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 typedef void emu_edge_vfix_func(uint8_t *dst, x86_reg dst_stride,
32  const uint8_t *src, x86_reg src_stride,
33  x86_reg start_y, x86_reg end_y, x86_reg bh);
34 typedef void emu_edge_vvar_func(uint8_t *dst, x86_reg dst_stride,
35  const uint8_t *src, x86_reg src_stride,
36  x86_reg start_y, x86_reg end_y, x86_reg bh,
37  x86_reg w);
38 
61 static emu_edge_vfix_func * const vfixtbl_sse2[22] = {
70 };
72 
73 typedef void emu_edge_hfix_func(uint8_t *dst, x86_reg dst_stride,
74  x86_reg start_x, x86_reg bh);
75 typedef void emu_edge_hvar_func(uint8_t *dst, x86_reg dst_stride,
76  x86_reg start_x, x86_reg n_words, x86_reg bh);
77 
89 static emu_edge_hfix_func * const hfixtbl_sse2[11] = {
94 };
96 #if HAVE_AVX2_EXTERNAL
97 extern emu_edge_hfix_func ff_emu_edge_hfix8_avx2;
98 extern emu_edge_hfix_func ff_emu_edge_hfix10_avx2;
99 extern emu_edge_hfix_func ff_emu_edge_hfix12_avx2;
100 extern emu_edge_hfix_func ff_emu_edge_hfix14_avx2;
101 extern emu_edge_hfix_func ff_emu_edge_hfix16_avx2;
102 extern emu_edge_hfix_func ff_emu_edge_hfix18_avx2;
103 extern emu_edge_hfix_func ff_emu_edge_hfix20_avx2;
104 extern emu_edge_hfix_func ff_emu_edge_hfix22_avx2;
105 static emu_edge_hfix_func * const hfixtbl_avx2[11] = {
107  ff_emu_edge_hfix8_avx2, ff_emu_edge_hfix10_avx2, ff_emu_edge_hfix12_avx2,
108  ff_emu_edge_hfix14_avx2, ff_emu_edge_hfix16_avx2, ff_emu_edge_hfix18_avx2,
109  ff_emu_edge_hfix20_avx2, ff_emu_edge_hfix22_avx2
110 };
111 extern emu_edge_hvar_func ff_emu_edge_hvar_avx2;
112 #endif
113 
114 static av_always_inline void emulated_edge_mc(uint8_t *dst, const uint8_t *src,
115  ptrdiff_t dst_stride,
116  ptrdiff_t src_stride,
117  x86_reg block_w, x86_reg block_h,
118  x86_reg src_x, x86_reg src_y,
119  x86_reg w, x86_reg h,
120  emu_edge_vfix_func * const *vfix_tbl,
121  emu_edge_vvar_func *v_extend_var,
122  emu_edge_hfix_func * const *hfix_tbl,
123  emu_edge_hvar_func *h_extend_var)
124 {
125  x86_reg start_y, start_x, end_y, end_x, src_y_add = 0, p;
126 
127  if (!w || !h)
128  return;
129 
130  av_assert2(block_w <= FFABS(dst_stride));
131 
132  if (src_y >= h) {
133  src -= src_y*src_stride;
134  src_y_add = h - 1;
135  src_y = h - 1;
136  } else if (src_y <= -block_h) {
137  src -= src_y*src_stride;
138  src_y_add = 1 - block_h;
139  src_y = 1 - block_h;
140  }
141  if (src_x >= w) {
142  src += w - 1 - src_x;
143  src_x = w - 1;
144  } else if (src_x <= -block_w) {
145  src += 1 - block_w - src_x;
146  src_x = 1 - block_w;
147  }
148 
149  start_y = FFMAX(0, -src_y);
150  start_x = FFMAX(0, -src_x);
151  end_y = FFMIN(block_h, h-src_y);
152  end_x = FFMIN(block_w, w-src_x);
153  av_assert2(start_x < end_x && block_w > 0);
154  av_assert2(start_y < end_y && block_h > 0);
155 
156  // fill in the to-be-copied part plus all above/below
157  src += (src_y_add + start_y) * src_stride + start_x;
