Remove gcc_fixes.h. It only contains workarounds for unsupported gcc versions.
[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 "dsputil_ppc.h"
25 #include "dsputil_altivec.h"
26 #include "util_altivec.h"
27 #include "types_altivec.h"
28
29 #define PUT_OP_U8_ALTIVEC(d, s, dst) d = s
30 #define AVG_OP_U8_ALTIVEC(d, s, dst) d = vec_avg(dst, s)
31
32 #define OP_U8_ALTIVEC                          PUT_OP_U8_ALTIVEC
33 #define PREFIX_h264_chroma_mc8_altivec         put_h264_chroma_mc8_altivec
34 #define PREFIX_h264_chroma_mc8_num             altivec_put_h264_chroma_mc8_num
35 #define PREFIX_h264_qpel16_h_lowpass_altivec   put_h264_qpel16_h_lowpass_altivec
36 #define PREFIX_h264_qpel16_h_lowpass_num       altivec_put_h264_qpel16_h_lowpass_num
37 #define PREFIX_h264_qpel16_v_lowpass_altivec   put_h264_qpel16_v_lowpass_altivec
38 #define PREFIX_h264_qpel16_v_lowpass_num       altivec_put_h264_qpel16_v_lowpass_num
39 #define PREFIX_h264_qpel16_hv_lowpass_altivec  put_h264_qpel16_hv_lowpass_altivec
40 #define PREFIX_h264_qpel16_hv_lowpass_num      altivec_put_h264_qpel16_hv_lowpass_num
41 #include "h264_template_altivec.c"
42 #undef OP_U8_ALTIVEC
43 #undef PREFIX_h264_chroma_mc8_altivec
44 #undef PREFIX_h264_chroma_mc8_num
45 #undef PREFIX_h264_qpel16_h_lowpass_altivec
46 #undef PREFIX_h264_qpel16_h_lowpass_num
47 #undef PREFIX_h264_qpel16_v_lowpass_altivec
48 #undef PREFIX_h264_qpel16_v_lowpass_num
49 #undef PREFIX_h264_qpel16_hv_lowpass_altivec
50 #undef PREFIX_h264_qpel16_hv_lowpass_num
51
52 #define OP_U8_ALTIVEC                          AVG_OP_U8_ALTIVEC
53 #define PREFIX_h264_chroma_mc8_altivec         avg_h264_chroma_mc8_altivec
54 #define PREFIX_h264_chroma_mc8_num             altivec_avg_h264_chroma_mc8_num
55 #define PREFIX_h264_qpel16_h_lowpass_altivec   avg_h264_qpel16_h_lowpass_altivec
56 #define PREFIX_h264_qpel16_h_lowpass_num       altivec_avg_h264_qpel16_h_lowpass_num
57 #define PREFIX_h264_qpel16_v_lowpass_altivec   avg_h264_qpel16_v_lowpass_altivec
58 #define PREFIX_h264_qpel16_v_lowpass_num       altivec_avg_h264_qpel16_v_lowpass_num
59 #define PREFIX_h264_qpel16_hv_lowpass_altivec  avg_h264_qpel16_hv_lowpass_altivec
60 #define PREFIX_h264_qpel16_hv_lowpass_num      altivec_avg_h264_qpel16_hv_lowpass_num
61 #include "h264_template_altivec.c"
62 #undef OP_U8_ALTIVEC
63 #undef PREFIX_h264_chroma_mc8_altivec
64 #undef PREFIX_h264_chroma_mc8_num
65 #undef PREFIX_h264_qpel16_h_lowpass_altivec
66 #undef PREFIX_h264_qpel16_h_lowpass_num
67 #undef PREFIX_h264_qpel16_v_lowpass_altivec
68 #undef PREFIX_h264_qpel16_v_lowpass_num
69 #undef PREFIX_h264_qpel16_hv_lowpass_altivec
70 #undef PREFIX_h264_qpel16_hv_lowpass_num
71
72 #define H264_MC(OPNAME, SIZE, CODETYPE) \
73 static void OPNAME ## h264_qpel ## SIZE ## _mc00_ ## CODETYPE (uint8_t *dst, uint8_t *src, int stride){\
74     OPNAME ## pixels ## SIZE ## _ ## CODETYPE(dst, src, stride, SIZE);\
75 }\
76 \
77 static void OPNAME ## h264_qpel ## SIZE ## _mc10_ ## CODETYPE(uint8_t *dst, uint8_t *src, int stride){ \
78     DECLARE_ALIGNED_16(uint8_t, half[SIZE*SIZE]);\
79     put_h264_qpel ## SIZE ## _h_lowpass_ ## CODETYPE(half, src, SIZE, stride);\
80     OPNAME ## pixels ## SIZE ## _l2_ ## CODETYPE(dst, src, half, stride, stride, SIZE);\
81 }\
82 \
83 static void OPNAME ## h264_qpel ## SIZE ## _mc20_ ## CODETYPE(uint8_t *dst, uint8_t *src, int stride){\
84     OPNAME ## h264_qpel ## SIZE ## _h_lowpass_ ## CODETYPE(dst, src, stride, stride);\
85 }\
86 \
87 static void OPNAME ## h264_qpel ## SIZE ## _mc30_ ## CODETYPE(uint8_t *dst, uint8_t *src, int stride){\
88     DECLARE_ALIGNED_16(uint8_t, half[SIZE*SIZE]);\
89     put_h264_qpel ## SIZE ## _h_lowpass_ ## CODETYPE(half, src, SIZE, stride);\
90     OPNAME ## pixels ## SIZE ## _l2_ ## CODETYPE(dst, src+1, half, stride, stride, SIZE);\
91 }\
92 \
93 static void OPNAME ## h264_qpel ## SIZE ## _mc01_ ## CODETYPE(uint8_t *dst, uint8_t *src, int stride){\
94     DECLARE_ALIGNED_16(uint8_t, half[SIZE*SIZE]);\
95     put_h264_qpel ## SIZE ## _v_lowpass_ ## CODETYPE(half, src, SIZE, stride);\
96     OPNAME ## pixels ## SIZE ## _l2_ ## CODETYPE(dst, src, half, stride, stride, SIZE);\
97 }\
98 \
99 static void OPNAME ## h264_qpel ## SIZE ## _mc02_ ## CODETYPE(uint8_t *dst, uint8_t *src, int stride){\
100     OPNAME ## h264_qpel ## SIZE ## _v_lowpass_ ## CODETYPE(dst, src, stride, stride);\
101 }\
102 \
103 static void OPNAME ## h264_qpel ## SIZE ## _mc03_ ## CODETYPE(uint8_t *dst, uint8_t *src, int stride){\
104     DECLARE_ALIGNED_16(uint8_t, half[SIZE*SIZE]);\
105     put_h264_qpel ## SIZE ## _v_lowpass_ ## CODETYPE(half, src, SIZE, stride);\
106     OPNAME ## pixels ## SIZE ## _l2_ ## CODETYPE(dst, src+stride, half, stride, stride, SIZE);\
107 }\
108 \
109 static void OPNAME ## h264_qpel ## SIZE ## _mc11_ ## CODETYPE(uint8_t *dst, uint8_t *src, int stride){\
110     DECLARE_ALIGNED_16(uint8_t, halfH[SIZE*SIZE]);\
111     DECLARE_ALIGNED_16(uint8_t, halfV[SIZE*SIZE]);\
112     put_h264_qpel ## SIZE ## _h_lowpass_ ## CODETYPE(halfH, src, SIZE, stride);\
113     put_h264_qpel ## SIZE ## _v_lowpass_ ## CODETYPE(halfV, src, SIZE, stride);\
114     OPNAME ## pixels ## SIZE ## _l2_ ## CODETYPE(dst, halfH, halfV, stride, SIZE, SIZE);\
115 }\
116 \
117 static void OPNAME ## h264_qpel ## SIZE ## _mc31_ ## CODETYPE(uint8_t *dst, uint8_t *src, int stride){\
118     DECLARE_ALIGNED_16(uint8_t, halfH[SIZE*SIZE]);\
119     DECLARE_ALIGNED_16(uint8_t, halfV[SIZE*SIZE]);\
120     put_h264_qpel ## SIZE ## _h_lowpass_ ## CODETYPE(halfH, src, SIZE, stride);\
121     put_h264_qpel ## SIZE ## _v_lowpass_ ## CODETYPE(halfV, src+1, SIZE, stride);\
122     OPNAME ## pixels ## SIZE ## _l2_ ## CODETYPE(dst, halfH, halfV, stride, SIZE, SIZE);\
123 }\
124 \
125 static void OPNAME ## h264_qpel ## SIZE ## _mc13_ ## CODETYPE(uint8_t *dst, uint8_t *src, int stride){\
126     DECLARE_ALIGNED_16(uint8_t, halfH[SIZE*SIZE]);\
127     DECLARE_ALIGNED_16(uint8_t, halfV[SIZE*SIZE]);\
128     put_h264_qpel ## SIZE ## _h_lowpass_ ## CODETYPE(halfH, src + stride, SIZE, stride);\
129     put_h264_qpel ## SIZE ## _v_lowpass_ ## CODETYPE(halfV, src, SIZE, stride);\
130     OPNAME ## pixels ## SIZE ## _l2_ ## CODETYPE(dst, halfH, halfV, stride, SIZE, SIZE);\
131 }\
132 \
133 static void OPNAME ## h264_qpel ## SIZE ## _mc33_ ## CODETYPE(uint8_t *dst, uint8_t *src, int stride){\
134     DECLARE_ALIGNED_16(uint8_t, halfH[SIZE*SIZE]);\
135     DECLARE_ALIGNED_16(uint8_t, halfV[SIZE*SIZE]);\
136     put_h264_qpel ## SIZE ## _h_lowpass_ ## CODETYPE(halfH, src + stride, SIZE, stride);\
137     put_h264_qpel ## SIZE ## _v_lowpass_ ## CODETYPE(halfV, src+1, SIZE, stride);\
138     OPNAME ## pixels ## SIZE ## _l2_ ## CODETYPE(dst, halfH, halfV, stride, SIZE, SIZE);\
139 }\
140 \
141 static void OPNAME ## h264_qpel ## SIZE ## _mc22_ ## CODETYPE(uint8_t *dst, uint8_t *src, int stride){\
142     DECLARE_ALIGNED_16(int16_t, tmp[SIZE*(SIZE+8)]);\
