Mon, 20 Feb 2012 18:01:22 +0100 |
Thomas Wuerthinger |
Merge.
|
Mon, 20 Feb 2012 17:58:46 +0100 |
Thomas Wuerthinger |
Visualizer nbproject changes.
|
Tue, 21 Feb 2012 16:35:42 +0100 |
Andreas Woess |
Don't emit moves for constants that are used only in frame states.
|
Tue, 21 Feb 2012 16:18:48 +0100 |
Andreas Woess |
Disable propagate types phase. Fix intrinsification of an invoke with a deoptimize node.
|
Mon, 20 Feb 2012 19:19:12 +0100 |
Gilles Duboscq |
Kill CallTargetNode and FrameState in killCFG to avoid keeping some invalid nodes alive
|
Mon, 20 Feb 2012 16:37:23 +0100 |
Lukas Stadler |
undo CanonicalizerPhase change
|
Mon, 20 Feb 2012 14:37:32 +0100 |
Lukas Stadler |
added Condition.implies(...)
|
Mon, 20 Feb 2012 14:34:51 +0100 |
Lukas Stadler |
Merge
|
Mon, 20 Feb 2012 14:29:51 +0100 |
Lukas Stadler |
experimental: added PiNode and PropagateTypesPhase
|
Mon, 20 Feb 2012 14:28:39 +0100 |
Lukas Stadler |
experimental: scheduling and unscheduling of the whole graph
|
Mon, 20 Feb 2012 14:27:35 +0100 |
Lukas Stadler |
experimental: "next" successor on all nodes
|
Mon, 20 Feb 2012 14:25:32 +0100 |
Lukas Stadler |
let StampFactory.alwaysDistinct(...) handle interface types
|
Mon, 20 Feb 2012 14:24:01 +0100 |
Lukas Stadler |
fixed unique call on already GVNed constant
|
Mon, 20 Feb 2012 14:22:23 +0100 |
Lukas Stadler |
use GraphOrder in CanonicalizerPhase (leads to fewer iterations)
|
Mon, 20 Feb 2012 14:20:28 +0100 |
Lukas Stadler |
separate GraphOrder from EscapeAnalysisPhase
|
Mon, 20 Feb 2012 14:18:38 +0100 |
Lukas Stadler |
small fix to PostOrderNodeIterator
|
Mon, 20 Feb 2012 12:30:58 +0100 |
Gilles Duboscq |
Synchronize createHotSpotTypeResolved (should fix EscapeAnalysis NPE)
|
Sun, 19 Feb 2012 04:28:40 +0100 |
Andreas Woess |
Fix NPE in DebugValueMap.group
|
Fri, 17 Feb 2012 16:03:09 -0800 |
Christian Haeubl |
added another variant for propagating loop frequencies
|
Fri, 17 Feb 2012 11:02:52 -0800 |
Christian Haeubl |
Merge
|
Fri, 17 Feb 2012 11:02:19 -0800 |
Christian Haeubl |
avoid code generation for typechecks that are only used for anchoring
|
Fri, 17 Feb 2012 10:34:34 -0800 |
Christian Haeubl |
fixed checkcast when inlining more than one method
|
Fri, 17 Feb 2012 10:30:39 -0800 |
Christian Haeubl |
fixed VS project files
|
Fri, 17 Feb 2012 19:24:53 +0100 |
Doug Simon |
Added missing file.
