FFmpeg
•All Data Structures Namespaces Files Functions Variables Typedefs Enumerations Enumerator Macros Modules Pages
hpeldsp_init.c
Go to the documentation of this file.
1 /*
2  * SIMD-optimized halfpel functions
3  * Copyright (c) 2000, 2001 Fabrice Bellard
4  * Copyright (c) 2002-2004 Michael Niedermayer <michaelni@gmx.at>
5  *
6  * This file is part of FFmpeg.
7  *
8  * FFmpeg is free software; you can redistribute it and/or
9  * modify it under the terms of the GNU Lesser General Public
10  * License as published by the Free Software Foundation; either
11  * version 2.1 of the License, or (at your option) any later version.
12  *
13  * FFmpeg is distributed in the hope that it will be useful,
14  * but WITHOUT ANY WARRANTY; without even the implied warranty of
15  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
16  * Lesser General Public License for more details.
17  *
18  * You should have received a copy of the GNU Lesser General Public
19  * License along with FFmpeg; if not, write to the Free Software
20  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
21  *
22  * MMX optimization by Nick Kurshev <nickols_k@mail.ru>
23  */
24 
25 #include <stddef.h>
26 #include <stdint.h>
27 
28 #include "libavutil/attributes.h"
29 #include "libavutil/cpu.h"
30 #include "libavutil/x86/cpu.h"
31 #include "libavcodec/avcodec.h"
32 #include "libavcodec/hpeldsp.h"
33 #include "libavcodec/pixels.h"
34 #include "fpel.h"
35 #include "hpeldsp.h"
36 
37 void ff_put_pixels8_x2_mmxext(uint8_t *block, const uint8_t *pixels,
38  ptrdiff_t line_size, int h);
39 void ff_put_pixels16_x2_mmxext(uint8_t *block, const uint8_t *pixels,
40  ptrdiff_t line_size, int h);
41 void ff_put_pixels16_x2_sse2(uint8_t *block, const uint8_t *pixels,
42  ptrdiff_t line_size, int h);
43 void ff_avg_pixels16_x2_sse2(uint8_t *block, const uint8_t *pixels,
44  ptrdiff_t line_size, int h);
45 void ff_put_pixels16_y2_sse2(uint8_t *block, const uint8_t *pixels,
46  ptrdiff_t line_size, int h);
47 void ff_avg_pixels16_y2_sse2(uint8_t *block, const uint8_t *pixels,
48  ptrdiff_t line_size, int h);
49 void ff_put_no_rnd_pixels8_x2_mmxext(uint8_t *block, const uint8_t *pixels,
50  ptrdiff_t line_size, int h);
51 void ff_put_no_rnd_pixels8_x2_exact_mmxext(uint8_t *block,
52  const uint8_t *pixels,
53  ptrdiff_t line_size, int h);
54 void ff_put_pixels8_y2_mmxext(uint8_t *block, const uint8_t *pixels,
55  ptrdiff_t line_size, int h);
56 void ff_put_no_rnd_pixels8_y2_mmxext(uint8_t *block, const uint8_t *pixels,
57  ptrdiff_t line_size, int h);
58 void ff_put_no_rnd_pixels8_y2_exact_mmxext(uint8_t *block,
59  const uint8_t *pixels,
60  ptrdiff_t line_size, int h);
61 void ff_avg_pixels8_x2_mmxext(uint8_t *block, const uint8_t *pixels,
62  ptrdiff_t line_size, int h);
63 void ff_avg_pixels8_y2_mmxext(uint8_t *block, const uint8_t *pixels,
64  ptrdiff_t line_size, int h);
65 void ff_avg_approx_pixels8_xy2_mmxext(uint8_t *block, const uint8_t *pixels,
66  ptrdiff_t line_size, int h);
67 
68 #define put_pixels8_mmx ff_put_pixels8_mmx
