PING: PATCH: PR middle-end/45234: [4.4/4.5/4.6 Regression] ICE in expand_call, at calls.c:2845 when passing aligned function argument from unaligned stack after alloca
H.J. Lu
hjl.tools@gmail.com
Thu Aug 26 17:53:00 GMT 2010
On Thu, Aug 26, 2010 at 9:19 AM, Dominique Dhumieres <dominiq@lps.ens.fr> wrote:
> On x86_64-apple-darwin10.4.0 this patch caused the following failures:
> FAIL: g++.dg/eh/async-unwind2.C execution test
> FAIL: gcc.c-torture/execute/struct-ret-1.c execution, -Os
> FAIL: libgomp.c++/collapse-2.C -O1 execution test
> FAIL: libgomp.c++/for-1.C -O1 execution test
> FAIL: libgomp.c++/for-8.C -O1 execution test
> at r163063 and with -m32.
>
I need to take pending_stack_adjust into account. I am testing it on Linux/ia32
and Linux/x86-64. Can you try this new patch on Darwin?
Thanks.
--
H.J.
-------------- next part --------------
gcc/
2010-08-26 H.J. Lu <hongjiu.lu@intel.com>
PR middle-end/45234
* calls.c (expand_call): Make sure that all variable sized
adjustments are multiple of preferred stack boundary after
stack alignment.
gcc/testsuite/
2010-08-26 H.J. Lu <hongjiu.lu@intel.com>
PR middle-end/45234
* gcc.dg/torture/stackalign/alloca-5.c: New.
diff --git a/gcc/calls.c b/gcc/calls.c
index cd0d9c5..87b5808 100644
--- a/gcc/calls.c
+++ b/gcc/calls.c
@@ -2377,6 +2377,19 @@ expand_call (tree exp, rtx target, int ignore)
preferred_unit_stack_boundary = preferred_stack_boundary / BITS_PER_UNIT;
+ if (SUPPORTS_STACK_ALIGNMENT)
+ {
+ /* All variable sized adjustments must be multiple of preferred
+ stack boundary. Stack alignment may change preferred stack
+ boundary after variable sized adjustments have been made. We
+ need to compensate it here. */
+ unsigned HOST_WIDE_INT delta
+ = ((stack_pointer_delta - pending_stack_adjust)
+ % preferred_unit_stack_boundary);
+ if (delta)
+ anti_adjust_stack (GEN_INT (preferred_unit_stack_boundary - delta));
+ }
+
/* We want to make two insn chains; one for a sibling call, the other
for a normal call. We will select one of the two chains after
initial RTL generation is complete. */
--- /dev/null 2010-07-23 13:04:30.193381062 -0700
+++ gcc/gcc/testsuite/gcc.dg/torture/stackalign/alloca-5.c 2010-08-10 06:29:57.308629935 -0700
@@ -0,0 +1,32 @@
+/* PR middle-end/45234 */
+/* { dg-do run { target { { i?86-*-* x86_64-*-* } && ilp32 } } } */
+/* { dg-options "-m32 -mincoming-stack-boundary=2 -mpreferred-stack-boundary=2" } */
+
+#include "check.h"
+
+void
+__attribute__ ((noinline))
+bar (__float128 f)
+{
+ check (&f, __alignof__(f));
+}
+
+int
+main (void)
+{
+ char *p = __builtin_alloca (6);
+
+ bar (0);
+
+ __builtin_strncpy (p, "good", 5);
+ if (__builtin_strncmp (p, "good", 5) != 0)
+ {
+#ifdef DEBUG
+ p[size] = '\0';
+ printf ("Failed: %s != good\n", p);
+#endif
+ abort ();
+ }
+
+ return 0;
+}
More information about the Gcc-patches
mailing list