mirror of
https://github.com/Unidata/netcdf-c.git
synced 2024-11-21 03:13:42 +08:00
961 lines
19 KiB
Plaintext
961 lines
19 KiB
Plaintext
dnl This is m4 source.
|
|
dnl Process using m4 to produce 'C' language file.
|
|
dnl
|
|
dnl If you see this line, you can ignore the next one.
|
|
/* Do not edit this file. It is produced from the corresponding .m4 source */
|
|
dnl
|
|
/*
|
|
* Copyright 2018, University Corporation for Atmospheric Research
|
|
* See netcdf/COPYRIGHT file for copying and redistribution conditions.
|
|
*/
|
|
/* "$Id: t_ncxx.m4,v 2.17 2005/05/20 14:48:14 russ Exp $" */
|
|
|
|
/*
|
|
* This program tests the aggregate external representation conversion
|
|
* functions "ncx_[pad_]{put,get}n_*()" declared in ncx.h.
|
|
* Unlike t_ncx.c, it only checks self consistency,
|
|
* not consistency with the xdr library.
|
|
*
|
|
* Link like this:
|
|
* cc t_ncxx.c ncx.o -o t_nxc
|
|
* (The xdr library is not needed.)
|
|
*
|
|
* If an assertion fails, there is a problem.
|
|
* Otherwise, the program is silent and has exit code 0.
|
|
*/
|
|
|
|
#if HAVE_CONFIG_H
|
|
#include <config.h>
|
|
#endif
|
|
|
|
#undef NDEBUG /* always active assert() in this file */
|
|
|
|
#include <stdio.h>
|
|
#include <string.h>
|
|
#include <limits.h>
|
|
/* alias poorly named limits.h macros */
|
|
#define SHORT_MAX SHRT_MAX
|
|
#define SHORT_MIN SHRT_MIN
|
|
#define USHORT_MAX USHRT_MAX
|
|
#include <assert.h>
|
|
#include "ncx.h"
|
|
#define X_SIZEOF_SCHAR X_SIZEOF_CHAR
|
|
#define X_LONG_MAX X_INT_MAX
|
|
#define X_LONG_MIN X_INT_MIN
|
|
|
|
#define XBSZ 1024
|
|
|
|
char ncxb[XBSZ];
|
|
char lbuf[XBSZ];
|
|
|
|
#define ArraySize(thang) (sizeof(thang)/sizeof(thang[0]))
|
|
#define eSizeOf(thang) ((size_t)(sizeof(thang[0])))
|
|
|
|
/*
|
|
* Some test data
|
|
* The ideas is that ncx_putn_type_type(...., types)
|
|
* should not return NC_ERANGE.
|
|
*/
|
|
|
|
#if SCHAR_MAX == X_SCHAR_MAX && SCHAR_MIN == X_SCHAR_MIN
|
|
static schar schars[] = {
|
|
SCHAR_MIN, SCHAR_MIN +1,
|
|
-1, 0, 1,
|
|
SCHAR_MAX - 1, SCHAR_MAX
|
|
};
|
|
#else
|
|
/* The implementation and this test assume 8 bit bytes. */
|
|
#error "Not 8 bit bytes ??"
|
|
#endif
|
|
|
|
static short shorts[] = {
|
|
#if SHORT_MAX <= X_SHORT_MAX
|
|
SHORT_MIN, SHORT_MIN + 1,
|
|
# if SCHAR_MAX < X_SHORT_MAX
|
|
SCHAR_MIN - 1, SCHAR_MIN, SCHAR_MIN + 1,
|
|
# endif
|
|
-1, 0, 1,
|
|
# if SCHAR_MAX < X_SHORT_MAX
|
|
SCHAR_MAX - 1, SCHAR_MAX, SCHAR_MAX + 1,
|
|
# endif
|
|
SHORT_MAX - 1, SHORT_MAX
|
|
#else
|
|
X_SHORT_MIN, X_SHORT_MIN + 1,
|
|
# if SCHAR_MAX < X_SHORT_MAX
|
|
SCHAR_MIN - 1, SCHAR_MIN, SCHAR_MIN + 1,
|
|
# endif
|
|
-1, 0, 1,
|
|
# if SCHAR_MAX < X_SHORT_MAX
|
|
SCHAR_MAX - 1, SCHAR_MAX, SCHAR_MAX + 1,
|
|
# endif
|
