------- Comment #1 from hjl dot tools at gmail dot com 2010-06-10 05:23 -------
This patch
---
iff --git a/gcc/config/i386/predicates.md b/gcc/config/i386/predicates.md
index a33d3af..6f569cc 100644
--- a/gcc/config/i386/predicates.md
+++ b/gcc/config/i386/predicates.md
@@ -820,7 +820,10 @@
/* All patterns using aligned_operand on memory operands ends up
in promoting memory operand to 64bit and thus causing memory mismatch.
*/
- if (TARGET_MEMORY_MISMATCH_STALL && !optimize_insn_for_size_p ())
+ if (TARGET_MEMORY_MISMATCH_STALL
+ && !optimize_insn_for_size_p ()
+ && (!MEM_EXPR (op)
+ || TREE_CODE (MEM_EXPR (op)) != PARM_DECL))
return 0;
/* Don't even try to do any aligned optimizations with volatiles. */
----
helps. I got
.file "call-1.c"
.text
.p2align 4,,15
.globl foo
.type foo, @function
foo:
pushl %ebp
movl %esp, %ebp
movl 12(%ebp), %eax
addl 8(%ebp), %eax
addl 16(%ebp), %eax
popl %ebp
ret
.size foo, .-foo
.p2align 4,,15
.globl bar
.type bar, @function
bar:
pushl %ebp
movl %esp, %ebp
movswl 12(%ebp), %eax << still read 2 bytes.
movswl 16(%ebp), %edx
movswl 8(%ebp), %ecx
movl %edx, 12(%ebp)
movl %ecx, 16(%ebp)
movl %eax, 8(%ebp)
popl %ebp
jmp foo
--
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=44490
