Libav
h264dsp.c
Go to the documentation of this file.
1 /*
2  * Copyright (c) 2004 Romain Dolbeau <romain@dolbeau.org>
3  *
4  * This file is part of Libav.
5  *
6  * Libav 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  * Libav 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 Libav; 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 <stdint.h>
24 #include <string.h>
25 
26 #include "libavutil/attributes.h"
27 #include "libavutil/cpu.h"
28 #include "libavutil/intreadwrite.h"
29 #include "libavutil/mem.h"
30 #include "libavutil/ppc/cpu.h"
33 
34 #include "libavcodec/h264dec.h"
35 #include "libavcodec/h264dsp.h"
36 
37 #if HAVE_ALTIVEC && HAVE_BIGENDIAN
38 
39 /****************************************************************************
40  * IDCT transform:
41  ****************************************************************************/
42 
43 #define VEC_1D_DCT(vb0,vb1,vb2,vb3,va0,va1,va2,va3) \
44  /* 1st stage */ \
45  vz0 = vec_add(vb0,vb2); /* temp[0] = Y[0] + Y[2] */ \
46  vz1 = vec_sub(vb0,vb2); /* temp[1] = Y[0] - Y[2] */ \
47  vz2 = vec_sra(vb1,vec_splat_u16(1)); \
48  vz2 = vec_sub(vz2,vb3); /* temp[2] = Y[1].1/2 - Y[3] */ \
49  vz3 = vec_sra(vb3,vec_splat_u16(1)); \
50  vz3 = vec_add(vb1,vz3); /* temp[3] = Y[1] + Y[3].1/2 */ \
51  /* 2nd stage: output */ \
52  va0 = vec_add(vz0,vz3); /* x[0] = temp[0] + temp[3] */ \
53  va1 = vec_add(vz1,vz2); /* x[1] = temp[1] + temp[2] */ \
54  va2 = vec_sub(vz1,vz2); /* x[2] = temp[1] - temp[2] */ \
55  va3 = vec_sub(vz0,vz3) /* x[3] = temp[0] - temp[3] */
56 
57 #define VEC_TRANSPOSE_4(a0,a1,a2,a3,b0,b1,b2,b3) \
58  b0 = vec_mergeh( a0, a0 ); \
59  b1 = vec_mergeh( a1, a0 ); \
60  b2 = vec_mergeh( a2, a0 ); \
61  b3 = vec_mergeh( a3, a0 ); \
62  a0 = vec_mergeh( b0, b2 ); \
63  a1 = vec_mergel( b0, b2 ); \
64  a2 = vec_mergeh( b1, b3 ); \
65  a3 = vec_mergel( b1, b3 ); \
66  b0 = vec_mergeh( a0, a2 ); \
67  b1 = vec_mergel( a0, a2 ); \
68  b2 = vec_mergeh( a1, a3 ); \
69  b3 = vec_mergel( a1, a3 )
70 
71 #define VEC_LOAD_U8_ADD_S16_STORE_U8(va) \
72  vdst_orig = vec_ld(0, dst); \
73  vdst = vec_perm(vdst_orig, zero_u8v, vdst_mask); \
74  vdst_ss = (vec_s16) vec_mergeh(zero_u8v, vdst); \
75  va = vec_add(va, vdst_ss); \
76  va_u8 = vec_packsu(va, zero_s16v); \
77  va_u32 = vec_splat((vec_u32)va_u8, 0); \
78  vec_ste(va_u32, element, (uint32_t*)dst);
79 
80 static void h264_idct_add_altivec(uint8_t *dst, int16_t *block, int stride)
81 {
82  vec_s16 va0, va1, va2, va3;
83  vec_s16 vz0, vz1, vz2, vz3;
84  vec_s16 vtmp0, vtmp1, vtmp2, vtmp3;
85  vec_u8 va_u8;
86  vec_u32 va_u32;
87  vec_s16 vdst_ss;
88  const vec_u16 v6us = vec_splat_u16(6);
89  vec_u8 vdst, vdst_orig;
90  vec_u8 vdst_mask = vec_lvsl(0, dst);
91  int element = ((unsigned long)dst & 0xf) >> 2;
92  LOAD_ZERO;
93 
94  block[0] += 32; /* add 32 as a DC-level for rounding */
95 
96  vtmp0 = vec_ld(0,block);
97  vtmp1 = vec_sld(vtmp0, vtmp0, 8);
98  vtmp2 = vec_ld(16,block);
99  vtmp3 = vec_sld(vtmp2, vtmp2, 8);
100  memset(block, 0, 16 * sizeof(int16_t));
101 
102  VEC_1D_DCT(vtmp0,vtmp1,vtmp2,vtmp3,va0,va1,va2,va3);
103  VEC_TRANSPOSE_4(va0,va1,va2,va3,vtmp0,vtmp1,vtmp2,vtmp3);
104  VEC_1D_DCT(vtmp0,vtmp1,vtmp2,vtmp3,va0,va1,va2,va3);
105 
106  va0 = vec_sra(va0,v6us);
107  va1 = vec_sra(va1,v6us);
108  va2 = vec_sra(va2,v6us);
109  va3 = vec_sra(va3,v6us);
110 
111  VEC_LOAD_U8_ADD_S16_STORE_U8(va0);
112  dst += stride;
113  VEC_LOAD_U8_ADD_S16_STORE_U8(va1);
114  dst += stride;
115  VEC_LOAD_U8_ADD_S16_STORE_U8(va2);
116  dst += stride;
117  VEC_LOAD_U8_ADD_S16_STORE_U8(va3);
118 }
119 
120 #define IDCT8_1D_ALTIVEC(s0, s1, s2, s3, s4, s5, s6, s7, d0, d1, d2, d3, d4, d5, d6, d7) {\
121  /* a0 = SRC(0) + SRC(4); */ \
122  vec_s16 a0v = vec_add(s0, s4); \
123  /* a2 = SRC(0) - SRC(4); */ \
124  vec_s16 a2v = vec_sub(s0, s4); \
125  /* a4 = (SRC(2)>>1) - SRC(6); */ \
126  vec_s16 a4v = vec_sub(vec_sra(s2, onev), s6); \
127  /* a6 = (SRC(6)>>1) + SRC(2); */ \
128  vec_s16 a6v = vec_add(vec_sra(s6, onev), s2); \
129  /* b0 = a0 + a6; */ \
130  vec_s16 b0v = vec_add(a0v, a6v); \
131  /* b2 = a2 + a4; */ \
132  vec_s16 b2v = vec_add(a2v, a4v); \
133  /* b4 = a2 - a4; */ \
134  vec_s16 b4v = vec_sub(a2v, a4v); \
135  /* b6 = a0 - a6; */ \
