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 "util_altivec.h"
  25
  26 // main steps of 8x8 transform
  27 #define STEP8(s0, s1, s2, s3, s4, s5, s6, s7, vec_rnd) \
  28 do { \
  29     t0 = vec_sl(vec_add(s0, s4), vec_2); \
  30     t0 = vec_add(vec_sl(t0, vec_1), t0); \
  31     t0 = vec_add(t0, vec_rnd); \
  32     t1 = vec_sl(vec_sub(s0, s4), vec_2); \
  33     t1 = vec_add(vec_sl(t1, vec_1), t1); \
  34     t1 = vec_add(t1, vec_rnd); \
  35     t2 = vec_add(vec_sl(s6, vec_2), vec_sl(s6, vec_1)); \
  36     t2 = vec_add(t2, vec_sl(s2, vec_4)); \
  37     t3 = vec_add(vec_sl(s2, vec_2), vec_sl(s2, vec_1)); \
  38     t3 = vec_sub(t3, vec_sl(s6, vec_4)); \
  39     t4 = vec_add(t0, t2); \
  40     t5 = vec_add(t1, t3); \
  41     t6 = vec_sub(t1, t3); \
  42     t7 = vec_sub(t0, t2); \
  43 \
  44     t0 = vec_sl(vec_add(s1, s3), vec_4); \
  45     t0 = vec_add(t0, vec_sl(s5, vec_3)); \
  46     t0 = vec_add(t0, vec_sl(s7, vec_2)); \
  47     t0 = vec_add(t0, vec_sub(s5, s3)); \
  48 \
  49     t1 = vec_sl(vec_sub(s1, s5), vec_4); \
  50     t1 = vec_sub(t1, vec_sl(s7, vec_3)); \
  51     t1 = vec_sub(t1, vec_sl(s3, vec_2)); \
  52     t1 = vec_sub(t1, vec_add(s1, s7)); \
  53 \
  54     t2 = vec_sl(vec_sub(s7, s3), vec_4); \
  55     t2 = vec_add(t2, vec_sl(s1, vec_3)); \
  56     t2 = vec_add(t2, vec_sl(s5, vec_2)); \
  57     t2 = vec_add(t2, vec_sub(s1, s7)); \
  58 \
  59     t3 = vec_sl(vec_sub(s5, s7), vec_4); \
  60     t3 = vec_sub(t3, vec_sl(s3, vec_3)); \
  61     t3 = vec_add(t3, vec_sl(s1, vec_2)); \
  62     t3 = vec_sub(t3, vec_add(s3, s5)); \
  63 \
  64     s0 = vec_add(t4, t0); \
  65     s1 = vec_add(t5, t1); \
  66     s2 = vec_add(t6, t2); \
  67     s3 = vec_add(t7, t3); \
  68     s4 = vec_sub(t7, t3); \
  69     s5 = vec_sub(t6, t2); \
  70     s6 = vec_sub(t5, t1); \
  71     s7 = vec_sub(t4, t0); \
  72 }while(0)
  73
  74 #define SHIFT_HOR8(s0, s1, s2, s3, s4, s5, s6, s7) \
  75 do { \
  76     s0 = vec_sra(s0, vec_3); \
  77     s1 = vec_sra(s1, vec_3); \
  78     s2 = vec_sra(s2, vec_3); \
  79     s3 = vec_sra(s3, vec_3); \
  80     s4 = vec_sra(s4, vec_3); \
  81     s5 = vec_sra(s5, vec_3); \
  82     s6 = vec_sra(s6, vec_3); \
  83     s7 = vec_sra(s7, vec_3); \
  84 }while(0)
  85
  86 #define SHIFT_VERT8(s0, s1, s2, s3, s4, s5, s6, s7) \
  87 do { \
  88     s0 = vec_sra(s0, vec_7); \
  89     s1 = vec_sra(s1, vec_7); \
  90     s2 = vec_sra(s2, vec_7); \
  91     s3 = vec_sra(s3, vec_7); \
  92     s4 = vec_sra(vec_add(s4, vec_1s), vec_7); \
  93     s5 = vec_sra(vec_add(s5, vec_1s), vec_7); \
  94     s6 = vec_sra(vec_add(s6, vec_1s), vec_7); \
  95     s7 = vec_sra(vec_add(s7, vec_1s), vec_7); \
  96 }while(0)
  97
  98 /* main steps of 4x4 transform */
  99 #define STEP4(s0, s1, s2, s3, vec_rnd) \
 100 do { \
 101     t1 = vec_add(vec_sl(s0, vec_4), s0); \
 102     t1 = vec_add(t1, vec_rnd); \
 103     t2 = vec_add(vec_sl(s2, vec_4), s2); \
