Andy Ayers [Tue, 10 May 2016 21:19:41 +0000 (14:19 -0700)]
Inliner: introduce ReplayPolicy
The ReplayPolicy reads an external script to determine which inlines
to perform. The script is the same Xml syntax that's produced by
the inliner when JitInlineDumpXml is enabled. This format can be edited
by hand or tool to force particular inlining patterns to occur. Methods
or calls sites not mentioned in the script are considered as noinline.
There's a bunch of work still left to make this fully robust, but in
testing it works well enough for my immediate use case that I'll hold
off on further polish until it's needed. But, for future reference,
here's a laundry list:
* Need better ways to identify methods. Token and hash are not enough.
* Need better ways to identify call sites. Callee token is not enough.
* Consider preparsing or mapping the script into memory.
* Consider caching node positions in the InlineContexts.
* Make it robust for multithreading.
* Handle the prejit root case somehow.
* Possibly allow overriding of inline attributes.
tijoytom [Tue, 10 May 2016 17:33:36 +0000 (10:33 -0700)]
Merge pull request #4793 from tijoytom/master
UTF8 Marshaling support(UnmanagedType.LPUTF8Str)
Hugh Bellamy [Tue, 10 May 2016 14:52:05 +0000 (15:52 +0100)]
Fix ArgumentNullException messages passed as parameter names (#2889)
Fixed several ArgumentNullExceptions that throw with the message as the parameter name, leading to potential confusion for developers.
Jonghyun Park [Tue, 10 May 2016 14:51:26 +0000 (23:51 +0900)]
Adjust relative offsets in UMThunkStub (#4888)
* Adjust relative offsets in UMThunkStub
This commit updates the relative offsets inside UMThunkStub according to the recent changes in
UMThunkStub (for stack unwinding).
* Uses 'UMThunkStub_StackArgsSize' instead of 'UMThunkStub_StackArgsOffset'
This commit revises 'UM2MThunk_WrapperHelper' to use 'UMThunkStub_StackArgsSize'
instead of 'UMThunkStub_StackArgsOffset'.
Jan Kotas [Tue, 10 May 2016 13:34:17 +0000 (06:34 -0700)]
Merge pull request #4713 from myungjoo/fix/4712
Match Inconsistent CMake/MSBuild Compiler Definitions 2
Jan Kotas [Tue, 10 May 2016 13:15:00 +0000 (06:15 -0700)]
Merge pull request #4870 from RussKeldorph/illicense
Update license header in most .il files.
Jan Kotas [Tue, 10 May 2016 13:10:55 +0000 (06:10 -0700)]
Merge pull request #4886 from dotnet-bot/from-tfs
Merge changes from TFS
Valery [Tue, 10 May 2016 12:57:47 +0000 (15:57 +0300)]
Fixed typo in jitinterface.cpp (#4889)
John Chen [Tue, 10 May 2016 06:20:49 +0000 (23:20 -0700)]
Merge pull request #4883 from JohnChen0/r2r
Fix CrossGen to skip some methods that require generic dictionary
John Chen [Tue, 10 May 2016 06:19:56 +0000 (23:19 -0700)]
Merge pull request #4829 from JohnChen0/master
Increase a test timeout
Jan Kotas [Tue, 10 May 2016 05:58:01 +0000 (22:58 -0700)]
JIT-EE interface changes to support CoreRT
- Add flags and constants for reverse PInvoke transitions (https://github.com/dotnet/corert/issues/611)
- Add new multi-dim array constructor that does not use varargs
[tfs-changeset: 1603336]
Matt Ellis [Tue, 10 May 2016 05:33:20 +0000 (22:33 -0700)]
Fix call to GetSortHandle (#4882)
GlobalizationNatvie_GetSortHandle takes a UTF8 encoded string for the
locale name we want to construct a handle to, which is passed to ICU to
open some locale data.
We converted the UTF-16 encoded locale name to UTF8 in managed code, but
neglected to actually ensure the resulting value was null terminated.