143     OPNAME ## h264_qpel ## SIZE ## _hv_lowpass_ ## CODETYPE(dst, tmp, src, stride, SIZE, stride);\
144 }\
145 \
146 static void OPNAME ## h264_qpel ## SIZE ## _mc21_ ## CODETYPE(uint8_t *dst, uint8_t *src, int stride){\
147     DECLARE_ALIGNED_16(uint8_t, halfH[SIZE*SIZE]);\
148     DECLARE_ALIGNED_16(uint8_t, halfHV[SIZE*SIZE]);\
149     DECLARE_ALIGNED_16(int16_t, tmp[SIZE*(SIZE+8)]);\
150     put_h264_qpel ## SIZE ## _h_lowpass_ ## CODETYPE(halfH, src, SIZE, stride);\
151     put_h264_qpel ## SIZE ## _hv_lowpass_ ## CODETYPE(halfHV, tmp, src, SIZE, SIZE, stride);\
152     OPNAME ## pixels ## SIZE ## _l2_ ## CODETYPE(dst, halfH, halfHV, stride, SIZE, SIZE);\
153 }\
154 \
155 static void OPNAME ## h264_qpel ## SIZE ## _mc23_ ## CODETYPE(uint8_t *dst, uint8_t *src, int stride){\
156     DECLARE_ALIGNED_16(uint8_t, halfH[SIZE*SIZE]);\
157     DECLARE_ALIGNED_16(uint8_t, halfHV[SIZE*SIZE]);\
158     DECLARE_ALIGNED_16(int16_t, tmp[SIZE*(SIZE+8)]);\
159     put_h264_qpel ## SIZE ## _h_lowpass_ ## CODETYPE(halfH, src + stride, SIZE, stride);\
160     put_h264_qpel ## SIZE ## _hv_lowpass_ ## CODETYPE(halfHV, tmp, src, SIZE, SIZE, stride);\
161     OPNAME ## pixels ## SIZE ## _l2_ ## CODETYPE(dst, halfH, halfHV, stride, SIZE, SIZE);\
162 }\
163 \
164 static void OPNAME ## h264_qpel ## SIZE ## _mc12_ ## CODETYPE(uint8_t *dst, uint8_t *src, int stride){\
165     DECLARE_ALIGNED_16(uint8_t, halfV[SIZE*SIZE]);\
166     DECLARE_ALIGNED_16(uint8_t, halfHV[SIZE*SIZE]);\
167     DECLARE_ALIGNED_16(int16_t, tmp[SIZE*(SIZE+8)]);\
168     put_h264_qpel ## SIZE ## _v_lowpass_ ## CODETYPE(halfV, src, SIZE, stride);\
169     put_h264_qpel ## SIZE ## _hv_lowpass_ ## CODETYPE(halfHV, tmp, src, SIZE, SIZE, stride);\
170     OPNAME ## pixels ## SIZE ## _l2_ ## CODETYPE(dst, halfV, halfHV, stride, SIZE, SIZE);\
171 }\
172 \
173 static void OPNAME ## h264_qpel ## SIZE ## _mc32_ ## CODETYPE(uint8_t *dst, uint8_t *src, int stride){\
174     DECLARE_ALIGNED_16(uint8_t, halfV[SIZE*SIZE]);\
175     DECLARE_ALIGNED_16(uint8_t, halfHV[SIZE*SIZE]);\
176     DECLARE_ALIGNED_16(int16_t, tmp[SIZE*(SIZE+8)]);\
177     put_h264_qpel ## SIZE ## _v_lowpass_ ## CODETYPE(halfV, src+1, SIZE, stride);\
178     put_h264_qpel ## SIZE ## _hv_lowpass_ ## CODETYPE(halfHV, tmp, src, SIZE, SIZE, stride);\
179     OPNAME ## pixels ## SIZE ## _l2_ ## CODETYPE(dst, halfV, halfHV, stride, SIZE, SIZE);\
180 }\
181
182 /* this code assume that stride % 16 == 0 */
183 void put_no_rnd_h264_chroma_mc8_altivec(uint8_t * dst, uint8_t * src, int stride, int h, int x, int y) {
184    DECLARE_ALIGNED_16(signed int, ABCD[4]) =
185                         {((8 - x) * (8 - y)),
186                          ((    x) * (8 - y)),
187                          ((8 - x) * (    y)),
188                          ((    x) * (    y))};
189     register int i;
190     vec_u8 fperm;
191     const vec_s32 vABCD = vec_ld(0, ABCD);
192     const vec_s16 vA = vec_splat((vec_s16)vABCD, 1);
193     const vec_s16 vB = vec_splat((vec_s16)vABCD, 3);
194     const vec_s16 vC = vec_splat((vec_s16)vABCD, 5);
195     const vec_s16 vD = vec_splat((vec_s16)vABCD, 7);
196     LOAD_ZERO;
197     const vec_s16 v28ss = vec_sub(vec_sl(vec_splat_s16(1),vec_splat_u16(5)),vec_splat_s16(4));
198     const vec_u16 v6us  = vec_splat_u16(6);
199     register int loadSecond     = (((unsigned long)src) % 16) <= 7 ? 0 : 1;
200     register int reallyBadAlign = (((unsigned long)src) % 16) == 15 ? 1 : 0;
201
202     vec_u8 vsrcAuc, av_uninit(vsrcBuc), vsrcperm0, vsrcperm1;
203     vec_u8 vsrc0uc, vsrc1uc;
204     vec_s16 vsrc0ssH, vsrc1ssH;
205     vec_u8 vsrcCuc, vsrc2uc, vsrc3uc;
206     vec_s16 vsrc2ssH, vsrc3ssH, psum;
207     vec_u8 vdst, ppsum, fsum;
208
209     if (((unsigned long)dst) % 16 == 0) {
210         fperm = (vec_u8){0x10, 0x11, 0x12, 0x13,
211                          0x14, 0x15, 0x16, 0x17,
212                          0x08, 0x09, 0x0A, 0x0B,
213                          0x0C, 0x0D, 0x0E, 0x0F};
214     } else {
215         fperm = (vec_u8){0x00, 0x01, 0x02, 0x03,
216                          0x04, 0x05, 0x06, 0x07,
217                          0x18, 0x19, 0x1A, 0x1B,
218                          0x1C, 0x1D, 0x1E, 0x1F};
219     }
220
221     vsrcAuc = vec_ld(0, src);
222
223     if (loadSecond)
224         vsrcBuc = vec_ld(16, src);
225     vsrcperm0 = vec_lvsl(0, src);
226     vsrcperm1 = vec_lvsl(1, src);
227
228     vsrc0uc = vec_perm(vsrcAuc, vsrcBuc, vsrcperm0);
229     if (reallyBadAlign)
230         vsrc1uc = vsrcBuc;
231     else
232         vsrc1uc = vec_perm(vsrcAuc, vsrcBuc, vsrcperm1);
233
234     vsrc0ssH = (vec_s16)vec_mergeh(zero_u8v, (vec_u8)vsrc0uc);
235     vsrc1ssH = (vec_s16)vec_mergeh(zero_u8v, (vec_u8)vsrc1uc);
236
237     if (!loadSecond) {// -> !reallyBadAlign
238         for (i = 0 ; i < h ; i++) {
239
240
241             vsrcCuc = vec_ld(stride + 0, src);
242
243             vsrc2uc = vec_perm(vsrcCuc, vsrcCuc, vsrcperm0);
244             vsrc3uc = vec_perm(vsrcCuc, vsrcCuc, vsrcperm1);
245
246             vsrc2ssH = (vec_s16)vec_mergeh(zero_u8v, (vec_u8)vsrc2uc);
247             vsrc3ssH = (vec_s16)vec_mergeh(zero_u8v, (vec_u8)vsrc3uc);
248
249             psum = vec_mladd(vA, vsrc0ssH, vec_splat_s16(0));
250             psum = vec_mladd(vB, vsrc1ssH, psum);
251             psum = vec_mladd(vC, vsrc2ssH, psum);
252             psum = vec_mladd(vD, vsrc3ssH, psum);
253             psum = vec_add(v28ss, psum);
254             psum = vec_sra(psum, v6us);
255
256             vdst = vec_ld(0, dst);
257             ppsum = (vec_u8)vec_packsu(psum, psum);
258             fsum = vec_perm(vdst, ppsum, fperm);
259
260             vec_st(fsum, 0, dst);
261
262             vsrc0ssH = vsrc2ssH;
263             vsrc1ssH = vsrc3ssH;
264
265             dst += stride;
266             src += stride;
267         }
268     } else {
269         vec_u8 vsrcDuc;
270         for (i = 0 ; i < h ; i++) {
271             vsrcCuc = vec_ld(stride + 0, src);
272             vsrcDuc = vec_ld(stride + 16, src);
273
274             vsrc2uc = vec_perm(vsrcCuc, vsrcDuc, vsrcperm0);
275             if (reallyBadAlign)
276                 vsrc3uc = vsrcDuc;
277             else
278                 vsrc3uc = vec_perm(vsrcCuc, vsrcDuc, vsrcperm1);
279
280             vsrc2ssH = (vec_s16)vec_mergeh(zero_u8v, (vec_u8)vsrc2uc);
281             vsrc3ssH = (vec_s16)vec_mergeh(zero_u8v, (vec_u8)vsrc3uc);
282
283             psum = vec_mladd(vA, vsrc0ssH, vec_splat_s16(0));
284             psum = vec_mladd(vB, vsrc1ssH, psum);
285             psum = vec_mladd(vC, vsrc2ssH, psum);
286             psum = vec_mladd(vD, vsrc3ssH, psum);
287             psum = vec_add(v28ss, psum);
288             psum = vec_sr(psum, v6us);
289
290             vdst = vec_ld(0, dst);
291             ppsum = (vec_u8)vec_pack(psum, psum);
292             fsum = vec_perm(vdst, ppsum, fperm);
293
294             vec_st(fsum, 0, dst);
295
296             vsrc0ssH = vsrc2ssH;
