Stephen Toub [Tue, 4 Dec 2018 01:39:02 +0000 (20:39 -0500)]
Finish (mostly) erradicating StringBuilder marshaling from corefx (#33780)
* Finish (mostly) erradicating StringBuilder marshaling from corefx
This should finish my quest of removing StringBuilder marshaling from coreclr and corefx, which I've strived to do because it often adds unnecessary overhead and often is done incorrectly because of intricacies in how the marshaling works; while not using it often leads to `unsafe` code at call sites, there's much less magic, and it affords the ability to optimize more if desired.
There are still three remaining [Out] StringBuilder's in Interop.Odbc.cs, which I've not removed because tests are limited and the call graph to these is non-trivial with StringBuilders passed through. There may also be a few straggling DllImports I missed while scouring interop that's not following our guidelines of being in src\Common\.
Other than those, the remaining StringBuilder usage I found in DllImports was for [In] only, and in those cases it's reasonable, as the call sites are building up StringBuilders and then passing them off to the call sites; converting those to use `char*` or similar could actually make them more expensive. I did, however, ensure they were properly annotated as [In], in order to make the intent clear and avoid potential marshaling costs for the unnecessary [Out] direction.
Where I was touching a function in one of these stray interop files that was duplicated elsewhere, I also consolidated it to a centralized location, but I've not in this PR done the cleanup work for the rest of the files.
* Address PR feedback
* Address further feedback
Signed-off-by: dotnet-bot <dotnet-bot@microsoft.com>
Jan Kotas [Tue, 4 Dec 2018 00:11:29 +0000 (16:11 -0800)]
Refactor all FCalls out of AppDomain.cs (#21337)
This saves the unmanaged->managed->unmanaged trip to initialize the assembly binder.
Includes small bits of unrelated cleanup.
Egor Bogatov [Mon, 3 Dec 2018 22:56:38 +0000 (01:56 +0300)]
Vectorize some Guid APIs (ctor, TryWriteBytes) (#21336)
* Optimize some Guid APIs
* get rid of WriteByteHelper
* use TryWrite instead of Write
* Optimize ctor `Guid(ReadOnlySpan<byte> b)` and remove `Equals` optimization (move to separate PR).
Egor Chesakov [Mon, 3 Dec 2018 22:51:28 +0000 (14:51 -0800)]
Remove Hostx86/arm CrossGen from building and Simplify build.sh command line arguments for Linux/arm cross build (Part 2) (#21034)
* Stop building and publishing Hostx86/arm crossgen on Linux/arm
* Remove -crosscomponent argument and stop using CAC_ROOTFS_DIR environment variable in build.sh
* Simplify the related logic in build.sh
* Don't need to specify crosscomponent in tests/scripts/run-pmi-diffs.py
* Don't set CAC_ROOTFS_DIR in buildpipeline, Jenkins files and in tests/scripts/run-pmi-diffs.py
* Adjust documentation
Carol Eidt [Mon, 3 Dec 2018 20:52:20 +0000 (12:52 -0800)]
Merge pull request #21264 from fiigii/x64only
Implement 64-bit-only hardware intrinsic
Jarret Shook [Mon, 3 Dec 2018 20:46:35 +0000 (12:46 -0800)]
Correct platorm reach for the Official build (#21310)
* Correct platorm reach for the Official build
Does the following:
1. Sets up a ci rule for master and a pr rule for master
2. Corrects the OSX queues
3. Corrects ubuntu internal queue
4. Converts all linux jobs to build using containers
5. Only runs official builds on: Pri1, release, bringing us to parity with old process
6. Fixes centos build-test
7. Adds a super-annoying groupname tag that avoids name mangling.
* make sure we only add triggers for internal builds
* Fix syntax
* Missing :
* Remove pr/ci
* Fix internal queue name
* Correct Public/internal choices.
* Add crossgen as well
* Correct indentation
* Correct checked pri1 tests
* Address pr feedback
* Add todo comments
Rui [Mon, 3 Dec 2018 20:33:51 +0000 (15:33 -0500)]
Clarify use of LoadFromAssemblyName method (#21256)
Tomáš Rylek [Mon, 3 Dec 2018 19:33:49 +0000 (20:33 +0100)]
Improvements for naked R2RDump output (#21317)
The initial impulse for this change was removing native offsets
from debug info that caused noise in naked R2R comparisons. Fixing
this required propagation of DumpOptions to some more call sites
prompting me to unify dumping logic to always use TextWriter
as the output medium. I have also fixed some minor bugs I noticed
around the naked disassembly.
