FFmpeg
vc1dsp_altivec.c
Go to the documentation of this file.
1 /*
2  * VC-1 and WMV3 decoder - DSP functions AltiVec-optimized
3  * Copyright (c) 2006 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 #include "config.h"
23 
24 #include "libavutil/attributes.h"
25 #include "libavutil/cpu.h"
26 #include "libavutil/ppc/cpu.h"
28 
29 #include "libavcodec/vc1dsp.h"
30 
31 #if HAVE_ALTIVEC
32 
33 // main steps of 8x8 transform
34 #define STEP8(s0, s1, s2, s3, s4, s5, s6, s7, vec_rnd) \
35 do { \
36  t0 = vec_sl(vec_add(s0, s4), vec_2); \
37  t0 = vec_add(vec_sl(t0, vec_1), t0); \
38  t0 = vec_add(t0, vec_rnd); \
39  t1 = vec_sl(vec_sub(s0, s4), vec_2); \
40  t1 = vec_add(vec_sl(t1, vec_1), t1); \
41  t1 = vec_add(t1, vec_rnd); \
42  t2 = vec_add(vec_sl(s6, vec_2), vec_sl(s6, vec_1)); \
43  t2 = vec_add(t2, vec_sl(s2, vec_4)); \
44  t3 = vec_add(vec_sl(s2, vec_2), vec_sl(s2, vec_1)); \
45  t3 = vec_sub(t3, vec_sl(s6, vec_4)); \
46  t4 = vec_add(t0, t2); \
47  t5 = vec_add(t1, t3); \
48  t6 = vec_sub(t1, t3); \
49  t7 = vec_sub(t0, t2); \
50 \
51  t0 = vec_sl(vec_add(s1, s3), vec_4); \
52  t0 = vec_add(t0, vec_sl(s5, vec_3)); \
53  t0 = vec_add(t0, vec_sl(s7, vec_2)); \
54  t0 = vec_add(t0, vec_sub(s5, s3)); \
55 \
56  t1 = vec_sl(vec_sub(s1, s5), vec_4); \
57  t1 = vec_sub(t1, vec_sl(s7, vec_3)); \
58  t1 = vec_sub(t1, vec_sl(s3, vec_2)); \
59  t1 = vec_sub(t1, vec_add(s1, s7)); \
60 \
61  t2 = vec_sl(vec_sub(s7, s3), vec_4); \
62  t2 = vec_add(t2, vec_sl(s1, vec_3)); \
63  t2 = vec_add(t2, vec_sl(s5, vec_2)); \
64  t2 = vec_add(t2, vec_sub(s1, s7)); \
65 \
66  t3 = vec_sl(vec_sub(s5, s7), vec_4); \
67  t3 = vec_sub(t3, vec_sl(s3, vec_3)); \
68  t3 = vec_add(t3, vec_sl(s1, vec_2)); \
69  t3 = vec_sub(t3, vec_add(s3, s5)); \
70 \
71  s0 = vec_add(t4, t0); \
72  s1 = vec_add(t5, t1); \
73  s2 = vec_add(t6, t2); \
74  s3 = vec_add(t7, t3); \
75  s4 = vec_sub(t7, t3); \
76  s5 = vec_sub(t6, t2); \
77  s6 = vec_sub(t5, t1); \
78  s7 = vec_sub(t4, t0); \
79 }while(0)
80 
81 #define SHIFT_HOR8(s0, s1, s2, s3, s4, s5, s6, s7) \
82 do { \
83  s0 = vec_sra(s0, vec_3); \
84  s1 = vec_sra(s1, vec_3); \
85  s2 = vec_sra(s2, vec_3); \
86  s3 = vec_sra(s3, vec_3); \
87  s4 = vec_sra(s4, vec_3); \
88  s5 = vec_sra(s5, vec_3); \
89  s6 = vec_sra(s6, vec_3); \
90  s7 = vec_sra(s7, vec_3); \
91 }while(0)
92 
93 #define SHIFT_VERT8(s0, s1, s2, s3, s4, s5, s6, s7) \
94 do { \
95  s0 = vec_sra(s0, vec_7); \
96  s1 = vec_sra(s1, vec_7); \
97  s2 = vec_sra(s2, vec_7); \
98  s3 = vec_sra(s3, vec_7); \
99  s4 = vec_sra(vec_add(s4, vec_1s), vec_7); \
100  s5 = vec_sra(vec_add(s5, vec_1s), vec_7); \
101  s6 = vec_sra(vec_add(s6, vec_1s), vec_7); \
102  s7 = vec_sra(vec_add(s7, vec_1s), vec_7); \
103 }while(0)
