_rv40_chroma_mc8_3dnow (uint8_t *dst, uint8_t *src,
1811 int stride, int h, int x, int y);
1813 void ff_put_h264_chroma_mc4_mmx (uint8_t *dst, uint8_t *src,
1814 int stride, int h, int x, int y);
1815 void ff_put_rv40_chroma_mc4_mmx (uint8_t *dst, uint8_t *src,
1816 int stride, int h, int x, int y);
1817 void ff_avg_h264_chroma_mc4_mmx2 (uint8_t *dst, uint8_t *src,
1818 int stride, int h, int x, int y);
1819 void ff_avg_rv40_chroma_mc4_mmx2 (uint8_t *dst, uint8_t *src,
1820 int stride, int h, int x, int y);
1821 void ff_avg_h264_chroma_mc4_3dnow (uint8_t *dst, uint8_t *src,
1822 int stride, int h, int x, int y);
1823 void ff_avg_rv40_chroma_mc4_3dnow (uint8_t *dst, uint8_t *src,
1824 int stride, int h, int x, int y);
1826 void ff_put_h264_chroma_mc2_mmx2 (uint8_t *dst, uint8_t *src,
1827 int stride, int h, int x, int y);
1828 void ff_avg_h264_chroma_mc2_mmx2 (uint8_t *dst, uint8_t *src,
1829 int stride, int h, int x, int y);
1831 void ff_put_h264_chroma_mc8_ssse3_rnd (uint8_t *dst, uint8_t *src,
1832 int stride, int h, int x, int y);
1833 void ff_put_h264_chroma_mc4_ssse3 (uint8_t *dst, uint8_t *src,
1834 int stride, int h, int x, int y);
1836 void ff_avg_h264_chroma_mc8_ssse3_rnd (uint8_t *dst, uint8_t *src,
1837 int stride, int h, int x, int y);
1838 void ff_avg_h264_chroma_mc4_ssse3 (uint8_t *dst, uint8_t *src,
1839 int stride, int h, int x, int y);
1841 #define CHROMA_MC(OP, NUM, DEPTH, OPT) \
1842 void ff_ ## OP ## _h264_chroma_mc ## NUM ## _ ## DEPTH ## _ ## OPT \
1843 (uint8_t *dst, uint8_t *src,\
1844 int stride, int h, int x, int y);
1846 CHROMA_MC(put, 2, 10, mmxext)
1847 CHROMA_MC(avg, 2, 10, mmxext)
1848 CHROMA_MC(put, 4, 10, mmxext)
1849 CHROMA_MC(avg, 4, 10, mmxext)
1850 CHROMA_MC(put, 8, 10, sse2)
1851 CHROMA_MC(avg, 8, 10, sse2)
1852 CHROMA_MC(put, 8, 10, avx)
1853 CHROMA_MC(avg, 8, 10, avx)
1856 void ff_put_cavs_qpel8_mc00_mmx2(uint8_t *dst, uint8_t *src, int stride) {
1857 put_pixels8_mmx(dst, src, stride, 8);
1859 void ff_avg_cavs_qpel8_mc00_mmx2(uint8_t *dst, uint8_t *src, int stride) {
1860 avg_pixels8_mmx(dst, src, stride, 8);
1862 void ff_put_cavs_qpel16_mc00_mmx2(uint8_t *dst, uint8_t *src, int stride) {
1863 put_pixels16_mmx(dst, src, stride, 16);
1865 void ff_avg_cavs_qpel16_mc00_mmx2(uint8_t *dst, uint8_t *src, int stride) {
1866 avg_pixels16_mmx(dst, src, stride, 16);
1870 void ff_put_vc1_mspel_mc00_mmx(uint8_t *dst, const uint8_t *src, int stride, int rnd) {
1871 put_pixels8_mmx(dst, src, stride, 8);
1873 void ff_avg_vc1_mspel_mc00_mmx2(uint8_t *dst, const uint8_t *src, int stride, int rnd) {
1874 avg_pixels8_mmx2(dst, src, stride, 8);
1877 /* XXX: those functions should be suppressed ASAP when all IDCTs are
1880 static void ff_libmpeg2mmx_idct_put(uint8_t *dest, int line_size, DCTELEM *block)
1882 ff_mmx_idct (block);
1883 ff_put_pixels_clamped_mmx(block, dest, line_size);
1885 static void ff_libmpeg2mmx_idct_add(uint8_t *dest, int line_size, DCTELEM *block)
1887 ff_mmx_idct (block);
1888 ff_add_pixels_clamped_mmx(block, dest, line_size);
1890 static void ff_libmpeg2mmx2_idct_put(uint8_t *dest, int line_size, DCTELEM *block)
1892 ff_mmxext_idct (block);
1893 ff_put_pixels_clamped_mmx(block, dest, line_size);
1895 static void ff_libmpeg2mmx2_idct_add(uint8_t *dest, int line_size, DCTELEM *block)
1897 ff_mmxext_idct (block);
1898 ff_add_pixels_clamped_mmx(block, dest, line_size);
1901 static void ff_idct_xvid_mmx_put(uint8_t *dest, int line_size, DCTELEM *block)
1903 ff_idct_xvid_mmx (block);
1904 ff_put_pixels_clamped_mmx(block, dest, line_size);
1906 static void ff_idct_xvid_mmx_add(uint8_t *dest, int line_size, DCTELEM *block)
1908 ff_idct_xvid_mmx (block);
1909 ff_add_pixels_clamped_mmx(block, dest, line_size);
1911 static void ff_idct_xvid_mmx2_put(uint8_t *dest, int line_size, DCTELEM *block)
1913 ff_idct_xvid_mmx2 (block);
1914 ff_put_pixels_clamped_mmx(block, dest, line_size);
1916 static void ff_idct_xvid_mmx2_add(uint8_t *dest, int line_size, DCTELEM *block)
1918 ff_idct_xvid_mmx2 (block);
1919 ff_add_pixels_clamped_mmx(block, dest, line_size);
1922 static void vorbis_inverse_coupling_3dnow(float *mag, float *ang, int blocksize)
1925 __asm__ volatile("pxor %%mm7, %%mm7":);
1926 for(i=0; i<blocksize; i+=2) {
1928 "movq %0, %%mm0 \n\t"
1929 "movq %1, %%mm1 \n\t"
1930 "movq %%mm0, %%mm2 \n\t"