Fixes #4784
Bruce Forstall [Tue, 10 May 2016 04:05:35 +0000 (21:05 -0700)]
Merge pull request #4865 from BruceForstall/JitDumpAssertPtrs
Don't print pointers to the JitDump
AlexGhiondea [Tue, 10 May 2016 01:33:32 +0000 (18:33 -0700)]
Merge pull request #4866 from AlexGhiondea/mscorlibUpdates
Create facade and reference assembly for mscorlib.
Andy Ayers [Tue, 10 May 2016 01:05:36 +0000 (18:05 -0700)]
Merge pull request #4867 from AndyAyersMS/FixBenchmarkNames
Fix class names for benchmarks
Rahul Kumar [Tue, 10 May 2016 01:04:29 +0000 (18:04 -0700)]
Merge pull request #4864 from dotnet-bot/from-tfs
Merge changes from TFS
John Chen (CLR) [Mon, 9 May 2016 23:38:54 +0000 (16:38 -0700)]
Fix CrossGen to skip some methods that require generic dictionary
Resolves issue #4801.
tijoytk [Mon, 9 May 2016 23:23:52 +0000 (16:23 -0700)]
Taking care of review comments.
Let out some of the review comments
since they are optimizations.
Russ Keldorph [Mon, 9 May 2016 21:25:36 +0000 (14:25 -0700)]
Update license header in most .il files.
Andy Ayers [Mon, 9 May 2016 20:21:07 +0000 (13:21 -0700)]
Fix class names for benchmarks
Fix a couple of class names for benchmarks so they more accurately
describe the benchmark.
Russ Keldorph [Mon, 9 May 2016 20:19:32 +0000 (13:19 -0700)]
Merge pull request #4862 from RussKeldorph/pending3
Fix conv_ovf_i8_i test
Alex Ghiondea [Mon, 9 May 2016 20:03:14 +0000 (13:03 -0700)]
Create the project that will generate the facade project
Alex Ghiondea [Fri, 6 May 2016 20:35:15 +0000 (13:35 -0700)]
Create a reference assembly for mscorlib
Bruce Forstall [Mon, 9 May 2016 20:03:50 +0000 (13:03 -0700)]
Don't print pointers to the JitDump; output the assertion bitvector as a string.
Carol Eidt [Mon, 9 May 2016 19:50:51 +0000 (12:50 -0700)]
Merge pull request #4852 from mikedn/lsra-reftype
Define LSRA's RefType enum via macros
Jan Vorlicek [Mon, 9 May 2016 19:09:55 +0000 (21:09 +0200)]
Let all signals be unblocked on Unix (#4863)
This change removes blocking of all of the 17 signals that we were
blocking. I have looked at the default behavior of their handlers
and we should be fine with it.
Jeremy Kuhne [Mon, 9 May 2016 18:50:47 +0000 (11:50 -0700)]
Fix desktop build break.
[tfs-changeset: 1603043]
John Chen (CLR) [Fri, 6 May 2016 17:26:46 +0000 (10:26 -0700)]
Increase a test timeout
Test case tests/src/GC/Regressions/v2.0-beta2/462651 occasionally fails
due to time out (see issue #3691). Mark it with IsLongRunningGCTest.
AlexGhiondea [Mon, 9 May 2016 17:28:44 +0000 (10:28 -0700)]
Merge pull request #4828 from AlexGhiondea/systemprivatecorelib
Create System.Private.CoreLib assembly
Rahul Kumar [Mon, 9 May 2016 17:26:50 +0000 (10:26 -0700)]
Merge pull request #4837 from rahku/crossgen
Helpers for ReadyToRun
Rahul Kumar [Thu, 5 May 2016 22:53:07 +0000 (15:53 -0700)]
[Arm64] Helpers for ReadyToRun
Mike Danes [Sun, 8 May 2016 17:41:31 +0000 (20:41 +0300)]
Define LSRA's RefType enum via macros
Russ Keldorph [Fri, 6 May 2016 23:27:37 +0000 (16:27 -0700)]
Fix conv_ovf_i8_i test
Test has at least four problems:
- It assumes native int = int32, so disable it for 64-bit
- It is just wrong in a couple of its assumptions about when overflow
happens during unsigned conversions.
