FFmpeg
videodsp_init.c
Go to the documentation of this file.
1 /*
2  * Copyright (C) 2002-2012 Michael Niedermayer
3  * Copyright (C) 2012 Ronald S. Bultje
4  *
5  * This file is part of FFmpeg.
6  *
7  * FFmpeg is free software; you can redistribute it and/or
8  * modify it under the terms of the GNU Lesser General Public
9  * License as published by the Free Software Foundation; either
10  * version 2.1 of the License, or (at your option) any later version.
11  *
12  * FFmpeg is distributed in the hope that it will be useful,
13  * but WITHOUT ANY WARRANTY; without even the implied warranty of
14  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15  * Lesser General Public License for more details.
16  *
17  * You should have received a copy of the GNU Lesser General Public
18  * License along with FFmpeg; if not, write to the Free Software
19  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
20  */
21 
22 #include "config.h"
23 #include "libavutil/attributes.h"
24 #include "libavutil/avassert.h"
25 #include "libavutil/common.h"
26 #include "libavutil/cpu.h"
27 #include "libavutil/x86/asm.h"
28 #include "libavutil/x86/cpu.h"
29 #include "libavcodec/videodsp.h"
30 
31 #if HAVE_X86ASM
32 typedef void emu_edge_vfix_func(uint8_t *dst, x86_reg dst_stride,
33  const uint8_t *src, x86_reg src_stride,
34  x86_reg start_y, x86_reg end_y, x86_reg bh);
35 typedef void emu_edge_vvar_func(uint8_t *dst, x86_reg dst_stride,
36  const uint8_t *src, x86_reg src_stride,
37  x86_reg start_y, x86_reg end_y, x86_reg bh,
38  x86_reg w);
39 
40 extern emu_edge_vfix_func ff_emu_edge_vfix1_mmx;
41 extern emu_edge_vfix_func ff_emu_edge_vfix2_mmx;
42 extern emu_edge_vfix_func ff_emu_edge_vfix3_mmx;
43 extern emu_edge_vfix_func ff_emu_edge_vfix4_mmx;
44 extern emu_edge_vfix_func ff_emu_edge_vfix5_mmx;
45 extern emu_edge_vfix_func ff_emu_edge_vfix6_mmx;
46 extern emu_edge_vfix_func ff_emu_edge_vfix7_mmx;
47 extern emu_edge_vfix_func ff_emu_edge_vfix8_mmx;
48 extern emu_edge_vfix_func ff_emu_edge_vfix9_mmx;
49 extern emu_edge_vfix_func ff_emu_edge_vfix10_mmx;
50 extern emu_edge_vfix_func ff_emu_edge_vfix11_mmx;
51 extern emu_edge_vfix_func ff_emu_edge_vfix12_mmx;
52 extern emu_edge_vfix_func ff_emu_edge_vfix13_mmx;
53 extern emu_edge_vfix_func ff_emu_edge_vfix14_mmx;
54 extern emu_edge_vfix_func ff_emu_edge_vfix15_mmx;
55 extern emu_edge_vfix_func ff_emu_edge_vfix16_mmx;
56 extern emu_edge_vfix_func ff_emu_edge_vfix17_mmx;
57 extern emu_edge_vfix_func ff_emu_edge_vfix18_mmx;
58 extern emu_edge_vfix_func ff_emu_edge_vfix19_mmx;
59 extern emu_edge_vfix_func ff_emu_edge_vfix20_mmx;
60 extern emu_edge_vfix_func ff_emu_edge_vfix21_mmx;
61 extern emu_edge_vfix_func ff_emu_edge_vfix22_mmx;
62 #if ARCH_X86_32
63 static emu_edge_vfix_func * const vfixtbl_mmx[22] = {
