annotate src/fftw-3.3.8/rdft/codelet-rdft.h @ 82:d0c2a83c1364

Add FFTW 3.3.8 source, and a Linux build
author Chris Cannam
date Tue, 19 Nov 2019 14:52:55 +0000
parents
children
rev   line source
Chris@82 1 /*
Chris@82 2 * Copyright (c) 2003, 2007-14 Matteo Frigo
Chris@82 3 * Copyright (c) 2003, 2007-14 Massachusetts Institute of Technology
Chris@82 4 *
Chris@82 5 * This program is free software; you can redistribute it and/or modify
Chris@82 6 * it under the terms of the GNU General Public License as published by
Chris@82 7 * the Free Software Foundation; either version 2 of the License, or
Chris@82 8 * (at your option) any later version.
Chris@82 9 *
Chris@82 10 * This program is distributed in the hope that it will be useful,
Chris@82 11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
Chris@82 12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
Chris@82 13 * GNU General Public License for more details.
Chris@82 14 *
Chris@82 15 * You should have received a copy of the GNU General Public License
Chris@82 16 * along with this program; if not, write to the Free Software
Chris@82 17 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
Chris@82 18 *
Chris@82 19 */
Chris@82 20
Chris@82 21 /*
Chris@82 22 * This header file must include every file or define every
Chris@82 23 * type or macro which is required to compile a codelet.
Chris@82 24 */
Chris@82 25
Chris@82 26 #ifndef __RDFT_CODELET_H__
Chris@82 27 #define __RDFT_CODELET_H__
Chris@82 28
Chris@82 29 #include "kernel/ifftw.h"
Chris@82 30
Chris@82 31 /**************************************************************
Chris@82 32 * types of codelets
Chris@82 33 **************************************************************/
Chris@82 34
Chris@82 35 /* FOOab, with a,b in {0,1}, denotes the FOO transform
Chris@82 36 where a/b say whether the input/output are shifted by
Chris@82 37 half a sample/slot. */
Chris@82 38
Chris@82 39 typedef enum {
Chris@82 40 R2HC00, R2HC01, R2HC10, R2HC11,
Chris@82 41 HC2R00, HC2R01, HC2R10, HC2R11,
Chris@82 42 DHT,
Chris@82 43 REDFT00, REDFT01, REDFT10, REDFT11, /* real-even == DCT's */
Chris@82 44 RODFT00, RODFT01, RODFT10, RODFT11 /* real-odd == DST's */
Chris@82 45 } rdft_kind;
Chris@82 46
Chris@82 47 /* standard R2HC/HC2R transforms are unshifted */
Chris@82 48 #define R2HC R2HC00
Chris@82 49 #define HC2R HC2R00
Chris@82 50
Chris@82 51 #define R2HCII R2HC01
Chris@82 52 #define HC2RIII HC2R10
Chris@82 53
Chris@82 54 /* (k) >= R2HC00 produces a warning under gcc because checking x >= 0
Chris@82 55 is superfluous for unsigned values...but it is needed because other
Chris@82 56 compilers (e.g. icc) may define the enum to be a signed int...grrr. */
Chris@82 57 #define R2HC_KINDP(k) ((k) >= R2HC00 && (k) <= R2HC11) /* uses kr2hc_genus */
Chris@82 58 #define HC2R_KINDP(k) ((k) >= HC2R00 && (k) <= HC2R11) /* uses khc2r_genus */
Chris@82 59
Chris@82 60 #define R2R_KINDP(k) ((k) >= DHT) /* uses kr2r_genus */
Chris@82 61
Chris@82 62 #define REDFT_KINDP(k) ((k) >= REDFT00 && (k) <= REDFT11)
Chris@82 63 #define RODFT_KINDP(k) ((k) >= RODFT00 && (k) <= RODFT11)
Chris@82 64 #define REODFT_KINDP(k) ((k) >= REDFT00 && (k) <= RODFT11)
Chris@82 65
Chris@82 66 /* codelets with real input (output) and complex output (input) */
Chris@82 67 typedef struct kr2c_desc_s kr2c_desc;
Chris@82 68
Chris@82 69 typedef struct {
Chris@82 70 rdft_kind kind;
Chris@82 71 INT vl;
Chris@82 72 } kr2c_genus;
Chris@82 73
Chris@82 74 struct kr2c_desc_s {
Chris@82 75 INT n; /* size of transform computed */
Chris@82 76 const char *nam;
Chris@82 77 opcnt ops;
Chris@82 78 const kr2c_genus *genus;
