libavcodec/ppc/vc1dsp_altivec.c

   1 /*
   2  * VC-1 and WMV3 decoder - DSP functions AltiVec-optimized
   3  * Copyright (c) 2006 Konstantin Shishkov
   4  *
   5  * This file is part of FFmpeg.
   6  *
   7  * FFmpeg is free software; you can redistribute it and/or
   8  * modify it under the terms of the GNU Lesser General Public
   9  * License as published by the Free Software Foundation; either
  10  * version 2.1 of the License, or (at your option) any later version.
  11  *
  12  * FFmpeg is distributed in the hope that it will be useful,
  13  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  14  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  15  * Lesser General Public License for more details.
  16  *
  17  * You should have received a copy of the GNU Lesser General Public
  18  * License along with FFmpeg; if not, write to the Free Software
  19  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
  20  */
  21
  22 #include "libavcodec/dsputil.h"
  23
  24 #include "gcc_fixes.h"
  25
  26 #include "util_altivec.h"
  27
  28 // main steps of 8x8 transform
  29 #define STEP8(s0, s1, s2, s3, s4, s5, s6, s7, vec_rnd) \
  30 do { \
  31     t0 = vec_sl(vec_add(s0, s4), vec_2); \
  32     t0 = vec_add(vec_sl(t0, vec_1), t0); \
  33     t0 = vec_add(t0, vec_rnd); \
  34     t1 = vec_sl(vec_sub(s0, s4), vec_2); \
  35     t1 = vec_add(vec_sl(t1, vec_1), t1); \
  36     t1 = vec_add(t1, vec_rnd); \
  37     t2 = vec_add(vec_sl(s6, vec_2), vec_sl(s6, vec_1)); \
  38     t2 = vec_add(t2, vec_sl(s2, vec_4)); \
  39     t3 = vec_add(vec_sl(s2, vec_2), vec_sl(s2, vec_1)); \
  40     t3 = vec_sub(t3, vec_sl(s6, vec_4)); \
  41     t4 = vec_add(t0, t2); \
  42     t5 = vec_add(t1, t3); \
  43     t6 = vec_sub(t1, t3); \
  44     t7 = vec_sub(t0, t2); \
  45 \
  46     t0 = vec_sl(vec_add(s1, s3), vec_4); \
  47     t0 = vec_add(t0, vec_sl(s5, vec_3)); \
  48     t0 = vec_add(t0, vec_sl(s7, vec_2)); \
  49     t0 = vec_add(t0, vec_sub(s5, s3)); \
  50 \
  51     t1 = vec_sl(vec_sub(s1, s5), vec_4); \
  52     t1 = vec_sub(t1, vec_sl(s7, vec_3)); \
  53     t1 = vec_sub(t1, vec_sl(s3, vec_2)); \
  54     t1 = vec_sub(t1, vec_add(s1, s7)); \
  55 \
  56     t2 = vec_sl(vec_sub(s7, s3), vec_4); \
  57     t2 = vec_add(t2, vec_sl(s1, vec_3)); \
  58     t2 = vec_add(t2, vec_sl(s5, vec_2)); \
  59     t2 = vec_add(t2, vec_sub(s1, s7)); \
  60 \
  61     t3 = vec_sl(vec_sub(s5, s7), vec_4); \
  62     t3 = vec_sub(t3, vec_sl(s3, vec_3)); \
  63     t3 = vec_add(t3, vec_sl(s1, vec_2)); \
  64     t3 = vec_sub(t3, vec_add(s3, s5)); \
  65 \
  66     s0 = vec_add(t4, t0); \
  67     s1 = vec_add(t5, t1); \
  68     s2 = vec_add(t6, t2); \
  69     s3 = vec_add(t7, t3); \
  70     s4 = vec_sub(t7, t3); \
  71     s5 = vec_sub(t6, t2); \
  72     s6 = vec_sub(t5, t1); \
  73     s7 = vec_sub(t4, t0); \
  74 }while(0)
  75
  76 #define SHIFT_HOR8(s0, s1, s2, s3, s4, s5, s6, s7) \
  77 do { \
  78     s0 = vec_sra(s0, vec_3); \
  79     s1 = vec_sra(s1, vec_3); \
  80     s2 = vec_sra(s2, vec_3); \
  81     s3 = vec_sra(s3, vec_3); \
  82     s4 = vec_sra(s4, vec_3); \
  83     s5 = vec_sra(s5, vec_3); \
  84     s6 = vec_sra(s6, vec_3); \
