VirtualBox

source: vbox/trunk/src/libs/ffmpeg-20060710/libavcodec/i386/dsputil_mmx.c@ 5776

Last change on this file since 5776 was 5776, checked in by vboxsync, 17 years ago

ffmpeg: exported to OSE
File size: 120.5 KB
Line 
1/*
2 * MMX optimized DSP utils
3 * Copyright (c) 2000, 2001 Fabrice Bellard.
4 * Copyright (c) 2002-2004 Michael Niedermayer <[email protected]>
5 *
6 * This library is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU Lesser General Public
8 * License as published by the Free Software Foundation; either
9 * version 2 of the License, or (at your option) any later version.
10 *
11 * This library is distributed in the hope that it will be useful,
12 * but WITHOUT ANY WARRANTY; without even the implied warranty of
13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
14 * Lesser General Public License for more details.
15 *
16 * You should have received a copy of the GNU Lesser General Public
17 * License along with this library; if not, write to the Free Software
18 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
19 *
20 * MMX optimization by Nick Kurshev <[email protected]>
21 */
22
23#include "../dsputil.h"
24#include "../simple_idct.h"
25#include "../mpegvideo.h"
26#include "x86_cpu.h"
27#include "mmx.h"
28
29//#undef NDEBUG
30//#include <assert.h>
31
32extern const uint8_t ff_h263_loop_filter_strength[32];
33extern void ff_idct_xvid_mmx(short *block);
34extern void ff_idct_xvid_mmx2(short *block);
35
36int mm_flags; /* multimedia extension flags */
37
38/* pixel operations */
39static const uint64_t mm_bone attribute_used __attribute__ ((aligned(8))) = 0x0101010101010101ULL;
40static const uint64_t mm_wone attribute_used __attribute__ ((aligned(8))) = 0x0001000100010001ULL;
41static const uint64_t mm_wtwo attribute_used __attribute__ ((aligned(8))) = 0x0002000200020002ULL;
42
43static const uint64_t ff_pw_20 attribute_used __attribute__ ((aligned(8))) = 0x0014001400140014ULL;
44static const uint64_t ff_pw_3 attribute_used __attribute__ ((aligned(8))) = 0x0003000300030003ULL;
45static const uint64_t ff_pw_4 attribute_used __attribute__ ((aligned(8))) = 0x0004000400040004ULL;
46static const uint64_t ff_pw_5 attribute_used __attribute__ ((aligned(8))) = 0x0005000500050005ULL;
47static const uint64_t ff_pw_8 attribute_used __attribute__ ((aligned(8))) = 0x0008000800080008ULL;
48static const uint64_t ff_pw_16 attribute_used __attribute__ ((aligned(8))) = 0x0010001000100010ULL;
49static const uint64_t ff_pw_32 attribute_used __attribute__ ((aligned(8))) = 0x0020002000200020ULL;
50static const uint64_t ff_pw_64 attribute_used __attribute__ ((aligned(8))) = 0x0040004000400040ULL;
51static const uint64_t ff_pw_15 attribute_used __attribute__ ((aligned(8))) = 0x000F000F000F000FULL;
52
53static const uint64_t ff_pb_3F attribute_used __attribute__ ((aligned(8))) = 0x3F3F3F3F3F3F3F3FULL;
54static const uint64_t ff_pb_FC attribute_used __attribute__ ((aligned(8))) = 0xFCFCFCFCFCFCFCFCULL;
55
56#if !defined(VBOX) || !defined(__DARWIN__)
57#define JUMPALIGN() __asm __volatile (".balign 8"::)
58#else
59#define JUMPALIGN() __asm __volatile (".align 3"::)
60#endif
61#define MOVQ_ZERO(regd) __asm __volatile ("pxor %%" #regd ", %%" #regd ::)
62
63#define MOVQ_WONE(regd) \
64 __asm __volatile ( \
65 "pcmpeqd %%" #regd ", %%" #regd " \n\t" \
66 "psrlw $15, %%" #regd ::)
67
68#define MOVQ_BFE(regd) \
69 __asm __volatile ( \
70 "pcmpeqd %%" #regd ", %%" #regd " \n\t"\
71 "paddb %%" #regd ", %%" #regd " \n\t" ::)
72
73#ifndef PIC
74#define MOVQ_BONE(regd) __asm __volatile ("movq %0, %%" #regd " \n\t" ::"m"(mm_bone))
75#define MOVQ_WTWO(regd) __asm __volatile ("movq %0, %%" #regd " \n\t" ::"m"(mm_wtwo))
76#else
77// for shared library it's better to use this way for accessing constants
78// pcmpeqd -> -1
79#define MOVQ_BONE(regd) \
80 __asm __volatile ( \
81 "pcmpeqd %%" #regd ", %%" #regd " \n\t" \
82 "psrlw $15, %%" #regd " \n\t" \
83 "packuswb %%" #regd ", %%" #regd " \n\t" ::)
84
85#define MOVQ_WTWO(regd) \
86 __asm __volatile ( \
87 "pcmpeqd %%" #regd ", %%" #regd " \n\t" \
88 "psrlw $15, %%" #regd " \n\t" \
89 "psllw $1, %%" #regd " \n\t"::)
90
91#endif
92
93// using regr as temporary and for the output result
94// first argument is unmodifed and second is trashed
95// regfe is supposed to contain 0xfefefefefefefefe
96#define PAVGB_MMX_NO_RND(rega, regb, regr, regfe) \
97 "movq " #rega ", " #regr " \n\t"\
98 "pand " #regb ", " #regr " \n\t"\
99 "pxor " #rega ", " #regb " \n\t"\
100 "pand " #regfe "," #regb " \n\t"\
101 "psrlq $1, " #regb " \n\t"\
102 "paddb " #regb ", " #regr " \n\t"
103
104#define PAVGB_MMX(rega, regb, regr, regfe) \
105 "movq " #rega ", " #regr " \n\t"\
106 "por " #regb ", " #regr " \n\t"\
107 "pxor " #rega ", " #regb " \n\t"\
108 "pand " #regfe "," #regb " \n\t"\
109 "psrlq $1, " #regb " \n\t"\
110 "psubb " #regb ", " #regr " \n\t"
111
112// mm6 is supposed to contain 0xfefefefefefefefe
113#define PAVGBP_MMX_NO_RND(rega, regb, regr, regc, regd, regp) \
114 "movq " #rega ", " #regr " \n\t"\
115 "movq " #regc ", " #regp " \n\t"\
116 "pand " #regb ", " #regr " \n\t"\
117 "pand " #regd ", " #regp " \n\t"\
118 "pxor " #rega ", " #regb " \n\t"\
119 "pxor " #regc ", " #regd " \n\t"\
120 "pand %%mm6, " #regb " \n\t"\
121 "pand %%mm6, " #regd " \n\t"\
122 "psrlq $1, " #regb " \n\t"\
123 "psrlq $1, " #regd " \n\t"\
124 "paddb " #regb ", " #regr " \n\t"\
125 "paddb " #regd ", " #regp " \n\t"
126
127#define PAVGBP_MMX(rega, regb, regr, regc, regd, regp) \
128 "movq " #rega ", " #regr " \n\t"\
129 "movq " #regc ", " #regp " \n\t"\
130 "por " #regb ", " #regr " \n\t"\
131 "por " #regd ", " #regp " \n\t"\
132 "pxor " #rega ", " #regb " \n\t"\
133 "pxor " #regc ", " #regd " \n\t"\
134 "pand %%mm6, " #regb " \n\t"\
135 "pand %%mm6, " #regd " \n\t"\
136 "psrlq $1, " #regd " \n\t"\
137 "psrlq $1, " #regb " \n\t"\
138 "psubb " #regb ", " #regr " \n\t"\
139 "psubb " #regd ", " #regp " \n\t"
140
141/***********************************/
142/* MMX no rounding */
143#define DEF(x, y) x ## _no_rnd_ ## y ##_mmx
144#define SET_RND MOVQ_WONE
145#define PAVGBP(a, b, c, d, e, f) PAVGBP_MMX_NO_RND(a, b, c, d, e, f)
146#define PAVGB(a, b, c, e) PAVGB_MMX_NO_RND(a, b, c, e)
147
148#include "dsputil_mmx_rnd.h"
149
150#undef DEF
151#undef SET_RND
152#undef PAVGBP
153#undef PAVGB
154/***********************************/
155/* MMX rounding */
156
157#define DEF(x, y) x ## _ ## y ##_mmx
158#define SET_RND MOVQ_WTWO
159#define PAVGBP(a, b, c, d, e, f) PAVGBP_MMX(a, b, c, d, e, f)
160#define PAVGB(a, b, c, e) PAVGB_MMX(a, b, c, e)
161
162#include "dsputil_mmx_rnd.h"
163
164#undef DEF
165#undef SET_RND
166#undef PAVGBP
167#undef PAVGB
168
169/***********************************/
170/* 3Dnow specific */
171
172#define DEF(x) x ## _3dnow
173/* for Athlons PAVGUSB is prefered */
174#define PAVGB "pavgusb"
175
176#include "dsputil_mmx_avg.h"
177
178#undef DEF
179#undef PAVGB
180
181/***********************************/
182/* MMX2 specific */
183
184#define DEF(x) x ## _mmx2
185
186/* Introduced only in MMX2 set */
187#define PAVGB "pavgb"
188
189#include "dsputil_mmx_avg.h"
190
191#undef DEF
192#undef PAVGB
193
194#define SBUTTERFLY(a,b,t,n)\
195 "movq " #a ", " #t " \n\t" /* abcd */\
196 "punpckl" #n " " #b ", " #a " \n\t" /* aebf */\
197 "punpckh" #n " " #b ", " #t " \n\t" /* cgdh */\
198
199/***********************************/
200/* standard MMX */
201
202#ifdef CONFIG_ENCODERS
203static void get_pixels_mmx(DCTELEM *block, const uint8_t *pixels, int line_size)
204{
205 asm volatile(
206 "mov $-128, %%"REG_a" \n\t"
207 "pxor %%mm7, %%mm7 \n\t"
208#if !defined(VBOX) || !defined(__DARWIN__)
209 ".balign 16 \n\t"
210#else
211 ".align 4 \n\t"
212#endif
213 "1: \n\t"
214 "movq (%0), %%mm0 \n\t"
215 "movq (%0, %2), %%mm2 \n\t"
216 "movq %%mm0, %%mm1 \n\t"
217 "movq %%mm2, %%mm3 \n\t"
218 "punpcklbw %%mm7, %%mm0 \n\t"
219 "punpckhbw %%mm7, %%mm1 \n\t"
220 "punpcklbw %%mm7, %%mm2 \n\t"
221 "punpckhbw %%mm7, %%mm3 \n\t"
222 "movq %%mm0, (%1, %%"REG_a") \n\t"
223 "movq %%mm1, 8(%1, %%"REG_a") \n\t"
224 "movq %%mm2, 16(%1, %%"REG_a") \n\t"
225 "movq %%mm3, 24(%1, %%"REG_a") \n\t"
226 "add %3, %0 \n\t"
227 "add $32, %%"REG_a" \n\t"
228 "js 1b \n\t"
229 : "+r" (pixels)
230 : "r" (block+64), "r" ((long)line_size), "r" ((long)line_size*2)
231 : "%"REG_a
232 );
233}
234
235static inline void diff_pixels_mmx(DCTELEM *block, const uint8_t *s1, const uint8_t *s2, int stride)
236{
237 asm volatile(
238 "pxor %%mm7, %%mm7 \n\t"
239 "mov $-128, %%"REG_a" \n\t"
240#if !defined(VBOX) || !defined(__DARWIN__)
241 ".balign 16 \n\t"
242#else
243 ".align 4 \n\t"
244#endif
245 "1: \n\t"
246 "movq (%0), %%mm0 \n\t"
247 "movq (%1), %%mm2 \n\t"
248 "movq %%mm0, %%mm1 \n\t"
249 "movq %%mm2, %%mm3 \n\t"
250 "punpcklbw %%mm7, %%mm0 \n\t"
251 "punpckhbw %%mm7, %%mm1 \n\t"
252 "punpcklbw %%mm7, %%mm2 \n\t"
253 "punpckhbw %%mm7, %%mm3 \n\t"
254 "psubw %%mm2, %%mm0 \n\t"
255 "psubw %%mm3, %%mm1 \n\t"
256 "movq %%mm0, (%2, %%"REG_a") \n\t"
257 "movq %%mm1, 8(%2, %%"REG_a") \n\t"
258 "add %3, %0 \n\t"
259 "add %3, %1 \n\t"
260 "add $16, %%"REG_a" \n\t"
261 "jnz 1b \n\t"
262 : "+r" (s1), "+r" (s2)
263 : "r" (block+64), "r" ((long)stride)
264 : "%"REG_a
265 );
266}
267#endif //CONFIG_ENCODERS
268
269void put_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels, int line_size)
270{
271 const DCTELEM *p;
272 uint8_t *pix;
273
274 /* read the pixels */
275 p = block;
276 pix = pixels;
277 /* unrolled loop */
278 __asm __volatile(
279 "movq %3, %%mm0 \n\t"
280 "movq 8%3, %%mm1 \n\t"
281 "movq 16%3, %%mm2 \n\t"
282 "movq 24%3, %%mm3 \n\t"
283 "movq 32%3, %%mm4 \n\t"
284 "movq 40%3, %%mm5 \n\t"
285 "movq 48%3, %%mm6 \n\t"
286 "movq 56%3, %%mm7 \n\t"
287 "packuswb %%mm1, %%mm0 \n\t"
288 "packuswb %%mm3, %%mm2 \n\t"
289 "packuswb %%mm5, %%mm4 \n\t"
290 "packuswb %%mm7, %%mm6 \n\t"
291 "movq %%mm0, (%0) \n\t"
292 "movq %%mm2, (%0, %1) \n\t"
293 "movq %%mm4, (%0, %1, 2) \n\t"
294 "movq %%mm6, (%0, %2) \n\t"
295 ::"r" (pix), "r" ((long)line_size), "r" ((long)line_size*3), "m"(*p)
296 :"memory");
297 pix += line_size*4;
298 p += 32;
299
300 // if here would be an exact copy of the code above
301 // compiler would generate some very strange code
302 // thus using "r"
303 __asm __volatile(
304 "movq (%3), %%mm0 \n\t"
305 "movq 8(%3), %%mm1 \n\t"
306 "movq 16(%3), %%mm2 \n\t"
307 "movq 24(%3), %%mm3 \n\t"
308 "movq 32(%3), %%mm4 \n\t"
309 "movq 40(%3), %%mm5 \n\t"
310 "movq 48(%3), %%mm6 \n\t"
311 "movq 56(%3), %%mm7 \n\t"
312 "packuswb %%mm1, %%mm0 \n\t"
313 "packuswb %%mm3, %%mm2 \n\t"
314 "packuswb %%mm5, %%mm4 \n\t"
315 "packuswb %%mm7, %%mm6 \n\t"
316 "movq %%mm0, (%0) \n\t"
317 "movq %%mm2, (%0, %1) \n\t"
318 "movq %%mm4, (%0, %1, 2) \n\t"
319 "movq %%mm6, (%0, %2) \n\t"
320 ::"r" (pix), "r" ((long)line_size), "r" ((long)line_size*3), "r"(p)
321 :"memory");
322}
323
324static DECLARE_ALIGNED_8(const unsigned char, vector128[8]) =
325 { 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80 };
326
327void put_signed_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels, int line_size)
328{
329 int i;
330
331 movq_m2r(*vector128, mm1);
332 for (i = 0; i < 8; i++) {
333 movq_m2r(*(block), mm0);
334 packsswb_m2r(*(block + 4), mm0);
335 block += 8;
336 paddb_r2r(mm1, mm0);
337 movq_r2m(mm0, *pixels);
338 pixels += line_size;
339 }
340}
341
342void add_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels, int line_size)
343{
344 const DCTELEM *p;
345 uint8_t *pix;
346 int i;
347
348 /* read the pixels */
349 p = block;
350 pix = pixels;
351 MOVQ_ZERO(mm7);
352 i = 4;
353 do {
354 __asm __volatile(
355 "movq (%2), %%mm0 \n\t"
356 "movq 8(%2), %%mm1 \n\t"
357 "movq 16(%2), %%mm2 \n\t"
358 "movq 24(%2), %%mm3 \n\t"
359 "movq %0, %%mm4 \n\t"
360 "movq %1, %%mm6 \n\t"
361 "movq %%mm4, %%mm5 \n\t"
362 "punpcklbw %%mm7, %%mm4 \n\t"
363 "punpckhbw %%mm7, %%mm5 \n\t"
364 "paddsw %%mm4, %%mm0 \n\t"
365 "paddsw %%mm5, %%mm1 \n\t"
366 "movq %%mm6, %%mm5 \n\t"
367 "punpcklbw %%mm7, %%mm6 \n\t"
368 "punpckhbw %%mm7, %%mm5 \n\t"
369 "paddsw %%mm6, %%mm2 \n\t"
370 "paddsw %%mm5, %%mm3 \n\t"
371 "packuswb %%mm1, %%mm0 \n\t"
372 "packuswb %%mm3, %%mm2 \n\t"
373 "movq %%mm0, %0 \n\t"
374 "movq %%mm2, %1 \n\t"
375 :"+m"(*pix), "+m"(*(pix+line_size))
376 :"r"(p)
377 :"memory");
378 pix += line_size*2;
379 p += 16;
380 } while (--i);
381}
382
383static void put_pixels4_mmx(uint8_t *block, const uint8_t *pixels, int line_size, int h)
384{
385 __asm __volatile(
386 "lea (%3, %3), %%"REG_a" \n\t"
387#if !defined(VBOX) || !defined(__DARWIN__)
388 ".balign 8 \n\t"
389#else
390 ".align 3 \n\t"
391#endif
392 "1: \n\t"
393 "movd (%1), %%mm0 \n\t"
394 "movd (%1, %3), %%mm1 \n\t"
395 "movd %%mm0, (%2) \n\t"
396 "movd %%mm1, (%2, %3) \n\t"
397 "add %%"REG_a", %1 \n\t"
398 "add %%"REG_a", %2 \n\t"
399 "movd (%1), %%mm0 \n\t"
400 "movd (%1, %3), %%mm1 \n\t"
401 "movd %%mm0, (%2) \n\t"
402 "movd %%mm1, (%2, %3) \n\t"
403 "add %%"REG_a", %1 \n\t"
404 "add %%"REG_a", %2 \n\t"
405 "subl $4, %0 \n\t"
406 "jnz 1b \n\t"
407 : "+g"(h), "+r" (pixels), "+r" (block)
408 : "r"((long)line_size)
409 : "%"REG_a, "memory"
410 );
411}
412
413static void put_pixels8_mmx(uint8_t *block, const uint8_t *pixels, int line_size, int h)
414{
415 __asm __volatile(
416 "lea (%3, %3), %%"REG_a" \n\t"
417#if !defined(VBOX) || !defined(__DARWIN__)
418 ".balign 8 \n\t"
419#else
420 ".align 3 \n\t"
421#endif
422 "1: \n\t"
423 "movq (%1), %%mm0 \n\t"
424 "movq (%1, %3), %%mm1 \n\t"
425 "movq %%mm0, (%2) \n\t"
426 "movq %%mm1, (%2, %3) \n\t"
427 "add %%"REG_a", %1 \n\t"
428 "add %%"REG_a", %2 \n\t"
429 "movq (%1), %%mm0 \n\t"
430 "movq (%1, %3), %%mm1 \n\t"
431 "movq %%mm0, (%2) \n\t"
432 "movq %%mm1, (%2, %3) \n\t"
433 "add %%"REG_a", %1 \n\t"
434 "add %%"REG_a", %2 \n\t"
