Mercurial > cgi-bin > hgwebdir.cgi > PR > Applications > VSs > VSs__H264__App

annotate libavcodec/ppc/h264_altivec.c @ 3:0b056460c67d

find changesets by author, revision, files, or words in the commit message
changed code to use VSs
author Nina Engelhardt <nengel@mailbox.tu-berlin.de>
date Mon, 29 Oct 2012 16:44:27 +0100
parents
children
rev   line source
nengel@2 1 /*
nengel@2 2 * Copyright (c) 2004 Romain Dolbeau <romain@dolbeau.org>
nengel@2 3 *
nengel@2 4 * This file is part of FFmpeg.
nengel@2 5 *
nengel@2 6 * FFmpeg is free software; you can redistribute it and/or
nengel@2 7 * modify it under the terms of the GNU Lesser General Public
nengel@2 8 * License as published by the Free Software Foundation; either
nengel@2 9 * version 2.1 of the License, or (at your option) any later version.
nengel@2 10 *
nengel@2 11 * FFmpeg is distributed in the hope that it will be useful,
nengel@2 12 * but WITHOUT ANY WARRANTY; without even the implied warranty of
nengel@2 13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
nengel@2 14 * Lesser General Public License for more details.
nengel@2 15 *
nengel@2 16 * You should have received a copy of the GNU Lesser General Public
nengel@2 17 * License along with FFmpeg; if not, write to the Free Software
nengel@2 18 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
nengel@2 19 */
nengel@2 20
nengel@2 21 #include "libavcodec/dsputil.h"
nengel@2 22 #include "libavcodec/h264_data.h"
nengel@2 23 #include "libavcodec/h264_dsp.h"
nengel@2 24
nengel@2 25 #include "dsputil_ppc.h"
nengel@2 26 #include "dsputil_altivec.h"
nengel@2 27 #include "util_altivec.h"
nengel@2 28 #include "types_altivec.h"
nengel@2 29
nengel@2 30 #define PUT_OP_U8_ALTIVEC(d, s, dst) d = s
nengel@2 31 #define AVG_OP_U8_ALTIVEC(d, s, dst) d = vec_avg(dst, s)
nengel@2 32
nengel@2 33 #define OP_U8_ALTIVEC PUT_OP_U8_ALTIVEC
nengel@2 34 #define PREFIX_h264_chroma_mc8_altivec put_h264_chroma_mc8_altivec
nengel@2 35 #define PREFIX_no_rnd_vc1_chroma_mc8_altivec put_no_rnd_vc1_chroma_mc8_altivec
nengel@2 36 #define PREFIX_h264_chroma_mc8_num altivec_put_h264_chroma_mc8_num
nengel@2 37 #define PREFIX_h264_qpel16_h_lowpass_altivec put_h264_qpel16_h_lowpass_altivec
nengel@2 38 #define PREFIX_h264_qpel16_h_lowpass_num altivec_put_h264_qpel16_h_lowpass_num
nengel@2 39 #define PREFIX_h264_qpel16_v_lowpass_altivec put_h264_qpel16_v_lowpass_altivec
nengel@2 40 #define PREFIX_h264_qpel16_v_lowpass_num altivec_put_h264_qpel16_v_lowpass_num
nengel@2 41 #define PREFIX_h264_qpel16_hv_lowpass_altivec put_h264_qpel16_hv_lowpass_altivec
nengel@2 42 #define PREFIX_h264_qpel16_hv_lowpass_num altivec_put_h264_qpel16_hv_lowpass_num
nengel@2 43 #include "h264_template_altivec.c"
nengel@2 44 #undef OP_U8_ALTIVEC
nengel@2 45 #undef PREFIX_h264_chroma_mc8_altivec
nengel@2 46 #undef PREFIX_no_rnd_vc1_chroma_mc8_altivec
nengel@2 47 #undef PREFIX_h264_chroma_mc8_num
nengel@2 48 #undef PREFIX_h264_qpel16_h_lowpass_altivec
nengel@2 49 #undef PREFIX_h264_qpel16_h_lowpass_num
nengel@2 50 #undef PREFIX_h264_qpel16_v_lowpass_altivec
nengel@2 51 #undef PREFIX_h264_qpel16_v_lowpass_num
nengel@2 52 #undef PREFIX_h264_qpel16_hv_lowpass_altivec
nengel@2 53 #undef PREFIX_h264_qpel16_hv_lowpass_num
nengel@2 54
nengel@2 55 #define OP_U8_ALTIVEC AVG_OP_U8_ALTIVEC
nengel@2 56 #define PREFIX_h264_chroma_mc8_altivec avg_h264_chroma_mc8_altivec
nengel@2 57 #define PREFIX_no_rnd_vc1_chroma_mc8_altivec avg_no_rnd_vc1_chroma_mc8_altivec
nengel@2 58 #define PREFIX_h264_chroma_mc8_num altivec_avg_h264_chroma_mc8_num
nengel@2 59 #define PREFIX_h264_qpel16_h_lowpass_altivec avg_h264_qpel16_h_lowpass_altivec
nengel@2 60 #define PREFIX_h264_qpel16_h_lowpass_num altivec_avg_h264_qpel16_h_lowpass_num
nengel@2 61 #define PREFIX_h264_qpel16_v_lowpass_altivec avg_h264_qpel16_v_lowpass_altivec
nengel@2 62 #define PREFIX_h264_qpel16_v_lowpass_num altivec_avg_h264_qpel16_v_lowpass_num
nengel@2 63 #define PREFIX_h264_qpel16_hv_lowpass_altivec avg_h264_qpel16_hv_lowpass_altivec
nengel@2 64 #define PREFIX_h264_qpel16_hv_lowpass_num altivec_avg_h264_qpel16_hv_lowpass_num
nengel@2 65 #include "h264_template_altivec.c"
nengel@2 66 #undef OP_U8_ALTIVEC
nengel@2 67 #undef PREFIX_h264_chroma_mc8_altivec
nengel@2 68 #undef PREFIX_no_rnd_vc1_chroma_mc8_altivec
nengel@2 69 #undef PREFIX_h264_chroma_mc8_num
nengel@2 70 #undef PREFIX_h264_qpel16_h_lowpass_altivec
nengel@2 71 #undef PREFIX_h264_qpel16_h_lowpass_num
nengel@2 72 #undef PREFIX_h264_qpel16_v_lowpass_altivec
nengel@2 73 #undef PREFIX_h264_qpel16_v_lowpass_num
nengel@2 74 #undef PREFIX_h264_qpel16_hv_lowpass_altivec
nengel@2 75 #undef PREFIX_h264_qpel16_hv_lowpass_num
nengel@2 76
nengel@2 77 #define H264_MC(OPNAME, SIZE, CODETYPE) \
nengel@2 78 static void OPNAME ## h264_qpel ## SIZE ## _mc00_ ## CODETYPE (uint8_t *dst, uint8_t *src, int stride){\
nengel@2 79 OPNAME ## pixels ## SIZE ## _ ## CODETYPE(dst, src, stride, SIZE);\
nengel@2 80 }\
nengel@2 81 \
nengel@2 82 static void OPNAME ## h264_qpel ## SIZE ## _mc10_ ## CODETYPE(uint8_t *dst, uint8_t *src, int stride){ \
nengel@2 83 DECLARE_ALIGNED(16, uint8_t, half)[SIZE*SIZE];\
nengel@2 84 put_h264_qpel ## SIZE ## _h_lowpass_ ## CODETYPE(half, src, SIZE, stride);\
nengel@2 85 OPNAME ## pixels ## SIZE ## _l2_ ## CODETYPE(dst, src, half, stride, stride, SIZE);\
nengel@2 86 }\
nengel@2 87 \
nengel@2 88 static void OPNAME ## h264_qpel ## SIZE ## _mc20_ ## CODETYPE(uint8_t *dst, uint8_t *src, int stride){\
nengel@2 89 OPNAME ## h264_qpel ## SIZE ## _h_lowpass_ ## CODETYPE(dst, src, stride, stride);\
nengel@2 90 }\
nengel@2 91 \
nengel@2 92 static void OPNAME ## h264_qpel ## SIZE ## _mc30_ ## CODETYPE(uint8_t *dst, uint8_t *src, int stride){\
nengel@2 93 DECLARE_ALIGNED(16, uint8_t, half)[SIZE*SIZE];\
nengel@2 94 put_h264_qpel ## SIZE ## _h_lowpass_ ## CODETYPE(half, src, SIZE, stride);\
nengel@2 95 OPNAME ## pixels ## SIZE ## _l2_ ## CODETYPE(dst, src+1, half, stride, stride, SIZE);\
nengel@2 96 }\
nengel@2 97 \
nengel@2 98 static void OPNAME ## h264_qpel ## SIZE ## _mc01_ ## CODETYPE(uint8_t *dst, uint8_t *src, int stride){\
nengel@2 99 DECLARE_ALIGNED(16, uint8_t, half)[SIZE*SIZE];\
nengel@2 100 put_h264_qpel ## SIZE ## _v_lowpass_ ## CODETYPE(half, src, SIZE, stride);\
