Chris@10: /*
Chris@10:  * Copyright (c) 2001 Matteo Frigo
Chris@10:  * Copyright (c) 2001 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: #ifndef __BENCH_USER_H__
Chris@10: #define __BENCH_USER_H__
Chris@10: 
Chris@10: #ifdef __cplusplus
Chris@10: extern "C" {
Chris@10: #endif                          /* __cplusplus */
Chris@10: 
Chris@10: /* benchmark program definitions for user code */
Chris@10: #include "config.h"
Chris@10: 
Chris@10: #if HAVE_STDDEF_H
Chris@10: #include <stddef.h>
Chris@10: #endif
Chris@10: 
Chris@10: #if HAVE_STDLIB_H
Chris@10: #include <stdlib.h>
Chris@10: #endif
Chris@10: 
Chris@10: #if defined(BENCHFFT_SINGLE)
Chris@10: typedef float bench_real;
Chris@10: #elif defined(BENCHFFT_LDOUBLE)
Chris@10: typedef long double bench_real;
Chris@10: #elif defined(BENCHFFT_QUAD)
Chris@10: typedef __float128 bench_real;
Chris@10: #else
Chris@10: typedef double bench_real;
Chris@10: #endif
Chris@10: 
Chris@10: typedef bench_real bench_complex[2];
Chris@10: 
Chris@10: #define c_re(c)  ((c)[0])
Chris@10: #define c_im(c)  ((c)[1])
Chris@10: 
Chris@10: #undef DOUBLE_PRECISION
Chris@10: #define DOUBLE_PRECISION (sizeof(bench_real) == sizeof(double))
Chris@10: #undef SINGLE_PRECISION
Chris@10: #define SINGLE_PRECISION (!DOUBLE_PRECISION && sizeof(bench_real) == sizeof(float))
Chris@10: #undef LDOUBLE_PRECISION
Chris@10: #define LDOUBLE_PRECISION (!DOUBLE_PRECISION && sizeof(bench_real) == sizeof(long double))
Chris@10: 
Chris@10: #undef QUAD_PRECISION
Chris@10: #ifdef BENCHFFT_QUAD
Chris@10: #define QUAD_PRECISION (!LDOUBLE_PRECISION && sizeof(bench_real) == sizeof(__float128))
Chris@10: #else
Chris@10: #define QUAD_PRECISION 0
Chris@10: #endif
Chris@10: 
Chris@10: typedef enum { PROBLEM_COMPLEX, PROBLEM_REAL, PROBLEM_R2R } problem_kind_t;
Chris@10: 
Chris@10: typedef enum {
Chris@10:      R2R_R2HC, R2R_HC2R, R2R_DHT,
Chris@10:      R2R_REDFT00, R2R_REDFT01, R2R_REDFT10, R2R_REDFT11,
Chris@10:      R2R_RODFT00, R2R_RODFT01, R2R_RODFT10, R2R_RODFT11
Chris@10: } r2r_kind_t;
Chris@10: 
Chris@10: typedef struct {
Chris@10:      int n;
Chris@10:      int is;			/* input stride */
Chris@10:      int os;			/* output stride */
Chris@10: } bench_iodim;
Chris@10: 
Chris@10: typedef struct {
Chris@10:      int rnk;
Chris@10:      bench_iodim *dims;
Chris@10: } bench_tensor;
Chris@10: 
Chris@10: bench_tensor *mktensor(int rnk);
Chris@10: void tensor_destroy(bench_tensor *sz);
Chris@10: int tensor_sz(const bench_tensor *sz);
Chris@10: bench_tensor *tensor_compress(const bench_tensor *sz);
Chris@10: int tensor_unitstridep(bench_tensor *t);
Chris@10: int tensor_rowmajorp(bench_tensor *t);
Chris@10: int tensor_real_rowmajorp(bench_tensor *t, int sign, int in_place);
Chris@10: bench_tensor *tensor_append(const bench_tensor *a, const bench_tensor *b);
Chris@10: bench_tensor *tensor_copy(const bench_tensor *sz);
Chris@10: bench_tensor *tensor_copy_sub(const bench_tensor *sz, int start_dim, int rnk);
Chris@10: bench_tensor *tensor_copy_swapio(const bench_tensor *sz);
Chris@10: void tensor_ibounds(bench_tensor *t, int *lbp, int *ubp);
Chris@10: void tensor_obounds(bench_tensor *t, int *lbp, int *ubp);
Chris@10: 
Chris@10: /*
Chris@10:   Definition of rank -infinity.
