tests/lib/libm/t_bit: Expand and tidy.

1. Instead of calling out VAX by name, use #ifdef NAN.

2. Verify signbit works on long double on all architectures, not just
   those with __HAVE_LONG_DOUBLE which means long double is _larger_
   than double.

3. Minor formatting tidying.
This commit is contained in:
riastradh 2024-05-06 18:41:23 +00:00
parent 9da594439c
commit 557aa1ae1a
1 changed files with 56 additions and 47 deletions

View File

@ -1,4 +1,4 @@
/* $NetBSD: t_bit.c,v 1.1 2019/04/26 08:52:16 maya Exp $ */
/* $NetBSD: t_bit.c,v 1.2 2024/05/06 18:41:23 riastradh Exp $ */
/*
* Written by Maya Rashish <maya@NetBSD.org>
@ -7,6 +7,9 @@
* Testing signbit{,f,l} function correctly
*/
#include <sys/cdefs.h>
__RCSID("$NetBSD: t_bit.c,v 1.2 2024/05/06 18:41:23 riastradh Exp $");
#include <atf-c.h>
#include <float.h>
#include <math.h>
@ -18,79 +21,85 @@ static const struct {
double input;
bool is_negative;
} values[] = {
{ -1, true},
{ -123, true},
{ -123E6, true},
#ifdef INFINITY
{ -INFINITY, true},
{ INFINITY, false},
#endif
{ 123E6, false},
{ 0, false},
{ -FLT_MIN, true},
{ FLT_MIN, false},
/*
{ -1, true },
{ -123, true },
{ -123E6, true },
{ -INFINITY, true },
{ INFINITY, false },
{ 123E6, false },
{ 0, false },
{ -0., true },
{ -FLT_MIN, true },
{ FLT_MIN, false },
/*
* Cannot be accurately represented as float,
* but sign should be preserved
*/
{ DBL_MAX, false},
{ -DBL_MAX, true},
{ DBL_MAX, false },
{ -DBL_MAX, true },
};
#ifdef __HAVE_LONG_DOUBLE
static const struct {
long double input;
bool is_negative;
} ldbl_values[] = {
{ -LDBL_MIN, true},
{ LDBL_MIN, false},
{ LDBL_MAX, false},
{ -LDBL_MAX, true},
{ -LDBL_MIN, true },
{ LDBL_MIN, false },
{ LDBL_MAX, false },
{ -LDBL_MAX, true },
};
#endif
ATF_TC(signbit);
ATF_TC_HEAD(signbit, tc)
{
atf_tc_set_md_var(tc, "descr","Check that signbit functions correctly");
atf_tc_set_md_var(tc, "descr",
"Check that signbit functions correctly");
}
ATF_TC_BODY(signbit, tc)
{
double iterator_d;
float iterator_f;
unsigned i;
for (unsigned int i = 0; i < __arraycount(values); i++) {
iterator_d = values[i].input;
iterator_f = (float) values[i].input;
if (signbit(iterator_f) != values[i].is_negative)
for (i = 0; i < __arraycount(values); i++) {
const float iterator_f = values[i].input;
const double iterator_d = values[i].input;
const long double iterator_l = values[i].input;
if (signbit(iterator_f) != values[i].is_negative) {
atf_tc_fail("%s:%d iteration %d signbitf is wrong"
" about the sign of %f", __func__,
__LINE__, i, iterator_f);
if (signbit(iterator_d) != values[i].is_negative)
" about the sign of %f", __func__, __LINE__, i,
iterator_f);
}
if (signbit(iterator_d) != values[i].is_negative) {
atf_tc_fail("%s:%d iteration %d signbit is wrong"
"about the sign of %f", __func__,
__LINE__,i, iterator_d);
#ifdef __HAVE_LONG_DOUBLE
long double iterator_l = values[i].input;
if (signbit(iterator_l) != values[i].is_negative)
"about the sign of %f", __func__, __LINE__, i,
iterator_d);
}
if (signbit(iterator_l) != values[i].is_negative) {
atf_tc_fail("%s:%d iteration %d signbitl is wrong"
" about the sign of %Lf", __func__,
__LINE__, i, iterator_l);
#endif
" about the sign of %Lf", __func__, __LINE__, i,
iterator_l);
}
}
#ifdef __HAVE_LONG_DOUBLE
for (unsigned int i = 0; i < __arraycount(ldbl_values); i++) {
if (signbit(ldbl_values[i].input) != ldbl_values[i].is_negative)
for (i = 0; i < __arraycount(ldbl_values); i++) {
if (signbit(ldbl_values[i].input) !=
ldbl_values[i].is_negative) {
atf_tc_fail("%s:%d iteration %d signbitl is"
"wrong about the sign of %Lf",
__func__, __LINE__, i,
ldbl_values[i].input);
"wrong about the sign of %Lf",
__func__, __LINE__, i,
ldbl_values[i].input);
}
}
#endif
#ifdef NAN
ATF_CHECK_EQ(signbit(copysignf(NAN, -1)), true);
ATF_CHECK_EQ(signbit(copysignf(NAN, +1)), false);
ATF_CHECK_EQ(signbit(copysign(NAN, -1)), true);
ATF_CHECK_EQ(signbit(copysign(NAN, +1)), false);
ATF_CHECK_EQ(signbit(copysignl(NAN, -1)), true);
ATF_CHECK_EQ(signbit(copysignl(NAN, +1)), false);
#endif
}
ATF_TP_ADD_TCS(tp)