nengel@2 101 OPNAME ## pixels ## SIZE ## _l2_ ## CODETYPE(dst, src, half, stride, stride, SIZE);\
nengel@2 102 }\
nengel@2 103 \
nengel@2 104 static void OPNAME ## h264_qpel ## SIZE ## _mc02_ ## CODETYPE(uint8_t *dst, uint8_t *src, int stride){\
nengel@2 105 OPNAME ## h264_qpel ## SIZE ## _v_lowpass_ ## CODETYPE(dst, src, stride, stride);\
nengel@2 106 }\
nengel@2 107 \
nengel@2 108 static void OPNAME ## h264_qpel ## SIZE ## _mc03_ ## CODETYPE(uint8_t *dst, uint8_t *src, int stride){\
nengel@2 109 DECLARE_ALIGNED(16, uint8_t, half)[SIZE*SIZE];\
nengel@2 110 put_h264_qpel ## SIZE ## _v_lowpass_ ## CODETYPE(half, src, SIZE, stride);\
nengel@2 111 OPNAME ## pixels ## SIZE ## _l2_ ## CODETYPE(dst, src+stride, half, stride, stride, SIZE);\
nengel@2 112 }\
nengel@2 113 \
nengel@2 114 static void OPNAME ## h264_qpel ## SIZE ## _mc11_ ## CODETYPE(uint8_t *dst, uint8_t *src, int stride){\
nengel@2 115 DECLARE_ALIGNED(16, uint8_t, halfH)[SIZE*SIZE];\
nengel@2 116 DECLARE_ALIGNED(16, uint8_t, halfV)[SIZE*SIZE];\
nengel@2 117 put_h264_qpel ## SIZE ## _h_lowpass_ ## CODETYPE(halfH, src, SIZE, stride);\
nengel@2 118 put_h264_qpel ## SIZE ## _v_lowpass_ ## CODETYPE(halfV, src, SIZE, stride);\
nengel@2 119 OPNAME ## pixels ## SIZE ## _l2_ ## CODETYPE(dst, halfH, halfV, stride, SIZE, SIZE);\
nengel@2 120 }\
nengel@2 121 \
nengel@2 122 static void OPNAME ## h264_qpel ## SIZE ## _mc31_ ## CODETYPE(uint8_t *dst, uint8_t *src, int stride){\
nengel@2 123 DECLARE_ALIGNED(16, uint8_t, halfH)[SIZE*SIZE];\
nengel@2 124 DECLARE_ALIGNED(16, uint8_t, halfV)[SIZE*SIZE];\
nengel@2 125 put_h264_qpel ## SIZE ## _h_lowpass_ ## CODETYPE(halfH, src, SIZE, stride);\
nengel@2 126 put_h264_qpel ## SIZE ## _v_lowpass_ ## CODETYPE(halfV, src+1, SIZE, stride);\
nengel@2 127 OPNAME ## pixels ## SIZE ## _l2_ ## CODETYPE(dst, halfH, halfV, stride, SIZE, SIZE);\
nengel@2 128 }\
nengel@2 129 \
nengel@2 130 static void OPNAME ## h264_qpel ## SIZE ## _mc13_ ## CODETYPE(uint8_t *dst, uint8_t *src, int stride){\
nengel@2 131 DECLARE_ALIGNED(16, uint8_t, halfH)[SIZE*SIZE];\
nengel@2 132 DECLARE_ALIGNED(16, uint8_t, halfV)[SIZE*SIZE];\
nengel@2 133 put_h264_qpel ## SIZE ## _h_lowpass_ ## CODETYPE(halfH, src + stride, SIZE, stride);\
nengel@2 134 put_h264_qpel ## SIZE ## _v_lowpass_ ## CODETYPE(halfV, src, SIZE, stride);\
nengel@2 135 OPNAME ## pixels ## SIZE ## _l2_ ## CODETYPE(dst, halfH, halfV, stride, SIZE, SIZE);\
nengel@2 136 }\
nengel@2 137 \
nengel@2 138 static void OPNAME ## h264_qpel ## SIZE ## _mc33_ ## CODETYPE(uint8_t *dst, uint8_t *src, int stride){\
nengel@2 139 DECLARE_ALIGNED(16, uint8_t, halfH)[SIZE*SIZE];\
nengel@2 140 DECLARE_ALIGNED(16, uint8_t, halfV)[SIZE*SIZE];\
nengel@2 141 put_h264_qpel ## SIZE ## _h_lowpass_ ## CODETYPE(halfH, src + stride, SIZE, stride);\
nengel@2 142 put_h264_qpel ## SIZE ## _v_lowpass_ ## CODETYPE(halfV, src+1, SIZE, stride);\
nengel@2 143 OPNAME ## pixels ## SIZE ## _l2_ ## CODETYPE(dst, halfH, halfV, stride, SIZE, SIZE);\
nengel@2 144 }\
nengel@2 145 \
nengel@2 146 static void OPNAME ## h264_qpel ## SIZE ## _mc22_ ## CODETYPE(uint8_t *dst, uint8_t *src, int stride){\
nengel@2 147 DECLARE_ALIGNED(16, int16_t, tmp)[SIZE*(SIZE+8)];\
nengel@2 148 OPNAME ## h264_qpel ## SIZE ## _hv_lowpass_ ## CODETYPE(dst, tmp, src, stride, SIZE, stride);\
nengel@2 149 }\
nengel@2 150 \
nengel@2 151 static void OPNAME ## h264_qpel ## SIZE ## _mc21_ ## CODETYPE(uint8_t *dst, uint8_t *src, int stride){\
nengel@2 152 DECLARE_ALIGNED(16, uint8_t, halfH)[SIZE*SIZE];\
nengel@2 153 DECLARE_ALIGNED(16, uint8_t, halfHV)[SIZE*SIZE];\
nengel@2 154 DECLARE_ALIGNED(16, int16_t, tmp)[SIZE*(SIZE+8)];\
nengel@2 155 put_h264_qpel ## SIZE ## _h_lowpass_ ## CODETYPE(halfH, src, SIZE, stride);\
nengel@2 156 put_h264_qpel ## SIZE ## _hv_lowpass_ ## CODETYPE(halfHV, tmp, src, SIZE, SIZE, stride);\
nengel@2 157 OPNAME ## pixels ## SIZE ## _l2_ ## CODETYPE(dst, halfH, halfHV, stride, SIZE, SIZE);\
nengel@2 158 }\
nengel@2 159 \
nengel@2 160 static void OPNAME ## h264_qpel ## SIZE ## _mc23_ ## CODETYPE(uint8_t *dst, uint8_t *src, int stride){\
nengel@2 161 DECLARE_ALIGNED(16, uint8_t, halfH)[SIZE*SIZE];\
nengel@2 162 DECLARE_ALIGNED(16, uint8_t, halfHV)[SIZE*SIZE];\
nengel@2 163 DECLARE_ALIGNED(16, int16_t, tmp)[SIZE*(SIZE+8)];\
nengel@2 164 put_h264_qpel ## SIZE ## _h_lowpass_ ## CODETYPE(halfH, src + stride, SIZE, stride);\
nengel@2 165 put_h264_qpel ## SIZE ## _hv_lowpass_ ## CODETYPE(halfHV, tmp, src, SIZE, SIZE, stride);\
nengel@2 166 OPNAME ## pixels ## SIZE ## _l2_ ## CODETYPE(dst, halfH, halfHV, stride, SIZE, SIZE);\
nengel@2 167 }\
nengel@2 168 \
nengel@2 169 static void OPNAME ## h264_qpel ## SIZE ## _mc12_ ## CODETYPE(uint8_t *dst, uint8_t *src, int stride){\
nengel@2 170 DECLARE_ALIGNED(16, uint8_t, halfV)[SIZE*SIZE];\
nengel@2 171 DECLARE_ALIGNED(16, uint8_t, halfHV)[SIZE*SIZE];\
nengel@2 172 DECLARE_ALIGNED(16, int16_t, tmp)[SIZE*(SIZE+8)];\
nengel@2 173 put_h264_qpel ## SIZE ## _v_lowpass_ ## CODETYPE(halfV, src, SIZE, stride);\
nengel@2 174 put_h264_qpel ## SIZE ## _hv_lowpass_ ## CODETYPE(halfHV, tmp, src, SIZE, SIZE, stride);\
nengel@2 175 OPNAME ## pixels ## SIZE ## _l2_ ## CODETYPE(dst, halfV, halfHV, stride, SIZE, SIZE);\
nengel@2 176 }\
nengel@2 177 \
nengel@2 178 static void OPNAME ## h264_qpel ## SIZE ## _mc32_ ## CODETYPE(uint8_t *dst, uint8_t *src, int stride){\
nengel@2 179 DECLARE_ALIGNED(16, uint8_t, halfV)[SIZE*SIZE];\
nengel@2 180 DECLARE_ALIGNED(16, uint8_t, halfHV)[SIZE*SIZE];\
nengel@2 181 DECLARE_ALIGNED(16, int16_t, tmp)[SIZE*(SIZE+8)];\
nengel@2 182 put_h264_qpel ## SIZE ## _v_lowpass_ ## CODETYPE(halfV, src+1, SIZE, stride);\
nengel@2 183 put_h264_qpel ## SIZE ## _hv_lowpass_ ## CODETYPE(halfHV, tmp, src, SIZE, SIZE, stride);\
nengel@2 184 OPNAME ## pixels ## SIZE ## _l2_ ## CODETYPE(dst, halfV, halfHV, stride, SIZE, SIZE);\
nengel@2 185 }\
nengel@2 186
nengel@2 187 static inline void put_pixels16_l2_altivec( uint8_t * dst, const uint8_t * src1,
nengel@2 188 const uint8_t * src2, int dst_stride,
nengel@2 189 int src_stride1, int h)
nengel@2 190 {
nengel@2 191 int i;
nengel@2 192 vec_u8 a, b, d, tmp1, tmp2, mask, mask_, edges, align;
nengel@2 193
nengel@2 194 mask_ = vec_lvsl(0, src2);