Thanks
Tomas
dotnet-maestro-bot [Mon, 3 Dec 2018 18:53:23 +0000 (10:53 -0800)]
Update BuildTools, CoreClr, CoreFx, CoreSetup to preview1-03430-01, preview-27203-03, preview.18603.1, preview-27203-01, respectively (#21291)
Andy Ayers [Mon, 3 Dec 2018 18:46:52 +0000 (10:46 -0800)]
JIT: fix overly aggressive type propagation from returns (#21316)
For quite a while now the jit has been propagating return types from
callees to the return spill temp. However this is only safe when the
callee has a single return site (or all return sites return the same
type).
Because return spill temps often end up getting assigned to still more
temps we haven't seen this overly aggressive type propgagation lead to
bugs, but now that we're tracking single def temps and doing more type
propagation during the late devirtualization callback, the fact that
these types are wrong has been exposed and can lead to incorrect
devirtualization.
The fix is to only consider the return spill temp as single def if the
callee has a single return site, and to check that the return spill temp
is single def before trying to propagate the type.
Fixes #21295.
Andrew Au [Mon, 3 Dec 2018 18:17:20 +0000 (10:17 -0800)]
Make sure the debug registers are available in the SavedRedirectContext (#21293)
Steve MacLean [Mon, 3 Dec 2018 17:49:31 +0000 (12:49 -0500)]
Remove IsNeutralDomain() (#21318)
* Remove IsNeutralDomain()
* PR feedback
Rafael Lillo [Mon, 3 Dec 2018 17:19:14 +0000 (15:19 -0200)]
Fixes abbreviated genitive (#21230)
* fixes abbreviation
* reverse .gitignore
* remove token lenght
Ilia [Mon, 3 Dec 2018 15:25:24 +0000 (18:25 +0300)]
Don't typedef intptr_t/uintptr_t when PAL_STDCPP_COMPAT is defined (#21165)
Bug: 20766
Michal Strehovský [Mon, 3 Dec 2018 08:15:26 +0000 (09:15 +0100)]
Make RyuJIT tolerate null CLASSID_RUNTIME_TYPE (#21331)
If we're compiling a class library that has no concept of reflection (and runtime types), EE cannot provide a class handle for this class. Make RyuJIT tolerant to null coming back.
Ben Adams [Mon, 3 Dec 2018 06:46:26 +0000 (06:46 +0000)]
Store CurrentThread in ThreadStatic (#21328)
Russ Keldorph [Mon, 3 Dec 2018 00:40:18 +0000 (16:40 -0800)]
Revert [publish symbols on FreeBSD (#21179)] (#21329)
This reverts commit
6120021023647ff881d791a033dba38a94fd1c2c.
Jan Kotas [Sat, 1 Dec 2018 21:09:21 +0000 (13:09 -0800)]
Rename CorElementType enum members to match the others definitions (#21324)
Maoni Stephens [Sat, 1 Dec 2018 20:38:36 +0000 (12:38 -0800)]
added info about the new .NET memory book (#21326)
dotnet-maestro[bot] [Sat, 1 Dec 2018 14:47:54 +0000 (14:47 +0000)]
Update dependencies from https://github.com/dotnet/arcade build 526 (#21321)
This change updates the following dependencies
- Microsoft.DotNet.Arcade.Sdk - 1.0.0-beta.18580.13
- Microsoft.DotNet.Helix.Sdk - 1.0.0-beta.18580.13
Jan Vorlicek [Sat, 1 Dec 2018 09:31:35 +0000 (10:31 +0100)]
Enable COM interop for collectible classes (#20919)
* Enable COM interop for collectible classes
* Modify DispatchInfo to use LoaderAllocator handles
The DispatchMemberInfo was using global handles to refer to the managed
MemberInfo instances. That doesn't work with unloadability.
This change modifies it to use handles allocated from LoaderAllocator.
