Mike Danes [Mon, 2 Jan 2017 09:05:31 +0000 (11:05 +0200)]
Move compare "decomposition" code
In the original implementation long compares were decomposed in LowerCompare and the rest of the work was done in TreeNodeInfoInitCmp, this restores that ordering.
This enables the and-cmp to test transform for the high part comparison of a long comparison on 32 bit. The low part is unfortunately not handled because the CMP and the AND are in different BBs.
Code like
mov eax, dword ptr [edi]
mov edi, dword ptr [edi+4]
and eax, 0xD1FFAB1E
and edi, 0xD1FFAB1E
jne SHORT G_M14457_IG11
test eax, eax
je G_M14457_IG15
becomes
mov eax, dword ptr [edi]
and eax, 0xD1FFAB1E
test dword ptr [edi+4], 0xD1FFAB1E
jne SHORT G_M14457_IG11
test eax, eax
je G_M14457_IG15
Mike Danes [Sun, 1 Jan 2017 18:02:06 +0000 (20:02 +0200)]
Move int/long compare tweak again
It turns out that loop cloning can generate small int/long comparisons as well and then this code can narrow down the comparison to small int. This requires setting the GTF_UNSIGNED flag so this code must run before the code that checks for small int comparisons.
No FX diffs.
To be sure move this long/int comparison tweak before anything else.
Mike Danes [Sun, 1 Jan 2017 14:40:51 +0000 (16:40 +0200)]
Comments
Mike Danes [Sun, 1 Jan 2017 13:58:09 +0000 (15:58 +0200)]
Simplify genCompareInt
- Remove redundancy around cmp/test selection
- Simplify type selection, the case when the types are identical is the most common so it should be first
- Add a bunch of asserts to check tha the selected type is correct.
Mike Danes [Sun, 1 Jan 2017 11:31:31 +0000 (13:31 +0200)]
Move int/long compare tweak to LowerCompare
Normally we should not get int/long compares but the JIT accidentally produces such compares sometimes.
Doing this in lowering requires to add a cast node but since this rarely happens the cost of adding a new node is not a concern.
FX diff shows a 151 bytes improvement without any regressions. All differences are due to narrowing int/long compares to int/int when the long operand is a constant.
Such narrowing usually avoids the need for a REX prefix and in some rare cases also allows a memory operand to be contained.
Mike Danes [Sun, 1 Jan 2017 09:37:52 +0000 (11:37 +0200)]
Move small unsigned type tweak to LowerCompare
This belongs in either LowerCompare or genCompareInt, it has nothing to do with TreeNodeInfoInitCmp.
Since this comparison change doesn't require codegen to actually generate a small compare let's move it to LowerCompare. It already tweaks types in a manner that requires changing the comparison type and having this there removes some redundancy.
No FX diffs.
Mike Danes [Wed, 28 Dec 2016 10:14:21 +0000 (12:14 +0200)]
Introduce OperIs
Checking a GenTree's oper is a common operation so having a shorter way of doing it should be useful, especially when considering the requirement to paranthesize code like x == a || x == b.
Mike Danes [Tue, 20 Dec 2016 21:30:35 +0000 (23:30 +0200)]
Better test narrowing
The previous implementation attemped to narrow down the test mask to have the same type as the memory operand. Appart from unnecessary complications that arise from sign extension the approach doesn't yield the best results in terms of code size.
FX diff shows a 586 bytes improvement without any regressions. All improvements are of the form
test byte ptr [rcx+64], 1
instead of the 3 byte longer
test dword ptr [rcx+64], 1
Mike Danes [Mon, 19 Dec 2016 21:15:01 +0000 (23:15 +0200)]
Handle cmp(and(x, y), 0)
There's no need to require y to be a constant. This restriction was an artifact of the old implementation using containment in lower/codegen to recognize this pattern.
This results in a couple of hundred AND instructions being converted to TEST. This is preferable as TEST can be macro-fused with the conditional branch that may follow it.
Occasionally this also saves a MOV because TEST doesn't update its first operand like AND does.
FX diff shows a 118 bytes improvement without any regressions.
