repositories / deb_ffmpeg.git / blob
? search:


da118a49b69358a9e598a9ea914c2a51189192fa
[deb_ffmpeg.git] / h264dsp.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 "config.h"
22 #include "libavutil/attributes.h"
23 #include "libavutil/cpu.h"
24 #include "libavutil/intreadwrite.h"
25 #include "libavutil/ppc/cpu.h"
26 #include "libavutil/ppc/types_altivec.h"
27 #include "libavutil/ppc/util_altivec.h"
28 #include "libavcodec/h264data.h"
29 #include "libavcodec/h264dsp.h"
30
31 #if HAVE_ALTIVEC
32
33 /****************************************************************************
34 * IDCT transform:
35 ****************************************************************************/
36
37 #define VEC_1D_DCT(vb0,vb1,vb2,vb3,va0,va1,va2,va3) \
38 /* 1st stage */ \
39 vz0 = vec_add(vb0,vb2); /* temp[0] = Y[0] + Y[2] */ \
40 vz1 = vec_sub(vb0,vb2); /* temp[1] = Y[0] - Y[2] */ \
41 vz2 = vec_sra(vb1,vec_splat_u16(1)); \
42 vz2 = vec_sub(vz2,vb3); /* temp[2] = Y[1].1/2 - Y[3] */ \
43 vz3 = vec_sra(vb3,vec_splat_u16(1)); \
44 vz3 = vec_add(vb1,vz3); /* temp[3] = Y[1] + Y[3].1/2 */ \
45 /* 2nd stage: output */ \
46 va0 = vec_add(vz0,vz3); /* x[0] = temp[0] + temp[3] */ \
47 va1 = vec_add(vz1,vz2); /* x[1] = temp[1] + temp[2] */ \
48 va2 = vec_sub(vz1,vz2); /* x[2] = temp[1] - temp[2] */ \
49 va3 = vec_sub(vz0,vz3) /* x[3] = temp[0] - temp[3] */
50
51 #define VEC_TRANSPOSE_4(a0,a1,a2,a3,b0,b1,b2,b3) \
52 b0 = vec_mergeh( a0, a0 ); \
53 b1 = vec_mergeh( a1, a0 ); \
54 b2 = vec_mergeh( a2, a0 ); \
55 b3 = vec_mergeh( a3, a0 ); \
56 a0 = vec_mergeh( b0, b2 ); \
57 a1 = vec_mergel( b0, b2 ); \
58 a2 = vec_mergeh( b1, b3 ); \
59 a3 = vec_mergel( b1, b3 ); \
60 b0 = vec_mergeh( a0, a2 ); \
61 b1 = vec_mergel( a0, a2 ); \
62 b2 = vec_mergeh( a1, a3 ); \
63 b3 = vec_mergel( a1, a3 )
64
65 #if HAVE_BIGENDIAN
66 #define vdst_load(d) \
67 vdst_orig = vec_ld(0, dst); \
68 vdst = vec_perm(vdst_orig, zero_u8v, vdst_mask);
69 #else
70 #define vdst_load(d) vdst = vec_vsx_ld(0, dst)
71 #endif
72
73 #define VEC_LOAD_U8_ADD_S16_STORE_U8(va) \
74 vdst_load(); \
75 vdst_ss = (vec_s16) VEC_MERGEH(zero_u8v, vdst); \
76 va = vec_add(va, vdst_ss); \
77 va_u8 = vec_packsu(va, zero_s16v); \
78 va_u32 = vec_splat((vec_u32)va_u8, 0); \
79 vec_ste(va_u32, element, (uint32_t*)dst);
80
81 static void h264_idct_add_altivec(uint8_t *dst, int16_t *block, int stride)
82 {
83 vec_s16 va0, va1, va2, va3;
84 vec_s16 vz0, vz1, vz2, vz3;
85 vec_s16 vtmp0, vtmp1, vtmp2, vtmp3;
86 vec_u8 va_u8;
87 vec_u32 va_u32;
88 vec_s16 vdst_ss;
89 const vec_u16 v6us = vec_splat_u16(6);
90 vec_u8 vdst, vdst_orig;
91 vec_u8 vdst_mask = vec_lvsl(0, dst);
92 int element = ((unsigned long)dst & 0xf) >> 2;
93 LOAD_ZERO;
94
95 block[0] += 32; /* add 32 as a DC-level for rounding */
96
97 vtmp0 = vec_ld(0,block);
98 vtmp1 = vec_sld(vtmp0, vtmp0, 8);
99 vtmp2 = vec_ld(16,block);
100 vtmp3 = vec_sld(vtmp2, vtmp2, 8);
101 memset(block, 0, 16 * sizeof(int16_t));
102
103 VEC_1D_DCT(vtmp0,vtmp1,vtmp2,vtmp3,va0,va1,va2,va3);
104 VEC_TRANSPOSE_4(va0,va1,va2,va3,vtmp0,vtmp1,vtmp2,vtmp3);
105 VEC_1D_DCT(vtmp0,vtmp1,vtmp2,vtmp3,va0,va1,va2,va3);
106
107 va0 = vec_sra(va0,v6us);
108 va1 = vec_sra(va1,v6us);
109 va2 = vec_sra(va2,v6us);
110 va3 = vec_sra(va3,v6us);
111
112 VEC_LOAD_U8_ADD_S16_STORE_U8(va0);
113 dst += stride;
114 VEC_LOAD_U8_ADD_S16_STORE_U8(va1);
115 dst += stride;
116 VEC_LOAD_U8_ADD_S16_STORE_U8(va2);
117 dst += stride;
118 VEC_LOAD_U8_ADD_S16_STORE_U8(va3);
119 }
120
