Chris@10
|
1 /*
|
Chris@10
|
2 * Copyright (c) 2003, 2007-11 Matteo Frigo
|
Chris@10
|
3 * Copyright (c) 2003, 2007-11 Massachusetts Institute of Technology
|
Chris@10
|
4 *
|
Chris@10
|
5 * This program is free software; you can redistribute it and/or modify
|
Chris@10
|
6 * it under the terms of the GNU General Public License as published by
|
Chris@10
|
7 * the Free Software Foundation; either version 2 of the License, or
|
Chris@10
|
8 * (at your option) any later version.
|
Chris@10
|
9 *
|
Chris@10
|
10 * This program is distributed in the hope that it will be useful,
|
Chris@10
|
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
|
Chris@10
|
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
|
Chris@10
|
13 * GNU General Public License for more details.
|
Chris@10
|
14 *
|
Chris@10
|
15 * You should have received a copy of the GNU General Public License
|
Chris@10
|
16 * along with this program; if not, write to the Free Software
|
Chris@10
|
17 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
|
Chris@10
|
18 *
|
Chris@10
|
19 */
|
Chris@10
|
20
|
Chris@10
|
21 /*
|
Chris@10
|
22 * This header file must include every file or define every
|
Chris@10
|
23 * type or macro which is required to compile a codelet.
|
Chris@10
|
24 */
|
Chris@10
|
25
|
Chris@10
|
26 #ifndef __RDFT_CODELET_H__
|
Chris@10
|
27 #define __RDFT_CODELET_H__
|
Chris@10
|
28
|
Chris@10
|
29 #include "ifftw.h"
|
Chris@10
|
30
|
Chris@10
|
31 /**************************************************************
|
Chris@10
|
32 * types of codelets
|
Chris@10
|
33 **************************************************************/
|
Chris@10
|
34
|
Chris@10
|
35 /* FOOab, with a,b in {0,1}, denotes the FOO transform
|
Chris@10
|
36 where a/b say whether the input/output are shifted by
|
Chris@10
|
37 half a sample/slot. */
|
Chris@10
|
38
|
Chris@10
|
39 typedef enum {
|
Chris@10
|
40 R2HC00, R2HC01, R2HC10, R2HC11,
|
Chris@10
|
41 HC2R00, HC2R01, HC2R10, HC2R11,
|
Chris@10
|
42 DHT,
|
Chris@10
|
43 REDFT00, REDFT01, REDFT10, REDFT11, /* real-even == DCT's */
|
Chris@10
|
44 RODFT00, RODFT01, RODFT10, RODFT11 /* real-odd == DST's */
|
Chris@10
|
45 } rdft_kind;
|
Chris@10
|
46
|
Chris@10
|
47 /* standard R2HC/HC2R transforms are unshifted */
|
Chris@10
|
48 #define R2HC R2HC00
|
Chris@10
|
49 #define HC2R HC2R00
|
Chris@10
|
50
|
Chris@10
|
51 #define R2HCII R2HC01
|
Chris@10
|
52 #define HC2RIII HC2R10
|
Chris@10
|
53
|
Chris@10
|
54 /* (k) >= R2HC00 produces a warning under gcc because checking x >= 0
|
Chris@10
|
55 is superfluous for unsigned values...but it is needed because other
|
Chris@10
|
56 compilers (e.g. icc) may define the enum to be a signed int...grrr. */
|
Chris@10
|
57 #define R2HC_KINDP(k) ((k) >= R2HC00 && (k) <= R2HC11) /* uses kr2hc_genus */
|
Chris@10
|
58 #define HC2R_KINDP(k) ((k) >= HC2R00 && (k) <= HC2R11) /* uses khc2r_genus */
|
Chris@10
|
59
|
Chris@10
|
60 #define R2R_KINDP(k) ((k) >= DHT) /* uses kr2r_genus */
|
Chris@10
|
61
|
Chris@10
|
62 #define REDFT_KINDP(k) ((k) >= REDFT00 && (k) <= REDFT11)
|
Chris@10
|
63 #define RODFT_KINDP(k) ((k) >= RODFT00 && (k) <= RODFT11)
|
Chris@10
|
64 #define REODFT_KINDP(k) ((k) >= REDFT00 && (k) <= RODFT11)
|
Chris@10
|
65
|
Chris@10
|
66 /* codelets with real input (output) and complex output (input) */
|
Chris@10
|
67 typedef struct kr2c_desc_s kr2c_desc;
|
Chris@10
|
68
|
Chris@10
|
69 typedef struct {
|
Chris@10
|
70 rdft_kind kind;
|
Chris@10
|
71 INT vl;
|
Chris@10
|
72 } kr2c_genus;
|
Chris@10
|
73
|
Chris@10
|
74 struct kr2c_desc_s {
|
Chris@10
|