69 #define put_pixels16_mmx ff_put_pixels16_mmx
70 #define put_pixels8_xy2_mmx ff_put_pixels8_xy2_mmx
71 #define put_no_rnd_pixels8_mmx ff_put_pixels8_mmx
72 #define put_no_rnd_pixels16_mmx ff_put_pixels16_mmx
73 
74 #if HAVE_INLINE_ASM
75 
76 /***********************************/
77 /* MMX no rounding */
78 #define DEF(x, y) x ## _no_rnd_ ## y ## _mmx
79 #define SET_RND MOVQ_WONE
80 #define STATIC static
81 
82 #include "rnd_template.c"
83 
84 #undef DEF
85 #undef SET_RND
86 #undef STATIC
87 
88 // this routine is 'slightly' suboptimal but mostly unused
89 static void avg_no_rnd_pixels8_xy2_mmx(uint8_t *block, const uint8_t *pixels,
90  ptrdiff_t line_size, int h)
91 {
92  MOVQ_ZERO(mm7);
93  MOVQ_WONE(mm6); // =2 for rnd and =1 for no_rnd version
94  __asm__ volatile(
95  "movq (%1), %%mm0 \n\t"
96  "movq 1(%1), %%mm4 \n\t"
97  "movq %%mm0, %%mm1 \n\t"
98  "movq %%mm4, %%mm5 \n\t"
99  "punpcklbw %%mm7, %%mm0 \n\t"
100  "punpcklbw %%mm7, %%mm4 \n\t"
101  "punpckhbw %%mm7, %%mm1 \n\t"
102  "punpckhbw %%mm7, %%mm5 \n\t"
103  "paddusw %%mm0, %%mm4 \n\t"
104  "paddusw %%mm1, %%mm5 \n\t"
105  "xor %%"FF_REG_a", %%"FF_REG_a" \n\t"
106  "add %3, %1 \n\t"
107  ".p2align 3 \n\t"
108  "1: \n\t"
109  "movq (%1, %%"FF_REG_a"), %%mm0 \n\t"
110  "movq 1(%1, %%"FF_REG_a"), %%mm2 \n\t"
111  "movq %%mm0, %%mm1 \n\t"
112  "movq %%mm2, %%mm3 \n\t"
113  "punpcklbw %%mm7, %%mm0 \n\t"
114  "punpcklbw %%mm7, %%mm2 \n\t"
115  "punpckhbw %%mm7, %%mm1 \n\t"
116  "punpckhbw %%mm7, %%mm3 \n\t"
117  "paddusw %%mm2, %%mm0 \n\t"
118  "paddusw %%mm3, %%mm1 \n\t"
119  "paddusw %%mm6, %%mm4 \n\t"
120  "paddusw %%mm6, %%mm5 \n\t"
121  "paddusw %%mm0, %%mm4 \n\t"
122  "paddusw %%mm1, %%mm5 \n\t"
123  "psrlw $2, %%mm4 \n\t"
124  "psrlw $2, %%mm5 \n\t"
125  "movq (%2, %%"FF_REG_a"), %%mm3 \n\t"
126  "packuswb %%mm5, %%mm4 \n\t"
127  "pcmpeqd %%mm2, %%mm2 \n\t"
128  "paddb %%mm2, %%mm2 \n\t"
129  PAVGB_MMX(%%mm3, %%mm4, %%mm5, %%mm2)
130  "movq %%mm5, (%2, %%"FF_REG_a") \n\t"
131  "add %3, %%"FF_REG_a" \n\t"
132 
133  "movq (%1, %%"FF_REG_a"), %%mm2 \n\t" // 0 <-> 2 1 <-> 3
134  "movq 1(%1, %%"FF_REG_a"), %%mm4 \n\t"
135  "movq %%mm2, %%mm3 \n\t"
136  "movq %%mm4, %%mm5 \n\t"
137  "punpcklbw %%mm7, %%mm2 \n\t"
138  "punpcklbw %%mm7, %%mm4 \n\t"
139  "punpckhbw %%mm7, %%mm3 \n\t"
140  "punpckhbw %%mm7, %%mm5 \n\t"
141  "paddusw %%mm2, %%mm4 \n\t"
142  "paddusw %%mm3, %%mm5 \n\t"
143  "paddusw %%mm6, %%mm0 \n\t"
144  "paddusw %%mm6, %%mm1 \n\t"
145  "paddusw %%mm4, %%mm0 \n\t"
146  "paddusw %%mm5, %%mm1 \n\t"
147  "psrlw $2, %%mm0 \n\t"
148  "psrlw $2, %%mm1 \n\t"
149  "movq (%2, %%"FF_REG_a"), %%mm3 \n\t"
150  "packuswb %%mm1, %%mm0 \n\t"
151  "pcmpeqd %%mm2, %%mm2 \n\t"
152  "paddb %%mm2, %%mm2 \n\t"
153  PAVGB_MMX(%%mm3, %%mm0, %%mm1, %%mm2)
154  "movq %%mm1, (%2, %%"FF_REG_a") \n\t"
155  "add %3, %%"FF_REG_a" \n\t"
156 
157  "subl $2, %0 \n\t"
158  "jnz 1b \n\t"
159  :"+g"(h), "+S"(pixels)
160  :"D"(block), "r"((x86_reg)line_size)
161  :FF_REG_a, "memory");
162 }
163 
