87fa6af3f9e0f34d21e2959a71443ab2b4791002
[libav.git] / libavcodec / ppc / h264_altivec.c
1 /*
2 * Copyright (c) 2004 Romain Dolbeau <romain@dolbeau.org>
3 *
4 * This library is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU Lesser General Public
6 * License as published by the Free Software Foundation; either
7 * version 2 of the License, or (at your option) any later version.
8 *
9 * This library is distributed in the hope that it will be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
12 * Lesser General Public License for more details.
13 *
14 * You should have received a copy of the GNU Lesser General Public
15 * License along with this library; if not, write to the Free Software
16 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
17 */
18
19 #include "../dsputil.h"
20
21 #include "gcc_fixes.h"
22
23 #include "dsputil_altivec.h"
24
25 #define PUT_OP_U8_ALTIVEC(d, s, dst) d = s
26 #define AVG_OP_U8_ALTIVEC(d, s, dst) d = vec_avg(dst, s)
27
28 #define OP_U8_ALTIVEC PUT_OP_U8_ALTIVEC
29 #define PREFIX_h264_chroma_mc8_altivec put_h264_chroma_mc8_altivec
30 #define PREFIX_h264_chroma_mc8_num altivec_put_h264_chroma_mc8_num
31 #define PREFIX_h264_qpel16_h_lowpass_altivec put_h264_qpel16_h_lowpass_altivec
32 #define PREFIX_h264_qpel16_h_lowpass_num altivec_put_h264_qpel16_h_lowpass_num
33 #define PREFIX_h264_qpel16_v_lowpass_altivec put_h264_qpel16_v_lowpass_altivec
34 #define PREFIX_h264_qpel16_v_lowpass_num altivec_put_h264_qpel16_v_lowpass_num
35 #define PREFIX_h264_qpel16_hv_lowpass_altivec put_h264_qpel16_hv_lowpass_altivec
36 #define PREFIX_h264_qpel16_hv_lowpass_num altivec_put_h264_qpel16_hv_lowpass_num
37 #include "h264_template_altivec.c"
38 #undef OP_U8_ALTIVEC
39 #undef PREFIX_h264_chroma_mc8_altivec
40 #undef PREFIX_h264_chroma_mc8_num
41 #undef PREFIX_h264_qpel16_h_lowpass_altivec
42 #undef PREFIX_h264_qpel16_h_lowpass_num
43 #undef PREFIX_h264_qpel16_v_lowpass_altivec
44 #undef PREFIX_h264_qpel16_v_lowpass_num
45 #undef PREFIX_h264_qpel16_hv_lowpass_altivec
46 #undef PREFIX_h264_qpel16_hv_lowpass_num
47
48 #define OP_U8_ALTIVEC AVG_OP_U8_ALTIVEC
49 #define PREFIX_h264_chroma_mc8_altivec avg_h264_chroma_mc8_altivec
50 #define PREFIX_h264_chroma_mc8_num altivec_avg_h264_chroma_mc8_num
51 #define PREFIX_h264_qpel16_h_lowpass_altivec avg_h264_qpel16_h_lowpass_altivec
52 #define PREFIX_h264_qpel16_h_lowpass_num altivec_avg_h264_qpel16_h_lowpass_num
53 #define PREFIX_h264_qpel16_v_lowpass_altivec avg_h264_qpel16_v_lowpass_altivec
54 #define PREFIX_h264_qpel16_v_lowpass_num altivec_avg_h264_qpel16_v_lowpass_num
55 #define PREFIX_h264_qpel16_hv_lowpass_altivec avg_h264_qpel16_hv_lowpass_altivec