  85     s7 = vec_sra(s7, vec_3); \
  86 }while(0)
  87
  88 #define SHIFT_VERT8(s0, s1, s2, s3, s4, s5, s6, s7) \
  89 do { \
  90     s0 = vec_sra(s0, vec_7); \
  91     s1 = vec_sra(s1, vec_7); \
  92     s2 = vec_sra(s2, vec_7); \
  93     s3 = vec_sra(s3, vec_7); \
  94     s4 = vec_sra(vec_add(s4, vec_1s), vec_7); \
  95     s5 = vec_sra(vec_add(s5, vec_1s), vec_7); \
  96     s6 = vec_sra(vec_add(s6, vec_1s), vec_7); \
  97     s7 = vec_sra(vec_add(s7, vec_1s), vec_7); \
  98 }while(0)
  99
 100 /* main steps of 4x4 transform */
 101 #define STEP4(s0, s1, s2, s3, vec_rnd) \
 102 do { \
 103     t1 = vec_add(vec_sl(s0, vec_4), s0); \
 104     t1 = vec_add(t1, vec_rnd); \
 105     t2 = vec_add(vec_sl(s2, vec_4), s2); \
 106     t0 = vec_add(t1, t2); \
 107     t1 = vec_sub(t1, t2); \
 108     t3 = vec_sl(vec_sub(s3, s1), vec_1); \
 109     t3 = vec_add(t3, vec_sl(t3, vec_2)); \
 110     t2 = vec_add(t3, vec_sl(s1, vec_5)); \
 111     t3 = vec_add(t3, vec_sl(s3, vec_3)); \
 112     t3 = vec_add(t3, vec_sl(s3, vec_2)); \
 113     s0 = vec_add(t0, t2); \
 114     s1 = vec_sub(t1, t3); \
 115     s2 = vec_add(t1, t3); \
 116     s3 = vec_sub(t0, t2); \
 117 }while (0)
 118
 119 #define SHIFT_HOR4(s0, s1, s2, s3) \
 120     s0 = vec_sra(s0, vec_3); \
 121     s1 = vec_sra(s1, vec_3); \
 122     s2 = vec_sra(s2, vec_3); \
 123     s3 = vec_sra(s3, vec_3);
 124
 125 #define SHIFT_VERT4(s0, s1, s2, s3) \
 126     s0 = vec_sra(s0, vec_7); \
 127     s1 = vec_sra(s1, vec_7); \
 128     s2 = vec_sra(s2, vec_7); \
 129     s3 = vec_sra(s3, vec_7);
 130
 131 /** Do inverse transform on 8x8 block
 132 */
 133 static void vc1_inv_trans_8x8_altivec(DCTELEM block[64])
 134 {
 135     vector signed short src0, src1, src2, src3, src4, src5, src6, src7;
 136     vector signed int s0, s1, s2, s3, s4, s5, s6, s7;
 137     vector signed int s8, s9, sA, sB, sC, sD, sE, sF;
 138     vector signed int t0, t1, t2, t3, t4, t5, t6, t7;
 139     const vector signed int vec_64 = vec_sl(vec_splat_s32(4), vec_splat_u32(4));
 140     const vector unsigned int vec_7 = vec_splat_u32(7);
 141     const vector unsigned int vec_4 = vec_splat_u32(4);
 142     const vector  signed int vec_4s = vec_splat_s32(4);
 143     const vector unsigned int vec_3 = vec_splat_u32(3);
 144     const vector unsigned int vec_2 = vec_splat_u32(2);
 145     const vector  signed int vec_1s = vec_splat_s32(1);
 146     const vector unsigned int vec_1 = vec_splat_u32(1);
 147
 148
 149     src0 = vec_ld(  0, block);
 150     src1 = vec_ld( 16, block);
 151     src2 = vec_ld( 32, block);
 152     src3 = vec_ld( 48, block);
 153     src4 = vec_ld( 64, block);
 154     src5 = vec_ld( 80, block);
 155     src6 = vec_ld( 96, block);
 156     src7 = vec_ld(112, block);
 157
 158     TRANSPOSE8(src0, src1, src2, src3, src4, src5, src6, src7);
 159     s0 = vec_unpackl(src0);
 160     s1 = vec_unpackl(src1);
 161     s2 = vec_unpackl(src2);
 162     s3 = vec_unpackl(src3);
 163     s4 = vec_unpackl(src4);
 164     s5 = vec_unpackl(src5);
 165     s6 = vec_unpackl(src6);
 166     s7 = vec_unpackl(src7);
 167     s8 = vec_unpackh(src0);
 168     s9 = vec_unpackh(src1);
