Tanner Gooding [Fri, 20 Jul 2018 17:47:47 +0000 (10:47 -0700)]
Realigning the HWIntrinsic template metadata tables
Tanner Gooding [Fri, 20 Jul 2018 17:32:47 +0000 (10:32 -0700)]
Replacing usages of random with TestLibrary.Generator in the HWIntrinsic template metadata
Tanner Gooding [Fri, 20 Jul 2018 17:19:33 +0000 (10:19 -0700)]
Adding TestLibrary.Generator methods for unsigned types
Mike McLaughlin [Tue, 24 Jul 2018 19:19:59 +0000 (12:19 -0700)]
Fix mscordbi metadata reader alignment bug on Linux. (#19070)
Works fine on Windows minidumps, but on Linux (via OpenVirtualProcess
for production breakpoints and future core dumps) the compiler's struct
alignment rules are different. On Windows, classes/structs are aligned
based on the largest field. On Linux, they are 4 byte aligned regardless
of the field sizes.
https://github.com/dotnet/coreclr/issues/17692
Ilia [Tue, 24 Jul 2018 16:49:16 +0000 (19:49 +0300)]
Release leaked ICorProfilerCallback9 in ~EEToProfInterfaceImpl (#19106)
Bug: 19093
Bruce Forstall [Tue, 24 Jul 2018 15:29:07 +0000 (08:29 -0700)]
Merge pull request #19099 from BruceForstall/Arm64DebugBuild
Change Windows arm32/arm64 Debug normal/innerloop jobs to NOT run tests
Russ Keldorph [Tue, 24 Jul 2018 15:25:59 +0000 (08:25 -0700)]
Revert "Revert "Fix some failures when using build-test.sh"" (#18919)
This reverts commit
184c3504e763ea7147505a8d49cb54b43cc82d78.
Sven Boemer [Tue, 24 Jul 2018 15:22:31 +0000 (08:22 -0700)]
Work around cmd command length limit in xunit Exec task (#19095)
On Windows, the Exec task passes the command to cmd, so long commands
run into the command length limit (see
https://github.com/Microsoft/msbuild/issues/2530). This workaround
shortens the xunit command line by replacing the path to the test
binary directory with an environment variable.
Aaron Robinson [Tue, 24 Jul 2018 01:40:17 +0000 (18:40 -0700)]
Add root MSBuild targets for Interop managed test assets (#19069)
Add root CMake file for Interop native test assets
Bruce Forstall [Mon, 23 Jul 2018 23:20:26 +0000 (16:20 -0700)]
Change Windows arm32/arm64 Debug normal/innerloop jobs to NOT run tests
We don't want these Debug jobs to run tests, because Debug testing is
slow, and we have few machine resources. It's a bit of a bug that we
are currently running tests, because the job is named with "Build", not
"Build and Test".
So make it do what the name says: Debug normal/innerloop arm32/arm64 jobs
will not run tests.
This is asymmetric, unfortunately. I might prefer that the Debug jobs get
generated as "Build and Test", but just not be triggered. (But be available
for triggering.) Or perhaps just not get generated. That might require a
separate set of just "Build" jobs that could be useful in some scenarios,
such as for what the current Debug "Build" only jobs are currently used for.
Egor Chesakov [Mon, 23 Jul 2018 21:15:07 +0000 (14:15 -0700)]
Add missing FlushInstructionCache to Thread::CommitGCStressInstructionUpdate (#18784)
Egor Chesakov [Mon, 23 Jul 2018 21:13:20 +0000 (14:13 -0700)]
Disable tests failing on ARM32 in https://github.com/dotnet/coreclr/pull/18784 (#19087)
Amy [Mon, 23 Jul 2018 20:35:57 +0000 (13:35 -0700)]
R2RDump - UnwindInfo for x86 (#18994)
* UnwindInfo for x86
* Update tests
mikedn [Mon, 23 Jul 2018 19:21:34 +0000 (22:21 +0300)]
Change gtExtractSideEffList to use GenTreeVisitor (#18257)
* Change gtExtractSideEffList to use GenTreeVisitor
The tree traversal logic in gtExtractSideEffList is basically an incomplete duplicate of GenTreeVisitor's traversal logic. It lacks handling for GT_ARR_ELEM and GT_ARR_OFFSET so if this are encountered any side effects their children may have are silently dropped.
In addition, side effect ordering was not always preserved. The comma list is built by prepending nodes to it so side effects have to be visited in reverse execution order. The old code did this only for simple opers, for special nodes such as GT_ARR_BOUNDS_CHECK normal execution order was used.
This actually complicates a bit the GenTreeVisitor implementation as side effects need to be first stored in an array. The number of side effects is usually small (<= 4) so this shouldn't be a problem.
* Use GTF_SIDE_EFFECT in optPrepareTreeForReplacement
Assertion propagation doesn't have any way to ensure that it is safe to remove class constructor calls so GTF_PERSISTENT_SIDE_EFFECTS_IN_CSE should not be used here.
Likewise, GTF_EXCEPT side effects must be preserved as well. VN doesn't always propagate exceptions so it's possible to end up with a tree that has a constant VN and contains exception side effects.