121 #define IDCT8_1D_ALTIVEC(s0, s1, s2, s3, s4, s5, s6, s7, d0, d1, d2, d3, d4, d5, d6, d7) {\
122 /* a0 = SRC(0) + SRC(4); */ \
123 vec_s16 a0v = vec_add(s0, s4); \
124 /* a2 = SRC(0) - SRC(4); */ \
125 vec_s16 a2v = vec_sub(s0, s4); \
126 /* a4 = (SRC(2)>>1) - SRC(6); */ \
127 vec_s16 a4v = vec_sub(vec_sra(s2, onev), s6); \
128 /* a6 = (SRC(6)>>1) + SRC(2); */ \
129 vec_s16 a6v = vec_add(vec_sra(s6, onev), s2); \
130 /* b0 = a0 + a6; */ \
131 vec_s16 b0v = vec_add(a0v, a6v); \
132 /* b2 = a2 + a4; */ \
133 vec_s16 b2v = vec_add(a2v, a4v); \
134 /* b4 = a2 - a4; */ \
135 vec_s16 b4v = vec_sub(a2v, a4v); \
136 /* b6 = a0 - a6; */ \
137 vec_s16 b6v = vec_sub(a0v, a6v); \
138 /* a1 = SRC(5) - SRC(3) - SRC(7) - (SRC(7)>>1); */ \
139 /* a1 = (SRC(5)-SRC(3)) - (SRC(7) + (SRC(7)>>1)); */ \
140 vec_s16 a1v = vec_sub( vec_sub(s5, s3), vec_add(s7, vec_sra(s7, onev)) ); \
141 /* a3 = SRC(7) + SRC(1) - SRC(3) - (SRC(3)>>1); */ \
142 /* a3 = (SRC(7)+SRC(1)) - (SRC(3) + (SRC(3)>>1)); */ \
143 vec_s16 a3v = vec_sub( vec_add(s7, s1), vec_add(s3, vec_sra(s3, onev)) );\
144 /* a5 = SRC(7) - SRC(1) + SRC(5) + (SRC(5)>>1); */ \
145 /* a5 = (SRC(7)-SRC(1)) + SRC(5) + (SRC(5)>>1); */ \
146 vec_s16 a5v = vec_add( vec_sub(s7, s1), vec_add(s5, vec_sra(s5, onev)) );\
147 /* a7 = SRC(5)+SRC(3) + SRC(1) + (SRC(1)>>1); */ \
148 vec_s16 a7v = vec_add( vec_add(s5, s3), vec_add(s1, vec_sra(s1, onev)) );\
149 /* b1 = (a7>>2) + a1; */ \
150 vec_s16 b1v = vec_add( vec_sra(a7v, twov), a1v); \
151 /* b3 = a3 + (a5>>2); */ \
152 vec_s16 b3v = vec_add(a3v, vec_sra(a5v, twov)); \
153 /* b5 = (a3>>2) - a5; */ \
154 vec_s16 b5v = vec_sub( vec_sra(a3v, twov), a5v); \
155 /* b7 = a7 - (a1>>2); */ \
156 vec_s16 b7v = vec_sub( a7v, vec_sra(a1v, twov)); \
157 /* DST(0, b0 + b7); */ \
158 d0 = vec_add(b0v, b7v); \
159 /* DST(1, b2 + b5); */ \
160 d1 = vec_add(b2v, b5v); \
161 /* DST(2, b4 + b3); */ \
162 d2 = vec_add(b4v, b3v); \
163 /* DST(3, b6 + b1); */ \
164 d3 = vec_add(b6v, b1v); \
165 /* DST(4, b6 - b1); */ \
166 d4 = vec_sub(b6v, b1v); \
167 /* DST(5, b4 - b3); */ \
168 d5 = vec_sub(b4v, b3v); \
169 /* DST(6, b2 - b5); */ \
170 d6 = vec_sub(b2v, b5v); \
171 /* DST(7, b0 - b7); */ \
172 d7 = vec_sub(b0v, b7v); \
173 }
174
175 #if HAVE_BIGENDIAN
176 #define GET_2PERM(ldv, stv, d) \
177 ldv = vec_lvsl(0, d); \
178 stv = vec_lvsr(8, d);
179 #define dstv_load(d) \
180 vec_u8 hv = vec_ld( 0, d ); \
181 vec_u8 lv = vec_ld( 7, d); \
182 vec_u8 dstv = vec_perm( hv, lv, (vec_u8)perm_ldv );
183 #define dest_unligned_store(d) \
184 vec_u8 edgehv; \
185 vec_u8 bodyv = vec_perm( idstsum8, idstsum8, perm_stv ); \
186 vec_u8 edgelv = vec_perm( sel, zero_u8v, perm_stv ); \
187 lv = vec_sel( lv, bodyv, edgelv ); \
188 vec_st( lv, 7, d ); \
189 hv = vec_ld( 0, d ); \
190 edgehv = vec_perm( zero_u8v, sel, perm_stv ); \
191 hv = vec_sel( hv, bodyv, edgehv ); \
192 vec_st( hv, 0, d );
193 #else
194
195 #define GET_2PERM(ldv, stv, d) {}
196 #define dstv_load(d) vec_u8 dstv = vec_vsx_ld(0, d)
197 #define dest_unligned_store(d)\
198 vec_u8 dst8 = vec_perm((vec_u8)idstsum8, dstv, vcprm(2,3,s2,s3));\
199 vec_vsx_st(dst8, 0, d)
200 #endif /* HAVE_BIGENDIAN */
201
202 #define ALTIVEC_STORE_SUM_CLIP(dest, idctv, perm_ldv, perm_stv, sel) { \
203 /* unaligned load */ \
204 dstv_load(dest); \
205 vec_s16 idct_sh6 = vec_sra(idctv, sixv); \
206 vec_u16 dst16 = (vec_u16)VEC_MERGEH(zero_u8v, dstv); \
207 vec_s16 idstsum = vec_adds(idct_sh6, (vec_s16)dst16); \
208 vec_u8 idstsum8 = vec_packsu(zero_s16v, idstsum); \