|
X_SHORT_MAX - 1, X_SHORT_MAX
|
|
#endif
|
|
};
|
|
|
|
static int ints[] = {
|
|
#if INT_MAX <= X_INT_MAX
|
|
INT_MIN, INT_MIN +1,
|
|
# if SHORT_MAX < X_INT_MAX
|
|
SHORT_MIN -1, SHORT_MIN, SHORT_MIN + 1,
|
|
# endif
|
|
# if SCHAR_MAX < X_INT_MAX
|
|
SCHAR_MIN - 1, SCHAR_MIN, SCHAR_MIN + 1,
|
|
# endif
|
|
-1, 0, 1,
|
|
# if SCHAR_MAX < X_INT_MAX
|
|
SCHAR_MAX - 1, SCHAR_MAX, SCHAR_MAX + 1,
|
|
# endif
|
|
# if SHORT_MAX < X_INT_MAX
|
|
SHORT_MAX - 1, SHORT_MAX, SHORT_MAX +1,
|
|
# endif
|
|
INT_MAX - 1, INT_MAX
|
|
#else
|
|
X_INT_MIN, X_INT_MIN +1,
|
|
# if SHORT_MAX < X_INT_MAX
|
|
SHORT_MIN -1, SHORT_MIN, SHORT_MIN + 1,
|
|
# endif
|
|
# if SCHAR_MAX < X_INT_MAX
|
|
SCHAR_MIN - 1, SCHAR_MIN, SCHAR_MIN + 1,
|
|
# endif
|
|
-1, 0, 1,
|
|
# if SCHAR_MAX < X_INT_MAX
|
|
SCHAR_MAX - 1, SCHAR_MAX, SCHAR_MAX + 1,
|
|
# endif
|
|
# if SHORT_MAX < X_INT_MAX
|
|
SHORT_MAX - 1, SHORT_MAX, SHORT_MAX +1,
|
|
# endif
|
|
X_INT_MAX - 1, X_INT_MAX
|
|
#endif /* INT */
|
|
};
|
|
|
|
|
|
/* N.B. only testing longs over X_INT range for now */
|
|
static long longs[] = {
|
|
#if LONG_MAX <= X_INT_MAX
|
|
LONG_MIN, LONG_MIN +1,
|
|
# if INT_MAX < X_INT_MAX
|
|
INT_MIN -1, INT_MIN, INT_MIN + 1,
|
|
# endif
|
|
# if SHORT_MAX < X_INT_MAX
|
|
SHORT_MIN -1, SHORT_MIN, SHORT_MIN + 1,
|
|
# endif
|
|
# if SCHAR_MAX < X_INT_MAX
|
|
SCHAR_MIN - 1, SCHAR_MIN, SCHAR_MIN + 1,
|
|
# endif
|
|
-1, 0, 1,
|
|
# if SCHAR_MAX < X_INT_MAX
|
|
SCHAR_MAX - 1, SCHAR_MAX, SCHAR_MAX + 1,
|
|
# endif
|
|
# if SHORT_MAX < X_INT_MAX
|
|
SHORT_MAX - 1, SHORT_MAX, SHORT_MAX +1,
|
|
# endif
|
|
# if INT_MAX < X_INT_MAX
|
|
INT_MAX -1, INT_MAX, INT_MAX + 1,
|
|
# endif
|
|
LONG_MAX - 1, LONG_MAX
|
|
#else
|
|
X_INT_MIN, X_INT_MIN +1,
|
|
# if SHORT_MAX < X_INT_MAX
|
|
SHORT_MIN -1, SHORT_MIN, SHORT_MIN + 1,
|
|
# endif
|
|
# if SCHAR_MAX < X_INT_MAX
|
|
SCHAR_MIN - 1, SCHAR_MIN, SCHAR_MIN + 1,
|
|
# endif
|
|
-1, 0, 1,
|
|
# if SCHAR_MAX < X_INT_MAX
|
|
SCHAR_MAX - 1, SCHAR_MAX, SCHAR_MAX + 1,
|
|
# endif
|
|
# if SHORT_MAX < X_INT_MAX
|
|
SHORT_MAX - 1, SHORT_MAX, SHORT_MAX +1,
|
|
# endif
|
|
X_INT_MAX - 1, X_INT_MAX
|
|
#endif
|
|
};
|
|
|
|
static float floats[] = {
|
|
-1.E9F,
|
|
-16777215, -16777214,
|
|
-999999,
|
|
-32769, -32768, -32767,
|
|
-129, -128, 127,
|
|
-1, 0, 1,
|
|
126, 127, 128,
|
|
32766, 32767, 32768,
|
|
999999,
|
|
16777214, 16777215, /* 2^24 -1 */
|
|
1.E9F
|
|
};
|
|
|
|
static double doubles[] = {
|
|
-1.E20,
|
|
-4503599627370495., -4503599627370494.,
|
|
-999999999999999.,
|
|
-1.E9,
|
|
-16777215, -16777214,
|
|
-999999,
|
|
-32769, -32768, -32767,
|
|
-129, -128, 127,
|
|
-1, 0, 1,
|
|
126, 127, 128,
|
|
32766, 32767, 32768,
|
|
999999,
|
|
16777214, 16777215, /* 2^24 -1 */
|
|
1.E9,
|
|
