2
* Copyright (c) 2004 Romain Dolbeau <romain@dolbeau.org>
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.
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.
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
19
#include "../dsputil.h"
21
#include "gcc_fixes.h"
23
#include "dsputil_altivec.h"
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)
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 "dsputil_h264_template_altivec.c"
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
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 "dsputil_h264_template_altivec.c"
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
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);\
73
static void OPNAME ## h264_qpel ## SIZE ## _mc10_ ## CODETYPE(uint8_t *dst, uint8_t *src, int stride){ \
74
DECLARE_ALIGNED_16(uint64_t, temp[SIZE*SIZE/8]);\
75
uint8_t * const half= (uint8_t*)temp;\
76
put_h264_qpel ## SIZE ## _h_lowpass_ ## CODETYPE(half, src, SIZE, stride);\
77
OPNAME ## pixels ## SIZE ## _l2_ ## CODETYPE(dst, src, half, stride, stride, SIZE);\
80
static void OPNAME ## h264_qpel ## SIZE ## _mc20_ ## CODETYPE(uint8_t *dst, uint8_t *src, int stride){\
81
OPNAME ## h264_qpel ## SIZE ## _h_lowpass_ ## CODETYPE(dst, src, stride, stride);\
84
static void OPNAME ## h264_qpel ## SIZE ## _mc30_ ## CODETYPE(uint8_t *dst, uint8_t *src, int stride){\
85
DECLARE_ALIGNED_16(uint64_t, temp[SIZE*SIZE/8]);\
86
uint8_t * const half= (uint8_t*)temp;\
87
put_h264_qpel ## SIZE ## _h_lowpass_ ## CODETYPE(half, src, SIZE, stride);\
88
OPNAME ## pixels ## SIZE ## _l2_ ## CODETYPE(dst, src+1, half, stride, stride, SIZE);\
91
static void OPNAME ## h264_qpel ## SIZE ## _mc01_ ## CODETYPE(uint8_t *dst, uint8_t *src, int stride){\
92
DECLARE_ALIGNED_16(uint64_t, temp[SIZE*SIZE/8]);\
93
uint8_t * const half= (uint8_t*)temp;\
94
put_h264_qpel ## SIZE ## _v_lowpass_ ## CODETYPE(half, src, SIZE, stride);\
95
OPNAME ## pixels ## SIZE ## _l2_ ## CODETYPE(dst, src, half, stride, stride, SIZE);\
98
static void OPNAME ## h264_qpel ## SIZE ## _mc02_ ## CODETYPE(uint8_t *dst, uint8_t *src, int stride){\
99
OPNAME ## h264_qpel ## SIZE ## _v_lowpass_ ## CODETYPE(dst, src, stride, stride);\
102
static void OPNAME ## h264_qpel ## SIZE ## _mc03_ ## CODETYPE(uint8_t *dst, uint8_t *src, int stride){\
103
DECLARE_ALIGNED_16(uint64_t, temp[SIZE*SIZE/8]);\
104
uint8_t * const half= (uint8_t*)temp;\
105
put_h264_qpel ## SIZE ## _v_lowpass_ ## CODETYPE(half, src, SIZE, stride);\
106
OPNAME ## pixels ## SIZE ## _l2_ ## CODETYPE(dst, src+stride, half, stride, stride, SIZE);\
109
static void OPNAME ## h264_qpel ## SIZE ## _mc11_ ## CODETYPE(uint8_t *dst, uint8_t *src, int stride){\
110
DECLARE_ALIGNED_16(uint64_t, temp[SIZE*SIZE/4]);\