435 "subl $4, %0 \n\t"
436 "jnz 1b \n\t"
437 : "+g"(h), "+r" (pixels), "+r" (block)
438 : "r"((long)line_size)
439 : "%"REG_a, "memory"
440 );
441}
442
443static void put_pixels16_mmx(uint8_t *block, const uint8_t *pixels, int line_size, int h)
444{
445 __asm __volatile(
446 "lea (%3, %3), %%"REG_a" \n\t"
447#if !defined(VBOX) || !defined(__DARWIN__)
448 ".balign 8 \n\t"
449#else
450 ".align 3 \n\t"
451#endif
452 "1: \n\t"
453 "movq (%1), %%mm0 \n\t"
454 "movq 8(%1), %%mm4 \n\t"
455 "movq (%1, %3), %%mm1 \n\t"
456 "movq 8(%1, %3), %%mm5 \n\t"
457 "movq %%mm0, (%2) \n\t"
458 "movq %%mm4, 8(%2) \n\t"
459 "movq %%mm1, (%2, %3) \n\t"
460 "movq %%mm5, 8(%2, %3) \n\t"
461 "add %%"REG_a", %1 \n\t"
462 "add %%"REG_a", %2 \n\t"
463 "movq (%1), %%mm0 \n\t"
464 "movq 8(%1), %%mm4 \n\t"
465 "movq (%1, %3), %%mm1 \n\t"
466 "movq 8(%1, %3), %%mm5 \n\t"
467 "movq %%mm0, (%2) \n\t"
468 "movq %%mm4, 8(%2) \n\t"
469 "movq %%mm1, (%2, %3) \n\t"
470 "movq %%mm5, 8(%2, %3) \n\t"
471 "add %%"REG_a", %1 \n\t"
472 "add %%"REG_a", %2 \n\t"
473 "subl $4, %0 \n\t"
474 "jnz 1b \n\t"
475 : "+g"(h), "+r" (pixels), "+r" (block)
476 : "r"((long)line_size)
477 : "%"REG_a, "memory"
478 );
479}
480
481static void clear_blocks_mmx(DCTELEM *blocks)
482{
483 __asm __volatile(
484 "pxor %%mm7, %%mm7 \n\t"
485 "mov $-128*6, %%"REG_a" \n\t"
486 "1: \n\t"
487 "movq %%mm7, (%0, %%"REG_a") \n\t"
488 "movq %%mm7, 8(%0, %%"REG_a") \n\t"
489 "movq %%mm7, 16(%0, %%"REG_a") \n\t"
490 "movq %%mm7, 24(%0, %%"REG_a") \n\t"
491 "add $32, %%"REG_a" \n\t"
492 " js 1b \n\t"
493 : : "r" (((uint8_t *)blocks)+128*6)
494 : "%"REG_a
495 );
496}
497
498#ifdef CONFIG_ENCODERS
499static int pix_sum16_mmx(uint8_t * pix, int line_size){
500 const int h=16;
501 int sum;
502 long index= -line_size*h;
503
504 __asm __volatile(
505 "pxor %%mm7, %%mm7 \n\t"
506 "pxor %%mm6, %%mm6 \n\t"
507 "1: \n\t"
508 "movq (%2, %1), %%mm0 \n\t"
509 "movq (%2, %1), %%mm1 \n\t"
510 "movq 8(%2, %1), %%mm2 \n\t"
511 "movq 8(%2, %1), %%mm3 \n\t"
512 "punpcklbw %%mm7, %%mm0 \n\t"
513 "punpckhbw %%mm7, %%mm1 \n\t"
514 "punpcklbw %%mm7, %%mm2 \n\t"
515 "punpckhbw %%mm7, %%mm3 \n\t"
516 "paddw %%mm0, %%mm1 \n\t"
517 "paddw %%mm2, %%mm3 \n\t"
518 "paddw %%mm1, %%mm3 \n\t"
519 "paddw %%mm3, %%mm6 \n\t"
520 "add %3, %1 \n\t"
521 " js 1b \n\t"
522 "movq %%mm6, %%mm5 \n\t"
523 "psrlq $32, %%mm6 \n\t"
524 "paddw %%mm5, %%mm6 \n\t"
525 "movq %%mm6, %%mm5 \n\t"
526 "psrlq $16, %%mm6 \n\t"
527 "paddw %%mm5, %%mm6 \n\t"
528 "movd %%mm6, %0 \n\t"
529 "andl $0xFFFF, %0 \n\t"
530 : "=&r" (sum), "+r" (index)
531 : "r" (pix - index), "r" ((long)line_size)
532 );
533
534 return sum;
535}
536#endif //CONFIG_ENCODERS
537
538static void add_bytes_mmx(uint8_t *dst, uint8_t *src, int w){
539 long i=0;
540 asm volatile(
541 "1: \n\t"
542 "movq (%1, %0), %%mm0 \n\t"
543 "movq (%2, %0), %%mm1 \n\t"
544 "paddb %%mm0, %%mm1 \n\t"
545 "movq %%mm1, (%2, %0) \n\t"
546 "movq 8(%1, %0), %%mm0 \n\t"
547 "movq 8(%2, %0), %%mm1 \n\t"
548 "paddb %%mm0, %%mm1 \n\t"
549 "movq %%mm1, 8(%2, %0) \n\t"
550 "add $16, %0 \n\t"
551 "cmp %3, %0 \n\t"
552 " jb 1b \n\t"
553 : "+r" (i)
554 : "r"(src), "r"(dst), "r"((long)w-15)
555 );
556 for(; i<w; i++)
557 dst[i+0] += src[i+0];
558}
559
560#define H263_LOOP_FILTER \
561 "pxor %%mm7, %%mm7 \n\t"\
562 "movq %0, %%mm0 \n\t"\
563 "movq %0, %%mm1 \n\t"\
564 "movq %3, %%mm2 \n\t"\
565 "movq %3, %%mm3 \n\t"\
566 "punpcklbw %%mm7, %%mm0 \n\t"\
567 "punpckhbw %%mm7, %%mm1 \n\t"\
568 "punpcklbw %%mm7, %%mm2 \n\t"\
569 "punpckhbw %%mm7, %%mm3 \n\t"\
570 "psubw %%mm2, %%mm0 \n\t"\
571 "psubw %%mm3, %%mm1 \n\t"\
572 "movq %1, %%mm2 \n\t"\
573 "movq %1, %%mm3 \n\t"\
574 "movq %2, %%mm4 \n\t"\
575 "movq %2, %%mm5 \n\t"\
576 "punpcklbw %%mm7, %%mm2 \n\t"\
577 "punpckhbw %%mm7, %%mm3 \n\t"\
578 "punpcklbw %%mm7, %%mm4 \n\t"\
579 "punpckhbw %%mm7, %%mm5 \n\t"\
580 "psubw %%mm2, %%mm4 \n\t"\
581 "psubw %%mm3, %%mm5 \n\t"\
582 "psllw $2, %%mm4 \n\t"\
583 "psllw $2, %%mm5 \n\t"\
584 "paddw %%mm0, %%mm4 \n\t"\
585 "paddw %%mm1, %%mm5 \n\t"\
586 "pxor %%mm6, %%mm6 \n\t"\
587 "pcmpgtw %%mm4, %%mm6 \n\t"\
588 "pcmpgtw %%mm5, %%mm7 \n\t"\
589 "pxor %%mm6, %%mm4 \n\t"\
590 "pxor %%mm7, %%mm5 \n\t"\
591 "psubw %%mm6, %%mm4 \n\t"\
592 "psubw %%mm7, %%mm5 \n\t"\
593 "psrlw $3, %%mm4 \n\t"\
594 "psrlw $3, %%mm5 \n\t"\
595 "packuswb %%mm5, %%mm4 \n\t"\
596 "packsswb %%mm7, %%mm6 \n\t"\
597 "pxor %%mm7, %%mm7 \n\t"\
598 "movd %4, %%mm2 \n\t"\
599 "punpcklbw %%mm2, %%mm2 \n\t"\
600 "punpcklbw %%mm2, %%mm2 \n\t"\
601 "punpcklbw %%mm2, %%mm2 \n\t"\
602 "psubusb %%mm4, %%mm2 \n\t"\
603 "movq %%mm2, %%mm3 \n\t"\
604 "psubusb %%mm4, %%mm3 \n\t"\
605 "psubb %%mm3, %%mm2 \n\t"\
606 "movq %1, %%mm3 \n\t"\
607 "movq %2, %%mm4 \n\t"\
608 "pxor %%mm6, %%mm3 \n\t"\
609 "pxor %%mm6, %%mm4 \n\t"\
610 "paddusb %%mm2, %%mm3 \n\t"\
611 "psubusb %%mm2, %%mm4 \n\t"\
612 "pxor %%mm6, %%mm3 \n\t"\
613 "pxor %%mm6, %%mm4 \n\t"\
614 "paddusb %%mm2, %%mm2 \n\t"\
615 "packsswb %%mm1, %%mm0 \n\t"\
616 "pcmpgtb %%mm0, %%mm7 \n\t"\
617 "pxor %%mm7, %%mm0 \n\t"\
618 "psubb %%mm7, %%mm0 \n\t"\
619 "movq %%mm0, %%mm1 \n\t"\
620 "psubusb %%mm2, %%mm0 \n\t"\
621 "psubb %%mm0, %%mm1 \n\t"\
622 "pand %5, %%mm1 \n\t"\
623 "psrlw $2, %%mm1 \n\t"\
624 "pxor %%mm7, %%mm1 \n\t"\
625 "psubb %%mm7, %%mm1 \n\t"\
626 "movq %0, %%mm5 \n\t"\
627 "movq %3, %%mm6 \n\t"\
628 "psubb %%mm1, %%mm5 \n\t"\
629 "paddb %%mm1, %%mm6 \n\t"
630
631static void h263_v_loop_filter_mmx(uint8_t *src, int stride, int qscale){
632 const int strength= ff_h263_loop_filter_strength[qscale];
633
634 asm volatile(
635
636 H263_LOOP_FILTER
637
638 "movq %%mm3, %1 \n\t"
639 "movq %%mm4, %2 \n\t"
640 "movq %%mm5, %0 \n\t"
641 "movq %%mm6, %3 \n\t"
642 : "+m" (*(uint64_t*)(src - 2*stride)),
643 "+m" (*(uint64_t*)(src - 1*stride)),
644 "+m" (*(uint64_t*)(src + 0*stride)),
645 "+m" (*(uint64_t*)(src + 1*stride))
646 : "g" (2*strength), "m"(ff_pb_FC)
647 );
648}
649
650static inline void transpose4x4(uint8_t *dst, uint8_t *src, int dst_stride, int src_stride){
651 asm volatile( //FIXME could save 1 instruction if done as 8x4 ...
652 "movd %4, %%mm0 \n\t"
653 "movd %5, %%mm1 \n\t"
654 "movd %6, %%mm2 \n\t"
655 "movd %7, %%mm3 \n\t"
656 "punpcklbw %%mm1, %%mm0 \n\t"
657 "punpcklbw %%mm3, %%mm2 \n\t"
658 "movq %%mm0, %%mm1 \n\t"
659 "punpcklwd %%mm2, %%mm0 \n\t"
660 "punpckhwd %%mm2, %%mm1 \n\t"
661 "movd %%mm0, %0 \n\t"
662 "punpckhdq %%mm0, %%mm0 \n\t"
663 "movd %%mm0, %1 \n\t"
664 "movd %%mm1, %2 \n\t"
665 "punpckhdq %%mm1, %%mm1 \n\t"
666 "movd %%mm1, %3 \n\t"
667
668 : "=m" (*(uint32_t*)(dst + 0*dst_stride)),
669 "=m" (*(uint32_t*)(dst + 1*dst_stride)),
670 "=m" (*(uint32_t*)(dst + 2*dst_stride)),
671 "=m" (*(uint32_t*)(dst + 3*dst_stride))
672 : "m" (*(uint32_t*)(src + 0*src_stride)),
673 "m" (*(uint32_t*)(src + 1*src_stride)),
674 "m" (*(uint32_t*)(src + 2*src_stride)),
675 "m" (*(uint32_t*)(src + 3*src_stride))
676 );
677}
678
679static void h263_h_loop_filter_mmx(uint8_t *src, int stride, int qscale){
680 const int strength= ff_h263_loop_filter_strength[qscale];
681 uint64_t temp[4] __attribute__ ((aligned(8)));
682 uint8_t *btemp= (uint8_t*)temp;
683
684 src -= 2;
685
686 transpose4x4(btemp , src , 8, stride);
687 transpose4x4(btemp+4, src + 4*stride, 8, stride);
688 asm volatile(
689 H263_LOOP_FILTER // 5 3 4 6
690
691 : "+m" (temp[0]),
692 "+m" (temp[1]),
693 "+m" (temp[2]),
694 "+m" (temp[3])
695 : "g" (2*strength), "m"(ff_pb_FC)
696 );
697
698 asm volatile(
699 "movq %%mm5, %%mm1 \n\t"
700 "movq %%mm4, %%mm0 \n\t"
701 "punpcklbw %%mm3, %%mm5 \n\t"
702 "punpcklbw %%mm6, %%mm4 \n\t"
703 "punpckhbw %%mm3, %%mm1 \n\t"
704 "punpckhbw %%mm6, %%mm0 \n\t"
705 "movq %%mm5, %%mm3 \n\t"
706 "movq %%mm1, %%mm6 \n\t"
707 "punpcklwd %%mm4, %%mm5 \n\t"
708 "punpcklwd %%mm0, %%mm1 \n\t"
709 "punpckhwd %%mm4, %%mm3 \n\t"
710 "punpckhwd %%mm0, %%mm6 \n\t"
711 "movd %%mm5, (%0) \n\t"
712 "punpckhdq %%mm5, %%mm5 \n\t"
713 "movd %%mm5, (%0,%2) \n\t"
714 "movd %%mm3, (%0,%2,2) \n\t"
715 "punpckhdq %%mm3, %%mm3 \n\t"
716 "movd %%mm3, (%0,%3) \n\t"
717 "movd %%mm1, (%1) \n\t"
718 "punpckhdq %%mm1, %%mm1 \n\t"
719 "movd %%mm1, (%1,%2) \n\t"
720 "movd %%mm6, (%1,%2,2) \n\t"
721 "punpckhdq %%mm6, %%mm6 \n\t"
722 "movd %%mm6, (%1,%3) \n\t"
723 :: "r" (src),
724 "r" (src + 4*stride),
725 "r" ((long) stride ),
726 "r" ((long)(3*stride))
727 );
728}
729
730#ifdef CONFIG_ENCODERS
731static int pix_norm1_mmx(uint8_t *pix, int line_size) {
732 int tmp;
733 asm volatile (
734 "movl $16,%%ecx\n"
735 "pxor %%mm0,%%mm0\n"
736 "pxor %%mm7,%%mm7\n"
737 "1:\n"
738 "movq (%0),%%mm2\n" /* mm2 = pix[0-7] */
739 "movq 8(%0),%%mm3\n" /* mm3 = pix[8-15] */
740
741 "movq %%mm2,%%mm1\n" /* mm1 = mm2 = pix[0-7] */
742
743 "punpckhbw %%mm0,%%mm1\n" /* mm1 = [pix4-7] */
744 "punpcklbw %%mm0,%%mm2\n" /* mm2 = [pix0-3] */
745
746 "movq %%mm3,%%mm4\n" /* mm4 = mm3 = pix[8-15] */
747 "punpckhbw %%mm0,%%mm3\n" /* mm3 = [pix12-15] */
748 "punpcklbw %%mm0,%%mm4\n" /* mm4 = [pix8-11] */
749
750 "pmaddwd %%mm1,%%mm1\n" /* mm1 = (pix0^2+pix1^2,pix2^2+pix3^2) */
751 "pmaddwd %%mm2,%%mm2\n" /* mm2 = (pix4^2+pix5^2,pix6^2+pix7^2) */
752
753 "pmaddwd %%mm3,%%mm3\n"
754 "pmaddwd %%mm4,%%mm4\n"
755
756 "paddd %%mm1,%%mm2\n" /* mm2 = (pix0^2+pix1^2+pix4^2+pix5^2,
757 pix2^2+pix3^2+pix6^2+pix7^2) */
758 "paddd %%mm3,%%mm4\n"
759 "paddd %%mm2,%%mm7\n"
760
761 "add %2, %0\n"
762 "paddd %%mm4,%%mm7\n"
763 "dec %%ecx\n"
764 "jnz 1b\n"
765
766 "movq %%mm7,%%mm1\n"
767 "psrlq $32, %%mm7\n" /* shift hi dword to lo */
768 "paddd %%mm7,%%mm1\n"
769 "movd %%mm1,%1\n"
770 : "+r" (pix), "=r"(tmp) : "r" ((long)line_size) : "%ecx" );
771 return tmp;
772}
773
774static int sse8_mmx(void *v, uint8_t * pix1, uint8_t * pix2, int line_size, int h) {
775 int tmp;
776 asm volatile (
777 "movl %4,%%ecx\n"
778 "shr $1,%%ecx\n"
779 "pxor %%mm0,%%mm0\n" /* mm0 = 0 */
780 "pxor %%mm7,%%mm7\n" /* mm7 holds the sum */
781 "1:\n"
782 "movq (%0),%%mm1\n" /* mm1 = pix1[0][0-7] */
783 "movq (%1),%%mm2\n" /* mm2 = pix2[0][0-7] */
784 "movq (%0,%3),%%mm3\n" /* mm3 = pix1[1][0-7] */
785 "movq (%1,%3),%%mm4\n" /* mm4 = pix2[1][0-7] */
786
787 /* todo: mm1-mm2, mm3-mm4 */
788 /* algo: substract mm1 from mm2 with saturation and vice versa */
789 /* OR the results to get absolute difference */
790 "movq %%mm1,%%mm5\n"
791 "movq %%mm3,%%mm6\n"
792 "psubusb %%mm2,%%mm1\n"
793 "psubusb %%mm4,%%mm3\n"
794 "psubusb %%mm5,%%mm2\n"
795 "psubusb %%mm6,%%mm4\n"
796
797 "por %%mm1,%%mm2\n"
798 "por %%mm3,%%mm4\n"
799
800 /* now convert to 16-bit vectors so we can square them */
801 "movq %%mm2,%%mm1\n"
802 "movq %%mm4,%%mm3\n"
803
804 "punpckhbw %%mm0,%%mm2\n"
805 "punpckhbw %%mm0,%%mm4\n"
806 "punpcklbw %%mm0,%%mm1\n" /* mm1 now spread over (mm1,mm2) */
807 "punpcklbw %%mm0,%%mm3\n" /* mm4 now spread over (mm3,mm4) */
808
809 "pmaddwd %%mm2,%%mm2\n"
810 "pmaddwd %%mm4,%%mm4\n"
811 "pmaddwd %%mm1,%%mm1\n"
812 "pmaddwd %%mm3,%%mm3\n"
813
814 "lea (%0,%3,2), %0\n" /* pix1 += 2*line_size */
815 "lea (%1,%3,2), %1\n" /* pix2 += 2*line_size */
816
817 "paddd %%mm2,%%mm1\n"
818 "paddd %%mm4,%%mm3\n"
819 "paddd %%mm1,%%mm7\n"
820 "paddd %%mm3,%%mm7\n"
821
822 "decl %%ecx\n"
823 "jnz 1b\n"
824
825 "movq %%mm7,%%mm1\n"
826 "psrlq $32, %%mm7\n" /* shift hi dword to lo */
827 "paddd %%mm7,%%mm1\n"
828 "movd %%mm1,%2\n"
829 : "+r" (pix1), "+r" (pix2), "=r"(tmp)
830 : "r" ((long)line_size) , "m" (h)
831 : "%ecx");
832 return tmp;
833}
834
835static int sse16_mmx(void *v, uint8_t * pix1, uint8_t * pix2, int line_size, int h) {
836 int tmp;
837 asm volatile (
838 "movl %4,%%ecx\n"
839 "pxor %%mm0,%%mm0\n" /* mm0 = 0 */
840 "pxor %%mm7,%%mm7\n" /* mm7 holds the sum */
841 "1:\n"
842 "movq (%0),%%mm1\n" /* mm1 = pix1[0-7] */
843 "movq (%1),%%mm2\n" /* mm2 = pix2[0-7] */
844 "movq 8(%0),%%mm3\n" /* mm3 = pix1[8-15] */
845 "movq 8(%1),%%mm4\n" /* mm4 = pix2[8-15] */
846
847 /* todo: mm1-mm2, mm3-mm4 */
848 /* algo: substract mm1 from mm2 with saturation and vice versa */
849 /* OR the results to get absolute difference */
850 "movq %%mm1,%%mm5\n"
851 "movq %%mm3,%%mm6\n"
852 "psubusb %%mm2,%%mm1\n"
853 "psubusb %%mm4,%%mm3\n"
854 "psubusb %%mm5,%%mm2\n"
855 "psubusb %%mm6,%%mm4\n"
856
857 "por %%mm1,%%mm2\n"
858 "por %%mm3,%%mm4\n"
859
860 /* now convert to 16-bit vectors so we can square them */
861 "movq %%mm2,%%mm1\n"
862 "movq %%mm4,%%mm3\n"
863
864 "punpckhbw %%mm0,%%mm2\n"
865 "punpckhbw %%mm0,%%mm4\n"
866 "punpcklbw %%mm0,%%mm1\n" /* mm1 now spread over (mm1,mm2) */
867 "punpcklbw %%mm0,%%mm3\n" /* mm4 now spread over (mm3,mm4) */
868
869 "pmaddwd %%mm2,%%mm2\n"
870 "pmaddwd %%mm4,%%mm4\n"
871 "pmaddwd %%mm1,%%mm1\n"
872 "pmaddwd %%mm3,%%mm3\n"
873
874 "add %3,%0\n"
875 "add %3,%1\n"
876
877 "paddd %%mm2,%%mm1\n"
878 "paddd %%mm4,%%mm3\n"
879 "paddd %%mm1,%%mm7\n"
880 "paddd %%mm3,%%mm7\n"
881
882 "decl %%ecx\n"
883 "jnz 1b\n"
884
885 "movq %%mm7,%%mm1\n"
886 "psrlq $32, %%mm7\n" /* shift hi dword to lo */
887 "paddd %%mm7,%%mm1\n"
888 "movd %%mm1,%2\n"
889 : "+r" (pix1), "+r" (pix2), "=r"(tmp)
890 : "r" ((long)line_size) , "m" (h)
891 : "%ecx");
892 return tmp;
893}
894
895static int sse16_sse2(void *v, uint8_t * pix1, uint8_t * pix2, int line_size, int h) {
896 int tmp;
897 asm volatile (
898 "shr $1,%2\n"
899 "pxor %%xmm0,%%xmm0\n" /* mm0 = 0 */
900 "pxor %%xmm7,%%xmm7\n" /* mm7 holds the sum */
901 "1:\n"
902 "movdqu (%0),%%xmm1\n" /* mm1 = pix1[0][0-15] */
903 "movdqu (%1),%%xmm2\n" /* mm2 = pix2[0][0-15] */
904 "movdqu (%0,%4),%%xmm3\n" /* mm3 = pix1[1][0-15] */
905 "movdqu (%1,%4),%%xmm4\n" /* mm4 = pix2[1][0-15] */
906
907 /* todo: mm1-mm2, mm3-mm4 */
908 /* algo: substract mm1 from mm2 with saturation and vice versa */
909 /* OR the results to get absolute difference */
910 "movdqa %%xmm1,%%xmm5\n"
911 "movdqa %%xmm3,%%xmm6\n"
