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

annotate libavcodec/ppc/idct_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) 2001 Michel Lespinasse
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 /*
nengel@2 22 * NOTE: This code is based on GPL code from the libmpeg2 project. The
nengel@2 23 * author, Michel Lespinasses, has given explicit permission to release
nengel@2 24 * under LGPL as part of FFmpeg.
nengel@2 25 */
nengel@2 26
nengel@2 27 /*
nengel@2 28 * FFmpeg integration by Dieter Shirley
nengel@2 29 *
nengel@2 30 * This file is a direct copy of the AltiVec IDCT module from the libmpeg2
nengel@2 31 * project. I've deleted all of the libmpeg2-specific code, renamed the
nengel@2 32 * functions and reordered the function parameters. The only change to the
nengel@2 33 * IDCT function itself was to factor out the partial transposition, and to
nengel@2 34 * perform a full transpose at the end of the function.
nengel@2 35 */
nengel@2 36
nengel@2 37
nengel@2 38 #include <stdlib.h> /* malloc(), free() */
nengel@2 39 #include <string.h>
nengel@2 40 #include "config.h"
nengel@2 41 #if HAVE_ALTIVEC_H
nengel@2 42 #include <altivec.h>
nengel@2 43 #endif
nengel@2 44 #include "libavcodec/dsputil.h"
nengel@2 45 #include "types_altivec.h"
nengel@2 46 #include "dsputil_ppc.h"
nengel@2 47 #include "dsputil_altivec.h"
nengel@2 48
nengel@2 49 #define IDCT_HALF \
nengel@2 50 /* 1st stage */ \
nengel@2 51 t1 = vec_mradds (a1, vx7, vx1 ); \
nengel@2 52 t8 = vec_mradds (a1, vx1, vec_subs (zero, vx7)); \
nengel@2 53 t7 = vec_mradds (a2, vx5, vx3); \
nengel@2 54 t3 = vec_mradds (ma2, vx3, vx5); \
nengel@2 55 \
nengel@2 56 /* 2nd stage */ \
nengel@2 57 t5 = vec_adds (vx0, vx4); \
nengel@2 58 t0 = vec_subs (vx0, vx4); \
nengel@2 59 t2 = vec_mradds (a0, vx6, vx2); \
nengel@2 60 t4 = vec_mradds (a0, vx2, vec_subs (zero, vx6)); \
nengel@2 61 t6 = vec_adds (t8, t3); \
nengel@2 62 t3 = vec_subs (t8, t3); \
nengel@2 63 t8 = vec_subs (t1, t7); \
nengel@2 64 t1 = vec_adds (t1, t7); \
nengel@2 65 \
nengel@2 66 /* 3rd stage */ \
nengel@2 67 t7 = vec_adds (t5, t2); \
nengel@2 68 t2 = vec_subs (t5, t2); \
nengel@2 69 t5 = vec_adds (t0, t4); \
nengel@2 70 t0 = vec_subs (t0, t4); \
nengel@2 71 t4 = vec_subs (t8, t3); \
nengel@2 72 t3 = vec_adds (t8, t3); \
nengel@2 73 \
nengel@2 74 /* 4th stage */ \
nengel@2 75 vy0 = vec_adds (t7, t1); \
nengel@2 76 vy7 = vec_subs (t7, t1); \
nengel@2 77 vy1 = vec_mradds (c4, t3, t5); \
nengel@2 78 vy6 = vec_mradds (mc4, t3, t5); \
nengel@2 79 vy2 = vec_mradds (c4, t4, t0); \
nengel@2 80 vy5 = vec_mradds (mc4, t4, t0); \
nengel@2 81 vy3 = vec_adds (t2, t6); \
nengel@2 82 vy4 = vec_subs (t2, t6);
nengel@2 83
nengel@2 84
nengel@2 85 #define IDCT \
nengel@2 86 vec_s16 vx0, vx1, vx2, vx3, vx4, vx5, vx6, vx7; \
nengel@2 87 vec_s16 vy0, vy1, vy2, vy3, vy4, vy5, vy6, vy7; \
nengel@2 88 vec_s16 a0, a1, a2, ma2, c4, mc4, zero, bias; \
nengel@2 89 vec_s16 t0, t1, t2, t3, t4, t5, t6, t7, t8; \
