Viktor Hofer [Thu, 19 Oct 2017 21:23:14 +0000 (23:23 +0200)]
Make ContractException public in impl assembly to support corert typeforwarding (#14589)
* Make ContractException public in impl assembly to support corert typeforwarding
Jan Kotas [Thu, 19 Oct 2017 21:20:03 +0000 (14:20 -0700)]
Fix condition for mustExpand flag (#14590)
The flag was getting overwritten when both CORINFO_FLG_INTRINSIC and CORINFO_FLG_JIT_INTRINSIC are set
Brian Sullivan [Thu, 19 Oct 2017 18:14:58 +0000 (11:14 -0700)]
Merge pull request #14582 from briansull/cleanup-iconhdl
Cleanup IconHandle nodes
Bruce Forstall [Thu, 19 Oct 2017 16:28:58 +0000 (09:28 -0700)]
Merge pull request #14570 from sdmaclea/PR-ARM64-FLAG_FEATURE_SIMD
[Arm64] Support flag FEATURE_SIMD
sergey ignatov [Thu, 19 Oct 2017 16:28:20 +0000 (19:28 +0300)]
[RyuJIT/ARM32] Support for CFI unwind info (#14447)
* [RyuJIT/ARM32] Support for CFI unwind info
Jim Ma [Thu, 19 Oct 2017 05:03:26 +0000 (13:03 +0800)]
Use ascii * instead of Chinese * character. (#14584)
Currently we have Chinese `*` characters at `src\classlibnative\bcltype\number.cpp` L289-294. That would cause a `C4819: Non-ASCII character in source` warning when building on Windows via Visual Studio (version: 15.4.0 (msvc 19.11.25547)).
This PR replaces the Chinese * characters with ascii *.
Fix #14556
Carol Eidt [Thu, 19 Oct 2017 04:17:43 +0000 (21:17 -0700)]
Merge pull request #14534 from CarolEidt/RegSelectCleanup
Refactor register selection heuristics
Koundinya Veluri [Thu, 19 Oct 2017 03:51:30 +0000 (20:51 -0700)]
Fix hill climbing float overflow (#14505)
Fix hill climbing float overflow
- When hill climbing finds that it wants to decrease the thread count but can't because the thread count is already the minimum, it instead tries to increase the sampling interval by a factor of up to 10 depending on how much it wanted to decrease the thread count
- The ratio was being used incorrectly (used max instead of min), and sometimes the ratio can be so large that the conversion to int after the float math overflows
- If something in the process enabled floating point exceptions, it may also crash
- There doesn't appear to be a clean way to disable hill climbing, added a config variable that disables it in case a workaround is necessary for some other reason in the future
- Fixed to avoid overflow in the math to what was probably intended
- There may be another bug in GetWaveComponent() that causes values of such high magnitude to be generated, I'll leave that investigation for when that in particular becomes a real issue
Carol Eidt [Thu, 19 Oct 2017 03:49:56 +0000 (20:49 -0700)]
Merge pull request #14573 from CarolEidt/Fix13751
Fix multiple issues in LSRA for TYP_DOUBLE:
Andy Ayers [Thu, 19 Oct 2017 02:42:52 +0000 (19:42 -0700)]
JIT: optimize stelem (ref) like we optimize stelem.ref (#14580)
In particular, avoid the array store check when storing null or values from
the same array.
Addresses first part of #14574.
Brian Sullivan [Thu, 19 Oct 2017 02:13:59 +0000 (19:13 -0700)]
Removed hnd1 and hnd2 arguments to gtNewIconEmb*HndNode methods
Brian Sullivan [Thu, 19 Oct 2017 01:58:20 +0000 (18:58 -0700)]
Clang format fixes
mikedn [Thu, 19 Oct 2017 01:34:55 +0000 (04:34 +0300)]
Ifdef out legacy uses of GT_ASG_op (#14384)
* Ifdef out legacy uses of GT_ASG_op
GT_ASG_op nodes are only generated when the legacy backend is used.