912 "psubusb %%xmm2,%%xmm1\n"
913 "psubusb %%xmm4,%%xmm3\n"
914 "psubusb %%xmm5,%%xmm2\n"
915 "psubusb %%xmm6,%%xmm4\n"
916
917 "por %%xmm1,%%xmm2\n"
918 "por %%xmm3,%%xmm4\n"
919
920 /* now convert to 16-bit vectors so we can square them */
921 "movdqa %%xmm2,%%xmm1\n"
922 "movdqa %%xmm4,%%xmm3\n"
923
924 "punpckhbw %%xmm0,%%xmm2\n"
925 "punpckhbw %%xmm0,%%xmm4\n"
926 "punpcklbw %%xmm0,%%xmm1\n" /* mm1 now spread over (mm1,mm2) */
927 "punpcklbw %%xmm0,%%xmm3\n" /* mm4 now spread over (mm3,mm4) */
928
929 "pmaddwd %%xmm2,%%xmm2\n"
930 "pmaddwd %%xmm4,%%xmm4\n"
931 "pmaddwd %%xmm1,%%xmm1\n"
932 "pmaddwd %%xmm3,%%xmm3\n"
933
934 "lea (%0,%4,2), %0\n" /* pix1 += 2*line_size */
935 "lea (%1,%4,2), %1\n" /* pix2 += 2*line_size */
936
937 "paddd %%xmm2,%%xmm1\n"
938 "paddd %%xmm4,%%xmm3\n"
939 "paddd %%xmm1,%%xmm7\n"
940 "paddd %%xmm3,%%xmm7\n"
941
942 "decl %2\n"
943 "jnz 1b\n"
944
945 "movdqa %%xmm7,%%xmm1\n"
946 "psrldq $8, %%xmm7\n" /* shift hi qword to lo */
947 "paddd %%xmm1,%%xmm7\n"
948 "movdqa %%xmm7,%%xmm1\n"
949 "psrldq $4, %%xmm7\n" /* shift hi dword to lo */
950 "paddd %%xmm1,%%xmm7\n"
951 "movd %%xmm7,%3\n"
952 : "+r" (pix1), "+r" (pix2), "+r"(h), "=r"(tmp)
953 : "r" ((long)line_size));
954 return tmp;
955}
956
957static int hf_noise8_mmx(uint8_t * pix1, int line_size, int h) {
958 int tmp;
959 asm volatile (
960 "movl %3,%%ecx\n"
961 "pxor %%mm7,%%mm7\n"
962 "pxor %%mm6,%%mm6\n"
963
964 "movq (%0),%%mm0\n"
965 "movq %%mm0, %%mm1\n"
966 "psllq $8, %%mm0\n"
967 "psrlq $8, %%mm1\n"
968 "psrlq $8, %%mm0\n"
969 "movq %%mm0, %%mm2\n"
970 "movq %%mm1, %%mm3\n"
971 "punpcklbw %%mm7,%%mm0\n"
972 "punpcklbw %%mm7,%%mm1\n"
973 "punpckhbw %%mm7,%%mm2\n"
974 "punpckhbw %%mm7,%%mm3\n"
975 "psubw %%mm1, %%mm0\n"
976 "psubw %%mm3, %%mm2\n"
977
978 "add %2,%0\n"
979
980 "movq (%0),%%mm4\n"
981 "movq %%mm4, %%mm1\n"
982 "psllq $8, %%mm4\n"
983 "psrlq $8, %%mm1\n"
984 "psrlq $8, %%mm4\n"
985 "movq %%mm4, %%mm5\n"
986 "movq %%mm1, %%mm3\n"
987 "punpcklbw %%mm7,%%mm4\n"
988 "punpcklbw %%mm7,%%mm1\n"
989 "punpckhbw %%mm7,%%mm5\n"
990 "punpckhbw %%mm7,%%mm3\n"
991 "psubw %%mm1, %%mm4\n"
992 "psubw %%mm3, %%mm5\n"
993 "psubw %%mm4, %%mm0\n"
994 "psubw %%mm5, %%mm2\n"
995 "pxor %%mm3, %%mm3\n"
996 "pxor %%mm1, %%mm1\n"
997 "pcmpgtw %%mm0, %%mm3\n\t"
998 "pcmpgtw %%mm2, %%mm1\n\t"
999 "pxor %%mm3, %%mm0\n"
1000 "pxor %%mm1, %%mm2\n"
1001 "psubw %%mm3, %%mm0\n"
1002 "psubw %%mm1, %%mm2\n"
1003 "paddw %%mm0, %%mm2\n"
1004 "paddw %%mm2, %%mm6\n"
1005
1006 "add %2,%0\n"
1007 "1:\n"
1008
1009 "movq (%0),%%mm0\n"
1010 "movq %%mm0, %%mm1\n"
1011 "psllq $8, %%mm0\n"
1012 "psrlq $8, %%mm1\n"
1013 "psrlq $8, %%mm0\n"
1014 "movq %%mm0, %%mm2\n"
1015 "movq %%mm1, %%mm3\n"
1016 "punpcklbw %%mm7,%%mm0\n"
1017 "punpcklbw %%mm7,%%mm1\n"
1018 "punpckhbw %%mm7,%%mm2\n"
1019 "punpckhbw %%mm7,%%mm3\n"
1020 "psubw %%mm1, %%mm0\n"
1021 "psubw %%mm3, %%mm2\n"
1022 "psubw %%mm0, %%mm4\n"
1023 "psubw %%mm2, %%mm5\n"
1024 "pxor %%mm3, %%mm3\n"
1025 "pxor %%mm1, %%mm1\n"
1026 "pcmpgtw %%mm4, %%mm3\n\t"
1027 "pcmpgtw %%mm5, %%mm1\n\t"
1028 "pxor %%mm3, %%mm4\n"
1029 "pxor %%mm1, %%mm5\n"
1030 "psubw %%mm3, %%mm4\n"
1031 "psubw %%mm1, %%mm5\n"
1032 "paddw %%mm4, %%mm5\n"
1033 "paddw %%mm5, %%mm6\n"
1034
1035 "add %2,%0\n"
1036
1037 "movq (%0),%%mm4\n"
1038 "movq %%mm4, %%mm1\n"
1039 "psllq $8, %%mm4\n"
1040 "psrlq $8, %%mm1\n"
1041 "psrlq $8, %%mm4\n"
1042 "movq %%mm4, %%mm5\n"
1043 "movq %%mm1, %%mm3\n"
1044 "punpcklbw %%mm7,%%mm4\n"
1045 "punpcklbw %%mm7,%%mm1\n"
1046 "punpckhbw %%mm7,%%mm5\n"
1047 "punpckhbw %%mm7,%%mm3\n"
1048 "psubw %%mm1, %%mm4\n"
1049 "psubw %%mm3, %%mm5\n"
1050 "psubw %%mm4, %%mm0\n"
1051 "psubw %%mm5, %%mm2\n"
1052 "pxor %%mm3, %%mm3\n"
1053 "pxor %%mm1, %%mm1\n"
1054 "pcmpgtw %%mm0, %%mm3\n\t"
1055 "pcmpgtw %%mm2, %%mm1\n\t"
1056 "pxor %%mm3, %%mm0\n"
1057 "pxor %%mm1, %%mm2\n"
1058 "psubw %%mm3, %%mm0\n"
1059 "psubw %%mm1, %%mm2\n"
1060 "paddw %%mm0, %%mm2\n"
1061 "paddw %%mm2, %%mm6\n"
1062
1063 "add %2,%0\n"
1064 "subl $2, %%ecx\n"
1065 " jnz 1b\n"
1066
1067 "movq %%mm6, %%mm0\n"
1068 "punpcklwd %%mm7,%%mm0\n"
1069 "punpckhwd %%mm7,%%mm6\n"
1070 "paddd %%mm0, %%mm6\n"
1071
1072 "movq %%mm6,%%mm0\n"
1073 "psrlq $32, %%mm6\n"
1074 "paddd %%mm6,%%mm0\n"
1075 "movd %%mm0,%1\n"
1076 : "+r" (pix1), "=r"(tmp)
1077 : "r" ((long)line_size) , "g" (h-2)
1078 : "%ecx");
1079 return tmp;
1080}
1081
1082static int hf_noise16_mmx(uint8_t * pix1, int line_size, int h) {
1083 int tmp;
1084 uint8_t * pix= pix1;
1085 asm volatile (
1086 "movl %3,%%ecx\n"
1087 "pxor %%mm7,%%mm7\n"
1088 "pxor %%mm6,%%mm6\n"
1089
1090 "movq (%0),%%mm0\n"
1091 "movq 1(%0),%%mm1\n"
1092 "movq %%mm0, %%mm2\n"
1093 "movq %%mm1, %%mm3\n"
1094 "punpcklbw %%mm7,%%mm0\n"
1095 "punpcklbw %%mm7,%%mm1\n"
1096 "punpckhbw %%mm7,%%mm2\n"
1097 "punpckhbw %%mm7,%%mm3\n"
1098 "psubw %%mm1, %%mm0\n"
1099 "psubw %%mm3, %%mm2\n"
1100
1101 "add %2,%0\n"
1102
1103 "movq (%0),%%mm4\n"
1104 "movq 1(%0),%%mm1\n"
1105 "movq %%mm4, %%mm5\n"
1106 "movq %%mm1, %%mm3\n"
1107 "punpcklbw %%mm7,%%mm4\n"
1108 "punpcklbw %%mm7,%%mm1\n"
1109 "punpckhbw %%mm7,%%mm5\n"
1110 "punpckhbw %%mm7,%%mm3\n"
1111 "psubw %%mm1, %%mm4\n"
1112 "psubw %%mm3, %%mm5\n"
1113 "psubw %%mm4, %%mm0\n"
1114 "psubw %%mm5, %%mm2\n"
1115 "pxor %%mm3, %%mm3\n"
1116 "pxor %%mm1, %%mm1\n"
1117 "pcmpgtw %%mm0, %%mm3\n\t"
1118 "pcmpgtw %%mm2, %%mm1\n\t"
1119 "pxor %%mm3, %%mm0\n"
1120 "pxor %%mm1, %%mm2\n"
1121 "psubw %%mm3, %%mm0\n"
1122 "psubw %%mm1, %%mm2\n"
1123 "paddw %%mm0, %%mm2\n"
1124 "paddw %%mm2, %%mm6\n"
1125
1126 "add %2,%0\n"
1127 "1:\n"
1128
1129 "movq (%0),%%mm0\n"
1130 "movq 1(%0),%%mm1\n"
1131 "movq %%mm0, %%mm2\n"
1132 "movq %%mm1, %%mm3\n"
1133 "punpcklbw %%mm7,%%mm0\n"
1134 "punpcklbw %%mm7,%%mm1\n"
1135 "punpckhbw %%mm7,%%mm2\n"
1136 "punpckhbw %%mm7,%%mm3\n"
1137 "psubw %%mm1, %%mm0\n"
1138 "psubw %%mm3, %%mm2\n"
1139 "psubw %%mm0, %%mm4\n"
1140 "psubw %%mm2, %%mm5\n"
1141 "pxor %%mm3, %%mm3\n"
1142 "pxor %%mm1, %%mm1\n"
1143 "pcmpgtw %%mm4, %%mm3\n\t"
1144 "pcmpgtw %%mm5, %%mm1\n\t"
1145 "pxor %%mm3, %%mm4\n"
1146 "pxor %%mm1, %%mm5\n"
1147 "psubw %%mm3, %%mm4\n"
1148 "psubw %%mm1, %%mm5\n"
1149 "paddw %%mm4, %%mm5\n"
1150 "paddw %%mm5, %%mm6\n"
1151
1152 "add %2,%0\n"
1153
1154 "movq (%0),%%mm4\n"
1155 "movq 1(%0),%%mm1\n"
1156 "movq %%mm4, %%mm5\n"
1157 "movq %%mm1, %%mm3\n"
1158 "punpcklbw %%mm7,%%mm4\n"
1159 "punpcklbw %%mm7,%%mm1\n"
1160 "punpckhbw %%mm7,%%mm5\n"
1161 "punpckhbw %%mm7,%%mm3\n"
1162 "psubw %%mm1, %%mm4\n"
1163 "psubw %%mm3, %%mm5\n"
1164 "psubw %%mm4, %%mm0\n"
1165 "psubw %%mm5, %%mm2\n"
1166 "pxor %%mm3, %%mm3\n"
1167 "pxor %%mm1, %%mm1\n"
1168 "pcmpgtw %%mm0, %%mm3\n\t"
1169 "pcmpgtw %%mm2, %%mm1\n\t"
1170 "pxor %%mm3, %%mm0\n"
1171 "pxor %%mm1, %%mm2\n"
1172 "psubw %%mm3, %%mm0\n"
1173 "psubw %%mm1, %%mm2\n"
1174 "paddw %%mm0, %%mm2\n"
1175 "paddw %%mm2, %%mm6\n"
1176
1177 "add %2,%0\n"
1178 "subl $2, %%ecx\n"
1179 " jnz 1b\n"
1180
1181 "movq %%mm6, %%mm0\n"
1182 "punpcklwd %%mm7,%%mm0\n"
1183 "punpckhwd %%mm7,%%mm6\n"
1184 "paddd %%mm0, %%mm6\n"
1185
1186 "movq %%mm6,%%mm0\n"
1187 "psrlq $32, %%mm6\n"
1188 "paddd %%mm6,%%mm0\n"
1189 "movd %%mm0,%1\n"
1190 : "+r" (pix1), "=r"(tmp)
1191 : "r" ((long)line_size) , "g" (h-2)
1192 : "%ecx");
1193 return tmp + hf_noise8_mmx(pix+8, line_size, h);
1194}
1195
1196static int nsse16_mmx(void *p, uint8_t * pix1, uint8_t * pix2, int line_size, int h) {
1197 MpegEncContext *c = p;
1198 int score1, score2;
1199
1200 if(c) score1 = c->dsp.sse[0](c, pix1, pix2, line_size, h);
1201 else score1 = sse16_mmx(c, pix1, pix2, line_size, h);
1202 score2= hf_noise16_mmx(pix1, line_size, h) - hf_noise16_mmx(pix2, line_size, h);
1203
1204 if(c) return score1 + ABS(score2)*c->avctx->nsse_weight;
1205 else return score1 + ABS(score2)*8;
1206}
1207
1208static int nsse8_mmx(void *p, uint8_t * pix1, uint8_t * pix2, int line_size, int h) {
1209 MpegEncContext *c = p;
1210 int score1= sse8_mmx(c, pix1, pix2, line_size, h);
1211 int score2= hf_noise8_mmx(pix1, line_size, h) - hf_noise8_mmx(pix2, line_size, h);
1212
1213 if(c) return score1 + ABS(score2)*c->avctx->nsse_weight;
1214 else return score1 + ABS(score2)*8;
1215}
1216
1217static int vsad_intra16_mmx(void *v, uint8_t * pix, uint8_t * dummy, int line_size, int h) {
1218 int tmp;
1219
1220 assert( (((int)pix) & 7) == 0);
1221 assert((line_size &7) ==0);
1222
1223#define SUM(in0, in1, out0, out1) \
1224 "movq (%0), %%mm2\n"\
1225 "movq 8(%0), %%mm3\n"\
1226 "add %2,%0\n"\
1227 "movq %%mm2, " #out0 "\n"\
1228 "movq %%mm3, " #out1 "\n"\
1229 "psubusb " #in0 ", %%mm2\n"\
1230 "psubusb " #in1 ", %%mm3\n"\
1231 "psubusb " #out0 ", " #in0 "\n"\
1232 "psubusb " #out1 ", " #in1 "\n"\
1233 "por %%mm2, " #in0 "\n"\
1234 "por %%mm3, " #in1 "\n"\
1235 "movq " #in0 ", %%mm2\n"\
1236 "movq " #in1 ", %%mm3\n"\
1237 "punpcklbw %%mm7, " #in0 "\n"\
1238 "punpcklbw %%mm7, " #in1 "\n"\
1239 "punpckhbw %%mm7, %%mm2\n"\
1240 "punpckhbw %%mm7, %%mm3\n"\
1241 "paddw " #in1 ", " #in0 "\n"\
1242 "paddw %%mm3, %%mm2\n"\
1243 "paddw %%mm2, " #in0 "\n"\
1244 "paddw " #in0 ", %%mm6\n"
1245
1246
1247 asm volatile (
1248 "movl %3,%%ecx\n"
1249 "pxor %%mm6,%%mm6\n"
1250 "pxor %%mm7,%%mm7\n"
1251 "movq (%0),%%mm0\n"
1252 "movq 8(%0),%%mm1\n"
1253 "add %2,%0\n"
1254 "subl $2, %%ecx\n"
1255 SUM(%%mm0, %%mm1, %%mm4, %%mm5)
1256 "1:\n"
1257
1258 SUM(%%mm4, %%mm5, %%mm0, %%mm1)
1259
1260 SUM(%%mm0, %%mm1, %%mm4, %%mm5)
1261
1262 "subl $2, %%ecx\n"
1263 "jnz 1b\n"
1264
1265 "movq %%mm6,%%mm0\n"
1266 "psrlq $32, %%mm6\n"
1267 "paddw %%mm6,%%mm0\n"
1268 "movq %%mm0,%%mm6\n"
1269 "psrlq $16, %%mm0\n"
1270 "paddw %%mm6,%%mm0\n"
1271 "movd %%mm0,%1\n"
1272 : "+r" (pix), "=r"(tmp)
1273 : "r" ((long)line_size) , "m" (h)
1274 : "%ecx");
1275 return tmp & 0xFFFF;
1276}
1277#undef SUM
1278
1279static int vsad_intra16_mmx2(void *v, uint8_t * pix, uint8_t * dummy, int line_size, int h) {
1280 int tmp;
1281
1282 assert( (((int)pix) & 7) == 0);
1283 assert((line_size &7) ==0);
1284
1285#define SUM(in0, in1, out0, out1) \
1286 "movq (%0), " #out0 "\n"\
1287 "movq 8(%0), " #out1 "\n"\
1288 "add %2,%0\n"\
1289 "psadbw " #out0 ", " #in0 "\n"\
1290 "psadbw " #out1 ", " #in1 "\n"\
1291 "paddw " #in1 ", " #in0 "\n"\
1292 "paddw " #in0 ", %%mm6\n"
1293
1294 asm volatile (
1295 "movl %3,%%ecx\n"
1296 "pxor %%mm6,%%mm6\n"
1297 "pxor %%mm7,%%mm7\n"
1298 "movq (%0),%%mm0\n"
1299 "movq 8(%0),%%mm1\n"
1300 "add %2,%0\n"
1301 "subl $2, %%ecx\n"
1302 SUM(%%mm0, %%mm1, %%mm4, %%mm5)
1303 "1:\n"
1304
1305 SUM(%%mm4, %%mm5, %%mm0, %%mm1)
1306
1307 SUM(%%mm0, %%mm1, %%mm4, %%mm5)
1308
1309 "subl $2, %%ecx\n"
1310 "jnz 1b\n"
1311
1312 "movd %%mm6,%1\n"
1313 : "+r" (pix), "=r"(tmp)
1314 : "r" ((long)line_size) , "m" (h)
1315 : "%ecx");
1316 return tmp;
1317}
1318#undef SUM
1319
1320static int vsad16_mmx(void *v, uint8_t * pix1, uint8_t * pix2, int line_size, int h) {
1321 int tmp;
1322
1323 assert( (((int)pix1) & 7) == 0);
1324 assert( (((int)pix2) & 7) == 0);
1325 assert((line_size &7) ==0);
1326
1327#define SUM(in0, in1, out0, out1) \
1328 "movq (%0),%%mm2\n"\
1329 "movq (%1)," #out0 "\n"\
1330 "movq 8(%0),%%mm3\n"\
1331 "movq 8(%1)," #out1 "\n"\
1332 "add %3,%0\n"\
1333 "add %3,%1\n"\
1334 "psubb " #out0 ", %%mm2\n"\
1335 "psubb " #out1 ", %%mm3\n"\
1336 "pxor %%mm7, %%mm2\n"\
1337 "pxor %%mm7, %%mm3\n"\
1338 "movq %%mm2, " #out0 "\n"\
1339 "movq %%mm3, " #out1 "\n"\
1340 "psubusb " #in0 ", %%mm2\n"\
1341 "psubusb " #in1 ", %%mm3\n"\
1342 "psubusb " #out0 ", " #in0 "\n"\
1343 "psubusb " #out1 ", " #in1 "\n"\
1344 "por %%mm2, " #in0 "\n"\
1345 "por %%mm3, " #in1 "\n"\
1346 "movq " #in0 ", %%mm2\n"\
1347 "movq " #in1 ", %%mm3\n"\
1348 "punpcklbw %%mm7, " #in0 "\n"\
1349 "punpcklbw %%mm7, " #in1 "\n"\
1350 "punpckhbw %%mm7, %%mm2\n"\
1351 "punpckhbw %%mm7, %%mm3\n"\
1352 "paddw " #in1 ", " #in0 "\n"\
1353 "paddw %%mm3, %%mm2\n"\
1354 "paddw %%mm2, " #in0 "\n"\
1355 "paddw " #in0 ", %%mm6\n"
1356
1357
1358 asm volatile (
1359 "movl %4,%%ecx\n"
1360 "pxor %%mm6,%%mm6\n"
1361 "pcmpeqw %%mm7,%%mm7\n"
1362 "psllw $15, %%mm7\n"
1363 "packsswb %%mm7, %%mm7\n"
1364 "movq (%0),%%mm0\n"
1365 "movq (%1),%%mm2\n"
1366 "movq 8(%0),%%mm1\n"
1367 "movq 8(%1),%%mm3\n"
1368 "add %3,%0\n"
1369 "add %3,%1\n"
1370 "subl $2, %%ecx\n"
1371 "psubb %%mm2, %%mm0\n"
1372 "psubb %%mm3, %%mm1\n"
1373 "pxor %%mm7, %%mm0\n"
1374 "pxor %%mm7, %%mm1\n"
1375 SUM(%%mm0, %%mm1, %%mm4, %%mm5)
1376 "1:\n"
1377
1378 SUM(%%mm4, %%mm5, %%mm0, %%mm1)
1379
1380 SUM(%%mm0, %%mm1, %%mm4, %%mm5)
1381
1382 "subl $2, %%ecx\n"
1383 "jnz 1b\n"
1384
1385 "movq %%mm6,%%mm0\n"
1386 "psrlq $32, %%mm6\n"
1387 "paddw %%mm6,%%mm0\n"
1388 "movq %%mm0,%%mm6\n"
1389 "psrlq $16, %%mm0\n"
1390 "paddw %%mm6,%%mm0\n"
1391 "movd %%mm0,%2\n"
1392 : "+r" (pix1), "+r" (pix2), "=r"(tmp)
1393 : "r" ((long)line_size) , "m" (h)
1394 : "%ecx");
1395 return tmp & 0x7FFF;
1396}
1397#undef SUM
1398
1399static int vsad16_mmx2(void *v, uint8_t * pix1, uint8_t * pix2, int line_size, int h) {
1400 int tmp;
1401
1402 assert( (((int)pix1) & 7) == 0);
1403 assert( (((int)pix2) & 7) == 0);
1404 assert((line_size &7) ==0);
1405
1406#define SUM(in0, in1, out0, out1) \
1407 "movq (%0)," #out0 "\n"\
1408 "movq (%1),%%mm2\n"\
1409 "movq 8(%0)," #out1 "\n"\
1410 "movq 8(%1),%%mm3\n"\
1411 "add %3,%0\n"\
1412 "add %3,%1\n"\
1413 "psubb %%mm2, " #out0 "\n"\
1414 "psubb %%mm3, " #out1 "\n"\
1415 "pxor %%mm7, " #out0 "\n"\
1416 "pxor %%mm7, " #out1 "\n"\
1417 "psadbw " #out0 ", " #in0 "\n"\
1418 "psadbw " #out1 ", " #in1 "\n"\
1419 "paddw " #in1 ", " #in0 "\n"\
1420 "paddw " #in0 ", %%mm6\n"
1421
1422 asm volatile (
1423 "movl %4,%%ecx\n"
1424 "pxor %%mm6,%%mm6\n"
1425 "pcmpeqw %%mm7,%%mm7\n"
1426 "psllw $15, %%mm7\n"
1427 "packsswb %%mm7, %%mm7\n"
1428 "movq (%0),%%mm0\n"
1429 "movq (%1),%%mm2\n"
1430 "movq 8(%0),%%mm1\n"