- On success, it returns 0xAAAA instead of the commonly accepted 100
- It has no copyright header
Also piggybacking a copyright fix for rem_r4.il that I missed earlier
Gaurav Khanna [Mon, 9 May 2016 14:36:36 +0000 (07:36 -0700)]
Merge pull request #4834 from gkhanna79/RemoveSxSJIT
Cleanup sxsJitStartup from codemanager
Davis Goodin [Mon, 9 May 2016 13:20:55 +0000 (08:20 -0500)]
Fix incremental build restore: move generated test_runtime project.json to bin (#4764)
* Move generated test_runtime project.json to bin.
* Update helix tasks to point to moved test_runtime files.
* Remove unnecessary TestRuntimeDependenciesJson property.
Jonghyun Park [Mon, 9 May 2016 09:46:10 +0000 (18:46 +0900)]
Do NOT overwrite the stack frame pointer (#4857)
CallEHFunclet currently overwrite the stack frame pointer (R7) while
recovering the registers.
This commit revises CallEHFunclet not to overwrite R7 when recoverint
the registers.
MyungJoo Ham [Mon, 9 May 2016 06:55:46 +0000 (15:55 +0900)]
Linux/ARM: fix stack alignment breaks (#4858)
ehhelper.S: the function no more requires dummy stack add.
memcpy.S: C_FUNC(memcpy) has misaligned stack.
(memcpy.S is not directly related with the issue #4779.
However, this requires the same fix with ehhelper.S)
Fix #4779
Signed-off-by: MyungJoo Ham <myungjoo.ham@samsung.com>
Russ Keldorph [Mon, 9 May 2016 04:44:05 +0000 (21:44 -0700)]
Merge pull request #4845 from RussKeldorph/pending
Fix some tests where the only problem was incorrect retval
Lubomir Litchev [Mon, 9 May 2016 01:08:08 +0000 (18:08 -0700)]
Update unix-test-instructions.md
Updated run coreclr tests instructions for Linux. Included the new semicolon separated paths for the coreFxBinDir.
Russ Keldorph [Fri, 6 May 2016 23:27:37 +0000 (16:27 -0700)]
Fix some tests where the only problem was incorrect retval
The following tests return values other than 100 on success. Modify the
harness to expect the values that they return:
```
JIT\jit64\regress\vsw\102754\test1\test1.cmd
JIT\Regression\CLR-x86-JIT\V1-M09\b16102\b16102\b16102.cmd
```
The following test is identical to `test1` in the same directory. Delete
it.
```
JIT\jit64\regress\vsw\102754\test2\test2.cmd
```
`rem_r4` returns a non-Unix-compatible 0xAAAA (>255). Change it to the
standard 100.
```
JIT\IL_Conformance\Old\Conformance_Base\rem_r4\rem_r4.cmd
```
The following were using the non-functional Environment.ExitCode.
Switched to Environment.Exit(Environment.ExitCode):
```
JIT\Methodical\switch\switch1\switch1.cmd
JIT\Methodical\switch\switch10\switch10.cmd
JIT\Methodical\switch\switch11\switch11.cmd
JIT\Methodical\switch\switch2\switch2.cmd
JIT\Methodical\switch\switch3\switch3.cmd
JIT\Methodical\switch\switch4\switch4.cmd
JIT\Methodical\switch\switch5\switch5.cmd
JIT\Methodical\switch\switch6\switch6.cmd
JIT\Methodical\switch\switch7\switch7.cmd
JIT\Methodical\switch\switch8\switch8.cmd
JIT\Methodical\switch\switch9\switch9.cmd
JIT\Methodical\tailcall\_il_dbgrecurse_ep_void\_il_dbgrecurse_ep_void.cmd
JIT\Methodical\tailcall\_il_dbgtest_void\_il_dbgtest_void.cmd
JIT\Methodical\tailcall\_il_relrecurse_ep_void\_il_relrecurse_ep_void.cmd
JIT\Methodical\tailcall\_il_reltest_void\_il_reltest_void.cmd
JIT\Regression\CLR-x86-JIT\V1-M12-Beta2\b65423\b65423\b65423.cmd
JIT\Regression\CLR-x86-JIT\V1.2-M01\b08046\b08046\b08046.cmd
```
Note that `b08046.cmd` is left disabled because it is still failing.
