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 "dsputil_mmx.h"
32 #include "idct_xvid.h"
37 /* pixel operations */
38 DECLARE_ALIGNED(8, const uint64_t, ff_bone) = 0x0101010101010101ULL;
39 DECLARE_ALIGNED(8, const uint64_t, ff_wtwo) = 0x0002000200020002ULL;
41 DECLARE_ALIGNED(16, const uint64_t, ff_pdw_80000000)[2] =
42 {0x8000000080000000ULL, 0x8000000080000000ULL};
44 DECLARE_ALIGNED(16, const xmm_reg, ff_pw_1 ) = {0x0001000100010001ULL, 0x0001000100010001ULL};
45 DECLARE_ALIGNED(8, const uint64_t, ff_pw_3 ) = 0x0003000300030003ULL;
46 DECLARE_ALIGNED(16, const xmm_reg, ff_pw_4 ) = {0x0004000400040004ULL, 0x0004000400040004ULL};
47 DECLARE_ALIGNED(16, const xmm_reg, ff_pw_5 ) = {0x0005000500050005ULL, 0x0005000500050005ULL};
48 DECLARE_ALIGNED(16, const xmm_reg, ff_pw_8 ) = {0x0008000800080008ULL, 0x0008000800080008ULL};
49 DECLARE_ALIGNED(16, const xmm_reg, ff_pw_9 ) = {0x0009000900090009ULL, 0x0009000900090009ULL};
50 DECLARE_ALIGNED(8, const uint64_t, ff_pw_15 ) = 0x000F000F000F000FULL;
51 DECLARE_ALIGNED(16, const xmm_reg, ff_pw_16 ) = {0x0010001000100010ULL, 0x0010001000100010ULL};
52 DECLARE_ALIGNED(16, const xmm_reg, ff_pw_17 ) = {0x0011001100110011ULL, 0x0011001100110011ULL};
53 DECLARE_ALIGNED(16, const xmm_reg, ff_pw_18 ) = {0x0012001200120012ULL, 0x0012001200120012ULL};
54 DECLARE_ALIGNED(8, const uint64_t, ff_pw_20 ) = 0x0014001400140014ULL;
55 DECLARE_ALIGNED(16, const xmm_reg, ff_pw_27 ) = {0x001B001B001B001BULL, 0x001B001B001B001BULL};
56 DECLARE_ALIGNED(16, const xmm_reg, ff_pw_28 ) = {0x001C001C001C001CULL, 0x001C001C001C001CULL};
57 DECLARE_ALIGNED(16, const xmm_reg, ff_pw_32 ) = {0x0020002000200020ULL, 0x0020002000200020ULL};
58 DECLARE_ALIGNED(8, const uint64_t, ff_pw_42 ) = 0x002A002A002A002AULL;
59 DECLARE_ALIGNED(8, const uint64_t, ff_pw_53 ) = 0x0035003500350035ULL;
60 DECLARE_ALIGNED(16, const xmm_reg, ff_pw_63 ) = {0x003F003F003F003FULL, 0x003F003F003F003FULL};
61 DECLARE_ALIGNED(16, const xmm_reg, ff_pw_64 ) = {0x0040004000400040ULL, 0x0040004000400040ULL};
62 DECLARE_ALIGNED(8, const uint64_t, ff_pw_96 ) = 0x0060006000600060ULL;
63 DECLARE_ALIGNED(8, const uint64_t, ff_pw_128) = 0x0080008000800080ULL;
64 DECLARE_ALIGNED(8, const uint64_t, ff_pw_255) = 0x00ff00ff00ff00ffULL;
66 DECLARE_ALIGNED(16, const xmm_reg, ff_pb_0 ) = {0x0000000000000000ULL, 0x0000000000000000ULL};
67 DECLARE_ALIGNED(16, const xmm_reg, ff_pb_1 ) = {0x0101010101010101ULL, 0x0101010101010101ULL};
68 DECLARE_ALIGNED(16, const xmm_reg, ff_pb_3 ) = {0x0303030303030303ULL, 0x0303030303030303ULL};
69 DECLARE_ALIGNED(16, const xmm_reg, ff_pb_4 ) = {0x0404040404040404ULL, 0x0404040404040404ULL};
70 DECLARE_ALIGNED(8, const uint64_t, ff_pb_7 ) = 0x0707070707070707ULL;
71 DECLARE_ALIGNED(8, const uint64_t, ff_pb_1F ) = 0x1F1F1F1F1F1F1F1FULL;
72 DECLARE_ALIGNED(8, const uint64_t, ff_pb_3F ) = 0x3F3F3F3F3F3F3F3FULL;
73 DECLARE_ALIGNED(16, const xmm_reg, ff_pb_80 ) = {0x8080808080808080ULL, 0x8080808080808080ULL};
74 DECLARE_ALIGNED(8, const uint64_t, ff_pb_81 ) = 0x8181818181818181ULL;
75 DECLARE_ALIGNED(16, const xmm_reg, ff_pb_A1 ) = {0xA1A1A1A1A1A1A1A1ULL, 0xA1A1A1A1A1A1A1A1ULL};
76 DECLARE_ALIGNED(16, const xmm_reg, ff_pb_F8 ) = {0xF8F8F8F8F8F8F8F8ULL, 0xF8F8F8F8F8F8F8F8ULL};
77 DECLARE_ALIGNED(8, const uint64_t, ff_pb_FC ) = 0xFCFCFCFCFCFCFCFCULL;
78 DECLARE_ALIGNED(16, const xmm_reg, ff_pb_FE ) = {0xFEFEFEFEFEFEFEFEULL, 0xFEFEFEFEFEFEFEFEULL};
80 DECLARE_ALIGNED(16, const double, ff_pd_1)[2] = { 1.0, 1.0 };
81 DECLARE_ALIGNED(16, const double, ff_pd_2)[2] = { 2.0, 2.0 };
83 #define JUMPALIGN() __asm__ volatile (ASMALIGN(3)::)
84 #define MOVQ_ZERO(regd) __asm__ volatile ("pxor %%" #regd ", %%" #regd ::)
86 #define MOVQ_BFE(regd) \
88 "pcmpeqd %%" #regd ", %%" #regd " \n\t"\
89 "paddb %%" #regd ", %%" #regd " \n\t" ::)
92 #define MOVQ_BONE(regd) __asm__ volatile ("movq %0, %%" #regd " \n\t" ::"m"(ff_bone))
93 #define MOVQ_WTWO(regd) __asm__ volatile ("movq %0, %%" #regd " \n\t" ::"m"(ff_wtwo))
95 // for shared library it's better to use this way for accessing constants
97 #define MOVQ_BONE(regd) \
99 "pcmpeqd %%" #regd ", %%" #regd " \n\t" \
100 "psrlw $15, %%" #regd " \n\t" \
101 "packuswb %%" #regd ", %%" #regd " \n\t" ::)
103 #define MOVQ_WTWO(regd) \
105 "pcmpeqd %%" #regd ", %%" #regd " \n\t" \
106 "psrlw $15, %%" #regd " \n\t" \
107 "psllw $1, %%" #regd " \n\t"::)
111 // using regr as temporary and for the output result
112 // first argument is unmodifed and second is trashed
113 // regfe is supposed to contain 0xfefefefefefefefe
114 #define PAVGB_MMX_NO_RND(rega, regb, regr, regfe) \
115 "movq " #rega ", " #regr " \n\t"\
116 "pand " #regb ", " #regr " \n\t"\
117 "pxor " #rega ", " #regb " \n\t"\
118 "pand " #regfe "," #regb " \n\t"\
119 "psrlq $1, " #regb " \n\t"\
120 "paddb " #regb ", " #regr " \n\t"
122 #define PAVGB_MMX(rega, regb, regr, regfe) \
123 "movq " #rega ", " #regr " \n\t"\
124 "por " #regb ", " #regr " \n\t"\
125 "pxor " #rega ", " #regb " \n\t"\
126 "pand " #regfe "," #regb " \n\t"\
127 "psrlq $1, " #regb " \n\t"\
128 "psubb " #regb ", " #regr " \n\t"
130 // mm6 is supposed to contain 0xfefefefefefefefe
131 #define PAVGBP_MMX_NO_RND(rega, regb, regr, regc, regd, regp) \
132 "movq " #rega ", " #regr " \n\t"\
133 "movq " #regc ", " #regp " \n\t"\
134 "pand " #regb ", " #regr " \n\t"\
135 "pand " #regd ", " #regp " \n\t"\
136 "pxor " #rega ", " #regb " \n\t"\
137 "pxor " #regc ", " #regd " \n\t"\
138 "pand %%mm6, " #regb " \n\t"\
139 "pand %%mm6, " #regd " \n\t"\
140 "psrlq $1, " #regb " \n\t"\
141 "psrlq $1, " #regd " \n\t"\
142 "paddb " #regb ", " #regr " \n\t"\
143 "paddb " #regd ", " #regp " \n\t"
145 #define PAVGBP_MMX(rega, regb, regr, regc, regd, regp) \
146 "movq " #rega ", " #regr " \n\t"\
147 "movq " #regc ", " #regp " \n\t"\
148 "por " #regb ", " #regr " \n\t"\
149 "por " #regd ", " #regp " \n\t"\
150 "pxor " #rega ", " #regb " \n\t"\
151 "pxor " #regc ", " #regd " \n\t"\
152 "pand %%mm6, " #regb " \n\t"\
153 "pand %%mm6, " #regd " \n\t"\
154 "psrlq $1, " #regd " \n\t"\
155 "psrlq $1, " #regb " \n\t"\
156 "psubb " #regb ", " #regr " \n\t"\
157 "psubb " #regd ", " #regp " \n\t"
159 /***********************************/
160 /* MMX no rounding */
161 #define DEF(x, y) x ## _no_rnd_ ## y ##_mmx
162 #define SET_RND MOVQ_WONE
163 #define PAVGBP(a, b, c, d, e, f) PAVGBP_MMX_NO_RND(a, b, c, d, e, f)
164 #define PAVGB(a, b, c, e) PAVGB_MMX_NO_RND(a, b, c, e)
165 #define OP_AVG(a, b, c, e) PAVGB_MMX(a, b, c, e)
167 #include "dsputil_mmx_rnd_template.c"
173 /***********************************/
176 #define DEF(x, y) x ## _ ## y ##_mmx
177 #define SET_RND MOVQ_WTWO
178 #define PAVGBP(a, b, c, d, e, f) PAVGBP_MMX(a, b, c, d, e, f)
179 #define PAVGB(a, b, c, e) PAVGB_MMX(a, b, c, e)
181 #include "dsputil_mmx_rnd_template.c"
189 /***********************************/
192 #define DEF(x) x ## _3dnow
193 #define PAVGB "pavgusb"
196 #include "dsputil_mmx_avg_template.c"
202 /***********************************/
205 #define DEF(x) x ## _mmx2
207 /* Introduced only in MMX2 set */
208 #define PAVGB "pavgb"
211 #include "dsputil_mmx_avg_template.c"
217 #define put_no_rnd_pixels16_mmx put_pixels16_mmx
218 #define put_no_rnd_pixels8_mmx put_pixels8_mmx
219 #define put_pixels16_mmx2 put_pixels16_mmx
220 #define put_pixels8_mmx2 put_pixels8_mmx
221 #define put_pixels4_mmx2 put_pixels4_mmx
222 #define put_no_rnd_pixels16_mmx2 put_no_rnd_pixels16_mmx
223 #define put_no_rnd_pixels8_mmx2 put_no_rnd_pixels8_mmx
224 #define put_pixels16_3dnow put_pixels16_mmx
225 #define put_pixels8_3dnow put_pixels8_mmx
226 #define put_pixels4_3dnow put_pixels4_mmx
227 #define put_no_rnd_pixels16_3dnow put_no_rnd_pixels16_mmx
228 #define put_no_rnd_pixels8_3dnow put_no_rnd_pixels8_mmx
230 /***********************************/
233 void ff_put_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels, int line_size)
238 /* read the pixels */
243 "movq %3, %%mm0 \n\t"
244 "movq 8%3, %%mm1 \n\t"
245 "movq 16%3, %%mm2 \n\t"
246 "movq 24%3, %%mm3 \n\t"
247 "movq 32%3, %%mm4 \n\t"
248 "movq 40%3, %%mm5 \n\t"
249 "movq 48%3, %%mm6 \n\t"
250 "movq 56%3, %%mm7 \n\t"
251 "packuswb %%mm1, %%mm0 \n\t"
252 "packuswb %%mm3, %%mm2 \n\t"
253 "packuswb %%mm5, %%mm4 \n\t"
254 "packuswb %%mm7, %%mm6 \n\t"
255 "movq %%mm0, (%0) \n\t"
256 "movq %%mm2, (%0, %1) \n\t"
257 "movq %%mm4, (%0, %1, 2) \n\t"
258 "movq %%mm6, (%0, %2) \n\t"
259 ::"r" (pix), "r" ((x86_reg)line_size), "r" ((x86_reg)line_size*3), "m"(*p)
264 // if here would be an exact copy of the code above
265 // compiler would generate some very strange code
268 "movq (%3), %%mm0 \n\t"
269 "movq 8(%3), %%mm1 \n\t"
270 "movq 16(%3), %%mm2 \n\t"
271 "movq 24(%3), %%mm3 \n\t"
272 "movq 32(%3), %%mm4 \n\t"
273 "movq 40(%3), %%mm5 \n\t"
274 "movq 48(%3), %%mm6 \n\t"
275 "movq 56(%3), %%mm7 \n\t"
276 "packuswb %%mm1, %%mm0 \n\t"
277 "packuswb %%mm3, %%mm2 \n\t"
278 "packuswb %%mm5, %%mm4 \n\t"
279 "packuswb %%mm7, %%mm6 \n\t"
280 "movq %%mm0, (%0) \n\t"
281 "movq %%mm2, (%0, %1) \n\t"
282 "movq %%mm4, (%0, %1, 2) \n\t"
283 "movq %%mm6, (%0, %2) \n\t"
284 ::"r" (pix), "r" ((x86_reg)line_size), "r" ((x86_reg)line_size*3), "r"(p)
288 DECLARE_ASM_CONST(8, uint8_t, ff_vector128)[8] =
289 { 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80 };
291 #define put_signed_pixels_clamped_mmx_half(off) \
292 "movq "#off"(%2), %%mm1 \n\t"\
293 "movq 16+"#off"(%2), %%mm2 \n\t"\
294 "movq 32+"#off"(%2), %%mm3 \n\t"\
295 "movq 48+"#off"(%2), %%mm4 \n\t"\
296 "packsswb 8+"#off"(%2), %%mm1 \n\t"\
297 "packsswb 24+"#off"(%2), %%mm2 \n\t"\
298 "packsswb 40+"#off"(%2), %%mm3 \n\t"\
299 "packsswb 56+"#off"(%2), %%mm4 \n\t"\
300 "paddb %%mm0, %%mm1 \n\t"\
301 "paddb %%mm0, %%mm2 \n\t"\
302 "paddb %%mm0, %%mm3 \n\t"\
303 "paddb %%mm0, %%mm4 \n\t"\
304 "movq %%mm1, (%0) \n\t"\
305 "movq %%mm2, (%0, %3) \n\t"\
306 "movq %%mm3, (%0, %3, 2) \n\t"\
307 "movq %%mm4, (%0, %1) \n\t"
309 void ff_put_signed_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels, int line_size)
311 x86_reg line_skip = line_size;
315 "movq "MANGLE(ff_vector128)", %%mm0 \n\t"
316 "lea (%3, %3, 2), %1 \n\t"
317 put_signed_pixels_clamped_mmx_half(0)
318 "lea (%0, %3, 4), %0 \n\t"
319 put_signed_pixels_clamped_mmx_half(64)
320 :"+&r" (pixels), "=&r" (line_skip3)
321 :"r" (block), "r"(line_skip)
325 void ff_add_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels, int line_size)
331 /* read the pixels */
338 "movq (%2), %%mm0 \n\t"
339 "movq 8(%2), %%mm1 \n\t"
340 "movq 16(%2), %%mm2 \n\t"
341 "movq 24(%2), %%mm3 \n\t"
342 "movq %0, %%mm4 \n\t"
343 "movq %1, %%mm6 \n\t"
344 "movq %%mm4, %%mm5 \n\t"
345 "punpcklbw %%mm7, %%mm4 \n\t"
346 "punpckhbw %%mm7, %%mm5 \n\t"
347 "paddsw %%mm4, %%mm0 \n\t"
348 "paddsw %%mm5, %%mm1 \n\t"
349 "movq %%mm6, %%mm5 \n\t"
350 "punpcklbw %%mm7, %%mm6 \n\t"
351 "punpckhbw %%mm7, %%mm5 \n\t"
352 "paddsw %%mm6, %%mm2 \n\t"
353 "paddsw %%mm5, %%mm3 \n\t"
354 "packuswb %%mm1, %%mm0 \n\t"
355 "packuswb %%mm3, %%mm2 \n\t"
356 "movq %%mm0, %0 \n\t"
357 "movq %%mm2, %1 \n\t"
358 :"+m"(*pix), "+m"(*(pix+line_size))
366 static void put_pixels4_mmx(uint8_t *block, const uint8_t *pixels, int line_size, int h)
369 "lea (%3, %3), %%"REG_a" \n\t"
372 "movd (%1), %%mm0 \n\t"
373 "movd (%1, %3), %%mm1 \n\t"
374 "movd %%mm0, (%2) \n\t"
375 "movd %%mm1, (%2, %3) \n\t"
376 "add %%"REG_a", %1 \n\t"
377 "add %%"REG_a", %2 \n\t"
378 "movd (%1), %%mm0 \n\t"
379 "movd (%1, %3), %%mm1 \n\t"
380 "movd %%mm0, (%2) \n\t"
381 "movd %%mm1, (%2, %3) \n\t"
382 "add %%"REG_a", %1 \n\t"
383 "add %%"REG_a", %2 \n\t"
386 : "+g"(h), "+r" (pixels), "+r" (block)
387 : "r"((x86_reg)line_size)
392 static void put_pixels8_mmx(uint8_t *block, const uint8_t *pixels, int line_size, int h)
395 "lea (%3, %3), %%"REG_a" \n\t"
398 "movq (%1), %%mm0 \n\t"
399 "movq (%1, %3), %%mm1 \n\t"
400 "movq %%mm0, (%2) \n\t"
401 "movq %%mm1, (%2, %3) \n\t"
402 "add %%"REG_a", %1 \n\t"
403 "add %%"REG_a", %2 \n\t"
404 "movq (%1), %%mm0 \n\t"
