projects / libav.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
h264: do not use 422 functions for monochrome
[libav.git] / libavcodec / ppc / h264dsp.c
1 /*
2 * Copyright (c) 2004 Romain Dolbeau <romain@dolbeau.org>
3 *
4 * This file is part of Libav.
5 *
6 * Libav 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 * Libav 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 Libav; if not, write to the Free Software
18 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
19 */
20
21 #include "config.h"
22 #include "libavutil/attributes.h"
23 #include "libavutil/cpu.h"
24 #include "libavutil/intreadwrite.h"
25 #include "libavutil/ppc/types_altivec.h"
26 #include "libavutil/ppc/util_altivec.h"
27 #include "libavcodec/h264data.h"
28 #include "libavcodec/h264dsp.h"
29
30 #if HAVE_ALTIVEC
31
32 /****************************************************************************
33 * IDCT transform:
34 ****************************************************************************/
35
36 #define VEC_1D_DCT(vb0,vb1,vb2,vb3,va0,va1,va2,va3) \
37 /* 1st stage */ \
38 vz0 = vec_add(vb0,vb2); /* temp[0] = Y[0] + Y[2] */ \
39 vz1 = vec_sub(vb0,vb2); /* temp[1] = Y[0] - Y[2] */ \
40 vz2 = vec_sra(vb1,vec_splat_u16(1)); \
41 vz2 = vec_sub(vz2,vb3); /* temp[2] = Y[1].1/2 - Y[3] */ \
42 vz3 = vec_sra(vb3,vec_splat_u16(1)); \
43 vz3 = vec_add(vb1,vz3); /* temp[3] = Y[1] + Y[3].1/2 */ \
44 /* 2nd stage: output */ \
45 va0 = vec_add(vz0,vz3); /* x[0] = temp[0] + temp[3] */ \
46 va1 = vec_add(vz1,vz2); /* x[1] = temp[1] + temp[2] */ \
47 va2 = vec_sub(vz1,vz2); /* x[2] = temp[1] - temp[2] */ \
48 va3 = vec_sub(vz0,vz3) /* x[3] = temp[0] - temp[3] */
49
50 #define VEC_TRANSPOSE_4(a0,a1,a2,a3,b0,b1,b2,b3) \
51 b0 = vec_mergeh( a0, a0 ); \
52 b1 = vec_mergeh( a1, a0 ); \
53 b2 = vec_mergeh( a2, a0 ); \
54 b3 = vec_mergeh( a3, a0 ); \
55 a0 = vec_mergeh( b0, b2 ); \
56 a1 = vec_mergel( b0, b2 ); \
57 a2 = vec_mergeh( b1, b3 ); \
58 a3 = vec_mergel( b1, b3 ); \
59 b0 = vec_mergeh( a0, a2 ); \
60 b1 = vec_mergel( a0, a2 ); \
61 b2 = vec_mergeh( a1, a3 ); \
62 b3 = vec_mergel( a1, a3 )
63
64 #define VEC_LOAD_U8_ADD_S16_STORE_U8(va) \
65 vdst_orig = vec_ld(0, dst); \
66 vdst = vec_perm(vdst_orig, zero_u8v, vdst_mask); \
67 vdst_ss = (vec_s16) vec_mergeh(zero_u8v, vdst); \
68 va = vec_add(va, vdst_ss); \
69 va_u8 = vec_packsu(va, zero_s16v); \
70 va_u32 = vec_splat((vec_u32)va_u8, 0); \
71 vec_ste(va_u32, element, (uint32_t*)dst);
72
73 static void h264_idct_add_altivec(uint8_t *dst, int16_t *block, int stride)
74 {
75 vec_s16 va0, va1, va2, va3;
76 vec_s16 vz0, vz1, vz2, vz3;
77 vec_s16 vtmp0, vtmp1, vtmp2, vtmp3;
78 vec_u8 va_u8;
79 vec_u32 va_u32;
80 vec_s16 vdst_ss;
81 const vec_u16 v6us = vec_splat_u16(6);
82 vec_u8 vdst, vdst_orig;
83 vec_u8 vdst_mask = vec_lvsl(0, dst);
84 int element = ((unsigned long)dst & 0xf) >> 2;
85 LOAD_ZERO;
86
87 block[0] += 32; /* add 32 as a DC-level for rounding */
88
89 vtmp0 = vec_ld(0,block);
90 vtmp1 = vec_sld(vtmp0, vtmp0, 8);
91 vtmp2 = vec_ld(16,block);
92 vtmp3 = vec_sld(vtmp2, vtmp2, 8);
93 memset(block, 0, 16 * sizeof(int16_t));
94
95 VEC_1D_DCT(vtmp0,vtmp1,vtmp2,vtmp3,va0,va1,va2,va3);
96 VEC_TRANSPOSE_4(va0,va1,va2,va3,vtmp0,vtmp1,vtmp2,vtmp3);
97 VEC_1D_DCT(vtmp0,vtmp1,vtmp2,vtmp3,va0,va1,va2,va3);
98
99 va0 = vec_sra(va0,v6us);
100 va1 = vec_sra(va1,v6us);
101 va2 = vec_sra(va2,v6us);
102 va3 = vec_sra(va3,v6us);
