h264chroma_template.c

   1 /*
   2  * Copyright (c) 2004 Romain Dolbeau <romain@dolbeau.org>
   3  *
   4  * This file is part of FFmpeg.
   5  *
   6  * FFmpeg is free software; you can redistribute it and/or
   7  * modify it under the terms of the GNU Lesser General Public
   8  * License as published by the Free Software Foundation; either
   9  * version 2.1 of the License, or (at your option) any later version.
  10  *
  11  * FFmpeg is distributed in the hope that it will be useful,
  12  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  13  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  14  * Lesser General Public License for more details.
  15  *
  16  * You should have received a copy of the GNU Lesser General Public
  17  * License along with FFmpeg; if not, write to the Free Software
  18  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
  19  */
  20
  21 #include "libavutil/mem.h"
  22 #include "libavutil/ppc/types_altivec.h"
  23 #include "libavutil/ppc/util_altivec.h"
  24
  25 /* this code assume that stride % 16 == 0 */
  26
  27 #define CHROMA_MC8_ALTIVEC_CORE(BIAS1, BIAS2) \
  28         vsrc2ssH = (vec_s16)VEC_MERGEH(zero_u8v,(vec_u8)vsrc2uc);\
  29         vsrc3ssH = (vec_s16)VEC_MERGEH(zero_u8v,(vec_u8)vsrc3uc);\
  30 \
  31         psum = vec_mladd(vA, vsrc0ssH, BIAS1);\
  32         psum = vec_mladd(vB, vsrc1ssH, psum);\
  33         psum = vec_mladd(vC, vsrc2ssH, psum);\
  34         psum = vec_mladd(vD, vsrc3ssH, psum);\
  35         psum = BIAS2(psum);\
  36         psum = vec_sr(psum, v6us);\
  37 \
  38         vdst = vec_ld(0, dst);\
  39         ppsum = (vec_u8)vec_pack(psum, psum);\
  40         vfdst = vec_perm(vdst, ppsum, fperm);\
  41 \
  42         OP_U8_ALTIVEC(fsum, vfdst, vdst);\
  43 \
  44         vec_st(fsum, 0, dst);\
  45 \
  46         vsrc0ssH = vsrc2ssH;\
  47         vsrc1ssH = vsrc3ssH;\
  48 \
  49         dst += stride;\
  50         src += stride;
  51
  52 #define CHROMA_MC8_ALTIVEC_CORE_SIMPLE \
  53 \
  54         vsrc0ssH = (vec_s16)VEC_MERGEH(zero_u8v,(vec_u8)vsrc0uc);\
  55         vsrc1ssH = (vec_s16)VEC_MERGEH(zero_u8v,(vec_u8)vsrc1uc);\
  56 \
  57         psum = vec_mladd(vA, vsrc0ssH, v32ss);\
  58         psum = vec_mladd(vE, vsrc1ssH, psum);\
  59         psum = vec_sr(psum, v6us);\
  60 \
  61         vdst = vec_ld(0, dst);\
  62         ppsum = (vec_u8)vec_pack(psum, psum);\
  63         vfdst = vec_perm(vdst, ppsum, fperm);\
  64 \
  65         OP_U8_ALTIVEC(fsum, vfdst, vdst);\
  66 \
  67         vec_st(fsum, 0, dst);\
  68 \
  69         dst += stride;\
  70         src += stride;
  71
  72 #define noop(a) a
  73 #define add28(a) vec_add(v28ss, a)
  74
  75 #if HAVE_BIGENDIAN
  76 #define GET_VSRC1(vs0, off, b, perm0, s){    \
  77     vec_u8 vsrcCuc, vsrcDuc;                 \
  78     vsrcCuc = vec_ld(off, s);                \
  79     if (loadSecond){                         \
  80         vsrcDuc = vec_ld(off + b, s);        \
  81     } else                                   \
  82         vsrcDuc = vsrcCuc;                   \
  83                                              \
  84     vs0 = vec_perm(vsrcCuc, vsrcDuc, perm0); \
  85 }
  86 #define GET_VSRC(vs0, vs1, off, b, perm0, perm1, s){ \