* Add tests for 18232
* CR feedback
Jan Kotas [Mon, 23 Jul 2018 19:17:56 +0000 (12:17 -0700)]
Partial R2R IBC fixes (#19073)
* Partial R2R IBC fixes
- Log use method code access in all cases, not just when the method is JITed
- Add workaround for CONTRACT_VIOLATION that shows up in checked builds when collecting IBC data
- Make /ReadyToRun switch work for CoreLib
Andy Ayers [Mon, 23 Jul 2018 18:47:17 +0000 (11:47 -0700)]
JIT: some lclvars related cleanup (#19077)
Consolidate various compiler globals used when setting local var ref
counts by folding them into the visitor:
* lvaMarkRefsCurBlock
* lvaMarkRefsCurStmt
* lvaMarkRefsWeight
Remove the largely vestigial `lvPrefReg` and associated methods to set
or modify this field. Haven't verified but this is likely a remmant of
the legacy backend.
In the one remaning use (lcl var sorting predicates), swap in `lvIsRegArg`
instead, which gets most of the same cases.
Aaron Robinson [Mon, 23 Jul 2018 10:30:57 +0000 (03:30 -0700)]
Remove hosthook api (#19079)
* Remove CallNeedsHostHook() API
* Remove IsHostHookEnabled() API and with it related dead code
* Remove code enabling host hooks (i.e. COMPlus_GenerateStubForHost)
Remove function declarations for creating host hooks
Update comments
Sung Yoon Whang [Mon, 23 Jul 2018 03:56:42 +0000 (20:56 -0700)]
Enable profiler attach on Windows (#18762)
* try building clr with prof attach enabled
* define ICLRProfiling interface
* profattach.dll now builds with CreateCLRProfiling export
* remove try catch
* basic attach working now
* build with profiler attach feature only on win
* Fix linux build
* cleanup
* more cleanup
* more cleanup
* remove profattach dll
* remove useless unix exports in mscorwks
* remove profattach from dll cmakelist
* Add back ifdef
* cleanup
* change LINUX to UNIX in clrdefinitions
* Fix broken checked builds
* Remove CLRProfilingClassFactory and metadata.h include from profattach.cpp
* remove useless extern C
* Add this back in
* adding ifndef DACCESS_COMPILE
* Try building with FWD define ICLRProfiling interface
* Test commit - removing additional definition from metahost.h to see if this will pass CI runs
* Address pr comments
Andy Ayers [Sun, 22 Jul 2018 16:12:10 +0000 (09:12 -0700)]
JIT: stateful local ref counts and weights (#19068)
Introduce a notion of state for local var ref counts and weighted ref counts.
Accesses and current state must agree.
State is invalid initially, enabled for an early period around bits of morph,
invalid again for a time, and then enabled normally once lvaMarkRefs is called.
Accesses normally specify RCS_NORMAL as the desired state, but in the accesses
of selected ref counts in morph, specify RCS_EARLY.
Revise how we decide if normal ref counting is active by changing
`lvaLocalVarRefCounted` into a method.
Update `gtIsLikelyRegVar` to not access ref counts when they're not in a valid
state.
Change weight APIs over to use `weight_t`.
Bruce Forstall [Sat, 21 Jul 2018 00:08:51 +0000 (17:08 -0700)]
Merge pull request #19049 from acmyu/issue19030
R2RDump - Skip R2RDumpTests in crossgen tests
Carol Eidt [Fri, 20 Jul 2018 23:17:04 +0000 (16:17 -0700)]
Merge pull request #19067 from CarolEidt/FixStructRefCnt
Remove incRefCnt from fgMakeOutgoingStructArgCopy
Brian Sullivan [Fri, 20 Jul 2018 22:36:25 +0000 (15:36 -0700)]
Merge pull request #19065 from briansull/fix_19022
The SIMD types must use the struct code path - Fixes issue 19022
Eugene Rozenfeld [Fri, 20 Jul 2018 22:35:47 +0000 (15:35 -0700)]
Improve narrowing of GT_AND nodes. (#18995)
This is a follow-up to to #18816 which resulted in a 6 byte regression in one of the
desktop SuperPMI methods. This change removes that regression and adds a number
of improved diffs.
If we are narrowing GT_AND to an unsigned type and one of the operands can be narrowed
into that type, the result of the GT_AND will also fit into that type and can be narrowed.
The same is true if one of the operands is an int const and can be narrowed into 'dsst'.
The change also ensures that we don't call optNarrowTree(false) more than once on each of the
GT_AND operands.
Aaron Robinson [Fri, 20 Jul 2018 22:27:28 +0000 (15:27 -0700)]
Update documentation for testing (#19063)
* Update documentation for testing
Andy Ayers [Fri, 20 Jul 2018 21:56:09 +0000 (14:56 -0700)]
JIT: handle implicit local var references via local var attribute bit (#19012)
Instead of relying on ref count bumps, add a new attribute bit to local
vars to indicate that they may have implicit references (prolog, epilog,
gc, eh) and may not have any IR references.
