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 Libav.
   6  *
   7  * Libav 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  * Libav 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 Libav; if not, write to the Free Software
  19  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
  20  */
  21
  22 #include "libavutil/attributes.h"
  23 #include "libavutil/ppc/types_altivec.h"
  24 #include "libavutil/ppc/util_altivec.h"
  25 #include "libavcodec/vc1dsp.h"
  26
  27 // main steps of 8x8 transform
  28 #define STEP8(s0, s1, s2, s3, s4, s5, s6, s7, vec_rnd) \
  29 do { \
  30     t0 = vec_sl(vec_add(s0, s4), vec_2); \
  31     t0 = vec_add(vec_sl(t0, vec_1), t0); \
  32     t0 = vec_add(t0, vec_rnd); \
  33     t1 = vec_sl(vec_sub(s0, s4), vec_2); \
  34     t1 = vec_add(vec_sl(t1, vec_1), t1); \
  35     t1 = vec_add(t1, vec_rnd); \
  36     t2 = vec_add(vec_sl(s6, vec_2), vec_sl(s6, vec_1)); \
  37     t2 = vec_add(t2, vec_sl(s2, vec_4)); \
  38     t3 = vec_add(vec_sl(s2, vec_2), vec_sl(s2, vec_1)); \
  39     t3 = vec_sub(t3, vec_sl(s6, vec_4)); \
  40     t4 = vec_add(t0, t2); \
  41     t5 = vec_add(t1, t3); \
  42     t6 = vec_sub(t1, t3); \
  43     t7 = vec_sub(t0, t2); \
  44 \
  45     t0 = vec_sl(vec_add(s1, s3), vec_4); \
  46     t0 = vec_add(t0, vec_sl(s5, vec_3)); \
  47     t0 = vec_add(t0, vec_sl(s7, vec_2)); \
  48     t0 = vec_add(t0, vec_sub(s5, s3)); \
  49 \
  50     t1 = vec_sl(vec_sub(s1, s5), vec_4); \
  51     t1 = vec_sub(t1, vec_sl(s7, vec_3)); \
  52     t1 = vec_sub(t1, vec_sl(s3, vec_2)); \
  53     t1 = vec_sub(t1, vec_add(s1, s7)); \
  54 \
  55     t2 = vec_sl(vec_sub(s7, s3), vec_4); \
  56     t2 = vec_add(t2, vec_sl(s1, vec_3)); \
  57     t2 = vec_add(t2, vec_sl(s5, vec_2)); \
  58     t2 = vec_add(t2, vec_sub(s1, s7)); \
  59 \
  60     t3 = vec_sl(vec_sub(s5, s7), vec_4); \
  61     t3 = vec_sub(t3, vec_sl(s3, vec_3)); \
  62     t3 = vec_add(t3, vec_sl(s1, vec_2)); \
  63     t3 = vec_sub(t3, vec_add(s3, s5)); \
  64 \
  65     s0 = vec_add(t4, t0); \
  66     s1 = vec_add(t5, t1); \
  67     s2 = vec_add(t6, t2); \
  68     s3 = vec_add(t7, t3); \
  69     s4 = vec_sub(t7, t3); \
  70     s5 = vec_sub(t6, t2); \
  71     s6 = vec_sub(t5, t1); \
  72     s7 = vec_sub(t4, t0); \
  73 }while(0)
  74
  75 #define SHIFT_HOR8(s0, s1, s2, s3, s4, s5, s6, s7) \
  76 do { \
  77     s0 = vec_sra(s0, vec_3); \
  78     s1 = vec_sra(s1, vec_3); \
  79     s2 = vec_sra(s2, vec_3); \
  80     s3 = vec_sra(s3, vec_3); \
  81     s4 = vec_sra(s4, vec_3); \
  82     s5 = vec_sra(s5, vec_3); \
  83     s6 = vec_sra(s6, vec_3); \
  84     s7 = vec_sra(s7, vec_3); \
  85 }while(0)
  86
  87 #define SHIFT_VERT8(s0, s1, s2, s3, s4, s5, s6, s7) \
  88 do { \
  89     s0 = vec_sra(s0, vec_7); \
  90     s1 = vec_sra(s1, vec_7); \
  91     s2 = vec_sra(s2, vec_7); \
  92     s3 = vec_sra(s3, vec_7); \
  93     s4 = vec_sra(vec_add(s4, vec_1s), vec_7); \
  94     s5 = vec_sra(vec_add(s5, vec_1s), vec_7); \
  95     s6 = vec_sra(vec_add(s6, vec_1s), vec_7); \
  96     s7 = vec_sra(vec_add(s7, vec_1s), vec_7); \
  97 }while(0)
  98
  99 /* main steps of 4x4 transform */
