On 13-06-27 12:22 PM, Vladimir Makarov wrote:
The following patch fixes a LRA bug for some experimental target.
Sorry, there is no a test case and the bug can not be reproduced by
other people until the target is made public.
Here is one more fix for another hard reg splitting bug.
The patch was successfully bootstrapped and tested on x86/x86-64.
Committed as rev. 200553.
2013-06-28 Vladimir Makarov <vmaka...@redhat.com>
* lra-constraints.c (need_for_split_p): Check call used hard regs
living through calls.
Index: lra-constraints.c
===================================================================
--- lra-constraints.c (revision 200552)
+++ lra-constraints.c (working copy)
@@ -4201,6 +4201,15 @@ need_for_split_p (HARD_REG_SET potential
&& (regno >= FIRST_PSEUDO_REGISTER
|| ! TEST_HARD_REG_BIT (eliminable_regset, hard_regno))
&& ! TEST_HARD_REG_BIT (lra_no_alloc_regs, hard_regno)
+ /* Don't split call clobbered hard regs living through
+ calls, otherwise we might have a check problem in the
+ assign sub-pass as in the most cases (exception is a
+ situation when lra_risky_transformations_p value is
+ true) the assign pass assumes that all pseudos living
+ through calls are assigned to call saved hard regs. */
+ && (regno >= FIRST_PSEUDO_REGISTER
+ || ! TEST_HARD_REG_BIT (call_used_reg_set, regno)
+ || usage_insns[regno].calls_num == calls_num)
/* We need at least 2 reloads to make pseudo splitting
profitable. We should provide hard regno splitting in
any case to solve 1st insn scheduling problem when
@@ -4799,7 +4808,7 @@ inherit_in_ebb (rtx head, rtx tail)
usage_insns invalidation. */
if ((dst_regno < FIRST_PSEUDO_REGISTER
|| reg_renumber[dst_regno] >= 0)
- && ! reg->subreg_p && reg->type != OP_IN)
+ && ! reg->subreg_p && reg->type != OP_IN)
{
HARD_REG_SET s;