297             vsrc1ssH = vsrc3ssH;
298
299             dst += stride;
300             src += stride;
301         }
302     }
303 }
304
305 static inline void put_pixels16_l2_altivec( uint8_t * dst, const uint8_t * src1,
306                                     const uint8_t * src2, int dst_stride,
307                                     int src_stride1, int h)
308 {
309     int i;
310     vec_u8 a, b, d, tmp1, tmp2, mask, mask_, edges, align;
311
312     mask_ = vec_lvsl(0, src2);
313
314     for (i = 0; i < h; i++) {
315
316         tmp1 = vec_ld(i * src_stride1, src1);
317         mask = vec_lvsl(i * src_stride1, src1);
318         tmp2 = vec_ld(i * src_stride1 + 15, src1);
319
320         a = vec_perm(tmp1, tmp2, mask);
321
322         tmp1 = vec_ld(i * 16, src2);
323         tmp2 = vec_ld(i * 16 + 15, src2);
324
325         b = vec_perm(tmp1, tmp2, mask_);
326
327         tmp1 = vec_ld(0, dst);
328         mask = vec_lvsl(0, dst);
329         tmp2 = vec_ld(15, dst);
330
331         d = vec_avg(a, b);
332
333         edges = vec_perm(tmp2, tmp1, mask);
334
335         align = vec_lvsr(0, dst);
336
337         tmp2 = vec_perm(d, edges, align);
338         tmp1 = vec_perm(edges, d, align);
339
340         vec_st(tmp2, 15, dst);
341         vec_st(tmp1, 0 , dst);
342
343         dst += dst_stride;
344     }
345 }
346
347 static inline void avg_pixels16_l2_altivec( uint8_t * dst, const uint8_t * src1,
348                                     const uint8_t * src2, int dst_stride,
349                                     int src_stride1, int h)
350 {
351     int i;
352     vec_u8 a, b, d, tmp1, tmp2, mask, mask_, edges, align;
353
354     mask_ = vec_lvsl(0, src2);
355
356     for (i = 0; i < h; i++) {
357
358         tmp1 = vec_ld(i * src_stride1, src1);
359         mask = vec_lvsl(i * src_stride1, src1);
360         tmp2 = vec_ld(i * src_stride1 + 15, src1);
361
362         a = vec_perm(tmp1, tmp2, mask);
363
364         tmp1 = vec_ld(i * 16, src2);
365         tmp2 = vec_ld(i * 16 + 15, src2);
366
367         b = vec_perm(tmp1, tmp2, mask_);
368
369         tmp1 = vec_ld(0, dst);
370         mask = vec_lvsl(0, dst);
371         tmp2 = vec_ld(15, dst);
372
373         d = vec_avg(vec_perm(tmp1, tmp2, mask), vec_avg(a, b));
374
375         edges = vec_perm(tmp2, tmp1, mask);
376
377         align = vec_lvsr(0, dst);
378
379         tmp2 = vec_perm(d, edges, align);
380         tmp1 = vec_perm(edges, d, align);
381
382         vec_st(tmp2, 15, dst);
383         vec_st(tmp1, 0 , dst);
384
385         dst += dst_stride;
386     }
387 }
388
389 /* Implemented but could be faster
390 #define put_pixels16_l2_altivec(d,s1,s2,ds,s1s,h) put_pixels16_l2(d,s1,s2,ds,s1s,16,h)
391 #define avg_pixels16_l2_altivec(d,s1,s2,ds,s1s,h) avg_pixels16_l2(d,s1,s2,ds,s1s,16,h)
392  */
393
394 H264_MC(put_, 16, altivec)
395 H264_MC(avg_, 16, altivec)
396
397
398 /****************************************************************************
399  * IDCT transform:
400  ****************************************************************************/
401
402 #define VEC_1D_DCT(vb0,vb1,vb2,vb3,va0,va1,va2,va3)               \
403     /* 1st stage */                                               \
404     vz0 = vec_add(vb0,vb2);       /* temp[0] = Y[0] + Y[2] */     \
405     vz1 = vec_sub(vb0,vb2);       /* temp[1] = Y[0] - Y[2] */     \
406     vz2 = vec_sra(vb1,vec_splat_u16(1));                          \
407     vz2 = vec_sub(vz2,vb3);       /* temp[2] = Y[1].1/2 - Y[3] */ \
408     vz3 = vec_sra(vb3,vec_splat_u16(1));                          \
409     vz3 = vec_add(vb1,vz3);       /* temp[3] = Y[1] + Y[3].1/2 */ \
410     /* 2nd stage: output */                                       \
411     va0 = vec_add(vz0,vz3);       /* x[0] = temp[0] + temp[3] */  \
412     va1 = vec_add(vz1,vz2);       /* x[1] = temp[1] + temp[2] */  \
413     va2 = vec_sub(vz1,vz2);       /* x[2] = temp[1] - temp[2] */  \
414     va3 = vec_sub(vz0,vz3)        /* x[3] = temp[0] - temp[3] */
415
416 #define VEC_TRANSPOSE_4(a0,a1,a2,a3,b0,b1,b2,b3) \
417     b0 = vec_mergeh( a0, a0 ); \
418     b1 = vec_mergeh( a1, a0 ); \
419     b2 = vec_mergeh( a2, a0 ); \
420     b3 = vec_mergeh( a3, a0 ); \
421     a0 = vec_mergeh( b0, b2 ); \
422     a1 = vec_mergel( b0, b2 ); \
423     a2 = vec_mergeh( b1, b3 ); \
424     a3 = vec_mergel( b1, b3 ); \
425     b0 = vec_mergeh( a0, a2 ); \
426     b1 = vec_mergel( a0, a2 ); \
427     b2 = vec_mergeh( a1, a3 ); \
428     b3 = vec_mergel( a1, a3 )
429
430 #define VEC_LOAD_U8_ADD_S16_STORE_U8(va)                      \
431     vdst_orig = vec_ld(0, dst);                               \
432     vdst = vec_perm(vdst_orig, zero_u8v, vdst_mask);          \
433     vdst_ss = (vec_s16) vec_mergeh(zero_u8v, vdst);         \
434     va = vec_add(va, vdst_ss);                                \
435     va_u8 = vec_packsu(va, zero_s16v);                        \
436     va_u32 = vec_splat((vec_u32)va_u8, 0);                  \
437     vec_ste(va_u32, element, (uint32_t*)dst);
438
439 static void ff_h264_idct_add_altivec(uint8_t *dst, DCTELEM *block, int stride)
440 {
441     vec_s16 va0, va1, va2, va3;
442     vec_s16 vz0, vz1, vz2, vz3;
443     vec_s16 vtmp0, vtmp1, vtmp2, vtmp3;
444     vec_u8 va_u8;
445     vec_u32 va_u32;
446     vec_s16 vdst_ss;
447     const vec_u16 v6us = vec_splat_u16(6);
448     vec_u8 vdst, vdst_orig;
449     vec_u8 vdst_mask = vec_lvsl(0, dst);
450     int element = ((unsigned long)dst & 0xf) >> 2;
451     LOAD_ZERO;
452
453     block[0] += 32;  /* add 32 as a DC-level for rounding */
454
455     vtmp0 = vec_ld(0,block);
456     vtmp1 = vec_sld(vtmp0, vtmp0, 8);
457     vtmp2 = vec_ld(16,block);
458     vtmp3 = vec_sld(vtmp2, vtmp2, 8);
459
460     VEC_1D_DCT(vtmp0,vtmp1,vtmp2,vtmp3,va0,va1,va2,va3);
461     VEC_TRANSPOSE_4(va0,va1,va2,va3,vtmp0,vtmp1,vtmp2,vtmp3);
462     VEC_1D_DCT(vtmp0,vtmp1,vtmp2,vtmp3,va0,va1,va2,va3);
463
464     va0 = vec_sra(va0,v6us);
465     va1 = vec_sra(va1,v6us);
466     va2 = vec_sra(va2,v6us);
467     va3 = vec_sra(va3,v6us);
468
469     VEC_LOAD_U8_ADD_S16_STORE_U8(va0);
470     dst += stride;
471     VEC_LOAD_U8_ADD_S16_STORE_U8(va1);
472     dst += stride;
473     VEC_LOAD_U8_ADD_S16_STORE_U8(va2);
474     dst += stride;
475     VEC_LOAD_U8_ADD_S16_STORE_U8(va3);
476 }
477
478 #define IDCT8_1D_ALTIVEC(s0, s1, s2, s3, s4, s5, s6, s7,  d0, d1, d2, d3, d4, d5, d6, d7) {\
479     /*        a0  = SRC(0) + SRC(4); */ \
480     vec_s16 a0v = vec_add(s0, s4);    \
481     /*        a2  = SRC(0) - SRC(4); */ \
482     vec_s16 a2v = vec_sub(s0, s4);    \
483     /*        a4  =           (SRC(2)>>1) - SRC(6); */ \
484     vec_s16 a4v = vec_sub(vec_sra(s2, onev), s6);    \
485     /*        a6  =           (SRC(6)>>1) + SRC(2); */ \
486     vec_s16 a6v = vec_add(vec_sra(s6, onev), s2);    \
487     /*        b0  =         a0 + a6; */ \
488     vec_s16 b0v = vec_add(a0v, a6v);  \