64  &ff_emu_edge_vfix1_mmx, &ff_emu_edge_vfix2_mmx, &ff_emu_edge_vfix3_mmx,
65  &ff_emu_edge_vfix4_mmx, &ff_emu_edge_vfix5_mmx, &ff_emu_edge_vfix6_mmx,
66  &ff_emu_edge_vfix7_mmx, &ff_emu_edge_vfix8_mmx, &ff_emu_edge_vfix9_mmx,
67  &ff_emu_edge_vfix10_mmx, &ff_emu_edge_vfix11_mmx, &ff_emu_edge_vfix12_mmx,
68  &ff_emu_edge_vfix13_mmx, &ff_emu_edge_vfix14_mmx, &ff_emu_edge_vfix15_mmx,
69  &ff_emu_edge_vfix16_mmx, &ff_emu_edge_vfix17_mmx, &ff_emu_edge_vfix18_mmx,
70  &ff_emu_edge_vfix19_mmx, &ff_emu_edge_vfix20_mmx, &ff_emu_edge_vfix21_mmx,
71  &ff_emu_edge_vfix22_mmx
72 };
73 #endif
74 extern emu_edge_vvar_func ff_emu_edge_vvar_mmx;
75 extern emu_edge_vfix_func ff_emu_edge_vfix16_sse;
76 extern emu_edge_vfix_func ff_emu_edge_vfix17_sse;
77 extern emu_edge_vfix_func ff_emu_edge_vfix18_sse;
78 extern emu_edge_vfix_func ff_emu_edge_vfix19_sse;
79 extern emu_edge_vfix_func ff_emu_edge_vfix20_sse;
80 extern emu_edge_vfix_func ff_emu_edge_vfix21_sse;
81 extern emu_edge_vfix_func ff_emu_edge_vfix22_sse;
82 static emu_edge_vfix_func * const vfixtbl_sse[22] = {
83  ff_emu_edge_vfix1_mmx, ff_emu_edge_vfix2_mmx, ff_emu_edge_vfix3_mmx,
84  ff_emu_edge_vfix4_mmx, ff_emu_edge_vfix5_mmx, ff_emu_edge_vfix6_mmx,
85  ff_emu_edge_vfix7_mmx, ff_emu_edge_vfix8_mmx, ff_emu_edge_vfix9_mmx,
86  ff_emu_edge_vfix10_mmx, ff_emu_edge_vfix11_mmx, ff_emu_edge_vfix12_mmx,
87  ff_emu_edge_vfix13_mmx, ff_emu_edge_vfix14_mmx, ff_emu_edge_vfix15_mmx,
88  ff_emu_edge_vfix16_sse, ff_emu_edge_vfix17_sse, ff_emu_edge_vfix18_sse,
89  ff_emu_edge_vfix19_sse, ff_emu_edge_vfix20_sse, ff_emu_edge_vfix21_sse,
90  ff_emu_edge_vfix22_sse
91 };
92 extern emu_edge_vvar_func ff_emu_edge_vvar_sse;
93 
94 typedef void emu_edge_hfix_func(uint8_t *dst, x86_reg dst_stride,
95  x86_reg start_x, x86_reg bh);
96 typedef void emu_edge_hvar_func(uint8_t *dst, x86_reg dst_stride,
97  x86_reg start_x, x86_reg n_words, x86_reg bh);
98 
99 extern emu_edge_hfix_func ff_emu_edge_hfix2_mmx;
100 extern emu_edge_hfix_func ff_emu_edge_hfix4_mmx;
101 extern emu_edge_hfix_func ff_emu_edge_hfix6_mmx;
102 extern emu_edge_hfix_func ff_emu_edge_hfix8_mmx;
103 extern emu_edge_hfix_func ff_emu_edge_hfix10_mmx;
104 extern emu_edge_hfix_func ff_emu_edge_hfix12_mmx;
105 extern emu_edge_hfix_func ff_emu_edge_hfix14_mmx;
106 extern emu_edge_hfix_func ff_emu_edge_hfix16_mmx;
107 extern emu_edge_hfix_func ff_emu_edge_hfix18_mmx;
108 extern emu_edge_hfix_func ff_emu_edge_hfix20_mmx;
109 extern emu_edge_hfix_func ff_emu_edge_hfix22_mmx;
110 #if ARCH_X86_32
111 static emu_edge_hfix_func * const hfixtbl_mmx[11] = {
112  ff_emu_edge_hfix2_mmx, ff_emu_edge_hfix4_mmx, ff_emu_edge_hfix6_mmx,
113  ff_emu_edge_hfix8_mmx, ff_emu_edge_hfix10_mmx, ff_emu_edge_hfix12_mmx,