1931 "movq %%mm1, %%mm3 \n\t"
1932 "pfcmpge %%mm7, %%mm2 \n\t" // m <= 0.0
1933 "pfcmpge %%mm7, %%mm3 \n\t" // a <= 0.0
1934 "pslld $31, %%mm2 \n\t" // keep only the sign bit
1935 "pxor %%mm2, %%mm1 \n\t"
1936 "movq %%mm3, %%mm4 \n\t"
1937 "pand %%mm1, %%mm3 \n\t"
1938 "pandn %%mm1, %%mm4 \n\t"
1939 "pfadd %%mm0, %%mm3 \n\t" // a = m + ((a<0) & (a ^ sign(m)))
1940 "pfsub %%mm4, %%mm0 \n\t" // m = m + ((a>0) & (a ^ sign(m)))
1941 "movq %%mm3, %1 \n\t"
1942 "movq %%mm0, %0 \n\t"
1943 :"+m"(mag[i]), "+m"(ang[i])
1947 __asm__ volatile("femms");
1949 static void vorbis_inverse_coupling_sse(float *mag, float *ang, int blocksize)
1954 "movaps %0, %%xmm5 \n\t"
1955 ::"m"(ff_pdw_80000000[0])
1957 for(i=0; i<blocksize; i+=4) {
1959 "movaps %0, %%xmm0 \n\t"
1960 "movaps %1, %%xmm1 \n\t"
1961 "xorps %%xmm2, %%xmm2 \n\t"
1962 "xorps %%xmm3, %%xmm3 \n\t"
1963 "cmpleps %%xmm0, %%xmm2 \n\t" // m <= 0.0
1964 "cmpleps %%xmm1, %%xmm3 \n\t" // a <= 0.0
1965 "andps %%xmm5, %%xmm2 \n\t" // keep only the sign bit
1966 "xorps %%xmm2, %%xmm1 \n\t"
1967 "movaps %%xmm3, %%xmm4 \n\t"
1968 "andps %%xmm1, %%xmm3 \n\t"
1969 "andnps %%xmm1, %%xmm4 \n\t"
1970 "addps %%xmm0, %%xmm3 \n\t" // a = m + ((a<0) & (a ^ sign(m)))
1971 "subps %%xmm4, %%xmm0 \n\t" // m = m + ((a>0) & (a ^ sign(m)))
1972 "movaps %%xmm3, %1 \n\t"
1973 "movaps %%xmm0, %0 \n\t"
1974 :"+m"(mag[i]), "+m"(ang[i])
1983 #define MIX5(mono,stereo)\
1985 "movss 0(%2), %%xmm5 \n"\
1986 "movss 8(%2), %%xmm6 \n"\
1987 "movss 24(%2), %%xmm7 \n"\
1988 "shufps $0, %%xmm5, %%xmm5 \n"\
1989 "shufps $0, %%xmm6, %%xmm6 \n"\
1990 "shufps $0, %%xmm7, %%xmm7 \n"\
1992 "movaps (%0,%1), %%xmm0 \n"\
1993 "movaps 0x400(%0,%1), %%xmm1 \n"\
1994 "movaps 0x800(%0,%1), %%xmm2 \n"\
1995 "movaps 0xc00(%0,%1), %%xmm3 \n"\
1996 "movaps 0x1000(%0,%1), %%xmm4 \n"\
1997 "mulps %%xmm5, %%xmm0 \n"\
1998 "mulps %%xmm6, %%xmm1 \n"\
1999 "mulps %%xmm5, %%xmm2 \n"\
2000 "mulps %%xmm7, %%xmm3 \n"\
2001 "mulps %%xmm7, %%xmm4 \n"\
2002 stereo("addps %%xmm1, %%xmm0 \n")\
2003 "addps %%xmm1, %%xmm2 \n"\
2004 "addps %%xmm3, %%xmm0 \n"\
2005 "addps %%xmm4, %%xmm2 \n"\
2006 mono("addps %%xmm2, %%xmm0 \n")\
2007 "movaps %%xmm0, (%0,%1) \n"\
2008 stereo("movaps %%xmm2, 0x400(%0,%1) \n")\
2012 :"r"(samples[0]+len), "r"(matrix)\
2013 :XMM_CLOBBERS("%xmm0", "%xmm1", "%xmm2", "%xmm3", \
2014 "%xmm4", "%xmm5", "%xmm6", "%xmm7",)\
2018 #define MIX_MISC(stereo)\
2021 "movaps (%3,%0), %%xmm0 \n"\
2022 stereo("movaps %%xmm0, %%xmm1 \n")\
2023 "mulps %%xmm4, %%xmm0 \n"\
2024 stereo("mulps %%xmm5, %%xmm1 \n")\
2025 "lea 1024(%3,%0), %1 \n"\
2028 "movaps (%1), %%xmm2 \n"\
2029 stereo("movaps %%xmm2, %%xmm3 \n")\
2030 "mulps (%4,%2), %%xmm2 \n"\
2031 stereo("mulps 16(%4,%2), %%xmm3 \n")\
2032 "addps %%xmm2, %%xmm0 \n"\
2033 stereo("addps %%xmm3, %%xmm1 \n")\
2037 "movaps %%xmm0, (%3,%0) \n"\
2038 stereo("movaps %%xmm1, 1024(%3,%0) \n")\
2041 :"+&r"(i), "=&r"(j), "=&r"(k)\
2042 :"r"(samples[0]+len), "r"(matrix_simd+in_ch), "g"((intptr_t)-32*(in_ch-1))\
2046 static void ac3_downmix_sse(float (*samples)[256], float (*matrix)[2], int out_ch, int in_ch, int len)
2048 int (*matrix_cmp)[2] = (int(*)[2])matrix;
2051 i = -len*sizeof(float);
2052 if(in_ch == 5 && out_ch == 2 && !(matrix_cmp[0][1]|matrix_cmp[2][0]|matrix_cmp[3][1]|matrix_cmp[4][0]|(matrix_cmp[1][0]^matrix_cmp[1][1])|(matrix_cmp[0][0]^matrix_cmp[2][1]))) {
2054 } else if(in_ch == 5 && out_ch == 1 && matrix_cmp[0][0]==matrix_cmp[2][0] && matrix_cmp[3][0]==matrix_cmp[4][0]) {
2057 DECLARE_ALIGNED(16, float, matrix_simd)[AC3_MAX_CHANNELS][2][4];
2058 j = 2*in_ch*sizeof(float);
2062 "movss (%2,%0), %%xmm4 \n"
2063 "movss 4(%2,%0), %%xmm5 \n"
2064 "shufps $0, %%xmm4, %%xmm4 \n"
2065 "shufps $0, %%xmm5, %%xmm5 \n"
2066 "movaps %%xmm4, (%1,%0,4) \n"
2067 "movaps %%xmm5, 16(%1,%0,4) \n"
2070 :"r"(matrix_simd), "r"(matrix)
2081 static void vector_fmul_3dnow(float *dst, const float *src0, const float *src1, int len){
2082 x86_reg i = (len-4)*4;
2085 "movq (%2,%0), %%mm0 \n\t"
2086 "movq 8(%2,%0), %%mm1 \n\t"
2087 "pfmul (%3,%0), %%mm0 \n\t"
2088 "pfmul 8(%3,%0), %%mm1 \n\t"
2089 "movq %%mm0, (%1,%0) \n\t"
2090 "movq %%mm1, 8(%1,%0) \n\t"
2095 :"r"(dst), "r"(src0), "r"(src1)
2099 static void vector_fmul_sse(float *dst, const float *src0, const float *src1, int len){
2100 x86_reg i = (len-8)*4;
