Hi JianXin,
Thanks for the review comments. I have attached the revised patch with these
changes. Let me know if it is fine.
Regards,
Shivaram
From: Jian-Xin Lai [mailto:laij...@gmail.com]
Sent: Thursday, September 08, 2011 7:25 AM
To: Rao, Shivarama
Cc: open64-devel
Subject: Re: [Open64-devel] Code review request (CG)
Also, it's better to move the definition of arcs close to its uses.
2011/9/8 Jian-Xin Lai <laij...@gmail.com<mailto:laij...@gmail.com>>
Hi Shivaram,
The true_arcs and ops are defined but never used. Could you please check that?
2011/8/24 Rao, Shivarama <shivarama....@amd.com<mailto:shivarama....@amd.com>>
Hi,
Could a gatekeeper please review the attached patch.
This patch fixes a problem in load folding optimization where the memory
location used in the load operation
is overwritten by a store operation. A check is added to prevent folding of
such load operations.
The test-case and the patch is attached with the mail.
Regards,
Shivaram
AMD Open Source Compiler Engineering
------------------------------------------------------------------------------
EMC VNX: the world's simplest storage, starting under $10K
The only unified storage solution that offers unified management
Up to 160% more powerful than alternatives and 25% more efficient.
Guaranteed. http://p.sf.net/sfu/emc-vnx-dev2dev
_______________________________________________
Open64-devel mailing list
Open64-devel@lists.sourceforge.net<mailto:Open64-devel@lists.sourceforge.net>
https://lists.sourceforge.net/lists/listinfo/open64-devel
--
Regards,
Lai Jian-Xin
--
Regards,
Lai Jian-Xin
Index: be/cg/x8664/ebo_special.cxx
===================================================================
--- be/cg/x8664/ebo_special.cxx (revision 3725)
+++ be/cg/x8664/ebo_special.cxx (working copy)
@@ -10983,7 +10983,42 @@
return FALSE;
if (index_loc >= 0 && !Pred_Opnd_Avail(op, loaded_tn_info, index_loc))
return FALSE;
+ // for this bb, obtain the dependence graph so that we can
+ // walk this op's expression tree to check dependencies.
+ // Make sure that load operand is not overwritten
+ // by a store operation before op using this dependency graph.
+ CG_DEP_Compute_Graph ( op->bb,
+ NO_ASSIGNED_REG_DEPS,
+ NON_CYCLIC,
+ NO_MEMREAD_ARCS,
+ INCLUDE_MEMIN_ARCS,
+ NO_CONTROL_ARCS,
+ NULL);
+
+
+ const INT hash_value = EBO_hash_op( load, NULL );
+ EBO_OP_INFO* opinfo = EBO_opinfo_table[hash_value];
+
+ while( opinfo != NULL ) {
+ OP* next_op = opinfo->in_op;
+ if( next_op == load ) break;
+ if( next_op && OP_store( next_op ) ) {
+ ARC_LIST *arcs;
+ for (arcs = OP_succs(load); arcs != NULL; arcs = ARC_LIST_rest(arcs)) {
+ ARC *arc = ARC_LIST_first(arcs);
+ if (ARC_kind(arc) != CG_DEP_MEMANTI) continue;
+ OP *succ_op = ARC_succ(arc);
+ if ((succ_op == next_op) && OP_Precedes(next_op, op)) {
+ CG_DEP_Delete_Graph (op->bb);
+ return FALSE;
+ }
+ }
+ }
+ opinfo = opinfo->same;
+ }
+ CG_DEP_Delete_Graph (op->bb);
+
TOP topcode;
if (base && offset && index && scale) {
new_op = Mk_OP (TOP_fmovddupxx,
------------------------------------------------------------------------------
Doing More with Less: The Next Generation Virtual Desktop
What are the key obstacles that have prevented many mid-market businesses
from deploying virtual desktops? How do next-generation virtual desktops
provide companies an easier-to-deploy, easier-to-manage and more affordable
virtual desktop model.http://www.accelacomm.com/jaw/sfnl/114/51426474/
_______________________________________________
Open64-devel mailing list
Open64-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/open64-devel