103
104 VEC_LOAD_U8_ADD_S16_STORE_U8(va0);
105 dst += stride;
106 VEC_LOAD_U8_ADD_S16_STORE_U8(va1);
107 dst += stride;
108 VEC_LOAD_U8_ADD_S16_STORE_U8(va2);
109 dst += stride;
110 VEC_LOAD_U8_ADD_S16_STORE_U8(va3);
111 }
112
113 #define IDCT8_1D_ALTIVEC(s0, s1, s2, s3, s4, s5, s6, s7, d0, d1, d2, d3, d4, d5, d6, d7) {\
114 /* a0 = SRC(0) + SRC(4); */ \
115 vec_s16 a0v = vec_add(s0, s4); \
116 /* a2 = SRC(0) - SRC(4); */ \
117 vec_s16 a2v = vec_sub(s0, s4); \
118 /* a4 = (SRC(2)>>1) - SRC(6); */ \
119 vec_s16 a4v = vec_sub(vec_sra(s2, onev), s6); \
120 /* a6 = (SRC(6)>>1) + SRC(2); */ \
121 vec_s16 a6v = vec_add(vec_sra(s6, onev), s2); \
122 /* b0 = a0 + a6; */ \
123 vec_s16 b0v = vec_add(a0v, a6v); \
124 /* b2 = a2 + a4; */ \
125 vec_s16 b2v = vec_add(a2v, a4v); \
126 /* b4 = a2 - a4; */ \
127 vec_s16 b4v = vec_sub(a2v, a4v); \
128 /* b6 = a0 - a6; */ \
129 vec_s16 b6v = vec_sub(a0v, a6v); \
130 /* a1 = SRC(5) - SRC(3) - SRC(7) - (SRC(7)>>1); */ \
131 /* a1 = (SRC(5)-SRC(3)) - (SRC(7) + (SRC(7)>>1)); */ \
132 vec_s16 a1v = vec_sub( vec_sub(s5, s3), vec_add(s7, vec_sra(s7, onev)) ); \
133 /* a3 = SRC(7) + SRC(1) - SRC(3) - (SRC(3)>>1); */ \
134 /* a3 = (SRC(7)+SRC(1)) - (SRC(3) + (SRC(3)>>1)); */ \
135 vec_s16 a3v = vec_sub( vec_add(s7, s1), vec_add(s3, vec_sra(s3, onev)) );\
136 /* a5 = SRC(7) - SRC(1) + SRC(5) + (SRC(5)>>1); */ \
137 /* a5 = (SRC(7)-SRC(1)) + SRC(5) + (SRC(5)>>1); */ \
138 vec_s16 a5v = vec_add( vec_sub(s7, s1), vec_add(s5, vec_sra(s5, onev)) );\
139 /* a7 = SRC(5)+SRC(3) + SRC(1) + (SRC(1)>>1); */ \
140 vec_s16 a7v = vec_add( vec_add(s5, s3), vec_add(s1, vec_sra(s1, onev)) );\
141 /* b1 = (a7>>2) + a1; */ \
142 vec_s16 b1v = vec_add( vec_sra(a7v, twov), a1v); \
143 /* b3 = a3 + (a5>>2); */ \
144 vec_s16 b3v = vec_add(a3v, vec_sra(a5v, twov)); \
145 /* b5 = (a3>>2) - a5; */ \
146 vec_s16 b5v = vec_sub( vec_sra(a3v, twov), a5v); \
147 /* b7 = a7 - (a1>>2); */ \
148 vec_s16 b7v = vec_sub( a7v, vec_sra(a1v, twov)); \
149 /* DST(0, b0 + b7); */ \
150 d0 = vec_add(b0v, b7v); \
151 /* DST(1, b2 + b5); */ \
152 d1 = vec_add(b2v, b5v); \
153 /* DST(2, b4 + b3); */ \
154 d2 = vec_add(b4v, b3v); \
155 /* DST(3, b6 + b1); */ \
156 d3 = vec_add(b6v, b1v); \
157 /* DST(4, b6 - b1); */ \
158 d4 = vec_sub(b6v, b1v); \
159 /* DST(5, b4 - b3); */ \
160 d5 = vec_sub(b4v, b3v); \
161 /* DST(6, b2 - b5); */ \
162 d6 = vec_sub(b2v, b5v); \
163 /* DST(7, b0 - b7); */ \
164 d7 = vec_sub(b0v, b7v); \
165 }
166
167 #define ALTIVEC_STORE_SUM_CLIP(dest, idctv, perm_ldv, perm_stv, sel) { \
168 /* unaligned load */ \
169 vec_u8 hv = vec_ld( 0, dest ); \
170 vec_u8 lv = vec_ld( 7, dest ); \
171 vec_u8 dstv = vec_perm( hv, lv, (vec_u8)perm_ldv ); \
172 vec_s16 idct_sh6 = vec_sra(idctv, sixv); \
173 vec_u16 dst16 = (vec_u16)vec_mergeh(zero_u8v, dstv); \
174 vec_s16 idstsum = vec_adds(idct_sh6, (vec_s16)dst16); \
175 vec_u8 idstsum8 = vec_packsu(zero_s16v, idstsum); \
176 vec_u8 edgehv; \
177 /* unaligned store */ \
178 vec_u8 bodyv = vec_perm( idstsum8, idstsum8, perm_stv );\
179 vec_u8 edgelv = vec_perm( sel, zero_u8v, perm_stv ); \
180 lv = vec_sel( lv, bodyv, edgelv ); \
181 vec_st( lv, 7, dest ); \
182 hv = vec_ld( 0, dest ); \
183 edgehv = vec_perm( zero_u8v, sel, perm_stv ); \
184 hv = vec_sel( hv, bodyv, edgehv ); \
185 vec_st( hv, 0, dest ); \
186 }
187
188 static void h264_idct8_add_altivec(uint8_t *dst, int16_t *dct, int stride)