405 "movq (%1, %3), %%mm1 \n\t"
406 "movq %%mm0, (%2) \n\t"
407 "movq %%mm1, (%2, %3) \n\t"
408 "add %%"REG_a", %1 \n\t"
409 "add %%"REG_a", %2 \n\t"
412 : "+g"(h), "+r" (pixels), "+r" (block)
413 : "r"((x86_reg)line_size)
418 static void put_pixels16_mmx(uint8_t *block, const uint8_t *pixels, int line_size, int h)
421 "lea (%3, %3), %%"REG_a" \n\t"
424 "movq (%1), %%mm0 \n\t"
425 "movq 8(%1), %%mm4 \n\t"
426 "movq (%1, %3), %%mm1 \n\t"
427 "movq 8(%1, %3), %%mm5 \n\t"
428 "movq %%mm0, (%2) \n\t"
429 "movq %%mm4, 8(%2) \n\t"
430 "movq %%mm1, (%2, %3) \n\t"
431 "movq %%mm5, 8(%2, %3) \n\t"
432 "add %%"REG_a", %1 \n\t"
433 "add %%"REG_a", %2 \n\t"
434 "movq (%1), %%mm0 \n\t"
435 "movq 8(%1), %%mm4 \n\t"
436 "movq (%1, %3), %%mm1 \n\t"
437 "movq 8(%1, %3), %%mm5 \n\t"
438 "movq %%mm0, (%2) \n\t"
439 "movq %%mm4, 8(%2) \n\t"
440 "movq %%mm1, (%2, %3) \n\t"
441 "movq %%mm5, 8(%2, %3) \n\t"
442 "add %%"REG_a", %1 \n\t"
443 "add %%"REG_a", %2 \n\t"
446 : "+g"(h), "+r" (pixels), "+r" (block)
447 : "r"((x86_reg)line_size)
452 static void put_pixels16_sse2(uint8_t *block, const uint8_t *pixels, int line_size, int h)
456 "movdqu (%1), %%xmm0 \n\t"
457 "movdqu (%1,%3), %%xmm1 \n\t"
458 "movdqu (%1,%3,2), %%xmm2 \n\t"
459 "movdqu (%1,%4), %%xmm3 \n\t"
460 "movdqa %%xmm0, (%2) \n\t"
461 "movdqa %%xmm1, (%2,%3) \n\t"
462 "movdqa %%xmm2, (%2,%3,2) \n\t"
463 "movdqa %%xmm3, (%2,%4) \n\t"
465 "lea (%1,%3,4), %1 \n\t"
466 "lea (%2,%3,4), %2 \n\t"
468 : "+g"(h), "+r" (pixels), "+r" (block)
469 : "r"((x86_reg)line_size), "r"((x86_reg)3L*line_size)
474 static void avg_pixels16_sse2(uint8_t *block, const uint8_t *pixels, int line_size, int h)
478 "movdqu (%1), %%xmm0 \n\t"
479 "movdqu (%1,%3), %%xmm1 \n\t"
480 "movdqu (%1,%3,2), %%xmm2 \n\t"
481 "movdqu (%1,%4), %%xmm3 \n\t"
482 "pavgb (%2), %%xmm0 \n\t"
483 "pavgb (%2,%3), %%xmm1 \n\t"
484 "pavgb (%2,%3,2), %%xmm2 \n\t"
485 "pavgb (%2,%4), %%xmm3 \n\t"
486 "movdqa %%xmm0, (%2) \n\t"
487 "movdqa %%xmm1, (%2,%3) \n\t"
488 "movdqa %%xmm2, (%2,%3,2) \n\t"
489 "movdqa %%xmm3, (%2,%4) \n\t"
491 "lea (%1,%3,4), %1 \n\t"
492 "lea (%2,%3,4), %2 \n\t"
494 : "+g"(h), "+r" (pixels), "+r" (block)
495 : "r"((x86_reg)line_size), "r"((x86_reg)3L*line_size)
500 #define CLEAR_BLOCKS(name,n) \
501 static void name(DCTELEM *blocks)\
504 "pxor %%mm7, %%mm7 \n\t"\
505 "mov %1, %%"REG_a" \n\t"\
507 "movq %%mm7, (%0, %%"REG_a") \n\t"\
508 "movq %%mm7, 8(%0, %%"REG_a") \n\t"\
509 "movq %%mm7, 16(%0, %%"REG_a") \n\t"\
510 "movq %%mm7, 24(%0, %%"REG_a") \n\t"\
511 "add $32, %%"REG_a" \n\t"\
513 : : "r" (((uint8_t *)blocks)+128*n),\
518 CLEAR_BLOCKS(clear_blocks_mmx, 6)
519 CLEAR_BLOCKS(clear_block_mmx, 1)
521 static void clear_block_sse(DCTELEM *block)
524 "xorps %%xmm0, %%xmm0 \n"
525 "movaps %%xmm0, (%0) \n"
526 "movaps %%xmm0, 16(%0) \n"
527 "movaps %%xmm0, 32(%0) \n"
528 "movaps %%xmm0, 48(%0) \n"
529 "movaps %%xmm0, 64(%0) \n"
530 "movaps %%xmm0, 80(%0) \n"
531 "movaps %%xmm0, 96(%0) \n"
532 "movaps %%xmm0, 112(%0) \n"
538 static void clear_blocks_sse(DCTELEM *blocks)
541 "xorps %%xmm0, %%xmm0 \n"
542 "mov %1, %%"REG_a" \n"
544 "movaps %%xmm0, (%0, %%"REG_a") \n"
545 "movaps %%xmm0, 16(%0, %%"REG_a") \n"
546 "movaps %%xmm0, 32(%0, %%"REG_a") \n"
547 "movaps %%xmm0, 48(%0, %%"REG_a") \n"
548 "movaps %%xmm0, 64(%0, %%"REG_a") \n"
549 "movaps %%xmm0, 80(%0, %%"REG_a") \n"
550 "movaps %%xmm0, 96(%0, %%"REG_a") \n"
551 "movaps %%xmm0, 112(%0, %%"REG_a") \n"
552 "add $128, %%"REG_a" \n"
554 : : "r" (((uint8_t *)blocks)+128*6),
560 static void add_bytes_mmx(uint8_t *dst, uint8_t *src, int w){
565 "movq (%1, %0), %%mm0 \n\t"
566 "movq (%2, %0), %%mm1 \n\t"
567 "paddb %%mm0, %%mm1 \n\t"
568 "movq %%mm1, (%2, %0) \n\t"
569 "movq 8(%1, %0), %%mm0 \n\t"
570 "movq 8(%2, %0), %%mm1 \n\t"
571 "paddb %%mm0, %%mm1 \n\t"
572 "movq %%mm1, 8(%2, %0) \n\t"
578 : "r"(src), "r"(dst), "r"((x86_reg)w-15)
581 dst[i+0] += src[i+0];
584 static void add_bytes_l2_mmx(uint8_t *dst, uint8_t *src1, uint8_t *src2, int w){
589 "movq (%2, %0), %%mm0 \n\t"
590 "movq 8(%2, %0), %%mm1 \n\t"
591 "paddb (%3, %0), %%mm0 \n\t"
592 "paddb 8(%3, %0), %%mm1 \n\t"
593 "movq %%mm0, (%1, %0) \n\t"
594 "movq %%mm1, 8(%1, %0) \n\t"
600 : "r"(dst), "r"(src1), "r"(src2), "r"((x86_reg)w-15)
603 dst[i] = src1[i] + src2[i];
606 #if HAVE_7REGS && HAVE_TEN_OPERANDS
607 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) {
610 int l = *left & 0xff;
611 int tl = *left_top & 0xff;
616 "movzbl (%3,%4), %2 \n"
629 "add (%6,%4), %b0 \n"
630 "mov %b0, (%5,%4) \n"
633 :"+&q"(l), "+&q"(tl), "=&r"(t), "=&q"(x), "+&r"(w2)
634 :"r"(dst+w), "r"(diff+w), "rm"(top+w)
641 #define H263_LOOP_FILTER \
642 "pxor %%mm7, %%mm7 \n\t"\
643 "movq %0, %%mm0 \n\t"\
644 "movq %0, %%mm1 \n\t"\
645 "movq %3, %%mm2 \n\t"\
646 "movq %3, %%mm3 \n\t"\
647 "punpcklbw %%mm7, %%mm0 \n\t"\
648 "punpckhbw %%mm7, %%mm1 \n\t"\
649 "punpcklbw %%mm7, %%mm2 \n\t"\
650 "punpckhbw %%mm7, %%mm3 \n\t"\
651 "psubw %%mm2, %%mm0 \n\t"\
652 "psubw %%mm3, %%mm1 \n\t"\
653 "movq %1, %%mm2 \n\t"\
654 "movq %1, %%mm3 \n\t"\
655 "movq %2, %%mm4 \n\t"\
656 "movq %2, %%mm5 \n\t"\
657 "punpcklbw %%mm7, %%mm2 \n\t"\
658 "punpckhbw %%mm7, %%mm3 \n\t"\
659 "punpcklbw %%mm7, %%mm4 \n\t"\
660 "punpckhbw %%mm7, %%mm5 \n\t"\
661 "psubw %%mm2, %%mm4 \n\t"\
662 "psubw %%mm3, %%mm5 \n\t"\
663 "psllw $2, %%mm4 \n\t"\
664 "psllw $2, %%mm5 \n\t"\
665 "paddw %%mm0, %%mm4 \n\t"\
666 "paddw %%mm1, %%mm5 \n\t"\
667 "pxor %%mm6, %%mm6 \n\t"\
668 "pcmpgtw %%mm4, %%mm6 \n\t"\
669 "pcmpgtw %%mm5, %%mm7 \n\t"\
670 "pxor %%mm6, %%mm4 \n\t"\
671 "pxor %%mm7, %%mm5 \n\t"\
672 "psubw %%mm6, %%mm4 \n\t"\
673 "psubw %%mm7, %%mm5 \n\t"\
674 "psrlw $3, %%mm4 \n\t"\
675 "psrlw $3, %%mm5 \n\t"\
676 "packuswb %%mm5, %%mm4 \n\t"\
677 "packsswb %%mm7, %%mm6 \n\t"\
678 "pxor %%mm7, %%mm7 \n\t"\
679 "movd %4, %%mm2 \n\t"\
680 "punpcklbw %%mm2, %%mm2 \n\t"\
681 "punpcklbw %%mm2, %%mm2 \n\t"\
682 "punpcklbw %%mm2, %%mm2 \n\t"\
683 "psubusb %%mm4, %%mm2 \n\t"\
684 "movq %%mm2, %%mm3 \n\t"\
685 "psubusb %%mm4, %%mm3 \n\t"\
686 "psubb %%mm3, %%mm2 \n\t"\
687 "movq %1, %%mm3 \n\t"\
688 "movq %2, %%mm4 \n\t"\
689 "pxor %%mm6, %%mm3 \n\t"\
690 "pxor %%mm6, %%mm4 \n\t"\
691 "paddusb %%mm2, %%mm3 \n\t"\
692 "psubusb %%mm2, %%mm4 \n\t"\
693 "pxor %%mm6, %%mm3 \n\t"\
694 "pxor %%mm6, %%mm4 \n\t"\
695 "paddusb %%mm2, %%mm2 \n\t"\
696 "packsswb %%mm1, %%mm0 \n\t"\
697 "pcmpgtb %%mm0, %%mm7 \n\t"\
698 "pxor %%mm7, %%mm0 \n\t"\
699 "psubb %%mm7, %%mm0 \n\t"\
700 "movq %%mm0, %%mm1 \n\t"\
701 "psubusb %%mm2, %%mm0 \n\t"\
702 "psubb %%mm0, %%mm1 \n\t"\
703 "pand %5, %%mm1 \n\t"\
704 "psrlw $2, %%mm1 \n\t"\
705 "pxor %%mm7, %%mm1 \n\t"\
706 "psubb %%mm7, %%mm1 \n\t"\
707 "movq %0, %%mm5 \n\t"\
708 "movq %3, %%mm6 \n\t"\
709 "psubb %%mm1, %%mm5 \n\t"\
710 "paddb %%mm1, %%mm6 \n\t"
712 static void h263_v_loop_filter_mmx(uint8_t *src, int stride, int qscale){
713 if(CONFIG_H263_DECODER || CONFIG_H263_ENCODER) {
714 const int strength= ff_h263_loop_filter_strength[qscale];
720 "movq %%mm3, %1 \n\t"
721 "movq %%mm4, %2 \n\t"
722 "movq %%mm5, %0 \n\t"
723 "movq %%mm6, %3 \n\t"
724 : "+m" (*(uint64_t*)(src - 2*stride)),
725 "+m" (*(uint64_t*)(src - 1*stride)),
726 "+m" (*(uint64_t*)(src + 0*stride)),
727 "+m" (*(uint64_t*)(src + 1*stride))
728 : "g" (2*strength), "m"(ff_pb_FC)
733 static void h263_h_loop_filter_mmx(uint8_t *src, int stride, int qscale){
734 if(CONFIG_H263_DECODER || CONFIG_H263_ENCODER) {
735 const int strength= ff_h263_loop_filter_strength[qscale];
736 DECLARE_ALIGNED(8, uint64_t, temp)[4];
737 uint8_t *btemp= (uint8_t*)temp;
741 transpose4x4(btemp , src , 8, stride);
742 transpose4x4(btemp+4, src + 4*stride, 8, stride);
744 H263_LOOP_FILTER // 5 3 4 6
750 : "g" (2*strength), "m"(ff_pb_FC)
754 "movq %%mm5, %%mm1 \n\t"
755 "movq %%mm4, %%mm0 \n\t"
756 "punpcklbw %%mm3, %%mm5 \n\t"
757 "punpcklbw %%mm6, %%mm4 \n\t"
758 "punpckhbw %%mm3, %%mm1 \n\t"
759 "punpckhbw %%mm6, %%mm0 \n\t"
760 "movq %%mm5, %%mm3 \n\t"
761 "movq %%mm1, %%mm6 \n\t"
762 "punpcklwd %%mm4, %%mm5 \n\t"
763 "punpcklwd %%mm0, %%mm1 \n\t"
764 "punpckhwd %%mm4, %%mm3 \n\t"
765 "punpckhwd %%mm0, %%mm6 \n\t"
766 "movd %%mm5, (%0) \n\t"
767 "punpckhdq %%mm5, %%mm5 \n\t"
768 "movd %%mm5, (%0,%2) \n\t"
769 "movd %%mm3, (%0,%2,2) \n\t"
770 "punpckhdq %%mm3, %%mm3 \n\t"
771 "movd %%mm3, (%0,%3) \n\t"
772 "movd %%mm1, (%1) \n\t"
773 "punpckhdq %%mm1, %%mm1 \n\t"
774 "movd %%mm1, (%1,%2) \n\t"
775 "movd %%mm6, (%1,%2,2) \n\t"
776 "punpckhdq %%mm6, %%mm6 \n\t"
777 "movd %%mm6, (%1,%3) \n\t"
779 "r" (src + 4*stride),
780 "r" ((x86_reg) stride ),
781 "r" ((x86_reg)(3*stride))
786 /* draw the edges of width 'w' of an image of size width, height
787 this mmx version can only handle w==8 || w==16 */
788 static void draw_edges_mmx(uint8_t *buf, int wrap, int width, int height, int w)
790 uint8_t *ptr, *last_line;
793 last_line = buf + (height - 1) * wrap;
800 "movd (%0), %%mm0 \n\t"
801 "punpcklbw %%mm0, %%mm0 \n\t"
802 "punpcklwd %%mm0, %%mm0 \n\t"
803 "punpckldq %%mm0, %%mm0 \n\t"
804 "movq %%mm0, -8(%0) \n\t"
805 "movq -8(%0, %2), %%mm1 \n\t"
806 "punpckhbw %%mm1, %%mm1 \n\t"
807 "punpckhwd %%mm1, %%mm1 \n\t"
808 "punpckhdq %%mm1, %%mm1 \n\t"
809 "movq %%mm1, (%0, %2) \n\t"
814 : "r" ((x86_reg)wrap), "r" ((x86_reg)width), "r" (ptr + wrap*height)
821 "movd (%0), %%mm0 \n\t"
822 "punpcklbw %%mm0, %%mm0 \n\t"
823 "punpcklwd %%mm0, %%mm0 \n\t"
824 "punpckldq %%mm0, %%mm0 \n\t"
825 "movq %%mm0, -8(%0) \n\t"
826 "movq %%mm0, -16(%0) \n\t"
827 "movq -8(%0, %2), %%mm1 \n\t"
828 "punpckhbw %%mm1, %%mm1 \n\t"
829 "punpckhwd %%mm1, %%mm1 \n\t"
830 "punpckhdq %%mm1, %%mm1 \n\t"
831 "movq %%mm1, (%0, %2) \n\t"
832 "movq %%mm1, 8(%0, %2) \n\t"
837 : "r" ((x86_reg)wrap), "r" ((x86_reg)width), "r" (ptr + wrap*height)
842 /* top and bottom (and hopefully also the corners) */
843 ptr= buf - (i + 1) * wrap - w;
846 "movq (%1, %0), %%mm0 \n\t"
847 "movq %%mm0, (%0) \n\t"
848 "movq %%mm0, (%0, %2) \n\t"
849 "movq %%mm0, (%0, %2, 2) \n\t"
850 "movq %%mm0, (%0, %3) \n\t"
855 : "r" ((x86_reg)buf - (x86_reg)ptr - w), "r" ((x86_reg)-wrap), "r" ((x86_reg)-wrap*3), "r" (ptr+width+2*w)
857 ptr= last_line + (i + 1) * wrap - w;
860 "movq (%1, %0), %%mm0 \n\t"
861 "movq %%mm0, (%0) \n\t"
862 "movq %%mm0, (%0, %2) \n\t"
863 "movq %%mm0, (%0, %2, 2) \n\t"
864 "movq %%mm0, (%0, %3) \n\t"
869 : "r" ((x86_reg)last_line - (x86_reg)ptr - w), "r" ((x86_reg)wrap), "r" ((x86_reg)wrap*3), "r" (ptr+width+2*w)
874 #define PAETH(cpu, abs3)\
875 static void add_png_paeth_prediction_##cpu(uint8_t *dst, uint8_t *src, uint8_t *top, int w, int bpp)\
880 "pxor %%mm7, %%mm7 \n"\
881 "movd (%1,%0), %%mm0 \n"\
882 "movd (%2,%0), %%mm1 \n"\
883 "punpcklbw %%mm7, %%mm0 \n"\
884 "punpcklbw %%mm7, %%mm1 \n"\
887 "movq %%mm1, %%mm2 \n"\
888 "movd (%2,%0), %%mm1 \n"\
889 "movq %%mm2, %%mm3 \n"\
890 "punpcklbw %%mm7, %%mm1 \n"\
891 "movq %%mm2, %%mm4 \n"\
892 "psubw %%mm1, %%mm3 \n"\
893 "psubw %%mm0, %%mm4 \n"\
894 "movq %%mm3, %%mm5 \n"\
895 "paddw %%mm4, %%mm5 \n"\
897 "movq %%mm4, %%mm6 \n"\
898 "pminsw %%mm5, %%mm6 \n"\
899 "pcmpgtw %%mm6, %%mm3 \n"\
900 "pcmpgtw %%mm5, %%mm4 \n"\
901 "movq %%mm4, %%mm6 \n"\
902 "pand %%mm3, %%mm4 \n"\
903 "pandn %%mm3, %%mm6 \n"\
904 "pandn %%mm0, %%mm3 \n"\
905 "movd (%3,%0), %%mm0 \n"\
906 "pand %%mm1, %%mm6 \n"\
907 "pand %%mm4, %%mm2 \n"\
908 "punpcklbw %%mm7, %%mm0 \n"\
910 "paddw %%mm6, %%mm0 \n"\
911 "paddw %%mm2, %%mm3 \n"\
912 "paddw %%mm3, %%mm0 \n"\
913 "pand %%mm5, %%mm0 \n"\
914 "movq %%mm0, %%mm3 \n"\
915 "packuswb %%mm3, %%mm3 \n"\
916 "movd %%mm3, (%1,%0) \n"\
921 :"r"(dst), "r"(top), "r"(src), "r"((x86_reg)bpp), "g"(end),\
928 "psubw %%mm5, %%mm7 \n"\
929 "pmaxsw %%mm7, %%mm5 \n"\
930 "pxor %%mm6, %%mm6 \n"\
931 "pxor %%mm7, %%mm7 \n"\
932 "psubw %%mm3, %%mm6 \n"\
933 "psubw %%mm4, %%mm7 \n"\
934 "pmaxsw %%mm6, %%mm3 \n"\
935 "pmaxsw %%mm7, %%mm4 \n"\
936 "pxor %%mm7, %%mm7 \n"
939 "pabsw %%mm3, %%mm3 \n"\
940 "pabsw %%mm4, %%mm4 \n"\
941 "pabsw %%mm5, %%mm5 \n"
