FFmpeg
 All Data Structures Files Functions Variables Typedefs Enumerations Enumerator Macros Groups Pages
h264_altivec.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 "libavutil/cpu.h"
22 #include "libavutil/intreadwrite.h"
23 #include "libavutil/ppc/types_altivec.h"
24 #include "libavutil/ppc/util_altivec.h"
25 #include "libavcodec/dsputil.h"
26 #include "libavcodec/h264data.h"
27 #include "libavcodec/h264dsp.h"
28 
29 #include "dsputil_altivec.h"
30 
31 #define PUT_OP_U8_ALTIVEC(d, s, dst) d = s
32 #define AVG_OP_U8_ALTIVEC(d, s, dst) d = vec_avg(dst, s)
33 
34 #define OP_U8_ALTIVEC PUT_OP_U8_ALTIVEC
35 #define PREFIX_h264_chroma_mc8_altivec put_h264_chroma_mc8_altivec
36 #define PREFIX_h264_chroma_mc8_num altivec_put_h264_chroma_mc8_num
37 #define PREFIX_h264_qpel16_h_lowpass_altivec put_h264_qpel16_h_lowpass_altivec
38 #define PREFIX_h264_qpel16_h_lowpass_num altivec_put_h264_qpel16_h_lowpass_num
39 #define PREFIX_h264_qpel16_v_lowpass_altivec put_h264_qpel16_v_lowpass_altivec
40 #define PREFIX_h264_qpel16_v_lowpass_num altivec_put_h264_qpel16_v_lowpass_num
41 #define PREFIX_h264_qpel16_hv_lowpass_altivec put_h264_qpel16_hv_lowpass_altivec
42 #define PREFIX_h264_qpel16_hv_lowpass_num altivec_put_h264_qpel16_hv_lowpass_num
43 #include "h264_altivec_template.c"
44 #undef OP_U8_ALTIVEC
45 #undef PREFIX_h264_chroma_mc8_altivec
46 #undef PREFIX_h264_chroma_mc8_num
47 #undef PREFIX_h264_qpel16_h_lowpass_altivec
48 #undef PREFIX_h264_qpel16_h_lowpass_num
49 #undef PREFIX_h264_qpel16_v_lowpass_altivec
50 #undef PREFIX_h264_qpel16_v_lowpass_num
51 #undef PREFIX_h264_qpel16_hv_lowpass_altivec
52 #undef PREFIX_h264_qpel16_hv_lowpass_num
53 
54 #define OP_U8_ALTIVEC AVG_OP_U8_ALTIVEC
55 #define PREFIX_h264_chroma_mc8_altivec avg_h264_chroma_mc8_altivec
56 #define PREFIX_h264_chroma_mc8_num altivec_avg_h264_chroma_mc8_num
57 #define PREFIX_h264_qpel16_h_lowpass_altivec avg_h264_qpel16_h_lowpass_altivec
58 #define PREFIX_h264_qpel16_h_lowpass_num altivec_avg_h264_qpel16_h_lowpass_num
59 #define PREFIX_h264_qpel16_v_lowpass_altivec avg_h264_qpel16_v_lowpass_altivec
60 #define PREFIX_h264_qpel16_v_lowpass_num altivec_avg_h264_qpel16_v_lowpass_num
61 #define PREFIX_h264_qpel16_hv_lowpass_altivec avg_h264_qpel16_hv_lowpass_altivec
62 #define PREFIX_h264_qpel16_hv_lowpass_num altivec_avg_h264_qpel16_hv_lowpass_num
63 #include "h264_altivec_template.c"
64 #undef OP_U8_ALTIVEC
65 #undef PREFIX_h264_chroma_mc8_altivec
66 #undef PREFIX_h264_chroma_mc8_num
67 #undef PREFIX_h264_qpel16_h_lowpass_altivec
68 #undef PREFIX_h264_qpel16_h_lowpass_num
69 #undef PREFIX_h264_qpel16_v_lowpass_altivec
70 #undef PREFIX_h264_qpel16_v_lowpass_num
71 #undef PREFIX_h264_qpel16_hv_lowpass_altivec
72 #undef PREFIX_h264_qpel16_hv_lowpass_num
73 
74 #define H264_MC(OPNAME, SIZE, CODETYPE) \
75 static void OPNAME ## h264_qpel ## SIZE ## _mc00_ ## CODETYPE (uint8_t *dst, uint8_t *src, int stride){\
76  ff_ ## OPNAME ## pixels ## SIZE ## _ ## CODETYPE(dst, src, stride, SIZE);\
77 }\
78 \
79 static void OPNAME ## h264_qpel ## SIZE ## _mc10_ ## CODETYPE(uint8_t *dst, uint8_t *src, int stride){ \
80  DECLARE_ALIGNED(16, uint8_t, half)[SIZE*SIZE];\
81  put_h264_qpel ## SIZE ## _h_lowpass_ ## CODETYPE(half, src, SIZE, stride);\
82  OPNAME ## pixels ## SIZE ## _l2_ ## CODETYPE(dst, src, half, stride, stride, SIZE);\
83 }\
84 \
85 static void OPNAME ## h264_qpel ## SIZE ## _mc20_ ## CODETYPE(uint8_t *dst, uint8_t *src, int stride){\
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, uint8_t *src, int stride){\
90  DECLARE_ALIGNED(16, uint8_t, half)[SIZE*SIZE];\
91  put_h264_qpel ## SIZE ## _h_lowpass_ ## CODETYPE(half, src, SIZE, stride);\
92  OPNAME ## pixels ## SIZE ## _l2_ ## CODETYPE(dst, src+1, half, stride, stride, SIZE);\
93 }\
94 \
95 static void OPNAME ## h264_qpel ## SIZE ## _mc01_ ## CODETYPE(uint8_t *dst, uint8_t *src, int stride){\
96  DECLARE_ALIGNED(16, uint8_t, half)[SIZE*SIZE];\
97  put_h264_qpel ## SIZE ## _v_lowpass_ ## CODETYPE(half, src, SIZE, stride);\
98  OPNAME ## pixels ## SIZE ## _l2_ ## CODETYPE(dst, src, half, stride, stride, SIZE);\
99 }\
100 \
101 static void OPNAME ## h264_qpel ## SIZE ## _mc02_ ## CODETYPE(uint8_t *dst, uint8_t *src, int stride){\
102  OPNAME ## h264_qpel ## SIZE ## _v_lowpass_ ## CODETYPE(dst, src, stride, stride);\
103 }\
104 \
105 static void OPNAME ## h264_qpel ## SIZE ## _mc03_ ## CODETYPE(uint8_t *dst, uint8_t *src, int stride){\
106  DECLARE_ALIGNED(16, uint8_t, half)[SIZE*SIZE];\
107  put_h264_qpel ## SIZE ## _v_lowpass_ ## CODETYPE(half, src, SIZE, stride);\
108  OPNAME ## pixels ## SIZE ## _l2_ ## CODETYPE(dst, src+stride, half, stride, stride, SIZE);\
109 }\
110 \
111 static void OPNAME ## h264_qpel ## SIZE ## _mc11_ ## CODETYPE(uint8_t *dst, uint8_t *src, int stride){\
112  DECLARE_ALIGNED(16, uint8_t, halfH)[SIZE*SIZE];\
113  DECLARE_ALIGNED(16, uint8_t, halfV)[SIZE*SIZE];\
114  put_h264_qpel ## SIZE ## _h_lowpass_ ## CODETYPE(halfH, src, SIZE, stride);\
