/*
	dct64_i386.c: DCT64, a C variant for i386

	copyright ?-2006 by the mpg123 project - free software under the terms of the LGPL 2.1
	see COPYING and AUTHORS files in distribution or http://mpg123.org
	initially written by Michael Hipp
*/

/*
 * Discrete Cosine Tansform (DCT) for subband synthesis
 * optimized for machines with no auto-increment. 
 * The performance is highly compiler dependend. Maybe
 * the dct64.c version for 'normal' processor may be faster
 * even for Intel processors.
 */

#include "mpg123lib_intern.h"

static void dct64_1(real *out0,real *out1,real *b1,real *b2,real *samples)
{
 {
  register real *costab = pnts[0];

  b1[0x00] = samples[0x00] + samples[0x1F];
  b1[0x01] = samples[0x01] + samples[0x1E];
  b1[0x1F] = REAL_MUL(samples[0x00] - samples[0x1F], costab[0x0]);
  b1[0x1E] = REAL_MUL(samples[0x01] - samples[0x1E], costab[0x1]);

  b1[0x02] = samples[0x02] + samples[0x1D];
  b1[0x03] = samples[0x03] + samples[0x1C];
  b1[0x1D] = REAL_MUL(samples[0x02] - samples[0x1D], costab[0x2]);
  b1[0x1C] = REAL_MUL(samples[0x03] - samples[0x1C], costab[0x3]);

  b1[0x04] = samples[0x04] + samples[0x1B];
  b1[0x05] = samples[0x05] + samples[0x1A];
  b1[0x1B] = REAL_MUL(samples[0x04] - samples[0x1B], costab[0x4]);
  b1[0x1A] = REAL_MUL(samples[0x05] - samples[0x1A], costab[0x5]);

  b1[0x06] = samples[0x06] + samples[0x19];
  b1[0x07] = samples[0x07] + samples[0x18];
  b1[0x19] = REAL_MUL(samples[0x06] - samples[0x19], costab[0x6]);
  b1[0x18] = REAL_MUL(samples[0x07] - samples[0x18], costab[0x7]);

  b1[0x08] = samples[0x08] + samples[0x17];
  b1[0x09] = samples[0x09] + samples[0x16];
  b1[0x17] = REAL_MUL(samples[0x08] - samples[0x17], costab[0x8]);
  b1[0x16] = REAL_MUL(samples[0x09] - samples[0x16], costab[0x9]);

  b1[0x0A] = samples[0x0A] + samples[0x15];
  b1[0x0B] = samples[0x0B] + samples[0x14];
  b1[0x15] = REAL_MUL(samples[0x0A] - samples[0x15], costab[0xA]);
  b1[0x14] = REAL_MUL(samples[0x0B] - samples[0x14], costab[0xB]);

  b1[0x0C] = samples[0x0C] + samples[0x13];
  b1[0x0D] = samples[0x0D] + samples[0x12];
  b1[0x13] = REAL_MUL(samples[0x0C] - samples[0x13], costab[0xC]);
  b1[0x12] = REAL_MUL(samples[0x0D] - samples[0x12], costab[0xD]);

  b1[0x0E] = samples[0x0E] + samples[0x11];
  b1[0x0F] = samples[0x0F] + samples[0x10];
  b1[0x11] = REAL_MUL(samples[0x0E] - samples[0x11], costab[0xE]);
  b1[0x10] = REAL_MUL(samples[0x0F] - samples[0x10], costab[0xF]);