 104     t0 = vec_add(t1, t2); \
 105     t1 = vec_sub(t1, t2); \
 106     t3 = vec_sl(vec_sub(s3, s1), vec_1); \
 107     t3 = vec_add(t3, vec_sl(t3, vec_2)); \
 108     t2 = vec_add(t3, vec_sl(s1, vec_5)); \
 109     t3 = vec_add(t3, vec_sl(s3, vec_3)); \
 110     t3 = vec_add(t3, vec_sl(s3, vec_2)); \
 111     s0 = vec_add(t0, t2); \
 112     s1 = vec_sub(t1, t3); \
 113     s2 = vec_add(t1, t3); \
 114     s3 = vec_sub(t0, t2); \
 115 }while (0)
 116
 117 #define SHIFT_HOR4(s0, s1, s2, s3) \
 118     s0 = vec_sra(s0, vec_3); \
 119     s1 = vec_sra(s1, vec_3); \
 120     s2 = vec_sra(s2, vec_3); \
 121     s3 = vec_sra(s3, vec_3);
 122
 123 #define SHIFT_VERT4(s0, s1, s2, s3) \
 124     s0 = vec_sra(s0, vec_7); \
 125     s1 = vec_sra(s1, vec_7); \
 126     s2 = vec_sra(s2, vec_7); \
 127     s3 = vec_sra(s3, vec_7);
 128
 129 /** Do inverse transform on 8x8 block
 130 */
 131 static void vc1_inv_trans_8x8_altivec(DCTELEM block[64])
 132 {
 133     vector signed short src0, src1, src2, src3, src4, src5, src6, src7;
 134     vector signed int s0, s1, s2, s3, s4, s5, s6, s7;
 135     vector signed int s8, s9, sA, sB, sC, sD, sE, sF;
 136     vector signed int t0, t1, t2, t3, t4, t5, t6, t7;
 137     const vector signed int vec_64 = vec_sl(vec_splat_s32(4), vec_splat_u32(4));
 138     const vector unsigned int vec_7 = vec_splat_u32(7);
 139     const vector unsigned int vec_4 = vec_splat_u32(4);
 140     const vector  signed int vec_4s = vec_splat_s32(4);
 141     const vector unsigned int vec_3 = vec_splat_u32(3);
 142     const vector unsigned int vec_2 = vec_splat_u32(2);
 143     const vector  signed int vec_1s = vec_splat_s32(1);
 144     const vector unsigned int vec_1 = vec_splat_u32(1);
 145
 146
 147     src0 = vec_ld(  0, block);
 148     src1 = vec_ld( 16, block);
 149     src2 = vec_ld( 32, block);
 150     src3 = vec_ld( 48, block);
 151     src4 = vec_ld( 64, block);
 152     src5 = vec_ld( 80, block);
 153     src6 = vec_ld( 96, block);
 154     src7 = vec_ld(112, block);
 155
 156     TRANSPOSE8(src0, src1, src2, src3, src4, src5, src6, src7);
 157     s0 = vec_unpackl(src0);
 158     s1 = vec_unpackl(src1);
 159     s2 = vec_unpackl(src2);
 160     s3 = vec_unpackl(src3);
 161     s4 = vec_unpackl(src4);
 162     s5 = vec_unpackl(src5);
 163     s6 = vec_unpackl(src6);
 164     s7 = vec_unpackl(src7);
 165     s8 = vec_unpackh(src0);
 166     s9 = vec_unpackh(src1);
 167     sA = vec_unpackh(src2);
 168     sB = vec_unpackh(src3);
 169     sC = vec_unpackh(src4);
 170     sD = vec_unpackh(src5);
 171     sE = vec_unpackh(src6);
 172     sF = vec_unpackh(src7);
 173     STEP8(s0, s1, s2, s3, s4, s5, s6, s7, vec_4s);
 174     SHIFT_HOR8(s0, s1, s2, s3, s4, s5, s6, s7);
 175     STEP8(s8, s9, sA, sB, sC, sD, sE, sF, vec_4s);
 176     SHIFT_HOR8(s8, s9, sA, sB, sC, sD, sE, sF);
 177     src0 = vec_pack(s8, s0);
 178     src1 = vec_pack(s9, s1);