999999999999999.,
|
|
4503599627370494., 4503599627370495., /* 2^53 -1 */
|
|
1.E20
|
|
};
|
|
|
|
static uchar uchars[] = {
|
|
0, 1,
|
|
UCHAR_MAX/2 -1, UCHAR_MAX/2, UCHAR_MAX/2 +1,
|
|
UCHAR_MAX - 1, UCHAR_MAX
|
|
};
|
|
|
|
/* End of test data */
|
|
|
|
dnl dnl dnl
|
|
dnl
|
|
dnl Macros
|
|
dnl
|
|
dnl dnl dnl
|
|
dnl
|
|
dnl Upcase(str)
|
|
dnl
|
|
define(`Upcase',dnl
|
|
`dnl
|
|
translit($1, abcdefghijklmnopqrstuvwxyz, ABCDEFGHIJKLMNOPQRSTUVWXYZ)')dnl
|
|
dnl dnl dnl
|
|
dnl
|
|
dnl Xsizeof(Xtype)
|
|
dnl
|
|
define(`Xsizeof', ``X_SIZEOF_'Upcase($1)')dnl
|
|
dnl dnl dnl
|
|
define(`XMin', ``X_'Upcase($1)`_MIN'')dnl
|
|
define(`XMax', ``X_'Upcase($1)`_MAX'')dnl
|
|
dnl dnl dnl
|
|
dnl
|
|
dnl T_PUTN(XType, Type)
|
|
dnl
|
|
define(`T_PUTN',dnl
|
|
`dnl
|
|
static void
|
|
t_putn_$1_$2(char *const buf)
|
|
{
|
|
char *xp = buf;
|
|
const $2 *tp = `$2's;
|
|
size_t nelems = ArraySize(`$2's);
|
|
int status = ncx_putn_$1_$2((void **)&xp, nelems, tp);
|
|
assert(xp == buf + nelems * Xsizeof($1));
|
|
{
|
|
size_t ii;
|
|
for(ii = 0; ii < nelems; ii++)
|
|
{
|
|
if((double) tp[ii] > XMax($1))
|
|
{
|
|
assert(status == NC_ERANGE);
|
|
return;
|
|
}
|
|
if((double) tp[ii] < XMin($1))
|
|
{
|
|
assert(status == NC_ERANGE);
|
|
return;
|
|
}
|
|
}
|
|
assert(status == 0);
|
|
}
|
|
}
|
|
')dnl
|
|
dnl dnl dnl
|
|
dnl
|
|
dnl T_PUTN_U(XType, Type)
|
|
dnl Doesn't make signed comparisons to unsigned type
|
|
dnl
|
|
define(`T_PUTN_U',dnl
|
|
`dnl
|
|
static void
|
|
t_putn_$1_$2(char *const buf)
|
|
{
|
|
char *xp = buf;
|
|
const $2 *tp = `$2's;
|
|
size_t nelems = ArraySize(`$2's);
|
|
int status = ncx_putn_$1_$2((void **)&xp, nelems, tp);
|
|
assert(xp == buf + nelems * Xsizeof($1));
|
|
{
|
|
size_t ii;
|
|
for(ii = 0; ii < nelems; ii++)
|
|
{
|
|
if((double) tp[ii] > XMax($1))
|
|
{
|
|
assert(status == NC_ERANGE);
|
|
return;
|
|
}
|
|
}
|
|
assert(status == 0);
|
|
}
|
|
}
|
|
')dnl
|
|
dnl dnl dnl
|
|
dnl
|
|
dnl T_PAD_PUTN(XType, Type)
|
|
dnl
|
|
define(`T_PAD_PUTN',dnl
|
|
`dnl
|
|
static void
|
|
t_pad_putn_$1_$2(char *const buf)
|
|
{
|
|
char *xp = buf;
|
|
const $2 *tp = `$2's;
|
|
size_t nelems = ArraySize(`$2's);
|
|
const char *end = buf + nelems * Xsizeof($1);
|
|
int status = ncx_pad_putn_$1_$2((void **)&xp, nelems, tp);
|
|
assert(xp >= end);
|
|
assert((xp - end) < 4);
|
|
assert((xp - buf)%4 == 0);
|
|
{
|
|
size_t ii;
|
|
for(ii = 0; ii < nelems; ii++)
|
|
{
|
|
if((double) tp[ii] > XMax($1))
|
|
{
|
|
assert(status == NC_ERANGE);
|
|
return;
|
|
}
|
|
if((double) tp[ii] < XMin($1))
|
|
{
|
|
assert(status == NC_ERANGE);
|
|
return;
|
|
}
|
|
}
|
|
assert(status == 0);
|
|
}
|
|
}
|
|
')dnl
|
|
dnl dnl dnl
|
|
dnl
|
|
dnl T_GETN(XType, Type)
|
|
dnl
|
|
define(`T_GETN',dnl
