vp8/decoder/x86/idct_blk_mmx.c

   1 /*
   2  *  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
   3  *
   4  *  Use of this source code is governed by a BSD-style license
   5  *  that can be found in the LICENSE file in the root of the source
   6  *  tree. An additional intellectual property rights grant can be found
   7  *  in the file PATENTS.  All contributing project authors may
   8  *  be found in the AUTHORS file in the root of the source tree.
   9  */
  10
  11 #include "vpx_ports/config.h"
  12 #include "vp8/common/idct.h"
  13 #include "vp8/decoder/dequantize.h"
  14
  15 void vp8_dequant_dc_idct_add_y_block_mmx
  16             (short *q, short *dq, unsigned char *pre,
  17              unsigned char *dst, int stride, char *eobs, short *dc)
  18 {
  19     int i;
  20
  21     for (i = 0; i < 4; i++)
  22     {
  23         if (eobs[0] > 1)
  24             vp8_dequant_dc_idct_add_mmx (q, dq, pre, dst, 16, stride, dc[0]);
  25         else
  26             vp8_dc_only_idct_add_mmx (dc[0], pre, dst, 16, stride);
  27
  28         if (eobs[1] > 1)
  29             vp8_dequant_dc_idct_add_mmx (q+16, dq, pre+4, dst+4, 16, stride, dc[1]);
  30         else
  31             vp8_dc_only_idct_add_mmx (dc[1], pre+4, dst+4, 16, stride);
  32
  33         if (eobs[2] > 1)
  34             vp8_dequant_dc_idct_add_mmx (q+32, dq, pre+8, dst+8, 16, stride, dc[2]);
  35         else
  36             vp8_dc_only_idct_add_mmx (dc[2], pre+8, dst+8, 16, stride);
  37
  38         if (eobs[3] > 1)
  39             vp8_dequant_dc_idct_add_mmx (q+48, dq, pre+12, dst+12, 16, stride, dc[3]);
  40         else
  41             vp8_dc_only_idct_add_mmx (dc[3], pre+12, dst+12, 16, stride);
  42
  43         q    += 64;
  44         dc   += 4;
  45         pre  += 64;
  46         dst  += 4*stride;
  47         eobs += 4;
  48     }
  49 }
  50
  51 void vp8_dequant_idct_add_y_block_mmx
  52             (short *q, short *dq, unsigned char *pre,
  53              unsigned char *dst, int stride, char *eobs)
  54 {
  55     int i;
  56
  57     for (i = 0; i < 4; i++)
  58     {
  59         if (eobs[0] > 1)
  60             vp8_dequant_idct_add_mmx (q, dq, pre, dst, 16, stride);
  61         else
  62         {
  63             vp8_dc_only_idct_add_mmx (q[0]*dq[0], pre, dst, 16, stride);
  64             ((int *)q)[0] = 0;
  65         }
  66
  67         if (eobs[1] > 1)
  68             vp8_dequant_idct_add_mmx (q+16, dq, pre+4, dst+4, 16, stride);
  69         else
  70         {
  71             vp8_dc_only_idct_add_mmx (q[16]*dq[0], pre+4, dst+4, 16, stride);
  72             ((int *)(q+16))[0] = 0;
  73         }
  74
  75         if (eobs[2] > 1)
  76             vp8_dequant_idct_add_mmx (q+32, dq, pre+8, dst+8, 16, stride);
  77         else
  78         {
  79             vp8_dc_only_idct_add_mmx (q[32]*dq[0], pre+8, dst+8, 16, stride);
  80             ((int *)(q+32))[0] = 0;
  81         }
  82
  83         if (eobs[3] > 1)
  84             vp8_dequant_idct_add_mmx (q+48, dq, pre+12, dst+12, 16, stride);
  85         else
  86         {
  87             vp8_dc_only_idct_add_mmx (q[48]*dq[0], pre+12, dst+12, 16, stride);
  88             ((int *)(q+48))[0] = 0;
  89         }
  90
  91         q    += 64;
  92         pre  += 64;
  93         dst  += 4*stride;
  94         eobs += 4;
  95     }
  96 }
  97
  98 void vp8_dequant_idct_add_uv_block_mmx
  99             (short *q, short *dq, unsigned char *pre,
 100              unsigned char *dstu, unsigned char *dstv, int stride, char *eobs)
 101 {
 102     int i;
 103
 104     for (i = 0; i < 2; i++)
 105     {
 106         if (eobs[0] > 1)
 107             vp8_dequant_idct_add_mmx (q, dq, pre, dstu, 8, stride);
 108         else
 109         {
 110             vp8_dc_only_idct_add_mmx (q[0]*dq[0], pre, dstu, 8, stride);
 111             ((int *)q)[0] = 0;
 112         }
 113
 114         if (eobs[1] > 1)
 115             vp8_dequant_idct_add_mmx (q+16, dq, pre+4, dstu+4, 8, stride);
 116         else
 117         {
 118             vp8_dc_only_idct_add_mmx (q[16]*dq[0], pre+4, dstu+4, 8, stride);
 119             ((int *)(q+16))[0] = 0;
 120         }
 121
 122         q    += 32;
 123         pre  += 32;
 124         dstu += 4*stride;
 125         eobs += 2;
 126     }
 127
 128     for (i = 0; i < 2; i++)
 129     {
 130         if (eobs[0] > 1)
 131             vp8_dequant_idct_add_mmx (q, dq, pre, dstv, 8, stride);
 132         else
 133         {
 134             vp8_dc_only_idct_add_mmx (q[0]*dq[0], pre, dstv, 8, stride);
 135             ((int *)q)[0] = 0;
 136         }
 137
 138         if (eobs[1] > 1)
 139             vp8_dequant_idct_add_mmx (q+16, dq, pre+4, dstv+4, 8, stride);
 140         else
 141         {
 142             vp8_dc_only_idct_add_mmx (q[16]*dq[0], pre+4, dstv+4, 8, stride);
 143             ((int *)(q+16))[0] = 0;
 144         }
 145
 146         q    += 32;
 147         pre  += 32;
 148         dstv += 4*stride;
 149         eobs += 2;
 150     }
 151 }