Hi!
As discussed in the PR, if a var isn't addressable and has gimple reg type,
I don't see any point to honor it's DECL_ALIGN, we only refer to the
var through SSA_NAME_VAR of SSA_NAMEs, nothing is allocated on the stack
immediately and the SSA_NAMEs are turned into pseudos for which we only care
about their modes and corresponding alignments if they need to be spilled to
stack.
Bootstrapped/regtested on x86_64-linux and i686-linux, ok for trunk?
2014-01-09 Jakub Jelinek <ja...@redhat.com>
PR middle-end/47735
* cfgexpand.c (expand_one_var): For SSA_NAMEs, if the underlying
var satisfies use_register_for_decl, just take into account type
alignment, rather than decl alignment.
* gcc.target/i386/pr47735.c: New test.
--- gcc/cfgexpand.c.jj 2014-01-08 19:37:33.630986939 +0100
+++ gcc/cfgexpand.c 2014-01-09 13:38:45.073324129 +0100
@@ -1215,8 +1215,11 @@ expand_one_var (tree var, bool toplevel,
we conservatively assume it will be on stack even if VAR is
eventually put into register after RA pass. For non-automatic
variables, which won't be on stack, we collect alignment of
- type and ignore user specified alignment. */
- if (TREE_STATIC (var) || DECL_EXTERNAL (var))
+ type and ignore user specified alignment. Similarly for
+ SSA_NAMEs for which use_register_for_decl returns true. */
+ if (TREE_STATIC (var)
+ || DECL_EXTERNAL (var)
+ || (TREE_CODE (origvar) == SSA_NAME && use_register_for_decl (var)))
align = MINIMUM_ALIGNMENT (TREE_TYPE (var),
TYPE_MODE (TREE_TYPE (var)),
TYPE_ALIGN (TREE_TYPE (var)));
--- gcc/testsuite/gcc.target/i386/pr47735.c.jj 2014-01-09 13:30:14.410941107
+0100
+++ gcc/testsuite/gcc.target/i386/pr47735.c 2014-01-09 13:28:45.000000000
+0100
@@ -0,0 +1,16 @@
+/* PR middle-end/47735 */
+/* { dg-do compile } */
+/* { dg-options "-O2 -fomit-frame-pointer" } */
+
+unsigned
+mulh (unsigned a, unsigned b)
+{
+ unsigned long long l __attribute__ ((aligned (32)))
+ = ((unsigned long long) a * (unsigned long long) b) >> 32;
+ return l;
+}
+
+/* No need to dynamically realign the stack here. */
+/* { dg-final { scan-assembler-not "and\[^\n\r]*%\[re\]sp" } } */
+/* Nor use a frame pointer. */
+/* { dg-final { scan-assembler-not "%\[re\]bp" } } */
Jakub