nengel@2 195
nengel@2 196 for (i = 0; i < h; i++) {
nengel@2 197
nengel@2 198 tmp1 = vec_ld(i * src_stride1, src1);
nengel@2 199 mask = vec_lvsl(i * src_stride1, src1);
nengel@2 200 tmp2 = vec_ld(i * src_stride1 + 15, src1);
nengel@2 201
nengel@2 202 a = vec_perm(tmp1, tmp2, mask);
nengel@2 203
nengel@2 204 tmp1 = vec_ld(i * 16, src2);
nengel@2 205 tmp2 = vec_ld(i * 16 + 15, src2);
nengel@2 206
nengel@2 207 b = vec_perm(tmp1, tmp2, mask_);
nengel@2 208
nengel@2 209 tmp1 = vec_ld(0, dst);
nengel@2 210 mask = vec_lvsl(0, dst);
nengel@2 211 tmp2 = vec_ld(15, dst);
nengel@2 212
nengel@2 213 d = vec_avg(a, b);
nengel@2 214
nengel@2 215 edges = vec_perm(tmp2, tmp1, mask);
nengel@2 216
nengel@2 217 align = vec_lvsr(0, dst);
nengel@2 218
nengel@2 219 tmp2 = vec_perm(d, edges, align);
nengel@2 220 tmp1 = vec_perm(edges, d, align);
nengel@2 221
nengel@2 222 vec_st(tmp2, 15, dst);
nengel@2 223 vec_st(tmp1, 0 , dst);
nengel@2 224
nengel@2 225 dst += dst_stride;
nengel@2 226 }
nengel@2 227 }
nengel@2 228
nengel@2 229 static inline void avg_pixels16_l2_altivec( uint8_t * dst, const uint8_t * src1,
nengel@2 230 const uint8_t * src2, int dst_stride,
nengel@2 231 int src_stride1, int h)
nengel@2 232 {
nengel@2 233 int i;
nengel@2 234 vec_u8 a, b, d, tmp1, tmp2, mask, mask_, edges, align;
nengel@2 235
nengel@2 236 mask_ = vec_lvsl(0, src2);
nengel@2 237
nengel@2 238 for (i = 0; i < h; i++) {
nengel@2 239
nengel@2 240 tmp1 = vec_ld(i * src_stride1, src1);
nengel@2 241 mask = vec_lvsl(i * src_stride1, src1);
nengel@2 242 tmp2 = vec_ld(i * src_stride1 + 15, src1);
nengel@2 243
nengel@2 244 a = vec_perm(tmp1, tmp2, mask);
nengel@2 245
nengel@2 246 tmp1 = vec_ld(i * 16, src2);
nengel@2 247 tmp2 = vec_ld(i * 16 + 15, src2);
nengel@2 248
nengel@2 249 b = vec_perm(tmp1, tmp2, mask_);
nengel@2 250
nengel@2 251 tmp1 = vec_ld(0, dst);
nengel@2 252 mask = vec_lvsl(0, dst);
nengel@2 253 tmp2 = vec_ld(15, dst);
nengel@2 254
nengel@2 255 d = vec_avg(vec_perm(tmp1, tmp2, mask), vec_avg(a, b));
nengel@2 256
nengel@2 257 edges = vec_perm(tmp2, tmp1, mask);
nengel@2 258
nengel@2 259 align = vec_lvsr(0, dst);
nengel@2 260
nengel@2 261 tmp2 = vec_perm(d, edges, align);
nengel@2 262 tmp1 = vec_perm(edges, d, align);
nengel@2 263
nengel@2 264 vec_st(tmp2, 15, dst);
nengel@2 265 vec_st(tmp1, 0 , dst);
nengel@2 266
nengel@2 267 dst += dst_stride;
nengel@2 268 }
nengel@2 269 }
nengel@2 270
nengel@2 271 /* Implemented but could be faster
nengel@2 272 #define put_pixels16_l2_altivec(d,s1,s2,ds,s1s,h) put_pixels16_l2(d,s1,s2,ds,s1s,16,h)
nengel@2 273 #define avg_pixels16_l2_altivec(d,s1,s2,ds,s1s,h) avg_pixels16_l2(d,s1,s2,ds,s1s,16,h)
nengel@2 274 */
nengel@2 275
nengel@2 276 H264_MC(put_, 16, altivec)
nengel@2 277 H264_MC(avg_, 16, altivec)
nengel@2 278
nengel@2 279
nengel@2 280 /****************************************************************************
nengel@2 281 * IDCT transform:
nengel@2 282 ****************************************************************************/
nengel@2 283
nengel@2 284 #define VEC_1D_DCT(vb0,vb1,vb2,vb3,va0,va1,va2,va3) \
nengel@2 285 /* 1st stage */ \
nengel@2 286 vz0 = vec_add(vb0,vb2); /* temp[0] = Y[0] + Y[2] */ \
nengel@2 287 vz1 = vec_sub(vb0,vb2); /* temp[1] = Y[0] - Y[2] */ \
nengel@2 288 vz2 = vec_sra(vb1,vec_splat_u16(1)); \
nengel@2 289 vz2 = vec_sub(vz2,vb3); /* temp[2] = Y[1].1/2 - Y[3] */ \
nengel@2 290 vz3 = vec_sra(vb3,vec_splat_u16(1)); \
nengel@2 291 vz3 = vec_add(vb1,vz3); /* temp[3] = Y[1] + Y[3].1/2 */ \
nengel@2 292 /* 2nd stage: output */ \
nengel@2 293 va0 = vec_add(vz0,vz3); /* x[0] = temp[0] + temp[3] */ \
nengel@2 294 va1 = vec_add(vz1,vz2); /* x[1] = temp[1] + temp[2] */ \
nengel@2 295 va2 = vec_sub(vz1,vz2); /* x[2] = temp[1] - temp[2] */ \
nengel@2 296 va3 = vec_sub(vz0,vz3) /* x[3] = temp[0] - temp[3] */
nengel@2 297
nengel@2 298 #define VEC_TRANSPOSE_4(a0,a1,a2,a3,b0,b1,b2,b3) \
nengel@2 299 b0 = vec_mergeh( a0, a0 ); \
nengel@2 300 b1 = vec_mergeh( a1, a0 ); \
nengel@2 301 b2 = vec_mergeh( a2, a0 ); \
nengel@2 302 b3 = vec_mergeh( a3, a0 ); \
nengel@2 303 a0 = vec_mergeh( b0, b2 ); \
nengel@2 304 a1 = vec_mergel( b0, b2 ); \
nengel@2 305 a2 = vec_mergeh( b1, b3 ); \
nengel@2 306 a3 = vec_mergel( b1, b3 ); \
nengel@2 307 b0 = vec_mergeh( a0, a2 ); \
nengel@2 308 b1 = vec_mergel( a0, a2 ); \
nengel@2 309 b2 = vec_mergeh( a1, a3 ); \
nengel@2 310 b3 = vec_mergel( a1, a3 )
nengel@2 311
nengel@2 312 #define VEC_LOAD_U8_ADD_S16_STORE_U8(va) \
nengel@2 313 vdst_orig = vec_ld(0, dst); \
nengel@2 314 vdst = vec_perm(vdst_orig, zero_u8v, vdst_mask); \
nengel@2 315 vdst_ss = (vec_s16) vec_mergeh(zero_u8v, vdst); \
nengel@2 316 va = vec_add(va, vdst_ss); \
nengel@2 317 va_u8 = vec_packsu(va, zero_s16v); \
nengel@2 318 va_u32 = vec_splat((vec_u32)va_u8, 0); \
nengel@2 319 vec_ste(va_u32, element, (uint32_t*)dst);
nengel@2 320
nengel@2 321 static void ff_h264_idct_add_altivec(uint8_t *dst, DCTELEM *block, int stride)
nengel@2 322 {
nengel@2 323 vec_s16 va0, va1, va2, va3;
nengel@2 324 vec_s16 vz0, vz1, vz2, vz3;
nengel@2 325 vec_s16 vtmp0, vtmp1, vtmp2, vtmp3;
nengel@2 326 vec_u8 va_u8;
nengel@2 327 vec_u32 va_u32;
nengel@2 328 vec_s16 vdst_ss;
nengel@2 329 const vec_u16 v6us = vec_splat_u16(6);
nengel@2 330 vec_u8 vdst, vdst_orig;
nengel@2 331 vec_u8 vdst_mask = vec_lvsl(0, dst);
nengel@2 332 int element = ((unsigned long)dst & 0xf) >> 2;
nengel@2 333 LOAD_ZERO;
nengel@2 334
nengel@2 335 block[0] += 32; /* add 32 as a DC-level for rounding */
nengel@2 336
nengel@2 337 vtmp0 = vec_ld(0,block);
nengel@2 338 vtmp1 = vec_sld(vtmp0, vtmp0, 8);
nengel@2 339 vtmp2 = vec_ld(16,block);
nengel@2 340 vtmp3 = vec_sld(vtmp2, vtmp2, 8);
nengel@2 341
nengel@2 342 VEC_1D_DCT(vtmp0,vtmp1,vtmp2,vtmp3,va0,va1,va2,va3);
nengel@2 343 VEC_TRANSPOSE_4(va0,va1,va2,va3,vtmp0,vtmp1,vtmp2,vtmp3);
nengel@2 344 VEC_1D_DCT(vtmp0,vtmp1,vtmp2,vtmp3,va0,va1,va2,va3);
nengel@2 345
nengel@2 346 va0 = vec_sra(va0,v6us);
nengel@2 347 va1 = vec_sra(va1,v6us);
nengel@2 348 va2 = vec_sra(va2,v6us);
nengel@2 349 va3 = vec_sra(va3,v6us);
nengel@2 350
nengel@2 351 VEC_LOAD_U8_ADD_S16_STORE_U8(va0);
nengel@2 352 dst += stride;
nengel@2 353 VEC_LOAD_U8_ADD_S16_STORE_U8(va1);
nengel@2 354 dst += stride;
nengel@2 355 VEC_LOAD_U8_ADD_S16_STORE_U8(va2);