2103 "movaps (%2,%0), %%xmm0 \n\t"
2104 "movaps 16(%2,%0), %%xmm1 \n\t"
2105 "mulps (%3,%0), %%xmm0 \n\t"
2106 "mulps 16(%3,%0), %%xmm1 \n\t"
2107 "movaps %%xmm0, (%1,%0) \n\t"
2108 "movaps %%xmm1, 16(%1,%0) \n\t"
2112 :"r"(dst), "r"(src0), "r"(src1)
2117 static void vector_fmul_reverse_3dnow2(float *dst, const float *src0, const float *src1, int len){
2118 x86_reg i = len*4-16;
2121 "pswapd 8(%1), %%mm0 \n\t"
2122 "pswapd (%1), %%mm1 \n\t"
2123 "pfmul (%3,%0), %%mm0 \n\t"
2124 "pfmul 8(%3,%0), %%mm1 \n\t"
2125 "movq %%mm0, (%2,%0) \n\t"
2126 "movq %%mm1, 8(%2,%0) \n\t"
2130 :"+r"(i), "+r"(src1)
2131 :"r"(dst), "r"(src0)
2133 __asm__ volatile("femms");
2135 static void vector_fmul_reverse_sse(float *dst, const float *src0, const float *src1, int len){
2136 x86_reg i = len*4-32;
2139 "movaps 16(%1), %%xmm0 \n\t"
2140 "movaps (%1), %%xmm1 \n\t"
2141 "shufps $0x1b, %%xmm0, %%xmm0 \n\t"
2142 "shufps $0x1b, %%xmm1, %%xmm1 \n\t"
2143 "mulps (%3,%0), %%xmm0 \n\t"
2144 "mulps 16(%3,%0), %%xmm1 \n\t"
2145 "movaps %%xmm0, (%2,%0) \n\t"
2146 "movaps %%xmm1, 16(%2,%0) \n\t"
2150 :"+r"(i), "+r"(src1)
2151 :"r"(dst), "r"(src0)
2155 static void vector_fmul_add_3dnow(float *dst, const float *src0, const float *src1,
2156 const float *src2, int len){
2157 x86_reg i = (len-4)*4;
2160 "movq (%2,%0), %%mm0 \n\t"
2161 "movq 8(%2,%0), %%mm1 \n\t"
2162 "pfmul (%3,%0), %%mm0 \n\t"
2163 "pfmul 8(%3,%0), %%mm1 \n\t"
2164 "pfadd (%4,%0), %%mm0 \n\t"
2165 "pfadd 8(%4,%0), %%mm1 \n\t"
2166 "movq %%mm0, (%1,%0) \n\t"
2167 "movq %%mm1, 8(%1,%0) \n\t"
2171 :"r"(dst), "r"(src0), "r"(src1), "r"(src2)
2174 __asm__ volatile("femms");
2176 static void vector_fmul_add_sse(float *dst, const float *src0, const float *src1,
2177 const float *src2, int len){
2178 x86_reg i = (len-8)*4;
2181 "movaps (%2,%0), %%xmm0 \n\t"
2182 "movaps 16(%2,%0), %%xmm1 \n\t"
2183 "mulps (%3,%0), %%xmm0 \n\t"
2184 "mulps 16(%3,%0), %%xmm1 \n\t"
2185 "addps (%4,%0), %%xmm0 \n\t"
2186 "addps 16(%4,%0), %%xmm1 \n\t"
2187 "movaps %%xmm0, (%1,%0) \n\t"
2188 "movaps %%xmm1, 16(%1,%0) \n\t"
2192 :"r"(dst), "r"(src0), "r"(src1), "r"(src2)
2198 static void vector_fmul_window_3dnow2(float *dst, const float *src0, const float *src1,
2199 const float *win, int len){
2201 x86_reg j = len*4-8;
2204 "pswapd (%5,%1), %%mm1 \n"
2205 "movq (%5,%0), %%mm0 \n"
2206 "pswapd (%4,%1), %%mm5 \n"
2207 "movq (%3,%0), %%mm4 \n"
2208 "movq %%mm0, %%mm2 \n"
2209 "movq %%mm1, %%mm3 \n"
2210 "pfmul %%mm4, %%mm2 \n" // src0[len+i]*win[len+i]
2211 "pfmul %%mm5, %%mm3 \n" // src1[ j]*win[len+j]
2212 "pfmul %%mm4, %%mm1 \n" // src0[len+i]*win[len+j]
2213 "pfmul %%mm5, %%mm0 \n" // src1[ j]*win[len+i]
2214 "pfadd %%mm3, %%mm2 \n"
2215 "pfsub %%mm0, %%mm1 \n"
2216 "pswapd %%mm2, %%mm2 \n"
2217 "movq %%mm1, (%2,%0) \n"
2218 "movq %%mm2, (%2,%1) \n"
2224 :"r"(dst+len), "r"(src0+len), "r"(src1), "r"(win+len)
2228 static void vector_fmul_window_sse(float *dst, const float *src0, const float *src1,
2229 const float *win, int len){
2231 x86_reg j = len*4-16;
2234 "movaps (%5,%1), %%xmm1 \n"
2235 "movaps (%5,%0), %%xmm0 \n"
2236 "movaps (%4,%1), %%xmm5 \n"
2237 "movaps (%3,%0), %%xmm4 \n"
2238 "shufps $0x1b, %%xmm1, %%xmm1 \n"
2239 "shufps $0x1b, %%xmm5, %%xmm5 \n"
2240 "movaps %%xmm0, %%xmm2 \n"
2241 "movaps %%xmm1, %%xmm3 \n"
2242 "mulps %%xmm4, %%xmm2 \n" // src0[len+i]*win[len+i]
2243 "mulps %%xmm5, %%xmm3 \n" // src1[ j]*win[len+j]
2244 "mulps %%xmm4, %%xmm1 \n" // src0[len+i]*win[len+j]
2245 "mulps %%xmm5, %%xmm0 \n" // src1[ j]*win[len+i]
2246 "addps %%xmm3, %%xmm2 \n"
2247 "subps %%xmm0, %%xmm1 \n"
2248 "shufps $0x1b, %%xmm2, %%xmm2 \n"
2249 "movaps %%xmm1, (%2,%0) \n"
2250 "movaps %%xmm2, (%2,%1) \n"
2255 :"r"(dst+len), "r"(src0+len), "r"(src1), "r"(win+len)
2258 #endif /* HAVE_6REGS */
2260 static void vector_clipf_sse(float *dst, const float *src, float min, float max,
2263 x86_reg i = (len-16)*4;
2265 "movss %3, %%xmm4 \n"
2266 "movss %4, %%xmm5 \n"
2267 "shufps $0, %%xmm4, %%xmm4 \n"
2268 "shufps $0, %%xmm5, %%xmm5 \n"
2270 "movaps (%2,%0), %%xmm0 \n\t" // 3/1 on intel
2271 "movaps 16(%2,%0), %%xmm1 \n\t"
2272 "movaps 32(%2,%0), %%xmm2 \n\t"
2273 "movaps 48(%2,%0), %%xmm3 \n\t"
2274 "maxps %%xmm4, %%xmm0 \n\t"
2275 "maxps %%xmm4, %%xmm1 \n\t"
2276 "maxps %%xmm4, %%xmm2 \n\t"
2277 "maxps %%xmm4, %%xmm3 \n\t"
2278 "minps %%xmm5, %%xmm0 \n\t"