 179     src2 = vec_pack(sA, s2);
 180     src3 = vec_pack(sB, s3);
 181     src4 = vec_pack(sC, s4);
 182     src5 = vec_pack(sD, s5);
 183     src6 = vec_pack(sE, s6);
 184     src7 = vec_pack(sF, s7);
 185     TRANSPOSE8(src0, src1, src2, src3, src4, src5, src6, src7);
 186
 187     s0 = vec_unpackl(src0);
 188     s1 = vec_unpackl(src1);
 189     s2 = vec_unpackl(src2);
 190     s3 = vec_unpackl(src3);
 191     s4 = vec_unpackl(src4);
 192     s5 = vec_unpackl(src5);
 193     s6 = vec_unpackl(src6);
 194     s7 = vec_unpackl(src7);
 195     s8 = vec_unpackh(src0);
 196     s9 = vec_unpackh(src1);
 197     sA = vec_unpackh(src2);
 198     sB = vec_unpackh(src3);
 199     sC = vec_unpackh(src4);
 200     sD = vec_unpackh(src5);
 201     sE = vec_unpackh(src6);
 202     sF = vec_unpackh(src7);
 203     STEP8(s0, s1, s2, s3, s4, s5, s6, s7, vec_64);
 204     SHIFT_VERT8(s0, s1, s2, s3, s4, s5, s6, s7);
 205     STEP8(s8, s9, sA, sB, sC, sD, sE, sF, vec_64);
 206     SHIFT_VERT8(s8, s9, sA, sB, sC, sD, sE, sF);
 207     src0 = vec_pack(s8, s0);
 208     src1 = vec_pack(s9, s1);
 209     src2 = vec_pack(sA, s2);
 210     src3 = vec_pack(sB, s3);
 211     src4 = vec_pack(sC, s4);
 212     src5 = vec_pack(sD, s5);
 213     src6 = vec_pack(sE, s6);
 214     src7 = vec_pack(sF, s7);
 215
 216     vec_st(src0,  0, block);
 217     vec_st(src1, 16, block);
 218     vec_st(src2, 32, block);
 219     vec_st(src3, 48, block);
 220     vec_st(src4, 64, block);
 221     vec_st(src5, 80, block);
 222     vec_st(src6, 96, block);
 223     vec_st(src7,112, block);
 224 }
 225
 226 /** Do inverse transform on 8x4 part of block
 227 */
 228 static void vc1_inv_trans_8x4_altivec(uint8_t *dest, int stride, DCTELEM *block)
 229 {
 230     vector signed short src0, src1, src2, src3, src4, src5, src6, src7;
 231     vector signed int s0, s1, s2, s3, s4, s5, s6, s7;
 232     vector signed int s8, s9, sA, sB, sC, sD, sE, sF;
 233     vector signed int t0, t1, t2, t3, t4, t5, t6, t7;
 234     const vector signed int vec_64 = vec_sl(vec_splat_s32(4), vec_splat_u32(4));
 235     const vector unsigned int vec_7 = vec_splat_u32(7);
 236     const vector unsigned int vec_5 = vec_splat_u32(5);
 237     const vector unsigned int vec_4 = vec_splat_u32(4);
 238     const vector  signed int vec_4s = vec_splat_s32(4);
 239     const vector unsigned int vec_3 = vec_splat_u32(3);
 240     const vector unsigned int vec_2 = vec_splat_u32(2);
 241     const vector unsigned int vec_1 = vec_splat_u32(1);
 242     vector unsigned char tmp;
 243     vector signed short tmp2, tmp3;
 244     vector unsigned char perm0, perm1, p0, p1, p;
 245
 246     src0 = vec_ld(  0, block);
 247     src1 = vec_ld( 16, block);
 248     src2 = vec_ld( 32, block);
 249     src3 = vec_ld( 48, block);
 250     src4 = vec_ld( 64, block);
 251     src5 = vec_ld( 80, block);
 252     src6 = vec_ld( 96, block);
 253     src7 = vec_ld(112, block);
 254