158  w = end_x - start_x;
159  if (w <= 22) {
160  vfix_tbl[w - 1](dst + start_x, dst_stride, src, src_stride,
161  start_y, end_y, block_h);
162  } else {
163  v_extend_var(dst + start_x, dst_stride, src, src_stride,
164  start_y, end_y, block_h, w);
165  }
166 
167  // fill left
168  if (start_x) {
169  if (start_x <= 22) {
170  hfix_tbl[(start_x - 1) >> 1](dst, dst_stride, start_x, block_h);
171  } else {
172  h_extend_var(dst, dst_stride,
173  start_x, (start_x + 1) >> 1, block_h);
174  }
175  }
176 
177  // fill right
178  p = block_w - end_x;
179  if (p) {
180  if (p <= 22) {
181  hfix_tbl[(p - 1) >> 1](dst + end_x - (p & 1), dst_stride,
182  -!(p & 1), block_h);
183  } else {
184  h_extend_var(dst + end_x - (p & 1), dst_stride,
185  -!(p & 1), (p + 1) >> 1, block_h);
186  }
187  }
188 }
189 
190 static av_noinline void emulated_edge_mc_sse2(uint8_t *buf, const uint8_t *src,
191  ptrdiff_t buf_stride,
192  ptrdiff_t src_stride,
193  int block_w, int block_h,
194  int src_x, int src_y, int w,
195  int h)
196 {
197  emulated_edge_mc(buf, src, buf_stride, src_stride, block_w, block_h,
198  src_x, src_y, w, h, vfixtbl_sse2, &ff_emu_edge_vvar_sse,
200 }
201 
202 #if HAVE_AVX2_EXTERNAL
203 static av_noinline void emulated_edge_mc_avx2(uint8_t *buf, const uint8_t *src,
204  ptrdiff_t buf_stride,
205  ptrdiff_t src_stride,
206  int block_w, int block_h,
207  int src_x, int src_y, int w,
208  int h)
209 {
210  emulated_edge_mc(buf, src, buf_stride, src_stride, block_w, block_h,
211  src_x, src_y, w, h, vfixtbl_sse2, &ff_emu_edge_vvar_sse,
212  hfixtbl_avx2, &ff_emu_edge_hvar_avx2);
213 }
214 #endif /* HAVE_AVX2_EXTERNAL */
215 
216 void ff_prefetch_mmxext(const uint8_t *buf, ptrdiff_t stride, int h);
217 
219 {
220  int cpu_flags = av_get_cpu_flags();
221 
222  if (EXTERNAL_MMXEXT(cpu_flags)) {
223  ctx->prefetch = ff_prefetch_mmxext;
224  }
225  if (EXTERNAL_SSE2(cpu_flags) && bpc <= 8) {
226  ctx->emulated_edge_mc = emulated_edge_mc_sse2;
227  }
228 #if HAVE_AVX2_EXTERNAL
229  if (EXTERNAL_AVX2(cpu_flags) && bpc <= 8) {
230  ctx->emulated_edge_mc = emulated_edge_mc_avx2;
231  }
232 #endif
233 }
cpu.h
ff_emu_edge_vvar_sse
emu_edge_vvar_func ff_emu_edge_vvar_sse
ff_emu_edge_hfix14_sse2
emu_edge_hfix_func ff_emu_edge_hfix14_sse2
ff_emu_edge_hfix10_sse2
emu_edge_hfix_func ff_emu_edge_hfix10_sse2
x86_reg
int x86_reg
Definition: asm.h:71
w
uint8_t w
Definition: llviddspenc.c:38
emu_edge_vvar_func
void emu_edge_vvar_func(uint8_t *dst, x86_reg dst_stride, const uint8_t *src, x86_reg src_stride, x86_reg start_y, x86_reg end_y, x86_reg bh, x86_reg w)
Definition: videodsp_init.c:34
FFMAX
#define FFMAX(a, b)
Definition: macros.h:47
ff_emu_edge_vfix4_sse2
emu_edge_vfix_func ff_emu_edge_vfix4_sse2
av_get_cpu_flags
int av_get_cpu_flags(void)
Return the flags which specify extensions supported by the CPU.