nengel@2 356 dst += stride;
nengel@2 357 VEC_LOAD_U8_ADD_S16_STORE_U8(va3);
nengel@2 358 }
nengel@2 359
nengel@2 360 #define IDCT8_1D_ALTIVEC(s0, s1, s2, s3, s4, s5, s6, s7, d0, d1, d2, d3, d4, d5, d6, d7) {\
nengel@2 361 /* a0 = SRC(0) + SRC(4); */ \
nengel@2 362 vec_s16 a0v = vec_add(s0, s4); \
nengel@2 363 /* a2 = SRC(0) - SRC(4); */ \
nengel@2 364 vec_s16 a2v = vec_sub(s0, s4); \
nengel@2 365 /* a4 = (SRC(2)>>1) - SRC(6); */ \
nengel@2 366 vec_s16 a4v = vec_sub(vec_sra(s2, onev), s6); \
nengel@2 367 /* a6 = (SRC(6)>>1) + SRC(2); */ \
nengel@2 368 vec_s16 a6v = vec_add(vec_sra(s6, onev), s2); \
nengel@2 369 /* b0 = a0 + a6; */ \
nengel@2 370 vec_s16 b0v = vec_add(a0v, a6v); \
nengel@2 371 /* b2 = a2 + a4; */ \
nengel@2 372 vec_s16 b2v = vec_add(a2v, a4v); \
nengel@2 373 /* b4 = a2 - a4; */ \
nengel@2 374 vec_s16 b4v = vec_sub(a2v, a4v); \
nengel@2 375 /* b6 = a0 - a6; */ \
nengel@2 376 vec_s16 b6v = vec_sub(a0v, a6v); \
nengel@2 377 /* a1 = SRC(5) - SRC(3) - SRC(7) - (SRC(7)>>1); */ \
nengel@2 378 /* a1 = (SRC(5)-SRC(3)) - (SRC(7) + (SRC(7)>>1)); */ \
nengel@2 379 vec_s16 a1v = vec_sub( vec_sub(s5, s3), vec_add(s7, vec_sra(s7, onev)) ); \
nengel@2 380 /* a3 = SRC(7) + SRC(1) - SRC(3) - (SRC(3)>>1); */ \
nengel@2 381 /* a3 = (SRC(7)+SRC(1)) - (SRC(3) + (SRC(3)>>1)); */ \
nengel@2 382 vec_s16 a3v = vec_sub( vec_add(s7, s1), vec_add(s3, vec_sra(s3, onev)) );\
nengel@2 383 /* a5 = SRC(7) - SRC(1) + SRC(5) + (SRC(5)>>1); */ \
nengel@2 384 /* a5 = (SRC(7)-SRC(1)) + SRC(5) + (SRC(5)>>1); */ \
nengel@2 385 vec_s16 a5v = vec_add( vec_sub(s7, s1), vec_add(s5, vec_sra(s5, onev)) );\
nengel@2 386 /* a7 = SRC(5)+SRC(3) + SRC(1) + (SRC(1)>>1); */ \
nengel@2 387 vec_s16 a7v = vec_add( vec_add(s5, s3), vec_add(s1, vec_sra(s1, onev)) );\
nengel@2 388 /* b1 = (a7>>2) + a1; */ \
nengel@2 389 vec_s16 b1v = vec_add( vec_sra(a7v, twov), a1v); \
nengel@2 390 /* b3 = a3 + (a5>>2); */ \
nengel@2 391 vec_s16 b3v = vec_add(a3v, vec_sra(a5v, twov)); \
nengel@2 392 /* b5 = (a3>>2) - a5; */ \
nengel@2 393 vec_s16 b5v = vec_sub( vec_sra(a3v, twov), a5v); \
nengel@2 394 /* b7 = a7 - (a1>>2); */ \
nengel@2 395 vec_s16 b7v = vec_sub( a7v, vec_sra(a1v, twov)); \
nengel@2 396 /* DST(0, b0 + b7); */ \
nengel@2 397 d0 = vec_add(b0v, b7v); \
nengel@2 398 /* DST(1, b2 + b5); */ \
nengel@2 399 d1 = vec_add(b2v, b5v); \
nengel@2 400 /* DST(2, b4 + b3); */ \
nengel@2 401 d2 = vec_add(b4v, b3v); \
nengel@2 402 /* DST(3, b6 + b1); */ \
nengel@2 403 d3 = vec_add(b6v, b1v); \
nengel@2 404 /* DST(4, b6 - b1); */ \
nengel@2 405 d4 = vec_sub(b6v, b1v); \
nengel@2 406 /* DST(5, b4 - b3); */ \
nengel@2 407 d5 = vec_sub(b4v, b3v); \
nengel@2 408 /* DST(6, b2 - b5); */ \
nengel@2 409 d6 = vec_sub(b2v, b5v); \
nengel@2 410 /* DST(7, b0 - b7); */ \
nengel@2 411 d7 = vec_sub(b0v, b7v); \
nengel@2 412 }
nengel@2 413
nengel@2 414 #define ALTIVEC_STORE_SUM_CLIP(dest, idctv, perm_ldv, perm_stv, sel) { \
nengel@2 415 /* unaligned load */ \
nengel@2 416 vec_u8 hv = vec_ld( 0, dest ); \
nengel@2 417 vec_u8 lv = vec_ld( 7, dest ); \
nengel@2 418 vec_u8 dstv = vec_perm( hv, lv, (vec_u8)perm_ldv ); \
nengel@2 419 vec_s16 idct_sh6 = vec_sra(idctv, sixv); \
nengel@2 420 vec_u16 dst16 = (vec_u16)vec_mergeh(zero_u8v, dstv); \
nengel@2 421 vec_s16 idstsum = vec_adds(idct_sh6, (vec_s16)dst16); \
nengel@2 422 vec_u8 idstsum8 = vec_packsu(zero_s16v, idstsum); \
nengel@2 423 vec_u8 edgehv; \
nengel@2 424 /* unaligned store */ \
nengel@2 425 vec_u8 bodyv = vec_perm( idstsum8, idstsum8, perm_stv );\
nengel@2 426 vec_u8 edgelv = vec_perm( sel, zero_u8v, perm_stv ); \
nengel@2 427 lv = vec_sel( lv, bodyv, edgelv ); \
nengel@2 428 vec_st( lv, 7, dest ); \
nengel@2 429 hv = vec_ld( 0, dest ); \
nengel@2 430 edgehv = vec_perm( zero_u8v, sel, perm_stv ); \
nengel@2 431 hv = vec_sel( hv, bodyv, edgehv ); \
nengel@2 432 vec_st( hv, 0, dest ); \
nengel@2 433 }
nengel@2 434
nengel@2 435 static void ff_h264_idct8_add_altivec( uint8_t *dst, DCTELEM *dct, int stride ) {
nengel@2 436 vec_s16 s0, s1, s2, s3, s4, s5, s6, s7;
nengel@2 437 vec_s16 d0, d1, d2, d3, d4, d5, d6, d7;
nengel@2 438 vec_s16 idct0, idct1, idct2, idct3, idct4, idct5, idct6, idct7;
nengel@2 439
nengel@2 440 vec_u8 perm_ldv = vec_lvsl(0, dst);
nengel@2 441 vec_u8 perm_stv = vec_lvsr(8, dst);
nengel@2 442
nengel@2 443 const vec_u16 onev = vec_splat_u16(1);
nengel@2 444 const vec_u16 twov = vec_splat_u16(2);
nengel@2 445 const vec_u16 sixv = vec_splat_u16(6);
nengel@2 446
nengel@2 447 const vec_u8 sel = (vec_u8) {0,0,0,0,0,0,0,0,-1,-1,-1,-1,-1,-1,-1,-1};
nengel@2 448 LOAD_ZERO;
nengel@2 449
nengel@2 450 dct[0] += 32; // rounding for the >>6 at the end
nengel@2 451
nengel@2 452 s0 = vec_ld(0x00, (int16_t*)dct);
nengel@2 453 s1 = vec_ld(0x10, (int16_t*)dct);
nengel@2 454 s2 = vec_ld(0x20, (int16_t*)dct);
nengel@2 455 s3 = vec_ld(0x30, (int16_t*)dct);
nengel@2 456 s4 = vec_ld(0x40, (int16_t*)dct);
nengel@2 457 s5 = vec_ld(0x50, (int16_t*)dct);
nengel@2 458 s6 = vec_ld(0x60, (int16_t*)dct);
nengel@2 459 s7 = vec_ld(0x70, (int16_t*)dct);
nengel@2 460
nengel@2 461 IDCT8_1D_ALTIVEC(s0, s1, s2, s3, s4, s5, s6, s7,
nengel@2 462 d0, d1, d2, d3, d4, d5, d6, d7);
nengel@2 463
nengel@2 464 TRANSPOSE8( d0, d1, d2, d3, d4, d5, d6, d7 );
nengel@2 465
nengel@2 466 IDCT8_1D_ALTIVEC(d0, d1, d2, d3, d4, d5, d6, d7,
nengel@2 467 idct0, idct1, idct2, idct3, idct4, idct5, idct6, idct7);
nengel@2 468
nengel@2 469 ALTIVEC_STORE_SUM_CLIP(&dst[0*stride], idct0, perm_ldv, perm_stv, sel);
nengel@2 470 ALTIVEC_STORE_SUM_CLIP(&dst[1*stride], idct1, perm_ldv, perm_stv, sel);
nengel@2 471 ALTIVEC_STORE_SUM_CLIP(&dst[2*stride], idct2, perm_ldv, perm_stv, sel);
nengel@2 472 ALTIVEC_STORE_SUM_CLIP(&dst[3*stride], idct3, perm_ldv, perm_stv, sel);
nengel@2 473 ALTIVEC_STORE_SUM_CLIP(&dst[4*stride], idct4, perm_ldv, perm_stv, sel);
nengel@2 474 ALTIVEC_STORE_SUM_CLIP(&dst[5*stride], idct5, perm_ldv, perm_stv, sel);
nengel@2 475 ALTIVEC_STORE_SUM_CLIP(&dst[6*stride], idct6, perm_ldv, perm_stv, sel);
nengel@2 476 ALTIVEC_STORE_SUM_CLIP(&dst[7*stride], idct7, perm_ldv, perm_stv, sel);
nengel@2 477 }
nengel@2 478
nengel@2 479 static av_always_inline void h264_idct_dc_add_internal(uint8_t *dst, DCTELEM *block, int stride, int size)