* Address feedback
* Cleanup gtOverflow/gtOverflowEx
Brian Sullivan [Thu, 19 Oct 2017 01:12:36 +0000 (18:12 -0700)]
Removed unused handle args
from gtNewIconHandleNode and gtNewIcomEmbHndNode
Stephen Toub [Thu, 19 Oct 2017 00:55:09 +0000 (20:55 -0400)]
Fix BinaryWriter/Reader span parameter names (#14577)
Per API review, they should match the corresponding overloads' existing parameter names.
Jan Kotas [Thu, 19 Oct 2017 00:54:18 +0000 (17:54 -0700)]
Allow both CORINFO_FLG_INTRINSIC and CORINFO_FLG_JIT_INTRINSIC to be set on potential intrinsics (#14566)
* Allow both CORINFO_FLG_INTRINSIC and CORINFO_FLG_JIT_INTRINSIC to be set on potential intrinsics
* Call getIntrinsicID for CORINFO_FLG_INTRINSIC only
Brian Sullivan [Thu, 19 Oct 2017 00:51:37 +0000 (17:51 -0700)]
Removed unused fields of Icon nodes
The union of gtIconHdl1, gtIconHdl2 and gtIconCPX, gtOIconCls is not used
Bruce Forstall [Wed, 18 Oct 2017 23:01:27 +0000 (16:01 -0700)]
Merge pull request #14572 from sdmaclea/PR-ARM64-SIMD-INSTRS
[Arm64] Add instructions needed by SIMD
Carol Eidt [Wed, 18 Oct 2017 21:31:05 +0000 (14:31 -0700)]
Fix multiple issues in LSRA for TYP_DOUBLE:
- updateAssignedInterval() needs to handle the case where the previously assigned interval was TYP_DOUBLE.
- Similarly, at block boundaries, if we are setting the register for a double interval, and the assigned interval is either null or is TYP_FLOAT ,we also need to unassign the other half of the register.
- LSRA only considers the valid double registers when allocating. When setting the candidates for the source of a return, it should only set the bit for the valid double register.
Fix #13751
Steve MacLean [Wed, 18 Oct 2017 18:02:19 +0000 (14:02 -0400)]
[Arm64] Support flag FEATURE_SIMD
Steve MacLean [Wed, 18 Oct 2017 18:25:50 +0000 (14:25 -0400)]
[Arm64] Add instructions needed by SIMD
William Godbe [Wed, 18 Oct 2017 20:15:02 +0000 (13:15 -0700)]
Merge pull request #14569 from wtgodbe/eqNe
Replace ne with eq in publish to azure steps
wtgodbe [Wed, 18 Oct 2017 18:03:13 +0000 (11:03 -0700)]
Replace ne with eq in publish to azure steps
Michelle McDaniel [Wed, 18 Oct 2017 17:41:31 +0000 (10:41 -0700)]
Merge pull request #14554 from adiaaida/reenablePGOLinux
Reenable PGO on Linux Release builds
Jan Kotas [Wed, 18 Oct 2017 09:38:55 +0000 (02:38 -0700)]
Improve thread statics performance (#14560)
- Disable code to handle multiple appdomains
- Use more efficient object array accessor
dotnet-maestro-bot [Wed, 18 Oct 2017 05:42:06 +0000 (00:42 -0500)]
Update BuildTools, CoreClr, CoreFx to prerelease-02118-01, preview1-25818-02, preview1-25818-01, respectively (#14558)
Carol Eidt [Wed, 18 Oct 2017 03:00:00 +0000 (20:00 -0700)]
Merge pull request #14547 from CarolEidt/Fix14539
Arm64: Use op2 type for LOCKADD
Sergey Andreenko [Wed, 18 Oct 2017 00:12:24 +0000 (17:12 -0700)]
format spmi sources. (#14545)
format spmi sources.
