16595bddee629cad807ef5a00f1390fa0e7415ec
[libav.git] / libavcodec / i386 / fft_3dn.c
1 /*
2 * FFT/MDCT transform with 3DNow! optimizations
3 * Copyright (c) 2006 Zuxy MENG Jie.
4 * Based on fft_sse.c copyright (c) 2002 Fabrice Bellard.
5 *
6 * This library 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 of the License, or (at your option) any later version.
10 *
11 * This library 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 this library; if not, write to the Free Software
18 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
19 */
20 #include "../dsputil.h"
21 #include <math.h>
22
23 #ifdef HAVE_MM3DNOW
24
25 #include <mm3dnow.h>
26
27 static const int p1m1[2] __attribute__((aligned(8))) =
28 { 0, 1 << 31 };
29
30 static const int m1p1[2] __attribute__((aligned(8))) =
31 { 1 << 31, 0 };
32
33 void ff_fft_calc_3dn(FFTContext *s, FFTComplex *z)
34 {
35 int ln = s->nbits;
36 int j, np, np2;
37 int nblocks, nloops;
38 register FFTComplex *p, *q;
39 FFTComplex *cptr, *cptr1;
40 int k;
41
42 np = 1 << ln;
43 /* FEMMS not a must here but recommended by AMD */
44 _m_femms();
45
46 {
47 __m64 *r, a0, a1, b0, b1, tmp, c;
48
49 r = (__m64 *)&z[0];
50 if (s->inverse)
51 c = *(__m64 *)m1p1;
52 else
53 c = *(__m64 *)p1m1;
54
55 j = (np >> 2);
56 do {
57 /* do the pass 0 butterfly */
58 a0 = _m_pfadd(r[0], r[1]);
59 a1 = _m_pfsub(r[0], r[1]);
60
61 /* do the pass 0 butterfly */
62 b0 = _m_pfadd(r[2], r[3]);
63 b1 = _m_pfsub(r[2], r[3]);
64
65 /* multiply third by -i */
66 tmp = _m_punpckhdq(b1, b1);
67 b1 = _m_punpckldq(b1, b1);
68 b1 = _m_punpckldq(tmp, b1);
69 b1 = _m_pxor(b1, c);
70
71 /* do the pass 1 butterfly */
72 r[0] = _m_pfadd(a0, b0);
73 r[1] = _m_pfadd(a1, b1);
74 r[2] = _m_pfsub(a0, b0);
75 r[3] = _m_pfsub(a1, b1);
76 r += 4;
77 } while (--j != 0);
78 }
79 /* pass 2 .. ln-1 */
80
81 nblocks = np >> 3;
82 nloops = 1 << 2;
83 np2 = np >> 1;
84
85 cptr1 = s->exptab1;
86 do {
87 p = z;
88 q = z + nloops;
89 j = nblocks;
90 do {
91 cptr = cptr1;
92 k = nloops >> 1;
93 do {
94 __m64 a0, a1, b0, b1, c0, c1, t10, t11, t20, t21;
95
96 a0 = *(__m64 *)&p[0];
97 a1 = *(__m64 *)&p[1];
98 b0 = *(__m64 *)&q[0];
99 b1 = *(__m64 *)&q[1];
100
101 /* complex mul */
102 c0 = *(__m64 *)&cptr[0];
103 c1 = *(__m64 *)&cptr[1];
104 /* cre*re cim*re */
105 t10 = _m_pfmul(c0, _m_punpckldq(b0, b0));
106 t11 = _m_pfmul(c1, _m_punpckldq(b1, b1));
107 c0 = *(__m64 *)&cptr[2];
108 c1 = *(__m64 *)&cptr[3];
109 /* -cim*im cre*im */
110 t20 = _m_pfmul(c0, _m_punpckhdq(b0, b0));
111 t21 = _m_pfmul(c1, _m_punpckhdq(b1, b1));
112 b0 = _m_pfadd(t10, t20);
113 b1 = _m_pfadd(t11, t21);
114
115 /* butterfly */
116 *(__m64 *)&p[0] = _m_pfadd(a0, b0);
117 *(__m64 *)&p[1] = _m_pfadd(a1, b1);
118 *(__m64 *)&q[0] = _m_pfsub(a0, b0);
119 *(__m64 *)&q[1] = _m_pfsub(a1, b1);
120
121 p += 2;
122 q += 2;
123 cptr += 4;
124 } while (--k);
125
126 p += nloops;
127 q += nloops;
128 } while (--j);
129 cptr1 += nloops * 2;
130 nblocks = nblocks >> 1;
131 nloops = nloops << 1;
132 } while (nblocks != 0);
133 _m_femms();
134 }
135
136 #endif