[Open64-devel] Bug fix for DO_LOOP_INFO constructor

2011-05-17 Thread Gilmore, Doug
We saw a hard-to-reproduce runtime regression in SPEC Leslie3 that was due the field Prefer_Fuse not being initialized in an instance of the DO_LOOP_INFO class that was constructed with: DO_LOOP_INFO::DO_LOOP_INFO(DO_LOOP_INFO *dli, MEM_POOL *pool) I attached a patch to fix this problem. Also I

Re: [Open64-devel] patch for better AMD64 ABI compatibility

2011-05-17 Thread David Coakley
1. Yes, we should be able to generalize and handle non-struct return cases too. However, the patch is well-tested as-is so I would prefer to make that improvement as a separate change and do more testing. 2. add_fake_parm creates a new VCALL with return type of void, so it will not add another fa

[Open64-devel] r3608 - trunk/osprey/be/opt

2011-05-17 Thread svn
Author: meiye Date: 2011-05-17 15:32:07 -0400 (Tue, 17 May 2011) New Revision: 3608 Modified: trunk/osprey/be/opt/opt_proactive.cxx trunk/osprey/be/opt/opt_proactive.h Log: Change interface for function calls returning 2 values. CR: Sun Chan Modified: trunk/osprey/be/opt/opt_proactive.cxx =

Re: [Open64-devel] patch for better AMD64 ABI compatibility

2011-05-17 Thread Yin Ma
Hi David, I looked at those code. I have some concerns on this implementations. 1. in wgen_decl.cxx/wgen_expr.cxx, the code only remove Mreturn.bar9_temp_0 if Need_Hidden_Parameter return true, which need first fake argument. Should we also remove temp_0 for the case that the return