209 /* unaligned store */ \
210 dest_unligned_store(dest);\
211 }
212
213 static void h264_idct8_add_altivec(uint8_t *dst, int16_t *dct, int stride)
214 {
215 vec_s16 s0, s1, s2, s3, s4, s5, s6, s7;
216 vec_s16 d0, d1, d2, d3, d4, d5, d6, d7;
217 vec_s16 idct0, idct1, idct2, idct3, idct4, idct5, idct6, idct7;
218
219 vec_u8 perm_ldv, perm_stv;
220 GET_2PERM(perm_ldv, perm_stv, dst);
221
222 const vec_u16 onev = vec_splat_u16(1);
223 const vec_u16 twov = vec_splat_u16(2);
224 const vec_u16 sixv = vec_splat_u16(6);
225
226 const vec_u8 sel = (vec_u8) {0,0,0,0,0,0,0,0,-1,-1,-1,-1,-1,-1,-1,-1};
227 LOAD_ZERO;
228
229 dct[0] += 32; // rounding for the >>6 at the end
230
231 s0 = vec_ld(0x00, (int16_t*)dct);
232 s1 = vec_ld(0x10, (int16_t*)dct);
233 s2 = vec_ld(0x20, (int16_t*)dct);
234 s3 = vec_ld(0x30, (int16_t*)dct);
235 s4 = vec_ld(0x40, (int16_t*)dct);
236 s5 = vec_ld(0x50, (int16_t*)dct);
237 s6 = vec_ld(0x60, (int16_t*)dct);
238 s7 = vec_ld(0x70, (int16_t*)dct);
239 memset(dct, 0, 64 * sizeof(int16_t));
240
241 IDCT8_1D_ALTIVEC(s0, s1, s2, s3, s4, s5, s6, s7,
242 d0, d1, d2, d3, d4, d5, d6, d7);
243
244 TRANSPOSE8( d0, d1, d2, d3, d4, d5, d6, d7 );
245
246 IDCT8_1D_ALTIVEC(d0, d1, d2, d3, d4, d5, d6, d7,
247 idct0, idct1, idct2, idct3, idct4, idct5, idct6, idct7);
248
249 ALTIVEC_STORE_SUM_CLIP(&dst[0*stride], idct0, perm_ldv, perm_stv, sel);
250 ALTIVEC_STORE_SUM_CLIP(&dst[1*stride], idct1, perm_ldv, perm_stv, sel);
251 ALTIVEC_STORE_SUM_CLIP(&dst[2*stride], idct2, perm_ldv, perm_stv, sel);
252 ALTIVEC_STORE_SUM_CLIP(&dst[3*stride], idct3, perm_ldv, perm_stv, sel);
253 ALTIVEC_STORE_SUM_CLIP(&dst[4*stride], idct4, perm_ldv, perm_stv, sel);
254 ALTIVEC_STORE_SUM_CLIP(&dst[5*stride], idct5, perm_ldv, perm_stv, sel);
255 ALTIVEC_STORE_SUM_CLIP(&dst[6*stride], idct6, perm_ldv, perm_stv, sel);
256 ALTIVEC_STORE_SUM_CLIP(&dst[7*stride], idct7, perm_ldv, perm_stv, sel);
257 }
258
259 static av_always_inline void h264_idct_dc_add_internal(uint8_t *dst, int16_t *block, int stride, int size)
260 {
261 vec_s16 dc16;
262 vec_u8 dcplus, dcminus, v0, v1, v2, v3, aligner;
263 vec_s32 v_dc32;
264 LOAD_ZERO;
265 DECLARE_ALIGNED(16, int, dc);
266 int i;
267
268 dc = (block[0] + 32) >> 6;
269 block[0] = 0;
270 v_dc32 = vec_lde(0, &dc);
271 dc16 = VEC_SPLAT16((vec_s16)v_dc32, 1);
272
273 if (size == 4)
274 dc16 = VEC_SLD16(dc16, zero_s16v, 8);
275 dcplus = vec_packsu(dc16, zero_s16v);
276 dcminus = vec_packsu(vec_sub(zero_s16v, dc16), zero_s16v);
277
278 aligner = vec_lvsr(0, dst);
279 #if !HAVE_BIGENDIAN
280 aligner = vec_perm(aligner, zero_u8v, vcswapc());
281 #endif
282 dcplus = vec_perm(dcplus, dcplus, aligner);
283 dcminus = vec_perm(dcminus, dcminus, aligner);
284
285 for (i = 0; i < size; i += 4) {
286 v0 = vec_ld(0, dst+0*stride);
287 v1 = vec_ld(0, dst+1*stride);
288 v2 = vec_ld(0, dst+2*stride);
289 v3 = vec_ld(0, dst+3*stride);
290
291 v0 = vec_adds(v0, dcplus);
292 v1 = vec_adds(v1, dcplus);
293 v2 = vec_adds(v2, dcplus);
294 v3 = vec_adds(v3, dcplus);
295
296 v0 = vec_subs(v0, dcminus);
297 v1 = vec_subs(v1, dcminus);
298 v2 = vec_subs(v2, dcminus);
299 v3 = vec_subs(v3, dcminus);
300
301 vec_st(v0, 0, dst+0*stride);
302 vec_st(v1, 0, dst+1*stride);
303 vec_st(v2, 0, dst+2*stride);
304 vec_st(v3, 0, dst+3*stride);
305
306 dst += 4*stride;
307 }
308 }
309
310 static void h264_idct_dc_add_altivec(uint8_t *dst, int16_t *block, int stride)
311 {
312 h264_idct_dc_add_internal(dst, block, stride, 4);
313 }
314