104 
105 /* main steps of 4x4 transform */
106 #define STEP4(s0, s1, s2, s3, vec_rnd) \
107 do { \
108  t1 = vec_add(vec_sl(s0, vec_4), s0); \
109  t1 = vec_add(t1, vec_rnd); \
110  t2 = vec_add(vec_sl(s2, vec_4), s2); \
111  t0 = vec_add(t1, t2); \
112  t1 = vec_sub(t1, t2); \
113  t3 = vec_sl(vec_sub(s3, s1), vec_1); \
114  t3 = vec_add(t3, vec_sl(t3, vec_2)); \
115  t2 = vec_add(t3, vec_sl(s1, vec_5)); \
116  t3 = vec_add(t3, vec_sl(s3, vec_3)); \
117  t3 = vec_add(t3, vec_sl(s3, vec_2)); \
118  s0 = vec_add(t0, t2); \
119  s1 = vec_sub(t1, t3); \
120  s2 = vec_add(t1, t3); \
121  s3 = vec_sub(t0, t2); \
122 }while (0)
123 
124 #define SHIFT_HOR4(s0, s1, s2, s3) \
125  s0 = vec_sra(s0, vec_3); \
126  s1 = vec_sra(s1, vec_3); \
127  s2 = vec_sra(s2, vec_3); \
128  s3 = vec_sra(s3, vec_3);
129 
130 #define SHIFT_VERT4(s0, s1, s2, s3) \
131  s0 = vec_sra(s0, vec_7); \
132  s1 = vec_sra(s1, vec_7); \
133  s2 = vec_sra(s2, vec_7); \
134  s3 = vec_sra(s3, vec_7);
135 
136 /** Do inverse transform on 8x8 block
137 */
138 static void vc1_inv_trans_8x8_altivec(int16_t block[64])
139 {
140  vector signed short src0, src1, src2, src3, src4, src5, src6, src7;
141  vector signed int s0, s1, s2, s3, s4, s5, s6, s7;
142  vector signed int s8, s9, sA, sB, sC, sD, sE, sF;
143  vector signed int t0, t1, t2, t3, t4, t5, t6, t7;
144  const vector signed int vec_64 = vec_sl(vec_splat_s32(4), vec_splat_u32(4));
145  const vector unsigned int vec_7 = vec_splat_u32(7);
146  const vector unsigned int vec_4 = vec_splat_u32(4);
147  const vector signed int vec_4s = vec_splat_s32(4);
148  const vector unsigned int vec_3 = vec_splat_u32(3);
149  const vector unsigned int vec_2 = vec_splat_u32(2);
150  const vector signed int vec_1s = vec_splat_s32(1);
151  const vector unsigned int vec_1 = vec_splat_u32(1);
152 
153  src0 = vec_ld( 0, block);
154  src1 = vec_ld( 16, block);
155  src2 = vec_ld( 32, block);
156  src3 = vec_ld( 48, block);
157  src4 = vec_ld( 64, block);
158  src5 = vec_ld( 80, block);
159  src6 = vec_ld( 96, block);
160  src7 = vec_ld(112, block);
161 
162  s0 = vec_unpackl(src0);
163  s1 = vec_unpackl(src1);
164  s2 = vec_unpackl(src2);
165  s3 = vec_unpackl(src3);
166  s4 = vec_unpackl(src4);
167  s5 = vec_unpackl(src5);
168  s6 = vec_unpackl(src6);
169  s7 = vec_unpackl(src7);
170  s8 = vec_unpackh(src0);
171  s9 = vec_unpackh(src1);
172  sA = vec_unpackh(src2);
173  sB = vec_unpackh(src3);
174  sC = vec_unpackh(src4);
175  sD = vec_unpackh(src5);
176  sE = vec_unpackh(src6);
177  sF = vec_unpackh(src7);
178  STEP8(s0, s1, s2, s3, s4, s5, s6, s7, vec_4s);
179  SHIFT_HOR8(s0, s1, s2, s3, s4, s5, s6, s7);
180  STEP8(s8, s9, sA, sB, sC, sD, sE, sF, vec_4s);
181  SHIFT_HOR8(s8, s9, sA, sB, sC, sD, sE, sF);