nengel@2 90 vec_u16 shift; \
nengel@2 91 \
nengel@2 92 c4 = vec_splat (constants[0], 0); \
nengel@2 93 a0 = vec_splat (constants[0], 1); \
nengel@2 94 a1 = vec_splat (constants[0], 2); \
nengel@2 95 a2 = vec_splat (constants[0], 3); \
nengel@2 96 mc4 = vec_splat (constants[0], 4); \
nengel@2 97 ma2 = vec_splat (constants[0], 5); \
nengel@2 98 bias = (vec_s16)vec_splat ((vec_s32)constants[0], 3); \
nengel@2 99 \
nengel@2 100 zero = vec_splat_s16 (0); \
nengel@2 101 shift = vec_splat_u16 (4); \
nengel@2 102 \
nengel@2 103 vx0 = vec_mradds (vec_sl (block[0], shift), constants[1], zero); \
nengel@2 104 vx1 = vec_mradds (vec_sl (block[1], shift), constants[2], zero); \
nengel@2 105 vx2 = vec_mradds (vec_sl (block[2], shift), constants[3], zero); \
nengel@2 106 vx3 = vec_mradds (vec_sl (block[3], shift), constants[4], zero); \
nengel@2 107 vx4 = vec_mradds (vec_sl (block[4], shift), constants[1], zero); \
nengel@2 108 vx5 = vec_mradds (vec_sl (block[5], shift), constants[4], zero); \
nengel@2 109 vx6 = vec_mradds (vec_sl (block[6], shift), constants[3], zero); \
nengel@2 110 vx7 = vec_mradds (vec_sl (block[7], shift), constants[2], zero); \
nengel@2 111 \
nengel@2 112 IDCT_HALF \
nengel@2 113 \
nengel@2 114 vx0 = vec_mergeh (vy0, vy4); \
nengel@2 115 vx1 = vec_mergel (vy0, vy4); \
nengel@2 116 vx2 = vec_mergeh (vy1, vy5); \
nengel@2 117 vx3 = vec_mergel (vy1, vy5); \
nengel@2 118 vx4 = vec_mergeh (vy2, vy6); \
nengel@2 119 vx5 = vec_mergel (vy2, vy6); \
nengel@2 120 vx6 = vec_mergeh (vy3, vy7); \
nengel@2 121 vx7 = vec_mergel (vy3, vy7); \
nengel@2 122 \
nengel@2 123 vy0 = vec_mergeh (vx0, vx4); \
nengel@2 124 vy1 = vec_mergel (vx0, vx4); \
nengel@2 125 vy2 = vec_mergeh (vx1, vx5); \
nengel@2 126 vy3 = vec_mergel (vx1, vx5); \
nengel@2 127 vy4 = vec_mergeh (vx2, vx6); \
nengel@2 128 vy5 = vec_mergel (vx2, vx6); \
nengel@2 129 vy6 = vec_mergeh (vx3, vx7); \
nengel@2 130 vy7 = vec_mergel (vx3, vx7); \
nengel@2 131 \
nengel@2 132 vx0 = vec_adds (vec_mergeh (vy0, vy4), bias); \
nengel@2 133 vx1 = vec_mergel (vy0, vy4); \
nengel@2 134 vx2 = vec_mergeh (vy1, vy5); \
nengel@2 135 vx3 = vec_mergel (vy1, vy5); \
nengel@2 136 vx4 = vec_mergeh (vy2, vy6); \
nengel@2 137 vx5 = vec_mergel (vy2, vy6); \
nengel@2 138 vx6 = vec_mergeh (vy3, vy7); \
nengel@2 139 vx7 = vec_mergel (vy3, vy7); \
nengel@2 140 \
nengel@2 141 IDCT_HALF \
nengel@2 142 \
nengel@2 143 shift = vec_splat_u16 (6); \
nengel@2 144 vx0 = vec_sra (vy0, shift); \
nengel@2 145 vx1 = vec_sra (vy1, shift); \
nengel@2 146 vx2 = vec_sra (vy2, shift); \
nengel@2 147 vx3 = vec_sra (vy3, shift); \
nengel@2 148 vx4 = vec_sra (vy4, shift); \
nengel@2 149 vx5 = vec_sra (vy5, shift); \
nengel@2 150 vx6 = vec_sra (vy6, shift); \
nengel@2 151 vx7 = vec_sra (vy7, shift);
nengel@2 152
nengel@2 153
nengel@2 154 static const vec_s16 constants[5] = {
nengel@2 155 {23170, 13573, 6518, 21895, -23170, -21895, 32, 31},
nengel@2 156 {16384, 22725, 21407, 19266, 16384, 19266, 21407, 22725},
