9d4c8435be93c25cfddb1e44c2cabd820d6b6c8b
[ffmpeg.git] / libavcodec / ppc / h264_altivec.c
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 "libavcodec/dsputil.h"
22 #include "libavcodec/h264data.h"
23
24 #include "gcc_fixes.h"
25
26 #include "dsputil_ppc.h"
27 #include "dsputil_altivec.h"
28 #include "util_altivec.h"
29 #include "types_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_template_altivec.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_template_altivec.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     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 /* this code assume that stride % 16 == 0 */
185 void put_no_rnd_h264_chroma_mc8_altivec(uint8_t * dst, uint8_t * src, int stride, int h, int x, int y) {
186    DECLARE_ALIGNED_16(signed int, ABCD[4]) =
187                         {((8 - x) * (8 - y)),
188                              ((x) * (8 - y)),
189                          ((8 - x) * (y)),
190                              ((x) * (y))};
191     register int i;
192     vec_u8_t fperm;
193     const vec_s32_t vABCD = vec_ld(0, ABCD);
194     const vec_s16_t vA = vec_splat((vec_s16_t)vABCD, 1);
195     const vec_s16_t vB = vec_splat((vec_s16_t)vABCD, 3);
196     const vec_s16_t vC = vec_splat((vec_s16_t)vABCD, 5);
197     const vec_s16_t vD = vec_splat((vec_s16_t)vABCD, 7);
198     LOAD_ZERO;
199     const vec_s16_t v28ss = vec_sub(vec_sl(vec_splat_s16(1),vec_splat_u16(5)),vec_splat_s16(4));
200     const vec_u16_t v6us  = vec_splat_u16(6);
201     register int loadSecond     = (((unsigned long)src) % 16) <= 7 ? 0 : 1;
202     register int reallyBadAlign = (((unsigned long)src) % 16) == 15 ? 1 : 0;
203
204     vec_u8_t vsrcAuc, vsrcBuc, vsrcperm0, vsrcperm1;
205     vec_u8_t vsrc0uc, vsrc1uc;
206     vec_s16_t vsrc0ssH, vsrc1ssH;
207     vec_u8_t vsrcCuc, vsrc2uc, vsrc3uc;
208     vec_s16_t vsrc2ssH, vsrc3ssH, psum;
209     vec_u8_t vdst, ppsum, fsum;
210
211     if (((unsigned long)dst) % 16 == 0) {
212         fperm = (vec_u8_t){0x10, 0x11, 0x12, 0x13,
213                            0x14, 0x15, 0x16, 0x17,
214                            0x08, 0x09, 0x0A, 0x0B,
215                            0x0C, 0x0D, 0x0E, 0x0F};
216     } else {
217         fperm = (vec_u8_t){0x00, 0x01, 0x02, 0x03,
218                            0x04, 0x05, 0x06, 0x07,
219                            0x18, 0x19, 0x1A, 0x1B,
220                            0x1C, 0x1D, 0x1E, 0x1F};
221     }
222
223     vsrcAuc = vec_ld(0, src);
224
225     if (loadSecond)
226         vsrcBuc = vec_ld(16, src);
227     vsrcperm0 = vec_lvsl(0, src);
228     vsrcperm1 = vec_lvsl(1, src);
229
230     vsrc0uc = vec_perm(vsrcAuc, vsrcBuc, vsrcperm0);
231     if (reallyBadAlign)
232         vsrc1uc = vsrcBuc;
233     else
234         vsrc1uc = vec_perm(vsrcAuc, vsrcBuc, vsrcperm1);
235
236     vsrc0ssH = (vec_s16_t)vec_mergeh(zero_u8v, (vec_u8_t)vsrc0uc);
237     vsrc1ssH = (vec_s16_t)vec_mergeh(zero_u8v, (vec_u8_t)vsrc1uc);
238
239     if (!loadSecond) {// -> !reallyBadAlign
240         for (i = 0 ; i < h ; i++) {
241
242
243             vsrcCuc = vec_ld(stride + 0, src);
244
245             vsrc2uc = vec_perm(vsrcCuc, vsrcCuc, vsrcperm0);
246             vsrc3uc = vec_perm(vsrcCuc, vsrcCuc, vsrcperm1);
247
248             vsrc2ssH = (vec_s16_t)vec_mergeh(zero_u8v, (vec_u8_t)vsrc2uc);
249             vsrc3ssH = (vec_s16_t)vec_mergeh(zero_u8v, (vec_u8_t)vsrc3uc);
250
251             psum = vec_mladd(vA, vsrc0ssH, vec_splat_s16(0));
252             psum = vec_mladd(vB, vsrc1ssH, psum);
253             psum = vec_mladd(vC, vsrc2ssH, psum);
254             psum = vec_mladd(vD, vsrc3ssH, psum);
255             psum = vec_add(v28ss, psum);
256             psum = vec_sra(psum, v6us);
257
258             vdst = vec_ld(0, dst);
259             ppsum = (vec_u8_t)vec_packsu(psum, psum);
260             fsum = vec_perm(vdst, ppsum, fperm);
261
262             vec_st(fsum, 0, dst);
263
264             vsrc0ssH = vsrc2ssH;
265             vsrc1ssH = vsrc3ssH;
266
267             dst += stride;
268             src += stride;
269         }
270     } else {
271         vec_u8_t vsrcDuc;
272         for (i = 0 ; i < h ; i++) {
273             vsrcCuc = vec_ld(stride + 0, src);
274             vsrcDuc = vec_ld(stride + 16, src);
275
276             vsrc2uc = vec_perm(vsrcCuc, vsrcDuc, vsrcperm0);
277             if (reallyBadAlign)
278                 vsrc3uc = vsrcDuc;
279             else
280                 vsrc3uc = vec_perm(vsrcCuc, vsrcDuc, vsrcperm1);
281
282             vsrc2ssH = (vec_s16_t)vec_mergeh(zero_u8v, (vec_u8_t)vsrc2uc);
283             vsrc3ssH = (vec_s16_t)vec_mergeh(zero_u8v, (vec_u8_t)vsrc3uc);
284
285             psum = vec_mladd(vA, vsrc0ssH, vec_splat_s16(0));
286             psum = vec_mladd(vB, vsrc1ssH, psum);
287             psum = vec_mladd(vC, vsrc2ssH, psum);
288             psum = vec_mladd(vD, vsrc3ssH, psum);