|
Fri, 17 Feb 2012 15:31:40 +0100 |
Gilles Duboscq |
Merge
|
Fri, 17 Feb 2012 14:47:57 +0100 |
Gilles Duboscq |
Merge
|
Fri, 17 Feb 2012 14:47:48 +0100 |
Gilles Duboscq |
Make it possible to pass jvm args in jtt and unittest
|
Fri, 17 Feb 2012 13:54:39 +0100 |
Gilles Duboscq |
Fix a problem with having a MemoryCheckpoint that is also a ControlSplit (InvokeWithException), the ANY_LOCATION can not be this control split in both branches because it may lead to scheduling problems : make ExceptionObject a MemoryCheckpoint
|
Fri, 17 Feb 2012 13:17:42 +0100 |
Gilles Duboscq |
check all framestates in LIRGen assert
|
Fri, 17 Feb 2012 12:37:29 +0100 |
Gilles Duboscq |
Merge
|
Fri, 17 Feb 2012 12:37:17 +0100 |
Gilles Duboscq |
Fix checkcast type in polymorphic inlining
|
Fri, 17 Feb 2012 11:37:44 +0100 |
Gilles Duboscq |
in esa mode, check that values in stateAfter are all ready after LIRGen of a ValudeNode
|
Thu, 16 Feb 2012 19:29:32 +0100 |
Gilles Duboscq |
Make assert in getInlineInfo output more information
|
Thu, 16 Feb 2012 18:59:18 +0100 |
Gilles Duboscq |
Merge
|
Thu, 16 Feb 2012 18:59:09 +0100 |
Gilles Duboscq |
Add checkcast for type-guarded inlining
|
Thu, 16 Feb 2012 18:24:31 +0100 |
Gilles Duboscq |
Fix merge
|
Thu, 16 Feb 2012 17:47:43 +0100 |
Gilles Duboscq |
Merge
|
Thu, 16 Feb 2012 17:47:14 +0100 |
Gilles Duboscq |
clenup unused floating nodes on killCFG
|
Thu, 16 Feb 2012 14:53:04 +0100 |
Gilles Duboscq |
Merge
|
Thu, 16 Feb 2012 14:43:05 +0100 |
Gilles Duboscq |
Add merge-before-loopend simplification for merges
|
Thu, 16 Feb 2012 13:48:47 +0100 |
Gilles Duboscq |
Transform some CiBailout into GraalInternalError
|
Thu, 16 Feb 2012 13:48:00 +0100 |
Gilles Duboscq |
Update visualiser's color filter
|
Thu, 16 Feb 2012 11:57:38 +0100 |
Gilles Duboscq |
Make it possible to have multiple LoopEnds per LoopBegin
|
Wed, 15 Feb 2012 20:09:25 +0100 |
Gilles Duboscq |
Make the graph object always available in context (helps debugging problems during LIRGen)
|
Wed, 15 Feb 2012 20:05:06 +0100 |
Gilles Duboscq |
Use parameters in graph group names for the visualizer so that successive compilation of methods with same name but different signature don't get fused
|
Tue, 14 Feb 2012 11:32:21 +0100 |
Gilles Duboscq |
Fix merge
|
Tue, 14 Feb 2012 11:28:57 +0100 |
Gilles Duboscq |
Merge
|
Mon, 13 Feb 2012 16:04:59 +0100 |
Gilles Duboscq |
move TypeProfileWidth definition inside the ifndef TIERED block
|
Fri, 17 Feb 2012 15:01:37 +0100 |
Doug Simon |
Added -g option to 'mx gate' to omit the server and client builds (i.e. -g == graal only).
|
Fri, 17 Feb 2012 14:07:32 +0100 |
Doug Simon |
Made GraalVM build work on Mac OS X.
|
Thu, 16 Feb 2012 16:32:14 -0800 |
Christian Haeubl |
Merge
|
Thu, 16 Feb 2012 16:31:27 -0800 |
Christian Haeubl |
disabled megamorphic calls, trying an inlining policy where invokes within loops do not have any advantages
|
Thu, 16 Feb 2012 14:02:08 -0800 |
Christian Haeubl |
disable megamorphic inlining for now
|
Thu, 16 Feb 2012 23:51:37 +0100 |
Doug Simon |
Small comment fix to test native code building guard in the gate.
|
Thu, 16 Feb 2012 22:58:26 +0100 |
Doug Simon |
Merge.
|
Thu, 16 Feb 2012 22:38:52 +0100 |
Doug Simon |
Added --omit-native-build to gate command so that native cleaning and building can be omitted for changesets that made no changes to the C/C++ code.