164 static void put_no_rnd_pixels8_x2_mmx(uint8_t *block, const uint8_t *pixels, ptrdiff_t line_size, int h)
165 {
166  MOVQ_BFE(mm6);
167  __asm__ volatile(
168  "lea (%3, %3), %%"FF_REG_a" \n\t"
169  ".p2align 3 \n\t"
170  "1: \n\t"
171  "movq (%1), %%mm0 \n\t"
172  "movq 1(%1), %%mm1 \n\t"
173  "movq (%1, %3), %%mm2 \n\t"
174  "movq 1(%1, %3), %%mm3 \n\t"
175  PAVGBP_MMX_NO_RND(%%mm0, %%mm1, %%mm4, %%mm2, %%mm3, %%mm5)
176  "movq %%mm4, (%2) \n\t"
177  "movq %%mm5, (%2, %3) \n\t"
178  "add %%"FF_REG_a", %1 \n\t"
179  "add %%"FF_REG_a", %2 \n\t"
180  "movq (%1), %%mm0 \n\t"
181  "movq 1(%1), %%mm1 \n\t"
182  "movq (%1, %3), %%mm2 \n\t"
183  "movq 1(%1, %3), %%mm3 \n\t"
184  PAVGBP_MMX_NO_RND(%%mm0, %%mm1, %%mm4, %%mm2, %%mm3, %%mm5)
185  "movq %%mm4, (%2) \n\t"
186  "movq %%mm5, (%2, %3) \n\t"
187  "add %%"FF_REG_a", %1 \n\t"
188  "add %%"FF_REG_a", %2 \n\t"
189  "subl $4, %0 \n\t"
190  "jnz 1b \n\t"
191  :"+g"(h), "+S"(pixels), "+D"(block)
192  :"r"((x86_reg)line_size)
193  :FF_REG_a, "memory");
194 }
195 
196 static void put_no_rnd_pixels16_x2_mmx(uint8_t *block, const uint8_t *pixels, ptrdiff_t line_size, int h)
197 {
198  MOVQ_BFE(mm6);
199  __asm__ volatile(
200  "lea (%3, %3), %%"FF_REG_a" \n\t"
201  ".p2align 3 \n\t"
202  "1: \n\t"
203  "movq (%1), %%mm0 \n\t"
204  "movq 1(%1), %%mm1 \n\t"
205  "movq (%1, %3), %%mm2 \n\t"
206  "movq 1(%1, %3), %%mm3 \n\t"
207  PAVGBP_MMX_NO_RND(%%mm0, %%mm1, %%mm4, %%mm2, %%mm3, %%mm5)
208  "movq %%mm4, (%2) \n\t"
209  "movq %%mm5, (%2, %3) \n\t"
210  "movq 8(%1), %%mm0 \n\t"
211  "movq 9(%1), %%mm1 \n\t"
212  "movq 8(%1, %3), %%mm2 \n\t"
213  "movq 9(%1, %3), %%mm3 \n\t"
214  PAVGBP_MMX_NO_RND(%%mm0, %%mm1, %%mm4, %%mm2, %%mm3, %%mm5)
215  "movq %%mm4, 8(%2) \n\t"
216  "movq %%mm5, 8(%2, %3) \n\t"
217  "add %%"FF_REG_a", %1 \n\t"
218  "add %%"FF_REG_a", %2 \n\t"
219  "movq (%1), %%mm0 \n\t"
220  "movq 1(%1), %%mm1 \n\t"
221  "movq (%1, %3), %%mm2 \n\t"
222  "movq 1(%1, %3), %%mm3 \n\t"
223  PAVGBP_MMX_NO_RND(%%mm0, %%mm1, %%mm4, %%mm2, %%mm3, %%mm5)
224  "movq %%mm4, (%2) \n\t"
225  "movq %%mm5, (%2, %3) \n\t"
226  "movq 8(%1), %%mm0 \n\t"
227  "movq 9(%1), %%mm1 \n\t"
228  "movq 8(%1, %3), %%mm2 \n\t"
229  "movq 9(%1, %3), %%mm3 \n\t"
230  PAVGBP_MMX_NO_RND(%%mm0, %%mm1, %%mm4, %%mm2, %%mm3, %%mm5)
231  "movq %%mm4, 8(%2) \n\t"
232  "movq %%mm5, 8(%2, %3) \n\t"
233  "add %%"FF_REG_a", %1 \n\t"
234  "add %%"FF_REG_a", %2 \n\t"
235  "subl $4, %0 \n\t"
236  "jnz 1b \n\t"
237  :"+g"(h), "+S"(pixels), "+D"(block)
238  :"r"((x86_reg)line_size)
239  :FF_REG_a, "memory");
240 }
241 
242 static void put_no_rnd_pixels8_y2_mmx(uint8_t *block, const uint8_t *pixels, ptrdiff_t line_size, int h)
243 {
244  MOVQ_BFE(mm6);
245  __asm__ volatile(
246  "lea (%3, %3), %%"FF_REG_a" \n\t"
247  "movq (%1), %%mm0 \n\t"
248  ".p2align 3 \n\t"
249  "1: \n\t"
250  "movq (%1, %3), %%mm1 \n\t"
251  "movq (%1, %%"FF_REG_a"),%%mm2\n\t"
252  PAVGBP_MMX_NO_RND(%%mm1, %%mm0, %%mm4, %%mm2, %%mm1, %%mm5)
253  "movq %%mm4, (%2) \n\t"
254  "movq %%mm5, (%2, %3) \n\t"
255  "add %%"FF_REG_a", %1 \n\t"
