FFmpeg
 All Data Structures Namespaces Files Functions Variables Typedefs Enumerations Enumerator Macros Groups Pages
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 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 #include "libavutil/attributes.h"
23 #include "libavutil/cpu.h"
24 #include "libavutil/intreadwrite.h"
25 #include "libavutil/ppc/types_altivec.h"
26 #include "libavutil/ppc/util_altivec.h"
27 #include "libavcodec/h264data.h"
28 #include "libavcodec/h264dsp.h"
29 
30 #if HAVE_ALTIVEC
31 
32 /****************************************************************************
33  * IDCT transform:
34  ****************************************************************************/
35 
36 #define VEC_1D_DCT(vb0,vb1,vb2,vb3,va0,va1,va2,va3) \
37  /* 1st stage */ \
38  vz0 = vec_add(vb0,vb2); /* temp[0] = Y[0] + Y[2] */ \
39  vz1 = vec_sub(vb0,vb2); /* temp[1] = Y[0] - Y[2] */ \
40  vz2 = vec_sra(vb1,vec_splat_u16(1)); \
41  vz2 = vec_sub(vz2,vb3); /* temp[2] = Y[1].1/2 - Y[3] */ \
42  vz3 = vec_sra(vb3,vec_splat_u16(1)); \
43  vz3 = vec_add(vb1,vz3); /* temp[3] = Y[1] + Y[3].1/2 */ \
44  /* 2nd stage: output */ \
45  va0 = vec_add(vz0,vz3); /* x[0] = temp[0] + temp[3] */ \
46  va1 = vec_add(vz1,vz2); /* x[1] = temp[1] + temp[2] */ \
47  va2 = vec_sub(vz1,vz2); /* x[2] = temp[1] - temp[2] */ \
48  va3 = vec_sub(vz0,vz3) /* x[3] = temp[0] - temp[3] */
49 
50 #define VEC_TRANSPOSE_4(a0,a1,a2,a3,b0,b1,b2,b3) \
51  b0 = vec_mergeh( a0, a0 ); \
52  b1 = vec_mergeh( a1, a0 ); \
53  b2 = vec_mergeh( a2, a0 ); \
54  b3 = vec_mergeh( a3, a0 ); \
55  a0 = vec_mergeh( b0, b2 ); \
56  a1 = vec_mergel( b0, b2 ); \
57  a2 = vec_mergeh( b1, b3 ); \
58  a3 = vec_mergel( b1, b3 ); \
59  b0 = vec_mergeh( a0, a2 ); \
60  b1 = vec_mergel( a0, a2 ); \
61  b2 = vec_mergeh( a1, a3 ); \
62  b3 = vec_mergel( a1, a3 )
63 
64 #define VEC_LOAD_U8_ADD_S16_STORE_U8(va) \
65  vdst_orig = vec_ld(0, dst); \
66  vdst = vec_perm(vdst_orig, zero_u8v, vdst_mask); \
67  vdst_ss = (vec_s16) vec_mergeh(zero_u8v, vdst); \
68  va = vec_add(va, vdst_ss); \
69  va_u8 = vec_packsu(va, zero_s16v); \
70  va_u32 = vec_splat((vec_u32)va_u8, 0); \
71  vec_ste(va_u32, element, (uint32_t*)dst);
72 
73 static void h264_idct_add_altivec(uint8_t *dst, int16_t *block, int stride)
74 {
75  vec_s16 va0, va1, va2, va3;
76  vec_s16 vz0, vz1, vz2, vz3;
77  vec_s16 vtmp0, vtmp1, vtmp2, vtmp3;
78  vec_u8 va_u8;
79  vec_u32 va_u32;
80  vec_s16 vdst_ss;
81  const vec_u16 v6us = vec_splat_u16(6);
82  vec_u8 vdst, vdst_orig;
83  vec_u8 vdst_mask = vec_lvsl(0, dst);
84  int element = ((unsigned long)dst & 0xf) >> 2;
85  LOAD_ZERO;
86 
87  block[0] += 32; /* add 32 as a DC-level for rounding */
88 
89  vtmp0 = vec_ld(0,block);
90  vtmp1 = vec_sld(vtmp0, vtmp0, 8);
91  vtmp2 = vec_ld(16,block);
92  vtmp3 = vec_sld(vtmp2, vtmp2, 8);
93  memset(block, 0, 16 * sizeof(int16_t));
94 
95  VEC_1D_DCT(vtmp0,vtmp1,vtmp2,vtmp3,va0,va1,va2,va3);
96  VEC_TRANSPOSE_4(va0,va1,va2,va3,vtmp0,vtmp1,vtmp2,vtmp3);
97  VEC_1D_DCT(vtmp0,vtmp1,vtmp2,vtmp3,va0,va1,va2,va3);
98 
99  va0 = vec_sra(va0,v6us);
100  va1 = vec_sra(va1,v6us);
101  va2 = vec_sra(va2,v6us);
102  va3 = vec_sra(va3,v6us);
103 
104  VEC_LOAD_U8_ADD_S16_STORE_U8(va0);
105  dst += stride;
106  VEC_LOAD_U8_ADD_S16_STORE_U8(va1);
107  dst += stride;
108  VEC_LOAD_U8_ADD_S16_STORE_U8(va2);
109  dst += stride;
110  VEC_LOAD_U8_ADD_S16_STORE_U8(va3);
111 }
112 