56 #define PREFIX_h264_qpel16_hv_lowpass_num altivec_avg_h264_qpel16_hv_lowpass_num
57 #include "h264_template_altivec.c"
58 #undef OP_U8_ALTIVEC
59 #undef PREFIX_h264_chroma_mc8_altivec
60 #undef PREFIX_h264_chroma_mc8_num
61 #undef PREFIX_h264_qpel16_h_lowpass_altivec
62 #undef PREFIX_h264_qpel16_h_lowpass_num
63 #undef PREFIX_h264_qpel16_v_lowpass_altivec
64 #undef PREFIX_h264_qpel16_v_lowpass_num
65 #undef PREFIX_h264_qpel16_hv_lowpass_altivec
66 #undef PREFIX_h264_qpel16_hv_lowpass_num
67
68 #define H264_MC(OPNAME, SIZE, CODETYPE) \
69 static void OPNAME ## h264_qpel ## SIZE ## _mc00_ ## CODETYPE (uint8_t *dst, uint8_t *src, int stride){\
70 OPNAME ## pixels ## SIZE ## _ ## CODETYPE(dst, src, stride, SIZE);\
71 }\
72 \
73 static void OPNAME ## h264_qpel ## SIZE ## _mc10_ ## CODETYPE(uint8_t *dst, uint8_t *src, int stride){ \
74 DECLARE_ALIGNED_16(uint8_t, half[SIZE*SIZE]);\
75 put_h264_qpel ## SIZE ## _h_lowpass_ ## CODETYPE(half, src, SIZE, stride);\
76 OPNAME ## pixels ## SIZE ## _l2_ ## CODETYPE(dst, src, half, stride, stride, SIZE);\
77 }\
78 \
79 static void OPNAME ## h264_qpel ## SIZE ## _mc20_ ## CODETYPE(uint8_t *dst, uint8_t *src, int stride){\
80 OPNAME ## h264_qpel ## SIZE ## _h_lowpass_ ## CODETYPE(dst, src, stride, stride);\
81 }\
82 \
83 static void OPNAME ## h264_qpel ## SIZE ## _mc30_ ## CODETYPE(uint8_t *dst, uint8_t *src, int stride){\
84 DECLARE_ALIGNED_16(uint8_t, half[SIZE*SIZE]);\
85 put_h264_qpel ## SIZE ## _h_lowpass_ ## CODETYPE(half, src, SIZE, stride);\
86 OPNAME ## pixels ## SIZE ## _l2_ ## CODETYPE(dst, src+1, half, stride, stride, SIZE);\
87 }\
88 \
89 static void OPNAME ## h264_qpel ## SIZE ## _mc01_ ## CODETYPE(uint8_t *dst, uint8_t *src, int stride){\
90 DECLARE_ALIGNED_16(uint8_t, half[SIZE*SIZE]);\
91 put_h264_qpel ## SIZE ## _v_lowpass_ ## CODETYPE(half, src, SIZE, stride);\
92 OPNAME ## pixels ## SIZE ## _l2_ ## CODETYPE(dst, src, half, stride, stride, SIZE);\
93 }\
94 \
95 static void OPNAME ## h264_qpel ## SIZE ## _mc02_ ## CODETYPE(uint8_t *dst, uint8_t *src, int stride){\
96 OPNAME ## h264_qpel ## SIZE ## _v_lowpass_ ## CODETYPE(dst, src, stride, stride);\
97 }\
98 \
99 static void OPNAME ## h264_qpel ## SIZE ## _mc03_ ## CODETYPE(uint8_t *dst, uint8_t *src, int stride){\
100 DECLARE_ALIGNED_16(uint8_t, half[SIZE*SIZE]);\
101 put_h264_qpel ## SIZE ## _v_lowpass_ ## CODETYPE(half, src, SIZE, stride);\
102 OPNAME ## pixels ## SIZE ## _l2_ ## CODETYPE(dst, src+stride, half, stride, stride, SIZE);\
103 }\
104 \
105 static void OPNAME ## h264_qpel ## SIZE ## _mc11_ ## CODETYPE(uint8_t *dst, uint8_t *src, int stride){\
106 DECLARE_ALIGNED_16(uint8_t, halfH[SIZE*SIZE]);\
107 DECLARE_ALIGNED_16(uint8_t, halfV[SIZE*SIZE]);\