Koundinya Veluri [Sun, 8 May 2016 18:45:14 +0000 (11:45 -0700)]
Merge pull request #4773 from kouvel/CardTableGrowFix3
Fix recently introduced timing issue when growing the card table
SaeHie Park [Sun, 8 May 2016 18:24:26 +0000 (03:24 +0900)]
ARM: Fix SOS stack dump command failure (#4774)
Dumping address pointer for 32bit system needs adjusting for current implementation
while it assumes %p argument is 64bit. This patch changes (ULONG64) casting to SOS_PTR()
so that it will be safe for both 64 and 32bit.
This will fix DumpStack and EEStack segmentation fault problem.
Related issue: #4672
Jan Kotas [Sun, 8 May 2016 12:02:50 +0000 (05:02 -0700)]
Disable flaky test (#4846)
Jan Kotas [Sun, 8 May 2016 04:21:50 +0000 (21:21 -0700)]
RyuJIT - throughput improvements, PInvoke transitions tweaks: (#4842)
* RyuJIT - throughput improvements, PInvoke transitions tweaks:
- Minor compilation throughput improvements:
- Avoid redundant calls to getMethodAttribs in impMarkInlineCandidate by reusing the value from CALL_INFO
- Avoid redundant calls to getFunctionEntryPoint in codegen by reusing the value computed during lower
- Stop checking CORINFO_FLG_NATIVE flag for inliner observations. It is dead flag, not set anywhere
- Tweaks for helper-based PInvoke transitions (used CoreRT only today):
- Stop emitting CORINFO_HELP_INIT_PINVOKE_FRAME. Just having the two simple begin/end helpers is better.
- Stop passing size of arguments in secret register.
- Enable direct calls for PInvoke targets
* Codereview feedback
Aditya Mandaleeka [Sat, 7 May 2016 20:51:11 +0000 (13:51 -0700)]
Merge pull request #4747 from krytarowski/netbsd-support-75
NetBSD: Fix build with LLVM-3.9
Aditya Mandaleeka [Sat, 7 May 2016 20:28:58 +0000 (13:28 -0700)]
Merge pull request #4843 from dotnet-bot/from-tfs
Merge changes from TFS
Jan Kotas [Sat, 7 May 2016 18:00:15 +0000 (11:00 -0700)]
Delete left-over uses of FeatureHostedBinder
[tfs-changeset: 1602735]
Bruce Forstall [Sat, 7 May 2016 15:52:07 +0000 (08:52 -0700)]
Merge pull request #4742 from lemmaa/fix-unit-test-script
Fix 'runtest.sh' script
Sivarv [Sat, 7 May 2016 15:20:57 +0000 (08:20 -0700)]
Merge pull request #4836 from sivarv/simdPromotion
Mark simd type vars as lvRegStruct accurately.
Rahul Kumar [Sat, 7 May 2016 14:21:37 +0000 (07:21 -0700)]
Fixes multidim array Get Accessor for enum datatype (#4808)
James Ko [Sat, 7 May 2016 14:20:18 +0000 (10:20 -0400)]
Use Buffer.BlockCopy in System.Reflection.Emit (#4812)
Thomas Klausner [Sat, 7 May 2016 14:14:55 +0000 (16:14 +0200)]
Fix a typo in NetBSD documentation. (#4841)
James Ko [Sat, 7 May 2016 14:14:06 +0000 (10:14 -0400)]
Buffer.BlockCopy: Avoid double-typechecks for arrays of same type (#4807)
Jan Vorlicek [Sat, 7 May 2016 08:47:59 +0000 (10:47 +0200)]
Remove FEATURE_HOSTED_BINDER definition (#4838)
The FEATURE_HOSTED_BINDER is always on so remove it from all the sources.
Tijoy Tom Kalathiparambil [Tue, 3 May 2016 21:25:22 +0000 (14:25 -0700)]
UTF8 Marshaling support(UnmanagedType.LPUTF8Str)
Usage: [MarshalAs(UnmanagedType.LPUTF8Str)] applied to string
and stringbuilder.