943 PAETH(mmx2, ABS3_MMX2)
945 PAETH(ssse3, ABS3_SSSE3)
948 #define QPEL_V_LOW(m3,m4,m5,m6, pw_20, pw_3, rnd, in0, in1, in2, in7, out, OP)\
949 "paddw " #m4 ", " #m3 " \n\t" /* x1 */\
950 "movq "MANGLE(ff_pw_20)", %%mm4 \n\t" /* 20 */\
951 "pmullw " #m3 ", %%mm4 \n\t" /* 20x1 */\
952 "movq "#in7", " #m3 " \n\t" /* d */\
953 "movq "#in0", %%mm5 \n\t" /* D */\
954 "paddw " #m3 ", %%mm5 \n\t" /* x4 */\
955 "psubw %%mm5, %%mm4 \n\t" /* 20x1 - x4 */\
956 "movq "#in1", %%mm5 \n\t" /* C */\
957 "movq "#in2", %%mm6 \n\t" /* B */\
958 "paddw " #m6 ", %%mm5 \n\t" /* x3 */\
959 "paddw " #m5 ", %%mm6 \n\t" /* x2 */\
960 "paddw %%mm6, %%mm6 \n\t" /* 2x2 */\
961 "psubw %%mm6, %%mm5 \n\t" /* -2x2 + x3 */\
962 "pmullw "MANGLE(ff_pw_3)", %%mm5 \n\t" /* -6x2 + 3x3 */\
963 "paddw " #rnd ", %%mm4 \n\t" /* x2 */\
964 "paddw %%mm4, %%mm5 \n\t" /* 20x1 - 6x2 + 3x3 - x4 */\
965 "psraw $5, %%mm5 \n\t"\
966 "packuswb %%mm5, %%mm5 \n\t"\
967 OP(%%mm5, out, %%mm7, d)
969 #define QPEL_BASE(OPNAME, ROUNDER, RND, OP_MMX2, OP_3DNOW)\
970 static void OPNAME ## mpeg4_qpel16_h_lowpass_mmx2(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h){\
974 "pxor %%mm7, %%mm7 \n\t"\
976 "movq (%0), %%mm0 \n\t" /* ABCDEFGH */\
977 "movq %%mm0, %%mm1 \n\t" /* ABCDEFGH */\
978 "movq %%mm0, %%mm2 \n\t" /* ABCDEFGH */\
979 "punpcklbw %%mm7, %%mm0 \n\t" /* 0A0B0C0D */\
980 "punpckhbw %%mm7, %%mm1 \n\t" /* 0E0F0G0H */\
981 "pshufw $0x90, %%mm0, %%mm5 \n\t" /* 0A0A0B0C */\
982 "pshufw $0x41, %%mm0, %%mm6 \n\t" /* 0B0A0A0B */\
983 "movq %%mm2, %%mm3 \n\t" /* ABCDEFGH */\
984 "movq %%mm2, %%mm4 \n\t" /* ABCDEFGH */\
985 "psllq $8, %%mm2 \n\t" /* 0ABCDEFG */\
986 "psllq $16, %%mm3 \n\t" /* 00ABCDEF */\
987 "psllq $24, %%mm4 \n\t" /* 000ABCDE */\
988 "punpckhbw %%mm7, %%mm2 \n\t" /* 0D0E0F0G */\
989 "punpckhbw %%mm7, %%mm3 \n\t" /* 0C0D0E0F */\
990 "punpckhbw %%mm7, %%mm4 \n\t" /* 0B0C0D0E */\
991 "paddw %%mm3, %%mm5 \n\t" /* b */\
992 "paddw %%mm2, %%mm6 \n\t" /* c */\
993 "paddw %%mm5, %%mm5 \n\t" /* 2b */\
994 "psubw %%mm5, %%mm6 \n\t" /* c - 2b */\
995 "pshufw $0x06, %%mm0, %%mm5 \n\t" /* 0C0B0A0A */\
996 "pmullw "MANGLE(ff_pw_3)", %%mm6 \n\t" /* 3c - 6b */\
997 "paddw %%mm4, %%mm0 \n\t" /* a */\
998 "paddw %%mm1, %%mm5 \n\t" /* d */\
999 "pmullw "MANGLE(ff_pw_20)", %%mm0 \n\t" /* 20a */\
1000 "psubw %%mm5, %%mm0 \n\t" /* 20a - d */\
1001 "paddw %6, %%mm6 \n\t"\
1002 "paddw %%mm6, %%mm0 \n\t" /* 20a - 6b + 3c - d */\
1003 "psraw $5, %%mm0 \n\t"\
1004 "movq %%mm0, %5 \n\t"\
1005 /* mm1=EFGH, mm2=DEFG, mm3=CDEF, mm4=BCDE, mm7=0 */\
1007 "movq 5(%0), %%mm0 \n\t" /* FGHIJKLM */\
1008 "movq %%mm0, %%mm5 \n\t" /* FGHIJKLM */\
1009 "movq %%mm0, %%mm6 \n\t" /* FGHIJKLM */\
1010 "psrlq $8, %%mm0 \n\t" /* GHIJKLM0 */\
1011 "psrlq $16, %%mm5 \n\t" /* HIJKLM00 */\
1012 "punpcklbw %%mm7, %%mm0 \n\t" /* 0G0H0I0J */\
1013 "punpcklbw %%mm7, %%mm5 \n\t" /* 0H0I0J0K */\
1014 "paddw %%mm0, %%mm2 \n\t" /* b */\
1015 "paddw %%mm5, %%mm3 \n\t" /* c */\
1016 "paddw %%mm2, %%mm2 \n\t" /* 2b */\
1017 "psubw %%mm2, %%mm3 \n\t" /* c - 2b */\
1018 "movq %%mm6, %%mm2 \n\t" /* FGHIJKLM */\
1019 "psrlq $24, %%mm6 \n\t" /* IJKLM000 */\
1020 "punpcklbw %%mm7, %%mm2 \n\t" /* 0F0G0H0I */\
1021 "punpcklbw %%mm7, %%mm6 \n\t" /* 0I0J0K0L */\
1022 "pmullw "MANGLE(ff_pw_3)", %%mm3 \n\t" /* 3c - 6b */\
1023 "paddw %%mm2, %%mm1 \n\t" /* a */\
1024 "paddw %%mm6, %%mm4 \n\t" /* d */\
1025 "pmullw "MANGLE(ff_pw_20)", %%mm1 \n\t" /* 20a */\
1026 "psubw %%mm4, %%mm3 \n\t" /* - 6b +3c - d */\
1027 "paddw %6, %%mm1 \n\t"\
1028 "paddw %%mm1, %%mm3 \n\t" /* 20a - 6b +3c - d */\
1029 "psraw $5, %%mm3 \n\t"\
1030 "movq %5, %%mm1 \n\t"\
1031 "packuswb %%mm3, %%mm1 \n\t"\
1032 OP_MMX2(%%mm1, (%1),%%mm4, q)\
1033 /* mm0= GHIJ, mm2=FGHI, mm5=HIJK, mm6=IJKL, mm7=0 */\
1035 "movq 9(%0), %%mm1 \n\t" /* JKLMNOPQ */\
1036 "movq %%mm1, %%mm4 \n\t" /* JKLMNOPQ */\
1037 "movq %%mm1, %%mm3 \n\t" /* JKLMNOPQ */\
1038 "psrlq $8, %%mm1 \n\t" /* KLMNOPQ0 */\
1039 "psrlq $16, %%mm4 \n\t" /* LMNOPQ00 */\
1040 "punpcklbw %%mm7, %%mm1 \n\t" /* 0K0L0M0N */\
1041 "punpcklbw %%mm7, %%mm4 \n\t" /* 0L0M0N0O */\
1042 "paddw %%mm1, %%mm5 \n\t" /* b */\
1043 "paddw %%mm4, %%mm0 \n\t" /* c */\
1044 "paddw %%mm5, %%mm5 \n\t" /* 2b */\
1045 "psubw %%mm5, %%mm0 \n\t" /* c - 2b */\
1046 "movq %%mm3, %%mm5 \n\t" /* JKLMNOPQ */\
1047 "psrlq $24, %%mm3 \n\t" /* MNOPQ000 */\
1048 "pmullw "MANGLE(ff_pw_3)", %%mm0 \n\t" /* 3c - 6b */\
1049 "punpcklbw %%mm7, %%mm3 \n\t" /* 0M0N0O0P */\
1050 "paddw %%mm3, %%mm2 \n\t" /* d */\
1051 "psubw %%mm2, %%mm0 \n\t" /* -6b + 3c - d */\
1052 "movq %%mm5, %%mm2 \n\t" /* JKLMNOPQ */\
1053 "punpcklbw %%mm7, %%mm2 \n\t" /* 0J0K0L0M */\
1054 "punpckhbw %%mm7, %%mm5 \n\t" /* 0N0O0P0Q */\
1055 "paddw %%mm2, %%mm6 \n\t" /* a */\
1056 "pmullw "MANGLE(ff_pw_20)", %%mm6 \n\t" /* 20a */\
1057 "paddw %6, %%mm0 \n\t"\
1058 "paddw %%mm6, %%mm0 \n\t" /* 20a - 6b + 3c - d */\
1059 "psraw $5, %%mm0 \n\t"\
1060 /* mm1=KLMN, mm2=JKLM, mm3=MNOP, mm4=LMNO, mm5=NOPQ mm7=0 */\
1062 "paddw %%mm5, %%mm3 \n\t" /* a */\
1063 "pshufw $0xF9, %%mm5, %%mm6 \n\t" /* 0O0P0Q0Q */\
1064 "paddw %%mm4, %%mm6 \n\t" /* b */\
1065 "pshufw $0xBE, %%mm5, %%mm4 \n\t" /* 0P0Q0Q0P */\
1066 "pshufw $0x6F, %%mm5, %%mm5 \n\t" /* 0Q0Q0P0O */\
1067 "paddw %%mm1, %%mm4 \n\t" /* c */\
1068 "paddw %%mm2, %%mm5 \n\t" /* d */\
1069 "paddw %%mm6, %%mm6 \n\t" /* 2b */\
1070 "psubw %%mm6, %%mm4 \n\t" /* c - 2b */\
1071 "pmullw "MANGLE(ff_pw_20)", %%mm3 \n\t" /* 20a */\
1072 "pmullw "MANGLE(ff_pw_3)", %%mm4 \n\t" /* 3c - 6b */\
1073 "psubw %%mm5, %%mm3 \n\t" /* -6b + 3c - d */\
1074 "paddw %6, %%mm4 \n\t"\
1075 "paddw %%mm3, %%mm4 \n\t" /* 20a - 6b + 3c - d */\
1076 "psraw $5, %%mm4 \n\t"\
1077 "packuswb %%mm4, %%mm0 \n\t"\
1078 OP_MMX2(%%mm0, 8(%1), %%mm4, q)\
1084 : "+a"(src), "+c"(dst), "+D"(h)\
1085 : "d"((x86_reg)srcStride), "S"((x86_reg)dstStride), /*"m"(ff_pw_20), "m"(ff_pw_3),*/ "m"(temp), "m"(ROUNDER)\
1090 static void OPNAME ## mpeg4_qpel16_h_lowpass_3dnow(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h){\
1093 /* quick HACK, XXX FIXME MUST be optimized */\
1096 temp[ 0]= (src[ 0]+src[ 1])*20 - (src[ 0]+src[ 2])*6 + (src[ 1]+src[ 3])*3 - (src[ 2]+src[ 4]);\
1097 temp[ 1]= (src[ 1]+src[ 2])*20 - (src[ 0]+src[ 3])*6 + (src[ 0]+src[ 4])*3 - (src[ 1]+src[ 5]);\
1098 temp[ 2]= (src[ 2]+src[ 3])*20 - (src[ 1]+src[ 4])*6 + (src[ 0]+src[ 5])*3 - (src[ 0]+src[ 6]);\
1099 temp[ 3]= (src[ 3]+src[ 4])*20 - (src[ 2]+src[ 5])*6 + (src[ 1]+src[ 6])*3 - (src[ 0]+src[ 7]);\
1100 temp[ 4]= (src[ 4]+src[ 5])*20 - (src[ 3]+src[ 6])*6 + (src[ 2]+src[ 7])*3 - (src[ 1]+src[ 8]);\
1101 temp[ 5]= (src[ 5]+src[ 6])*20 - (src[ 4]+src[ 7])*6 + (src[ 3]+src[ 8])*3 - (src[ 2]+src[ 9]);\
1102 temp[ 6]= (src[ 6]+src[ 7])*20 - (src[ 5]+src[ 8])*6 + (src[ 4]+src[ 9])*3 - (src[ 3]+src[10]);\
1103 temp[ 7]= (src[ 7]+src[ 8])*20 - (src[ 6]+src[ 9])*6 + (src[ 5]+src[10])*3 - (src[ 4]+src[11]);\
1104 temp[ 8]= (src[ 8]+src[ 9])*20 - (src[ 7]+src[10])*6 + (src[ 6]+src[11])*3 - (src[ 5]+src[12]);\
1105 temp[ 9]= (src[ 9]+src[10])*20 - (src[ 8]+src[11])*6 + (src[ 7]+src[12])*3 - (src[ 6]+src[13]);\
1106 temp[10]= (src[10]+src[11])*20 - (src[ 9]+src[12])*6 + (src[ 8]+src[13])*3 - (src[ 7]+src[14]);\
1107 temp[11]= (src[11]+src[12])*20 - (src[10]+src[13])*6 + (src[ 9]+src[14])*3 - (src[ 8]+src[15]);\
1108 temp[12]= (src[12]+src[13])*20 - (src[11]+src[14])*6 + (src[10]+src[15])*3 - (src[ 9]+src[16]);\
1109 temp[13]= (src[13]+src[14])*20 - (src[12]+src[15])*6 + (src[11]+src[16])*3 - (src[10]+src[16]);\
1110 temp[14]= (src[14]+src[15])*20 - (src[13]+src[16])*6 + (src[12]+src[16])*3 - (src[11]+src[15]);\
1111 temp[15]= (src[15]+src[16])*20 - (src[14]+src[16])*6 + (src[13]+src[15])*3 - (src[12]+src[14]);\
1113 "movq (%0), %%mm0 \n\t"\
1114 "movq 8(%0), %%mm1 \n\t"\
1115 "paddw %2, %%mm0 \n\t"\
1116 "paddw %2, %%mm1 \n\t"\
1117 "psraw $5, %%mm0 \n\t"\
1118 "psraw $5, %%mm1 \n\t"\
1119 "packuswb %%mm1, %%mm0 \n\t"\
1120 OP_3DNOW(%%mm0, (%1), %%mm1, q)\
1121 "movq 16(%0), %%mm0 \n\t"\
1122 "movq 24(%0), %%mm1 \n\t"\
1123 "paddw %2, %%mm0 \n\t"\
1124 "paddw %2, %%mm1 \n\t"\
1125 "psraw $5, %%mm0 \n\t"\
1126 "psraw $5, %%mm1 \n\t"\
1127 "packuswb %%mm1, %%mm0 \n\t"\
1128 OP_3DNOW(%%mm0, 8(%1), %%mm1, q)\
1129 :: "r"(temp), "r"(dst), "m"(ROUNDER)\
1137 static void OPNAME ## mpeg4_qpel8_h_lowpass_mmx2(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h){\
1139 "pxor %%mm7, %%mm7 \n\t"\
1141 "movq (%0), %%mm0 \n\t" /* ABCDEFGH */\
1142 "movq %%mm0, %%mm1 \n\t" /* ABCDEFGH */\
1143 "movq %%mm0, %%mm2 \n\t" /* ABCDEFGH */\
1144 "punpcklbw %%mm7, %%mm0 \n\t" /* 0A0B0C0D */\
1145 "punpckhbw %%mm7, %%mm1 \n\t" /* 0E0F0G0H */\
1146 "pshufw $0x90, %%mm0, %%mm5 \n\t" /* 0A0A0B0C */\
1147 "pshufw $0x41, %%mm0, %%mm6 \n\t" /* 0B0A0A0B */\
1148 "movq %%mm2, %%mm3 \n\t" /* ABCDEFGH */\
1149 "movq %%mm2, %%mm4 \n\t" /* ABCDEFGH */\
1150 "psllq $8, %%mm2 \n\t" /* 0ABCDEFG */\
1151 "psllq $16, %%mm3 \n\t" /* 00ABCDEF */\
1152 "psllq $24, %%mm4 \n\t" /* 000ABCDE */\
1153 "punpckhbw %%mm7, %%mm2 \n\t" /* 0D0E0F0G */\
1154 "punpckhbw %%mm7, %%mm3 \n\t" /* 0C0D0E0F */\
1155 "punpckhbw %%mm7, %%mm4 \n\t" /* 0B0C0D0E */\
1156 "paddw %%mm3, %%mm5 \n\t" /* b */\
1157 "paddw %%mm2, %%mm6 \n\t" /* c */\
1158 "paddw %%mm5, %%mm5 \n\t" /* 2b */\
1159 "psubw %%mm5, %%mm6 \n\t" /* c - 2b */\
1160 "pshufw $0x06, %%mm0, %%mm5 \n\t" /* 0C0B0A0A */\
1161 "pmullw "MANGLE(ff_pw_3)", %%mm6 \n\t" /* 3c - 6b */\
1162 "paddw %%mm4, %%mm0 \n\t" /* a */\
1163 "paddw %%mm1, %%mm5 \n\t" /* d */\
1164 "pmullw "MANGLE(ff_pw_20)", %%mm0 \n\t" /* 20a */\
1165 "psubw %%mm5, %%mm0 \n\t" /* 20a - d */\
1166 "paddw %5, %%mm6 \n\t"\
1167 "paddw %%mm6, %%mm0 \n\t" /* 20a - 6b + 3c - d */\
1168 "psraw $5, %%mm0 \n\t"\
1169 /* mm1=EFGH, mm2=DEFG, mm3=CDEF, mm4=BCDE, mm7=0 */\
1171 "movd 5(%0), %%mm5 \n\t" /* FGHI */\
1172 "punpcklbw %%mm7, %%mm5 \n\t" /* 0F0G0H0I */\
1173 "pshufw $0xF9, %%mm5, %%mm6 \n\t" /* 0G0H0I0I */\
1174 "paddw %%mm5, %%mm1 \n\t" /* a */\
1175 "paddw %%mm6, %%mm2 \n\t" /* b */\
1176 "pshufw $0xBE, %%mm5, %%mm6 \n\t" /* 0H0I0I0H */\
1177 "pshufw $0x6F, %%mm5, %%mm5 \n\t" /* 0I0I0H0G */\
1178 "paddw %%mm6, %%mm3 \n\t" /* c */\
1179 "paddw %%mm5, %%mm4 \n\t" /* d */\
1180 "paddw %%mm2, %%mm2 \n\t" /* 2b */\
1181 "psubw %%mm2, %%mm3 \n\t" /* c - 2b */\
1182 "pmullw "MANGLE(ff_pw_20)", %%mm1 \n\t" /* 20a */\
1183 "pmullw "MANGLE(ff_pw_3)", %%mm3 \n\t" /* 3c - 6b */\
1184 "psubw %%mm4, %%mm3 \n\t" /* -6b + 3c - d */\
1185 "paddw %5, %%mm1 \n\t"\
1186 "paddw %%mm1, %%mm3 \n\t" /* 20a - 6b + 3c - d */\
1187 "psraw $5, %%mm3 \n\t"\
1188 "packuswb %%mm3, %%mm0 \n\t"\
1189 OP_MMX2(%%mm0, (%1), %%mm4, q)\
1195 : "+a"(src), "+c"(dst), "+d"(h)\
1196 : "S"((x86_reg)srcStride), "D"((x86_reg)dstStride), /*"m"(ff_pw_20), "m"(ff_pw_3),*/ "m"(ROUNDER)\
1201 static void OPNAME ## mpeg4_qpel8_h_lowpass_3dnow(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h){\
1204 /* quick HACK, XXX FIXME MUST be optimized */\
1207 temp[ 0]= (src[ 0]+src[ 1])*20 - (src[ 0]+src[ 2])*6 + (src[ 1]+src[ 3])*3 - (src[ 2]+src[ 4]);\
1208 temp[ 1]= (src[ 1]+src[ 2])*20 - (src[ 0]+src[ 3])*6 + (src[ 0]+src[ 4])*3 - (src[ 1]+src[ 5]);\
1209 temp[ 2]= (src[ 2]+src[ 3])*20 - (src[ 1]+src[ 4])*6 + (src[ 0]+src[ 5])*3 - (src[ 0]+src[ 6]);\
1210 temp[ 3]= (src[ 3]+src[ 4])*20 - (src[ 2]+src[ 5])*6 + (src[ 1]+src[ 6])*3 - (src[ 0]+src[ 7]);\
1211 temp[ 4]= (src[ 4]+src[ 5])*20 - (src[ 3]+src[ 6])*6 + (src[ 2]+src[ 7])*3 - (src[ 1]+src[ 8]);\
1212 temp[ 5]= (src[ 5]+src[ 6])*20 - (src[ 4]+src[ 7])*6 + (src[ 3]+src[ 8])*3 - (src[ 2]+src[ 8]);\
1213 temp[ 6]= (src[ 6]+src[ 7])*20 - (src[ 5]+src[ 8])*6 + (src[ 4]+src[ 8])*3 - (src[ 3]+src[ 7]);\
1214 temp[ 7]= (src[ 7]+src[ 8])*20 - (src[ 6]+src[ 8])*6 + (src[ 5]+src[ 7])*3 - (src[ 4]+src[ 6]);\
1216 "movq (%0), %%mm0 \n\t"\
1217 "movq 8(%0), %%mm1 \n\t"\
1218 "paddw %2, %%mm0 \n\t"\
1219 "paddw %2, %%mm1 \n\t"\
1220 "psraw $5, %%mm0 \n\t"\
1221 "psraw $5, %%mm1 \n\t"\
1222 "packuswb %%mm1, %%mm0 \n\t"\