|
|
`dnl
|
|
static void
|
|
t_getn_$1_$2(const char *const buf)
|
|
{
|
|
const char *xp = buf;
|
|
const $2 *tp = `$2's;
|
|
$2 *lp = ($2 *)lbuf;
|
|
size_t nelems = ArraySize(`$2's);
|
|
int status = ncx_getn_$1_$2((const void **)&xp, nelems, lp);
|
|
assert(xp == buf + nelems * Xsizeof($1));
|
|
assert(status == 0);
|
|
|
|
{
|
|
size_t ii;
|
|
for(ii = 0; ii < nelems; ii++)
|
|
{
|
|
if(((double)tp[ii] <= XMax($1))
|
|
&& ((double)tp[ii] >= XMin($1)))
|
|
{
|
|
assert(tp[ii] == lp[ii]);
|
|
}
|
|
}
|
|
}
|
|
}
|
|
')dnl
|
|
dnl dnl dnl
|
|
dnl
|
|
dnl T_GETN_U(XType, Type)
|
|
dnl Doesn't make signed comparisons to unsigned type
|
|
dnl
|
|
define(`T_GETN_U',dnl
|
|
`dnl
|
|
static void
|
|
t_getn_$1_$2(const char *const buf)
|
|
{
|
|
const char *xp = buf;
|
|
const $2 *tp = `$2's;
|
|
$2 *lp = ($2 *)lbuf;
|
|
size_t nelems = ArraySize(`$2's);
|
|
int status = ncx_getn_$1_$2((const void **)&xp, nelems, lp);
|
|
assert(xp == buf + nelems * Xsizeof($1));
|
|
assert(status == 0);
|
|
|
|
{
|
|
size_t ii;
|
|
for(ii = 0; ii < nelems; ii++)
|
|
{
|
|
if((double) tp[ii] <= XMax($1))
|
|
{
|
|
assert(tp[ii] == lp[ii]);
|
|
}
|
|
}
|
|
}
|
|
}
|
|
')dnl
|
|
dnl dnl dnl
|
|
dnl
|
|
dnl T_PAD_GETN(XType, Type)
|
|
dnl
|
|
define(`T_PAD_GETN',dnl
|
|
`dnl
|
|
static void
|
|
t_pad_getn_$1_$2(const char *const buf)
|
|
{
|
|
const char *xp = buf;
|
|
const $2 *tp = `$2's;
|
|
$2 *lp = ($2 *)lbuf;
|
|
size_t nelems = ArraySize(`$2's);
|
|
const char *end = buf + nelems * Xsizeof($1);
|
|
int status = ncx_pad_getn_$1_$2((const void **)&xp, nelems, lp);
|
|
assert(xp >= end);
|
|
assert((xp - end) < 4);
|
|
assert((xp - buf)%4 == 0);
|
|
assert(status == 0);
|
|
|
|
{
|
|
size_t ii;
|
|
for(ii = 0; ii < nelems; ii++)
|
|
{
|
|
if(((double) tp[ii] <= XMax($1))
|
|
&& ((double) tp[ii] >= XMin($1)))
|
|
{
|
|
assert(tp[ii] == lp[ii]);
|
|
}
|
|
}
|
|
}
|
|
}
|
|
')dnl
|
|
|
|
dnl dnl dnl
|
|
dnl
|
|
dnl Declare & define test routines
|
|
dnl
|
|
dnl dnl dnl
|
|
|
|
T_PUTN(schar, schar)
|
|
dnl T_PUTN(schar, uchar) replaced by special case code.
|
|
dnl - we don't return conversion errors putting uchar to schar.
|
|
static void
|
|
t_putn_schar_uchar(char *const buf)
|
|
{
|
|
char *xp = buf;
|
|
const uchar *tp = uchars;
|
|
size_t nelems = ArraySize(schars);
|
|
int status = ncx_putn_schar_uchar((void **)&xp, nelems, tp);
|
|
assert(xp == buf + nelems * X_SIZEOF_SCHAR);
|
|
assert(status == 0);
|
|
}
|
|
|
|
T_PUTN(schar, short)
|
|
T_PUTN(schar, int)
|
|
T_PUTN(schar, long)
|
|
T_PUTN(schar, float)
|
|
T_PUTN(schar, double)
|
|
|
|
T_PAD_PUTN(schar, schar)
|
|
dnl T_PAD_PUTN(schar, uchar) replaced by special case code.
|
|
dnl - we don't return conversion errors putting uchar to schar.