108 put_h264_qpel ## SIZE ## _h_lowpass_ ## CODETYPE(halfH, src, SIZE, stride);\
109 put_h264_qpel ## SIZE ## _v_lowpass_ ## CODETYPE(halfV, src, SIZE, stride);\
110 OPNAME ## pixels ## SIZE ## _l2_ ## CODETYPE(dst, halfH, halfV, stride, SIZE, SIZE);\
111 }\
112 \
113 static void OPNAME ## h264_qpel ## SIZE ## _mc31_ ## CODETYPE(uint8_t *dst, uint8_t *src, int stride){\
114 DECLARE_ALIGNED_16(uint8_t, halfH[SIZE*SIZE]);\
115 DECLARE_ALIGNED_16(uint8_t, halfV[SIZE*SIZE]);\
116 put_h264_qpel ## SIZE ## _h_lowpass_ ## CODETYPE(halfH, src, SIZE, stride);\
117 put_h264_qpel ## SIZE ## _v_lowpass_ ## CODETYPE(halfV, src+1, SIZE, stride);\
118 OPNAME ## pixels ## SIZE ## _l2_ ## CODETYPE(dst, halfH, halfV, stride, SIZE, SIZE);\
119 }\
120 \
121 static void OPNAME ## h264_qpel ## SIZE ## _mc13_ ## CODETYPE(uint8_t *dst, uint8_t *src, int stride){\
122 DECLARE_ALIGNED_16(uint8_t, halfH[SIZE*SIZE]);\
123 DECLARE_ALIGNED_16(uint8_t, halfV[SIZE*SIZE]);\
124 put_h264_qpel ## SIZE ## _h_lowpass_ ## CODETYPE(halfH, src + stride, SIZE, stride);\
125 put_h264_qpel ## SIZE ## _v_lowpass_ ## CODETYPE(halfV, src, SIZE, stride);\
126 OPNAME ## pixels ## SIZE ## _l2_ ## CODETYPE(dst, halfH, halfV, stride, SIZE, SIZE);\
127 }\
128 \
129 static void OPNAME ## h264_qpel ## SIZE ## _mc33_ ## CODETYPE(uint8_t *dst, uint8_t *src, int stride){\
130 DECLARE_ALIGNED_16(uint8_t, halfH[SIZE*SIZE]);\
131 DECLARE_ALIGNED_16(uint8_t, halfV[SIZE*SIZE]);\
132 put_h264_qpel ## SIZE ## _h_lowpass_ ## CODETYPE(halfH, src + stride, SIZE, stride);\
133 put_h264_qpel ## SIZE ## _v_lowpass_ ## CODETYPE(halfV, src+1, SIZE, stride);\
134 OPNAME ## pixels ## SIZE ## _l2_ ## CODETYPE(dst, halfH, halfV, stride, SIZE, SIZE);\
135 }\
136 \
137 static void OPNAME ## h264_qpel ## SIZE ## _mc22_ ## CODETYPE(uint8_t *dst, uint8_t *src, int stride){\
138 DECLARE_ALIGNED_16(int16_t, tmp[SIZE*(SIZE+8)]);\
139 OPNAME ## h264_qpel ## SIZE ## _hv_lowpass_ ## CODETYPE(dst, tmp, src, stride, SIZE, stride);\
140 }\
141 \
142 static void OPNAME ## h264_qpel ## SIZE ## _mc21_ ## CODETYPE(uint8_t *dst, uint8_t *src, int stride){\
143 DECLARE_ALIGNED_16(uint8_t, halfH[SIZE*SIZE]);\
144 DECLARE_ALIGNED_16(uint8_t, halfHV[SIZE*SIZE]);\
145 DECLARE_ALIGNED_16(int16_t, tmp[SIZE*(SIZE+8)]);\
146 put_h264_qpel ## SIZE ## _h_lowpass_ ## CODETYPE(halfH, src, SIZE, stride);\
147 put_h264_qpel ## SIZE ## _hv_lowpass_ ## CODETYPE(halfHV, tmp, src, SIZE, SIZE, stride);\
148 OPNAME ## pixels ## SIZE ## _l2_ ## CODETYPE(dst, halfH, halfHV, stride, SIZE, SIZE);\
149 }\
150 \
151 static void OPNAME ## h264_qpel ## SIZE ## _mc23_ ## CODETYPE(uint8_t *dst, uint8_t *src, int stride){\
152 DECLARE_ALIGNED_16(uint8_t, halfH[SIZE*SIZE]);\