115  put_h264_qpel ## SIZE ## _v_lowpass_ ## CODETYPE(halfV, src, SIZE, stride);\
116  OPNAME ## pixels ## SIZE ## _l2_ ## CODETYPE(dst, halfH, halfV, stride, SIZE, SIZE);\
117 }\
118 \
119 static void OPNAME ## h264_qpel ## SIZE ## _mc31_ ## CODETYPE(uint8_t *dst, uint8_t *src, int stride){\
120  DECLARE_ALIGNED(16, uint8_t, halfH)[SIZE*SIZE];\
121  DECLARE_ALIGNED(16, uint8_t, halfV)[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);\
125 }\
126 \
127 static void OPNAME ## h264_qpel ## SIZE ## _mc13_ ## CODETYPE(uint8_t *dst, uint8_t *src, int stride){\
128  DECLARE_ALIGNED(16, uint8_t, halfH)[SIZE*SIZE];\
129  DECLARE_ALIGNED(16, uint8_t, halfV)[SIZE*SIZE];\
130  put_h264_qpel ## SIZE ## _h_lowpass_ ## CODETYPE(halfH, src + stride, SIZE, stride);\
131  put_h264_qpel ## SIZE ## _v_lowpass_ ## CODETYPE(halfV, src, SIZE, stride);\
132  OPNAME ## pixels ## SIZE ## _l2_ ## CODETYPE(dst, halfH, halfV, stride, SIZE, SIZE);\
133 }\
134 \
135 static void OPNAME ## h264_qpel ## SIZE ## _mc33_ ## CODETYPE(uint8_t *dst, uint8_t *src, int stride){\
136  DECLARE_ALIGNED(16, uint8_t, halfH)[SIZE*SIZE];\
137  DECLARE_ALIGNED(16, uint8_t, halfV)[SIZE*SIZE];\
138  put_h264_qpel ## SIZE ## _h_lowpass_ ## CODETYPE(halfH, src + stride, SIZE, stride);\
139  put_h264_qpel ## SIZE ## _v_lowpass_ ## CODETYPE(halfV, src+1, SIZE, stride);\
140  OPNAME ## pixels ## SIZE ## _l2_ ## CODETYPE(dst, halfH, halfV, stride, SIZE, SIZE);\
141 }\
142 \
143 static void OPNAME ## h264_qpel ## SIZE ## _mc22_ ## CODETYPE(uint8_t *dst, uint8_t *src, int stride){\
144  DECLARE_ALIGNED(16, int16_t, tmp)[SIZE*(SIZE+8)];\
145  OPNAME ## h264_qpel ## SIZE ## _hv_lowpass_ ## CODETYPE(dst, tmp, src, stride, SIZE, stride);\
146 }\
147 \
148 static void OPNAME ## h264_qpel ## SIZE ## _mc21_ ## CODETYPE(uint8_t *dst, uint8_t *src, int stride){\
149  DECLARE_ALIGNED(16, uint8_t, halfH)[SIZE*SIZE];\
150  DECLARE_ALIGNED(16, uint8_t, halfHV)[SIZE*SIZE];\
151  DECLARE_ALIGNED(16, int16_t, tmp)[SIZE*(SIZE+8)];\
152  put_h264_qpel ## SIZE ## _h_lowpass_ ## CODETYPE(halfH, src, SIZE, stride);\
153  put_h264_qpel ## SIZE ## _hv_lowpass_ ## CODETYPE(halfHV, tmp, src, SIZE, SIZE, stride);\
154  OPNAME ## pixels ## SIZE ## _l2_ ## CODETYPE(dst, halfH, halfHV, stride, SIZE, SIZE);\
155 }\
156 \
157 static void OPNAME ## h264_qpel ## SIZE ## _mc23_ ## CODETYPE(uint8_t *dst, uint8_t *src, int stride){\
158  DECLARE_ALIGNED(16, uint8_t, halfH)[SIZE*SIZE];\
159  DECLARE_ALIGNED(16, uint8_t, halfHV)[SIZE*SIZE];\
160  DECLARE_ALIGNED(16, int16_t, tmp)[SIZE*(SIZE+8)];\
161  put_h264_qpel ## SIZE ## _h_lowpass_ ## CODETYPE(halfH, src + stride, SIZE, stride);\
162  put_h264_qpel ## SIZE ## _hv_lowpass_ ## CODETYPE(halfHV, tmp, src, SIZE, SIZE, stride);\
163  OPNAME ## pixels ## SIZE ## _l2_ ## CODETYPE(dst, halfH, halfHV, stride, SIZE, SIZE);\
164 }\
165 \
166 static void OPNAME ## h264_qpel ## SIZE ## _mc12_ ## CODETYPE(uint8_t *dst, uint8_t *src, int stride){\
167  DECLARE_ALIGNED(16, uint8_t, halfV)[SIZE*SIZE];\
168  DECLARE_ALIGNED(16, uint8_t, halfHV)[SIZE*SIZE];\
169  DECLARE_ALIGNED(16, int16_t, tmp)[SIZE*(SIZE+8)];\
170  put_h264_qpel ## SIZE ## _v_lowpass_ ## CODETYPE(halfV, src, SIZE, stride);\
171  put_h264_qpel ## SIZE ## _hv_lowpass_ ## CODETYPE(halfHV, tmp, src, SIZE, SIZE, stride);\
172  OPNAME ## pixels ## SIZE ## _l2_ ## CODETYPE(dst, halfV, halfHV, stride, SIZE, SIZE);\
173 }\
174 \
175 static void OPNAME ## h264_qpel ## SIZE ## _mc32_ ## CODETYPE(uint8_t *dst, uint8_t *src, int stride){\
176  DECLARE_ALIGNED(16, uint8_t, halfV)[SIZE*SIZE];\
177  DECLARE_ALIGNED(16, uint8_t, halfHV)[SIZE*SIZE];\
178  DECLARE_ALIGNED(16, int16_t, tmp)[SIZE*(SIZE+8)];\
179  put_h264_qpel ## SIZE ## _v_lowpass_ ## CODETYPE(halfV, src+1, SIZE, stride);\
180  put_h264_qpel ## SIZE ## _hv_lowpass_ ## CODETYPE(halfHV, tmp, src, SIZE, SIZE, stride);\
181  OPNAME ## pixels ## SIZE ## _l2_ ## CODETYPE(dst, halfV, halfHV, stride, SIZE, SIZE);\
182 }\
183 
184 static inline void put_pixels16_l2_altivec( uint8_t * dst, const uint8_t * src1,
185  const uint8_t * src2, int dst_stride,
186  int src_stride1, int h)
187 {
188  int i;
189  vec_u8 a, b, d, tmp1, tmp2, mask, mask_, edges, align;
190 
191  mask_ = vec_lvsl(0, src2);
192 
193  for (i = 0; i < h; i++) {
194 
195  tmp1 = vec_ld(i * src_stride1, src1);
196  mask = vec_lvsl(i * src_stride1, src1);
197  tmp2 = vec_ld(i * src_stride1 + 15, src1);
198 
199  a = vec_perm(tmp1, tmp2, mask);
200 
201  tmp1 = vec_ld(i * 16, src2);
202  tmp2 = vec_ld(i * 16 + 15, src2);
203 
204  b = vec_perm(tmp1, tmp2, mask_);
205 
206  tmp1 = vec_ld(0, dst);
207  mask = vec_lvsl(0, dst);
208  tmp2 = vec_ld(15, dst);
209 
210  d = vec_avg(a, b);
211 
212  edges = vec_perm(tmp2, tmp1, mask);
213 
214  align = vec_lvsr(0, dst);
215 
216  tmp2 = vec_perm(d, edges, align);
217  tmp1 = vec_perm(edges, d, align);
218 
219  vec_st(tmp2, 15, dst);
220  vec_st(tmp1, 0 , dst);
221 
222  dst += dst_stride;
223  }
224 }
225 