289             psum = vec_add(v28ss, psum);
290             psum = vec_sr(psum, v6us);
291
292             vdst = vec_ld(0, dst);
293             ppsum = (vec_u8_t)vec_pack(psum, psum);
294             fsum = vec_perm(vdst, ppsum, fperm);
295
296             vec_st(fsum, 0, dst);
297
298             vsrc0ssH = vsrc2ssH;
299             vsrc1ssH = vsrc3ssH;
300
301             dst += stride;
302             src += stride;
303         }
304     }
305 }
306
307 static inline void put_pixels16_l2_altivec( uint8_t * dst, const uint8_t * src1,
308                                     const uint8_t * src2, int dst_stride,
309                                     int src_stride1, int h)
310 {
311     int i;
312     vec_u8_t a, b, d, tmp1, tmp2, mask, mask_, edges, align;
313
314     mask_ = vec_lvsl(0, src2);
315
316     for (i = 0; i < h; i++) {
317
318         tmp1 = vec_ld(i * src_stride1, src1);
319         mask = vec_lvsl(i * src_stride1, src1);
320         tmp2 = vec_ld(i * src_stride1 + 15, src1);
321
322         a = vec_perm(tmp1, tmp2, mask);
323
324         tmp1 = vec_ld(i * 16, src2);
325         tmp2 = vec_ld(i * 16 + 15, src2);
326
327         b = vec_perm(tmp1, tmp2, mask_);
328
329         tmp1 = vec_ld(0, dst);
330         mask = vec_lvsl(0, dst);
331         tmp2 = vec_ld(15, dst);
332
333         d = vec_avg(a, b);
334
335         edges = vec_perm(tmp2, tmp1, mask);
336
337         align = vec_lvsr(0, dst);
338
339         tmp2 = vec_perm(d, edges, align);
340         tmp1 = vec_perm(edges, d, align);
341
342         vec_st(tmp2, 15, dst);
343         vec_st(tmp1, 0 , dst);
344
345         dst += dst_stride;
346     }
347 }
348
349 static inline void avg_pixels16_l2_altivec( uint8_t * dst, const uint8_t * src1,
350                                     const uint8_t * src2, int dst_stride,
351                                     int src_stride1, int h)
352 {
353     int i;
354     vec_u8_t a, b, d, tmp1, tmp2, mask, mask_, edges, align;
355
356     mask_ = vec_lvsl(0, src2);
357
358     for (i = 0; i < h; i++) {
359
360         tmp1 = vec_ld(i * src_stride1, src1);
361         mask = vec_lvsl(i * src_stride1, src1);
362         tmp2 = vec_ld(i * src_stride1 + 15, src1);
363
364         a = vec_perm(tmp1, tmp2, mask);
365
366         tmp1 = vec_ld(i * 16, src2);
367         tmp2 = vec_ld(i * 16 + 15, src2);
368
369         b = vec_perm(tmp1, tmp2, mask_);
370
371         tmp1 = vec_ld(0, dst);
372         mask = vec_lvsl(0, dst);
373         tmp2 = vec_ld(15, dst);
374
375         d = vec_avg(vec_perm(tmp1, tmp2, mask), vec_avg(a, b));
376
377         edges = vec_perm(tmp2, tmp1, mask);
378
379         align = vec_lvsr(0, dst);
380
381         tmp2 = vec_perm(d, edges, align);
382         tmp1 = vec_perm(edges, d, align);
383
384         vec_st(tmp2, 15, dst);
385         vec_st(tmp1, 0 , dst);
386
387         dst += dst_stride;
388     }
389 }
390
391 /* Implemented but could be faster
392 #define put_pixels16_l2_altivec(d,s1,s2,ds,s1s,h) put_pixels16_l2(d,s1,s2,ds,s1s,16,h)
393 #define avg_pixels16_l2_altivec(d,s1,s2,ds,s1s,h) avg_pixels16_l2(d,s1,s2,ds,s1s,16,h)
394  */
395
396 H264_MC(put_, 16, altivec)
397 H264_MC(avg_, 16, altivec)
398
399
400 /****************************************************************************
401  * IDCT transform:
402  ****************************************************************************/
403
404 #define VEC_1D_DCT(vb0,vb1,vb2,vb3,va0,va1,va2,va3)               \
405     /* 1st stage */                                               \
406     vz0 = vec_add(vb0,vb2);       /* temp[0] = Y[0] + Y[2] */     \
407     vz1 = vec_sub(vb0,vb2);       /* temp[1] = Y[0] - Y[2] */     \
408     vz2 = vec_sra(vb1,vec_splat_u16(1));                          \
409     vz2 = vec_sub(vz2,vb3);       /* temp[2] = Y[1].1/2 - Y[3] */ \
410     vz3 = vec_sra(vb3,vec_splat_u16(1));                          \
411     vz3 = vec_add(vb1,vz3);       /* temp[3] = Y[1] + Y[3].1/2 */ \
412     /* 2nd stage: output */                                       \
413     va0 = vec_add(vz0,vz3);       /* x[0] = temp[0] + temp[3] */  \
414     va1 = vec_add(vz1,vz2);       /* x[1] = temp[1] + temp[2] */  \
415     va2 = vec_sub(vz1,vz2);       /* x[2] = temp[1] - temp[2] */  \
416     va3 = vec_sub(vz0,vz3)        /* x[3] = temp[0] - temp[3] */
417
418 #define VEC_TRANSPOSE_4(a0,a1,a2,a3,b0,b1,b2,b3) \
419     b0 = vec_mergeh( a0, a0 ); \
420     b1 = vec_mergeh( a1, a0 ); \
421     b2 = vec_mergeh( a2, a0 ); \
422     b3 = vec_mergeh( a3, a0 ); \
423     a0 = vec_mergeh( b0, b2 ); \
424     a1 = vec_mergel( b0, b2 ); \
425     a2 = vec_mergeh( b1, b3 ); \
426     a3 = vec_mergel( b1, b3 ); \
427     b0 = vec_mergeh( a0, a2 ); \
428     b1 = vec_mergel( a0, a2 ); \
429     b2 = vec_mergeh( a1, a3 ); \
430     b3 = vec_mergel( a1, a3 )
431