256  "add %%"FF_REG_a", %2 \n\t"
257  "movq (%1, %3), %%mm1 \n\t"
258  "movq (%1, %%"FF_REG_a"),%%mm0\n\t"
259  PAVGBP_MMX_NO_RND(%%mm1, %%mm2, %%mm4, %%mm0, %%mm1, %%mm5)
260  "movq %%mm4, (%2) \n\t"
261  "movq %%mm5, (%2, %3) \n\t"
262  "add %%"FF_REG_a", %1 \n\t"
263  "add %%"FF_REG_a", %2 \n\t"
264  "subl $4, %0 \n\t"
265  "jnz 1b \n\t"
266  :"+g"(h), "+S"(pixels), "+D"(block)
267  :"r"((x86_reg)line_size)
268  :FF_REG_a, "memory");
269 }
270 
271 static void avg_no_rnd_pixels16_x2_mmx(uint8_t *block, const uint8_t *pixels, ptrdiff_t line_size, int h)
272 {
273  MOVQ_BFE(mm6);
274  __asm__ volatile(
275  ".p2align 3 \n\t"
276  "1: \n\t"
277  "movq (%1), %%mm0 \n\t"
278  "movq 1(%1), %%mm1 \n\t"
279  "movq (%2), %%mm3 \n\t"
280  PAVGB_MMX_NO_RND(%%mm0, %%mm1, %%mm2, %%mm6)
281  PAVGB_MMX(%%mm3, %%mm2, %%mm0, %%mm6)
282  "movq %%mm0, (%2) \n\t"
283  "movq 8(%1), %%mm0 \n\t"
284  "movq 9(%1), %%mm1 \n\t"
285  "movq 8(%2), %%mm3 \n\t"
286  PAVGB_MMX_NO_RND(%%mm0, %%mm1, %%mm2, %%mm6)
287  PAVGB_MMX(%%mm3, %%mm2, %%mm0, %%mm6)
288  "movq %%mm0, 8(%2) \n\t"
289  "add %3, %1 \n\t"
290  "add %3, %2 \n\t"
291  "subl $1, %0 \n\t"
292  "jnz 1b \n\t"
293  :"+g"(h), "+S"(pixels), "+D"(block)
294  :"r"((x86_reg)line_size)
295  :"memory");
296 }
297 
298 static void avg_no_rnd_pixels8_y2_mmx(uint8_t *block, const uint8_t *pixels, ptrdiff_t line_size, int h)
299 {
300  MOVQ_BFE(mm6);
301  __asm__ volatile(
302  "lea (%3, %3), %%"FF_REG_a" \n\t"
303  "movq (%1), %%mm0 \n\t"
304  ".p2align 3 \n\t"
305  "1: \n\t"
306  "movq (%1, %3), %%mm1 \n\t"
307  "movq (%1, %%"FF_REG_a"), %%mm2 \n\t"
308  PAVGBP_MMX_NO_RND(%%mm1, %%mm0, %%mm4, %%mm2, %%mm1, %%mm5)
309  "movq (%2), %%mm3 \n\t"
310  PAVGB_MMX(%%mm3, %%mm4, %%mm0, %%mm6)
311  "movq (%2, %3), %%mm3 \n\t"
312  PAVGB_MMX(%%mm3, %%mm5, %%mm1, %%mm6)
313  "movq %%mm0, (%2) \n\t"
314  "movq %%mm1, (%2, %3) \n\t"
315  "add %%"FF_REG_a", %1 \n\t"
316  "add %%"FF_REG_a", %2 \n\t"
317 
318  "movq (%1, %3), %%mm1 \n\t"
319  "movq (%1, %%"FF_REG_a"), %%mm0 \n\t"
320  PAVGBP_MMX_NO_RND(%%mm1, %%mm2, %%mm4, %%mm0, %%mm1, %%mm5)
321  "movq (%2), %%mm3 \n\t"
322  PAVGB_MMX(%%mm3, %%mm4, %%mm2, %%mm6)
323  "movq (%2, %3), %%mm3 \n\t"
324  PAVGB_MMX(%%mm3, %%mm5, %%mm1, %%mm6)
325  "movq %%mm2, (%2) \n\t"
326  "movq %%mm1, (%2, %3) \n\t"
327  "add %%"FF_REG_a", %1 \n\t"
328  "add %%"FF_REG_a", %2 \n\t"
329 
330  "subl $4, %0 \n\t"
331  "jnz 1b \n\t"
332  :"+g"(h), "+S"(pixels), "+D"(block)
333  :"r"((x86_reg)line_size)
334  :FF_REG_a, "memory");
335 }
336 
337 #if HAVE_MMX
338 CALL_2X_PIXELS(avg_no_rnd_pixels16_y2_mmx, avg_no_rnd_pixels8_y2_mmx, 8)
339 CALL_2X_PIXELS(put_no_rnd_pixels16_y2_mmx, put_no_rnd_pixels8_y2_mmx, 8)
340 
341 CALL_2X_PIXELS(avg_no_rnd_pixels16_xy2_mmx, avg_no_rnd_pixels8_xy2_mmx, 8)
342 CALL_2X_PIXELS(put_no_rnd_pixels16_xy2_mmx, put_no_rnd_pixels8_xy2_mmx, 8)
343 #endif
344 
345 /***********************************/
346 /* MMX rounding */
347 
348 #define SET_RND MOVQ_WTWO
349 #define DEF(x, y) ff_ ## x ## _ ## y ## _mmx
350 #define STATIC
351 