136  vec_s16 b6v = vec_sub(a0v, a6v); \
137  /* a1 = SRC(5) - SRC(3) - SRC(7) - (SRC(7)>>1); */ \
138  /* a1 = (SRC(5)-SRC(3)) - (SRC(7) + (SRC(7)>>1)); */ \
139  vec_s16 a1v = vec_sub( vec_sub(s5, s3), vec_add(s7, vec_sra(s7, onev)) ); \
140  /* a3 = SRC(7) + SRC(1) - SRC(3) - (SRC(3)>>1); */ \
141  /* a3 = (SRC(7)+SRC(1)) - (SRC(3) + (SRC(3)>>1)); */ \
142  vec_s16 a3v = vec_sub( vec_add(s7, s1), vec_add(s3, vec_sra(s3, onev)) );\
143  /* a5 = SRC(7) - SRC(1) + SRC(5) + (SRC(5)>>1); */ \
144  /* a5 = (SRC(7)-SRC(1)) + SRC(5) + (SRC(5)>>1); */ \
145  vec_s16 a5v = vec_add( vec_sub(s7, s1), vec_add(s5, vec_sra(s5, onev)) );\
146  /* a7 = SRC(5)+SRC(3) + SRC(1) + (SRC(1)>>1); */ \
147  vec_s16 a7v = vec_add( vec_add(s5, s3), vec_add(s1, vec_sra(s1, onev)) );\
148  /* b1 = (a7>>2) + a1; */ \
149  vec_s16 b1v = vec_add( vec_sra(a7v, twov), a1v); \
150  /* b3 = a3 + (a5>>2); */ \
151  vec_s16 b3v = vec_add(a3v, vec_sra(a5v, twov)); \
152  /* b5 = (a3>>2) - a5; */ \
153  vec_s16 b5v = vec_sub( vec_sra(a3v, twov), a5v); \
154  /* b7 = a7 - (a1>>2); */ \
155  vec_s16 b7v = vec_sub( a7v, vec_sra(a1v, twov)); \
156  /* DST(0, b0 + b7); */ \
157  d0 = vec_add(b0v, b7v); \
158  /* DST(1, b2 + b5); */ \
159  d1 = vec_add(b2v, b5v); \
160  /* DST(2, b4 + b3); */ \
161  d2 = vec_add(b4v, b3v); \
162  /* DST(3, b6 + b1); */ \
163  d3 = vec_add(b6v, b1v); \
164  /* DST(4, b6 - b1); */ \
165  d4 = vec_sub(b6v, b1v); \
166  /* DST(5, b4 - b3); */ \
167  d5 = vec_sub(b4v, b3v); \
168  /* DST(6, b2 - b5); */ \
169  d6 = vec_sub(b2v, b5v); \
170  /* DST(7, b0 - b7); */ \
171  d7 = vec_sub(b0v, b7v); \
172 }
173 
174 #define ALTIVEC_STORE_SUM_CLIP(dest, idctv, perm_ldv, perm_stv, sel) { \
175  /* unaligned load */ \
176  vec_u8 hv = vec_ld( 0, dest ); \
177  vec_u8 lv = vec_ld( 7, dest ); \
178  vec_u8 dstv = vec_perm( hv, lv, (vec_u8)perm_ldv ); \
179  vec_s16 idct_sh6 = vec_sra(idctv, sixv); \
180  vec_u16 dst16 = (vec_u16)vec_mergeh(zero_u8v, dstv); \
181  vec_s16 idstsum = vec_adds(idct_sh6, (vec_s16)dst16); \
182  vec_u8 idstsum8 = vec_packsu(zero_s16v, idstsum); \
183  vec_u8 edgehv; \
184  /* unaligned store */ \
185  vec_u8 bodyv = vec_perm( idstsum8, idstsum8, perm_stv );\
186  vec_u8 edgelv = vec_perm( sel, zero_u8v, perm_stv ); \
187  lv = vec_sel( lv, bodyv, edgelv ); \
188  vec_st( lv, 7, dest ); \
189  hv = vec_ld( 0, dest ); \
190  edgehv = vec_perm( zero_u8v, sel, perm_stv ); \
191  hv = vec_sel( hv, bodyv, edgehv ); \
192  vec_st( hv, 0, dest ); \
193  }
194 
195 static void h264_idct8_add_altivec(uint8_t *dst, int16_t *dct, int stride)
196 {
197  vec_s16 s0, s1, s2, s3, s4, s5, s6, s7;
198  vec_s16 d0, d1, d2, d3, d4, d5, d6, d7;
199  vec_s16 idct0, idct1, idct2, idct3, idct4, idct5, idct6, idct7;
200 
201  vec_u8 perm_ldv = vec_lvsl(0, dst);
202  vec_u8 perm_stv = vec_lvsr(8, dst);
203 
204  const vec_u16 onev = vec_splat_u16(1);
205  const vec_u16 twov = vec_splat_u16(2);
206  const vec_u16 sixv = vec_splat_u16(6);
207 
208  const vec_u8 sel = (vec_u8) {0,0,0,0,0,0,0,0,-1,-1,-1,-1,-1,-1,-1,-1};
209  LOAD_ZERO;
210 
211  dct[0] += 32; // rounding for the >>6 at the end
212 
213  s0 = vec_ld(0x00, (int16_t*)dct);
214  s1 = vec_ld(0x10, (int16_t*)dct);
215  s2 = vec_ld(0x20, (int16_t*)dct);
216  s3 = vec_ld(0x30, (int16_t*)dct);
217  s4 = vec_ld(0x40, (int16_t*)dct);
218  s5 = vec_ld(0x50, (int16_t*)dct);
219  s6 = vec_ld(0x60, (int16_t*)dct);
220  s7 = vec_ld(0x70, (int16_t*)dct);
221  memset(dct, 0, 64 * sizeof(int16_t));
222 
223  IDCT8_1D_ALTIVEC(s0, s1, s2, s3, s4, s5, s6, s7,
224  d0, d1, d2, d3, d4, d5, d6, d7);
225 
226  TRANSPOSE8( d0, d1, d2, d3, d4, d5, d6, d7 );
227 
228  IDCT8_1D_ALTIVEC(d0, d1, d2, d3, d4, d5, d6, d7,
229  idct0, idct1, idct2, idct3, idct4, idct5, idct6, idct7);
230 
231  ALTIVEC_STORE_SUM_CLIP(&dst[0*stride], idct0, perm_ldv, perm_stv, sel);
232  ALTIVEC_STORE_SUM_CLIP(&dst[1*stride], idct1, perm_ldv, perm_stv, sel);
233  ALTIVEC_STORE_SUM_CLIP(&dst[2*stride], idct2, perm_ldv, perm_stv, sel);
234  ALTIVEC_STORE_SUM_CLIP(&dst[3*stride], idct3, perm_ldv, perm_stv, sel);
235  ALTIVEC_STORE_SUM_CLIP(&dst[4*stride], idct4, perm_ldv, perm_stv, sel);
236  ALTIVEC_STORE_SUM_CLIP(&dst[5*stride], idct5, perm_ldv, perm_stv, sel);
