1 /*
2 * Copyright (c) 2004 Romain Dolbeau <romain@dolbeau.org>
3 *
4 * This file is part of FFmpeg.
5 *
6 * FFmpeg is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU Lesser General Public
8 * License as published by the Free Software Foundation; either
9 * version 2.1 of the License, or (at your option) any later version.
10 *
11 * FFmpeg is distributed in the hope that it will be useful,
12 * but WITHOUT ANY WARRANTY; without even the implied warranty of
13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
14 * Lesser General Public License for more details.
15 *
16 * You should have received a copy of the GNU Lesser General Public
17 * License along with FFmpeg; if not, write to the Free Software
18 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
19 */
20
21 #include "config.h"
22
23 #include "libavutil/attributes.h"
24 #include "libavutil/cpu.h"
25 #include "libavutil/intreadwrite.h"
26 #include "libavutil/mem_internal.h"
27 #include "libavutil/ppc/cpu.h"
28 #include "libavutil/ppc/util_altivec.h"
29
30 #include "libavcodec/h264qpel.h"
31
32 #include "hpeldsp_altivec.h"
33
34 #if HAVE_ALTIVEC
35
36 #define PUT_OP_U8_ALTIVEC(d, s, dst) d = s
37 #define AVG_OP_U8_ALTIVEC(d, s, dst) d = vec_avg(dst, s)
38
39 #define OP_U8_ALTIVEC PUT_OP_U8_ALTIVEC
40 #define PREFIX_h264_qpel16_h_lowpass_altivec put_h264_qpel16_h_lowpass_altivec
41 #define PREFIX_h264_qpel16_h_lowpass_num altivec_put_h264_qpel16_h_lowpass_num
42 #define PREFIX_h264_qpel16_v_lowpass_altivec put_h264_qpel16_v_lowpass_altivec
43 #define PREFIX_h264_qpel16_v_lowpass_num altivec_put_h264_qpel16_v_lowpass_num
44 #define PREFIX_h264_qpel16_hv_lowpass_altivec put_h264_qpel16_hv_lowpass_altivec
45 #define PREFIX_h264_qpel16_hv_lowpass_num altivec_put_h264_qpel16_hv_lowpass_num
46 #include "h264qpel_template.c"
47 #undef OP_U8_ALTIVEC
48 #undef PREFIX_h264_qpel16_h_lowpass_altivec
49 #undef PREFIX_h264_qpel16_h_lowpass_num
50 #undef PREFIX_h264_qpel16_v_lowpass_altivec
51 #undef PREFIX_h264_qpel16_v_lowpass_num
52 #undef PREFIX_h264_qpel16_hv_lowpass_altivec
53 #undef PREFIX_h264_qpel16_hv_lowpass_num
54
55 #define OP_U8_ALTIVEC AVG_OP_U8_ALTIVEC
56 #define PREFIX_h264_qpel16_h_lowpass_altivec avg_h264_qpel16_h_lowpass_altivec
57 #define PREFIX_h264_qpel16_h_lowpass_num altivec_avg_h264_qpel16_h_lowpass_num
58 #define PREFIX_h264_qpel16_v_lowpass_altivec avg_h264_qpel16_v_lowpass_altivec
59 #define PREFIX_h264_qpel16_v_lowpass_num altivec_avg_h264_qpel16_v_lowpass_num
60 #define PREFIX_h264_qpel16_hv_lowpass_altivec avg_h264_qpel16_hv_lowpass_altivec
61 #define PREFIX_h264_qpel16_hv_lowpass_num altivec_avg_h264_qpel16_hv_lowpass_num
62 #include "h264qpel_template.c"
63 #undef OP_U8_ALTIVEC
64 #undef PREFIX_h264_qpel16_h_lowpass_altivec
65 #undef PREFIX_h264_qpel16_h_lowpass_num