113 #define IDCT8_1D_ALTIVEC(s0, s1, s2, s3, s4, s5, s6, s7, d0, d1, d2, d3, d4, d5, d6, d7) {\
114  /* a0 = SRC(0) + SRC(4); */ \
115  vec_s16 a0v = vec_add(s0, s4); \
116  /* a2 = SRC(0) - SRC(4); */ \
117  vec_s16 a2v = vec_sub(s0, s4); \
118  /* a4 = (SRC(2)>>1) - SRC(6); */ \
119  vec_s16 a4v = vec_sub(vec_sra(s2, onev), s6); \
120  /* a6 = (SRC(6)>>1) + SRC(2); */ \
121  vec_s16 a6v = vec_add(vec_sra(s6, onev), s2); \
122  /* b0 = a0 + a6; */ \
123  vec_s16 b0v = vec_add(a0v, a6v); \
124  /* b2 = a2 + a4; */ \
125  vec_s16 b2v = vec_add(a2v, a4v); \
126  /* b4 = a2 - a4; */ \
127  vec_s16 b4v = vec_sub(a2v, a4v); \
128  /* b6 = a0 - a6; */ \
129  vec_s16 b6v = vec_sub(a0v, a6v); \
130  /* a1 = SRC(5) - SRC(3) - SRC(7) - (SRC(7)>>1); */ \
131  /* a1 = (SRC(5)-SRC(3)) - (SRC(7) + (SRC(7)>>1)); */ \
132  vec_s16 a1v = vec_sub( vec_sub(s5, s3), vec_add(s7, vec_sra(s7, onev)) ); \
133  /* a3 = SRC(7) + SRC(1) - SRC(3) - (SRC(3)>>1); */ \
134  /* a3 = (SRC(7)+SRC(1)) - (SRC(3) + (SRC(3)>>1)); */ \
135  vec_s16 a3v = vec_sub( vec_add(s7, s1), vec_add(s3, vec_sra(s3, onev)) );\
136  /* a5 = SRC(7) - SRC(1) + SRC(5) + (SRC(5)>>1); */ \
137  /* a5 = (SRC(7)-SRC(1)) + SRC(5) + (SRC(5)>>1); */ \
138  vec_s16 a5v = vec_add( vec_sub(s7, s1), vec_add(s5, vec_sra(s5, onev)) );\
139  /* a7 = SRC(5)+SRC(3) + SRC(1) + (SRC(1)>>1); */ \
140  vec_s16 a7v = vec_add( vec_add(s5, s3), vec_add(s1, vec_sra(s1, onev)) );\
141  /* b1 = (a7>>2) + a1; */ \
142  vec_s16 b1v = vec_add( vec_sra(a7v, twov), a1v); \
143  /* b3 = a3 + (a5>>2); */ \
144  vec_s16 b3v = vec_add(a3v, vec_sra(a5v, twov)); \
145  /* b5 = (a3>>2) - a5; */ \
146  vec_s16 b5v = vec_sub( vec_sra(a3v, twov), a5v); \
147  /* b7 = a7 - (a1>>2); */ \
148  vec_s16 b7v = vec_sub( a7v, vec_sra(a1v, twov)); \
149  /* DST(0, b0 + b7); */ \
150  d0 = vec_add(b0v, b7v); \
151  /* DST(1, b2 + b5); */ \
152  d1 = vec_add(b2v, b5v); \
153  /* DST(2, b4 + b3); */ \
154  d2 = vec_add(b4v, b3v); \
155  /* DST(3, b6 + b1); */ \
156  d3 = vec_add(b6v, b1v); \
157  /* DST(4, b6 - b1); */ \
158  d4 = vec_sub(b6v, b1v); \
159  /* DST(5, b4 - b3); */ \
160  d5 = vec_sub(b4v, b3v); \
161  /* DST(6, b2 - b5); */ \
162  d6 = vec_sub(b2v, b5v); \
163  /* DST(7, b0 - b7); */ \
164  d7 = vec_sub(b0v, b7v); \
165 }
166 
167 #define ALTIVEC_STORE_SUM_CLIP(dest, idctv, perm_ldv, perm_stv, sel) { \
168  /* unaligned load */ \
169  vec_u8 hv = vec_ld( 0, dest ); \
170  vec_u8 lv = vec_ld( 7, dest ); \
171  vec_u8 dstv = vec_perm( hv, lv, (vec_u8)perm_ldv ); \
172  vec_s16 idct_sh6 = vec_sra(idctv, sixv); \
173  vec_u16 dst16 = (vec_u16)vec_mergeh(zero_u8v, dstv); \
174  vec_s16 idstsum = vec_adds(idct_sh6, (vec_s16)dst16); \
175  vec_u8 idstsum8 = vec_packsu(zero_s16v, idstsum); \
176  vec_u8 edgehv; \
177  /* unaligned store */ \
178  vec_u8 bodyv = vec_perm( idstsum8, idstsum8, perm_stv );\
179  vec_u8 edgelv = vec_perm( sel, zero_u8v, perm_stv ); \
180  lv = vec_sel( lv, bodyv, edgelv ); \
181  vec_st( lv, 7, dest ); \
182  hv = vec_ld( 0, dest ); \
183  edgehv = vec_perm( zero_u8v, sel, perm_stv ); \
184  hv = vec_sel( hv, bodyv, edgehv ); \
185  vec_st( hv, 0, dest ); \
186  }
187 
188 static void h264_idct8_add_altivec(uint8_t *dst, int16_t *dct, int stride)
189 {
190  vec_s16 s0, s1, s2, s3, s4, s5, s6, s7;
191  vec_s16 d0, d1, d2, d3, d4, d5, d6, d7;
192  vec_s16 idct0, idct1, idct2, idct3, idct4, idct5, idct6, idct7;
193 
194  vec_u8 perm_ldv = vec_lvsl(0, dst);
195  vec_u8 perm_stv = vec_lvsr(8, dst);
