Re: [PATCH 2/2 v3][IRA,LRA] Fix PR86939, IRA incorrectly creates an interference between a pseudo register and a hard register

[Peter Bergner <bergner at linux dot ibm dot com>]

On 11/6/18 6:23 AM, Renlin Li wrote:
> I just did a bootstrap again with everything up to r264897 which is Oct 6.
> it produce the ICE I mentioned on the PR87899.
> 
> The first combiner patch on Oct 22.

Do the testsuite results (for disable-bootstrap builds) differ between
r264896 and r264897?  If so, that would be much easier to track down.

If not, maybe the following patch could help to narrow down which gcc
source file(s) are being miscompiled by allowing you to disable the
special handling of copy conflicts with an option?  The option default
(ie, not using the option or -fno-ira-copies-conflict) is the same behavior
as now and -fira-copies-conflict would make things behave like they did
before my patch.

Peter


Index: gcc/common.opt
===================================================================
--- gcc/common.opt	(revision 265402)
+++ gcc/common.opt	(working copy)
@@ -1761,6 +1761,10 @@ Enum(ira_region) String(all) Value(IRA_R
 EnumValue
 Enum(ira_region) String(mixed) Value(IRA_REGION_MIXED)
 
+fira-copies-conflict
+Common Report Var(flag_ira_copies_conflict) Init(0) Optimization
+Make pseudos connected by a copy conflict
+
 fira-hoist-pressure
 Common Report Var(flag_ira_hoist_pressure) Init(1) Optimization
 Use IRA based register pressure calculation
Index: gcc/ira-lives.c
===================================================================
--- gcc/ira-lives.c	(revision 265402)
+++ gcc/ira-lives.c	(working copy)
@@ -1066,7 +1066,7 @@ non_conflicting_reg_copy_p (rtx_insn *in
 {
   /* Reload has issues with overlapping pseudos being assigned to the
      same hard register, so don't allow it.  See PR87600 for details.  */
-  if (!targetm.lra_p ())
+  if (flag_ira_copies_conflict || !targetm.lra_p ())
     return NULL_RTX;
 
   rtx set = single_set (insn);