Mike Danes [Mon, 19 Dec 2016 19:53:56 +0000 (21:53 +0200)]
Stop narrowing cmp(cast<ubyte>(icon), 0)
This isn't useful and requires special code to narrow the cast operand. The expression is constant and compiling such an expression shows that JIT does fold it so it doesn't reach lowering.
No FX diffs.
Mike Danes [Mon, 19 Dec 2016 19:12:37 +0000 (21:12 +0200)]
Introduce genTypeValueFitsIn
Mike Danes [Mon, 19 Dec 2016 06:36:02 +0000 (08:36 +0200)]
Handle small memory compares against 0
The previous commit missed the TYP_BOOL case that the original code handled. But then the original code failed to do this thinking that "test" will be used. But that means we end up with a "movzx" as well:
movzx rax, byte ptr [rsi+24]
test eax, eax
instead of just
cmp byte ptr [rsi+24], 0
The Intel manual actually recomends against using the "cmp mem, imm" form but not if other instructions need to be added. This may warrant further investigation though.
FX diff shows a 7579 bytes improvement without any regressions.
Mike Danes [Sun, 18 Dec 2016 11:54:21 +0000 (13:54 +0200)]
Contain "op1" from "cmp op1, icon" only if types are identical
LowerCompare already adjust the types to match when possible so there's no point in comparing type sizes.
No FX diffs.
Mike Danes [Sat, 17 Dec 2016 20:28:07 +0000 (22:28 +0200)]
Remove unused code
Mike Danes [Sat, 17 Dec 2016 19:51:59 +0000 (21:51 +0200)]
Move and-cmp-test transform from TreeNodeInfoInitCmp to LowerCompare
FX diff shows a 32 bytes improvement without any regressions. Previously the first GT_AND operand wasn't marked reg optional and sometimes it ended up being loaded into a register:
mov eax, dword ptr [rsp+B0H]
test eax, 1
Not anymore:
test dword ptr [rsp+B0H], 1
Also, we now generate TEST instead of AND even in cases where the second AND operand isn't a containable constant.
Mike Danes [Sat, 17 Dec 2016 18:49:50 +0000 (20:49 +0200)]
Add GT_TEST_EQ and GT_TEST_NE
Mike Danes [Sat, 17 Dec 2016 18:26:13 +0000 (20:26 +0200)]
Move and-cmp-nonzero transform from TreeNodeInfoInitCmp to LowerCompare
FX diff shows a 34 byte improvement without any regressions. In the original code this transform was unnecessarily blocked by an uncontainable GT_AND bit mask:
mov rax, 0x80000000
and rax, qword ptr [rsp+08H]
mov rdx, 0x80000000
cmp rax, rdx
sete al
now generates
mov rax, 0x8000000000000000
and rax, qword ptr [rsp+08H]
setne al
Mike Danes [Sat, 17 Dec 2016 17:47:17 +0000 (19:47 +0200)]
Remove unused GTF_RELOP_SMALL
Mike Danes [Sat, 17 Dec 2016 17:36:21 +0000 (19:36 +0200)]
Move cast elision from TreeNodeInfoInitCmp to LowerCompare
No FX diffs
Mike Danes [Sat, 17 Dec 2016 16:28:42 +0000 (18:28 +0200)]
Move narrowing from TreeNodeInfoInitCmp to LowerCompare
FX diff shows a 1274 bytes improvement without any regressions. The original code didn't narrow the compare when op2 was 0. Because of this we had
movzx eax, byte ptr [ebx]
test eax, eax
instead of the shorter
cmp byte ptr[ebx], 0
Pat Gavlin [Tue, 17 Jan 2017 17:36:11 +0000 (09:36 -0800)]
Merge pull request #8925 from pgavlin/HeapVerifyIncompatibleTests
Mark a small number of tests heap-verify-incompatible.