196 
197  const vec_u16 onev = vec_splat_u16(1);
198  const vec_u16 twov = vec_splat_u16(2);
199  const vec_u16 sixv = vec_splat_u16(6);
200 
201  const vec_u8 sel = (vec_u8) {0,0,0,0,0,0,0,0,-1,-1,-1,-1,-1,-1,-1,-1};
202  LOAD_ZERO;
203 
204  dct[0] += 32; // rounding for the >>6 at the end
205 
206  s0 = vec_ld(0x00, (int16_t*)dct);
207  s1 = vec_ld(0x10, (int16_t*)dct);
208  s2 = vec_ld(0x20, (int16_t*)dct);
209  s3 = vec_ld(0x30, (int16_t*)dct);
210  s4 = vec_ld(0x40, (int16_t*)dct);
211  s5 = vec_ld(0x50, (int16_t*)dct);
212  s6 = vec_ld(0x60, (int16_t*)dct);
213  s7 = vec_ld(0x70, (int16_t*)dct);
214  memset(dct, 0, 64 * sizeof(int16_t));
215 
216  IDCT8_1D_ALTIVEC(s0, s1, s2, s3, s4, s5, s6, s7,
217  d0, d1, d2, d3, d4, d5, d6, d7);
218 
219  TRANSPOSE8( d0, d1, d2, d3, d4, d5, d6, d7 );
220 
221  IDCT8_1D_ALTIVEC(d0, d1, d2, d3, d4, d5, d6, d7,
222  idct0, idct1, idct2, idct3, idct4, idct5, idct6, idct7);
223 
224  ALTIVEC_STORE_SUM_CLIP(&dst[0*stride], idct0, perm_ldv, perm_stv, sel);
225  ALTIVEC_STORE_SUM_CLIP(&dst[1*stride], idct1, perm_ldv, perm_stv, sel);
226  ALTIVEC_STORE_SUM_CLIP(&dst[2*stride], idct2, perm_ldv, perm_stv, sel);
227  ALTIVEC_STORE_SUM_CLIP(&dst[3*stride], idct3, perm_ldv, perm_stv, sel);
228  ALTIVEC_STORE_SUM_CLIP(&dst[4*stride], idct4, perm_ldv, perm_stv, sel);
229  ALTIVEC_STORE_SUM_CLIP(&dst[5*stride], idct5, perm_ldv, perm_stv, sel);
230  ALTIVEC_STORE_SUM_CLIP(&dst[6*stride], idct6, perm_ldv, perm_stv, sel);
231  ALTIVEC_STORE_SUM_CLIP(&dst[7*stride], idct7, perm_ldv, perm_stv, sel);
232 }
233 
234 static av_always_inline void h264_idct_dc_add_internal(uint8_t *dst, int16_t *block, int stride, int size)
235 {
236  vec_s16 dc16;
237  vec_u8 dcplus, dcminus, v0, v1, v2, v3, aligner;
238  LOAD_ZERO;
239  DECLARE_ALIGNED(16, int, dc);
240  int i;
241 
242  dc = (block[0] + 32) >> 6;
243  block[0] = 0;
244  dc16 = vec_splat((vec_s16) vec_lde(0, &dc), 1);
245 
246  if (size == 4)
247  dc16 = vec_sld(dc16, zero_s16v, 8);
248  dcplus = vec_packsu(dc16, zero_s16v);
249  dcminus = vec_packsu(vec_sub(zero_s16v, dc16), zero_s16v);
250 
251  aligner = vec_lvsr(0, dst);
252  dcplus = vec_perm(dcplus, dcplus, aligner);
253  dcminus = vec_perm(dcminus, dcminus, aligner);
254 
255  for (i = 0; i < size; i += 4) {
256  v0 = vec_ld(0, dst+0*stride);
257  v1 = vec_ld(0, dst+1*stride);
258  v2 = vec_ld(0, dst+2*stride);
259  v3 = vec_ld(0, dst+3*stride);
260 
261  v0 = vec_adds(v0, dcplus);
262  v1 = vec_adds(v1, dcplus);
263  v2 = vec_adds(v2, dcplus);
264  v3 = vec_adds(v3, dcplus);
265 
266  v0 = vec_subs(v0, dcminus);
267  v1 = vec_subs(v1, dcminus);
268  v2 = vec_subs(v2, dcminus);
269  v3 = vec_subs(v3, dcminus);
270 
271  vec_st(v0, 0, dst+0*stride);
272  vec_st(v1, 0, dst+1*stride);
273  vec_st(v2, 0, dst+2*stride);
274  vec_st(v3, 0, dst+3*stride);
275 
276  dst += 4*stride;
277  }
278 }
279 
280 static void h264_idct_dc_add_altivec(uint8_t *dst, int16_t *block, int stride)
281 {
282  h264_idct_dc_add_internal(dst, block, stride, 4);
283 }
284 
285 static void h264_idct8_dc_add_altivec(uint8_t *dst, int16_t *block, int stride)
286 {
287  h264_idct_dc_add_internal(dst, block, stride, 8);
288 }
289 
290 static void h264_idct_add16_altivec(uint8_t *dst, const int *block_offset,
291  int16_t *block, int stride,
292  const uint8_t nnzc[15 * 8])
293 {
294  int i;
295  for(i=0; i<16; i++){
296  int nnz = nnzc[ scan8[i] ];
297  if(nnz){