|
|
static void
|
|
t_pad_putn_schar_uchar(char *const buf)
|
|
{
|
|
char *xp = buf;
|
|
const uchar *tp = uchars;
|
|
size_t nelems = ArraySize(uchars);
|
|
const char *end = buf + nelems * X_SIZEOF_SCHAR;
|
|
int status = ncx_pad_putn_schar_uchar((void **)&xp, nelems, tp);
|
|
assert(xp >= end);
|
|
assert((xp - end) < 4);
|
|
assert((xp - buf)%4 == 0);
|
|
assert(status == 0);
|
|
}
|
|
|
|
T_PAD_PUTN(schar, short)
|
|
T_PAD_PUTN(schar, int)
|
|
T_PAD_PUTN(schar, long)
|
|
T_PAD_PUTN(schar, float)
|
|
T_PAD_PUTN(schar, double)
|
|
|
|
T_PUTN(short, schar)
|
|
T_PUTN_U(short, uchar)
|
|
T_PUTN(short, short)
|
|
T_PUTN(short, int)
|
|
T_PUTN(short, long)
|
|
T_PUTN(short, float)
|
|
T_PUTN(short, double)
|
|
|
|
T_PAD_PUTN(short, schar)
|
|
dnl T_PAD_PUTN(short, uchar)
|
|
dnl Don't make signed comparisons to usigned type
|
|
static void
|
|
t_pad_putn_short_uchar(char *const buf)
|
|
{
|
|
char *xp = buf;
|
|
const uchar *tp = uchars;
|
|
size_t nelems = ArraySize(uchars);
|
|
const char *end = buf + nelems * X_SIZEOF_SHORT;
|
|
int status = ncx_pad_putn_short_uchar((void **)&xp, nelems, tp);
|
|
assert(xp >= end);
|
|
assert((xp - end) < 4);
|
|
assert((xp - buf)%4 == 0);
|
|
{
|
|
size_t ii;
|
|
for(ii = 0; ii < nelems; ii++)
|
|
{
|
|
if(tp[ii] > X_SHORT_MAX)
|
|
{
|
|
assert(status == NC_ERANGE);
|
|
return;
|
|
}
|
|
}
|
|
assert(status == 0);
|
|
}
|
|
}
|
|
|
|
T_PAD_PUTN(short, short)
|
|
T_PAD_PUTN(short, int)
|
|
T_PAD_PUTN(short, long)
|
|
T_PAD_PUTN(short, float)
|
|
T_PAD_PUTN(short, double)
|
|
|
|
T_PUTN(int, schar)
|
|
T_PUTN_U(int, uchar)
|
|
T_PUTN(int, short)
|
|
T_PUTN(int, int)
|
|
T_PUTN(int, long)
|
|
T_PUTN(int, float)
|
|
T_PUTN(int, double)
|
|
|
|
T_PUTN(float, schar)
|
|
T_PUTN_U(float, uchar)
|
|
T_PUTN(float, short)
|
|
T_PUTN(float, int)
|
|
T_PUTN(float, long)
|
|
T_PUTN(float, float)
|
|
T_PUTN(float, double)
|
|
|
|
T_PUTN(double, schar)
|
|
T_PUTN_U(double, uchar)
|
|
T_PUTN(double, short)
|
|
T_PUTN(double, int)
|
|
T_PUTN(double, long)
|
|
T_PUTN(double, float)
|
|
T_PUTN(double, double)
|
|
|
|
|
|
|
|
T_GETN(schar, schar)
|
|
dnl T_GETN(schar, uchar)
|
|
dnl - we don't return conversion errors gettin schar to uchar.
|
|
static void
|
|
t_getn_schar_uchar(const char *const buf)
|
|
{
|
|
const char *xp = buf;
|
|
const uchar *tp = uchars;
|
|
uchar *lp = (uchar *)lbuf;
|
|
size_t nelems = ArraySize(schars);
|
|
int status = ncx_getn_schar_uchar((const void **)&xp, nelems, lp);
|
|
assert(xp == buf + nelems * X_SIZEOF_SCHAR);
|
|
assert(status == 0);
|
|
|
|
{
|
|
size_t ii;
|
|
for(ii = 0; ii < nelems; ii++)
|
|
{
|
|
assert(tp[ii] == lp[ii]);
|
|
}
|
|
}
|
|
}
|
|
|
|
T_GETN(schar, short)
|
|
T_GETN(schar, int)
|
|
T_GETN(schar, long)
|
|
T_GETN(schar, float)
|
|
T_GETN(schar, double)
|
|
|
|
T_PAD_GETN(schar, schar)
|
|
dnl T_PAD_GETN(schar, uchar)
|
|
dnl - we don't return conversion errors gettin schar to uchar.