66 #undef PREFIX_h264_qpel16_v_lowpass_altivec
67 #undef PREFIX_h264_qpel16_v_lowpass_num
68 #undef PREFIX_h264_qpel16_hv_lowpass_altivec
69 #undef PREFIX_h264_qpel16_hv_lowpass_num
70
71 #define H264_MC(OPNAME, SIZE, CODETYPE) \
72 static void OPNAME ## h264_qpel ## SIZE ## _mc00_ ## CODETYPE (uint8_t *dst, const uint8_t *src, ptrdiff_t stride)\
73 {\
74 ff_ ## OPNAME ## pixels ## SIZE ## _ ## CODETYPE(dst, src, stride, SIZE);\
75 }\
76 \
77 static void OPNAME ## h264_qpel ## SIZE ## _mc10_ ## CODETYPE(uint8_t *dst, const uint8_t *src, ptrdiff_t stride)\
78 { \
79 DECLARE_ALIGNED(16, uint8_t, half)[SIZE*SIZE];\
80 put_h264_qpel ## SIZE ## _h_lowpass_ ## CODETYPE(half, src, SIZE, stride);\
81 OPNAME ## pixels ## SIZE ## _l2_ ## CODETYPE(dst, src, half, stride, stride, SIZE);\
82 }\
83 \
84 static void OPNAME ## h264_qpel ## SIZE ## _mc20_ ## CODETYPE(uint8_t *dst, const uint8_t *src, ptrdiff_t stride)\
85 {\
86 OPNAME ## h264_qpel ## SIZE ## _h_lowpass_ ## CODETYPE(dst, src, stride, stride);\
87 }\
88 \
89 static void OPNAME ## h264_qpel ## SIZE ## _mc30_ ## CODETYPE(uint8_t *dst, const uint8_t *src, ptrdiff_t stride)\
90 {\
91 DECLARE_ALIGNED(16, uint8_t, half)[SIZE*SIZE];\
92 put_h264_qpel ## SIZE ## _h_lowpass_ ## CODETYPE(half, src, SIZE, stride);\
93 OPNAME ## pixels ## SIZE ## _l2_ ## CODETYPE(dst, src+1, half, stride, stride, SIZE);\
94 }\
95 \
96 static void OPNAME ## h264_qpel ## SIZE ## _mc01_ ## CODETYPE(uint8_t *dst, const uint8_t *src, ptrdiff_t stride)\
97 {\
98 DECLARE_ALIGNED(16, uint8_t, half)[SIZE*SIZE];\
99 put_h264_qpel ## SIZE ## _v_lowpass_ ## CODETYPE(half, src, SIZE, stride);\
100 OPNAME ## pixels ## SIZE ## _l2_ ## CODETYPE(dst, src, half, stride, stride, SIZE);\
101 }\
102 \
103 static void OPNAME ## h264_qpel ## SIZE ## _mc02_ ## CODETYPE(uint8_t *dst, const uint8_t *src, ptrdiff_t stride)\
104 {\
105 OPNAME ## h264_qpel ## SIZE ## _v_lowpass_ ## CODETYPE(dst, src, stride, stride);\
106 }\
107 \
108 static void OPNAME ## h264_qpel ## SIZE ## _mc03_ ## CODETYPE(uint8_t *dst, const uint8_t *src, ptrdiff_t stride)\
109 {\
110 DECLARE_ALIGNED(16, uint8_t, half)[SIZE*SIZE];\
111 put_h264_qpel ## SIZE ## _v_lowpass_ ## CODETYPE(half, src, SIZE, stride);\
112 OPNAME ## pixels ## SIZE ## _l2_ ## CODETYPE(dst, src+stride, half, stride, stride, SIZE);\
113 }\
114 \
115 static void OPNAME ## h264_qpel ## SIZE ## _mc11_ ## CODETYPE(uint8_t *dst, const uint8_t *src, ptrdiff_t stride)\
116 {\
117 DECLARE_ALIGNED(16, uint8_t, halfH)[SIZE*SIZE];\
118 DECLARE_ALIGNED(16, uint8_t, halfV)[SIZE*SIZE];\
119 put_h264_qpel ## SIZE ## _h_lowpass_ ## CODETYPE(halfH, src, SIZE, stride);\
120 put_h264_qpel ## SIZE ## _v_lowpass_ ## CODETYPE(halfV, src, SIZE, stride);\
121 OPNAME ## pixels ## SIZE ## _l2_ ## CODETYPE(dst, halfH, halfV, stride, SIZE, SIZE);\