* Disable COM interop for WinRT types
* Remove collectible check from IsTypeVisibleFromCom. That fixes three
new COM interop tests
* Add collectible check to GetComClassFactory when we check for
unsupported interop with WinRT
* Add COM unloadability tests
Add two tests to test COM unloadability:
* One for using native COM server from managed COM client
* One for using managed COM objects from native client
* Add unloading test for IUnknownTest
* Disable NETClientPrimitivesInALC on Win ARM
The NETClientPrimitives is disabled there too.
Aaron Robinson [Sat, 1 Dec 2018 03:30:37 +0000 (19:30 -0800)]
Change target 'Target' and position in build steps for response file generation (#21315)
Fei Peng [Thu, 29 Nov 2018 00:25:48 +0000 (16:25 -0800)]
Migrate 64-bit-only intrinsic tests to new APIs
Fei Peng [Thu, 15 Nov 2018 23:00:26 +0000 (15:00 -0800)]
Migrate FormattingHelpers.CountDigits to using Lzcnt.X64
Fei Peng [Thu, 29 Nov 2018 00:25:14 +0000 (16:25 -0800)]
Implement 64-bit-only intrinsic
Steve MacLean [Fri, 30 Nov 2018 23:15:26 +0000 (18:15 -0500)]
Fix crossgen segfault
Steve MacLean [Fri, 30 Nov 2018 22:18:46 +0000 (17:18 -0500)]
Address PR feedback
Steve MacLean [Fri, 30 Nov 2018 03:14:11 +0000 (22:14 -0500)]
Remove dead code
Carol Eidt [Fri, 30 Nov 2018 23:30:03 +0000 (15:30 -0800)]
Merge pull request #21311 from CarolEidt/FixLvaUpdateClassDump
Fix dump in LvaUpdateClass
Aaron Robinson [Fri, 30 Nov 2018 23:01:04 +0000 (15:01 -0800)]
Reenable TypeEquivalence tests (#21309)
Fei Peng [Thu, 15 Nov 2018 22:44:55 +0000 (14:44 -0800)]
Remove legacy 64-bit intrinsic APIs
Andrew Au [Fri, 30 Nov 2018 21:40:44 +0000 (13:40 -0800)]
Eliminate the requirement that abort requires no lock held on the current thread (#21206)
Carol Eidt [Fri, 30 Nov 2018 20:42:15 +0000 (12:42 -0800)]
Fix dump in LvaUpdateClass
It was failing, presumably because of too many or too large args, so break it up a abit.
Aaron Robinson [Fri, 30 Nov 2018 18:06:28 +0000 (10:06 -0800)]
Update how CoreDisTools is discovered (#21261)
* Update Disassembler logic to search for CoreDisTools next to binary _or_ under the path specified by CORE_ROOT
* Update GCCover critical section
Stephen Toub [Fri, 30 Nov 2018 15:16:25 +0000 (10:16 -0500)]
A few more Enum perf improvements (#21284)
* A few more Enum perf improvements
I started out to just remove the unnecessary boxing allocation that happens when you do enum.ToString("D") or, if the value doesn't map to a named enum entry, just enum.ToString() (there's an internal boxing of the value that happens in addition to any boxing that happens of the enum itself). As part of that, I added a ValueToString method that replaces the GetValue().ToString() calls that appear several times, and in writing that ValueToString, I opted to avoid pinning, instead using Unsafe.As. Once I did that, I then removed pinning everywhere else it was being done, standardizing on Unsafe.As.