Chris@82 79 };
Chris@82 80
Chris@82 81 typedef void (*kr2c) (R *R0, R *R1, R *Cr, R *Ci,
Chris@82 82 stride rs, stride csr, stride csi,
Chris@82 83 INT vl, INT ivs, INT ovs);
Chris@82 84 void X(kr2c_register)(planner *p, kr2c codelet, const kr2c_desc *desc);
Chris@82 85
Chris@82 86 /* half-complex to half-complex DIT/DIF codelets: */
Chris@82 87 typedef struct hc2hc_desc_s hc2hc_desc;
Chris@82 88
Chris@82 89 typedef struct {
Chris@82 90 rdft_kind kind;
Chris@82 91 INT vl;
Chris@82 92 } hc2hc_genus;
Chris@82 93
Chris@82 94 struct hc2hc_desc_s {
Chris@82 95 INT radix;
Chris@82 96 const char *nam;
Chris@82 97 const tw_instr *tw;
Chris@82 98 const hc2hc_genus *genus;
Chris@82 99 opcnt ops;
Chris@82 100 };
Chris@82 101
Chris@82 102 typedef void (*khc2hc) (R *rioarray, R *iioarray, const R *W,
Chris@82 103 stride rs, INT mb, INT me, INT ms);
Chris@82 104 void X(khc2hc_register)(planner *p, khc2hc codelet, const hc2hc_desc *desc);
Chris@82 105
Chris@82 106 /* half-complex to rdft2-complex DIT/DIF codelets: */
Chris@82 107 typedef struct hc2c_desc_s hc2c_desc;
Chris@82 108
Chris@82 109 typedef enum {
Chris@82 110 HC2C_VIA_RDFT,
Chris@82 111 HC2C_VIA_DFT
Chris@82 112 } hc2c_kind;
Chris@82 113
Chris@82 114 typedef struct {
Chris@82 115 int (*okp)(
Chris@82 116 const R *Rp, const R *Ip, const R *Rm, const R *Im,
Chris@82 117 INT rs, INT mb, INT me, INT ms,
Chris@82 118 const planner *plnr);
Chris@82 119 rdft_kind kind;
Chris@82 120 INT vl;
Chris@82 121 } hc2c_genus;
Chris@82 122
Chris@82 123 struct hc2c_desc_s {
Chris@82 124 INT radix;
Chris@82 125 const char *nam;
Chris@82 126 const tw_instr *tw;
Chris@82 127 const hc2c_genus *genus;
Chris@82 128 opcnt ops;
Chris@82 129 };
Chris@82 130
Chris@82 131 typedef void (*khc2c) (R *Rp, R *Ip, R *Rm, R *Im, const R *W,
Chris@82 132 stride rs, INT mb, INT me, INT ms);
Chris@82 133 void X(khc2c_register)(planner *p, khc2c codelet, const hc2c_desc *desc,
Chris@82 134 hc2c_kind hc2ckind);
Chris@82 135
Chris@82 136 extern const solvtab X(solvtab_rdft_r2cf);
Chris@82 137 extern const solvtab X(solvtab_rdft_r2cb);
Chris@82 138 extern const solvtab X(solvtab_rdft_sse2);
Chris@82 139 extern const solvtab X(solvtab_rdft_avx);
Chris@82 140 extern const solvtab X(solvtab_rdft_avx_128_fma);
Chris@82 141 extern const solvtab X(solvtab_rdft_avx2);
Chris@82 142 extern const solvtab X(solvtab_rdft_avx2_128);
Chris@82 143 extern const solvtab X(solvtab_rdft_avx512);
Chris@82 144 extern const solvtab X(solvtab_rdft_kcvi);
Chris@82 145 extern const solvtab X(solvtab_rdft_altivec);
Chris@82 146 extern const solvtab X(solvtab_rdft_vsx);
Chris@82 147 extern const solvtab X(solvtab_rdft_neon);
Chris@82 148 extern const solvtab X(solvtab_rdft_generic_simd128);
Chris@82 149 extern const solvtab X(solvtab_rdft_generic_simd256);
Chris@82 150
Chris@82 151 /* real-input & output DFT-like codelets (DHT, etc.) */
Chris@82 152 typedef struct kr2r_desc_s kr2r_desc;
Chris@82 153
Chris@82 154 typedef struct {
Chris@82 155 INT vl;
Chris@82 156 } kr2r_genus;
Chris@82 157
Chris@82 158 struct kr2r_desc_s {
Chris@82 159 INT n; /* size of transform computed */
Chris@82 160 const char *nam;
Chris@82 161 opcnt ops;
Chris@82 162 const kr2r_genus *genus;
Chris@82 163 rdft_kind kind;
Chris@82 164 };
Chris@82 165
Chris@82 166 typedef void (*kr2r) (const R *I, R *O, stride is, stride os,
Chris@82 167 INT vl, INT ivs, INT ovs);
Chris@82 168 void X(kr2r_register)(planner *p, kr2r codelet, const kr2r_desc *desc);
Chris@82 169
Chris@82 170 extern const solvtab X(solvtab_rdft_r2r);
Chris@82 171
Chris@82 172 #endif /* __RDFT_CODELET_H__ */