237  ALTIVEC_STORE_SUM_CLIP(&dst[6*stride], idct6, perm_ldv, perm_stv, sel);
238  ALTIVEC_STORE_SUM_CLIP(&dst[7*stride], idct7, perm_ldv, perm_stv, sel);
239 }
240 
241 static av_always_inline void h264_idct_dc_add_internal(uint8_t *dst, int16_t *block, int stride, int size)
242 {
243  vec_s16 dc16;
244  vec_u8 dcplus, dcminus, v0, v1, v2, v3, aligner;
245  LOAD_ZERO;
246  DECLARE_ALIGNED(16, int, dc);
247  int i;
248 
249  dc = (block[0] + 32) >> 6;
250  block[0] = 0;
251  dc16 = vec_splat((vec_s16) vec_lde(0, &dc), 1);
252 
253  if (size == 4)
254  dc16 = vec_sld(dc16, zero_s16v, 8);
255  dcplus = vec_packsu(dc16, zero_s16v);
256  dcminus = vec_packsu(vec_sub(zero_s16v, dc16), zero_s16v);
257 
258  aligner = vec_lvsr(0, dst);
259  dcplus = vec_perm(dcplus, dcplus, aligner);
260  dcminus = vec_perm(dcminus, dcminus, aligner);
261 
262  for (i = 0; i < size; i += 4) {
263  v0 = vec_ld(0, dst+0*stride);
264  v1 = vec_ld(0, dst+1*stride);
265  v2 = vec_ld(0, dst+2*stride);
266  v3 = vec_ld(0, dst+3*stride);
267 
268  v0 = vec_adds(v0, dcplus);
269  v1 = vec_adds(v1, dcplus);
270  v2 = vec_adds(v2, dcplus);
271  v3 = vec_adds(v3, dcplus);
272 
273  v0 = vec_subs(v0, dcminus);
274  v1 = vec_subs(v1, dcminus);
275  v2 = vec_subs(v2, dcminus);
276  v3 = vec_subs(v3, dcminus);
277 
278  vec_st(v0, 0, dst+0*stride);
279  vec_st(v1, 0, dst+1*stride);
280  vec_st(v2, 0, dst+2*stride);
281  vec_st(v3, 0, dst+3*stride);
282 
283  dst += 4*stride;
284  }
285 }
286 
287 static void h264_idct_dc_add_altivec(uint8_t *dst, int16_t *block, int stride)
288 {
289  h264_idct_dc_add_internal(dst, block, stride, 4);
290 }
291 
292 static void h264_idct8_dc_add_altivec(uint8_t *dst, int16_t *block, int stride)
293 {
294  h264_idct_dc_add_internal(dst, block, stride, 8);
295 }
296 
297 static void h264_idct_add16_altivec(uint8_t *dst, const int *block_offset,
298  int16_t *block, int stride,
299  const uint8_t nnzc[15 * 8])
300 {
301  int i;
302  for(i=0; i<16; i++){
303  int nnz = nnzc[ scan8[i] ];
304  if(nnz){
305  if(nnz==1 && block[i*16]) h264_idct_dc_add_altivec(dst + block_offset[i], block + i*16, stride);
306  else h264_idct_add_altivec(dst + block_offset[i], block + i*16, stride);
307  }
308  }
309 }
310 
311 static void h264_idct_add16intra_altivec(uint8_t *dst, const int *block_offset,
312  int16_t *block, int stride,
313  const uint8_t nnzc[15 * 8])
314 {
315  int i;
316  for(i=0; i<16; i++){
317  if(nnzc[ scan8[i] ]) h264_idct_add_altivec(dst + block_offset[i], block + i*16, stride);
318  else if(block[i*16]) h264_idct_dc_add_altivec(dst + block_offset[i], block + i*16, stride);
319  }
320 }
321 
322 static void h264_idct8_add4_altivec(uint8_t *dst, const int *block_offset,
323  int16_t *block, int stride,
324  const uint8_t nnzc[15 * 8])
325 {
326  int i;
327  for(i=0; i<16; i+=4){
328  int nnz = nnzc[ scan8[i] ];
329  if(nnz){
330  if(nnz==1 && block[i*16]) h264_idct8_dc_add_altivec(dst + block_offset[i], block + i*16, stride);
331  else h264_idct8_add_altivec(dst + block_offset[i], block + i*16, stride);
332  }
333  }
334 }
335 
336 static void h264_idct_add8_altivec(uint8_t **dest, const int *block_offset,
337  int16_t *block, int stride,
338  const uint8_t nnzc[15 * 8])
339 {
340  int i, j;
341  for (j = 1; j < 3; j++) {
342  for(i = j * 16; i < j * 16 + 4; i++){
343  if(nnzc[ scan8[i] ])
344  h264_idct_add_altivec(dest[j-1] + block_offset[i], block + i*16, stride);
345  else if(block[i*16])
346  h264_idct_dc_add_altivec(dest[j-1] + block_offset[i], block + i*16, stride);
347  }
348  }
349 }
350 
351 #define transpose4x16(r0, r1, r2, r3) { \
352  register vec_u8 r4; \
353  register vec_u8 r5; \
354  register vec_u8 r6; \
355  register vec_u8 r7; \
356  \
357  r4 = vec_mergeh(r0, r2); /*0, 2 set 0*/ \
358  r5 = vec_mergel(r0, r2); /*0, 2 set 1*/ \
359  r6 = vec_mergeh(r1, r3); /*1, 3 set 0*/ \
360  r7 = vec_mergel(r1, r3); /*1, 3 set 1*/ \
361  \
362  r0 = vec_mergeh(r4, r6); /*all set 0*/ \
363  r1 = vec_mergel(r4, r6); /*all set 1*/ \
364  r2 = vec_mergeh(r5, r7); /*all set 2*/ \
365  r3 = vec_mergel(r5, r7); /*all set 3*/ \
366 }
367 
368 static inline void write16x4(uint8_t *dst, int dst_stride,
369  register vec_u8 r0, register vec_u8 r1,
370  register vec_u8 r2, register vec_u8 r3) {