114  ff_emu_edge_hfix14_mmx, ff_emu_edge_hfix16_mmx, ff_emu_edge_hfix18_mmx,
115  ff_emu_edge_hfix20_mmx, ff_emu_edge_hfix22_mmx
116 };
117 #endif
118 extern emu_edge_hvar_func ff_emu_edge_hvar_mmx;
119 extern emu_edge_hfix_func ff_emu_edge_hfix16_sse2;
120 extern emu_edge_hfix_func ff_emu_edge_hfix18_sse2;
121 extern emu_edge_hfix_func ff_emu_edge_hfix20_sse2;
122 extern emu_edge_hfix_func ff_emu_edge_hfix22_sse2;
123 static emu_edge_hfix_func * const hfixtbl_sse2[11] = {
124  ff_emu_edge_hfix2_mmx, ff_emu_edge_hfix4_mmx, ff_emu_edge_hfix6_mmx,
125  ff_emu_edge_hfix8_mmx, ff_emu_edge_hfix10_mmx, ff_emu_edge_hfix12_mmx,
126  ff_emu_edge_hfix14_mmx, ff_emu_edge_hfix16_sse2, ff_emu_edge_hfix18_sse2,
127  ff_emu_edge_hfix20_sse2, ff_emu_edge_hfix22_sse2
128 };
129 extern emu_edge_hvar_func ff_emu_edge_hvar_sse2;
130 #if HAVE_AVX2_EXTERNAL
131 extern emu_edge_hfix_func ff_emu_edge_hfix8_avx2;
132 extern emu_edge_hfix_func ff_emu_edge_hfix10_avx2;
133 extern emu_edge_hfix_func ff_emu_edge_hfix12_avx2;
134 extern emu_edge_hfix_func ff_emu_edge_hfix14_avx2;
135 extern emu_edge_hfix_func ff_emu_edge_hfix16_avx2;
136 extern emu_edge_hfix_func ff_emu_edge_hfix18_avx2;
137 extern emu_edge_hfix_func ff_emu_edge_hfix20_avx2;
138 extern emu_edge_hfix_func ff_emu_edge_hfix22_avx2;
139 static emu_edge_hfix_func * const hfixtbl_avx2[11] = {
140  ff_emu_edge_hfix2_mmx, ff_emu_edge_hfix4_mmx, ff_emu_edge_hfix6_mmx,
141  ff_emu_edge_hfix8_avx2, ff_emu_edge_hfix10_avx2, ff_emu_edge_hfix12_avx2,
142  ff_emu_edge_hfix14_avx2, ff_emu_edge_hfix16_avx2, ff_emu_edge_hfix18_avx2,
143  ff_emu_edge_hfix20_avx2, ff_emu_edge_hfix22_avx2
144 };
145 extern emu_edge_hvar_func ff_emu_edge_hvar_avx2;
146 #endif
147 
148 static av_always_inline void emulated_edge_mc(uint8_t *dst, const uint8_t *src,
149  ptrdiff_t dst_stride,
150  ptrdiff_t src_stride,
151  x86_reg block_w, x86_reg block_h,
152  x86_reg src_x, x86_reg src_y,
153  x86_reg w, x86_reg h,
154  emu_edge_vfix_func * const *vfix_tbl,
155  emu_edge_vvar_func *v_extend_var,
156  emu_edge_hfix_func * const *hfix_tbl,
157  emu_edge_hvar_func *h_extend_var)
158 {
159  x86_reg start_y, start_x, end_y, end_x, src_y_add = 0, p;
160 
161  if (!w || !h)
162  return;
163 
164  av_assert2(block_w <= FFABS(dst_stride));
165 
166  if (src_y >= h) {
167  src -= src_y*src_stride;
168  src_y_add = h - 1;
169  src_y = h - 1;
170  } else if (src_y <= -block_h) {
171  src -= src_y*src_stride;
172  src_y_add = 1 - block_h;
173  src_y = 1 - block_h;
174  }
175  if (src_x >= w) {
176  src += w - 1 - src_x;
177  src_x = w - 1;
178  } else if (src_x <= -block_w) {
179  src += 1 - block_w - src_x;
180  src_x = 1 - block_w;
181  }
182 
183  start_y = FFMAX(0, -src_y);
184  start_x = FFMAX(0, -src_x);
185  end_y = FFMIN(block_h, h-src_y);