 }


 {
  register real *costab = pnts[1];

  b2[0x00] = b1[0x00] + b1[0x0F]; 
  b2[0x01] = b1[0x01] + b1[0x0E]; 
  b2[0x0F] = REAL_MUL(b1[0x00] - b1[0x0F], costab[0]);
  b2[0x0E] = REAL_MUL(b1[0x01] - b1[0x0E], costab[1]);

  b2[0x02] = b1[0x02] + b1[0x0D]; 
  b2[0x03] = b1[0x03] + b1[0x0C]; 
  b2[0x0D] = REAL_MUL(b1[0x02] - b1[0x0D], costab[2]);
  b2[0x0C] = REAL_MUL(b1[0x03] - b1[0x0C], costab[3]);

  b2[0x04] = b1[0x04] + b1[0x0B]; 
  b2[0x05] = b1[0x05] + b1[0x0A]; 
  b2[0x0B] = REAL_MUL(b1[0x04] - b1[0x0B], costab[4]);
  b2[0x0A] = REAL_MUL(b1[0x05] - b1[0x0A], costab[5]);

  b2[0x06] = b1[0x06] + b1[0x09]; 
  b2[0x07] = b1[0x07] + b1[0x08]; 
  b2[0x09] = REAL_MUL(b1[0x06] - b1[0x09], costab[6]);
  b2[0x08] = REAL_MUL(b1[0x07] - b1[0x08], costab[7]);

  /* */

  b2[0x10] = b1[0x10] + b1[0x1F];
  b2[0x11] = b1[0x11] + b1[0x1E];
  b2[0x1F] = REAL_MUL(b1[0x1F] - b1[0x10], costab[0]);
  b2[0x1E] = REAL_MUL(b1[0x1E] - b1[0x11], costab[1]);

  b2[0x12] = b1[0x12] + b1[0x1D];
  b2[0x13] = b1[0x13] + b1[0x1C];
  b2[0x1D] = REAL_MUL(b1[0x1D] - b1[0x12], costab[2]);
  b2[0x1C] = REAL_MUL(b1[0x1C] - b1[0x13], costab[3]);

  b2[0x14] = b1[0x14] + b1[0x1B];
  b2[0x15] = b1[0x15] + b1[0x1A];
  b2[0x1B] = REAL_MUL(b1[0x1B] - b1[0x14], costab[4]);
  b2[0x1A] = REAL_MUL(b1[0x1A] - b1[0x15], costab[5]);

  b2[0x16] = b1[0x16] + b1[0x19];
  b2[0x17] = b1[0x17] + b1[0x18];
  b2[0x19] = REAL_MUL(b1[0x19] - b1[0x16], costab[6]);
  b2[0x18] = REAL_MUL(b1[0x18] - b1[0x17], costab[7]);
 }

 {
  register real *costab = pnts[2];

  b1[0x00] = b2[0x00] + b2[0x07];
  b1[0x07] = REAL_MUL(b2[0x00] - b2[0x07], costab[0]);
  b1[0x01] = b2[0x01] + b2[0x06];
  b1[0x06] = REAL_MUL(b2[0x01] - b2[0x06], costab[1]);
  b1[0x02] = b2[0x02] + b2[0x05];
  b1[0x05] = REAL_MUL(b2[0x02] - b2[0x05], costab[2]);
  b1[0x03] = b2[0x03] + b2[0x04];
  b1[0x04] = REAL_MUL(b2[0x03] - b2[0x04], costab[3]);

  b1[0x08] = b2[0x08] + b2[0x0F];
  b1[0x0F] = REAL_MUL(b2[0x0F] - b2[0x08], costab[0]);
  b1[0x09] = b2[0x09] + b2[0x0E];
  b1[0x0E] = REAL_MUL(b2[0x0E] - b2[0x09], costab[1]);
  b1[0x0A] = b2[0x0A] + b2[0x0D];
  b1[0x0D] = REAL_MUL(b2[0x0D] - b2[0x0A], costab[2]);
  b1[0x0B] = b2[0x0B] + b2[0x0C];
  b1[0x0C] = REAL_MUL(b2[0x0C] - b2[0x0B], costab[3]);

  b1[0x10] = b2[0x10] + b2[0x17];
  b1[0x17] = REAL_MUL(b2[0x10] - b2[0x17], costab[0]);
  b1[0x11] = b2[0x11] + b2[0x16];
  b1[0x16] = REAL_MUL(b2[0x11] - b2[0x16], costab[1]);
  b1[0x12] = b2[0x12] + b2[0x15];
  b1[0x15] = REAL_MUL(b2[0x12] - b2[0x15], costab[2]);
  b1[0x13] = b2[0x13] + b2[0x14];
  b1[0x14] = REAL_MUL(b2[0x13] - b2[0x14], costab[3]);