298  if(nnz==1 && block[i*16]) h264_idct_dc_add_altivec(dst + block_offset[i], block + i*16, stride);
299  else h264_idct_add_altivec(dst + block_offset[i], block + i*16, stride);
300  }
301  }
302 }
303 
304 static void h264_idct_add16intra_altivec(uint8_t *dst, const int *block_offset,
305  int16_t *block, int stride,
306  const uint8_t nnzc[15 * 8])
307 {
308  int i;
309  for(i=0; i<16; i++){
310  if(nnzc[ scan8[i] ]) h264_idct_add_altivec(dst + block_offset[i], block + i*16, stride);
311  else if(block[i*16]) h264_idct_dc_add_altivec(dst + block_offset[i], block + i*16, stride);
312  }
313 }
314 
315 static void h264_idct8_add4_altivec(uint8_t *dst, const int *block_offset,
316  int16_t *block, int stride,
317  const uint8_t nnzc[15 * 8])
318 {
319  int i;
320  for(i=0; i<16; i+=4){
321  int nnz = nnzc[ scan8[i] ];
322  if(nnz){
323  if(nnz==1 && block[i*16]) h264_idct8_dc_add_altivec(dst + block_offset[i], block + i*16, stride);
324  else h264_idct8_add_altivec(dst + block_offset[i], block + i*16, stride);
325  }
326  }
327 }
328 
329 static void h264_idct_add8_altivec(uint8_t **dest, const int *block_offset,
330  int16_t *block, int stride,
331  const uint8_t nnzc[15 * 8])
332 {
333  int i, j;
334  for (j = 1; j < 3; j++) {
335  for(i = j * 16; i < j * 16 + 4; i++){
336  if(nnzc[ scan8[i] ])
337  h264_idct_add_altivec(dest[j-1] + block_offset[i], block + i*16, stride);
338  else if(block[i*16])
339  h264_idct_dc_add_altivec(dest[j-1] + block_offset[i], block + i*16, stride);
340  }
341  }
342 }
343 
344 #define transpose4x16(r0, r1, r2, r3) { \
345  register vec_u8 r4; \
346  register vec_u8 r5; \
347  register vec_u8 r6; \
348  register vec_u8 r7; \
349  \
350  r4 = vec_mergeh(r0, r2); /*0, 2 set 0*/ \
351  r5 = vec_mergel(r0, r2); /*0, 2 set 1*/ \
352  r6 = vec_mergeh(r1, r3); /*1, 3 set 0*/ \
353  r7 = vec_mergel(r1, r3); /*1, 3 set 1*/ \
354  \
355  r0 = vec_mergeh(r4, r6); /*all set 0*/ \
356  r1 = vec_mergel(r4, r6); /*all set 1*/ \
357  r2 = vec_mergeh(r5, r7); /*all set 2*/ \
358  r3 = vec_mergel(r5, r7); /*all set 3*/ \
359 }
360 
361 static inline void write16x4(uint8_t *dst, int dst_stride,
362  register vec_u8 r0, register vec_u8 r1,
363  register vec_u8 r2, register vec_u8 r3) {
364  DECLARE_ALIGNED(16, unsigned char, result)[64];
365  uint32_t *src_int = (uint32_t *)result, *dst_int = (uint32_t *)dst;
366  int int_dst_stride = dst_stride/4;
367 
368  vec_st(r0, 0, result);
369  vec_st(r1, 16, result);
370  vec_st(r2, 32, result);
371  vec_st(r3, 48, result);
372  /* FIXME: there has to be a better way!!!! */
373  *dst_int = *src_int;
374  *(dst_int+ int_dst_stride) = *(src_int + 1);
375  *(dst_int+ 2*int_dst_stride) = *(src_int + 2);
376  *(dst_int+ 3*int_dst_stride) = *(src_int + 3);
377  *(dst_int+ 4*int_dst_stride) = *(src_int + 4);
378  *(dst_int+ 5*int_dst_stride) = *(src_int + 5);
379  *(dst_int+ 6*int_dst_stride) = *(src_int + 6);
380  *(dst_int+ 7*int_dst_stride) = *(src_int + 7);
381  *(dst_int+ 8*int_dst_stride) = *(src_int + 8);
382  *(dst_int+ 9*int_dst_stride) = *(src_int + 9);
383  *(dst_int+10*int_dst_stride) = *(src_int + 10);
384  *(dst_int+11*int_dst_stride) = *(src_int + 11);
385  *(dst_int+12*int_dst_stride) = *(src_int + 12);
386  *(dst_int+13*int_dst_stride) = *(src_int + 13);
387  *(dst_int+14*int_dst_stride) = *(src_int + 14);
388  *(dst_int+15*int_dst_stride) = *(src_int + 15);
389 }
390 
391 /** @brief performs a 6x16 transpose of data in src, and stores it to dst