186  end_x = FFMIN(block_w, w-src_x);
187  av_assert2(start_x < end_x && block_w > 0);
188  av_assert2(start_y < end_y && block_h > 0);
189 
190  // fill in the to-be-copied part plus all above/below
191  src += (src_y_add + start_y) * src_stride + start_x;
192  w = end_x - start_x;
193  if (w <= 22) {
194  vfix_tbl[w - 1](dst + start_x, dst_stride, src, src_stride,
195  start_y, end_y, block_h);
196  } else {
197  v_extend_var(dst + start_x, dst_stride, src, src_stride,
198  start_y, end_y, block_h, w);
199  }
200 
201  // fill left
202  if (start_x) {
203  if (start_x <= 22) {
204  hfix_tbl[(start_x - 1) >> 1](dst, dst_stride, start_x, block_h);
205  } else {
206  h_extend_var(dst, dst_stride,
207  start_x, (start_x + 1) >> 1, block_h);
208  }
209  }
210 
211  // fill right
212  p = block_w - end_x;
213  if (p) {
214  if (p <= 22) {
215  hfix_tbl[(p - 1) >> 1](dst + end_x - (p & 1), dst_stride,
216  -!(p & 1), block_h);
217  } else {
218  h_extend_var(dst + end_x - (p & 1), dst_stride,
219  -!(p & 1), (p + 1) >> 1, block_h);
220  }
221  }
222 }
223 
224 #if ARCH_X86_32
225 static av_noinline void emulated_edge_mc_mmx(uint8_t *buf, const uint8_t *src,
226  ptrdiff_t buf_stride,
227  ptrdiff_t src_stride,
228  int block_w, int block_h,
229  int src_x, int src_y, int w, int h)
230 {
231  emulated_edge_mc(buf, src, buf_stride, src_stride, block_w, block_h,
232  src_x, src_y, w, h, vfixtbl_mmx, &ff_emu_edge_vvar_mmx,
233  hfixtbl_mmx, &ff_emu_edge_hvar_mmx);
234 }
235 
236 static av_noinline void emulated_edge_mc_sse(uint8_t *buf, const uint8_t *src,
237  ptrdiff_t buf_stride,
238  ptrdiff_t src_stride,
239  int block_w, int block_h,
240  int src_x, int src_y, int w, int h)
241 {
242  emulated_edge_mc(buf, src, buf_stride, src_stride, block_w, block_h,
243  src_x, src_y, w, h, vfixtbl_sse, &ff_emu_edge_vvar_sse,
244  hfixtbl_mmx, &ff_emu_edge_hvar_mmx);
245 }
246 #endif
247 
248 static av_noinline void emulated_edge_mc_sse2(uint8_t *buf, const uint8_t *src,
249  ptrdiff_t buf_stride,
250  ptrdiff_t src_stride,
251  int block_w, int block_h,
252  int src_x, int src_y, int w,
253  int h)
254 {
255  emulated_edge_mc(buf, src, buf_stride, src_stride, block_w, block_h,
256  src_x, src_y, w, h, vfixtbl_sse, &ff_emu_edge_vvar_sse,
257  hfixtbl_sse2, &ff_emu_edge_hvar_sse2);
258 }
259 
260 #if HAVE_AVX2_EXTERNAL
261 static av_noinline void emulated_edge_mc_avx2(uint8_t *buf, const uint8_t *src,
262  ptrdiff_t buf_stride,
263  ptrdiff_t src_stride,
264  int block_w, int block_h,
265  int src_x, int src_y, int w,
266  int h)
267 {
268  emulated_edge_mc(buf, src, buf_stride, src_stride, block_w, block_h,
269  src_x, src_y, w, h, vfixtbl_sse, &ff_emu_edge_vvar_sse,
270  hfixtbl_avx2, &ff_emu_edge_hvar_avx2);
271 }
272 #endif /* HAVE_AVX2_EXTERNAL */
