I'm leaving for vacation and don't have time to review this till next week.
Maybe another gatekeeper could review it?
________________________________
From: 朱庆 [mailto:zqing1...@gmail.com]
Sent: Tuesday, August 31, 2010 12:57 AM
To: Mike Murphy
Cc: open64-devel@lists.sourceforge.net
Subject: Re: [Open64-devel] Assertion failure in wn_lower phase
New patch with your advice, just delete when replace those nodes.
Index: wn_lower.cxx
===================================================================
--- wn_lower.cxx (revision 3327)
+++ wn_lower.cxx (working copy)
@@ -2400,8 +2400,6 @@
WN_Sqrt(desc, WN_kid0(tree)));
lower_complex_expr(block, div, actions, realpart, imagpart);
-
- WN_Delete(tree);
}
break;
@@ -2416,7 +2414,6 @@
*realpart = WN_Paren(type, *realpart);
*imagpart = WN_Paren(type, *imagpart);
- WN_Delete(tree);
}
break;
@@ -2589,7 +2586,6 @@
*realpart = WN_LdidPreg(type, realpartN);
*imagpart = WN_LdidPreg(type, imagpartN);
}
- WN_Delete(tree);
}
break;
@@ -2817,7 +2813,6 @@
*realpart = WN_LdidPreg(type, realpartN);
*imagpart = WN_LdidPreg(type, imagpartN);
}
- WN_Delete(tree);
}
break;
@@ -2846,8 +2841,6 @@
realpart, &iz);
*imagpart = WN_Neg(type, iz);
-
- WN_Delete(tree);
}
break;
case INTRN_V16C8MPY_ADDSUB:
@@ -8041,6 +8034,7 @@
if (imagexp == NULL && realTY == WN_rtype(WN_kid0(tree)))
{
imagexp = WN_Floatconst(realTY, 0.0);
+ WN_Delete(WN_kid0(tree));
WN_kid0(tree) = WN_CreateExp2( OPR_PAIR, WN_desc(tree), MTYPE_V,
realexp, imagexp);
}
@@ -8467,7 +8461,7 @@
eval = WN_CreateEval(realexp);
WN_Set_Linenum(eval, current_srcpos); // Bug 1268
WN_INSERT_BlockLast(block, eval);
-
+ WN_Delete(child);
child = imagexp;
}
else if (Action(LOWER_QUAD) && MTYPE_is_quad(WN_rtype(child)))
---------
Thanks,
zhuqing
在 2010年8月27日 上午7:33，Mike Murphy
<mmur...@nvidia.com<mailto:mmur...@nvidia.com>>写道：
In several places you do
WN_kid0(tree) ==NULL;
which is one too many =.
The code around the original problem looks wrong too, as if you move
lower_complex_expr down, then line 8041 which references imageexp is wrong.
The usual rule is that the when you replace a node you should delete. To
always delete the node after every call seems potentially wrong, if the node
was still referenced (admittedly not the common case). The code for this
lower_complex_expr does not seem consistent which is confusing and thus is
causing the problem you are seeing. It looks to me like lower_complex_expr
should not be doing any deletes, but I wouldn't automatically do the deletes
after every call. It does look like lines 8044 and 8088 should be deletes,
plus 8465.
________________________________
From: 朱庆 [mailto:zqing1...@gmail.com<mailto:zqing1...@gmail.com>]
Sent: Wednesday, August 25, 2010 7:38 PM
To: Mike Murphy
Cc:
open64-devel@lists.sourceforge.net<mailto:open64-devel@lists.sourceforge.net>
Subject: Re: [Open64-devel] Assertion failure in wn_lower phase
Yes, you are right, I 've added the new patch in the attchment to deleted all
the whirl node after they were lowered. Could you help reviewing it again?
Thanks,
Zhuqing
在 2010年8月19日 下午1:07，Mike Murphy
<mmur...@nvidia.com<mailto:mmur...@nvidia.com>>写道：
Aren't there multiple calls to lower_complex_expr? In this case maybe only
deleting at the caller is right, but if other uses do not do the delete, and
you don't delete in the callee, then you will never delete.
________________________________
From: 朱庆 [mailto:zqing1...@gmail.com<mailto:zqing1...@gmail.com>]
Sent: Wednesday, August 18, 2010 9:23 PM
To:
open64-devel@lists.sourceforge.net<mailto:open64-devel@lists.sourceforge.net>
Subject: [Open64-devel] Assertion failure in wn_lower phase
Hi, All
Can gatekeeper help review this fix?
This bug is happens in wn_lower phase, more detailed info please refer to
https://bugs.open64.net/show_bug.cgi?id=591.
Attached is the patch.
The failure is due to one whirl node address attached by two nodes.
The main cause is that in wn_lower.cxx:7741:lower_complex_expr(block,
WN_kid0(tree), actions, &realexp, &imagexp), WN_kid0(tree) was deleted in the
function,
, howerver in 7783, the node deleted again , so if there are other new node
created between line 7741and 7783, it may reuse the deleted node address, and
will be wrongly deleted by line 7783.
The fix is modified function lower_complex_expr, delete all "WN_Delete(tree);".
7741 lower_complex_expr(block, WN_kid0(tree), actions, &realexp,
&imagexp);
7742 ......
7762 realexpN = AssignExpr(block, realexp, realTY);
7763 realexp_copy = WN_LdidPreg(realTY, realexpN);
7764 .......
7783 WN_Delete(WN_kid0(tree));
7784 WN_kid0(tree) = WN_Ldid(MTYPE_F8, 0, c4temp_st,
MTYPE_To_TY(MTYPE_F8));
Best Regards,
Zhu Qing
________________________________
This email message is for the sole use of the intended recipient(s) and may
contain confidential information. Any unauthorized review, use, disclosure or
distribution is prohibited. If you are not the intended recipient, please
contact the sender by reply email and destroy all copies of the original
message.
________________________________
------------------------------------------------------------------------------
This SF.net Dev2Dev email is sponsored by:
Show off your parallel programming skills.
Enter the Intel(R) Threading Challenge 2010.
http://p.sf.net/sfu/intel-thread-sfd
_______________________________________________
Open64-devel mailing list
Open64-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/open64-devel