226 static inline void avg_pixels16_l2_altivec( uint8_t * dst, const uint8_t * src1,
227  const uint8_t * src2, int dst_stride,
228  int src_stride1, int h)
229 {
230  int i;
231  vec_u8 a, b, d, tmp1, tmp2, mask, mask_, edges, align;
232 
233  mask_ = vec_lvsl(0, src2);
234 
235  for (i = 0; i < h; i++) {
236 
237  tmp1 = vec_ld(i * src_stride1, src1);
238  mask = vec_lvsl(i * src_stride1, src1);
239  tmp2 = vec_ld(i * src_stride1 + 15, src1);
240 
241  a = vec_perm(tmp1, tmp2, mask);
242 
243  tmp1 = vec_ld(i * 16, src2);
244  tmp2 = vec_ld(i * 16 + 15, src2);
245 
246  b = vec_perm(tmp1, tmp2, mask_);
247 
248  tmp1 = vec_ld(0, dst);
249  mask = vec_lvsl(0, dst);
250  tmp2 = vec_ld(15, dst);
251 
252  d = vec_avg(vec_perm(tmp1, tmp2, mask), vec_avg(a, b));
253 
254  edges = vec_perm(tmp2, tmp1, mask);
255 
256  align = vec_lvsr(0, dst);
257 
258  tmp2 = vec_perm(d, edges, align);
259  tmp1 = vec_perm(edges, d, align);
260 
261  vec_st(tmp2, 15, dst);
262  vec_st(tmp1, 0 , dst);
263 
264  dst += dst_stride;
265  }
266 }
267 
268 /* Implemented but could be faster
269 #define put_pixels16_l2_altivec(d,s1,s2,ds,s1s,h) put_pixels16_l2(d,s1,s2,ds,s1s,16,h)
270 #define avg_pixels16_l2_altivec(d,s1,s2,ds,s1s,h) avg_pixels16_l2(d,s1,s2,ds,s1s,16,h)
271  */
272 
273 H264_MC(put_, 16, altivec)
274 H264_MC(avg_, 16, altivec)
275 
276 
277 /****************************************************************************
278  * IDCT transform:
279  ****************************************************************************/
280 
281 #define VEC_1D_DCT(vb0,vb1,vb2,vb3,va0,va1,va2,va3) \
282  /* 1st stage */ \
283  vz0 = vec_add(vb0,vb2); /* temp[0] = Y[0] + Y[2] */ \
284  vz1 = vec_sub(vb0,vb2); /* temp[1] = Y[0] - Y[2] */ \
285  vz2 = vec_sra(vb1,vec_splat_u16(1)); \
286  vz2 = vec_sub(vz2,vb3); /* temp[2] = Y[1].1/2 - Y[3] */ \
287  vz3 = vec_sra(vb3,vec_splat_u16(1)); \
288  vz3 = vec_add(vb1,vz3); /* temp[3] = Y[1] + Y[3].1/2 */ \
289  /* 2nd stage: output */ \
290  va0 = vec_add(vz0,vz3); /* x[0] = temp[0] + temp[3] */ \
291  va1 = vec_add(vz1,vz2); /* x[1] = temp[1] + temp[2] */ \
292  va2 = vec_sub(vz1,vz2); /* x[2] = temp[1] - temp[2] */ \
293  va3 = vec_sub(vz0,vz3) /* x[3] = temp[0] - temp[3] */
294 
295 #define VEC_TRANSPOSE_4(a0,a1,a2,a3,b0,b1,b2,b3) \
296  b0 = vec_mergeh( a0, a0 ); \
297  b1 = vec_mergeh( a1, a0 ); \
298  b2 = vec_mergeh( a2, a0 ); \
299  b3 = vec_mergeh( a3, a0 ); \
300  a0 = vec_mergeh( b0, b2 ); \
301  a1 = vec_mergel( b0, b2 ); \
302  a2 = vec_mergeh( b1, b3 ); \
303  a3 = vec_mergel( b1, b3 ); \
304  b0 = vec_mergeh( a0, a2 ); \
305  b1 = vec_mergel( a0, a2 ); \
306  b2 = vec_mergeh( a1, a3 ); \
307  b3 = vec_mergel( a1, a3 )
308 
309 #define VEC_LOAD_U8_ADD_S16_STORE_U8(va) \
310  vdst_orig = vec_ld(0, dst); \
311  vdst = vec_perm(vdst_orig, zero_u8v, vdst_mask); \
312  vdst_ss = (vec_s16) vec_mergeh(zero_u8v, vdst); \
313  va = vec_add(va, vdst_ss); \
314  va_u8 = vec_packsu(va, zero_s16v); \
315  va_u32 = vec_splat((vec_u32)va_u8, 0); \
316  vec_ste(va_u32, element, (uint32_t*)dst);
317 
318 static void ff_h264_idct_add_altivec(uint8_t *dst, DCTELEM *block, int stride)
319 {
320  vec_s16 va0, va1, va2, va3;
321  vec_s16 vz0, vz1, vz2, vz3;
322  vec_s16 vtmp0, vtmp1, vtmp2, vtmp3;
323  vec_u8 va_u8;
324  vec_u32 va_u32;
325  vec_s16 vdst_ss;
326  const vec_u16 v6us = vec_splat_u16(6);
327  vec_u8 vdst, vdst_orig;
328  vec_u8 vdst_mask = vec_lvsl(0, dst);
329  int element = ((unsigned long)dst & 0xf) >> 2;
330  LOAD_ZERO;
331 
332  block[0] += 32; /* add 32 as a DC-level for rounding */
333 
334  vtmp0 = vec_ld(0,block);
335  vtmp1 = vec_sld(vtmp0, vtmp0, 8);
336  vtmp2 = vec_ld(16,block);
337  vtmp3 = vec_sld(vtmp2, vtmp2, 8);
338 
339  VEC_1D_DCT(vtmp0,vtmp1,vtmp2,vtmp3,va0,va1,va2,va3);
340  VEC_TRANSPOSE_4(va0,va1,va2,va3,vtmp0,vtmp1,vtmp2,vtmp3);
341  VEC_1D_DCT(vtmp0,vtmp1,vtmp2,vtmp3,va0,va1,va2,va3);
342 
343  va0 = vec_sra(va0,v6us);
344  va1 = vec_sra(va1,v6us);
345  va2 = vec_sra(va2,v6us);
346  va3 = vec_sra(va3,v6us);
347 
348  VEC_LOAD_U8_ADD_S16_STORE_U8(va0);
349  dst += stride;
350  VEC_LOAD_U8_ADD_S16_STORE_U8(va1);
351  dst += stride;
352  VEC_LOAD_U8_ADD_S16_STORE_U8(va2);
353  dst += stride;
354  VEC_LOAD_U8_ADD_S16_STORE_U8(va3);
355 }
356 
357 #define IDCT8_1D_ALTIVEC(s0, s1, s2, s3, s4, s5, s6, s7, d0, d1, d2, d3, d4, d5, d6, d7) {\
358  /* a0 = SRC(0) + SRC(4); */ \
359  vec_s16 a0v = vec_add(s0, s4); \
360  /* a2 = SRC(0) - SRC(4); */ \
361  vec_s16 a2v = vec_sub(s0, s4); \
362  /* a4 = (SRC(2)>>1) - SRC(6); */ \
363  vec_s16 a4v = vec_sub(vec_sra(s2, onev), s6); \
364  /* a6 = (SRC(6)>>1) + SRC(2); */ \
365  vec_s16 a6v = vec_add(vec_sra(s6, onev), s2); \
366  /* b0 = a0 + a6; */ \
367  vec_s16 b0v = vec_add(a0v, a6v); \
368  /* b2 = a2 + a4; */ \
369  vec_s16 b2v = vec_add(a2v, a4v); \
370  /* b4 = a2 - a4; */ \
371  vec_s16 b4v = vec_sub(a2v, a4v); \
372  /* b6 = a0 - a6; */ \
373  vec_s16 b6v = vec_sub(a0v, a6v); \
374  /* a1 = SRC(5) - SRC(3) - SRC(7) - (SRC(7)>>1); */ \