273 #endif /* HAVE_X86ASM */
274 
275 void ff_prefetch_mmxext(uint8_t *buf, ptrdiff_t stride, int h);
276 void ff_prefetch_3dnow(uint8_t *buf, ptrdiff_t stride, int h);
277 
279 {
280 #if HAVE_X86ASM
281  int cpu_flags = av_get_cpu_flags();
282 
283 #if ARCH_X86_32
284  if (EXTERNAL_MMX(cpu_flags) && bpc <= 8) {
285  ctx->emulated_edge_mc = emulated_edge_mc_mmx;
286  }
288  ctx->prefetch = ff_prefetch_3dnow;
289  }
290 #endif /* ARCH_X86_32 */
291  if (EXTERNAL_MMXEXT(cpu_flags)) {
292  ctx->prefetch = ff_prefetch_mmxext;
293  }
294 #if ARCH_X86_32
295  if (EXTERNAL_SSE(cpu_flags) && bpc <= 8) {
296  ctx->emulated_edge_mc = emulated_edge_mc_sse;
297  }
298 #endif /* ARCH_X86_32 */
299  if (EXTERNAL_SSE2(cpu_flags) && bpc <= 8) {
300  ctx->emulated_edge_mc = emulated_edge_mc_sse2;
301  }
302 #if HAVE_AVX2_EXTERNAL
303  if (EXTERNAL_AVX2(cpu_flags) && bpc <= 8) {
304  ctx->emulated_edge_mc = emulated_edge_mc_avx2;
305  }
306 #endif
307 #endif /* HAVE_X86ASM */
308 }
stride
int stride
Definition: mace.c:144
cpu.h
w
uint8_t w
Definition: llviddspenc.c:38
EXTERNAL_AMD3DNOW
#define EXTERNAL_AMD3DNOW(flags)
Definition: cpu.h:54
FFMAX
#define FFMAX(a, b)
Definition: macros.h:47
av_get_cpu_flags
int av_get_cpu_flags(void)
Return the flags which specify extensions supported by the CPU.
Definition: cpu.c:96
cpu_flags
static atomic_int cpu_flags
Definition: cpu.c:50
av_noinline
#define av_noinline
Definition: attributes.h:72
EXTERNAL_AVX2
#define EXTERNAL_AVX2(flags)
Definition: cpu.h:78
avassert.h
av_cold
#define av_cold
Definition: attributes.h:90
EXTERNAL_SSE
#define EXTERNAL_SSE(flags)
Definition: cpu.h:58
ctx
AVFormatContext * ctx
Definition: movenc.c:48
ff_videodsp_init_x86
av_cold void ff_videodsp_init_x86(VideoDSPContext *ctx, int bpc)
Definition: videodsp_init.c:278
FFABS
#define FFABS(a)
Absolute value, Note, INT_MIN / INT64_MIN result in undefined behavior as they are not representable ...
Definition: common.h:65
ff_prefetch_3dnow
void ff_prefetch_3dnow(uint8_t *buf, ptrdiff_t stride, int h)
src
#define src
Definition: vp8dsp.c:255
cpu.h
asm.h
attributes.h
EXTERNAL_SSE2
#define EXTERNAL_SSE2(flags)
Definition: cpu.h:59
av_assert2
#define av_assert2(cond)
assert() equivalent, that does lie in speed critical code.
Definition: avassert.h:64
common.h
av_always_inline
#define av_always_inline
Definition: attributes.h:49
FFMIN
#define FFMIN(a, b)
Definition: macros.h:49
VideoDSPContext
Definition: videodsp.h:41
videodsp.h
x86_reg
int x86_reg
Definition: asm.h:72
h
h
Definition: vp9dsp_template.c:2038
EXTERNAL_MMX
#define EXTERNAL_MMX(flags)
Definition: cpu.h:56
ff_prefetch_mmxext
void ff_prefetch_mmxext(uint8_t *buf, ptrdiff_t stride, int h)
EXTERNAL_MMXEXT
#define EXTERNAL_MMXEXT(flags)
Definition: cpu.h:57