 169     sA = vec_unpackh(src2);
 170     sB = vec_unpackh(src3);
 171     sC = vec_unpackh(src4);
 172     sD = vec_unpackh(src5);
 173     sE = vec_unpackh(src6);
 174     sF = vec_unpackh(src7);
 175     STEP8(s0, s1, s2, s3, s4, s5, s6, s7, vec_4s);
 176     SHIFT_HOR8(s0, s1, s2, s3, s4, s5, s6, s7);
 177     STEP8(s8, s9, sA, sB, sC, sD, sE, sF, vec_4s);
 178     SHIFT_HOR8(s8, s9, sA, sB, sC, sD, sE, sF);
 179     src0 = vec_pack(s8, s0);
 180     src1 = vec_pack(s9, s1);
 181     src2 = vec_pack(sA, s2);
 182     src3 = vec_pack(sB, s3);
 183     src4 = vec_pack(sC, s4);
 184     src5 = vec_pack(sD, s5);
 185     src6 = vec_pack(sE, s6);
 186     src7 = vec_pack(sF, s7);
 187     TRANSPOSE8(src0, src1, src2, src3, src4, src5, src6, src7);
 188
 189     s0 = vec_unpackl(src0);
 190     s1 = vec_unpackl(src1);
 191     s2 = vec_unpackl(src2);
 192     s3 = vec_unpackl(src3);
 193     s4 = vec_unpackl(src4);
 194     s5 = vec_unpackl(src5);
 195     s6 = vec_unpackl(src6);
 196     s7 = vec_unpackl(src7);
 197     s8 = vec_unpackh(src0);
 198     s9 = vec_unpackh(src1);
 199     sA = vec_unpackh(src2);
 200     sB = vec_unpackh(src3);
 201     sC = vec_unpackh(src4);
 202     sD = vec_unpackh(src5);
 203     sE = vec_unpackh(src6);
 204     sF = vec_unpackh(src7);
 205     STEP8(s0, s1, s2, s3, s4, s5, s6, s7, vec_64);
 206     SHIFT_VERT8(s0, s1, s2, s3, s4, s5, s6, s7);
 207     STEP8(s8, s9, sA, sB, sC, sD, sE, sF, vec_64);
 208     SHIFT_VERT8(s8, s9, sA, sB, sC, sD, sE, sF);
 209     src0 = vec_pack(s8, s0);
 210     src1 = vec_pack(s9, s1);
 211     src2 = vec_pack(sA, s2);
 212     src3 = vec_pack(sB, s3);
 213     src4 = vec_pack(sC, s4);
 214     src5 = vec_pack(sD, s5);
 215     src6 = vec_pack(sE, s6);
 216     src7 = vec_pack(sF, s7);
 217
 218     vec_st(src0,  0, block);
 219     vec_st(src1, 16, block);
 220     vec_st(src2, 32, block);
 221     vec_st(src3, 48, block);
 222     vec_st(src4, 64, block);
 223     vec_st(src5, 80, block);
 224     vec_st(src6, 96, block);
 225     vec_st(src7,112, block);
 226 }
 227
 228 /** Do inverse transform on 8x4 part of block
 229 */
 230 static void vc1_inv_trans_8x4_altivec(uint8_t *dest, int stride, DCTELEM *block)
 231 {
 232     vector signed short src0, src1, src2, src3, src4, src5, src6, src7;
 233     vector signed int s0, s1, s2, s3, s4, s5, s6, s7;
 234     vector signed int s8, s9, sA, sB, sC, sD, sE, sF;
 235     vector signed int t0, t1, t2, t3, t4, t5, t6, t7;
 236     const vector signed int vec_64 = vec_sl(vec_splat_s32(4), vec_splat_u32(4));
 237     const vector unsigned int vec_7 = vec_splat_u32(7);
 238     const vector unsigned int vec_5 = vec_splat_u32(5);
 239     const vector unsigned int vec_4 = vec_splat_u32(4);
 240     const vector  signed int vec_4s = vec_splat_s32(4);
 241     const vector unsigned int vec_3 = vec_splat_u32(3);
 242     const vector unsigned int vec_2 = vec_splat_u32(2);
 243     const vector unsigned int vec_1 = vec_splat_u32(1);
 244     vector unsigned char tmp;
 245     vector signed short tmp2, tmp3;
 246     vector unsigned char perm0, perm1, p0, p1, p;
 247
 248     src0 = vec_ld(  0, block);
 249     src1 = vec_ld( 16, block);
