2 * Copyright (c) 2004 Romain Dolbeau <romain@dolbeau.org>
4 * This file is part of FFmpeg.
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.
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.
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
23 #include "gcc_fixes.h"
25 #include "dsputil_altivec.h"
26 #include "types_altivec.h"
28 #define PUT_OP_U8_ALTIVEC(d, s, dst) d = s
29 #define AVG_OP_U8_ALTIVEC(d, s, dst) d = vec_avg(dst, s)
31 #define OP_U8_ALTIVEC PUT_OP_U8_ALTIVEC
32 #define PREFIX_h264_chroma_mc8_altivec put_h264_chroma_mc8_altivec
33 #define PREFIX_h264_chroma_mc8_num altivec_put_h264_chroma_mc8_num
34 #define PREFIX_h264_qpel16_h_lowpass_altivec put_h264_qpel16_h_lowpass_altivec
35 #define PREFIX_h264_qpel16_h_lowpass_num altivec_put_h264_qpel16_h_lowpass_num
36 #define PREFIX_h264_qpel16_v_lowpass_altivec put_h264_qpel16_v_lowpass_altivec
37 #define PREFIX_h264_qpel16_v_lowpass_num altivec_put_h264_qpel16_v_lowpass_num
38 #define PREFIX_h264_qpel16_hv_lowpass_altivec put_h264_qpel16_hv_lowpass_altivec
39 #define PREFIX_h264_qpel16_hv_lowpass_num altivec_put_h264_qpel16_hv_lowpass_num
40 #include "h264_template_altivec.c"
42 #undef PREFIX_h264_chroma_mc8_altivec
43 #undef PREFIX_h264_chroma_mc8_num
44 #undef PREFIX_h264_qpel16_h_lowpass_altivec
45 #undef PREFIX_h264_qpel16_h_lowpass_num
46 #undef PREFIX_h264_qpel16_v_lowpass_altivec
47 #undef PREFIX_h264_qpel16_v_lowpass_num
48 #undef PREFIX_h264_qpel16_hv_lowpass_altivec
49 #undef PREFIX_h264_qpel16_hv_lowpass_num
51 #define OP_U8_ALTIVEC AVG_OP_U8_ALTIVEC
52 #define PREFIX_h264_chroma_mc8_altivec avg_h264_chroma_mc8_altivec
53 #define PREFIX_h264_chroma_mc8_num altivec_avg_h264_chroma_mc8_num
54 #define PREFIX_h264_qpel16_h_lowpass_altivec avg_h264_qpel16_h_lowpass_altivec
55 #define PREFIX_h264_qpel16_h_lowpass_num altivec_avg_h264_qpel16_h_lowpass_num
56 #define PREFIX_h264_qpel16_v_lowpass_altivec avg_h264_qpel16_v_lowpass_altivec
57 #define PREFIX_h264_qpel16_v_lowpass_num altivec_avg_h264_qpel16_v_lowpass_num
58 #define PREFIX_h264_qpel16_hv_lowpass_altivec avg_h264_qpel16_hv_lowpass_altivec
59 #define PREFIX_h264_qpel16_hv_lowpass_num altivec_avg_h264_qpel16_hv_lowpass_num
60 #include "h264_template_altivec.c"
62 #undef PREFIX_h264_chroma_mc8_altivec
63 #undef PREFIX_h264_chroma_mc8_num
64 #undef PREFIX_h264_qpel16_h_lowpass_altivec
65 #undef PREFIX_h264_qpel16_h_lowpass_num
66 #undef PREFIX_h264_qpel16_v_lowpass_altivec
67 #undef PREFIX_h264_qpel16_v_lowpass_num
68 #undef PREFIX_h264_qpel16_hv_lowpass_altivec
69 #undef PREFIX_h264_qpel16_hv_lowpass_num
71 #define H264_MC(OPNAME, SIZE, CODETYPE) \