375  /* a1 = (SRC(5)-SRC(3)) - (SRC(7) + (SRC(7)>>1)); */ \
376  vec_s16 a1v = vec_sub( vec_sub(s5, s3), vec_add(s7, vec_sra(s7, onev)) ); \
377  /* a3 = SRC(7) + SRC(1) - SRC(3) - (SRC(3)>>1); */ \
378  /* a3 = (SRC(7)+SRC(1)) - (SRC(3) + (SRC(3)>>1)); */ \
379  vec_s16 a3v = vec_sub( vec_add(s7, s1), vec_add(s3, vec_sra(s3, onev)) );\
380  /* a5 = SRC(7) - SRC(1) + SRC(5) + (SRC(5)>>1); */ \
381  /* a5 = (SRC(7)-SRC(1)) + SRC(5) + (SRC(5)>>1); */ \
382  vec_s16 a5v = vec_add( vec_sub(s7, s1), vec_add(s5, vec_sra(s5, onev)) );\
383  /* a7 = SRC(5)+SRC(3) + SRC(1) + (SRC(1)>>1); */ \
384  vec_s16 a7v = vec_add( vec_add(s5, s3), vec_add(s1, vec_sra(s1, onev)) );\
385  /* b1 = (a7>>2) + a1; */ \
386  vec_s16 b1v = vec_add( vec_sra(a7v, twov), a1v); \
387  /* b3 = a3 + (a5>>2); */ \
388  vec_s16 b3v = vec_add(a3v, vec_sra(a5v, twov)); \
389  /* b5 = (a3>>2) - a5; */ \
390  vec_s16 b5v = vec_sub( vec_sra(a3v, twov), a5v); \
391  /* b7 = a7 - (a1>>2); */ \
392  vec_s16 b7v = vec_sub( a7v, vec_sra(a1v, twov)); \
393  /* DST(0, b0 + b7); */ \
394  d0 = vec_add(b0v, b7v); \
395  /* DST(1, b2 + b5); */ \
396  d1 = vec_add(b2v, b5v); \
397  /* DST(2, b4 + b3); */ \
398  d2 = vec_add(b4v, b3v); \
399  /* DST(3, b6 + b1); */ \
400  d3 = vec_add(b6v, b1v); \
401  /* DST(4, b6 - b1); */ \
402  d4 = vec_sub(b6v, b1v); \
403  /* DST(5, b4 - b3); */ \
404  d5 = vec_sub(b4v, b3v); \
405  /* DST(6, b2 - b5); */ \
406  d6 = vec_sub(b2v, b5v); \
407  /* DST(7, b0 - b7); */ \
408  d7 = vec_sub(b0v, b7v); \
409 }
410 
411 #define ALTIVEC_STORE_SUM_CLIP(dest, idctv, perm_ldv, perm_stv, sel) { \
412  /* unaligned load */ \
413  vec_u8 hv = vec_ld( 0, dest ); \
414  vec_u8 lv = vec_ld( 7, dest ); \
415  vec_u8 dstv = vec_perm( hv, lv, (vec_u8)perm_ldv ); \
416  vec_s16 idct_sh6 = vec_sra(idctv, sixv); \
417  vec_u16 dst16 = (vec_u16)vec_mergeh(zero_u8v, dstv); \
418  vec_s16 idstsum = vec_adds(idct_sh6, (vec_s16)dst16); \
419  vec_u8 idstsum8 = vec_packsu(zero_s16v, idstsum); \
420  vec_u8 edgehv; \
421  /* unaligned store */ \
422  vec_u8 bodyv = vec_perm( idstsum8, idstsum8, perm_stv );\
423  vec_u8 edgelv = vec_perm( sel, zero_u8v, perm_stv ); \
424  lv = vec_sel( lv, bodyv, edgelv ); \
425  vec_st( lv, 7, dest ); \
426  hv = vec_ld( 0, dest ); \
427  edgehv = vec_perm( zero_u8v, sel, perm_stv ); \
428  hv = vec_sel( hv, bodyv, edgehv ); \
429  vec_st( hv, 0, dest ); \
430  }
431 
432 static void ff_h264_idct8_add_altivec( uint8_t *dst, DCTELEM *dct, int stride ) {
433  vec_s16 s0, s1, s2, s3, s4, s5, s6, s7;
434  vec_s16 d0, d1, d2, d3, d4, d5, d6, d7;
435  vec_s16 idct0, idct1, idct2, idct3, idct4, idct5, idct6, idct7;
436 
437  vec_u8 perm_ldv = vec_lvsl(0, dst);
438  vec_u8 perm_stv = vec_lvsr(8, dst);
439 
440  const vec_u16 onev = vec_splat_u16(1);
441  const vec_u16 twov = vec_splat_u16(2);
442  const vec_u16 sixv = vec_splat_u16(6);
443 
444  const vec_u8 sel = (vec_u8) {0,0,0,0,0,0,0,0,-1,-1,-1,-1,-1,-1,-1,-1};
445  LOAD_ZERO;
446 
447  dct[0] += 32; // rounding for the >>6 at the end
448 
449  s0 = vec_ld(0x00, (int16_t*)dct);
450  s1 = vec_ld(0x10, (int16_t*)dct);
451  s2 = vec_ld(0x20, (int16_t*)dct);
452  s3 = vec_ld(0x30, (int16_t*)dct);
453  s4 = vec_ld(0x40, (int16_t*)dct);
454  s5 = vec_ld(0x50, (int16_t*)dct);
455  s6 = vec_ld(0x60, (int16_t*)dct);
456  s7 = vec_ld(0x70, (int16_t*)dct);
457 
458  IDCT8_1D_ALTIVEC(s0, s1, s2, s3, s4, s5, s6, s7,
459  d0, d1, d2, d3, d4, d5, d6, d7);
460 
461  TRANSPOSE8( d0, d1, d2, d3, d4, d5, d6, d7 );
462 
463  IDCT8_1D_ALTIVEC(d0, d1, d2, d3, d4, d5, d6, d7,
464  idct0, idct1, idct2, idct3, idct4, idct5, idct6, idct7);
465 
466  ALTIVEC_STORE_SUM_CLIP(&dst[0*stride], idct0, perm_ldv, perm_stv, sel);
467  ALTIVEC_STORE_SUM_CLIP(&dst[1*stride], idct1, perm_ldv, perm_stv, sel);
468  ALTIVEC_STORE_SUM_CLIP(&dst[2*stride], idct2, perm_ldv, perm_stv, sel);
469  ALTIVEC_STORE_SUM_CLIP(&dst[3*stride], idct3, perm_ldv, perm_stv, sel);
470  ALTIVEC_STORE_SUM_CLIP(&dst[4*stride], idct4, perm_ldv, perm_stv, sel);
471  ALTIVEC_STORE_SUM_CLIP(&dst[5*stride], idct5, perm_ldv, perm_stv, sel);
472  ALTIVEC_STORE_SUM_CLIP(&dst[6*stride], idct6, perm_ldv, perm_stv, sel);
473  ALTIVEC_STORE_SUM_CLIP(&dst[7*stride], idct7, perm_ldv, perm_stv, sel);
474 }
475 
476 static av_always_inline void h264_idct_dc_add_internal(uint8_t *dst, DCTELEM *block, int stride, int size)
477 {
478  vec_s16 dc16;
479  vec_u8 dcplus, dcminus, v0, v1, v2, v3, aligner;
480  LOAD_ZERO;
481  DECLARE_ALIGNED(16, int, dc);
482  int i;
483 
484  dc = (block[0] + 32) >> 6;
485  dc16 = vec_splat((vec_s16) vec_lde(0, &dc), 1);
486 
487  if (size == 4)
488  dc16 = vec_sld(dc16, zero_s16v, 8);
489  dcplus = vec_packsu(dc16, zero_s16v);
490  dcminus = vec_packsu(vec_sub(zero_s16v, dc16), zero_s16v);
491 
492  aligner = vec_lvsr(0, dst);
493  dcplus = vec_perm(dcplus, dcplus, aligner);
494  dcminus = vec_perm(dcminus, dcminus, aligner);
495 
496  for (i = 0; i < size; i += 4) {
497  v0 = vec_ld(0, dst+0*stride);