 100 #define STEP4(s0, s1, s2, s3, vec_rnd) \
 101 do { \
 102     t1 = vec_add(vec_sl(s0, vec_4), s0); \
 103     t1 = vec_add(t1, vec_rnd); \
 104     t2 = vec_add(vec_sl(s2, vec_4), s2); \
 105     t0 = vec_add(t1, t2); \
 106     t1 = vec_sub(t1, t2); \
 107     t3 = vec_sl(vec_sub(s3, s1), vec_1); \
 108     t3 = vec_add(t3, vec_sl(t3, vec_2)); \
 109     t2 = vec_add(t3, vec_sl(s1, vec_5)); \
 110     t3 = vec_add(t3, vec_sl(s3, vec_3)); \
 111     t3 = vec_add(t3, vec_sl(s3, vec_2)); \
 112     s0 = vec_add(t0, t2); \
 113     s1 = vec_sub(t1, t3); \
 114     s2 = vec_add(t1, t3); \
 115     s3 = vec_sub(t0, t2); \
 116 }while (0)
 117
 118 #define SHIFT_HOR4(s0, s1, s2, s3) \
 119     s0 = vec_sra(s0, vec_3); \
 120     s1 = vec_sra(s1, vec_3); \
 121     s2 = vec_sra(s2, vec_3); \
 122     s3 = vec_sra(s3, vec_3);
 123
 124 #define SHIFT_VERT4(s0, s1, s2, s3) \
 125     s0 = vec_sra(s0, vec_7); \
 126     s1 = vec_sra(s1, vec_7); \
 127     s2 = vec_sra(s2, vec_7); \
 128     s3 = vec_sra(s3, vec_7);
 129
 130 /** Do inverse transform on 8x8 block
 131 */
 132 static void vc1_inv_trans_8x8_altivec(int16_t block[64])
 133 {
 134     vector signed short src0, src1, src2, src3, src4, src5, src6, src7;
 135     vector signed int s0, s1, s2, s3, s4, s5, s6, s7;
 136     vector signed int s8, s9, sA, sB, sC, sD, sE, sF;
 137     vector signed int t0, t1, t2, t3, t4, t5, t6, t7;
 138     const vector signed int vec_64 = vec_sl(vec_splat_s32(4), vec_splat_u32(4));
 139     const vector unsigned int vec_7 = vec_splat_u32(7);
 140     const vector unsigned int vec_4 = vec_splat_u32(4);
 141     const vector  signed int vec_4s = vec_splat_s32(4);
 142     const vector unsigned int vec_3 = vec_splat_u32(3);
 143     const vector unsigned int vec_2 = vec_splat_u32(2);
 144     const vector  signed int vec_1s = vec_splat_s32(1);
 145     const vector unsigned int vec_1 = vec_splat_u32(1);
 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     s0 = vec_unpackl(src0);
 157     s1 = vec_unpackl(src1);
 158     s2 = vec_unpackl(src2);
 159     s3 = vec_unpackl(src3);
 160     s4 = vec_unpackl(src4);
 161     s5 = vec_unpackl(src5);
 162     s6 = vec_unpackl(src6);
 163     s7 = vec_unpackl(src7);
 164     s8 = vec_unpackh(src0);
 165     s9 = vec_unpackh(src1);
 166     sA = vec_unpackh(src2);
 167     sB = vec_unpackh(src3);
 168     sC = vec_unpackh(src4);
 169     sD = vec_unpackh(src5);
 170     sE = vec_unpackh(src6);
 171     sF = vec_unpackh(src7);
 172     STEP8(s0, s1, s2, s3, s4, s5, s6, s7, vec_4s);
 173     SHIFT_HOR8(s0, s1, s2, s3, s4, s5, s6, s7);
 174     STEP8(s8, s9, sA, sB, sC, sD, sE, sF, vec_4s);
 175     SHIFT_HOR8(s8, s9, sA, sB, sC, sD, sE, sF);
 176     src0 = vec_pack(s8, s0);
 177     src1 = vec_pack(s9, s1);
 178     src2 = vec_pack(sA, s2);
 179     src3 = vec_pack(sB, s3);
 180     src4 = vec_pack(sC, s4);
 181     src5 = vec_pack(sD, s5);
 182     src6 = vec_pack(sE, s6);
 183     src7 = vec_pack(sF, s7);
 184     TRANSPOSE8(src0, src1, src2, src3, src4, src5, src6, src7);
 185
 186     s0 = vec_unpackl(src0);
 187     s1 = vec_unpackl(src1);
 188     s2 = vec_unpackl(src2);
 189     s3 = vec_unpackl(src3);