189 {
190 vec_s16 s0, s1, s2, s3, s4, s5, s6, s7;
191 vec_s16 d0, d1, d2, d3, d4, d5, d6, d7;
192 vec_s16 idct0, idct1, idct2, idct3, idct4, idct5, idct6, idct7;
193
194 vec_u8 perm_ldv = vec_lvsl(0, dst);
195 vec_u8 perm_stv = vec_lvsr(8, dst);
196
197 const vec_u16 onev = vec_splat_u16(1);
198 const vec_u16 twov = vec_splat_u16(2);
199 const vec_u16 sixv = vec_splat_u16(6);
200
201 const vec_u8 sel = (vec_u8) {0,0,0,0,0,0,0,0,-1,-1,-1,-1,-1,-1,-1,-1};
202 LOAD_ZERO;
203
204 dct[0] += 32; // rounding for the >>6 at the end
205
206 s0 = vec_ld(0x00, (int16_t*)dct);
207 s1 = vec_ld(0x10, (int16_t*)dct);
208 s2 = vec_ld(0x20, (int16_t*)dct);
209 s3 = vec_ld(0x30, (int16_t*)dct);
210 s4 = vec_ld(0x40, (int16_t*)dct);
211 s5 = vec_ld(0x50, (int16_t*)dct);
212 s6 = vec_ld(0x60, (int16_t*)dct);
213 s7 = vec_ld(0x70, (int16_t*)dct);
214 memset(dct, 0, 64 * sizeof(int16_t));
215
216 IDCT8_1D_ALTIVEC(s0, s1, s2, s3, s4, s5, s6, s7,
217 d0, d1, d2, d3, d4, d5, d6, d7);
218
219 TRANSPOSE8( d0, d1, d2, d3, d4, d5, d6, d7 );
220
221 IDCT8_1D_ALTIVEC(d0, d1, d2, d3, d4, d5, d6, d7,
222 idct0, idct1, idct2, idct3, idct4, idct5, idct6, idct7);
223
224 ALTIVEC_STORE_SUM_CLIP(&dst[0*stride], idct0, perm_ldv, perm_stv, sel);
225 ALTIVEC_STORE_SUM_CLIP(&dst[1*stride], idct1, perm_ldv, perm_stv, sel);
226 ALTIVEC_STORE_SUM_CLIP(&dst[2*stride], idct2, perm_ldv, perm_stv, sel);
227 ALTIVEC_STORE_SUM_CLIP(&dst[3*stride], idct3, perm_ldv, perm_stv, sel);
228 ALTIVEC_STORE_SUM_CLIP(&dst[4*stride], idct4, perm_ldv, perm_stv, sel);
229 ALTIVEC_STORE_SUM_CLIP(&dst[5*stride], idct5, perm_ldv, perm_stv, sel);
230 ALTIVEC_STORE_SUM_CLIP(&dst[6*stride], idct6, perm_ldv, perm_stv, sel);
231 ALTIVEC_STORE_SUM_CLIP(&dst[7*stride], idct7, perm_ldv, perm_stv, sel);
232 }
233
234 static av_always_inline void h264_idct_dc_add_internal(uint8_t *dst, int16_t *block, int stride, int size)
235 {
236 vec_s16 dc16;
237 vec_u8 dcplus, dcminus, v0, v1, v2, v3, aligner;
238 LOAD_ZERO;
239 DECLARE_ALIGNED(16, int, dc);
240 int i;
241
242 dc = (block[0] + 32) >> 6;
243 block[0] = 0;
244 dc16 = vec_splat((vec_s16) vec_lde(0, &dc), 1);
245
246 if (size == 4)
247 dc16 = vec_sld(dc16, zero_s16v, 8);
248 dcplus = vec_packsu(dc16, zero_s16v);
249 dcminus = vec_packsu(vec_sub(zero_s16v, dc16), zero_s16v);
250
251 aligner = vec_lvsr(0, dst);
252 dcplus = vec_perm(dcplus, dcplus, aligner);
253 dcminus = vec_perm(dcminus, dcminus, aligner);
254
255 for (i = 0; i < size; i += 4) {
256 v0 = vec_ld(0, dst+0*stride);
257 v1 = vec_ld(0, dst+1*stride);
258 v2 = vec_ld(0, dst+2*stride);
259 v3 = vec_ld(0, dst+3*stride);
260
261 v0 = vec_adds(v0, dcplus);
262 v1 = vec_adds(v1, dcplus);
263 v2 = vec_adds(v2, dcplus);
264 v3 = vec_adds(v3, dcplus);
265
266 v0 = vec_subs(v0, dcminus);
267 v1 = vec_subs(v1, dcminus);
268 v2 = vec_subs(v2, dcminus);
269 v3 = vec_subs(v3, dcminus);
270
271 vec_st(v0, 0, dst+0*stride);
272 vec_st(v1, 0, dst+1*stride);
273 vec_st(v2, 0, dst+2*stride);
274 vec_st(v3, 0, dst+3*stride);
275
276 dst += 4*stride;
277 }
278 }
279
280 static void h264_idct_dc_add_altivec(uint8_t *dst, int16_t *block, int stride)
281 {
282 h264_idct_dc_add_internal(dst, block, stride, 4);
283 }
284
285 static void h264_idct8_dc_add_altivec(uint8_t *dst, int16_t *block, int stride)
286 {
287 h264_idct_dc_add_internal(dst, block, stride, 8);
288 }
289
290 static void h264_idct_add16_altivec(uint8_t *dst, const int *block_offset,