72 static void OPNAME ## h264_qpel ## SIZE ## _mc00_ ## CODETYPE (uint8_t *dst, uint8_t *src, int stride){\
73 OPNAME ## pixels ## SIZE ## _ ## CODETYPE(dst, src, stride, SIZE);\
76 static void OPNAME ## h264_qpel ## SIZE ## _mc10_ ## CODETYPE(uint8_t *dst, uint8_t *src, int stride){ \
77 DECLARE_ALIGNED_16(uint8_t, half[SIZE*SIZE]);\
78 put_h264_qpel ## SIZE ## _h_lowpass_ ## CODETYPE(half, src, SIZE, stride);\
79 OPNAME ## pixels ## SIZE ## _l2_ ## CODETYPE(dst, src, half, stride, stride, SIZE);\
82 static void OPNAME ## h264_qpel ## SIZE ## _mc20_ ## CODETYPE(uint8_t *dst, uint8_t *src, int stride){\
83 OPNAME ## h264_qpel ## SIZE ## _h_lowpass_ ## CODETYPE(dst, src, stride, stride);\
86 static void OPNAME ## h264_qpel ## SIZE ## _mc30_ ## CODETYPE(uint8_t *dst, uint8_t *src, int stride){\
87 DECLARE_ALIGNED_16(uint8_t, half[SIZE*SIZE]);\
88 put_h264_qpel ## SIZE ## _h_lowpass_ ## CODETYPE(half, src, SIZE, stride);\
89 OPNAME ## pixels ## SIZE ## _l2_ ## CODETYPE(dst, src+1, half, stride, stride, SIZE);\
92 static void OPNAME ## h264_qpel ## SIZE ## _mc01_ ## CODETYPE(uint8_t *dst, uint8_t *src, int stride){\
93 DECLARE_ALIGNED_16(uint8_t, half[SIZE*SIZE]);\
94 put_h264_qpel ## SIZE ## _v_lowpass_ ## CODETYPE(half, src, SIZE, stride);\
95 OPNAME ## pixels ## SIZE ## _l2_ ## CODETYPE(dst, src, half, stride, stride, SIZE);\
98 static void OPNAME ## h264_qpel ## SIZE ## _mc02_ ## CODETYPE(uint8_t *dst, uint8_t *src, int stride){\
99 OPNAME ## h264_qpel ## SIZE ## _v_lowpass_ ## CODETYPE(dst, src, stride, stride);\
102 static void OPNAME ## h264_qpel ## SIZE ## _mc03_ ## CODETYPE(uint8_t *dst, uint8_t *src, int stride){\
103 DECLARE_ALIGNED_16(uint8_t, half[SIZE*SIZE]);\
104 put_h264_qpel ## SIZE ## _v_lowpass_ ## CODETYPE(half, src, SIZE, stride);\
105 OPNAME ## pixels ## SIZE ## _l2_ ## CODETYPE(dst, src+stride, half, stride, stride, SIZE);\
108 static void OPNAME ## h264_qpel ## SIZE ## _mc11_ ## CODETYPE(uint8_t *dst, uint8_t *src, int stride){\
109 DECLARE_ALIGNED_16(uint8_t, halfH[SIZE*SIZE]);\
110 DECLARE_ALIGNED_16(uint8_t, halfV[SIZE*SIZE]);\
111 put_h264_qpel ## SIZE ## _h_lowpass_ ## CODETYPE(halfH, src, SIZE, stride);\
112 put_h264_qpel ## SIZE ## _v_lowpass_ ## CODETYPE(halfV, src, SIZE, stride);\
113 OPNAME ## pixels ## SIZE ## _l2_ ## CODETYPE(dst, halfH, halfV, stride, SIZE, SIZE);\
116 static void OPNAME ## h264_qpel ## SIZE ## _mc31_ ## CODETYPE(uint8_t *dst, uint8_t *src, int stride){\
117 DECLARE_ALIGNED_16(uint8_t, halfH[SIZE*SIZE]);\
118 DECLARE_ALIGNED_16(uint8_t, halfV[SIZE*SIZE]);\
119 put_h264_qpel ## SIZE ## _h_lowpass_ ## CODETYPE(halfH, src, SIZE, stride);\
120 put_h264_qpel ## SIZE ## _v_lowpass_ ## CODETYPE(halfV, src+1, SIZE, stride);\