315 static void h264_idct8_dc_add_altivec(uint8_t *dst, int16_t *block, int stride)
316 {
317 h264_idct_dc_add_internal(dst, block, stride, 8);
318 }
319
320 static void h264_idct_add16_altivec(uint8_t *dst, const int *block_offset,
321 int16_t *block, int stride,
322 const uint8_t nnzc[15 * 8])
323 {
324 int i;
325 for(i=0; i<16; i++){
326 int nnz = nnzc[ scan8[i] ];
327 if(nnz){
328 if(nnz==1 && block[i*16]) h264_idct_dc_add_altivec(dst + block_offset[i], block + i*16, stride);
329 else h264_idct_add_altivec(dst + block_offset[i], block + i*16, stride);
330 }
331 }
332 }
333
334 static void h264_idct_add16intra_altivec(uint8_t *dst, const int *block_offset,
335 int16_t *block, int stride,
336 const uint8_t nnzc[15 * 8])
337 {
338 int i;
339 for(i=0; i<16; i++){
340 if(nnzc[ scan8[i] ]) h264_idct_add_altivec(dst + block_offset[i], block + i*16, stride);
341 else if(block[i*16]) h264_idct_dc_add_altivec(dst + block_offset[i], block + i*16, stride);
342 }
343 }
344
345 static void h264_idct8_add4_altivec(uint8_t *dst, const int *block_offset,
346 int16_t *block, int stride,
347 const uint8_t nnzc[15 * 8])
348 {
349 int i;
350 for(i=0; i<16; i+=4){
351 int nnz = nnzc[ scan8[i] ];
352 if(nnz){
353 if(nnz==1 && block[i*16]) h264_idct8_dc_add_altivec(dst + block_offset[i], block + i*16, stride);
354 else h264_idct8_add_altivec(dst + block_offset[i], block + i*16, stride);
355 }
356 }
357 }
358
359 static void h264_idct_add8_altivec(uint8_t **dest, const int *block_offset,
360 int16_t *block, int stride,
361 const uint8_t nnzc[15 * 8])
362 {
363 int i, j;
364 for (j = 1; j < 3; j++) {
365 for(i = j * 16; i < j * 16 + 4; i++){
366 if(nnzc[ scan8[i] ])
367 h264_idct_add_altivec(dest[j-1] + block_offset[i], block + i*16, stride);
368 else if(block[i*16])
369 h264_idct_dc_add_altivec(dest[j-1] + block_offset[i], block + i*16, stride);
370 }
371 }
372 }
373
374 #define transpose4x16(r0, r1, r2, r3) { \
375 register vec_u8 r4; \
376 register vec_u8 r5; \
377 register vec_u8 r6; \
378 register vec_u8 r7; \
379 \
380 r4 = vec_mergeh(r0, r2); /*0, 2 set 0*/ \
381 r5 = vec_mergel(r0, r2); /*0, 2 set 1*/ \
382 r6 = vec_mergeh(r1, r3); /*1, 3 set 0*/ \
383 r7 = vec_mergel(r1, r3); /*1, 3 set 1*/ \
384 \
385 r0 = vec_mergeh(r4, r6); /*all set 0*/ \
386 r1 = vec_mergel(r4, r6); /*all set 1*/ \
387 r2 = vec_mergeh(r5, r7); /*all set 2*/ \
388 r3 = vec_mergel(r5, r7); /*all set 3*/ \
389 }
390
391 static inline void write16x4(uint8_t *dst, int dst_stride,
392 register vec_u8 r0, register vec_u8 r1,
393 register vec_u8 r2, register vec_u8 r3) {
394 DECLARE_ALIGNED(16, unsigned char, result)[64];
395 uint32_t *src_int = (uint32_t *)result, *dst_int = (uint32_t *)dst;
396 int int_dst_stride = dst_stride/4;
397
398 vec_st(r0, 0, result);
399 vec_st(r1, 16, result);
400 vec_st(r2, 32, result);
401 vec_st(r3, 48, result);
402 /* FIXME: there has to be a better way!!!! */
403 *dst_int = *src_int;
404 *(dst_int+ int_dst_stride) = *(src_int + 1);
405 *(dst_int+ 2*int_dst_stride) = *(src_int + 2);
406 *(dst_int+ 3*int_dst_stride) = *(src_int + 3);
407 *(dst_int+ 4*int_dst_stride) = *(src_int + 4);
408 *(dst_int+ 5*int_dst_stride) = *(src_int + 5);
409 *(dst_int+ 6*int_dst_stride) = *(src_int + 6);
410 *(dst_int+ 7*int_dst_stride) = *(src_int + 7);
411 *(dst_int+ 8*int_dst_stride) = *(src_int + 8);
412 *(dst_int+ 9*int_dst_stride) = *(src_int + 9);
413 *(dst_int+10*int_dst_stride) = *(src_int + 10);
414 *(dst_int+11*int_dst_stride) = *(src_int + 11);