352 #include "rnd_template.c"
353 
354 #undef NO_AVG
355 #undef DEF
356 #undef SET_RND
357 
358 #if HAVE_MMX
359 CALL_2X_PIXELS(put_pixels16_xy2_mmx, ff_put_pixels8_xy2_mmx, 8)
360 #endif
361 
362 #endif /* HAVE_INLINE_ASM */
363 
364 
365 #if HAVE_X86ASM
366 
367 #define HPELDSP_AVG_PIXELS16(CPUEXT) \
368  CALL_2X_PIXELS(put_no_rnd_pixels16_x2 ## CPUEXT, ff_put_no_rnd_pixels8_x2 ## CPUEXT, 8) \
369  CALL_2X_PIXELS(put_pixels16_y2 ## CPUEXT, ff_put_pixels8_y2 ## CPUEXT, 8) \
370  CALL_2X_PIXELS(put_no_rnd_pixels16_y2 ## CPUEXT, ff_put_no_rnd_pixels8_y2 ## CPUEXT, 8) \
371  CALL_2X_PIXELS(avg_pixels16_x2 ## CPUEXT, ff_avg_pixels8_x2 ## CPUEXT, 8) \
372  CALL_2X_PIXELS(avg_pixels16_y2 ## CPUEXT, ff_avg_pixels8_y2 ## CPUEXT, 8) \
373  CALL_2X_PIXELS(avg_pixels16_xy2 ## CPUEXT, ff_avg_pixels8_xy2 ## CPUEXT, 8) \
374  CALL_2X_PIXELS(avg_approx_pixels16_xy2## CPUEXT, ff_avg_approx_pixels8_xy2## CPUEXT, 8)
375 
376 HPELDSP_AVG_PIXELS16(_mmxext)
377 
378 #endif /* HAVE_X86ASM */
379 
380 #define SET_HPEL_FUNCS_EXT(PFX, IDX, SIZE, CPU) \
381  if (HAVE_MMX_EXTERNAL) \
382  c->PFX ## _pixels_tab IDX [0] = PFX ## _pixels ## SIZE ## _ ## CPU
383 
384 #define SET_HPEL_FUNCS03(PFX, IDX, SIZE, CPU) \
385  do { \
386  SET_HPEL_FUNCS_EXT(PFX, IDX, SIZE, CPU); \
387  c->PFX ## _pixels_tab IDX [3] = PFX ## _pixels ## SIZE ## _xy2_ ## CPU; \
388  } while (0)
389 #define SET_HPEL_FUNCS12(PFX, IDX, SIZE, CPU) \
390  do { \
391  c->PFX ## _pixels_tab IDX [1] = PFX ## _pixels ## SIZE ## _x2_ ## CPU; \
392  c->PFX ## _pixels_tab IDX [2] = PFX ## _pixels ## SIZE ## _y2_ ## CPU; \
393  } while (0)
394 #define SET_HPEL_FUNCS(PFX, IDX, SIZE, CPU) \
395  do { \
396  SET_HPEL_FUNCS03(PFX, IDX, SIZE, CPU); \
397  SET_HPEL_FUNCS12(PFX, IDX, SIZE, CPU); \
398  } while (0)
399 
400 static void hpeldsp_init_mmx(HpelDSPContext *c, int flags)
401 {
402 #if HAVE_MMX_INLINE
403  SET_HPEL_FUNCS03(put, [0], 16, mmx);
404  SET_HPEL_FUNCS(put_no_rnd, [0], 16, mmx);
405  SET_HPEL_FUNCS12(avg_no_rnd, , 16, mmx);
406  c->avg_no_rnd_pixels_tab[3] = avg_no_rnd_pixels16_xy2_mmx;
407  SET_HPEL_FUNCS03(put, [1], 8, mmx);
408  SET_HPEL_FUNCS(put_no_rnd, [1], 8, mmx);
409 #endif
410 }
411 
412 static void hpeldsp_init_mmxext(HpelDSPContext *c, int flags)
413 {
414 #if HAVE_MMXEXT_EXTERNAL
415  c->put_pixels_tab[0][1] = ff_put_pixels16_x2_mmxext;
416  c->put_pixels_tab[0][2] = put_pixels16_y2_mmxext;
417 
418  c->avg_pixels_tab[0][0] = ff_avg_pixels16_mmxext;
419  c->avg_pixels_tab[0][1] = avg_pixels16_x2_mmxext;
420  c->avg_pixels_tab[0][2] = avg_pixels16_y2_mmxext;
421  c->avg_pixels_tab[0][3] = avg_pixels16_xy2_mmxext;
422 
423  c->put_pixels_tab[1][1] = ff_put_pixels8_x2_mmxext;
424  c->put_pixels_tab[1][2] = ff_put_pixels8_y2_mmxext;
425 
426  c->avg_pixels_tab[1][0] = ff_avg_pixels8_mmxext;
427  c->avg_pixels_tab[1][1] = ff_avg_pixels8_x2_mmxext;
428  c->avg_pixels_tab[1][2] = ff_avg_pixels8_y2_mmxext;
429  c->avg_pixels_tab[1][3] = ff_avg_pixels8_xy2_mmxext;
430 
431  c->put_no_rnd_pixels_tab[1][1] = ff_put_no_rnd_pixels8_x2_exact_mmxext;