121 OPNAME ## pixels ## SIZE ## _l2_ ## CODETYPE(dst, halfH, halfV, stride, SIZE, SIZE);\
124 static void OPNAME ## h264_qpel ## SIZE ## _mc13_ ## CODETYPE(uint8_t *dst, uint8_t *src, int stride){\
125 DECLARE_ALIGNED_16(uint8_t, halfH[SIZE*SIZE]);\
126 DECLARE_ALIGNED_16(uint8_t, halfV[SIZE*SIZE]);\
127 put_h264_qpel ## SIZE ## _h_lowpass_ ## CODETYPE(halfH, src + stride, SIZE, stride);\
128 put_h264_qpel ## SIZE ## _v_lowpass_ ## CODETYPE(halfV, src, SIZE, stride);\
129 OPNAME ## pixels ## SIZE ## _l2_ ## CODETYPE(dst, halfH, halfV, stride, SIZE, SIZE);\
132 static void OPNAME ## h264_qpel ## SIZE ## _mc33_ ## CODETYPE(uint8_t *dst, uint8_t *src, int stride){\
133 DECLARE_ALIGNED_16(uint8_t, halfH[SIZE*SIZE]);\
134 DECLARE_ALIGNED_16(uint8_t, halfV[SIZE*SIZE]);\
135 put_h264_qpel ## SIZE ## _h_lowpass_ ## CODETYPE(halfH, src + stride, SIZE, stride);\
136 put_h264_qpel ## SIZE ## _v_lowpass_ ## CODETYPE(halfV, src+1, SIZE, stride);\
137 OPNAME ## pixels ## SIZE ## _l2_ ## CODETYPE(dst, halfH, halfV, stride, SIZE, SIZE);\
140 static void OPNAME ## h264_qpel ## SIZE ## _mc22_ ## CODETYPE(uint8_t *dst, uint8_t *src, int stride){\
141 DECLARE_ALIGNED_16(int16_t, tmp[SIZE*(SIZE+8)]);\
142 OPNAME ## h264_qpel ## SIZE ## _hv_lowpass_ ## CODETYPE(dst, tmp, src, stride, SIZE, stride);\
145 static void OPNAME ## h264_qpel ## SIZE ## _mc21_ ## CODETYPE(uint8_t *dst, uint8_t *src, int stride){\
146 DECLARE_ALIGNED_16(uint8_t, halfH[SIZE*SIZE]);\
147 DECLARE_ALIGNED_16(uint8_t, halfHV[SIZE*SIZE]);\
148 DECLARE_ALIGNED_16(int16_t, tmp[SIZE*(SIZE+8)]);\
149 put_h264_qpel ## SIZE ## _h_lowpass_ ## CODETYPE(halfH, src, SIZE, stride);\
150 put_h264_qpel ## SIZE ## _hv_lowpass_ ## CODETYPE(halfHV, tmp, src, SIZE, SIZE, stride);\
151 OPNAME ## pixels ## SIZE ## _l2_ ## CODETYPE(dst, halfH, halfHV, stride, SIZE, SIZE);\
154 static void OPNAME ## h264_qpel ## SIZE ## _mc23_ ## CODETYPE(uint8_t *dst, uint8_t *src, int stride){\
155 DECLARE_ALIGNED_16(uint8_t, halfH[SIZE*SIZE]);\
156 DECLARE_ALIGNED_16(uint8_t, halfHV[SIZE*SIZE]);\
157 DECLARE_ALIGNED_16(int16_t, tmp[SIZE*(SIZE+8)]);\
158 put_h264_qpel ## SIZE ## _h_lowpass_ ## CODETYPE(halfH, src + stride, SIZE, stride);\
159 put_h264_qpel ## SIZE ## _hv_lowpass_ ## CODETYPE(halfHV, tmp, src, SIZE, SIZE, stride);\
160 OPNAME ## pixels ## SIZE ## _l2_ ## CODETYPE(dst, halfH, halfHV, stride, SIZE, SIZE);\
163 static void OPNAME ## h264_qpel ## SIZE ## _mc12_ ## CODETYPE(uint8_t *dst, uint8_t *src, int stride){\
164 DECLARE_ALIGNED_16(uint8_t, halfV[SIZE*SIZE]);\
165 DECLARE_ALIGNED_16(uint8_t, halfHV[SIZE*SIZE]);\
166 DECLARE_ALIGNED_16(int16_t, tmp[SIZE*(SIZE+8)]);\