Use this attribute bit to ensure that the ref count and weighted ref count for
such variables are never reported as zero, and as a result that these variables
end up being allocated and reportable.
This is another preparatory step for #18969 and frees the jit to recompute
explicit ref counts via an IR scan without having to special case the counts
for these variables.
The jit can no longer describe implicit counts other than 1 and implicit weights
otehr than BB_UNITY_WEIGHT, but that currently doesn't seem to be very important.
The new bit fits into an existing padding void so LclVarDsc remains at 128 bytes
(for windows x64).
Stephen Toub [Fri, 20 Jul 2018 21:11:24 +0000 (17:11 -0400)]
Fix unnnecessary reference to DateTimeFormatInfo.DecimalSeparator (#19064)
dotnet-maestro-bot [Fri, 20 Jul 2018 12:47:21 +0000 (05:47 -0700)]
Update BuildTools, CoreClr, CoreFx, PgoData to preview1-03020-01, preview1-26720-05, preview1-26720-01, master-
20180720-0221, respectively
Aaron Robinson [Fri, 20 Jul 2018 20:07:25 +0000 (13:07 -0700)]
Disable manual COM registration logic on ARM32 (#19042)
* Add advapi32.lib to the list of libs to link against on ARM.
Remove conditional build on ARM in source.
* Manually edit ARM and ARM64 test lists.
Konstantin Baladurin [Fri, 20 Jul 2018 18:19:08 +0000 (21:19 +0300)]
Zapper::CompileAssembly: save NI file atomically (#19059)
It can get rid of possible problems with corrupted NI files if crossgen
will be killed during image saving.
Carol Eidt [Fri, 20 Jul 2018 14:24:39 +0000 (07:24 -0700)]
Remove incRefCnt from fgMakeOutgoingStructArgCopy
Carol Eidt [Fri, 20 Jul 2018 14:20:05 +0000 (07:20 -0700)]
Merge pull request #19040 from CarolEidt/DiffFix
JitDump improvements
Gleb Balykov [Fri, 20 Jul 2018 14:13:49 +0000 (17:13 +0300)]
Add Tizen 5.0 RID (#19057)
Stephen Toub [Fri, 20 Jul 2018 10:56:57 +0000 (03:56 -0700)]
Improve throughput of TimeSpan.ToString/TryFormat with "g"/"G" (#19051)
* Improve throughput of TimeSpan.ToString/TryFormat with "g"/"G"
TimeSpan has three standard formats: "c", "g", and "G". Yesterday I updated its implementation with throughput improvements for "c" (the default) based on porting the design from Utf8Formatter; this PR does so for "g"/"G".
Initially I wasn't going to handle "g"/"G" as they factor in culture (Utf8Formatter doesn't), but even with accessing the current culture there are still significant wins to be had. I was also going to keep the "c" and "g"/"G" implementations separate, to avoid bogging down the default "c" formatting with additional conditions needed to support "g"/"G", but the overhead incurred for that turns out to be minimal enough that it's worth keeping one implementation rather than two mostly-similar ones... the impact on "c" is mostly within noise.
This PR makes a significant throughput improvement for "g"/"G" formatting. It also removes several unnecessary allocations, such that TryFormat with "g"/"G" is now allocation-free (and ToString just allocates the asked-for string).
* Address PR feedback
Aaron Robinson [Fri, 20 Jul 2018 07:48:07 +0000 (00:48 -0700)]
Minor interop code changes (#19005)
* Remove SpecialIsGenericTypeVisibleFromCom() from header file
* Clean up if check resulting from remoting code removal
Phil Garcia [Fri, 20 Jul 2018 07:28:39 +0000 (00:28 -0700)]
Fixed missing readonly attribute (#19052)
Aaron Robinson [Fri, 20 Jul 2018 02:42:48 +0000 (19:42 -0700)]
Add issues with COM tests in the helix build. (#19047)
Presently COM tests does not appear to run properly on Windows.Nano but
there is no way to special case that OS so disabling them on all builds.
Amy Yu [Fri, 20 Jul 2018 00:01:50 +0000 (17:01 -0700)]
Skip R2RDumpTests in crossgen tests, test unsupported on ARM
Brian Sullivan [Fri, 20 Jul 2018 00:56:34 +0000 (17:56 -0700)]
Fixes issue 19022
The SIMD types need to use the struct path when we are building
the zero-initializations in fgMorphRecursiveFastTailCallIntoLoop
Tanner Gooding [Thu, 19 Jul 2018 19:16:06 +0000 (12:16 -0700)]
Removing WINCORESYS usages from the test projects
Noah Falk [Thu, 19 Jul 2018 23:36:34 +0000 (16:36 -0700)]
Enable the no tiered compilation innerloop PR legs (#18953)
Enable the no tiered compilation innerloop PR legs
Tanner Gooding [Thu, 19 Jul 2018 18:30:24 +0000 (11:30 -0700)]
Adding a project reference for CoreCLRTestLibrary to the hwintrinsic test dir.props
Tanner Gooding [Thu, 19 Jul 2018 18:00:46 +0000 (11:00 -0700)]
Regenerating the HWIntrinsic tests from the updated templates
Tanner Gooding [Thu, 19 Jul 2018 17:30:32 +0000 (10:30 -0700)]
Updating the templated HWIntrinsic tests to use `TestLibrary.TestFramework` for logging.