2279 "minps %%xmm5, %%xmm1 \n\t"
2280 "minps %%xmm5, %%xmm2 \n\t"
2281 "minps %%xmm5, %%xmm3 \n\t"
2282 "movaps %%xmm0, (%1,%0) \n\t"
2283 "movaps %%xmm1, 16(%1,%0) \n\t"
2284 "movaps %%xmm2, 32(%1,%0) \n\t"
2285 "movaps %%xmm3, 48(%1,%0) \n\t"
2289 :"r"(dst), "r"(src), "m"(min), "m"(max)
2294 void ff_vp3_idct_mmx(int16_t *input_data);
2295 void ff_vp3_idct_put_mmx(uint8_t *dest, int line_size, DCTELEM *block);
2296 void ff_vp3_idct_add_mmx(uint8_t *dest, int line_size, DCTELEM *block);
2298 void ff_vp3_idct_dc_add_mmx2(uint8_t *dest, int line_size, const DCTELEM *block);
2300 void ff_vp3_v_loop_filter_mmx2(uint8_t *src, int stride, int *bounding_values);
2301 void ff_vp3_h_loop_filter_mmx2(uint8_t *src, int stride, int *bounding_values);
2303 void ff_vp3_idct_sse2(int16_t *input_data);
2304 void ff_vp3_idct_put_sse2(uint8_t *dest, int line_size, DCTELEM *block);
2305 void ff_vp3_idct_add_sse2(uint8_t *dest, int line_size, DCTELEM *block);
2307 int32_t ff_scalarproduct_int16_mmx2(const int16_t *v1, const int16_t *v2, int order, int shift);
2308 int32_t ff_scalarproduct_int16_sse2(const int16_t *v1, const int16_t *v2, int order, int shift);
2309 int32_t ff_scalarproduct_and_madd_int16_mmx2(int16_t *v1, const int16_t *v2, const int16_t *v3, int order, int mul);
2310 int32_t ff_scalarproduct_and_madd_int16_sse2(int16_t *v1, const int16_t *v2, const int16_t *v3, int order, int mul);
2311 int32_t ff_scalarproduct_and_madd_int16_ssse3(int16_t *v1, const int16_t *v2, const int16_t *v3, int order, int mul);
2313 void ff_apply_window_int16_mmxext (int16_t *output, const int16_t *input,
2314 const int16_t *window, unsigned int len);
2315 void ff_apply_window_int16_mmxext_ba (int16_t *output, const int16_t *input,
2316 const int16_t *window, unsigned int len);
2317 void ff_apply_window_int16_sse2 (int16_t *output, const int16_t *input,
2318 const int16_t *window, unsigned int len);
2319 void ff_apply_window_int16_sse2_ba (int16_t *output, const int16_t *input,
2320 const int16_t *window, unsigned int len);
2321 void ff_apply_window_int16_ssse3 (int16_t *output, const int16_t *input,
2322 const int16_t *window, unsigned int len);
2323 void ff_apply_window_int16_ssse3_atom(int16_t *output, const int16_t *input,
2324 const int16_t *window, unsigned int len);
2326 void ff_add_hfyu_median_prediction_mmx2(uint8_t *dst, const uint8_t *top, const uint8_t *diff, int w, int *left, int *left_top);
2327 int ff_add_hfyu_left_prediction_ssse3(uint8_t *dst, const uint8_t *src, int w, int left);
2328 int ff_add_hfyu_left_prediction_sse4(uint8_t *dst, const uint8_t *src, int w, int left);
2330 float ff_scalarproduct_float_sse(const float *v1, const float *v2, int order);
2332 void ff_vector_clip_int32_mmx (int32_t *dst, const int32_t *src, int32_t min,
2333 int32_t max, unsigned int len);
2334 void ff_vector_clip_int32_sse2 (int32_t *dst, const int32_t *src, int32_t min,
2335 int32_t max, unsigned int len);
2336 void ff_vector_clip_int32_sse2_int(int32_t *dst, const int32_t *src, int32_t min,
2337 int32_t max, unsigned int len);
2338 void ff_vector_clip_int32_sse41 (int32_t *dst, const int32_t *src, int32_t min,
2339 int32_t max, unsigned int len);
2341 void dsputil_init_mmx(DSPContext* c, AVCodecContext *avctx)
2343 int mm_flags = av_get_cpu_flags();
2344 const int high_bit_depth = avctx->codec_id == CODEC_ID_H264 && avctx->bits_per_raw_sample > 8;
2345 const int bit_depth = avctx->bits_per_raw_sample;
2347 if (avctx->dsp_mask) {
2348 if (avctx->dsp_mask & AV_CPU_FLAG_FORCE)
2349 mm_flags |= (avctx->dsp_mask & 0xffff);
2351 mm_flags &= ~(avctx->dsp_mask & 0xffff);
2355 av_log(avctx, AV_LOG_INFO, "libavcodec: CPU flags:");
2356 if (mm_flags & AV_CPU_FLAG_MMX)
2357 av_log(avctx, AV_LOG_INFO, " mmx");
2358 if (mm_flags & AV_CPU_FLAG_MMX2)
2359 av_log(avctx, AV_LOG_INFO, " mmx2");
2360 if (mm_flags & AV_CPU_FLAG_3DNOW)
2361 av_log(avctx, AV_LOG_INFO, " 3dnow");
2362 if (mm_flags & AV_CPU_FLAG_SSE)
2363 av_log(avctx, AV_LOG_INFO, " sse");
2364 if (mm_flags & AV_CPU_FLAG_SSE2)
2365 av_log(avctx, AV_LOG_INFO, " sse2");
2366 av_log(avctx, AV_LOG_INFO, "\n");
2369 if (mm_flags & AV_CPU_FLAG_MMX) {
2370 const int idct_algo= avctx->idct_algo;
2372 if(avctx->lowres==0){
2373 if(idct_algo==FF_IDCT_AUTO || idct_algo==FF_IDCT_SIMPLEMMX){
