rv40dsp_init.c
Go to the documentation of this file.
1 /*
2  * RV40 decoder motion compensation functions x86-optimised
3  * Copyright (c) 2008 Konstantin Shishkov
4  *
5  * This file is part of Libav.
6  *
7  * Libav 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  * Libav 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 Libav; if not, write to the Free Software
19  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
20  */
21 
29 #include "libavcodec/rv34dsp.h"
30 #include "libavutil/mem.h"
31 #include "libavutil/x86/cpu.h"
32 #include "dsputil_mmx.h"
33 
34 #if HAVE_YASM
35 void ff_put_rv40_chroma_mc8_mmx (uint8_t *dst, uint8_t *src,
36  int stride, int h, int x, int y);
37 void ff_avg_rv40_chroma_mc8_mmxext(uint8_t *dst, uint8_t *src,
38  int stride, int h, int x, int y);
39 void ff_avg_rv40_chroma_mc8_3dnow(uint8_t *dst, uint8_t *src,
40  int stride, int h, int x, int y);
41 
42 void ff_put_rv40_chroma_mc4_mmx (uint8_t *dst, uint8_t *src,
43  int stride, int h, int x, int y);
44 void ff_avg_rv40_chroma_mc4_mmxext(uint8_t *dst, uint8_t *src,
45  int stride, int h, int x, int y);
46 void ff_avg_rv40_chroma_mc4_3dnow(uint8_t *dst, uint8_t *src,
47  int stride, int h, int x, int y);
48 
49 #define DECLARE_WEIGHT(opt) \
50 void ff_rv40_weight_func_rnd_16_##opt(uint8_t *dst, uint8_t *src1, uint8_t *src2, \
51  int w1, int w2, ptrdiff_t stride); \
52 void ff_rv40_weight_func_rnd_8_##opt (uint8_t *dst, uint8_t *src1, uint8_t *src2, \
53  int w1, int w2, ptrdiff_t stride); \
54 void ff_rv40_weight_func_nornd_16_##opt(uint8_t *dst, uint8_t *src1, uint8_t *src2, \
55  int w1, int w2, ptrdiff_t stride); \
56 void ff_rv40_weight_func_nornd_8_##opt (uint8_t *dst, uint8_t *src1, uint8_t *src2, \
57  int w1, int w2, ptrdiff_t stride);
58 DECLARE_WEIGHT(mmxext)
59 DECLARE_WEIGHT(sse2)
60 DECLARE_WEIGHT(ssse3)
61 
70 #define QPEL_FUNC_DECL(OP, SIZE, PH, PV, OPT) \
71 static void OP ## rv40_qpel ##SIZE ##_mc ##PH ##PV ##OPT(uint8_t *dst, \
72  uint8_t *src, \
73  int stride) \
74 { \
75  int i; \
76  if (PH && PV) { \
77  DECLARE_ALIGNED(16, uint8_t, tmp)[SIZE * (SIZE + 5)]; \
78  uint8_t *tmpptr = tmp + SIZE * 2; \
79  src -= stride * 2; \
80  \
81  for (i = 0; i < SIZE; i += LOOPSIZE) \
82  ff_put_rv40_qpel_h ##OPT(tmp + i, SIZE, src + i, stride, \
83  SIZE + 5, HCOFF(PH)); \
84  for (i = 0; i < SIZE; i += LOOPSIZE) \
85  ff_ ##OP ##rv40_qpel_v ##OPT(dst + i, stride, tmpptr + i, \
86  SIZE, SIZE, VCOFF(PV)); \
87  } else if (PV) { \
88  for (i = 0; i < SIZE; i += LOOPSIZE) \
89  ff_ ##OP ##rv40_qpel_v ## OPT(dst + i, stride, src + i, \
90  stride, SIZE, VCOFF(PV)); \
91  } else { \
92  for (i = 0; i < SIZE; i += LOOPSIZE) \
93  ff_ ##OP ##rv40_qpel_h ## OPT(dst + i, stride, src + i, \
94  stride, SIZE, HCOFF(PH)); \
95  } \
96 };
97 
100 #define QPEL_FUNCS_DECL(OP, PH, PV, OPT) \
101  QPEL_FUNC_DECL(OP, 8, PH, PV, OPT) \
102  QPEL_FUNC_DECL(OP, 16, PH, PV, OPT)