489     /*        b2  =         a2 + a4; */ \
490     vec_s16 b2v = vec_add(a2v, a4v);  \
491     /*        b4  =         a2 - a4; */ \
492     vec_s16 b4v = vec_sub(a2v, a4v);  \
493     /*        b6  =         a0 - a6; */ \
494     vec_s16 b6v = vec_sub(a0v, a6v);  \
495     /* a1 =  SRC(5) - SRC(3) - SRC(7) - (SRC(7)>>1); */ \
496     /*        a1 =             (SRC(5)-SRC(3)) -  (SRC(7)  +  (SRC(7)>>1)); */ \
497     vec_s16 a1v = vec_sub( vec_sub(s5, s3), vec_add(s7, vec_sra(s7, onev)) ); \
498     /* a3 =  SRC(7) + SRC(1) - SRC(3) - (SRC(3)>>1); */ \
499     /*        a3 =             (SRC(7)+SRC(1)) -  (SRC(3)  +  (SRC(3)>>1)); */ \
500     vec_s16 a3v = vec_sub( vec_add(s7, s1), vec_add(s3, vec_sra(s3, onev)) );\
501     /* a5 =  SRC(7) - SRC(1) + SRC(5) + (SRC(5)>>1); */ \
502     /*        a5 =             (SRC(7)-SRC(1)) +   SRC(5) +   (SRC(5)>>1); */ \
503     vec_s16 a5v = vec_add( vec_sub(s7, s1), vec_add(s5, vec_sra(s5, onev)) );\
504     /*        a7 =                SRC(5)+SRC(3) +  SRC(1) +   (SRC(1)>>1); */ \
505     vec_s16 a7v = vec_add( vec_add(s5, s3), vec_add(s1, vec_sra(s1, onev)) );\
506     /*        b1 =                  (a7>>2)  +  a1; */ \
507     vec_s16 b1v = vec_add( vec_sra(a7v, twov), a1v); \
508     /*        b3 =          a3 +        (a5>>2); */ \
509     vec_s16 b3v = vec_add(a3v, vec_sra(a5v, twov)); \
510     /*        b5 =                  (a3>>2)  -   a5; */ \
511     vec_s16 b5v = vec_sub( vec_sra(a3v, twov), a5v); \
512     /*        b7 =           a7 -        (a1>>2); */ \
513     vec_s16 b7v = vec_sub( a7v, vec_sra(a1v, twov)); \
514     /* DST(0,    b0 + b7); */ \
515     d0 = vec_add(b0v, b7v); \
516     /* DST(1,    b2 + b5); */ \
517     d1 = vec_add(b2v, b5v); \
518     /* DST(2,    b4 + b3); */ \
519     d2 = vec_add(b4v, b3v); \
520     /* DST(3,    b6 + b1); */ \
521     d3 = vec_add(b6v, b1v); \
522     /* DST(4,    b6 - b1); */ \
523     d4 = vec_sub(b6v, b1v); \
524     /* DST(5,    b4 - b3); */ \
525     d5 = vec_sub(b4v, b3v); \
526     /* DST(6,    b2 - b5); */ \
527     d6 = vec_sub(b2v, b5v); \
528     /* DST(7,    b0 - b7); */ \
529     d7 = vec_sub(b0v, b7v); \
530 }
531
532 #define ALTIVEC_STORE_SUM_CLIP(dest, idctv, perm_ldv, perm_stv, sel) { \
533     /* unaligned load */                                       \
534     vec_u8 hv = vec_ld( 0, dest );                           \
535     vec_u8 lv = vec_ld( 7, dest );                           \
536     vec_u8 dstv   = vec_perm( hv, lv, (vec_u8)perm_ldv );  \
537     vec_s16 idct_sh6 = vec_sra(idctv, sixv);                 \
538     vec_u16 dst16 = (vec_u16)vec_mergeh(zero_u8v, dstv);   \
539     vec_s16 idstsum = vec_adds(idct_sh6, (vec_s16)dst16);  \
540     vec_u8 idstsum8 = vec_packsu(zero_s16v, idstsum);        \
541     vec_u8 edgehv;                                           \
542     /* unaligned store */                                      \
543     vec_u8 bodyv  = vec_perm( idstsum8, idstsum8, perm_stv );\
544     vec_u8 edgelv = vec_perm( sel, zero_u8v, perm_stv );     \
545     lv    = vec_sel( lv, bodyv, edgelv );                      \
546     vec_st( lv, 7, dest );                                     \
547     hv    = vec_ld( 0, dest );                                 \
548     edgehv = vec_perm( zero_u8v, sel, perm_stv );              \
549     hv    = vec_sel( hv, bodyv, edgehv );                      \
550     vec_st( hv, 0, dest );                                     \
551  }
552
553 void ff_h264_idct8_add_altivec( uint8_t *dst, DCTELEM *dct, int stride ) {
554     vec_s16 s0, s1, s2, s3, s4, s5, s6, s7;
555     vec_s16 d0, d1, d2, d3, d4, d5, d6, d7;
556     vec_s16 idct0, idct1, idct2, idct3, idct4, idct5, idct6, idct7;
557
558     vec_u8 perm_ldv = vec_lvsl(0, dst);
559     vec_u8 perm_stv = vec_lvsr(8, dst);
560
561     const vec_u16 onev = vec_splat_u16(1);
562     const vec_u16 twov = vec_splat_u16(2);
563     const vec_u16 sixv = vec_splat_u16(6);
564
565     const vec_u8 sel = (vec_u8) {0,0,0,0,0,0,0,0,-1,-1,-1,-1,-1,-1,-1,-1};
566     LOAD_ZERO;
567
568     dct[0] += 32; // rounding for the >>6 at the end
569
570     s0 = vec_ld(0x00, (int16_t*)dct);
571     s1 = vec_ld(0x10, (int16_t*)dct);
572     s2 = vec_ld(0x20, (int16_t*)dct);
573     s3 = vec_ld(0x30, (int16_t*)dct);
574     s4 = vec_ld(0x40, (int16_t*)dct);
575     s5 = vec_ld(0x50, (int16_t*)dct);
576     s6 = vec_ld(0x60, (int16_t*)dct);
577     s7 = vec_ld(0x70, (int16_t*)dct);
578
579     IDCT8_1D_ALTIVEC(s0, s1, s2, s3, s4, s5, s6, s7,
580                      d0, d1, d2, d3, d4, d5, d6, d7);
581
582     TRANSPOSE8( d0,  d1,  d2,  d3,  d4,  d5,  d6, d7 );
583
584     IDCT8_1D_ALTIVEC(d0,  d1,  d2,  d3,  d4,  d5,  d6, d7,
585                      idct0, idct1, idct2, idct3, idct4, idct5, idct6, idct7);
586
587     ALTIVEC_STORE_SUM_CLIP(&dst[0*stride], idct0, perm_ldv, perm_stv, sel);
588     ALTIVEC_STORE_SUM_CLIP(&dst[1*stride], idct1, perm_ldv, perm_stv, sel);
589     ALTIVEC_STORE_SUM_CLIP(&dst[2*stride], idct2, perm_ldv, perm_stv, sel);
590     ALTIVEC_STORE_SUM_CLIP(&dst[3*stride], idct3, perm_ldv, perm_stv, sel);
591     ALTIVEC_STORE_SUM_CLIP(&dst[4*stride], idct4, perm_ldv, perm_stv, sel);
592     ALTIVEC_STORE_SUM_CLIP(&dst[5*stride], idct5, perm_ldv, perm_stv, sel);
593     ALTIVEC_STORE_SUM_CLIP(&dst[6*stride], idct6, perm_ldv, perm_stv, sel);
594     ALTIVEC_STORE_SUM_CLIP(&dst[7*stride], idct7, perm_ldv, perm_stv, sel);
595 }
596
597 static av_always_inline void h264_idct_dc_add_internal(uint8_t *dst, DCTELEM *block, int stride, int size)
598 {
599     vec_s16 dc16;
600     vec_u8 dcplus, dcminus, v0, v1, v2, v3, aligner;
601     LOAD_ZERO;
602     DECLARE_ALIGNED_16(int, dc);
603     int i;
604
605     dc = (block[0] + 32) >> 6;
606     dc16 = vec_splat((vec_s16) vec_lde(0, &dc), 1);
607
608     if (size == 4)
609         dc16 = vec_sld(dc16, zero_s16v, 8);
610     dcplus = vec_packsu(dc16, zero_s16v);
611     dcminus = vec_packsu(vec_sub(zero_s16v, dc16), zero_s16v);
612
613     aligner = vec_lvsr(0, dst);
614     dcplus = vec_perm(dcplus, dcplus, aligner);
615     dcminus = vec_perm(dcminus, dcminus, aligner);
616
617     for (i = 0; i < size; i += 4) {
618         v0 = vec_ld(0, dst+0*stride);
619         v1 = vec_ld(0, dst+1*stride);
620         v2 = vec_ld(0, dst+2*stride);
621         v3 = vec_ld(0, dst+3*stride);
622
623         v0 = vec_adds(v0, dcplus);
624         v1 = vec_adds(v1, dcplus);
625         v2 = vec_adds(v2, dcplus);
626         v3 = vec_adds(v3, dcplus);
627
628         v0 = vec_subs(v0, dcminus);
629         v1 = vec_subs(v1, dcminus);
630         v2 = vec_subs(v2, dcminus);
631         v3 = vec_subs(v3, dcminus);
632
633         vec_st(v0, 0, dst+0*stride);
634         vec_st(v1, 0, dst+1*stride);
635         vec_st(v2, 0, dst+2*stride);
636         vec_st(v3, 0, dst+3*stride);
637
638         dst += 4*stride;
639     }
640 }
641
642 static void h264_idct_dc_add_altivec(uint8_t *dst, DCTELEM *block, int stride)