111
uint8_t * const halfH= (uint8_t*)temp;\
112
uint8_t * const halfV= ((uint8_t*)temp) + SIZE*SIZE;\
113
put_h264_qpel ## SIZE ## _h_lowpass_ ## CODETYPE(halfH, src, SIZE, stride);\
114
put_h264_qpel ## SIZE ## _v_lowpass_ ## CODETYPE(halfV, src, SIZE, stride);\
115
OPNAME ## pixels ## SIZE ## _l2_ ## CODETYPE(dst, halfH, halfV, stride, SIZE, SIZE);\
118
static void OPNAME ## h264_qpel ## SIZE ## _mc31_ ## CODETYPE(uint8_t *dst, uint8_t *src, int stride){\
119
DECLARE_ALIGNED_16(uint64_t, temp[SIZE*SIZE/4]);\
120
uint8_t * const halfH= (uint8_t*)temp;\
121
uint8_t * const halfV= ((uint8_t*)temp) + SIZE*SIZE;\
122
put_h264_qpel ## SIZE ## _h_lowpass_ ## CODETYPE(halfH, src, SIZE, stride);\
123
put_h264_qpel ## SIZE ## _v_lowpass_ ## CODETYPE(halfV, src+1, SIZE, stride);\
124
OPNAME ## pixels ## SIZE ## _l2_ ## CODETYPE(dst, halfH, halfV, stride, SIZE, SIZE);\
127
static void OPNAME ## h264_qpel ## SIZE ## _mc13_ ## CODETYPE(uint8_t *dst, uint8_t *src, int stride){\
128
DECLARE_ALIGNED_16(uint64_t, temp[SIZE*SIZE/4]);\
129
uint8_t * const halfH= (uint8_t*)temp;\
130
uint8_t * const halfV= ((uint8_t*)temp) + SIZE*SIZE;\
131
put_h264_qpel ## SIZE ## _h_lowpass_ ## CODETYPE(halfH, src + stride, SIZE, stride);\
132
put_h264_qpel ## SIZE ## _v_lowpass_ ## CODETYPE(halfV, src, SIZE, stride);\
133
OPNAME ## pixels ## SIZE ## _l2_ ## CODETYPE(dst, halfH, halfV, stride, SIZE, SIZE);\
136
static void OPNAME ## h264_qpel ## SIZE ## _mc33_ ## CODETYPE(uint8_t *dst, uint8_t *src, int stride){\
137
DECLARE_ALIGNED_16(uint64_t, temp[SIZE*SIZE/4]);\
138
uint8_t * const halfH= (uint8_t*)temp;\
139
uint8_t * const halfV= ((uint8_t*)temp) + SIZE*SIZE;\
140
put_h264_qpel ## SIZE ## _h_lowpass_ ## CODETYPE(halfH, src + stride, SIZE, stride);\
141
put_h264_qpel ## SIZE ## _v_lowpass_ ## CODETYPE(halfV, src+1, SIZE, stride);\
142
OPNAME ## pixels ## SIZE ## _l2_ ## CODETYPE(dst, halfH, halfV, stride, SIZE, SIZE);\
145
static void OPNAME ## h264_qpel ## SIZE ## _mc22_ ## CODETYPE(uint8_t *dst, uint8_t *src, int stride){\
146
DECLARE_ALIGNED_16(uint64_t, temp[SIZE*(SIZE+8)/4]);\
147
int16_t * const tmp= (int16_t*)temp;\
148
OPNAME ## h264_qpel ## SIZE ## _hv_lowpass_ ## CODETYPE(dst, tmp, src, stride, SIZE, stride);\
151
static void OPNAME ## h264_qpel ## SIZE ## _mc21_ ## CODETYPE(uint8_t *dst, uint8_t *src, int stride){\
152
DECLARE_ALIGNED_16(uint64_t, temp[SIZE*(SIZE+8)/4 + SIZE*SIZE/4]);\
153
uint8_t * const halfH= (uint8_t*)temp;\
154
uint8_t * const halfHV= ((uint8_t*)temp) + SIZE*SIZE;\
155
int16_t * const tmp= ((int16_t*)temp) + SIZE*SIZE;\
156
put_h264_qpel ## SIZE ## _h_lowpass_ ## CODETYPE(halfH, src, SIZE, stride);\
157
put_h264_qpel ## SIZE ## _hv_lowpass_ ## CODETYPE(halfHV, tmp, src, SIZE, SIZE, stride);\