* Address PR feedback
dotnet-maestro[bot] [Fri, 30 Nov 2018 14:48:20 +0000 (14:48 +0000)]
Update dependency files (#21211)
* Update dependencies from https://github.com/dotnet/arcade build 459
This change updates the following dependencies
- Microsoft.DotNet.Arcade.Sdk - 1.0.0-beta.18576.5
- Microsoft.DotNet.Helix.Sdk - 1.0.0-beta.18576.5
* Update dependencies from https://github.com/dotnet/arcade build 491
This change updates the following dependencies
- Microsoft.DotNet.Arcade.Sdk - 1.0.0-beta.18578.9
- Microsoft.DotNet.Helix.Sdk - 1.0.0-beta.18578.9
* Update dependencies from https://github.com/dotnet/arcade build 508
This change updates the following dependencies
- Microsoft.DotNet.Arcade.Sdk - 1.0.0-beta.18579.9
- Microsoft.DotNet.Helix.Sdk - 1.0.0-beta.18579.9
David Mason [Fri, 30 Nov 2018 09:25:57 +0000 (01:25 -0800)]
Fix perf issues with GetCurrentThreadHomeHeapNumber (#21150)
Aaron Robinson [Fri, 30 Nov 2018 06:51:08 +0000 (22:51 -0800)]
Disable TypeEquivalence tests due to flaky x86 build (#21294)
* Remove includes of SDK imports from prop and targets files
* Disable TypeEquivalence test due to failure on x86
Jan Kotas [Fri, 30 Nov 2018 02:39:36 +0000 (18:39 -0800)]
Undo accidental change
Levi Broderick [Mon, 19 Nov 2018 23:07:08 +0000 (15:07 -0800)]
Add EnumerateRunes() ref APIs and unit tests (dotnet/corefx#33504)
Signed-off-by: dotnet-bot <dotnet-bot@microsoft.com>
Jan Kotas [Sat, 24 Nov 2018 19:53:33 +0000 (11:53 -0800)]
Delete CORECLR ifdef (dotnet/corert#6610)
This ifdef should not be needed anymore.
Signed-off-by: dotnet-bot <dotnet-bot@microsoft.com>
John Doe [Thu, 22 Nov 2018 22:23:08 +0000 (14:23 -0800)]
Typos (dotnet/corert#6605)
* accross -> across
* adress -> address
* Appartment -> Apartment
* persective -> perspective
* paramaterized -> parameterized
* Langage -> Language
* miminum -> minimum
* Curret -> Current
* currnet -> current
* definately -> definitely
Signed-off-by: dotnet-bot <dotnet-bot@microsoft.com>
Michal Strehovský [Thu, 22 Nov 2018 16:01:17 +0000 (17:01 +0100)]
Properly report intrinsic fields (dotnet/corert#6602)
* Properly report intrinsic fields
We were hitting an assert in RyuJIT because we were reporting IntPtr.Zero as an intrinsic even if we were not trying to `CORINFO_ACCESS_GET`.
I fixed that part and also went ahead to port the rest of getFieldIntrinsic from CoreCLR. We'll need this anyway.
* Add intrinsic fields
Signed-off-by: dotnet-bot <dotnet-bot@microsoft.com>
Juan Hoyos [Fri, 30 Nov 2018 02:19:08 +0000 (18:19 -0800)]
Merge pull request #21287 from hoyosjs/fix-tls-daccast
Properly DACize TLB access for thread local statics
Alexej Liebenthal [Fri, 30 Nov 2018 00:58:38 +0000 (01:58 +0100)]
Added TryParse to StandardFormat (#21216)
Brian Robbins [Fri, 30 Nov 2018 00:39:41 +0000 (16:39 -0800)]
Defer Initialization of FrameworkEventSource During EventPipeController Initialization (#21255)
dotnet-maestro-bot [Thu, 29 Nov 2018 23:09:56 +0000 (15:09 -0800)]
Update CoreClr, CoreFx, CoreSetup, PgoData to preview-27129-03, preview.18579.2, preview-27129-02, master-
20181124-0042, respectively (#21178)
Juan Sebastian Hoyos Ayala [Thu, 29 Nov 2018 23:06:21 +0000 (15:06 -0800)]
Properly DACize TLB access for thread local statics
Fei Peng [Thu, 29 Nov 2018 23:02:24 +0000 (15:02 -0800)]
Add SuperPMI code for the new getMethodNameFromMetadata
Fei Peng [Wed, 21 Nov 2018 01:05:25 +0000 (17:05 -0800)]
Update JIT/EE getMethodNameFromMetadata to get enclosing class name
Stephen Toub [Thu, 29 Nov 2018 21:01:02 +0000 (16:01 -0500)]
Consolidate validation code in Enum.ToObject overloads (#21280)
Unnecessary duplication. Moving it into a shared method has no measurable impact on ToObject performance, which is dominated by the boxing costs.
