The following patch fixes the P1 PR. The details can be found on
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=65123
The patch was bootstrapped and tested on x86-64.
There is no new test case as the PR is reported against a test in GCC
testuite.
Committed as rev.220946.
2015-02-24 Vladimir Makarov <vmaka...@redhat.com>
PR rtl-optimization/65123
* lra-remat.c (operand_to_remat): Check hard regs in insn
definition too.
Index: lra-remat.c
===================================================================
--- lra-remat.c (revision 220916)
+++ lra-remat.c (working copy)
@@ -459,6 +459,16 @@ operand_to_remat (rtx_insn *insn)
reg2 = reg2->next)
if (reg2->type == OP_OUT && reg->regno == reg2->regno)
return -1;
+ if (reg->regno < FIRST_PSEUDO_REGISTER)
+ for (struct lra_insn_reg *reg2 = static_id->hard_regs;
+ reg2 != NULL;
+ reg2 = reg2->next)
+ if (reg2->type == OP_OUT
+ && reg->regno <= reg2->regno
+ && (reg2->regno
+ < (reg->regno
+ + hard_regno_nregs[reg->regno][reg->biggest_mode])))
+ return -1;
}
/* Find the rematerialization operand. */
int nop = static_id->n_operands;