2374 c->idct_put= ff_simple_idct_put_mmx;
2375 c->idct_add= ff_simple_idct_add_mmx;
2376 c->idct = ff_simple_idct_mmx;
2377 c->idct_permutation_type= FF_SIMPLE_IDCT_PERM;
2379 }else if(idct_algo==FF_IDCT_LIBMPEG2MMX){
2380 if(mm_flags & AV_CPU_FLAG_MMX2){
2381 c->idct_put= ff_libmpeg2mmx2_idct_put;
2382 c->idct_add= ff_libmpeg2mmx2_idct_add;
2383 c->idct = ff_mmxext_idct;
2385 c->idct_put= ff_libmpeg2mmx_idct_put;
2386 c->idct_add= ff_libmpeg2mmx_idct_add;
2387 c->idct = ff_mmx_idct;
2389 c->idct_permutation_type= FF_LIBMPEG2_IDCT_PERM;
2391 }else if((CONFIG_VP3_DECODER || CONFIG_VP5_DECODER || CONFIG_VP6_DECODER) &&
2392 idct_algo==FF_IDCT_VP3 && HAVE_YASM){
2393 if(mm_flags & AV_CPU_FLAG_SSE2){
2394 c->idct_put= ff_vp3_idct_put_sse2;
2395 c->idct_add= ff_vp3_idct_add_sse2;
2396 c->idct = ff_vp3_idct_sse2;
2397 c->idct_permutation_type= FF_TRANSPOSE_IDCT_PERM;
2399 c->idct_put= ff_vp3_idct_put_mmx;
2400 c->idct_add= ff_vp3_idct_add_mmx;
2401 c->idct = ff_vp3_idct_mmx;
2402 c->idct_permutation_type= FF_PARTTRANS_IDCT_PERM;
2404 }else if(idct_algo==FF_IDCT_CAVS){
2405 c->idct_permutation_type= FF_TRANSPOSE_IDCT_PERM;
2406 }else if(idct_algo==FF_IDCT_XVIDMMX){
2407 if(mm_flags & AV_CPU_FLAG_SSE2){
2408 c->idct_put= ff_idct_xvid_sse2_put;
2409 c->idct_add= ff_idct_xvid_sse2_add;
2410 c->idct = ff_idct_xvid_sse2;
2411 c->idct_permutation_type= FF_SSE2_IDCT_PERM;
2412 }else if(mm_flags & AV_CPU_FLAG_MMX2){
2413 c->idct_put= ff_idct_xvid_mmx2_put;
2414 c->idct_add= ff_idct_xvid_mmx2_add;
2415 c->idct = ff_idct_xvid_mmx2;
2417 c->idct_put= ff_idct_xvid_mmx_put;
2418 c->idct_add= ff_idct_xvid_mmx_add;
2419 c->idct = ff_idct_xvid_mmx;
2424 c->put_pixels_clamped = ff_put_pixels_clamped_mmx;
2425 c->put_signed_pixels_clamped = ff_put_signed_pixels_clamped_mmx;
2426 c->add_pixels_clamped = ff_add_pixels_clamped_mmx;
2427 if (!high_bit_depth) {
2428 c->clear_block = clear_block_mmx;
2429 c->clear_blocks = clear_blocks_mmx;
2430 if ((mm_flags & AV_CPU_FLAG_SSE) &&
2431 !(CONFIG_MPEG_XVMC_DECODER && avctx->xvmc_acceleration > 1)){
2432 /* XvMCCreateBlocks() may not allocate 16-byte aligned blocks */
2433 c->clear_block = clear_block_sse;
2434 c->clear_blocks = clear_blocks_sse;
2438 #define SET_HPEL_FUNCS(PFX, IDX, SIZE, CPU) \
2439 c->PFX ## _pixels_tab[IDX][0] = PFX ## _pixels ## SIZE ## _ ## CPU; \
2440 c->PFX ## _pixels_tab[IDX][1] = PFX ## _pixels ## SIZE ## _x2_ ## CPU; \
2441 c->PFX ## _pixels_tab[IDX][2] = PFX ## _pixels ## SIZE ## _y2_ ## CPU; \
2442 c->PFX ## _pixels_tab[IDX][3] = PFX ## _pixels ## SIZE ## _xy2_ ## CPU
2444 if (!high_bit_depth) {
2445 SET_HPEL_FUNCS(put, 0, 16, mmx);
2446 SET_HPEL_FUNCS(put_no_rnd, 0, 16, mmx);
2447 SET_HPEL_FUNCS(avg, 0, 16, mmx);
2448 SET_HPEL_FUNCS(avg_no_rnd, 0, 16, mmx);
2449 SET_HPEL_FUNCS(put, 1, 8, mmx);
2450 SET_HPEL_FUNCS(put_no_rnd, 1, 8, mmx);
2451 SET_HPEL_FUNCS(avg, 1, 8, mmx);
2452 SET_HPEL_FUNCS(avg_no_rnd, 1, 8, mmx);
2455 #if ARCH_X86_32 || !HAVE_YASM
2458 #if ARCH_X86_32 && HAVE_YASM
2459 if (!high_bit_depth)
2460 c->emulated_edge_mc = emulated_edge_mc_mmx;
2463 c->add_bytes= add_bytes_mmx;
2465 if (!high_bit_depth)
2466 c->draw_edges = draw_edges_mmx;
2468 if (CONFIG_H263_DECODER || CONFIG_H263_ENCODER) {
2469 c->h263_v_loop_filter= h263_v_loop_filter_mmx;
2470 c->h263_h_loop_filter= h263_h_loop_filter_mmx;
2474 if (!high_bit_depth) {
2475 c->put_h264_chroma_pixels_tab[0]= ff_put_h264_chroma_mc8_mmx_rnd;
2476 c->put_h264_chroma_pixels_tab[1]= ff_put_h264_chroma_mc4_mmx;
2479 c->put_rv40_chroma_pixels_tab[0]= ff_put_rv40_chroma_mc8_mmx;
2480 c->put_rv40_chroma_pixels_tab[1]= ff_put_rv40_chroma_mc4_mmx;
2482 c->vector_clip_int32 = ff_vector_clip_int32_mmx;
2485 if (mm_flags & AV_CPU_FLAG_MMX2) {
2486 c->prefetch = prefetch_mmx2;
2488 if (!high_bit_depth) {
2489 c->put_pixels_tab[0][1] = put_pixels16_x2_mmx2;
2490 c->put_pixels_tab[0][2] = put_pixels16_y2_mmx2;
2492 c->avg_pixels_tab[0][0] = avg_pixels16_mmx2;
2493 c->avg_pixels_tab[0][1] = avg_pixels16_x2_mmx2;
2494 c->avg_pixels_tab[0][2] = avg_pixels16_y2_mmx2;
2496 c->put_pixels_tab[1][1] = put_pixels8_x2_mmx2;
2497 c->put_pixels_tab[1][2] = put_pixels8_y2_mmx2;
2499 c->avg_pixels_tab[1][0] = avg_pixels8_mmx2;
2500 c->avg_pixels_tab[1][1] = avg_pixels8_x2_mmx2;
2501 c->avg_pixels_tab[1][2] = avg_pixels8_y2_mmx2;
2504 if(!(avctx->flags & CODEC_FLAG_BITEXACT)){