182  src0 = vec_pack(s8, s0);
183  src1 = vec_pack(s9, s1);
184  src2 = vec_pack(sA, s2);
185  src3 = vec_pack(sB, s3);
186  src4 = vec_pack(sC, s4);
187  src5 = vec_pack(sD, s5);
188  src6 = vec_pack(sE, s6);
189  src7 = vec_pack(sF, s7);
190  TRANSPOSE8(src0, src1, src2, src3, src4, src5, src6, src7);
191 
192  s0 = vec_unpackl(src0);
193  s1 = vec_unpackl(src1);
194  s2 = vec_unpackl(src2);
195  s3 = vec_unpackl(src3);
196  s4 = vec_unpackl(src4);
197  s5 = vec_unpackl(src5);
198  s6 = vec_unpackl(src6);
199  s7 = vec_unpackl(src7);
200  s8 = vec_unpackh(src0);
201  s9 = vec_unpackh(src1);
202  sA = vec_unpackh(src2);
203  sB = vec_unpackh(src3);
204  sC = vec_unpackh(src4);
205  sD = vec_unpackh(src5);
206  sE = vec_unpackh(src6);
207  sF = vec_unpackh(src7);
208  STEP8(s0, s1, s2, s3, s4, s5, s6, s7, vec_64);
209  SHIFT_VERT8(s0, s1, s2, s3, s4, s5, s6, s7);
210  STEP8(s8, s9, sA, sB, sC, sD, sE, sF, vec_64);
211  SHIFT_VERT8(s8, s9, sA, sB, sC, sD, sE, sF);
212  src0 = vec_pack(s8, s0);
213  src1 = vec_pack(s9, s1);
214  src2 = vec_pack(sA, s2);
215  src3 = vec_pack(sB, s3);
216  src4 = vec_pack(sC, s4);
217  src5 = vec_pack(sD, s5);
218  src6 = vec_pack(sE, s6);
219  src7 = vec_pack(sF, s7);
220 
221  vec_st(src0, 0, block);
222  vec_st(src1, 16, block);
223  vec_st(src2, 32, block);
224  vec_st(src3, 48, block);
225  vec_st(src4, 64, block);
226  vec_st(src5, 80, block);
227  vec_st(src6, 96, block);
228  vec_st(src7,112, block);
229 }
230 
231 /** Do inverse transform on 8x4 part of block
232 */
233 static void vc1_inv_trans_8x4_altivec(uint8_t *dest, ptrdiff_t stride,
234  int16_t *block)
235 {
236  vector signed short src0, src1, src2, src3, src4, src5, src6, src7;
237  vector signed int s0, s1, s2, s3, s4, s5, s6, s7;
238  vector signed int s8, s9, sA, sB, sC, sD, sE, sF;
239  vector signed int t0, t1, t2, t3, t4, t5, t6, t7;
240  const vector signed int vec_64 = vec_sl(vec_splat_s32(4), vec_splat_u32(4));
241  const vector unsigned int vec_7 = vec_splat_u32(7);
242  const vector unsigned int vec_5 = vec_splat_u32(5);
243  const vector unsigned int vec_4 = vec_splat_u32(4);
244  const vector signed int vec_4s = vec_splat_s32(4);
245  const vector unsigned int vec_3 = vec_splat_u32(3);
246  const vector unsigned int vec_2 = vec_splat_u32(2);
247  const vector unsigned int vec_1 = vec_splat_u32(1);
248  vector unsigned char tmp;
249  vector signed short tmp2, tmp3;
250  vector unsigned char perm0, perm1, p0, p1, p;
251 
252  src0 = vec_ld( 0, block);
253  src1 = vec_ld( 16, block);
254  src2 = vec_ld( 32, block);
255  src3 = vec_ld( 48, block);
256  src4 = vec_ld( 64, block);
257  src5 = vec_ld( 80, block);
258  src6 = vec_ld( 96, block);