|
|
static void
|
|
t_pad_getn_schar_uchar(const char *const buf)
|
|
{
|
|
const char *xp = buf;
|
|
const uchar *tp = uchars;
|
|
uchar *lp = (uchar *)lbuf;
|
|
size_t nelems = ArraySize(schars);
|
|
const char *end = buf + nelems * X_SIZEOF_SCHAR;
|
|
int status = ncx_pad_getn_schar_uchar((const void **)&xp, nelems, lp);
|
|
assert(xp >= end);
|
|
assert((xp - end) < 4);
|
|
assert((xp - buf)%4 == 0);
|
|
assert(status == 0);
|
|
|
|
{
|
|
size_t ii;
|
|
for(ii = 0; ii < nelems; ii++)
|
|
{
|
|
assert(tp[ii] == lp[ii]);
|
|
}
|
|
}
|
|
}
|
|
|
|
T_PAD_GETN(schar, short)
|
|
T_PAD_GETN(schar, int)
|
|
T_PAD_GETN(schar, long)
|
|
T_PAD_GETN(schar, float)
|
|
T_PAD_GETN(schar, double)
|
|
|
|
T_GETN(short, schar)
|
|
T_GETN_U(short, uchar)
|
|
T_GETN(short, short)
|
|
T_GETN(short, int)
|
|
T_GETN(short, long)
|
|
T_GETN(short, float)
|
|
T_GETN(short, double)
|
|
|
|
T_PAD_GETN(short, schar)
|
|
dnl T_PAD_GETN(short, uchar)
|
|
dnl Don't make signed comparisons to usigned type
|
|
static void
|
|
t_pad_getn_short_uchar(const char *const buf)
|
|
{
|
|
const char *xp = buf;
|
|
const uchar *tp = uchars;
|
|
uchar *lp = (uchar *)lbuf;
|
|
size_t nelems = ArraySize(uchars);
|
|
const char *end = buf + nelems * X_SIZEOF_SHORT;
|
|
int status = ncx_pad_getn_short_uchar((const void **)&xp, nelems, lp);
|
|
assert(xp >= end);
|
|
assert((xp - end) < 4);
|
|
assert((xp - buf)%4 == 0);
|
|
assert(status == 0);
|
|
|
|
{
|
|
size_t ii;
|
|
for(ii = 0; ii < nelems; ii++)
|
|
{
|
|
if((tp[ii] <= X_SHORT_MAX))
|
|
{
|
|
assert(tp[ii] == lp[ii]);
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
T_PAD_GETN(short, short)
|
|
T_PAD_GETN(short, int)
|
|
T_PAD_GETN(short, long)
|
|
T_PAD_GETN(short, float)
|
|
T_PAD_GETN(short, double)
|
|
|
|
T_GETN(int, schar)
|
|
T_GETN_U(int, uchar)
|
|
T_GETN(int, short)
|
|
T_GETN(int, int)
|
|
T_GETN(int, long)
|
|
T_GETN(int, float)
|
|
T_GETN(int, double)
|
|
|
|
T_GETN(float, schar)
|
|
T_GETN_U(float, uchar)
|
|
T_GETN(float, short)
|
|
dnl T_GETN(float, int)
|
|
dnl Exact conversion of int to x_float is limited by external float mantissa
|
|
static void
|
|
t_getn_float_int(const char *const buf)
|
|
{
|
|
const char *xp = buf;
|
|
const int *tp = ints;
|
|
int *lp = (int *)lbuf;
|
|
size_t nelems = ArraySize(ints);
|
|
int status = ncx_getn_float_int((const void **)&xp, nelems, lp);
|
|
assert(xp == buf + nelems * X_SIZEOF_FLOAT);
|
|
/* If the system rounds up can get NC_ERANGE */
|
|
assert(status == 0 || status == NC_ERANGE);
|
|
|
|
{
|
|
size_t ii;
|
|
for(ii = 0; ii < nelems; ii++)
|
|
{
|
|
/* limited by x_float mantissa nbits */
|
|
if((tp[ii] <= 16777215)
|
|
&& (tp[ii] >= -16777215))
|
|
{
|
|
assert(tp[ii] == lp[ii]);
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
dnl T_GETN(float, long)
|
|
dnl Exact conversion of long to x_float is limited by external float mantissa
|
|
static void
|
|
t_getn_float_long(const char *const buf)
|
|
{
|
|
const char *xp = buf;
|
|
const long *tp = longs;
|
|
long *lp = (long *)lbuf;
|
|
