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/mem_internal.h"
33 #include "libavutil/x86/cpu.h"
34 #include "hpeldsp.h"
35
36 #define DEFINE_FN(op, size, insn) \
37 static void op##_rv40_qpel##size##_mc33_##insn(uint8_t *dst, const uint8_t *src, \
38 ptrdiff_t stride) \
39 { \
40 ff_##op##_pixels##size##_xy2_##insn(dst, src, stride, size); \
41 }
42
43 #if HAVE_X86ASM
44 void ff_put_rv40_chroma_mc8_mmx (uint8_t *dst, uint8_t *src,
45 ptrdiff_t stride, int h, int x, int y);
46 void ff_avg_rv40_chroma_mc8_mmxext(uint8_t *dst, uint8_t *src,
47 ptrdiff_t stride, int h, int x, int y);
48 void ff_avg_rv40_chroma_mc8_3dnow(uint8_t *dst, uint8_t *src,
49 ptrdiff_t stride, int h, int x, int y);
50
51 void ff_put_rv40_chroma_mc4_mmx (uint8_t *dst, uint8_t *src,
52 ptrdiff_t stride, int h, int x, int y);
53 void ff_avg_rv40_chroma_mc4_mmxext(uint8_t *dst, uint8_t *src,
54 ptrdiff_t stride, int h, int x, int y);
55 void ff_avg_rv40_chroma_mc4_3dnow(uint8_t *dst, uint8_t *src,
56 ptrdiff_t stride, int h, int x, int y);
57
58 #define DECLARE_WEIGHT(opt) \
59 void ff_rv40_weight_func_rnd_16_##opt(uint8_t *dst, uint8_t *src1, uint8_t *src2, \
60 int w1, int w2, ptrdiff_t stride); \
61 void ff_rv40_weight_func_rnd_8_##opt (uint8_t *dst, uint8_t *src1, uint8_t *src2, \
62 int w1, int w2, ptrdiff_t stride); \
63 void ff_rv40_weight_func_nornd_16_##opt(uint8_t *dst, uint8_t *src1, uint8_t *src2, \
64 int w1, int w2, ptrdiff_t stride); \
65 void ff_rv40_weight_func_nornd_8_##opt (uint8_t *dst, uint8_t *src1, uint8_t *src2, \
66 int w1, int w2, ptrdiff_t stride);
67 DECLARE_WEIGHT(mmxext)
DECLARE_WEIGHT(sse2)68 DECLARE_WEIGHT(sse2)
69 DECLARE_WEIGHT(ssse3)
70
71 /** @{ */
72 /**
73 * Define one qpel function.
74 * LOOPSIZE must be already set to the number of pixels processed per
75 * iteration in the inner loop of the called functions.
76 * COFF(x) must be already defined so as to provide the offset into any
77 * array of coeffs used by the called function for the qpel position x.
78 */
79 #define QPEL_FUNC_DECL(OP, SIZE, PH, PV, OPT) \
80 static void OP ## rv40_qpel ##SIZE ##_mc ##PH ##PV ##OPT(uint8_t *dst, \
81 const uint8_t *src, \
82 ptrdiff_t stride) \
83 { \
84 int i; \
85 if (PH && PV) { \
86 LOCAL_ALIGNED(16, uint8_t, tmp, [SIZE * (SIZE + 5)]); \
87 uint8_t *tmpptr = tmp + SIZE * 2; \
88 src -= stride * 2; \
89 \
90 for (i = 0; i < SIZE; i += LOOPSIZE) \
91 ff_put_rv40_qpel_h ##OPT(tmp + i, SIZE, src + i, stride, \
92 SIZE + 5, HCOFF(PH)); \
93 for (i = 0; i < SIZE; i += LOOPSIZE) \
94 ff_ ##OP ##rv40_qpel_v ##OPT(dst + i, stride, tmpptr + i, \
95 SIZE, SIZE, VCOFF(PV)); \
96 } else if (PV) { \
97 for (i = 0; i < SIZE; i += LOOPSIZE) \
98 ff_ ##OP ##rv40_qpel_v ## OPT(dst + i, stride, src + i, \
99 stride, SIZE, VCOFF(PV)); \
100 } else { \
101 for (i = 0; i < SIZE; i += LOOPSIZE) \
102 ff_ ##OP ##rv40_qpel_h ## OPT(dst + i, stride, src + i, \
103 stride, SIZE, HCOFF(PH)); \
104 } \
105 }
106
107 /** Declare functions for sizes 8 and 16 and given operations