2505 if (!high_bit_depth) {
2506 c->put_no_rnd_pixels_tab[0][1] = put_no_rnd_pixels16_x2_mmx2;
2507 c->put_no_rnd_pixels_tab[0][2] = put_no_rnd_pixels16_y2_mmx2;
2508 c->put_no_rnd_pixels_tab[1][1] = put_no_rnd_pixels8_x2_mmx2;
2509 c->put_no_rnd_pixels_tab[1][2] = put_no_rnd_pixels8_y2_mmx2;
2510 c->avg_pixels_tab[0][3] = avg_pixels16_xy2_mmx2;
2511 c->avg_pixels_tab[1][3] = avg_pixels8_xy2_mmx2;
2514 if (CONFIG_VP3_DECODER && HAVE_YASM) {
2515 c->vp3_v_loop_filter= ff_vp3_v_loop_filter_mmx2;
2516 c->vp3_h_loop_filter= ff_vp3_h_loop_filter_mmx2;
2519 if (CONFIG_VP3_DECODER && HAVE_YASM) {
2520 c->vp3_idct_dc_add = ff_vp3_idct_dc_add_mmx2;
2523 if (CONFIG_VP3_DECODER
2524 && (avctx->codec_id == CODEC_ID_VP3 || avctx->codec_id == CODEC_ID_THEORA)) {
2525 c->put_no_rnd_pixels_tab[1][1] = put_no_rnd_pixels8_x2_exact_mmx2;
2526 c->put_no_rnd_pixels_tab[1][2] = put_no_rnd_pixels8_y2_exact_mmx2;
2529 #define SET_QPEL_FUNCS(PFX, IDX, SIZE, CPU, PREFIX) \
2530 c->PFX ## _pixels_tab[IDX][ 0] = PREFIX ## PFX ## SIZE ## _mc00_ ## CPU; \
2531 c->PFX ## _pixels_tab[IDX][ 1] = PREFIX ## PFX ## SIZE ## _mc10_ ## CPU; \
2532 c->PFX ## _pixels_tab[IDX][ 2] = PREFIX ## PFX ## SIZE ## _mc20_ ## CPU; \
2533 c->PFX ## _pixels_tab[IDX][ 3] = PREFIX ## PFX ## SIZE ## _mc30_ ## CPU; \
2534 c->PFX ## _pixels_tab[IDX][ 4] = PREFIX ## PFX ## SIZE ## _mc01_ ## CPU; \
2535 c->PFX ## _pixels_tab[IDX][ 5] = PREFIX ## PFX ## SIZE ## _mc11_ ## CPU; \
2536 c->PFX ## _pixels_tab[IDX][ 6] = PREFIX ## PFX ## SIZE ## _mc21_ ## CPU; \
2537 c->PFX ## _pixels_tab[IDX][ 7] = PREFIX ## PFX ## SIZE ## _mc31_ ## CPU; \
2538 c->PFX ## _pixels_tab[IDX][ 8] = PREFIX ## PFX ## SIZE ## _mc02_ ## CPU; \
2539 c->PFX ## _pixels_tab[IDX][ 9] = PREFIX ## PFX ## SIZE ## _mc12_ ## CPU; \
2540 c->PFX ## _pixels_tab[IDX][10] = PREFIX ## PFX ## SIZE ## _mc22_ ## CPU; \
2541 c->PFX ## _pixels_tab[IDX][11] = PREFIX ## PFX ## SIZE ## _mc32_ ## CPU; \
2542 c->PFX ## _pixels_tab[IDX][12] = PREFIX ## PFX ## SIZE ## _mc03_ ## CPU; \
2543 c->PFX ## _pixels_tab[IDX][13] = PREFIX ## PFX ## SIZE ## _mc13_ ## CPU; \
2544 c->PFX ## _pixels_tab[IDX][14] = PREFIX ## PFX ## SIZE ## _mc23_ ## CPU; \
2545 c->PFX ## _pixels_tab[IDX][15] = PREFIX ## PFX ## SIZE ## _mc33_ ## CPU
2547 SET_QPEL_FUNCS(put_qpel, 0, 16, mmx2, );
2548 SET_QPEL_FUNCS(put_qpel, 1, 8, mmx2, );
2549 SET_QPEL_FUNCS(put_no_rnd_qpel, 0, 16, mmx2, );
2550 SET_QPEL_FUNCS(put_no_rnd_qpel, 1, 8, mmx2, );
2551 SET_QPEL_FUNCS(avg_qpel, 0, 16, mmx2, );
2552 SET_QPEL_FUNCS(avg_qpel, 1, 8, mmx2, );
2554 if (!high_bit_depth) {
2555 SET_QPEL_FUNCS(put_h264_qpel, 0, 16, mmx2, );
2556 SET_QPEL_FUNCS(put_h264_qpel, 1, 8, mmx2, );
2557 SET_QPEL_FUNCS(put_h264_qpel, 2, 4, mmx2, );
2558 SET_QPEL_FUNCS(avg_h264_qpel, 0, 16, mmx2, );
2559 SET_QPEL_FUNCS(avg_h264_qpel, 1, 8, mmx2, );
2560 SET_QPEL_FUNCS(avg_h264_qpel, 2, 4, mmx2, );
2562 else if (bit_depth == 10) {
2565 SET_QPEL_FUNCS(avg_h264_qpel, 0, 16, 10_mmxext, ff_);
2566 SET_QPEL_FUNCS(put_h264_qpel, 0, 16, 10_mmxext, ff_);
2567 SET_QPEL_FUNCS(put_h264_qpel, 1, 8, 10_mmxext, ff_);
2568 SET_QPEL_FUNCS(avg_h264_qpel, 1, 8, 10_mmxext, ff_);
2570 SET_QPEL_FUNCS(put_h264_qpel, 2, 4, 10_mmxext, ff_);
2571 SET_QPEL_FUNCS(avg_h264_qpel, 2, 4, 10_mmxext, ff_);
2575 SET_QPEL_FUNCS(put_2tap_qpel, 0, 16, mmx2, );
2576 SET_QPEL_FUNCS(put_2tap_qpel, 1, 8, mmx2, );
2577 SET_QPEL_FUNCS(avg_2tap_qpel, 0, 16, mmx2, );
2578 SET_QPEL_FUNCS(avg_2tap_qpel, 1, 8, mmx2, );
2581 c->avg_rv40_chroma_pixels_tab[0]= ff_avg_rv40_chroma_mc8_mmx2;
2582 c->avg_rv40_chroma_pixels_tab[1]= ff_avg_rv40_chroma_mc4_mmx2;
2584 if (!high_bit_depth) {
2585 c->avg_h264_chroma_pixels_tab[0]= ff_avg_h264_chroma_mc8_mmx2_rnd;
2586 c->avg_h264_chroma_pixels_tab[1]= ff_avg_h264_chroma_mc4_mmx2;
2587 c->avg_h264_chroma_pixels_tab[2]= ff_avg_h264_chroma_mc2_mmx2;
2588 c->put_h264_chroma_pixels_tab[2]= ff_put_h264_chroma_mc2_mmx2;
2590 if (bit_depth == 10) {
2591 c->put_h264_chroma_pixels_tab[2]= ff_put_h264_chroma_mc2_10_mmxext;
2592 c->avg_h264_chroma_pixels_tab[2]= ff_avg_h264_chroma_mc2_10_mmxext;
2593 c->put_h264_chroma_pixels_tab[1]= ff_put_h264_chroma_mc4_10_mmxext;
2594 c->avg_h264_chroma_pixels_tab[1]= ff_avg_h264_chroma_mc4_10_mmxext;
2597 c->add_hfyu_median_prediction = ff_add_hfyu_median_prediction_mmx2;