Implementation mostly use Encoding.UTF8 API to do the byte buffer
to string roundtripping. Introducing two new marshalers,
UTF8StringMarshaler and UTF8BufferMarshaler which handle string
and StringBuilder respectively. [Out] StringBuilder marshaling use
builder capacity as the buffer size ie (builder. Capacity + 1) *3
which is enough for any UTF8 char in BMP plane, infact Encoding.UTF8
mscorlib APIs use the same length.All marshaling flags(ThrowOnUnmapable,
defaultchar) are ignored since they do not make sense in UTF16 to UTD8
context.
The public contracts are not yet updated, the public contracts and
public marshaling API (Marshal.PtrToStringUtf8 and StringToHGlobalUtf8)
will be added once the implementation is in. The marshal api are anyway
going to be a wrapper around Encoding.GetBytes and GetChars.
sivarv [Fri, 6 May 2016 21:36:58 +0000 (14:36 -0700)]
Mark simd type vars as lvRegStruct accurately.
Pat Gavlin [Fri, 6 May 2016 23:32:18 +0000 (16:32 -0700)]
Merge pull request #4803 from pgavlin/RootBinDir
Add a `bindir` parameter to build.sh.
Pat Gavlin [Fri, 6 May 2016 23:23:26 +0000 (16:23 -0700)]
Merge pull request #4831 from pgavlin/gh2300
Move b12011 into testsUnsupportedOutsideWindows.
Russ Keldorph [Fri, 6 May 2016 22:57:36 +0000 (15:57 -0700)]
Merge pull request #4826 from RussKeldorph/vararg
Move Windows-only vararg/arglist tests to the appropriate exclusion file
Andy Ayers [Fri, 6 May 2016 22:22:03 +0000 (15:22 -0700)]
Merge pull request #4815 from AndyAyersMS/LegacyPolicyInlineLimit
Inliner: add JitInlineLimit check to LegacyPolicy under debug
Maoni Stephens [Fri, 6 May 2016 21:53:23 +0000 (14:53 -0700)]
Merge pull request #4770 from Maoni0/affi
enable affinity for GC threads and some bug fixes
Jan Vorlicek [Fri, 6 May 2016 21:51:25 +0000 (23:51 +0200)]
Fix Windows x86 exception handling issue (#4830)
This change fixes an exception handling issue that happens on x86 on Windows
when exception is raised in System.RuntimeType.MakeGenericType.
The problem was caused by GCPROTECT_HOLDER macro in RuntimeTypeHandle::GetTypeByName
that causes popping of GCFrame (and zeroing its m_next field) that's in the middle of the thread's frames list during
the stack unwinding.
That breaks the list and when UnwindFrames happen later and tries to walk the stack, the StackFrameIterator::NextRaw
asserts when checking the validity of the list.
The fix is to move the keepAlive to the managed caller of the RuntimeTypeHandle::GetTypeByName QCall, which removes
the need for the GCPROTECT_HOLDER.
Since it was the only usage of that holder and of the underlying FrameWithCookieHolder class, I've removed those.
In addition to that, I've modified COMModule::GetType and AssemblyNative::GetType to use the same pattern, since
they could also suffer from the problem the GCPROTECT_HOLDER was attempting to fix.
Gaurav Khanna [Fri, 6 May 2016 21:28:38 +0000 (14:28 -0700)]
Cleanup sxsJitStartup from codemanager
Sejong Oh [Fri, 6 May 2016 21:25:40 +0000 (14:25 -0700)]
Merge pull request #4690 from sejongoh/use_rax_for_ret_buf
Return RetBuf address via RAX for Windows 64
Sivarv [Fri, 6 May 2016 21:17:50 +0000 (14:17 -0700)]
Merge pull request #4795 from sivarv/simdret
[x64 Unix] Avoid SIMD types returned in two return registers from requiring to go through memory.
Kyungwoo Lee [Fri, 6 May 2016 20:45:36 +0000 (13:45 -0700)]
Merge pull request #4791 from kyulee1/fasttailcall
ARM64: Enable Fast Tail Call
Kyungwoo Lee [Wed, 4 May 2016 20:26:33 +0000 (13:26 -0700)]
ARM64: Enable Fast Tail Call
Fixes https://github.com/dotnet/coreclr/issues/4420
Like x64, enable fast tail call but tail call opt is still off.
This means explicit tail call is converted to epilog + jmp IP0.