1223 OP_3DNOW(%%mm0, (%1), %%mm1, q)\
1224 :: "r"(temp), "r"(dst), "m"(ROUNDER)\
1232 #define QPEL_OP(OPNAME, ROUNDER, RND, OP, MMX)\
1234 static void OPNAME ## mpeg4_qpel16_v_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
1235 uint64_t temp[17*4];\
1236 uint64_t *temp_ptr= temp;\
1241 "pxor %%mm7, %%mm7 \n\t"\
1243 "movq (%0), %%mm0 \n\t"\
1244 "movq (%0), %%mm1 \n\t"\
1245 "movq 8(%0), %%mm2 \n\t"\
1246 "movq 8(%0), %%mm3 \n\t"\
1247 "punpcklbw %%mm7, %%mm0 \n\t"\
1248 "punpckhbw %%mm7, %%mm1 \n\t"\
1249 "punpcklbw %%mm7, %%mm2 \n\t"\
1250 "punpckhbw %%mm7, %%mm3 \n\t"\
1251 "movq %%mm0, (%1) \n\t"\
1252 "movq %%mm1, 17*8(%1) \n\t"\
1253 "movq %%mm2, 2*17*8(%1) \n\t"\
1254 "movq %%mm3, 3*17*8(%1) \n\t"\
1259 : "+r" (src), "+r" (temp_ptr), "+r"(count)\
1260 : "r" ((x86_reg)srcStride)\
1267 /*FIXME reorder for speed */\
1269 /*"pxor %%mm7, %%mm7 \n\t"*/\
1271 "movq (%0), %%mm0 \n\t"\
1272 "movq 8(%0), %%mm1 \n\t"\
1273 "movq 16(%0), %%mm2 \n\t"\
1274 "movq 24(%0), %%mm3 \n\t"\
1275 QPEL_V_LOW(%%mm0, %%mm1, %%mm2, %%mm3, %5, %6, %5, 16(%0), 8(%0), (%0), 32(%0), (%1), OP)\
1276 QPEL_V_LOW(%%mm1, %%mm2, %%mm3, %%mm0, %5, %6, %5, 8(%0), (%0), (%0), 40(%0), (%1, %3), OP)\
1278 QPEL_V_LOW(%%mm2, %%mm3, %%mm0, %%mm1, %5, %6, %5, (%0), (%0), 8(%0), 48(%0), (%1), OP)\
1280 QPEL_V_LOW(%%mm3, %%mm0, %%mm1, %%mm2, %5, %6, %5, (%0), 8(%0), 16(%0), 56(%0), (%1, %3), OP)\
1282 QPEL_V_LOW(%%mm0, %%mm1, %%mm2, %%mm3, %5, %6, %5, 8(%0), 16(%0), 24(%0), 64(%0), (%1), OP)\
1283 QPEL_V_LOW(%%mm1, %%mm2, %%mm3, %%mm0, %5, %6, %5, 16(%0), 24(%0), 32(%0), 72(%0), (%1, %3), OP)\
1285 QPEL_V_LOW(%%mm2, %%mm3, %%mm0, %%mm1, %5, %6, %5, 24(%0), 32(%0), 40(%0), 80(%0), (%1), OP)\
1286 QPEL_V_LOW(%%mm3, %%mm0, %%mm1, %%mm2, %5, %6, %5, 32(%0), 40(%0), 48(%0), 88(%0), (%1, %3), OP)\
1288 QPEL_V_LOW(%%mm0, %%mm1, %%mm2, %%mm3, %5, %6, %5, 40(%0), 48(%0), 56(%0), 96(%0), (%1), OP)\
1289 QPEL_V_LOW(%%mm1, %%mm2, %%mm3, %%mm0, %5, %6, %5, 48(%0), 56(%0), 64(%0),104(%0), (%1, %3), OP)\
1291 QPEL_V_LOW(%%mm2, %%mm3, %%mm0, %%mm1, %5, %6, %5, 56(%0), 64(%0), 72(%0),112(%0), (%1), OP)\
1292 QPEL_V_LOW(%%mm3, %%mm0, %%mm1, %%mm2, %5, %6, %5, 64(%0), 72(%0), 80(%0),120(%0), (%1, %3), OP)\
1294 QPEL_V_LOW(%%mm0, %%mm1, %%mm2, %%mm3, %5, %6, %5, 72(%0), 80(%0), 88(%0),128(%0), (%1), OP)\
1296 QPEL_V_LOW(%%mm1, %%mm2, %%mm3, %%mm0, %5, %6, %5, 80(%0), 88(%0), 96(%0),128(%0), (%1, %3), OP)\
1298 QPEL_V_LOW(%%mm2, %%mm3, %%mm0, %%mm1, %5, %6, %5, 88(%0), 96(%0),104(%0),120(%0), (%1), OP)\
1299 QPEL_V_LOW(%%mm3, %%mm0, %%mm1, %%mm2, %5, %6, %5, 96(%0),104(%0),112(%0),112(%0), (%1, %3), OP)\
1301 "add $136, %0 \n\t"\
1306 : "+r"(temp_ptr), "+r"(dst), "+g"(count)\
1307 : "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)\
1312 static void OPNAME ## mpeg4_qpel8_v_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
1313 uint64_t temp[9*2];\
1314 uint64_t *temp_ptr= temp;\
1319 "pxor %%mm7, %%mm7 \n\t"\
1321 "movq (%0), %%mm0 \n\t"\
1322 "movq (%0), %%mm1 \n\t"\
1323 "punpcklbw %%mm7, %%mm0 \n\t"\
1324 "punpckhbw %%mm7, %%mm1 \n\t"\
1325 "movq %%mm0, (%1) \n\t"\
1326 "movq %%mm1, 9*8(%1) \n\t"\
1331 : "+r" (src), "+r" (temp_ptr), "+r"(count)\
1332 : "r" ((x86_reg)srcStride)\
1339 /*FIXME reorder for speed */\
1341 /*"pxor %%mm7, %%mm7 \n\t"*/\
1343 "movq (%0), %%mm0 \n\t"\
1344 "movq 8(%0), %%mm1 \n\t"\
1345 "movq 16(%0), %%mm2 \n\t"\
1346 "movq 24(%0), %%mm3 \n\t"\
1347 QPEL_V_LOW(%%mm0, %%mm1, %%mm2, %%mm3, %5, %6, %5, 16(%0), 8(%0), (%0), 32(%0), (%1), OP)\
1348 QPEL_V_LOW(%%mm1, %%mm2, %%mm3, %%mm0, %5, %6, %5, 8(%0), (%0), (%0), 40(%0), (%1, %3), OP)\
1350 QPEL_V_LOW(%%mm2, %%mm3, %%mm0, %%mm1, %5, %6, %5, (%0), (%0), 8(%0), 48(%0), (%1), OP)\
1352 QPEL_V_LOW(%%mm3, %%mm0, %%mm1, %%mm2, %5, %6, %5, (%0), 8(%0), 16(%0), 56(%0), (%1, %3), OP)\
1354 QPEL_V_LOW(%%mm0, %%mm1, %%mm2, %%mm3, %5, %6, %5, 8(%0), 16(%0), 24(%0), 64(%0), (%1), OP)\
1356 QPEL_V_LOW(%%mm1, %%mm2, %%mm3, %%mm0, %5, %6, %5, 16(%0), 24(%0), 32(%0), 64(%0), (%1, %3), OP)\
1358 QPEL_V_LOW(%%mm2, %%mm3, %%mm0, %%mm1, %5, %6, %5, 24(%0), 32(%0), 40(%0), 56(%0), (%1), OP)\
1359 QPEL_V_LOW(%%mm3, %%mm0, %%mm1, %%mm2, %5, %6, %5, 32(%0), 40(%0), 48(%0), 48(%0), (%1, %3), OP)\
1366 : "+r"(temp_ptr), "+r"(dst), "+g"(count)\
1367 : "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)\
1372 static void OPNAME ## qpel8_mc00_ ## MMX (uint8_t *dst, uint8_t *src, int stride){\
1373 OPNAME ## pixels8_ ## MMX(dst, src, stride, 8);\
1376 static void OPNAME ## qpel8_mc10_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1378 uint8_t * const half= (uint8_t*)temp;\
1379 put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(half, src, 8, stride, 8);\
1380 OPNAME ## pixels8_l2_ ## MMX(dst, src, half, stride, stride, 8);\
1383 static void OPNAME ## qpel8_mc20_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1384 OPNAME ## mpeg4_qpel8_h_lowpass_ ## MMX(dst, src, stride, stride, 8);\
1387 static void OPNAME ## qpel8_mc30_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1389 uint8_t * const half= (uint8_t*)temp;\
1390 put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(half, src, 8, stride, 8);\
1391 OPNAME ## pixels8_l2_ ## MMX(dst, src+1, half, stride, stride, 8);\
1394 static void OPNAME ## qpel8_mc01_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1396 uint8_t * const half= (uint8_t*)temp;\
1397 put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(half, src, 8, stride);\
1398 OPNAME ## pixels8_l2_ ## MMX(dst, src, half, stride, stride, 8);\
1401 static void OPNAME ## qpel8_mc02_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1402 OPNAME ## mpeg4_qpel8_v_lowpass_ ## MMX(dst, src, stride, stride);\
1405 static void OPNAME ## qpel8_mc03_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1407 uint8_t * const half= (uint8_t*)temp;\
1408 put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(half, src, 8, stride);\
1409 OPNAME ## pixels8_l2_ ## MMX(dst, src+stride, half, stride, stride, 8);\
1411 static void OPNAME ## qpel8_mc11_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1412 uint64_t half[8 + 9];\
1413 uint8_t * const halfH= ((uint8_t*)half) + 64;\
1414 uint8_t * const halfHV= ((uint8_t*)half);\
1415 put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, stride, 9);\
1416 put ## RND ## pixels8_l2_ ## MMX(halfH, src, halfH, 8, stride, 9);\
1417 put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(halfHV, halfH, 8, 8);\
1418 OPNAME ## pixels8_l2_ ## MMX(dst, halfH, halfHV, stride, 8, 8);\
1420 static void OPNAME ## qpel8_mc31_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1421 uint64_t half[8 + 9];\
1422 uint8_t * const halfH= ((uint8_t*)half) + 64;\
1423 uint8_t * const halfHV= ((uint8_t*)half);\
1424 put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, stride, 9);\
1425 put ## RND ## pixels8_l2_ ## MMX(halfH, src+1, halfH, 8, stride, 9);\
1426 put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(halfHV, halfH, 8, 8);\
1427 OPNAME ## pixels8_l2_ ## MMX(dst, halfH, halfHV, stride, 8, 8);\
1429 static void OPNAME ## qpel8_mc13_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1430 uint64_t half[8 + 9];\
1431 uint8_t * const halfH= ((uint8_t*)half) + 64;\
1432 uint8_t * const halfHV= ((uint8_t*)half);\
1433 put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, stride, 9);\
1434 put ## RND ## pixels8_l2_ ## MMX(halfH, src, halfH, 8, stride, 9);\
1435 put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(halfHV, halfH, 8, 8);\
1436 OPNAME ## pixels8_l2_ ## MMX(dst, halfH+8, halfHV, stride, 8, 8);\
1438 static void OPNAME ## qpel8_mc33_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1439 uint64_t half[8 + 9];\
1440 uint8_t * const halfH= ((uint8_t*)half) + 64;\
1441 uint8_t * const halfHV= ((uint8_t*)half);\
1442 put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, stride, 9);\
1443 put ## RND ## pixels8_l2_ ## MMX(halfH, src+1, halfH, 8, stride, 9);\
1444 put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(halfHV, halfH, 8, 8);\
1445 OPNAME ## pixels8_l2_ ## MMX(dst, halfH+8, halfHV, stride, 8, 8);\
1447 static void OPNAME ## qpel8_mc21_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1448 uint64_t half[8 + 9];\
1449 uint8_t * const halfH= ((uint8_t*)half) + 64;\
1450 uint8_t * const halfHV= ((uint8_t*)half);\
1451 put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, stride, 9);\
1452 put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(halfHV, halfH, 8, 8);\
1453 OPNAME ## pixels8_l2_ ## MMX(dst, halfH, halfHV, stride, 8, 8);\
1455 static void OPNAME ## qpel8_mc23_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1456 uint64_t half[8 + 9];\
1457 uint8_t * const halfH= ((uint8_t*)half) + 64;\
1458 uint8_t * const halfHV= ((uint8_t*)half);\
1459 put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, stride, 9);\
1460 put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(halfHV, halfH, 8, 8);\
1461 OPNAME ## pixels8_l2_ ## MMX(dst, halfH+8, halfHV, stride, 8, 8);\
1463 static void OPNAME ## qpel8_mc12_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1464 uint64_t half[8 + 9];\
1465 uint8_t * const halfH= ((uint8_t*)half);\
1466 put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, stride, 9);\
1467 put ## RND ## pixels8_l2_ ## MMX(halfH, src, halfH, 8, stride, 9);\
1468 OPNAME ## mpeg4_qpel8_v_lowpass_ ## MMX(dst, halfH, stride, 8);\
1470 static void OPNAME ## qpel8_mc32_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1471 uint64_t half[8 + 9];\
1472 uint8_t * const halfH= ((uint8_t*)half);\
1473 put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, stride, 9);\
1474 put ## RND ## pixels8_l2_ ## MMX(halfH, src+1, halfH, 8, stride, 9);\
1475 OPNAME ## mpeg4_qpel8_v_lowpass_ ## MMX(dst, halfH, stride, 8);\
1477 static void OPNAME ## qpel8_mc22_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1479 uint8_t * const halfH= ((uint8_t*)half);\
1480 put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, stride, 9);\
1481 OPNAME ## mpeg4_qpel8_v_lowpass_ ## MMX(dst, halfH, stride, 8);\
1483 static void OPNAME ## qpel16_mc00_ ## MMX (uint8_t *dst, uint8_t *src, int stride){\
1484 OPNAME ## pixels16_ ## MMX(dst, src, stride, 16);\
1487 static void OPNAME ## qpel16_mc10_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1489 uint8_t * const half= (uint8_t*)temp;\
1490 put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(half, src, 16, stride, 16);\
1491 OPNAME ## pixels16_l2_ ## MMX(dst, src, half, stride, stride, 16);\
1494 static void OPNAME ## qpel16_mc20_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1495 OPNAME ## mpeg4_qpel16_h_lowpass_ ## MMX(dst, src, stride, stride, 16);\
1498 static void OPNAME ## qpel16_mc30_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1500 uint8_t * const half= (uint8_t*)temp;\
1501 put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(half, src, 16, stride, 16);\
1502 OPNAME ## pixels16_l2_ ## MMX(dst, src+1, half, stride, stride, 16);\
1505 static void OPNAME ## qpel16_mc01_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1507 uint8_t * const half= (uint8_t*)temp;\
1508 put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(half, src, 16, stride);\
1509 OPNAME ## pixels16_l2_ ## MMX(dst, src, half, stride, stride, 16);\
1512 static void OPNAME ## qpel16_mc02_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1513 OPNAME ## mpeg4_qpel16_v_lowpass_ ## MMX(dst, src, stride, stride);\
1516 static void OPNAME ## qpel16_mc03_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1518 uint8_t * const half= (uint8_t*)temp;\
1519 put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(half, src, 16, stride);\
1520 OPNAME ## pixels16_l2_ ## MMX(dst, src+stride, half, stride, stride, 16);\
1522 static void OPNAME ## qpel16_mc11_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1523 uint64_t half[16*2 + 17*2];\
1524 uint8_t * const halfH= ((uint8_t*)half) + 256;\
1525 uint8_t * const halfHV= ((uint8_t*)half);\
