Chris@10: /* Chris@10: * Copyright (c) 2003, 2007-11 Matteo Frigo Chris@10: * Copyright (c) 2003, 2007-11 Massachusetts Institute of Technology Chris@10: * Chris@10: * This program is free software; you can redistribute it and/or modify Chris@10: * it under the terms of the GNU General Public License as published by Chris@10: * the Free Software Foundation; either version 2 of the License, or Chris@10: * (at your option) any later version. Chris@10: * Chris@10: * This program is distributed in the hope that it will be useful, Chris@10: * but WITHOUT ANY WARRANTY; without even the implied warranty of Chris@10: * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the Chris@10: * GNU General Public License for more details. Chris@10: * Chris@10: * You should have received a copy of the GNU General Public License Chris@10: * along with this program; if not, write to the Free Software Chris@10: * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA Chris@10: * Chris@10: */ Chris@10: Chris@10: /* Chris@10: * This header file must include every file or define every Chris@10: * type or macro which is required to compile a codelet. Chris@10: */ Chris@10: Chris@10: #ifndef __RDFT_CODELET_H__ Chris@10: #define __RDFT_CODELET_H__ Chris@10: Chris@10: #include "ifftw.h" Chris@10: Chris@10: /************************************************************** Chris@10: * types of codelets Chris@10: **************************************************************/ Chris@10: Chris@10: /* FOOab, with a,b in {0,1}, denotes the FOO transform Chris@10: where a/b say whether the input/output are shifted by Chris@10: half a sample/slot. */ Chris@10: Chris@10: typedef enum { Chris@10: R2HC00, R2HC01, R2HC10, R2HC11, Chris@10: HC2R00, HC2R01, HC2R10, HC2R11, Chris@10: DHT, Chris@10: REDFT00, REDFT01, REDFT10, REDFT11, /* real-even == DCT's */ Chris@10: RODFT00, RODFT01, RODFT10, RODFT11 /* real-odd == DST's */ Chris@10: } rdft_kind; Chris@10: Chris@10: /* standard R2HC/HC2R transforms are unshifted */ Chris@10: #define R2HC R2HC00 Chris@10: #define HC2R HC2R00 Chris@10: Chris@10: #define R2HCII R2HC01 Chris@10: #define HC2RIII HC2R10 Chris@10: Chris@10: /* (k) >= R2HC00 produces a warning under gcc because checking x >= 0 Chris@10: is superfluous for unsigned values...but it is needed because other Chris@10: compilers (e.g. icc) may define the enum to be a signed int...grrr. */ Chris@10: #define R2HC_KINDP(k) ((k) >= R2HC00 && (k) <= R2HC11) /* uses kr2hc_genus */ Chris@10: #define HC2R_KINDP(k) ((k) >= HC2R00 && (k) <= HC2R11) /* uses khc2r_genus */ Chris@10: Chris@10: #define R2R_KINDP(k) ((k) >= DHT) /* uses kr2r_genus */ Chris@10: Chris@10: #define REDFT_KINDP(k) ((k) >= REDFT00 && (k) <= REDFT11) Chris@10: #define RODFT_KINDP(k) ((k) >= RODFT00 && (k) <= RODFT11) Chris@10: #define REODFT_KINDP(k) ((k) >= REDFT00 && (k) <= RODFT11) Chris@10: Chris@10: /* codelets with real input (output) and complex output (input) */ Chris@10: typedef struct kr2c_desc_s kr2c_desc; Chris@10: Chris@10: typedef struct { Chris@10: rdft_kind kind; Chris@10: INT vl; Chris@10: } kr2c_genus; Chris@10: Chris@10: struct kr2c_desc_s { Chris@10: INT n; /* size of transform computed */ Chris@10: const char *nam; Chris@10: opcnt ops; Chris@10: const kr2c_genus *genus; Chris@10: }; Chris@10: Chris@10: typedef void (*kr2c) (R *R0, R *R1, R *Cr, R *Ci, Chris@10: stride rs, stride csr, stride csi, Chris@10: INT vl, INT ivs, INT ovs); Chris@10: void X(kr2c_register)(planner *p, kr2c codelet, const kr2c_desc *desc); Chris@10: Chris@10: /* half-complex to half-complex DIT/DIF codelets: */ Chris@10: typedef struct hc2hc_desc_s hc2hc_desc; Chris@10: Chris@10: typedef struct { Chris@10: rdft_kind kind; Chris@10: INT vl; Chris@10: } hc2hc_genus; Chris@10: Chris@10: struct hc2hc_desc_s { Chris@10: INT radix; Chris@10: const char *nam; Chris@10: const tw_instr *tw; Chris@10: const hc2hc_genus *genus; Chris@10: opcnt ops; Chris@10: }; Chris@10: Chris@10: typedef void (*khc2hc) (R *rioarray, R *iioarray, const R *W, Chris@10: stride rs, INT mb, INT me, INT ms); Chris@10: void X(khc2hc_register)(planner *p, khc2hc codelet, const hc2hc_desc *desc); Chris@10: Chris@10: /* half-complex to rdft2-complex DIT/DIF codelets: */ Chris@10: typedef struct hc2c_desc_s hc2c_desc; Chris@10: Chris@10: typedef enum { Chris@10: HC2C_VIA_RDFT, Chris@10: HC2C_VIA_DFT Chris@10: } hc2c_kind; Chris@10: Chris@10: typedef struct { Chris@10: int (*okp)( Chris@10: const R *Rp, const R *Ip, const R *Rm, const R *Im, Chris@10: INT rs, INT mb, INT me, INT ms, Chris@10: const planner *plnr); Chris@10: rdft_kind kind; Chris@10: INT vl; Chris@10: } hc2c_genus; Chris@10: Chris@10: struct hc2c_desc_s { Chris@10: INT radix; Chris@10: const char *nam; Chris@10: const tw_instr *tw; Chris@10: const hc2c_genus *genus; Chris@10: opcnt ops; Chris@10: }; Chris@10: Chris@10: typedef void (*khc2c) (R *Rp, R *Ip, R *Rm, R *Im, const R *W, Chris@10: stride rs, INT mb, INT me, INT ms); Chris@10: void X(khc2c_register)(planner *p, khc2c codelet, const hc2c_desc *desc, Chris@10: hc2c_kind hc2ckind); Chris@10: Chris@10: extern const solvtab X(solvtab_rdft_r2cf); Chris@10: extern const solvtab X(solvtab_rdft_r2cb); Chris@10: extern const solvtab X(solvtab_rdft_sse2); Chris@10: extern const solvtab X(solvtab_rdft_avx); Chris@10: extern const solvtab X(solvtab_rdft_altivec); Chris@10: extern const solvtab X(solvtab_rdft_neon); Chris@10: Chris@10: /* real-input & output DFT-like codelets (DHT, etc.) */ Chris@10: typedef struct kr2r_desc_s kr2r_desc; Chris@10: Chris@10: typedef struct { Chris@10: INT vl; Chris@10: } kr2r_genus; Chris@10: Chris@10: struct kr2r_desc_s { Chris@10: INT n; /* size of transform computed */ Chris@10: const char *nam; Chris@10: opcnt ops; Chris@10: const kr2r_genus *genus; Chris@10: rdft_kind kind; Chris@10: }; Chris@10: Chris@10: typedef void (*kr2r) (const R *I, R *O, stride is, stride os, Chris@10: INT vl, INT ivs, INT ovs); Chris@10: void X(kr2r_register)(planner *p, kr2r codelet, const kr2r_desc *desc); Chris@10: Chris@10: extern const solvtab X(solvtab_rdft_r2r); Chris@10: Chris@10: #endif /* __RDFT_CODELET_H__ */