Sven Boemer [Thu, 19 Jul 2018 20:44:16 +0000 (13:44 -0700)]
Build xunit wrappers the same way on windows and unix (#18695)
* Initial change to allow build wrappers and runtest.py
* Build xunit wrappers on unix
The generated wrapper needs to target netcoreapp on unix. I had to
exclude assets from the xunit package and introduce a dependency on
the private corefx bits, to resolve a dependency conflict in which the
generated wrapper was depending on an older System.Runtime.dll than
the helper library.
I also disabled binclash logging, because the wrapper build binplaces
the helper library to the same location multiple times. I couldn't
find a simple way to disable binclash logging for the wrapper build
only, since that requires passing an empty switch to run.exe, and bash
word splitting makes this nontrivial from build-test.
* Correctly generate TestEnv xplat
Note that this will still require changes to the test wrapper
to actually source the TestEnv on unix
* Build xunit wrappers using SDK
* Target netcoreapp2.0 in xunit wrappers
This way, the wrappers can build even if the 2.1 SDK isn't installed on
the machine.
* Restore to packages directory for xunit wrappers
* Move common properties out to dir.common.props
When building wrappers using the SDK, we need some basic
properties (like the build os/arch/config, and the output directories)
to be set. I factored out properties used by both the old test build
and the new SDK-project test build.
At first I tried using Directory.Build.props (which is automatically
imported by the SDK), but our test build already imports SDK targets
in various places, so this was resulting in duplicate
imports. Instead, I used dir.common.props, and made the imports
explicit.
* Remove desktop-specific test wrapper csproj
* Pass build os/arch/type and logsdir to msbuild from runtest.py
* Remove xunit wrapper helper library from traversal build
* Fix parameter passing in build-test.sh
Use bash arrays to pass parameters for the build command. This makes
it possible to pass arguments with spaces to build_Tests_internal. We
use this to disable binclashlogging selectively (for the xunit wrapper
build only).
* Clean up factored .props files
* Undo runtest.sh changes
* Use latest xunit console runner everywhere
* Remove extra StaticDependency on xunit.runner.console
* Eliminate tests/src/dir.common.props, and rename dir.sdkbuild.props
tests/src/dir.common.props was only used for the desktop-specific
xunit wrapper helper library. There's no need for it any more, so its
properties have been moved into tests/src/dir.props.
dir.sdkbuild.props has been renamed to dir.common.props, since it
contains properties used by SDK projects and buildtools projects.
This change also re-enables the test build.
* Reintroduce dir.sdkbuild.props as a place for SDK-only props
With this, some properties shared by SDK projects can go in a global
location. The TargetFramework is shared by all SDK projects in the
test tree.
This change also uses a property for the xunit package directory that
contains the xunit.console.dll we copy to core_root.
* Add xml namespace to dir.common.props
This fixes a failure in the windows build.
* Satisfy xunit analyzer
* Satisfy xunit analyzer again
* Use SDK msbuild to build wrappers
On windows, the use of run.exe, config.json, and msbuild.cmd uses
msbuild.exe on the path. This change will build wrappers using the
local SDK via "dotnet msbuild", bypassing run.exe. Run.exe will go
away entirely with the move from buildtools to arcade, so other build
invocatios should follow suit.
* Remove Microsoft.CSharp.Core.targets workaround
UseBuildTools used to be true all the time. Now that we are building
wrappers on core, UseBuildTools becomes false. However, the rest of
the runtest.proj expects to build using buildtools, so we keep
UseBuildTools true until we switch to arcade.
The CSharpCoreTargetsPath was imported when running on core only. This
used to happen only on unix, but now it also happens when building
runtest.proj for the xunit wrappers on windows. On unix, this targets
file was a symlink to itself to work around some buildtools logic that
expected the file to exist. This workaround no longer appears
necessary, and on windows, this was never used in the first place, so
this change removes it.
* Remove UseRoslynCompilers prop and unify roslyn import
UseRoslynCompilers was introduced in buildtools by
https://github.com/dotnet/buildtools/pull/947, with different
behaviors on windows/unix. It was removed by
https://github.com/dotnet/buildtools/pull/1974, so we can unify our
roslyn imports now.
* Don't copy xunit dlls to corefx test host
The corefx tests run on specific versions of xunit dlls, defined in
CoreFX.depproj. We want to use these versions in the test host, not
those in CORE_ROOT, so exclude these from being copied to the test host
directory. This fixes the failing corefx tests.
* Don't pass run.exe arguments through build-test.cmd in test pipeline
These arguments get passed along to the xunit wrapper build as
unprocessed build args. They need to work for "dotnet msbuild" (used
for the wrapper build) as well as for run.exe.
* Fix parameter passing of priority arg in build-test.cmd
UnprocessedBuildArgs should contain arguments in the format expected by
msbuild, not by run.exe.