|
Thu, 16 Feb 2012 18:30:39 +0100 |
Andreas Woess |
Fix possible NullPointerException in HotSpotMethodResolvedImpl.toJava
|
Thu, 16 Feb 2012 17:08:32 +0100 |
Lukas Stadler |
evacuate guards when deleting control splits
|
Tue, 14 Feb 2012 18:00:32 -0800 |
Christian Haeubl |
set default inlining policy to the greedy one
|
Tue, 14 Feb 2012 15:01:36 -0800 |
Christian Haeubl |
fixed visual studio project files
|
Tue, 14 Feb 2012 14:01:37 -0800 |
Christian Haeubl |
minor change
|
Tue, 14 Feb 2012 14:00:23 -0800 |
Christian Haeubl |
Merge
|
Tue, 14 Feb 2012 13:53:25 -0800 |
Christian Haeubl |
bugfixes, added bytecode complexity estimation
|
Tue, 14 Feb 2012 20:19:29 +0100 |
Andreas Woess |
Canonicalize invoke kind to static when the exact receiver type is known, regardless of whether it is constant.
|
Tue, 14 Feb 2012 10:21:38 -0800 |
Christian Haeubl |
Merge
|
Mon, 13 Feb 2012 18:40:54 -0800 |
Christian Haeubl |
changed DynamicSizeBasedInliningPolicy to use log(probability) for frequently executed loops, added check for SmallCompiledCodeSize to WeightBasedInliningPolicy.
|
Mon, 13 Feb 2012 16:28:21 -0800 |
Christian Haeubl |
added assertion to ensure that methods are only inlined if they cannot be intrinsified
|
Mon, 13 Feb 2012 15:39:20 -0800 |
Christian Haeubl |
Fixed computation of exact type information in PhiStampPhase.
|
Tue, 14 Feb 2012 16:51:06 +0100 |
Lukas Stadler |
remove jtt tests from the gate (they fail in fastdebug)
|
Tue, 14 Feb 2012 16:27:59 +0100 |
Lukas Stadler |
add jtt tests, converted from the maxine repository using ConvertJTT
|
Tue, 14 Feb 2012 15:40:53 +0100 |
Lukas Stadler |
small fix to source code formatting (spaces in array initializers)
|
Mon, 13 Feb 2012 23:47:06 +0100 |
Andreas Woess |
Create exception dispatch even if catch type is not initialized yet.
|
Tue, 14 Feb 2012 12:29:11 +0100 |
Doug Simon |
Modified JDK replication process such that a copy of the default VM is made. This prevents issues with the replicated JDK being used to bootstrap a HotSpot build.
|
Mon, 13 Feb 2012 23:15:53 +0100 |
Doug Simon |
Made DebugInformationRecorder::describe_scope() take both a methodHandle _and_ a ciMethod* parameter to avoid creating handles in scopes where it is not allowed.
|
Mon, 13 Feb 2012 23:13:37 +0100 |
Doug Simon |
Made all vm builds go into subdirectories, even product builds to simplify building the various types of VMs (server, client and graal).
|
Mon, 13 Feb 2012 13:49:12 -0800 |
Christian Wimmer |
Use kind of input operand (instead of result operand) to determine the type of moves, because an Object constant can be loaded to a Long register for unsafe operation.
|
Mon, 13 Feb 2012 13:47:41 -0800 |
Christian Wimmer |
make Dump option work when IdealGraphVisualizer is not running. Print a warning message that connection is not possible instead of exiting the VM.
|
Mon, 13 Feb 2012 17:01:14 +0100 |
Andreas Woess |
Fix boxing elimination for boxed constants.
|
Mon, 13 Feb 2012 14:02:30 +0100 |
Doug Simon |
Fixed regressions in building and running debug verison of HotSpot Client.
|
Mon, 13 Feb 2012 14:01:49 +0100 |
Doug Simon |
Added debug build of HotSpot Client to the gate.
|
Mon, 13 Feb 2012 10:33:25 +0100 |
Doug Simon |
Removed unnecessary casts.
|
Mon, 13 Feb 2012 10:27:38 +0100 |
Doug Simon |
Clarified mx help text on how to build/run [Graal|Client|Server].
|
Sun, 12 Feb 2012 23:37:47 +0100 |
Andreas Woess |
Visualizer: Fix range slider scrollbar. Change app.name to graalvisualizer.
|
Sun, 12 Feb 2012 23:21:54 +0100 |
Andreas Woess |
Visualizer: Fix edge tooltip showing wrong node ids. GraalEdgeColorFilter: Don't change style of dashed lines.