498  v1 = vec_ld(0, dst+1*stride);
499  v2 = vec_ld(0, dst+2*stride);
500  v3 = vec_ld(0, dst+3*stride);
501 
502  v0 = vec_adds(v0, dcplus);
503  v1 = vec_adds(v1, dcplus);
504  v2 = vec_adds(v2, dcplus);
505  v3 = vec_adds(v3, dcplus);
506 
507  v0 = vec_subs(v0, dcminus);
508  v1 = vec_subs(v1, dcminus);
509  v2 = vec_subs(v2, dcminus);
510  v3 = vec_subs(v3, dcminus);
511 
512  vec_st(v0, 0, dst+0*stride);
513  vec_st(v1, 0, dst+1*stride);
514  vec_st(v2, 0, dst+2*stride);
515  vec_st(v3, 0, dst+3*stride);
516 
517  dst += 4*stride;
518  }
519 }
520 
521 static void h264_idct_dc_add_altivec(uint8_t *dst, DCTELEM *block, int stride)
522 {
523  h264_idct_dc_add_internal(dst, block, stride, 4);
524 }
525 
526 static void ff_h264_idct8_dc_add_altivec(uint8_t *dst, DCTELEM *block, int stride)
527 {
528  h264_idct_dc_add_internal(dst, block, stride, 8);
529 }
530 
531 static void ff_h264_idct_add16_altivec(uint8_t *dst, const int *block_offset, DCTELEM *block, int stride, const uint8_t nnzc[15*8]){
532  int i;
533  for(i=0; i<16; i++){
534  int nnz = nnzc[ scan8[i] ];
535  if(nnz){
536  if(nnz==1 && block[i*16]) h264_idct_dc_add_altivec(dst + block_offset[i], block + i*16, stride);
537  else ff_h264_idct_add_altivec(dst + block_offset[i], block + i*16, stride);
538  }
539  }
540 }
541 
542 static void ff_h264_idct_add16intra_altivec(uint8_t *dst, const int *block_offset, DCTELEM *block, int stride, const uint8_t nnzc[15*8]){
543  int i;
544  for(i=0; i<16; i++){
545  if(nnzc[ scan8[i] ]) ff_h264_idct_add_altivec(dst + block_offset[i], block + i*16, stride);
546  else if(block[i*16]) h264_idct_dc_add_altivec(dst + block_offset[i], block + i*16, stride);
547  }
548 }
549 
550 static void ff_h264_idct8_add4_altivec(uint8_t *dst, const int *block_offset, DCTELEM *block, int stride, const uint8_t nnzc[15*8]){
551  int i;
552  for(i=0; i<16; i+=4){
553  int nnz = nnzc[ scan8[i] ];
554  if(nnz){
555  if(nnz==1 && block[i*16]) ff_h264_idct8_dc_add_altivec(dst + block_offset[i], block + i*16, stride);
556  else ff_h264_idct8_add_altivec (dst + block_offset[i], block + i*16, stride);
557  }
558  }
559 }
560 
561 static void ff_h264_idct_add8_altivec(uint8_t **dest, const int *block_offset, DCTELEM *block, int stride, const uint8_t nnzc[15*8]){
562  int i, j;
563  for (j = 1; j < 3; j++) {
564  for(i = j * 16; i < j * 16 + 4; i++){
565  if(nnzc[ scan8[i] ])
566  ff_h264_idct_add_altivec(dest[j-1] + block_offset[i], block + i*16, stride);
567  else if(block[i*16])
568  h264_idct_dc_add_altivec(dest[j-1] + block_offset[i], block + i*16, stride);
569  }
570  }
571 }
572 
573 #define transpose4x16(r0, r1, r2, r3) { \
574  register vec_u8 r4; \
575  register vec_u8 r5; \
576  register vec_u8 r6; \
577  register vec_u8 r7; \
578  \
579  r4 = vec_mergeh(r0, r2); /*0, 2 set 0*/ \
580  r5 = vec_mergel(r0, r2); /*0, 2 set 1*/ \
581  r6 = vec_mergeh(r1, r3); /*1, 3 set 0*/ \
582  r7 = vec_mergel(r1, r3); /*1, 3 set 1*/ \
583  \
584  r0 = vec_mergeh(r4, r6); /*all set 0*/ \
585  r1 = vec_mergel(r4, r6); /*all set 1*/ \
586  r2 = vec_mergeh(r5, r7); /*all set 2*/ \
587  r3 = vec_mergel(r5, r7); /*all set 3*/ \
588 }
589 
590 static inline void write16x4(uint8_t *dst, int dst_stride,
591  register vec_u8 r0, register vec_u8 r1,
592  register vec_u8 r2, register vec_u8 r3) {
593  DECLARE_ALIGNED(16, unsigned char, result)[64];
594  uint32_t *src_int = (uint32_t *)result, *dst_int = (uint32_t *)dst;
595  int int_dst_stride = dst_stride/4;
596 
597  vec_st(r0, 0, result);
598  vec_st(r1, 16, result);
599  vec_st(r2, 32, result);
600  vec_st(r3, 48, result);
601  /* FIXME: there has to be a better way!!!! */
602  *dst_int = *src_int;
603  *(dst_int+ int_dst_stride) = *(src_int + 1);
604  *(dst_int+ 2*int_dst_stride) = *(src_int + 2);
605  *(dst_int+ 3*int_dst_stride) = *(src_int + 3);
606  *(dst_int+ 4*int_dst_stride) = *(src_int + 4);
607  *(dst_int+ 5*int_dst_stride) = *(src_int + 5);
608  *(dst_int+ 6*int_dst_stride) = *(src_int + 6);
609  *(dst_int+ 7*int_dst_stride) = *(src_int + 7);
610  *(dst_int+ 8*int_dst_stride) = *(src_int + 8);
611  *(dst_int+ 9*int_dst_stride) = *(src_int + 9);
612  *(dst_int+10*int_dst_stride) = *(src_int + 10);
613  *(dst_int+11*int_dst_stride) = *(src_int + 11);
614  *(dst_int+12*int_dst_stride) = *(src_int + 12);
615  *(dst_int+13*int_dst_stride) = *(src_int + 13);
616  *(dst_int+14*int_dst_stride) = *(src_int + 14);
617  *(dst_int+15*int_dst_stride) = *(src_int + 15);
618 }
619 
620 /** @brief performs a 6x16 transpose of data in src, and stores it to dst
621  @todo FIXME: see if we can't spare some vec_lvsl() by them factorizing
622  out of unaligned_load() */
623 #define readAndTranspose16x6(src, src_stride, r8, r9, r10, r11, r12, r13) {\
624  register vec_u8 r0 = unaligned_load(0, src); \
625  register vec_u8 r1 = unaligned_load( src_stride, src); \
626  register vec_u8 r2 = unaligned_load(2* src_stride, src); \
627  register vec_u8 r3 = unaligned_load(3* src_stride, src); \