nengel@2 480 {
nengel@2 481 vec_s16 dc16;
nengel@2 482 vec_u8 dcplus, dcminus, v0, v1, v2, v3, aligner;
nengel@2 483 LOAD_ZERO;
nengel@2 484 DECLARE_ALIGNED(16, int, dc);
nengel@2 485 int i;
nengel@2 486
nengel@2 487 dc = (block[0] + 32) >> 6;
nengel@2 488 dc16 = vec_splat((vec_s16) vec_lde(0, &dc), 1);
nengel@2 489
nengel@2 490 if (size == 4)
nengel@2 491 dc16 = vec_sld(dc16, zero_s16v, 8);
nengel@2 492 dcplus = vec_packsu(dc16, zero_s16v);
nengel@2 493 dcminus = vec_packsu(vec_sub(zero_s16v, dc16), zero_s16v);
nengel@2 494
nengel@2 495 aligner = vec_lvsr(0, dst);
nengel@2 496 dcplus = vec_perm(dcplus, dcplus, aligner);
nengel@2 497 dcminus = vec_perm(dcminus, dcminus, aligner);
nengel@2 498
nengel@2 499 for (i = 0; i < size; i += 4) {
nengel@2 500 v0 = vec_ld(0, dst+0*stride);
nengel@2 501 v1 = vec_ld(0, dst+1*stride);
nengel@2 502 v2 = vec_ld(0, dst+2*stride);
nengel@2 503 v3 = vec_ld(0, dst+3*stride);
nengel@2 504
nengel@2 505 v0 = vec_adds(v0, dcplus);
nengel@2 506 v1 = vec_adds(v1, dcplus);
nengel@2 507 v2 = vec_adds(v2, dcplus);
nengel@2 508 v3 = vec_adds(v3, dcplus);
nengel@2 509
nengel@2 510 v0 = vec_subs(v0, dcminus);
nengel@2 511 v1 = vec_subs(v1, dcminus);
nengel@2 512 v2 = vec_subs(v2, dcminus);
nengel@2 513 v3 = vec_subs(v3, dcminus);
nengel@2 514
nengel@2 515 vec_st(v0, 0, dst+0*stride);
nengel@2 516 vec_st(v1, 0, dst+1*stride);
nengel@2 517 vec_st(v2, 0, dst+2*stride);
nengel@2 518 vec_st(v3, 0, dst+3*stride);
nengel@2 519
nengel@2 520 dst += 4*stride;
nengel@2 521 }
nengel@2 522 }
nengel@2 523
nengel@2 524 static void h264_idct_dc_add_altivec(uint8_t *dst, DCTELEM *block, int stride)
nengel@2 525 {
nengel@2 526 h264_idct_dc_add_internal(dst, block, stride, 4);
nengel@2 527 }
nengel@2 528
nengel@2 529 static void ff_h264_idct8_dc_add_altivec(uint8_t *dst, DCTELEM *block, int stride)
nengel@2 530 {
nengel@2 531 h264_idct_dc_add_internal(dst, block, stride, 8);
nengel@2 532 }
nengel@2 533
nengel@2 534 static void ff_h264_idct_add16_altivec(uint8_t *dst, const int *block_offset, DCTELEM *block, int stride, const uint8_t nnzc[6*8]){
nengel@2 535 int i;
nengel@2 536 for(i=0; i<16; i++){
nengel@2 537 int nnz = nnzc[ scan8[i] ];
nengel@2 538 if(nnz){
nengel@2 539 if(nnz==1 && block[i*16]) h264_idct_dc_add_altivec(dst + block_offset[i], block + i*16, stride);
nengel@2 540 else ff_h264_idct_add_altivec(dst + block_offset[i], block + i*16, stride);
nengel@2 541 }
nengel@2 542 }
nengel@2 543 }
nengel@2 544
nengel@2 545 static void ff_h264_idct_add16intra_altivec(uint8_t *dst, const int *block_offset, DCTELEM *block, int stride, const uint8_t nnzc[6*8]){
nengel@2 546 int i;
nengel@2 547 for(i=0; i<16; i++){
nengel@2 548 if(nnzc[ scan8[i] ]) ff_h264_idct_add_altivec(dst + block_offset[i], block + i*16, stride);
nengel@2 549 else if(block[i*16]) h264_idct_dc_add_altivec(dst + block_offset[i], block + i*16, stride);
nengel@2 550 }
nengel@2 551 }
nengel@2 552
nengel@2 553 static void ff_h264_idct8_add4_altivec(uint8_t *dst, const int *block_offset, DCTELEM *block, int stride, const uint8_t nnzc[6*8]){
nengel@2 554 int i;
nengel@2 555 for(i=0; i<16; i+=4){
nengel@2 556 int nnz = nnzc[ scan8[i] ];
nengel@2 557 if(nnz){
nengel@2 558 if(nnz==1 && block[i*16]) ff_h264_idct8_dc_add_altivec(dst + block_offset[i], block + i*16, stride);
nengel@2 559 else ff_h264_idct8_add_altivec (dst + block_offset[i], block + i*16, stride);
nengel@2 560 }
nengel@2 561 }
nengel@2 562 }
nengel@2 563
nengel@2 564 static void ff_h264_idct_add8_altivec(uint8_t **dest, const int *block_offset, DCTELEM *block, int stride, const uint8_t nnzc[6*8]){
nengel@2 565 int i;
nengel@2 566 for(i=16; i<16+8; i++){
nengel@2 567 if(nnzc[ scan8[i] ])
nengel@2 568 ff_h264_idct_add_altivec(dest[(i&4)>>2] + block_offset[i], block + i*16, stride);
nengel@2 569 else if(block[i*16])
nengel@2 570 h264_idct_dc_add_altivec(dest[(i&4)>>2] + block_offset[i], block + i*16, stride);
nengel@2 571 }
nengel@2 572 }
nengel@2 573
nengel@2 574 #define transpose4x16(r0, r1, r2, r3) { \
nengel@2 575 register vec_u8 r4; \
nengel@2 576 register vec_u8 r5; \
nengel@2 577 register vec_u8 r6; \
nengel@2 578 register vec_u8 r7; \
nengel@2 579 \
nengel@2 580 r4 = vec_mergeh(r0, r2); /*0, 2 set 0*/ \
nengel@2 581 r5 = vec_mergel(r0, r2); /*0, 2 set 1*/ \
nengel@2 582 r6 = vec_mergeh(r1, r3); /*1, 3 set 0*/ \
nengel@2 583 r7 = vec_mergel(r1, r3); /*1, 3 set 1*/ \
nengel@2 584 \
nengel@2 585 r0 = vec_mergeh(r4, r6); /*all set 0*/ \
nengel@2 586 r1 = vec_mergel(r4, r6); /*all set 1*/ \
nengel@2 587 r2 = vec_mergeh(r5, r7); /*all set 2*/ \
nengel@2 588 r3 = vec_mergel(r5, r7); /*all set 3*/ \
nengel@2 589 }
nengel@2 590
nengel@2 591 static inline void write16x4(uint8_t *dst, int dst_stride,
nengel@2 592 register vec_u8 r0, register vec_u8 r1,
nengel@2 593 register vec_u8 r2, register vec_u8 r3) {
nengel@2 594 DECLARE_ALIGNED(16, unsigned char, result)[64];
nengel@2 595 uint32_t *src_int = (uint32_t *)result, *dst_int = (uint32_t *)dst;
nengel@2 596 int int_dst_stride = dst_stride/4;
nengel@2 597
nengel@2 598 vec_st(r0, 0, result);
nengel@2 599 vec_st(r1, 16, result);
nengel@2 600 vec_st(r2, 32, result);
nengel@2 601 vec_st(r3, 48, result);
nengel@2 602 /* FIXME: there has to be a better way!!!! */
nengel@2 603 *dst_int = *src_int;
nengel@2 604 *(dst_int+ int_dst_stride) = *(src_int + 1);
nengel@2 605 *(dst_int+ 2*int_dst_stride) = *(src_int + 2);
nengel@2 606 *(dst_int+ 3*int_dst_stride) = *(src_int + 3);
nengel@2 607 *(dst_int+ 4*int_dst_stride) = *(src_int + 4);
nengel@2 608 *(dst_int+ 5*int_dst_stride) = *(src_int + 5);
nengel@2 609 *(dst_int+ 6*int_dst_stride) = *(src_int + 6);
nengel@2 610 *(dst_int+ 7*int_dst_stride) = *(src_int + 7);
nengel@2 611 *(dst_int+ 8*int_dst_stride) = *(src_int + 8);
nengel@2 612 *(dst_int+ 9*int_dst_stride) = *(src_int + 9);
nengel@2 613 *(dst_int+10*int_dst_stride) = *(src_int + 10);
nengel@2 614 *(dst_int+11*int_dst_stride) = *(src_int + 11);
nengel@2 615 *(dst_int+12*int_dst_stride) = *(src_int + 12);
nengel@2 616 *(dst_int+13*int_dst_stride) = *(src_int + 13);
nengel@2 617 *(dst_int+14*int_dst_stride) = *(src_int + 14);
nengel@2 618 *(dst_int+15*int_dst_stride) = *(src_int + 15);
nengel@2 619 }
nengel@2 620