Tail call recursive/tail call via helper is still not enabled yet.
Pat Gavlin [Fri, 6 May 2016 20:08:19 +0000 (13:08 -0700)]
Move b12011 into testsUnsupportedOutsideWindows.
This test requires COM interop.
Pat Gavlin [Thu, 5 May 2016 22:13:40 +0000 (15:13 -0700)]
Add a `bindir` parameter to build.sh.
This parameter can be used to redirect the usual output path from
`./bin` to a different directory.
Bruce Forstall [Fri, 6 May 2016 19:48:04 +0000 (12:48 -0700)]
Merge pull request #4814 from BruceForstall/FixGcRefAssert
Fix GC ref frame offset range assert
Bruce Forstall [Fri, 6 May 2016 19:47:11 +0000 (12:47 -0700)]
Merge pull request #4811 from BruceForstall/FixFunctionTrace
Fix COMPlus_JitFunctionTrace=1 to work better with NYI
Alex Ghiondea [Fri, 6 May 2016 19:11:27 +0000 (12:11 -0700)]
Package System.Private.CoreLib in the Microsoft.NETCore.Runtime.CoreCLR package.
Alex Ghiondea [Fri, 6 May 2016 19:09:50 +0000 (12:09 -0700)]
Create project that builds System.Private.CoreLib assembly.
This will contain the same sources as mscorlib but will have a different name.
Bruce Forstall [Fri, 6 May 2016 19:09:12 +0000 (12:09 -0700)]
Merge pull request #4822 from dotnet-bot/from-tfs
Merge changes from TFS
Maoni0 [Wed, 4 May 2016 01:29:46 +0000 (18:29 -0700)]
enable affinity and numa for GC and some bug fixes
double the threshold for balancing onto a remote NUMA node - this was proven to be best with the asp.net benchmark.
enable cpu group
Andy Ayers [Fri, 6 May 2016 18:45:46 +0000 (11:45 -0700)]
Merge pull request #4806 from AndyAyersMS/FixInlineDump
Fix inline dump format issue
Russ Keldorph [Fri, 6 May 2016 14:58:47 +0000 (07:58 -0700)]
Move Windows-only vararg/arglist tests to the appropriate exclusion file
Our position on vararg/arglist is that it is supported on Windows only, so
I'm repurposing #1440 to simply express that position in the exclusion
files rather than removing the tests.
Fixes #1440
John Chen [Fri, 6 May 2016 06:59:13 +0000 (23:59 -0700)]
Merge pull request #4797 from JohnChen0/master
Remove an assert in AssemblySpec::EmitToken
Bruce Forstall [Fri, 6 May 2016 03:21:57 +0000 (20:21 -0700)]
Merge pull request #4810 from BruceForstall/Fix4666
Mark EDX as trashed by the no-gc helper calls
Bruce Forstall [Fri, 6 May 2016 03:07:13 +0000 (20:07 -0700)]
Merge pull request #4819 from BruceForstall/FixRyujitX86CheckedTestRun
Disable tests which fail in a RyuJIT/x86 Checked build test run
Matt Ellis [Fri, 6 May 2016 01:27:21 +0000 (18:27 -0700)]
Merge pull request #4821 from ellismg/fix-unix-builds
Fix conditional in Unix build definition
Matt Ellis [Fri, 6 May 2016 01:12:44 +0000 (18:12 -0700)]
Fix conditional in Unix build definition
Bruce Forstall [Fri, 6 May 2016 00:57:42 +0000 (17:57 -0700)]
Disable tests which fail in a RyuJIT/x86 Checked build test run
Bruce Forstall [Thu, 5 May 2016 23:44:24 +0000 (16:44 -0700)]
Fix COMPlus_JitFunctionTrace=1 to work better with NYI
Bruce Forstall [Fri, 6 May 2016 00:21:27 +0000 (17:21 -0700)]
Fix GC ref frame offset range assert
In emitSetFrameRangeGCRs(), on x86, there is the following assert:
assert(offsHi < 0);
This is saying that the highest offset of a stack frame gc reference
is negative w.r.t. the frame pointer. This is not quite true: the
highest offset is exclusive of the range (the lowest offset is inclusive).