1526 put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16, stride, 17);\
1527 put ## RND ## pixels16_l2_ ## MMX(halfH, src, halfH, 16, stride, 17);\
1528 put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(halfHV, halfH, 16, 16);\
1529 OPNAME ## pixels16_l2_ ## MMX(dst, halfH, halfHV, stride, 16, 16);\
1531 static void OPNAME ## qpel16_mc31_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1532 uint64_t half[16*2 + 17*2];\
1533 uint8_t * const halfH= ((uint8_t*)half) + 256;\
1534 uint8_t * const halfHV= ((uint8_t*)half);\
1535 put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16, stride, 17);\
1536 put ## RND ## pixels16_l2_ ## MMX(halfH, src+1, halfH, 16, stride, 17);\
1537 put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(halfHV, halfH, 16, 16);\
1538 OPNAME ## pixels16_l2_ ## MMX(dst, halfH, halfHV, stride, 16, 16);\
1540 static void OPNAME ## qpel16_mc13_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1541 uint64_t half[16*2 + 17*2];\
1542 uint8_t * const halfH= ((uint8_t*)half) + 256;\
1543 uint8_t * const halfHV= ((uint8_t*)half);\
1544 put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16, stride, 17);\
1545 put ## RND ## pixels16_l2_ ## MMX(halfH, src, halfH, 16, stride, 17);\
1546 put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(halfHV, halfH, 16, 16);\
1547 OPNAME ## pixels16_l2_ ## MMX(dst, halfH+16, halfHV, stride, 16, 16);\
1549 static void OPNAME ## qpel16_mc33_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1550 uint64_t half[16*2 + 17*2];\
1551 uint8_t * const halfH= ((uint8_t*)half) + 256;\
1552 uint8_t * const halfHV= ((uint8_t*)half);\
1553 put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16, stride, 17);\
1554 put ## RND ## pixels16_l2_ ## MMX(halfH, src+1, halfH, 16, stride, 17);\
1555 put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(halfHV, halfH, 16, 16);\
1556 OPNAME ## pixels16_l2_ ## MMX(dst, halfH+16, halfHV, stride, 16, 16);\
1558 static void OPNAME ## qpel16_mc21_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1559 uint64_t half[16*2 + 17*2];\
1560 uint8_t * const halfH= ((uint8_t*)half) + 256;\
1561 uint8_t * const halfHV= ((uint8_t*)half);\
1562 put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16, stride, 17);\
1563 put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(halfHV, halfH, 16, 16);\
1564 OPNAME ## pixels16_l2_ ## MMX(dst, halfH, halfHV, stride, 16, 16);\
1566 static void OPNAME ## qpel16_mc23_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1567 uint64_t half[16*2 + 17*2];\
1568 uint8_t * const halfH= ((uint8_t*)half) + 256;\
1569 uint8_t * const halfHV= ((uint8_t*)half);\
1570 put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16, stride, 17);\
1571 put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(halfHV, halfH, 16, 16);\
1572 OPNAME ## pixels16_l2_ ## MMX(dst, halfH+16, halfHV, stride, 16, 16);\
1574 static void OPNAME ## qpel16_mc12_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1575 uint64_t half[17*2];\
1576 uint8_t * const halfH= ((uint8_t*)half);\
1577 put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16, stride, 17);\
1578 put ## RND ## pixels16_l2_ ## MMX(halfH, src, halfH, 16, stride, 17);\
1579 OPNAME ## mpeg4_qpel16_v_lowpass_ ## MMX(dst, halfH, stride, 16);\
1581 static void OPNAME ## qpel16_mc32_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1582 uint64_t half[17*2];\
1583 uint8_t * const halfH= ((uint8_t*)half);\
1584 put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16, stride, 17);\
1585 put ## RND ## pixels16_l2_ ## MMX(halfH, src+1, halfH, 16, stride, 17);\
1586 OPNAME ## mpeg4_qpel16_v_lowpass_ ## MMX(dst, halfH, stride, 16);\
1588 static void OPNAME ## qpel16_mc22_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1589 uint64_t half[17*2];\
1590 uint8_t * const halfH= ((uint8_t*)half);\
1591 put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16, stride, 17);\
1592 OPNAME ## mpeg4_qpel16_v_lowpass_ ## MMX(dst, halfH, stride, 16);\
1595 #define PUT_OP(a,b,temp, size) "mov" #size " " #a ", " #b " \n\t"
1596 #define AVG_3DNOW_OP(a,b,temp, size) \
1597 "mov" #size " " #b ", " #temp " \n\t"\
1598 "pavgusb " #temp ", " #a " \n\t"\
1599 "mov" #size " " #a ", " #b " \n\t"
1600 #define AVG_MMX2_OP(a,b,temp, size) \
1601 "mov" #size " " #b ", " #temp " \n\t"\
1602 "pavgb " #temp ", " #a " \n\t"\
1603 "mov" #size " " #a ", " #b " \n\t"
1605 QPEL_BASE(put_ , ff_pw_16, _ , PUT_OP, PUT_OP)
1606 QPEL_BASE(avg_ , ff_pw_16, _ , AVG_MMX2_OP, AVG_3DNOW_OP)
1607 QPEL_BASE(put_no_rnd_, ff_pw_15, _no_rnd_, PUT_OP, PUT_OP)
1608 QPEL_OP(put_ , ff_pw_16, _ , PUT_OP, 3dnow)
1609 QPEL_OP(avg_ , ff_pw_16, _ , AVG_3DNOW_OP, 3dnow)
1610 QPEL_OP(put_no_rnd_, ff_pw_15, _no_rnd_, PUT_OP, 3dnow)
1611 QPEL_OP(put_ , ff_pw_16, _ , PUT_OP, mmx2)
1612 QPEL_OP(avg_ , ff_pw_16, _ , AVG_MMX2_OP, mmx2)
1613 QPEL_OP(put_no_rnd_, ff_pw_15, _no_rnd_, PUT_OP, mmx2)
1615 /***********************************/
1616 /* bilinear qpel: not compliant to any spec, only for -lavdopts fast */
1618 #define QPEL_2TAP_XY(OPNAME, SIZE, MMX, XY, HPEL)\
1619 static void OPNAME ## 2tap_qpel ## SIZE ## _mc ## XY ## _ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1620 OPNAME ## pixels ## SIZE ## HPEL(dst, src, stride, SIZE);\
1622 #define QPEL_2TAP_L3(OPNAME, SIZE, MMX, XY, S0, S1, S2)\
1623 static void OPNAME ## 2tap_qpel ## SIZE ## _mc ## XY ## _ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1624 OPNAME ## 2tap_qpel ## SIZE ## _l3_ ## MMX(dst, src+S0, stride, SIZE, S1, S2);\
1627 #define QPEL_2TAP(OPNAME, SIZE, MMX)\
1628 QPEL_2TAP_XY(OPNAME, SIZE, MMX, 20, _x2_ ## MMX)\
1629 QPEL_2TAP_XY(OPNAME, SIZE, MMX, 02, _y2_ ## MMX)\
1630 QPEL_2TAP_XY(OPNAME, SIZE, MMX, 22, _xy2_mmx)\
1631 static const qpel_mc_func OPNAME ## 2tap_qpel ## SIZE ## _mc00_ ## MMX =\
1632 OPNAME ## qpel ## SIZE ## _mc00_ ## MMX;\
1633 static const qpel_mc_func OPNAME ## 2tap_qpel ## SIZE ## _mc21_ ## MMX =\
1634 OPNAME ## 2tap_qpel ## SIZE ## _mc20_ ## MMX;\
1635 static const qpel_mc_func OPNAME ## 2tap_qpel ## SIZE ## _mc12_ ## MMX =\
1636 OPNAME ## 2tap_qpel ## SIZE ## _mc02_ ## MMX;\
1637 static void OPNAME ## 2tap_qpel ## SIZE ## _mc32_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1638 OPNAME ## pixels ## SIZE ## _y2_ ## MMX(dst, src+1, stride, SIZE);\
1640 static void OPNAME ## 2tap_qpel ## SIZE ## _mc23_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1641 OPNAME ## pixels ## SIZE ## _x2_ ## MMX(dst, src+stride, stride, SIZE);\
1643 QPEL_2TAP_L3(OPNAME, SIZE, MMX, 10, 0, 1, 0)\
1644 QPEL_2TAP_L3(OPNAME, SIZE, MMX, 30, 1, -1, 0)\
1645 QPEL_2TAP_L3(OPNAME, SIZE, MMX, 01, 0, stride, 0)\
1646 QPEL_2TAP_L3(OPNAME, SIZE, MMX, 03, stride, -stride, 0)\
1647 QPEL_2TAP_L3(OPNAME, SIZE, MMX, 11, 0, stride, 1)\
1648 QPEL_2TAP_L3(OPNAME, SIZE, MMX, 31, 1, stride, -1)\
1649 QPEL_2TAP_L3(OPNAME, SIZE, MMX, 13, stride, -stride, 1)\
1650 QPEL_2TAP_L3(OPNAME, SIZE, MMX, 33, stride+1, -stride, -1)\
1652 QPEL_2TAP(put_, 16, mmx2)
1653 QPEL_2TAP(avg_, 16, mmx2)
1654 QPEL_2TAP(put_, 8, mmx2)
1655 QPEL_2TAP(avg_, 8, mmx2)
1656 QPEL_2TAP(put_, 16, 3dnow)
1657 QPEL_2TAP(avg_, 16, 3dnow)
1658 QPEL_2TAP(put_, 8, 3dnow)
1659 QPEL_2TAP(avg_, 8, 3dnow)
1663 static void just_return(void) { return; }
1666 static void gmc_mmx(uint8_t *dst, uint8_t *src, int stride, int h, int ox, int oy,
1667 int dxx, int dxy, int dyx, int dyy, int shift, int r, int width, int height){
1669 const int ix = ox>>(16+shift);
1670 const int iy = oy>>(16+shift);
1671 const int oxs = ox>>4;
1672 const int oys = oy>>4;
1673 const int dxxs = dxx>>4;
1674 const int dxys = dxy>>4;
1675 const int dyxs = dyx>>4;
1676 const int dyys = dyy>>4;
1677 const uint16_t r4[4] = {r,r,r,r};
1678 const uint16_t dxy4[4] = {dxys,dxys,dxys,dxys};
1679 const uint16_t dyy4[4] = {dyys,dyys,dyys,dyys};
1680 const uint64_t shift2 = 2*shift;
1681 uint8_t edge_buf[(h+1)*stride];
1684 const int dxw = (dxx-(1<<(16+shift)))*(w-1);
1685 const int dyh = (dyy-(1<<(16+shift)))*(h-1);
1686 const int dxh = dxy*(h-1);
1687 const int dyw = dyx*(w-1);
1688 if( // non-constant fullpel offset (3% of blocks)
1689 ((ox^(ox+dxw)) | (ox^(ox+dxh)) | (ox^(ox+dxw+dxh)) |
1690 (oy^(oy+dyw)) | (oy^(oy+dyh)) | (oy^(oy+dyw+dyh))) >> (16+shift)
1691 // uses more than 16 bits of subpel mv (only at huge resolution)
1692 || (dxx|dxy|dyx|dyy)&15 )
1694 //FIXME could still use mmx for some of the rows
1695 ff_gmc_c(dst, src, stride, h, ox, oy, dxx, dxy, dyx, dyy, shift, r, width, height);
1699 src += ix + iy*stride;
1700 if( (unsigned)ix >= width-w ||
1701 (unsigned)iy >= height-h )
1703 ff_emulated_edge_mc(edge_buf, src, stride, w+1, h+1, ix, iy, width, height);
1708 "movd %0, %%mm6 \n\t"
1709 "pxor %%mm7, %%mm7 \n\t"
1710 "punpcklwd %%mm6, %%mm6 \n\t"
1711 "punpcklwd %%mm6, %%mm6 \n\t"
1715 for(x=0; x<w; x+=4){
1716 uint16_t dx4[4] = { oxs - dxys + dxxs*(x+0),
1717 oxs - dxys + dxxs*(x+1),
1718 oxs - dxys + dxxs*(x+2),
1719 oxs - dxys + dxxs*(x+3) };
1720 uint16_t dy4[4] = { oys - dyys + dyxs*(x+0),
1721 oys - dyys + dyxs*(x+1),
1722 oys - dyys + dyxs*(x+2),
1723 oys - dyys + dyxs*(x+3) };
1727 "movq %0, %%mm4 \n\t"
1728 "movq %1, %%mm5 \n\t"
1729 "paddw %2, %%mm4 \n\t"
1730 "paddw %3, %%mm5 \n\t"
1731 "movq %%mm4, %0 \n\t"
1732 "movq %%mm5, %1 \n\t"
1733 "psrlw $12, %%mm4 \n\t"
1734 "psrlw $12, %%mm5 \n\t"
1735 : "+m"(*dx4), "+m"(*dy4)
1736 : "m"(*dxy4), "m"(*dyy4)
1740 "movq %%mm6, %%mm2 \n\t"
1741 "movq %%mm6, %%mm1 \n\t"
1742 "psubw %%mm4, %%mm2 \n\t"
1743 "psubw %%mm5, %%mm1 \n\t"
1744 "movq %%mm2, %%mm0 \n\t"
1745 "movq %%mm4, %%mm3 \n\t"
1746 "pmullw %%mm1, %%mm0 \n\t" // (s-dx)*(s-dy)
1747 "pmullw %%mm5, %%mm3 \n\t" // dx*dy
1748 "pmullw %%mm5, %%mm2 \n\t" // (s-dx)*dy
1749 "pmullw %%mm4, %%mm1 \n\t" // dx*(s-dy)
1751 "movd %4, %%mm5 \n\t"
1752 "movd %3, %%mm4 \n\t"
1753 "punpcklbw %%mm7, %%mm5 \n\t"
1754 "punpcklbw %%mm7, %%mm4 \n\t"
1755 "pmullw %%mm5, %%mm3 \n\t" // src[1,1] * dx*dy
1756 "pmullw %%mm4, %%mm2 \n\t" // src[0,1] * (s-dx)*dy
1758 "movd %2, %%mm5 \n\t"
1759 "movd %1, %%mm4 \n\t"
1760 "punpcklbw %%mm7, %%mm5 \n\t"
1761 "punpcklbw %%mm7, %%mm4 \n\t"
1762 "pmullw %%mm5, %%mm1 \n\t" // src[1,0] * dx*(s-dy)
1763 "pmullw %%mm4, %%mm0 \n\t" // src[0,0] * (s-dx)*(s-dy)
1764 "paddw %5, %%mm1 \n\t"
1765 "paddw %%mm3, %%mm2 \n\t"
1766 "paddw %%mm1, %%mm0 \n\t"
1767 "paddw %%mm2, %%mm0 \n\t"
1769 "psrlw %6, %%mm0 \n\t"
1770 "packuswb %%mm0, %%mm0 \n\t"
1771 "movd %%mm0, %0 \n\t"
1773 : "=m"(dst[x+y*stride])
1774 : "m"(src[0]), "m"(src[1]),
1775 "m"(src[stride]), "m"(src[stride+1]),
1776 "m"(*r4), "m"(shift2)
1784 #define PREFETCH(name, op) \
1785 static void name(void *mem, int stride, int h){\
1786 const uint8_t *p= mem;\
1788 __asm__ volatile(#op" %0" :: "m"(*p));\
1792 PREFETCH(prefetch_mmx2, prefetcht0)
1793 PREFETCH(prefetch_3dnow, prefetch)
1796 #include "h264_qpel_mmx.c"
1798 void ff_put_h264_chroma_mc8_mmx_rnd (uint8_t *dst, uint8_t *src,
1799 int stride, int h, int x, int y);
1800 void ff_put_vc1_chroma_mc8_mmx_nornd (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_vc1_chroma_mc8_mmx2_nornd (uint8_t *dst, uint8_t *src,
1807 int stride, int h, int x, int y);
1808 void ff_avg_rv40_chroma_mc8_mmx2 (uint8_t *dst, uint8_t *src,
1809 int stride, int h, int x, int y);
1810 void ff_avg_h264_chroma_mc8_3dnow_rnd (uint8_t *dst, uint8_t *src,
1811 int stride, int h, int x, int y);
1812 void ff_avg_vc1_chroma_mc8_3dnow_nornd(uint8_t *dst, uint8_t *src,
1813 int stride, int h, int x, int y);
1814 void ff_avg_rv40_chroma_mc8_3dnow (uint8_t *dst, uint8_t *src,
1815 int stride, int h, int x, int y);
1817 void ff_put_h264_chroma_mc4_mmx (uint8_t *dst, uint8_t *src,
1818 int stride, int h, int x, int y);
1819 void ff_put_rv40_chroma_mc4_mmx (uint8_t *dst, uint8_t *src,
1820 int stride, int h, int x, int y);
1821 void ff_avg_h264_chroma_mc4_mmx2 (uint8_t *dst, uint8_t *src,
1822 int stride, int h, int x, int y);
