floraachy
/
OpenBLAS
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Fix complex support for MSVC headers

This commit is contained in:
Isuru Fernando 2017-07-28 11:50:29 +05:30
parent dc24914415
commit ca17b4b75c
5 changed files with 40 additions and 42 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

50
common.h
View File

@ -495,6 +495,31 @@ static void __inline blas_lock(volatile BLASULONG *address){
#define MMAP_POLICY (MAP_PRIVATE | MAP_ANONYMOUS) #define MMAP_POLICY (MAP_PRIVATE | MAP_ANONYMOUS)
#endif #endif
/* C99 supports complex floating numbers natively, which GCC also offers as an
extension since version 3.0. If neither are available, use a compatible
structure as fallback (see Clause 6.2.5.13 of the C99 standard). */
#if ((defined(__STDC_IEC_559_COMPLEX__) || __STDC_VERSION__ >= 199901L || \
(__GNUC__ >= 3 && !defined(__cplusplus))) && !(defined(FORCE_OPENBLAS_COMPLEX_STRUCT))) && !defined(_MSC_VER)
#define OPENBLAS_COMPLEX_C99
#ifndef __cplusplus
#include <complex.h>
#endif
typedef float _Complex openblas_complex_float;
typedef double _Complex openblas_complex_double;
typedef xdouble _Complex openblas_complex_xdouble;
#define openblas_make_complex_float(real, imag) ((real) + ((imag) * _Complex_I))
#define openblas_make_complex_double(real, imag) ((real) + ((imag) * _Complex_I))
#define openblas_make_complex_xdouble(real, imag) ((real) + ((imag) * _Complex_I))
#else
#define OPENBLAS_COMPLEX_STRUCT
typedef struct { float real, imag; } openblas_complex_float;
typedef struct { double real, imag; } openblas_complex_double;
typedef struct { xdouble real, imag; } openblas_complex_xdouble;
#define openblas_make_complex_float(real, imag) {(real), (imag)}
#define openblas_make_complex_double(real, imag) {(real), (imag)}
#define openblas_make_complex_xdouble(real, imag) {(real), (imag)}
#endif
#include "param.h" #include "param.h"
#include "common_param.h" #include "common_param.h"
@ -524,31 +549,6 @@ static void __inline blas_lock(volatile BLASULONG *address){
#include <stdio.h> #include <stdio.h>
#endif // NOINCLUDE #endif // NOINCLUDE
/* C99 supports complex floating numbers natively, which GCC also offers as an
extension since version 3.0. If neither are available, use a compatible
structure as fallback (see Clause 6.2.5.13 of the C99 standard). */
#if ((defined(__STDC_IEC_559_COMPLEX__) || __STDC_VERSION__ >= 199901L || \
(__GNUC__ >= 3 && !defined(__cplusplus))) && !(defined(FORCE_OPENBLAS_COMPLEX_STRUCT)))
#define OPENBLAS_COMPLEX_C99
#ifndef __cplusplus
#include <complex.h>
#endif
typedef float _Complex openblas_complex_float;
typedef double _Complex openblas_complex_double;
typedef xdouble _Complex openblas_complex_xdouble;
#define openblas_make_complex_float(real, imag) ((real) + ((imag) * _Complex_I))
#define openblas_make_complex_double(real, imag) ((real) + ((imag) * _Complex_I))
#define openblas_make_complex_xdouble(real, imag) ((real) + ((imag) * _Complex_I))
#else
#define OPENBLAS_COMPLEX_STRUCT
typedef struct { float real, imag; } openblas_complex_float;
typedef struct { double real, imag; } openblas_complex_double;
typedef struct { xdouble real, imag; } openblas_complex_xdouble;
#define openblas_make_complex_float(real, imag) {(real), (imag)}
#define openblas_make_complex_double(real, imag) {(real), (imag)}
#define openblas_make_complex_xdouble(real, imag) {(real), (imag)}
#endif
#ifdef XDOUBLE #ifdef XDOUBLE
#define OPENBLAS_COMPLEX_FLOAT openblas_complex_xdouble #define OPENBLAS_COMPLEX_FLOAT openblas_complex_xdouble
#define OPENBLAS_MAKE_COMPLEX_FLOAT(r,i) openblas_make_complex_xdouble(r,i) #define OPENBLAS_MAKE_COMPLEX_FLOAT(r,i) openblas_make_complex_xdouble(r,i)