628  register vec_u8 r4 = unaligned_load(4* src_stride, src); \
629  register vec_u8 r5 = unaligned_load(5* src_stride, src); \
630  register vec_u8 r6 = unaligned_load(6* src_stride, src); \
631  register vec_u8 r7 = unaligned_load(7* src_stride, src); \
632  register vec_u8 r14 = unaligned_load(14*src_stride, src); \
633  register vec_u8 r15 = unaligned_load(15*src_stride, src); \
634  \
635  r8 = unaligned_load( 8*src_stride, src); \
636  r9 = unaligned_load( 9*src_stride, src); \
637  r10 = unaligned_load(10*src_stride, src); \
638  r11 = unaligned_load(11*src_stride, src); \
639  r12 = unaligned_load(12*src_stride, src); \
640  r13 = unaligned_load(13*src_stride, src); \
641  \
642  /*Merge first pairs*/ \
643  r0 = vec_mergeh(r0, r8); /*0, 8*/ \
644  r1 = vec_mergeh(r1, r9); /*1, 9*/ \
645  r2 = vec_mergeh(r2, r10); /*2,10*/ \
646  r3 = vec_mergeh(r3, r11); /*3,11*/ \
647  r4 = vec_mergeh(r4, r12); /*4,12*/ \
648  r5 = vec_mergeh(r5, r13); /*5,13*/ \
649  r6 = vec_mergeh(r6, r14); /*6,14*/ \
650  r7 = vec_mergeh(r7, r15); /*7,15*/ \
651  \
652  /*Merge second pairs*/ \
653  r8 = vec_mergeh(r0, r4); /*0,4, 8,12 set 0*/ \
654  r9 = vec_mergel(r0, r4); /*0,4, 8,12 set 1*/ \
655  r10 = vec_mergeh(r1, r5); /*1,5, 9,13 set 0*/ \
656  r11 = vec_mergel(r1, r5); /*1,5, 9,13 set 1*/ \
657  r12 = vec_mergeh(r2, r6); /*2,6,10,14 set 0*/ \
658  r13 = vec_mergel(r2, r6); /*2,6,10,14 set 1*/ \
659  r14 = vec_mergeh(r3, r7); /*3,7,11,15 set 0*/ \
660  r15 = vec_mergel(r3, r7); /*3,7,11,15 set 1*/ \
661  \
662  /*Third merge*/ \
663  r0 = vec_mergeh(r8, r12); /*0,2,4,6,8,10,12,14 set 0*/ \
664  r1 = vec_mergel(r8, r12); /*0,2,4,6,8,10,12,14 set 1*/ \
665  r2 = vec_mergeh(r9, r13); /*0,2,4,6,8,10,12,14 set 2*/ \
666  r4 = vec_mergeh(r10, r14); /*1,3,5,7,9,11,13,15 set 0*/ \
667  r5 = vec_mergel(r10, r14); /*1,3,5,7,9,11,13,15 set 1*/ \
668  r6 = vec_mergeh(r11, r15); /*1,3,5,7,9,11,13,15 set 2*/ \
669  /* Don't need to compute 3 and 7*/ \
670  \
671  /*Final merge*/ \
672  r8 = vec_mergeh(r0, r4); /*all set 0*/ \
673  r9 = vec_mergel(r0, r4); /*all set 1*/ \
674  r10 = vec_mergeh(r1, r5); /*all set 2*/ \
675  r11 = vec_mergel(r1, r5); /*all set 3*/ \
676  r12 = vec_mergeh(r2, r6); /*all set 4*/ \
677  r13 = vec_mergel(r2, r6); /*all set 5*/ \
678  /* Don't need to compute 14 and 15*/ \
679  \
680 }
681 
682 // out: o = |x-y| < a
683 static inline vec_u8 diff_lt_altivec ( register vec_u8 x,
684  register vec_u8 y,
685  register vec_u8 a) {
686 
687  register vec_u8 diff = vec_subs(x, y);
688  register vec_u8 diffneg = vec_subs(y, x);
689  register vec_u8 o = vec_or(diff, diffneg); /* |x-y| */
690  o = (vec_u8)vec_cmplt(o, a);
691  return o;
692 }
693 
694 static inline vec_u8 h264_deblock_mask ( register vec_u8 p0,
695  register vec_u8 p1,
696  register vec_u8 q0,
697  register vec_u8 q1,
698  register vec_u8 alpha,
699  register vec_u8 beta) {
700 
701  register vec_u8 mask;
702  register vec_u8 tempmask;
703 
704  mask = diff_lt_altivec(p0, q0, alpha);
705  tempmask = diff_lt_altivec(p1, p0, beta);
706  mask = vec_and(mask, tempmask);
707  tempmask = diff_lt_altivec(q1, q0, beta);
708  mask = vec_and(mask, tempmask);
709 
710  return mask;
711 }
712 
713 // out: newp1 = clip((p2 + ((p0 + q0 + 1) >> 1)) >> 1, p1-tc0, p1+tc0)
714 static inline vec_u8 h264_deblock_q1(register vec_u8 p0,
715  register vec_u8 p1,
716  register vec_u8 p2,
717  register vec_u8 q0,
718  register vec_u8 tc0) {
719 
720  register vec_u8 average = vec_avg(p0, q0);
721  register vec_u8 temp;
722  register vec_u8 uncliped;
723  register vec_u8 ones;
724  register vec_u8 max;
725  register vec_u8 min;
726  register vec_u8 newp1;
727 
728  temp = vec_xor(average, p2);
729  average = vec_avg(average, p2); /*avg(p2, avg(p0, q0)) */
730  ones = vec_splat_u8(1);
731  temp = vec_and(temp, ones); /*(p2^avg(p0, q0)) & 1 */
732  uncliped = vec_subs(average, temp); /*(p2+((p0+q0+1)>>1))>>1 */
733  max = vec_adds(p1, tc0);
734  min = vec_subs(p1, tc0);
735  newp1 = vec_max(min, uncliped);
736  newp1 = vec_min(max, newp1);
737  return newp1;
738 }
739 
740 #define h264_deblock_p0_q0(p0, p1, q0, q1, tc0masked) { \
741  \
742  const vec_u8 A0v = vec_sl(vec_splat_u8(10), vec_splat_u8(4)); \
743  \
744  register vec_u8 pq0bit = vec_xor(p0,q0); \
745  register vec_u8 q1minus; \
746  register vec_u8 p0minus; \
747  register vec_u8 stage1; \
748  register vec_u8 stage2; \
749  register vec_u8 vec160; \
750  register vec_u8 delta; \
751  register vec_u8 deltaneg; \
752  \
753  q1minus = vec_nor(q1, q1); /* 255 - q1 */ \
754  stage1 = vec_avg(p1, q1minus); /* (p1 - q1 + 256)>>1 */ \
755  stage2 = vec_sr(stage1, vec_splat_u8(1)); /* (p1 - q1 + 256)>>2 = 64 + (p1 - q1) >> 2 */ \
756  p0minus = vec_nor(p0, p0); /* 255 - p0 */ \
757  stage1 = vec_avg(q0, p0minus); /* (q0 - p0 + 256)>>1 */ \
758  pq0bit = vec_and(pq0bit, vec_splat_u8(1)); \
759  stage2 = vec_avg(stage2, pq0bit); /* 32 + ((q0 - p0)&1 + (p1 - q1) >> 2 + 1) >> 1 */ \