I found a test case where the 'this' pointer, a gc ref, was located
immediately below the frame pointer, in range [ebp-4,ebp), so offsHi==0.
The reason we don't hit this more is because this assert is only in
the JitDump code (under #ifdef DEBUG and 'if (verbose)'). (This
also seems wrong.)
I changed the assert to (offsHi <= 0).
Jeremy Kuhne [Fri, 6 May 2016 00:20:27 +0000 (17:20 -0700)]
This change ports the support files from NetFxDev1. It doesn't change behavior yet. Adding a feature for the app context switch for path compat as we don't intend to use it in core.
[tfs-changeset: 1602284]
Andy Ayers [Thu, 5 May 2016 23:48:53 +0000 (16:48 -0700)]
Inliner: add JitInlineLimit check to LegacyPolicy under debug
Adds the ability to limit the number of inlines done by the LegacyPolicy.
Useful in trying to binary search for inlines that cause correctness
issues.
This limit impacts inlining in all methods, so effective isolation may
also require use of JitNoInlineRange to hone in on the root method that
is the source of problems.
Bruce Forstall [Thu, 5 May 2016 23:35:44 +0000 (16:35 -0700)]
Mark EDX as trashed by the no-gc helper calls
Fixes #4666
Andy Ayers [Thu, 5 May 2016 21:51:33 +0000 (14:51 -0700)]
Fix inline dump format issue
Lost some implict casts to float when I updated SIZE_SCALE to an enum
constant, so now make the casts explicit.
Just impacts dump output.
Also, add a newline to LegacyPolicy dump to keep messages distinct.
Sejong Oh [Fri, 29 Apr 2016 21:05:16 +0000 (14:05 -0700)]
Return RetBuf address via RAX for Windows 64
The AMD64 native calling convention requires return buffer address to be
returned by callee in RAX. This changeset makes JIT follow this rule.
Especially, in case that a managed method returns a struct value to its
native caller, the caller reads a pointer to the return value from RAX.
Asm diffs show expected differences such as 'mov rax, reg' at the end of
functions and some different register assignments.
Jarret Shook [Thu, 5 May 2016 21:18:17 +0000 (14:18 -0700)]
Merge pull request #4800 from jashook/x86_ci_fix
Fix x86 netci changes to build correctly.
sivarv [Thu, 5 May 2016 01:32:46 +0000 (18:32 -0700)]
Avoid SIMD type return values from requiring to go through
memory for returning.
jashook [Thu, 5 May 2016 15:54:21 +0000 (08:54 -0700)]
Fixup netci changes to build correctly.
Also more verbosely show which jit and arch the job is.
Bruce Forstall [Thu, 5 May 2016 16:01:57 +0000 (09:01 -0700)]
Merge pull request #4782 from BruceForstall/NyiLongExTree
RyuJIT/x86: convert SBCG to NYI for multi-level expression trees
Davis Goodin [Thu, 5 May 2016 14:28:41 +0000 (09:28 -0500)]
Merge pull request #4792 from dagood/remove-core_root-workaround
Remove CORE_ROOT workaround
Sung-Jae Lee [Thu, 5 May 2016 14:08:27 +0000 (23:08 +0900)]
Following the review comments
John Chen (CLR) [Thu, 5 May 2016 06:11:15 +0000 (23:11 -0700)]
Remove an assert in AssemblySpec::EmitToken
AssemblySpec::EmitToken has a PRECONDITION that prohibits emitting AssemblyRef
tokens to WinMD files, except inside CrossGen. However, there are scenarios,
such as XML serialization, that need to emit such tokens. Thus this PRECONDITION
is removed.
Jarret Shook [Thu, 5 May 2016 05:08:39 +0000 (22:08 -0700)]
Merge pull request #4790 from jashook/x86_netci_two
Add ryujit x86 pr trigger.
Rama krishnan Raghupathy [Thu, 5 May 2016 02:30:08 +0000 (19:30 -0700)]
Merge pull request #4767 from ramarag/VSD_ResolveWorker_indirection
[Arm64] Initialize indirection cell for delegates
Rama Krishnan Raghupathy [Wed, 4 May 2016 01:16:47 +0000 (18:16 -0700)]
[Arm64] Initialize indirection cell for delegates