259  src7 = vec_ld(112, block);
260 
261  TRANSPOSE8(src0, src1, src2, src3, src4, src5, src6, src7);
262  s0 = vec_unpackl(src0);
263  s1 = vec_unpackl(src1);
264  s2 = vec_unpackl(src2);
265  s3 = vec_unpackl(src3);
266  s4 = vec_unpackl(src4);
267  s5 = vec_unpackl(src5);
268  s6 = vec_unpackl(src6);
269  s7 = vec_unpackl(src7);
270  s8 = vec_unpackh(src0);
271  s9 = vec_unpackh(src1);
272  sA = vec_unpackh(src2);
273  sB = vec_unpackh(src3);
274  sC = vec_unpackh(src4);
275  sD = vec_unpackh(src5);
276  sE = vec_unpackh(src6);
277  sF = vec_unpackh(src7);
278  STEP8(s0, s1, s2, s3, s4, s5, s6, s7, vec_4s);
279  SHIFT_HOR8(s0, s1, s2, s3, s4, s5, s6, s7);
280  STEP8(s8, s9, sA, sB, sC, sD, sE, sF, vec_4s);
281  SHIFT_HOR8(s8, s9, sA, sB, sC, sD, sE, sF);
282  src0 = vec_pack(s8, s0);
283  src1 = vec_pack(s9, s1);
284  src2 = vec_pack(sA, s2);
285  src3 = vec_pack(sB, s3);
286  src4 = vec_pack(sC, s4);
287  src5 = vec_pack(sD, s5);
288  src6 = vec_pack(sE, s6);
289  src7 = vec_pack(sF, s7);
290  TRANSPOSE8(src0, src1, src2, src3, src4, src5, src6, src7);
291 
292  s0 = vec_unpackh(src0);
293  s1 = vec_unpackh(src1);
294  s2 = vec_unpackh(src2);
295  s3 = vec_unpackh(src3);
296  s8 = vec_unpackl(src0);
297  s9 = vec_unpackl(src1);
298  sA = vec_unpackl(src2);
299  sB = vec_unpackl(src3);
300  STEP4(s0, s1, s2, s3, vec_64);
301  SHIFT_VERT4(s0, s1, s2, s3);
302  STEP4(s8, s9, sA, sB, vec_64);
303  SHIFT_VERT4(s8, s9, sA, sB);
304  src0 = vec_pack(s0, s8);
305  src1 = vec_pack(s1, s9);
306  src2 = vec_pack(s2, sA);
307  src3 = vec_pack(s3, sB);
308 
309 #if HAVE_BIGENDIAN
310  p0 = vec_lvsl (0, dest);
311  p1 = vec_lvsl (stride, dest);
312  p = vec_splat_u8 (-1);
313  perm0 = vec_mergeh (p, p0);
314  perm1 = vec_mergeh (p, p1);
315 #define GET_TMP2(dst, p) \
316  tmp = vec_ld (0, dest); \
317  tmp2 = (vector signed short)vec_perm (tmp, vec_splat_u8(0), p);
318 #else
319 #define GET_TMP2(dst,p) \
320  tmp = vec_vsx_ld (0, dst); \
321  tmp2 = (vector signed short)vec_mergeh (tmp, vec_splat_u8(0));
322 #endif
323 
324 #define ADD(dest,src,perm) \
325  GET_TMP2(dest, perm); \
326  tmp3 = vec_adds (tmp2, src); \
327  tmp = vec_packsu (tmp3, tmp3); \
328  vec_ste ((vector unsigned int)tmp, 0, (unsigned int *)dest); \
329  vec_ste ((vector unsigned int)tmp, 4, (unsigned int *)dest);
330 
331  ADD (dest, src0, perm0) dest += stride;
332  ADD (dest, src1, perm1) dest += stride;
333  ADD (dest, src2, perm0) dest += stride;
334  ADD (dest, src3, perm1)
335 }
336 
337 #define PUT_OP_U8_ALTIVEC(d, s, dst) d = s
338 #define AVG_OP_U8_ALTIVEC(d, s, dst) d = vec_avg(dst, s)
339 
340 #define OP_U8_ALTIVEC PUT_OP_U8_ALTIVEC
341 #define PREFIX_no_rnd_vc1_chroma_mc8_altivec put_no_rnd_vc1_chroma_mc8_altivec