Stephen Toub [Thu, 29 Nov 2018 20:59:30 +0000 (15:59 -0500)]
Improve performance of Enum.{Try}Parse (#21214)
* Improve performance of Enum.{Try}Parse
In particular for the generic overloads, where this avoids all per-operation allocations when commonly used underlying types are used (for example, this improves when the underlying enum type is int but not double).
* Address PR feedback
Aaron Robinson [Thu, 29 Nov 2018 20:44:04 +0000 (12:44 -0800)]
Enable type equivalence (#21265)
* Enable TypeEquivalence feature for Windows platform
* Basic test - verified test exercises TypeEquivalence code paths
Tanner Gooding [Thu, 29 Nov 2018 20:15:43 +0000 (12:15 -0800)]
Implement the Count property, the IEquatable and IFormattable interfaces, and the standard object overrides on Vector64/128/256<T> (#21259)
* Exposing the Count property on Vector64/128/256<T>
* Implementing IEquatable, IFormattable, and the object overrides on Vector64/128/256<T>
* Use StringBuilderCache, HashCode, and the provided format specifier
* Use StringBuilderCache.GetStringAndRelease and get the correct element when formatting Vector64/128/256<T>
Jeremy Koritzinsky [Thu, 29 Nov 2018 19:57:36 +0000 (11:57 -0800)]
Add more advanced SafeArray tests (#21063)
* Port simple SafeArray tests over from .NET Framework's test suite.
* Clean up native calling conventions.
* Add basic field array -> field SAFEARRAY* tests.
* Fix some hiding stdcall calling convention decorators.
* PR Feedback.
* Rewrite native part of SafeArray tests
* Wrote the managed side of the tests. Non-passing ones are currently commented out.
* Fix up tests and enable marshalling a string[] as a SAFEARRAY of LPSTR or LPWSTR.
* Add blittable records test with a return parameter.
* Add returning non-blittable record safearray test.
* Fix which typedef we're using
* Clean up
* Fix/cleanup RecordNative
* Replace "" with <> for platformdefines.h include.
* Use macro for calling convention.
* PR Feedback.
* Remove residual "Helpers.h" include.
* Fix remaining incorrect calling conventions.
* One more calling convention fix.
Jan Vorlicek [Thu, 29 Nov 2018 18:06:34 +0000 (19:06 +0100)]
Fix build parallelism on Windows with NUMA (#21278)
When multiple NUMA nodes are enabled on the machine where coreclr is
built, we incorrectly detect the number of cores that we use for build
parallelism of the native part of the build (NumberOfCores) as only
a number in the last NUMA node.
The reason is that the `wmic cpu get NumberOfCores /value` returns
multiple lines, one per each NUMA node.
This change fixes it by summing values from all the lines.
Tanner Gooding [Thu, 29 Nov 2018 17:46:34 +0000 (09:46 -0800)]
Don't treat the Vector64/128/256<T> helper methods as intrinsic if featureSIMD is disabled (#21274)
Bruce Forstall [Thu, 29 Nov 2018 17:46:18 +0000 (09:46 -0800)]
Merge pull request #21267 from BruceForstall/FixExclusion
Fix ObjectStackAllocationTests on Linux under minopts
Tomas Weinfurt [Thu, 29 Nov 2018 17:31:46 +0000 (09:31 -0800)]
publish symbols on FreeBSD (#21179)
* enable build-id on FreeBSD
* start publishing symbols for freebsd
* enable FEATURE_PERFMAP on FreeBSD
Tanner Gooding [Thu, 29 Nov 2018 15:59:27 +0000 (07:59 -0800)]
Updating genSIMDZero to only use `xorps` (#21249)
Andy Ayers [Thu, 29 Nov 2018 08:32:55 +0000 (00:32 -0800)]
Track single def locals in importer (#21251)
* Defer setting lvSingleDef until lvaMarkLocalVars
Since this information isn't used early on in the jit, defer setting this bit
until later.
* Track single-def locals in the importer
and use this information to enable type updates during late devirtualization.
This gets some cases where the exact type materializes late (perhaps via a
chain of copies).