371  DECLARE_ALIGNED(16, unsigned char, result)[64];
372  uint32_t *src_int = (uint32_t *)result, *dst_int = (uint32_t *)dst;
373  int int_dst_stride = dst_stride/4;
374 
375  vec_st(r0, 0, result);
376  vec_st(r1, 16, result);
377  vec_st(r2, 32, result);
378  vec_st(r3, 48, result);
379  /* FIXME: there has to be a better way!!!! */
380  *dst_int = *src_int;
381  *(dst_int+ int_dst_stride) = *(src_int + 1);
382  *(dst_int+ 2*int_dst_stride) = *(src_int + 2);
383  *(dst_int+ 3*int_dst_stride) = *(src_int + 3);
384  *(dst_int+ 4*int_dst_stride) = *(src_int + 4);
385  *(dst_int+ 5*int_dst_stride) = *(src_int + 5);
386  *(dst_int+ 6*int_dst_stride) = *(src_int + 6);
387  *(dst_int+ 7*int_dst_stride) = *(src_int + 7);
388  *(dst_int+ 8*int_dst_stride) = *(src_int + 8);
389  *(dst_int+ 9*int_dst_stride) = *(src_int + 9);
390  *(dst_int+10*int_dst_stride) = *(src_int + 10);
391  *(dst_int+11*int_dst_stride) = *(src_int + 11);
392  *(dst_int+12*int_dst_stride) = *(src_int + 12);
393  *(dst_int+13*int_dst_stride) = *(src_int + 13);
394  *(dst_int+14*int_dst_stride) = *(src_int + 14);
395  *(dst_int+15*int_dst_stride) = *(src_int + 15);
396 }
397 
401 #define readAndTranspose16x6(src, src_stride, r8, r9, r10, r11, r12, r13) {\
402  register vec_u8 r0 = unaligned_load(0, src); \
403  register vec_u8 r1 = unaligned_load( src_stride, src); \
404  register vec_u8 r2 = unaligned_load(2* src_stride, src); \
405  register vec_u8 r3 = unaligned_load(3* src_stride, src); \
406  register vec_u8 r4 = unaligned_load(4* src_stride, src); \
407  register vec_u8 r5 = unaligned_load(5* src_stride, src); \
408  register vec_u8 r6 = unaligned_load(6* src_stride, src); \
409  register vec_u8 r7 = unaligned_load(7* src_stride, src); \
410  register vec_u8 r14 = unaligned_load(14*src_stride, src); \
411  register vec_u8 r15 = unaligned_load(15*src_stride, src); \
412  \
413  r8 = unaligned_load( 8*src_stride, src); \
414  r9 = unaligned_load( 9*src_stride, src); \
415  r10 = unaligned_load(10*src_stride, src); \
416  r11 = unaligned_load(11*src_stride, src); \
417  r12 = unaligned_load(12*src_stride, src); \
418  r13 = unaligned_load(13*src_stride, src); \
419  \
420  /*Merge first pairs*/ \
421  r0 = vec_mergeh(r0, r8); /*0, 8*/ \
422  r1 = vec_mergeh(r1, r9); /*1, 9*/ \
423  r2 = vec_mergeh(r2, r10); /*2,10*/ \
424  r3 = vec_mergeh(r3, r11); /*3,11*/ \
425  r4 = vec_mergeh(r4, r12); /*4,12*/ \
426  r5 = vec_mergeh(r5, r13); /*5,13*/ \
427  r6 = vec_mergeh(r6, r14); /*6,14*/ \
428  r7 = vec_mergeh(r7, r15); /*7,15*/ \
429  \
430  /*Merge second pairs*/ \
431  r8 = vec_mergeh(r0, r4); /*0,4, 8,12 set 0*/ \
432  r9 = vec_mergel(r0, r4); /*0,4, 8,12 set 1*/ \
433  r10 = vec_mergeh(r1, r5); /*1,5, 9,13 set 0*/ \
434  r11 = vec_mergel(r1, r5); /*1,5, 9,13 set 1*/ \
435  r12 = vec_mergeh(r2, r6); /*2,6,10,14 set 0*/ \
436  r13 = vec_mergel(r2, r6); /*2,6,10,14 set 1*/ \
437  r14 = vec_mergeh(r3, r7); /*3,7,11,15 set 0*/ \
438  r15 = vec_mergel(r3, r7); /*3,7,11,15 set 1*/ \
439  \
440  /*Third merge*/ \
441  r0 = vec_mergeh(r8, r12); /*0,2,4,6,8,10,12,14 set 0*/ \
442  r1 = vec_mergel(r8, r12); /*0,2,4,6,8,10,12,14 set 1*/ \
443  r2 = vec_mergeh(r9, r13); /*0,2,4,6,8,10,12,14 set 2*/ \
444  r4 = vec_mergeh(r10, r14); /*1,3,5,7,9,11,13,15 set 0*/ \
445  r5 = vec_mergel(r10, r14); /*1,3,5,7,9,11,13,15 set 1*/ \
446  r6 = vec_mergeh(r11, r15); /*1,3,5,7,9,11,13,15 set 2*/ \
447  /* Don't need to compute 3 and 7*/ \
448  \
449  /*Final merge*/ \
450  r8 = vec_mergeh(r0, r4); /*all set 0*/ \
451  r9 = vec_mergel(r0, r4); /*all set 1*/ \
452  r10 = vec_mergeh(r1, r5); /*all set 2*/ \
453  r11 = vec_mergel(r1, r5); /*all set 3*/ \
454  r12 = vec_mergeh(r2, r6); /*all set 4*/ \
455  r13 = vec_mergel(r2, r6); /*all set 5*/ \
456  /* Don't need to compute 14 and 15*/ \
457  \
458 }
459 
460 // out: o = |x-y| < a
461 static inline vec_u8 diff_lt_altivec ( register vec_u8 x,
462  register vec_u8 y,
463  register vec_u8 a) {
464 
465  register vec_u8 diff = vec_subs(x, y);
466  register vec_u8 diffneg = vec_subs(y, x);
467  register vec_u8 o = vec_or(diff, diffneg); /* |x-y| */
468  o = (vec_u8)vec_cmplt(o, a);
469  return o;
470 }
471 
472 static inline vec_u8 h264_deblock_mask ( register vec_u8 p0,