* Fix parameter passing of unprocessed args in build-test.cmd
The "--" syntax is used by run.exe to pass everything following to
msbuild directly. It should not be a part of unprocessed args.
* Pass TargetsWindowsArg to wrapper build in build-test.cmd
Helix builds tests on windows and runs them on unix using the xunit
wrappers. When cross-building the wrappers like this, TargetsWindows
is set to false by the test build pipeline. This variable ensures that
the wrapper uses correct directory separators when invoking the test
.sh file.
* Pass BuildTestsAgainstPackages arg to exclude unix tests
Helix builds xunit wrappers on windows, and runs them on unix. The
BuildTestsAgainstPackages should currently be set to true in the
windows wrapper build to properly filter the .cmd files based on
exclusions in issues.targets.
Jan Kotas [Wed, 18 Jul 2018 21:02:13 +0000 (14:02 -0700)]
Fix desktop build break
Signed-off-by: dotnet-bot <dotnet-bot@microsoft.com>
Carol Eidt [Thu, 19 Jul 2018 19:11:20 +0000 (12:11 -0700)]
JitDump improvements
- Reduce the size of "diffable" values (I've been encountering a number of diffs on x86; 2 bits seems to take care of it)
- Dump a note when we start morphing args for a call.
Carol Eidt [Thu, 19 Jul 2018 18:08:33 +0000 (11:08 -0700)]
Merge pull request #18999 from CarolEidt/Fix16359
Arm64: Fix handling of IP0 & IP1
Bruce Forstall [Thu, 19 Jul 2018 18:04:49 +0000 (11:04 -0700)]
Merge pull request #19032 from BruceForstall/DisableR2rDumpTestForJitStress
Disable r2rdump test for JIT stress currently
Bruce Forstall [Thu, 19 Jul 2018 17:10:02 +0000 (10:10 -0700)]
Disable r2rdump test for JIT stress currently
Tracked by issue #19011
Bruce Forstall [Thu, 19 Jul 2018 16:02:24 +0000 (09:02 -0700)]
Merge pull request #19004 from BruceForstall/FixX86UbuntuJobs
Disable failing Ubuntu x86 jobs
Kai Ruhnau [Thu, 19 Jul 2018 15:41:15 +0000 (17:41 +0200)]
Use mcontext_t instead of sigcontext (#18983)
The type of uc_mcontext has always been `mcontext_t`, but earlier versions of Glibc just typedef'ed this to `sigcontext`.
Newer versions of Glibc (since 2.27) [introduced an explicit struct type](https://sourceware.org/git/?p=glibc.git;a=blobdiff;f=sysdeps/unix/sysv/linux/arm/sys/ucontext.h;h=
192d1bdeac3e62e13110f68614f7af624047c3a9;hp=
2abceef2a4ca9bbb6e42208eaee548228b041e5b;hb=
4fa9b3bfe6759c82beb4b043a54a3598ca467289;hpb=
5898f4548efdcd7c0fd437a74eeb80facc51a117) and now compilation fails.
Jan Kotas [Thu, 19 Jul 2018 15:29:36 +0000 (08:29 -0700)]
Add MemoryMarshal.AsRef implementation (#19021)
Contributes to https://github.com/dotnet/corefx/issues/30613
sunnycase [Thu, 19 Jul 2018 08:05:40 +0000 (16:05 +0800)]
Fix opinfo: InlineR should only read 8 bytes (#18801)
Noah Falk [Thu, 19 Jul 2018 07:28:31 +0000 (00:28 -0700)]
Eliminate arm64 contract asserts (#19015)
These methods don't trigger GC but were previously marked as if they did. Correcting the marking will eliminate spurious asserts.
dotnet-maestro-bot [Thu, 19 Jul 2018 07:16:55 +0000 (00:16 -0700)]
Update CoreClr to preview1-26719-01 (#19016)
Hugh Bellamy [Thu, 19 Jul 2018 03:29:30 +0000 (04:29 +0100)]
Implement GetComObjectData and SetComObjectData (#18900)
* Cleanup Marshal.cs
* Cleanup __ComObject
* Implement GetComObjectData and SetComObjectData
* Build fixes
* PR feedback and more marshal cleanup
* More feedback
* Disable corefx tests
* Address PR feedback
* Fix build breaking typo
* Test disable
Aaron Robinson [Thu, 19 Jul 2018 00:52:02 +0000 (17:52 -0700)]
Basic .NET client tests (#18843)
- Add support for a testing primitive marshalling with a native COM server
and .NET client. This uses RegFree COM for activation.
- Remove ClassicCOM tests
- Bring back tests for activation via reflection
Bruce Forstall [Wed, 18 Jul 2018 23:33:54 +0000 (16:33 -0700)]
Disable failing Ubuntu x86 jobs
Also add a minimal fix that fixes a current problem, but isn't
expected to help fix the entire issue.
Failures tracked by https://github.com/dotnet/coreclr/issues/19003.