  b1[0x18] = b2[0x18] + b2[0x1F];
  b1[0x1F] = REAL_MUL(b2[0x1F] - b2[0x18], costab[0]);
  b1[0x19] = b2[0x19] + b2[0x1E];
  b1[0x1E] = REAL_MUL(b2[0x1E] - b2[0x19], costab[1]);
  b1[0x1A] = b2[0x1A] + b2[0x1D];
  b1[0x1D] = REAL_MUL(b2[0x1D] - b2[0x1A], costab[2]);
  b1[0x1B] = b2[0x1B] + b2[0x1C];
  b1[0x1C] = REAL_MUL(b2[0x1C] - b2[0x1B], costab[3]);
 }

 {
  register real const cos0 = pnts[3][0];
  register real const cos1 = pnts[3][1];

  b2[0x00] = b1[0x00] + b1[0x03];
  b2[0x03] = REAL_MUL(b1[0x00] - b1[0x03], cos0);
  b2[0x01] = b1[0x01] + b1[0x02];
  b2[0x02] = REAL_MUL(b1[0x01] - b1[0x02], cos1);

  b2[0x04] = b1[0x04] + b1[0x07];
  b2[0x07] = REAL_MUL(b1[0x07] - b1[0x04], cos0);
  b2[0x05] = b1[0x05] + b1[0x06];
  b2[0x06] = REAL_MUL(b1[0x06] - b1[0x05], cos1);

  b2[0x08] = b1[0x08] + b1[0x0B];
  b2[0x0B] = REAL_MUL(b1[0x08] - b1[0x0B], cos0);
  b2[0x09] = b1[0x09] + b1[0x0A];
  b2[0x0A] = REAL_MUL(b1[0x09] - b1[0x0A], cos1);
  
  b2[0x0C] = b1[0x0C] + b1[0x0F];
  b2[0x0F] = REAL_MUL(b1[0x0F] - b1[0x0C], cos0);
  b2[0x0D] = b1[0x0D] + b1[0x0E];
  b2[0x0E] = REAL_MUL(b1[0x0E] - b1[0x0D], cos1);

  b2[0x10] = b1[0x10] + b1[0x13];
  b2[0x13] = REAL_MUL(b1[0x10] - b1[0x13], cos0);
  b2[0x11] = b1[0x11] + b1[0x12];
  b2[0x12] = REAL_MUL(b1[0x11] - b1[0x12], cos1);

  b2[0x14] = b1[0x14] + b1[0x17];
  b2[0x17] = REAL_MUL(b1[0x17] - b1[0x14], cos0);
  b2[0x15] = b1[0x15] + b1[0x16];
  b2[0x16] = REAL_MUL(b1[0x16] - b1[0x15], cos1);

  b2[0x18] = b1[0x18] + b1[0x1B];
  b2[0x1B] = REAL_MUL(b1[0x18] - b1[0x1B], cos0);
  b2[0x19] = b1[0x19] + b1[0x1A];
  b2[0x1A] = REAL_MUL(b1[0x19] - b1[0x1A], cos1);

  b2[0x1C] = b1[0x1C] + b1[0x1F];
  b2[0x1F] = REAL_MUL(b1[0x1F] - b1[0x1C], cos0);
  b2[0x1D] = b1[0x1D] + b1[0x1E];
  b2[0x1E] = REAL_MUL(b1[0x1E] - b1[0x1D], cos1);
 }