122 }\
123 \
124 static void OPNAME ## h264_qpel ## SIZE ## _mc31_ ## CODETYPE(uint8_t *dst, const uint8_t *src, ptrdiff_t stride)\
125 {\
126 DECLARE_ALIGNED(16, uint8_t, halfH)[SIZE*SIZE];\
127 DECLARE_ALIGNED(16, uint8_t, halfV)[SIZE*SIZE];\
128 put_h264_qpel ## SIZE ## _h_lowpass_ ## CODETYPE(halfH, src, SIZE, stride);\
129 put_h264_qpel ## SIZE ## _v_lowpass_ ## CODETYPE(halfV, src+1, SIZE, stride);\
130 OPNAME ## pixels ## SIZE ## _l2_ ## CODETYPE(dst, halfH, halfV, stride, SIZE, SIZE);\
131 }\
132 \
133 static void OPNAME ## h264_qpel ## SIZE ## _mc13_ ## CODETYPE(uint8_t *dst, const uint8_t *src, ptrdiff_t stride)\
134 {\
135 DECLARE_ALIGNED(16, uint8_t, halfH)[SIZE*SIZE];\
136 DECLARE_ALIGNED(16, uint8_t, halfV)[SIZE*SIZE];\
137 put_h264_qpel ## SIZE ## _h_lowpass_ ## CODETYPE(halfH, src + stride, SIZE, stride);\
138 put_h264_qpel ## SIZE ## _v_lowpass_ ## CODETYPE(halfV, src, SIZE, stride);\
139 OPNAME ## pixels ## SIZE ## _l2_ ## CODETYPE(dst, halfH, halfV, stride, SIZE, SIZE);\
140 }\
141 \
142 static void OPNAME ## h264_qpel ## SIZE ## _mc33_ ## CODETYPE(uint8_t *dst, const uint8_t *src, ptrdiff_t stride)\
143 {\
144 DECLARE_ALIGNED(16, uint8_t, halfH)[SIZE*SIZE];\
145 DECLARE_ALIGNED(16, uint8_t, halfV)[SIZE*SIZE];\
146 put_h264_qpel ## SIZE ## _h_lowpass_ ## CODETYPE(halfH, src + stride, SIZE, stride);\
147 put_h264_qpel ## SIZE ## _v_lowpass_ ## CODETYPE(halfV, src+1, SIZE, stride);\
148 OPNAME ## pixels ## SIZE ## _l2_ ## CODETYPE(dst, halfH, halfV, stride, SIZE, SIZE);\
149 }\
150 \
151 static void OPNAME ## h264_qpel ## SIZE ## _mc22_ ## CODETYPE(uint8_t *dst, const uint8_t *src, ptrdiff_t stride)\
152 {\
153 DECLARE_ALIGNED(16, int16_t, tmp)[SIZE*(SIZE+8)];\
154 OPNAME ## h264_qpel ## SIZE ## _hv_lowpass_ ## CODETYPE(dst, tmp, src, stride, SIZE, stride);\
155 }\
156 \
157 static void OPNAME ## h264_qpel ## SIZE ## _mc21_ ## CODETYPE(uint8_t *dst, const uint8_t *src, ptrdiff_t stride)\
158 {\
159 DECLARE_ALIGNED(16, uint8_t, halfH)[SIZE*SIZE];\
160 DECLARE_ALIGNED(16, uint8_t, halfHV)[SIZE*SIZE];\
161 DECLARE_ALIGNED(16, int16_t, tmp)[SIZE*(SIZE+8)];\
162 put_h264_qpel ## SIZE ## _h_lowpass_ ## CODETYPE(halfH, src, SIZE, stride);\
163 put_h264_qpel ## SIZE ## _hv_lowpass_ ## CODETYPE(halfHV, tmp, src, SIZE, SIZE, stride);\
164 OPNAME ## pixels ## SIZE ## _l2_ ## CODETYPE(dst, halfH, halfHV, stride, SIZE, SIZE);\
165 }\
166 \
167 static void OPNAME ## h264_qpel ## SIZE ## _mc23_ ## CODETYPE(uint8_t *dst, const uint8_t *src, ptrdiff_t stride)\
168 {\
169 DECLARE_ALIGNED(16, uint8_t, halfH)[SIZE*SIZE];\
170 DECLARE_ALIGNED(16, uint8_t, halfHV)[SIZE*SIZE];\
171 DECLARE_ALIGNED(16, int16_t, tmp)[SIZE*(SIZE+8)];\
172 put_h264_qpel ## SIZE ## _h_lowpass_ ## CODETYPE(halfH, src + stride, SIZE, stride);\
