FFmpeg
 All Data Structures Namespaces Files Functions Variables Typedefs Enumerations Enumerator Macros Groups Pages
hpeldsp_init.c
Go to the documentation of this file.
1 /*
2  * MMX optimized DSP utils
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/cpu.h"
26 #include "libavutil/x86/asm.h"
27 #include "libavutil/x86/cpu.h"
28 #include "libavcodec/hpeldsp.h"
29 #include "dsputil_x86.h"
30 
31 void ff_put_pixels8_x2_mmxext(uint8_t *block, const uint8_t *pixels,
32  ptrdiff_t line_size, int h);
33 void ff_put_pixels8_x2_3dnow(uint8_t *block, const uint8_t *pixels,
34  ptrdiff_t line_size, int h);
35 void ff_put_pixels16_x2_mmxext(uint8_t *block, const uint8_t *pixels,
36  ptrdiff_t line_size, int h);
37 void ff_put_pixels16_x2_3dnow(uint8_t *block, const uint8_t *pixels,
38  ptrdiff_t line_size, int h);
39 void ff_put_no_rnd_pixels8_x2_mmxext(uint8_t *block, const uint8_t *pixels,
40  ptrdiff_t line_size, int h);
41 void ff_put_no_rnd_pixels8_x2_3dnow(uint8_t *block, const uint8_t *pixels,
42  ptrdiff_t line_size, int h);
43 void ff_put_no_rnd_pixels8_x2_exact_mmxext(uint8_t *block,
44  const uint8_t *pixels,
45  ptrdiff_t line_size, int h);
46 void ff_put_no_rnd_pixels8_x2_exact_3dnow(uint8_t *block,
47  const uint8_t *pixels,
48  ptrdiff_t line_size, int h);
49 void ff_put_pixels8_y2_mmxext(uint8_t *block, const uint8_t *pixels,
50  ptrdiff_t line_size, int h);
51 void ff_put_pixels8_y2_3dnow(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_3dnow(uint8_t *block, const uint8_t *pixels,
56  ptrdiff_t line_size, int h);
57 void ff_put_no_rnd_pixels8_y2_exact_mmxext(uint8_t *block,
58  const uint8_t *pixels,
59  ptrdiff_t line_size, int h);
60 void ff_put_no_rnd_pixels8_y2_exact_3dnow(uint8_t *block,
61  const uint8_t *pixels,
62  ptrdiff_t line_size, int h);
63 void ff_avg_pixels8_3dnow(uint8_t *block, const uint8_t *pixels,
64  ptrdiff_t line_size, int h);
65 void ff_avg_pixels8_x2_mmxext(uint8_t *block, const uint8_t *pixels,
66  ptrdiff_t line_size, int h);
67 void ff_avg_pixels8_x2_3dnow(uint8_t *block, const uint8_t *pixels,
68  ptrdiff_t line_size, int h);
69 void ff_avg_pixels8_y2_mmxext(uint8_t *block, const uint8_t *pixels,
70  ptrdiff_t line_size, int h);
71 void ff_avg_pixels8_y2_3dnow(uint8_t *block, const uint8_t *pixels,
72  ptrdiff_t line_size, int h);
73 void ff_avg_pixels8_xy2_mmxext(uint8_t *block, const uint8_t *pixels,
74  ptrdiff_t line_size, int h);
75 void ff_avg_pixels8_xy2_3dnow(uint8_t *block, const uint8_t *pixels,
76  ptrdiff_t line_size, int h);
77 
78 #define avg_pixels8_mmx ff_avg_pixels8_mmx
79 #define avg_pixels8_x2_mmx ff_avg_pixels8_x2_mmx
80 #define avg_pixels16_mmx ff_avg_pixels16_mmx
81 #define avg_pixels8_xy2_mmx ff_avg_pixels8_xy2_mmx
82 #define avg_pixels16_xy2_mmx ff_avg_pixels16_xy2_mmx
83 #define put_pixels8_mmx ff_put_pixels8_mmx
84 #define put_pixels16_mmx ff_put_pixels16_mmx
85 #define put_pixels8_xy2_mmx ff_put_pixels8_xy2_mmx
86 #define put_pixels16_xy2_mmx ff_put_pixels16_xy2_mmx
87 #define avg_no_rnd_pixels16_mmx ff_avg_pixels16_mmx