Brian Sullivan [Tue, 17 Oct 2017 23:34:37 +0000 (16:34 -0700)]
Merge pull request #14553 from briansull/fix-warn
Fix for x86 desktop build break
Carol Eidt [Tue, 17 Oct 2017 20:35:14 +0000 (13:35 -0700)]
Arm64: Use op2 type for LOCKADD
Fix #14539
William Godbe [Tue, 17 Oct 2017 22:11:19 +0000 (15:11 -0700)]
Merge pull request #14555 from wtgodbe/AzureBlobSymPkg
Use C#-friendly path for indexing symbol packages
wtgodbe [Tue, 17 Oct 2017 22:10:21 +0000 (15:10 -0700)]
Use C#-friendly path for indexing symbol packages
Michelle McDaniel [Tue, 17 Oct 2017 22:06:34 +0000 (15:06 -0700)]
Reenable PGO on Linux Release builds
Now that all the machines have been updated to have llvm 3.9 with PGO
support, reenable PGO builds.
Brian Sullivan [Tue, 17 Oct 2017 22:03:22 +0000 (15:03 -0700)]
Fix for x86 desktop build
Fix warning C4389: '==' : signed/unsigned mismatch
Stephen Toub [Tue, 17 Oct 2017 21:37:57 +0000 (17:37 -0400)]
Merge pull request #14541 from stephentoub/taskbeginwait
Avoid TaskWaitBegin/End events in Task.Wait on already completed task
Justin Van Patten [Tue, 17 Oct 2017 21:23:24 +0000 (14:23 -0700)]
GetEnvironmentVariable: Avoid StringBuilder marshaling overhead (#14502)
* GetEnvironmentVariable: Avoid StringBuilder marshaling overhead
Try a stack allocated buffer first, then fallback to using the shared
array pool.
Koundinya Veluri [Tue, 17 Oct 2017 20:52:52 +0000 (13:52 -0700)]
CLRLifoSemaphore cleanup (#14535)
- Removed volatile loads. They don't actually help with anything. Should help a bit on arm, haven't tested.
- I had added them initially out of pattern and could have left them out. There was some concern before that without a volatile load a compiler could replace local uses with a memory load and that would change the meaning of what is intended, but that doesn't apply to these kind of loops because it would be incorrect to do so, the concern may apply to loops like the following:
```c#
while(true)
{
Counts counts = m_counts;
if(InterlockedCompareExchange(&m_counts, counts + 1, counts) == counts)
break;
}
```
Where if the 3rd argument is replaced by the compiler with m_counts, it may yield an incorrect result. It's an invalid optimization but anyway that can't be done in the type of loops used in this code because the initial value used for subsequent loop iterations is the result of the compare-exchange operation and not the memory location.
- No need to decrement the count of waiters woken upon timeout. I had initially copied that part of the code from SemaphoreSlim, which needs it because it uses a Monitor that doesn't provide the same guarantees as the wait objects used here. It's not needed here.
- No change to perf on x64, I plan on testing arm as part of issue https://github.com/dotnet/coreclr/issues/14067 once I have some time and get a machine.
Stephen Toub [Tue, 17 Oct 2017 20:01:57 +0000 (16:01 -0400)]
Avoid TaskWaitBegin/End events in Task.Wait on already completed task
We are unnecessarily firing the TaskWaitBegin/End EventSource event when the task has already completed if it's faulted/canceled. We should only fire it if the task hasn't completed by the time we check.
Bruce Forstall [Tue, 17 Oct 2017 18:28:23 +0000 (11:28 -0700)]
Merge pull request #14493 from alpencolt/armel-cross-build
[RyuJIT/armel] cross build armel on x86 host
Bruce Forstall [Tue, 17 Oct 2017 18:21:15 +0000 (11:21 -0700)]
Merge pull request #14334 from BruceForstall/AddTestEnvToWindows
Fix corefx testing invocation
Sergey Andreenko [Tue, 17 Oct 2017 02:46:37 +0000 (19:46 -0700)]
move ReplaceWith to cpp
to make DEBUG_DESTROY_NODE visible.
Sergey Andreenko [Thu, 12 Oct 2017 01:28:35 +0000 (18:28 -0700)]
Refactor CopyFrom
Sergey Andreenko [Thu, 12 Oct 2017 00:55:25 +0000 (17:55 -0700)]
add check that each tree in the method has its own unique gtTreeId.