153 DECLARE_ALIGNED_16(uint8_t, halfHV[SIZE*SIZE]);\
154 DECLARE_ALIGNED_16(int16_t, tmp[SIZE*(SIZE+8)]);\
155 put_h264_qpel ## SIZE ## _h_lowpass_ ## CODETYPE(halfH, src + stride, SIZE, stride);\
156 put_h264_qpel ## SIZE ## _hv_lowpass_ ## CODETYPE(halfHV, tmp, src, SIZE, SIZE, stride);\
157 OPNAME ## pixels ## SIZE ## _l2_ ## CODETYPE(dst, halfH, halfHV, stride, SIZE, SIZE);\
158 }\
159 \
160 static void OPNAME ## h264_qpel ## SIZE ## _mc12_ ## CODETYPE(uint8_t *dst, uint8_t *src, int stride){\
161 DECLARE_ALIGNED_16(uint8_t, halfV[SIZE*SIZE]);\
162 DECLARE_ALIGNED_16(uint8_t, halfHV[SIZE*SIZE]);\
163 DECLARE_ALIGNED_16(int16_t, tmp[SIZE*(SIZE+8)]);\
164 put_h264_qpel ## SIZE ## _v_lowpass_ ## CODETYPE(halfV, src, SIZE, stride);\
165 put_h264_qpel ## SIZE ## _hv_lowpass_ ## CODETYPE(halfHV, tmp, src, SIZE, SIZE, stride);\
166 OPNAME ## pixels ## SIZE ## _l2_ ## CODETYPE(dst, halfV, halfHV, stride, SIZE, SIZE);\
167 }\
168 \
169 static void OPNAME ## h264_qpel ## SIZE ## _mc32_ ## CODETYPE(uint8_t *dst, uint8_t *src, int stride){\
170 DECLARE_ALIGNED_16(uint8_t, halfV[SIZE*SIZE]);\
171 DECLARE_ALIGNED_16(uint8_t, halfHV[SIZE*SIZE]);\
172 DECLARE_ALIGNED_16(int16_t, tmp[SIZE*(SIZE+8)]);\
173 put_h264_qpel ## SIZE ## _v_lowpass_ ## CODETYPE(halfV, src+1, SIZE, stride);\
174 put_h264_qpel ## SIZE ## _hv_lowpass_ ## CODETYPE(halfHV, tmp, src, SIZE, SIZE, stride);\
175 OPNAME ## pixels ## SIZE ## _l2_ ## CODETYPE(dst, halfV, halfHV, stride, SIZE, SIZE);\
176 }\
177
178 static inline void put_pixels16_l2_altivec( uint8_t * dst, const uint8_t * src1,
179 const uint8_t * src2, int dst_stride,
180 int src_stride1, int h)
181 {
182 int i;
183 vector unsigned char a, b, d, tmp1, tmp2, mask, mask_, edges, align;
184
185 mask_ = vec_lvsl(0, src2);
186
187 for (i = 0; i < h; i++) {
188
189 tmp1 = vec_ld(i * src_stride1, src1);
190 mask = vec_lvsl(i * src_stride1, src1);
191 tmp2 = vec_ld(i * src_stride1 + 15, src1);
192
193 a = vec_perm(tmp1, tmp2, mask);
194
195 tmp1 = vec_ld(i * 16, src2);
196 tmp2 = vec_ld(i * 16 + 15, src2);
197
198 b = vec_perm(tmp1, tmp2, mask_);
199
200 tmp1 = vec_ld(0, dst);
201 mask = vec_lvsl(0, dst);
202 tmp2 = vec_ld(15, dst);
203
204 d = vec_avg(a, b);
205
206 edges = vec_perm(tmp2, tmp1, mask);
207
208 align = vec_lvsr(0, dst);
209
210 tmp1 = vec_perm(edges, d, align);
211 tmp2 = vec_perm(d, edges, align);
212
213 vec_st(tmp2, 15, dst);
214 vec_st(tmp1, 0 , dst);
215
216 dst += dst_stride;
217 }
218 }
219
220 static inline void avg_pixels16_l2_altivec( uint8_t * dst, const uint8_t * src1,
221 const uint8_t * src2, int dst_stride,
222 int src_stride1, int h)
223 {
224 int i;
225 vector unsigned char a, b, d, tmp1, tmp2, mask, mask_, edges, align;