88 #define put_no_rnd_pixels8_mmx ff_put_pixels8_mmx
89 #define put_no_rnd_pixels16_mmx ff_put_pixels16_mmx
90 
91 #if HAVE_INLINE_ASM
92 
93 /***********************************/
94 /* MMX no rounding */
95 #define DEF(x, y) x ## _no_rnd_ ## y ## _mmx
96 #define SET_RND MOVQ_WONE
97 #define PAVGBP(a, b, c, d, e, f) PAVGBP_MMX_NO_RND(a, b, c, d, e, f)
98 #define PAVGB(a, b, c, e) PAVGB_MMX_NO_RND(a, b, c, e)
99 #define STATIC static
100 
101 #include "rnd_template.c"
102 #include "hpeldsp_rnd_template.c"
103 
104 #undef DEF
105 #undef SET_RND
106 #undef PAVGBP
107 #undef PAVGB
108 #undef STATIC
109 
110 PIXELS16(static, avg_no_rnd, , _y2, _mmx)
111 PIXELS16(static, put_no_rnd, , _y2, _mmx)
112 
113 PIXELS16(static, avg_no_rnd, , _xy2, _mmx)
114 PIXELS16(static, put_no_rnd, , _xy2, _mmx)
115 
116 /***********************************/
117 /* MMX rounding */
118 
119 #define DEF(x, y) x ## _ ## y ## _mmx
120 #define SET_RND MOVQ_WTWO
121 #define PAVGBP(a, b, c, d, e, f) PAVGBP_MMX(a, b, c, d, e, f)
122 #define PAVGB(a, b, c, e) PAVGB_MMX(a, b, c, e)
123 
124 #include "hpeldsp_rnd_template.c"
125 
126 #undef DEF
127 #undef SET_RND
128 #undef PAVGBP
129 #undef PAVGB
130 
131 PIXELS16(static, avg, , _y2, _mmx)
132 PIXELS16(static, put, , _y2, _mmx)
133 
134 #endif /* HAVE_INLINE_ASM */
135 
136 
137 #if HAVE_YASM
138 
139 #define HPELDSP_AVG_PIXELS16(CPUEXT) \
140  PIXELS16(static, put_no_rnd, ff_, _x2, CPUEXT) \
141  PIXELS16(static, put, ff_, _y2, CPUEXT) \
142  PIXELS16(static, put_no_rnd, ff_, _y2, CPUEXT) \
143  PIXELS16(static, avg, ff_, , CPUEXT) \
144  PIXELS16(static, avg, ff_, _x2, CPUEXT) \
145  PIXELS16(static, avg, ff_, _y2, CPUEXT) \
146  PIXELS16(static, avg, ff_, _xy2, CPUEXT)
147 
148 HPELDSP_AVG_PIXELS16(_3dnow)
149 HPELDSP_AVG_PIXELS16(_mmxext)
150 
151 #endif /* HAVE_YASM */
152 
153 #define SET_HPEL_FUNCS(PFX, IDX, SIZE, CPU) \
154  do { \
155  c->PFX ## _pixels_tab IDX [0] = PFX ## _pixels ## SIZE ## _ ## CPU; \
156  c->PFX ## _pixels_tab IDX [1] = PFX ## _pixels ## SIZE ## _x2_ ## CPU; \
157  c->PFX ## _pixels_tab IDX [2] = PFX ## _pixels ## SIZE ## _y2_ ## CPU; \
158  c->PFX ## _pixels_tab IDX [3] = PFX ## _pixels ## SIZE ## _xy2_ ## CPU; \
159  } while (0)
160 
161 static void hpeldsp_init_mmx(HpelDSPContext *c, int flags, int cpu_flags)
162 {
163 #if HAVE_MMX_INLINE
164  SET_HPEL_FUNCS(put, [0], 16, mmx);
165  SET_HPEL_FUNCS(put_no_rnd, [0], 16, mmx);
166  SET_HPEL_FUNCS(avg, [0], 16, mmx);
167  SET_HPEL_FUNCS(avg_no_rnd, , 16, mmx);
168  SET_HPEL_FUNCS(put, [1], 8, mmx);
169  SET_HPEL_FUNCS(put_no_rnd, [1], 8, mmx);
170  SET_HPEL_FUNCS(avg, [1], 8, mmx);
171 #endif /* HAVE_MMX_INLINE */
172 }
173 
174 static void hpeldsp_init_mmxext(HpelDSPContext *c, int flags, int cpu_flags)
175 {
176 #if HAVE_MMXEXT_EXTERNAL
177  c->put_pixels_tab[0][1] = ff_put_pixels16_x2_mmxext;
178  c->put_pixels_tab[0][2] = put_pixels16_y2_mmxext;
179 
180  c->avg_pixels_tab[0][0] = avg_pixels16_mmxext;
181  c->avg_pixels_tab[0][1] = avg_pixels16_x2_mmxext;
182  c->avg_pixels_tab[0][2] = avg_pixels16_y2_mmxext;
183 
184  c->put_pixels_tab[1][1] = ff_put_pixels8_x2_mmxext;