108 * and qpel position. */
109 #define QPEL_FUNCS_DECL(OP, PH, PV, OPT) \
110 QPEL_FUNC_DECL(OP, 8, PH, PV, OPT) \
111 QPEL_FUNC_DECL(OP, 16, PH, PV, OPT)
112
113 /** Declare all functions for all sizes and qpel positions */
114 #define QPEL_MC_DECL(OP, OPT) \
115 void ff_ ##OP ##rv40_qpel_h ##OPT(uint8_t *dst, ptrdiff_t dstStride, \
116 const uint8_t *src, \
117 ptrdiff_t srcStride, \
118 int len, int m); \
119 void ff_ ##OP ##rv40_qpel_v ##OPT(uint8_t *dst, ptrdiff_t dstStride, \
120 const uint8_t *src, \
121 ptrdiff_t srcStride, \
122 int len, int m); \
123 QPEL_FUNCS_DECL(OP, 0, 1, OPT) \
124 QPEL_FUNCS_DECL(OP, 0, 3, OPT) \
125 QPEL_FUNCS_DECL(OP, 1, 0, OPT) \
126 QPEL_FUNCS_DECL(OP, 1, 1, OPT) \
127 QPEL_FUNCS_DECL(OP, 1, 2, OPT) \
128 QPEL_FUNCS_DECL(OP, 1, 3, OPT) \
129 QPEL_FUNCS_DECL(OP, 2, 1, OPT) \
130 QPEL_FUNCS_DECL(OP, 2, 2, OPT) \
131 QPEL_FUNCS_DECL(OP, 2, 3, OPT) \
132 QPEL_FUNCS_DECL(OP, 3, 0, OPT) \
133 QPEL_FUNCS_DECL(OP, 3, 1, OPT) \
134 QPEL_FUNCS_DECL(OP, 3, 2, OPT)
135 /** @} */
136
137 #define LOOPSIZE 8
138 #define HCOFF(x) (32 * ((x) - 1))
139 #define VCOFF(x) (32 * ((x) - 1))
140 QPEL_MC_DECL(put_, _ssse3)
141 QPEL_MC_DECL(avg_, _ssse3)
142
143 #undef LOOPSIZE
144 #undef HCOFF
145 #undef VCOFF
146 #define LOOPSIZE 8
147 #define HCOFF(x) (64 * ((x) - 1))
148 #define VCOFF(x) (64 * ((x) - 1))
149 QPEL_MC_DECL(put_, _sse2)
150 QPEL_MC_DECL(avg_, _sse2)
151
152 #if ARCH_X86_32
153 #undef LOOPSIZE
154 #undef HCOFF
155 #undef VCOFF
156 #define LOOPSIZE 4
157 #define HCOFF(x) (64 * ((x) - 1))
158 #define VCOFF(x) (64 * ((x) - 1))
159
160 QPEL_MC_DECL(put_, _mmx)
161
162 #define ff_put_rv40_qpel_h_mmxext ff_put_rv40_qpel_h_mmx
163 #define ff_put_rv40_qpel_v_mmxext ff_put_rv40_qpel_v_mmx
164 QPEL_MC_DECL(avg_, _mmxext)
165
166 #define ff_put_rv40_qpel_h_3dnow ff_put_rv40_qpel_h_mmx
167 #define ff_put_rv40_qpel_v_3dnow ff_put_rv40_qpel_v_mmx
168 QPEL_MC_DECL(avg_, _3dnow)
169 #endif
170
171 /** @{ */
172 /** Set one function */
173 #define QPEL_FUNC_SET(OP, SIZE, PH, PV, OPT) \
174 c-> OP ## pixels_tab[2 - SIZE / 8][4 * PV + PH] = OP ## rv40_qpel ##SIZE ## _mc ##PH ##PV ##OPT;
175
176 /** Set functions put and avg for sizes 8 and 16 and a given qpel position */
177 #define QPEL_FUNCS_SET(OP, PH, PV, OPT) \
178 QPEL_FUNC_SET(OP, 8, PH, PV, OPT) \
179 QPEL_FUNC_SET(OP, 16, PH, PV, OPT)
180
181 /** Set all functions for all sizes and qpel positions */
182 #define QPEL_MC_SET(OP, OPT) \
183 QPEL_FUNCS_SET (OP, 0, 1, OPT) \
184 QPEL_FUNCS_SET (OP, 0, 3, OPT) \
185 QPEL_FUNCS_SET (OP, 1, 0, OPT) \
186 QPEL_FUNCS_SET (OP, 1, 1, OPT) \
187 QPEL_FUNCS_SET (OP, 1, 2, OPT) \
188 QPEL_FUNCS_SET (OP, 1, 3, OPT) \
189 QPEL_FUNCS_SET (OP, 2, 1, OPT) \
190 QPEL_FUNCS_SET (OP, 2, 2, OPT) \
191 QPEL_FUNCS_SET (OP, 2, 3, OPT) \
192 QPEL_FUNCS_SET (OP, 3, 0, OPT) \
193 QPEL_FUNCS_SET (OP, 3, 1, OPT) \
194 QPEL_FUNCS_SET (OP, 3, 2, OPT)
195 /** @} */
196
197 DEFINE_FN(put, 8, ssse3)
198
199 DEFINE_FN(put, 16, sse2)
200 DEFINE_FN(put, 16, ssse3)