291 int16_t *block, int stride,
292 const uint8_t nnzc[15 * 8])
293 {
294 int i;
295 for(i=0; i<16; i++){
296 int nnz = nnzc[ scan8[i] ];
297 if(nnz){
298 if(nnz==1 && block[i*16]) h264_idct_dc_add_altivec(dst + block_offset[i], block + i*16, stride);
299 else h264_idct_add_altivec(dst + block_offset[i], block + i*16, stride);
300 }
301 }
302 }
303
304 static void h264_idct_add16intra_altivec(uint8_t *dst, const int *block_offset,
305 int16_t *block, int stride,
306 const uint8_t nnzc[15 * 8])
307 {
308 int i;
309 for(i=0; i<16; i++){
310 if(nnzc[ scan8[i] ]) h264_idct_add_altivec(dst + block_offset[i], block + i*16, stride);
311 else if(block[i*16]) h264_idct_dc_add_altivec(dst + block_offset[i], block + i*16, stride);
312 }
313 }
314
315 static void h264_idct8_add4_altivec(uint8_t *dst, const int *block_offset,
316 int16_t *block, int stride,
317 const uint8_t nnzc[15 * 8])
318 {
319 int i;
320 for(i=0; i<16; i+=4){
321 int nnz = nnzc[ scan8[i] ];
322 if(nnz){
323 if(nnz==1 && block[i*16]) h264_idct8_dc_add_altivec(dst + block_offset[i], block + i*16, stride);
324 else h264_idct8_add_altivec(dst + block_offset[i], block + i*16, stride);
325 }
326 }
327 }
328
329 static void h264_idct_add8_altivec(uint8_t **dest, const int *block_offset,
330 int16_t *block, int stride,
331 const uint8_t nnzc[15 * 8])
332 {
333 int i, j;
334 for (j = 1; j < 3; j++) {
335 for(i = j * 16; i < j * 16 + 4; i++){
336 if(nnzc[ scan8[i] ])
337 h264_idct_add_altivec(dest[j-1] + block_offset[i], block + i*16, stride);
338 else if(block[i*16])
339 h264_idct_dc_add_altivec(dest[j-1] + block_offset[i], block + i*16, stride);
340 }
341 }
342 }
343
344 #define transpose4x16(r0, r1, r2, r3) { \
345 register vec_u8 r4; \
346 register vec_u8 r5; \
347 register vec_u8 r6; \
348 register vec_u8 r7; \
349 \
350 r4 = vec_mergeh(r0, r2); /*0, 2 set 0*/ \
351 r5 = vec_mergel(r0, r2); /*0, 2 set 1*/ \
352 r6 = vec_mergeh(r1, r3); /*1, 3 set 0*/ \
353 r7 = vec_mergel(r1, r3); /*1, 3 set 1*/ \
354 \
355 r0 = vec_mergeh(r4, r6); /*all set 0*/ \
356 r1 = vec_mergel(r4, r6); /*all set 1*/ \
357 r2 = vec_mergeh(r5, r7); /*all set 2*/ \
358 r3 = vec_mergel(r5, r7); /*all set 3*/ \
359 }
360
361 static inline void write16x4(uint8_t *dst, int dst_stride,
362 register vec_u8 r0, register vec_u8 r1,
363 register vec_u8 r2, register vec_u8 r3) {
364 DECLARE_ALIGNED(16, unsigned char, result)[64];
365 uint32_t *src_int = (uint32_t *)result, *dst_int = (uint32_t *)dst;
366 int int_dst_stride = dst_stride/4;
367
368 vec_st(r0, 0, result);
369 vec_st(r1, 16, result);
370 vec_st(r2, 32, result);
371 vec_st(r3, 48, result);
372 /* FIXME: there has to be a better way!!!! */
373 *dst_int = *src_int;
374 *(dst_int+ int_dst_stride) = *(src_int + 1);
375 *(dst_int+ 2*int_dst_stride) = *(src_int + 2);
376 *(dst_int+ 3*int_dst_stride) = *(src_int + 3);
377 *(dst_int+ 4*int_dst_stride) = *(src_int + 4);
378 *(dst_int+ 5*int_dst_stride) = *(src_int + 5);
379 *(dst_int+ 6*int_dst_stride) = *(src_int + 6);
380 *(dst_int+ 7*int_dst_stride) = *(src_int + 7);
381 *(dst_int+ 8*int_dst_stride) = *(src_int + 8);
382 *(dst_int+ 9*int_dst_stride) = *(src_int + 9);
383 *(dst_int+10*int_dst_stride) = *(src_int + 10);
384 *(dst_int+11*int_dst_stride) = *(src_int + 11);
385 *(dst_int+12*int_dst_stride) = *(src_int + 12);
386 *(dst_int+13*int_dst_stride) = *(src_int + 13);