473  register vec_u8 p1,
474  register vec_u8 q0,
475  register vec_u8 q1,
476  register vec_u8 alpha,
477  register vec_u8 beta) {
478 
479  register vec_u8 mask;
480  register vec_u8 tempmask;
481 
482  mask = diff_lt_altivec(p0, q0, alpha);
483  tempmask = diff_lt_altivec(p1, p0, beta);
484  mask = vec_and(mask, tempmask);
485  tempmask = diff_lt_altivec(q1, q0, beta);
486  mask = vec_and(mask, tempmask);
487 
488  return mask;
489 }
490 
491 // out: newp1 = clip((p2 + ((p0 + q0 + 1) >> 1)) >> 1, p1-tc0, p1+tc0)
492 static inline vec_u8 h264_deblock_q1(register vec_u8 p0,
493  register vec_u8 p1,
494  register vec_u8 p2,
495  register vec_u8 q0,
496  register vec_u8 tc0) {
497 
498  register vec_u8 average = vec_avg(p0, q0);
499  register vec_u8 temp;
500  register vec_u8 uncliped;
501  register vec_u8 ones;
502  register vec_u8 max;
503  register vec_u8 min;
504  register vec_u8 newp1;
505 
506  temp = vec_xor(average, p2);
507  average = vec_avg(average, p2); /*avg(p2, avg(p0, q0)) */
508  ones = vec_splat_u8(1);
509  temp = vec_and(temp, ones); /*(p2^avg(p0, q0)) & 1 */
510  uncliped = vec_subs(average, temp); /*(p2+((p0+q0+1)>>1))>>1 */
511  max = vec_adds(p1, tc0);
512  min = vec_subs(p1, tc0);
513  newp1 = vec_max(min, uncliped);
514  newp1 = vec_min(max, newp1);
515  return newp1;
516 }
517 
518 #define h264_deblock_p0_q0(p0, p1, q0, q1, tc0masked) { \
519  \
520  const vec_u8 A0v = vec_sl(vec_splat_u8(10), vec_splat_u8(4)); \
521  \
522  register vec_u8 pq0bit = vec_xor(p0,q0); \
523  register vec_u8 q1minus; \
524  register vec_u8 p0minus; \
525  register vec_u8 stage1; \
526  register vec_u8 stage2; \
527  register vec_u8 vec160; \
528  register vec_u8 delta; \
529  register vec_u8 deltaneg; \
530  \
531  q1minus = vec_nor(q1, q1); /* 255 - q1 */ \
532  stage1 = vec_avg(p1, q1minus); /* (p1 - q1 + 256)>>1 */ \
533  stage2 = vec_sr(stage1, vec_splat_u8(1)); /* (p1 - q1 + 256)>>2 = 64 + (p1 - q1) >> 2 */ \
534  p0minus = vec_nor(p0, p0); /* 255 - p0 */ \
535  stage1 = vec_avg(q0, p0minus); /* (q0 - p0 + 256)>>1 */ \
536  pq0bit = vec_and(pq0bit, vec_splat_u8(1)); \
537  stage2 = vec_avg(stage2, pq0bit); /* 32 + ((q0 - p0)&1 + (p1 - q1) >> 2 + 1) >> 1 */ \
538  stage2 = vec_adds(stage2, stage1); /* 160 + ((p0 - q0) + (p1 - q1) >> 2 + 1) >> 1 */ \
539  vec160 = vec_ld(0, &A0v); \
540  deltaneg = vec_subs(vec160, stage2); /* -d */ \
541  delta = vec_subs(stage2, vec160); /* d */ \
542  deltaneg = vec_min(tc0masked, deltaneg); \
543  delta = vec_min(tc0masked, delta); \
544  p0 = vec_subs(p0, deltaneg); \
545  q0 = vec_subs(q0, delta); \
546  p0 = vec_adds(p0, delta); \
547  q0 = vec_adds(q0, deltaneg); \
548 }
549 
550 #define h264_loop_filter_luma_altivec(p2, p1, p0, q0, q1, q2, alpha, beta, tc0) { \
551  DECLARE_ALIGNED(16, unsigned char, temp)[16]; \
552  register vec_u8 alphavec; \
553  register vec_u8 betavec; \
554  register vec_u8 mask; \
555  register vec_u8 p1mask; \
556  register vec_u8 q1mask; \
557  register vector signed char tc0vec; \
558  register vec_u8 finaltc0; \
559  register vec_u8 tc0masked; \
560  register vec_u8 newp1; \
561  register vec_u8 newq1; \
562  \
563  temp[0] = alpha; \
564  temp[1] = beta; \
565  alphavec = vec_ld(0, temp); \
566  betavec = vec_splat(alphavec, 0x1); \
567  alphavec = vec_splat(alphavec, 0x0); \
568  mask = h264_deblock_mask(p0, p1, q0, q1, alphavec, betavec); /*if in block */ \
569  \
570  AV_COPY32(temp, tc0); \
571  tc0vec = vec_ld(0, (signed char*)temp); \
572  tc0vec = vec_mergeh(tc0vec, tc0vec); \
573  tc0vec = vec_mergeh(tc0vec, tc0vec); \
574  mask = vec_and(mask, vec_cmpgt(tc0vec, vec_splat_s8(-1))); /* if tc0[i] >= 0 */ \
575  finaltc0 = vec_and((vec_u8)tc0vec, mask); /* tc = tc0 */ \
576  \
577  p1mask = diff_lt_altivec(p2, p0, betavec); \
578  p1mask = vec_and(p1mask, mask); /* if ( |p2 - p0| < beta) */ \
579  tc0masked = vec_and(p1mask, (vec_u8)tc0vec); \
580  finaltc0 = vec_sub(finaltc0, p1mask); /* tc++ */ \
581  newp1 = h264_deblock_q1(p0, p1, p2, q0, tc0masked); \
582  /*end if*/ \
583  \
584  q1mask = diff_lt_altivec(q2, q0, betavec); \
585  q1mask = vec_and(q1mask, mask); /* if ( |q2 - q0| < beta ) */\
586  tc0masked = vec_and(q1mask, (vec_u8)tc0vec); \