size_t nelems = ArraySize(longs);
|
|
int status = ncx_getn_float_long((const void **)&xp, nelems, lp);
|
|
assert(xp == buf + nelems * X_SIZEOF_FLOAT);
|
|
/* If the system rounds up can get NC_ERANGE */
|
|
assert(status == 0 || status == NC_ERANGE);
|
|
|
|
{
|
|
size_t ii;
|
|
for(ii = 0; ii < nelems; ii++)
|
|
{
|
|
/* limited by x_float mantissa nbits */
|
|
if((tp[ii] <= 16777215)
|
|
&& (tp[ii] >= 16777215))
|
|
{
|
|
if(tp[ii] != lp[ii])
|
|
(void) fprintf(stderr,
|
|
"%.9e != %.9e float_float (diff %.9e)\n",
|
|
(double)tp[ii], (double)lp[ii],
|
|
(double)(tp[ii] - lp[ii]));
|
|
}
|
|
}
|
|
}
|
|
}
|
|
T_GETN(float, float)
|
|
dnl T_GETN(float, double)
|
|
dnl Exact conversion of double to x_float is limited by external float mantissa
|
|
static void
|
|
t_getn_float_double(const char *const buf)
|
|
{
|
|
const char *xp = buf;
|
|
const double *tp = doubles;
|
|
double *lp = (double *)lbuf;
|
|
size_t nelems = ArraySize(doubles);
|
|
int status = ncx_getn_float_double((const void **)&xp, nelems, lp);
|
|
assert(xp == buf + nelems * X_SIZEOF_FLOAT);
|
|
assert(status == 0);
|
|
|
|
{
|
|
size_t ii;
|
|
for(ii = 0; ii < nelems; ii++)
|
|
{
|
|
if((tp[ii] <= X_FLOAT_MAX)
|
|
&& (tp[ii] >= X_FLOAT_MIN))
|
|
{
|
|
if(((float)tp[ii]) != lp[ii])
|
|
{
|
|
if(tp[ii] != 0)
|
|
{
|
|
double eps = (tp[ii] - lp[ii])/tp[ii];
|
|
if(eps > 1.19209290E-07F) /* X_FLT_EPSILON */
|
|
{
|
|
(void) fprintf(stderr,
|
|
"%.9e != %.9e float_double (eps %.9e)\n",
|
|
tp[ii], lp[ii], eps);
|
|
}
|
|
}
|
|
else
|
|
{
|
|
(void) fprintf(stderr,
|
|
"%.9e != %.9e float_double (diff %.9e)\n",
|
|
tp[ii], lp[ii], tp[ii] - lp[ii]);
|
|
|
|
}
|
|
}
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
|
|
T_GETN(double, schar)
|
|
T_GETN_U(double, uchar)
|
|
T_GETN(double, short)
|
|
T_GETN(double, int)
|
|
T_GETN(double, long)
|
|
T_GETN(double, float)
|
|
T_GETN(double, double)
|
|
|
|
|
|
#if defined(_CRAYIEEE) && !defined(_CRAYMPP) /* T90 */
|
|
#include <signal.h>
|
|
#endif /* T90 */
|
|
|
|
int
|
|
main(int ac, char *av[])
|
|
{
|
|
|
|
#if defined(_CRAYIEEE) && !defined(_CRAYMPP) /* T90 */
|
|
/*
|
|
* Some of the extreme test assignments in this program trigger
|
|
* floating point exceptions on CRAY T90
|
|
*/
|
|
(void) signal(SIGFPE, SIG_IGN);
|
|
#endif /* T90 */
|
|
|
|
/* x_schar */
|
|
t_putn_schar_schar(ncxb);
|
|
t_getn_schar_schar(ncxb);
|
|
|
|
t_putn_schar_uchar(ncxb);
|
|
t_getn_schar_uchar(ncxb);
|
|
|
|
t_putn_schar_short(ncxb);
|
|
t_getn_schar_short(ncxb);
|
|
|
|
t_putn_schar_int(ncxb);
|
|
t_getn_schar_int(ncxb);
|
|
|
|
t_putn_schar_long(ncxb);
|
|
t_getn_schar_long(ncxb);
|
|
|
|
t_putn_schar_float(ncxb);
|
|
t_getn_schar_float(ncxb);
|
|
|
|
t_putn_schar_double(ncxb);
|
|
t_getn_schar_double(ncxb);
|
|
|
|
/* pad x_schar */
|
|
t_pad_putn_schar_schar(ncxb);
|
|
t_getn_schar_schar(ncxb);
|
|
t_pad_getn_schar_schar(ncxb);
|
|
|
|
t_pad_putn_schar_uchar(ncxb);
|
|
t_getn_schar_uchar(ncxb);
|
|
t_pad_getn_schar_uchar(ncxb);
|
|
|
|
t_pad_putn_schar_short(ncxb);
|
|
t_getn_schar_short(ncxb);
