FFmpeg
 All Data Structures Namespaces Files Functions Variables Typedefs Enumerations Enumerator Macros Groups Pages
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 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 /**
23  * @file
24  * RV40 decoder motion compensation functions x86-optimised
25  * 2,0 and 0,2 have h264 equivalents.
26  * 3,3 is bugged in the rv40 format and maps to _xy2 version
27  */
28 
29 #include "libavcodec/rv34dsp.h"
30 #include "libavutil/attributes.h"
31 #include "libavutil/mem.h"
32 #include "libavutil/x86/cpu.h"
33 #include "dsputil_x86.h"
34 
35 #if HAVE_YASM
36 void ff_put_rv40_chroma_mc8_mmx (uint8_t *dst, uint8_t *src,
37  int stride, int h, int x, int y);
38 void ff_avg_rv40_chroma_mc8_mmxext(uint8_t *dst, uint8_t *src,
39  int stride, int h, int x, int y);
40 void ff_avg_rv40_chroma_mc8_3dnow(uint8_t *dst, uint8_t *src,
41  int stride, int h, int x, int y);
42 
43 void ff_put_rv40_chroma_mc4_mmx (uint8_t *dst, uint8_t *src,
44  int stride, int h, int x, int y);
45 void ff_avg_rv40_chroma_mc4_mmxext(uint8_t *dst, uint8_t *src,
46  int stride, int h, int x, int y);
47 void ff_avg_rv40_chroma_mc4_3dnow(uint8_t *dst, uint8_t *src,
48  int stride, int h, int x, int y);
49 
50 #define DECLARE_WEIGHT(opt) \
51 void ff_rv40_weight_func_rnd_16_##opt(uint8_t *dst, uint8_t *src1, uint8_t *src2, \
52  int w1, int w2, ptrdiff_t stride); \
53 void ff_rv40_weight_func_rnd_8_##opt (uint8_t *dst, uint8_t *src1, uint8_t *src2, \
54  int w1, int w2, ptrdiff_t stride); \
55 void ff_rv40_weight_func_nornd_16_##opt(uint8_t *dst, uint8_t *src1, uint8_t *src2, \
56  int w1, int w2, ptrdiff_t stride); \
57 void ff_rv40_weight_func_nornd_8_##opt (uint8_t *dst, uint8_t *src1, uint8_t *src2, \
58  int w1, int w2, ptrdiff_t stride);
59 DECLARE_WEIGHT(mmxext)
60 DECLARE_WEIGHT(sse2)
61 DECLARE_WEIGHT(ssse3)
62 
63 /** @{ */
64 /**
65  * Define one qpel function.
66  * LOOPSIZE must be already set to the number of pixels processed per
67  * iteration in the inner loop of the called functions.
68  * COFF(x) must be already defined so as to provide the offset into any
69  * array of coeffs used by the called function for the qpel position x.
70  */
71 #define QPEL_FUNC_DECL(OP, SIZE, PH, PV, OPT) \
72 static void OP ## rv40_qpel ##SIZE ##_mc ##PH ##PV ##OPT(uint8_t *dst, \
73  uint8_t *src, \
74  ptrdiff_t stride) \
75 { \
76  int i; \
77  if (PH && PV) { \
78  DECLARE_ALIGNED(16, uint8_t, tmp)[SIZE * (SIZE + 5)]; \
79  uint8_t *tmpptr = tmp + SIZE * 2; \
80  src -= stride * 2; \
81  \
82  for (i = 0; i < SIZE; i += LOOPSIZE) \
83  ff_put_rv40_qpel_h ##OPT(tmp + i, SIZE, src + i, stride, \
84  SIZE + 5, HCOFF(PH)); \
85  for (i = 0; i < SIZE; i += LOOPSIZE) \
86  ff_ ##OP ##rv40_qpel_v ##OPT(dst + i, stride, tmpptr + i, \
87  SIZE, SIZE, VCOFF(PV)); \
88  } else if (PV) { \
89  for (i = 0; i < SIZE; i += LOOPSIZE) \
90  ff_ ##OP ##rv40_qpel_v ## OPT(dst + i, stride, src + i, \
91  stride, SIZE, VCOFF(PV)); \
92  } else { \