  87     vec_u8 vsrcCuc, vsrcDuc;                         \
  88     vsrcCuc = vec_ld(off, s);                        \
  89     if (loadSecond){                                 \
  90         vsrcDuc = vec_ld(off + b, s);                \
  91     } else                                           \
  92         vsrcDuc = vsrcCuc;                           \
  93                                                      \
  94     vs0 = vec_perm(vsrcCuc, vsrcDuc, perm0);         \
  95     if (reallyBadAlign){                             \
  96         vs1 = vsrcDuc;                               \
  97     } else                                           \
  98         vs1 = vec_perm(vsrcCuc, vsrcDuc, perm1);     \
  99  }
 100
 101 #else
 102
 103 #define GET_VSRC1(vs0, off, b, perm0, s){            \
 104     vs0 = vec_vsx_ld(off, s);                        \
 105  }
 106 #define GET_VSRC(vs0, vs1, off, b, perm0, perm1, s){ \
 107     vs0 = vec_vsx_ld(off, s);                        \
 108     vs1 = vec_vsx_ld(off + 1, s);                    \
 109  }
 110 #endif /* HAVE_BIGENDIAN */
 111
 112 #ifdef PREFIX_h264_chroma_mc8_altivec
 113 static void PREFIX_h264_chroma_mc8_altivec(uint8_t * dst, uint8_t * src,
 114                                     int stride, int h, int x, int y) {
 115     DECLARE_ALIGNED(16, signed int, ABCD)[4] =
 116                         {((8 - x) * (8 - y)),
 117                          ((    x) * (8 - y)),
 118                          ((8 - x) * (    y)),
 119                          ((    x) * (    y))};
 120     register int i;
 121     vec_u8 fperm;
 122     LOAD_ZERO;
 123     const vec_s32 vABCD = vec_ld(0, ABCD);
 124     const vec_s16 vA = VEC_SPLAT16(vABCD, 1);
 125     const vec_s16 vB = VEC_SPLAT16(vABCD, 3);
 126     const vec_s16 vC = VEC_SPLAT16(vABCD, 5);
 127     const vec_s16 vD = VEC_SPLAT16(vABCD, 7);
 128     const vec_s16 v32ss = vec_sl(vec_splat_s16(1),vec_splat_u16(5));
 129     const vec_u16 v6us = vec_splat_u16(6);
 130
 131     vec_u8 vsrcperm0, vsrcperm1;
 132     vec_u8 vsrc0uc, vsrc1uc;
 133     vec_s16 vsrc0ssH, vsrc1ssH;
 134     vec_u8 vsrc2uc, vsrc3uc;
 135     vec_s16 vsrc2ssH, vsrc3ssH, psum;
 136     vec_u8 vdst, ppsum, vfdst, fsum;
 137 #if HAVE_BIGENDIAN
 138     register int loadSecond = (((unsigned long)src) % 16) <= 7 ? 0 : 1;
 139     register int reallyBadAlign = (((unsigned long)src) % 16) == 15 ? 1 : 0;
 140     vsrcperm0 = vec_lvsl(0, src);
 141     vsrcperm1 = vec_lvsl(1, src);
 142 #endif
 143
 144     if (((unsigned long)dst) % 16 == 0) {
 145         fperm = (vec_u8){0x10, 0x11, 0x12, 0x13,
 146                          0x14, 0x15, 0x16, 0x17,
 147                          0x08, 0x09, 0x0A, 0x0B,
 148                          0x0C, 0x0D, 0x0E, 0x0F};
 149     } else {
 150         fperm = (vec_u8){0x00, 0x01, 0x02, 0x03,
 151                          0x04, 0x05, 0x06, 0x07,
 152                          0x18, 0x19, 0x1A, 0x1B,
 153                          0x1C, 0x1D, 0x1E, 0x1F};
 154     }
 155
 156     GET_VSRC(vsrc0uc, vsrc1uc, 0, 16, vsrcperm0, vsrcperm1, src);
 157
 158     vsrc0ssH = (vec_s16)VEC_MERGEH(zero_u8v,(vec_u8)vsrc0uc);
 159     vsrc1ssH = (vec_s16)VEC_MERGEH(zero_u8v,(vec_u8)vsrc1uc);
 160
 161     if (ABCD[3]) {
 162         for (i = 0 ; i < h ; i++) {