167 put_h264_qpel ## SIZE ## _v_lowpass_ ## CODETYPE(halfV, src, SIZE, stride);\
168 put_h264_qpel ## SIZE ## _hv_lowpass_ ## CODETYPE(halfHV, tmp, src, SIZE, SIZE, stride);\
169 OPNAME ## pixels ## SIZE ## _l2_ ## CODETYPE(dst, halfV, halfHV, stride, SIZE, SIZE);\
172 static void OPNAME ## h264_qpel ## SIZE ## _mc32_ ## CODETYPE(uint8_t *dst, uint8_t *src, int stride){\
173 DECLARE_ALIGNED_16(uint8_t, halfV[SIZE*SIZE]);\
174 DECLARE_ALIGNED_16(uint8_t, halfHV[SIZE*SIZE]);\
175 DECLARE_ALIGNED_16(int16_t, tmp[SIZE*(SIZE+8)]);\
176 put_h264_qpel ## SIZE ## _v_lowpass_ ## CODETYPE(halfV, src+1, SIZE, stride);\
177 put_h264_qpel ## SIZE ## _hv_lowpass_ ## CODETYPE(halfHV, tmp, src, SIZE, SIZE, stride);\
178 OPNAME ## pixels ## SIZE ## _l2_ ## CODETYPE(dst, halfV, halfHV, stride, SIZE, SIZE);\
181 /* this code assume that stride % 16 == 0 */
182 void put_no_rnd_h264_chroma_mc8_altivec(uint8_t * dst, uint8_t * src, int stride, int h, int x, int y) {
183 DECLARE_ALIGNED_16(signed int, ABCD[4]) =
184 {((8 - x) * (8 - y)),
189 vector unsigned char fperm;
190 const vector signed int vABCD = vec_ld(0, ABCD);
191 const vector signed short vA = vec_splat((vector signed short)vABCD, 1);
192 const vector signed short vB = vec_splat((vector signed short)vABCD, 3);
193 const vector signed short vC = vec_splat((vector signed short)vABCD, 5);
194 const vector signed short vD = vec_splat((vector signed short)vABCD, 7);
195 const vector signed int vzero = vec_splat_s32(0);
196 const vector signed short v28ss = vec_sub(vec_sl(vec_splat_s16(1),vec_splat_u16(5)),vec_splat_s16(4));
197 const vector unsigned short v6us = vec_splat_u16(6);
198 register int loadSecond = (((unsigned long)src) % 16) <= 7 ? 0 : 1;
199 register int reallyBadAlign = (((unsigned long)src) % 16) == 15 ? 1 : 0;
201 vector unsigned char vsrcAuc, vsrcBuc, vsrcperm0, vsrcperm1;
202 vector unsigned char vsrc0uc, vsrc1uc;
203 vector signed short vsrc0ssH, vsrc1ssH;
204 vector unsigned char vsrcCuc, vsrc2uc, vsrc3uc;
205 vector signed short vsrc2ssH, vsrc3ssH, psum;
206 vector unsigned char vdst, ppsum, fsum;
208 if (((unsigned long)dst) % 16 == 0) {
209 fperm = (vector unsigned char)AVV(0x10, 0x11, 0x12, 0x13,
210 0x14, 0x15, 0x16, 0x17,
211 0x08, 0x09, 0x0A, 0x0B,
212 0x0C, 0x0D, 0x0E, 0x0F);
214 fperm = (vector unsigned char)AVV(0x00, 0x01, 0x02, 0x03,
215 0x04, 0x05, 0x06, 0x07,
216 0x18, 0x19, 0x1A, 0x1B,
217 0x1C, 0x1D, 0x1E, 0x1F);
220 vsrcAuc = vec_ld(0, src);
223 vsrcBuc = vec_ld(16, src);
224 vsrcperm0 = vec_lvsl(0, src);
225 vsrcperm1 = vec_lvsl(1, src);
227 vsrc0uc = vec_perm(vsrcAuc, vsrcBuc, vsrcperm0);
231 vsrc1uc = vec_perm(vsrcAuc, vsrcBuc, vsrcperm1);