643 {
644     h264_idct_dc_add_internal(dst, block, stride, 4);
645 }
646
647 static void ff_h264_idct8_dc_add_altivec(uint8_t *dst, DCTELEM *block, int stride)
648 {
649     h264_idct_dc_add_internal(dst, block, stride, 8);
650 }
651
652 static void ff_h264_idct_add16_altivec(uint8_t *dst, const int *block_offset, DCTELEM *block, int stride, const uint8_t nnzc[6*8]){
653     int i;
654     for(i=0; i<16; i++){
655         int nnz = nnzc[ scan8[i] ];
656         if(nnz){
657             if(nnz==1 && block[i*16]) h264_idct_dc_add_altivec(dst + block_offset[i], block + i*16, stride);
658             else                      ff_h264_idct_add_altivec(dst + block_offset[i], block + i*16, stride);
659         }
660     }
661 }
662
663 static void ff_h264_idct_add16intra_altivec(uint8_t *dst, const int *block_offset, DCTELEM *block, int stride, const uint8_t nnzc[6*8]){
664     int i;
665     for(i=0; i<16; i++){
666         if(nnzc[ scan8[i] ]) ff_h264_idct_add_altivec(dst + block_offset[i], block + i*16, stride);
667         else if(block[i*16]) h264_idct_dc_add_altivec(dst + block_offset[i], block + i*16, stride);
668     }
669 }
670
671 static void ff_h264_idct8_add4_altivec(uint8_t *dst, const int *block_offset, DCTELEM *block, int stride, const uint8_t nnzc[6*8]){
672     int i;
673     for(i=0; i<16; i+=4){
674         int nnz = nnzc[ scan8[i] ];
675         if(nnz){
676             if(nnz==1 && block[i*16]) ff_h264_idct8_dc_add_altivec(dst + block_offset[i], block + i*16, stride);
677             else                      ff_h264_idct8_add_altivec   (dst + block_offset[i], block + i*16, stride);
678         }
679     }
680 }
681
682 static void ff_h264_idct_add8_altivec(uint8_t **dest, const int *block_offset, DCTELEM *block, int stride, const uint8_t nnzc[6*8]){
683     int i;
684     for(i=16; i<16+8; i++){
685         if(nnzc[ scan8[i] ])
686             ff_h264_idct_add_altivec(dest[(i&4)>>2] + block_offset[i], block + i*16, stride);
687         else if(block[i*16])
688             h264_idct_dc_add_altivec(dest[(i&4)>>2] + block_offset[i], block + i*16, stride);
689     }
690 }
691
692 #define transpose4x16(r0, r1, r2, r3) {      \
693     register vec_u8 r4;                    \
694     register vec_u8 r5;                    \
695     register vec_u8 r6;                    \
696     register vec_u8 r7;                    \
697                                              \
698     r4 = vec_mergeh(r0, r2);  /*0, 2 set 0*/ \
699     r5 = vec_mergel(r0, r2);  /*0, 2 set 1*/ \
700     r6 = vec_mergeh(r1, r3);  /*1, 3 set 0*/ \
701     r7 = vec_mergel(r1, r3);  /*1, 3 set 1*/ \
702                                              \
703     r0 = vec_mergeh(r4, r6);  /*all set 0*/  \
704     r1 = vec_mergel(r4, r6);  /*all set 1*/  \
705     r2 = vec_mergeh(r5, r7);  /*all set 2*/  \
706     r3 = vec_mergel(r5, r7);  /*all set 3*/  \
707 }
708
709 static inline void write16x4(uint8_t *dst, int dst_stride,
710                              register vec_u8 r0, register vec_u8 r1,
711                              register vec_u8 r2, register vec_u8 r3) {
712     DECLARE_ALIGNED_16(unsigned char, result[64]);
713     uint32_t *src_int = (uint32_t *)result, *dst_int = (uint32_t *)dst;
714     int int_dst_stride = dst_stride/4;
715
716     vec_st(r0, 0, result);
717     vec_st(r1, 16, result);
718     vec_st(r2, 32, result);
719     vec_st(r3, 48, result);
720     /* FIXME: there has to be a better way!!!! */
721     *dst_int = *src_int;
722     *(dst_int+   int_dst_stride) = *(src_int + 1);
723     *(dst_int+ 2*int_dst_stride) = *(src_int + 2);
724     *(dst_int+ 3*int_dst_stride) = *(src_int + 3);
725     *(dst_int+ 4*int_dst_stride) = *(src_int + 4);
726     *(dst_int+ 5*int_dst_stride) = *(src_int + 5);
727     *(dst_int+ 6*int_dst_stride) = *(src_int + 6);
728     *(dst_int+ 7*int_dst_stride) = *(src_int + 7);
729     *(dst_int+ 8*int_dst_stride) = *(src_int + 8);
730     *(dst_int+ 9*int_dst_stride) = *(src_int + 9);
731     *(dst_int+10*int_dst_stride) = *(src_int + 10);
732     *(dst_int+11*int_dst_stride) = *(src_int + 11);
733     *(dst_int+12*int_dst_stride) = *(src_int + 12);
734     *(dst_int+13*int_dst_stride) = *(src_int + 13);
735     *(dst_int+14*int_dst_stride) = *(src_int + 14);
736     *(dst_int+15*int_dst_stride) = *(src_int + 15);
737 }
738
739 /** \brief performs a 6x16 transpose of data in src, and stores it to dst
740     \todo FIXME: see if we can't spare some vec_lvsl() by them factorizing
741     out of unaligned_load() */
742 #define readAndTranspose16x6(src, src_stride, r8, r9, r10, r11, r12, r13) {\
743     register vec_u8 r0  = unaligned_load(0,             src);            \
744     register vec_u8 r1  = unaligned_load(   src_stride, src);            \
745     register vec_u8 r2  = unaligned_load(2* src_stride, src);            \
746     register vec_u8 r3  = unaligned_load(3* src_stride, src);            \
747     register vec_u8 r4  = unaligned_load(4* src_stride, src);            \
748     register vec_u8 r5  = unaligned_load(5* src_stride, src);            \
749     register vec_u8 r6  = unaligned_load(6* src_stride, src);            \
750     register vec_u8 r7  = unaligned_load(7* src_stride, src);            \
751     register vec_u8 r14 = unaligned_load(14*src_stride, src);            \
752     register vec_u8 r15 = unaligned_load(15*src_stride, src);            \
753                                                                            \
754     r8  = unaligned_load( 8*src_stride, src);                              \
755     r9  = unaligned_load( 9*src_stride, src);                              \
756     r10 = unaligned_load(10*src_stride, src);                              \
757     r11 = unaligned_load(11*src_stride, src);                              \
758     r12 = unaligned_load(12*src_stride, src);                              \
759     r13 = unaligned_load(13*src_stride, src);                              \
760                                                                            \
761     /*Merge first pairs*/                                                  \
762     r0 = vec_mergeh(r0, r8);    /*0, 8*/                                   \
763     r1 = vec_mergeh(r1, r9);    /*1, 9*/                                   \
764     r2 = vec_mergeh(r2, r10);   /*2,10*/                                   \
765     r3 = vec_mergeh(r3, r11);   /*3,11*/                                   \
766     r4 = vec_mergeh(r4, r12);   /*4,12*/                                   \
767     r5 = vec_mergeh(r5, r13);   /*5,13*/                                   \
768     r6 = vec_mergeh(r6, r14);   /*6,14*/                                   \
769     r7 = vec_mergeh(r7, r15);   /*7,15*/                                   \
770                                                                            \
771     /*Merge second pairs*/                                                 \