587  finaltc0 = vec_sub(finaltc0, q1mask); /* tc++ */ \
588  newq1 = h264_deblock_q1(p0, q1, q2, q0, tc0masked); \
589  /*end if*/ \
590  \
591  h264_deblock_p0_q0(p0, p1, q0, q1, finaltc0); \
592  p1 = newp1; \
593  q1 = newq1; \
594 }
595 
596 static void h264_v_loop_filter_luma_altivec(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0) {
597 
598  if ((tc0[0] & tc0[1] & tc0[2] & tc0[3]) >= 0) {
599  register vec_u8 p2 = vec_ld(-3*stride, pix);
600  register vec_u8 p1 = vec_ld(-2*stride, pix);
601  register vec_u8 p0 = vec_ld(-1*stride, pix);
602  register vec_u8 q0 = vec_ld(0, pix);
603  register vec_u8 q1 = vec_ld(stride, pix);
604  register vec_u8 q2 = vec_ld(2*stride, pix);
605  h264_loop_filter_luma_altivec(p2, p1, p0, q0, q1, q2, alpha, beta, tc0);
606  vec_st(p1, -2*stride, pix);
607  vec_st(p0, -1*stride, pix);
608  vec_st(q0, 0, pix);
609  vec_st(q1, stride, pix);
610  }
611 }
612 
613 static void h264_h_loop_filter_luma_altivec(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0) {
614 
615  register vec_u8 line0, line1, line2, line3, line4, line5;
616  if ((tc0[0] & tc0[1] & tc0[2] & tc0[3]) < 0)
617  return;
618  readAndTranspose16x6(pix-3, stride, line0, line1, line2, line3, line4, line5);
619  h264_loop_filter_luma_altivec(line0, line1, line2, line3, line4, line5, alpha, beta, tc0);
620  transpose4x16(line1, line2, line3, line4);
621  write16x4(pix-2, stride, line1, line2, line3, line4);
622 }
623 
624 static av_always_inline
625 void weight_h264_W_altivec(uint8_t *block, int stride, int height,
626  int log2_denom, int weight, int offset, int w)
627 {
628  int y, aligned;
629  vec_u8 vblock;
630  vec_s16 vtemp, vweight, voffset, v0, v1;
631  vec_u16 vlog2_denom;
632  DECLARE_ALIGNED(16, int32_t, temp)[4];
633  LOAD_ZERO;
634 
635  offset <<= log2_denom;
636  if(log2_denom) offset += 1<<(log2_denom-1);
637  temp[0] = log2_denom;
638  temp[1] = weight;
639  temp[2] = offset;
640 
641  vtemp = (vec_s16)vec_ld(0, temp);
642  vlog2_denom = (vec_u16)vec_splat(vtemp, 1);
643  vweight = vec_splat(vtemp, 3);
644  voffset = vec_splat(vtemp, 5);
645  aligned = !((unsigned long)block & 0xf);
646 
647  for (y = 0; y < height; y++) {
648  vblock = vec_ld(0, block);
649 
650  v0 = (vec_s16)vec_mergeh(zero_u8v, vblock);
651  v1 = (vec_s16)vec_mergel(zero_u8v, vblock);
652 
653  if (w == 16 || aligned) {
654  v0 = vec_mladd(v0, vweight, zero_s16v);
655  v0 = vec_adds(v0, voffset);
656  v0 = vec_sra(v0, vlog2_denom);
657  }
658  if (w == 16 || !aligned) {
659  v1 = vec_mladd(v1, vweight, zero_s16v);
660  v1 = vec_adds(v1, voffset);
661  v1 = vec_sra(v1, vlog2_denom);
662  }
663  vblock = vec_packsu(v0, v1);
664  vec_st(vblock, 0, block);
665 
666  block += stride;
667  }
668 }
669 
670 static av_always_inline
671 void biweight_h264_W_altivec(uint8_t *dst, uint8_t *src, int stride, int height,
672  int log2_denom, int weightd, int weights, int offset, int w)
673 {
674  int y, dst_aligned, src_aligned;
675  vec_u8 vsrc, vdst;
676  vec_s16 vtemp, vweights, vweightd, voffset, v0, v1, v2, v3;
677  vec_u16 vlog2_denom;
678  DECLARE_ALIGNED(16, int32_t, temp)[4];
679  LOAD_ZERO;
680 
681  offset = ((offset + 1) | 1) << log2_denom;
682  temp[0] = log2_denom+1;
683  temp[1] = weights;
684  temp[2] = weightd;
685  temp[3] = offset;
686 
687  vtemp = (vec_s16)vec_ld(0, temp);
688  vlog2_denom = (vec_u16)vec_splat(vtemp, 1);
689  vweights = vec_splat(vtemp, 3);
690  vweightd = vec_splat(vtemp, 5);
691  voffset = vec_splat(vtemp, 7);
692  dst_aligned = !((unsigned long)dst & 0xf);
693  src_aligned = !((unsigned long)src & 0xf);
694 
695  for (y = 0; y < height; y++) {
696  vdst = vec_ld(0, dst);
697  vsrc = vec_ld(0, src);
698 
699  v0 = (vec_s16)vec_mergeh(zero_u8v, vdst);
700  v1 = (vec_s16)vec_mergel(zero_u8v, vdst);
701  v2 = (vec_s16)vec_mergeh(zero_u8v, vsrc);
702  v3 = (vec_s16)vec_mergel(zero_u8v, vsrc);
703 
704  if (w == 8) {
705  if (src_aligned)
706  v3 = v2;
707  else
708  v2 = v3;
709  }
710 
711  if (w == 16 || dst_aligned) {
712  v0 = vec_mladd(v0, vweightd, zero_s16v);
713  v2 = vec_mladd(v2, vweights, zero_s16v);
714 
715  v0 = vec_adds(v0, voffset);
716  v0 = vec_adds(v0, v2);
717  v0 = vec_sra(v0, vlog2_denom);
718  }
719  if (w == 16 || !dst_aligned) {