432 #define VEC_LOAD_U8_ADD_S16_STORE_U8(va)                      \
433     vdst_orig = vec_ld(0, dst);                               \
434     vdst = vec_perm(vdst_orig, zero_u8v, vdst_mask);          \
435     vdst_ss = (vec_s16_t) vec_mergeh(zero_u8v, vdst);         \
436     va = vec_add(va, vdst_ss);                                \
437     va_u8 = vec_packsu(va, zero_s16v);                        \
438     va_u32 = vec_splat((vec_u32_t)va_u8, 0);                  \
439     vec_ste(va_u32, element, (uint32_t*)dst);
440
441 static void ff_h264_idct_add_altivec(uint8_t *dst, DCTELEM *block, int stride)
442 {
443     vec_s16_t va0, va1, va2, va3;
444     vec_s16_t vz0, vz1, vz2, vz3;
445     vec_s16_t vtmp0, vtmp1, vtmp2, vtmp3;
446     vec_u8_t va_u8;
447     vec_u32_t va_u32;
448     vec_s16_t vdst_ss;
449     const vec_u16_t v6us = vec_splat_u16(6);
450     vec_u8_t vdst, vdst_orig;
451     vec_u8_t vdst_mask = vec_lvsl(0, dst);
452     int element = ((unsigned long)dst & 0xf) >> 2;
453     LOAD_ZERO;
454
455     block[0] += 32;  /* add 32 as a DC-level for rounding */
456
457     vtmp0 = vec_ld(0,block);
458     vtmp1 = vec_sld(vtmp0, vtmp0, 8);
459     vtmp2 = vec_ld(16,block);
460     vtmp3 = vec_sld(vtmp2, vtmp2, 8);
461
462     VEC_1D_DCT(vtmp0,vtmp1,vtmp2,vtmp3,va0,va1,va2,va3);
463     VEC_TRANSPOSE_4(va0,va1,va2,va3,vtmp0,vtmp1,vtmp2,vtmp3);
464     VEC_1D_DCT(vtmp0,vtmp1,vtmp2,vtmp3,va0,va1,va2,va3);
465
466     va0 = vec_sra(va0,v6us);
467     va1 = vec_sra(va1,v6us);
468     va2 = vec_sra(va2,v6us);
469     va3 = vec_sra(va3,v6us);
470
471     VEC_LOAD_U8_ADD_S16_STORE_U8(va0);
472     dst += stride;
473     VEC_LOAD_U8_ADD_S16_STORE_U8(va1);
474     dst += stride;
475     VEC_LOAD_U8_ADD_S16_STORE_U8(va2);
476     dst += stride;
477     VEC_LOAD_U8_ADD_S16_STORE_U8(va3);
478 }
479
480 #define IDCT8_1D_ALTIVEC(s0, s1, s2, s3, s4, s5, s6, s7,  d0, d1, d2, d3, d4, d5, d6, d7) {\
481     /*        a0  = SRC(0) + SRC(4); */ \
482     vec_s16_t a0v = vec_add(s0, s4);    \
483     /*        a2  = SRC(0) - SRC(4); */ \
484     vec_s16_t a2v = vec_sub(s0, s4);    \
485     /*        a4  =           (SRC(2)>>1) - SRC(6); */ \
486     vec_s16_t a4v = vec_sub(vec_sra(s2, onev), s6);    \
487     /*        a6  =           (SRC(6)>>1) + SRC(2); */ \
488     vec_s16_t a6v = vec_add(vec_sra(s6, onev), s2);    \
489     /*        b0  =         a0 + a6; */ \
490     vec_s16_t b0v = vec_add(a0v, a6v);  \
491     /*        b2  =         a2 + a4; */ \
492     vec_s16_t b2v = vec_add(a2v, a4v);  \
493     /*        b4  =         a2 - a4; */ \
494     vec_s16_t b4v = vec_sub(a2v, a4v);  \
495     /*        b6  =         a0 - a6; */ \
496     vec_s16_t b6v = vec_sub(a0v, a6v);  \
497     /* a1 =  SRC(5) - SRC(3) - SRC(7) - (SRC(7)>>1); */ \
498     /*        a1 =             (SRC(5)-SRC(3)) -  (SRC(7)  +  (SRC(7)>>1)); */ \
499     vec_s16_t a1v = vec_sub( vec_sub(s5, s3), vec_add(s7, vec_sra(s7, onev)) ); \
500     /* a3 =  SRC(7) + SRC(1) - SRC(3) - (SRC(3)>>1); */ \
501     /*        a3 =             (SRC(7)+SRC(1)) -  (SRC(3)  +  (SRC(3)>>1)); */ \
502     vec_s16_t a3v = vec_sub( vec_add(s7, s1), vec_add(s3, vec_sra(s3, onev)) );\
503     /* a5 =  SRC(7) - SRC(1) + SRC(5) + (SRC(5)>>1); */ \
504     /*        a5 =             (SRC(7)-SRC(1)) +   SRC(5) +   (SRC(5)>>1); */ \
505     vec_s16_t a5v = vec_add( vec_sub(s7, s1), vec_add(s5, vec_sra(s5, onev)) );\
506     /*        a7 =                SRC(5)+SRC(3) +  SRC(1) +   (SRC(1)>>1); */ \
507     vec_s16_t a7v = vec_add( vec_add(s5, s3), vec_add(s1, vec_sra(s1, onev)) );\
508     /*        b1 =                  (a7>>2)  +  a1; */ \
509     vec_s16_t b1v = vec_add( vec_sra(a7v, twov), a1v); \
510     /*        b3 =          a3 +        (a5>>2); */ \
511     vec_s16_t b3v = vec_add(a3v, vec_sra(a5v, twov)); \
512     /*        b5 =                  (a3>>2)  -   a5; */ \
513     vec_s16_t b5v = vec_sub( vec_sra(a3v, twov), a5v); \
514     /*        b7 =           a7 -        (a1>>2); */ \
515     vec_s16_t b7v = vec_sub( a7v, vec_sra(a1v, twov)); \
516     /* DST(0,    b0 + b7); */ \
517     d0 = vec_add(b0v, b7v); \
518     /* DST(1,    b2 + b5); */ \
519     d1 = vec_add(b2v, b5v); \
520     /* DST(2,    b4 + b3); */ \
521     d2 = vec_add(b4v, b3v); \
522     /* DST(3,    b6 + b1); */ \
523     d3 = vec_add(b6v, b1v); \
524     /* DST(4,    b6 - b1); */ \
525     d4 = vec_sub(b6v, b1v); \
526     /* DST(5,    b4 - b3); */ \
527     d5 = vec_sub(b4v, b3v); \
528     /* DST(6,    b2 - b5); */ \
529     d6 = vec_sub(b2v, b5v); \
530     /* DST(7,    b0 - b7); */ \
531     d7 = vec_sub(b0v, b7v); \
532 }
533
534 #define ALTIVEC_STORE_SUM_CLIP(dest, idctv, perm_ldv, perm_stv, sel) { \
535     /* unaligned load */                                       \
536     vec_u8_t hv = vec_ld( 0, dest );                           \
537     vec_u8_t lv = vec_ld( 7, dest );                           \
