FFmpeg
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 "libavutil/attributes.h"
26 #include "libavutil/cpu.h"
27 #include "libavutil/x86/cpu.h"
28 #include "libavcodec/avcodec.h"
29 #include "libavcodec/hpeldsp.h"
30 #include "libavcodec/pixels.h"
31 #include "fpel.h"
32 #include "hpeldsp.h"
33 
34 void ff_put_pixels8_x2_mmxext(uint8_t *block, const uint8_t *pixels,
35  ptrdiff_t line_size, int h);
36 void ff_put_pixels16_x2_mmxext(uint8_t *block, const uint8_t *pixels,
37  ptrdiff_t line_size, int h);
38 void ff_put_pixels16_x2_sse2(uint8_t *block, const uint8_t *pixels,
39  ptrdiff_t line_size, int h);
40 void ff_avg_pixels16_x2_sse2(uint8_t *block, const uint8_t *pixels,
41  ptrdiff_t line_size, int h);
42 void ff_put_pixels16_y2_sse2(uint8_t *block, const uint8_t *pixels,
43  ptrdiff_t line_size, int h);
44 void ff_avg_pixels16_y2_sse2(uint8_t *block, const uint8_t *pixels,
45  ptrdiff_t line_size, int h);
46 void ff_put_no_rnd_pixels8_x2_mmxext(uint8_t *block, const uint8_t *pixels,
47  ptrdiff_t line_size, int h);
48 void ff_put_no_rnd_pixels8_x2_exact_mmxext(uint8_t *block,
49  const uint8_t *pixels,
50  ptrdiff_t line_size, int h);
51 void ff_put_pixels8_y2_mmxext(uint8_t *block, const uint8_t *pixels,
52  ptrdiff_t line_size, int h);
53 void ff_put_no_rnd_pixels8_y2_mmxext(uint8_t *block, const uint8_t *pixels,
54  ptrdiff_t line_size, int h);
55 void ff_put_no_rnd_pixels8_y2_exact_mmxext(uint8_t *block,
56  const uint8_t *pixels,
57  ptrdiff_t line_size, int h);
58 void ff_avg_pixels8_x2_mmxext(uint8_t *block, const uint8_t *pixels,
59  ptrdiff_t line_size, int h);
60 void ff_avg_pixels8_y2_mmxext(uint8_t *block, const uint8_t *pixels,
61  ptrdiff_t line_size, int h);
62 void ff_avg_approx_pixels8_xy2_mmxext(uint8_t *block, const uint8_t *pixels,
63  ptrdiff_t line_size, int h);
64 
65 #define put_pixels8_mmx ff_put_pixels8_mmx
66 #define put_pixels16_mmx ff_put_pixels16_mmx
67 #define put_pixels8_xy2_mmx ff_put_pixels8_xy2_mmx
68 #define put_no_rnd_pixels8_mmx ff_put_pixels8_mmx
69 #define put_no_rnd_pixels16_mmx ff_put_pixels16_mmx
70 
71 #if HAVE_INLINE_ASM
72 
73 /***********************************/
74 /* MMX no rounding */
75 #define DEF(x, y) x ## _no_rnd_ ## y ## _mmx
76 #define SET_RND MOVQ_WONE
77 #define PAVGBP(a, b, c, d, e, f) PAVGBP_MMX_NO_RND(a, b, c, d, e, f)
78 #define PAVGB(a, b, c, e) PAVGB_MMX_NO_RND(a, b, c, e)
79 #define STATIC static
80 
81 #include "rnd_template.c"
82 #include "hpeldsp_rnd_template.c"
83 
84 #undef DEF
85 #undef SET_RND
86 #undef PAVGBP
87 #undef PAVGB
88 #undef STATIC
89 
90 #if HAVE_MMX
91 CALL_2X_PIXELS(avg_no_rnd_pixels16_y2_mmx, avg_no_rnd_pixels8_y2_mmx, 8)
92 CALL_2X_PIXELS(put_no_rnd_pixels16_y2_mmx, put_no_rnd_pixels8_y2_mmx, 8)
93 
94 CALL_2X_PIXELS(avg_no_rnd_pixels16_xy2_mmx, avg_no_rnd_pixels8_xy2_mmx, 8)
95 CALL_2X_PIXELS(put_no_rnd_pixels16_xy2_mmx, put_no_rnd_pixels8_xy2_mmx, 8)
96 #endif
97 
98 /***********************************/
99 /* MMX rounding */
100 
101 #define SET_RND MOVQ_WTWO
102 #define DEF(x, y) ff_ ## x ## _ ## y ## _mmx
103 #define STATIC
104 #define NO_AVG
105 
106 #include "rnd_template.c"
107 
108 #undef NO_AVG
109 #undef DEF
110 #undef SET_RND
111 
112 #if HAVE_MMX
113 CALL_2X_PIXELS(put_pixels16_xy2_mmx, ff_put_pixels8_xy2_mmx, 8)
114 #endif
115 