 {
  register real const cos0 = pnts[4][0];

  b1[0x00] = b2[0x00] + b2[0x01];
  b1[0x01] = REAL_MUL(b2[0x00] - b2[0x01], cos0);
  b1[0x02] = b2[0x02] + b2[0x03];
  b1[0x03] = REAL_MUL(b2[0x03] - b2[0x02], cos0);
  b1[0x02] += b1[0x03];

  b1[0x04] = b2[0x04] + b2[0x05];
  b1[0x05] = REAL_MUL(b2[0x04] - b2[0x05], cos0);
  b1[0x06] = b2[0x06] + b2[0x07];
  b1[0x07] = REAL_MUL(b2[0x07] - b2[0x06], cos0);
  b1[0x06] += b1[0x07];
  b1[0x04] += b1[0x06];
  b1[0x06] += b1[0x05];
  b1[0x05] += b1[0x07];

  b1[0x08] = b2[0x08] + b2[0x09];
  b1[0x09] = REAL_MUL(b2[0x08] - b2[0x09], cos0);
  b1[0x0A] = b2[0x0A] + b2[0x0B];
  b1[0x0B] = REAL_MUL(b2[0x0B] - b2[0x0A], cos0);
  b1[0x0A] += b1[0x0B];

  b1[0x0C] = b2[0x0C] + b2[0x0D];
  b1[0x0D] = REAL_MUL(b2[0x0C] - b2[0x0D], cos0);
  b1[0x0E] = b2[0x0E] + b2[0x0F];
  b1[0x0F] = REAL_MUL(b2[0x0F] - b2[0x0E], cos0);
  b1[0x0E] += b1[0x0F];
  b1[0x0C] += b1[0x0E];
  b1[0x0E] += b1[0x0D];
  b1[0x0D] += b1[0x0F];

  b1[0x10] = b2[0x10] + b2[0x11];
  b1[0x11] = REAL_MUL(b2[0x10] - b2[0x11], cos0);
  b1[0x12] = b2[0x12] + b2[0x13];
  b1[0x13] = REAL_MUL(b2[0x13] - b2[0x12], cos0);
  b1[0x12] += b1[0x13];

  b1[0x14] = b2[0x14] + b2[0x15];
  b1[0x15] = REAL_MUL(b2[0x14] - b2[0x15], cos0);
  b1[0x16] = b2[0x16] + b2[0x17];
  b1[0x17] = REAL_MUL(b2[0x17] - b2[0x16], cos0);
  b1[0x16] += b1[0x17];
  b1[0x14] += b1[0x16];
  b1[0x16] += b1[0x15];
  b1[0x15] += b1[0x17];

  b1[0x18] = b2[0x18] + b2[0x19];
  b1[0x19] = REAL_MUL(b2[0x18] - b2[0x19], cos0);
  b1[0x1A] = b2[0x1A] + b2[0x1B];
  b1[0x1B] = REAL_MUL(b2[0x1B] - b2[0x1A], cos0);
  b1[0x1A] += b1[0x1B];

  b1[0x1C] = b2[0x1C] + b2[0x1D];
  b1[0x1D] = REAL_MUL(b2[0x1C] - b2[0x1D], cos0);
  b1[0x1E] = b2[0x1E] + b2[0x1F];
  b1[0x1F] = REAL_MUL(b2[0x1F] - b2[0x1E], cos0);
  b1[0x1E] += b1[0x1F];
  b1[0x1C] += b1[0x1E];
  b1[0x1E] += b1[0x1D];
  b1[0x1D] += b1[0x1F];
 }

 out0[0x10*16] = REAL_SCALE_DCT64(b1[0x00]);
 out0[0x10*12] = REAL_SCALE_DCT64(b1[0x04]);
 out0[0x10* 8] = REAL_SCALE_DCT64(b1[0x02]);
 out0[0x10* 4] = REAL_SCALE_DCT64(b1[0x06]);
 out0[0x10* 0] = REAL_SCALE_DCT64(b1[0x01]);
 out1[0x10* 0] = REAL_SCALE_DCT64(b1[0x01]);
 out1[0x10* 4] = REAL_SCALE_DCT64(b1[0x05]);
 out1[0x10* 8] = REAL_SCALE_DCT64(b1[0x03]);
 out1[0x10*12] = REAL_SCALE_DCT64(b1[0x07]);