 250     src2 = vec_ld( 32, block);
 251     src3 = vec_ld( 48, block);
 252     src4 = vec_ld( 64, block);
 253     src5 = vec_ld( 80, block);
 254     src6 = vec_ld( 96, block);
 255     src7 = vec_ld(112, block);
 256
 257     TRANSPOSE8(src0, src1, src2, src3, src4, src5, src6, src7);
 258     s0 = vec_unpackl(src0);
 259     s1 = vec_unpackl(src1);
 260     s2 = vec_unpackl(src2);
 261     s3 = vec_unpackl(src3);
 262     s4 = vec_unpackl(src4);
 263     s5 = vec_unpackl(src5);
 264     s6 = vec_unpackl(src6);
 265     s7 = vec_unpackl(src7);
 266     s8 = vec_unpackh(src0);
 267     s9 = vec_unpackh(src1);
 268     sA = vec_unpackh(src2);
 269     sB = vec_unpackh(src3);
 270     sC = vec_unpackh(src4);
 271     sD = vec_unpackh(src5);
 272     sE = vec_unpackh(src6);
 273     sF = vec_unpackh(src7);
 274     STEP8(s0, s1, s2, s3, s4, s5, s6, s7, vec_4s);
 275     SHIFT_HOR8(s0, s1, s2, s3, s4, s5, s6, s7);
 276     STEP8(s8, s9, sA, sB, sC, sD, sE, sF, vec_4s);
 277     SHIFT_HOR8(s8, s9, sA, sB, sC, sD, sE, sF);
 278     src0 = vec_pack(s8, s0);
 279     src1 = vec_pack(s9, s1);
 280     src2 = vec_pack(sA, s2);
 281     src3 = vec_pack(sB, s3);
 282     src4 = vec_pack(sC, s4);
 283     src5 = vec_pack(sD, s5);
 284     src6 = vec_pack(sE, s6);
 285     src7 = vec_pack(sF, s7);
 286     TRANSPOSE8(src0, src1, src2, src3, src4, src5, src6, src7);
 287
 288     s0 = vec_unpackh(src0);
 289     s1 = vec_unpackh(src1);
 290     s2 = vec_unpackh(src2);
 291     s3 = vec_unpackh(src3);
 292     s8 = vec_unpackl(src0);
 293     s9 = vec_unpackl(src1);
 294     sA = vec_unpackl(src2);
 295     sB = vec_unpackl(src3);
 296     STEP4(s0, s1, s2, s3, vec_64);
 297     SHIFT_VERT4(s0, s1, s2, s3);
 298     STEP4(s8, s9, sA, sB, vec_64);
 299     SHIFT_VERT4(s8, s9, sA, sB);
 300     src0 = vec_pack(s0, s8);
 301     src1 = vec_pack(s1, s9);
 302     src2 = vec_pack(s2, sA);
 303     src3 = vec_pack(s3, sB);
 304
 305     p0 = vec_lvsl (0, dest);
 306     p1 = vec_lvsl (stride, dest);
 307     p = vec_splat_u8 (-1);
 308     perm0 = vec_mergeh (p, p0);
 309     perm1 = vec_mergeh (p, p1);
 310
 311 #define ADD(dest,src,perm)                                              \
 312     /* *(uint64_t *)&tmp = *(uint64_t *)dest; */                        \
 313     tmp = vec_ld (0, dest);                                             \
 314     tmp2 = (vector signed short)vec_perm (tmp, vec_splat_u8(0), perm);  \
 315     tmp3 = vec_adds (tmp2, src);                                        \
 316     tmp = vec_packsu (tmp3, tmp3);                                      \
 317     vec_ste ((vector unsigned int)tmp, 0, (unsigned int *)dest);        \
 318     vec_ste ((vector unsigned int)tmp, 4, (unsigned int *)dest);
 319
 320     ADD (dest, src0, perm0)      dest += stride;
 321     ADD (dest, src1, perm1)      dest += stride;
 322     ADD (dest, src2, perm0)      dest += stride;
 323     ADD (dest, src3, perm1)
 324 }
 325
 326
 327 void vc1dsp_init_altivec(DSPContext* dsp, AVCodecContext *avctx) {
 328     dsp->vc1_inv_trans_8x8 = vc1_inv_trans_8x8_altivec;
 329     dsp->vc1_inv_trans_8x4 = vc1_inv_trans_8x4_altivec;
 330 }