538     vec_u8_t dstv   = vec_perm( hv, lv, (vec_u8_t)perm_ldv );  \
539     vec_s16_t idct_sh6 = vec_sra(idctv, sixv);                 \
540     vec_u16_t dst16 = (vec_u16_t)vec_mergeh(zero_u8v, dstv);   \
541     vec_s16_t idstsum = vec_adds(idct_sh6, (vec_s16_t)dst16);  \
542     vec_u8_t idstsum8 = vec_packsu(zero_s16v, idstsum);        \
543     vec_u8_t edgehv;                                           \
544     /* unaligned store */                                      \
545     vec_u8_t bodyv  = vec_perm( idstsum8, idstsum8, perm_stv );\
546     vec_u8_t edgelv = vec_perm( sel, zero_u8v, perm_stv );     \
547     lv    = vec_sel( lv, bodyv, edgelv );                      \
548     vec_st( lv, 7, dest );                                     \
549     hv    = vec_ld( 0, dest );                                 \
550     edgehv = vec_perm( zero_u8v, sel, perm_stv );              \
551     hv    = vec_sel( hv, bodyv, edgehv );                      \
552     vec_st( hv, 0, dest );                                     \
553  }
554
555 void ff_h264_idct8_add_altivec( uint8_t *dst, DCTELEM *dct, int stride ) {
556     vec_s16_t s0, s1, s2, s3, s4, s5, s6, s7;
557     vec_s16_t d0, d1, d2, d3, d4, d5, d6, d7;
558     vec_s16_t idct0, idct1, idct2, idct3, idct4, idct5, idct6, idct7;
559
560     vec_u8_t perm_ldv = vec_lvsl(0, dst);
561     vec_u8_t perm_stv = vec_lvsr(8, dst);
562
563     const vec_u16_t onev = vec_splat_u16(1);
564     const vec_u16_t twov = vec_splat_u16(2);
565     const vec_u16_t sixv = vec_splat_u16(6);
566
567     const vec_u8_t sel = (vec_u8_t) {0,0,0,0,0,0,0,0,-1,-1,-1,-1,-1,-1,-1,-1};
568     LOAD_ZERO;
569
570     dct[0] += 32; // rounding for the >>6 at the end
571
572     s0 = vec_ld(0x00, (int16_t*)dct);
573     s1 = vec_ld(0x10, (int16_t*)dct);
574     s2 = vec_ld(0x20, (int16_t*)dct);
575     s3 = vec_ld(0x30, (int16_t*)dct);
576     s4 = vec_ld(0x40, (int16_t*)dct);
577     s5 = vec_ld(0x50, (int16_t*)dct);
578     s6 = vec_ld(0x60, (int16_t*)dct);
579     s7 = vec_ld(0x70, (int16_t*)dct);
580
581     IDCT8_1D_ALTIVEC(s0, s1, s2, s3, s4, s5, s6, s7,
582                      d0, d1, d2, d3, d4, d5, d6, d7);
583
584     TRANSPOSE8( d0,  d1,  d2,  d3,  d4,  d5,  d6, d7 );
585
586     IDCT8_1D_ALTIVEC(d0,  d1,  d2,  d3,  d4,  d5,  d6, d7,
587                      idct0, idct1, idct2, idct3, idct4, idct5, idct6, idct7);
588
589     ALTIVEC_STORE_SUM_CLIP(&dst[0*stride], idct0, perm_ldv, perm_stv, sel);
590     ALTIVEC_STORE_SUM_CLIP(&dst[1*stride], idct1, perm_ldv, perm_stv, sel);
591     ALTIVEC_STORE_SUM_CLIP(&dst[2*stride], idct2, perm_ldv, perm_stv, sel);
592     ALTIVEC_STORE_SUM_CLIP(&dst[3*stride], idct3, perm_ldv, perm_stv, sel);
593     ALTIVEC_STORE_SUM_CLIP(&dst[4*stride], idct4, perm_ldv, perm_stv, sel);
594     ALTIVEC_STORE_SUM_CLIP(&dst[5*stride], idct5, perm_ldv, perm_stv, sel);
595     ALTIVEC_STORE_SUM_CLIP(&dst[6*stride], idct6, perm_ldv, perm_stv, sel);
596     ALTIVEC_STORE_SUM_CLIP(&dst[7*stride], idct7, perm_ldv, perm_stv, sel);
597 }
598
599 // TODO: implement this in AltiVec
600 static void ff_h264_idct8_dc_add_altivec(uint8_t *dst, DCTELEM *block, int stride) {
601     int i, j;
602     uint8_t *cm = ff_cropTbl + MAX_NEG_CROP;
603     int dc = (block[0] + 32) >> 6;
604     for( j = 0; j < 8; j++ )
605     {
606         for( i = 0; i < 8; i++ )
607             dst[i] = cm[ dst[i] + dc ];
608         dst += stride;
609     }
610 }
611
612 static void ff_h264_idct8_add4_altivec(uint8_t *dst, const int *block_offset, DCTELEM *block, int stride, const uint8_t nnzc[6*8]){
613     int i;
614     for(i=0; i<16; i+=4){
615         int nnz = nnzc[ scan8[i] ];
616         if(nnz){
617             if(nnz==1 && block[i*16]) ff_h264_idct8_dc_add_altivec(dst + block_offset[i], block + i*16, stride);
618             else                      ff_h264_idct8_add_altivec   (dst + block_offset[i], block + i*16, stride);
619         }
620     }
621 }
622
623 #define transpose4x16(r0, r1, r2, r3) {      \
624     register vec_u8_t r4;                    \
625     register vec_u8_t r5;                    \
626     register vec_u8_t r6;                    \
627     register vec_u8_t r7;                    \
628                                              \
629     r4 = vec_mergeh(r0, r2);  /*0, 2 set 0*/ \
630     r5 = vec_mergel(r0, r2);  /*0, 2 set 1*/ \
631     r6 = vec_mergeh(r1, r3);  /*1, 3 set 0*/ \
632     r7 = vec_mergel(r1, r3);  /*1, 3 set 1*/ \
633                                              \
634     r0 = vec_mergeh(r4, r6);  /*all set 0*/  \
635     r1 = vec_mergel(r4, r6);  /*all set 1*/  \
636     r2 = vec_mergeh(r5, r7);  /*all set 2*/  \
637     r3 = vec_mergel(r5, r7);  /*all set 3*/  \
638 }
639
640 static inline void write16x4(uint8_t *dst, int dst_stride,
641                              register vec_u8_t r0, register vec_u8_t r1,
