src/imageeffects_mmx.cpp

   1 /*
   2   Copyright (c) 2006 Paolo Capriotti <p.capriotti@gmail.com>
   3             (c) 2006 Maurizio Monge <maurizio.monge@kdemail.net>
   4
   5   This program is free software; you can redistribute it and/or modify
   6   it under the terms of the GNU General Public License as published by
   7   the Free Software Foundation; either version 2 of the License, or
   8   (at your option) any later version.
   9 */
  10
  11 #include <math.h>
  12 #include <inttypes.h>
  13 #include <mmintrin.h>
  14 #include <QImage>
  15
  16
  17 namespace ImageEffects {
  18
  19 union vec4s
  20 {
  21   uint16_t i[4];
  22   __m64 v;
  23 };
  24
  25 static inline void blur_mmx(void *px, __m64& v, __m64& alpha)
  26 {
  27   uint64_t z1 = 0ULL;
  28   uint64_t z2 = 0ULL;
  29   asm(
  30     "movd %[pixel], %[t1]\n"
  31     "punpcklbw %[t1], %[t2]\n"
  32     "psrlw $1, %[t2]\n"
  33     "psubw %[accum], %[t2]\n"
  34     "pmulhw %[alpha], %[t2]\n"
  35     "psllw $1, %[t2]\n"
  36     "paddw %[t2], %[accum]\n"
  37     "movq %[accum], %[t1]\n"
  38     "psrlw $7, %[t1]\n"
  39        // "pand %[mask], %[t1]\n"
  40     "packuswb %[t1], %[t1]\n"
  41     "movd %[t1], %[pixel]\n"
  42     :  [pixel] "+m"(*(uint32_t*)px)
  43      , [accum] "+y"(v)
  44      , [t1] "+y"(z1)
  45      , [t2] "+y"(z2)
  46     :  [alpha] "y"(alpha)
  47      // , [mask] "y"(0x00ff00ff00ff00ffULL)
  48   );
  49 }
  50
  51 void expblur_mmx( QImage &img, int radius )
  52 {
  53   if(radius<1)
  54     return;
  55
  56   /* Calculate the alpha such that 90% of
  57      the kernel is within the radius.
  58      (Kernel extends to infinity)
  59   */
  60   uint16_t alpha = (uint16_t)((1<<15)*(1.0f-expf(-2.3f/(radius+1.f))));
  61
  62   vec4s a;
  63   QRgb *ptr = (QRgb *)img.bits();
  64   int h = img.height();
  65   int w = img.width();
  66   int hw = (img.height()-1)*img.width();
  67   for(int i=0;i<4;i++)
  68     a.i[i] = alpha;
  69
  70   for(int row=0;row<h;row++)
  71   {
  72     vec4s z;
  73     uint8_t *cptr = (uint8_t*)(ptr+row*w);
  74     for(int i=0;i<4;i++)
  75       z.i[i] = cptr[i]<<7;
  76
  77     for(int index=1; index<w; index++)
  78       blur_mmx(&cptr[index*4], z.v, a.v);
  79
  80     for(int index=w-2; index>=0; index--)
  81       blur_mmx(&cptr[index*4], z.v, a.v);
  82   }
  83
  84   for(int col=0;col<w;col++)
  85   {
  86     vec4s z;
  87     uint8_t *cptr = (uint8_t*)(ptr+col);
  88
  89     for(int i=0;i<4;i++)
  90       z.i[i] = cptr[i]<<7;
  91
  92     for(int index=w; index<hw; index+=w)
  93       blur_mmx(&cptr[index*4], z.v, a.v);
  94
  95     for(int index=hw-w; index>=0; index-=w)
  96       blur_mmx(&cptr[index*4], z.v, a.v);
  97   }
  98
  99   asm("emms");
 100   return;
 101 }
 102
 103 } //end namespace ImageEffects