116 #endif /* HAVE_INLINE_ASM */
117 
118 
119 #if HAVE_X86ASM
120 
121 #define HPELDSP_AVG_PIXELS16(CPUEXT) \
122  CALL_2X_PIXELS(put_no_rnd_pixels16_x2 ## CPUEXT, ff_put_no_rnd_pixels8_x2 ## CPUEXT, 8) \
123  CALL_2X_PIXELS(put_pixels16_y2 ## CPUEXT, ff_put_pixels8_y2 ## CPUEXT, 8) \
124  CALL_2X_PIXELS(put_no_rnd_pixels16_y2 ## CPUEXT, ff_put_no_rnd_pixels8_y2 ## CPUEXT, 8) \
125  CALL_2X_PIXELS(avg_pixels16 ## CPUEXT, ff_avg_pixels8 ## CPUEXT, 8) \
126  CALL_2X_PIXELS(avg_pixels16_x2 ## CPUEXT, ff_avg_pixels8_x2 ## CPUEXT, 8) \
127  CALL_2X_PIXELS(avg_pixels16_y2 ## CPUEXT, ff_avg_pixels8_y2 ## CPUEXT, 8) \
128  CALL_2X_PIXELS(avg_pixels16_xy2 ## CPUEXT, ff_avg_pixels8_xy2 ## CPUEXT, 8) \
129  CALL_2X_PIXELS(avg_approx_pixels16_xy2## CPUEXT, ff_avg_approx_pixels8_xy2## CPUEXT, 8)
130 
131 HPELDSP_AVG_PIXELS16(_mmxext)
132 
133 #endif /* HAVE_X86ASM */
134 
135 #define SET_HPEL_FUNCS_EXT(PFX, IDX, SIZE, CPU) \
136  if (HAVE_MMX_EXTERNAL) \
137  c->PFX ## _pixels_tab IDX [0] = PFX ## _pixels ## SIZE ## _ ## CPU
138 
139 #define SET_HPEL_FUNCS03(PFX, IDX, SIZE, CPU) \
140  do { \
141  SET_HPEL_FUNCS_EXT(PFX, IDX, SIZE, CPU); \
142  c->PFX ## _pixels_tab IDX [3] = PFX ## _pixels ## SIZE ## _xy2_ ## CPU; \
143  } while (0)
144 #define SET_HPEL_FUNCS12(PFX, IDX, SIZE, CPU) \
145  do { \
146  c->PFX ## _pixels_tab IDX [1] = PFX ## _pixels ## SIZE ## _x2_ ## CPU; \
147  c->PFX ## _pixels_tab IDX [2] = PFX ## _pixels ## SIZE ## _y2_ ## CPU; \
148  } while (0)
149 #define SET_HPEL_FUNCS(PFX, IDX, SIZE, CPU) \
150  do { \
151  SET_HPEL_FUNCS03(PFX, IDX, SIZE, CPU); \
152  SET_HPEL_FUNCS12(PFX, IDX, SIZE, CPU); \
153  } while (0)
154 
155 static void hpeldsp_init_mmx(HpelDSPContext *c, int flags)
156 {
157 #if HAVE_MMX_INLINE
158  SET_HPEL_FUNCS03(put, [0], 16, mmx);
159  SET_HPEL_FUNCS(put_no_rnd, [0], 16, mmx);
160  SET_HPEL_FUNCS12(avg_no_rnd, , 16, mmx);
161  c->avg_no_rnd_pixels_tab[3] = avg_no_rnd_pixels16_xy2_mmx;
162  SET_HPEL_FUNCS03(put, [1], 8, mmx);
163  SET_HPEL_FUNCS(put_no_rnd, [1], 8, mmx);
164 #endif
165 }
166 
167 static void hpeldsp_init_mmxext(HpelDSPContext *c, int flags)
168 {
169 #if HAVE_MMXEXT_EXTERNAL
170  c->put_pixels_tab[0][1] = ff_put_pixels16_x2_mmxext;
171  c->put_pixels_tab[0][2] = put_pixels16_y2_mmxext;
172 
173  c->avg_pixels_tab[0][0] = avg_pixels16_mmxext;
174  c->avg_pixels_tab[0][1] = avg_pixels16_x2_mmxext;
175  c->avg_pixels_tab[0][2] = avg_pixels16_y2_mmxext;
176  c->avg_pixels_tab[0][3] = avg_pixels16_xy2_mmxext;
177 
178  c->put_pixels_tab[1][1] = ff_put_pixels8_x2_mmxext;
179  c->put_pixels_tab[1][2] = ff_put_pixels8_y2_mmxext;
180 
181  c->avg_pixels_tab[1][0] = ff_avg_pixels8_mmxext;
182  c->avg_pixels_tab[1][1] = ff_avg_pixels8_x2_mmxext;
183  c->avg_pixels_tab[1][2] = ff_avg_pixels8_y2_mmxext;
184  c->avg_pixels_tab[1][3] = ff_avg_pixels8_xy2_mmxext;
185 
186  c->put_no_rnd_pixels_tab[1][1] = ff_put_no_rnd_pixels8_x2_exact_mmxext;
187  c->put_no_rnd_pixels_tab[1][2] = ff_put_no_rnd_pixels8_y2_exact_mmxext;
188 
189  c->avg_no_rnd_pixels_tab[0] = ff_avg_pixels16_mmxext;
190 
191  if (!(flags & AV_CODEC_FLAG_BITEXACT)) {
192  c->put_no_rnd_pixels_tab[0][1] = put_no_rnd_pixels16_x2_mmxext;
193  c->put_no_rnd_pixels_tab[0][2] = put_no_rnd_pixels16_y2_mmxext;
194  c->put_no_rnd_pixels_tab[1][1] = ff_put_no_rnd_pixels8_x2_mmxext;