772     r8  = vec_mergeh(r0, r4);   /*0,4, 8,12 set 0*/                        \
773     r9  = vec_mergel(r0, r4);   /*0,4, 8,12 set 1*/                        \
774     r10 = vec_mergeh(r1, r5);   /*1,5, 9,13 set 0*/                        \
775     r11 = vec_mergel(r1, r5);   /*1,5, 9,13 set 1*/                        \
776     r12 = vec_mergeh(r2, r6);   /*2,6,10,14 set 0*/                        \
777     r13 = vec_mergel(r2, r6);   /*2,6,10,14 set 1*/                        \
778     r14 = vec_mergeh(r3, r7);   /*3,7,11,15 set 0*/                        \
779     r15 = vec_mergel(r3, r7);   /*3,7,11,15 set 1*/                        \
780                                                                            \
781     /*Third merge*/                                                        \
782     r0 = vec_mergeh(r8,  r12);  /*0,2,4,6,8,10,12,14 set 0*/               \
783     r1 = vec_mergel(r8,  r12);  /*0,2,4,6,8,10,12,14 set 1*/               \
784     r2 = vec_mergeh(r9,  r13);  /*0,2,4,6,8,10,12,14 set 2*/               \
785     r4 = vec_mergeh(r10, r14);  /*1,3,5,7,9,11,13,15 set 0*/               \
786     r5 = vec_mergel(r10, r14);  /*1,3,5,7,9,11,13,15 set 1*/               \
787     r6 = vec_mergeh(r11, r15);  /*1,3,5,7,9,11,13,15 set 2*/               \
788     /* Don't need to compute 3 and 7*/                                     \
789                                                                            \
790     /*Final merge*/                                                        \
791     r8  = vec_mergeh(r0, r4);   /*all set 0*/                              \
792     r9  = vec_mergel(r0, r4);   /*all set 1*/                              \
793     r10 = vec_mergeh(r1, r5);   /*all set 2*/                              \
794     r11 = vec_mergel(r1, r5);   /*all set 3*/                              \
795     r12 = vec_mergeh(r2, r6);   /*all set 4*/                              \
796     r13 = vec_mergel(r2, r6);   /*all set 5*/                              \
797     /* Don't need to compute 14 and 15*/                                   \
798                                                                            \
799 }
800
801 // out: o = |x-y| < a
802 static inline vec_u8 diff_lt_altivec ( register vec_u8 x,
803                                          register vec_u8 y,
804                                          register vec_u8 a) {
805
806     register vec_u8 diff = vec_subs(x, y);
807     register vec_u8 diffneg = vec_subs(y, x);
808     register vec_u8 o = vec_or(diff, diffneg); /* |x-y| */
809     o = (vec_u8)vec_cmplt(o, a);
810     return o;
811 }
812
813 static inline vec_u8 h264_deblock_mask ( register vec_u8 p0,
814                                            register vec_u8 p1,
815                                            register vec_u8 q0,
816                                            register vec_u8 q1,
817                                            register vec_u8 alpha,
818                                            register vec_u8 beta) {
819
820     register vec_u8 mask;
821     register vec_u8 tempmask;
822
823     mask = diff_lt_altivec(p0, q0, alpha);
824     tempmask = diff_lt_altivec(p1, p0, beta);
825     mask = vec_and(mask, tempmask);
826     tempmask = diff_lt_altivec(q1, q0, beta);
827     mask = vec_and(mask, tempmask);
828
829     return mask;
830 }
831
832 // out: newp1 = clip((p2 + ((p0 + q0 + 1) >> 1)) >> 1, p1-tc0, p1+tc0)
833 static inline vec_u8 h264_deblock_q1(register vec_u8 p0,
834                                        register vec_u8 p1,
835                                        register vec_u8 p2,
836                                        register vec_u8 q0,
837                                        register vec_u8 tc0) {
838
839     register vec_u8 average = vec_avg(p0, q0);
840     register vec_u8 temp;
841     register vec_u8 uncliped;
842     register vec_u8 ones;
843     register vec_u8 max;
844     register vec_u8 min;
845     register vec_u8 newp1;
846
847     temp = vec_xor(average, p2);
848     average = vec_avg(average, p2);     /*avg(p2, avg(p0, q0)) */
849     ones = vec_splat_u8(1);
850     temp = vec_and(temp, ones);         /*(p2^avg(p0, q0)) & 1 */
851     uncliped = vec_subs(average, temp); /*(p2+((p0+q0+1)>>1))>>1 */
852     max = vec_adds(p1, tc0);
853     min = vec_subs(p1, tc0);
854     newp1 = vec_max(min, uncliped);
855     newp1 = vec_min(max, newp1);
856     return newp1;
857 }
858
859 #define h264_deblock_p0_q0(p0, p1, q0, q1, tc0masked) {                                           \
860                                                                                                   \
861     const vec_u8 A0v = vec_sl(vec_splat_u8(10), vec_splat_u8(4));                               \
862                                                                                                   \
863     register vec_u8 pq0bit = vec_xor(p0,q0);                                                    \
864     register vec_u8 q1minus;                                                                    \
865     register vec_u8 p0minus;                                                                    \
866     register vec_u8 stage1;                                                                     \
867     register vec_u8 stage2;                                                                     \
868     register vec_u8 vec160;                                                                     \
869     register vec_u8 delta;                                                                      \
870     register vec_u8 deltaneg;                                                                   \
871                                                                                                   \
872     q1minus = vec_nor(q1, q1);                 /* 255 - q1 */                                     \
873     stage1 = vec_avg(p1, q1minus);             /* (p1 - q1 + 256)>>1 */                           \
874     stage2 = vec_sr(stage1, vec_splat_u8(1));  /* (p1 - q1 + 256)>>2 = 64 + (p1 - q1) >> 2 */     \
875     p0minus = vec_nor(p0, p0);                 /* 255 - p0 */                                     \
876     stage1 = vec_avg(q0, p0minus);             /* (q0 - p0 + 256)>>1 */                           \
877     pq0bit = vec_and(pq0bit, vec_splat_u8(1));                                                    \
878     stage2 = vec_avg(stage2, pq0bit);          /* 32 + ((q0 - p0)&1 + (p1 - q1) >> 2 + 1) >> 1 */ \
879     stage2 = vec_adds(stage2, stage1);         /* 160 + ((p0 - q0) + (p1 - q1) >> 2 + 1) >> 1 */  \
880     vec160 = vec_ld(0, &A0v);                                                                     \
881     deltaneg = vec_subs(vec160, stage2);       /* -d */                                           \
882     delta = vec_subs(stage2, vec160);          /* d */                                            \