415 *(dst_int+12*int_dst_stride) = *(src_int + 12);
416 *(dst_int+13*int_dst_stride) = *(src_int + 13);
417 *(dst_int+14*int_dst_stride) = *(src_int + 14);
418 *(dst_int+15*int_dst_stride) = *(src_int + 15);
419 }
420
421 /** @brief performs a 6x16 transpose of data in src, and stores it to dst
422 @todo FIXME: see if we can't spare some vec_lvsl() by them factorizing
423 out of unaligned_load() */
424 #define readAndTranspose16x6(src, src_stride, r8, r9, r10, r11, r12, r13) {\
425 register vec_u8 r0 = unaligned_load(0, src); \
426 register vec_u8 r1 = unaligned_load( src_stride, src); \
427 register vec_u8 r2 = unaligned_load(2* src_stride, src); \
428 register vec_u8 r3 = unaligned_load(3* src_stride, src); \
429 register vec_u8 r4 = unaligned_load(4* src_stride, src); \
430 register vec_u8 r5 = unaligned_load(5* src_stride, src); \
431 register vec_u8 r6 = unaligned_load(6* src_stride, src); \
432 register vec_u8 r7 = unaligned_load(7* src_stride, src); \
433 register vec_u8 r14 = unaligned_load(14*src_stride, src); \
434 register vec_u8 r15 = unaligned_load(15*src_stride, src); \
435 \
436 r8 = unaligned_load( 8*src_stride, src); \
437 r9 = unaligned_load( 9*src_stride, src); \
438 r10 = unaligned_load(10*src_stride, src); \
439 r11 = unaligned_load(11*src_stride, src); \
440 r12 = unaligned_load(12*src_stride, src); \
441 r13 = unaligned_load(13*src_stride, src); \
442 \
443 /*Merge first pairs*/ \
444 r0 = vec_mergeh(r0, r8); /*0, 8*/ \
445 r1 = vec_mergeh(r1, r9); /*1, 9*/ \
446 r2 = vec_mergeh(r2, r10); /*2,10*/ \
447 r3 = vec_mergeh(r3, r11); /*3,11*/ \
448 r4 = vec_mergeh(r4, r12); /*4,12*/ \
449 r5 = vec_mergeh(r5, r13); /*5,13*/ \
450 r6 = vec_mergeh(r6, r14); /*6,14*/ \
451 r7 = vec_mergeh(r7, r15); /*7,15*/ \
452 \
453 /*Merge second pairs*/ \
454 r8 = vec_mergeh(r0, r4); /*0,4, 8,12 set 0*/ \
455 r9 = vec_mergel(r0, r4); /*0,4, 8,12 set 1*/ \
456 r10 = vec_mergeh(r1, r5); /*1,5, 9,13 set 0*/ \
457 r11 = vec_mergel(r1, r5); /*1,5, 9,13 set 1*/ \
458 r12 = vec_mergeh(r2, r6); /*2,6,10,14 set 0*/ \
459 r13 = vec_mergel(r2, r6); /*2,6,10,14 set 1*/ \
460 r14 = vec_mergeh(r3, r7); /*3,7,11,15 set 0*/ \
461 r15 = vec_mergel(r3, r7); /*3,7,11,15 set 1*/ \
462 \
463 /*Third merge*/ \
464 r0 = vec_mergeh(r8, r12); /*0,2,4,6,8,10,12,14 set 0*/ \
465 r1 = vec_mergel(r8, r12); /*0,2,4,6,8,10,12,14 set 1*/ \
466 r2 = vec_mergeh(r9, r13); /*0,2,4,6,8,10,12,14 set 2*/ \
467 r4 = vec_mergeh(r10, r14); /*1,3,5,7,9,11,13,15 set 0*/ \
468 r5 = vec_mergel(r10, r14); /*1,3,5,7,9,11,13,15 set 1*/ \
469 r6 = vec_mergeh(r11, r15); /*1,3,5,7,9,11,13,15 set 2*/ \
470 /* Don't need to compute 3 and 7*/ \
471 \
472 /*Final merge*/ \
473 r8 = vec_mergeh(r0, r4); /*all set 0*/ \
474 r9 = vec_mergel(r0, r4); /*all set 1*/ \
475 r10 = vec_mergeh(r1, r5); /*all set 2*/ \
476 r11 = vec_mergel(r1, r5); /*all set 3*/ \
477 r12 = vec_mergeh(r2, r6); /*all set 4*/ \
478 r13 = vec_mergel(r2, r6); /*all set 5*/ \
479 /* Don't need to compute 14 and 15*/ \
480 \
481 }
482
483 // out: o = |x-y| < a
484 static inline vec_u8 diff_lt_altivec ( register vec_u8 x,
485 register vec_u8 y,
486 register vec_u8 a) {
487
488 register vec_u8 diff = vec_subs(x, y);
489 register vec_u8 diffneg = vec_subs(y, x);
490 register vec_u8 o = vec_or(diff, diffneg); /* |x-y| */
491 o = (vec_u8)vec_cmplt(o, a);
492 return o;
493 }
494
495 static inline vec_u8 h264_deblock_mask ( register vec_u8 p0,
496 register vec_u8 p1,
497 register vec_u8 q0,
498 register vec_u8 q1,