1431 "movq 8(%1),%%mm3\n"
1432 "add %3,%0\n"
1433 "add %3,%1\n"
1434 "subl $2, %%ecx\n"
1435 "psubb %%mm2, %%mm0\n"
1436 "psubb %%mm3, %%mm1\n"
1437 "pxor %%mm7, %%mm0\n"
1438 "pxor %%mm7, %%mm1\n"
1439 SUM(%%mm0, %%mm1, %%mm4, %%mm5)
1440 "1:\n"
1441
1442 SUM(%%mm4, %%mm5, %%mm0, %%mm1)
1443
1444 SUM(%%mm0, %%mm1, %%mm4, %%mm5)
1445
1446 "subl $2, %%ecx\n"
1447 "jnz 1b\n"
1448
1449 "movd %%mm6,%2\n"
1450 : "+r" (pix1), "+r" (pix2), "=r"(tmp)
1451 : "r" ((long)line_size) , "m" (h)
1452 : "%ecx");
1453 return tmp;
1454}
1455#undef SUM
1456
1457static void diff_bytes_mmx(uint8_t *dst, uint8_t *src1, uint8_t *src2, int w){
1458 long i=0;
1459 asm volatile(
1460 "1: \n\t"
1461 "movq (%2, %0), %%mm0 \n\t"
1462 "movq (%1, %0), %%mm1 \n\t"
1463 "psubb %%mm0, %%mm1 \n\t"
1464 "movq %%mm1, (%3, %0) \n\t"
1465 "movq 8(%2, %0), %%mm0 \n\t"
1466 "movq 8(%1, %0), %%mm1 \n\t"
1467 "psubb %%mm0, %%mm1 \n\t"
1468 "movq %%mm1, 8(%3, %0) \n\t"
1469 "add $16, %0 \n\t"
1470 "cmp %4, %0 \n\t"
1471 " jb 1b \n\t"
1472 : "+r" (i)
1473 : "r"(src1), "r"(src2), "r"(dst), "r"((long)w-15)
1474 );
1475 for(; i<w; i++)
1476 dst[i+0] = src1[i+0]-src2[i+0];
1477}
1478
1479static void sub_hfyu_median_prediction_mmx2(uint8_t *dst, uint8_t *src1, uint8_t *src2, int w, int *left, int *left_top){
1480 long i=0;
1481 uint8_t l, lt;
1482
1483 asm volatile(
1484 "1: \n\t"
1485 "movq -1(%1, %0), %%mm0 \n\t" // LT
1486 "movq (%1, %0), %%mm1 \n\t" // T
1487 "movq -1(%2, %0), %%mm2 \n\t" // L
1488 "movq (%2, %0), %%mm3 \n\t" // X
1489 "movq %%mm2, %%mm4 \n\t" // L
1490 "psubb %%mm0, %%mm2 \n\t"
1491 "paddb %%mm1, %%mm2 \n\t" // L + T - LT
1492 "movq %%mm4, %%mm5 \n\t" // L
1493 "pmaxub %%mm1, %%mm4 \n\t" // max(T, L)
1494 "pminub %%mm5, %%mm1 \n\t" // min(T, L)
1495 "pminub %%mm2, %%mm4 \n\t"
1496 "pmaxub %%mm1, %%mm4 \n\t"
1497 "psubb %%mm4, %%mm3 \n\t" // dst - pred
1498 "movq %%mm3, (%3, %0) \n\t"
1499 "add $8, %0 \n\t"
1500 "cmp %4, %0 \n\t"
1501 " jb 1b \n\t"
1502 : "+r" (i)
1503 : "r"(src1), "r"(src2), "r"(dst), "r"((long)w)
1504 );
1505
1506 l= *left;
1507 lt= *left_top;
1508
1509 dst[0]= src2[0] - mid_pred(l, src1[0], (l + src1[0] - lt)&0xFF);
1510
1511 *left_top= src1[w-1];
1512 *left = src2[w-1];
1513}
1514
1515#define LBUTTERFLY2(a1,b1,a2,b2)\
1516 "paddw " #b1 ", " #a1 " \n\t"\
1517 "paddw " #b2 ", " #a2 " \n\t"\
1518 "paddw " #b1 ", " #b1 " \n\t"\
1519 "paddw " #b2 ", " #b2 " \n\t"\
1520 "psubw " #a1 ", " #b1 " \n\t"\
1521 "psubw " #a2 ", " #b2 " \n\t"
1522
1523#define HADAMARD48\
1524 LBUTTERFLY2(%%mm0, %%mm1, %%mm2, %%mm3)\
1525 LBUTTERFLY2(%%mm4, %%mm5, %%mm6, %%mm7)\
1526 LBUTTERFLY2(%%mm0, %%mm2, %%mm1, %%mm3)\
1527 LBUTTERFLY2(%%mm4, %%mm6, %%mm5, %%mm7)\
1528 LBUTTERFLY2(%%mm0, %%mm4, %%mm1, %%mm5)\
1529 LBUTTERFLY2(%%mm2, %%mm6, %%mm3, %%mm7)\
1530
1531#define MMABS(a,z)\
1532 "pxor " #z ", " #z " \n\t"\
1533 "pcmpgtw " #a ", " #z " \n\t"\
1534 "pxor " #z ", " #a " \n\t"\
1535 "psubw " #z ", " #a " \n\t"
1536
1537#define MMABS_SUM(a,z, sum)\
1538 "pxor " #z ", " #z " \n\t"\
1539 "pcmpgtw " #a ", " #z " \n\t"\
1540 "pxor " #z ", " #a " \n\t"\
1541 "psubw " #z ", " #a " \n\t"\
1542 "paddusw " #a ", " #sum " \n\t"
1543
1544#define MMABS_MMX2(a,z)\
1545 "pxor " #z ", " #z " \n\t"\
1546 "psubw " #a ", " #z " \n\t"\
1547 "pmaxsw " #z ", " #a " \n\t"
1548
1549#define MMABS_SUM_MMX2(a,z, sum)\
1550 "pxor " #z ", " #z " \n\t"\
1551 "psubw " #a ", " #z " \n\t"\
1552 "pmaxsw " #z ", " #a " \n\t"\
1553 "paddusw " #a ", " #sum " \n\t"
1554
1555#define TRANSPOSE4(a,b,c,d,t)\
1556 SBUTTERFLY(a,b,t,wd) /* a=aebf t=cgdh */\
1557 SBUTTERFLY(c,d,b,wd) /* c=imjn b=kolp */\
1558 SBUTTERFLY(a,c,d,dq) /* a=aeim d=bfjn */\
1559 SBUTTERFLY(t,b,c,dq) /* t=cgko c=dhlp */
1560
1561#define LOAD4(o, a, b, c, d)\
1562 "movq "#o"(%1), " #a " \n\t"\
1563 "movq "#o"+16(%1), " #b " \n\t"\
1564 "movq "#o"+32(%1), " #c " \n\t"\
1565 "movq "#o"+48(%1), " #d " \n\t"
1566
1567#define STORE4(o, a, b, c, d)\
1568 "movq "#a", "#o"(%1) \n\t"\
1569 "movq "#b", "#o"+16(%1) \n\t"\
1570 "movq "#c", "#o"+32(%1) \n\t"\
1571 "movq "#d", "#o"+48(%1) \n\t"\
1572
1573static int hadamard8_diff_mmx(void *s, uint8_t *src1, uint8_t *src2, int stride, int h){
1574 DECLARE_ALIGNED_8(uint64_t, temp[16]);
1575 int sum=0;
1576
1577 assert(h==8);
1578
1579 diff_pixels_mmx((DCTELEM*)temp, src1, src2, stride);
1580
1581 asm volatile(
1582 LOAD4(0 , %%mm0, %%mm1, %%mm2, %%mm3)
1583 LOAD4(64, %%mm4, %%mm5, %%mm6, %%mm7)
1584
1585 HADAMARD48
1586
1587 "movq %%mm7, 112(%1) \n\t"
1588
1589 TRANSPOSE4(%%mm0, %%mm1, %%mm2, %%mm3, %%mm7)
1590 STORE4(0 , %%mm0, %%mm3, %%mm7, %%mm2)
1591
1592 "movq 112(%1), %%mm7 \n\t"
1593 TRANSPOSE4(%%mm4, %%mm5, %%mm6, %%mm7, %%mm0)
1594 STORE4(64, %%mm4, %%mm7, %%mm0, %%mm6)
1595
1596 LOAD4(8 , %%mm0, %%mm1, %%mm2, %%mm3)
1597 LOAD4(72, %%mm4, %%mm5, %%mm6, %%mm7)
1598
1599 HADAMARD48
1600
1601 "movq %%mm7, 120(%1) \n\t"
1602
1603 TRANSPOSE4(%%mm0, %%mm1, %%mm2, %%mm3, %%mm7)
1604 STORE4(8 , %%mm0, %%mm3, %%mm7, %%mm2)
1605
1606 "movq 120(%1), %%mm7 \n\t"
1607 TRANSPOSE4(%%mm4, %%mm5, %%mm6, %%mm7, %%mm0)
1608 "movq %%mm7, %%mm5 \n\t"//FIXME remove
1609 "movq %%mm6, %%mm7 \n\t"
1610 "movq %%mm0, %%mm6 \n\t"
1611// STORE4(72, %%mm4, %%mm7, %%mm0, %%mm6) //FIXME remove
1612
1613 LOAD4(64, %%mm0, %%mm1, %%mm2, %%mm3)
1614// LOAD4(72, %%mm4, %%mm5, %%mm6, %%mm7)
1615
1616 HADAMARD48
1617 "movq %%mm7, 64(%1) \n\t"
1618 MMABS(%%mm0, %%mm7)
1619 MMABS_SUM(%%mm1, %%mm7, %%mm0)
1620 MMABS_SUM(%%mm2, %%mm7, %%mm0)
1621 MMABS_SUM(%%mm3, %%mm7, %%mm0)
1622 MMABS_SUM(%%mm4, %%mm7, %%mm0)
1623 MMABS_SUM(%%mm5, %%mm7, %%mm0)
1624 MMABS_SUM(%%mm6, %%mm7, %%mm0)
1625 "movq 64(%1), %%mm1 \n\t"
1626 MMABS_SUM(%%mm1, %%mm7, %%mm0)
1627 "movq %%mm0, 64(%1) \n\t"
1628
1629 LOAD4(0 , %%mm0, %%mm1, %%mm2, %%mm3)
1630 LOAD4(8 , %%mm4, %%mm5, %%mm6, %%mm7)
1631
1632 HADAMARD48
1633 "movq %%mm7, (%1) \n\t"
1634 MMABS(%%mm0, %%mm7)
1635 MMABS_SUM(%%mm1, %%mm7, %%mm0)
1636 MMABS_SUM(%%mm2, %%mm7, %%mm0)
1637 MMABS_SUM(%%mm3, %%mm7, %%mm0)
1638 MMABS_SUM(%%mm4, %%mm7, %%mm0)
1639 MMABS_SUM(%%mm5, %%mm7, %%mm0)
1640 MMABS_SUM(%%mm6, %%mm7, %%mm0)
1641 "movq (%1), %%mm1 \n\t"
1642 MMABS_SUM(%%mm1, %%mm7, %%mm0)
1643 "movq 64(%1), %%mm1 \n\t"
1644 MMABS_SUM(%%mm1, %%mm7, %%mm0)
1645
1646 "movq %%mm0, %%mm1 \n\t"
1647 "psrlq $32, %%mm0 \n\t"
1648 "paddusw %%mm1, %%mm0 \n\t"
1649 "movq %%mm0, %%mm1 \n\t"
1650 "psrlq $16, %%mm0 \n\t"
1651 "paddusw %%mm1, %%mm0 \n\t"
1652 "movd %%mm0, %0 \n\t"
1653
1654 : "=r" (sum)
1655 : "r"(temp)
1656 );
1657 return sum&0xFFFF;
1658}
1659
1660static int hadamard8_diff_mmx2(void *s, uint8_t *src1, uint8_t *src2, int stride, int h){
1661 DECLARE_ALIGNED_8(uint64_t, temp[16]);
1662 int sum=0;
1663
1664 assert(h==8);
1665
1666 diff_pixels_mmx((DCTELEM*)temp, src1, src2, stride);
1667
1668 asm volatile(
1669 LOAD4(0 , %%mm0, %%mm1, %%mm2, %%mm3)
1670 LOAD4(64, %%mm4, %%mm5, %%mm6, %%mm7)
1671
1672 HADAMARD48
1673
1674 "movq %%mm7, 112(%1) \n\t"
1675
1676 TRANSPOSE4(%%mm0, %%mm1, %%mm2, %%mm3, %%mm7)
1677 STORE4(0 , %%mm0, %%mm3, %%mm7, %%mm2)
1678
1679 "movq 112(%1), %%mm7 \n\t"
1680 TRANSPOSE4(%%mm4, %%mm5, %%mm6, %%mm7, %%mm0)
1681 STORE4(64, %%mm4, %%mm7, %%mm0, %%mm6)
1682
1683 LOAD4(8 , %%mm0, %%mm1, %%mm2, %%mm3)
1684 LOAD4(72, %%mm4, %%mm5, %%mm6, %%mm7)
1685
1686 HADAMARD48
1687
1688 "movq %%mm7, 120(%1) \n\t"
1689
1690 TRANSPOSE4(%%mm0, %%mm1, %%mm2, %%mm3, %%mm7)
1691 STORE4(8 , %%mm0, %%mm3, %%mm7, %%mm2)
1692
1693 "movq 120(%1), %%mm7 \n\t"
1694 TRANSPOSE4(%%mm4, %%mm5, %%mm6, %%mm7, %%mm0)
1695 "movq %%mm7, %%mm5 \n\t"//FIXME remove
1696 "movq %%mm6, %%mm7 \n\t"
1697 "movq %%mm0, %%mm6 \n\t"
1698// STORE4(72, %%mm4, %%mm7, %%mm0, %%mm6) //FIXME remove
1699
1700 LOAD4(64, %%mm0, %%mm1, %%mm2, %%mm3)
1701// LOAD4(72, %%mm4, %%mm5, %%mm6, %%mm7)
1702
1703 HADAMARD48
1704 "movq %%mm7, 64(%1) \n\t"
1705 MMABS_MMX2(%%mm0, %%mm7)
1706 MMABS_SUM_MMX2(%%mm1, %%mm7, %%mm0)
1707 MMABS_SUM_MMX2(%%mm2, %%mm7, %%mm0)
1708 MMABS_SUM_MMX2(%%mm3, %%mm7, %%mm0)
1709 MMABS_SUM_MMX2(%%mm4, %%mm7, %%mm0)
1710 MMABS_SUM_MMX2(%%mm5, %%mm7, %%mm0)
1711 MMABS_SUM_MMX2(%%mm6, %%mm7, %%mm0)
1712 "movq 64(%1), %%mm1 \n\t"
1713 MMABS_SUM_MMX2(%%mm1, %%mm7, %%mm0)
1714 "movq %%mm0, 64(%1) \n\t"
1715
1716 LOAD4(0 , %%mm0, %%mm1, %%mm2, %%mm3)
1717 LOAD4(8 , %%mm4, %%mm5, %%mm6, %%mm7)
1718
1719 HADAMARD48
1720 "movq %%mm7, (%1) \n\t"
1721 MMABS_MMX2(%%mm0, %%mm7)
1722 MMABS_SUM_MMX2(%%mm1, %%mm7, %%mm0)
1723 MMABS_SUM_MMX2(%%mm2, %%mm7, %%mm0)
1724 MMABS_SUM_MMX2(%%mm3, %%mm7, %%mm0)
1725 MMABS_SUM_MMX2(%%mm4, %%mm7, %%mm0)
1726 MMABS_SUM_MMX2(%%mm5, %%mm7, %%mm0)
1727 MMABS_SUM_MMX2(%%mm6, %%mm7, %%mm0)
1728 "movq (%1), %%mm1 \n\t"
1729 MMABS_SUM_MMX2(%%mm1, %%mm7, %%mm0)
1730 "movq 64(%1), %%mm1 \n\t"
1731 MMABS_SUM_MMX2(%%mm1, %%mm7, %%mm0)
1732
1733 "pshufw $0x0E, %%mm0, %%mm1 \n\t"
1734 "paddusw %%mm1, %%mm0 \n\t"
1735 "pshufw $0x01, %%mm0, %%mm1 \n\t"
1736 "paddusw %%mm1, %%mm0 \n\t"
1737 "movd %%mm0, %0 \n\t"
1738
1739 : "=r" (sum)
1740 : "r"(temp)
1741 );
1742 return sum&0xFFFF;
1743}
1744
1745
1746WARPER8_16_SQ(hadamard8_diff_mmx, hadamard8_diff16_mmx)
1747WARPER8_16_SQ(hadamard8_diff_mmx2, hadamard8_diff16_mmx2)
1748#endif //CONFIG_ENCODERS
1749
1750#define put_no_rnd_pixels8_mmx(a,b,c,d) put_pixels8_mmx(a,b,c,d)
1751#define put_no_rnd_pixels16_mmx(a,b,c,d) put_pixels16_mmx(a,b,c,d)
1752
1753#define QPEL_V_LOW(m3,m4,m5,m6, pw_20, pw_3, rnd, in0, in1, in2, in7, out, OP)\
1754 "paddw " #m4 ", " #m3 " \n\t" /* x1 */\
1755 "movq "MANGLE(ff_pw_20)", %%mm4 \n\t" /* 20 */\
1756 "pmullw " #m3 ", %%mm4 \n\t" /* 20x1 */\
1757 "movq "#in7", " #m3 " \n\t" /* d */\
1758 "movq "#in0", %%mm5 \n\t" /* D */\
1759 "paddw " #m3 ", %%mm5 \n\t" /* x4 */\
1760 "psubw %%mm5, %%mm4 \n\t" /* 20x1 - x4 */\
1761 "movq "#in1", %%mm5 \n\t" /* C */\
1762 "movq "#in2", %%mm6 \n\t" /* B */\
1763 "paddw " #m6 ", %%mm5 \n\t" /* x3 */\
1764 "paddw " #m5 ", %%mm6 \n\t" /* x2 */\
1765 "paddw %%mm6, %%mm6 \n\t" /* 2x2 */\
1766 "psubw %%mm6, %%mm5 \n\t" /* -2x2 + x3 */\
1767 "pmullw "MANGLE(ff_pw_3)", %%mm5 \n\t" /* -6x2 + 3x3 */\
1768 "paddw " #rnd ", %%mm4 \n\t" /* x2 */\
1769 "paddw %%mm4, %%mm5 \n\t" /* 20x1 - 6x2 + 3x3 - x4 */\
1770 "psraw $5, %%mm5 \n\t"\
1771 "packuswb %%mm5, %%mm5 \n\t"\
1772 OP(%%mm5, out, %%mm7, d)
1773
1774#define QPEL_BASE(OPNAME, ROUNDER, RND, OP_MMX2, OP_3DNOW)\
1775static void OPNAME ## mpeg4_qpel16_h_lowpass_mmx2(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h){\
1776 uint64_t temp;\
1777\
1778 asm volatile(\
1779 "pxor %%mm7, %%mm7 \n\t"\
1780 "1: \n\t"\
1781 "movq (%0), %%mm0 \n\t" /* ABCDEFGH */\
1782 "movq %%mm0, %%mm1 \n\t" /* ABCDEFGH */\
1783 "movq %%mm0, %%mm2 \n\t" /* ABCDEFGH */\
1784 "punpcklbw %%mm7, %%mm0 \n\t" /* 0A0B0C0D */\
1785 "punpckhbw %%mm7, %%mm1 \n\t" /* 0E0F0G0H */\
1786 "pshufw $0x90, %%mm0, %%mm5 \n\t" /* 0A0A0B0C */\
1787 "pshufw $0x41, %%mm0, %%mm6 \n\t" /* 0B0A0A0B */\
1788 "movq %%mm2, %%mm3 \n\t" /* ABCDEFGH */\
1789 "movq %%mm2, %%mm4 \n\t" /* ABCDEFGH */\
1790 "psllq $8, %%mm2 \n\t" /* 0ABCDEFG */\
1791 "psllq $16, %%mm3 \n\t" /* 00ABCDEF */\
1792 "psllq $24, %%mm4 \n\t" /* 000ABCDE */\
1793 "punpckhbw %%mm7, %%mm2 \n\t" /* 0D0E0F0G */\
1794 "punpckhbw %%mm7, %%mm3 \n\t" /* 0C0D0E0F */\
1795 "punpckhbw %%mm7, %%mm4 \n\t" /* 0B0C0D0E */\
1796 "paddw %%mm3, %%mm5 \n\t" /* b */\
1797 "paddw %%mm2, %%mm6 \n\t" /* c */\
1798 "paddw %%mm5, %%mm5 \n\t" /* 2b */\
1799 "psubw %%mm5, %%mm6 \n\t" /* c - 2b */\
1800 "pshufw $0x06, %%mm0, %%mm5 \n\t" /* 0C0B0A0A */\
1801 "pmullw "MANGLE(ff_pw_3)", %%mm6 \n\t" /* 3c - 6b */\
1802 "paddw %%mm4, %%mm0 \n\t" /* a */\
1803 "paddw %%mm1, %%mm5 \n\t" /* d */\
1804 "pmullw "MANGLE(ff_pw_20)", %%mm0 \n\t" /* 20a */\
1805 "psubw %%mm5, %%mm0 \n\t" /* 20a - d */\
1806 "paddw %6, %%mm6 \n\t"\
1807 "paddw %%mm6, %%mm0 \n\t" /* 20a - 6b + 3c - d */\
1808 "psraw $5, %%mm0 \n\t"\
1809 "movq %%mm0, %5 \n\t"\
1810 /* mm1=EFGH, mm2=DEFG, mm3=CDEF, mm4=BCDE, mm7=0 */\
1811 \
1812 "movq 5(%0), %%mm0 \n\t" /* FGHIJKLM */\
1813 "movq %%mm0, %%mm5 \n\t" /* FGHIJKLM */\
1814 "movq %%mm0, %%mm6 \n\t" /* FGHIJKLM */\
1815 "psrlq $8, %%mm0 \n\t" /* GHIJKLM0 */\
1816 "psrlq $16, %%mm5 \n\t" /* HIJKLM00 */\
1817 "punpcklbw %%mm7, %%mm0 \n\t" /* 0G0H0I0J */\
1818 "punpcklbw %%mm7, %%mm5 \n\t" /* 0H0I0J0K */\
1819 "paddw %%mm0, %%mm2 \n\t" /* b */\
1820 "paddw %%mm5, %%mm3 \n\t" /* c */\
1821 "paddw %%mm2, %%mm2 \n\t" /* 2b */\
1822 "psubw %%mm2, %%mm3 \n\t" /* c - 2b */\
1823 "movq %%mm6, %%mm2 \n\t" /* FGHIJKLM */\
1824 "psrlq $24, %%mm6 \n\t" /* IJKLM000 */\
1825 "punpcklbw %%mm7, %%mm2 \n\t" /* 0F0G0H0I */\
1826 "punpcklbw %%mm7, %%mm6 \n\t" /* 0I0J0K0L */\
1827 "pmullw "MANGLE(ff_pw_3)", %%mm3 \n\t" /* 3c - 6b */\
1828 "paddw %%mm2, %%mm1 \n\t" /* a */\
1829 "paddw %%mm6, %%mm4 \n\t" /* d */\
1830 "pmullw "MANGLE(ff_pw_20)", %%mm1 \n\t" /* 20a */\
1831 "psubw %%mm4, %%mm3 \n\t" /* - 6b +3c - d */\
1832 "paddw %6, %%mm1 \n\t"\
1833 "paddw %%mm1, %%mm3 \n\t" /* 20a - 6b +3c - d */\
1834 "psraw $5, %%mm3 \n\t"\
1835 "movq %5, %%mm1 \n\t"\
1836 "packuswb %%mm3, %%mm1 \n\t"\
1837 OP_MMX2(%%mm1, (%1),%%mm4, q)\
1838 /* mm0= GHIJ, mm2=FGHI, mm5=HIJK, mm6=IJKL, mm7=0 */\
1839 \
1840 "movq 9(%0), %%mm1 \n\t" /* JKLMNOPQ */\
1841 "movq %%mm1, %%mm4 \n\t" /* JKLMNOPQ */\
1842 "movq %%mm1, %%mm3 \n\t" /* JKLMNOPQ */\
1843 "psrlq $8, %%mm1 \n\t" /* KLMNOPQ0 */\
1844 "psrlq $16, %%mm4 \n\t" /* LMNOPQ00 */\
1845 "punpcklbw %%mm7, %%mm1 \n\t" /* 0K0L0M0N */\
1846 "punpcklbw %%mm7, %%mm4 \n\t" /* 0L0M0N0O */\
1847 "paddw %%mm1, %%mm5 \n\t" /* b */\
1848 "paddw %%mm4, %%mm0 \n\t" /* c */\
1849 "paddw %%mm5, %%mm5 \n\t" /* 2b */\
1850 "psubw %%mm5, %%mm0 \n\t" /* c - 2b */\
1851 "movq %%mm3, %%mm5 \n\t" /* JKLMNOPQ */\
1852 "psrlq $24, %%mm3 \n\t" /* MNOPQ000 */\
1853 "pmullw "MANGLE(ff_pw_3)", %%mm0 \n\t" /* 3c - 6b */\