392  @todo FIXME: see if we can't spare some vec_lvsl() by them factorizing
393  out of unaligned_load() */
394 #define readAndTranspose16x6(src, src_stride, r8, r9, r10, r11, r12, r13) {\
395  register vec_u8 r0 = unaligned_load(0, src); \
396  register vec_u8 r1 = unaligned_load( src_stride, src); \
397  register vec_u8 r2 = unaligned_load(2* src_stride, src); \
398  register vec_u8 r3 = unaligned_load(3* src_stride, src); \
399  register vec_u8 r4 = unaligned_load(4* src_stride, src); \
400  register vec_u8 r5 = unaligned_load(5* src_stride, src); \
401  register vec_u8 r6 = unaligned_load(6* src_stride, src); \
402  register vec_u8 r7 = unaligned_load(7* src_stride, src); \
403  register vec_u8 r14 = unaligned_load(14*src_stride, src); \
404  register vec_u8 r15 = unaligned_load(15*src_stride, src); \
405  \
406  r8 = unaligned_load( 8*src_stride, src); \
407  r9 = unaligned_load( 9*src_stride, src); \
408  r10 = unaligned_load(10*src_stride, src); \
409  r11 = unaligned_load(11*src_stride, src); \
410  r12 = unaligned_load(12*src_stride, src); \
411  r13 = unaligned_load(13*src_stride, src); \
412  \
413  /*Merge first pairs*/ \
414  r0 = vec_mergeh(r0, r8); /*0, 8*/ \
415  r1 = vec_mergeh(r1, r9); /*1, 9*/ \
416  r2 = vec_mergeh(r2, r10); /*2,10*/ \
417  r3 = vec_mergeh(r3, r11); /*3,11*/ \
418  r4 = vec_mergeh(r4, r12); /*4,12*/ \
419  r5 = vec_mergeh(r5, r13); /*5,13*/ \
420  r6 = vec_mergeh(r6, r14); /*6,14*/ \
421  r7 = vec_mergeh(r7, r15); /*7,15*/ \
422  \
423  /*Merge second pairs*/ \
424  r8 = vec_mergeh(r0, r4); /*0,4, 8,12 set 0*/ \
425  r9 = vec_mergel(r0, r4); /*0,4, 8,12 set 1*/ \
426  r10 = vec_mergeh(r1, r5); /*1,5, 9,13 set 0*/ \
427  r11 = vec_mergel(r1, r5); /*1,5, 9,13 set 1*/ \
428  r12 = vec_mergeh(r2, r6); /*2,6,10,14 set 0*/ \
429  r13 = vec_mergel(r2, r6); /*2,6,10,14 set 1*/ \
430  r14 = vec_mergeh(r3, r7); /*3,7,11,15 set 0*/ \
431  r15 = vec_mergel(r3, r7); /*3,7,11,15 set 1*/ \
432  \
433  /*Third merge*/ \
434  r0 = vec_mergeh(r8, r12); /*0,2,4,6,8,10,12,14 set 0*/ \
435  r1 = vec_mergel(r8, r12); /*0,2,4,6,8,10,12,14 set 1*/ \
436  r2 = vec_mergeh(r9, r13); /*0,2,4,6,8,10,12,14 set 2*/ \
437  r4 = vec_mergeh(r10, r14); /*1,3,5,7,9,11,13,15 set 0*/ \
438  r5 = vec_mergel(r10, r14); /*1,3,5,7,9,11,13,15 set 1*/ \
439  r6 = vec_mergeh(r11, r15); /*1,3,5,7,9,11,13,15 set 2*/ \
440  /* Don't need to compute 3 and 7*/ \
441  \
442  /*Final merge*/ \
443  r8 = vec_mergeh(r0, r4); /*all set 0*/ \
444  r9 = vec_mergel(r0, r4); /*all set 1*/ \
445  r10 = vec_mergeh(r1, r5); /*all set 2*/ \
446  r11 = vec_mergel(r1, r5); /*all set 3*/ \
447  r12 = vec_mergeh(r2, r6); /*all set 4*/ \
448  r13 = vec_mergel(r2, r6); /*all set 5*/ \
449  /* Don't need to compute 14 and 15*/ \
450  \
451 }
452 
453 // out: o = |x-y| < a
454 static inline vec_u8 diff_lt_altivec ( register vec_u8 x,
455  register vec_u8 y,
456  register vec_u8 a) {
457 
458  register vec_u8 diff = vec_subs(x, y);
459  register vec_u8 diffneg = vec_subs(y, x);
460  register vec_u8 o = vec_or(diff, diffneg); /* |x-y| */
461  o = (vec_u8)vec_cmplt(o, a);
462  return o;
463 }
464 
465 static inline vec_u8 h264_deblock_mask ( register vec_u8 p0,
466  register vec_u8 p1,
467  register vec_u8 q0,
468  register vec_u8 q1,
469  register vec_u8 alpha,
470  register vec_u8 beta) {
471 
472  register vec_u8 mask;
473  register vec_u8 tempmask;
474 
475  mask = diff_lt_altivec(p0, q0, alpha);