93  for (i = 0; i < SIZE; i += LOOPSIZE) \
94  ff_ ##OP ##rv40_qpel_h ## OPT(dst + i, stride, src + i, \
95  stride, SIZE, HCOFF(PH)); \
96  } \
97 };
98 
99 /** Declare functions for sizes 8 and 16 and given operations
100  * and qpel position. */
101 #define QPEL_FUNCS_DECL(OP, PH, PV, OPT) \
102  QPEL_FUNC_DECL(OP, 8, PH, PV, OPT) \
103  QPEL_FUNC_DECL(OP, 16, PH, PV, OPT)
104 
105 /** Declare all functions for all sizes and qpel positions */
106 #define QPEL_MC_DECL(OP, OPT) \
107 void ff_ ##OP ##rv40_qpel_h ##OPT(uint8_t *dst, ptrdiff_t dstStride, \
108  const uint8_t *src, \
109  ptrdiff_t srcStride, \
110  int len, int m); \
111 void ff_ ##OP ##rv40_qpel_v ##OPT(uint8_t *dst, ptrdiff_t dstStride, \
112  const uint8_t *src, \
113  ptrdiff_t srcStride, \
114  int len, int m); \
115 QPEL_FUNCS_DECL(OP, 0, 1, OPT) \
116 QPEL_FUNCS_DECL(OP, 0, 3, OPT) \
117 QPEL_FUNCS_DECL(OP, 1, 0, OPT) \
118 QPEL_FUNCS_DECL(OP, 1, 1, OPT) \
119 QPEL_FUNCS_DECL(OP, 1, 2, OPT) \
120 QPEL_FUNCS_DECL(OP, 1, 3, OPT) \
121 QPEL_FUNCS_DECL(OP, 2, 1, OPT) \
122 QPEL_FUNCS_DECL(OP, 2, 2, OPT) \
123 QPEL_FUNCS_DECL(OP, 2, 3, OPT) \
124 QPEL_FUNCS_DECL(OP, 3, 0, OPT) \
125 QPEL_FUNCS_DECL(OP, 3, 1, OPT) \
126 QPEL_FUNCS_DECL(OP, 3, 2, OPT)
127 /** @} */
128 
129 #define LOOPSIZE 8
130 #define HCOFF(x) (32 * (x - 1))
131 #define VCOFF(x) (32 * (x - 1))
132 QPEL_MC_DECL(put_, _ssse3)
133 QPEL_MC_DECL(avg_, _ssse3)
134 
135 #undef LOOPSIZE
136 #undef HCOFF
137 #undef VCOFF
138 #define LOOPSIZE 8
139 #define HCOFF(x) (64 * (x - 1))
140 #define VCOFF(x) (64 * (x - 1))
141 QPEL_MC_DECL(put_, _sse2)
142 QPEL_MC_DECL(avg_, _sse2)
143 
144 #if ARCH_X86_32
145 #undef LOOPSIZE
146 #undef HCOFF
147 #undef VCOFF
148 #define LOOPSIZE 4
149 #define HCOFF(x) (64 * (x - 1))
150 #define VCOFF(x) (64 * (x - 1))
151 
152 QPEL_MC_DECL(put_, _mmx)
153 
154 #define ff_put_rv40_qpel_h_mmxext ff_put_rv40_qpel_h_mmx
155 #define ff_put_rv40_qpel_v_mmxext ff_put_rv40_qpel_v_mmx
156 QPEL_MC_DECL(avg_, _mmxext)
157 
158 #define ff_put_rv40_qpel_h_3dnow ff_put_rv40_qpel_h_mmx
159 #define ff_put_rv40_qpel_v_3dnow ff_put_rv40_qpel_v_mmx
160 QPEL_MC_DECL(avg_, _3dnow)
161 #endif
162 
163 /** @{ */
164 /** Set one function */
165 #define QPEL_FUNC_SET(OP, SIZE, PH, PV, OPT) \
166  c-> OP ## pixels_tab[2 - SIZE / 8][4 * PV + PH] = OP ## rv40_qpel ##SIZE ## _mc ##PH ##PV ##OPT;
167 
168 /** Set functions put and avg for sizes 8 and 16 and a given qpel position */
169 #define QPEL_FUNCS_SET(OP, PH, PV, OPT) \
170  QPEL_FUNC_SET(OP, 8, PH, PV, OPT) \
171  QPEL_FUNC_SET(OP, 16, PH, PV, OPT)
172 
173 /** Set all functions for all sizes and qpel positions */
174 #define QPEL_MC_SET(OP, OPT) \
175 QPEL_FUNCS_SET (OP, 0, 1, OPT) \
176 QPEL_FUNCS_SET (OP, 0, 3, OPT) \
177 QPEL_FUNCS_SET (OP, 1, 0, OPT) \
178 QPEL_FUNCS_SET (OP, 1, 1, OPT) \
179 QPEL_FUNCS_SET (OP, 1, 2, OPT) \
180 QPEL_FUNCS_SET (OP, 1, 3, OPT) \
181 QPEL_FUNCS_SET (OP, 2, 1, OPT) \
182 QPEL_FUNCS_SET (OP, 2, 2, OPT) \
183 QPEL_FUNCS_SET (OP, 2, 3, OPT) \