1854 "punpcklbw %%mm7, %%mm3 \n\t" /* 0M0N0O0P */\
1855 "paddw %%mm3, %%mm2 \n\t" /* d */\
1856 "psubw %%mm2, %%mm0 \n\t" /* -6b + 3c - d */\
1857 "movq %%mm5, %%mm2 \n\t" /* JKLMNOPQ */\
1858 "punpcklbw %%mm7, %%mm2 \n\t" /* 0J0K0L0M */\
1859 "punpckhbw %%mm7, %%mm5 \n\t" /* 0N0O0P0Q */\
1860 "paddw %%mm2, %%mm6 \n\t" /* a */\
1861 "pmullw "MANGLE(ff_pw_20)", %%mm6 \n\t" /* 20a */\
1862 "paddw %6, %%mm0 \n\t"\
1863 "paddw %%mm6, %%mm0 \n\t" /* 20a - 6b + 3c - d */\
1864 "psraw $5, %%mm0 \n\t"\
1865 /* mm1=KLMN, mm2=JKLM, mm3=MNOP, mm4=LMNO, mm5=NOPQ mm7=0 */\
1866 \
1867 "paddw %%mm5, %%mm3 \n\t" /* a */\
1868 "pshufw $0xF9, %%mm5, %%mm6 \n\t" /* 0O0P0Q0Q */\
1869 "paddw %%mm4, %%mm6 \n\t" /* b */\
1870 "pshufw $0xBE, %%mm5, %%mm4 \n\t" /* 0P0Q0Q0P */\
1871 "pshufw $0x6F, %%mm5, %%mm5 \n\t" /* 0Q0Q0P0O */\
1872 "paddw %%mm1, %%mm4 \n\t" /* c */\
1873 "paddw %%mm2, %%mm5 \n\t" /* d */\
1874 "paddw %%mm6, %%mm6 \n\t" /* 2b */\
1875 "psubw %%mm6, %%mm4 \n\t" /* c - 2b */\
1876 "pmullw "MANGLE(ff_pw_20)", %%mm3 \n\t" /* 20a */\
1877 "pmullw "MANGLE(ff_pw_3)", %%mm4 \n\t" /* 3c - 6b */\
1878 "psubw %%mm5, %%mm3 \n\t" /* -6b + 3c - d */\
1879 "paddw %6, %%mm4 \n\t"\
1880 "paddw %%mm3, %%mm4 \n\t" /* 20a - 6b + 3c - d */\
1881 "psraw $5, %%mm4 \n\t"\
1882 "packuswb %%mm4, %%mm0 \n\t"\
1883 OP_MMX2(%%mm0, 8(%1), %%mm4, q)\
1884 \
1885 "add %3, %0 \n\t"\
1886 "add %4, %1 \n\t"\
1887 "decl %2 \n\t"\
1888 " jnz 1b \n\t"\
1889 : "+a"(src), "+c"(dst), "+m"(h)\
1890 : "d"((long)srcStride), "S"((long)dstStride), /*"m"(ff_pw_20), "m"(ff_pw_3),*/ "m"(temp), "m"(ROUNDER)\
1891 : "memory"\
1892 );\
1893}\
1894\
1895static void OPNAME ## mpeg4_qpel16_h_lowpass_3dnow(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h){\
1896 int i;\
1897 int16_t temp[16];\
1898 /* quick HACK, XXX FIXME MUST be optimized */\
1899 for(i=0; i<h; i++)\
1900 {\
1901 temp[ 0]= (src[ 0]+src[ 1])*20 - (src[ 0]+src[ 2])*6 + (src[ 1]+src[ 3])*3 - (src[ 2]+src[ 4]);\
1902 temp[ 1]= (src[ 1]+src[ 2])*20 - (src[ 0]+src[ 3])*6 + (src[ 0]+src[ 4])*3 - (src[ 1]+src[ 5]);\
1903 temp[ 2]= (src[ 2]+src[ 3])*20 - (src[ 1]+src[ 4])*6 + (src[ 0]+src[ 5])*3 - (src[ 0]+src[ 6]);\
1904 temp[ 3]= (src[ 3]+src[ 4])*20 - (src[ 2]+src[ 5])*6 + (src[ 1]+src[ 6])*3 - (src[ 0]+src[ 7]);\
1905 temp[ 4]= (src[ 4]+src[ 5])*20 - (src[ 3]+src[ 6])*6 + (src[ 2]+src[ 7])*3 - (src[ 1]+src[ 8]);\
1906 temp[ 5]= (src[ 5]+src[ 6])*20 - (src[ 4]+src[ 7])*6 + (src[ 3]+src[ 8])*3 - (src[ 2]+src[ 9]);\
1907 temp[ 6]= (src[ 6]+src[ 7])*20 - (src[ 5]+src[ 8])*6 + (src[ 4]+src[ 9])*3 - (src[ 3]+src[10]);\
1908 temp[ 7]= (src[ 7]+src[ 8])*20 - (src[ 6]+src[ 9])*6 + (src[ 5]+src[10])*3 - (src[ 4]+src[11]);\
1909 temp[ 8]= (src[ 8]+src[ 9])*20 - (src[ 7]+src[10])*6 + (src[ 6]+src[11])*3 - (src[ 5]+src[12]);\
1910 temp[ 9]= (src[ 9]+src[10])*20 - (src[ 8]+src[11])*6 + (src[ 7]+src[12])*3 - (src[ 6]+src[13]);\
1911 temp[10]= (src[10]+src[11])*20 - (src[ 9]+src[12])*6 + (src[ 8]+src[13])*3 - (src[ 7]+src[14]);\
1912 temp[11]= (src[11]+src[12])*20 - (src[10]+src[13])*6 + (src[ 9]+src[14])*3 - (src[ 8]+src[15]);\
1913 temp[12]= (src[12]+src[13])*20 - (src[11]+src[14])*6 + (src[10]+src[15])*3 - (src[ 9]+src[16]);\
1914 temp[13]= (src[13]+src[14])*20 - (src[12]+src[15])*6 + (src[11]+src[16])*3 - (src[10]+src[16]);\
1915 temp[14]= (src[14]+src[15])*20 - (src[13]+src[16])*6 + (src[12]+src[16])*3 - (src[11]+src[15]);\
1916 temp[15]= (src[15]+src[16])*20 - (src[14]+src[16])*6 + (src[13]+src[15])*3 - (src[12]+src[14]);\
1917 asm volatile(\
1918 "movq (%0), %%mm0 \n\t"\
1919 "movq 8(%0), %%mm1 \n\t"\
1920 "paddw %2, %%mm0 \n\t"\
1921 "paddw %2, %%mm1 \n\t"\
1922 "psraw $5, %%mm0 \n\t"\
1923 "psraw $5, %%mm1 \n\t"\
1924 "packuswb %%mm1, %%mm0 \n\t"\
1925 OP_3DNOW(%%mm0, (%1), %%mm1, q)\
1926 "movq 16(%0), %%mm0 \n\t"\
1927 "movq 24(%0), %%mm1 \n\t"\
1928 "paddw %2, %%mm0 \n\t"\
1929 "paddw %2, %%mm1 \n\t"\
1930 "psraw $5, %%mm0 \n\t"\
1931 "psraw $5, %%mm1 \n\t"\
1932 "packuswb %%mm1, %%mm0 \n\t"\
1933 OP_3DNOW(%%mm0, 8(%1), %%mm1, q)\
1934 :: "r"(temp), "r"(dst), "m"(ROUNDER)\
1935 : "memory"\
1936 );\
1937 dst+=dstStride;\
1938 src+=srcStride;\
1939 }\
1940}\
1941\
1942static void OPNAME ## mpeg4_qpel8_h_lowpass_mmx2(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h){\
1943 uint64_t temp;\
1944\
1945 asm volatile(\
1946 "pxor %%mm7, %%mm7 \n\t"\
1947 "1: \n\t"\
1948 "movq (%0), %%mm0 \n\t" /* ABCDEFGH */\
1949 "movq %%mm0, %%mm1 \n\t" /* ABCDEFGH */\
1950 "movq %%mm0, %%mm2 \n\t" /* ABCDEFGH */\
1951 "punpcklbw %%mm7, %%mm0 \n\t" /* 0A0B0C0D */\
1952 "punpckhbw %%mm7, %%mm1 \n\t" /* 0E0F0G0H */\
1953 "pshufw $0x90, %%mm0, %%mm5 \n\t" /* 0A0A0B0C */\
1954 "pshufw $0x41, %%mm0, %%mm6 \n\t" /* 0B0A0A0B */\
1955 "movq %%mm2, %%mm3 \n\t" /* ABCDEFGH */\
1956 "movq %%mm2, %%mm4 \n\t" /* ABCDEFGH */\
1957 "psllq $8, %%mm2 \n\t" /* 0ABCDEFG */\
1958 "psllq $16, %%mm3 \n\t" /* 00ABCDEF */\
1959 "psllq $24, %%mm4 \n\t" /* 000ABCDE */\
1960 "punpckhbw %%mm7, %%mm2 \n\t" /* 0D0E0F0G */\
1961 "punpckhbw %%mm7, %%mm3 \n\t" /* 0C0D0E0F */\
1962 "punpckhbw %%mm7, %%mm4 \n\t" /* 0B0C0D0E */\
1963 "paddw %%mm3, %%mm5 \n\t" /* b */\
1964 "paddw %%mm2, %%mm6 \n\t" /* c */\
1965 "paddw %%mm5, %%mm5 \n\t" /* 2b */\
1966 "psubw %%mm5, %%mm6 \n\t" /* c - 2b */\
1967 "pshufw $0x06, %%mm0, %%mm5 \n\t" /* 0C0B0A0A */\
1968 "pmullw "MANGLE(ff_pw_3)", %%mm6 \n\t" /* 3c - 6b */\
1969 "paddw %%mm4, %%mm0 \n\t" /* a */\
1970 "paddw %%mm1, %%mm5 \n\t" /* d */\
1971 "pmullw "MANGLE(ff_pw_20)", %%mm0 \n\t" /* 20a */\
1972 "psubw %%mm5, %%mm0 \n\t" /* 20a - d */\
1973 "paddw %6, %%mm6 \n\t"\
1974 "paddw %%mm6, %%mm0 \n\t" /* 20a - 6b + 3c - d */\
1975 "psraw $5, %%mm0 \n\t"\
1976 /* mm1=EFGH, mm2=DEFG, mm3=CDEF, mm4=BCDE, mm7=0 */\
1977 \
1978 "movd 5(%0), %%mm5 \n\t" /* FGHI */\
1979 "punpcklbw %%mm7, %%mm5 \n\t" /* 0F0G0H0I */\
1980 "pshufw $0xF9, %%mm5, %%mm6 \n\t" /* 0G0H0I0I */\
1981 "paddw %%mm5, %%mm1 \n\t" /* a */\
1982 "paddw %%mm6, %%mm2 \n\t" /* b */\
1983 "pshufw $0xBE, %%mm5, %%mm6 \n\t" /* 0H0I0I0H */\
1984 "pshufw $0x6F, %%mm5, %%mm5 \n\t" /* 0I0I0H0G */\
1985 "paddw %%mm6, %%mm3 \n\t" /* c */\
1986 "paddw %%mm5, %%mm4 \n\t" /* d */\
1987 "paddw %%mm2, %%mm2 \n\t" /* 2b */\
1988 "psubw %%mm2, %%mm3 \n\t" /* c - 2b */\
1989 "pmullw "MANGLE(ff_pw_20)", %%mm1 \n\t" /* 20a */\
1990 "pmullw "MANGLE(ff_pw_3)", %%mm3 \n\t" /* 3c - 6b */\
1991 "psubw %%mm4, %%mm3 \n\t" /* -6b + 3c - d */\
1992 "paddw %6, %%mm1 \n\t"\
1993 "paddw %%mm1, %%mm3 \n\t" /* 20a - 6b + 3c - d */\
1994 "psraw $5, %%mm3 \n\t"\
1995 "packuswb %%mm3, %%mm0 \n\t"\
1996 OP_MMX2(%%mm0, (%1), %%mm4, q)\
1997 \
1998 "add %3, %0 \n\t"\
1999 "add %4, %1 \n\t"\
2000 "decl %2 \n\t"\
2001 " jnz 1b \n\t"\
2002 : "+a"(src), "+c"(dst), "+m"(h)\
2003 : "S"((long)srcStride), "D"((long)dstStride), /*"m"(ff_pw_20), "m"(ff_pw_3),*/ "m"(temp), "m"(ROUNDER)\
2004 : "memory"\
2005 );\
2006}\
2007\
2008static void OPNAME ## mpeg4_qpel8_h_lowpass_3dnow(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h){\
2009 int i;\
2010 int16_t temp[8];\
2011 /* quick HACK, XXX FIXME MUST be optimized */\
2012 for(i=0; i<h; i++)\
2013 {\
2014 temp[ 0]= (src[ 0]+src[ 1])*20 - (src[ 0]+src[ 2])*6 + (src[ 1]+src[ 3])*3 - (src[ 2]+src[ 4]);\
2015 temp[ 1]= (src[ 1]+src[ 2])*20 - (src[ 0]+src[ 3])*6 + (src[ 0]+src[ 4])*3 - (src[ 1]+src[ 5]);\
2016 temp[ 2]= (src[ 2]+src[ 3])*20 - (src[ 1]+src[ 4])*6 + (src[ 0]+src[ 5])*3 - (src[ 0]+src[ 6]);\
2017 temp[ 3]= (src[ 3]+src[ 4])*20 - (src[ 2]+src[ 5])*6 + (src[ 1]+src[ 6])*3 - (src[ 0]+src[ 7]);\
2018 temp[ 4]= (src[ 4]+src[ 5])*20 - (src[ 3]+src[ 6])*6 + (src[ 2]+src[ 7])*3 - (src[ 1]+src[ 8]);\
2019 temp[ 5]= (src[ 5]+src[ 6])*20 - (src[ 4]+src[ 7])*6 + (src[ 3]+src[ 8])*3 - (src[ 2]+src[ 8]);\
2020 temp[ 6]= (src[ 6]+src[ 7])*20 - (src[ 5]+src[ 8])*6 + (src[ 4]+src[ 8])*3 - (src[ 3]+src[ 7]);\
2021 temp[ 7]= (src[ 7]+src[ 8])*20 - (src[ 6]+src[ 8])*6 + (src[ 5]+src[ 7])*3 - (src[ 4]+src[ 6]);\
2022 asm volatile(\
2023 "movq (%0), %%mm0 \n\t"\
2024 "movq 8(%0), %%mm1 \n\t"\
2025 "paddw %2, %%mm0 \n\t"\
2026 "paddw %2, %%mm1 \n\t"\
2027 "psraw $5, %%mm0 \n\t"\
2028 "psraw $5, %%mm1 \n\t"\
2029 "packuswb %%mm1, %%mm0 \n\t"\
2030 OP_3DNOW(%%mm0, (%1), %%mm1, q)\
2031 :: "r"(temp), "r"(dst), "m"(ROUNDER)\
2032 :"memory"\
2033 );\
2034 dst+=dstStride;\
2035 src+=srcStride;\
2036 }\
2037}
2038
2039#define QPEL_OP(OPNAME, ROUNDER, RND, OP, MMX)\
2040\
2041static void OPNAME ## mpeg4_qpel16_v_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
2042 uint64_t temp[17*4];\
2043 uint64_t *temp_ptr= temp;\
2044 int count= 17;\
2045\
2046 /*FIXME unroll */\
2047 asm volatile(\
2048 "pxor %%mm7, %%mm7 \n\t"\
2049 "1: \n\t"\
2050 "movq (%0), %%mm0 \n\t"\
2051 "movq (%0), %%mm1 \n\t"\
2052 "movq 8(%0), %%mm2 \n\t"\
2053 "movq 8(%0), %%mm3 \n\t"\
2054 "punpcklbw %%mm7, %%mm0 \n\t"\
2055 "punpckhbw %%mm7, %%mm1 \n\t"\
2056 "punpcklbw %%mm7, %%mm2 \n\t"\
2057 "punpckhbw %%mm7, %%mm3 \n\t"\
2058 "movq %%mm0, (%1) \n\t"\
2059 "movq %%mm1, 17*8(%1) \n\t"\
2060 "movq %%mm2, 2*17*8(%1) \n\t"\
2061 "movq %%mm3, 3*17*8(%1) \n\t"\
2062 "add $8, %1 \n\t"\
2063 "add %3, %0 \n\t"\
2064 "decl %2 \n\t"\
2065 " jnz 1b \n\t"\
2066 : "+r" (src), "+r" (temp_ptr), "+r"(count)\
2067 : "r" ((long)srcStride)\
2068 : "memory"\
2069 );\
2070 \
2071 temp_ptr= temp;\
2072 count=4;\
2073 \
2074/*FIXME reorder for speed */\
2075 asm volatile(\
2076 /*"pxor %%mm7, %%mm7 \n\t"*/\
2077 "1: \n\t"\
2078 "movq (%0), %%mm0 \n\t"\
2079 "movq 8(%0), %%mm1 \n\t"\
2080 "movq 16(%0), %%mm2 \n\t"\
2081 "movq 24(%0), %%mm3 \n\t"\
2082 QPEL_V_LOW(%%mm0, %%mm1, %%mm2, %%mm3, %5, %6, %5, 16(%0), 8(%0), (%0), 32(%0), (%1), OP)\
2083 QPEL_V_LOW(%%mm1, %%mm2, %%mm3, %%mm0, %5, %6, %5, 8(%0), (%0), (%0), 40(%0), (%1, %3), OP)\
2084 "add %4, %1 \n\t"\
2085 QPEL_V_LOW(%%mm2, %%mm3, %%mm0, %%mm1, %5, %6, %5, (%0), (%0), 8(%0), 48(%0), (%1), OP)\
2086 \
2087 QPEL_V_LOW(%%mm3, %%mm0, %%mm1, %%mm2, %5, %6, %5, (%0), 8(%0), 16(%0), 56(%0), (%1, %3), OP)\
2088 "add %4, %1 \n\t"\
2089 QPEL_V_LOW(%%mm0, %%mm1, %%mm2, %%mm3, %5, %6, %5, 8(%0), 16(%0), 24(%0), 64(%0), (%1), OP)\
2090 QPEL_V_LOW(%%mm1, %%mm2, %%mm3, %%mm0, %5, %6, %5, 16(%0), 24(%0), 32(%0), 72(%0), (%1, %3), OP)\
2091 "add %4, %1 \n\t"\
2092 QPEL_V_LOW(%%mm2, %%mm3, %%mm0, %%mm1, %5, %6, %5, 24(%0), 32(%0), 40(%0), 80(%0), (%1), OP)\
2093 QPEL_V_LOW(%%mm3, %%mm0, %%mm1, %%mm2, %5, %6, %5, 32(%0), 40(%0), 48(%0), 88(%0), (%1, %3), OP)\
2094 "add %4, %1 \n\t"\
2095 QPEL_V_LOW(%%mm0, %%mm1, %%mm2, %%mm3, %5, %6, %5, 40(%0), 48(%0), 56(%0), 96(%0), (%1), OP)\
2096 QPEL_V_LOW(%%mm1, %%mm2, %%mm3, %%mm0, %5, %6, %5, 48(%0), 56(%0), 64(%0),104(%0), (%1, %3), OP)\
2097 "add %4, %1 \n\t"\
2098 QPEL_V_LOW(%%mm2, %%mm3, %%mm0, %%mm1, %5, %6, %5, 56(%0), 64(%0), 72(%0),112(%0), (%1), OP)\
2099 QPEL_V_LOW(%%mm3, %%mm0, %%mm1, %%mm2, %5, %6, %5, 64(%0), 72(%0), 80(%0),120(%0), (%1, %3), OP)\
2100 "add %4, %1 \n\t"\
2101 QPEL_V_LOW(%%mm0, %%mm1, %%mm2, %%mm3, %5, %6, %5, 72(%0), 80(%0), 88(%0),128(%0), (%1), OP)\
2102 \
2103 QPEL_V_LOW(%%mm1, %%mm2, %%mm3, %%mm0, %5, %6, %5, 80(%0), 88(%0), 96(%0),128(%0), (%1, %3), OP)\
2104 "add %4, %1 \n\t" \
2105 QPEL_V_LOW(%%mm2, %%mm3, %%mm0, %%mm1, %5, %6, %5, 88(%0), 96(%0),104(%0),120(%0), (%1), OP)\
2106 QPEL_V_LOW(%%mm3, %%mm0, %%mm1, %%mm2, %5, %6, %5, 96(%0),104(%0),112(%0),112(%0), (%1, %3), OP)\
2107 \
2108 "add $136, %0 \n\t"\
2109 "add %6, %1 \n\t"\
2110 "decl %2 \n\t"\
2111 " jnz 1b \n\t"\
2112 \
2113 : "+r"(temp_ptr), "+r"(dst), "+g"(count)\
2114 : "r"((long)dstStride), "r"(2*(long)dstStride), /*"m"(ff_pw_20), "m"(ff_pw_3),*/ "m"(ROUNDER), "g"(4-14*(long)dstStride)\
2115 :"memory"\
2116 );\
2117}\
2118\
2119static void OPNAME ## mpeg4_qpel8_v_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
2120 uint64_t temp[9*2];\
2121 uint64_t *temp_ptr= temp;\
2122 int count= 9;\
2123\
2124 /*FIXME unroll */\
2125 asm volatile(\
2126 "pxor %%mm7, %%mm7 \n\t"\
2127 "1: \n\t"\
2128 "movq (%0), %%mm0 \n\t"\
2129 "movq (%0), %%mm1 \n\t"\
2130 "punpcklbw %%mm7, %%mm0 \n\t"\
2131 "punpckhbw %%mm7, %%mm1 \n\t"\
2132 "movq %%mm0, (%1) \n\t"\
2133 "movq %%mm1, 9*8(%1) \n\t"\
2134 "add $8, %1 \n\t"\
2135 "add %3, %0 \n\t"\
2136 "decl %2 \n\t"\
2137 " jnz 1b \n\t"\
2138 : "+r" (src), "+r" (temp_ptr), "+r"(count)\
2139 : "r" ((long)srcStride)\
2140 : "memory"\
2141 );\
2142 \
2143 temp_ptr= temp;\
2144 count=2;\
2145 \
2146/*FIXME reorder for speed */\
2147 asm volatile(\
2148 /*"pxor %%mm7, %%mm7 \n\t"*/\
2149 "1: \n\t"\
2150 "movq (%0), %%mm0 \n\t"\
2151 "movq 8(%0), %%mm1 \n\t"\
2152 "movq 16(%0), %%mm2 \n\t"\
2153 "movq 24(%0), %%mm3 \n\t"\
2154 QPEL_V_LOW(%%mm0, %%mm1, %%mm2, %%mm3, %5, %6, %5, 16(%0), 8(%0), (%0), 32(%0), (%1), OP)\
2155 QPEL_V_LOW(%%mm1, %%mm2, %%mm3, %%mm0, %5, %6, %5, 8(%0), (%0), (%0), 40(%0), (%1, %3), OP)\
2156 "add %4, %1 \n\t"\
2157 QPEL_V_LOW(%%mm2, %%mm3, %%mm0, %%mm1, %5, %6, %5, (%0), (%0), 8(%0), 48(%0), (%1), OP)\
2158 \
2159 QPEL_V_LOW(%%mm3, %%mm0, %%mm1, %%mm2, %5, %6, %5, (%0), 8(%0), 16(%0), 56(%0), (%1, %3), OP)\
2160 "add %4, %1 \n\t"\
2161 QPEL_V_LOW(%%mm0, %%mm1, %%mm2, %%mm3, %5, %6, %5, 8(%0), 16(%0), 24(%0), 64(%0), (%1), OP)\
2162 \
2163 QPEL_V_LOW(%%mm1, %%mm2, %%mm3, %%mm0, %5, %6, %5, 16(%0), 24(%0), 32(%0), 64(%0), (%1, %3), OP)\
2164 "add %4, %1 \n\t"\
2165 QPEL_V_LOW(%%mm2, %%mm3, %%mm0, %%mm1, %5, %6, %5, 24(%0), 32(%0), 40(%0), 56(%0), (%1), OP)\
2166 QPEL_V_LOW(%%mm3, %%mm0, %%mm1, %%mm2, %5, %6, %5, 32(%0), 40(%0), 48(%0), 48(%0), (%1, %3), OP)\
2167 \
2168 "add $72, %0 \n\t"\
2169 "add %6, %1 \n\t"\
2170 "decl %2 \n\t"\
2171 " jnz 1b \n\t"\
2172 \
2173 : "+r"(temp_ptr), "+r"(dst), "+g"(count)\
2174 : "r"((long)dstStride), "r"(2*(long)dstStride), /*"m"(ff_pw_20), "m"(ff_pw_3),*/ "m"(ROUNDER), "g"(4-6*(long)dstStride)\