1823 void ff_avg_rv40_chroma_mc4_mmx2 (uint8_t *dst, uint8_t *src,
1824 int stride, int h, int x, int y);
1825 void ff_avg_h264_chroma_mc4_3dnow (uint8_t *dst, uint8_t *src,
1826 int stride, int h, int x, int y);
1827 void ff_avg_rv40_chroma_mc4_3dnow (uint8_t *dst, uint8_t *src,
1828 int stride, int h, int x, int y);
1830 void ff_put_h264_chroma_mc2_mmx2 (uint8_t *dst, uint8_t *src,
1831 int stride, int h, int x, int y);
1832 void ff_avg_h264_chroma_mc2_mmx2 (uint8_t *dst, uint8_t *src,
1833 int stride, int h, int x, int y);
1835 void ff_put_h264_chroma_mc8_ssse3_rnd (uint8_t *dst, uint8_t *src,
1836 int stride, int h, int x, int y);
1837 void ff_put_vc1_chroma_mc8_ssse3_nornd(uint8_t *dst, uint8_t *src,
1838 int stride, int h, int x, int y);
1839 void ff_put_h264_chroma_mc4_ssse3 (uint8_t *dst, uint8_t *src,
1840 int stride, int h, int x, int y);
1842 void ff_avg_h264_chroma_mc8_ssse3_rnd (uint8_t *dst, uint8_t *src,
1843 int stride, int h, int x, int y);
1844 void ff_avg_vc1_chroma_mc8_ssse3_nornd(uint8_t *dst, uint8_t *src,
1845 int stride, int h, int x, int y);
1846 void ff_avg_h264_chroma_mc4_ssse3 (uint8_t *dst, uint8_t *src,
1847 int stride, int h, int x, int y);
1851 void ff_put_cavs_qpel8_mc00_mmx2(uint8_t *dst, uint8_t *src, int stride) {
1852 put_pixels8_mmx(dst, src, stride, 8);
1854 void ff_avg_cavs_qpel8_mc00_mmx2(uint8_t *dst, uint8_t *src, int stride) {
1855 avg_pixels8_mmx(dst, src, stride, 8);
1857 void ff_put_cavs_qpel16_mc00_mmx2(uint8_t *dst, uint8_t *src, int stride) {
1858 put_pixels16_mmx(dst, src, stride, 16);
1860 void ff_avg_cavs_qpel16_mc00_mmx2(uint8_t *dst, uint8_t *src, int stride) {
1861 avg_pixels16_mmx(dst, src, stride, 16);
1865 void ff_put_vc1_mspel_mc00_mmx(uint8_t *dst, const uint8_t *src, int stride, int rnd) {
1866 put_pixels8_mmx(dst, src, stride, 8);
1868 void ff_avg_vc1_mspel_mc00_mmx2(uint8_t *dst, const uint8_t *src, int stride, int rnd) {
1869 avg_pixels8_mmx2(dst, src, stride, 8);
1872 /* XXX: those functions should be suppressed ASAP when all IDCTs are
1875 static void ff_libmpeg2mmx_idct_put(uint8_t *dest, int line_size, DCTELEM *block)
1877 ff_mmx_idct (block);
1878 ff_put_pixels_clamped_mmx(block, dest, line_size);
1880 static void ff_libmpeg2mmx_idct_add(uint8_t *dest, int line_size, DCTELEM *block)
1882 ff_mmx_idct (block);
1883 ff_add_pixels_clamped_mmx(block, dest, line_size);
1885 static void ff_libmpeg2mmx2_idct_put(uint8_t *dest, int line_size, DCTELEM *block)
1887 ff_mmxext_idct (block);
1888 ff_put_pixels_clamped_mmx(block, dest, line_size);
1890 static void ff_libmpeg2mmx2_idct_add(uint8_t *dest, int line_size, DCTELEM *block)
1892 ff_mmxext_idct (block);
1893 ff_add_pixels_clamped_mmx(block, dest, line_size);
1896 static void ff_idct_xvid_mmx_put(uint8_t *dest, int line_size, DCTELEM *block)
1898 ff_idct_xvid_mmx (block);
1899 ff_put_pixels_clamped_mmx(block, dest, line_size);
1901 static void ff_idct_xvid_mmx_add(uint8_t *dest, int line_size, DCTELEM *block)
1903 ff_idct_xvid_mmx (block);
1904 ff_add_pixels_clamped_mmx(block, dest, line_size);
1906 static void ff_idct_xvid_mmx2_put(uint8_t *dest, int line_size, DCTELEM *block)
1908 ff_idct_xvid_mmx2 (block);
1909 ff_put_pixels_clamped_mmx(block, dest, line_size);
1911 static void ff_idct_xvid_mmx2_add(uint8_t *dest, int line_size, DCTELEM *block)
1913 ff_idct_xvid_mmx2 (block);
1914 ff_add_pixels_clamped_mmx(block, dest, line_size);
1917 static void vorbis_inverse_coupling_3dnow(float *mag, float *ang, int blocksize)
1920 __asm__ volatile("pxor %%mm7, %%mm7":);
1921 for(i=0; i<blocksize; i+=2) {
1923 "movq %0, %%mm0 \n\t"
1924 "movq %1, %%mm1 \n\t"
1925 "movq %%mm0, %%mm2 \n\t"
1926 "movq %%mm1, %%mm3 \n\t"
1927 "pfcmpge %%mm7, %%mm2 \n\t" // m <= 0.0
1928 "pfcmpge %%mm7, %%mm3 \n\t" // a <= 0.0
1929 "pslld $31, %%mm2 \n\t" // keep only the sign bit
1930 "pxor %%mm2, %%mm1 \n\t"
1931 "movq %%mm3, %%mm4 \n\t"
1932 "pand %%mm1, %%mm3 \n\t"
1933 "pandn %%mm1, %%mm4 \n\t"
1934 "pfadd %%mm0, %%mm3 \n\t" // a = m + ((a<0) & (a ^ sign(m)))
1935 "pfsub %%mm4, %%mm0 \n\t" // m = m + ((a>0) & (a ^ sign(m)))
1936 "movq %%mm3, %1 \n\t"
1937 "movq %%mm0, %0 \n\t"
1938 :"+m"(mag[i]), "+m"(ang[i])
1942 __asm__ volatile("femms");
1944 static void vorbis_inverse_coupling_sse(float *mag, float *ang, int blocksize)
1949 "movaps %0, %%xmm5 \n\t"
1950 ::"m"(ff_pdw_80000000[0])
1952 for(i=0; i<blocksize; i+=4) {
1954 "movaps %0, %%xmm0 \n\t"
1955 "movaps %1, %%xmm1 \n\t"
1956 "xorps %%xmm2, %%xmm2 \n\t"
1957 "xorps %%xmm3, %%xmm3 \n\t"
1958 "cmpleps %%xmm0, %%xmm2 \n\t" // m <= 0.0
1959 "cmpleps %%xmm1, %%xmm3 \n\t" // a <= 0.0
1960 "andps %%xmm5, %%xmm2 \n\t" // keep only the sign bit
1961 "xorps %%xmm2, %%xmm1 \n\t"
1962 "movaps %%xmm3, %%xmm4 \n\t"
1963 "andps %%xmm1, %%xmm3 \n\t"
1964 "andnps %%xmm1, %%xmm4 \n\t"
1965 "addps %%xmm0, %%xmm3 \n\t" // a = m + ((a<0) & (a ^ sign(m)))
1966 "subps %%xmm4, %%xmm0 \n\t" // m = m + ((a>0) & (a ^ sign(m)))
1967 "movaps %%xmm3, %1 \n\t"
1968 "movaps %%xmm0, %0 \n\t"
1969 :"+m"(mag[i]), "+m"(ang[i])
1978 #define MIX5(mono,stereo)\
1980 "movss 0(%2), %%xmm5 \n"\
1981 "movss 8(%2), %%xmm6 \n"\
1982 "movss 24(%2), %%xmm7 \n"\
1983 "shufps $0, %%xmm5, %%xmm5 \n"\
1984 "shufps $0, %%xmm6, %%xmm6 \n"\
1985 "shufps $0, %%xmm7, %%xmm7 \n"\
1987 "movaps (%0,%1), %%xmm0 \n"\
1988 "movaps 0x400(%0,%1), %%xmm1 \n"\
1989 "movaps 0x800(%0,%1), %%xmm2 \n"\
1990 "movaps 0xc00(%0,%1), %%xmm3 \n"\
1991 "movaps 0x1000(%0,%1), %%xmm4 \n"\
1992 "mulps %%xmm5, %%xmm0 \n"\
1993 "mulps %%xmm6, %%xmm1 \n"\
1994 "mulps %%xmm5, %%xmm2 \n"\
1995 "mulps %%xmm7, %%xmm3 \n"\
1996 "mulps %%xmm7, %%xmm4 \n"\
1997 stereo("addps %%xmm1, %%xmm0 \n")\
1998 "addps %%xmm1, %%xmm2 \n"\
1999 "addps %%xmm3, %%xmm0 \n"\
2000 "addps %%xmm4, %%xmm2 \n"\
2001 mono("addps %%xmm2, %%xmm0 \n")\
2002 "movaps %%xmm0, (%0,%1) \n"\
2003 stereo("movaps %%xmm2, 0x400(%0,%1) \n")\
2007 :"r"(samples[0]+len), "r"(matrix)\
2008 :XMM_CLOBBERS("%xmm0", "%xmm1", "%xmm2", "%xmm3", \
2009 "%xmm4", "%xmm5", "%xmm6", "%xmm7",)\
2013 #define MIX_MISC(stereo)\
2016 "movaps (%3,%0), %%xmm0 \n"\
2017 stereo("movaps %%xmm0, %%xmm1 \n")\
2018 "mulps %%xmm4, %%xmm0 \n"\
2019 stereo("mulps %%xmm5, %%xmm1 \n")\
2020 "lea 1024(%3,%0), %1 \n"\
2023 "movaps (%1), %%xmm2 \n"\
2024 stereo("movaps %%xmm2, %%xmm3 \n")\
2025 "mulps (%4,%2), %%xmm2 \n"\
2026 stereo("mulps 16(%4,%2), %%xmm3 \n")\
2027 "addps %%xmm2, %%xmm0 \n"\
2028 stereo("addps %%xmm3, %%xmm1 \n")\
2032 "movaps %%xmm0, (%3,%0) \n"\
2033 stereo("movaps %%xmm1, 1024(%3,%0) \n")\
2036 :"+&r"(i), "=&r"(j), "=&r"(k)\
2037 :"r"(samples[0]+len), "r"(matrix_simd+in_ch), "g"((intptr_t)-32*(in_ch-1))\
2041 static void ac3_downmix_sse(float (*samples)[256], float (*matrix)[2], int out_ch, int in_ch, int len)
2043 int (*matrix_cmp)[2] = (int(*)[2])matrix;
2046 i = -len*sizeof(float);
2047 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]))) {
2049 } else if(in_ch == 5 && out_ch == 1 && matrix_cmp[0][0]==matrix_cmp[2][0] && matrix_cmp[3][0]==matrix_cmp[4][0]) {
2052 DECLARE_ALIGNED(16, float, matrix_simd)[in_ch][2][4];
2053 j = 2*in_ch*sizeof(float);
2057 "movss (%2,%0), %%xmm4 \n"
2058 "movss 4(%2,%0), %%xmm5 \n"
2059 "shufps $0, %%xmm4, %%xmm4 \n"
2060 "shufps $0, %%xmm5, %%xmm5 \n"
2061 "movaps %%xmm4, (%1,%0,4) \n"
2062 "movaps %%xmm5, 16(%1,%0,4) \n"
2065 :"r"(matrix_simd), "r"(matrix)
2076 static void vector_fmul_3dnow(float *dst, const float *src, int len){
2077 x86_reg i = (len-4)*4;
2080 "movq (%1,%0), %%mm0 \n\t"
2081 "movq 8(%1,%0), %%mm1 \n\t"
2082 "pfmul (%2,%0), %%mm0 \n\t"
2083 "pfmul 8(%2,%0), %%mm1 \n\t"
2084 "movq %%mm0, (%1,%0) \n\t"
2085 "movq %%mm1, 8(%1,%0) \n\t"
2094 static void vector_fmul_sse(float *dst, const float *src, int len){
2095 x86_reg i = (len-8)*4;
2098 "movaps (%1,%0), %%xmm0 \n\t"
2099 "movaps 16(%1,%0), %%xmm1 \n\t"
2100 "mulps (%2,%0), %%xmm0 \n\t"
2101 "mulps 16(%2,%0), %%xmm1 \n\t"
2102 "movaps %%xmm0, (%1,%0) \n\t"
2103 "movaps %%xmm1, 16(%1,%0) \n\t"
2112 static void vector_fmul_reverse_3dnow2(float *dst, const float *src0, const float *src1, int len){
2113 x86_reg i = len*4-16;
2116 "pswapd 8(%1), %%mm0 \n\t"
2117 "pswapd (%1), %%mm1 \n\t"
2118 "pfmul (%3,%0), %%mm0 \n\t"
2119 "pfmul 8(%3,%0), %%mm1 \n\t"
2120 "movq %%mm0, (%2,%0) \n\t"
2121 "movq %%mm1, 8(%2,%0) \n\t"
2125 :"+r"(i), "+r"(src1)
2126 :"r"(dst), "r"(src0)
2128 __asm__ volatile("femms");
2130 static void vector_fmul_reverse_sse(float *dst, const float *src0, const float *src1, int len){
2131 x86_reg i = len*4-32;
2134 "movaps 16(%1), %%xmm0 \n\t"
2135 "movaps (%1), %%xmm1 \n\t"
2136 "shufps $0x1b, %%xmm0, %%xmm0 \n\t"
2137 "shufps $0x1b, %%xmm1, %%xmm1 \n\t"
2138 "mulps (%3,%0), %%xmm0 \n\t"
2139 "mulps 16(%3,%0), %%xmm1 \n\t"
2140 "movaps %%xmm0, (%2,%0) \n\t"
2141 "movaps %%xmm1, 16(%2,%0) \n\t"
2145 :"+r"(i), "+r"(src1)
2146 :"r"(dst), "r"(src0)
2150 static void vector_fmul_add_3dnow(float *dst, const float *src0, const float *src1,
2151 const float *src2, int len){
2152 x86_reg i = (len-4)*4;
2155 "movq (%2,%0), %%mm0 \n\t"
2156 "movq 8(%2,%0), %%mm1 \n\t"
2157 "pfmul (%3,%0), %%mm0 \n\t"
2158 "pfmul 8(%3,%0), %%mm1 \n\t"
2159 "pfadd (%4,%0), %%mm0 \n\t"
2160 "pfadd 8(%4,%0), %%mm1 \n\t"
2161 "movq %%mm0, (%1,%0) \n\t"
2162 "movq %%mm1, 8(%1,%0) \n\t"
2166 :"r"(dst), "r"(src0), "r"(src1), "r"(src2)
2169 __asm__ volatile("femms");
2171 static void vector_fmul_add_sse(float *dst, const float *src0, const float *src1,
2172 const float *src2, int len){
2173 x86_reg i = (len-8)*4;
2176 "movaps (%2,%0), %%xmm0 \n\t"
2177 "movaps 16(%2,%0), %%xmm1 \n\t"
2178 "mulps (%3,%0), %%xmm0 \n\t"
2179 "mulps 16(%3,%0), %%xmm1 \n\t"
2180 "addps (%4,%0), %%xmm0 \n\t"
2181 "addps 16(%4,%0), %%xmm1 \n\t"
2182 "movaps %%xmm0, (%1,%0) \n\t"
2183 "movaps %%xmm1, 16(%1,%0) \n\t"
2187 :"r"(dst), "r"(src0), "r"(src1), "r"(src2)
2192 static void vector_fmul_window_3dnow2(float *dst, const float *src0, const float *src1,
2193 const float *win, float add_bias, int len){
2197 x86_reg j = len*4-8;
2200 "pswapd (%5,%1), %%mm1 \n"
2201 "movq (%5,%0), %%mm0 \n"
2202 "pswapd (%4,%1), %%mm5 \n"
2203 "movq (%3,%0), %%mm4 \n"
2204 "movq %%mm0, %%mm2 \n"
2205 "movq %%mm1, %%mm3 \n"
2206 "pfmul %%mm4, %%mm2 \n" // src0[len+i]*win[len+i]
2207 "pfmul %%mm5, %%mm3 \n" // src1[ j]*win[len+j]
2208 "pfmul %%mm4, %%mm1 \n" // src0[len+i]*win[len+j]
2209 "pfmul %%mm5, %%mm0 \n" // src1[ j]*win[len+i]
2210 "pfadd %%mm3, %%mm2 \n"
2211 "pfsub %%mm0, %%mm1 \n"
2212 "pswapd %%mm2, %%mm2 \n"
2213 "movq %%mm1, (%2,%0) \n"
2214 "movq %%mm2, (%2,%1) \n"
2220 :"r"(dst+len), "r"(src0+len), "r"(src1), "r"(win+len)
2224 ff_vector_fmul_window_c(dst, src0, src1, win, add_bias, len);
2227 static void vector_fmul_window_sse(float *dst, const float *src0, const float *src1,
2228 const float *win, float add_bias, int len){
2232 x86_reg j = len*4-16;
2235 "movaps (%5,%1), %%xmm1 \n"
2236 "movaps (%5,%0), %%xmm0 \n"
2237 "movaps (%4,%1), %%xmm5 \n"
2238 "movaps (%3,%0), %%xmm4 \n"
2239 "shufps $0x1b, %%xmm1, %%xmm1 \n"
2240 "shufps $0x1b, %%xmm5, %%xmm5 \n"
2241 "movaps %%xmm0, %%xmm2 \n"
2242 "movaps %%xmm1, %%xmm3 \n"
2243 "mulps %%xmm4, %%xmm2 \n" // src0[len+i]*win[len+i]
2244 "mulps %%xmm5, %%xmm3 \n" // src1[ j]*win[len+j]
2245 "mulps %%xmm4, %%xmm1 \n" // src0[len+i]*win[len+j]
2246 "mulps %%xmm5, %%xmm0 \n" // src1[ j]*win[len+i]
2247 "addps %%xmm3, %%xmm2 \n"
2248 "subps %%xmm0, %%xmm1 \n"
2249 "shufps $0x1b, %%xmm2, %%xmm2 \n"
2250 "movaps %%xmm1, (%2,%0) \n"
2251 "movaps %%xmm2, (%2,%1) \n"
2256 :"r"(dst+len), "r"(src0+len), "r"(src1), "r"(win+len)
2260 ff_vector_fmul_window_c(dst, src0, src1, win, add_bias, len);
2263 static void int32_to_float_fmul_scalar_sse(float *dst, const int *src, float mul, int len)
2267 "movss %3, %%xmm4 \n"
2268 "shufps $0, %%xmm4, %%xmm4 \n"
2270 "cvtpi2ps (%2,%0), %%xmm0 \n"
2271 "cvtpi2ps 8(%2,%0), %%xmm1 \n"
2272 "cvtpi2ps 16(%2,%0), %%xmm2 \n"
2273 "cvtpi2ps 24(%2,%0), %%xmm3 \n"