642                              register vec_u8_t r2, register vec_u8_t r3) {
643     DECLARE_ALIGNED_16(unsigned char, result[64]);
644     uint32_t *src_int = (uint32_t *)result, *dst_int = (uint32_t *)dst;
645     int int_dst_stride = dst_stride/4;
646
647     vec_st(r0, 0, result);
648     vec_st(r1, 16, result);
649     vec_st(r2, 32, result);
650     vec_st(r3, 48, result);
651     /* FIXME: there has to be a better way!!!! */
652     *dst_int = *src_int;
653     *(dst_int+   int_dst_stride) = *(src_int + 1);
654     *(dst_int+ 2*int_dst_stride) = *(src_int + 2);
655     *(dst_int+ 3*int_dst_stride) = *(src_int + 3);
656     *(dst_int+ 4*int_dst_stride) = *(src_int + 4);
657     *(dst_int+ 5*int_dst_stride) = *(src_int + 5);
658     *(dst_int+ 6*int_dst_stride) = *(src_int + 6);
659     *(dst_int+ 7*int_dst_stride) = *(src_int + 7);
660     *(dst_int+ 8*int_dst_stride) = *(src_int + 8);
661     *(dst_int+ 9*int_dst_stride) = *(src_int + 9);
662     *(dst_int+10*int_dst_stride) = *(src_int + 10);
663     *(dst_int+11*int_dst_stride) = *(src_int + 11);
664     *(dst_int+12*int_dst_stride) = *(src_int + 12);
665     *(dst_int+13*int_dst_stride) = *(src_int + 13);
666     *(dst_int+14*int_dst_stride) = *(src_int + 14);
667     *(dst_int+15*int_dst_stride) = *(src_int + 15);
668 }
669
670 /** \brief performs a 6x16 transpose of data in src, and stores it to dst
671     \todo FIXME: see if we can't spare some vec_lvsl() by them factorizing
672     out of unaligned_load() */
673 #define readAndTranspose16x6(src, src_stride, r8, r9, r10, r11, r12, r13) {\
674     register vec_u8_t r0  = unaligned_load(0,             src);            \
675     register vec_u8_t r1  = unaligned_load(   src_stride, src);            \
676     register vec_u8_t r2  = unaligned_load(2* src_stride, src);            \
677     register vec_u8_t r3  = unaligned_load(3* src_stride, src);            \
678     register vec_u8_t r4  = unaligned_load(4* src_stride, src);            \
679     register vec_u8_t r5  = unaligned_load(5* src_stride, src);            \
680     register vec_u8_t r6  = unaligned_load(6* src_stride, src);            \
681     register vec_u8_t r7  = unaligned_load(7* src_stride, src);            \
682     register vec_u8_t r14 = unaligned_load(14*src_stride, src);            \
683     register vec_u8_t r15 = unaligned_load(15*src_stride, src);            \
684                                                                            \
685     r8  = unaligned_load( 8*src_stride, src);                              \
686     r9  = unaligned_load( 9*src_stride, src);                              \
687     r10 = unaligned_load(10*src_stride, src);                              \
688     r11 = unaligned_load(11*src_stride, src);                              \
689     r12 = unaligned_load(12*src_stride, src);                              \
690     r13 = unaligned_load(13*src_stride, src);                              \
691                                                                            \
692     /*Merge first pairs*/                                                  \
693     r0 = vec_mergeh(r0, r8);    /*0, 8*/                                   \
694     r1 = vec_mergeh(r1, r9);    /*1, 9*/                                   \
695     r2 = vec_mergeh(r2, r10);   /*2,10*/                                   \
696     r3 = vec_mergeh(r3, r11);   /*3,11*/                                   \
697     r4 = vec_mergeh(r4, r12);   /*4,12*/                                   \
698     r5 = vec_mergeh(r5, r13);   /*5,13*/                                   \
699     r6 = vec_mergeh(r6, r14);   /*6,14*/                                   \
700     r7 = vec_mergeh(r7, r15);   /*7,15*/                                   \
701                                                                            \
702     /*Merge second pairs*/                                                 \
703     r8  = vec_mergeh(r0, r4);   /*0,4, 8,12 set 0*/                        \
704     r9  = vec_mergel(r0, r4);   /*0,4, 8,12 set 1*/                        \
705     r10 = vec_mergeh(r1, r5);   /*1,5, 9,13 set 0*/                        \
706     r11 = vec_mergel(r1, r5);   /*1,5, 9,13 set 1*/                        \
707     r12 = vec_mergeh(r2, r6);   /*2,6,10,14 set 0*/                        \
708     r13 = vec_mergel(r2, r6);   /*2,6,10,14 set 1*/                        \
709     r14 = vec_mergeh(r3, r7);   /*3,7,11,15 set 0*/                        \
710     r15 = vec_mergel(r3, r7);   /*3,7,11,15 set 1*/                        \
711                                                                            \
712     /*Third merge*/                                                        \
713     r0 = vec_mergeh(r8,  r12);  /*0,2,4,6,8,10,12,14 set 0*/               \