 255     TRANSPOSE8(src0, src1, src2, src3, src4, src5, src6, src7);
 256     s0 = vec_unpackl(src0);
 257     s1 = vec_unpackl(src1);
 258     s2 = vec_unpackl(src2);
 259     s3 = vec_unpackl(src3);
 260     s4 = vec_unpackl(src4);
 261     s5 = vec_unpackl(src5);
 262     s6 = vec_unpackl(src6);
 263     s7 = vec_unpackl(src7);
 264     s8 = vec_unpackh(src0);
 265     s9 = vec_unpackh(src1);
 266     sA = vec_unpackh(src2);
 267     sB = vec_unpackh(src3);
 268     sC = vec_unpackh(src4);
 269     sD = vec_unpackh(src5);
 270     sE = vec_unpackh(src6);
 271     sF = vec_unpackh(src7);
 272     STEP8(s0, s1, s2, s3, s4, s5, s6, s7, vec_4s);
 273     SHIFT_HOR8(s0, s1, s2, s3, s4, s5, s6, s7);
 274     STEP8(s8, s9, sA, sB, sC, sD, sE, sF, vec_4s);
 275     SHIFT_HOR8(s8, s9, sA, sB, sC, sD, sE, sF);
 276     src0 = vec_pack(s8, s0);
 277     src1 = vec_pack(s9, s1);
 278     src2 = vec_pack(sA, s2);
 279     src3 = vec_pack(sB, s3);
 280     src4 = vec_pack(sC, s4);
 281     src5 = vec_pack(sD, s5);
 282     src6 = vec_pack(sE, s6);
 283     src7 = vec_pack(sF, s7);
 284     TRANSPOSE8(src0, src1, src2, src3, src4, src5, src6, src7);
 285
 286     s0 = vec_unpackh(src0);
 287     s1 = vec_unpackh(src1);
 288     s2 = vec_unpackh(src2);
 289     s3 = vec_unpackh(src3);
 290     s8 = vec_unpackl(src0);
 291     s9 = vec_unpackl(src1);
 292     sA = vec_unpackl(src2);
 293     sB = vec_unpackl(src3);
 294     STEP4(s0, s1, s2, s3, vec_64);
 295     SHIFT_VERT4(s0, s1, s2, s3);
 296     STEP4(s8, s9, sA, sB, vec_64);
 297     SHIFT_VERT4(s8, s9, sA, sB);
 298     src0 = vec_pack(s0, s8);
 299     src1 = vec_pack(s1, s9);
 300     src2 = vec_pack(s2, sA);
 301     src3 = vec_pack(s3, sB);
 302
 303     p0 = vec_lvsl (0, dest);
 304     p1 = vec_lvsl (stride, dest);
 305     p = vec_splat_u8 (-1);
 306     perm0 = vec_mergeh (p, p0);
 307     perm1 = vec_mergeh (p, p1);
 308
 309 #define ADD(dest,src,perm)                                              \
 310     /* *(uint64_t *)&tmp = *(uint64_t *)dest; */                        \
 311     tmp = vec_ld (0, dest);                                             \
 312     tmp2 = (vector signed short)vec_perm (tmp, vec_splat_u8(0), perm);  \
 313     tmp3 = vec_adds (tmp2, src);                                        \
 314     tmp = vec_packsu (tmp3, tmp3);                                      \
 315     vec_ste ((vector unsigned int)tmp, 0, (unsigned int *)dest);        \
 316     vec_ste ((vector unsigned int)tmp, 4, (unsigned int *)dest);
 317
 318     ADD (dest, src0, perm0)      dest += stride;
 319     ADD (dest, src1, perm1)      dest += stride;
 320     ADD (dest, src2, perm0)      dest += stride;
 321     ADD (dest, src3, perm1)
 322 }
 323
 324
 325 void vc1dsp_init_altivec(DSPContext* dsp, AVCodecContext *avctx) {
 326     dsp->vc1_inv_trans_8x8 = vc1_inv_trans_8x8_altivec;
 327     dsp->vc1_inv_trans_8x4 = vc1_inv_trans_8x4_altivec;
 328 }