2600 if( mm_flags&AV_CPU_FLAG_3DNOW )
2601 c->add_hfyu_median_prediction = add_hfyu_median_prediction_cmov;
2604 } else if (mm_flags & AV_CPU_FLAG_3DNOW) {
2605 c->prefetch = prefetch_3dnow;
2607 if (!high_bit_depth) {
2608 c->put_pixels_tab[0][1] = put_pixels16_x2_3dnow;
2609 c->put_pixels_tab[0][2] = put_pixels16_y2_3dnow;
2611 c->avg_pixels_tab[0][0] = avg_pixels16_3dnow;
2612 c->avg_pixels_tab[0][1] = avg_pixels16_x2_3dnow;
2613 c->avg_pixels_tab[0][2] = avg_pixels16_y2_3dnow;
2615 c->put_pixels_tab[1][1] = put_pixels8_x2_3dnow;
2616 c->put_pixels_tab[1][2] = put_pixels8_y2_3dnow;
2618 c->avg_pixels_tab[1][0] = avg_pixels8_3dnow;
2619 c->avg_pixels_tab[1][1] = avg_pixels8_x2_3dnow;
2620 c->avg_pixels_tab[1][2] = avg_pixels8_y2_3dnow;
2622 if(!(avctx->flags & CODEC_FLAG_BITEXACT)){
2623 c->put_no_rnd_pixels_tab[0][1] = put_no_rnd_pixels16_x2_3dnow;
2624 c->put_no_rnd_pixels_tab[0][2] = put_no_rnd_pixels16_y2_3dnow;
2625 c->put_no_rnd_pixels_tab[1][1] = put_no_rnd_pixels8_x2_3dnow;
2626 c->put_no_rnd_pixels_tab[1][2] = put_no_rnd_pixels8_y2_3dnow;
2627 c->avg_pixels_tab[0][3] = avg_pixels16_xy2_3dnow;
2628 c->avg_pixels_tab[1][3] = avg_pixels8_xy2_3dnow;
2632 if (CONFIG_VP3_DECODER
2633 && (avctx->codec_id == CODEC_ID_VP3 || avctx->codec_id == CODEC_ID_THEORA)) {
2634 c->put_no_rnd_pixels_tab[1][1] = put_no_rnd_pixels8_x2_exact_3dnow;
2635 c->put_no_rnd_pixels_tab[1][2] = put_no_rnd_pixels8_y2_exact_3dnow;
2638 SET_QPEL_FUNCS(put_qpel, 0, 16, 3dnow, );
2639 SET_QPEL_FUNCS(put_qpel, 1, 8, 3dnow, );
2640 SET_QPEL_FUNCS(put_no_rnd_qpel, 0, 16, 3dnow, );
2641 SET_QPEL_FUNCS(put_no_rnd_qpel, 1, 8, 3dnow, );
2642 SET_QPEL_FUNCS(avg_qpel, 0, 16, 3dnow, );
2643 SET_QPEL_FUNCS(avg_qpel, 1, 8, 3dnow, );
2645 if (!high_bit_depth) {
2646 SET_QPEL_FUNCS(put_h264_qpel, 0, 16, 3dnow, );
2647 SET_QPEL_FUNCS(put_h264_qpel, 1, 8, 3dnow, );
2648 SET_QPEL_FUNCS(put_h264_qpel, 2, 4, 3dnow, );
2649 SET_QPEL_FUNCS(avg_h264_qpel, 0, 16, 3dnow, );
2650 SET_QPEL_FUNCS(avg_h264_qpel, 1, 8, 3dnow, );
2651 SET_QPEL_FUNCS(avg_h264_qpel, 2, 4, 3dnow, );
2654 SET_QPEL_FUNCS(put_2tap_qpel, 0, 16, 3dnow, );
2655 SET_QPEL_FUNCS(put_2tap_qpel, 1, 8, 3dnow, );
2656 SET_QPEL_FUNCS(avg_2tap_qpel, 0, 16, 3dnow, );
2657 SET_QPEL_FUNCS(avg_2tap_qpel, 1, 8, 3dnow, );
2660 if (!high_bit_depth) {
2661 c->avg_h264_chroma_pixels_tab[0]= ff_avg_h264_chroma_mc8_3dnow_rnd;
2662 c->avg_h264_chroma_pixels_tab[1]= ff_avg_h264_chroma_mc4_3dnow;
2665 c->avg_rv40_chroma_pixels_tab[0]= ff_avg_rv40_chroma_mc8_3dnow;
2666 c->avg_rv40_chroma_pixels_tab[1]= ff_avg_rv40_chroma_mc4_3dnow;
2671 #define H264_QPEL_FUNCS(x, y, CPU)\
2672 c->put_h264_qpel_pixels_tab[0][x+y*4] = put_h264_qpel16_mc##x##y##_##CPU;\
2673 c->put_h264_qpel_pixels_tab[1][x+y*4] = put_h264_qpel8_mc##x##y##_##CPU;\
2674 c->avg_h264_qpel_pixels_tab[0][x+y*4] = avg_h264_qpel16_mc##x##y##_##CPU;\
2675 c->avg_h264_qpel_pixels_tab[1][x+y*4] = avg_h264_qpel8_mc##x##y##_##CPU;
2676 if((mm_flags & AV_CPU_FLAG_SSE2) && !(mm_flags & AV_CPU_FLAG_3DNOW)){
2677 // these functions are slower than mmx on AMD, but faster on Intel
2678 if (!high_bit_depth) {
2679 c->put_pixels_tab[0][0] = put_pixels16_sse2;
2680 c->put_no_rnd_pixels_tab[0][0] = put_pixels16_sse2;
2681 c->avg_pixels_tab[0][0] = avg_pixels16_sse2;
2682 H264_QPEL_FUNCS(0, 0, sse2);
2685 if(mm_flags & AV_CPU_FLAG_SSE2){
2686 if (!high_bit_depth) {
2687 H264_QPEL_FUNCS(0, 1, sse2);
2688 H264_QPEL_FUNCS(0, 2, sse2);
2689 H264_QPEL_FUNCS(0, 3, sse2);
2690 H264_QPEL_FUNCS(1, 1, sse2);
2691 H264_QPEL_FUNCS(1, 2, sse2);
2692 H264_QPEL_FUNCS(1, 3, sse2);
2693 H264_QPEL_FUNCS(2, 1, sse2);
2694 H264_QPEL_FUNCS(2, 2, sse2);
2695 H264_QPEL_FUNCS(2, 3, sse2);
2696 H264_QPEL_FUNCS(3, 1, sse2);
2697 H264_QPEL_FUNCS(3, 2, sse2);
2698 H264_QPEL_FUNCS(3, 3, sse2);
2701 #define H264_QPEL_FUNCS_10(x, y, CPU)\
2702 c->put_h264_qpel_pixels_tab[0][x+y*4] = ff_put_h264_qpel16_mc##x##y##_10_##CPU;\
2703 c->put_h264_qpel_pixels_tab[1][x+y*4] = ff_put_h264_qpel8_mc##x##y##_10_##CPU;\
2704 c->avg_h264_qpel_pixels_tab[0][x+y*4] = ff_avg_h264_qpel16_mc##x##y##_10_##CPU;\
2705 c->avg_h264_qpel_pixels_tab[1][x+y*4] = ff_avg_h264_qpel8_mc##x##y##_10_##CPU;
2706 if (bit_depth == 10) {
2707 SET_QPEL_FUNCS(put_h264_qpel, 0, 16, 10_sse2, ff_);
2708 SET_QPEL_FUNCS(put_h264_qpel, 1, 8, 10_sse2, ff_);