2274 "movlhps %%xmm1, %%xmm0 \n"
2275 "movlhps %%xmm3, %%xmm2 \n"
2276 "mulps %%xmm4, %%xmm0 \n"
2277 "mulps %%xmm4, %%xmm2 \n"
2278 "movaps %%xmm0, (%1,%0) \n"
2279 "movaps %%xmm2, 16(%1,%0) \n"
2283 :"r"(dst+len), "r"(src+len), "m"(mul)
2287 static void int32_to_float_fmul_scalar_sse2(float *dst, const int *src, float mul, int len)
2291 "movss %3, %%xmm4 \n"
2292 "shufps $0, %%xmm4, %%xmm4 \n"
2294 "cvtdq2ps (%2,%0), %%xmm0 \n"
2295 "cvtdq2ps 16(%2,%0), %%xmm1 \n"
2296 "mulps %%xmm4, %%xmm0 \n"
2297 "mulps %%xmm4, %%xmm1 \n"
2298 "movaps %%xmm0, (%1,%0) \n"
2299 "movaps %%xmm1, 16(%1,%0) \n"
2303 :"r"(dst+len), "r"(src+len), "m"(mul)
2307 static void vector_clipf_sse(float *dst, const float *src, float min, float max,
2310 x86_reg i = (len-16)*4;
2312 "movss %3, %%xmm4 \n"
2313 "movss %4, %%xmm5 \n"
2314 "shufps $0, %%xmm4, %%xmm4 \n"
2315 "shufps $0, %%xmm5, %%xmm5 \n"
2317 "movaps (%2,%0), %%xmm0 \n\t" // 3/1 on intel
2318 "movaps 16(%2,%0), %%xmm1 \n\t"
2319 "movaps 32(%2,%0), %%xmm2 \n\t"
2320 "movaps 48(%2,%0), %%xmm3 \n\t"
2321 "maxps %%xmm4, %%xmm0 \n\t"
2322 "maxps %%xmm4, %%xmm1 \n\t"
2323 "maxps %%xmm4, %%xmm2 \n\t"
2324 "maxps %%xmm4, %%xmm3 \n\t"
2325 "minps %%xmm5, %%xmm0 \n\t"
2326 "minps %%xmm5, %%xmm1 \n\t"
2327 "minps %%xmm5, %%xmm2 \n\t"
2328 "minps %%xmm5, %%xmm3 \n\t"
2329 "movaps %%xmm0, (%1,%0) \n\t"
2330 "movaps %%xmm1, 16(%1,%0) \n\t"
2331 "movaps %%xmm2, 32(%1,%0) \n\t"
2332 "movaps %%xmm3, 48(%1,%0) \n\t"
2336 :"r"(dst), "r"(src), "m"(min), "m"(max)
2341 static void float_to_int16_3dnow(int16_t *dst, const float *src, long len){
2342 x86_reg reglen = len;
2343 // not bit-exact: pf2id uses different rounding than C and SSE
2346 "lea (%2,%0,2) , %2 \n\t"
2350 "pf2id (%2,%0,2) , %%mm0 \n\t"
2351 "pf2id 8(%2,%0,2) , %%mm1 \n\t"
2352 "pf2id 16(%2,%0,2) , %%mm2 \n\t"
2353 "pf2id 24(%2,%0,2) , %%mm3 \n\t"
2354 "packssdw %%mm1 , %%mm0 \n\t"
2355 "packssdw %%mm3 , %%mm2 \n\t"
2356 "movq %%mm0 , (%1,%0) \n\t"
2357 "movq %%mm2 , 8(%1,%0) \n\t"
2361 :"+r"(reglen), "+r"(dst), "+r"(src)
2364 static void float_to_int16_sse(int16_t *dst, const float *src, long len){
2365 x86_reg reglen = len;
2368 "lea (%2,%0,2) , %2 \n\t"
2372 "cvtps2pi (%2,%0,2) , %%mm0 \n\t"
2373 "cvtps2pi 8(%2,%0,2) , %%mm1 \n\t"
2374 "cvtps2pi 16(%2,%0,2) , %%mm2 \n\t"
2375 "cvtps2pi 24(%2,%0,2) , %%mm3 \n\t"
2376 "packssdw %%mm1 , %%mm0 \n\t"
2377 "packssdw %%mm3 , %%mm2 \n\t"
2378 "movq %%mm0 , (%1,%0) \n\t"
2379 "movq %%mm2 , 8(%1,%0) \n\t"
2383 :"+r"(reglen), "+r"(dst), "+r"(src)
2387 static void float_to_int16_sse2(int16_t *dst, const float *src, long len){
2388 x86_reg reglen = len;
2391 "lea (%2,%0,2) , %2 \n\t"
2395 "cvtps2dq (%2,%0,2) , %%xmm0 \n\t"
2396 "cvtps2dq 16(%2,%0,2) , %%xmm1 \n\t"
2397 "packssdw %%xmm1 , %%xmm0 \n\t"
2398 "movdqa %%xmm0 , (%1,%0) \n\t"
2401 :"+r"(reglen), "+r"(dst), "+r"(src)
2405 void ff_vp3_idct_mmx(int16_t *input_data);
2406 void ff_vp3_idct_put_mmx(uint8_t *dest, int line_size, DCTELEM *block);
2407 void ff_vp3_idct_add_mmx(uint8_t *dest, int line_size, DCTELEM *block);
2409 void ff_vp3_idct_dc_add_mmx2(uint8_t *dest, int line_size, const DCTELEM *block);
2411 void ff_vp3_v_loop_filter_mmx2(uint8_t *src, int stride, int *bounding_values);
2412 void ff_vp3_h_loop_filter_mmx2(uint8_t *src, int stride, int *bounding_values);
2414 void ff_vp3_idct_sse2(int16_t *input_data);
2415 void ff_vp3_idct_put_sse2(uint8_t *dest, int line_size, DCTELEM *block);
2416 void ff_vp3_idct_add_sse2(uint8_t *dest, int line_size, DCTELEM *block);
2418 void ff_float_to_int16_interleave6_sse(int16_t *dst, const float **src, int len);
2419 void ff_float_to_int16_interleave6_3dnow(int16_t *dst, const float **src, int len);
2420 void ff_float_to_int16_interleave6_3dn2(int16_t *dst, const float **src, int len);
2421 int32_t ff_scalarproduct_int16_mmx2(const int16_t *v1, const int16_t *v2, int order, int shift);
2422 int32_t ff_scalarproduct_int16_sse2(const int16_t *v1, const int16_t *v2, int order, int shift);
2423 int32_t ff_scalarproduct_and_madd_int16_mmx2(int16_t *v1, const int16_t *v2, const int16_t *v3, int order, int mul);
2424 int32_t ff_scalarproduct_and_madd_int16_sse2(int16_t *v1, const int16_t *v2, const int16_t *v3, int order, int mul);
2425 int32_t ff_scalarproduct_and_madd_int16_ssse3(int16_t *v1, const int16_t *v2, const int16_t *v3, int order, int mul);
2426 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);
2427 int ff_add_hfyu_left_prediction_ssse3(uint8_t *dst, const uint8_t *src, int w, int left);
2428 int ff_add_hfyu_left_prediction_sse4(uint8_t *dst, const uint8_t *src, int w, int left);
2431 #define ff_float_to_int16_interleave6_sse(a,b,c) float_to_int16_interleave_misc_sse(a,b,c,6)
2432 #define ff_float_to_int16_interleave6_3dnow(a,b,c) float_to_int16_interleave_misc_3dnow(a,b,c,6)
2433 #define ff_float_to_int16_interleave6_3dn2(a,b,c) float_to_int16_interleave_misc_3dnow(a,b,c,6)
2435 #define ff_float_to_int16_interleave6_sse2 ff_float_to_int16_interleave6_sse
2437 #define FLOAT_TO_INT16_INTERLEAVE(cpu, body) \
2438 /* gcc pessimizes register allocation if this is in the same function as float_to_int16_interleave_sse2*/\
2439 static av_noinline void float_to_int16_interleave_misc_##cpu(int16_t *dst, const float **src, long len, int channels){\
2440 DECLARE_ALIGNED(16, int16_t, tmp)[len];\
2442 for(c=0; c<channels; c++){\
2443 float_to_int16_##cpu(tmp, src[c], len);\
2444 for(i=0, j=c; i<len; i++, j+=channels)\
2449 static void float_to_int16_interleave_##cpu(int16_t *dst, const float **src, long len, int channels){\
2451 float_to_int16_##cpu(dst, src[0], len);\
2452 else if(channels==2){\
2453 x86_reg reglen = len; \
2454 const float *src0 = src[0];\
2455 const float *src1 = src[1];\
2463 :"+r"(reglen), "+r"(dst), "+r"(src0), "+r"(src1)\
2465 }else if(channels==6){\
2466 ff_float_to_int16_interleave6_##cpu(dst, src, len);\
2468 float_to_int16_interleave_misc_##cpu(dst, src, len, channels);\
2471 FLOAT_TO_INT16_INTERLEAVE(3dnow,
2473 "pf2id (%2,%0), %%mm0 \n"
2474 "pf2id 8(%2,%0), %%mm1 \n"
2475 "pf2id (%3,%0), %%mm2 \n"
2476 "pf2id 8(%3,%0), %%mm3 \n"
2477 "packssdw %%mm1, %%mm0 \n"
2478 "packssdw %%mm3, %%mm2 \n"
2479 "movq %%mm0, %%mm1 \n"
2480 "punpcklwd %%mm2, %%mm0 \n"
2481 "punpckhwd %%mm2, %%mm1 \n"
2482 "movq %%mm0, (%1,%0)\n"
2483 "movq %%mm1, 8(%1,%0)\n"
2489 FLOAT_TO_INT16_INTERLEAVE(sse,
2491 "cvtps2pi (%2,%0), %%mm0 \n"
2492 "cvtps2pi 8(%2,%0), %%mm1 \n"
2493 "cvtps2pi (%3,%0), %%mm2 \n"
2494 "cvtps2pi 8(%3,%0), %%mm3 \n"
2495 "packssdw %%mm1, %%mm0 \n"
2496 "packssdw %%mm3, %%mm2 \n"
2497 "movq %%mm0, %%mm1 \n"
2498 "punpcklwd %%mm2, %%mm0 \n"
2499 "punpckhwd %%mm2, %%mm1 \n"
2500 "movq %%mm0, (%1,%0)\n"
2501 "movq %%mm1, 8(%1,%0)\n"
2507 FLOAT_TO_INT16_INTERLEAVE(sse2,
2509 "cvtps2dq (%2,%0), %%xmm0 \n"
2510 "cvtps2dq (%3,%0), %%xmm1 \n"
2511 "packssdw %%xmm1, %%xmm0 \n"
2512 "movhlps %%xmm0, %%xmm1 \n"
2513 "punpcklwd %%xmm1, %%xmm0 \n"
2514 "movdqa %%xmm0, (%1,%0) \n"
2519 static void float_to_int16_interleave_3dn2(int16_t *dst, const float **src, long len, int channels){
2521 ff_float_to_int16_interleave6_3dn2(dst, src, len);
2523 float_to_int16_interleave_3dnow(dst, src, len, channels);
2526 float ff_scalarproduct_float_sse(const float *v1, const float *v2, int order);
2528 void dsputil_init_mmx(DSPContext* c, AVCodecContext *avctx)
2530 int mm_flags = av_get_cpu_flags();
2532 if (avctx->dsp_mask) {
2533 if (avctx->dsp_mask & AV_CPU_FLAG_FORCE)
2534 mm_flags |= (avctx->dsp_mask & 0xffff);
2536 mm_flags &= ~(avctx->dsp_mask & 0xffff);
2540 av_log(avctx, AV_LOG_INFO, "libavcodec: CPU flags:");
2541 if (mm_flags & AV_CPU_FLAG_MMX)
2542 av_log(avctx, AV_LOG_INFO, " mmx");
2543 if (mm_flags & AV_CPU_FLAG_MMX2)
2544 av_log(avctx, AV_LOG_INFO, " mmx2");
2545 if (mm_flags & AV_CPU_FLAG_3DNOW)
2546 av_log(avctx, AV_LOG_INFO, " 3dnow");
2547 if (mm_flags & AV_CPU_FLAG_SSE)
2548 av_log(avctx, AV_LOG_INFO, " sse");
2549 if (mm_flags & AV_CPU_FLAG_SSE2)
2550 av_log(avctx, AV_LOG_INFO, " sse2");
2551 av_log(avctx, AV_LOG_INFO, "\n");
2554 if (mm_flags & AV_CPU_FLAG_MMX) {
2555 const int idct_algo= avctx->idct_algo;
2557 if(avctx->lowres==0){
2558 if(idct_algo==FF_IDCT_AUTO || idct_algo==FF_IDCT_SIMPLEMMX){
2559 c->idct_put= ff_simple_idct_put_mmx;
2560 c->idct_add= ff_simple_idct_add_mmx;
2561 c->idct = ff_simple_idct_mmx;
2562 c->idct_permutation_type= FF_SIMPLE_IDCT_PERM;
2564 }else if(idct_algo==FF_IDCT_LIBMPEG2MMX){
2565 if(mm_flags & AV_CPU_FLAG_MMX2){
2566 c->idct_put= ff_libmpeg2mmx2_idct_put;
2567 c->idct_add= ff_libmpeg2mmx2_idct_add;
2568 c->idct = ff_mmxext_idct;
2570 c->idct_put= ff_libmpeg2mmx_idct_put;
2571 c->idct_add= ff_libmpeg2mmx_idct_add;
2572 c->idct = ff_mmx_idct;
2574 c->idct_permutation_type= FF_LIBMPEG2_IDCT_PERM;
2576 }else if((CONFIG_VP3_DECODER || CONFIG_VP5_DECODER || CONFIG_VP6_DECODER) &&
2577 idct_algo==FF_IDCT_VP3 && HAVE_YASM){
2578 if(mm_flags & AV_CPU_FLAG_SSE2){
2579 c->idct_put= ff_vp3_idct_put_sse2;
2580 c->idct_add= ff_vp3_idct_add_sse2;
2581 c->idct = ff_vp3_idct_sse2;
2582 c->idct_permutation_type= FF_TRANSPOSE_IDCT_PERM;
2584 c->idct_put= ff_vp3_idct_put_mmx;
2585 c->idct_add= ff_vp3_idct_add_mmx;
2586 c->idct = ff_vp3_idct_mmx;
2587 c->idct_permutation_type= FF_PARTTRANS_IDCT_PERM;
2589 }else if(idct_algo==FF_IDCT_CAVS){
2590 c->idct_permutation_type= FF_TRANSPOSE_IDCT_PERM;
2591 }else if(idct_algo==FF_IDCT_XVIDMMX){
2592 if(mm_flags & AV_CPU_FLAG_SSE2){
2593 c->idct_put= ff_idct_xvid_sse2_put;
2594 c->idct_add= ff_idct_xvid_sse2_add;
2595 c->idct = ff_idct_xvid_sse2;
2596 c->idct_permutation_type= FF_SSE2_IDCT_PERM;
2597 }else if(mm_flags & AV_CPU_FLAG_MMX2){
2598 c->idct_put= ff_idct_xvid_mmx2_put;
2599 c->idct_add= ff_idct_xvid_mmx2_add;
2600 c->idct = ff_idct_xvid_mmx2;
2602 c->idct_put= ff_idct_xvid_mmx_put;
2603 c->idct_add= ff_idct_xvid_mmx_add;
2604 c->idct = ff_idct_xvid_mmx;
2609 c->put_pixels_clamped = ff_put_pixels_clamped_mmx;
2610 c->put_signed_pixels_clamped = ff_put_signed_pixels_clamped_mmx;
2611 c->add_pixels_clamped = ff_add_pixels_clamped_mmx;
2612 c->clear_block = clear_block_mmx;
2613 c->clear_blocks = clear_blocks_mmx;
2614 if ((mm_flags & AV_CPU_FLAG_SSE) &&
2615 !(CONFIG_MPEG_XVMC_DECODER && avctx->xvmc_acceleration > 1)){
2616 /* XvMCCreateBlocks() may not allocate 16-byte aligned blocks */
2617 c->clear_block = clear_block_sse;
2618 c->clear_blocks = clear_blocks_sse;
2621 #define SET_HPEL_FUNCS(PFX, IDX, SIZE, CPU) \
2622 c->PFX ## _pixels_tab[IDX][0] = PFX ## _pixels ## SIZE ## _ ## CPU; \
2623 c->PFX ## _pixels_tab[IDX][1] = PFX ## _pixels ## SIZE ## _x2_ ## CPU; \
2624 c->PFX ## _pixels_tab[IDX][2] = PFX ## _pixels ## SIZE ## _y2_ ## CPU; \
2625 c->PFX ## _pixels_tab[IDX][3] = PFX ## _pixels ## SIZE ## _xy2_ ## CPU
2627 SET_HPEL_FUNCS(put, 0, 16, mmx);
2628 SET_HPEL_FUNCS(put_no_rnd, 0, 16, mmx);
2629 SET_HPEL_FUNCS(avg, 0, 16, mmx);
2630 SET_HPEL_FUNCS(avg_no_rnd, 0, 16, mmx);
2631 SET_HPEL_FUNCS(put, 1, 8, mmx);
2632 SET_HPEL_FUNCS(put_no_rnd, 1, 8, mmx);
2633 SET_HPEL_FUNCS(avg, 1, 8, mmx);
2634 SET_HPEL_FUNCS(avg_no_rnd, 1, 8, mmx);
2638 c->add_bytes= add_bytes_mmx;
2639 c->add_bytes_l2= add_bytes_l2_mmx;
2641 c->draw_edges = draw_edges_mmx;
2643 if (CONFIG_H263_DECODER || CONFIG_H263_ENCODER) {
2644 c->h263_v_loop_filter= h263_v_loop_filter_mmx;
2645 c->h263_h_loop_filter= h263_h_loop_filter_mmx;
2649 c->put_h264_chroma_pixels_tab[0]= ff_put_h264_chroma_mc8_mmx_rnd;
2650 c->put_h264_chroma_pixels_tab[1]= ff_put_h264_chroma_mc4_mmx;
2651 c->put_no_rnd_vc1_chroma_pixels_tab[0]= ff_put_vc1_chroma_mc8_mmx_nornd;
2653 c->put_rv40_chroma_pixels_tab[0]= ff_put_rv40_chroma_mc8_mmx;
2654 c->put_rv40_chroma_pixels_tab[1]= ff_put_rv40_chroma_mc4_mmx;
2657 if (mm_flags & AV_CPU_FLAG_MMX2) {
2658 c->prefetch = prefetch_mmx2;
2660 c->put_pixels_tab[0][1] = put_pixels16_x2_mmx2;
2661 c->put_pixels_tab[0][2] = put_pixels16_y2_mmx2;
2663 c->avg_pixels_tab[0][0] = avg_pixels16_mmx2;