760  stage2 = vec_adds(stage2, stage1); /* 160 + ((p0 - q0) + (p1 - q1) >> 2 + 1) >> 1 */ \
761  vec160 = vec_ld(0, &A0v); \
762  deltaneg = vec_subs(vec160, stage2); /* -d */ \
763  delta = vec_subs(stage2, vec160); /* d */ \
764  deltaneg = vec_min(tc0masked, deltaneg); \
765  delta = vec_min(tc0masked, delta); \
766  p0 = vec_subs(p0, deltaneg); \
767  q0 = vec_subs(q0, delta); \
768  p0 = vec_adds(p0, delta); \
769  q0 = vec_adds(q0, deltaneg); \
770 }
771 
772 #define h264_loop_filter_luma_altivec(p2, p1, p0, q0, q1, q2, alpha, beta, tc0) { \
773  DECLARE_ALIGNED(16, unsigned char, temp)[16]; \
774  register vec_u8 alphavec; \
775  register vec_u8 betavec; \
776  register vec_u8 mask; \
777  register vec_u8 p1mask; \
778  register vec_u8 q1mask; \
779  register vector signed char tc0vec; \
780  register vec_u8 finaltc0; \
781  register vec_u8 tc0masked; \
782  register vec_u8 newp1; \
783  register vec_u8 newq1; \
784  \
785  temp[0] = alpha; \
786  temp[1] = beta; \
787  alphavec = vec_ld(0, temp); \
788  betavec = vec_splat(alphavec, 0x1); \
789  alphavec = vec_splat(alphavec, 0x0); \
790  mask = h264_deblock_mask(p0, p1, q0, q1, alphavec, betavec); /*if in block */ \
791  \
792  AV_COPY32(temp, tc0); \
793  tc0vec = vec_ld(0, (signed char*)temp); \
794  tc0vec = vec_mergeh(tc0vec, tc0vec); \
795  tc0vec = vec_mergeh(tc0vec, tc0vec); \
796  mask = vec_and(mask, vec_cmpgt(tc0vec, vec_splat_s8(-1))); /* if tc0[i] >= 0 */ \
797  finaltc0 = vec_and((vec_u8)tc0vec, mask); /* tc = tc0 */ \
798  \
799  p1mask = diff_lt_altivec(p2, p0, betavec); \
800  p1mask = vec_and(p1mask, mask); /* if ( |p2 - p0| < beta) */ \
801  tc0masked = vec_and(p1mask, (vec_u8)tc0vec); \
802  finaltc0 = vec_sub(finaltc0, p1mask); /* tc++ */ \
803  newp1 = h264_deblock_q1(p0, p1, p2, q0, tc0masked); \
804  /*end if*/ \
805  \
806  q1mask = diff_lt_altivec(q2, q0, betavec); \
807  q1mask = vec_and(q1mask, mask); /* if ( |q2 - q0| < beta ) */\
808  tc0masked = vec_and(q1mask, (vec_u8)tc0vec); \
809  finaltc0 = vec_sub(finaltc0, q1mask); /* tc++ */ \
810  newq1 = h264_deblock_q1(p0, q1, q2, q0, tc0masked); \
811  /*end if*/ \
812  \
813  h264_deblock_p0_q0(p0, p1, q0, q1, finaltc0); \
814  p1 = newp1; \
815  q1 = newq1; \
816 }
817 
818 static void h264_v_loop_filter_luma_altivec(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0) {
819 
820  if ((tc0[0] & tc0[1] & tc0[2] & tc0[3]) >= 0) {
821  register vec_u8 p2 = vec_ld(-3*stride, pix);
822  register vec_u8 p1 = vec_ld(-2*stride, pix);
823  register vec_u8 p0 = vec_ld(-1*stride, pix);
824  register vec_u8 q0 = vec_ld(0, pix);
825  register vec_u8 q1 = vec_ld(stride, pix);
826  register vec_u8 q2 = vec_ld(2*stride, pix);
827  h264_loop_filter_luma_altivec(p2, p1, p0, q0, q1, q2, alpha, beta, tc0);
828  vec_st(p1, -2*stride, pix);
829  vec_st(p0, -1*stride, pix);
830  vec_st(q0, 0, pix);
831  vec_st(q1, stride, pix);
832  }
833 }
834 
835 static void h264_h_loop_filter_luma_altivec(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0) {
836 
837  register vec_u8 line0, line1, line2, line3, line4, line5;
838  if ((tc0[0] & tc0[1] & tc0[2] & tc0[3]) < 0)
839  return;
840  readAndTranspose16x6(pix-3, stride, line0, line1, line2, line3, line4, line5);
841  h264_loop_filter_luma_altivec(line0, line1, line2, line3, line4, line5, alpha, beta, tc0);
842  transpose4x16(line1, line2, line3, line4);
843  write16x4(pix-2, stride, line1, line2, line3, line4);
844 }
845 
846 static av_always_inline
847 void weight_h264_W_altivec(uint8_t *block, int stride, int height,
848  int log2_denom, int weight, int offset, int w)
849 {
850  int y, aligned;
851  vec_u8 vblock;
852  vec_s16 vtemp, vweight, voffset, v0, v1;
853  vec_u16 vlog2_denom;
854  DECLARE_ALIGNED(16, int32_t, temp)[4];
855  LOAD_ZERO;
856 
857  offset <<= log2_denom;
858  if(log2_denom) offset += 1<<(log2_denom-1);
859  temp[0] = log2_denom;
860  temp[1] = weight;
861  temp[2] = offset;
862 
863  vtemp = (vec_s16)vec_ld(0, temp);
864  vlog2_denom = (vec_u16)vec_splat(vtemp, 1);
865  vweight = vec_splat(vtemp, 3);
866  voffset = vec_splat(vtemp, 5);
867  aligned = !((unsigned long)block & 0xf);
868 
869  for (y = 0; y < height; y++) {
870  vblock = vec_ld(0, block);
871 
872  v0 = (vec_s16)vec_mergeh(zero_u8v, vblock);
873  v1 = (vec_s16)vec_mergel(zero_u8v, vblock);
874 
875  if (w == 16 || aligned) {
876  v0 = vec_mladd(v0, vweight, zero_s16v);
877  v0 = vec_adds(v0, voffset);
878  v0 = vec_sra(v0, vlog2_denom);
879  }
880  if (w == 16 || !aligned) {
881  v1 = vec_mladd(v1, vweight, zero_s16v);
882  v1 = vec_adds(v1, voffset);
883  v1 = vec_sra(v1, vlog2_denom);
884  }
885  vblock = vec_packsu(v0, v1);
886  vec_st(vblock, 0, block);
887 
888  block += stride;
889  }
890 }
891 
892 static av_always_inline
893 void biweight_h264_W_altivec(uint8_t *dst, uint8_t *src, int stride, int height,
894  int log2_denom, int weightd, int weights, int offset, int w)
895 {
896  int y, dst_aligned, src_aligned;
897  vec_u8 vsrc, vdst;