 163             GET_VSRC(vsrc2uc, vsrc3uc, stride, 16, vsrcperm0, vsrcperm1, src);
 164             CHROMA_MC8_ALTIVEC_CORE(v32ss, noop);
 165         }
 166     } else {
 167         const vec_s16 vE = vec_add(vB, vC);
 168         if (ABCD[2]) { // x == 0 B == 0
 169             for (i = 0 ; i < h ; i++) {
 170                 GET_VSRC1(vsrc1uc, stride, 15, vsrcperm0, src);
 171                 CHROMA_MC8_ALTIVEC_CORE_SIMPLE;
 172                 vsrc0uc = vsrc1uc;
 173             }
 174         } else { // y == 0 C == 0
 175             for (i = 0 ; i < h ; i++) {
 176                GET_VSRC(vsrc0uc, vsrc1uc, 0, 15, vsrcperm0, vsrcperm1, src);
 177                CHROMA_MC8_ALTIVEC_CORE_SIMPLE;
 178             }
 179         }
 180     }
 181 }
 182 #endif
 183
 184 /* this code assume that stride % 16 == 0 */
 185 #ifdef PREFIX_no_rnd_vc1_chroma_mc8_altivec
 186 static void PREFIX_no_rnd_vc1_chroma_mc8_altivec(uint8_t * dst, uint8_t * src, int stride, int h, int x, int y) {
 187    DECLARE_ALIGNED(16, signed int, ABCD)[4] =
 188                         {((8 - x) * (8 - y)),
 189                          ((    x) * (8 - y)),
 190                          ((8 - x) * (    y)),
 191                          ((    x) * (    y))};
 192     register int i;
 193     vec_u8 fperm;
 194     LOAD_ZERO;
 195     const vec_s32 vABCD = vec_ld(0, ABCD);
 196     const vec_s16 vA = VEC_SPLAT16(vABCD, 1);
 197     const vec_s16 vB = VEC_SPLAT16(vABCD, 3);
 198     const vec_s16 vC = VEC_SPLAT16(vABCD, 5);
 199     const vec_s16 vD = VEC_SPLAT16(vABCD, 7);
 200     const vec_s16 v28ss = vec_sub(vec_sl(vec_splat_s16(1),vec_splat_u16(5)),vec_splat_s16(4));
 201     const vec_u16 v6us  = vec_splat_u16(6);
 202
 203     vec_u8 vsrcperm0, vsrcperm1;
 204     vec_u8 vsrc0uc, vsrc1uc;
 205     vec_s16 vsrc0ssH, vsrc1ssH;
 206     vec_u8 vsrc2uc, vsrc3uc;
 207     vec_s16 vsrc2ssH, vsrc3ssH, psum;
 208     vec_u8 vdst, ppsum, vfdst, fsum;
 209 #if HAVE_BIGENDIAN
 210     register int loadSecond     = (((unsigned long)src) % 16) <= 7 ? 0 : 1;
 211     register int reallyBadAlign = (((unsigned long)src) % 16) == 15 ? 1 : 0;
 212     vsrcperm0 = vec_lvsl(0, src);
 213     vsrcperm1 = vec_lvsl(1, src);
 214 #endif
 215
 216     if (((unsigned long)dst) % 16 == 0) {
 217         fperm = (vec_u8){0x10, 0x11, 0x12, 0x13,
 218                          0x14, 0x15, 0x16, 0x17,
 219                          0x08, 0x09, 0x0A, 0x0B,
 220                          0x0C, 0x0D, 0x0E, 0x0F};
 221     } else {
 222         fperm = (vec_u8){0x00, 0x01, 0x02, 0x03,
 223                          0x04, 0x05, 0x06, 0x07,
 224                          0x18, 0x19, 0x1A, 0x1B,
 225                          0x1C, 0x1D, 0x1E, 0x1F};
 226     }
 227
 228     GET_VSRC(vsrc0uc, vsrc1uc, 0, 16, vsrcperm0, vsrcperm1, src);
 229
 230     vsrc0ssH = (vec_s16)VEC_MERGEH(zero_u8v, (vec_u8)vsrc0uc);
 231     vsrc1ssH = (vec_s16)VEC_MERGEH(zero_u8v, (vec_u8)vsrc1uc);
 232
 233     for (i = 0 ; i < h ; i++) {
 234         GET_VSRC(vsrc2uc, vsrc3uc, stride, 16, vsrcperm0, vsrcperm1, src);
 235         CHROMA_MC8_ALTIVEC_CORE(vec_splat_s16(0), add28);
 236     }
 237 }
 238 #endif
 239
 240 #undef noop
 241 #undef add28
 242 #undef CHROMA_MC8_ALTIVEC_CORE