342 #include "h264chroma_template.c"
343 #undef OP_U8_ALTIVEC
344 #undef PREFIX_no_rnd_vc1_chroma_mc8_altivec
345 
346 #define OP_U8_ALTIVEC AVG_OP_U8_ALTIVEC
347 #define PREFIX_no_rnd_vc1_chroma_mc8_altivec avg_no_rnd_vc1_chroma_mc8_altivec
348 #include "h264chroma_template.c"
349 #undef OP_U8_ALTIVEC
350 #undef PREFIX_no_rnd_vc1_chroma_mc8_altivec
351 
352 #endif /* HAVE_ALTIVEC */
353 
355 {
356 #if HAVE_ALTIVEC
358  return;
359 
360  dsp->vc1_inv_trans_8x8 = vc1_inv_trans_8x8_altivec;
361  dsp->vc1_inv_trans_8x4 = vc1_inv_trans_8x4_altivec;
362  dsp->put_no_rnd_vc1_chroma_pixels_tab[0] = put_no_rnd_vc1_chroma_mc8_altivec;
363  dsp->avg_no_rnd_vc1_chroma_pixels_tab[0] = avg_no_rnd_vc1_chroma_mc8_altivec;
364 #endif /* HAVE_ALTIVEC */
365 }
s5
#define s5
Definition: regdef.h:42
vc1dsp.h
src1
const pixel * src1
Definition: h264pred_template.c:421
VC1DSPContext::avg_no_rnd_vc1_chroma_pixels_tab
h264_chroma_mc_func avg_no_rnd_vc1_chroma_pixels_tab[3]
Definition: vc1dsp.h:64
tmp
static uint8_t tmp[11]
Definition: aes_ctr.c:28
t0
#define t0
Definition: regdef.h:28
t1
#define t1
Definition: regdef.h:29
av_get_cpu_flags
int av_get_cpu_flags(void)
Return the flags which specify extensions supported by the CPU.
Definition: cpu.c:103
VC1DSPContext::put_no_rnd_vc1_chroma_pixels_tab
h264_chroma_mc_func put_no_rnd_vc1_chroma_pixels_tab[3]
Definition: vc1dsp.h:63
s3
#define s3
Definition: regdef.h:40
av_cold
#define av_cold
Definition: attributes.h:90
s1
#define s1
Definition: regdef.h:38
t7
#define t7
Definition: regdef.h:35
ADD
#define ADD(a, b)
Definition: dct32_template.c:123
VC1DSPContext::vc1_inv_trans_8x4
void(* vc1_inv_trans_8x4)(uint8_t *dest, ptrdiff_t stride, int16_t *block)
Definition: vc1dsp.h:38
t5
#define t5
Definition: regdef.h:33
s6
#define s6
Definition: regdef.h:43
t6
#define t6
Definition: regdef.h:34
s2
#define s2
Definition: regdef.h:39
PPC_ALTIVEC
#define PPC_ALTIVEC(flags)
Definition: cpu.h:25
cpu.h
attributes.h
VC1DSPContext::vc1_inv_trans_8x8
void(* vc1_inv_trans_8x8)(int16_t *b)
Definition: vc1dsp.h:37
t4
#define t4
Definition: regdef.h:32
t3
#define t3
Definition: regdef.h:31
src2
const pixel * src2
Definition: h264pred_template.c:422
s4
#define s4
Definition: regdef.h:41
stride
#define stride
Definition: h264pred_template.c:537
VC1DSPContext
Definition: vc1dsp.h:35
t2
#define t2
Definition: regdef.h:30
src0
const pixel *const src0
Definition: h264pred_template.c:420
s0
#define s0
Definition: regdef.h:37
util_altivec.h
ff_vc1dsp_init_ppc
av_cold void ff_vc1dsp_init_ppc(VC1DSPContext *dsp)
Definition: vc1dsp_altivec.c:354
cpu.h
block
The exact code depends on how similar the blocks are and how related they are to the block
Definition: filter_design.txt:207