2175 : "memory"\
2176 );\
2177}\
2178\
2179static void OPNAME ## qpel8_mc00_ ## MMX (uint8_t *dst, uint8_t *src, int stride){\
2180 OPNAME ## pixels8_mmx(dst, src, stride, 8);\
2181}\
2182\
2183static void OPNAME ## qpel8_mc10_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
2184 uint64_t temp[8];\
2185 uint8_t * const half= (uint8_t*)temp;\
2186 put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(half, src, 8, stride, 8);\
2187 OPNAME ## pixels8_l2_ ## MMX(dst, src, half, stride, stride, 8);\
2188}\
2189\
2190static void OPNAME ## qpel8_mc20_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
2191 OPNAME ## mpeg4_qpel8_h_lowpass_ ## MMX(dst, src, stride, stride, 8);\
2192}\
2193\
2194static void OPNAME ## qpel8_mc30_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
2195 uint64_t temp[8];\
2196 uint8_t * const half= (uint8_t*)temp;\
2197 put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(half, src, 8, stride, 8);\
2198 OPNAME ## pixels8_l2_ ## MMX(dst, src+1, half, stride, stride, 8);\
2199}\
2200\
2201static void OPNAME ## qpel8_mc01_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
2202 uint64_t temp[8];\
2203 uint8_t * const half= (uint8_t*)temp;\
2204 put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(half, src, 8, stride);\
2205 OPNAME ## pixels8_l2_ ## MMX(dst, src, half, stride, stride, 8);\
2206}\
2207\
2208static void OPNAME ## qpel8_mc02_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
2209 OPNAME ## mpeg4_qpel8_v_lowpass_ ## MMX(dst, src, stride, stride);\
2210}\
2211\
2212static void OPNAME ## qpel8_mc03_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
2213 uint64_t temp[8];\
2214 uint8_t * const half= (uint8_t*)temp;\
2215 put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(half, src, 8, stride);\
2216 OPNAME ## pixels8_l2_ ## MMX(dst, src+stride, half, stride, stride, 8);\
2217}\
2218static void OPNAME ## qpel8_mc11_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
2219 uint64_t half[8 + 9];\
2220 uint8_t * const halfH= ((uint8_t*)half) + 64;\
2221 uint8_t * const halfHV= ((uint8_t*)half);\
2222 put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, stride, 9);\
2223 put ## RND ## pixels8_l2_ ## MMX(halfH, src, halfH, 8, stride, 9);\
2224 put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(halfHV, halfH, 8, 8);\
2225 OPNAME ## pixels8_l2_ ## MMX(dst, halfH, halfHV, stride, 8, 8);\
2226}\
2227static void OPNAME ## qpel8_mc31_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
2228 uint64_t half[8 + 9];\
2229 uint8_t * const halfH= ((uint8_t*)half) + 64;\
2230 uint8_t * const halfHV= ((uint8_t*)half);\
2231 put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, stride, 9);\
2232 put ## RND ## pixels8_l2_ ## MMX(halfH, src+1, halfH, 8, stride, 9);\
2233 put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(halfHV, halfH, 8, 8);\
2234 OPNAME ## pixels8_l2_ ## MMX(dst, halfH, halfHV, stride, 8, 8);\
2235}\
2236static void OPNAME ## qpel8_mc13_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
2237 uint64_t half[8 + 9];\
2238 uint8_t * const halfH= ((uint8_t*)half) + 64;\
2239 uint8_t * const halfHV= ((uint8_t*)half);\
2240 put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, stride, 9);\
2241 put ## RND ## pixels8_l2_ ## MMX(halfH, src, halfH, 8, stride, 9);\
2242 put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(halfHV, halfH, 8, 8);\
2243 OPNAME ## pixels8_l2_ ## MMX(dst, halfH+8, halfHV, stride, 8, 8);\
2244}\
2245static void OPNAME ## qpel8_mc33_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
2246 uint64_t half[8 + 9];\
2247 uint8_t * const halfH= ((uint8_t*)half) + 64;\
2248 uint8_t * const halfHV= ((uint8_t*)half);\
2249 put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, stride, 9);\
2250 put ## RND ## pixels8_l2_ ## MMX(halfH, src+1, halfH, 8, stride, 9);\
2251 put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(halfHV, halfH, 8, 8);\
2252 OPNAME ## pixels8_l2_ ## MMX(dst, halfH+8, halfHV, stride, 8, 8);\
2253}\
2254static void OPNAME ## qpel8_mc21_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
2255 uint64_t half[8 + 9];\
2256 uint8_t * const halfH= ((uint8_t*)half) + 64;\
2257 uint8_t * const halfHV= ((uint8_t*)half);\
2258 put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, stride, 9);\
2259 put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(halfHV, halfH, 8, 8);\
2260 OPNAME ## pixels8_l2_ ## MMX(dst, halfH, halfHV, stride, 8, 8);\
2261}\
2262static void OPNAME ## qpel8_mc23_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
2263 uint64_t half[8 + 9];\
2264 uint8_t * const halfH= ((uint8_t*)half) + 64;\
2265 uint8_t * const halfHV= ((uint8_t*)half);\
2266 put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, stride, 9);\
2267 put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(halfHV, halfH, 8, 8);\
2268 OPNAME ## pixels8_l2_ ## MMX(dst, halfH+8, halfHV, stride, 8, 8);\
2269}\
2270static void OPNAME ## qpel8_mc12_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
2271 uint64_t half[8 + 9];\
2272 uint8_t * const halfH= ((uint8_t*)half);\
2273 put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, stride, 9);\
2274 put ## RND ## pixels8_l2_ ## MMX(halfH, src, halfH, 8, stride, 9);\
2275 OPNAME ## mpeg4_qpel8_v_lowpass_ ## MMX(dst, halfH, stride, 8);\
2276}\
2277static void OPNAME ## qpel8_mc32_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
2278 uint64_t half[8 + 9];\
2279 uint8_t * const halfH= ((uint8_t*)half);\
2280 put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, stride, 9);\
2281 put ## RND ## pixels8_l2_ ## MMX(halfH, src+1, halfH, 8, stride, 9);\
2282 OPNAME ## mpeg4_qpel8_v_lowpass_ ## MMX(dst, halfH, stride, 8);\
2283}\
2284static void OPNAME ## qpel8_mc22_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
2285 uint64_t half[9];\
2286 uint8_t * const halfH= ((uint8_t*)half);\
2287 put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, stride, 9);\
2288 OPNAME ## mpeg4_qpel8_v_lowpass_ ## MMX(dst, halfH, stride, 8);\
2289}\
2290static void OPNAME ## qpel16_mc00_ ## MMX (uint8_t *dst, uint8_t *src, int stride){\
2291 OPNAME ## pixels16_mmx(dst, src, stride, 16);\
2292}\
2293\
2294static void OPNAME ## qpel16_mc10_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
2295 uint64_t temp[32];\
2296 uint8_t * const half= (uint8_t*)temp;\
2297 put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(half, src, 16, stride, 16);\
2298 OPNAME ## pixels16_l2_ ## MMX(dst, src, half, stride, stride, 16);\
2299}\
2300\
2301static void OPNAME ## qpel16_mc20_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
2302 OPNAME ## mpeg4_qpel16_h_lowpass_ ## MMX(dst, src, stride, stride, 16);\
2303}\
2304\
2305static void OPNAME ## qpel16_mc30_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
2306 uint64_t temp[32];\
2307 uint8_t * const half= (uint8_t*)temp;\
2308 put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(half, src, 16, stride, 16);\
2309 OPNAME ## pixels16_l2_ ## MMX(dst, src+1, half, stride, stride, 16);\
2310}\
2311\
2312static void OPNAME ## qpel16_mc01_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
2313 uint64_t temp[32];\
2314 uint8_t * const half= (uint8_t*)temp;\
2315 put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(half, src, 16, stride);\
2316 OPNAME ## pixels16_l2_ ## MMX(dst, src, half, stride, stride, 16);\
2317}\
2318\
2319static void OPNAME ## qpel16_mc02_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
2320 OPNAME ## mpeg4_qpel16_v_lowpass_ ## MMX(dst, src, stride, stride);\
2321}\
2322\
2323static void OPNAME ## qpel16_mc03_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
2324 uint64_t temp[32];\
2325 uint8_t * const half= (uint8_t*)temp;\
2326 put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(half, src, 16, stride);\
2327 OPNAME ## pixels16_l2_ ## MMX(dst, src+stride, half, stride, stride, 16);\
2328}\
2329static void OPNAME ## qpel16_mc11_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
2330 uint64_t half[16*2 + 17*2];\
2331 uint8_t * const halfH= ((uint8_t*)half) + 256;\
2332 uint8_t * const halfHV= ((uint8_t*)half);\
2333 put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16, stride, 17);\
2334 put ## RND ## pixels16_l2_ ## MMX(halfH, src, halfH, 16, stride, 17);\
2335 put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(halfHV, halfH, 16, 16);\
2336 OPNAME ## pixels16_l2_ ## MMX(dst, halfH, halfHV, stride, 16, 16);\
2337}\
2338static void OPNAME ## qpel16_mc31_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
2339 uint64_t half[16*2 + 17*2];\
2340 uint8_t * const halfH= ((uint8_t*)half) + 256;\
2341 uint8_t * const halfHV= ((uint8_t*)half);\
2342 put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16, stride, 17);\
2343 put ## RND ## pixels16_l2_ ## MMX(halfH, src+1, halfH, 16, stride, 17);\
2344 put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(halfHV, halfH, 16, 16);\
2345 OPNAME ## pixels16_l2_ ## MMX(dst, halfH, halfHV, stride, 16, 16);\
2346}\
2347static void OPNAME ## qpel16_mc13_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
2348 uint64_t half[16*2 + 17*2];\
2349 uint8_t * const halfH= ((uint8_t*)half) + 256;\
2350 uint8_t * const halfHV= ((uint8_t*)half);\
2351 put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16, stride, 17);\
2352 put ## RND ## pixels16_l2_ ## MMX(halfH, src, halfH, 16, stride, 17);\
2353 put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(halfHV, halfH, 16, 16);\
2354 OPNAME ## pixels16_l2_ ## MMX(dst, halfH+16, halfHV, stride, 16, 16);\
2355}\
2356static void OPNAME ## qpel16_mc33_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
2357 uint64_t half[16*2 + 17*2];\
2358 uint8_t * const halfH= ((uint8_t*)half) + 256;\
2359 uint8_t * const halfHV= ((uint8_t*)half);\
2360 put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16, stride, 17);\
2361 put ## RND ## pixels16_l2_ ## MMX(halfH, src+1, halfH, 16, stride, 17);\
2362 put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(halfHV, halfH, 16, 16);\
2363 OPNAME ## pixels16_l2_ ## MMX(dst, halfH+16, halfHV, stride, 16, 16);\
2364}\
2365static void OPNAME ## qpel16_mc21_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
2366 uint64_t half[16*2 + 17*2];\
2367 uint8_t * const halfH= ((uint8_t*)half) + 256;\
2368 uint8_t * const halfHV= ((uint8_t*)half);\
2369 put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16, stride, 17);\
2370 put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(halfHV, halfH, 16, 16);\
2371 OPNAME ## pixels16_l2_ ## MMX(dst, halfH, halfHV, stride, 16, 16);\
2372}\
2373static void OPNAME ## qpel16_mc23_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
2374 uint64_t half[16*2 + 17*2];\
2375 uint8_t * const halfH= ((uint8_t*)half) + 256;\
2376 uint8_t * const halfHV= ((uint8_t*)half);\
2377 put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16, stride, 17);\
2378 put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(halfHV, halfH, 16, 16);\
2379 OPNAME ## pixels16_l2_ ## MMX(dst, halfH+16, halfHV, stride, 16, 16);\
2380}\
2381static void OPNAME ## qpel16_mc12_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
2382 uint64_t half[17*2];\
2383 uint8_t * const halfH= ((uint8_t*)half);\
2384 put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16, stride, 17);\
2385 put ## RND ## pixels16_l2_ ## MMX(halfH, src, halfH, 16, stride, 17);\
2386 OPNAME ## mpeg4_qpel16_v_lowpass_ ## MMX(dst, halfH, stride, 16);\
2387}\
2388static void OPNAME ## qpel16_mc32_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
2389 uint64_t half[17*2];\
2390 uint8_t * const halfH= ((uint8_t*)half);\
2391 put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16, stride, 17);\
2392 put ## RND ## pixels16_l2_ ## MMX(halfH, src+1, halfH, 16, stride, 17);\
2393 OPNAME ## mpeg4_qpel16_v_lowpass_ ## MMX(dst, halfH, stride, 16);\
2394}\
2395static void OPNAME ## qpel16_mc22_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
2396 uint64_t half[17*2];\
2397 uint8_t * const halfH= ((uint8_t*)half);\
2398 put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16, stride, 17);\
2399 OPNAME ## mpeg4_qpel16_v_lowpass_ ## MMX(dst, halfH, stride, 16);\
2400}
2401
2402#define PUT_OP(a,b,temp, size) "mov" #size " " #a ", " #b " \n\t"
2403#define AVG_3DNOW_OP(a,b,temp, size) \
2404"mov" #size " " #b ", " #temp " \n\t"\
2405"pavgusb " #temp ", " #a " \n\t"\
2406"mov" #size " " #a ", " #b " \n\t"
2407#define AVG_MMX2_OP(a,b,temp, size) \
2408"mov" #size " " #b ", " #temp " \n\t"\
2409"pavgb " #temp ", " #a " \n\t"\
2410"mov" #size " " #a ", " #b " \n\t"
2411
2412QPEL_BASE(put_ , ff_pw_16, _ , PUT_OP, PUT_OP)
2413QPEL_BASE(avg_ , ff_pw_16, _ , AVG_MMX2_OP, AVG_3DNOW_OP)
2414QPEL_BASE(put_no_rnd_, ff_pw_15, _no_rnd_, PUT_OP, PUT_OP)
2415QPEL_OP(put_ , ff_pw_16, _ , PUT_OP, 3dnow)
2416QPEL_OP(avg_ , ff_pw_16, _ , AVG_3DNOW_OP, 3dnow)
2417QPEL_OP(put_no_rnd_, ff_pw_15, _no_rnd_, PUT_OP, 3dnow)
2418QPEL_OP(put_ , ff_pw_16, _ , PUT_OP, mmx2)
2419QPEL_OP(avg_ , ff_pw_16, _ , AVG_MMX2_OP, mmx2)
2420QPEL_OP(put_no_rnd_, ff_pw_15, _no_rnd_, PUT_OP, mmx2)
2421
2422#if 0
2423static void just_return() { return; }
2424#endif
2425
2426#define SET_QPEL_FUNC(postfix1, postfix2) \
2427 c->put_ ## postfix1 = put_ ## postfix2;\
2428 c->put_no_rnd_ ## postfix1 = put_no_rnd_ ## postfix2;\
2429 c->avg_ ## postfix1 = avg_ ## postfix2;
2430
2431static void gmc_mmx(uint8_t *dst, uint8_t *src, int stride, int h, int ox, int oy,
2432 int dxx, int dxy, int dyx, int dyy, int shift, int r, int width, int height){
2433 const int w = 8;
2434 const int ix = ox>>(16+shift);
2435 const int iy = oy>>(16+shift);
2436 const int oxs = ox>>4;
2437 const int oys = oy>>4;
2438 const int dxxs = dxx>>4;