714     r1 = vec_mergel(r8,  r12);  /*0,2,4,6,8,10,12,14 set 1*/               \
715     r2 = vec_mergeh(r9,  r13);  /*0,2,4,6,8,10,12,14 set 2*/               \
716     r4 = vec_mergeh(r10, r14);  /*1,3,5,7,9,11,13,15 set 0*/               \
717     r5 = vec_mergel(r10, r14);  /*1,3,5,7,9,11,13,15 set 1*/               \
718     r6 = vec_mergeh(r11, r15);  /*1,3,5,7,9,11,13,15 set 2*/               \
719     /* Don't need to compute 3 and 7*/                                     \
720                                                                            \
721     /*Final merge*/                                                        \
722     r8  = vec_mergeh(r0, r4);   /*all set 0*/                              \
723     r9  = vec_mergel(r0, r4);   /*all set 1*/                              \
724     r10 = vec_mergeh(r1, r5);   /*all set 2*/                              \
725     r11 = vec_mergel(r1, r5);   /*all set 3*/                              \
726     r12 = vec_mergeh(r2, r6);   /*all set 4*/                              \
727     r13 = vec_mergel(r2, r6);   /*all set 5*/                              \
728     /* Don't need to compute 14 and 15*/                                   \
729                                                                            \
730 }
731
732 // out: o = |x-y| < a
733 static inline vec_u8_t diff_lt_altivec ( register vec_u8_t x,
734                                          register vec_u8_t y,
735                                          register vec_u8_t a) {
736
737     register vec_u8_t diff = vec_subs(x, y);
738     register vec_u8_t diffneg = vec_subs(y, x);
739     register vec_u8_t o = vec_or(diff, diffneg); /* |x-y| */
740     o = (vec_u8_t)vec_cmplt(o, a);
741     return o;
742 }
743
744 static inline vec_u8_t h264_deblock_mask ( register vec_u8_t p0,
745                                            register vec_u8_t p1,
746                                            register vec_u8_t q0,
747                                            register vec_u8_t q1,
748                                            register vec_u8_t alpha,
749                                            register vec_u8_t beta) {
750
751     register vec_u8_t mask;
752     register vec_u8_t tempmask;
753
754     mask = diff_lt_altivec(p0, q0, alpha);
755     tempmask = diff_lt_altivec(p1, p0, beta);
756     mask = vec_and(mask, tempmask);
757     tempmask = diff_lt_altivec(q1, q0, beta);
758     mask = vec_and(mask, tempmask);
759
760     return mask;
761 }
762
763 // out: newp1 = clip((p2 + ((p0 + q0 + 1) >> 1)) >> 1, p1-tc0, p1+tc0)
764 static inline vec_u8_t h264_deblock_q1(register vec_u8_t p0,
765                                        register vec_u8_t p1,
766                                        register vec_u8_t p2,
767                                        register vec_u8_t q0,
768                                        register vec_u8_t tc0) {
769
770     register vec_u8_t average = vec_avg(p0, q0);
771     register vec_u8_t temp;
772     register vec_u8_t uncliped;
773     register vec_u8_t ones;
774     register vec_u8_t max;
775     register vec_u8_t min;
776     register vec_u8_t newp1;
777
778     temp = vec_xor(average, p2);
779     average = vec_avg(average, p2);     /*avg(p2, avg(p0, q0)) */
780     ones = vec_splat_u8(1);
781     temp = vec_and(temp, ones);         /*(p2^avg(p0, q0)) & 1 */
782     uncliped = vec_subs(average, temp); /*(p2+((p0+q0+1)>>1))>>1 */
783     max = vec_adds(p1, tc0);
784     min = vec_subs(p1, tc0);
785     newp1 = vec_max(min, uncliped);
786     newp1 = vec_min(max, newp1);
787     return newp1;
788 }
789
790 #define h264_deblock_p0_q0(p0, p1, q0, q1, tc0masked) {                                           \
791                                                                                                   \
792     const vec_u8_t A0v = vec_sl(vec_splat_u8(10), vec_splat_u8(4));                               \
793                                                                                                   \
794     register vec_u8_t pq0bit = vec_xor(p0,q0);                                                    \
795     register vec_u8_t q1minus;                                                                    \
796     register vec_u8_t p0minus;                                                                    \
797     register vec_u8_t stage1;                                                                     \
798     register vec_u8_t stage2;                                                                     \
799     register vec_u8_t vec160;                                                                     \
800     register vec_u8_t delta;                                                                      \
801     register vec_u8_t deltaneg;                                                                   \