103 
105 #define QPEL_MC_DECL(OP, OPT) \
106 void ff_ ##OP ##rv40_qpel_h ##OPT(uint8_t *dst, ptrdiff_t dstStride, \
107  const uint8_t *src, \
108  ptrdiff_t srcStride, \
109  int len, int m); \
110 void ff_ ##OP ##rv40_qpel_v ##OPT(uint8_t *dst, ptrdiff_t dstStride, \
111  const uint8_t *src, \
112  ptrdiff_t srcStride, \
113  int len, int m); \
114 QPEL_FUNCS_DECL(OP, 0, 1, OPT) \
115 QPEL_FUNCS_DECL(OP, 0, 3, OPT) \
116 QPEL_FUNCS_DECL(OP, 1, 0, OPT) \
117 QPEL_FUNCS_DECL(OP, 1, 1, OPT) \
118 QPEL_FUNCS_DECL(OP, 1, 2, OPT) \
119 QPEL_FUNCS_DECL(OP, 1, 3, OPT) \
120 QPEL_FUNCS_DECL(OP, 2, 1, OPT) \
121 QPEL_FUNCS_DECL(OP, 2, 2, OPT) \
122 QPEL_FUNCS_DECL(OP, 2, 3, OPT) \
123 QPEL_FUNCS_DECL(OP, 3, 0, OPT) \
124 QPEL_FUNCS_DECL(OP, 3, 1, OPT) \
125 QPEL_FUNCS_DECL(OP, 3, 2, OPT)
126 
128 #define LOOPSIZE 8
129 #define HCOFF(x) (32 * (x - 1))
130 #define VCOFF(x) (32 * (x - 1))
131 QPEL_MC_DECL(put_, _ssse3)
132 QPEL_MC_DECL(avg_, _ssse3)
133 
134 #undef LOOPSIZE
135 #undef HCOFF
136 #undef VCOFF
137 #define LOOPSIZE 8
138 #define HCOFF(x) (64 * (x - 1))
139 #define VCOFF(x) (64 * (x - 1))
140 QPEL_MC_DECL(put_, _sse2)
141 QPEL_MC_DECL(avg_, _sse2)
142 
143 #if ARCH_X86_32
144 #undef LOOPSIZE
145 #undef HCOFF
146 #undef VCOFF
147 #define LOOPSIZE 4
148 #define HCOFF(x) (64 * (x - 1))
149 #define VCOFF(x) (64 * (x - 1))
150 
151 QPEL_MC_DECL(put_, _mmx)
152 
153 #define ff_put_rv40_qpel_h_mmxext ff_put_rv40_qpel_h_mmx
154 #define ff_put_rv40_qpel_v_mmxext ff_put_rv40_qpel_v_mmx
155 QPEL_MC_DECL(avg_, _mmxext)
156 
157 #define ff_put_rv40_qpel_h_3dnow ff_put_rv40_qpel_h_mmx
158 #define ff_put_rv40_qpel_v_3dnow ff_put_rv40_qpel_v_mmx
159 QPEL_MC_DECL(avg_, _3dnow)
160 #endif
161 
164 #define QPEL_FUNC_SET(OP, SIZE, PH, PV, OPT) \
165  c-> OP ## pixels_tab[2 - SIZE / 8][4 * PV + PH] = OP ## rv40_qpel ##SIZE ## _mc ##PH ##PV ##OPT;
166 
168 #define QPEL_FUNCS_SET(OP, PH, PV, OPT) \
169  QPEL_FUNC_SET(OP, 8, PH, PV, OPT) \
170  QPEL_FUNC_SET(OP, 16, PH, PV, OPT)
171 
173 #define QPEL_MC_SET(OP, OPT) \
174 QPEL_FUNCS_SET (OP, 0, 1, OPT) \
175 QPEL_FUNCS_SET (OP, 0, 3, OPT) \
176 QPEL_FUNCS_SET (OP, 1, 0, OPT) \
177 QPEL_FUNCS_SET (OP, 1, 1, OPT) \
178 QPEL_FUNCS_SET (OP, 1, 2, OPT) \
179 QPEL_FUNCS_SET (OP, 1, 3, OPT) \
180 QPEL_FUNCS_SET (OP, 2, 1, OPT) \
181 QPEL_FUNCS_SET (OP, 2, 2, OPT) \
182 QPEL_FUNCS_SET (OP, 2, 3, OPT) \
183 QPEL_FUNCS_SET (OP, 3, 0, OPT) \
184 QPEL_FUNCS_SET (OP, 3, 1, OPT) \
185 QPEL_FUNCS_SET (OP, 3, 2, OPT)
186 
188 #endif /* HAVE_YASM */
189 
191 {
192 #if HAVE_YASM
193  int mm_flags = av_get_cpu_flags();
194 
195  if (EXTERNAL_MMX(mm_flags)) {
196  c->put_chroma_pixels_tab[0] = ff_put_rv40_chroma_mc8_mmx;
197  c->put_chroma_pixels_tab[1] = ff_put_rv40_chroma_mc4_mmx;
198 #if HAVE_MMX_INLINE
203 #endif /* HAVE_MMX_INLINE */
204 #if ARCH_X86_32
205  QPEL_MC_SET(put_, _mmx)
206 #endif
207  }
208  if (EXTERNAL_MMXEXT(mm_flags)) {