 190     s4 = vec_unpackl(src4);
 191     s5 = vec_unpackl(src5);
 192     s6 = vec_unpackl(src6);
 193     s7 = vec_unpackl(src7);
 194     s8 = vec_unpackh(src0);
 195     s9 = vec_unpackh(src1);
 196     sA = vec_unpackh(src2);
 197     sB = vec_unpackh(src3);
 198     sC = vec_unpackh(src4);
 199     sD = vec_unpackh(src5);
 200     sE = vec_unpackh(src6);
 201     sF = vec_unpackh(src7);
 202     STEP8(s0, s1, s2, s3, s4, s5, s6, s7, vec_64);
 203     SHIFT_VERT8(s0, s1, s2, s3, s4, s5, s6, s7);
 204     STEP8(s8, s9, sA, sB, sC, sD, sE, sF, vec_64);
 205     SHIFT_VERT8(s8, s9, sA, sB, sC, sD, sE, sF);
 206     src0 = vec_pack(s8, s0);
 207     src1 = vec_pack(s9, s1);
 208     src2 = vec_pack(sA, s2);
 209     src3 = vec_pack(sB, s3);
 210     src4 = vec_pack(sC, s4);
 211     src5 = vec_pack(sD, s5);
 212     src6 = vec_pack(sE, s6);
 213     src7 = vec_pack(sF, s7);
 214
 215     vec_st(src0,  0, block);
 216     vec_st(src1, 16, block);
 217     vec_st(src2, 32, block);
 218     vec_st(src3, 48, block);
 219     vec_st(src4, 64, block);
 220     vec_st(src5, 80, block);
 221     vec_st(src6, 96, block);
 222     vec_st(src7,112, block);
 223 }
 224
 225 /** Do inverse transform on 8x4 part of block
 226 */
 227 static void vc1_inv_trans_8x4_altivec(uint8_t *dest, int stride, int16_t *block)
 228 {
 229     vector signed short src0, src1, src2, src3, src4, src5, src6, src7;
 230     vector signed int s0, s1, s2, s3, s4, s5, s6, s7;
 231     vector signed int s8, s9, sA, sB, sC, sD, sE, sF;
 232     vector signed int t0, t1, t2, t3, t4, t5, t6, t7;
 233     const vector signed int vec_64 = vec_sl(vec_splat_s32(4), vec_splat_u32(4));
 234     const vector unsigned int vec_7 = vec_splat_u32(7);
 235     const vector unsigned int vec_5 = vec_splat_u32(5);
 236     const vector unsigned int vec_4 = vec_splat_u32(4);
 237     const vector  signed int vec_4s = vec_splat_s32(4);
 238     const vector unsigned int vec_3 = vec_splat_u32(3);
 239     const vector unsigned int vec_2 = vec_splat_u32(2);
 240     const vector unsigned int vec_1 = vec_splat_u32(1);
 241     vector unsigned char tmp;
 242     vector signed short tmp2, tmp3;
 243     vector unsigned char perm0, perm1, p0, p1, p;
 244
 245     src0 = vec_ld(  0, block);
 246     src1 = vec_ld( 16, block);
 247     src2 = vec_ld( 32, block);
 248     src3 = vec_ld( 48, block);
 249     src4 = vec_ld( 64, block);
 250     src5 = vec_ld( 80, block);
 251     src6 = vec_ld( 96, block);
 252     src7 = vec_ld(112, block);
 253
 254     TRANSPOSE8(src0, src1, src2, src3, src4, src5, src6, src7);
 255     s0 = vec_unpackl(src0);
 256     s1 = vec_unpackl(src1);
 257     s2 = vec_unpackl(src2);
 258     s3 = vec_unpackl(src3);
 259     s4 = vec_unpackl(src4);
 260     s5 = vec_unpackl(src5);
 261     s6 = vec_unpackl(src6);
 262     s7 = vec_unpackl(src7);
 263     s8 = vec_unpackh(src0);
 264     s9 = vec_unpackh(src1);
 265     sA = vec_unpackh(src2);
 266     sB = vec_unpackh(src3);
 267     sC = vec_unpackh(src4);
 268     sD = vec_unpackh(src5);
 269     sE = vec_unpackh(src6);
 270     sF = vec_unpackh(src7);
 271     STEP8(s0, s1, s2, s3, s4, s5, s6, s7, vec_4s);
