| Fri, 05 Aug 2011 18:44:32 +0200 |
Gilles Duboscq |
Merga
|
| Fri, 05 Aug 2011 17:40:35 +0200 |
Gilles Duboscq |
Reintroduce Conditional (IfOp) using the new infrastructure, fix the Cmov support in the backend, use x86 cmov when possible, make it work with floating point comparison conditions, make MaterializeNode a subclass of Conditional, fix nasty bug related to introducing LIR jumps in an existing LIRBlock in MaterializeNode/Conditional
|
| Wed, 03 Aug 2011 12:07:11 +0200 |
Lukas Stadler |
merge
|
| Wed, 03 Aug 2011 11:51:47 +0200 |
Lukas Stadler |
more work on new node structure, executes scimark, fop, avrora, luindex, lusearch, pmd, h2 and xalan
|
| Thu, 28 Jul 2011 08:18:43 +0200 |
Lukas Stadler |
changes to Node structure (incomplete)
|
| Wed, 27 Jul 2011 17:24:11 -0700 |
Thomas Wuerthinger |
Fixed compiler warnings.
|
| Wed, 13 Jul 2011 15:08:49 +0200 |
Gilles Duboscq |
Fix for loop inversion now runs tests, fop, lusearch, eclipse, avrora and scimark
|
| Tue, 12 Jul 2011 17:54:32 +0200 |
Gilles Duboscq |
Fix broken code in exemples, Fix regression and bug in peeling/inverting
|
| Tue, 12 Jul 2011 13:10:33 +0200 |
Gilles Duboscq |
Finish implementation of loop inversion
|
| Fri, 08 Jul 2011 18:00:39 +0200 |
Gilles Duboscq |
revert outer framesate scheduling change
|
| Fri, 08 Jul 2011 15:00:16 +0200 |
Gilles Duboscq |
Merge
|
| Fri, 08 Jul 2011 14:49:01 +0200 |
Gilles Duboscq |
Fix for peeling spliting : inner framestates which are attached to nodes after the coloring can not be colored
|
| Fri, 08 Jul 2011 13:38:38 +0200 |
Gilles Duboscq |
More precise inOrBefore, make both inOrBefore and inOrAfter accessible on Loop, compute inOrAfter, inOrBefore and full loop nodes only if needed
|
| Thu, 07 Jul 2011 18:21:30 +0200 |
Gilles Duboscq |
Canonicalize & DCE after peeling, fix for spliting in peeling
|
| Thu, 07 Jul 2011 16:04:21 +0200 |
Gilles Duboscq |
Removed assertion in EdgeMoveOptimizer that is not valid anymore because of guards
|
| Wed, 06 Jul 2011 16:31:18 +0200 |
Lukas Stadler |
added PhiType.(Value, Memory, Virtual), removed Phi.isDead, phis are now deleted immediately when they become invalid
|
| Sun, 03 Jul 2011 11:05:05 +0200 |
Gilles Duboscq |
Fix bug in peeling in updating the parent loops after peeling. Added verify to If
|
| Fri, 01 Jul 2011 18:37:54 +0200 |
Gilles Duboscq |
Fixed various peeling bugs (can use nodes which are not Placeholders as loop exits)
|
| Fri, 01 Jul 2011 12:56:52 +0200 |
Gilles Duboscq |
Make NewInstance a FixedWithNext to avoid it from floating too much (could be hoisted out of loops for exemple). Fixes for loop peeling
|
| Thu, 30 Jun 2011 10:07:35 +0200 |
Gilles Duboscq |
Fix some coloring bug, fix to keep more Placeholders at loop exits, fix for loop nodes computation
|
| Wed, 29 Jun 2011 12:23:13 +0200 |
Gilles Duboscq |
Fix loop peeling exit values fixup
|
| Tue, 28 Jun 2011 17:17:51 +0200 |
Gilles Duboscq |
Fix for dead Phis in loop nodes detection
|
| Tue, 28 Jun 2011 16:13:32 +0200 |
Gilles Duboscq |
Fix multiple bugs in loop peeling
|
| Tue, 28 Jun 2011 10:10:47 +0200 |
Gilles Duboscq |
Loop peeling work
|
| Tue, 21 Jun 2011 16:58:34 +0200 |
Gilles Duboscq |
Compute Loop information, keep Placeholder on loop exits, added some utility functions for data/cfg traversal
|