2709 SET_QPEL_FUNCS(avg_h264_qpel, 0, 16, 10_sse2, ff_);
2710 SET_QPEL_FUNCS(avg_h264_qpel, 1, 8, 10_sse2, ff_);
2711 H264_QPEL_FUNCS_10(1, 0, sse2_cache64)
2712 H264_QPEL_FUNCS_10(2, 0, sse2_cache64)
2713 H264_QPEL_FUNCS_10(3, 0, sse2_cache64)
2715 c->put_h264_chroma_pixels_tab[0]= ff_put_h264_chroma_mc8_10_sse2;
2716 c->avg_h264_chroma_pixels_tab[0]= ff_avg_h264_chroma_mc8_10_sse2;
2721 if(mm_flags & AV_CPU_FLAG_SSSE3){
2722 if (!high_bit_depth) {
2723 H264_QPEL_FUNCS(1, 0, ssse3);
2724 H264_QPEL_FUNCS(1, 1, ssse3);
2725 H264_QPEL_FUNCS(1, 2, ssse3);
2726 H264_QPEL_FUNCS(1, 3, ssse3);
2727 H264_QPEL_FUNCS(2, 0, ssse3);
2728 H264_QPEL_FUNCS(2, 1, ssse3);
2729 H264_QPEL_FUNCS(2, 2, ssse3);
2730 H264_QPEL_FUNCS(2, 3, ssse3);
2731 H264_QPEL_FUNCS(3, 0, ssse3);
2732 H264_QPEL_FUNCS(3, 1, ssse3);
2733 H264_QPEL_FUNCS(3, 2, ssse3);
2734 H264_QPEL_FUNCS(3, 3, ssse3);
2737 else if (bit_depth == 10) {
2738 H264_QPEL_FUNCS_10(1, 0, ssse3_cache64)
2739 H264_QPEL_FUNCS_10(2, 0, ssse3_cache64)
2740 H264_QPEL_FUNCS_10(3, 0, ssse3_cache64)
2742 if (!high_bit_depth) {
2743 c->put_h264_chroma_pixels_tab[0]= ff_put_h264_chroma_mc8_ssse3_rnd;
2744 c->avg_h264_chroma_pixels_tab[0]= ff_avg_h264_chroma_mc8_ssse3_rnd;
2745 c->put_h264_chroma_pixels_tab[1]= ff_put_h264_chroma_mc4_ssse3;
2746 c->avg_h264_chroma_pixels_tab[1]= ff_avg_h264_chroma_mc4_ssse3;
2748 c->add_hfyu_left_prediction = ff_add_hfyu_left_prediction_ssse3;
2749 if (mm_flags & AV_CPU_FLAG_SSE4) // not really sse4, just slow on Conroe
2750 c->add_hfyu_left_prediction = ff_add_hfyu_left_prediction_sse4;
2755 if(mm_flags & AV_CPU_FLAG_3DNOW){
2756 c->vorbis_inverse_coupling = vorbis_inverse_coupling_3dnow;
2757 c->vector_fmul = vector_fmul_3dnow;
2759 if(mm_flags & AV_CPU_FLAG_3DNOWEXT){
2760 c->vector_fmul_reverse = vector_fmul_reverse_3dnow2;
2762 c->vector_fmul_window = vector_fmul_window_3dnow2;
2765 if(mm_flags & AV_CPU_FLAG_MMX2){
2767 c->scalarproduct_int16 = ff_scalarproduct_int16_mmx2;
2768 c->scalarproduct_and_madd_int16 = ff_scalarproduct_and_madd_int16_mmx2;
2769 if (avctx->flags & CODEC_FLAG_BITEXACT) {
2770 c->apply_window_int16 = ff_apply_window_int16_mmxext_ba;
2772 c->apply_window_int16 = ff_apply_window_int16_mmxext;
2776 if(mm_flags & AV_CPU_FLAG_SSE){
2777 c->vorbis_inverse_coupling = vorbis_inverse_coupling_sse;
2778 c->ac3_downmix = ac3_downmix_sse;
2779 c->vector_fmul = vector_fmul_sse;
2780 c->vector_fmul_reverse = vector_fmul_reverse_sse;
2781 c->vector_fmul_add = vector_fmul_add_sse;
2783 c->vector_fmul_window = vector_fmul_window_sse;
2785 c->vector_clipf = vector_clipf_sse;
2787 c->scalarproduct_float = ff_scalarproduct_float_sse;
2790 if(mm_flags & AV_CPU_FLAG_3DNOW)
2791 c->vector_fmul_add = vector_fmul_add_3dnow; // faster than sse
2792 if(mm_flags & AV_CPU_FLAG_SSE2){
2794 c->scalarproduct_int16 = ff_scalarproduct_int16_sse2;
2795 c->scalarproduct_and_madd_int16 = ff_scalarproduct_and_madd_int16_sse2;
2796 if (mm_flags & AV_CPU_FLAG_ATOM) {
2797 c->vector_clip_int32 = ff_vector_clip_int32_sse2_int;
2799 c->vector_clip_int32 = ff_vector_clip_int32_sse2;
2801 if (avctx->flags & CODEC_FLAG_BITEXACT) {
2802 c->apply_window_int16 = ff_apply_window_int16_sse2_ba;
2804 if (!(mm_flags & AV_CPU_FLAG_SSE2SLOW)) {
2805 c->apply_window_int16 = ff_apply_window_int16_sse2;
2809 if (!high_bit_depth)
2810 c->emulated_edge_mc = emulated_edge_mc_sse;
2814 if (mm_flags & AV_CPU_FLAG_SSSE3) {
2816 if (mm_flags & AV_CPU_FLAG_ATOM) {
2817 c->apply_window_int16 = ff_apply_window_int16_ssse3_atom;
2819 c->apply_window_int16 = ff_apply_window_int16_ssse3;
2821 if (!(mm_flags & (AV_CPU_FLAG_SSE42|AV_CPU_FLAG_3DNOW))) { // cachesplit
2822 c->scalarproduct_and_madd_int16 = ff_scalarproduct_and_madd_int16_ssse3;
2827 if (mm_flags & AV_CPU_FLAG_SSE4 && HAVE_SSE) {
2829 c->vector_clip_int32 = ff_vector_clip_int32_sse41;
2833 #if HAVE_AVX && HAVE_YASM
2834 if (mm_flags & AV_CPU_FLAG_AVX) {
2835 if (bit_depth == 10) {
2836 //AVX implies !cache64.
2837 //TODO: Port cache(32|64) detection from x264.
2838 H264_QPEL_FUNCS_10(1, 0, sse2)
2839 H264_QPEL_FUNCS_10(2, 0, sse2)
2840 H264_QPEL_FUNCS_10(3, 0, sse2)
2842 c->put_h264_chroma_pixels_tab[0]= ff_put_h264_chroma_mc8_10_avx;
2843 c->avg_h264_chroma_pixels_tab[0]= ff_avg_h264_chroma_mc8_10_avx;
2849 if (CONFIG_ENCODERS)
2850 dsputilenc_init_mmx(c, avctx);