 272     SHIFT_HOR8(s0, s1, s2, s3, s4, s5, s6, s7);
 273     STEP8(s8, s9, sA, sB, sC, sD, sE, sF, vec_4s);
 274     SHIFT_HOR8(s8, s9, sA, sB, sC, sD, sE, sF);
 275     src0 = vec_pack(s8, s0);
 276     src1 = vec_pack(s9, s1);
 277     src2 = vec_pack(sA, s2);
 278     src3 = vec_pack(sB, s3);
 279     src4 = vec_pack(sC, s4);
 280     src5 = vec_pack(sD, s5);
 281     src6 = vec_pack(sE, s6);
 282     src7 = vec_pack(sF, s7);
 283     TRANSPOSE8(src0, src1, src2, src3, src4, src5, src6, src7);
 284
 285     s0 = vec_unpackh(src0);
 286     s1 = vec_unpackh(src1);
 287     s2 = vec_unpackh(src2);
 288     s3 = vec_unpackh(src3);
 289     s8 = vec_unpackl(src0);
 290     s9 = vec_unpackl(src1);
 291     sA = vec_unpackl(src2);
 292     sB = vec_unpackl(src3);
 293     STEP4(s0, s1, s2, s3, vec_64);
 294     SHIFT_VERT4(s0, s1, s2, s3);
 295     STEP4(s8, s9, sA, sB, vec_64);
 296     SHIFT_VERT4(s8, s9, sA, sB);
 297     src0 = vec_pack(s0, s8);
 298     src1 = vec_pack(s1, s9);
 299     src2 = vec_pack(s2, sA);
 300     src3 = vec_pack(s3, sB);
 301
 302     p0 = vec_lvsl (0, dest);
 303     p1 = vec_lvsl (stride, dest);
 304     p = vec_splat_u8 (-1);
 305     perm0 = vec_mergeh (p, p0);
 306     perm1 = vec_mergeh (p, p1);
 307
 308 #define ADD(dest,src,perm)                                              \
 309     /* *(uint64_t *)&tmp = *(uint64_t *)dest; */                        \
 310     tmp = vec_ld (0, dest);                                             \
 311     tmp2 = (vector signed short)vec_perm (tmp, vec_splat_u8(0), perm);  \
 312     tmp3 = vec_adds (tmp2, src);                                        \
 313     tmp = vec_packsu (tmp3, tmp3);                                      \
 314     vec_ste ((vector unsigned int)tmp, 0, (unsigned int *)dest);        \
 315     vec_ste ((vector unsigned int)tmp, 4, (unsigned int *)dest);
 316
 317     ADD (dest, src0, perm0)      dest += stride;
 318     ADD (dest, src1, perm1)      dest += stride;
 319     ADD (dest, src2, perm0)      dest += stride;
 320     ADD (dest, src3, perm1)
 321 }
 322
 323 #define PUT_OP_U8_ALTIVEC(d, s, dst) d = s
 324 #define AVG_OP_U8_ALTIVEC(d, s, dst) d = vec_avg(dst, s)
 325
 326 #define OP_U8_ALTIVEC                          PUT_OP_U8_ALTIVEC
 327 #define PREFIX_no_rnd_vc1_chroma_mc8_altivec   put_no_rnd_vc1_chroma_mc8_altivec
 328 #include "h264chroma_template.c"
 329 #undef OP_U8_ALTIVEC
 330 #undef PREFIX_no_rnd_vc1_chroma_mc8_altivec
 331
 332 #define OP_U8_ALTIVEC                          AVG_OP_U8_ALTIVEC
 333 #define PREFIX_no_rnd_vc1_chroma_mc8_altivec   avg_no_rnd_vc1_chroma_mc8_altivec
 334 #include "h264chroma_template.c"
 335 #undef OP_U8_ALTIVEC
 336 #undef PREFIX_no_rnd_vc1_chroma_mc8_altivec
 337
 338 av_cold void ff_vc1dsp_init_altivec(VC1DSPContext *dsp)
 339 {
 340     if (!(av_get_cpu_flags() & AV_CPU_FLAG_ALTIVEC))
 341         return;
 342
 343     dsp->vc1_inv_trans_8x8 = vc1_inv_trans_8x8_altivec;
 344     dsp->vc1_inv_trans_8x4 = vc1_inv_trans_8x4_altivec;
 345     dsp->put_no_rnd_vc1_chroma_pixels_tab[0] = put_no_rnd_vc1_chroma_mc8_altivec;
 346     dsp->avg_no_rnd_vc1_chroma_pixels_tab[0] = avg_no_rnd_vc1_chroma_mc8_altivec;
 347 }