Carol Eidt [Tue, 17 Jan 2017 16:59:59 +0000 (08:59 -0800)]
Merge pull request #8908 from CarolEidt/Contained
Don't treat spill temps as contained
Toni Solarin-Sodara [Tue, 17 Jan 2017 16:39:38 +0000 (17:39 +0100)]
Use nameof in place of hardcoded variable names (#8968)
Fixes #8967
chunseoklee [Tue, 17 Jan 2017 14:48:00 +0000 (23:48 +0900)]
Fix armel toolchain setting (#8961)
This PR fixes build fail for armel(debian)
Signed-off-by: chunseoklee <chunseok.lee@samsung.com>
Hyung-Kyu Choi [Tue, 17 Jan 2017 14:45:38 +0000 (23:45 +0900)]
Update description related to ARM rootfs (#8962)
Update description in script and documentation related to ARM rootfs.
1. Change arm-softfp to armel to apply recent changes.
2. Change UbuntuCodeName to LinuxCodeName
For arm and armel, rootfs are available from three different Linux distro.
For arm, Ubuntu is available.
For armel, Debian and Tizen are available.
Signed-off-by: Hyung-Kyu Choi <hk0110.choi@samsung.com>
Hyeongseok Oh [Tue, 17 Jan 2017 12:41:58 +0000 (21:41 +0900)]
[ARM32/Linx] cross-architecture build: restrict build project list (#8918)
* [ARM32/Linx] cross-architecture build: restrict build project list
Restrict build project for ARM32/Linux cross-architecture
We can build these projects with PR #8866 and #8917. (release build)
- crossgen
- clrjit
- protojit
* modify crosscomponents.cmake
Add list items when not Linux, instead remove in Linux
Jonghyun Park [Mon, 16 Jan 2017 11:59:00 +0000 (20:59 +0900)]
[x86/Linux] Port ''ExceptionTracker::UpdateNonvolatileRegisters' (#8915)
mikedn [Mon, 16 Jan 2017 04:49:52 +0000 (06:49 +0200)]
Remove unused mscoree files and code (#8956)
mikedn [Mon, 16 Jan 2017 04:49:18 +0000 (06:49 +0200)]
Remove mscorpe cruft (#8957)
Jan Kotas [Sat, 14 Jan 2017 08:33:15 +0000 (00:33 -0800)]
Merge pull request #8953 from kouvel/InterruptInFinally
Enable thread interrupt in finally blocks, remove some invalid asserts
Koundinya Veluri [Sat, 14 Jan 2017 05:18:34 +0000 (21:18 -0800)]
Remove invalid asserts in threads.cpp, DoAppropriateWaitWorker(int, ...)
These cases are actually possible because Windows APIs are inconsistent in their behavior when invalid handles are concerned. Depending on the invalid handle's value, a multi-wait can error with ERROR_INVALID_HANDLE, and a single-wait on an invalid handle could return WAIT_TIMEOUT.
Sean Gillespie [Sat, 14 Jan 2017 03:06:05 +0000 (19:06 -0800)]
Enable Standalone GC builds in the CI on OSX and Linux (#8950)
Koundinya Veluri [Sat, 14 Jan 2017 00:34:36 +0000 (16:34 -0800)]
Enable thread interrupt in finally blocks
Thread abort was not allowed in finally blocks, but the same logic was being applied to thread interrupt as well. There is nothing special about thread interrupt that requires it to not work in finally blocks.
Cesar Blum Silveira [Sat, 14 Jan 2017 00:09:46 +0000 (16:09 -0800)]
Check if xsave is enabled by OS before calling xgetbv in XmmYmmStateSupport (#8939)
* Check if xsave is enabled by OS before calling xgetbv in XmmYmmStateSupport. Fix #8903
* Add ebx to clobbered registers.
Carol Eidt [Fri, 9 Dec 2016 19:22:41 +0000 (11:22 -0800)]
Don't treat spill temps as contained
Make Contained refer only to whether code is only generated for a node as part of its parent.
Spill temps that are used from memory are not contained, as code is generated for them.
Add methods to indicate whether an operand is used from register, memory (including spill),
or spill specifically, for use by codegen in distinguishing these cases.
Matt Ellis [Fri, 13 Jan 2017 22:15:21 +0000 (14:15 -0800)]
Merge pull request #8938 from ellismg/fix-corefx-hash-and-artifacts-for-ci
Use a fixed version of CoreFX for testing
Jan Vorlicek [Fri, 13 Jan 2017 22:11:54 +0000 (23:11 +0100)]
Fixes to make runtime work on ARM64 Linux (#8947)
This is a result of attempt to bring up CoreCLR on ARM64 Android.