720  v1 = vec_mladd(v1, vweightd, zero_s16v);
721  v3 = vec_mladd(v3, vweights, zero_s16v);
722 
723  v1 = vec_adds(v1, voffset);
724  v1 = vec_adds(v1, v3);
725  v1 = vec_sra(v1, vlog2_denom);
726  }
727  vdst = vec_packsu(v0, v1);
728  vec_st(vdst, 0, dst);
729 
730  dst += stride;
731  src += stride;
732  }
733 }
734 
735 #define H264_WEIGHT(W) \
736 static void weight_h264_pixels ## W ## _altivec(uint8_t *block, int stride, int height, \
737  int log2_denom, int weight, int offset) \
738 { \
739  weight_h264_W_altivec(block, stride, height, log2_denom, weight, offset, W); \
740 }\
741 static void biweight_h264_pixels ## W ## _altivec(uint8_t *dst, uint8_t *src, int stride, int height, \
742  int log2_denom, int weightd, int weights, int offset) \
743 { \
744  biweight_h264_W_altivec(dst, src, stride, height, log2_denom, weightd, weights, offset, W); \
745 }
746 
747 H264_WEIGHT(16)
748 H264_WEIGHT( 8)
749 #endif /* HAVE_ALTIVEC */
750 
751 av_cold void ff_h264dsp_init_ppc(H264DSPContext *c, const int bit_depth,
752  const int chroma_format_idc)
753 {
754 #if HAVE_ALTIVEC && HAVE_BIGENDIAN
756  return;
757 
758  if (bit_depth == 8) {
759  c->h264_idct_add = h264_idct_add_altivec;
760  if (chroma_format_idc <= 1)
761  c->h264_idct_add8 = h264_idct_add8_altivec;
762  c->h264_idct_add16 = h264_idct_add16_altivec;
763  c->h264_idct_add16intra = h264_idct_add16intra_altivec;
764  c->h264_idct_dc_add= h264_idct_dc_add_altivec;
765  c->h264_idct8_dc_add = h264_idct8_dc_add_altivec;
766  c->h264_idct8_add = h264_idct8_add_altivec;
767  c->h264_idct8_add4 = h264_idct8_add4_altivec;
768  c->h264_v_loop_filter_luma= h264_v_loop_filter_luma_altivec;
769  c->h264_h_loop_filter_luma= h264_h_loop_filter_luma_altivec;
770 
771  c->weight_h264_pixels_tab[0] = weight_h264_pixels16_altivec;
772  c->weight_h264_pixels_tab[1] = weight_h264_pixels8_altivec;
773  c->biweight_h264_pixels_tab[0] = biweight_h264_pixels16_altivec;
774  c->biweight_h264_pixels_tab[1] = biweight_h264_pixels8_altivec;
775  }
776 #endif /* HAVE_ALTIVEC */
777 }
int size
memory handling functions
#define DECLARE_ALIGNED(n, t, v)
Definition: mem.h:58
int stride
Definition: mace.c:144
H.264 DSP functions.
Macro definitions for various function/variable attributes.
static int16_t block[64]
Definition: dct.c:97
uint8_t
#define av_cold
Definition: attributes.h:66
h264_weight_func weight_h264_pixels_tab[4]
Definition: h264dsp.h:43
void(* h264_idct_add16intra)(uint8_t *dst, const int *blockoffset, int16_t *block, int stride, const uint8_t nnzc[15 *8])
Definition: h264dsp.h:98
#define vec_s16
Definition: types_altivec.h:30
#define zero_s16v
Definition: types_altivec.h:43
void(* h264_idct_add)(uint8_t *dst, int16_t *block, int stride)
Definition: h264dsp.h:80
void(* h264_idct8_dc_add)(uint8_t *dst, int16_t *block, int stride)
Definition: h264dsp.h:86
#define src
Definition: vp8dsp.c:254
static const uint16_t mask[17]
Definition: lzw.c:38
#define PPC_ALTIVEC(flags)
Definition: cpu.h:26
#define vec_u16
Definition: types_altivec.h:29
#define H264_WEIGHT(W)
#define LOAD_ZERO
Definition: types_altivec.h:38
static void idct6(int pre_mant[6])
Calculate 6-point IDCT of the pre-mantissas.
Definition: eac3dec.c:168
void(* h264_idct_add16)(uint8_t *dst, const int *blockoffset, int16_t *block, int stride, const uint8_t nnzc[15 *8])
Definition: h264dsp.h:89
h264_biweight_func biweight_h264_pixels_tab[4]
Definition: h264dsp.h:44
Context for storing H.264 DSP functions.
Definition: h264dsp.h:41
int32_t
#define vec_u32
Definition: types_altivec.h:31
H.264 / AVC / MPEG-4 part10 codec.
void(* h264_v_loop_filter_luma)(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0)
Definition: h264dsp.h:47
av_cold void ff_h264dsp_init_ppc(H264DSPContext *c, const int bit_depth, const int chroma_format_idc)
Definition: h264dsp.c:751
void(* h264_idct8_add)(uint8_t *dst, int16_t *block, int stride)
Definition: h264dsp.h:82
#define vec_u8
Definition: types_altivec.h:27
static const uint8_t scan8[16 *3+3]
Definition: h264dec.h:622
void(* h264_idct8_add4)(uint8_t *dst, const int *blockoffset, int16_t *block, int stride, const uint8_t nnzc[15 *8])
Definition: h264dsp.h:92
int av_get_cpu_flags(void)
Return the flags which specify extensions supported by the CPU.
