The following patch fixes
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=58048
LRA has a correct behaviour (as reload) on the test with -O0. But with
-O2 a constant propagation into asm operands resulted in LRA cycling
instead of error reporting.
The patch was successfully bootstrapped and tested on x86/x86-64.
Committed as rev. 201454.
2013-08-02 Vladimir Makarov <[email protected]>
PR rtl-optimization/58048
* lra-constraints.c (process_alt_operands): Don't check asm
operand on register.
2013-08-02 Vladimir Makarov <[email protected]>
PR rtl-optimization/58048
* gcc.target/i386/pr58048.c: New.
Index: lra-constraints.c
===================================================================
--- lra-constraints.c (revision 201438)
+++ lra-constraints.c (working copy)
@@ -1892,7 +1892,7 @@ process_alt_operands (int only_alternati
/* For asms, verify that the class for this alternative is
possible
for the mode that is specified. */
- if (!no_regs_p && REG_P (op) && INSN_CODE (curr_insn) < 0)
+ if (!no_regs_p && INSN_CODE (curr_insn) < 0)
{
int i;
for (i = 0; i < FIRST_PSEUDO_REGISTER; i++)
Index: testsuite/gcc.target/i386/pr58048.c
===================================================================
--- testsuite/gcc.target/i386/pr58048.c (revision 0)
+++ testsuite/gcc.target/i386/pr58048.c (working copy)
@@ -0,0 +1,11 @@
+/* PR target/58048 */
+/* { dg-do compile } */
+/* { dg-options "-O2" } */
+
+void
+div3 (void)
+{
+ double tmp1;
+
+ asm volatile ("fscale":"=t" (tmp1):"0" (0), "u" (0)); /* { dg-error
"inconsistent operand constraints in an 'asm'" } */
+}