2439 const int dxys = dxy>>4;
2440 const int dyxs = dyx>>4;
2441 const int dyys = dyy>>4;
2442 const uint16_t r4[4] = {r,r,r,r};
2443 const uint16_t dxy4[4] = {dxys,dxys,dxys,dxys};
2444 const uint16_t dyy4[4] = {dyys,dyys,dyys,dyys};
2445 const uint64_t shift2 = 2*shift;
2446 uint8_t edge_buf[(h+1)*stride];
2447 int x, y;
2448
2449 const int dxw = (dxx-(1<<(16+shift)))*(w-1);
2450 const int dyh = (dyy-(1<<(16+shift)))*(h-1);
2451 const int dxh = dxy*(h-1);
2452 const int dyw = dyx*(w-1);
2453 if( // non-constant fullpel offset (3% of blocks)
2454 (ox^(ox+dxw) | ox^(ox+dxh) | ox^(ox+dxw+dxh) |
2455 oy^(oy+dyw) | oy^(oy+dyh) | oy^(oy+dyw+dyh)) >> (16+shift)
2456 // uses more than 16 bits of subpel mv (only at huge resolution)
2457 || (dxx|dxy|dyx|dyy)&15 )
2458 {
2459 //FIXME could still use mmx for some of the rows
2460 ff_gmc_c(dst, src, stride, h, ox, oy, dxx, dxy, dyx, dyy, shift, r, width, height);
2461 return;
2462 }
2463
2464 src += ix + iy*stride;
2465 if( (unsigned)ix >= width-w ||
2466 (unsigned)iy >= height-h )
2467 {
2468 ff_emulated_edge_mc(edge_buf, src, stride, w+1, h+1, ix, iy, width, height);
2469 src = edge_buf;
2470 }
2471
2472 asm volatile(
2473 "movd %0, %%mm6 \n\t"
2474 "pxor %%mm7, %%mm7 \n\t"
2475 "punpcklwd %%mm6, %%mm6 \n\t"
2476 "punpcklwd %%mm6, %%mm6 \n\t"
2477 :: "r"(1<<shift)
2478 );
2479
2480 for(x=0; x<w; x+=4){
2481 uint16_t dx4[4] = { oxs - dxys + dxxs*(x+0),
2482 oxs - dxys + dxxs*(x+1),
2483 oxs - dxys + dxxs*(x+2),
2484 oxs - dxys + dxxs*(x+3) };
2485 uint16_t dy4[4] = { oys - dyys + dyxs*(x+0),
2486 oys - dyys + dyxs*(x+1),
2487 oys - dyys + dyxs*(x+2),
2488 oys - dyys + dyxs*(x+3) };
2489
2490 for(y=0; y<h; y++){
2491 asm volatile(
2492 "movq %0, %%mm4 \n\t"
2493 "movq %1, %%mm5 \n\t"
2494 "paddw %2, %%mm4 \n\t"
2495 "paddw %3, %%mm5 \n\t"
2496 "movq %%mm4, %0 \n\t"
2497 "movq %%mm5, %1 \n\t"
2498 "psrlw $12, %%mm4 \n\t"
2499 "psrlw $12, %%mm5 \n\t"
2500 : "+m"(*dx4), "+m"(*dy4)
2501 : "m"(*dxy4), "m"(*dyy4)
2502 );
2503
2504 asm volatile(
2505 "movq %%mm6, %%mm2 \n\t"
2506 "movq %%mm6, %%mm1 \n\t"
2507 "psubw %%mm4, %%mm2 \n\t"
2508 "psubw %%mm5, %%mm1 \n\t"
2509 "movq %%mm2, %%mm0 \n\t"
2510 "movq %%mm4, %%mm3 \n\t"
2511 "pmullw %%mm1, %%mm0 \n\t" // (s-dx)*(s-dy)
2512 "pmullw %%mm5, %%mm3 \n\t" // dx*dy
2513 "pmullw %%mm5, %%mm2 \n\t" // (s-dx)*dy
2514 "pmullw %%mm4, %%mm1 \n\t" // dx*(s-dy)
2515
2516 "movd %4, %%mm5 \n\t"
2517 "movd %3, %%mm4 \n\t"
2518 "punpcklbw %%mm7, %%mm5 \n\t"
2519 "punpcklbw %%mm7, %%mm4 \n\t"
2520 "pmullw %%mm5, %%mm3 \n\t" // src[1,1] * dx*dy
2521 "pmullw %%mm4, %%mm2 \n\t" // src[0,1] * (s-dx)*dy
2522
2523 "movd %2, %%mm5 \n\t"
2524 "movd %1, %%mm4 \n\t"
2525 "punpcklbw %%mm7, %%mm5 \n\t"
2526 "punpcklbw %%mm7, %%mm4 \n\t"
2527 "pmullw %%mm5, %%mm1 \n\t" // src[1,0] * dx*(s-dy)
2528 "pmullw %%mm4, %%mm0 \n\t" // src[0,0] * (s-dx)*(s-dy)
2529 "paddw %5, %%mm1 \n\t"
2530 "paddw %%mm3, %%mm2 \n\t"
2531 "paddw %%mm1, %%mm0 \n\t"
2532 "paddw %%mm2, %%mm0 \n\t"
2533
2534 "psrlw %6, %%mm0 \n\t"
2535 "packuswb %%mm0, %%mm0 \n\t"
2536 "movd %%mm0, %0 \n\t"
2537
2538 : "=m"(dst[x+y*stride])
2539 : "m"(src[0]), "m"(src[1]),
2540 "m"(src[stride]), "m"(src[stride+1]),
2541 "m"(*r4), "m"(shift2)
2542 );
2543 src += stride;
2544 }
2545 src += 4-h*stride;
2546 }
2547}
2548
2549static int try_8x8basis_mmx(int16_t rem[64], int16_t weight[64], int16_t basis[64], int scale){
2550 long i=0;
2551
2552 assert(ABS(scale) < 256);
2553 scale<<= 16 + 1 - BASIS_SHIFT + RECON_SHIFT;
2554
2555 asm volatile(
2556 "pcmpeqw %%mm6, %%mm6 \n\t" // -1w
2557 "psrlw $15, %%mm6 \n\t" // 1w
2558 "pxor %%mm7, %%mm7 \n\t"
2559 "movd %4, %%mm5 \n\t"
2560 "punpcklwd %%mm5, %%mm5 \n\t"
2561 "punpcklwd %%mm5, %%mm5 \n\t"
2562 "1: \n\t"
2563 "movq (%1, %0), %%mm0 \n\t"
2564 "movq 8(%1, %0), %%mm1 \n\t"
2565 "pmulhw %%mm5, %%mm0 \n\t"
2566 "pmulhw %%mm5, %%mm1 \n\t"
2567 "paddw %%mm6, %%mm0 \n\t"
2568 "paddw %%mm6, %%mm1 \n\t"
2569 "psraw $1, %%mm0 \n\t"
2570 "psraw $1, %%mm1 \n\t"
2571 "paddw (%2, %0), %%mm0 \n\t"
2572 "paddw 8(%2, %0), %%mm1 \n\t"
2573 "psraw $6, %%mm0 \n\t"
2574 "psraw $6, %%mm1 \n\t"
2575 "pmullw (%3, %0), %%mm0 \n\t"
2576 "pmullw 8(%3, %0), %%mm1 \n\t"
2577 "pmaddwd %%mm0, %%mm0 \n\t"
2578 "pmaddwd %%mm1, %%mm1 \n\t"
2579 "paddd %%mm1, %%mm0 \n\t"
2580 "psrld $4, %%mm0 \n\t"
2581 "paddd %%mm0, %%mm7 \n\t"
2582 "add $16, %0 \n\t"
2583 "cmp $128, %0 \n\t" //FIXME optimize & bench
2584 " jb 1b \n\t"
2585 "movq %%mm7, %%mm6 \n\t"
2586 "psrlq $32, %%mm7 \n\t"
2587 "paddd %%mm6, %%mm7 \n\t"
2588 "psrld $2, %%mm7 \n\t"
2589 "movd %%mm7, %0 \n\t"
2590
2591 : "+r" (i)
2592 : "r"(basis), "r"(rem), "r"(weight), "g"(scale)
2593 );
2594 return i;
2595}
2596
2597static void add_8x8basis_mmx(int16_t rem[64], int16_t basis[64], int scale){
2598 long i=0;
2599
2600 if(ABS(scale) < 256){
2601 scale<<= 16 + 1 - BASIS_SHIFT + RECON_SHIFT;
2602 asm volatile(
2603 "pcmpeqw %%mm6, %%mm6 \n\t" // -1w
2604 "psrlw $15, %%mm6 \n\t" // 1w
2605 "movd %3, %%mm5 \n\t"
2606 "punpcklwd %%mm5, %%mm5 \n\t"
2607 "punpcklwd %%mm5, %%mm5 \n\t"
2608 "1: \n\t"
2609 "movq (%1, %0), %%mm0 \n\t"
2610 "movq 8(%1, %0), %%mm1 \n\t"
2611 "pmulhw %%mm5, %%mm0 \n\t"
2612 "pmulhw %%mm5, %%mm1 \n\t"
2613 "paddw %%mm6, %%mm0 \n\t"
2614 "paddw %%mm6, %%mm1 \n\t"
2615 "psraw $1, %%mm0 \n\t"
2616 "psraw $1, %%mm1 \n\t"
2617 "paddw (%2, %0), %%mm0 \n\t"
2618 "paddw 8(%2, %0), %%mm1 \n\t"
2619 "movq %%mm0, (%2, %0) \n\t"
2620 "movq %%mm1, 8(%2, %0) \n\t"
2621 "add $16, %0 \n\t"
2622 "cmp $128, %0 \n\t" //FIXME optimize & bench
2623 " jb 1b \n\t"
2624
2625 : "+r" (i)
2626 : "r"(basis), "r"(rem), "g"(scale)
2627 );
2628 }else{
2629 for(i=0; i<8*8; i++){
2630 rem[i] += (basis[i]*scale + (1<<(BASIS_SHIFT - RECON_SHIFT-1)))>>(BASIS_SHIFT - RECON_SHIFT);
2631 }
2632 }
2633}
2634
2635#define PREFETCH(name, op) \
2636void name(void *mem, int stride, int h){\
2637 const uint8_t *p= mem;\
2638 do{\
2639 asm volatile(#op" %0" :: "m"(*p));\
2640 p+= stride;\
2641 }while(--h);\
2642}
2643PREFETCH(prefetch_mmx2, prefetcht0)
2644PREFETCH(prefetch_3dnow, prefetch)
2645#undef PREFETCH
2646
2647#include "h264dsp_mmx.c"
2648
2649/* external functions, from idct_mmx.c */
2650void ff_mmx_idct(DCTELEM *block);
2651void ff_mmxext_idct(DCTELEM *block);
2652
2653void ff_vp3_idct_sse2(int16_t *input_data);
2654void ff_vp3_idct_mmx(int16_t *data);
2655void ff_vp3_dsp_init_mmx(void);
2656
2657/* XXX: those functions should be suppressed ASAP when all IDCTs are
2658 converted */
2659static void ff_libmpeg2mmx_idct_put(uint8_t *dest, int line_size, DCTELEM *block)
2660{
2661 ff_mmx_idct (block);
2662 put_pixels_clamped_mmx(block, dest, line_size);
2663}
2664static void ff_libmpeg2mmx_idct_add(uint8_t *dest, int line_size, DCTELEM *block)
2665{
2666 ff_mmx_idct (block);
2667 add_pixels_clamped_mmx(block, dest, line_size);
2668}
2669static void ff_libmpeg2mmx2_idct_put(uint8_t *dest, int line_size, DCTELEM *block)
2670{
2671 ff_mmxext_idct (block);
2672 put_pixels_clamped_mmx(block, dest, line_size);
2673}
2674static void ff_libmpeg2mmx2_idct_add(uint8_t *dest, int line_size, DCTELEM *block)
2675{
2676 ff_mmxext_idct (block);
2677 add_pixels_clamped_mmx(block, dest, line_size);
2678}
2679static void ff_vp3_idct_put_sse2(uint8_t *dest, int line_size, DCTELEM *block)
2680{
2681 ff_vp3_idct_sse2(block);
2682 put_signed_pixels_clamped_mmx(block, dest, line_size);
2683}
2684static void ff_vp3_idct_add_sse2(uint8_t *dest, int line_size, DCTELEM *block)
2685{
2686 ff_vp3_idct_sse2(block);
2687 add_pixels_clamped_mmx(block, dest, line_size);
2688}
2689static void ff_vp3_idct_put_mmx(uint8_t *dest, int line_size, DCTELEM *block)
2690{
2691 ff_vp3_idct_mmx(block);
2692 put_signed_pixels_clamped_mmx(block, dest, line_size);
2693}
2694static void ff_vp3_idct_add_mmx(uint8_t *dest, int line_size, DCTELEM *block)
2695{
2696 ff_vp3_idct_mmx(block);
2697 add_pixels_clamped_mmx(block, dest, line_size);
2698}
2699
2700#ifdef CONFIG_SNOW_ENCODER
2701extern void ff_snow_horizontal_compose97i_sse2(DWTELEM *b, int width);
2702extern void ff_snow_horizontal_compose97i_mmx(DWTELEM *b, int width);
2703extern void ff_snow_vertical_compose97i_sse2(DWTELEM *b0, DWTELEM *b1, DWTELEM *b2, DWTELEM *b3, DWTELEM *b4, DWTELEM *b5, int width);
2704extern void ff_snow_vertical_compose97i_mmx(DWTELEM *b0, DWTELEM *b1, DWTELEM *b2, DWTELEM *b3, DWTELEM *b4, DWTELEM *b5, int width);
2705extern void ff_snow_inner_add_yblock_sse2(uint8_t *obmc, const int obmc_stride, uint8_t * * block, int b_w, int b_h,
2706 int src_x, int src_y, int src_stride, slice_buffer * sb, int add, uint8_t * dst8);
2707extern void ff_snow_inner_add_yblock_mmx(uint8_t *obmc, const int obmc_stride, uint8_t * * block, int b_w, int b_h,
2708 int src_x, int src_y, int src_stride, slice_buffer * sb, int add, uint8_t * dst8);
2709#endif
2710
2711void dsputil_init_mmx(DSPContext* c, AVCodecContext *avctx)
2712{
2713 mm_flags = mm_support();
2714
2715 if (avctx->dsp_mask) {
2716 if (avctx->dsp_mask & FF_MM_FORCE)
2717 mm_flags |= (avctx->dsp_mask & 0xffff);
2718 else
2719 mm_flags &= ~(avctx->dsp_mask & 0xffff);
2720 }
2721
2722#if 0
2723 av_log(avctx, AV_LOG_INFO, "libavcodec: CPU flags:");
2724 if (mm_flags & MM_MMX)
2725 av_log(avctx, AV_LOG_INFO, " mmx");
2726 if (mm_flags & MM_MMXEXT)
2727 av_log(avctx, AV_LOG_INFO, " mmxext");
2728 if (mm_flags & MM_3DNOW)
2729 av_log(avctx, AV_LOG_INFO, " 3dnow");
2730 if (mm_flags & MM_SSE)
2731 av_log(avctx, AV_LOG_INFO, " sse");
2732 if (mm_flags & MM_SSE2)
2733 av_log(avctx, AV_LOG_INFO, " sse2");
2734 av_log(avctx, AV_LOG_INFO, "\n");
2735#endif
2736
2737 if (mm_flags & MM_MMX) {
2738 const int idct_algo= avctx->idct_algo;
2739
2740#ifdef CONFIG_ENCODERS
2741 const int dct_algo = avctx->dct_algo;
2742 if(dct_algo==FF_DCT_AUTO || dct_algo==FF_DCT_MMX){
2743 if(mm_flags & MM_SSE2){
2744 c->fdct = ff_fdct_sse2;
2745 }else if(mm_flags & MM_MMXEXT){
2746 c->fdct = ff_fdct_mmx2;
2747 }else{
2748 c->fdct = ff_fdct_mmx;
2749 }
2750 }
2751#endif //CONFIG_ENCODERS
2752 if(avctx->lowres==0){
2753 if(idct_algo==FF_IDCT_AUTO || idct_algo==FF_IDCT_SIMPLEMMX){
2754 c->idct_put= ff_simple_idct_put_mmx;
2755 c->idct_add= ff_simple_idct_add_mmx;
2756 c->idct = ff_simple_idct_mmx;
2757 c->idct_permutation_type= FF_SIMPLE_IDCT_PERM;
2758 }else if(idct_algo==FF_IDCT_LIBMPEG2MMX){
2759 if(mm_flags & MM_MMXEXT){
2760 c->idct_put= ff_libmpeg2mmx2_idct_put;
2761 c->idct_add= ff_libmpeg2mmx2_idct_add;
2762 c->idct = ff_mmxext_idct;
2763 }else{
2764 c->idct_put= ff_libmpeg2mmx_idct_put;
2765 c->idct_add= ff_libmpeg2mmx_idct_add;
2766 c->idct = ff_mmx_idct;
2767 }
2768 c->idct_permutation_type= FF_LIBMPEG2_IDCT_PERM;
2769 }else if(idct_algo==FF_IDCT_VP3){
2770 if(mm_flags & MM_SSE2){
2771 c->idct_put= ff_vp3_idct_put_sse2;
2772 c->idct_add= ff_vp3_idct_add_sse2;
2773 c->idct = ff_vp3_idct_sse2;
2774 c->idct_permutation_type= FF_TRANSPOSE_IDCT_PERM;
2775 }else{
2776 ff_vp3_dsp_init_mmx();
2777 c->idct_put= ff_vp3_idct_put_mmx;
2778 c->idct_add= ff_vp3_idct_add_mmx;
2779 c->idct = ff_vp3_idct_mmx;
2780 c->idct_permutation_type= FF_PARTTRANS_IDCT_PERM;
2781 }
2782 }
2783 }
2784
2785#ifdef CONFIG_ENCODERS
2786 c->get_pixels = get_pixels_mmx;
2787 c->diff_pixels = diff_pixels_mmx;
2788#endif //CONFIG_ENCODERS
2789 c->put_pixels_clamped = put_pixels_clamped_mmx;
2790 c->put_signed_pixels_clamped = put_signed_pixels_clamped_mmx;
2791 c->add_pixels_clamped = add_pixels_clamped_mmx;
2792 c->clear_blocks = clear_blocks_mmx;
2793#ifdef CONFIG_ENCODERS
2794 c->pix_sum = pix_sum16_mmx;
2795#endif //CONFIG_ENCODERS
2796
2797 c->put_pixels_tab[0][0] = put_pixels16_mmx;
2798 c->put_pixels_tab[0][1] = put_pixels16_x2_mmx;
2799 c->put_pixels_tab[0][2] = put_pixels16_y2_mmx;
2800 c->put_pixels_tab[0][3] = put_pixels16_xy2_mmx;
2801
2802 c->put_no_rnd_pixels_tab[0][0] = put_pixels16_mmx;
2803 c->put_no_rnd_pixels_tab[0][1] = put_no_rnd_pixels16_x2_mmx;
2804 c->put_no_rnd_pixels_tab[0][2] = put_no_rnd_pixels16_y2_mmx;
2805 c->put_no_rnd_pixels_tab[0][3] = put_no_rnd_pixels16_xy2_mmx;
2806
2807 c->avg_pixels_tab[0][0] = avg_pixels16_mmx;
2808 c->avg_pixels_tab[0][1] = avg_pixels16_x2_mmx;
2809 c->avg_pixels_tab[0][2] = avg_pixels16_y2_mmx;
2810 c->avg_pixels_tab[0][3] = avg_pixels16_xy2_mmx;
2811
2812 c->avg_no_rnd_pixels_tab[0][0] = avg_no_rnd_pixels16_mmx;
2813 c->avg_no_rnd_pixels_tab[0][1] = avg_no_rnd_pixels16_x2_mmx;
2814 c->avg_no_rnd_pixels_tab[0][2] = avg_no_rnd_pixels16_y2_mmx;
2815 c->avg_no_rnd_pixels_tab[0][3] = avg_no_rnd_pixels16_xy2_mmx;
2816
2817 c->put_pixels_tab[1][0] = put_pixels8_mmx;
2818 c->put_pixels_tab[1][1] = put_pixels8_x2_mmx;
2819 c->put_pixels_tab[1][2] = put_pixels8_y2_mmx;
2820 c->put_pixels_tab[1][3] = put_pixels8_xy2_mmx;
2821
2822 c->put_no_rnd_pixels_tab[1][0] = put_pixels8_mmx;
2823 c->put_no_rnd_pixels_tab[1][1] = put_no_rnd_pixels8_x2_mmx;
2824 c->put_no_rnd_pixels_tab[1][2] = put_no_rnd_pixels8_y2_mmx;
2825 c->put_no_rnd_pixels_tab[1][3] = put_no_rnd_pixels8_xy2_mmx;
2826
2827 c->avg_pixels_tab[1][0] = avg_pixels8_mmx;
2828 c->avg_pixels_tab[1][1] = avg_pixels8_x2_mmx;
2829 c->avg_pixels_tab[1][2] = avg_pixels8_y2_mmx;
2830 c->avg_pixels_tab[1][3] = avg_pixels8_xy2_mmx;
2831
2832 c->avg_no_rnd_pixels_tab[1][0] = avg_no_rnd_pixels8_mmx;
2833 c->avg_no_rnd_pixels_tab[1][1] = avg_no_rnd_pixels8_x2_mmx;
2834 c->avg_no_rnd_pixels_tab[1][2] = avg_no_rnd_pixels8_y2_mmx;
2835 c->avg_no_rnd_pixels_tab[1][3] = avg_no_rnd_pixels8_xy2_mmx;
2836
2837 c->gmc= gmc_mmx;
2838
2839 c->add_bytes= add_bytes_mmx;
2840#ifdef CONFIG_ENCODERS
2841 c->diff_bytes= diff_bytes_mmx;
2842
2843 c->hadamard8_diff[0]= hadamard8_diff16_mmx;
2844 c->hadamard8_diff[1]= hadamard8_diff_mmx;
2845
2846 c->pix_norm1 = pix_norm1_mmx;