The bring up is on hold now, but I want to check in the changes
that added ARM64 asm helpers and fixed general Linux ARM64 issues.
William Godbe [Fri, 13 Jan 2017 22:06:17 +0000 (14:06 -0800)]
Merge pull request #8934 from wtgodbe/localpkg
Stop build from leaving localpkg cache in src
wtgodbe [Fri, 13 Jan 2017 00:35:45 +0000 (16:35 -0800)]
Stop build from leaving localpkg cache in src
Matt Ellis [Fri, 13 Jan 2017 01:37:15 +0000 (17:37 -0800)]
Use a fixed version of CoreFX for testing
CoreFX is going to be merging changes soon that will break how we
consume them to do our testing. To give us time to react, we'll fix
the version of the repository we build to a commit before the
changes. We'll also download artifacts from a saved build (produced
before the change took place) so the layout is as we expect.
The issue tracking cleaning this up is #8937
Andy Ayers [Fri, 13 Jan 2017 18:21:44 +0000 (10:21 -0800)]
Merge pull request #8935 from AndyAyersMS/FixSmAssert
JIT: remove match accounting from inliner state machine
Koundinya Veluri [Fri, 13 Jan 2017 16:45:57 +0000 (08:45 -0800)]
Fix Mutex.ReleaseMutex exception type for compatibility (#8941)
Dan Moseley [Fri, 13 Jan 2017 04:39:03 +0000 (20:39 -0800)]
Remove #if false blocks (#8933)
Andy Ayers [Fri, 13 Jan 2017 00:59:00 +0000 (16:59 -0800)]
JIT: remove match accounting from inliner state machine
The inliner's code-size estimating state machine keeps count of
matches, but the count was only used in an assert that checked
that the count did not overflow.
The assert fired when jit stress drove the inliner to evaluate a
huge method as a potential inline candidate and the count reached
the overflow value.
This change removes the counting and the related assert.
Closes #8932.
Gaurav Khanna [Fri, 13 Jan 2017 00:19:17 +0000 (16:19 -0800)]
Merge pull request #8931 from dotnet-bot/master-UpdateDependencies
Update CoreClr to beta-24912-04 (master)
Carol Eidt [Thu, 12 Jan 2017 23:36:04 +0000 (15:36 -0800)]
Merge pull request #8928 from CarolEidt/Fix8898
Fix putArgStk dstCount and ConsumeReg errors
sandreenko [Thu, 12 Jan 2017 22:03:34 +0000 (14:03 -0800)]
Merge pull request #8806 from sandreenko/array-bounds-check
Сhange children order in GenTreeBoundsChk. Fix #8077
mikedn [Thu, 12 Jan 2017 20:59:10 +0000 (22:59 +0200)]
Prefer using Array.Length as upper for loop limit (#8923)
The JIT can't eliminate range checks if it can't "see" Length and uses loop cloning which generates a lot of code. Even in cases where not all range checks can be eliminated and loop cloning is used anyway it's still preferable to have fewer range checks.
For example, SortExceptions is ~140 bytes shorter after this change, despite the fact that loop cloning is still being used.
dotnet-bot [Thu, 12 Jan 2017 20:57:32 +0000 (20:57 +0000)]
Update CoreClr to beta-24912-04
Carol Eidt [Thu, 12 Jan 2017 20:15:16 +0000 (12:15 -0800)]
Fix putArgStk dstCount and ConsumeReg errors
The LastConsumedNode used in genCheckConsumeNode was not initialized for arm64.
Fixing this exposed several places where nodes were being consumed twice or in the wrong order.
In addition, since GT_PUTARG_STK doesn't define a register, its dstCount needs to be zero. This is enabled by checking IsValue() instead of type of TYP_VOID for the default case of TreeNodeInfoInit. This was missed for both arm and arm64.
Pat Gavlin [Thu, 12 Jan 2017 18:41:54 +0000 (10:41 -0800)]
Mark a small number of tests heap-verify-incompatible.
These tests simply take too long to execute when run with heap
verification.