Definition: cpu.c:109
ff_emu_edge_vfix2_sse2
emu_edge_vfix_func ff_emu_edge_vfix2_sse2
cpu_flags
static atomic_int cpu_flags
Definition: cpu.c:56
emu_edge_hvar_func
void emu_edge_hvar_func(uint8_t *dst, x86_reg dst_stride, x86_reg start_x, x86_reg n_words, x86_reg bh)
Definition: videodsp_init.c:75
ff_emu_edge_vfix19_sse2
emu_edge_vfix_func ff_emu_edge_vfix19_sse2
ff_emu_edge_vfix15_sse2
emu_edge_vfix_func ff_emu_edge_vfix15_sse2
emulated_edge_mc_sse2
static av_noinline void emulated_edge_mc_sse2(uint8_t *buf, const uint8_t *src, ptrdiff_t buf_stride, ptrdiff_t src_stride, int block_w, int block_h, int src_x, int src_y, int w, int h)
Definition: videodsp_init.c:190
av_noinline
#define av_noinline
Definition: attributes.h:88
ff_emu_edge_hfix22_sse2
emu_edge_hfix_func ff_emu_edge_hfix22_sse2
EXTERNAL_AVX2
#define EXTERNAL_AVX2(flags)
Definition: cpu.h:71
avassert.h
av_cold
#define av_cold
Definition: attributes.h:106
ff_emu_edge_hfix6_sse2
emu_edge_hfix_func ff_emu_edge_hfix6_sse2
emu_edge_vfix_func
void emu_edge_vfix_func(uint8_t *dst, x86_reg dst_stride, const uint8_t *src, x86_reg src_stride, x86_reg start_y, x86_reg end_y, x86_reg bh)
Definition: videodsp_init.c:31
ctx
AVFormatContext * ctx
Definition: movenc.c:49
ff_videodsp_init_x86
av_cold void ff_videodsp_init_x86(VideoDSPContext *ctx, int bpc)
Definition: videodsp_init.c:218
FFABS
#define FFABS(a)
Absolute value, Note, INT_MIN / INT64_MIN result in undefined behavior as they are not representable ...
Definition: common.h:74
ff_emu_edge_hvar_sse2
emu_edge_hvar_func ff_emu_edge_hvar_sse2
ff_prefetch_mmxext
void ff_prefetch_mmxext(const uint8_t *buf, ptrdiff_t stride, int h)
ff_emu_edge_vfix12_sse2
emu_edge_vfix_func ff_emu_edge_vfix12_sse2
ff_emu_edge_vfix5_sse2
emu_edge_vfix_func ff_emu_edge_vfix5_sse2
ff_emu_edge_vfix3_sse2
emu_edge_vfix_func ff_emu_edge_vfix3_sse2
asm.h
emu_edge_hfix_func
void emu_edge_hfix_func(uint8_t *dst, x86_reg dst_stride, x86_reg start_x, x86_reg bh)
Definition: videodsp_init.c:73
emulated_edge_mc
static av_always_inline void emulated_edge_mc(uint8_t *dst, const uint8_t *src, ptrdiff_t dst_stride, ptrdiff_t src_stride, x86_reg block_w, x86_reg block_h, x86_reg src_x, x86_reg src_y, x86_reg w, x86_reg h, emu_edge_vfix_func *const *vfix_tbl, emu_edge_vvar_func *v_extend_var, emu_edge_hfix_func *const *hfix_tbl, emu_edge_hvar_func *h_extend_var)
Definition: videodsp_init.c:114
ff_emu_edge_hfix20_sse2
emu_edge_hfix_func ff_emu_edge_hfix20_sse2
ff_emu_edge_vfix11_sse2
emu_edge_vfix_func ff_emu_edge_vfix11_sse2
ff_emu_edge_vfix17_sse2
emu_edge_vfix_func ff_emu_edge_vfix17_sse2
ff_emu_edge_vfix20_sse2
emu_edge_vfix_func ff_emu_edge_vfix20_sse2
dst
uint8_t ptrdiff_t const uint8_t ptrdiff_t int intptr_t intptr_t int int16_t * dst
Definition: dsp.h:87
cpu.h
ff_emu_edge_vfix6_sse2
emu_edge_vfix_func ff_emu_edge_vfix6_sse2
attributes.h
EXTERNAL_SSE2
#define EXTERNAL_SSE2(flags)
Definition: cpu.h:52
ff_emu_edge_vfix1_sse2
emu_edge_vfix_func ff_emu_edge_vfix1_sse2
ff_emu_edge_vfix7_sse2
emu_edge_vfix_func ff_emu_edge_vfix7_sse2
av_assert2
#define av_assert2(cond)
assert() equivalent, that does lie in speed critical code.