#if 1
 out0[0x10*14] = REAL_SCALE_DCT64(b1[0x08] + b1[0x0C]);
 out0[0x10*10] = REAL_SCALE_DCT64(b1[0x0C] + b1[0x0a]);
 out0[0x10* 6] = REAL_SCALE_DCT64(b1[0x0A] + b1[0x0E]);
 out0[0x10* 2] = REAL_SCALE_DCT64(b1[0x0E] + b1[0x09]);
 out1[0x10* 2] = REAL_SCALE_DCT64(b1[0x09] + b1[0x0D]);
 out1[0x10* 6] = REAL_SCALE_DCT64(b1[0x0D] + b1[0x0B]);
 out1[0x10*10] = REAL_SCALE_DCT64(b1[0x0B] + b1[0x0F]);
 out1[0x10*14] = REAL_SCALE_DCT64(b1[0x0F]);
#else
 b1[0x08] += b1[0x0C];
 out0[0x10*14] = REAL_SCALE_DCT64(b1[0x08]);
 b1[0x0C] += b1[0x0a];
 out0[0x10*10] = REAL_SCALE_DCT64(b1[0x0C]);
 b1[0x0A] += b1[0x0E];
 out0[0x10* 6] = REAL_SCALE_DCT64(b1[0x0A]);
 b1[0x0E] += b1[0x09];
 out0[0x10* 2] = REAL_SCALE_DCT64(b1[0x0E]);
 b1[0x09] += b1[0x0D];
 out1[0x10* 2] = REAL_SCALE_DCT64(b1[0x09]);
 b1[0x0D] += b1[0x0B];
 out1[0x10* 6] = REAL_SCALE_DCT64(b1[0x0D]);
 b1[0x0B] += b1[0x0F];
 out1[0x10*10] = REAL_SCALE_DCT64(b1[0x0B]);
 out1[0x10*14] = REAL_SCALE_DCT64(b1[0x0F]);
#endif

 { 
 real tmp;
 tmp = b1[0x18] + b1[0x1C];
 out0[0x10*15] = REAL_SCALE_DCT64(tmp + b1[0x10]);
 out0[0x10*13] = REAL_SCALE_DCT64(tmp + b1[0x14]);
 tmp = b1[0x1C] + b1[0x1A];
 out0[0x10*11] = REAL_SCALE_DCT64(tmp + b1[0x14]);
 out0[0x10* 9] = REAL_SCALE_DCT64(tmp + b1[0x12]);
 tmp = b1[0x1A] + b1[0x1E];
 out0[0x10* 7] = REAL_SCALE_DCT64(tmp + b1[0x12]);
 out0[0x10* 5] = REAL_SCALE_DCT64(tmp + b1[0x16]);
 tmp = b1[0x1E] + b1[0x19];
 out0[0x10* 3] = REAL_SCALE_DCT64(tmp + b1[0x16]);
 out0[0x10* 1] = REAL_SCALE_DCT64(tmp + b1[0x11]);
 tmp = b1[0x19] + b1[0x1D];
 out1[0x10* 1] = REAL_SCALE_DCT64(tmp + b1[0x11]);
 out1[0x10* 3] = REAL_SCALE_DCT64(tmp + b1[0x15]); 
 tmp = b1[0x1D] + b1[0x1B];
 out1[0x10* 5] = REAL_SCALE_DCT64(tmp + b1[0x15]);
 out1[0x10* 7] = REAL_SCALE_DCT64(tmp + b1[0x13]);
 tmp = b1[0x1B] + b1[0x1F];
 out1[0x10* 9] = REAL_SCALE_DCT64(tmp + b1[0x13]);
 out1[0x10*11] = REAL_SCALE_DCT64(tmp + b1[0x17]);
 out1[0x10*13] = REAL_SCALE_DCT64(b1[0x17] + b1[0x1F]);
 out1[0x10*15] = REAL_SCALE_DCT64(b1[0x1F]);
 }
}

/*
 * the call via dct64 is a trick to force GCC to use
 * (new) registers for the b1,b2 pointer to the bufs[xx] field
 */
void dct64_i386(real *a,real *b,real *c)
{
  real bufs[0x40];
  dct64_1(a,b,bufs,bufs+0x20,c);
}