12
common_param.h
View File

@ -333,8 +333,8 @@ BLASLONG (*icamin_k)(BLASLONG, float *, BLASLONG);
float (*cnrm2_k) (BLASLONG, float *, BLASLONG); float (*cnrm2_k) (BLASLONG, float *, BLASLONG);
float (*casum_k) (BLASLONG, float *, BLASLONG); float (*casum_k) (BLASLONG, float *, BLASLONG);
int (*ccopy_k) (BLASLONG, float *, BLASLONG, float *, BLASLONG); int (*ccopy_k) (BLASLONG, float *, BLASLONG, float *, BLASLONG);
float _Complex (*cdotu_k) (BLASLONG, float *, BLASLONG, float *, BLASLONG); openblas_complex_float (*cdotu_k) (BLASLONG, float *, BLASLONG, float *, BLASLONG);
float _Complex (*cdotc_k) (BLASLONG, float *, BLASLONG, float *, BLASLONG); openblas_complex_float (*cdotc_k) (BLASLONG, float *, BLASLONG, float *, BLASLONG);
int (*csrot_k) (BLASLONG, float *, BLASLONG, float *, BLASLONG, float, float); int (*csrot_k) (BLASLONG, float *, BLASLONG, float *, BLASLONG, float, float);
int (*caxpy_k) (BLASLONG, BLASLONG, BLASLONG, float, float, float *, BLASLONG, float *, BLASLONG, float *, BLASLONG); int (*caxpy_k) (BLASLONG, BLASLONG, BLASLONG, float, float, float *, BLASLONG, float *, BLASLONG, float *, BLASLONG);
@ -496,8 +496,8 @@ BLASLONG (*izamin_k)(BLASLONG, double *, BLASLONG);
double (*znrm2_k) (BLASLONG, double *, BLASLONG); double (*znrm2_k) (BLASLONG, double *, BLASLONG);
double (*zasum_k) (BLASLONG, double *, BLASLONG); double (*zasum_k) (BLASLONG, double *, BLASLONG);
int (*zcopy_k) (BLASLONG, double *, BLASLONG, double *, BLASLONG); int (*zcopy_k) (BLASLONG, double *, BLASLONG, double *, BLASLONG);
double _Complex (*zdotu_k) (BLASLONG, double *, BLASLONG, double *, BLASLONG); openblas_complex_double (*zdotu_k) (BLASLONG, double *, BLASLONG, double *, BLASLONG);
double _Complex (*zdotc_k) (BLASLONG, double *, BLASLONG, double *, BLASLONG); openblas_complex_double (*zdotc_k) (BLASLONG, double *, BLASLONG, double *, BLASLONG);
int (*zdrot_k) (BLASLONG, double *, BLASLONG, double *, BLASLONG, double, double); int (*zdrot_k) (BLASLONG, double *, BLASLONG, double *, BLASLONG, double, double);
int (*zaxpy_k) (BLASLONG, BLASLONG, BLASLONG, double, double, double *, BLASLONG, double *, BLASLONG, double *, BLASLONG); int (*zaxpy_k) (BLASLONG, BLASLONG, BLASLONG, double, double, double *, BLASLONG, double *, BLASLONG, double *, BLASLONG);
@ -661,8 +661,8 @@ BLASLONG (*ixamin_k)(BLASLONG, xdouble *, BLASLONG);
xdouble (*xnrm2_k) (BLASLONG, xdouble *, BLASLONG); xdouble (*xnrm2_k) (BLASLONG, xdouble *, BLASLONG);
xdouble (*xasum_k) (BLASLONG, xdouble *, BLASLONG); xdouble (*xasum_k) (BLASLONG, xdouble *, BLASLONG);
int (*xcopy_k) (BLASLONG, xdouble *, BLASLONG, xdouble *, BLASLONG); int (*xcopy_k) (BLASLONG, xdouble *, BLASLONG, xdouble *, BLASLONG);
xdouble _Complex (*xdotu_k) (BLASLONG, xdouble *, BLASLONG, xdouble *, BLASLONG); openblas_complex_xdouble (*xdotu_k) (BLASLONG, xdouble *, BLASLONG, xdouble *, BLASLONG);
xdouble _Complex (*xdotc_k) (BLASLONG, xdouble *, BLASLONG, xdouble *, BLASLONG); openblas_complex_xdouble (*xdotc_k) (BLASLONG, xdouble *, BLASLONG, xdouble *, BLASLONG);
int (*xqrot_k) (BLASLONG, xdouble *, BLASLONG, xdouble *, BLASLONG, xdouble, xdouble); int (*xqrot_k) (BLASLONG, xdouble *, BLASLONG, xdouble *, BLASLONG, xdouble, xdouble);
int (*xaxpy_k) (BLASLONG, BLASLONG, BLASLONG, xdouble, xdouble, xdouble *, BLASLONG, xdouble *, BLASLONG, xdouble *, BLASLONG); int (*xaxpy_k) (BLASLONG, BLASLONG, BLASLONG, xdouble, xdouble, xdouble *, BLASLONG, xdouble *, BLASLONG, xdouble *, BLASLONG);