Chris@10:   This definition has the property that if you want rank 0 or 1,
Chris@10:   you can simply test for rank <= 1.  This is a common case.
Chris@10:  
Chris@10:   A tensor of rank -infinity has size 0.
Chris@10: */
Chris@10: #define RNK_MINFTY  ((int)(((unsigned) -1) >> 1))
Chris@10: #define FINITE_RNK(rnk) ((rnk) != RNK_MINFTY)
Chris@10: 
Chris@10: typedef struct {
Chris@10:      problem_kind_t kind;
Chris@10:      r2r_kind_t *k;
Chris@10:      bench_tensor *sz;
Chris@10:      bench_tensor *vecsz;
Chris@10:      int sign;
Chris@10:      int in_place;
Chris@10:      int destroy_input;
Chris@10:      int split;
Chris@10:      void *in, *out;
Chris@10:      void *inphys, *outphys;
Chris@10:      int iphyssz, ophyssz;
Chris@10:      char *pstring;
Chris@10:      void *userinfo; /* user can store whatever */
Chris@10:      int scrambled_in, scrambled_out; /* hack for MPI */
Chris@10: 
Chris@10:      /* internal hack so that we can use verifier in FFTW test program */
Chris@10:      void *ini, *outi; /* if nonzero, point to imag. parts for dft */
Chris@10: 
Chris@10:      /* another internal hack to avoid passing around too many parameters */
Chris@10:      double setup_time;
Chris@10: } bench_problem;
Chris@10: 
Chris@10: extern int verbose;
Chris@10: 
Chris@10: extern int no_speed_allocation;
Chris@10: 
Chris@10: extern int always_pad_real;
Chris@10: 
Chris@10: #define LIBBENCH_TIMER 0
Chris@10: #define USER_TIMER 1
Chris@10: #define BENCH_NTIMERS 2
Chris@10: extern void timer_start(int which_timer);
Chris@10: extern double timer_stop(int which_timer);
Chris@10: 
Chris@10: extern int can_do(bench_problem *p);
Chris@10: extern void setup(bench_problem *p);
Chris@10: extern void doit(int iter, bench_problem *p);
Chris@10: extern void done(bench_problem *p);
Chris@10: extern void main_init(int *argc, char ***argv);
Chris@10: extern void cleanup(void);
Chris@10: extern void verify(const char *param, int rounds, double tol);
Chris@10: extern void useropt(const char *arg);
Chris@10: 
Chris@10: extern void verify_problem(bench_problem *p, int rounds, double tol);
Chris@10: 
Chris@10: extern void problem_alloc(bench_problem *p);
Chris@10: extern void problem_free(bench_problem *p);
Chris@10: extern void problem_zero(bench_problem *p);
Chris@10: extern void problem_destroy(bench_problem *p);
Chris@10: 
Chris@10: extern int power_of_two(int n);
Chris@10: extern int log_2(int n);
Chris@10: 
Chris@10: 
Chris@10: #define CASSIGN(out, in) (c_re(out) = c_re(in), c_im(out) = c_im(in))
Chris@10: 
Chris@10: bench_tensor *verify_pack(const bench_tensor *sz, int s);
Chris@10: 
Chris@10: typedef struct {
Chris@10:      double l;
Chris@10:      double i;
Chris@10:      double s;
Chris@10: } errors;
Chris@10: 
Chris@10: void verify_dft(bench_problem *p, int rounds, double tol, errors *e);
Chris@10: void verify_rdft2(bench_problem *p, int rounds, double tol, errors *e);
Chris@10: void verify_r2r(bench_problem *p, int rounds, double tol, errors *e);
Chris@10: 
Chris@10: /**************************************************************/
Chris@10: /* routines to override */
Chris@10: 
Chris@10: extern void after_problem_ccopy_from(bench_problem *p, bench_real *ri, bench_real *ii);
Chris@10: extern void after_problem_ccopy_to(bench_problem *p, bench_real *ro, bench_real *io);