Sergey Andreenko [Wed, 11 Oct 2017 21:50:23 +0000 (14:50 -0700)]
extract fgDebugCheckStmtsList
Stephen Toub [Tue, 17 Oct 2017 16:48:45 +0000 (12:48 -0400)]
Merge pull request #14537 from dotnet-maestro-bot/master-UpdateDependencies
Update CoreFx to preview1-25817-02 (master)
Viktor Hofer [Tue, 17 Oct 2017 15:25:39 +0000 (17:25 +0200)]
Merge pull request #14311 from ViktorHofer/SerializationExceptions
Make coreclr exceptions serializable and add typeforwards
dotnet-maestro-bot [Tue, 17 Oct 2017 14:23:43 +0000 (07:23 -0700)]
Update CoreFx to preview1-25817-02
Sujin Kim [Tue, 17 Oct 2017 13:33:51 +0000 (22:33 +0900)]
[RyuJIT/ARM32] Add the optimization case on CreateDictionaryLookupHelper (#13933)
* Implement optimization case for CreateDictionaryLookupHelper
Signed-off-by: Hyung-Kyu Choi <hk0110.choi@samsung.com>
* Reenable mainv1/mainv2 tests
Alexander Soldatov [Fri, 13 Oct 2017 16:14:35 +0000 (19:14 +0300)]
[RyuJIT/armel] cross build armel on x86 host.
Compile ARM soft-fp version of libclrjit.so on x86 platform.
For running cross-jit execute as:
COMPlus_AltJit=* COMPlus_AltJitName=libprotononjit.so ./corerun ...
Sergey Andreenko [Tue, 17 Oct 2017 06:40:43 +0000 (23:40 -0700)]
protect strlen from nullptr (#14533)
Carol Eidt [Tue, 17 Oct 2017 05:07:15 +0000 (22:07 -0700)]
Merge pull request #14350 from CarolEidt/LsraInfoCleanup
Cleanup of Lowering & LsraInfo
Stephen Toub [Tue, 17 Oct 2017 02:22:35 +0000 (22:22 -0400)]
Merge pull request #14527 from stephentoub/timer_partition
Reduce Timer lock contention
Carol Eidt [Tue, 17 Oct 2017 02:06:01 +0000 (19:06 -0700)]
Refactor register selection heuristics
This is in preparation for tuning these for both throughput and code quality.
dotnet-maestro-bot [Tue, 17 Oct 2017 01:42:41 +0000 (20:42 -0500)]
Update BuildTools to prerelease-02116-01 (#14530)
Aditya Mandaleeka [Tue, 17 Oct 2017 01:36:58 +0000 (18:36 -0700)]
Merge pull request #14529 from adityamandaleeka/r2r_exception_race
Fix deadlock in ReadyToRun scenarios between EH and code heap deletion
Brian Sullivan [Tue, 17 Oct 2017 00:47:59 +0000 (17:47 -0700)]
Merge pull request #14501 from briansull/FitsIn
Use FitsIn8 when setting GT_INT_CNS costs
Sergey Andreenko [Tue, 17 Oct 2017 00:46:05 +0000 (17:46 -0700)]
Fix optEarlyPropRewriteTree (#14480)
Return ans use the right tree from optEarlyPropRewriteTree
Forbid large dst nodes.
* fix typo
Roman Artemev [Mon, 16 Oct 2017 23:20:45 +0000 (16:20 -0700)]
Fixed stack probing for RHEL 6.9 (Linux kernel 2.6.32) (#14481)
Fixed stack probing for RHEL 6.9 (Linux kernel 2.6.32) (#14481)
Brian Sullivan [Mon, 16 Oct 2017 22:33:37 +0000 (15:33 -0700)]
Updated GenTree::IsIntCnsFitsInI32() tio use FitsInI32()
Aditya Mandaleeka [Mon, 16 Oct 2017 21:58:50 +0000 (14:58 -0700)]
Fix deadlock caused by R2R method info lookups during code heap
deletion.