802                                                                                                   \
803     q1minus = vec_nor(q1, q1);                 /* 255 - q1 */                                     \
804     stage1 = vec_avg(p1, q1minus);             /* (p1 - q1 + 256)>>1 */                           \
805     stage2 = vec_sr(stage1, vec_splat_u8(1));  /* (p1 - q1 + 256)>>2 = 64 + (p1 - q1) >> 2 */     \
806     p0minus = vec_nor(p0, p0);                 /* 255 - p0 */                                     \
807     stage1 = vec_avg(q0, p0minus);             /* (q0 - p0 + 256)>>1 */                           \
808     pq0bit = vec_and(pq0bit, vec_splat_u8(1));                                                    \
809     stage2 = vec_avg(stage2, pq0bit);          /* 32 + ((q0 - p0)&1 + (p1 - q1) >> 2 + 1) >> 1 */ \
810     stage2 = vec_adds(stage2, stage1);         /* 160 + ((p0 - q0) + (p1 - q1) >> 2 + 1) >> 1 */  \
811     vec160 = vec_ld(0, &A0v);                                                                     \
812     deltaneg = vec_subs(vec160, stage2);       /* -d */                                           \
813     delta = vec_subs(stage2, vec160);          /* d */                                            \
814     deltaneg = vec_min(tc0masked, deltaneg);                                                      \
815     delta = vec_min(tc0masked, delta);                                                            \
816     p0 = vec_subs(p0, deltaneg);                                                                  \
817     q0 = vec_subs(q0, delta);                                                                     \
818     p0 = vec_adds(p0, delta);                                                                     \
819     q0 = vec_adds(q0, deltaneg);                                                                  \
820 }
821
822 #define h264_loop_filter_luma_altivec(p2, p1, p0, q0, q1, q2, alpha, beta, tc0) {            \
823     DECLARE_ALIGNED_16(unsigned char, temp[16]);                                             \
824     register vec_u8_t alphavec;                                                              \
825     register vec_u8_t betavec;                                                               \
826     register vec_u8_t mask;                                                                  \
827     register vec_u8_t p1mask;                                                                \
828     register vec_u8_t q1mask;                                                                \
829     register vector signed   char tc0vec;                                                    \
830     register vec_u8_t finaltc0;                                                              \
831     register vec_u8_t tc0masked;                                                             \
832     register vec_u8_t newp1;                                                                 \
833     register vec_u8_t newq1;                                                                 \
834                                                                                              \
835     temp[0] = alpha;                                                                         \
836     temp[1] = beta;                                                                          \
837     alphavec = vec_ld(0, temp);                                                              \
838     betavec = vec_splat(alphavec, 0x1);                                                      \
839     alphavec = vec_splat(alphavec, 0x0);                                                     \
840     mask = h264_deblock_mask(p0, p1, q0, q1, alphavec, betavec); /*if in block */            \
841                                                                                              \
842     *((int *)temp) = *((int *)tc0);                                                          \
843     tc0vec = vec_ld(0, (signed char*)temp);                                                  \
844     tc0vec = vec_mergeh(tc0vec, tc0vec);                                                     \
845     tc0vec = vec_mergeh(tc0vec, tc0vec);                                                     \
846     mask = vec_and(mask, vec_cmpgt(tc0vec, vec_splat_s8(-1)));  /* if tc0[i] >= 0 */         \
847     finaltc0 = vec_and((vec_u8_t)tc0vec, mask);     /* tc = tc0 */                           \
848                                                                                              \
849     p1mask = diff_lt_altivec(p2, p0, betavec);                                               \
850     p1mask = vec_and(p1mask, mask);                             /* if ( |p2 - p0| < beta) */ \
851     tc0masked = vec_and(p1mask, (vec_u8_t)tc0vec);                                           \