432  c->put_no_rnd_pixels_tab[1][2] = ff_put_no_rnd_pixels8_y2_exact_mmxext;
433 
434  c->avg_no_rnd_pixels_tab[0] = ff_avg_pixels16_mmxext;
435 
436  if (!(flags & AV_CODEC_FLAG_BITEXACT)) {
437  c->put_no_rnd_pixels_tab[0][1] = put_no_rnd_pixels16_x2_mmxext;
438  c->put_no_rnd_pixels_tab[0][2] = put_no_rnd_pixels16_y2_mmxext;
439  c->put_no_rnd_pixels_tab[1][1] = ff_put_no_rnd_pixels8_x2_mmxext;
440  c->put_no_rnd_pixels_tab[1][2] = ff_put_no_rnd_pixels8_y2_mmxext;
441 
442  c->avg_pixels_tab[0][3] = avg_approx_pixels16_xy2_mmxext;
443  c->avg_pixels_tab[1][3] = ff_avg_approx_pixels8_xy2_mmxext;
444  }
445 #endif /* HAVE_MMXEXT_EXTERNAL */
446 }
447 
448 static void hpeldsp_init_sse2_fast(HpelDSPContext *c, int flags)
449 {
450 #if HAVE_SSE2_EXTERNAL
451  c->put_pixels_tab[0][0] = ff_put_pixels16_sse2;
452  c->put_no_rnd_pixels_tab[0][0] = ff_put_pixels16_sse2;
453  c->put_pixels_tab[0][1] = ff_put_pixels16_x2_sse2;
454  c->put_pixels_tab[0][2] = ff_put_pixels16_y2_sse2;
455  c->put_pixels_tab[0][3] = ff_put_pixels16_xy2_sse2;
456  c->avg_pixels_tab[0][0] = ff_avg_pixels16_sse2;
457  c->avg_pixels_tab[0][1] = ff_avg_pixels16_x2_sse2;
458  c->avg_pixels_tab[0][2] = ff_avg_pixels16_y2_sse2;
459  c->avg_pixels_tab[0][3] = ff_avg_pixels16_xy2_sse2;
460  c->avg_no_rnd_pixels_tab[0] = ff_avg_pixels16_sse2;
461 #endif /* HAVE_SSE2_EXTERNAL */
462 }
463 
464 static void hpeldsp_init_ssse3(HpelDSPContext *c, int flags)
465 {
466 #if HAVE_SSSE3_EXTERNAL
467  c->put_pixels_tab[0][3] = ff_put_pixels16_xy2_ssse3;
468  c->avg_pixels_tab[0][3] = ff_avg_pixels16_xy2_ssse3;
469  c->put_pixels_tab[1][3] = ff_put_pixels8_xy2_ssse3;
470  c->avg_pixels_tab[1][3] = ff_avg_pixels8_xy2_ssse3;
471 #endif
472 }
473 
474 av_cold void ff_hpeldsp_init_x86(HpelDSPContext *c, int flags)
475 {
476  int cpu_flags = av_get_cpu_flags();
477 
478  if (INLINE_MMX(cpu_flags))
479  hpeldsp_init_mmx(c, flags);
480 
481  if (EXTERNAL_MMXEXT(cpu_flags))
482  hpeldsp_init_mmxext(c, flags);
483 
484  if (EXTERNAL_SSE2_FAST(cpu_flags))
485  hpeldsp_init_sse2_fast(c, flags);
486 
487  if (EXTERNAL_SSSE3(cpu_flags))
488  hpeldsp_init_ssse3(c, flags);
489 }
flags
const SwsFlags flags[]
Definition: swscale.c:61
INLINE_MMX
#define INLINE_MMX(flags)
Definition: cpu.h:87
PAVGB_MMX
#define PAVGB_MMX(rega, regb, regr, regfe)
Definition: inline_asm.h:63
cpu.h
EXTERNAL_SSE2_FAST
#define EXTERNAL_SSE2_FAST(flags)
Definition: cpu.h:60
ff_put_pixels16_y2_sse2
void ff_put_pixels16_y2_sse2(uint8_t *block, const uint8_t *pixels, ptrdiff_t line_size, int h)
ff_put_no_rnd_pixels8_x2_exact_mmxext
void ff_put_no_rnd_pixels8_x2_exact_mmxext(uint8_t *block, const uint8_t *pixels, ptrdiff_t line_size, int h)
ff_put_pixels8_xy2_ssse3
void ff_put_pixels8_xy2_ssse3(uint8_t *block, const uint8_t *pixels, ptrdiff_t line_size, int h)
x86_reg
int x86_reg
Definition: asm.h:72
pixels.h
MOVQ_BFE
#define MOVQ_BFE(regd)
Definition: inline_asm.h:34
av_get_cpu_flags
int av_get_cpu_flags(void)
Return the flags which specify extensions supported by the CPU.