499 register vec_u8 alpha,
500 register vec_u8 beta) {
501
502 register vec_u8 mask;
503 register vec_u8 tempmask;
504
505 mask = diff_lt_altivec(p0, q0, alpha);
506 tempmask = diff_lt_altivec(p1, p0, beta);
507 mask = vec_and(mask, tempmask);
508 tempmask = diff_lt_altivec(q1, q0, beta);
509 mask = vec_and(mask, tempmask);
510
511 return mask;
512 }
513
514 // out: newp1 = clip((p2 + ((p0 + q0 + 1) >> 1)) >> 1, p1-tc0, p1+tc0)
515 static inline vec_u8 h264_deblock_q1(register vec_u8 p0,
516 register vec_u8 p1,
517 register vec_u8 p2,
518 register vec_u8 q0,
519 register vec_u8 tc0) {
520
521 register vec_u8 average = vec_avg(p0, q0);
522 register vec_u8 temp;
523 register vec_u8 uncliped;
524 register vec_u8 ones;
525 register vec_u8 max;
526 register vec_u8 min;
527 register vec_u8 newp1;
528
529 temp = vec_xor(average, p2);
530 average = vec_avg(average, p2); /*avg(p2, avg(p0, q0)) */
531 ones = vec_splat_u8(1);
532 temp = vec_and(temp, ones); /*(p2^avg(p0, q0)) & 1 */
533 uncliped = vec_subs(average, temp); /*(p2+((p0+q0+1)>>1))>>1 */
534 max = vec_adds(p1, tc0);
535 min = vec_subs(p1, tc0);
536 newp1 = vec_max(min, uncliped);
537 newp1 = vec_min(max, newp1);
538 return newp1;
539 }
540
541 #define h264_deblock_p0_q0(p0, p1, q0, q1, tc0masked) { \
542 \
543 const vec_u8 A0v = vec_sl(vec_splat_u8(10), vec_splat_u8(4)); \
544 \
545 register vec_u8 pq0bit = vec_xor(p0,q0); \
546 register vec_u8 q1minus; \
547 register vec_u8 p0minus; \
548 register vec_u8 stage1; \
549 register vec_u8 stage2; \
550 register vec_u8 vec160; \
551 register vec_u8 delta; \
552 register vec_u8 deltaneg; \
553 \
554 q1minus = vec_nor(q1, q1); /* 255 - q1 */ \
555 stage1 = vec_avg(p1, q1minus); /* (p1 - q1 + 256)>>1 */ \
556 stage2 = vec_sr(stage1, vec_splat_u8(1)); /* (p1 - q1 + 256)>>2 = 64 + (p1 - q1) >> 2 */ \
557 p0minus = vec_nor(p0, p0); /* 255 - p0 */ \
558 stage1 = vec_avg(q0, p0minus); /* (q0 - p0 + 256)>>1 */ \
559 pq0bit = vec_and(pq0bit, vec_splat_u8(1)); \
560 stage2 = vec_avg(stage2, pq0bit); /* 32 + ((q0 - p0)&1 + (p1 - q1) >> 2 + 1) >> 1 */ \
561 stage2 = vec_adds(stage2, stage1); /* 160 + ((p0 - q0) + (p1 - q1) >> 2 + 1) >> 1 */ \
562 vec160 = vec_ld(0, &A0v); \
563 deltaneg = vec_subs(vec160, stage2); /* -d */ \
564 delta = vec_subs(stage2, vec160); /* d */ \
565 deltaneg = vec_min(tc0masked, deltaneg); \
566 delta = vec_min(tc0masked, delta); \
567 p0 = vec_subs(p0, deltaneg); \
568 q0 = vec_subs(q0, delta); \
569 p0 = vec_adds(p0, delta); \
570 q0 = vec_adds(q0, deltaneg); \
571 }
572
573 #define h264_loop_filter_luma_altivec(p2, p1, p0, q0, q1, q2, alpha, beta, tc0) { \
574 DECLARE_ALIGNED(16, unsigned char, temp)[16]; \
575 register vec_u8 alphavec; \
576 register vec_u8 betavec; \
577 register vec_u8 mask; \
578 register vec_u8 p1mask; \
579 register vec_u8 q1mask; \
580 register vector signed char tc0vec; \
581 register vec_u8 finaltc0; \
582 register vec_u8 tc0masked; \
583 register vec_u8 newp1; \
584 register vec_u8 newq1; \
585 \
586 temp[0] = alpha; \
587 temp[1] = beta; \
588 alphavec = vec_ld(0, temp); \
589 betavec = vec_splat(alphavec, 0x1); \
590 alphavec = vec_splat(alphavec, 0x0); \
591 mask = h264_deblock_mask(p0, p1, q0, q1, alphavec, betavec); /*if in block */ \
592 \
593 AV_COPY32(temp, tc0); \
594 tc0vec = vec_ld(0, (signed char*)temp); \
595 tc0vec = vec_mergeh(tc0vec, tc0vec); \
596 tc0vec = vec_mergeh(tc0vec, tc0vec); \
597 mask = vec_and(mask, vec_cmpgt(tc0vec, vec_splat_s8(-1))); /* if tc0[i] >= 0 */ \