9
kernel/x86_64/cdot.c
View File

@ -91,16 +91,15 @@ static void cdot_kernel_16(BLASLONG n, FLOAT *x, FLOAT *y, FLOAT *d)
#endif #endif
FLOAT _Complex CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x, FLOAT *y, BLASLONG inc_y) OPENBLAS_COMPLEX_FLOAT CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x, FLOAT *y, BLASLONG inc_y)
{ {
BLASLONG i; BLASLONG i;
BLASLONG ix,iy; BLASLONG ix,iy;
FLOAT _Complex result;
FLOAT dot[8] = { 0.0, 0.0, 0.0 , 0.0, 0.0, 0.0, 0.0, 0.0 } ; FLOAT dot[8] = { 0.0, 0.0, 0.0 , 0.0, 0.0, 0.0, 0.0, 0.0 } ;
if ( n <= 0 ) if ( n <= 0 )
{ {
result = OPENBLAS_MAKE_COMPLEX_FLOAT (0.0, 0.0) ; OPENBLAS_COMPLEX_FLOAT result = OPENBLAS_MAKE_COMPLEX_FLOAT (0.0, 0.0) ;
return(result); return(result);
} }
@ -160,11 +159,11 @@ FLOAT _Complex CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x, FLOAT *y, BLASLONG in
} }
#if !defined(CONJ) #if !defined(CONJ)
result = OPENBLAS_MAKE_COMPLEX_FLOAT (dot[0]-dot[1], dot[4]+dot[5]) ; OPENBLAS_COMPLEX_FLOAT result = OPENBLAS_MAKE_COMPLEX_FLOAT (dot[0]-dot[1], dot[4]+dot[5]) ;
// CREAL(result) = dot[0] - dot[1]; // CREAL(result) = dot[0] - dot[1];
// CIMAG(result) = dot[4] + dot[5]; // CIMAG(result) = dot[4] + dot[5];
#else #else
result = OPENBLAS_MAKE_COMPLEX_FLOAT (dot[0]+dot[1], dot[4]-dot[5]) ; OPENBLAS_COMPLEX_FLOAT result = OPENBLAS_MAKE_COMPLEX_FLOAT (dot[0]+dot[1], dot[4]-dot[5]) ;
// CREAL(result) = dot[0] + dot[1]; // CREAL(result) = dot[0] + dot[1];
// CIMAG(result) = dot[4] - dot[5]; // CIMAG(result) = dot[4] - dot[5];