883     deltaneg = vec_min(tc0masked, deltaneg);                                                      \
884     delta = vec_min(tc0masked, delta);                                                            \
885     p0 = vec_subs(p0, deltaneg);                                                                  \
886     q0 = vec_subs(q0, delta);                                                                     \
887     p0 = vec_adds(p0, delta);                                                                     \
888     q0 = vec_adds(q0, deltaneg);                                                                  \
889 }
890
891 #define h264_loop_filter_luma_altivec(p2, p1, p0, q0, q1, q2, alpha, beta, tc0) {            \
892     DECLARE_ALIGNED_16(unsigned char, temp[16]);                                             \
893     register vec_u8 alphavec;                                                              \
894     register vec_u8 betavec;                                                               \
895     register vec_u8 mask;                                                                  \
896     register vec_u8 p1mask;                                                                \
897     register vec_u8 q1mask;                                                                \
898     register vector signed   char tc0vec;                                                    \
899     register vec_u8 finaltc0;                                                              \
900     register vec_u8 tc0masked;                                                             \
901     register vec_u8 newp1;                                                                 \
902     register vec_u8 newq1;                                                                 \
903                                                                                              \
904     temp[0] = alpha;                                                                         \
905     temp[1] = beta;                                                                          \
906     alphavec = vec_ld(0, temp);                                                              \
907     betavec = vec_splat(alphavec, 0x1);                                                      \
908     alphavec = vec_splat(alphavec, 0x0);                                                     \
909     mask = h264_deblock_mask(p0, p1, q0, q1, alphavec, betavec); /*if in block */            \
910                                                                                              \
911     *((int *)temp) = *((int *)tc0);                                                          \
912     tc0vec = vec_ld(0, (signed char*)temp);                                                  \
913     tc0vec = vec_mergeh(tc0vec, tc0vec);                                                     \
914     tc0vec = vec_mergeh(tc0vec, tc0vec);                                                     \
915     mask = vec_and(mask, vec_cmpgt(tc0vec, vec_splat_s8(-1)));  /* if tc0[i] >= 0 */         \
916     finaltc0 = vec_and((vec_u8)tc0vec, mask);     /* tc = tc0 */                           \
917                                                                                              \
918     p1mask = diff_lt_altivec(p2, p0, betavec);                                               \
919     p1mask = vec_and(p1mask, mask);                             /* if ( |p2 - p0| < beta) */ \
920     tc0masked = vec_and(p1mask, (vec_u8)tc0vec);                                           \
921     finaltc0 = vec_sub(finaltc0, p1mask);                       /* tc++ */                   \
922     newp1 = h264_deblock_q1(p0, p1, p2, q0, tc0masked);                                      \
923     /*end if*/                                                                               \
924                                                                                              \
925     q1mask = diff_lt_altivec(q2, q0, betavec);                                               \
926     q1mask = vec_and(q1mask, mask);                             /* if ( |q2 - q0| < beta ) */\
927     tc0masked = vec_and(q1mask, (vec_u8)tc0vec);                                           \
928     finaltc0 = vec_sub(finaltc0, q1mask);                       /* tc++ */                   \
929     newq1 = h264_deblock_q1(p0, q1, q2, q0, tc0masked);                                      \
930     /*end if*/                                                                               \
931                                                                                              \
932     h264_deblock_p0_q0(p0, p1, q0, q1, finaltc0);                                            \
933     p1 = newp1;                                                                              \
934     q1 = newq1;                                                                              \
935 }
936
937 static void h264_v_loop_filter_luma_altivec(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0) {
938
939     if ((tc0[0] & tc0[1] & tc0[2] & tc0[3]) >= 0) {
940         register vec_u8 p2 = vec_ld(-3*stride, pix);
941         register vec_u8 p1 = vec_ld(-2*stride, pix);
942         register vec_u8 p0 = vec_ld(-1*stride, pix);
943         register vec_u8 q0 = vec_ld(0, pix);
944         register vec_u8 q1 = vec_ld(stride, pix);
945         register vec_u8 q2 = vec_ld(2*stride, pix);
946         h264_loop_filter_luma_altivec(p2, p1, p0, q0, q1, q2, alpha, beta, tc0);
947         vec_st(p1, -2*stride, pix);
948         vec_st(p0, -1*stride, pix);
949         vec_st(q0, 0, pix);
950         vec_st(q1, stride, pix);
951     }
952 }
953
954 static void h264_h_loop_filter_luma_altivec(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0) {
955
956     register vec_u8 line0, line1, line2, line3, line4, line5;
957     if ((tc0[0] & tc0[1] & tc0[2] & tc0[3]) < 0)
958         return;
959     readAndTranspose16x6(pix-3, stride, line0, line1, line2, line3, line4, line5);
960     h264_loop_filter_luma_altivec(line0, line1, line2, line3, line4, line5, alpha, beta, tc0);
961     transpose4x16(line1, line2, line3, line4);
962     write16x4(pix-2, stride, line1, line2, line3, line4);
963 }
964
965 static av_always_inline
966 void weight_h264_WxH_altivec(uint8_t *block, int stride, int log2_denom, int weight, int offset, int w, int h)
967 {
968     int y, aligned;
969     vec_u8 vblock;
970     vec_s16 vtemp, vweight, voffset, v0, v1;
971     vec_u16 vlog2_denom;
972     DECLARE_ALIGNED_16(int32_t, temp[4]);
973     LOAD_ZERO;
974
975     offset <<= log2_denom;
976     if(log2_denom) offset += 1<<(log2_denom-1);
977     temp[0] = log2_denom;
978     temp[1] = weight;
979     temp[2] = offset;
980
981     vtemp = (vec_s16)vec_ld(0, temp);
982     vlog2_denom = (vec_u16)vec_splat(vtemp, 1);
983     vweight = vec_splat(vtemp, 3);
