3
* Discrete Cosine Tansform (DCT) for subband synthesis
5
* -funroll-loops (for gcc) will remove the loops for better performance
6
* using loops in the source-code enhances readabillity
10
* TODO: write an optimized version for the down-sampling modes
11
* (in these modes the bands 16-31 (2:1) or 8-31 (4:1) are zero
16
void dct64(real *out0,real *out1,real *samples)
22
register real *b1,*b2,*bs,*costab;
30
*bs++ = (*b1++ + *--b2);
32
*bs++ = REAL_MUL((*--b2 - *b1++), *--costab);
40
*bs++ = (*b1++ + *--b2);
42
*bs++ = REAL_MUL((*--b2 - *b1++), *--costab);
46
*bs++ = (*b1++ + *--b2);
48
*bs++ = REAL_MUL((*b1++ - *--b2), *--costab);
59
*bs++ = (*b1++ + *--b2);
61
*bs++ = REAL_MUL((*--b2 - *b1++), costab[i]);
64
*bs++ = (*b1++ + *--b2);
66
*bs++ = REAL_MUL((*b1++ - *--b2), costab[i]);
76
*bs++ = (*b1++ + *--b2);
77
*bs++ = (*b1++ + *--b2);
78
*bs++ = REAL_MUL((*--b2 - *b1++), costab[1]);
79
*bs++ = REAL_MUL((*--b2 - *b1++), costab[0]);
81
*bs++ = (*b1++ + *--b2);
82
*bs++ = (*b1++ + *--b2);
83
*bs++ = REAL_MUL((*b1++ - *--b2), costab[1]);
84
*bs++ = REAL_MUL((*b1++ - *--b2), costab[0]);
95
*bs++ = REAL_MUL((v0 - v1), (*costab));
98
*bs++ = REAL_MUL((v1 - v0), (*costab));
108
for(b1=bufs,i=8;i;i--,b1+=4)
111
for(b1=bufs,i=4;i;i--,b1+=8)
118
for(b1=bufs,i=2;i;i--,b1+=16)
131
out0[0x10*16] = bufs[0];
132
out0[0x10*15] = bufs[16+0] + bufs[16+8];
133
out0[0x10*14] = bufs[8];
134
out0[0x10*13] = bufs[16+8] + bufs[16+4];
135
out0[0x10*12] = bufs[4];
136
out0[0x10*11] = bufs[16+4] + bufs[16+12];
137
out0[0x10*10] = bufs[12];
138
out0[0x10* 9] = bufs[16+12] + bufs[16+2];
139
out0[0x10* 8] = bufs[2];
140
out0[0x10* 7] = bufs[16+2] + bufs[16+10];
141
out0[0x10* 6] = bufs[10];
142
out0[0x10* 5] = bufs[16+10] + bufs[16+6];
143
out0[0x10* 4] = bufs[6];
144
out0[0x10* 3] = bufs[16+6] + bufs[16+14];
145
out0[0x10* 2] = bufs[14];
146
out0[0x10* 1] = bufs[16+14] + bufs[16+1];
147
out0[0x10* 0] = bufs[1];
149
out1[0x10* 0] = bufs[1];
150
out1[0x10* 1] = bufs[16+1] + bufs[16+9];
151
out1[0x10* 2] = bufs[9];
152
out1[0x10* 3] = bufs[16+9] + bufs[16+5];
153
out1[0x10* 4] = bufs[5];
154
out1[0x10* 5] = bufs[16+5] + bufs[16+13];
155
out1[0x10* 6] = bufs[13];
156
out1[0x10* 7] = bufs[16+13] + bufs[16+3];
157
out1[0x10* 8] = bufs[3];
158
out1[0x10* 9] = bufs[16+3] + bufs[16+11];
159
out1[0x10*10] = bufs[11];
160
out1[0x10*11] = bufs[16+11] + bufs[16+7];
161
out1[0x10*12] = bufs[7];
162
out1[0x10*13] = bufs[16+7] + bufs[16+15];
163
out1[0x10*14] = bufs[15];
164
out1[0x10*15] = bufs[16+15];