Definition: cpu.c:109
cpu_flags
static atomic_int cpu_flags
Definition: cpu.c:56
SET_HPEL_FUNCS12
#define SET_HPEL_FUNCS12(PFX, IDX, SIZE, CPU)
Definition: hpeldsp_init.c:389
ff_avg_pixels8_xy2_mmxext
void ff_avg_pixels8_xy2_mmxext(uint8_t *block, const uint8_t *pixels, ptrdiff_t line_size, int h)
hpeldsp_init_ssse3
static void hpeldsp_init_ssse3(HpelDSPContext *c, int flags)
Definition: hpeldsp_init.c:464
ff_put_no_rnd_pixels8_y2_exact_mmxext
void ff_put_no_rnd_pixels8_y2_exact_mmxext(uint8_t *block, const uint8_t *pixels, ptrdiff_t line_size, int h)
rnd_template.c
ff_hpeldsp_init_x86
av_cold void ff_hpeldsp_init_x86(HpelDSPContext *c, int flags)
Definition: hpeldsp_init.c:474
ff_avg_pixels8_y2_mmxext
void ff_avg_pixels8_y2_mmxext(uint8_t *block, const uint8_t *pixels, ptrdiff_t line_size, int h)
MOVQ_ZERO
#define MOVQ_ZERO(regd)
Definition: inline_asm.h:32
ff_put_pixels8_xy2_mmx
void ff_put_pixels8_xy2_mmx(uint8_t *block, const uint8_t *pixels, ptrdiff_t line_size, int h)
av_cold
#define av_cold
Definition: attributes.h:90
ff_avg_pixels8_x2_mmxext
void ff_avg_pixels8_x2_mmxext(uint8_t *block, const uint8_t *pixels, ptrdiff_t line_size, int h)
hpeldsp_init_mmxext
static void hpeldsp_init_mmxext(HpelDSPContext *c, int flags)
Definition: hpeldsp_init.c:412
ff_avg_pixels8_mmxext
void ff_avg_pixels8_mmxext(uint8_t *block, const uint8_t *pixels, ptrdiff_t line_size, int h)
SET_HPEL_FUNCS03
#define SET_HPEL_FUNCS03(PFX, IDX, SIZE, CPU)
Definition: hpeldsp_init.c:384
ff_put_pixels16_sse2
void ff_put_pixels16_sse2(uint8_t *block, const uint8_t *pixels, ptrdiff_t line_size, int h)
ff_avg_pixels16_y2_sse2
void ff_avg_pixels16_y2_sse2(uint8_t *block, const uint8_t *pixels, ptrdiff_t line_size, int h)
ff_avg_pixels16_xy2_sse2
void ff_avg_pixels16_xy2_sse2(uint8_t *block, const uint8_t *pixels, ptrdiff_t line_size, int h)
MOVQ_WONE
#define MOVQ_WONE(regd)
Definition: inline_asm.h:26
PAVGBP_MMX_NO_RND
#define PAVGBP_MMX_NO_RND(rega, regb, regr, regc, regd, regp)
Definition: inline_asm.h:72
hpeldsp_init_sse2_fast
static void hpeldsp_init_sse2_fast(HpelDSPContext *c, int flags)
Definition: hpeldsp_init.c:448
c
Undefined Behavior In the C some operations are like signed integer dereferencing freed accessing outside allocated Undefined Behavior must not occur in a C it is not safe even if the output of undefined operations is unused The unsafety may seem nit picking but Optimizing compilers have in fact optimized code on the assumption that no undefined Behavior occurs Optimizing code based on wrong assumptions can and has in some cases lead to effects beyond the output of computations The signed integer overflow problem in speed critical code Code which is highly optimized and works with signed integers sometimes has the problem that often the output of the computation does not c
Definition: undefined.txt:32
ff_avg_approx_pixels8_xy2_mmxext
void ff_avg_approx_pixels8_xy2_mmxext(uint8_t *block, const uint8_t *pixels, ptrdiff_t line_size, int h)
ff_put_pixels16_xy2_ssse3
void ff_put_pixels16_xy2_ssse3(uint8_t *block, const uint8_t *pixels, ptrdiff_t line_size, int h)
HpelDSPContext
Half-pel DSP context.