Swaroop Sridhar [Thu, 29 Nov 2018 07:09:30 +0000 (23:09 -0800)]
Fix #21209 (#21237)
This commit has two changes to Native Load Library Tests fix the failures in:
https://github.com/dotnet/coreclr/issues/21209
1) Windows: The test tried to load a DLL from Win32 directory which doesn't exist
in Windows server 2016. So changed to use a file that exists.
2) Linux: Disable the test when run against packages -- because in this configuration
the tests are built on Windows, and the native DLL is built separately and copied
over to the CORE_ROOT directory instead of the test directory.
[Several other tests are disabled with the comment "Issue building native DLL" on Linux]
John Salem [Thu, 29 Nov 2018 06:34:48 +0000 (22:34 -0800)]
Add IsCollectible property to Memberinfo and MethodInfo (#21155)
Stephen Toub [Thu, 29 Nov 2018 04:04:08 +0000 (23:04 -0500)]
Improve Enum.ToString perf for [Flags] enums (#21254)
Two main changes:
- Rather than using a StringBuilder to insert the strings for all of the consistuent values, we track the constituent values in a span, summing the expected length, and then new up a string and copy the results directly into it.
- When there's a single value that matches the supplied value, we just return the cached string rather than allocating a new one.
Andrew Au [Thu, 29 Nov 2018 02:23:00 +0000 (18:23 -0800)]
Remove obsolete assert (#21260)
Jan Kotas [Thu, 29 Nov 2018 02:18:04 +0000 (18:18 -0800)]
Delete code related to LoaderOptimization and SharedDomain (#21031)
Bruce Forstall [Thu, 29 Nov 2018 01:16:23 +0000 (17:16 -0800)]
Fix ObjectStackAllocationTests on Linux under minopts
Fix the capitalization of COMPlus_JITMinOpts variable, which
is case-sensitive on Linux.
Fixes #21266
Ben Adams [Wed, 28 Nov 2018 21:45:16 +0000 (21:45 +0000)]
Queue ValueTaskAwaiter IAsyncStateMachineBox directly to ThreadPool (#21159)
* Queue ValueTaskAwaiter IAsyncStateMachineBox directly to ThreadPool
* Invert the dependency
* Move to UnsafeQueueUserWorkItem
* MRVTSC queue null or Deafult EC to UnsafeQUWI
* Revert MRVTSC change
* Add comment and validation
* Use s_invokeAsyncStateMachineBox for AsTask
* nits
* nits 2
* Rever ValueTask
* nits
Russ Keldorph [Wed, 28 Nov 2018 18:43:50 +0000 (10:43 -0800)]
Update Helix queues used for testing
Andy Ayers [Wed, 28 Nov 2018 20:51:46 +0000 (12:51 -0800)]
Merge pull request #21217 from AndyAyersMS/Fix21051
JIT: retype some return expressions
Andy Ayers [Wed, 28 Nov 2018 20:51:04 +0000 (12:51 -0800)]
Merge pull request #21226 from AndyAyersMS/HandleMoreGeneralIndirInJmpAddr
JIT: handle general indir case for GT_JMP address
Jan Kotas [Wed, 28 Nov 2018 19:19:48 +0000 (11:19 -0800)]
Switch to preemptive mode before calling COM (#21238)
* Switch to preemptive mode before calling COM
COM can call back the runtime via RCW that leads to contract violations
Fixes #20373
* Fix one places
Jan Kotas [Wed, 28 Nov 2018 19:18:54 +0000 (11:18 -0800)]
Delete test dev10_630880 (#21239)
* Revert "Disable test Dev10_630880"
This reverts commit
4fb76de35184fb80054f29779f51af82a4aa7ad8.
* Delete test dev10_630880
This test has been exercising internal CoreLib details that have been deleted.
Brian Sullivan [Wed, 28 Nov 2018 18:17:45 +0000 (10:17 -0800)]
Merge pull request #21204 from briansull/issue_20185
Change VNUnpackExc to always write an exception set
Sergey Andreenko [Wed, 28 Nov 2018 17:36:09 +0000 (09:36 -0800)]
exclude tests\src\Interop\COM\NativeClients\Primitives from GCStress runs. (#21222)
* exclude tests\src\Interop\COM\NativeClients\Primitives.csproj from GCStress runs.
