This is the mail archive of the
gcc-patches@gcc.gnu.org
mailing list for the GCC project.
[committed] testcase for pr26483, ia64 libffi denorm failure
- From: James E Wilson <wilson at specifix dot com>
- To: gcc-patches at gcc dot gnu dot org
- Date: Fri, 07 Apr 2006 16:14:09 -0700
- Subject: [committed] testcase for pr26483, ia64 libffi denorm failure
I have checked in this testcase. It reproduces the problem reported in
PR26498, which is a case where denorm values get accidentally rounded by
libffi on IA-64. I have an unfinished patch in the PR, which I will try
to finish this weekend. I tested this new testcase on ia64-linux
unpatched, ia64-linux patched, x86_64-linux, and x86_64-linux with
-m32. It failed for the first one, and worked for the other 3, which is
what I would expect.
--
Jim Wilson, GNU Tools Support, http://www.specifix.com
2006-04-07 James E Wilson <wilson@specifix.com>
* testsuite/libffi.call/float4.c: New testcase.
Index: testsuite/libffi.call/float4.c
===================================================================
--- testsuite/libffi.call/float4.c (revision 0)
+++ testsuite/libffi.call/float4.c (revision 0)
@@ -0,0 +1,60 @@
+/* Area: ffi_call
+ Purpose: Check denorm double value.
+ Limitations: none.
+ PR: PR26483.
+ Originator: From the original ffitest.c */
+
+/* { dg-do run } */
+#include "ffitest.h"
+#include "float.h"
+
+typedef union
+{
+ double d;
+ unsigned char c[sizeof (double)];
+} value_type;
+
+#define CANARY 0xba
+
+static double dblit(double d)
+{
+ return d;
+}
+
+int main (void)
+{
+ ffi_cif cif;
+ ffi_type *args[MAX_ARGS];
+ void *values[MAX_ARGS];
+ double d;
+ value_type result[2];
+ unsigned int i;
+
+ args[0] = &ffi_type_double;
+ values[0] = &d;
+
+ /* Initialize the cif */
+ CHECK(ffi_prep_cif(&cif, FFI_DEFAULT_ABI, 1,
+ &ffi_type_double, args) == FFI_OK);
+
+ d = DBL_MIN / 2;
+
+ /* Put a canary in the return array. This is a regression test for
+ a buffer overrun. */
+ memset(result[1].c, CANARY, sizeof (double));
+
+ ffi_call(&cif, FFI_FN(dblit), &result[0].d, values);
+
+ /* The standard delta check doesn't work for denorms. Since we didn't do
+ any arithmetic, we should get the original result back, and hence an
+ exact check should be OK here. */
+
+ CHECK(result[0].d == dblit(d));
+
+ /* Check the canary. */
+ for (i = 0; i < sizeof (double); ++i)
+ CHECK(result[1].c[i] == CANARY);
+
+ exit(0);
+
+}