Gaurav Khanna [Thu, 12 Jan 2017 18:11:44 +0000 (10:11 -0800)]
Merge pull request #8909 from gkhanna79/Docker1404
Add optional argument to skip unmount for rootfs
Carol Eidt [Thu, 12 Jan 2017 16:54:49 +0000 (08:54 -0800)]
Merge pull request #8840 from CarolEidt/Fix8824
Fix initialization of resolution sets
Jonghyun Park [Thu, 12 Jan 2017 15:43:54 +0000 (00:43 +0900)]
[x86/Linux] Port 'ExceptionTracker::ProcessOSExceptionNotification' (#8912)
Sivarv [Thu, 12 Jan 2017 07:11:05 +0000 (23:11 -0800)]
Merge pull request #8588 from litian2025/AVX_SSE
Remove AVX/SSE transition penalties
Li Tian [Thu, 12 Jan 2017 05:08:11 +0000 (21:08 -0800)]
fix format
dotnet bot [Thu, 12 Jan 2017 04:55:50 +0000 (20:55 -0800)]
Update CoreClr, CoreFx to beta-24911-02, beta-24911-08, respectively (#8855)
Li Tian [Thu, 12 Jan 2017 04:47:48 +0000 (20:47 -0800)]
Add more comments to document remove AVX/SSE transition penalty logic
Joseph Tremoulet [Thu, 12 Jan 2017 02:24:40 +0000 (21:24 -0500)]
Merge pull request #8905 from JosephTremoulet/OneMoreHeapUse
Remove bogus fgCurHeapDef guard
Russell C Hadley [Thu, 12 Jan 2017 01:19:58 +0000 (17:19 -0800)]
Merge pull request #8879 from russellhadley/span-intrinsics
Initial implementation of two field span.
Carol Eidt [Sat, 7 Jan 2017 02:21:31 +0000 (18:21 -0800)]
Fix initialization of resolution sets
This was causing a failure during dumping.
Fixing it exposed another dumping failure in crossgen of System.Private.Corelib.dll on x86, due to a dead node. I've fixed the assert, but I believe the dead node should have been eliminated. I filed issue #8839 for this.
Mike McLaughlin [Thu, 12 Jan 2017 00:40:49 +0000 (16:40 -0800)]
Fix issue #8139 amd64 decoder bug. (#8901)
Ignore modrm == 0 which ignores "inc dword ptr []" instructions.
Also fixed minor EEHeap formatting bug on Linux.
Drew Scoggins [Thu, 12 Jan 2017 00:36:41 +0000 (16:36 -0800)]
Merge pull request #8907 from DrewScoggins/PullRequestFix
Fix pull request title automation
Maarten Mensink [Thu, 12 Jan 2017 00:28:18 +0000 (01:28 +0100)]
Add clang 3.7 support and remove __FakePrologName to fix build (#8904)
* add clang 3.7 support
* Removing
__FakePrologName="DelayLoad_Helper\suffix\()_FakeProlog"
based on https://github.com/dotnet/coreclr/issues/4332#issuecomment-
271990909
Drew Scoggins [Thu, 12 Jan 2017 00:24:43 +0000 (16:24 -0800)]
Fix pull request title automation
When I added the ability to give a custom name to a run that was kicked
off using the PR leg I did it by adding an additional parameter. When
using a variable name there I need to use the dollar sign syntax of
groovy and not an environment variable.
Gaurav Khanna [Thu, 12 Jan 2017 00:03:51 +0000 (16:03 -0800)]
Add optional argument to skip unmount for rootfs
Pat Gavlin [Wed, 11 Jan 2017 23:44:28 +0000 (15:44 -0800)]
Merge pull request #8897 from pgavlin/GH8860.1
Fix #8860.
Pat Gavlin [Wed, 11 Jan 2017 23:15:31 +0000 (15:15 -0800)]
Format code.
Pat Gavlin [Wed, 11 Jan 2017 23:07:44 +0000 (15:07 -0800)]
Merge pull request #8899 from pgavlin/FixTotalMemoryTest
Fix the GC total memory test.
