libavcodec/ppc/fmtconvert_altivec.c

   1 /*
   2  * Copyright (c) 2006 Luca Barbato <lu_zero@gentoo.org>
   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 "libavcodec/fmtconvert.h"
  22
  23 #include "libavutil/ppc/util_altivec.h"
  24 #include "libavutil/attributes.h"
  25 #include "libavutil/mem.h"
  26 #include "dsputil_altivec.h"
  27
  28 static void int32_to_float_fmul_scalar_altivec(float *dst, const int *src,
  29                                                float mul, int len)
  30 {
  31     union {
  32         vector float v;
  33         float s[4];
  34     } mul_u;
  35     int i;
  36     vector float src1, src2, dst1, dst2, mul_v, zero;
  37
  38     zero = (vector float)vec_splat_u32(0);
  39     mul_u.s[0] = mul;
  40     mul_v = vec_splat(mul_u.v, 0);
  41
  42     for (i = 0; i < len; i += 8) {
  43         src1 = vec_ctf(vec_ld(0,  src+i), 0);
  44         src2 = vec_ctf(vec_ld(16, src+i), 0);
  45         dst1 = vec_madd(src1, mul_v, zero);
  46         dst2 = vec_madd(src2, mul_v, zero);
  47         vec_st(dst1,  0, dst+i);
  48         vec_st(dst2, 16, dst+i);
  49     }
  50 }
  51
  52
  53 static vector signed short float_to_int16_one_altivec(const float *src)
  54 {
  55     vector float s0 = vec_ld(0, src);
  56     vector float s1 = vec_ld(16, src);
  57     vector signed int t0 = vec_cts(s0, 0);
  58     vector signed int t1 = vec_cts(s1, 0);
  59     return vec_packs(t0,t1);
  60 }
  61
  62 static void float_to_int16_altivec(int16_t *dst, const float *src, long len)
  63 {
  64     int i;
  65     vector signed short d0, d1, d;
  66     vector unsigned char align;
  67     if (((long)dst) & 15) { //FIXME
  68         for (i = 0; i < len - 7; i += 8) {
  69             d0 = vec_ld(0, dst+i);
  70             d  = float_to_int16_one_altivec(src + i);
  71             d1 = vec_ld(15, dst+i);
  72             d1 = vec_perm(d1, d0, vec_lvsl(0, dst + i));
  73             align = vec_lvsr(0, dst + i);
  74             d0 = vec_perm(d1, d, align);
  75             d1 = vec_perm(d, d1, align);
  76             vec_st(d0,  0, dst + i);
  77             vec_st(d1, 15, dst + i);
  78         }
  79     } else {
  80         for (i = 0; i < len - 7; i += 8) {
  81             d = float_to_int16_one_altivec(src + i);
  82             vec_st(d, 0, dst + i);
  83         }
  84     }
  85 }
  86
  87 #define VSTE_INC(dst, v, elem, inc) do {                \
  88         vector signed short s = vec_splat(v, elem);     \
  89         vec_ste(s, 0, dst);                             \
  90         dst += inc;                                     \
  91     } while (0)
  92
  93 static void float_to_int16_stride_altivec(int16_t *dst, const float *src,
  94                                           long len, int stride)
  95 {
  96     int i;
  97     vector signed short d;
  98
  99     for (i = 0; i < len - 7; i += 8) {
 100         d = float_to_int16_one_altivec(src + i);
 101         VSTE_INC(dst, d, 0, stride);
 102         VSTE_INC(dst, d, 1, stride);
 103         VSTE_INC(dst, d, 2, stride);
 104         VSTE_INC(dst, d, 3, stride);
 105         VSTE_INC(dst, d, 4, stride);
 106         VSTE_INC(dst, d, 5, stride);
 107         VSTE_INC(dst, d, 6, stride);
 108         VSTE_INC(dst, d, 7, stride);
 109     }
 110 }
 111
 112 static void float_to_int16_interleave_altivec(int16_t *dst, const float **src,
 113                                               long len, int channels)
 114 {
 115     int i;
 116     vector signed short d0, d1, d2, c0, c1, t0, t1;
 117     vector unsigned char align;
 118
 119     if (channels == 1)
 120         float_to_int16_altivec(dst, src[0], len);
 121     else {
 122         if (channels == 2) {
 123             if (((long)dst) & 15) {
 124                 for (i = 0; i < len - 7; i += 8) {
 125                     d0 = vec_ld(0,  dst + i);
 126                     t0 = float_to_int16_one_altivec(src[0] + i);
 127                     d1 = vec_ld(31, dst + i);
 128                     t1 = float_to_int16_one_altivec(src[1] + i);
 129                     c0 = vec_mergeh(t0, t1);
 130                     c1 = vec_mergel(t0, t1);
 131                     d2 = vec_perm(d1, d0, vec_lvsl(0, dst + i));
 132                     align = vec_lvsr(0, dst + i);
 133                     d0 = vec_perm(d2, c0, align);
 134                     d1 = vec_perm(c0, c1, align);
 135                     vec_st(d0,  0, dst + i);
 136                     d0 = vec_perm(c1, d2, align);
 137                     vec_st(d1, 15, dst + i);
 138                     vec_st(d0, 31, dst + i);
 139                     dst += 8;
 140                 }
 141             } else {
 142                 for (i = 0; i < len - 7; i += 8) {
 143                     t0 = float_to_int16_one_altivec(src[0] + i);
 144                     t1 = float_to_int16_one_altivec(src[1] + i);
 145                     d0 = vec_mergeh(t0, t1);
 146                     d1 = vec_mergel(t0, t1);
 147                     vec_st(d0,  0, dst + i);
 148                     vec_st(d1, 16, dst + i);
 149                     dst += 8;
 150                 }
 151             }
 152         } else {
 153             for (i = 0; i < channels; i++)
 154                 float_to_int16_stride_altivec(dst + i, src[i], len, channels);
 155         }
 156     }
 157 }
 158
 159 av_cold void ff_fmt_convert_init_altivec(FmtConvertContext *c, AVCodecContext *avctx)
 160 {
 161     c->int32_to_float_fmul_scalar = int32_to_float_fmul_scalar_altivec;
 162     if (!(avctx->flags & CODEC_FLAG_BITEXACT)) {
 163         c->float_to_int16 = float_to_int16_altivec;
 164         c->float_to_int16_interleave = float_to_int16_interleave_altivec;
 165     }
 166 }