Definition: hpeldsp.h:45
ff_put_pixels16_x2_sse2
void ff_put_pixels16_x2_sse2(uint8_t *block, const uint8_t *pixels, ptrdiff_t line_size, int h)
hpeldsp_init_mmx
static void hpeldsp_init_mmx(HpelDSPContext *c, int flags)
Definition: hpeldsp_init.c:400
ff_avg_pixels16_sse2
void ff_avg_pixels16_sse2(uint8_t *block, const uint8_t *pixels, ptrdiff_t line_size, int h)
cpu.h
ff_avg_pixels16_mmxext
void ff_avg_pixels16_mmxext(uint8_t *block, const uint8_t *pixels, ptrdiff_t line_size, int h)
ff_put_pixels16_x2_mmxext
void ff_put_pixels16_x2_mmxext(uint8_t *block, const uint8_t *pixels, ptrdiff_t line_size, int h)
attributes.h
ff_put_pixels16_xy2_sse2
void ff_put_pixels16_xy2_sse2(uint8_t *block, const uint8_t *pixels, ptrdiff_t line_size, int h)
ff_put_pixels8_x2_mmxext
void ff_put_pixels8_x2_mmxext(uint8_t *block, const uint8_t *pixels, ptrdiff_t line_size, int h)
avcodec.h
ff_avg_pixels16_xy2_ssse3
void ff_avg_pixels16_xy2_ssse3(uint8_t *block, const uint8_t *pixels, ptrdiff_t line_size, int h)
__asm__
__asm__(".macro parse_r var r\n\t" "\\var = -1\n\t" _IFC_REG(0) _IFC_REG(1) _IFC_REG(2) _IFC_REG(3) _IFC_REG(4) _IFC_REG(5) _IFC_REG(6) _IFC_REG(7) _IFC_REG(8) _IFC_REG(9) _IFC_REG(10) _IFC_REG(11) _IFC_REG(12) _IFC_REG(13) _IFC_REG(14) _IFC_REG(15) _IFC_REG(16) _IFC_REG(17) _IFC_REG(18) _IFC_REG(19) _IFC_REG(20) _IFC_REG(21) _IFC_REG(22) _IFC_REG(23) _IFC_REG(24) _IFC_REG(25) _IFC_REG(26) _IFC_REG(27) _IFC_REG(28) _IFC_REG(29) _IFC_REG(30) _IFC_REG(31) ".iflt \\var\n\t" ".error \"Unable to parse register name \\r\"\n\t" ".endif\n\t" ".endm")
SET_HPEL_FUNCS
#define SET_HPEL_FUNCS(PFX, IDX, SIZE, CPU)
Definition: hpeldsp_init.c:394
CALL_2X_PIXELS
#define CALL_2X_PIXELS(a, b, n)
Definition: pixels.h:34
ff_avg_pixels16_x2_sse2
void ff_avg_pixels16_x2_sse2(uint8_t *block, const uint8_t *pixels, ptrdiff_t line_size, int h)
ff_put_pixels8_y2_mmxext
void ff_put_pixels8_y2_mmxext(uint8_t *block, const uint8_t *pixels, ptrdiff_t line_size, int h)
fpel.h
AV_CODEC_FLAG_BITEXACT
#define AV_CODEC_FLAG_BITEXACT
Use only bitexact stuff (except (I)DCT).
Definition: avcodec.h:322
ff_put_no_rnd_pixels8_y2_mmxext
void ff_put_no_rnd_pixels8_y2_mmxext(uint8_t *block, const uint8_t *pixels, ptrdiff_t line_size, int h)
hpeldsp.h
block
The exact code depends on how similar the blocks are and how related they are to the block
Definition: filter_design.txt:207
ff_avg_pixels8_xy2_ssse3
void ff_avg_pixels8_xy2_ssse3(uint8_t *block, const uint8_t *pixels, ptrdiff_t line_size, int h)
h
h
Definition: vp9dsp_template.c:2070
EXTERNAL_SSSE3
#define EXTERNAL_SSSE3(flags)
Definition: cpu.h:65
PAVGB_MMX_NO_RND
#define PAVGB_MMX_NO_RND(rega, regb, regr, regfe)
Definition: inline_asm.h:55
ff_put_no_rnd_pixels8_x2_mmxext
void ff_put_no_rnd_pixels8_x2_mmxext(uint8_t *block, const uint8_t *pixels, ptrdiff_t line_size, int h)
EXTERNAL_MMXEXT
#define EXTERNAL_MMXEXT(flags)
Definition: cpu.h:57
Generated on Mon Jul 14 2025 19:22:27 for FFmpeg by   doxygen 1.8.17