2847 c->sse[0] = (mm_flags & MM_SSE2) ? sse16_sse2 : sse16_mmx;
2848 c->sse[1] = sse8_mmx;
2849 c->vsad[4]= vsad_intra16_mmx;
2850
2851 c->nsse[0] = nsse16_mmx;
2852 c->nsse[1] = nsse8_mmx;
2853 if(!(avctx->flags & CODEC_FLAG_BITEXACT)){
2854 c->vsad[0] = vsad16_mmx;
2855 }
2856
2857 if(!(avctx->flags & CODEC_FLAG_BITEXACT)){
2858 c->try_8x8basis= try_8x8basis_mmx;
2859 }
2860 c->add_8x8basis= add_8x8basis_mmx;
2861
2862#endif //CONFIG_ENCODERS
2863
2864 c->h263_v_loop_filter= h263_v_loop_filter_mmx;
2865 c->h263_h_loop_filter= h263_h_loop_filter_mmx;
2866 c->put_h264_chroma_pixels_tab[0]= put_h264_chroma_mc8_mmx;
2867 c->put_h264_chroma_pixels_tab[1]= put_h264_chroma_mc4_mmx;
2868
2869 c->h264_idct_dc_add=
2870 c->h264_idct_add= ff_h264_idct_add_mmx;
2871 c->h264_idct8_dc_add=
2872 c->h264_idct8_add= ff_h264_idct8_add_mmx;
2873
2874 if (mm_flags & MM_MMXEXT) {
2875 c->prefetch = prefetch_mmx2;
2876
2877 c->put_pixels_tab[0][1] = put_pixels16_x2_mmx2;
2878 c->put_pixels_tab[0][2] = put_pixels16_y2_mmx2;
2879
2880 c->avg_pixels_tab[0][0] = avg_pixels16_mmx2;
2881 c->avg_pixels_tab[0][1] = avg_pixels16_x2_mmx2;
2882 c->avg_pixels_tab[0][2] = avg_pixels16_y2_mmx2;
2883
2884 c->put_pixels_tab[1][1] = put_pixels8_x2_mmx2;
2885 c->put_pixels_tab[1][2] = put_pixels8_y2_mmx2;
2886
2887 c->avg_pixels_tab[1][0] = avg_pixels8_mmx2;
2888 c->avg_pixels_tab[1][1] = avg_pixels8_x2_mmx2;
2889 c->avg_pixels_tab[1][2] = avg_pixels8_y2_mmx2;
2890
2891#ifdef CONFIG_ENCODERS
2892 c->hadamard8_diff[0]= hadamard8_diff16_mmx2;
2893 c->hadamard8_diff[1]= hadamard8_diff_mmx2;
2894 c->vsad[4]= vsad_intra16_mmx2;
2895#endif //CONFIG_ENCODERS
2896
2897 c->h264_idct_dc_add= ff_h264_idct_dc_add_mmx2;
2898 c->h264_idct8_dc_add= ff_h264_idct8_dc_add_mmx2;
2899
2900 if(!(avctx->flags & CODEC_FLAG_BITEXACT)){
2901 c->put_no_rnd_pixels_tab[0][1] = put_no_rnd_pixels16_x2_mmx2;
2902 c->put_no_rnd_pixels_tab[0][2] = put_no_rnd_pixels16_y2_mmx2;
2903 c->put_no_rnd_pixels_tab[1][1] = put_no_rnd_pixels8_x2_mmx2;
2904 c->put_no_rnd_pixels_tab[1][2] = put_no_rnd_pixels8_y2_mmx2;
2905 c->avg_pixels_tab[0][3] = avg_pixels16_xy2_mmx2;
2906 c->avg_pixels_tab[1][3] = avg_pixels8_xy2_mmx2;
2907#ifdef CONFIG_ENCODERS
2908 c->vsad[0] = vsad16_mmx2;
2909#endif //CONFIG_ENCODERS
2910 }
2911
2912#if 1
2913 SET_QPEL_FUNC(qpel_pixels_tab[0][ 0], qpel16_mc00_mmx2)
2914 SET_QPEL_FUNC(qpel_pixels_tab[0][ 1], qpel16_mc10_mmx2)
2915 SET_QPEL_FUNC(qpel_pixels_tab[0][ 2], qpel16_mc20_mmx2)
2916 SET_QPEL_FUNC(qpel_pixels_tab[0][ 3], qpel16_mc30_mmx2)
2917 SET_QPEL_FUNC(qpel_pixels_tab[0][ 4], qpel16_mc01_mmx2)
2918 SET_QPEL_FUNC(qpel_pixels_tab[0][ 5], qpel16_mc11_mmx2)
2919 SET_QPEL_FUNC(qpel_pixels_tab[0][ 6], qpel16_mc21_mmx2)
2920 SET_QPEL_FUNC(qpel_pixels_tab[0][ 7], qpel16_mc31_mmx2)
2921 SET_QPEL_FUNC(qpel_pixels_tab[0][ 8], qpel16_mc02_mmx2)
2922 SET_QPEL_FUNC(qpel_pixels_tab[0][ 9], qpel16_mc12_mmx2)
2923 SET_QPEL_FUNC(qpel_pixels_tab[0][10], qpel16_mc22_mmx2)
2924 SET_QPEL_FUNC(qpel_pixels_tab[0][11], qpel16_mc32_mmx2)
2925 SET_QPEL_FUNC(qpel_pixels_tab[0][12], qpel16_mc03_mmx2)
2926 SET_QPEL_FUNC(qpel_pixels_tab[0][13], qpel16_mc13_mmx2)
2927 SET_QPEL_FUNC(qpel_pixels_tab[0][14], qpel16_mc23_mmx2)
2928 SET_QPEL_FUNC(qpel_pixels_tab[0][15], qpel16_mc33_mmx2)
2929 SET_QPEL_FUNC(qpel_pixels_tab[1][ 0], qpel8_mc00_mmx2)
2930 SET_QPEL_FUNC(qpel_pixels_tab[1][ 1], qpel8_mc10_mmx2)
2931 SET_QPEL_FUNC(qpel_pixels_tab[1][ 2], qpel8_mc20_mmx2)
2932 SET_QPEL_FUNC(qpel_pixels_tab[1][ 3], qpel8_mc30_mmx2)
2933 SET_QPEL_FUNC(qpel_pixels_tab[1][ 4], qpel8_mc01_mmx2)
2934 SET_QPEL_FUNC(qpel_pixels_tab[1][ 5], qpel8_mc11_mmx2)
2935 SET_QPEL_FUNC(qpel_pixels_tab[1][ 6], qpel8_mc21_mmx2)
2936 SET_QPEL_FUNC(qpel_pixels_tab[1][ 7], qpel8_mc31_mmx2)
2937 SET_QPEL_FUNC(qpel_pixels_tab[1][ 8], qpel8_mc02_mmx2)
2938 SET_QPEL_FUNC(qpel_pixels_tab[1][ 9], qpel8_mc12_mmx2)
2939 SET_QPEL_FUNC(qpel_pixels_tab[1][10], qpel8_mc22_mmx2)
2940 SET_QPEL_FUNC(qpel_pixels_tab[1][11], qpel8_mc32_mmx2)
2941 SET_QPEL_FUNC(qpel_pixels_tab[1][12], qpel8_mc03_mmx2)
2942 SET_QPEL_FUNC(qpel_pixels_tab[1][13], qpel8_mc13_mmx2)
2943 SET_QPEL_FUNC(qpel_pixels_tab[1][14], qpel8_mc23_mmx2)
2944 SET_QPEL_FUNC(qpel_pixels_tab[1][15], qpel8_mc33_mmx2)
2945#endif
2946
2947//FIXME 3dnow too
2948#define dspfunc(PFX, IDX, NUM) \
2949 c->PFX ## _pixels_tab[IDX][ 0] = PFX ## NUM ## _mc00_mmx2; \
2950 c->PFX ## _pixels_tab[IDX][ 1] = PFX ## NUM ## _mc10_mmx2; \
2951 c->PFX ## _pixels_tab[IDX][ 2] = PFX ## NUM ## _mc20_mmx2; \
2952 c->PFX ## _pixels_tab[IDX][ 3] = PFX ## NUM ## _mc30_mmx2; \
2953 c->PFX ## _pixels_tab[IDX][ 4] = PFX ## NUM ## _mc01_mmx2; \
2954 c->PFX ## _pixels_tab[IDX][ 5] = PFX ## NUM ## _mc11_mmx2; \
2955 c->PFX ## _pixels_tab[IDX][ 6] = PFX ## NUM ## _mc21_mmx2; \
2956 c->PFX ## _pixels_tab[IDX][ 7] = PFX ## NUM ## _mc31_mmx2; \
2957 c->PFX ## _pixels_tab[IDX][ 8] = PFX ## NUM ## _mc02_mmx2; \
2958 c->PFX ## _pixels_tab[IDX][ 9] = PFX ## NUM ## _mc12_mmx2; \
2959 c->PFX ## _pixels_tab[IDX][10] = PFX ## NUM ## _mc22_mmx2; \
2960 c->PFX ## _pixels_tab[IDX][11] = PFX ## NUM ## _mc32_mmx2; \
2961 c->PFX ## _pixels_tab[IDX][12] = PFX ## NUM ## _mc03_mmx2; \
2962 c->PFX ## _pixels_tab[IDX][13] = PFX ## NUM ## _mc13_mmx2; \
2963 c->PFX ## _pixels_tab[IDX][14] = PFX ## NUM ## _mc23_mmx2; \
2964 c->PFX ## _pixels_tab[IDX][15] = PFX ## NUM ## _mc33_mmx2
2965
2966 dspfunc(put_h264_qpel, 0, 16);
2967 dspfunc(put_h264_qpel, 1, 8);
2968 dspfunc(put_h264_qpel, 2, 4);
2969 dspfunc(avg_h264_qpel, 0, 16);
2970 dspfunc(avg_h264_qpel, 1, 8);
2971 dspfunc(avg_h264_qpel, 2, 4);
2972#undef dspfunc
2973
2974 c->avg_h264_chroma_pixels_tab[0]= avg_h264_chroma_mc8_mmx2;
2975 c->avg_h264_chroma_pixels_tab[1]= avg_h264_chroma_mc4_mmx2;
2976 c->avg_h264_chroma_pixels_tab[2]= avg_h264_chroma_mc2_mmx2;
2977 c->put_h264_chroma_pixels_tab[2]= put_h264_chroma_mc2_mmx2;
2978 c->h264_v_loop_filter_luma= h264_v_loop_filter_luma_mmx2;
2979 c->h264_h_loop_filter_luma= h264_h_loop_filter_luma_mmx2;
2980 c->h264_v_loop_filter_chroma= h264_v_loop_filter_chroma_mmx2;
2981 c->h264_h_loop_filter_chroma= h264_h_loop_filter_chroma_mmx2;
2982 c->h264_v_loop_filter_chroma_intra= h264_v_loop_filter_chroma_intra_mmx2;
2983 c->h264_h_loop_filter_chroma_intra= h264_h_loop_filter_chroma_intra_mmx2;
2984
2985 c->weight_h264_pixels_tab[0]= ff_h264_weight_16x16_mmx2;
2986 c->weight_h264_pixels_tab[1]= ff_h264_weight_16x8_mmx2;
2987 c->weight_h264_pixels_tab[2]= ff_h264_weight_8x16_mmx2;
2988 c->weight_h264_pixels_tab[3]= ff_h264_weight_8x8_mmx2;
2989 c->weight_h264_pixels_tab[4]= ff_h264_weight_8x4_mmx2;
2990 c->weight_h264_pixels_tab[5]= ff_h264_weight_4x8_mmx2;
2991 c->weight_h264_pixels_tab[6]= ff_h264_weight_4x4_mmx2;
2992 c->weight_h264_pixels_tab[7]= ff_h264_weight_4x2_mmx2;
2993
2994 c->biweight_h264_pixels_tab[0]= ff_h264_biweight_16x16_mmx2;
2995 c->biweight_h264_pixels_tab[1]= ff_h264_biweight_16x8_mmx2;
2996 c->biweight_h264_pixels_tab[2]= ff_h264_biweight_8x16_mmx2;
2997 c->biweight_h264_pixels_tab[3]= ff_h264_biweight_8x8_mmx2;
2998 c->biweight_h264_pixels_tab[4]= ff_h264_biweight_8x4_mmx2;
2999 c->biweight_h264_pixels_tab[5]= ff_h264_biweight_4x8_mmx2;
3000 c->biweight_h264_pixels_tab[6]= ff_h264_biweight_4x4_mmx2;
3001 c->biweight_h264_pixels_tab[7]= ff_h264_biweight_4x2_mmx2;
3002
3003#ifdef CONFIG_ENCODERS
3004 c->sub_hfyu_median_prediction= sub_hfyu_median_prediction_mmx2;
3005#endif //CONFIG_ENCODERS
3006 } else if (mm_flags & MM_3DNOW) {
3007 c->prefetch = prefetch_3dnow;
3008
3009 c->put_pixels_tab[0][1] = put_pixels16_x2_3dnow;
3010 c->put_pixels_tab[0][2] = put_pixels16_y2_3dnow;
3011
3012 c->avg_pixels_tab[0][0] = avg_pixels16_3dnow;
3013 c->avg_pixels_tab[0][1] = avg_pixels16_x2_3dnow;
3014 c->avg_pixels_tab[0][2] = avg_pixels16_y2_3dnow;
3015
3016 c->put_pixels_tab[1][1] = put_pixels8_x2_3dnow;
3017 c->put_pixels_tab[1][2] = put_pixels8_y2_3dnow;
3018
3019 c->avg_pixels_tab[1][0] = avg_pixels8_3dnow;
3020 c->avg_pixels_tab[1][1] = avg_pixels8_x2_3dnow;
3021 c->avg_pixels_tab[1][2] = avg_pixels8_y2_3dnow;
3022
3023 if(!(avctx->flags & CODEC_FLAG_BITEXACT)){
3024 c->put_no_rnd_pixels_tab[0][1] = put_no_rnd_pixels16_x2_3dnow;
3025 c->put_no_rnd_pixels_tab[0][2] = put_no_rnd_pixels16_y2_3dnow;
3026 c->put_no_rnd_pixels_tab[1][1] = put_no_rnd_pixels8_x2_3dnow;
3027 c->put_no_rnd_pixels_tab[1][2] = put_no_rnd_pixels8_y2_3dnow;
3028 c->avg_pixels_tab[0][3] = avg_pixels16_xy2_3dnow;
3029 c->avg_pixels_tab[1][3] = avg_pixels8_xy2_3dnow;
3030 }
3031
3032 SET_QPEL_FUNC(qpel_pixels_tab[0][ 0], qpel16_mc00_3dnow)
3033 SET_QPEL_FUNC(qpel_pixels_tab[0][ 1], qpel16_mc10_3dnow)
3034 SET_QPEL_FUNC(qpel_pixels_tab[0][ 2], qpel16_mc20_3dnow)
3035 SET_QPEL_FUNC(qpel_pixels_tab[0][ 3], qpel16_mc30_3dnow)
3036 SET_QPEL_FUNC(qpel_pixels_tab[0][ 4], qpel16_mc01_3dnow)
3037 SET_QPEL_FUNC(qpel_pixels_tab[0][ 5], qpel16_mc11_3dnow)
3038 SET_QPEL_FUNC(qpel_pixels_tab[0][ 6], qpel16_mc21_3dnow)
3039 SET_QPEL_FUNC(qpel_pixels_tab[0][ 7], qpel16_mc31_3dnow)
3040 SET_QPEL_FUNC(qpel_pixels_tab[0][ 8], qpel16_mc02_3dnow)
3041 SET_QPEL_FUNC(qpel_pixels_tab[0][ 9], qpel16_mc12_3dnow)
3042 SET_QPEL_FUNC(qpel_pixels_tab[0][10], qpel16_mc22_3dnow)
3043 SET_QPEL_FUNC(qpel_pixels_tab[0][11], qpel16_mc32_3dnow)
3044 SET_QPEL_FUNC(qpel_pixels_tab[0][12], qpel16_mc03_3dnow)
3045 SET_QPEL_FUNC(qpel_pixels_tab[0][13], qpel16_mc13_3dnow)
3046 SET_QPEL_FUNC(qpel_pixels_tab[0][14], qpel16_mc23_3dnow)
3047 SET_QPEL_FUNC(qpel_pixels_tab[0][15], qpel16_mc33_3dnow)
3048 SET_QPEL_FUNC(qpel_pixels_tab[1][ 0], qpel8_mc00_3dnow)
3049 SET_QPEL_FUNC(qpel_pixels_tab[1][ 1], qpel8_mc10_3dnow)
3050 SET_QPEL_FUNC(qpel_pixels_tab[1][ 2], qpel8_mc20_3dnow)
3051 SET_QPEL_FUNC(qpel_pixels_tab[1][ 3], qpel8_mc30_3dnow)
3052 SET_QPEL_FUNC(qpel_pixels_tab[1][ 4], qpel8_mc01_3dnow)
3053 SET_QPEL_FUNC(qpel_pixels_tab[1][ 5], qpel8_mc11_3dnow)
3054 SET_QPEL_FUNC(qpel_pixels_tab[1][ 6], qpel8_mc21_3dnow)
3055 SET_QPEL_FUNC(qpel_pixels_tab[1][ 7], qpel8_mc31_3dnow)
3056 SET_QPEL_FUNC(qpel_pixels_tab[1][ 8], qpel8_mc02_3dnow)
3057 SET_QPEL_FUNC(qpel_pixels_tab[1][ 9], qpel8_mc12_3dnow)
3058 SET_QPEL_FUNC(qpel_pixels_tab[1][10], qpel8_mc22_3dnow)
3059 SET_QPEL_FUNC(qpel_pixels_tab[1][11], qpel8_mc32_3dnow)
3060 SET_QPEL_FUNC(qpel_pixels_tab[1][12], qpel8_mc03_3dnow)
3061 SET_QPEL_FUNC(qpel_pixels_tab[1][13], qpel8_mc13_3dnow)
3062 SET_QPEL_FUNC(qpel_pixels_tab[1][14], qpel8_mc23_3dnow)
3063 SET_QPEL_FUNC(qpel_pixels_tab[1][15], qpel8_mc33_3dnow)
3064
3065#define dspfunc(PFX, IDX, NUM) \
3066 c->PFX ## _pixels_tab[IDX][ 0] = PFX ## NUM ## _mc00_3dnow; \
3067 c->PFX ## _pixels_tab[IDX][ 1] = PFX ## NUM ## _mc10_3dnow; \
3068 c->PFX ## _pixels_tab[IDX][ 2] = PFX ## NUM ## _mc20_3dnow; \
3069 c->PFX ## _pixels_tab[IDX][ 3] = PFX ## NUM ## _mc30_3dnow; \
3070 c->PFX ## _pixels_tab[IDX][ 4] = PFX ## NUM ## _mc01_3dnow; \
3071 c->PFX ## _pixels_tab[IDX][ 5] = PFX ## NUM ## _mc11_3dnow; \
3072 c->PFX ## _pixels_tab[IDX][ 6] = PFX ## NUM ## _mc21_3dnow; \
3073 c->PFX ## _pixels_tab[IDX][ 7] = PFX ## NUM ## _mc31_3dnow; \
3074 c->PFX ## _pixels_tab[IDX][ 8] = PFX ## NUM ## _mc02_3dnow; \
3075 c->PFX ## _pixels_tab[IDX][ 9] = PFX ## NUM ## _mc12_3dnow; \
3076 c->PFX ## _pixels_tab[IDX][10] = PFX ## NUM ## _mc22_3dnow; \
3077 c->PFX ## _pixels_tab[IDX][11] = PFX ## NUM ## _mc32_3dnow; \
3078 c->PFX ## _pixels_tab[IDX][12] = PFX ## NUM ## _mc03_3dnow; \
3079 c->PFX ## _pixels_tab[IDX][13] = PFX ## NUM ## _mc13_3dnow; \
3080 c->PFX ## _pixels_tab[IDX][14] = PFX ## NUM ## _mc23_3dnow; \
3081 c->PFX ## _pixels_tab[IDX][15] = PFX ## NUM ## _mc33_3dnow
3082
3083 dspfunc(put_h264_qpel, 0, 16);
3084 dspfunc(put_h264_qpel, 1, 8);
3085 dspfunc(put_h264_qpel, 2, 4);
3086 dspfunc(avg_h264_qpel, 0, 16);
3087 dspfunc(avg_h264_qpel, 1, 8);
3088 dspfunc(avg_h264_qpel, 2, 4);
3089
3090 c->avg_h264_chroma_pixels_tab[0]= avg_h264_chroma_mc8_3dnow;
3091 c->avg_h264_chroma_pixels_tab[1]= avg_h264_chroma_mc4_3dnow;
3092 }
3093
3094#ifdef CONFIG_SNOW_ENCODER
3095 if(mm_flags & MM_SSE2){
3096 c->horizontal_compose97i = ff_snow_horizontal_compose97i_sse2;
3097 c->vertical_compose97i = ff_snow_vertical_compose97i_sse2;
3098 c->inner_add_yblock = ff_snow_inner_add_yblock_sse2;
3099 }
3100 else{
3101 c->horizontal_compose97i = ff_snow_horizontal_compose97i_mmx;
3102 c->vertical_compose97i = ff_snow_vertical_compose97i_mmx;
3103 c->inner_add_yblock = ff_snow_inner_add_yblock_mmx;
3104 }
3105#endif
3106 }
3107
3108#ifdef CONFIG_ENCODERS
3109 dsputil_init_pix_mmx(c, avctx);
3110#endif //CONFIG_ENCODERS
3111#if 0
3112 // for speed testing
3113 get_pixels = just_return;
3114 put_pixels_clamped = just_return;
3115 add_pixels_clamped = just_return;
3116
3117 pix_abs16x16 = just_return;
3118 pix_abs16x16_x2 = just_return;
3119 pix_abs16x16_y2 = just_return;
3120 pix_abs16x16_xy2 = just_return;
3121
3122 put_pixels_tab[0] = just_return;
3123 put_pixels_tab[1] = just_return;
3124 put_pixels_tab[2] = just_return;
3125 put_pixels_tab[3] = just_return;
3126
3127 put_no_rnd_pixels_tab[0] = just_return;
3128 put_no_rnd_pixels_tab[1] = just_return;
3129 put_no_rnd_pixels_tab[2] = just_return;
3130 put_no_rnd_pixels_tab[3] = just_return;
3131
3132 avg_pixels_tab[0] = just_return;
3133 avg_pixels_tab[1] = just_return;
3134 avg_pixels_tab[2] = just_return;
3135 avg_pixels_tab[3] = just_return;
3136
3137 avg_no_rnd_pixels_tab[0] = just_return;
3138 avg_no_rnd_pixels_tab[1] = just_return;
3139 avg_no_rnd_pixels_tab[2] = just_return;
3140 avg_no_rnd_pixels_tab[3] = just_return;
3141
3142 //av_fdct = just_return;
3143 //ff_idct = just_return;
3144#endif
3145}
Note: See TracBrowser for help on using the repository browser.

© 2025 Oracle Support Privacy / Do Not Sell My Info Terms of Use Trademark Policy Automated Access Etiquette