2664 c->avg_pixels_tab[0][1] = avg_pixels16_x2_mmx2;
2665 c->avg_pixels_tab[0][2] = avg_pixels16_y2_mmx2;
2667 c->put_pixels_tab[1][1] = put_pixels8_x2_mmx2;
2668 c->put_pixels_tab[1][2] = put_pixels8_y2_mmx2;
2670 c->avg_pixels_tab[1][0] = avg_pixels8_mmx2;
2671 c->avg_pixels_tab[1][1] = avg_pixels8_x2_mmx2;
2672 c->avg_pixels_tab[1][2] = avg_pixels8_y2_mmx2;
2674 if(!(avctx->flags & CODEC_FLAG_BITEXACT)){
2675 c->put_no_rnd_pixels_tab[0][1] = put_no_rnd_pixels16_x2_mmx2;
2676 c->put_no_rnd_pixels_tab[0][2] = put_no_rnd_pixels16_y2_mmx2;
2677 c->put_no_rnd_pixels_tab[1][1] = put_no_rnd_pixels8_x2_mmx2;
2678 c->put_no_rnd_pixels_tab[1][2] = put_no_rnd_pixels8_y2_mmx2;
2679 c->avg_pixels_tab[0][3] = avg_pixels16_xy2_mmx2;
2680 c->avg_pixels_tab[1][3] = avg_pixels8_xy2_mmx2;
2682 if (CONFIG_VP3_DECODER && HAVE_YASM) {
2683 c->vp3_v_loop_filter= ff_vp3_v_loop_filter_mmx2;
2684 c->vp3_h_loop_filter= ff_vp3_h_loop_filter_mmx2;
2687 if (CONFIG_VP3_DECODER && HAVE_YASM) {
2688 c->vp3_idct_dc_add = ff_vp3_idct_dc_add_mmx2;
2691 if (CONFIG_VP3_DECODER
2692 && (avctx->codec_id == CODEC_ID_VP3 || avctx->codec_id == CODEC_ID_THEORA)) {
2693 c->put_no_rnd_pixels_tab[1][1] = put_no_rnd_pixels8_x2_exact_mmx2;
2694 c->put_no_rnd_pixels_tab[1][2] = put_no_rnd_pixels8_y2_exact_mmx2;
2697 #define SET_QPEL_FUNCS(PFX, IDX, SIZE, CPU) \
2698 c->PFX ## _pixels_tab[IDX][ 0] = PFX ## SIZE ## _mc00_ ## CPU; \
2699 c->PFX ## _pixels_tab[IDX][ 1] = PFX ## SIZE ## _mc10_ ## CPU; \
2700 c->PFX ## _pixels_tab[IDX][ 2] = PFX ## SIZE ## _mc20_ ## CPU; \
2701 c->PFX ## _pixels_tab[IDX][ 3] = PFX ## SIZE ## _mc30_ ## CPU; \
2702 c->PFX ## _pixels_tab[IDX][ 4] = PFX ## SIZE ## _mc01_ ## CPU; \
2703 c->PFX ## _pixels_tab[IDX][ 5] = PFX ## SIZE ## _mc11_ ## CPU; \
2704 c->PFX ## _pixels_tab[IDX][ 6] = PFX ## SIZE ## _mc21_ ## CPU; \
2705 c->PFX ## _pixels_tab[IDX][ 7] = PFX ## SIZE ## _mc31_ ## CPU; \
2706 c->PFX ## _pixels_tab[IDX][ 8] = PFX ## SIZE ## _mc02_ ## CPU; \
2707 c->PFX ## _pixels_tab[IDX][ 9] = PFX ## SIZE ## _mc12_ ## CPU; \
2708 c->PFX ## _pixels_tab[IDX][10] = PFX ## SIZE ## _mc22_ ## CPU; \
2709 c->PFX ## _pixels_tab[IDX][11] = PFX ## SIZE ## _mc32_ ## CPU; \
2710 c->PFX ## _pixels_tab[IDX][12] = PFX ## SIZE ## _mc03_ ## CPU; \
2711 c->PFX ## _pixels_tab[IDX][13] = PFX ## SIZE ## _mc13_ ## CPU; \
2712 c->PFX ## _pixels_tab[IDX][14] = PFX ## SIZE ## _mc23_ ## CPU; \
2713 c->PFX ## _pixels_tab[IDX][15] = PFX ## SIZE ## _mc33_ ## CPU
2715 SET_QPEL_FUNCS(put_qpel, 0, 16, mmx2);
2716 SET_QPEL_FUNCS(put_qpel, 1, 8, mmx2);
2717 SET_QPEL_FUNCS(put_no_rnd_qpel, 0, 16, mmx2);
2718 SET_QPEL_FUNCS(put_no_rnd_qpel, 1, 8, mmx2);
2719 SET_QPEL_FUNCS(avg_qpel, 0, 16, mmx2);
2720 SET_QPEL_FUNCS(avg_qpel, 1, 8, mmx2);
2722 SET_QPEL_FUNCS(put_h264_qpel, 0, 16, mmx2);
2723 SET_QPEL_FUNCS(put_h264_qpel, 1, 8, mmx2);
2724 SET_QPEL_FUNCS(put_h264_qpel, 2, 4, mmx2);
2725 SET_QPEL_FUNCS(avg_h264_qpel, 0, 16, mmx2);
2726 SET_QPEL_FUNCS(avg_h264_qpel, 1, 8, mmx2);
2727 SET_QPEL_FUNCS(avg_h264_qpel, 2, 4, mmx2);
2729 SET_QPEL_FUNCS(put_2tap_qpel, 0, 16, mmx2);
2730 SET_QPEL_FUNCS(put_2tap_qpel, 1, 8, mmx2);
2731 SET_QPEL_FUNCS(avg_2tap_qpel, 0, 16, mmx2);
2732 SET_QPEL_FUNCS(avg_2tap_qpel, 1, 8, mmx2);
2735 c->avg_rv40_chroma_pixels_tab[0]= ff_avg_rv40_chroma_mc8_mmx2;
2736 c->avg_rv40_chroma_pixels_tab[1]= ff_avg_rv40_chroma_mc4_mmx2;
2738 c->avg_no_rnd_vc1_chroma_pixels_tab[0]= ff_avg_vc1_chroma_mc8_mmx2_nornd;
2740 c->avg_h264_chroma_pixels_tab[0]= ff_avg_h264_chroma_mc8_mmx2_rnd;
2741 c->avg_h264_chroma_pixels_tab[1]= ff_avg_h264_chroma_mc4_mmx2;
2742 c->avg_h264_chroma_pixels_tab[2]= ff_avg_h264_chroma_mc2_mmx2;
2743 c->put_h264_chroma_pixels_tab[2]= ff_put_h264_chroma_mc2_mmx2;
2745 c->add_hfyu_median_prediction = ff_add_hfyu_median_prediction_mmx2;
2747 #if HAVE_7REGS && HAVE_TEN_OPERANDS
2748 if( mm_flags&AV_CPU_FLAG_3DNOW )
2749 c->add_hfyu_median_prediction = add_hfyu_median_prediction_cmov;
2752 if (CONFIG_VC1_DECODER)
2753 ff_vc1dsp_init_mmx(c, avctx);
2755 c->add_png_paeth_prediction= add_png_paeth_prediction_mmx2;
2756 } else if (mm_flags & AV_CPU_FLAG_3DNOW) {
2757 c->prefetch = prefetch_3dnow;
2759 c->put_pixels_tab[0][1] = put_pixels16_x2_3dnow;
2760 c->put_pixels_tab[0][2] = put_pixels16_y2_3dnow;
2762 c->avg_pixels_tab[0][0] = avg_pixels16_3dnow;
2763 c->avg_pixels_tab[0][1] = avg_pixels16_x2_3dnow;
2764 c->avg_pixels_tab[0][2] = avg_pixels16_y2_3dnow;
2766 c->put_pixels_tab[1][1] = put_pixels8_x2_3dnow;
2767 c->put_pixels_tab[1][2] = put_pixels8_y2_3dnow;
2769 c->avg_pixels_tab[1][0] = avg_pixels8_3dnow;
2770 c->avg_pixels_tab[1][1] = avg_pixels8_x2_3dnow;
2771 c->avg_pixels_tab[1][2] = avg_pixels8_y2_3dnow;
2773 if(!(avctx->flags & CODEC_FLAG_BITEXACT)){
2774 c->put_no_rnd_pixels_tab[0][1] = put_no_rnd_pixels16_x2_3dnow;
2775 c->put_no_rnd_pixels_tab[0][2] = put_no_rnd_pixels16_y2_3dnow;
2776 c->put_no_rnd_pixels_tab[1][1] = put_no_rnd_pixels8_x2_3dnow;
2777 c->put_no_rnd_pixels_tab[1][2] = put_no_rnd_pixels8_y2_3dnow;
2778 c->avg_pixels_tab[0][3] = avg_pixels16_xy2_3dnow;
2779 c->avg_pixels_tab[1][3] = avg_pixels8_xy2_3dnow;
2782 if (CONFIG_VP3_DECODER
2783 && (avctx->codec_id == CODEC_ID_VP3 || avctx->codec_id == CODEC_ID_THEORA)) {
2784 c->put_no_rnd_pixels_tab[1][1] = put_no_rnd_pixels8_x2_exact_3dnow;
2785 c->put_no_rnd_pixels_tab[1][2] = put_no_rnd_pixels8_y2_exact_3dnow;
2788 SET_QPEL_FUNCS(put_qpel, 0, 16, 3dnow);
2789 SET_QPEL_FUNCS(put_qpel, 1, 8, 3dnow);
2790 SET_QPEL_FUNCS(put_no_rnd_qpel, 0, 16, 3dnow);
2791 SET_QPEL_FUNCS(put_no_rnd_qpel, 1, 8, 3dnow);
2792 SET_QPEL_FUNCS(avg_qpel, 0, 16, 3dnow);
2793 SET_QPEL_FUNCS(avg_qpel, 1, 8, 3dnow);
2795 SET_QPEL_FUNCS(put_h264_qpel, 0, 16, 3dnow);
2796 SET_QPEL_FUNCS(put_h264_qpel, 1, 8, 3dnow);
2797 SET_QPEL_FUNCS(put_h264_qpel, 2, 4, 3dnow);
2798 SET_QPEL_FUNCS(avg_h264_qpel, 0, 16, 3dnow);
2799 SET_QPEL_FUNCS(avg_h264_qpel, 1, 8, 3dnow);
2800 SET_QPEL_FUNCS(avg_h264_qpel, 2, 4, 3dnow);
2802 SET_QPEL_FUNCS(put_2tap_qpel, 0, 16, 3dnow);
2803 SET_QPEL_FUNCS(put_2tap_qpel, 1, 8, 3dnow);
2804 SET_QPEL_FUNCS(avg_2tap_qpel, 0, 16, 3dnow);
2805 SET_QPEL_FUNCS(avg_2tap_qpel, 1, 8, 3dnow);
2808 c->avg_h264_chroma_pixels_tab[0]= ff_avg_h264_chroma_mc8_3dnow_rnd;
2809 c->avg_h264_chroma_pixels_tab[1]= ff_avg_h264_chroma_mc4_3dnow;
2811 c->avg_no_rnd_vc1_chroma_pixels_tab[0]= ff_avg_vc1_chroma_mc8_3dnow_nornd;
2813 c->avg_rv40_chroma_pixels_tab[0]= ff_avg_rv40_chroma_mc8_3dnow;
2814 c->avg_rv40_chroma_pixels_tab[1]= ff_avg_rv40_chroma_mc4_3dnow;
2819 #define H264_QPEL_FUNCS(x, y, CPU)\
2820 c->put_h264_qpel_pixels_tab[0][x+y*4] = put_h264_qpel16_mc##x##y##_##CPU;\
2821 c->put_h264_qpel_pixels_tab[1][x+y*4] = put_h264_qpel8_mc##x##y##_##CPU;\
2822 c->avg_h264_qpel_pixels_tab[0][x+y*4] = avg_h264_qpel16_mc##x##y##_##CPU;\
2823 c->avg_h264_qpel_pixels_tab[1][x+y*4] = avg_h264_qpel8_mc##x##y##_##CPU;
2824 if((mm_flags & AV_CPU_FLAG_SSE2) && !(mm_flags & AV_CPU_FLAG_3DNOW)){
2825 // these functions are slower than mmx on AMD, but faster on Intel
2826 c->put_pixels_tab[0][0] = put_pixels16_sse2;
2827 c->put_no_rnd_pixels_tab[0][0] = put_pixels16_sse2;
2828 c->avg_pixels_tab[0][0] = avg_pixels16_sse2;
2829 H264_QPEL_FUNCS(0, 0, sse2);
2831 if(mm_flags & AV_CPU_FLAG_SSE2){
2832 H264_QPEL_FUNCS(0, 1, sse2);
2833 H264_QPEL_FUNCS(0, 2, sse2);
2834 H264_QPEL_FUNCS(0, 3, sse2);
2835 H264_QPEL_FUNCS(1, 1, sse2);
2836 H264_QPEL_FUNCS(1, 2, sse2);
2837 H264_QPEL_FUNCS(1, 3, sse2);
2838 H264_QPEL_FUNCS(2, 1, sse2);
2839 H264_QPEL_FUNCS(2, 2, sse2);
2840 H264_QPEL_FUNCS(2, 3, sse2);
2841 H264_QPEL_FUNCS(3, 1, sse2);
2842 H264_QPEL_FUNCS(3, 2, sse2);
2843 H264_QPEL_FUNCS(3, 3, sse2);
2846 if(mm_flags & AV_CPU_FLAG_SSSE3){
2847 H264_QPEL_FUNCS(1, 0, ssse3);
2848 H264_QPEL_FUNCS(1, 1, ssse3);
2849 H264_QPEL_FUNCS(1, 2, ssse3);
2850 H264_QPEL_FUNCS(1, 3, ssse3);
2851 H264_QPEL_FUNCS(2, 0, ssse3);
2852 H264_QPEL_FUNCS(2, 1, ssse3);
2853 H264_QPEL_FUNCS(2, 2, ssse3);
2854 H264_QPEL_FUNCS(2, 3, ssse3);
2855 H264_QPEL_FUNCS(3, 0, ssse3);
2856 H264_QPEL_FUNCS(3, 1, ssse3);
2857 H264_QPEL_FUNCS(3, 2, ssse3);
2858 H264_QPEL_FUNCS(3, 3, ssse3);
2859 c->add_png_paeth_prediction= add_png_paeth_prediction_ssse3;
2861 c->put_no_rnd_vc1_chroma_pixels_tab[0]= ff_put_vc1_chroma_mc8_ssse3_nornd;
2862 c->avg_no_rnd_vc1_chroma_pixels_tab[0]= ff_avg_vc1_chroma_mc8_ssse3_nornd;
2863 c->put_h264_chroma_pixels_tab[0]= ff_put_h264_chroma_mc8_ssse3_rnd;
2864 c->avg_h264_chroma_pixels_tab[0]= ff_avg_h264_chroma_mc8_ssse3_rnd;
2865 c->put_h264_chroma_pixels_tab[1]= ff_put_h264_chroma_mc4_ssse3;
2866 c->avg_h264_chroma_pixels_tab[1]= ff_avg_h264_chroma_mc4_ssse3;
2867 c->add_hfyu_left_prediction = ff_add_hfyu_left_prediction_ssse3;
2868 if (mm_flags & AV_CPU_FLAG_SSE4) // not really sse4, just slow on Conroe
2869 c->add_hfyu_left_prediction = ff_add_hfyu_left_prediction_sse4;
2874 if(mm_flags & AV_CPU_FLAG_3DNOW){
2875 c->vorbis_inverse_coupling = vorbis_inverse_coupling_3dnow;
2876 c->vector_fmul = vector_fmul_3dnow;
2877 if(!(avctx->flags & CODEC_FLAG_BITEXACT)){
2878 c->float_to_int16 = float_to_int16_3dnow;
2879 c->float_to_int16_interleave = float_to_int16_interleave_3dnow;
2882 if(mm_flags & AV_CPU_FLAG_3DNOWEXT){
2883 c->vector_fmul_reverse = vector_fmul_reverse_3dnow2;
2884 c->vector_fmul_window = vector_fmul_window_3dnow2;
2885 if(!(avctx->flags & CODEC_FLAG_BITEXACT)){
2886 c->float_to_int16_interleave = float_to_int16_interleave_3dn2;
2889 if(mm_flags & AV_CPU_FLAG_MMX2){
2891 c->scalarproduct_int16 = ff_scalarproduct_int16_mmx2;
2892 c->scalarproduct_and_madd_int16 = ff_scalarproduct_and_madd_int16_mmx2;
2895 if(mm_flags & AV_CPU_FLAG_SSE){
2896 c->vorbis_inverse_coupling = vorbis_inverse_coupling_sse;
2897 c->ac3_downmix = ac3_downmix_sse;
2898 c->vector_fmul = vector_fmul_sse;
2899 c->vector_fmul_reverse = vector_fmul_reverse_sse;
2900 c->vector_fmul_add = vector_fmul_add_sse;
2901 c->vector_fmul_window = vector_fmul_window_sse;
2902 c->int32_to_float_fmul_scalar = int32_to_float_fmul_scalar_sse;
2903 c->vector_clipf = vector_clipf_sse;
2904 c->float_to_int16 = float_to_int16_sse;
2905 c->float_to_int16_interleave = float_to_int16_interleave_sse;
2907 c->scalarproduct_float = ff_scalarproduct_float_sse;
2910 if(mm_flags & AV_CPU_FLAG_3DNOW)
2911 c->vector_fmul_add = vector_fmul_add_3dnow; // faster than sse
2912 if(mm_flags & AV_CPU_FLAG_SSE2){
2913 c->int32_to_float_fmul_scalar = int32_to_float_fmul_scalar_sse2;
2914 c->float_to_int16 = float_to_int16_sse2;
2915 c->float_to_int16_interleave = float_to_int16_interleave_sse2;
2917 c->scalarproduct_int16 = ff_scalarproduct_int16_sse2;
2918 c->scalarproduct_and_madd_int16 = ff_scalarproduct_and_madd_int16_sse2;
2921 if((mm_flags & AV_CPU_FLAG_SSSE3) && !(mm_flags & (AV_CPU_FLAG_SSE42|AV_CPU_FLAG_3DNOW)) && HAVE_YASM) // cachesplit
2922 c->scalarproduct_and_madd_int16 = ff_scalarproduct_and_madd_int16_ssse3;
2925 if (CONFIG_ENCODERS)
2926 dsputilenc_init_mmx(c, avctx);
2929 // for speed testing
2930 get_pixels = just_return;
2931 put_pixels_clamped = just_return;
2932 add_pixels_clamped = just_return;
2934 pix_abs16x16 = just_return;
2935 pix_abs16x16_x2 = just_return;
2936 pix_abs16x16_y2 = just_return;
2937 pix_abs16x16_xy2 = just_return;
2939 put_pixels_tab[0] = just_return;
2940 put_pixels_tab[1] = just_return;
2941 put_pixels_tab[2] = just_return;
2942 put_pixels_tab[3] = just_return;
2944 put_no_rnd_pixels_tab[0] = just_return;
2945 put_no_rnd_pixels_tab[1] = just_return;
2946 put_no_rnd_pixels_tab[2] = just_return;
2947 put_no_rnd_pixels_tab[3] = just_return;
2949 avg_pixels_tab[0] = just_return;
2950 avg_pixels_tab[1] = just_return;
2951 avg_pixels_tab[2] = just_return;
2952 avg_pixels_tab[3] = just_return;
2954 avg_no_rnd_pixels_tab[0] = just_return;
2955 avg_no_rnd_pixels_tab[1] = just_return;
2956 avg_no_rnd_pixels_tab[2] = just_return;
2957 avg_no_rnd_pixels_tab[3] = just_return;
2959 //av_fdct = just_return;
2960 //ff_idct = just_return;