233 vsrc0ssH = (vector signed short)vec_mergeh((vector unsigned char)vzero,
234 (vector unsigned char)vsrc0uc);
235 vsrc1ssH = (vector signed short)vec_mergeh((vector unsigned char)vzero,
236 (vector unsigned char)vsrc1uc);
238 if (!loadSecond) {// -> !reallyBadAlign
239 for (i = 0 ; i < h ; i++) {
242 vsrcCuc = vec_ld(stride + 0, src);
244 vsrc2uc = vec_perm(vsrcCuc, vsrcCuc, vsrcperm0);
245 vsrc3uc = vec_perm(vsrcCuc, vsrcCuc, vsrcperm1);
247 vsrc2ssH = (vector signed short)vec_mergeh((vector unsigned char)vzero,
248 (vector unsigned char)vsrc2uc);
249 vsrc3ssH = (vector signed short)vec_mergeh((vector unsigned char)vzero,
250 (vector unsigned char)vsrc3uc);
252 psum = vec_mladd(vA, vsrc0ssH, vec_splat_s16(0));
253 psum = vec_mladd(vB, vsrc1ssH, psum);
254 psum = vec_mladd(vC, vsrc2ssH, psum);
255 psum = vec_mladd(vD, vsrc3ssH, psum);
256 psum = vec_add(v28ss, psum);
257 psum = vec_sra(psum, v6us);
259 vdst = vec_ld(0, dst);
260 ppsum = (vector unsigned char)vec_packsu(psum, psum);
261 fsum = vec_perm(vdst, ppsum, fperm);
263 vec_st(fsum, 0, dst);
272 vector unsigned char vsrcDuc;
273 for (i = 0 ; i < h ; i++) {
274 vsrcCuc = vec_ld(stride + 0, src);
275 vsrcDuc = vec_ld(stride + 16, src);
277 vsrc2uc = vec_perm(vsrcCuc, vsrcDuc, vsrcperm0);
281 vsrc3uc = vec_perm(vsrcCuc, vsrcDuc, vsrcperm1);
283 vsrc2ssH = (vector signed short)vec_mergeh((vector unsigned char)vzero,
284 (vector unsigned char)vsrc2uc);
285 vsrc3ssH = (vector signed short)vec_mergeh((vector unsigned char)vzero,
286 (vector unsigned char)vsrc3uc);
288 psum = vec_mladd(vA, vsrc0ssH, vec_splat_s16(0));
289 psum = vec_mladd(vB, vsrc1ssH, psum);
290 psum = vec_mladd(vC, vsrc2ssH, psum);
291 psum = vec_mladd(vD, vsrc3ssH, psum);
292 psum = vec_add(v28ss, psum);
293 psum = vec_sr(psum, v6us);
295 vdst = vec_ld(0, dst);
296 ppsum = (vector unsigned char)vec_pack(psum, psum);
297 fsum = vec_perm(vdst, ppsum, fperm);
299 vec_st(fsum, 0, dst);
310 static inline void put_pixels16_l2_altivec( uint8_t * dst, const uint8_t * src1,
311 const uint8_t * src2, int dst_stride,
312 int src_stride1, int h)
315 vector unsigned char a, b, d, tmp1, tmp2, mask, mask_, edges, align;
317 mask_ = vec_lvsl(0, src2);
319 for (i = 0; i < h; i++) {
321 tmp1 = vec_ld(i * src_stride1, src1);
322 mask = vec_lvsl(i * src_stride1, src1);
323 tmp2 = vec_ld(i * src_stride1 + 15, src1);
325 a = vec_perm(tmp1, tmp2, mask);
327 tmp1 = vec_ld(i * 16, src2);
328 tmp2 = vec_ld(i * 16 + 15, src2);
330 b = vec_perm(tmp1, tmp2, mask_);
332 tmp1 = vec_ld(0, dst);
333 mask = vec_lvsl(0, dst);
334 tmp2 = vec_ld(15, dst);
338 edges = vec_perm(tmp2, tmp1, mask);