898  vec_s16 vtemp, vweights, vweightd, voffset, v0, v1, v2, v3;
899  vec_u16 vlog2_denom;
900  DECLARE_ALIGNED(16, int32_t, temp)[4];
901  LOAD_ZERO;
902 
903  offset = ((offset + 1) | 1) << log2_denom;
904  temp[0] = log2_denom+1;
905  temp[1] = weights;
906  temp[2] = weightd;
907  temp[3] = offset;
908 
909  vtemp = (vec_s16)vec_ld(0, temp);
910  vlog2_denom = (vec_u16)vec_splat(vtemp, 1);
911  vweights = vec_splat(vtemp, 3);
912  vweightd = vec_splat(vtemp, 5);
913  voffset = vec_splat(vtemp, 7);
914  dst_aligned = !((unsigned long)dst & 0xf);
915  src_aligned = !((unsigned long)src & 0xf);
916 
917  for (y = 0; y < height; y++) {
918  vdst = vec_ld(0, dst);
919  vsrc = vec_ld(0, src);
920 
921  v0 = (vec_s16)vec_mergeh(zero_u8v, vdst);
922  v1 = (vec_s16)vec_mergel(zero_u8v, vdst);
923  v2 = (vec_s16)vec_mergeh(zero_u8v, vsrc);
924  v3 = (vec_s16)vec_mergel(zero_u8v, vsrc);
925 
926  if (w == 8) {
927  if (src_aligned)
928  v3 = v2;
929  else
930  v2 = v3;
931  }
932 
933  if (w == 16 || dst_aligned) {
934  v0 = vec_mladd(v0, vweightd, zero_s16v);
935  v2 = vec_mladd(v2, vweights, zero_s16v);
936 
937  v0 = vec_adds(v0, voffset);
938  v0 = vec_adds(v0, v2);
939  v0 = vec_sra(v0, vlog2_denom);
940  }
941  if (w == 16 || !dst_aligned) {
942  v1 = vec_mladd(v1, vweightd, zero_s16v);
943  v3 = vec_mladd(v3, vweights, zero_s16v);
944 
945  v1 = vec_adds(v1, voffset);
946  v1 = vec_adds(v1, v3);
947  v1 = vec_sra(v1, vlog2_denom);
948  }
949  vdst = vec_packsu(v0, v1);
950  vec_st(vdst, 0, dst);
951 
952  dst += stride;
953  src += stride;
954  }
955 }
956 
957 #define H264_WEIGHT(W) \
958 static void ff_weight_h264_pixels ## W ## _altivec(uint8_t *block, int stride, int height, \
959  int log2_denom, int weight, int offset){ \
960  weight_h264_W_altivec(block, stride, height, log2_denom, weight, offset, W); \
961 }\
962 static void ff_biweight_h264_pixels ## W ## _altivec(uint8_t *dst, uint8_t *src, int stride, int height, \
963  int log2_denom, int weightd, int weights, int offset){ \
964  biweight_h264_W_altivec(dst, src, stride, height, log2_denom, weightd, weights, offset, W); \
965 }
966 
967 H264_WEIGHT(16)
968 H264_WEIGHT( 8)
969 
970 void ff_dsputil_h264_init_ppc(DSPContext* c, AVCodecContext *avctx) {
971  const int high_bit_depth = avctx->bits_per_raw_sample > 8;
972 
973  if (av_get_cpu_flags() & AV_CPU_FLAG_ALTIVEC) {
974  if (!high_bit_depth) {
975  c->put_h264_chroma_pixels_tab[0] = put_h264_chroma_mc8_altivec;
976  c->avg_h264_chroma_pixels_tab[0] = avg_h264_chroma_mc8_altivec;
977 
978 #define dspfunc(PFX, IDX, NUM) \
979  c->PFX ## _pixels_tab[IDX][ 0] = PFX ## NUM ## _mc00_altivec; \
980  c->PFX ## _pixels_tab[IDX][ 1] = PFX ## NUM ## _mc10_altivec; \
981  c->PFX ## _pixels_tab[IDX][ 2] = PFX ## NUM ## _mc20_altivec; \
982  c->PFX ## _pixels_tab[IDX][ 3] = PFX ## NUM ## _mc30_altivec; \
983  c->PFX ## _pixels_tab[IDX][ 4] = PFX ## NUM ## _mc01_altivec; \
984  c->PFX ## _pixels_tab[IDX][ 5] = PFX ## NUM ## _mc11_altivec; \
985  c->PFX ## _pixels_tab[IDX][ 6] = PFX ## NUM ## _mc21_altivec; \
986  c->PFX ## _pixels_tab[IDX][ 7] = PFX ## NUM ## _mc31_altivec; \
987  c->PFX ## _pixels_tab[IDX][ 8] = PFX ## NUM ## _mc02_altivec; \
988  c->PFX ## _pixels_tab[IDX][ 9] = PFX ## NUM ## _mc12_altivec; \
989  c->PFX ## _pixels_tab[IDX][10] = PFX ## NUM ## _mc22_altivec; \
990  c->PFX ## _pixels_tab[IDX][11] = PFX ## NUM ## _mc32_altivec; \
991  c->PFX ## _pixels_tab[IDX][12] = PFX ## NUM ## _mc03_altivec; \
992  c->PFX ## _pixels_tab[IDX][13] = PFX ## NUM ## _mc13_altivec; \
993  c->PFX ## _pixels_tab[IDX][14] = PFX ## NUM ## _mc23_altivec; \
994  c->PFX ## _pixels_tab[IDX][15] = PFX ## NUM ## _mc33_altivec
995 
996  dspfunc(put_h264_qpel, 0, 16);
997  dspfunc(avg_h264_qpel, 0, 16);
998 #undef dspfunc
999  }
1000  }
1001 }
1002 
1003 void ff_h264dsp_init_ppc(H264DSPContext *c, const int bit_depth, const int chroma_format_idc)
1004 {
1005  if (av_get_cpu_flags() & AV_CPU_FLAG_ALTIVEC) {
1006  if (bit_depth == 8) {
1007  c->h264_idct_add = ff_h264_idct_add_altivec;
1008  if (chroma_format_idc == 1)
1009  c->h264_idct_add8 = ff_h264_idct_add8_altivec;
1010  c->h264_idct_add16 = ff_h264_idct_add16_altivec;
1011  c->h264_idct_add16intra = ff_h264_idct_add16intra_altivec;
1012  c->h264_idct_dc_add= h264_idct_dc_add_altivec;
1013  c->h264_idct8_dc_add = ff_h264_idct8_dc_add_altivec;
1014  c->h264_idct8_add = ff_h264_idct8_add_altivec;
1015  c->h264_idct8_add4 = ff_h264_idct8_add4_altivec;
1016  c->h264_v_loop_filter_luma= h264_v_loop_filter_luma_altivec;
1017  c->h264_h_loop_filter_luma= h264_h_loop_filter_luma_altivec;
1018 
1019  c->weight_h264_pixels_tab[0] = ff_weight_h264_pixels16_altivec;
1020  c->weight_h264_pixels_tab[1] = ff_weight_h264_pixels8_altivec;
1021  c->biweight_h264_pixels_tab[0] = ff_biweight_h264_pixels16_altivec;
1022  c->biweight_h264_pixels_tab[1] = ff_biweight_h264_pixels8_altivec;
1023  }
1024  }
1025 }
Generated on Sat May 25 2013 03:58:39 for FFmpeg by   doxygen 1.8.2