598 finaltc0 = vec_and((vec_u8)tc0vec, mask); /* tc = tc0 */ \
599 \
600 p1mask = diff_lt_altivec(p2, p0, betavec); \
601 p1mask = vec_and(p1mask, mask); /* if ( |p2 - p0| < beta) */ \
602 tc0masked = vec_and(p1mask, (vec_u8)tc0vec); \
603 finaltc0 = vec_sub(finaltc0, p1mask); /* tc++ */ \
604 newp1 = h264_deblock_q1(p0, p1, p2, q0, tc0masked); \
605 /*end if*/ \
606 \
607 q1mask = diff_lt_altivec(q2, q0, betavec); \
608 q1mask = vec_and(q1mask, mask); /* if ( |q2 - q0| < beta ) */\
609 tc0masked = vec_and(q1mask, (vec_u8)tc0vec); \
610 finaltc0 = vec_sub(finaltc0, q1mask); /* tc++ */ \
611 newq1 = h264_deblock_q1(p0, q1, q2, q0, tc0masked); \
612 /*end if*/ \
613 \
614 h264_deblock_p0_q0(p0, p1, q0, q1, finaltc0); \
615 p1 = newp1; \
616 q1 = newq1; \
617 }
618
619 static void h264_v_loop_filter_luma_altivec(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0) {
620
621 if ((tc0[0] & tc0[1] & tc0[2] & tc0[3]) >= 0) {
622 register vec_u8 p2 = vec_ld(-3*stride, pix);
623 register vec_u8 p1 = vec_ld(-2*stride, pix);
624 register vec_u8 p0 = vec_ld(-1*stride, pix);
625 register vec_u8 q0 = vec_ld(0, pix);
626 register vec_u8 q1 = vec_ld(stride, pix);
627 register vec_u8 q2 = vec_ld(2*stride, pix);
628 h264_loop_filter_luma_altivec(p2, p1, p0, q0, q1, q2, alpha, beta, tc0);
629 vec_st(p1, -2*stride, pix);
630 vec_st(p0, -1*stride, pix);
631 vec_st(q0, 0, pix);
632 vec_st(q1, stride, pix);
633 }
634 }
635
636 static void h264_h_loop_filter_luma_altivec(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0) {
637
638 register vec_u8 line0, line1, line2, line3, line4, line5;
639 if ((tc0[0] & tc0[1] & tc0[2] & tc0[3]) < 0)
640 return;
641 readAndTranspose16x6(pix-3, stride, line0, line1, line2, line3, line4, line5);
642 h264_loop_filter_luma_altivec(line0, line1, line2, line3, line4, line5, alpha, beta, tc0);
643 transpose4x16(line1, line2, line3, line4);
644 write16x4(pix-2, stride, line1, line2, line3, line4);
645 }
646
647 static av_always_inline
648 void weight_h264_W_altivec(uint8_t *block, int stride, int height,
649 int log2_denom, int weight, int offset, int w)
650 {
651 int y, aligned;
652 vec_u8 vblock;
653 vec_s16 vtemp, vweight, voffset, v0, v1;
654 vec_u16 vlog2_denom;
655 DECLARE_ALIGNED(16, int32_t, temp)[4];
656 LOAD_ZERO;
657
658 offset <<= log2_denom;
659 if(log2_denom) offset += 1<<(log2_denom-1);
660 temp[0] = log2_denom;
661 temp[1] = weight;
662 temp[2] = offset;
663
664 vtemp = (vec_s16)vec_ld(0, temp);
665 #if !HAVE_BIGENDIAN
666 vtemp =(vec_s16)vec_perm(vtemp, vtemp, vcswapi2s(0,1,2,3));
667 #endif
668 vlog2_denom = (vec_u16)vec_splat(vtemp, 1);
669 vweight = vec_splat(vtemp, 3);
670 voffset = vec_splat(vtemp, 5);
671 aligned = !((unsigned long)block & 0xf);
672
673 for (y = 0; y < height; y++) {
674 vblock = vec_ld(0, block);
675
676 v0 = (vec_s16)VEC_MERGEH(zero_u8v, vblock);
677 v1 = (vec_s16)VEC_MERGEL(zero_u8v, vblock);
678
679 if (w == 16 || aligned) {
680 v0 = vec_mladd(v0, vweight, zero_s16v);
681 v0 = vec_adds(v0, voffset);
682 v0 = vec_sra(v0, vlog2_denom);
683 }
684 if (w == 16 || !aligned) {
685 v1 = vec_mladd(v1, vweight, zero_s16v);
686 v1 = vec_adds(v1, voffset);
687 v1 = vec_sra(v1, vlog2_denom);
688 }
689 vblock = vec_packsu(v0, v1);
690 vec_st(vblock, 0, block);
691
692 block += stride;
693 }
694 }
695
696 static av_always_inline
697 void biweight_h264_W_altivec(uint8_t *dst, uint8_t *src, int stride, int height,
698 int log2_denom, int weightd, int weights, int offset, int w)