nengel@2 621 /** \brief performs a 6x16 transpose of data in src, and stores it to dst
nengel@2 622 \todo FIXME: see if we can't spare some vec_lvsl() by them factorizing
nengel@2 623 out of unaligned_load() */
nengel@2 624 #define readAndTranspose16x6(src, src_stride, r8, r9, r10, r11, r12, r13) {\
nengel@2 625 register vec_u8 r0 = unaligned_load(0, src); \
nengel@2 626 register vec_u8 r1 = unaligned_load( src_stride, src); \
nengel@2 627 register vec_u8 r2 = unaligned_load(2* src_stride, src); \
nengel@2 628 register vec_u8 r3 = unaligned_load(3* src_stride, src); \
nengel@2 629 register vec_u8 r4 = unaligned_load(4* src_stride, src); \
nengel@2 630 register vec_u8 r5 = unaligned_load(5* src_stride, src); \
nengel@2 631 register vec_u8 r6 = unaligned_load(6* src_stride, src); \
nengel@2 632 register vec_u8 r7 = unaligned_load(7* src_stride, src); \
nengel@2 633 register vec_u8 r14 = unaligned_load(14*src_stride, src); \
nengel@2 634 register vec_u8 r15 = unaligned_load(15*src_stride, src); \
nengel@2 635 \
nengel@2 636 r8 = unaligned_load( 8*src_stride, src); \
nengel@2 637 r9 = unaligned_load( 9*src_stride, src); \
nengel@2 638 r10 = unaligned_load(10*src_stride, src); \
nengel@2 639 r11 = unaligned_load(11*src_stride, src); \
nengel@2 640 r12 = unaligned_load(12*src_stride, src); \
nengel@2 641 r13 = unaligned_load(13*src_stride, src); \
nengel@2 642 \
nengel@2 643 /*Merge first pairs*/ \
nengel@2 644 r0 = vec_mergeh(r0, r8); /*0, 8*/ \
nengel@2 645 r1 = vec_mergeh(r1, r9); /*1, 9*/ \
nengel@2 646 r2 = vec_mergeh(r2, r10); /*2,10*/ \
nengel@2 647 r3 = vec_mergeh(r3, r11); /*3,11*/ \
nengel@2 648 r4 = vec_mergeh(r4, r12); /*4,12*/ \
nengel@2 649 r5 = vec_mergeh(r5, r13); /*5,13*/ \
nengel@2 650 r6 = vec_mergeh(r6, r14); /*6,14*/ \
nengel@2 651 r7 = vec_mergeh(r7, r15); /*7,15*/ \
nengel@2 652 \
nengel@2 653 /*Merge second pairs*/ \
nengel@2 654 r8 = vec_mergeh(r0, r4); /*0,4, 8,12 set 0*/ \
nengel@2 655 r9 = vec_mergel(r0, r4); /*0,4, 8,12 set 1*/ \
nengel@2 656 r10 = vec_mergeh(r1, r5); /*1,5, 9,13 set 0*/ \
nengel@2 657 r11 = vec_mergel(r1, r5); /*1,5, 9,13 set 1*/ \
nengel@2 658 r12 = vec_mergeh(r2, r6); /*2,6,10,14 set 0*/ \
nengel@2 659 r13 = vec_mergel(r2, r6); /*2,6,10,14 set 1*/ \
nengel@2 660 r14 = vec_mergeh(r3, r7); /*3,7,11,15 set 0*/ \
nengel@2 661 r15 = vec_mergel(r3, r7); /*3,7,11,15 set 1*/ \
nengel@2 662 \
nengel@2 663 /*Third merge*/ \
nengel@2 664 r0 = vec_mergeh(r8, r12); /*0,2,4,6,8,10,12,14 set 0*/ \
nengel@2 665 r1 = vec_mergel(r8, r12); /*0,2,4,6,8,10,12,14 set 1*/ \
nengel@2 666 r2 = vec_mergeh(r9, r13); /*0,2,4,6,8,10,12,14 set 2*/ \
nengel@2 667 r4 = vec_mergeh(r10, r14); /*1,3,5,7,9,11,13,15 set 0*/ \
nengel@2 668 r5 = vec_mergel(r10, r14); /*1,3,5,7,9,11,13,15 set 1*/ \
nengel@2 669 r6 = vec_mergeh(r11, r15); /*1,3,5,7,9,11,13,15 set 2*/ \
nengel@2 670 /* Don't need to compute 3 and 7*/ \
nengel@2 671 \
nengel@2 672 /*Final merge*/ \
nengel@2 673 r8 = vec_mergeh(r0, r4); /*all set 0*/ \
nengel@2 674 r9 = vec_mergel(r0, r4); /*all set 1*/ \
nengel@2 675 r10 = vec_mergeh(r1, r5); /*all set 2*/ \
nengel@2 676 r11 = vec_mergel(r1, r5); /*all set 3*/ \
nengel@2 677 r12 = vec_mergeh(r2, r6); /*all set 4*/ \
nengel@2 678 r13 = vec_mergel(r2, r6); /*all set 5*/ \
nengel@2 679 /* Don't need to compute 14 and 15*/ \
nengel@2 680 \
nengel@2 681 }
nengel@2 682
nengel@2 683 // out: o = |x-y| < a
nengel@2 684 static inline vec_u8 diff_lt_altivec ( register vec_u8 x,
nengel@2 685 register vec_u8 y,
nengel@2 686 register vec_u8 a) {
nengel@2 687
nengel@2 688 register vec_u8 diff = vec_subs(x, y);
nengel@2 689 register vec_u8 diffneg = vec_subs(y, x);
nengel@2 690 register vec_u8 o = vec_or(diff, diffneg); /* |x-y| */
nengel@2 691 o = (vec_u8)vec_cmplt(o, a);
nengel@2 692 return o;
nengel@2 693 }
nengel@2 694
nengel@2 695 static inline vec_u8 h264_deblock_mask ( register vec_u8 p0,
nengel@2 696 register vec_u8 p1,
nengel@2 697 register vec_u8 q0,
nengel@2 698 register vec_u8 q1,
nengel@2 699 register vec_u8 alpha,
nengel@2 700 register vec_u8 beta) {
nengel@2 701
nengel@2 702 register vec_u8 mask;
nengel@2 703 register vec_u8 tempmask;
nengel@2 704
nengel@2 705 mask = diff_lt_altivec(p0, q0, alpha);
nengel@2 706 tempmask = diff_lt_altivec(p1, p0, beta);
nengel@2 707 mask = vec_and(mask, tempmask);
nengel@2 708 tempmask = diff_lt_altivec(q1, q0, beta);
nengel@2 709 mask = vec_and(mask, tempmask);
nengel@2 710
nengel@2 711 return mask;
nengel@2 712 }
nengel@2 713
nengel@2 714 // out: newp1 = clip((p2 + ((p0 + q0 + 1) >> 1)) >> 1, p1-tc0, p1+tc0)
nengel@2 715 static inline vec_u8 h264_deblock_q1(register vec_u8 p0,
nengel@2 716 register vec_u8 p1,
nengel@2 717 register vec_u8 p2,
nengel@2 718 register vec_u8 q0,
nengel@2 719 register vec_u8 tc0) {
nengel@2 720
nengel@2 721 register vec_u8 average = vec_avg(p0, q0);
nengel@2 722 register vec_u8 temp;
nengel@2 723 register vec_u8 uncliped;
nengel@2 724 register vec_u8 ones;
nengel@2 725 register vec_u8 max;
nengel@2 726 register vec_u8 min;
nengel@2 727 register vec_u8 newp1;
nengel@2 728
nengel@2 729 temp = vec_xor(average, p2);
nengel@2 730 average = vec_avg(average, p2); /*avg(p2, avg(p0, q0)) */
nengel@2 731 ones = vec_splat_u8(1);
nengel@2 732 temp = vec_and(temp, ones); /*(p2^avg(p0, q0)) & 1 */
nengel@2 733 uncliped = vec_subs(average, temp); /*(p2+((p0+q0+1)>>1))>>1 */
nengel@2 734 max = vec_adds(p1, tc0);
nengel@2 735 min = vec_subs(p1, tc0);
nengel@2 736 newp1 = vec_max(min, uncliped);
nengel@2 737 newp1 = vec_min(max, newp1);
nengel@2 738 return newp1;
nengel@2 739 }
nengel@2 740
nengel@2 741 #define h264_deblock_p0_q0(p0, p1, q0, q1, tc0masked) { \
nengel@2 742 \
nengel@2 743 const vec_u8 A0v = vec_sl(vec_splat_u8(10), vec_splat_u8(4)); \
nengel@2 744 \
nengel@2 745 register vec_u8 pq0bit = vec_xor(p0,q0); \
nengel@2 746 register vec_u8 q1minus; \
nengel@2 747 register vec_u8 p0minus; \
nengel@2 748 register vec_u8 stage1; \
nengel@2 749 register vec_u8 stage2; \
nengel@2 750 register vec_u8 vec160; \
nengel@2 751 register vec_u8 delta; \
nengel@2 752 register vec_u8 deltaneg; \
nengel@2 753 \
nengel@2 754 q1minus = vec_nor(q1, q1); /* 255 - q1 */ \
nengel@2 755 stage1 = vec_avg(p1, q1minus); /* (p1 - q1 + 256)>>1 */ \