Stephen Toub [Wed, 18 Jul 2018 23:10:42 +0000 (16:10 -0700)]
Improve TimeSpan.ToString/TryFormat throughput for default format (#18990)
Jakob Botsch Nielsen [Wed, 18 Jul 2018 22:57:17 +0000 (00:57 +0200)]
Properly type morphed NEG nodes (#18837)
* Properly type optimized NEG nodes
When the JIT was morphing trees like '-1 * expr', it would turn the
multiplication into a NEG node with the same type as its right operand.
This is a problem when the right operand was a small type like TYP_UBYTE
because the NEG node always produces a widened result. This could cause
problems when the negation was fed into something that depended on the
type, such as a cast to TYP_UBYTE: here the JIT would conclude that the
cast could be dropped, and end up producing a widened result.
The solution is to give the tree the actual type of the NEG node.
Also add a test for this case and for a similar case of '0 - expr',
which already had a fix.
Fix #18780
* Address PR feedback
* Clarify comment
Andy Ayers [Wed, 18 Jul 2018 21:35:08 +0000 (14:35 -0700)]
JIT: force all local var ref counts to be accessed via API (#18979)
This is a preparatory change for auditing and controlling how local
variable ref counts are observed and manipulated.
See #18969 for context.
No diffs seen locally. No TP impact expected.
There is a small chance we may see some asserts in broader testing
as there were places in original code where local ref counts were
incremented without checking for possible overflows. The new APIs
will assert for overflow cases.
dotnet-maestro-bot [Wed, 18 Jul 2018 20:46:37 +0000 (13:46 -0700)]
Update CoreClr, PgoData to preview1-26718-05, master-
20180718-0214, respectively (#18978)
Carol Eidt [Wed, 18 Jul 2018 20:41:59 +0000 (13:41 -0700)]
Arm64: Fix handling of IP0 & IP1
These are sometimes mandated to be allocated to certain nodes, but they were not in the list of allocatable registers (REG_VAR_ORDER). This led to special handling in LSRA, which, it turns out, was incomplete. This resulted in failures for JitStressRegs=0x200.
Based on the discussion in #14607, this adds IP0 to RBM_CALLEE_TRASH_NOGC, and fixes the REG_VAR_ORDER.
Fix #14607, Fix #16359, Fix #17861
Bruce Forstall [Wed, 18 Jul 2018 19:29:21 +0000 (12:29 -0700)]
Merge pull request #18970 from BruceForstall/FixAltjitCorefxTests
Fix corefx altjit test jobs
Amy [Wed, 18 Jul 2018 19:26:36 +0000 (12:26 -0700)]
R2RDump - Test infrastructure (#18745)
* Test infrastructure for R2RDump, compare xml and output diff
* Add R2RDumpTest to build-tests
* Fix errors causing tests to fail
* Create XUnitWrapper for R2RDumpTest
* Generate readytorun binaries from source code, copy expected xml output files to executable's directory
* Test R2RDump through commandline instead of calling R2RDump functions
* Fix errors
* Prevent duplicate xml tags
* Read test xml from string instead of file
* Fix test bugs
* Call dotnet r2rdump from msbuild
* Fix errors
* Use right slash for paths in bash
* Use different expected xml for different architectures
* R2RDumpTests for non-Windows
* Add more test cases
* Fix errors
* Supress warnings, avoid error when parsing x86 images
* Add license headers
Jan Kotas [Wed, 18 Jul 2018 19:00:42 +0000 (12:00 -0700)]
Revert "Improve narrowing of GT_AND nodes. (#18916)" (#18975)
This reverts commit
c0bad3c66218f61dd3e4f151b9d2eae22c6d88d2.
Jay Krell [Wed, 18 Jul 2018 19:00:19 +0000 (12:00 -0700)]
Use FD_CLOEXEC instead of 1. (#18976)
Carol Eidt [Wed, 18 Jul 2018 18:01:46 +0000 (11:01 -0700)]
Merge pull request #18941 from CarolEidt/Fix18884
Kill RCX when used by shift
Jan Kotas [Wed, 18 Jul 2018 17:42:00 +0000 (10:42 -0700)]
Use api-set instead of combase.dll (#18981)
Marco Rossignoli [Wed, 18 Jul 2018 10:43:02 +0000 (12:43 +0200)]
Remove unuseful null check on Activator.CreateInstance (#18806)
Pent Ploompuu [Wed, 18 Jul 2018 10:14:49 +0000 (13:14 +0300)]
Optimize {Try}FormatDecimal (#18935)
Jan Kotas [Wed, 18 Jul 2018 10:13:15 +0000 (03:13 -0700)]
Delete Span ifdefs that are no longer needed (#18982)
Bruce Forstall [Tue, 17 Jul 2018 21:32:33 +0000 (14:32 -0700)]
Fix corefx altjit test jobs
Recently, the corefx build switched from using the VS installed msbuild
to using "dotnet msbuild" for doing builds. This changed the way
command-line parsing was handled.
If we continue to pass the WithoutCategories argument on the command-line,
it ends up requiring an extraordinary number of "escape" backslashes
to preserve exactly the right number of backslashes that will be passed
through to the actual invocation of msbuild. That seems complex and also
not robust to future change.
Instead, construct a response file with the simple required string,
that will be passed directly through to msbuild.