Michelle McDaniel [Wed, 11 Jan 2017 23:06:30 +0000 (15:06 -0800)]
Merge pull request #8900 from adiaaida/fix8550
Copy lvExactSize to shadowVar
Li Tian [Wed, 11 Jan 2017 22:48:20 +0000 (14:48 -0800)]
use assert insteaf of if statement
Jonghyun Park [Wed, 11 Jan 2017 22:22:00 +0000 (07:22 +0900)]
Rewrite CalleeSavedRegisters for x86 (#8871)
This commit revises CalleeSavedRegisters for x86 using
ENUM_CALLEE_SAVED_REGISTERS as in x64.
Jan Kotas [Wed, 7 Dec 2016 05:30:54 +0000 (21:30 -0800)]
Initial implementation of two field span.
Implements two field span struct which is comprised of a byref field
that may be an interior pointer to a managed object, or a native
pointer indicating the start of the span, and a length field which
describes the span of access.
Since there is no MSIL operation which assign a byref field, the jit
gets involved and treats the constructor and getter of a special struct
called ByReference that contains an declared IntPtr. This special
struct is then used as a field in the span implementation and recognized
by the runtime as a field that may contain a GC pointer. In
implementation, the ctor of ByReference is converted into an assignment
value is returned by a reverse assignment.
Since there are some dependencies on CoreFX for the span implementation
local testing of the implementation has been done using the
BasicSpanTest.cs in the CoreCLR tests. Once this is checked in and is
propagated to CoreFX the apporopate code in the packages will be enabled
and then may be referenced in CoreCLR tests. At that time more span
tests will be added.
Additional comments and fixes based on code review added.
Sven Boemer [Wed, 11 Jan 2017 21:32:56 +0000 (13:32 -0800)]
Merge pull request #8570 from sbomer/dasm_automation
Fix some problems with the ubuntu dasm output
Andy Ayers [Wed, 11 Jan 2017 21:20:57 +0000 (13:20 -0800)]
Merge pull request #8551 from AndyAyersMS/FinallyCloning
Finally cloning
Joseph Tremoulet [Wed, 11 Jan 2017 20:59:29 +0000 (15:59 -0500)]
Remove bogus fgCurHeapDef guard
We should set fgCurHeapUse when visiting a heap use regardless of whether
fgCurHeapDef is set, because we haven't proven that the def writes all
memory that the use might read. This should have been included in #8757
but somehow slipped through.
Michelle McDaniel [Wed, 11 Jan 2017 20:48:55 +0000 (12:48 -0800)]
Copy lvExactSize to shadowVar
For TYP_SIMD12 vars, we need to copy the lvExactSize to the shadowVar
so that we have the right size in lower.
Fixes #8550.
Pat Gavlin [Wed, 11 Jan 2017 20:34:21 +0000 (12:34 -0800)]
Fix the GC total memory test.
This test was relying upon the result of GC.GetTotalMemory() returning a
greater number after allocating a large array. However, the array was
not kept live past the call to GC.GetTotalMemory, which resulted in this
test failing under GCStress=0xC (in which a GC is performed after each
instruction). This change adds the requisite call to GC.KeepAlive to
keep the allocated array live enough to be observed by
GC.GetTotalMemory.
Pat Gavlin [Wed, 11 Jan 2017 19:42:30 +0000 (11:42 -0800)]
Fix #8860.
These failure was an assertion under GC stress due to registers that
were reported as live while containing an invalid GC reference. The root
cause of this assertion was the use of SIMD instructions that did not
kill any GC ref contained in their destination register. This change fixes
the bug by properly killing any GC ref in the destination register for
such an instruction, which required two changes:
- Instructions with the format `IF_RRW_RRW_CNS` were never killing the
destination register. The code that handles this format has been
updated to do so when necessary.
- `pextrw` and `pmovmskb` were not handled correctly in
`emitInsCanOnlyWriteSSE2OrAVXReg`. These instructions may write to
GPRs, so that function must return false when they are passed as
input.