nengel@2 756 stage2 = vec_sr(stage1, vec_splat_u8(1)); /* (p1 - q1 + 256)>>2 = 64 + (p1 - q1) >> 2 */ \
nengel@2 757 p0minus = vec_nor(p0, p0); /* 255 - p0 */ \
nengel@2 758 stage1 = vec_avg(q0, p0minus); /* (q0 - p0 + 256)>>1 */ \
nengel@2 759 pq0bit = vec_and(pq0bit, vec_splat_u8(1)); \
nengel@2 760 stage2 = vec_avg(stage2, pq0bit); /* 32 + ((q0 - p0)&1 + (p1 - q1) >> 2 + 1) >> 1 */ \
nengel@2 761 stage2 = vec_adds(stage2, stage1); /* 160 + ((p0 - q0) + (p1 - q1) >> 2 + 1) >> 1 */ \
nengel@2 762 vec160 = vec_ld(0, &A0v); \
nengel@2 763 deltaneg = vec_subs(vec160, stage2); /* -d */ \
nengel@2 764 delta = vec_subs(stage2, vec160); /* d */ \
nengel@2 765 deltaneg = vec_min(tc0masked, deltaneg); \
nengel@2 766 delta = vec_min(tc0masked, delta); \
nengel@2 767 p0 = vec_subs(p0, deltaneg); \
nengel@2 768 q0 = vec_subs(q0, delta); \
nengel@2 769 p0 = vec_adds(p0, delta); \
nengel@2 770 q0 = vec_adds(q0, deltaneg); \
nengel@2 771 }
nengel@2 772
nengel@2 773 #define h264_loop_filter_luma_altivec(p2, p1, p0, q0, q1, q2, alpha, beta, tc0) { \
nengel@2 774 DECLARE_ALIGNED(16, unsigned char, temp)[16]; \
nengel@2 775 register vec_u8 alphavec; \
nengel@2 776 register vec_u8 betavec; \
nengel@2 777 register vec_u8 mask; \
nengel@2 778 register vec_u8 p1mask; \
nengel@2 779 register vec_u8 q1mask; \
nengel@2 780 register vector signed char tc0vec; \
nengel@2 781 register vec_u8 finaltc0; \
nengel@2 782 register vec_u8 tc0masked; \
nengel@2 783 register vec_u8 newp1; \
nengel@2 784 register vec_u8 newq1; \
nengel@2 785 \
nengel@2 786 temp[0] = alpha; \
nengel@2 787 temp[1] = beta; \
nengel@2 788 alphavec = vec_ld(0, temp); \
nengel@2 789 betavec = vec_splat(alphavec, 0x1); \
nengel@2 790 alphavec = vec_splat(alphavec, 0x0); \
nengel@2 791 mask = h264_deblock_mask(p0, p1, q0, q1, alphavec, betavec); /*if in block */ \
nengel@2 792 \
nengel@2 793 *((int *)temp) = *((int *)tc0); \
nengel@2 794 tc0vec = vec_ld(0, (signed char*)temp); \
nengel@2 795 tc0vec = vec_mergeh(tc0vec, tc0vec); \
nengel@2 796 tc0vec = vec_mergeh(tc0vec, tc0vec); \
nengel@2 797 mask = vec_and(mask, vec_cmpgt(tc0vec, vec_splat_s8(-1))); /* if tc0[i] >= 0 */ \
nengel@2 798 finaltc0 = vec_and((vec_u8)tc0vec, mask); /* tc = tc0 */ \
nengel@2 799 \
nengel@2 800 p1mask = diff_lt_altivec(p2, p0, betavec); \
nengel@2 801 p1mask = vec_and(p1mask, mask); /* if ( |p2 - p0| < beta) */ \
nengel@2 802 tc0masked = vec_and(p1mask, (vec_u8)tc0vec); \
nengel@2 803 finaltc0 = vec_sub(finaltc0, p1mask); /* tc++ */ \
nengel@2 804 newp1 = h264_deblock_q1(p0, p1, p2, q0, tc0masked); \
nengel@2 805 /*end if*/ \
nengel@2 806 \
nengel@2 807 q1mask = diff_lt_altivec(q2, q0, betavec); \
nengel@2 808 q1mask = vec_and(q1mask, mask); /* if ( |q2 - q0| < beta ) */\
nengel@2 809 tc0masked = vec_and(q1mask, (vec_u8)tc0vec); \
nengel@2 810 finaltc0 = vec_sub(finaltc0, q1mask); /* tc++ */ \
nengel@2 811 newq1 = h264_deblock_q1(p0, q1, q2, q0, tc0masked); \
nengel@2 812 /*end if*/ \
nengel@2 813 \
nengel@2 814 h264_deblock_p0_q0(p0, p1, q0, q1, finaltc0); \
nengel@2 815 p1 = newp1; \
nengel@2 816 q1 = newq1; \
nengel@2 817 }
nengel@2 818
nengel@2 819 static void h264_v_loop_filter_luma_altivec(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0) {
nengel@2 820
nengel@2 821 if ((tc0[0] & tc0[1] & tc0[2] & tc0[3]) >= 0) {
nengel@2 822 register vec_u8 p2 = vec_ld(-3*stride, pix);
nengel@2 823 register vec_u8 p1 = vec_ld(-2*stride, pix);
nengel@2 824 register vec_u8 p0 = vec_ld(-1*stride, pix);
nengel@2 825 register vec_u8 q0 = vec_ld(0, pix);
nengel@2 826 register vec_u8 q1 = vec_ld(stride, pix);
nengel@2 827 register vec_u8 q2 = vec_ld(2*stride, pix);
nengel@2 828 h264_loop_filter_luma_altivec(p2, p1, p0, q0, q1, q2, alpha, beta, tc0);
nengel@2 829 vec_st(p1, -2*stride, pix);
nengel@2 830 vec_st(p0, -1*stride, pix);
nengel@2 831 vec_st(q0, 0, pix);
nengel@2 832 vec_st(q1, stride, pix);
nengel@2 833 }
nengel@2 834 }
nengel@2 835
nengel@2 836 static void h264_h_loop_filter_luma_altivec(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0) {
nengel@2 837
nengel@2 838 register vec_u8 line0, line1, line2, line3, line4, line5;
nengel@2 839 if ((tc0[0] & tc0[1] & tc0[2] & tc0[3]) < 0)
nengel@2 840 return;
nengel@2 841 readAndTranspose16x6(pix-3, stride, line0, line1, line2, line3, line4, line5);
nengel@2 842 h264_loop_filter_luma_altivec(line0, line1, line2, line3, line4, line5, alpha, beta, tc0);
nengel@2 843 transpose4x16(line1, line2, line3, line4);
nengel@2 844 write16x4(pix-2, stride, line1, line2, line3, line4);
nengel@2 845 }
nengel@2 846
nengel@2 847 static av_always_inline
nengel@2 848 void weight_h264_WxH_altivec(uint8_t *block, int stride, int log2_denom, int weight, int offset, int w, int h)
nengel@2 849 {
nengel@2 850 int y, aligned;
nengel@2 851 vec_u8 vblock;
nengel@2 852 vec_s16 vtemp, vweight, voffset, v0, v1;
nengel@2 853 vec_u16 vlog2_denom;
nengel@2 854 DECLARE_ALIGNED(16, int32_t, temp)[4];
nengel@2 855 LOAD_ZERO;
nengel@2 856
nengel@2 857 offset <<= log2_denom;
nengel@2 858 if(log2_denom) offset += 1<<(log2_denom-1);
nengel@2 859 temp[0] = log2_denom;
nengel@2 860 temp[1] = weight;
nengel@2 861 temp[2] = offset;
nengel@2 862
nengel@2 863 vtemp = (vec_s16)vec_ld(0, temp);
nengel@2 864 vlog2_denom = (vec_u16)vec_splat(vtemp, 1);
nengel@2 865 vweight = vec_splat(vtemp, 3);
nengel@2 866 voffset = vec_splat(vtemp, 5);
nengel@2 867 aligned = !((unsigned long)block & 0xf);
nengel@2 868
nengel@2 869 for (y=0; y<h; y++) {
nengel@2 870 vblock = vec_ld(0, block);
nengel@2 871
nengel@2 872 v0 = (vec_s16)vec_mergeh(zero_u8v, vblock);
nengel@2 873 v1 = (vec_s16)vec_mergel(zero_u8v, vblock);
nengel@2 874
nengel@2 875 if (w == 16 || aligned) {
nengel@2 876 v0 = vec_mladd(v0, vweight, zero_s16v);
nengel@2 877 v0 = vec_adds(v0, voffset);
nengel@2 878 v0 = vec_sra(v0, vlog2_denom);
nengel@2 879 }
nengel@2 880 if (w == 16 || !aligned) {
nengel@2 881 v1 = vec_mladd(v1, vweight, zero_s16v);
nengel@2 882 v1 = vec_adds(v1, voffset);
nengel@2 883 v1 = vec_sra(v1, vlog2_denom);
nengel@2 884 }
nengel@2 885 vblock = vec_packsu(v0, v1);
nengel@2 886 vec_st(vblock, 0, block);
nengel@2 887
nengel@2 888 block += stride;
nengel@2 889 }
nengel@2 890 }
nengel@2 891
nengel@2 892 static av_always_inline
