8bd7b89d01d58dc3d798d2b2c192340ae8d0479a
[libav.git] / libavcodec / i386 / fft_3dn.c
1 /*
2 * FFT/MDCT transform with 3DNow! optimizations
3 * Copyright (c) 2006 Zuxy MENG Jie, Loren Merritt
4 * Based on fft_sse.c copyright (c) 2002 Fabrice Bellard.
5 *
6 * This file is part of FFmpeg.
7 *
8 * FFmpeg is free software; you can redistribute it and/or
9 * modify it under the terms of the GNU Lesser General Public
10 * License as published by the Free Software Foundation; either
11 * version 2.1 of the License, or (at your option) any later version.
12 *
13 * FFmpeg is distributed in the hope that it will be useful,
14 * but WITHOUT ANY WARRANTY; without even the implied warranty of
15 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
16 * Lesser General Public License for more details.
17 *
18 * You should have received a copy of the GNU Lesser General Public
19 * License along with FFmpeg; if not, write to the Free Software
20 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
21 */
22
23 #include "libavutil/x86_cpu.h"
24 #include "libavcodec/dsputil.h"
25
26 static const int p1m1[2] __attribute__((aligned(8))) =
27 { 0, 1 << 31 };
28
29 static const int m1p1[2] __attribute__((aligned(8))) =
30 { 1 << 31, 0 };
31
32 void ff_fft_calc_3dn(FFTContext *s, FFTComplex *z)
33 {
34 int ln = s->nbits;
35 long j;
36 x86_reg i;
37 long nblocks, nloops;
38 FFTComplex *p, *cptr;
39
40 asm volatile(
41 /* FEMMS is not a must here but recommended by AMD */
42 "femms \n\t"
43 "movq %0, %%mm7 \n\t"
44 ::"m"(*(s->inverse ? m1p1 : p1m1))
45 );
46
47 i = 8 << ln;
48 asm volatile(
49 "1: \n\t"
50 "sub $32, %0 \n\t"
51 "movq (%0,%1), %%mm0 \n\t"
52 "movq 16(%0,%1), %%mm1 \n\t"
53 "movq 8(%0,%1), %%mm2 \n\t"
54 "movq 24(%0,%1), %%mm3 \n\t"
55 "movq %%mm0, %%mm4 \n\t"
56 "movq %%mm1, %%mm5 \n\t"
57 "pfadd %%mm2, %%mm0 \n\t"
58 "pfadd %%mm3, %%mm1 \n\t"
59 "pfsub %%mm2, %%mm4 \n\t"
60 "pfsub %%mm3, %%mm5 \n\t"
61 "movq %%mm0, %%mm2 \n\t"
62 "punpckldq %%mm5, %%mm6 \n\t"
63 "punpckhdq %%mm6, %%mm5 \n\t"
64 "movq %%mm4, %%mm3 \n\t"
65 "pxor %%mm7, %%mm5 \n\t"
66 "pfadd %%mm1, %%mm0 \n\t"
67 "pfadd %%mm5, %%mm4 \n\t"
68 "pfsub %%mm1, %%mm2 \n\t"
69 "pfsub %%mm5, %%mm3 \n\t"
70 "movq %%mm0, (%0,%1) \n\t"
71 "movq %%mm4, 8(%0,%1) \n\t"
72 "movq %%mm2, 16(%0,%1) \n\t"
73 "movq %%mm3, 24(%0,%1) \n\t"
74 "jg 1b \n\t"
75 :"+r"(i)
76 :"r"(z)
77 );
78 /* pass 2 .. ln-1 */
79
80 nblocks = 1 << (ln-3);
81 nloops = 1 << 2;
82 cptr = s->exptab1;
83 do {
84 p = z;
85 j = nblocks;
86 do {
87 i = nloops*8;
88 asm volatile(
89 "1: \n\t"
90 "sub $16, %0 \n\t"
91 "movq (%1,%0), %%mm0 \n\t"
92 "movq 8(%1,%0), %%mm1 \n\t"
93 "movq (%2,%0), %%mm2 \n\t"
94 "movq 8(%2,%0), %%mm3 \n\t"
95 "movq %%mm2, %%mm4 \n\t"
96 "movq %%mm3, %%mm5 \n\t"
97 "punpckldq %%mm2, %%mm2 \n\t"
98 "punpckldq %%mm3, %%mm3 \n\t"
99 "punpckhdq %%mm4, %%mm4 \n\t"
100 "punpckhdq %%mm5, %%mm5 \n\t"
101 "pfmul (%3,%0,2), %%mm2 \n\t" // cre*re cim*re
102 "pfmul 8(%3,%0,2), %%mm3 \n\t"
103 "pfmul 16(%3,%0,2), %%mm4 \n\t" // -cim*im cre*im
104 "pfmul 24(%3,%0,2), %%mm5 \n\t"
105 "pfadd %%mm2, %%mm4 \n\t" // cre*re-cim*im cim*re+cre*im
106 "pfadd %%mm3, %%mm5 \n\t"
107 "movq %%mm0, %%mm2 \n\t"
108 "movq %%mm1, %%mm3 \n\t"
109 "pfadd %%mm4, %%mm0 \n\t"
110 "pfadd %%mm5, %%mm1 \n\t"
111 "pfsub %%mm4, %%mm2 \n\t"
112 "pfsub %%mm5, %%mm3 \n\t"
113 "movq %%mm0, (%1,%0) \n\t"
114 "movq %%mm1, 8(%1,%0) \n\t"
115 "movq %%mm2, (%2,%0) \n\t"
116 "movq %%mm3, 8(%2,%0) \n\t"
117 "jg 1b \n\t"
118 :"+r"(i)
119 :"r"(p), "r"(p + nloops), "r"(cptr)
120 );
121 p += nloops*2;
122 } while (--j);
123 cptr += nloops*2;
124 nblocks >>= 1;
125 nloops <<= 1;
126 } while (nblocks != 0);
127 asm volatile("femms");
128 }