158
OPNAME ## pixels ## SIZE ## _l2_ ## CODETYPE(dst, halfH, halfHV, stride, SIZE, SIZE);\
161
static void OPNAME ## h264_qpel ## SIZE ## _mc23_ ## CODETYPE(uint8_t *dst, uint8_t *src, int stride){\
162
DECLARE_ALIGNED_16(uint64_t, temp[SIZE*(SIZE+8)/4 + SIZE*SIZE/4]);\
163
uint8_t * const halfH= (uint8_t*)temp;\
164
uint8_t * const halfHV= ((uint8_t*)temp) + SIZE*SIZE;\
165
int16_t * const tmp= ((int16_t*)temp) + SIZE*SIZE;\
166
put_h264_qpel ## SIZE ## _h_lowpass_ ## CODETYPE(halfH, src + stride, SIZE, stride);\
167
put_h264_qpel ## SIZE ## _hv_lowpass_ ## CODETYPE(halfHV, tmp, src, SIZE, SIZE, stride);\
168
OPNAME ## pixels ## SIZE ## _l2_ ## CODETYPE(dst, halfH, halfHV, stride, SIZE, SIZE);\
171
static void OPNAME ## h264_qpel ## SIZE ## _mc12_ ## CODETYPE(uint8_t *dst, uint8_t *src, int stride){\
172
DECLARE_ALIGNED_16(uint64_t, temp[SIZE*(SIZE+8)/4 + SIZE*SIZE/4]);\
173
uint8_t * const halfV= (uint8_t*)temp;\
174
uint8_t * const halfHV= ((uint8_t*)temp) + SIZE*SIZE;\
175
int16_t * const tmp= ((int16_t*)temp) + SIZE*SIZE;\
176
put_h264_qpel ## SIZE ## _v_lowpass_ ## CODETYPE(halfV, src, SIZE, stride);\
177
put_h264_qpel ## SIZE ## _hv_lowpass_ ## CODETYPE(halfHV, tmp, src, SIZE, SIZE, stride);\
178
OPNAME ## pixels ## SIZE ## _l2_ ## CODETYPE(dst, halfV, halfHV, stride, SIZE, SIZE);\
181
static void OPNAME ## h264_qpel ## SIZE ## _mc32_ ## CODETYPE(uint8_t *dst, uint8_t *src, int stride){\
182
DECLARE_ALIGNED_16(uint64_t, temp[SIZE*(SIZE+8)/4 + SIZE*SIZE/4]);\
183
uint8_t * const halfV= (uint8_t*)temp;\
184
uint8_t * const halfHV= ((uint8_t*)temp) + SIZE*SIZE;\
185
int16_t * const tmp= ((int16_t*)temp) + SIZE*SIZE;\
186
put_h264_qpel ## SIZE ## _v_lowpass_ ## CODETYPE(halfV, src+1, SIZE, stride);\
187
put_h264_qpel ## SIZE ## _hv_lowpass_ ## CODETYPE(halfHV, tmp, src, SIZE, SIZE, stride);\
188
OPNAME ## pixels ## SIZE ## _l2_ ## CODETYPE(dst, halfV, halfHV, stride, SIZE, SIZE);\
191
static inline void put_pixels16_l2_altivec( uint8_t * dst, const uint8_t * src1,
192
const uint8_t * src2, int dst_stride,
193
int src_stride1, int h)
196
vector unsigned char a, b, d, tmp1, tmp2, mask, mask_, edges, align;
198
mask_ = vec_lvsl(0, src2);
200
for (i = 0; i < h; i++) {
202
tmp1 = vec_ld(i * src_stride1, src1);
203
mask = vec_lvsl(i * src_stride1, src1);
204
tmp2 = vec_ld(i * src_stride1 + 15, src1);
206
a = vec_perm(tmp1, tmp2, mask);
208
tmp1 = vec_ld(i * 16, src2);
209
tmp2 = vec_ld(i * 16 + 15, src2);
211
b = vec_perm(tmp1, tmp2, mask_);
213
tmp1 = vec_ld(0, dst);
214
mask = vec_lvsl(0, dst);
215
tmp2 = vec_ld(15, dst);
219
edges = vec_perm(tmp2, tmp1, mask);
221
align = vec_lvsr(0, dst);
223
tmp1 = vec_perm(edges, d, align);
224
tmp2 = vec_perm(d, edges, align);