* fix alignment
Tanner Gooding [Wed, 28 Nov 2018 11:54:12 +0000 (03:54 -0800)]
Fixing up Utf8Parser.TryParseNumber to not fail for overly large exponents (#21233)
Tanner Gooding [Wed, 28 Nov 2018 11:52:43 +0000 (03:52 -0800)]
Implement the S.R.I.VectorXXX `get_Zero` and `As` helper methods as JIT intrinsics (#21198)
* Remove ARM64_SIMD_StaticCast intrinsic and the x86 TwoTypeGeneric flag
* Implementing the `S.R.Intrinsic.VectorXXX.As` methods as runtime intrinsics
* Implementing the get_Zero method on the S.R.Intrinsic.VectorXXX types as runtime intrinsics
Jan Vorlicek [Wed, 28 Nov 2018 08:55:35 +0000 (09:55 +0100)]
Improve performance of cgroup access (#21229)
Currently, we create a CGroup instance on each request for getting
used or total physical memory. This has an extra cost of finding
filesystem paths of the current cgroup files. I have found that
if we do the initialization just once, the performance of getting
the used or total memory in a tight loop improves 22 times.
Accidentally, I was also looking into a perf regression in the
ByteMark.BenchBitOps test that was observed in the past, seemingly
related to the recent change to the way we get the used memory.
And I've found that the benchmark results improve two fold with
the change in this commit.
This change was made both in PAL and in the standalone GC.
Jan Vorlicek [Wed, 28 Nov 2018 05:36:59 +0000 (06:36 +0100)]
Fix ILStubCache allocation for collectible assemblies (#21188)
The ILStubCache was being allocated per domain unless the domain was a
compilation AppDomain. This is wrong for collectible assemblies, since
after an assembly is collected, the cache keeps stale entries referring
to already deleted MethodTables.
The fix is to make ILStubChange per LoaderAllocator instead (and keep
the per module instances for compilation AppDomain).
Bruce Forstall [Wed, 28 Nov 2018 03:39:04 +0000 (19:39 -0800)]
Merge pull request #21234 from BruceForstall/Disable630880
Disable test Dev10_630880
Ilia [Wed, 28 Nov 2018 01:34:29 +0000 (04:34 +0300)]
Move all the math functions undef #ifndef PAL_STDCPP_COMPAT (#21170)
Bug: 20786
Bruce Forstall [Wed, 28 Nov 2018 01:04:36 +0000 (17:04 -0800)]
Disable test Dev10_630880
It was made obsolete by https://github.com/dotnet/coreclr/pull/21157
Brian Sullivan [Wed, 28 Nov 2018 00:40:58 +0000 (16:40 -0800)]
Code review feedback
Bruce Forstall [Wed, 28 Nov 2018 00:33:25 +0000 (16:33 -0800)]
Disable failing Linux/arm32 corefx tests (#21225)
* Disable failing Linux/arm32 corefx tests
Dan Moseley [Tue, 27 Nov 2018 21:24:57 +0000 (13:24 -0800)]
Update changing-corelib.md (#21220)
* Update changing-corelib.md
@jkotas
* Update changing-corelib.md
Pent Ploompuu [Tue, 27 Nov 2018 21:24:30 +0000 (23:24 +0200)]
Diff-able disassembly fixes (#20863)
Andy Ayers [Tue, 27 Nov 2018 21:11:02 +0000 (13:11 -0800)]
JIT: handle general indir case for GT_JMP address
Some upcoming changes to reduce tiering overhead require that directly
invoked virtual methods be called indirectly via their slot, so that the
method body can be updated and callers patched up by patching the method
table slot.
Existing code for x64 implicitly assumes that a GT_JMP indirect target address
is near enough to the call site that a 32 bit RIP-relative displacement will
work. We can ensure this is true by always generating a reloc (and hence
potentially a jump stub) -- unless the target happens to fit in 32 bits and
so can be addressed absolutely.