|
|
t_pad_getn_schar_short(ncxb);
|
|
|
|
t_pad_putn_schar_int(ncxb);
|
|
t_getn_schar_int(ncxb);
|
|
t_pad_getn_schar_int(ncxb);
|
|
|
|
t_pad_putn_schar_long(ncxb);
|
|
t_getn_schar_long(ncxb);
|
|
t_pad_getn_schar_long(ncxb);
|
|
|
|
t_pad_putn_schar_float(ncxb);
|
|
t_getn_schar_float(ncxb);
|
|
t_pad_getn_schar_float(ncxb);
|
|
|
|
t_pad_putn_schar_double(ncxb);
|
|
t_getn_schar_double(ncxb);
|
|
t_pad_getn_schar_double(ncxb);
|
|
|
|
/* x_short */
|
|
t_putn_short_schar(ncxb);
|
|
t_getn_short_schar(ncxb);
|
|
|
|
t_putn_short_uchar(ncxb);
|
|
t_getn_short_uchar(ncxb);
|
|
|
|
t_putn_short_short(ncxb);
|
|
t_getn_short_short(ncxb);
|
|
|
|
t_putn_short_int(ncxb);
|
|
t_getn_short_int(ncxb);
|
|
|
|
t_putn_short_long(ncxb);
|
|
t_getn_short_long(ncxb);
|
|
|
|
t_putn_short_float(ncxb);
|
|
t_getn_short_float(ncxb);
|
|
|
|
t_putn_short_double(ncxb);
|
|
t_getn_short_double(ncxb);
|
|
|
|
/* pad x_short */
|
|
t_pad_putn_short_schar(ncxb);
|
|
t_getn_short_schar(ncxb);
|
|
t_pad_getn_short_schar(ncxb);
|
|
|
|
t_pad_putn_short_uchar(ncxb);
|
|
t_getn_short_uchar(ncxb);
|
|
t_pad_getn_short_uchar(ncxb);
|
|
|
|
t_pad_putn_short_short(ncxb);
|
|
t_getn_short_short(ncxb);
|
|
t_pad_getn_short_short(ncxb);
|
|
|
|
t_pad_putn_short_int(ncxb);
|
|
t_getn_short_int(ncxb);
|
|
t_pad_getn_short_int(ncxb);
|
|
|
|
t_pad_putn_short_long(ncxb);
|
|
t_getn_short_long(ncxb);
|
|
t_pad_getn_short_long(ncxb);
|
|
|
|
t_pad_putn_short_float(ncxb);
|
|
t_getn_short_float(ncxb);
|
|
t_pad_getn_short_float(ncxb);
|
|
|
|
t_pad_putn_short_double(ncxb);
|
|
t_getn_short_double(ncxb);
|
|
t_pad_getn_short_double(ncxb);
|
|
|
|
/* x_int */
|
|
t_putn_int_schar(ncxb);
|
|
t_getn_int_schar(ncxb);
|
|
|
|
t_putn_int_uchar(ncxb);
|
|
t_getn_int_uchar(ncxb);
|
|
|
|
t_putn_int_short(ncxb);
|
|
t_getn_int_short(ncxb);
|
|
|
|
t_putn_int_int(ncxb);
|
|
t_getn_int_int(ncxb);
|
|
|
|
t_putn_int_long(ncxb);
|
|
t_getn_int_long(ncxb);
|
|
|
|
t_putn_int_float(ncxb);
|
|
t_getn_int_float(ncxb);
|
|
|
|
t_putn_int_double(ncxb);
|
|
t_getn_int_double(ncxb);
|
|
|
|
/* x_float */
|
|
t_putn_float_schar(ncxb);
|
|
t_getn_float_schar(ncxb);
|
|
|
|
t_putn_float_uchar(ncxb);
|
|
t_getn_float_uchar(ncxb);
|
|
|
|
t_putn_float_short(ncxb);
|
|
t_getn_float_short(ncxb);
|
|
|
|
t_putn_float_int(ncxb);
|
|
t_getn_float_int(ncxb);
|
|
|
|
t_putn_float_long(ncxb);
|
|
t_getn_float_long(ncxb);
|
|
|
|
t_putn_float_float(ncxb);
|
|
t_getn_float_float(ncxb);
|
|
|
|
t_putn_float_double(ncxb);
|
|
t_getn_float_double(ncxb);
|
|
|
|
/* x_double */
|
|
t_putn_double_schar(ncxb);
|
|
t_getn_double_schar(ncxb);
|
|
|
|
t_putn_double_uchar(ncxb);
|
|
t_getn_double_uchar(ncxb);
|
|
|
|
t_putn_double_short(ncxb);
|
|
t_getn_double_short(ncxb);
|
|
|
|
t_putn_double_int(ncxb);
|
|
t_getn_double_int(ncxb);
|
|
|
|
t_putn_double_long(ncxb);
|
|
t_getn_double_long(ncxb);
|
|
|
|
t_putn_double_float(ncxb);
|
|
t_getn_double_float(ncxb);
|
|
|
|
t_putn_double_double(ncxb);
|
|
t_getn_double_double(ncxb);
|
|
|
|
return 0;
|
|
}
|