185  c->put_pixels_tab[1][2] = ff_put_pixels8_y2_mmxext;
186 
187  c->avg_pixels_tab[1][0] = ff_avg_pixels8_mmxext;
188  c->avg_pixels_tab[1][1] = ff_avg_pixels8_x2_mmxext;
189  c->avg_pixels_tab[1][2] = ff_avg_pixels8_y2_mmxext;
190 
191  if (!(flags & 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_pixels16_xy2_mmxext;
198  c->avg_pixels_tab[1][3] = ff_avg_pixels8_xy2_mmxext;
199  }
200 
201  if (flags & CODEC_FLAG_BITEXACT && CONFIG_VP3_DECODER) {
202  c->put_no_rnd_pixels_tab[1][1] = ff_put_no_rnd_pixels8_x2_exact_mmxext;
203  c->put_no_rnd_pixels_tab[1][2] = ff_put_no_rnd_pixels8_y2_exact_mmxext;
204  }
205 #endif /* HAVE_MMXEXT_EXTERNAL */
206 }
207 
208 static void hpeldsp_init_3dnow(HpelDSPContext *c, int flags, int cpu_flags)
209 {
210 #if HAVE_AMD3DNOW_EXTERNAL
211  c->put_pixels_tab[0][1] = ff_put_pixels16_x2_3dnow;
212  c->put_pixels_tab[0][2] = put_pixels16_y2_3dnow;
213 
214  c->avg_pixels_tab[0][0] = avg_pixels16_3dnow;
215  c->avg_pixels_tab[0][1] = avg_pixels16_x2_3dnow;
216  c->avg_pixels_tab[0][2] = avg_pixels16_y2_3dnow;
217 
218  c->put_pixels_tab[1][1] = ff_put_pixels8_x2_3dnow;
219  c->put_pixels_tab[1][2] = ff_put_pixels8_y2_3dnow;
220 
221  c->avg_pixels_tab[1][0] = ff_avg_pixels8_3dnow;
222  c->avg_pixels_tab[1][1] = ff_avg_pixels8_x2_3dnow;
223  c->avg_pixels_tab[1][2] = ff_avg_pixels8_y2_3dnow;
224 
225  if (!(flags & CODEC_FLAG_BITEXACT)){
226  c->put_no_rnd_pixels_tab[0][1] = put_no_rnd_pixels16_x2_3dnow;
227  c->put_no_rnd_pixels_tab[0][2] = put_no_rnd_pixels16_y2_3dnow;
228  c->put_no_rnd_pixels_tab[1][1] = ff_put_no_rnd_pixels8_x2_3dnow;
229  c->put_no_rnd_pixels_tab[1][2] = ff_put_no_rnd_pixels8_y2_3dnow;
230 
231  c->avg_pixels_tab[0][3] = avg_pixels16_xy2_3dnow;
232  c->avg_pixels_tab[1][3] = ff_avg_pixels8_xy2_3dnow;
233  }
234 
235  if (flags & CODEC_FLAG_BITEXACT && CONFIG_VP3_DECODER) {
236  c->put_no_rnd_pixels_tab[1][1] = ff_put_no_rnd_pixels8_x2_exact_3dnow;
237  c->put_no_rnd_pixels_tab[1][2] = ff_put_no_rnd_pixels8_y2_exact_3dnow;
238  }
239 #endif /* HAVE_AMD3DNOW_EXTERNAL */
240 }
241 
242 static void hpeldsp_init_sse2(HpelDSPContext *c, int flags, int cpu_flags)
243 {
244 #if HAVE_SSE2_EXTERNAL
245  if (!(cpu_flags & AV_CPU_FLAG_SSE2SLOW)) {
246  // these functions are slower than mmx on AMD, but faster on Intel
247  c->put_pixels_tab[0][0] = ff_put_pixels16_sse2;
248  c->put_no_rnd_pixels_tab[0][0] = ff_put_pixels16_sse2;
249  c->avg_pixels_tab[0][0] = ff_avg_pixels16_sse2;
250  }
251 #endif /* HAVE_SSE2_EXTERNAL */
252 }
253 
254 void ff_hpeldsp_init_x86(HpelDSPContext *c, int flags)
255 {
256  int cpu_flags = av_get_cpu_flags();
257 
258  if (INLINE_MMX(cpu_flags))
259  hpeldsp_init_mmx(c, flags, cpu_flags);
260 
261  if (EXTERNAL_MMXEXT(cpu_flags))
262  hpeldsp_init_mmxext(c, flags, cpu_flags);
263 
264  if (EXTERNAL_AMD3DNOW(cpu_flags))
265  hpeldsp_init_3dnow(c, flags, cpu_flags);
266 
267  if (EXTERNAL_SSE2(cpu_flags))
268  hpeldsp_init_sse2(c, flags, cpu_flags);
269 }
Generated on Sat Jan 25 2014 19:51:58 for FFmpeg by   doxygen 1.8.2