Definition: avassert.h:68
common.h
av_always_inline
#define av_always_inline
Definition: attributes.h:63
FFMIN
#define FFMIN(a, b)
Definition: macros.h:49
ff_emu_edge_hfix18_sse2
emu_edge_hfix_func ff_emu_edge_hfix18_sse2
hfixtbl_sse2
static emu_edge_hfix_func *const hfixtbl_sse2[11]
Definition: videodsp_init.c:89
ff_emu_edge_hfix16_sse2
emu_edge_hfix_func ff_emu_edge_hfix16_sse2
ff_emu_edge_vfix8_sse2
emu_edge_vfix_func ff_emu_edge_vfix8_sse2
ff_emu_edge_hfix4_sse2
emu_edge_hfix_func ff_emu_edge_hfix4_sse2
ff_emu_edge_vfix21_sse2
emu_edge_vfix_func ff_emu_edge_vfix21_sse2
ff_emu_edge_vfix10_sse2
emu_edge_vfix_func ff_emu_edge_vfix10_sse2
ff_emu_edge_vfix14_sse2
emu_edge_vfix_func ff_emu_edge_vfix14_sse2
ff_emu_edge_vfix13_sse2
emu_edge_vfix_func ff_emu_edge_vfix13_sse2
vfixtbl_sse2
static emu_edge_vfix_func *const vfixtbl_sse2[22]
Definition: videodsp_init.c:61
ff_emu_edge_vfix16_sse2
emu_edge_vfix_func ff_emu_edge_vfix16_sse2
Windows::Graphics::DirectX::Direct3D11::p
IDirect3DDxgiInterfaceAccess _COM_Outptr_ void ** p
Definition: vsrc_gfxcapture_winrt.hpp:53
VideoDSPContext
Definition: videodsp.h:40
ff_emu_edge_vfix18_sse2
emu_edge_vfix_func ff_emu_edge_vfix18_sse2
ff_emu_edge_hfix8_sse2
emu_edge_hfix_func ff_emu_edge_hfix8_sse2
videodsp.h
ff_emu_edge_vfix22_sse2
emu_edge_vfix_func ff_emu_edge_vfix22_sse2
h
h
Definition: vp9dsp_template.c:2070
stride
#define stride
Definition: h264pred_template.c:536
ff_emu_edge_vfix9_sse2
emu_edge_vfix_func ff_emu_edge_vfix9_sse2
EXTERNAL_MMXEXT
#define EXTERNAL_MMXEXT(flags)
Definition: cpu.h:50
ff_emu_edge_hfix2_sse2
emu_edge_hfix_func ff_emu_edge_hfix2_sse2
src
#define src
Definition: vp8dsp.c:248
ff_emu_edge_hfix12_sse2
emu_edge_hfix_func ff_emu_edge_hfix12_sse2