852     finaltc0 = vec_sub(finaltc0, p1mask);                       /* tc++ */                   \
853     newp1 = h264_deblock_q1(p0, p1, p2, q0, tc0masked);                                      \
854     /*end if*/                                                                               \
855                                                                                              \
856     q1mask = diff_lt_altivec(q2, q0, betavec);                                               \
857     q1mask = vec_and(q1mask, mask);                             /* if ( |q2 - q0| < beta ) */\
858     tc0masked = vec_and(q1mask, (vec_u8_t)tc0vec);                                           \
859     finaltc0 = vec_sub(finaltc0, q1mask);                       /* tc++ */                   \
860     newq1 = h264_deblock_q1(p0, q1, q2, q0, tc0masked);                                      \
861     /*end if*/                                                                               \
862                                                                                              \
863     h264_deblock_p0_q0(p0, p1, q0, q1, finaltc0);                                            \
864     p1 = newp1;                                                                              \
865     q1 = newq1;                                                                              \
866 }
867
868 static void h264_v_loop_filter_luma_altivec(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0) {
869
870     if ((tc0[0] & tc0[1] & tc0[2] & tc0[3]) >= 0) {
871         register vec_u8_t p2 = vec_ld(-3*stride, pix);
872         register vec_u8_t p1 = vec_ld(-2*stride, pix);
873         register vec_u8_t p0 = vec_ld(-1*stride, pix);
874         register vec_u8_t q0 = vec_ld(0, pix);
875         register vec_u8_t q1 = vec_ld(stride, pix);
876         register vec_u8_t q2 = vec_ld(2*stride, pix);
877         h264_loop_filter_luma_altivec(p2, p1, p0, q0, q1, q2, alpha, beta, tc0);
878         vec_st(p1, -2*stride, pix);
879         vec_st(p0, -1*stride, pix);
880         vec_st(q0, 0, pix);
881         vec_st(q1, stride, pix);
882     }
883 }
884
885 static void h264_h_loop_filter_luma_altivec(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0) {
886
887     register vec_u8_t line0, line1, line2, line3, line4, line5;
888     if ((tc0[0] & tc0[1] & tc0[2] & tc0[3]) < 0)
889         return;
890     readAndTranspose16x6(pix-3, stride, line0, line1, line2, line3, line4, line5);
891     h264_loop_filter_luma_altivec(line0, line1, line2, line3, line4, line5, alpha, beta, tc0);
892     transpose4x16(line1, line2, line3, line4);
893     write16x4(pix-2, stride, line1, line2, line3, line4);
894 }
895
896 void dsputil_h264_init_ppc(DSPContext* c, AVCodecContext *avctx) {
897
898     if (has_altivec()) {
899         c->put_h264_chroma_pixels_tab[0] = put_h264_chroma_mc8_altivec;
900         c->put_no_rnd_h264_chroma_pixels_tab[0] = put_no_rnd_h264_chroma_mc8_altivec;
901         c->avg_h264_chroma_pixels_tab[0] = avg_h264_chroma_mc8_altivec;
902 /* ff_h264_idct_add_altivec may be re-enabled once AltiVec versions of
903    h264_idct_add16, h264_idct_add16intra, h264_idct_add8 are implemented
904         c->h264_idct_add = ff_h264_idct_add_altivec;
905 */
906         c->h264_idct8_add = ff_h264_idct8_add_altivec;
907         c->h264_idct8_add4 = ff_h264_idct8_add4_altivec;
908         c->h264_v_loop_filter_luma= h264_v_loop_filter_luma_altivec;
909         c->h264_h_loop_filter_luma= h264_h_loop_filter_luma_altivec;
910
911 #define dspfunc(PFX, IDX, NUM) \
912         c->PFX ## _pixels_tab[IDX][ 0] = PFX ## NUM ## _mc00_altivec; \
913         c->PFX ## _pixels_tab[IDX][ 1] = PFX ## NUM ## _mc10_altivec; \
914         c->PFX ## _pixels_tab[IDX][ 2] = PFX ## NUM ## _mc20_altivec; \
915         c->PFX ## _pixels_tab[IDX][ 3] = PFX ## NUM ## _mc30_altivec; \
916         c->PFX ## _pixels_tab[IDX][ 4] = PFX ## NUM ## _mc01_altivec; \
917         c->PFX ## _pixels_tab[IDX][ 5] = PFX ## NUM ## _mc11_altivec; \
918         c->PFX ## _pixels_tab[IDX][ 6] = PFX ## NUM ## _mc21_altivec; \
919         c->PFX ## _pixels_tab[IDX][ 7] = PFX ## NUM ## _mc31_altivec; \
920         c->PFX ## _pixels_tab[IDX][ 8] = PFX ## NUM ## _mc02_altivec; \
921         c->PFX ## _pixels_tab[IDX][ 9] = PFX ## NUM ## _mc12_altivec; \
922         c->PFX ## _pixels_tab[IDX][10] = PFX ## NUM ## _mc22_altivec; \
923         c->PFX ## _pixels_tab[IDX][11] = PFX ## NUM ## _mc32_altivec; \
924         c->PFX ## _pixels_tab[IDX][12] = PFX ## NUM ## _mc03_altivec; \
925         c->PFX ## _pixels_tab[IDX][13] = PFX ## NUM ## _mc13_altivec; \
926         c->PFX ## _pixels_tab[IDX][14] = PFX ## NUM ## _mc23_altivec; \
927         c->PFX ## _pixels_tab[IDX][15] = PFX ## NUM ## _mc33_altivec
928
929         dspfunc(put_h264_qpel, 0, 16);
930         dspfunc(avg_h264_qpel, 0, 16);
931 #undef dspfunc
932     }
933 }