226
vec_st(tmp2, 15, dst);
227
vec_st(tmp1, 0 , dst);
233
static inline void avg_pixels16_l2_altivec( uint8_t * dst, const uint8_t * src1,
234
const uint8_t * src2, int dst_stride,
235
int src_stride1, int h)
238
vector unsigned char a, b, d, tmp1, tmp2, mask, mask_, edges, align;
240
mask_ = vec_lvsl(0, src2);
242
for (i = 0; i < h; i++) {
244
tmp1 = vec_ld(i * src_stride1, src1);
245
mask = vec_lvsl(i * src_stride1, src1);
246
tmp2 = vec_ld(i * src_stride1 + 15, src1);
248
a = vec_perm(tmp1, tmp2, mask);
250
tmp1 = vec_ld(i * 16, src2);
251
tmp2 = vec_ld(i * 16 + 15, src2);
253
b = vec_perm(tmp1, tmp2, mask_);
255
tmp1 = vec_ld(0, dst);
256
mask = vec_lvsl(0, dst);
257
tmp2 = vec_ld(15, dst);
259
d = vec_avg(vec_perm(tmp1, tmp2, mask), vec_avg(a, b));
261
edges = vec_perm(tmp2, tmp1, mask);
263
align = vec_lvsr(0, dst);
265
tmp1 = vec_perm(edges, d, align);
266
tmp2 = vec_perm(d, edges, align);
268
vec_st(tmp2, 15, dst);
269
vec_st(tmp1, 0 , dst);
275
/* Implemented but could be faster
276
#define put_pixels16_l2_altivec(d,s1,s2,ds,s1s,h) put_pixels16_l2(d,s1,s2,ds,s1s,16,h)
277
#define avg_pixels16_l2_altivec(d,s1,s2,ds,s1s,h) avg_pixels16_l2(d,s1,s2,ds,s1s,16,h)
280
H264_MC(put_, 16, altivec)
281
H264_MC(avg_, 16, altivec)
283
void dsputil_h264_init_ppc(DSPContext* c, AVCodecContext *avctx) {
287
c->put_h264_chroma_pixels_tab[0] = put_h264_chroma_mc8_altivec;
288
c->avg_h264_chroma_pixels_tab[0] = avg_h264_chroma_mc8_altivec;
290
#define dspfunc(PFX, IDX, NUM) \
291
c->PFX ## _pixels_tab[IDX][ 0] = PFX ## NUM ## _mc00_altivec; \
292
c->PFX ## _pixels_tab[IDX][ 1] = PFX ## NUM ## _mc10_altivec; \
293
c->PFX ## _pixels_tab[IDX][ 2] = PFX ## NUM ## _mc20_altivec; \
294
c->PFX ## _pixels_tab[IDX][ 3] = PFX ## NUM ## _mc30_altivec; \
295
c->PFX ## _pixels_tab[IDX][ 4] = PFX ## NUM ## _mc01_altivec; \
296
c->PFX ## _pixels_tab[IDX][ 5] = PFX ## NUM ## _mc11_altivec; \
297
c->PFX ## _pixels_tab[IDX][ 6] = PFX ## NUM ## _mc21_altivec; \
298
c->PFX ## _pixels_tab[IDX][ 7] = PFX ## NUM ## _mc31_altivec; \
299
c->PFX ## _pixels_tab[IDX][ 8] = PFX ## NUM ## _mc02_altivec; \
300
c->PFX ## _pixels_tab[IDX][ 9] = PFX ## NUM ## _mc12_altivec; \
301
c->PFX ## _pixels_tab[IDX][10] = PFX ## NUM ## _mc22_altivec; \
302
c->PFX ## _pixels_tab[IDX][11] = PFX ## NUM ## _mc32_altivec; \
303
c->PFX ## _pixels_tab[IDX][12] = PFX ## NUM ## _mc03_altivec; \
304
c->PFX ## _pixels_tab[IDX][13] = PFX ## NUM ## _mc13_altivec; \
305
c->PFX ## _pixels_tab[IDX][14] = PFX ## NUM ## _mc23_altivec; \
306
c->PFX ## _pixels_tab[IDX][15] = PFX ## NUM ## _mc33_altivec
308
dspfunc(put_h264_qpel, 0, 16);
309
dspfunc(avg_h264_qpel, 0, 16);
313
#endif /* HAVE_ALTIVEC */
315
// Non-AltiVec PPC optimisations