Module Name: src
Committed By: dsl
Date: Sun Mar 2 22:40:45 UTC 2014
Modified Files:
src/tests/lib/libm: t_acos.c
Log Message:
Include the subtest number in any error output.
Also temporarily print the rounding mode.
I think that acos(-1) is ending up with the wrong sign because the test
is being run with 'round towards -ve infinity' set.
I think it getting set somewhere and causing this test to fail.
The acos() code probably needs fixing - it shouldn't depend on the
round mode like this. But first I want to know if this if the error.
To generate a diff of this commit:
cvs rdiff -u -r1.5 -r1.6 src/tests/lib/libm/t_acos.c
Please note that diffs are not public domain; they are subject to the
copyright notices on the relevant files.
Modified files:
Index: src/tests/lib/libm/t_acos.c
diff -u src/tests/lib/libm/t_acos.c:1.5 src/tests/lib/libm/t_acos.c:1.6
--- src/tests/lib/libm/t_acos.c:1.5 Sat Mar 1 21:08:39 2014
+++ src/tests/lib/libm/t_acos.c Sun Mar 2 22:40:45 2014
@@ -1,4 +1,4 @@
-/* $NetBSD: t_acos.c,v 1.5 2014/03/01 21:08:39 dsl Exp $ */
+/* $NetBSD: t_acos.c,v 1.6 2014/03/02 22:40:45 dsl Exp $ */
/*-
* Copyright (c) 2011 The NetBSD Foundation, Inc.
@@ -31,29 +31,32 @@
#include <atf-c.h>
#include <math.h>
+#include <fenv.h>
/*
* Check result of fn(arg) is correct within the bounds.
* Should be ok to do the checks using 'double' for 'float' functions.
*/
-#define T_LIBM_CHECK(fn, arg, expect, epsilon) do { \
+#define T_LIBM_CHECK(subtest, fn, arg, expect, epsilon) do { \
double r = fn(arg); \
double e = fabs(r - expect); \
if (e > epsilon) \
- atf_tc_fail_nonfatal(#fn "(%g) is %g not %g (error %g > %g)", \
- arg, r, expect, e, epsilon); \
+ atf_tc_fail_nonfatal( \
+ "subtest %zu: " #fn "(%g) is %g not %g (error %g > %g), roundmode %x", \
+ subtest, arg, r, expect, e, epsilon, fegetround()); \
} while (0)
/* Check that the result of fn(arg) is NaN */
#ifndef __vax__
-#define T_LIBM_CHECK_NAN(fn, arg) do { \
+#define T_LIBM_CHECK_NAN(subtest, fn, arg) do { \
double r = fn(arg); \
if (!isnan(r)) \
- atf_tc_fail_nonfatal(#fn "(%g) is %g not NaN", arg, r); \
+ atf_tc_fail_nonfatal("subtest %zu: " #fn "(%g) is %g not NaN", \
+ subtest, arg, r); \
} while (0)
#else
/* vax doesn't support NaN */
-#define T_LIBM_CHECK_NAN(fn, arg) (void)(arg)
+#define T_LIBM_CHECK_NAN(subtest, fn, arg) (void)(arg)
#endif
#define AFT_LIBM_TEST(name, description) \
@@ -78,11 +81,11 @@ AFT_LIBM_TEST(acos_nan, "Test acos/acosf
size_t i;
for (i = 0; i < __arraycount(x); i++) {
- T_LIBM_CHECK_NAN(acos, x[i]);
+ T_LIBM_CHECK_NAN(i, acos, x[i]);
if (i < 2)
/* Values are too small for float */
continue;
- T_LIBM_CHECK_NAN(acosf, x[i]);
+ T_LIBM_CHECK_NAN(i, acosf, x[i]);
}
}
@@ -104,14 +107,14 @@ AFT_LIBM_TEST(acos_inrange, "Test acos/a
size_t i;
/*
- * Note that acos(x) might be calcualted as atan((1-x*x)/x).
- * This means that acos(-1) is atan(-0.0), if the sign is lost
- * the value will be 0 (atan(+0)) not M_PI.
+ * Note that acos(x) might be calculated as atan2(sqrt(1-x*x),x).
+ * This means that acos(-1) is atan2(+0,-1), if the sign is wrong
+ * the value will be -M_PI (atan2(-0,-1)) not M_PI.
*/
for (i = 0; i < __arraycount(values); i++) {
- T_LIBM_CHECK(acos, values[i].x, values[i].y, 1.0e-15);
- T_LIBM_CHECK(acosf, values[i].x, values[i].y, 1.0e-5);
+ T_LIBM_CHECK(i, acos, values[i].x, values[i].y, 1.0e-15);
+ T_LIBM_CHECK(i, acosf, values[i].x, values[i].y, 1.0e-5);
}
}