9
kernel/x86_64/zdot.c
View File

@ -86,18 +86,17 @@ static void zdot_kernel_8(BLASLONG n, FLOAT *x, FLOAT *y, FLOAT *d)
#endif #endif
FLOAT _Complex CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x, FLOAT *y, BLASLONG inc_y) OPENBLAS_COMPLEX_FLOAT CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x, FLOAT *y, BLASLONG inc_y)
{ {
BLASLONG i; BLASLONG i;
BLASLONG ix,iy; BLASLONG ix,iy;
FLOAT _Complex result;
FLOAT dot[4] = { 0.0, 0.0, 0.0 , 0.0 } ; FLOAT dot[4] = { 0.0, 0.0, 0.0 , 0.0 } ;
if ( n <= 0 ) if ( n <= 0 )
{ {
// CREAL(result) = 0.0 ; // CREAL(result) = 0.0 ;
// CIMAG(result) = 0.0 ; // CIMAG(result) = 0.0 ;
result=OPENBLAS_MAKE_COMPLEX_FLOAT(0.0,0.0); OPENBLAS_COMPLEX_FLOAT result=OPENBLAS_MAKE_COMPLEX_FLOAT(0.0,0.0);
return(result); return(result);
} }
@ -151,11 +150,11 @@ FLOAT _Complex CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x, FLOAT *y, BLASLONG in
} }
#if !defined(CONJ) #if !defined(CONJ)
result=OPENBLAS_MAKE_COMPLEX_FLOAT(dot[0]-dot[1],dot[2]+dot[3]); OPENBLAS_COMPLEX_FLOAT result=OPENBLAS_MAKE_COMPLEX_FLOAT(dot[0]-dot[1],dot[2]+dot[3]);
// CREAL(result) = dot[0] - dot[1]; // CREAL(result) = dot[0] - dot[1];
// CIMAG(result) = dot[2] + dot[3]; // CIMAG(result) = dot[2] + dot[3];
#else #else
result=OPENBLAS_MAKE_COMPLEX_FLOAT(dot[0]+dot[1],dot[2]-dot[3]); OPENBLAS_COMPLEX_FLOAT result=OPENBLAS_MAKE_COMPLEX_FLOAT(dot[0]+dot[1],dot[2]-dot[3]);
// CREAL(result) = dot[0] + dot[1]; // CREAL(result) = dot[0] + dot[1];
// CIMAG(result) = dot[2] - dot[3]; // CIMAG(result) = dot[2] - dot[3];

2
openblas_config_template.h
View File

@ -59,7 +59,7 @@ typedef int blasint;
extension since version 3.0. If neither are available, use a compatible extension since version 3.0. If neither are available, use a compatible
structure as fallback (see Clause 6.2.5.13 of the C99 standard). */ structure as fallback (see Clause 6.2.5.13 of the C99 standard). */
#if ((defined(__STDC_IEC_559_COMPLEX__) || __STDC_VERSION__ >= 199901L || \ #if ((defined(__STDC_IEC_559_COMPLEX__) || __STDC_VERSION__ >= 199901L || \
(__GNUC__ >= 3 && !defined(__cplusplus))) && !(defined(FORCE_OPENBLAS_COMPLEX_STRUCT))) (__GNUC__ >= 3 && !defined(__cplusplus))) && !(defined(FORCE_OPENBLAS_COMPLEX_STRUCT))) && !defined(_MSC_VER)
#define OPENBLAS_COMPLEX_C99 #define OPENBLAS_COMPLEX_C99
#ifndef __cplusplus #ifndef __cplusplus
#include <complex.h> #include <complex.h>