226
227 mask_ = vec_lvsl(0, src2);
228
229 for (i = 0; i < h; i++) {
230
231 tmp1 = vec_ld(i * src_stride1, src1);
232 mask = vec_lvsl(i * src_stride1, src1);
233 tmp2 = vec_ld(i * src_stride1 + 15, src1);
234
235 a = vec_perm(tmp1, tmp2, mask);
236
237 tmp1 = vec_ld(i * 16, src2);
238 tmp2 = vec_ld(i * 16 + 15, src2);
239
240 b = vec_perm(tmp1, tmp2, mask_);
241
242 tmp1 = vec_ld(0, dst);
243 mask = vec_lvsl(0, dst);
244 tmp2 = vec_ld(15, dst);
245
246 d = vec_avg(vec_perm(tmp1, tmp2, mask), vec_avg(a, b));
247
248 edges = vec_perm(tmp2, tmp1, mask);
249
250 align = vec_lvsr(0, dst);
251
252 tmp1 = vec_perm(edges, d, align);
253 tmp2 = vec_perm(d, edges, align);
254
255 vec_st(tmp2, 15, dst);
256 vec_st(tmp1, 0 , dst);
257
258 dst += dst_stride;
259 }
260 }
261
262 /* Implemented but could be faster
263 #define put_pixels16_l2_altivec(d,s1,s2,ds,s1s,h) put_pixels16_l2(d,s1,s2,ds,s1s,16,h)
264 #define avg_pixels16_l2_altivec(d,s1,s2,ds,s1s,h) avg_pixels16_l2(d,s1,s2,ds,s1s,16,h)
265 */
266
267 H264_MC(put_, 16, altivec)
268 H264_MC(avg_, 16, altivec)
269
270 void dsputil_h264_init_ppc(DSPContext* c, AVCodecContext *avctx) {
271
272 #ifdef HAVE_ALTIVEC
273 if (has_altivec()) {
274 c->put_h264_chroma_pixels_tab[0] = put_h264_chroma_mc8_altivec;
275 c->avg_h264_chroma_pixels_tab[0] = avg_h264_chroma_mc8_altivec;
276
277 #define dspfunc(PFX, IDX, NUM) \
278 c->PFX ## _pixels_tab[IDX][ 0] = PFX ## NUM ## _mc00_altivec; \
279 c->PFX ## _pixels_tab[IDX][ 1] = PFX ## NUM ## _mc10_altivec; \
280 c->PFX ## _pixels_tab[IDX][ 2] = PFX ## NUM ## _mc20_altivec; \
281 c->PFX ## _pixels_tab[IDX][ 3] = PFX ## NUM ## _mc30_altivec; \
282 c->PFX ## _pixels_tab[IDX][ 4] = PFX ## NUM ## _mc01_altivec; \
283 c->PFX ## _pixels_tab[IDX][ 5] = PFX ## NUM ## _mc11_altivec; \
284 c->PFX ## _pixels_tab[IDX][ 6] = PFX ## NUM ## _mc21_altivec; \
285 c->PFX ## _pixels_tab[IDX][ 7] = PFX ## NUM ## _mc31_altivec; \
286 c->PFX ## _pixels_tab[IDX][ 8] = PFX ## NUM ## _mc02_altivec; \
287 c->PFX ## _pixels_tab[IDX][ 9] = PFX ## NUM ## _mc12_altivec; \
288 c->PFX ## _pixels_tab[IDX][10] = PFX ## NUM ## _mc22_altivec; \
289 c->PFX ## _pixels_tab[IDX][11] = PFX ## NUM ## _mc32_altivec; \
290 c->PFX ## _pixels_tab[IDX][12] = PFX ## NUM ## _mc03_altivec; \
291 c->PFX ## _pixels_tab[IDX][13] = PFX ## NUM ## _mc13_altivec; \
292 c->PFX ## _pixels_tab[IDX][14] = PFX ## NUM ## _mc23_altivec; \
293 c->PFX ## _pixels_tab[IDX][15] = PFX ## NUM ## _mc33_altivec
294
295 dspfunc(put_h264_qpel, 0, 16);
296 dspfunc(avg_h264_qpel, 0, 16);
297 #undef dspfunc
298
299 } else
300 #endif /* HAVE_ALTIVEC */
301 {
302 // Non-AltiVec PPC optimisations
303
304 // ... pending ...
305 }
306 }