387 *(dst_int+14*int_dst_stride) = *(src_int + 14);
388 *(dst_int+15*int_dst_stride) = *(src_int + 15);
389 }
390
391 /** @brief performs a 6x16 transpose of data in src, and stores it to dst
392 @todo FIXME: see if we can't spare some vec_lvsl() by them factorizing
393 out of unaligned_load() */
394 #define readAndTranspose16x6(src, src_stride, r8, r9, r10, r11, r12, r13) {\
395 register vec_u8 r0 = unaligned_load(0, src); \
396 register vec_u8 r1 = unaligned_load( src_stride, src); \
397 register vec_u8 r2 = unaligned_load(2* src_stride, src); \
398 register vec_u8 r3 = unaligned_load(3* src_stride, src); \
399 register vec_u8 r4 = unaligned_load(4* src_stride, src); \
400 register vec_u8 r5 = unaligned_load(5* src_stride, src); \
401 register vec_u8 r6 = unaligned_load(6* src_stride, src); \
402 register vec_u8 r7 = unaligned_load(7* src_stride, src); \
403 register vec_u8 r14 = unaligned_load(14*src_stride, src); \
404 register vec_u8 r15 = unaligned_load(15*src_stride, src); \
405 \
406 r8 = unaligned_load( 8*src_stride, src); \
407 r9 = unaligned_load( 9*src_stride, src); \
408 r10 = unaligned_load(10*src_stride, src); \
409 r11 = unaligned_load(11*src_stride, src); \
410 r12 = unaligned_load(12*src_stride, src); \
411 r13 = unaligned_load(13*src_stride, src); \
412 \
413 /*Merge first pairs*/ \
414 r0 = vec_mergeh(r0, r8); /*0, 8*/ \
415 r1 = vec_mergeh(r1, r9); /*1, 9*/ \
416 r2 = vec_mergeh(r2, r10); /*2,10*/ \
417 r3 = vec_mergeh(r3, r11); /*3,11*/ \
418 r4 = vec_mergeh(r4, r12); /*4,12*/ \
419 r5 = vec_mergeh(r5, r13); /*5,13*/ \
420 r6 = vec_mergeh(r6, r14); /*6,14*/ \
421 r7 = vec_mergeh(r7, r15); /*7,15*/ \
422 \
423 /*Merge second pairs*/ \
424 r8 = vec_mergeh(r0, r4); /*0,4, 8,12 set 0*/ \
425 r9 = vec_mergel(r0, r4); /*0,4, 8,12 set 1*/ \
426 r10 = vec_mergeh(r1, r5); /*1,5, 9,13 set 0*/ \
427 r11 = vec_mergel(r1, r5); /*1,5, 9,13 set 1*/ \
428 r12 = vec_mergeh(r2, r6); /*2,6,10,14 set 0*/ \
429 r13 = vec_mergel(r2, r6); /*2,6,10,14 set 1*/ \
430 r14 = vec_mergeh(r3, r7); /*3,7,11,15 set 0*/ \
431 r15 = vec_mergel(r3, r7); /*3,7,11,15 set 1*/ \
432 \
433 /*Third merge*/ \
434 r0 = vec_mergeh(r8, r12); /*0,2,4,6,8,10,12,14 set 0*/ \
435 r1 = vec_mergel(r8, r12); /*0,2,4,6,8,10,12,14 set 1*/ \
436 r2 = vec_mergeh(r9, r13); /*0,2,4,6,8,10,12,14 set 2*/ \
437 r4 = vec_mergeh(r10, r14); /*1,3,5,7,9,11,13,15 set 0*/ \
438 r5 = vec_mergel(r10, r14); /*1,3,5,7,9,11,13,15 set 1*/ \
439 r6 = vec_mergeh(r11, r15); /*1,3,5,7,9,11,13,15 set 2*/ \
440 /* Don't need to compute 3 and 7*/ \
441 \
442 /*Final merge*/ \
443 r8 = vec_mergeh(r0, r4); /*all set 0*/ \
444 r9 = vec_mergel(r0, r4); /*all set 1*/ \
445 r10 = vec_mergeh(r1, r5); /*all set 2*/ \
446 r11 = vec_mergel(r1, r5); /*all set 3*/ \
447 r12 = vec_mergeh(r2, r6); /*all set 4*/ \
448 r13 = vec_mergel(r2, r6); /*all set 5*/ \
449 /* Don't need to compute 14 and 15*/ \
450 \
451 }
452
453 // out: o = |x-y| < a
454 static inline vec_u8 diff_lt_altivec ( register vec_u8 x,
455 register vec_u8 y,
456 register vec_u8 a) {
457
458 register vec_u8 diff = vec_subs(x, y);
459 register vec_u8 diffneg = vec_subs(y, x);
460 register vec_u8 o = vec_or(diff, diffneg); /* |x-y| */
461 o = (vec_u8)vec_cmplt(o, a);
462 return o;
463 }
464
465 static inline vec_u8 h264_deblock_mask ( register vec_u8 p0,
466 register vec_u8 p1,
467 register vec_u8 q0,
468 register vec_u8 q1,
469 register vec_u8 alpha,