Andy Ayers [Tue, 27 Nov 2018 19:38:56 +0000 (11:38 -0800)]
remove workaround comments
Carol Eidt [Tue, 27 Nov 2018 19:12:52 +0000 (11:12 -0800)]
Merge pull request #21151 from CarolEidt/Fix21080
Use LclFld for full-width cpblk of different types
Andy Ayers [Sat, 17 Nov 2018 18:26:25 +0000 (10:26 -0800)]
JIT: retype some return expressions
The jit is fairly tolerant of byref/native int mismatches for inline arguments
and return values. And some of the new unsafe helpers do this kind of
reinterpretation across call boundaries as well. This sometimes leads the jit
to lose track of byrefs.
A general fix for this is likely somewhat involved. For now we simply detect if
we're about to lose track of a byref when substituting a particular kind of
return expression, and retype the expression to a byref.
Fixes #21051.
Stephen Toub [Tue, 27 Nov 2018 14:27:06 +0000 (09:27 -0500)]
Use local array in ConcurrentQueue<T> for small perf improvement (#21192)
Enqueue/TryDequeue/TryPeek are all repeatedly accessing the same array from a readonly field. Changing them to instead access that same array but from a cached local results in an ~10% throughput boost on a microbenchmark that does uncontended reads/writes of objects from/to the queue. (I also tried using a ref local to point directly to the target slot in the array, but that actually resulted in a measurable regression.)
Andy Ayers [Tue, 27 Nov 2018 08:36:35 +0000 (00:36 -0800)]
Improve perf for Index based span indexers (#21196)
First, evaluate the actual index before invoking the underlying int indexer,
so that the jit doesn't think the span is address taken.
Second, improve the jit's ability to remove some redundant comparisons like
those that arise in computing the actual index.
Chuck Ries [Tue, 27 Nov 2018 06:06:34 +0000 (22:06 -0800)]
Fix a use after free for Managed Data BP (#21205)
ShimProxyCallback::DataBreakpoint::DataBreakpointEvent was holding onto a bare
BYTE* for the CONTEXT rather than copying the buffer and taking ownership. Due to
lifetime, this resulted in a use after free. Apparently in retail code we got lucky
and this worked enough of the time that we never noticed it.
Levi Broderick [Tue, 27 Nov 2018 05:08:10 +0000 (21:08 -0800)]
Fix Rune.ToUpper / ToLower under GlobalizationMode.Invariant (#21203)
Jan Vorlicek [Tue, 27 Nov 2018 03:15:45 +0000 (04:15 +0100)]
Disable noncollectible alc finalization (#21189)
* Disable noncollectible AssemblyLoadContext finalization
The constructor of AssemblyLoadContext was missing a call to
GC.SuppressFinalize for non-collectible AssemblyLoadContext. That means
that the finalizer can be executed and unloading being initialized even
for non-collectible AssemblyLoadContext.
This change fixes that.
* SuppressFinalize non-collectible AssemblyLoadContext
The finalizer for the non-collectible AssemblyLoadContext is never
called and so it should not be in the finalization queue.
* Handle finalizer call without constructor executed
Check if the constructor was executed before we run the finalizer code.
Jan Kotas [Tue, 27 Nov 2018 01:55:10 +0000 (17:55 -0800)]
Disable domain neutral loading (#21156)
Tanner Gooding [Tue, 27 Nov 2018 01:54:36 +0000 (17:54 -0800)]
Deleting a format.patch file that got checked in (#21202)
Jan Kotas [Tue, 27 Nov 2018 01:06:35 +0000 (17:06 -0800)]
Update documentation for changing CoreLib (#21201)
Carol Eidt [Wed, 21 Nov 2018 22:12:03 +0000 (14:12 -0800)]
Use LclFld for full-width cpblk of different types
This issue arose because the source tree was a struct with a single double field, and the destination was a struct with a single long field. Copy prop replaced the lclVar in the source, which was under a `IND(ADDR)` with a lclVar that was not address taken, although it was marked `GTF_DONT_CSE`. Replacing the src tree with a `GT_LCL_FLD` addresses this issue.
Also, ensure that the target is marked `DoNotEnregister` if it isn't referenced as the same size/type.
Fix #21080
Fix #21064
Brian Sullivan [Tue, 27 Nov 2018 00:31:37 +0000 (16:31 -0800)]
Merge pull request #21184 from mikedn/load-norm
Fix normalize on load handling in assertion propagation