Egor Chesakov [Tue, 17 Jul 2018 23:22:27 +0000 (16:22 -0700)]
Enable CoreFx tests disabled in https://github.com/dotnet/coreclr/issues/17753 (#18967)
Pent Ploompuu [Tue, 17 Jul 2018 21:59:00 +0000 (00:59 +0300)]
Clean-up number.cpp (#18964)
Jan Kotas [Tue, 17 Jul 2018 19:11:33 +0000 (12:11 -0700)]
Add pooling for JIT scratch memory (#18924)
Fixes #3408
dotnet-maestro-bot [Tue, 17 Jul 2018 19:10:53 +0000 (12:10 -0700)]
Update CoreClr, CoreFx, PgoData to preview1-26717-04, preview1-26717-04, master-
20180717-0139, respectively (master) (#18961)
* Update CoreClr, CoreFx, PgoData to preview1-26717-04, preview1-26717-04, master-
20180717-0139, respectively
* Disabled outdated tests
Aaron Robinson [Tue, 17 Jul 2018 18:12:02 +0000 (11:12 -0700)]
Update NativeCallable attribute tests (#18946)
* Update NativeCallable attribute tests
Vance Morrison [Tue, 17 Jul 2018 17:42:28 +0000 (10:42 -0700)]
Merge pull request #18938 from vancem/EventSourceShutdown
Add back in code deleted in .NET Core port.
Carol Eidt [Mon, 16 Jul 2018 18:37:40 +0000 (11:37 -0700)]
Kill RCX when used by shift
RCX must be explicitly killed. Otherwise, if there's a case of a def/use conflict - as in this test case where the shift amount is defined by a divide that must go in RAX, it won't be explicitly assigned to RCX,.
Also, the handling of conflicts must not use the register assignment of the def on the use if it conflicts with the use register requirements, and vice versa.
Fix #18884
Carol Eidt [Tue, 17 Jul 2018 15:54:42 +0000 (08:54 -0700)]
Merge pull request #18891 from CarolEidt/Fix18153
Ensure BuildUse uses the correct reg number
Pent Ploompuu [Tue, 17 Jul 2018 15:41:39 +0000 (18:41 +0300)]
Move Decimal to shared (#18948)
* Move Decimal to shared
* Remove DecimalCanonicalize{Internal}
Stephen Toub [Tue, 17 Jul 2018 13:00:11 +0000 (06:00 -0700)]
Improve throughput of {U}Int32/64.{Try}Parse for hex (#18952)
* Improve throughput of {U}Int32/64.{Try}Parse for hex
- Copies/modifies the recently added TryParse32/64IntegerStyle routines for hex in order to provide more optimized parsing.
- Since AllowHexSpecifier can't be used with any other styles (other than allowing leading/trailing whitespace), deletes existing code paths for parsing hex.
- As I was copying code around, I kept getting bitten by different names used for the same thing, so I also standardized some parameter names in various helper functions.
* Clean up
- Remove a bunch of unnecessary 'unsafe' keywords
- Rephrase some while(true) loops as do/while loops for consistency and to avoid some repetition
- Remove an unnecessary wrapper function
dotnet-maestro-bot [Tue, 17 Jul 2018 11:27:23 +0000 (04:27 -0700)]
Update BuildTools, CoreClr, CoreFx, PgoData to preview1-03013-03, preview1-26716-04, preview1-26717-02, master-
20180716-0324, respectively (master) (#18927)
* Update BuildTools, CoreClr, CoreFx, PgoData to preview1-03013-03, preview1-26716-04, preview1-26717-02, master-
20180716-0324, respectively
* Update EventPipe StackWalker usage to specify ALLOW_INVALID_OBJECTS.
Jan Kotas [Tue, 17 Jul 2018 08:51:57 +0000 (01:51 -0700)]
Delete redundant broken offsetof definitions (#18947)
Fixes build breaks with latest Visual Studio Preview.
Fixes #18338
Koundinya Veluri [Tue, 17 Jul 2018 05:04:07 +0000 (22:04 -0700)]
Apply tiering's call counting delay more broadly (#18610)
Apply tiering's call counting delay more broadly
Issues
- When some time passes between process startup and first significant use of the app, startup perf with tiering can be slower because the call counting delay is no longer in effect
- This is especially true when the process is affinitized to one cpu
Fixes
- Initiate and prolong the call counting delay upon tier 0 activity (jitting or r2r code lookup for a new method)
- Stop call counting for a called method when the delay is in effect
- Stop (and don't start) tier 1 jitting when the delay is in effect
- After the delay resume call counting and tier 1 jitting
- If the process is affinitized to one cpu at process startup, multiply the delay by 10
No change in benchmarks.
Carol Eidt [Tue, 17 Jul 2018 04:45:47 +0000 (21:45 -0700)]
Use `FEATURE_ARG_SPLIT` to qualify handling of `GT_PUTARG_SPLIT`
Andy Ayers [Tue, 17 Jul 2018 01:48:24 +0000 (18:48 -0700)]
JIT: fix value numbering to handle GT_NULLCHECK more generally (#18942)
With the advent of #18819 we may now see GT_NULLCHECK nodes with operands
that also can cause exceptions.