470 register vec_u8 beta) {
471
472 register vec_u8 mask;
473 register vec_u8 tempmask;
474
475 mask = diff_lt_altivec(p0, q0, alpha);
476 tempmask = diff_lt_altivec(p1, p0, beta);
477 mask = vec_and(mask, tempmask);
478 tempmask = diff_lt_altivec(q1, q0, beta);
479 mask = vec_and(mask, tempmask);
480
481 return mask;
482 }
483
484 // out: newp1 = clip((p2 + ((p0 + q0 + 1) >> 1)) >> 1, p1-tc0, p1+tc0)
485 static inline vec_u8 h264_deblock_q1(register vec_u8 p0,
486 register vec_u8 p1,
487 register vec_u8 p2,
488 register vec_u8 q0,
489 register vec_u8 tc0) {
490
491 register vec_u8 average = vec_avg(p0, q0);
492 register vec_u8 temp;
493 register vec_u8 uncliped;
494 register vec_u8 ones;
495 register vec_u8 max;
496 register vec_u8 min;
497 register vec_u8 newp1;
498
499 temp = vec_xor(average, p2);
500 average = vec_avg(average, p2); /*avg(p2, avg(p0, q0)) */
501 ones = vec_splat_u8(1);
502 temp = vec_and(temp, ones); /*(p2^avg(p0, q0)) & 1 */
503 uncliped = vec_subs(average, temp); /*(p2+((p0+q0+1)>>1))>>1 */
504 max = vec_adds(p1, tc0);
505 min = vec_subs(p1, tc0);
506 newp1 = vec_max(min, uncliped);
507 newp1 = vec_min(max, newp1);
508 return newp1;
509 }
510
511 #define h264_deblock_p0_q0(p0, p1, q0, q1, tc0masked) { \
512 \
513 const vec_u8 A0v = vec_sl(vec_splat_u8(10), vec_splat_u8(4)); \
514 \
515 register vec_u8 pq0bit = vec_xor(p0,q0); \
516 register vec_u8 q1minus; \
517 register vec_u8 p0minus; \
518 register vec_u8 stage1; \
519 register vec_u8 stage2; \
520 register vec_u8 vec160; \
521 register vec_u8 delta; \
522 register vec_u8 deltaneg; \
523 \
524 q1minus = vec_nor(q1, q1); /* 255 - q1 */ \
525 stage1 = vec_avg(p1, q1minus); /* (p1 - q1 + 256)>>1 */ \
526 stage2 = vec_sr(stage1, vec_splat_u8(1)); /* (p1 - q1 + 256)>>2 = 64 + (p1 - q1) >> 2 */ \
527 p0minus = vec_nor(p0, p0); /* 255 - p0 */ \
528 stage1 = vec_avg(q0, p0minus); /* (q0 - p0 + 256)>>1 */ \
529 pq0bit = vec_and(pq0bit, vec_splat_u8(1)); \
530 stage2 = vec_avg(stage2, pq0bit); /* 32 + ((q0 - p0)&1 + (p1 - q1) >> 2 + 1) >> 1 */ \
531 stage2 = vec_adds(stage2, stage1); /* 160 + ((p0 - q0) + (p1 - q1) >> 2 + 1) >> 1 */ \
532 vec160 = vec_ld(0, &A0v); \
533 deltaneg = vec_subs(vec160, stage2); /* -d */ \
534 delta = vec_subs(stage2, vec160); /* d */ \
535 deltaneg = vec_min(tc0masked, deltaneg); \
536 delta = vec_min(tc0masked, delta); \
537 p0 = vec_subs(p0, deltaneg); \
538 q0 = vec_subs(q0, delta); \
539 p0 = vec_adds(p0, delta); \
540 q0 = vec_adds(q0, deltaneg); \
541 }
542
543 #define h264_loop_filter_luma_altivec(p2, p1, p0, q0, q1, q2, alpha, beta, tc0) { \
544 DECLARE_ALIGNED(16, unsigned char, temp)[16]; \
545 register vec_u8 alphavec; \
546 register vec_u8 betavec; \
547 register vec_u8 mask; \
548 register vec_u8 p1mask; \
549 register vec_u8 q1mask; \
550 register vector signed char tc0vec; \
551 register vec_u8 finaltc0; \
552 register vec_u8 tc0masked; \
553 register vec_u8 newp1; \
554 register vec_u8 newq1; \
555 \
556 temp[0] = alpha; \
557 temp[1] = beta; \
558 alphavec = vec_ld(0, temp); \
559 betavec = vec_splat(alphavec, 0x1); \
560 alphavec = vec_splat(alphavec, 0x0); \
561 mask = h264_deblock_mask(p0, p1, q0, q1, alphavec, betavec); /*if in block */ \
562 \
563 AV_COPY32(temp, tc0); \
564 tc0vec = vec_ld(0, (signed char*)temp); \
565 tc0vec = vec_mergeh(tc0vec, tc0vec); \
566 tc0vec = vec_mergeh(tc0vec, tc0vec); \