340 align = vec_lvsr(0, dst);
342 tmp2 = vec_perm(d, edges, align);
343 tmp1 = vec_perm(edges, d, align);
345 vec_st(tmp2, 15, dst);
346 vec_st(tmp1, 0 , dst);
352 static inline void avg_pixels16_l2_altivec( uint8_t * dst, const uint8_t * src1,
353 const uint8_t * src2, int dst_stride,
354 int src_stride1, int h)
357 vector unsigned char a, b, d, tmp1, tmp2, mask, mask_, edges, align;
359 mask_ = vec_lvsl(0, src2);
361 for (i = 0; i < h; i++) {
363 tmp1 = vec_ld(i * src_stride1, src1);
364 mask = vec_lvsl(i * src_stride1, src1);
365 tmp2 = vec_ld(i * src_stride1 + 15, src1);
367 a = vec_perm(tmp1, tmp2, mask);
369 tmp1 = vec_ld(i * 16, src2);
370 tmp2 = vec_ld(i * 16 + 15, src2);
372 b = vec_perm(tmp1, tmp2, mask_);
374 tmp1 = vec_ld(0, dst);
375 mask = vec_lvsl(0, dst);
376 tmp2 = vec_ld(15, dst);
378 d = vec_avg(vec_perm(tmp1, tmp2, mask), vec_avg(a, b));
380 edges = vec_perm(tmp2, tmp1, mask);
382 align = vec_lvsr(0, dst);
384 tmp2 = vec_perm(d, edges, align);
385 tmp1 = vec_perm(edges, d, align);
387 vec_st(tmp2, 15, dst);
388 vec_st(tmp1, 0 , dst);
394 /* Implemented but could be faster
395 #define put_pixels16_l2_altivec(d,s1,s2,ds,s1s,h) put_pixels16_l2(d,s1,s2,ds,s1s,16,h)
396 #define avg_pixels16_l2_altivec(d,s1,s2,ds,s1s,h) avg_pixels16_l2(d,s1,s2,ds,s1s,16,h)
399 H264_MC(put_, 16, altivec)
400 H264_MC(avg_, 16, altivec)
403 /****************************************************************************
405 ****************************************************************************/
407 #define IDCT8_1D_ALTIVEC(s0, s1, s2, s3, s4, s5, s6, s7, d0, d1, d2, d3, d4, d5, d6, d7) {\
408 /* a0 = SRC(0) + SRC(4); */ \
409 vec_s16_t a0v = vec_add(s0, s4); \
410 /* a2 = SRC(0) - SRC(4); */ \
411 vec_s16_t a2v = vec_sub(s0, s4); \
412 /* a4 = (SRC(2)>>1) - SRC(6); */ \
413 vec_s16_t a4v = vec_sub(vec_sra(s2, onev), s6); \
414 /* a6 = (SRC(6)>>1) + SRC(2); */ \
415 vec_s16_t a6v = vec_add(vec_sra(s6, onev), s2); \
416 /* b0 = a0 + a6; */ \
417 vec_s16_t b0v = vec_add(a0v, a6v); \
418 /* b2 = a2 + a4; */ \
419 vec_s16_t b2v = vec_add(a2v, a4v); \
420 /* b4 = a2 - a4; */ \
421 vec_s16_t b4v = vec_sub(a2v, a4v); \
422 /* b6 = a0 - a6; */ \
423 vec_s16_t b6v = vec_sub(a0v, a6v); \
424 /* a1 = SRC(5) - SRC(3) - SRC(7) - (SRC(7)>>1); */ \
425 /* a1 = (SRC(5)-SRC(3)) - (SRC(7) + (SRC(7)>>1)); */ \
426 vec_s16_t a1v = vec_sub( vec_sub(s5, s3), vec_add(s7, vec_sra(s7, onev)) ); \
427 /* a3 = SRC(7) + SRC(1) - SRC(3) - (SRC(3)>>1); */ \
428 /* a3 = (SRC(7)+SRC(1)) - (SRC(3) + (SRC(3)>>1)); */ \
429 vec_s16_t a3v = vec_sub( vec_add(s7, s1), vec_add(s3, vec_sra(s3, onev)) );\
430 /* a5 = SRC(7) - SRC(1) + SRC(5) + (SRC(5)>>1); */ \