Definition: cpu.c:47
uint8_t pi<< 24) CONV_FUNC_GROUP(AV_SAMPLE_FMT_FLT, float, AV_SAMPLE_FMT_U8, uint8_t,(*(const uint8_t *) pi - 0x80) *(1.0f/(1<< 7))) CONV_FUNC_GROUP(AV_SAMPLE_FMT_DBL, double, AV_SAMPLE_FMT_U8, uint8_t,(*(const uint8_t *) pi - 0x80) *(1.0/(1<< 7))) CONV_FUNC_GROUP(AV_SAMPLE_FMT_U8, uint8_t, AV_SAMPLE_FMT_S16, int16_t,(*(const int16_t *) pi >> 8)+0x80) CONV_FUNC_GROUP(AV_SAMPLE_FMT_FLT, float, AV_SAMPLE_FMT_S16, int16_t, *(const int16_t *) pi *(1.0f/(1<< 15))) CONV_FUNC_GROUP(AV_SAMPLE_FMT_DBL, double, AV_SAMPLE_FMT_S16, int16_t, *(const int16_t *) pi *(1.0/(1<< 15))) CONV_FUNC_GROUP(AV_SAMPLE_FMT_U8, uint8_t, AV_SAMPLE_FMT_S32, int32_t,(*(const int32_t *) pi >> 24)+0x80) CONV_FUNC_GROUP(AV_SAMPLE_FMT_FLT, float, AV_SAMPLE_FMT_S32, int32_t, *(const int32_t *) pi *(1.0f/(1U<< 31))) CONV_FUNC_GROUP(AV_SAMPLE_FMT_DBL, double, AV_SAMPLE_FMT_S32, int32_t, *(const int32_t *) pi *(1.0/(1U<< 31))) CONV_FUNC_GROUP(AV_SAMPLE_FMT_U8, uint8_t, AV_SAMPLE_FMT_FLT, float, av_clip_uint8(lrintf(*(const float *) pi *(1<< 7))+0x80)) CONV_FUNC_GROUP(AV_SAMPLE_FMT_S16, int16_t, AV_SAMPLE_FMT_FLT, float, av_clip_int16(lrintf(*(const float *) pi *(1<< 15)))) CONV_FUNC_GROUP(AV_SAMPLE_FMT_S32, int32_t, AV_SAMPLE_FMT_FLT, float, av_clipl_int32(llrintf(*(const float *) pi *(1U<< 31)))) CONV_FUNC_GROUP(AV_SAMPLE_FMT_U8, uint8_t, AV_SAMPLE_FMT_DBL, double, av_clip_uint8(lrint(*(const double *) pi *(1<< 7))+0x80)) CONV_FUNC_GROUP(AV_SAMPLE_FMT_S16, int16_t, AV_SAMPLE_FMT_DBL, double, av_clip_int16(lrint(*(const double *) pi *(1<< 15)))) CONV_FUNC_GROUP(AV_SAMPLE_FMT_S32, int32_t, AV_SAMPLE_FMT_DBL, double, av_clipl_int32(llrint(*(const double *) pi *(1U<< 31)))) #define SET_CONV_FUNC_GROUP(ofmt, ifmt) static void set_generic_function(AudioConvert *ac) { } void ff_audio_convert_free(AudioConvert **ac) { if(! *ac) return;ff_dither_free(&(*ac) ->dc);av_freep(ac);} AudioConvert *ff_audio_convert_alloc(AVAudioResampleContext *avr, enum AVSampleFormat out_fmt, enum AVSampleFormat in_fmt, int channels, int sample_rate, int apply_map) { AudioConvert *ac;int in_planar, out_planar;ac=av_mallocz(sizeof(*ac));if(!ac) return NULL;ac->avr=avr;ac->out_fmt=out_fmt;ac->in_fmt=in_fmt;ac->channels=channels;ac->apply_map=apply_map;if(avr->dither_method !=AV_RESAMPLE_DITHER_NONE &&av_get_packed_sample_fmt(out_fmt)==AV_SAMPLE_FMT_S16 &&av_get_bytes_per_sample(in_fmt) > 2) { ac->dc=ff_dither_alloc(avr, out_fmt, in_fmt, channels, sample_rate, apply_map);if(!ac->dc) { av_free(ac);return NULL;} return ac;} in_planar=ff_sample_fmt_is_planar(in_fmt, channels);out_planar=ff_sample_fmt_is_planar(out_fmt, channels);if(in_planar==out_planar) { ac->func_type=CONV_FUNC_TYPE_FLAT;ac->planes=in_planar ? ac->channels :1;} else if(in_planar) ac->func_type=CONV_FUNC_TYPE_INTERLEAVE;else ac->func_type=CONV_FUNC_TYPE_DEINTERLEAVE;set_generic_function(ac);if(ARCH_AARCH64) ff_audio_convert_init_aarch64(ac);if(ARCH_ARM) ff_audio_convert_init_arm(ac);if(ARCH_X86) ff_audio_convert_init_x86(ac);return ac;} int ff_audio_convert(AudioConvert *ac, AudioData *out, AudioData *in) { int use_generic=1;int len=in->nb_samples;int p;if(ac->dc) { av_log(ac->avr, AV_LOG_TRACE, "%d samples - audio_convert: %s to %s (dithered)\", len, av_get_sample_fmt_name(ac->in_fmt), av_get_sample_fmt_name(ac->out_fmt));return ff_convert_dither(ac-> dc
Contains misc utility macros and inline functions.
void(* h264_h_loop_filter_luma)(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0)
Definition: h264dsp.h:49
int height
Definition: gxfenc.c:72
void(* h264_idct_dc_add)(uint8_t *dst, int16_t *block, int stride)
Definition: h264dsp.h:84
#define av_always_inline
Definition: attributes.h:40
void(* h264_idct_add8)(uint8_t **dst, const int *blockoffset, int16_t *block, int stride, const uint8_t nnzc[15 *8])
Definition: h264dsp.h:95
float min
#define zero_u8v
Definition: types_altivec.h:40