libavcodec/ppc/vp3dsp_altivec.c

   1 /*
   2  * Copyright (C) 2009 David Conrad
   3  *
   4  * This file is part of Libav.
   5  *
   6  * Libav 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  * Libav 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 Libav; if not, write to the Free Software
  18  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
  19  */
  20
  21 #include <string.h>
  22
  23 #include "config.h"
  24 #include "libavutil/attributes.h"
  25 #include "libavutil/cpu.h"
  26 #include "libavcodec/vp3dsp.h"
  27
  28 #if HAVE_ALTIVEC
  29
  30 #include "libavutil/ppc/types_altivec.h"
  31 #include "libavutil/ppc/util_altivec.h"
  32 #include "dsputil_altivec.h"
  33
  34 static const vec_s16 constants =
  35     {0, 64277, 60547, 54491, 46341, 36410, 25080, 12785};
  36 static const vec_u8 interleave_high =
  37     {0, 1, 16, 17, 4, 5, 20, 21, 8, 9, 24, 25, 12, 13, 28, 29};
  38
  39 #define IDCT_START \
  40     vec_s16 A, B, C, D, Ad, Bd, Cd, Dd, E, F, G, H;\
  41     vec_s16 Ed, Gd, Add, Bdd, Fd, Hd;\
  42     vec_s16 eight = vec_splat_s16(8);\
  43     vec_u16 four = vec_splat_u16(4);\
  44 \
  45     vec_s16 C1 = vec_splat(constants, 1);\
  46     vec_s16 C2 = vec_splat(constants, 2);\
  47     vec_s16 C3 = vec_splat(constants, 3);\
  48     vec_s16 C4 = vec_splat(constants, 4);\
  49     vec_s16 C5 = vec_splat(constants, 5);\
  50     vec_s16 C6 = vec_splat(constants, 6);\
  51     vec_s16 C7 = vec_splat(constants, 7);\
  52 \
  53     vec_s16 b0 = vec_ld(0x00, block);\
  54     vec_s16 b1 = vec_ld(0x10, block);\
  55     vec_s16 b2 = vec_ld(0x20, block);\
  56     vec_s16 b3 = vec_ld(0x30, block);\
  57     vec_s16 b4 = vec_ld(0x40, block);\
  58     vec_s16 b5 = vec_ld(0x50, block);\
  59     vec_s16 b6 = vec_ld(0x60, block);\
  60     vec_s16 b7 = vec_ld(0x70, block);
  61
  62 // these functions do (a*C)>>16
  63 // things are tricky because a is signed, but C unsigned.
  64 // M15 is used if C fits in 15 bit unsigned (C6,C7)
  65 // M16 is used if C requires 16 bits unsigned
  66 static inline vec_s16 M15(vec_s16 a, vec_s16 C)
  67 {
  68     return (vec_s16)vec_perm(vec_mule(a,C), vec_mulo(a,C), interleave_high);
  69 }
  70 static inline vec_s16 M16(vec_s16 a, vec_s16 C)
  71 {
  72     return vec_add(a, M15(a, C));
  73 }
  74
  75 #define IDCT_1D(ADD, SHIFT)\
  76     A = vec_add(M16(b1, C1), M15(b7, C7));\
  77     B = vec_sub(M15(b1, C7), M16(b7, C1));\
  78     C = vec_add(M16(b3, C3), M16(b5, C5));\
  79     D = vec_sub(M16(b5, C3), M16(b3, C5));\
  80 \
  81     Ad = M16(vec_sub(A, C), C4);\
  82     Bd = M16(vec_sub(B, D), C4);\
  83 \
  84     Cd = vec_add(A, C);\
  85     Dd = vec_add(B, D);\
  86 \
  87     E = ADD(M16(vec_add(b0, b4), C4));\
  88     F = ADD(M16(vec_sub(b0, b4), C4));\
  89 \
  90     G = vec_add(M16(b2, C2), M15(b6, C6));\
  91     H = vec_sub(M15(b2, C6), M16(b6, C2));\
  92 \
  93     Ed = vec_sub(E, G);\