431 /* a5 = (SRC(7)-SRC(1)) + SRC(5) + (SRC(5)>>1); */ \
432 vec_s16_t a5v = vec_add( vec_sub(s7, s1), vec_add(s5, vec_sra(s5, onev)) );\
433 /* a7 = SRC(5)+SRC(3) + SRC(1) + (SRC(1)>>1); */ \
434 vec_s16_t a7v = vec_add( vec_add(s5, s3), vec_add(s1, vec_sra(s1, onev)) );\
435 /* b1 = (a7>>2) + a1; */ \
436 vec_s16_t b1v = vec_add( vec_sra(a7v, twov), a1v); \
437 /* b3 = a3 + (a5>>2); */ \
438 vec_s16_t b3v = vec_add(a3v, vec_sra(a5v, twov)); \
439 /* b5 = (a3>>2) - a5; */ \
440 vec_s16_t b5v = vec_sub( vec_sra(a3v, twov), a5v); \
441 /* b7 = a7 - (a1>>2); */ \
442 vec_s16_t b7v = vec_sub( a7v, vec_sra(a1v, twov)); \
443 /* DST(0, b0 + b7); */ \
444 d0 = vec_add(b0v, b7v); \
445 /* DST(1, b2 + b5); */ \
446 d1 = vec_add(b2v, b5v); \
447 /* DST(2, b4 + b3); */ \
448 d2 = vec_add(b4v, b3v); \
449 /* DST(3, b6 + b1); */ \
450 d3 = vec_add(b6v, b1v); \
451 /* DST(4, b6 - b1); */ \
452 d4 = vec_sub(b6v, b1v); \
453 /* DST(5, b4 - b3); */ \
454 d5 = vec_sub(b4v, b3v); \
455 /* DST(6, b2 - b5); */ \
456 d6 = vec_sub(b2v, b5v); \
457 /* DST(7, b0 - b7); */ \
458 d7 = vec_sub(b0v, b7v); \
461 #define ALTIVEC_STORE_SUM_CLIP(dest, idctv, perm_ldv, perm_stv, sel) { \
462 /* unaligned load */ \
463 vec_u8_t hv = vec_ld( 0, dest ); \
464 vec_u8_t lv = vec_ld( 7, dest ); \
465 vec_u8_t dstv = vec_perm( hv, lv, (vec_u8_t)perm_ldv ); \
466 vec_s16_t idct_sh6 = vec_sra(idctv, sixv); \
467 vec_u16_t dst16 = (vec_u16_t)vec_mergeh(zero_u8v, dstv); \
468 vec_s16_t idstsum = vec_adds(idct_sh6, (vec_s16_t)dst16); \
469 vec_u8_t idstsum8 = vec_packsu(zero_s16v, idstsum); \
471 /* unaligned store */ \
472 vec_u8_t bodyv = vec_perm( idstsum8, idstsum8, perm_stv );\
473 vec_u8_t edgelv = vec_perm( sel, zero_u8v, perm_stv ); \
474 lv = vec_sel( lv, bodyv, edgelv ); \
475 vec_st( lv, 7, dest ); \
476 hv = vec_ld( 0, dest ); \
477 edgehv = vec_perm( zero_u8v, sel, perm_stv ); \
478 hv = vec_sel( hv, bodyv, edgehv ); \
479 vec_st( hv, 0, dest ); \
482 void ff_h264_idct8_add_altivec( uint8_t *dst, DCTELEM *dct, int stride ) {
483 vec_s16_t s0, s1, s2, s3, s4, s5, s6, s7;
484 vec_s16_t d0, d1, d2, d3, d4, d5, d6, d7;
485 vec_s16_t idct0, idct1, idct2, idct3, idct4, idct5, idct6, idct7;
487 vec_u8_t perm_ldv = vec_lvsl(0, dst);
488 vec_u8_t perm_stv = vec_lvsr(8, dst);
490 const vec_u16_t onev = vec_splat_u16(1);
491 const vec_u16_t twov = vec_splat_u16(2);
492 const vec_u16_t sixv = vec_splat_u16(6);
494 const vec_u8_t sel = (vec_u8_t) AVV(0,0,0,0,0,0,0,0,
495 -1,-1,-1,-1,-1,-1,-1,-1);
498 dct[0] += 32; // rounding for the >>6 at the end
500 s0 = vec_ld(0x00, (int16_t*)dct);
501 s1 = vec_ld(0x10, (int16_t*)dct);
502 s2 = vec_ld(0x20, (int16_t*)dct);