173 put_h264_qpel ## SIZE ## _hv_lowpass_ ## CODETYPE(halfHV, tmp, src, SIZE, SIZE, stride);\
174 OPNAME ## pixels ## SIZE ## _l2_ ## CODETYPE(dst, halfH, halfHV, stride, SIZE, SIZE);\
175 }\
176 \
177 static void OPNAME ## h264_qpel ## SIZE ## _mc12_ ## CODETYPE(uint8_t *dst, const uint8_t *src, ptrdiff_t stride)\
178 {\
179 DECLARE_ALIGNED(16, uint8_t, halfV)[SIZE*SIZE];\
180 DECLARE_ALIGNED(16, uint8_t, halfHV)[SIZE*SIZE];\
181 DECLARE_ALIGNED(16, int16_t, tmp)[SIZE*(SIZE+8)];\
182 put_h264_qpel ## SIZE ## _v_lowpass_ ## CODETYPE(halfV, src, SIZE, stride);\
183 put_h264_qpel ## SIZE ## _hv_lowpass_ ## CODETYPE(halfHV, tmp, src, SIZE, SIZE, stride);\
184 OPNAME ## pixels ## SIZE ## _l2_ ## CODETYPE(dst, halfV, halfHV, stride, SIZE, SIZE);\
185 }\
186 \
187 static void OPNAME ## h264_qpel ## SIZE ## _mc32_ ## CODETYPE(uint8_t *dst, const uint8_t *src, ptrdiff_t stride)\
188 {\
189 DECLARE_ALIGNED(16, uint8_t, halfV)[SIZE*SIZE];\
190 DECLARE_ALIGNED(16, uint8_t, halfHV)[SIZE*SIZE];\
191 DECLARE_ALIGNED(16, int16_t, tmp)[SIZE*(SIZE+8)];\
192 put_h264_qpel ## SIZE ## _v_lowpass_ ## CODETYPE(halfV, src+1, SIZE, stride);\
193 put_h264_qpel ## SIZE ## _hv_lowpass_ ## CODETYPE(halfHV, tmp, src, SIZE, SIZE, stride);\
194 OPNAME ## pixels ## SIZE ## _l2_ ## CODETYPE(dst, halfV, halfHV, stride, SIZE, SIZE);\
195 }\
196
197 #if HAVE_BIGENDIAN
198 #define put_unligned_store(s, dest) { \
199 tmp1 = vec_ld(0, dest); \
200 mask = vec_lvsl(0, dest); \
201 tmp2 = vec_ld(15, dest); \
202 edges = vec_perm(tmp2, tmp1, mask); \
203 align = vec_lvsr(0, dest); \
204 tmp2 = vec_perm(s, edges, align); \
205 tmp1 = vec_perm(edges, s, align); \
206 vec_st(tmp2, 15, dest); \
207 vec_st(tmp1, 0 , dest); \
208 }
209 #else
210 #define put_unligned_store(s, dest) vec_vsx_st(s, 0, dest);
211 #endif /* HAVE_BIGENDIAN */
212
put_pixels16_l2_altivec(uint8_t * dst,const uint8_t * src1,const uint8_t * src2,int dst_stride,int src_stride1,int h)213 static inline void put_pixels16_l2_altivec( uint8_t * dst, const uint8_t * src1,
214 const uint8_t * src2, int dst_stride,
215 int src_stride1, int h)
216 {
217 int i;
218 vec_u8 a, b, d, mask_;
219 #if HAVE_BIGENDIAN
220 vec_u8 tmp1, tmp2, mask, edges, align;
221 mask_ = vec_lvsl(0, src2);
222 #endif
223
224 for (i = 0; i < h; i++) {
225 a = unaligned_load(i * src_stride1, src1);
226 b = load_with_perm_vec(i * 16, src2, mask_);
227 d = vec_avg(a, b);
228 put_unligned_store(d, dst);
229 dst += dst_stride;
230 }
231 }
232
233 #if HAVE_BIGENDIAN
234 #define avg_unligned_store(s, dest){ \
235 tmp1 = vec_ld(0, dest); \
236 mask = vec_lvsl(0, dest); \
237 tmp2 = vec_ld(15, dest); \
238 a = vec_avg(vec_perm(tmp1, tmp2, mask), s); \
239 edges = vec_perm(tmp2, tmp1, mask); \
240 align = vec_lvsr(0, dest); \
241 tmp2 = vec_perm(a, edges, align); \