476  tempmask = diff_lt_altivec(p1, p0, beta);
477  mask = vec_and(mask, tempmask);
478  tempmask = diff_lt_altivec(q1, q0, beta);
479  mask = vec_and(mask, tempmask);
480 
481  return mask;
482 }
483 
484 // out: newp1 = clip((p2 + ((p0 + q0 + 1) >> 1)) >> 1, p1-tc0, p1+tc0)
485 static inline vec_u8 h264_deblock_q1(register vec_u8 p0,
486  register vec_u8 p1,
487  register vec_u8 p2,
488  register vec_u8 q0,
489  register vec_u8 tc0) {
490 
491  register vec_u8 average = vec_avg(p0, q0);
492  register vec_u8 temp;
493  register vec_u8 uncliped;
494  register vec_u8 ones;
495  register vec_u8 max;
496  register vec_u8 min;
497  register vec_u8 newp1;
498 
499  temp = vec_xor(average, p2);
500  average = vec_avg(average, p2); /*avg(p2, avg(p0, q0)) */
501  ones = vec_splat_u8(1);
502  temp = vec_and(temp, ones); /*(p2^avg(p0, q0)) & 1 */
503  uncliped = vec_subs(average, temp); /*(p2+((p0+q0+1)>>1))>>1 */
504  max = vec_adds(p1, tc0);
505  min = vec_subs(p1, tc0);
506  newp1 = vec_max(min, uncliped);
507  newp1 = vec_min(max, newp1);
508  return newp1;
509 }
510 
511 #define h264_deblock_p0_q0(p0, p1, q0, q1, tc0masked) { \
512  \
513  const vec_u8 A0v = vec_sl(vec_splat_u8(10), vec_splat_u8(4)); \
514  \
515  register vec_u8 pq0bit = vec_xor(p0,q0); \
516  register vec_u8 q1minus; \
517  register vec_u8 p0minus; \
518  register vec_u8 stage1; \
519  register vec_u8 stage2; \
520  register vec_u8 vec160; \
521  register vec_u8 delta; \
522  register vec_u8 deltaneg; \
523  \
524  q1minus = vec_nor(q1, q1); /* 255 - q1 */ \
525  stage1 = vec_avg(p1, q1minus); /* (p1 - q1 + 256)>>1 */ \
526  stage2 = vec_sr(stage1, vec_splat_u8(1)); /* (p1 - q1 + 256)>>2 = 64 + (p1 - q1) >> 2 */ \
527  p0minus = vec_nor(p0, p0); /* 255 - p0 */ \
528  stage1 = vec_avg(q0, p0minus); /* (q0 - p0 + 256)>>1 */ \
529  pq0bit = vec_and(pq0bit, vec_splat_u8(1)); \
530  stage2 = vec_avg(stage2, pq0bit); /* 32 + ((q0 - p0)&1 + (p1 - q1) >> 2 + 1) >> 1 */ \
531  stage2 = vec_adds(stage2, stage1); /* 160 + ((p0 - q0) + (p1 - q1) >> 2 + 1) >> 1 */ \
532  vec160 = vec_ld(0, &A0v); \
533  deltaneg = vec_subs(vec160, stage2); /* -d */ \
534  delta = vec_subs(stage2, vec160); /* d */ \
535  deltaneg = vec_min(tc0masked, deltaneg); \
536  delta = vec_min(tc0masked, delta); \
537  p0 = vec_subs(p0, deltaneg); \
538  q0 = vec_subs(q0, delta); \
539  p0 = vec_adds(p0, delta); \
540  q0 = vec_adds(q0, deltaneg); \
541 }
542 
543 #define h264_loop_filter_luma_altivec(p2, p1, p0, q0, q1, q2, alpha, beta, tc0) { \
544  DECLARE_ALIGNED(16, unsigned char, temp)[16]; \
545  register vec_u8 alphavec; \
546  register vec_u8 betavec; \
547  register vec_u8 mask; \
548  register vec_u8 p1mask; \
549  register vec_u8 q1mask; \
550  register vector signed char tc0vec; \
551  register vec_u8 finaltc0; \
552  register vec_u8 tc0masked; \
553  register vec_u8 newp1; \
554  register vec_u8 newq1; \
555  \
556  temp[0] = alpha; \
557  temp[1] = beta; \
558  alphavec = vec_ld(0, temp); \
559  betavec = vec_splat(alphavec, 0x1); \
560  alphavec = vec_splat(alphavec, 0x0); \
561  mask = h264_deblock_mask(p0, p1, q0, q1, alphavec, betavec); /*if in block */ \
562  \
563  AV_COPY32(temp, tc0); \
564  tc0vec = vec_ld(0, (signed char*)temp); \
565  tc0vec = vec_mergeh(tc0vec, tc0vec); \
566  tc0vec = vec_mergeh(tc0vec, tc0vec); \
567  mask = vec_and(mask, vec_cmpgt(tc0vec, vec_splat_s8(-1))); /* if tc0[i] >= 0 */ \
568  finaltc0 = vec_and((vec_u8)tc0vec, mask); /* tc = tc0 */ \