503 s3 = vec_ld(0x30, (int16_t*)dct);
504 s4 = vec_ld(0x40, (int16_t*)dct);
505 s5 = vec_ld(0x50, (int16_t*)dct);
506 s6 = vec_ld(0x60, (int16_t*)dct);
507 s7 = vec_ld(0x70, (int16_t*)dct);
509 IDCT8_1D_ALTIVEC(s0, s1, s2, s3, s4, s5, s6, s7,
510 d0, d1, d2, d3, d4, d5, d6, d7);
512 TRANSPOSE8( d0, d1, d2, d3, d4, d5, d6, d7 );
514 IDCT8_1D_ALTIVEC(d0, d1, d2, d3, d4, d5, d6, d7,
515 idct0, idct1, idct2, idct3, idct4, idct5, idct6, idct7);
517 ALTIVEC_STORE_SUM_CLIP(&dst[0*stride], idct0, perm_ldv, perm_stv, sel);
518 ALTIVEC_STORE_SUM_CLIP(&dst[1*stride], idct1, perm_ldv, perm_stv, sel);
519 ALTIVEC_STORE_SUM_CLIP(&dst[2*stride], idct2, perm_ldv, perm_stv, sel);
520 ALTIVEC_STORE_SUM_CLIP(&dst[3*stride], idct3, perm_ldv, perm_stv, sel);
521 ALTIVEC_STORE_SUM_CLIP(&dst[4*stride], idct4, perm_ldv, perm_stv, sel);
522 ALTIVEC_STORE_SUM_CLIP(&dst[5*stride], idct5, perm_ldv, perm_stv, sel);
523 ALTIVEC_STORE_SUM_CLIP(&dst[6*stride], idct6, perm_ldv, perm_stv, sel);
524 ALTIVEC_STORE_SUM_CLIP(&dst[7*stride], idct7, perm_ldv, perm_stv, sel);
527 void dsputil_h264_init_ppc(DSPContext* c, AVCodecContext *avctx) {
531 c->put_h264_chroma_pixels_tab[0] = put_h264_chroma_mc8_altivec;
532 c->put_no_rnd_h264_chroma_pixels_tab[0] = put_no_rnd_h264_chroma_mc8_altivec;
533 c->avg_h264_chroma_pixels_tab[0] = avg_h264_chroma_mc8_altivec;
534 c->h264_idct8_add = ff_h264_idct8_add_altivec;
536 #define dspfunc(PFX, IDX, NUM) \
537 c->PFX ## _pixels_tab[IDX][ 0] = PFX ## NUM ## _mc00_altivec; \
538 c->PFX ## _pixels_tab[IDX][ 1] = PFX ## NUM ## _mc10_altivec; \
539 c->PFX ## _pixels_tab[IDX][ 2] = PFX ## NUM ## _mc20_altivec; \
540 c->PFX ## _pixels_tab[IDX][ 3] = PFX ## NUM ## _mc30_altivec; \
541 c->PFX ## _pixels_tab[IDX][ 4] = PFX ## NUM ## _mc01_altivec; \
542 c->PFX ## _pixels_tab[IDX][ 5] = PFX ## NUM ## _mc11_altivec; \
543 c->PFX ## _pixels_tab[IDX][ 6] = PFX ## NUM ## _mc21_altivec; \
544 c->PFX ## _pixels_tab[IDX][ 7] = PFX ## NUM ## _mc31_altivec; \
545 c->PFX ## _pixels_tab[IDX][ 8] = PFX ## NUM ## _mc02_altivec; \
546 c->PFX ## _pixels_tab[IDX][ 9] = PFX ## NUM ## _mc12_altivec; \
547 c->PFX ## _pixels_tab[IDX][10] = PFX ## NUM ## _mc22_altivec; \
548 c->PFX ## _pixels_tab[IDX][11] = PFX ## NUM ## _mc32_altivec; \
549 c->PFX ## _pixels_tab[IDX][12] = PFX ## NUM ## _mc03_altivec; \
550 c->PFX ## _pixels_tab[IDX][13] = PFX ## NUM ## _mc13_altivec; \
551 c->PFX ## _pixels_tab[IDX][14] = PFX ## NUM ## _mc23_altivec; \
552 c->PFX ## _pixels_tab[IDX][15] = PFX ## NUM ## _mc33_altivec
554 dspfunc(put_h264_qpel, 0, 16);
555 dspfunc(avg_h264_qpel, 0, 16);
559 #endif /* HAVE_ALTIVEC */
561 // Non-AltiVec PPC optimisations