75 INT n; /* size of transform computed */
|
Chris@10
|
76 const char *nam;
|
Chris@10
|
77 opcnt ops;
|
Chris@10
|
78 const kr2c_genus *genus;
|
Chris@10
|
79 };
|
Chris@10
|
80
|
Chris@10
|
81 typedef void (*kr2c) (R *R0, R *R1, R *Cr, R *Ci,
|
Chris@10
|
82 stride rs, stride csr, stride csi,
|
Chris@10
|
83 INT vl, INT ivs, INT ovs);
|
Chris@10
|
84 void X(kr2c_register)(planner *p, kr2c codelet, const kr2c_desc *desc);
|
Chris@10
|
85
|
Chris@10
|
86 /* half-complex to half-complex DIT/DIF codelets: */
|
Chris@10
|
87 typedef struct hc2hc_desc_s hc2hc_desc;
|
Chris@10
|
88
|
Chris@10
|
89 typedef struct {
|
Chris@10
|
90 rdft_kind kind;
|
Chris@10
|
91 INT vl;
|
Chris@10
|
92 } hc2hc_genus;
|
Chris@10
|
93
|
Chris@10
|
94 struct hc2hc_desc_s {
|
Chris@10
|
95 INT radix;
|
Chris@10
|
96 const char *nam;
|
Chris@10
|
97 const tw_instr *tw;
|
Chris@10
|
98 const hc2hc_genus *genus;
|
Chris@10
|
99 opcnt ops;
|
Chris@10
|
100 };
|
Chris@10
|
101
|
Chris@10
|
102 typedef void (*khc2hc) (R *rioarray, R *iioarray, const R *W,
|
Chris@10
|
103 stride rs, INT mb, INT me, INT ms);
|
Chris@10
|
104 void X(khc2hc_register)(planner *p, khc2hc codelet, const hc2hc_desc *desc);
|
Chris@10
|
105
|
Chris@10
|
106 /* half-complex to rdft2-complex DIT/DIF codelets: */
|
Chris@10
|
107 typedef struct hc2c_desc_s hc2c_desc;
|
Chris@10
|
108
|
Chris@10
|
109 typedef enum {
|
Chris@10
|
110 HC2C_VIA_RDFT,
|
Chris@10
|
111 HC2C_VIA_DFT
|
Chris@10
|
112 } hc2c_kind;
|
Chris@10
|
113
|
Chris@10
|
114 typedef struct {
|
Chris@10
|
115 int (*okp)(
|
Chris@10
|
116 const R *Rp, const R *Ip, const R *Rm, const R *Im,
|
Chris@10
|
117 INT rs, INT mb, INT me, INT ms,
|
Chris@10
|
118 const planner *plnr);
|
Chris@10
|
119 rdft_kind kind;
|
Chris@10
|
120 INT vl;
|
Chris@10
|
121 } hc2c_genus;
|
Chris@10
|
122
|
Chris@10
|
123 struct hc2c_desc_s {
|
Chris@10
|
124 INT radix;
|
Chris@10
|
125 const char *nam;
|
Chris@10
|
126 const tw_instr *tw;
|
Chris@10
|
127 const hc2c_genus *genus;
|
Chris@10
|
128 opcnt ops;
|
Chris@10
|
129 };
|
Chris@10
|
130
|
Chris@10
|
131 typedef void (*khc2c) (R *Rp, R *Ip, R *Rm, R *Im, const R *W,
|
Chris@10
|
132 stride rs, INT mb, INT me, INT ms);
|
Chris@10
|
133 void X(khc2c_register)(planner *p, khc2c codelet, const hc2c_desc *desc,
|
Chris@10
|
134 hc2c_kind hc2ckind);
|
Chris@10
|
135
|
Chris@10
|
136 extern const solvtab X(solvtab_rdft_r2cf);
|
Chris@10
|
137 extern const solvtab X(solvtab_rdft_r2cb);
|
Chris@10
|
138 extern const solvtab X(solvtab_rdft_sse2);
|
Chris@10
|
139 extern const solvtab X(solvtab_rdft_avx);
|
Chris@10
|
140 extern const solvtab X(solvtab_rdft_altivec);
|
Chris@10
|
141 extern const solvtab X(solvtab_rdft_neon);
|
Chris@10
|
142
|
Chris@10
|
143 /* real-input & output DFT-like codelets (DHT, etc.) */
|
Chris@10
|
144 typedef struct kr2r_desc_s kr2r_desc;
|
Chris@10
|
145
|
Chris@10
|
146 typedef struct {
|
Chris@10
|
147 INT vl;
|
Chris@10
|
148 } kr2r_genus;
|
Chris@10
|
149
|
Chris@10
|
150 struct kr2r_desc_s {
|
Chris@10
|
151 INT n; /* size of transform computed */
|
Chris@10
|
152 const char *nam;
|
Chris@10
|
153 opcnt ops;
|
Chris@10
|
154 const kr2r_genus *genus;
|
Chris@10
|
155 rdft_kind kind;
|
Chris@10
|
156 };
|
Chris@10
|
157
|
Chris@10
|
158 typedef void (*kr2r) (const R *I, R *O, stride is, stride os,
|
Chris@10
|
159 INT vl, INT ivs, INT ovs);
|
Chris@10
|
160 void X(kr2r_register)(planner *p, kr2r codelet, const kr2r_desc *desc);
|
Chris@10
|
161
|
Chris@10
|
162 extern const solvtab X(solvtab_rdft_r2r);
|
Chris@10
|
163
|
Chris@10
|
164 #endif /* __RDFT_CODELET_H__ */
|