569  \
570  p1mask = diff_lt_altivec(p2, p0, betavec); \
571  p1mask = vec_and(p1mask, mask); /* if ( |p2 - p0| < beta) */ \
572  tc0masked = vec_and(p1mask, (vec_u8)tc0vec); \
573  finaltc0 = vec_sub(finaltc0, p1mask); /* tc++ */ \
574  newp1 = h264_deblock_q1(p0, p1, p2, q0, tc0masked); \
575  /*end if*/ \
576  \
577  q1mask = diff_lt_altivec(q2, q0, betavec); \
578  q1mask = vec_and(q1mask, mask); /* if ( |q2 - q0| < beta ) */\
579  tc0masked = vec_and(q1mask, (vec_u8)tc0vec); \
580  finaltc0 = vec_sub(finaltc0, q1mask); /* tc++ */ \
581  newq1 = h264_deblock_q1(p0, q1, q2, q0, tc0masked); \
582  /*end if*/ \
583  \
584  h264_deblock_p0_q0(p0, p1, q0, q1, finaltc0); \
585  p1 = newp1; \
586  q1 = newq1; \
587 }
588 
589 static void h264_v_loop_filter_luma_altivec(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0) {
590 
591  if ((tc0[0] & tc0[1] & tc0[2] & tc0[3]) >= 0) {
592  register vec_u8 p2 = vec_ld(-3*stride, pix);
593  register vec_u8 p1 = vec_ld(-2*stride, pix);
594  register vec_u8 p0 = vec_ld(-1*stride, pix);
595  register vec_u8 q0 = vec_ld(0, pix);
596  register vec_u8 q1 = vec_ld(stride, pix);
597  register vec_u8 q2 = vec_ld(2*stride, pix);
598  h264_loop_filter_luma_altivec(p2, p1, p0, q0, q1, q2, alpha, beta, tc0);
599  vec_st(p1, -2*stride, pix);
600  vec_st(p0, -1*stride, pix);
601  vec_st(q0, 0, pix);
602  vec_st(q1, stride, pix);
603  }
604 }
605 
606 static void h264_h_loop_filter_luma_altivec(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0) {
607 
608  register vec_u8 line0, line1, line2, line3, line4, line5;
609  if ((tc0[0] & tc0[1] & tc0[2] & tc0[3]) < 0)
610  return;
611  readAndTranspose16x6(pix-3, stride, line0, line1, line2, line3, line4, line5);
612  h264_loop_filter_luma_altivec(line0, line1, line2, line3, line4, line5, alpha, beta, tc0);
613  transpose4x16(line1, line2, line3, line4);
614  write16x4(pix-2, stride, line1, line2, line3, line4);
615 }
616 
617 static av_always_inline
618 void weight_h264_W_altivec(uint8_t *block, int stride, int height,
619  int log2_denom, int weight, int offset, int w)
620 {
621  int y, aligned;
622  vec_u8 vblock;
623  vec_s16 vtemp, vweight, voffset, v0, v1;
624  vec_u16 vlog2_denom;
625  DECLARE_ALIGNED(16, int32_t, temp)[4];
626  LOAD_ZERO;
627 
628  offset <<= log2_denom;
629  if(log2_denom) offset += 1<<(log2_denom-1);
630  temp[0] = log2_denom;
631  temp[1] = weight;
632  temp[2] = offset;
633 
634  vtemp = (vec_s16)vec_ld(0, temp);
635  vlog2_denom = (vec_u16)vec_splat(vtemp, 1);
636  vweight = vec_splat(vtemp, 3);
637  voffset = vec_splat(vtemp, 5);
638  aligned = !((unsigned long)block & 0xf);
639 
640  for (y = 0; y < height; y++) {
641  vblock = vec_ld(0, block);
642 
643  v0 = (vec_s16)vec_mergeh(zero_u8v, vblock);
644  v1 = (vec_s16)vec_mergel(zero_u8v, vblock);
645 
646  if (w == 16 || aligned) {
647  v0 = vec_mladd(v0, vweight, zero_s16v);
648  v0 = vec_adds(v0, voffset);
649  v0 = vec_sra(v0, vlog2_denom);
650  }
651  if (w == 16 || !aligned) {
652  v1 = vec_mladd(v1, vweight, zero_s16v);
653  v1 = vec_adds(v1, voffset);
654  v1 = vec_sra(v1, vlog2_denom);
655  }
656  vblock = vec_packsu(v0, v1);
657  vec_st(vblock, 0, block);
658 
659  block += stride;
660  }
661 }
662 
663 static av_always_inline
664 void biweight_h264_W_altivec(uint8_t *dst, uint8_t *src, int stride, int height,
665  int log2_denom, int weightd, int weights, int offset, int w)
666 {
667  int y, dst_aligned, src_aligned;