699 {
700 int y, dst_aligned, src_aligned;
701 vec_u8 vsrc, vdst;
702 vec_s16 vtemp, vweights, vweightd, voffset, v0, v1, v2, v3;
703 vec_u16 vlog2_denom;
704 DECLARE_ALIGNED(16, int32_t, temp)[4];
705 LOAD_ZERO;
706
707 offset = ((offset + 1) | 1) << log2_denom;
708 temp[0] = log2_denom+1;
709 temp[1] = weights;
710 temp[2] = weightd;
711 temp[3] = offset;
712
713 vtemp = (vec_s16)vec_ld(0, temp);
714 #if !HAVE_BIGENDIAN
715 vtemp =(vec_s16)vec_perm(vtemp, vtemp, vcswapi2s(0,1,2,3));
716 #endif
717 vlog2_denom = (vec_u16)vec_splat(vtemp, 1);
718 vweights = vec_splat(vtemp, 3);
719 vweightd = vec_splat(vtemp, 5);
720 voffset = vec_splat(vtemp, 7);
721 dst_aligned = !((unsigned long)dst & 0xf);
722 src_aligned = !((unsigned long)src & 0xf);
723
724 for (y = 0; y < height; y++) {
725 vdst = vec_ld(0, dst);
726 vsrc = vec_ld(0, src);
727
728 v0 = (vec_s16)VEC_MERGEH(zero_u8v, vdst);
729 v1 = (vec_s16)VEC_MERGEL(zero_u8v, vdst);
730 v2 = (vec_s16)VEC_MERGEH(zero_u8v, vsrc);
731 v3 = (vec_s16)VEC_MERGEL(zero_u8v, vsrc);
732
733 if (w == 8) {
734 if (src_aligned)
735 v3 = v2;
736 else
737 v2 = v3;
738 }
739
740 if (w == 16 || dst_aligned) {
741 v0 = vec_mladd(v0, vweightd, zero_s16v);
742 v2 = vec_mladd(v2, vweights, zero_s16v);
743
744 v0 = vec_adds(v0, voffset);
745 v0 = vec_adds(v0, v2);
746 v0 = vec_sra(v0, vlog2_denom);
747 }
748 if (w == 16 || !dst_aligned) {
749 v1 = vec_mladd(v1, vweightd, zero_s16v);
750 v3 = vec_mladd(v3, vweights, zero_s16v);
751
752 v1 = vec_adds(v1, voffset);
753 v1 = vec_adds(v1, v3);
754 v1 = vec_sra(v1, vlog2_denom);
755 }
756 vdst = vec_packsu(v0, v1);
757 vec_st(vdst, 0, dst);
758
759 dst += stride;
760 src += stride;
761 }
762 }
763
764 #define H264_WEIGHT(W) \
765 static void weight_h264_pixels ## W ## _altivec(uint8_t *block, int stride, int height, \
766 int log2_denom, int weight, int offset) \
767 { \
768 weight_h264_W_altivec(block, stride, height, log2_denom, weight, offset, W); \
769 }\
770 static void biweight_h264_pixels ## W ## _altivec(uint8_t *dst, uint8_t *src, int stride, int height, \
771 int log2_denom, int weightd, int weights, int offset) \
772 { \
773 biweight_h264_W_altivec(dst, src, stride, height, log2_denom, weightd, weights, offset, W); \
774 }
775
776 H264_WEIGHT(16)
777 H264_WEIGHT( 8)
778 #endif /* HAVE_ALTIVEC */
779
780 av_cold void ff_h264dsp_init_ppc(H264DSPContext *c, const int bit_depth,
781 const int chroma_format_idc)
782 {
783 #if HAVE_ALTIVEC
784 if (!PPC_ALTIVEC(av_get_cpu_flags()))
785 return;
786
787 if (bit_depth == 8) {
788 c->h264_idct_add = h264_idct_add_altivec;
789 if (chroma_format_idc <= 1)
790 c->h264_idct_add8 = h264_idct_add8_altivec;
791 c->h264_idct_add16 = h264_idct_add16_altivec;
792 c->h264_idct_add16intra = h264_idct_add16intra_altivec;
793 c->h264_idct_dc_add= h264_idct_dc_add_altivec;
794 c->h264_idct8_dc_add = h264_idct8_dc_add_altivec;
795 c->h264_idct8_add = h264_idct8_add_altivec;
796 c->h264_idct8_add4 = h264_idct8_add4_altivec;
797 c->h264_v_loop_filter_luma= h264_v_loop_filter_luma_altivec;
798 c->h264_h_loop_filter_luma= h264_h_loop_filter_luma_altivec;
799
800 c->weight_h264_pixels_tab[0] = weight_h264_pixels16_altivec;
801 c->weight_h264_pixels_tab[1] = weight_h264_pixels8_altivec;
802 c->biweight_h264_pixels_tab[0] = biweight_h264_pixels16_altivec;
803 c->biweight_h264_pixels_tab[1] = biweight_h264_pixels8_altivec;
804 }
805 #endif /* HAVE_ALTIVEC */
806 }
ffmpeg Piment Noir Debian package repository