195  c->put_no_rnd_pixels_tab[1][2] = ff_put_no_rnd_pixels8_y2_mmxext;
196 
197  c->avg_pixels_tab[0][3] = avg_approx_pixels16_xy2_mmxext;
198  c->avg_pixels_tab[1][3] = ff_avg_approx_pixels8_xy2_mmxext;
199  }
200 #endif /* HAVE_MMXEXT_EXTERNAL */
201 }
202 
203 static void hpeldsp_init_sse2_fast(HpelDSPContext *c, int flags)
204 {
205 #if HAVE_SSE2_EXTERNAL
206  c->put_pixels_tab[0][0] = ff_put_pixels16_sse2;
207  c->put_no_rnd_pixels_tab[0][0] = ff_put_pixels16_sse2;
208  c->put_pixels_tab[0][1] = ff_put_pixels16_x2_sse2;
209  c->put_pixels_tab[0][2] = ff_put_pixels16_y2_sse2;
210  c->put_pixels_tab[0][3] = ff_put_pixels16_xy2_sse2;
211  c->avg_pixels_tab[0][0] = ff_avg_pixels16_sse2;
212  c->avg_pixels_tab[0][1] = ff_avg_pixels16_x2_sse2;
213  c->avg_pixels_tab[0][2] = ff_avg_pixels16_y2_sse2;
214  c->avg_pixels_tab[0][3] = ff_avg_pixels16_xy2_sse2;
215  c->avg_no_rnd_pixels_tab[0] = ff_avg_pixels16_sse2;
216 #endif /* HAVE_SSE2_EXTERNAL */
217 }
218 
219 static void hpeldsp_init_ssse3(HpelDSPContext *c, int flags)
220 {
221 #if HAVE_SSSE3_EXTERNAL
222  c->put_pixels_tab[0][3] = ff_put_pixels16_xy2_ssse3;
223  c->avg_pixels_tab[0][3] = ff_avg_pixels16_xy2_ssse3;
224  c->put_pixels_tab[1][3] = ff_put_pixels8_xy2_ssse3;
225  c->avg_pixels_tab[1][3] = ff_avg_pixels8_xy2_ssse3;
226 #endif
227 }
228 
229 av_cold void ff_hpeldsp_init_x86(HpelDSPContext *c, int flags)
230 {
231  int cpu_flags = av_get_cpu_flags();
232 
233  if (INLINE_MMX(cpu_flags))
234  hpeldsp_init_mmx(c, flags);
235 
236  if (EXTERNAL_MMXEXT(cpu_flags))
237  hpeldsp_init_mmxext(c, flags);
238 
239  if (EXTERNAL_SSE2_FAST(cpu_flags))
240  hpeldsp_init_sse2_fast(c, flags);
241 
242  if (EXTERNAL_SSSE3(cpu_flags))
243  hpeldsp_init_ssse3(c, flags);
244 }
INLINE_MMX
#define INLINE_MMX(flags)
Definition: cpu.h:87
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)
pixels.h
av_get_cpu_flags
int av_get_cpu_flags(void)
Return the flags which specify extensions supported by the CPU.
Definition: cpu.c:107
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:144
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:219
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:229
ff_avg_pixels8_y2_mmxext
void ff_avg_pixels8_y2_mmxext(uint8_t *block, const uint8_t *pixels, ptrdiff_t line_size, int h)
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:167
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:139
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)
hpeldsp_rnd_template.c
ff_avg_pixels16_xy2_sse2
void ff_avg_pixels16_xy2_sse2(uint8_t *block, const uint8_t *pixels, ptrdiff_t line_size, int h)
hpeldsp_init_sse2_fast
static void hpeldsp_init_sse2_fast(HpelDSPContext *c, int flags)
Definition: hpeldsp_init.c:203
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:155
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)
SET_HPEL_FUNCS
#define SET_HPEL_FUNCS(PFX, IDX, SIZE, CPU)
Definition: hpeldsp_init.c:149
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:342
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
flags
#define flags(name, subs,...)
Definition: cbs_av1.c:482
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
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 Sun Dec 1 2024 19:22:15 for FFmpeg by   doxygen 1.8.17