242 tmp1 = vec_perm(edges, a, align); \
243 vec_st(tmp2, 15, dest); \
244 vec_st(tmp1, 0 , dest); \
245 }
246 #else
247 #define avg_unligned_store(s, dest){ \
248 a = vec_avg(vec_vsx_ld(0, dst), s); \
249 vec_vsx_st(a, 0, dst); \
250 }
251 #endif /* HAVE_BIGENDIAN */
252
avg_pixels16_l2_altivec(uint8_t * dst,const uint8_t * src1,const uint8_t * src2,int dst_stride,int src_stride1,int h)253 static inline void avg_pixels16_l2_altivec( uint8_t * dst, const uint8_t * src1,
254 const uint8_t * src2, int dst_stride,
255 int src_stride1, int h)
256 {
257 int i;
258 vec_u8 a, b, d, mask_;
259
260 #if HAVE_BIGENDIAN
261 vec_u8 tmp1, tmp2, mask, edges, align;
262 mask_ = vec_lvsl(0, src2);
263 #endif
264
265 for (i = 0; i < h; i++) {
266 a = unaligned_load(i * src_stride1, src1);
267 b = load_with_perm_vec(i * 16, src2, mask_);
268 d = vec_avg(a, b);
269 avg_unligned_store(d, dst);
270 dst += dst_stride;
271 }
272 }
273
274 /* Implemented but could be faster
275 #define put_pixels16_l2_altivec(d,s1,s2,ds,s1s,h) put_pixels16_l2(d,s1,s2,ds,s1s,16,h)
276 #define avg_pixels16_l2_altivec(d,s1,s2,ds,s1s,h) avg_pixels16_l2(d,s1,s2,ds,s1s,16,h)
277 */
278
279 H264_MC(put_, 16, altivec)
280 H264_MC(avg_, 16, altivec)
281 #endif /* HAVE_ALTIVEC */
282
ff_h264qpel_init_ppc(H264QpelContext * c,int bit_depth)283 av_cold void ff_h264qpel_init_ppc(H264QpelContext *c, int bit_depth)
284 {
285 #if HAVE_ALTIVEC
286 const int high_bit_depth = bit_depth > 8;
287
288 if (!PPC_ALTIVEC(av_get_cpu_flags()))
289 return;
290
291 if (!high_bit_depth) {
292 #define dspfunc(PFX, IDX, NUM) \
293 c->PFX ## _pixels_tab[IDX][ 0] = PFX ## NUM ## _mc00_altivec; \
294 c->PFX ## _pixels_tab[IDX][ 1] = PFX ## NUM ## _mc10_altivec; \
295 c->PFX ## _pixels_tab[IDX][ 2] = PFX ## NUM ## _mc20_altivec; \
296 c->PFX ## _pixels_tab[IDX][ 3] = PFX ## NUM ## _mc30_altivec; \
297 c->PFX ## _pixels_tab[IDX][ 4] = PFX ## NUM ## _mc01_altivec; \
298 c->PFX ## _pixels_tab[IDX][ 5] = PFX ## NUM ## _mc11_altivec; \
299 c->PFX ## _pixels_tab[IDX][ 6] = PFX ## NUM ## _mc21_altivec; \
300 c->PFX ## _pixels_tab[IDX][ 7] = PFX ## NUM ## _mc31_altivec; \
301 c->PFX ## _pixels_tab[IDX][ 8] = PFX ## NUM ## _mc02_altivec; \
302 c->PFX ## _pixels_tab[IDX][ 9] = PFX ## NUM ## _mc12_altivec; \
303 c->PFX ## _pixels_tab[IDX][10] = PFX ## NUM ## _mc22_altivec; \
304 c->PFX ## _pixels_tab[IDX][11] = PFX ## NUM ## _mc32_altivec; \
305 c->PFX ## _pixels_tab[IDX][12] = PFX ## NUM ## _mc03_altivec; \
306 c->PFX ## _pixels_tab[IDX][13] = PFX ## NUM ## _mc13_altivec; \
307 c->PFX ## _pixels_tab[IDX][14] = PFX ## NUM ## _mc23_altivec; \
308 c->PFX ## _pixels_tab[IDX][15] = PFX ## NUM ## _mc33_altivec
309
310 dspfunc(put_h264_qpel, 0, 16);
311 dspfunc(avg_h264_qpel, 0, 16);
312 #undef dspfunc
313 }
314 #endif /* HAVE_ALTIVEC */
315 }
316