Noah Falk [Mon, 16 Oct 2017 22:18:43 +0000 (15:18 -0700)]
Fix stepping with tiered jitting
Fix #14426. Added a more flexible IL master breakpoint that can:
a) bind to native offset 0 of each jitted code body
b) use a MethodDescFilter so that only jitted code for one generic instance receives breakpoints
The remaining change is simply to switch step-in and trace stepping to use that new breakpoint binding behavior instead of the code version specific binding behavior that was used before. This ensures that even if the code version changes after the trace occurs we will still have a breakpoint on the alternate code versions and complete the stepping operation.
Brian Sullivan [Mon, 16 Oct 2017 22:17:36 +0000 (15:17 -0700)]
Merge branch 'master' into FitsIn
William Godbe [Mon, 16 Oct 2017 22:17:17 +0000 (15:17 -0700)]
Merge pull request #14522 from dotnet/wtgodbe-patch-1
Parameterize PublishFlatContainer in x86 build
Brian Sullivan [Mon, 16 Oct 2017 22:16:31 +0000 (15:16 -0700)]
Review feedback changes
Sergey Andreenko [Mon, 16 Oct 2017 22:07:23 +0000 (15:07 -0700)]
spill GT_RET_EXPR inside fat calli candidates. (#14520)
spill ret expr
Jan Kotas [Mon, 16 Oct 2017 22:00:33 +0000 (15:00 -0700)]
Delete dead code (#14521)
Carol Eidt [Mon, 16 Oct 2017 21:58:53 +0000 (14:58 -0700)]
PR Feedback
Jacek Blaszczynski [Mon, 16 Oct 2017 21:56:59 +0000 (23:56 +0200)]
Optimize native build of coreclr repo - 3x less memory , 11% build time improvement (#14509)
* Fix native build overparallelism
* Fix native build parallelization problems
* Allow overriding all build workaround parameters
* Fix cleanup ommisions and local environment variable naming
* Address code review feedback
* Fix nul spelling and add standard message prefix
* Simplify settings for CL compiler parallelism
* Address code review feedback - remove whitespace
Stephen Toub [Mon, 16 Oct 2017 21:36:09 +0000 (17:36 -0400)]
Address PR feedback
Viktor Hofer [Mon, 16 Oct 2017 18:29:40 +0000 (20:29 +0200)]
Adjusting field name changes in native code
Stephen Toub [Mon, 16 Oct 2017 19:16:31 +0000 (15:16 -0400)]
Change Task.Delay to use TimerQueueTimer instead of Timer
Timer is just a wrapper for a TimerHolder which is just a wrapper for a TimerQueueTimer. We don't actually want the behavior provided by TimerHolder (to allow the underlying timer to be collected while it's in use) and explicitly work around that, and for that pleasure we're paying two additional allocations. Just skip directly to the inner type.
Stephen Toub [Mon, 16 Oct 2017 19:09:52 +0000 (15:09 -0400)]
Queue TimerQueueTimer directly to ThreadPool
Eliminate a wrapper ThreadPoolWorkItem that's unnecessary. Saves an allocation ever time a Timer fires.
Stephen Toub [Mon, 16 Oct 2017 18:52:31 +0000 (14:52 -0400)]
Reduce Timer lock contention
Timer currently uses a global lock to protect a single "queue" of Timers, and any operation to create/change/fire/delete a Timer takes that same lock. This leads to a scalability problem for code that operates on lots of timers.
This change partitions this single queue into N queues, which all operate independently. In the .NET Framework implementation, there's already logically such a split with one queue/lock per AppDomain, and this change utilizes the same underlying support in the runtime. As such, contention is distributed across the N locks, helping scalability.