184 QPEL_FUNCS_SET (OP, 3, 0, OPT) \
185 QPEL_FUNCS_SET (OP, 3, 1, OPT) \
186 QPEL_FUNCS_SET (OP, 3, 2, OPT)
187 /** @} */
188 
189 #endif /* HAVE_YASM */
190 
191 #if HAVE_MMX_INLINE
192 static void put_rv40_qpel8_mc33_mmx(uint8_t *dst, uint8_t *src,
193  ptrdiff_t stride)
194 {
195  ff_put_pixels8_xy2_mmx(dst, src, stride, 8);
196 }
197 static void put_rv40_qpel16_mc33_mmx(uint8_t *dst, uint8_t *src,
198  ptrdiff_t stride)
199 {
200  ff_put_pixels16_xy2_mmx(dst, src, stride, 16);
201 }
202 static void avg_rv40_qpel8_mc33_mmx(uint8_t *dst, uint8_t *src,
203  ptrdiff_t stride)
204 {
205  ff_avg_pixels8_xy2_mmx(dst, src, stride, 8);
206 }
207 static void avg_rv40_qpel16_mc33_mmx(uint8_t *dst, uint8_t *src,
208  ptrdiff_t stride)
209 {
210  ff_avg_pixels16_xy2_mmx(dst, src, stride, 16);
211 }
212 #endif /* HAVE_MMX_INLINE */
213 
215 {
216 #if HAVE_YASM
217  int mm_flags = av_get_cpu_flags();
218 
219  if (EXTERNAL_MMX(mm_flags)) {
220  c->put_chroma_pixels_tab[0] = ff_put_rv40_chroma_mc8_mmx;
221  c->put_chroma_pixels_tab[1] = ff_put_rv40_chroma_mc4_mmx;
222 #if HAVE_MMX_INLINE
223  c->put_pixels_tab[0][15] = put_rv40_qpel16_mc33_mmx;
224  c->put_pixels_tab[1][15] = put_rv40_qpel8_mc33_mmx;
225  c->avg_pixels_tab[0][15] = avg_rv40_qpel16_mc33_mmx;
226  c->avg_pixels_tab[1][15] = avg_rv40_qpel8_mc33_mmx;
227 #endif /* HAVE_MMX_INLINE */
228 #if ARCH_X86_32
229  QPEL_MC_SET(put_, _mmx)
230 #endif
231  }
232  if (EXTERNAL_MMXEXT(mm_flags)) {
233  c->avg_chroma_pixels_tab[0] = ff_avg_rv40_chroma_mc8_mmxext;
234  c->avg_chroma_pixels_tab[1] = ff_avg_rv40_chroma_mc4_mmxext;
235  c->rv40_weight_pixels_tab[0][0] = ff_rv40_weight_func_rnd_16_mmxext;
236  c->rv40_weight_pixels_tab[0][1] = ff_rv40_weight_func_rnd_8_mmxext;
237  c->rv40_weight_pixels_tab[1][0] = ff_rv40_weight_func_nornd_16_mmxext;
238  c->rv40_weight_pixels_tab[1][1] = ff_rv40_weight_func_nornd_8_mmxext;
239 #if ARCH_X86_32
240  QPEL_MC_SET(avg_, _mmxext)
241 #endif
242  } else if (EXTERNAL_AMD3DNOW(mm_flags)) {
243  c->avg_chroma_pixels_tab[0] = ff_avg_rv40_chroma_mc8_3dnow;
244  c->avg_chroma_pixels_tab[1] = ff_avg_rv40_chroma_mc4_3dnow;
245 #if ARCH_X86_32
246  QPEL_MC_SET(avg_, _3dnow)
247 #endif
248  }
249  if (EXTERNAL_SSE2(mm_flags)) {
250  c->rv40_weight_pixels_tab[0][0] = ff_rv40_weight_func_rnd_16_sse2;
251  c->rv40_weight_pixels_tab[0][1] = ff_rv40_weight_func_rnd_8_sse2;
252  c->rv40_weight_pixels_tab[1][0] = ff_rv40_weight_func_nornd_16_sse2;
253  c->rv40_weight_pixels_tab[1][1] = ff_rv40_weight_func_nornd_8_sse2;
254  QPEL_MC_SET(put_, _sse2)
255  QPEL_MC_SET(avg_, _sse2)
256  }
257  if (EXTERNAL_SSSE3(mm_flags)) {
258  c->rv40_weight_pixels_tab[0][0] = ff_rv40_weight_func_rnd_16_ssse3;
259  c->rv40_weight_pixels_tab[0][1] = ff_rv40_weight_func_rnd_8_ssse3;
260  c->rv40_weight_pixels_tab[1][0] = ff_rv40_weight_func_nornd_16_ssse3;
261  c->rv40_weight_pixels_tab[1][1] = ff_rv40_weight_func_nornd_8_ssse3;
262  QPEL_MC_SET(put_, _ssse3)
263  QPEL_MC_SET(avg_, _ssse3)
264  }
265 #endif /* HAVE_YASM */
266 }