567 mask = vec_and(mask, vec_cmpgt(tc0vec, vec_splat_s8(-1))); /* if tc0[i] >= 0 */ \
568 finaltc0 = vec_and((vec_u8)tc0vec, mask); /* tc = tc0 */ \
569 \
570 p1mask = diff_lt_altivec(p2, p0, betavec); \
571 p1mask = vec_and(p1mask, mask); /* if ( |p2 - p0| < beta) */ \
572 tc0masked = vec_and(p1mask, (vec_u8)tc0vec); \
573 finaltc0 = vec_sub(finaltc0, p1mask); /* tc++ */ \
574 newp1 = h264_deblock_q1(p0, p1, p2, q0, tc0masked); \
575 /*end if*/ \
576 \
577 q1mask = diff_lt_altivec(q2, q0, betavec); \
578 q1mask = vec_and(q1mask, mask); /* if ( |q2 - q0| < beta ) */\
579 tc0masked = vec_and(q1mask, (vec_u8)tc0vec); \
580 finaltc0 = vec_sub(finaltc0, q1mask); /* tc++ */ \
581 newq1 = h264_deblock_q1(p0, q1, q2, q0, tc0masked); \
582 /*end if*/ \
583 \
584 h264_deblock_p0_q0(p0, p1, q0, q1, finaltc0); \
585 p1 = newp1; \
586 q1 = newq1; \
587 }
588
589 static void h264_v_loop_filter_luma_altivec(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0) {
590
591 if ((tc0[0] & tc0[1] & tc0[2] & tc0[3]) >= 0) {
592 register vec_u8 p2 = vec_ld(-3*stride, pix);
593 register vec_u8 p1 = vec_ld(-2*stride, pix);
594 register vec_u8 p0 = vec_ld(-1*stride, pix);
595 register vec_u8 q0 = vec_ld(0, pix);
596 register vec_u8 q1 = vec_ld(stride, pix);
597 register vec_u8 q2 = vec_ld(2*stride, pix);
598 h264_loop_filter_luma_altivec(p2, p1, p0, q0, q1, q2, alpha, beta, tc0);
599 vec_st(p1, -2*stride, pix);
600 vec_st(p0, -1*stride, pix);
601 vec_st(q0, 0, pix);
602 vec_st(q1, stride, pix);
603 }
604 }
605
606 static void h264_h_loop_filter_luma_altivec(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0) {
607
608 register vec_u8 line0, line1, line2, line3, line4, line5;
609 if ((tc0[0] & tc0[1] & tc0[2] & tc0[3]) < 0)
610 return;
611 readAndTranspose16x6(pix-3, stride, line0, line1, line2, line3, line4, line5);
612 h264_loop_filter_luma_altivec(line0, line1, line2, line3, line4, line5, alpha, beta, tc0);
613 transpose4x16(line1, line2, line3, line4);
614 write16x4(pix-2, stride, line1, line2, line3, line4);
615 }
616
617 static av_always_inline
618 void weight_h264_W_altivec(uint8_t *block, int stride, int height,
619 int log2_denom, int weight, int offset, int w)
620 {
621 int y, aligned;
622 vec_u8 vblock;
623 vec_s16 vtemp, vweight, voffset, v0, v1;
624 vec_u16 vlog2_denom;
625 DECLARE_ALIGNED(16, int32_t, temp)[4];
626 LOAD_ZERO;
627
628 offset <<= log2_denom;
629 if(log2_denom) offset += 1<<(log2_denom-1);
630 temp[0] = log2_denom;
631 temp[1] = weight;
632 temp[2] = offset;
633
634 vtemp = (vec_s16)vec_ld(0, temp);
635 vlog2_denom = (vec_u16)vec_splat(vtemp, 1);
636 vweight = vec_splat(vtemp, 3);
637 voffset = vec_splat(vtemp, 5);
638 aligned = !((unsigned long)block & 0xf);
639
640 for (y = 0; y < height; y++) {
641 vblock = vec_ld(0, block);
642
643 v0 = (vec_s16)vec_mergeh(zero_u8v, vblock);
644 v1 = (vec_s16)vec_mergel(zero_u8v, vblock);
645
646 if (w == 16 || aligned) {
647 v0 = vec_mladd(v0, vweight, zero_s16v);
648 v0 = vec_adds(v0, voffset);
649 v0 = vec_sra(v0, vlog2_denom);
650 }
651 if (w == 16 || !aligned) {
652 v1 = vec_mladd(v1, vweight, zero_s16v);
653 v1 = vec_adds(v1, voffset);
654 v1 = vec_sra(v1, vlog2_denom);
655 }
656 vblock = vec_packsu(v0, v1);
657 vec_st(vblock, 0, block);
658
659 block += stride;
660 }
661 }
662
663 static av_always_inline
664 void biweight_h264_W_altivec(uint8_t *dst, uint8_t *src, int stride, int height,