Handle this in value numbering.
Closes #18937.
Carol Eidt [Mon, 16 Jul 2018 23:31:00 +0000 (16:31 -0700)]
PR Feedback
Tanner Gooding [Sun, 15 Jul 2018 15:04:45 +0000 (08:04 -0700)]
Adding tests for the Avx2 AlignRight, Blend, and ConvertTo* hwintrinsics.
Tanner Gooding [Thu, 12 Jul 2018 14:40:30 +0000 (07:40 -0700)]
Implementing the Avx2 AlignRight, Blend, and ConvertTo* hwintrinsics.
Bruce Forstall [Mon, 16 Jul 2018 21:41:19 +0000 (14:41 -0700)]
Merge pull request #18917 from BruceForstall/dev/unix_test_workflow
Enable Windows ARM64 corefx jobs in the CI
Noah Falk [Mon, 16 Jul 2018 20:48:37 +0000 (13:48 -0700)]
Add no tiered compilation test jobs (#18864)
* Add no tiered compilation test jobs
With the upcoming switch to make tiered compilation on by default we want to preserve some JIT testing that will not enable tiered compilation.
1) Add 'no_tiered_compilation_pri0' jit stress job, logically the non-tiered variant of current 'Innerloop' jobs
- This job will run on all PRs for Checked x (Windows x86, Windows x64, Ubuntu arm) - but not scheduled that way for the moment so it can be tested post-checkin
2) Add 'no_tiered_compilation' jit stress job, logically the non-tiered variant of the current 'normal' job
- This job has the same periodic schedule as other jit stress jobs
3) Add 'r2r_no_tiered_compilation' r2r stress job, logically the non-tiered variant of the current 'r2r' job
- This job has the same periodic schedule as other r2r stress jobs
Misc adjustments - Added isPri0TestScenario because 'innerloop' is no longer the only scenario name which has this property + eliminated various comments that implied 'innerloop' IFF pri0
Carol Eidt [Thu, 12 Jul 2018 19:51:06 +0000 (12:51 -0700)]
Ensure BuildUse uses the correct reg number
`BuildUse` was setting the regNumber for all of the uses of a multi-reg node to the main/first regNum. This was missed because this results in a def/use conflict on that reg (the def was set correctly), which is generally resolved in favor of the def. The exception is when there is a kill of the register in between, in which case the use tries to allocate the register its been assigned, causing the `farthestRefPhysRegRecord != nullptr` assert (aka "a register can't be found to spill").
This fixes the specific issue, and adds additional asserts to identify future/additional such issues.
The new asserts depend upon all the regNums being appropriately when/if any are set, which wasn't always the case prior to register allocation.
Fix #18153
Matt Warren [Mon, 16 Jul 2018 18:50:49 +0000 (19:50 +0100)]
Fixing markdown list formatting (#18936)
Jan Kotas [Mon, 16 Jul 2018 18:40:26 +0000 (11:40 -0700)]
Fix disabled test name (#18932)
System.ComponentModel.Composition.Tests use friends visibility. The fixed drop does not work well for testing of the live bits.
Eugene Rozenfeld [Mon, 16 Jul 2018 18:24:54 +0000 (11:24 -0700)]
Improve narrowing of GT_AND nodes. (#18916)
This is a follow-up to to #18816 which resulted in a 6 byte regression in one of the
desktop SuperPMI methods. This change removes that regression and adds a number
of improved diffs.
If we are narrowing GT_AND to an unsigned type and one of the operands can be narrowed
into that type, the result of the GT_AND will also fit into that type and can be narrowed.
The same is true if one of the operands is an int const and can be narrowed into 'dsst'.
The change also ensures that we don't call optNarrowTree(false) more than once on each of the
GT_AND operands.
Egor Chesakov [Mon, 16 Jul 2018 16:41:25 +0000 (09:41 -0700)]
Mark RuntimeEventSourceTest as GCStressIncompatible (#18911)
Vance Morrison [Mon, 16 Jul 2018 16:40:31 +0000 (09:40 -0700)]
Add back in code deleted in .NET Core port.
This fixes shutdown race bug that CoreCLR port reintroduced.
Basically the fix needed some Appdomain APIs that were probably
not in .NET Core 1.0 so they were removed. However they are
back in .NET Core 2.0 so it is trivial to simply 'do the right thing', and
put the shutdown logic back.
See comment in front of DisposeOnShutdown for more.
Stephen Toub [Mon, 16 Jul 2018 13:59:55 +0000 (06:59 -0700)]
Improve throughput of UInt32/UInt64 integer parsing (#18930)
As was recently done for Int32/Int64, ports the Utf8Parser approach to parsing to UInt32/UIn64.{Try}Parse, specifically for NumberStyles.Integer (the default).
Also fixes an issue discovered in the previous Int32/Int64 changes, where if the input both has an overflow and has a formatting error (e.g. Int32.Parse("12345678910blah"), we would end up throwing whichever error was hit first, which is a change from .NET Core 2.1 and netfx. The FormatException needs to be preferred over the OverflowException, which just means we can't bail early when overflow is detected.