Chris@10: extern void after_problem_hccopy_from(bench_problem *p, bench_real *ri, bench_real *ii);
Chris@10: extern void after_problem_hccopy_to(bench_problem *p, bench_real *ro, bench_real *io);
Chris@10: extern void after_problem_rcopy_from(bench_problem *p, bench_real *ri);
Chris@10: extern void after_problem_rcopy_to(bench_problem *p, bench_real *ro);
Chris@10: extern void bench_exit(int status);
Chris@10: extern double bench_cost_postprocess(double cost);
Chris@10: 
Chris@10: /**************************************************************
Chris@10:  * malloc
Chris@10:  **************************************************************/
Chris@10: extern void *bench_malloc(size_t size);
Chris@10: extern void bench_free(void *ptr);
Chris@10: extern void bench_free0(void *ptr);
Chris@10: 
Chris@10: /**************************************************************
Chris@10:  * alloca
Chris@10:  **************************************************************/
Chris@10: #ifdef HAVE_ALLOCA_H
Chris@10: #include <alloca.h>
Chris@10: #endif
Chris@10: 
Chris@10: /**************************************************************
Chris@10:  * assert
Chris@10:  **************************************************************/
Chris@10: extern void bench_assertion_failed(const char *s, int line, const char *file);
Chris@10: #define BENCH_ASSERT(ex)						 \
Chris@10:       (void)((ex) || (bench_assertion_failed(#ex, __LINE__, __FILE__), 0))
Chris@10: 
Chris@10: #define UNUSED(x) (void)x
Chris@10: 
Chris@10: /***************************************
Chris@10:  * Documentation strings
Chris@10:  ***************************************/
Chris@10: struct bench_doc {
Chris@10:      const char *key;
Chris@10:      const char *val;
Chris@10:      const char *(*f)(void);
Chris@10: };
Chris@10: 
Chris@10: extern struct bench_doc bench_doc[];
Chris@10: 
Chris@10: #ifdef CC
Chris@10: #define CC_DOC BENCH_DOC("cc", CC)
Chris@10: #elif defined(BENCH_CC)
Chris@10: #define CC_DOC BENCH_DOC("cc", BENCH_CC)
Chris@10: #else
Chris@10: #define CC_DOC /* none */
Chris@10: #endif
Chris@10: 
Chris@10: #ifdef CXX
Chris@10: #define CXX_DOC BENCH_DOC("cxx", CXX)
Chris@10: #elif defined(BENCH_CXX)
Chris@10: #define CXX_DOC BENCH_DOC("cxx", BENCH_CXX)
Chris@10: #else
Chris@10: #define CXX_DOC /* none */
Chris@10: #endif
Chris@10: 
Chris@10: #ifdef F77
Chris@10: #define F77_DOC BENCH_DOC("f77", F77)
Chris@10: #elif defined(BENCH_F77)
Chris@10: #define F77_DOC BENCH_DOC("f77", BENCH_F77)
Chris@10: #else
Chris@10: #define F77_DOC /* none */
Chris@10: #endif
Chris@10: 
Chris@10: #ifdef F90
Chris@10: #define F90_DOC BENCH_DOC("f90", F90)
Chris@10: #elif defined(BENCH_F90)
Chris@10: #define F90_DOC BENCH_DOC("f90", BENCH_F90)
Chris@10: #else
Chris@10: #define F90_DOC /* none */
Chris@10: #endif
Chris@10: 
Chris@10: #define BEGIN_BENCH_DOC						\
Chris@10: struct bench_doc bench_doc[] = {				\
Chris@10:     CC_DOC							\
Chris@10:     CXX_DOC							\
Chris@10:     F77_DOC							\
Chris@10:     F90_DOC
Chris@10: 
Chris@10: #define BENCH_DOC(key, val) { key, val, 0 },
Chris@10: #define BENCH_DOCF(key, f) { key, 0, f },
Chris@10: 
Chris@10: #define END_BENCH_DOC				\
Chris@10:      {0, 0, 0}};
Chris@10: 
Chris@10: #ifdef __cplusplus
Chris@10: }                               /* extern "C" */
Chris@10: #endif                          /* __cplusplus */
Chris@10:     
Chris@10: #endif /* __BENCH_USER_H__ */