665 int log2_denom, int weightd, int weights, int offset, int w)
666 {
667 int y, dst_aligned, src_aligned;
668 vec_u8 vsrc, vdst;
669 vec_s16 vtemp, vweights, vweightd, voffset, v0, v1, v2, v3;
670 vec_u16 vlog2_denom;
671 DECLARE_ALIGNED(16, int32_t, temp)[4];
672 LOAD_ZERO;
673
674 offset = ((offset + 1) | 1) << log2_denom;
675 temp[0] = log2_denom+1;
676 temp[1] = weights;
677 temp[2] = weightd;
678 temp[3] = offset;
679
680 vtemp = (vec_s16)vec_ld(0, temp);
681 vlog2_denom = (vec_u16)vec_splat(vtemp, 1);
682 vweights = vec_splat(vtemp, 3);
683 vweightd = vec_splat(vtemp, 5);
684 voffset = vec_splat(vtemp, 7);
685 dst_aligned = !((unsigned long)dst & 0xf);
686 src_aligned = !((unsigned long)src & 0xf);
687
688 for (y = 0; y < height; y++) {
689 vdst = vec_ld(0, dst);
690 vsrc = vec_ld(0, src);
691
692 v0 = (vec_s16)vec_mergeh(zero_u8v, vdst);
693 v1 = (vec_s16)vec_mergel(zero_u8v, vdst);
694 v2 = (vec_s16)vec_mergeh(zero_u8v, vsrc);
695 v3 = (vec_s16)vec_mergel(zero_u8v, vsrc);
696
697 if (w == 8) {
698 if (src_aligned)
699 v3 = v2;
700 else
701 v2 = v3;
702 }
703
704 if (w == 16 || dst_aligned) {
705 v0 = vec_mladd(v0, vweightd, zero_s16v);
706 v2 = vec_mladd(v2, vweights, zero_s16v);
707
708 v0 = vec_adds(v0, voffset);
709 v0 = vec_adds(v0, v2);
710 v0 = vec_sra(v0, vlog2_denom);
711 }
712 if (w == 16 || !dst_aligned) {
713 v1 = vec_mladd(v1, vweightd, zero_s16v);
714 v3 = vec_mladd(v3, vweights, zero_s16v);
715
716 v1 = vec_adds(v1, voffset);
717 v1 = vec_adds(v1, v3);
718 v1 = vec_sra(v1, vlog2_denom);
719 }
720 vdst = vec_packsu(v0, v1);
721 vec_st(vdst, 0, dst);
722
723 dst += stride;
724 src += stride;
725 }
726 }
727
728 #define H264_WEIGHT(W) \
729 static void weight_h264_pixels ## W ## _altivec(uint8_t *block, int stride, int height, \
730 int log2_denom, int weight, int offset) \
731 { \
732 weight_h264_W_altivec(block, stride, height, log2_denom, weight, offset, W); \
733 }\
734 static void biweight_h264_pixels ## W ## _altivec(uint8_t *dst, uint8_t *src, int stride, int height, \
735 int log2_denom, int weightd, int weights, int offset) \
736 { \
737 biweight_h264_W_altivec(dst, src, stride, height, log2_denom, weightd, weights, offset, W); \
738 }
739
740 H264_WEIGHT(16)
741 H264_WEIGHT( 8)
742 #endif /* HAVE_ALTIVEC */
743
744 av_cold void ff_h264dsp_init_ppc(H264DSPContext *c, const int bit_depth,
745 const int chroma_format_idc)
746 {
747 #if HAVE_ALTIVEC
748 if (!(av_get_cpu_flags() & AV_CPU_FLAG_ALTIVEC))
749 return;
750
751 if (bit_depth == 8) {
752 c->h264_idct_add = h264_idct_add_altivec;
753 if (chroma_format_idc <= 1)
754 c->h264_idct_add8 = h264_idct_add8_altivec;
755 c->h264_idct_add16 = h264_idct_add16_altivec;
756 c->h264_idct_add16intra = h264_idct_add16intra_altivec;
757 c->h264_idct_dc_add= h264_idct_dc_add_altivec;
758 c->h264_idct8_dc_add = h264_idct8_dc_add_altivec;
759 c->h264_idct8_add = h264_idct8_add_altivec;
760 c->h264_idct8_add4 = h264_idct8_add4_altivec;
761 c->h264_v_loop_filter_luma= h264_v_loop_filter_luma_altivec;
762 c->h264_h_loop_filter_luma= h264_h_loop_filter_luma_altivec;
763
764 c->weight_h264_pixels_tab[0] = weight_h264_pixels16_altivec;
765 c->weight_h264_pixels_tab[1] = weight_h264_pixels8_altivec;
766 c->biweight_h264_pixels_tab[0] = biweight_h264_pixels16_altivec;
767 c->biweight_h264_pixels_tab[1] = biweight_h264_pixels8_altivec;
768 }
769 #endif /* HAVE_ALTIVEC */
770 }
Libav master