Brian Sullivan [Mon, 16 Oct 2017 18:11:20 +0000 (11:11 -0700)]
Merge branch 'master' into FitsIn
William Godbe [Mon, 16 Oct 2017 17:59:31 +0000 (10:59 -0700)]
Parameterize PublishFlatContainer in x86 build
This was missed in my last change to parameterize PublishFlatContainer across the board, and is what was causing the failures in today's pipe builds. We only want to publish non-flat for Release builds, but were hardcoding PublishFlatContainer=true for all builds in the x86 definition.
dotnet-maestro-bot [Mon, 16 Oct 2017 17:13:57 +0000 (12:13 -0500)]
Update CoreFx to preview1-25816-01 (#14516)
Michelle McDaniel [Mon, 16 Oct 2017 16:39:25 +0000 (09:39 -0700)]
Merge pull request #14519 from adiaaida/fixThroughputScript
Fix call to submission.py in run-throughput-perf.py
Michelle McDaniel [Mon, 16 Oct 2017 16:03:42 +0000 (09:03 -0700)]
Fix call to submission.py in run-throughput-perf.py
In run-throughput-perf.py, we were calling submission.py with --arch,
but submission.py requires -arch or --architecture. This recently made
submission.py start failing in throughput jobs.
This change also fixes run-throughput-perf so that if submission.py or
upload.py fail, the script fails so we see these jobs as failing in the
lab.
Sergey Andreenko [Mon, 16 Oct 2017 15:47:46 +0000 (08:47 -0700)]
Forbid to clone GT_RET_EXPR (#14495)
* Forbid to clone GT_RET_EXPR
It potentially creates non-unique trees.
Justin Van Patten [Mon, 16 Oct 2017 05:49:53 +0000 (22:49 -0700)]
Environment.SystemDirectory: Avoid StringBuilder overhead (#14513)
* Environment.SystemDirectory: Avoid StringBuilder overhead
Use a stack allocated buffer, with fallback to a char array.
dotnet-maestro-bot [Sun, 15 Oct 2017 22:21:29 +0000 (17:21 -0500)]
Update CoreFx to preview1-25815-02 (#14511)
dotnet bot [Sat, 14 Oct 2017 20:08:26 +0000 (13:08 -0700)]
Delete contract annotations (dotnet/corert#4722) (#14507)
Signed-off-by: dotnet-bot <dotnet-bot@microsoft.com>
dotnet-maestro-bot [Sat, 14 Oct 2017 16:27:07 +0000 (11:27 -0500)]
Update BuildTools, CoreFx, PgoData to prerelease-02113-01, preview1-25814-02, master-
20171012-0049, respectively (#14476)
Justin Van Patten [Sat, 14 Oct 2017 14:39:37 +0000 (07:39 -0700)]
Delete EnvironmentAugments.GetEnvironmentVariables (#14500)
CoreFX has been using EnumerateEnvironmentVariables for a while now.
Justin Van Patten [Sat, 14 Oct 2017 14:39:02 +0000 (07:39 -0700)]
Remove temp unsafe stackalloc workarounds for Span (#14503)
Justin Van Patten [Sat, 14 Oct 2017 14:38:46 +0000 (07:38 -0700)]
Delete unnecessary Debug.Assert (#14504)
Swaroop Sridhar [Mon, 9 Oct 2017 19:42:12 +0000 (12:42 -0700)]
Fix ILLink perf benchmark
This commit implements the following two changes to fix the
failure in ILLink perf test.
1) Fix Roslyn benchmark
Use ILLink.Tasks package to build the Roslyn linked benchmark,
now that Roslyn can target netcoreapp2.0.
2) Disable Musicstore temporarily
Disabling MusicStore build temporarily, since there's a build failure.
Brian Sullivan [Sat, 14 Oct 2017 03:02:43 +0000 (20:02 -0700)]
Merge pull request #14465 from sdmaclea/PR-ARM64-DUP-DV_2D
[ARM64] Fix INS_dup DV_2D encoding
Stephen Toub [Sat, 14 Oct 2017 01:52:26 +0000 (21:52 -0400)]
Add EventSource event for dropped async state machines (#14497)
One of the main bugs we've seen with async methods in production are cases where an async method awaits something that's never completed and is then just dropped. This in turn allows the async state machine to be collected, such that subsequent dumps have no usable information to determine what hung and where.
