Use full path for #includes from another directory.
[libav.git] / libavcodec / ppc / fft_altivec.c
1 /*
2 * FFT/IFFT transforms
3 * AltiVec-enabled
4 * Copyright (c) 2003 Romain Dolbeau <romain@dolbeau.org>
5 * Based on code Copyright (c) 2002 Fabrice Bellard.
6 *
7 * This file is part of FFmpeg.
8 *
9 * FFmpeg is free software; you can redistribute it and/or
10 * modify it under the terms of the GNU Lesser General Public
11 * License as published by the Free Software Foundation; either
12 * version 2.1 of the License, or (at your option) any later version.
13 *
14 * FFmpeg is distributed in the hope that it will be useful,
15 * but WITHOUT ANY WARRANTY; without even the implied warranty of
16 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
17 * Lesser General Public License for more details.
18 *
19 * You should have received a copy of the GNU Lesser General Public
20 * License along with FFmpeg; if not, write to the Free Software
21 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
22 */
23 #include "libavcodec/dsputil.h"
24
25 #include "gcc_fixes.h"
26
27 #include "dsputil_ppc.h"
28 #include "util_altivec.h"
29 /*
30 those three macros are from libavcodec/fft.c
31 and are required for the reference C code
32 */
33 /* butter fly op */
34 #define BF(pre, pim, qre, qim, pre1, pim1, qre1, qim1) \
35 {\
36 FFTSample ax, ay, bx, by;\
37 bx=pre1;\
38 by=pim1;\
39 ax=qre1;\
40 ay=qim1;\
41 pre = (bx + ax);\
42 pim = (by + ay);\
43 qre = (bx - ax);\
44 qim = (by - ay);\
45 }
46 #define MUL16(a,b) ((a) * (b))
47 #define CMUL(pre, pim, are, aim, bre, bim) \
48 {\
49 pre = (MUL16(are, bre) - MUL16(aim, bim));\
50 pim = (MUL16(are, bim) + MUL16(bre, aim));\
51 }
52
53
54 /**
55 * Do a complex FFT with the parameters defined in ff_fft_init(). The
56 * input data must be permuted before with s->revtab table. No
57 * 1.0/sqrt(n) normalization is done.
58 * AltiVec-enabled
59 * This code assumes that the 'z' pointer is 16 bytes-aligned
60 * It also assumes all FFTComplex are 8 bytes-aligned pair of float
61 * The code is exactly the same as the SSE version, except
62 * that successive MUL + ADD/SUB have been merged into
63 * fused multiply-add ('vec_madd' in altivec)
64 */
65 void ff_fft_calc_altivec(FFTContext *s, FFTComplex *z)
66 {
67 POWERPC_PERF_DECLARE(altivec_fft_num, s->nbits >= 6);
68 register const vector float vczero = (const vector float)vec_splat_u32(0.);
69
70 int ln = s->nbits;
71 int j, np, np2;
72 int nblocks, nloops;
73 register FFTComplex *p, *q;
74 FFTComplex *cptr, *cptr1;
75 int k;
76
77 POWERPC_PERF_START_COUNT(altivec_fft_num, s->nbits >= 6);
78
79 np = 1 << ln;
80
81 {
82 vector float *r, a, b, a1, c1, c2;
83
84 r = (vector float *)&z[0];
85
86 c1 = vcii(p,p,n,n);
87
88 if (s->inverse)
89 {
90 c2 = vcii(p,p,n,p);
91 }
92 else
93 {
94 c2 = vcii(p,p,p,n);
95 }
96
97 j = (np >> 2);
98 do {
99 a = vec_ld(0, r);
100 a1 = vec_ld(sizeof(vector float), r);
101
102 b = vec_perm(a,a,vcprmle(1,0,3,2));
103 a = vec_madd(a,c1,b);
104 /* do the pass 0 butterfly */
105
106 b = vec_perm(a1,a1,vcprmle(1,0,3,2));
107 b = vec_madd(a1,c1,b);
108 /* do the pass 0 butterfly */
109
110 /* multiply third by -i */
111 b = vec_perm(b,b,vcprmle(2,3,1,0));
112
113 /* do the pass 1 butterfly */
114 vec_st(vec_madd(b,c2,a), 0, r);
115 vec_st(vec_nmsub(b,c2,a), sizeof(vector float), r);
116
117 r += 2;
118 } while (--j != 0);
119 }
120 /* pass 2 .. ln-1 */
121
122 nblocks = np >> 3;
123 nloops = 1 << 2;
124 np2 = np >> 1;
125
126 cptr1 = s->exptab1;
127 do {
128 p = z;
129 q = z + nloops;
130 j = nblocks;
131 do {
132 cptr = cptr1;
133 k = nloops >> 1;
134 do {
135 vector float a,b,c,t1;
136
137 a = vec_ld(0, (float*)p);
138 b = vec_ld(0, (float*)q);
139
140 /* complex mul */
141 c = vec_ld(0, (float*)cptr);
142 /* cre*re cim*re */
143 t1 = vec_madd(c, vec_perm(b,b,vcprmle(2,2,0,0)),vczero);
144 c = vec_ld(sizeof(vector float), (float*)cptr);
145 /* -cim*im cre*im */
146 b = vec_madd(c, vec_perm(b,b,vcprmle(3,3,1,1)),t1);
147
148 /* butterfly */
149 vec_st(vec_add(a,b), 0, (float*)p);
150 vec_st(vec_sub(a,b), 0, (float*)q);
151
152 p += 2;
153 q += 2;
154 cptr += 4;
155 } while (--k);
156
157 p += nloops;
158 q += nloops;
159 } while (--j);
160 cptr1 += nloops * 2;
161 nblocks = nblocks >> 1;
162 nloops = nloops << 1;
163 } while (nblocks != 0);
164
165 POWERPC_PERF_STOP_COUNT(altivec_fft_num, s->nbits >= 6);
166 }