668  vec_u8 vsrc, vdst;
669  vec_s16 vtemp, vweights, vweightd, voffset, v0, v1, v2, v3;
670  vec_u16 vlog2_denom;
671  DECLARE_ALIGNED(16, int32_t, temp)[4];
672  LOAD_ZERO;
673 
674  offset = ((offset + 1) | 1) << log2_denom;
675  temp[0] = log2_denom+1;
676  temp[1] = weights;
677  temp[2] = weightd;
678  temp[3] = offset;
679 
680  vtemp = (vec_s16)vec_ld(0, temp);
681  vlog2_denom = (vec_u16)vec_splat(vtemp, 1);
682  vweights = vec_splat(vtemp, 3);
683  vweightd = vec_splat(vtemp, 5);
684  voffset = vec_splat(vtemp, 7);
685  dst_aligned = !((unsigned long)dst & 0xf);
686  src_aligned = !((unsigned long)src & 0xf);
687 
688  for (y = 0; y < height; y++) {
689  vdst = vec_ld(0, dst);
690  vsrc = vec_ld(0, src);
691 
692  v0 = (vec_s16)vec_mergeh(zero_u8v, vdst);
693  v1 = (vec_s16)vec_mergel(zero_u8v, vdst);
694  v2 = (vec_s16)vec_mergeh(zero_u8v, vsrc);
695  v3 = (vec_s16)vec_mergel(zero_u8v, vsrc);
696 
697  if (w == 8) {
698  if (src_aligned)
699  v3 = v2;
700  else
701  v2 = v3;
702  }
703 
704  if (w == 16 || dst_aligned) {
705  v0 = vec_mladd(v0, vweightd, zero_s16v);
706  v2 = vec_mladd(v2, vweights, zero_s16v);
707 
708  v0 = vec_adds(v0, voffset);
709  v0 = vec_adds(v0, v2);
710  v0 = vec_sra(v0, vlog2_denom);
711  }
712  if (w == 16 || !dst_aligned) {
713  v1 = vec_mladd(v1, vweightd, zero_s16v);
714  v3 = vec_mladd(v3, vweights, zero_s16v);
715 
716  v1 = vec_adds(v1, voffset);
717  v1 = vec_adds(v1, v3);
718  v1 = vec_sra(v1, vlog2_denom);
719  }
720  vdst = vec_packsu(v0, v1);
721  vec_st(vdst, 0, dst);
722 
723  dst += stride;
724  src += stride;
725  }
726 }
727 
728 #define H264_WEIGHT(W) \
729 static void weight_h264_pixels ## W ## _altivec(uint8_t *block, int stride, int height, \
730  int log2_denom, int weight, int offset) \
731 { \
732  weight_h264_W_altivec(block, stride, height, log2_denom, weight, offset, W); \
733 }\
734 static void biweight_h264_pixels ## W ## _altivec(uint8_t *dst, uint8_t *src, int stride, int height, \
735  int log2_denom, int weightd, int weights, int offset) \
736 { \
737  biweight_h264_W_altivec(dst, src, stride, height, log2_denom, weightd, weights, offset, W); \
738 }
739 
740 H264_WEIGHT(16)
741 H264_WEIGHT( 8)
742 #endif /* HAVE_ALTIVEC */
743 
744 av_cold void ff_h264dsp_init_ppc(H264DSPContext *c, const int bit_depth,
745  const int chroma_format_idc)
746 {
747 #if HAVE_ALTIVEC
748  if (av_get_cpu_flags() & AV_CPU_FLAG_ALTIVEC) {
749  if (bit_depth == 8) {
750  c->h264_idct_add = h264_idct_add_altivec;
751  if (chroma_format_idc == 1)
752  c->h264_idct_add8 = h264_idct_add8_altivec;
753  c->h264_idct_add16 = h264_idct_add16_altivec;
754  c->h264_idct_add16intra = h264_idct_add16intra_altivec;
755  c->h264_idct_dc_add= h264_idct_dc_add_altivec;
756  c->h264_idct8_dc_add = h264_idct8_dc_add_altivec;
757  c->h264_idct8_add = h264_idct8_add_altivec;
758  c->h264_idct8_add4 = h264_idct8_add4_altivec;
759  c->h264_v_loop_filter_luma= h264_v_loop_filter_luma_altivec;
760  c->h264_h_loop_filter_luma= h264_h_loop_filter_luma_altivec;
761 
762  c->weight_h264_pixels_tab[0] = weight_h264_pixels16_altivec;
763  c->weight_h264_pixels_tab[1] = weight_h264_pixels8_altivec;
764  c->biweight_h264_pixels_tab[0] = biweight_h264_pixels16_altivec;
765  c->biweight_h264_pixels_tab[1] = biweight_h264_pixels8_altivec;
766  }
767  }
768 #endif /* HAVE_ALTIVEC */
769 }
Generated on Wed Jul 10 2013 23:47:59 for FFmpeg by   doxygen 1.8.2