nengel@2 893 void biweight_h264_WxH_altivec(uint8_t *dst, uint8_t *src, int stride, int log2_denom,
nengel@2 894 int weightd, int weights, int offset, int w, int h)
nengel@2 895 {
nengel@2 896 int y, dst_aligned, src_aligned;
nengel@2 897 vec_u8 vsrc, vdst;
nengel@2 898 vec_s16 vtemp, vweights, vweightd, voffset, v0, v1, v2, v3;
nengel@2 899 vec_u16 vlog2_denom;
nengel@2 900 DECLARE_ALIGNED(16, int32_t, temp)[4];
nengel@2 901 LOAD_ZERO;
nengel@2 902
nengel@2 903 offset = ((offset + 1) | 1) << log2_denom;
nengel@2 904 temp[0] = log2_denom+1;
nengel@2 905 temp[1] = weights;
nengel@2 906 temp[2] = weightd;
nengel@2 907 temp[3] = offset;
nengel@2 908
nengel@2 909 vtemp = (vec_s16)vec_ld(0, temp);
nengel@2 910 vlog2_denom = (vec_u16)vec_splat(vtemp, 1);
nengel@2 911 vweights = vec_splat(vtemp, 3);
nengel@2 912 vweightd = vec_splat(vtemp, 5);
nengel@2 913 voffset = vec_splat(vtemp, 7);
nengel@2 914 dst_aligned = !((unsigned long)dst & 0xf);
nengel@2 915 src_aligned = !((unsigned long)src & 0xf);
nengel@2 916
nengel@2 917 for (y=0; y<h; y++) {
nengel@2 918 vdst = vec_ld(0, dst);
nengel@2 919 vsrc = vec_ld(0, src);
nengel@2 920
nengel@2 921 v0 = (vec_s16)vec_mergeh(zero_u8v, vdst);
nengel@2 922 v1 = (vec_s16)vec_mergel(zero_u8v, vdst);
nengel@2 923 v2 = (vec_s16)vec_mergeh(zero_u8v, vsrc);
nengel@2 924 v3 = (vec_s16)vec_mergel(zero_u8v, vsrc);
nengel@2 925
nengel@2 926 if (w == 8) {
nengel@2 927 if (src_aligned)
nengel@2 928 v3 = v2;
nengel@2 929 else
nengel@2 930 v2 = v3;
nengel@2 931 }
nengel@2 932
nengel@2 933 if (w == 16 || dst_aligned) {
nengel@2 934 v0 = vec_mladd(v0, vweightd, zero_s16v);
nengel@2 935 v2 = vec_mladd(v2, vweights, zero_s16v);
nengel@2 936
nengel@2 937 v0 = vec_adds(v0, voffset);
nengel@2 938 v0 = vec_adds(v0, v2);
nengel@2 939 v0 = vec_sra(v0, vlog2_denom);
nengel@2 940 }
nengel@2 941 if (w == 16 || !dst_aligned) {
nengel@2 942 v1 = vec_mladd(v1, vweightd, zero_s16v);
nengel@2 943 v3 = vec_mladd(v3, vweights, zero_s16v);
nengel@2 944
nengel@2 945 v1 = vec_adds(v1, voffset);
nengel@2 946 v1 = vec_adds(v1, v3);
nengel@2 947 v1 = vec_sra(v1, vlog2_denom);
nengel@2 948 }
nengel@2 949 vdst = vec_packsu(v0, v1);
nengel@2 950 vec_st(vdst, 0, dst);
nengel@2 951
nengel@2 952 dst += stride;
nengel@2 953 src += stride;
nengel@2 954 }
nengel@2 955 }
nengel@2 956
nengel@2 957 #define H264_WEIGHT(W,H) \
nengel@2 958 static void ff_weight_h264_pixels ## W ## x ## H ## _altivec(uint8_t *block, int stride, int log2_denom, int weight, int offset){ \
nengel@2 959 weight_h264_WxH_altivec(block, stride, log2_denom, weight, offset, W, H); \
nengel@2 960 }\
nengel@2 961 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){ \
nengel@2 962 biweight_h264_WxH_altivec(dst, src, stride, log2_denom, weightd, weights, offset, W, H); \
nengel@2 963 }
nengel@2 964
nengel@2 965 H264_WEIGHT(16,16)
nengel@2 966 H264_WEIGHT(16, 8)
nengel@2 967 H264_WEIGHT( 8,16)
nengel@2 968 H264_WEIGHT( 8, 8)
nengel@2 969 H264_WEIGHT( 8, 4)
nengel@2 970
nengel@2 971 void dsputil_h264_init_ppc(DSPContext* c) {
nengel@2 972 c->put_h264_chroma_pixels_tab[0] = put_h264_chroma_mc8_altivec;
nengel@2 973 c->avg_h264_chroma_pixels_tab[0] = avg_h264_chroma_mc8_altivec;
nengel@2 974
nengel@2 975 #define dspfunc(PFX, IDX, NUM) \
nengel@2 976 c->PFX ## _pixels_tab[IDX][ 0] = PFX ## NUM ## _mc00_altivec; \
nengel@2 977 c->PFX ## _pixels_tab[IDX][ 1] = PFX ## NUM ## _mc10_altivec; \
nengel@2 978 c->PFX ## _pixels_tab[IDX][ 2] = PFX ## NUM ## _mc20_altivec; \
nengel@2 979 c->PFX ## _pixels_tab[IDX][ 3] = PFX ## NUM ## _mc30_altivec; \
nengel@2 980 c->PFX ## _pixels_tab[IDX][ 4] = PFX ## NUM ## _mc01_altivec; \
nengel@2 981 c->PFX ## _pixels_tab[IDX][ 5] = PFX ## NUM ## _mc11_altivec; \
nengel@2 982 c->PFX ## _pixels_tab[IDX][ 6] = PFX ## NUM ## _mc21_altivec; \
nengel@2 983 c->PFX ## _pixels_tab[IDX][ 7] = PFX ## NUM ## _mc31_altivec; \
nengel@2 984 c->PFX ## _pixels_tab[IDX][ 8] = PFX ## NUM ## _mc02_altivec; \
nengel@2 985 c->PFX ## _pixels_tab[IDX][ 9] = PFX ## NUM ## _mc12_altivec; \
nengel@2 986 c->PFX ## _pixels_tab[IDX][10] = PFX ## NUM ## _mc22_altivec; \
nengel@2 987 c->PFX ## _pixels_tab[IDX][11] = PFX ## NUM ## _mc32_altivec; \
nengel@2 988 c->PFX ## _pixels_tab[IDX][12] = PFX ## NUM ## _mc03_altivec; \
nengel@2 989 c->PFX ## _pixels_tab[IDX][13] = PFX ## NUM ## _mc13_altivec; \
nengel@2 990 c->PFX ## _pixels_tab[IDX][14] = PFX ## NUM ## _mc23_altivec; \
nengel@2 991 c->PFX ## _pixels_tab[IDX][15] = PFX ## NUM ## _mc33_altivec
nengel@2 992
nengel@2 993 dspfunc(put_h264_qpel, 0, 16);
nengel@2 994 dspfunc(avg_h264_qpel, 0, 16);
nengel@2 995 #undef dspfunc
nengel@2 996 }
nengel@2 997
nengel@2 998 void ff_h264dsp_init_ppc(H264DSPContext *c){
nengel@2 999 c->h264_idct_dc_add= h264_idct_dc_add_altivec;
nengel@2 1000 c->h264_idct_add = ff_h264_idct_add_altivec;
nengel@2 1001 c->h264_idct_add8 = ff_h264_idct_add8_altivec;
nengel@2 1002 c->h264_idct_add16 = ff_h264_idct_add16_altivec;
nengel@2 1003 c->h264_idct_add16intra = ff_h264_idct_add16intra_altivec;
nengel@2 1004
nengel@2 1005 c->h264_idct8_dc_add = ff_h264_idct8_dc_add_altivec;
nengel@2 1006 c->h264_idct8_add = ff_h264_idct8_add_altivec;
nengel@2 1007 c->h264_idct8_add4 = ff_h264_idct8_add4_altivec;
nengel@2 1008 c->h264_v_loop_filter_luma= h264_v_loop_filter_luma_altivec;
nengel@2 1009 c->h264_h_loop_filter_luma= h264_h_loop_filter_luma_altivec;
nengel@2 1010
nengel@2 1011 c->weight_h264_pixels_tab[0] = ff_weight_h264_pixels16x16_altivec;
nengel@2 1012 c->weight_h264_pixels_tab[1] = ff_weight_h264_pixels16x8_altivec;
nengel@2 1013 c->weight_h264_pixels_tab[2] = ff_weight_h264_pixels8x16_altivec;
nengel@2 1014 c->weight_h264_pixels_tab[3] = ff_weight_h264_pixels8x8_altivec;
nengel@2 1015 c->weight_h264_pixels_tab[4] = ff_weight_h264_pixels8x4_altivec;
nengel@2 1016 c->biweight_h264_pixels_tab[0] = ff_biweight_h264_pixels16x16_altivec;
nengel@2 1017 c->biweight_h264_pixels_tab[1] = ff_biweight_h264_pixels16x8_altivec;
nengel@2 1018 c->biweight_h264_pixels_tab[2] = ff_biweight_h264_pixels8x16_altivec;
nengel@2 1019 c->biweight_h264_pixels_tab[3] = ff_biweight_h264_pixels8x8_altivec;
nengel@2 1020 c->biweight_h264_pixels_tab[4] = ff_biweight_h264_pixels8x4_altivec;
nengel@2 1021 }