  94     Gd = vec_add(E, G);\
  95 \
  96     Add = vec_add(F, Ad);\
  97     Bdd = vec_sub(Bd, H);\
  98 \
  99     Fd = vec_sub(F, Ad);\
 100     Hd = vec_add(Bd, H);\
 101 \
 102     b0 = SHIFT(vec_add(Gd, Cd));\
 103     b7 = SHIFT(vec_sub(Gd, Cd));\
 104 \
 105     b1 = SHIFT(vec_add(Add, Hd));\
 106     b2 = SHIFT(vec_sub(Add, Hd));\
 107 \
 108     b3 = SHIFT(vec_add(Ed, Dd));\
 109     b4 = SHIFT(vec_sub(Ed, Dd));\
 110 \
 111     b5 = SHIFT(vec_add(Fd, Bdd));\
 112     b6 = SHIFT(vec_sub(Fd, Bdd));
 113
 114 #define NOP(a) a
 115 #define ADD8(a) vec_add(a, eight)
 116 #define SHIFT4(a) vec_sra(a, four)
 117
 118 static void vp3_idct_put_altivec(uint8_t *dst, int stride, int16_t block[64])
 119 {
 120     vec_u8 t;
 121     IDCT_START
 122
 123     // pixels are signed; so add 128*16 in addition to the normal 8
 124     vec_s16 v2048 = vec_sl(vec_splat_s16(1), vec_splat_u16(11));
 125     eight = vec_add(eight, v2048);
 126
 127     IDCT_1D(NOP, NOP)
 128     TRANSPOSE8(b0, b1, b2, b3, b4, b5, b6, b7);
 129     IDCT_1D(ADD8, SHIFT4)
 130
 131 #define PUT(a)\
 132     t = vec_packsu(a, a);\
 133     vec_ste((vec_u32)t, 0, (unsigned int *)dst);\
 134     vec_ste((vec_u32)t, 4, (unsigned int *)dst);
 135
 136     PUT(b0)     dst += stride;
 137     PUT(b1)     dst += stride;
 138     PUT(b2)     dst += stride;
 139     PUT(b3)     dst += stride;
 140     PUT(b4)     dst += stride;
 141     PUT(b5)     dst += stride;
 142     PUT(b6)     dst += stride;
 143     PUT(b7)
 144     memset(block, 0, sizeof(*block) * 64);
 145 }
 146
 147 static void vp3_idct_add_altivec(uint8_t *dst, int stride, int16_t block[64])
 148 {
 149     LOAD_ZERO;
 150     vec_u8 t, vdst;
 151     vec_s16 vdst_16;
 152     vec_u8 vdst_mask = vec_mergeh(vec_splat_u8(-1), vec_lvsl(0, dst));
 153
 154     IDCT_START
 155
 156     IDCT_1D(NOP, NOP)
 157     TRANSPOSE8(b0, b1, b2, b3, b4, b5, b6, b7);
 158     IDCT_1D(ADD8, SHIFT4)
 159
 160 #define ADD(a)\
 161     vdst = vec_ld(0, dst);\
 162     vdst_16 = (vec_s16)vec_perm(vdst, zero_u8v, vdst_mask);\
 163     vdst_16 = vec_adds(a, vdst_16);\
 164     t = vec_packsu(vdst_16, vdst_16);\
 165     vec_ste((vec_u32)t, 0, (unsigned int *)dst);\
 166     vec_ste((vec_u32)t, 4, (unsigned int *)dst);
 167
 168     ADD(b0)     dst += stride;
 169     ADD(b1)     dst += stride;
 170     ADD(b2)     dst += stride;
 171     ADD(b3)     dst += stride;
 172     ADD(b4)     dst += stride;
 173     ADD(b5)     dst += stride;
 174     ADD(b6)     dst += stride;
 175     ADD(b7)
 176     memset(block, 0, sizeof(*block) * 64);
 177 }
 178
 179 #endif /* HAVE_ALTIVEC */
 180
 181 av_cold void ff_vp3dsp_init_ppc(VP3DSPContext *c, int flags)
 182 {
 183 #if HAVE_ALTIVEC
 184     if (av_get_cpu_flags() & AV_CPU_FLAG_ALTIVEC) {
 185         c->idct_put  = vp3_idct_put_altivec;
 186         c->idct_add  = vp3_idct_add_altivec;
 187         c->idct_perm = FF_TRANSPOSE_IDCT_PERM;
 188     }
 189 #endif
 190 }