|
Sat, 11 Feb 2012 23:35:44 +0100 |
Andreas Woess |
(Windows) Add Graal build configurations (with GRAAL defined) and fix build command.
|
Sat, 11 Feb 2012 11:15:32 -0800 |
Christian Haeubl |
bugfix
|
Sat, 11 Feb 2012 11:09:20 -0800 |
Christian Haeubl |
bugfix
|
Sat, 11 Feb 2012 10:30:46 -0800 |
Christian Haeubl |
added another inlining policy, added option to disable propagation of loop frequencies
|
Fri, 10 Feb 2012 17:59:35 -0800 |
Christian Haeubl |
better allocation of methodData's extraData so that exceptions and deoptimizations can be recorded more accurately
|
Fri, 10 Feb 2012 17:03:51 -0800 |
Christian Haeubl |
reverted default inlining policy
|
Fri, 10 Feb 2012 16:56:01 -0800 |
Christian Haeubl |
bugfix
|
Fri, 10 Feb 2012 14:37:38 -0800 |
Christian Haeubl |
active client compiler inlining policy for one run on the benchmark server
|
Fri, 10 Feb 2012 14:36:19 -0800 |
Christian Haeubl |
avoid recursive inlining, escape analysis does no longer restart inlining with level 0, bugfixes
|
Fri, 10 Feb 2012 11:14:51 -0800 |
Christian Haeubl |
added compilation time statistics to product mode
|
Fri, 10 Feb 2012 10:16:19 -0800 |
Christian Haeubl |
Merge
|
Fri, 10 Feb 2012 10:07:48 -0800 |
Christian Haeubl |
enabled nmethod statistics for product build
|
Fri, 10 Feb 2012 17:09:39 +0100 |
Gilles Duboscq |
Merge
|
Fri, 10 Feb 2012 17:04:03 +0100 |
Gilles Duboscq |
Make it possible again to build a real client libjvm, drop the UseGraal flag.
|
Fri, 10 Feb 2012 02:22:23 +0100 |
Gilles Duboscq |
Make graph dumping a bit more robust when there is no method, enable debug in the startCompiler method, add context and scope for snippets installation
|
Thu, 09 Feb 2012 13:50:52 +0100 |
Gilles Duboscq |
Merge
|
Thu, 09 Feb 2012 13:21:42 +0100 |
Gilles Duboscq |
Fix an other NPE in CFGPrinterObserver
|
Fri, 10 Feb 2012 13:58:04 +0100 |
Lukas Stadler |
explicitly exclude com.oracle.max.graal.tests from graal class path to make JUnit tests work
|
Thu, 09 Feb 2012 18:36:11 -0800 |
Christian Haeubl |
Merge
|
Thu, 09 Feb 2012 18:35:47 -0800 |
Christian Haeubl |
added option to summarize debug values
|
Thu, 09 Feb 2012 17:03:44 -0800 |
Christian Haeubl |
fixed another endless recompilation
|
Thu, 09 Feb 2012 13:26:51 -0800 |
Christian Haeubl |
changed profiling maturity so that profiling information is used more likely
|
Thu, 09 Feb 2012 17:37:53 -0800 |
Christian Wimmer |
Lower NormalizeCompareNode in the front end to avoid LIR operation.
|
Thu, 09 Feb 2012 23:54:33 +0100 |
Thomas Wuerthinger |
Fixed a bug in the canonicalizer of the UnboxNode (brought up by Truffle).
|
Thu, 09 Feb 2012 21:26:26 +0100 |
Thomas Wuerthinger |
Added experimental ConvertDeoptimizeToGuardPhase.
|
Thu, 09 Feb 2012 20:05:59 +0100 |
Thomas Wuerthinger |
Allow to intrinsify an invoke with a deoptimize node. Make debug output more relaxed wrt to null in FrameMap.method
|
Thu, 09 Feb 2012 18:20:56 +0100 |
Thomas Wuerthinger |
Make IntegerAddNode non-final.
|
Thu, 09 Feb 2012 13:39:07 +0100 |
Thomas Wuerthinger |
Fix unit test cases for new way of creating exception edges for invokes (i.e. only create edge if there was at least 1 exception at that place).
|