This is the mail archive of the gcc-patches@gcc.gnu.org mailing list for the GCC project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

[PATCH]: Testcase for buggy MPFR versions


This testcase checks for a bug in MPFR that exists in mpfr-2.2.0.  There
exist some fixes for mpfr-2.2.0 in a "cumulative patch", however this bug
and several others are not fixed by that patch.  The problem is however
addressed in mpfr-2.2.1.

Since many people were against a hard requirement for mpfr-2.2.1 in gcc,
this testcase helps illustrate to the user that they may wish to upgrade
their MPFR package without stopping them from building gcc.

Note: the existing builtin-sin-mpfr-1.c test is similar but slightly
different in that the underlying sin bug it tests *is* in fact fixed by
the cumulative patch.  The pow one is not.

Tested with mainline linked with mpfr-2.2.0, mpfr-2.2.0+cumulative patch
and mpfr-2.2.1.  The testcase fails in the first two cases and passes in
the latter one (as expected).

Okay for mainline?

		Thanks,
		--Kaveh



2006-12-17  Kaveh R. Ghazi  <ghazi@caip.rutgers.edu>

	* gcc.dg/torture/builtin-pow-mpfr-1.c: New test.

diff -rup orig/egcc-SVN20061216/gcc/testsuite/gcc.dg/torture/builtin-pow-mpfr-1.c egcc-SVN20061216/gcc/testsuite/gcc.dg/torture/builtin-pow-mpfr-1.c
--- orig/egcc-SVN20061216/gcc/testsuite/gcc.dg/torture/builtin-pow-mpfr-1.c	2006-12-17 19:25:00.560310022 -0500
+++ egcc-SVN20061216/gcc/testsuite/gcc.dg/torture/builtin-pow-mpfr-1.c	2006-12-17 19:23:47.683436338 -0500
@@ -0,0 +1,19 @@
+/* Version 2.2.0 of MPFR had bugs in pow underflow/overflow.  This
+   test checks to see if that buggy version was installed.  The
+   problem is NOT fixed with the version 2.2.0 "cumulative patch".
+   However it is fixed in version 2.2.1 and presumably later MPFR
+   versions.
+
+   Origin: Kaveh R. Ghazi 12/17/2006.  */
+
+/* { dg-do compile } */
+/* { dg-options "-fdump-tree-original" } */
+
+extern double testit()
+{
+  /* This underflows and therefore gcc should not fold it.  */
+  return __builtin_pow (0.11, 1.0e38);
+}
+
+/* { dg-final { scan-tree-dump "pow" "original" } } */
+/* { dg-final { cleanup-tree-dump "original" } } */


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]