209  c->avg_chroma_pixels_tab[0] = ff_avg_rv40_chroma_mc8_mmxext;
210  c->avg_chroma_pixels_tab[1] = ff_avg_rv40_chroma_mc4_mmxext;
211  c->rv40_weight_pixels_tab[0][0] = ff_rv40_weight_func_rnd_16_mmxext;
212  c->rv40_weight_pixels_tab[0][1] = ff_rv40_weight_func_rnd_8_mmxext;
213  c->rv40_weight_pixels_tab[1][0] = ff_rv40_weight_func_nornd_16_mmxext;
214  c->rv40_weight_pixels_tab[1][1] = ff_rv40_weight_func_nornd_8_mmxext;
215 #if ARCH_X86_32
216  QPEL_MC_SET(avg_, _mmxext)
217 #endif
218  } else if (EXTERNAL_AMD3DNOW(mm_flags)) {
219  c->avg_chroma_pixels_tab[0] = ff_avg_rv40_chroma_mc8_3dnow;
220  c->avg_chroma_pixels_tab[1] = ff_avg_rv40_chroma_mc4_3dnow;
221 #if ARCH_X86_32
222  QPEL_MC_SET(avg_, _3dnow)
223 #endif
224  }
225  if (EXTERNAL_SSE2(mm_flags)) {
226  c->rv40_weight_pixels_tab[0][0] = ff_rv40_weight_func_rnd_16_sse2;
227  c->rv40_weight_pixels_tab[0][1] = ff_rv40_weight_func_rnd_8_sse2;
228  c->rv40_weight_pixels_tab[1][0] = ff_rv40_weight_func_nornd_16_sse2;
229  c->rv40_weight_pixels_tab[1][1] = ff_rv40_weight_func_nornd_8_sse2;
230  QPEL_MC_SET(put_, _sse2)
231  QPEL_MC_SET(avg_, _sse2)
232  }
233  if (EXTERNAL_SSSE3(mm_flags)) {
234  c->rv40_weight_pixels_tab[0][0] = ff_rv40_weight_func_rnd_16_ssse3;
235  c->rv40_weight_pixels_tab[0][1] = ff_rv40_weight_func_rnd_8_ssse3;
236  c->rv40_weight_pixels_tab[1][0] = ff_rv40_weight_func_nornd_16_ssse3;
237  c->rv40_weight_pixels_tab[1][1] = ff_rv40_weight_func_nornd_8_ssse3;
238  QPEL_MC_SET(put_, _ssse3)
239  QPEL_MC_SET(avg_, _ssse3)
240  }
241 #endif /* HAVE_YASM */
242 }
qpel_mc_func put_pixels_tab[4][16]
Definition: rv34dsp.h:57
#define EXTERNAL_MMX(flags)
Definition: cpu.h:33
memory handling functions
int stride
Definition: mace.c:144
uint8_t
rv40_weight_func rv40_weight_pixels_tab[2][2]
Biweight functions, first dimension is transform size (16/8), second is whether the weight is prescal...
Definition: rv34dsp.h:66
#define EXTERNAL_SSE2(flags)
Definition: cpu.h:36
void ff_avg_rv40_qpel16_mc33_mmx(uint8_t *block, uint8_t *pixels, int line_size)
qpel_mc_func avg_pixels_tab[4][16]
Definition: rv34dsp.h:58
RV30/40 decoder motion compensation functions.
#define EXTERNAL_SSSE3(flags)
Definition: cpu.h:38
int av_get_cpu_flags(void)
Return the flags which specify extensions supported by the CPU.
Definition: cpu.c:25
void ff_avg_rv40_qpel8_mc33_mmx(uint8_t *block, uint8_t *pixels, int line_size)
#define EXTERNAL_MMXEXT(flags)
Definition: cpu.h:34
void ff_put_rv40_qpel16_mc33_mmx(uint8_t *block, uint8_t *pixels, int line_size)
void ff_put_rv40_qpel8_mc33_mmx(uint8_t *block, uint8_t *pixels, int line_size)
h264_chroma_mc_func avg_chroma_pixels_tab[3]
Definition: rv34dsp.h:60
void ff_rv40dsp_init_x86(RV34DSPContext *c, DSPContext *dsp)
Definition: rv40dsp_init.c:190
h264_chroma_mc_func put_chroma_pixels_tab[3]
Definition: rv34dsp.h:59
#define EXTERNAL_AMD3DNOW(flags)
Definition: cpu.h:31
DSPContext.
Definition: dsputil.h:194