984     voffset = vec_splat(vtemp, 5);
985     aligned = !((unsigned long)block & 0xf);
986
987     for (y=0; y<h; y++) {
988         vblock = vec_ld(0, block);
989
990         v0 = (vec_s16)vec_mergeh(zero_u8v, vblock);
991         v1 = (vec_s16)vec_mergel(zero_u8v, vblock);
992
993         if (w == 16 || aligned) {
994             v0 = vec_mladd(v0, vweight, zero_s16v);
995             v0 = vec_adds(v0, voffset);
996             v0 = vec_sra(v0, vlog2_denom);
997         }
998         if (w == 16 || !aligned) {
999             v1 = vec_mladd(v1, vweight, zero_s16v);
1000             v1 = vec_adds(v1, voffset);
1001             v1 = vec_sra(v1, vlog2_denom);
1002         }
1003         vblock = vec_packsu(v0, v1);
1004         vec_st(vblock, 0, block);
1005
1006         block += stride;
1007     }
1008 }
1009
1010 static av_always_inline
1011 void biweight_h264_WxH_altivec(uint8_t *dst, uint8_t *src, int stride, int log2_denom,
1012                                int weightd, int weights, int offset, int w, int h)
1013 {
1014     int y, dst_aligned, src_aligned;
1015     vec_u8 vsrc, vdst;
1016     vec_s16 vtemp, vweights, vweightd, voffset, v0, v1, v2, v3;
1017     vec_u16 vlog2_denom;
1018     DECLARE_ALIGNED_16(int32_t, temp[4]);
1019     LOAD_ZERO;
1020
1021     offset = ((offset + 1) | 1) << log2_denom;
1022     temp[0] = log2_denom+1;
1023     temp[1] = weights;
1024     temp[2] = weightd;
1025     temp[3] = offset;
1026
1027     vtemp = (vec_s16)vec_ld(0, temp);
1028     vlog2_denom = (vec_u16)vec_splat(vtemp, 1);
1029     vweights = vec_splat(vtemp, 3);
1030     vweightd = vec_splat(vtemp, 5);
1031     voffset = vec_splat(vtemp, 7);
1032     dst_aligned = !((unsigned long)dst & 0xf);
1033     src_aligned = !((unsigned long)src & 0xf);
1034
1035     for (y=0; y<h; y++) {
1036         vdst = vec_ld(0, dst);
1037         vsrc = vec_ld(0, src);
1038
1039         v0 = (vec_s16)vec_mergeh(zero_u8v, vdst);
1040         v1 = (vec_s16)vec_mergel(zero_u8v, vdst);
1041         v2 = (vec_s16)vec_mergeh(zero_u8v, vsrc);
1042         v3 = (vec_s16)vec_mergel(zero_u8v, vsrc);
1043
1044         if (w == 8) {
1045             if (src_aligned)
1046                 v3 = v2;
1047             else
1048                 v2 = v3;
1049         }
1050
1051         if (w == 16 || dst_aligned) {
1052             v0 = vec_mladd(v0, vweightd, zero_s16v);
1053             v2 = vec_mladd(v2, vweights, zero_s16v);
1054
1055             v0 = vec_adds(v0, voffset);
1056             v0 = vec_adds(v0, v2);
1057             v0 = vec_sra(v0, vlog2_denom);
1058         }
1059         if (w == 16 || !dst_aligned) {
1060             v1 = vec_mladd(v1, vweightd, zero_s16v);
1061             v3 = vec_mladd(v3, vweights, zero_s16v);
1062
1063             v1 = vec_adds(v1, voffset);
1064             v1 = vec_adds(v1, v3);
1065             v1 = vec_sra(v1, vlog2_denom);
1066         }
1067         vdst = vec_packsu(v0, v1);
1068         vec_st(vdst, 0, dst);
1069
1070         dst += stride;
1071         src += stride;
1072     }
1073 }
1074
1075 #define H264_WEIGHT(W,H) \
1076 static void ff_weight_h264_pixels ## W ## x ## H ## _altivec(uint8_t *block, int stride, int log2_denom, int weight, int offset){ \
1077     weight_h264_WxH_altivec(block, stride, log2_denom, weight, offset, W, H); \
1078 }\
1079 static void ff_biweight_h264_pixels ## W ## x ## H ## _altivec(uint8_t *dst, uint8_t *src, int stride, int log2_denom, int weightd, int weights, int offset){ \
1080     biweight_h264_WxH_altivec(dst, src, stride, log2_denom, weightd, weights, offset, W, H); \
1081 }
1082
1083 H264_WEIGHT(16,16)
1084 H264_WEIGHT(16, 8)
1085 H264_WEIGHT( 8,16)
1086 H264_WEIGHT( 8, 8)
1087 H264_WEIGHT( 8, 4)
1088
1089 void dsputil_h264_init_ppc(DSPContext* c, AVCodecContext *avctx) {
1090
1091     if (has_altivec()) {
1092         c->put_h264_chroma_pixels_tab[0] = put_h264_chroma_mc8_altivec;
1093         c->put_no_rnd_h264_chroma_pixels_tab[0] = put_no_rnd_h264_chroma_mc8_altivec;
1094         c->avg_h264_chroma_pixels_tab[0] = avg_h264_chroma_mc8_altivec;
1095         c->h264_idct_add = ff_h264_idct_add_altivec;
1096         c->h264_idct_add8 = ff_h264_idct_add8_altivec;
1097         c->h264_idct_add16 = ff_h264_idct_add16_altivec;
1098         c->h264_idct_add16intra = ff_h264_idct_add16intra_altivec;
1099         c->h264_idct_dc_add= h264_idct_dc_add_altivec;
1100         c->h264_idct8_dc_add = ff_h264_idct8_dc_add_altivec;
1101         c->h264_idct8_add = ff_h264_idct8_add_altivec;
1102         c->h264_idct8_add4 = ff_h264_idct8_add4_altivec;
1103         c->h264_v_loop_filter_luma= h264_v_loop_filter_luma_altivec;
1104         c->h264_h_loop_filter_luma= h264_h_loop_filter_luma_altivec;
1105
1106 #define dspfunc(PFX, IDX, NUM) \
1107         c->PFX ## _pixels_tab[IDX][ 0] = PFX ## NUM ## _mc00_altivec; \
1108         c->PFX ## _pixels_tab[IDX][ 1] = PFX ## NUM ## _mc10_altivec; \
1109         c->PFX ## _pixels_tab[IDX][ 2] = PFX ## NUM ## _mc20_altivec; \
1110         c->PFX ## _pixels_tab[IDX][ 3] = PFX ## NUM ## _mc30_altivec; \
1111         c->PFX ## _pixels_tab[IDX][ 4] = PFX ## NUM ## _mc01_altivec; \
1112         c->PFX ## _pixels_tab[IDX][ 5] = PFX ## NUM ## _mc11_altivec; \
1113         c->PFX ## _pixels_tab[IDX][ 6] = PFX ## NUM ## _mc21_altivec; \
1114         c->PFX ## _pixels_tab[IDX][ 7] = PFX ## NUM ## _mc31_altivec; \
1115         c->PFX ## _pixels_tab[IDX][ 8] = PFX ## NUM ## _mc02_altivec; \
1116         c->PFX ## _pixels_tab[IDX][ 9] = PFX ## NUM ## _mc12_altivec; \
1117         c->PFX ## _pixels_tab[IDX][10] = PFX ## NUM ## _mc22_altivec; \
1118         c->PFX ## _pixels_tab[IDX][11] = PFX ## NUM ## _mc32_altivec; \
1119         c->PFX ## _pixels_tab[IDX][12] = PFX ## NUM ## _mc03_altivec; \
1120         c->PFX ## _pixels_tab[IDX][13] = PFX ## NUM ## _mc13_altivec; \
1121         c->PFX ## _pixels_tab[IDX][14] = PFX ## NUM ## _mc23_altivec; \
1122         c->PFX ## _pixels_tab[IDX][15] = PFX ## NUM ## _mc33_altivec
1123
1124         dspfunc(put_h264_qpel, 0, 16);
1125         dspfunc(avg_h264_qpel, 0, 16);
1126 #undef dspfunc
1127
1128         c->weight_h264_pixels_tab[0] = ff_weight_h264_pixels16x16_altivec;
1129         c->weight_h264_pixels_tab[1] = ff_weight_h264_pixels16x8_altivec;
1130         c->weight_h264_pixels_tab[2] = ff_weight_h264_pixels8x16_altivec;
1131         c->weight_h264_pixels_tab[3] = ff_weight_h264_pixels8x8_altivec;
1132         c->weight_h264_pixels_tab[4] = ff_weight_h264_pixels8x4_altivec;
1133         c->biweight_h264_pixels_tab[0] = ff_biweight_h264_pixels16x16_altivec;
1134         c->biweight_h264_pixels_tab[1] = ff_biweight_h264_pixels16x8_altivec;
1135         c->biweight_h264_pixels_tab[2] = ff_biweight_h264_pixels8x16_altivec;
1136         c->biweight_h264_pixels_tab[3] = ff_biweight_h264_pixels8x8_altivec;
1137         c->biweight_h264_pixels_tab[4] = ff_biweight_h264_pixels8x4_altivec;
1138     }
1139 }