201
202 DEFINE_FN(avg, 8, mmxext)
203 DEFINE_FN(avg, 8, ssse3)
204
205 DEFINE_FN(avg, 16, sse2)
206 DEFINE_FN(avg, 16, ssse3)
207 #endif /* HAVE_X86ASM */
208
209 #if HAVE_MMX_INLINE
210 DEFINE_FN(put, 8, mmx)
211 DEFINE_FN(avg, 8, mmx)
212 DEFINE_FN(put, 16, mmx)
213 DEFINE_FN(avg, 16, mmx)
214 #endif
215
216 av_cold void ff_rv40dsp_init_x86(RV34DSPContext *c)
217 {
218 av_unused int cpu_flags = av_get_cpu_flags();
219
220 #if HAVE_MMX_INLINE
221 if (INLINE_MMX(cpu_flags)) {
222 c->put_pixels_tab[0][15] = put_rv40_qpel16_mc33_mmx;
223 c->put_pixels_tab[1][15] = put_rv40_qpel8_mc33_mmx;
224 c->avg_pixels_tab[0][15] = avg_rv40_qpel16_mc33_mmx;
225 c->avg_pixels_tab[1][15] = avg_rv40_qpel8_mc33_mmx;
226 }
227 #endif /* HAVE_MMX_INLINE */
228
229 #if HAVE_X86ASM
230 if (EXTERNAL_MMX(cpu_flags)) {
231 c->put_chroma_pixels_tab[0] = ff_put_rv40_chroma_mc8_mmx;
232 c->put_chroma_pixels_tab[1] = ff_put_rv40_chroma_mc4_mmx;
233 #if ARCH_X86_32
234 QPEL_MC_SET(put_, _mmx)
235 #endif
236 }
237 if (EXTERNAL_AMD3DNOW(cpu_flags)) {
238 c->avg_chroma_pixels_tab[0] = ff_avg_rv40_chroma_mc8_3dnow;
239 c->avg_chroma_pixels_tab[1] = ff_avg_rv40_chroma_mc4_3dnow;
240 #if ARCH_X86_32
241 QPEL_MC_SET(avg_, _3dnow)
242 #endif
243 }
244 if (EXTERNAL_MMXEXT(cpu_flags)) {
245 c->avg_pixels_tab[1][15] = avg_rv40_qpel8_mc33_mmxext;
246 c->avg_chroma_pixels_tab[0] = ff_avg_rv40_chroma_mc8_mmxext;
247 c->avg_chroma_pixels_tab[1] = ff_avg_rv40_chroma_mc4_mmxext;
248 c->rv40_weight_pixels_tab[0][0] = ff_rv40_weight_func_rnd_16_mmxext;
249 c->rv40_weight_pixels_tab[0][1] = ff_rv40_weight_func_rnd_8_mmxext;
250 c->rv40_weight_pixels_tab[1][0] = ff_rv40_weight_func_nornd_16_mmxext;
251 c->rv40_weight_pixels_tab[1][1] = ff_rv40_weight_func_nornd_8_mmxext;
252 #if ARCH_X86_32
253 QPEL_MC_SET(avg_, _mmxext)
254 #endif
255 }
256 if (EXTERNAL_SSE2(cpu_flags)) {
257 c->put_pixels_tab[0][15] = put_rv40_qpel16_mc33_sse2;
258 c->avg_pixels_tab[0][15] = avg_rv40_qpel16_mc33_sse2;
259 c->rv40_weight_pixels_tab[0][0] = ff_rv40_weight_func_rnd_16_sse2;
260 c->rv40_weight_pixels_tab[0][1] = ff_rv40_weight_func_rnd_8_sse2;
261 c->rv40_weight_pixels_tab[1][0] = ff_rv40_weight_func_nornd_16_sse2;
262 c->rv40_weight_pixels_tab[1][1] = ff_rv40_weight_func_nornd_8_sse2;
263 QPEL_MC_SET(put_, _sse2)
264 QPEL_MC_SET(avg_, _sse2)
265 }
266 if (EXTERNAL_SSSE3(cpu_flags)) {
267 c->put_pixels_tab[0][15] = put_rv40_qpel16_mc33_ssse3;
268 c->put_pixels_tab[1][15] = put_rv40_qpel8_mc33_ssse3;
269 c->avg_pixels_tab[0][15] = avg_rv40_qpel16_mc33_ssse3;
270 c->avg_pixels_tab[1][15] = avg_rv40_qpel8_mc33_ssse3;
271 c->rv40_weight_pixels_tab[0][0] = ff_rv40_weight_func_rnd_16_ssse3;
272 c->rv40_weight_pixels_tab[0][1] = ff_rv40_weight_func_rnd_8_ssse3;
273 c->rv40_weight_pixels_tab[1][0] = ff_rv40_weight_func_nornd_16_ssse3;
274 c->rv40_weight_pixels_tab[1][1] = ff_rv40_weight_func_nornd_8_ssse3;
275 QPEL_MC_SET(put_, _ssse3)
276 QPEL_MC_SET(avg_, _ssse3)
277 }
278 #endif /* HAVE_X86ASM */
279 }
280