To help diagnose such issues, this commit adds an EventSource event that fires when an async state machine is collected without having been completed. When that happens, it dumps a textual representation of itself into a string and sends that to an event.
Doing this depends on the recent changes that were made to how we box async state machines to the heap. Now that we do so with an explicit strongly-typed "box" instead of a normal runtime box, when the event source event is enabled, we can instead instantiate a derived and finalizable box, where the finalizer is responsible for firing the event, and we then suppress finalization when the async method completes. The only async method state machines that are then finalized are those that are allowed to be collected before they're completed.
Brian Sullivan [Fri, 13 Oct 2017 23:22:18 +0000 (16:22 -0700)]
Use IsIntCnsFitsIn8
When classifying the size of integer constants use new method FitsInI8() and existing method FitsInI32
José Rivero [Sat, 14 Oct 2017 00:42:55 +0000 (17:42 -0700)]
Disabling legacy banckend perf CQ and scenario runs (#14468)
- Recent changes to enforce data validation uncovered bug on the legacy backend runs. Now, we do not allow to upload tests with no performance data.
CQ run bug tracked by: #14463
JitBench run bug tracked by: #14477
Jan Kotas [Fri, 13 Oct 2017 20:25:47 +0000 (13:25 -0700)]
Merge pull request dotnet/corert#4719 from dotnet/nmirror
Merge nmirror to master
Signed-off-by: dotnet-bot <dotnet-bot@microsoft.com>
Andy Ayers [Fri, 13 Oct 2017 23:05:44 +0000 (16:05 -0700)]
Updates to the jit bench harness (#14467)
Updates to the jit bench harness
Introduce a number of new options:
* use-existing-setup to reuse a previously set up environment
* iterations to control number of iterations
* tiering to enable tiered jit
* disable-r2r to disable loading of R2R images
* disable-ngen to disable loading of NGEN images
* minopts to force jit to generate minopts code
Also start tracking average response time. This is not yet "uploaded"
since the underlying data reported by the app needs some updates.
Andy Ayers [Fri, 13 Oct 2017 22:55:06 +0000 (15:55 -0700)]
Optimize type casts (#14420)
JIT: optimize type casts
Implement the jit interface compareTypesForEquality method
to handle casts from known types to known types, and from
shared types to certain interface types.
Call this method in the jit for castclass and isinst, using
`gtGetClassHandle` to obtain the from type. Optimize sucessful
casts and unsuccessful isinsts when the from type is known
exactly.
Undo part of the type-equality based optimization/workaround
in the AsyncMethodBuilder code that was introduced in #14178
in favor of interface checks. There is more here that can
be done here before this issue is entirely closed and I will
look at this subsequently.
This optimization allows the jit to remove boxes that are
used solely to feed type casts, and so closes #12877.
Carol Eidt [Fri, 13 Oct 2017 20:35:42 +0000 (13:35 -0700)]
Fix handling of SIMD12
It turns out that it is problematic to change the type of a SIMD12 PUTARG_STK operand. Instead, at codegen time, use the number of slots on the PUTARG_STK to identify the SIMD12 case.
Also, add some tests extracted from the corefx System.Numerics.Vectors tests.
William Godbe [Fri, 13 Oct 2017 19:55:42 +0000 (12:55 -0700)]
Merge pull request #14498 from wtgodbe/CloudDropAccountName
Unify CloudDropAccountName
wtgodbe [Fri, 13 Oct 2017 19:55:22 +0000 (12:55 -0700)]
Unify CloudDropAccountName
William Godbe [Fri, 13 Oct 2017 18:45:45 +0000 (11:45 -0700)]
Merge pull request #14496 from wtgodbe/WorkingDir
Correctly set working folder in publish
wtgodbe [Fri, 13 Oct 2017 18:45:10 +0000 (11:45 -0700)]
Correctly set working folder in publish
Hanjoung Lee [Fri, 13 Oct 2017 14:18:14 +0000 (23:18 +0900)]
Fix compile error with crossgen CreatePDB on Unix (#14488)
- Fix wchar string literals
- Fix init order of class fields