nengel@2 157 {22725, 31521, 29692, 26722, 22725, 26722, 29692, 31521},
nengel@2 158 {21407, 29692, 27969, 25172, 21407, 25172, 27969, 29692},
nengel@2 159 {19266, 26722, 25172, 22654, 19266, 22654, 25172, 26722}
nengel@2 160 };
nengel@2 161
nengel@2 162 void idct_put_altivec(uint8_t* dest, int stride, int16_t *blk)
nengel@2 163 {
nengel@2 164 POWERPC_PERF_DECLARE(altivec_idct_put_num, 1);
nengel@2 165 vec_s16 *block = (vec_s16*)blk;
nengel@2 166 vec_u8 tmp;
nengel@2 167
nengel@2 168 #if CONFIG_POWERPC_PERF
nengel@2 169 POWERPC_PERF_START_COUNT(altivec_idct_put_num, 1);
nengel@2 170 #endif
nengel@2 171 IDCT
nengel@2 172
nengel@2 173 #define COPY(dest,src) \
nengel@2 174 tmp = vec_packsu (src, src); \
nengel@2 175 vec_ste ((vec_u32)tmp, 0, (unsigned int *)dest); \
nengel@2 176 vec_ste ((vec_u32)tmp, 4, (unsigned int *)dest);
nengel@2 177
nengel@2 178 COPY (dest, vx0) dest += stride;
nengel@2 179 COPY (dest, vx1) dest += stride;
nengel@2 180 COPY (dest, vx2) dest += stride;
nengel@2 181 COPY (dest, vx3) dest += stride;
nengel@2 182 COPY (dest, vx4) dest += stride;
nengel@2 183 COPY (dest, vx5) dest += stride;
nengel@2 184 COPY (dest, vx6) dest += stride;
nengel@2 185 COPY (dest, vx7)
nengel@2 186
nengel@2 187 POWERPC_PERF_STOP_COUNT(altivec_idct_put_num, 1);
nengel@2 188 }
nengel@2 189
nengel@2 190 void idct_add_altivec(uint8_t* dest, int stride, int16_t *blk)
nengel@2 191 {
nengel@2 192 POWERPC_PERF_DECLARE(altivec_idct_add_num, 1);
nengel@2 193 vec_s16 *block = (vec_s16*)blk;
nengel@2 194 vec_u8 tmp;
nengel@2 195 vec_s16 tmp2, tmp3;
nengel@2 196 vec_u8 perm0;
nengel@2 197 vec_u8 perm1;
nengel@2 198 vec_u8 p0, p1, p;
nengel@2 199
nengel@2 200 #if CONFIG_POWERPC_PERF
nengel@2 201 POWERPC_PERF_START_COUNT(altivec_idct_add_num, 1);
nengel@2 202 #endif
nengel@2 203
nengel@2 204 IDCT
nengel@2 205
nengel@2 206 p0 = vec_lvsl (0, dest);
nengel@2 207 p1 = vec_lvsl (stride, dest);
nengel@2 208 p = vec_splat_u8 (-1);
nengel@2 209 perm0 = vec_mergeh (p, p0);
nengel@2 210 perm1 = vec_mergeh (p, p1);
nengel@2 211
nengel@2 212 #define ADD(dest,src,perm) \
nengel@2 213 /* *(uint64_t *)&tmp = *(uint64_t *)dest; */ \
nengel@2 214 tmp = vec_ld (0, dest); \
nengel@2 215 tmp2 = (vec_s16)vec_perm (tmp, (vec_u8)zero, perm); \
nengel@2 216 tmp3 = vec_adds (tmp2, src); \
nengel@2 217 tmp = vec_packsu (tmp3, tmp3); \
nengel@2 218 vec_ste ((vec_u32)tmp, 0, (unsigned int *)dest); \
nengel@2 219 vec_ste ((vec_u32)tmp, 4, (unsigned int *)dest);
nengel@2 220
nengel@2 221 ADD (dest, vx0, perm0) dest += stride;
nengel@2 222 ADD (dest, vx1, perm1) dest += stride;
nengel@2 223 ADD (dest, vx2, perm0) dest += stride;
nengel@2 224 ADD (dest, vx3, perm1) dest += stride;
nengel@2 225 ADD (dest, vx4, perm0) dest += stride;
nengel@2 226 ADD (dest, vx5, perm1) dest += stride;
nengel@2 227 ADD (dest, vx6, perm0) dest += stride;
nengel@2 228 ADD (dest, vx7, perm1)
nengel@2 229
nengel@2 230 POWERPC_PERF_STOP_COUNT(altivec_idct_add_num, 1);
nengel@2 231 }
nengel@2 232