Joseph Tremoulet [Wed, 11 Jan 2017 19:47:11 +0000 (14:47 -0500)]
Merge pull request #8880 from JosephTremoulet/SimplifyHeapPhiArg
Simplify HeapPhiArg
Thiago Camargos Lopes [Wed, 11 Jan 2017 19:12:22 +0000 (17:12 -0200)]
Add parameterless overloads of TrimStart/TrimEnd (#8834)
Add parameterless overloads of TrimStart/TrimEnd
Stephen Toub [Wed, 11 Jan 2017 17:35:06 +0000 (12:35 -0500)]
Merge pull request #8892 from stephentoub/enumeq_ser
Fix deserialization of EnumEqualityComparer
Hyeongseok Oh [Wed, 11 Jan 2017 14:37:21 +0000 (23:37 +0900)]
[ARM32/Linux] Fix cross-architecture build error: T_CONTEXT (#8888)
confusion of T_CONTEXT and CONTEXT in projects
- cee_dac
- gc_wks
related issue: #8885
Stephen Toub [Wed, 11 Jan 2017 13:58:53 +0000 (08:58 -0500)]
Fix deserialization of EnumEqualityComparer
Its deserialization ctor is getting removed by the rewriter, causing deserialization of its instances to fail.
Stephen Toub [Wed, 11 Jan 2017 13:52:58 +0000 (08:52 -0500)]
Merge pull request #8868 from stephentoub/linkedcts_array
Remove array allocation from CTS.CreateLinkedTokenSource
Li Tian [Wed, 11 Jan 2017 07:32:57 +0000 (23:32 -0800)]
fix format error
Li Tian [Wed, 11 Jan 2017 06:09:47 +0000 (22:09 -0800)]
fix comments, assertion failure in crossgen mscorlib
Andy Ayers [Thu, 1 Dec 2016 02:16:17 +0000 (18:16 -0800)]
JIT: Finally Optimizations
Adds two optimization for try-finallys: empty finally removal and
finally cloning.
Empty finally removal identifies trivially empty finally clauses and
removes the entire try-finally EH region. COde in the try is "promoted"
to be in the parent EH region (or method region). Empty finallys often
appear after inlining empty Dispose methods. Removing a try-finally with
an empty finally both reduces code size and improves code speed.
Finally cloning duplicates the code for the finally and 'inlines' it
along one of the normal exit paths from the try. This improves code
speed in the typical case where there is no exception raised while
the try is active. It generally increases code size slightly. However,
finallys are rare enough that the overall code size increase across
all methods is quite small. The jit will clone most finallys, provided
they are not too large, and are not contained in or contain other EH
constructs. If a try contains multiple exit paths only the final "fall
through" path will be optimized.
These optimizations are enabled for all target architectures. Finally
cloning is currently disabled for desktop CLR because more work is needed
to support thread abort.
More details on both optimizations can be found in the design document
added as part of this commit.
In debug builds, finally cloning can be selectively disabled or enabled
by setting COMPlus_JitEnableFinallyCloning to 0 or 1 respectively. This
config setting can thus be used override the default behavior (cloning
enabled for CoreCLR, disabled otherwise) for diagnostic or testing purposes.
Closes #1505. Closes #8065.
sandreenko [Wed, 11 Jan 2017 01:12:52 +0000 (17:12 -0800)]
Merge pull request #8862 from sandreenko/ReversePInvoke-encoding-in-gc
SetReversePInvokeFrameSlot into gcInfo for reversePInvoke
Sedar Gokbulut [Wed, 11 Jan 2017 00:54:16 +0000 (16:54 -0800)]
Add Microsoft.NETCore.Native package (2nd attempt) (#8882)
* Adding Microsoft.NETCore.Native package (#8791)
* "Adding Microsoft.NETCore.Native package"
* Adding generic linux configuration to Microsoft.NETCore.Native package
* Fixing the OSX pkgproj of Microsoft.NETCore.Native to include S.Globalization.Native.dylib and not .so
* Fixing the osx release build.
Li Tian [Wed, 11 Jan 2017 00:52:34 +0000 (16:52 -0800)]
rename, use getSIMDInstructionSet()
Jonghyun Park [Wed, 11 Jan 2017 00:48:43 +0000 (09:48 +0900)]
[x86/Linux] Fix WIN64EXCEPTIONS build error (#8629)
* Move GetUnwindInfo and GetNumberOfUnwindInfos into the real code header
This commit fixes #8342.
* Use WIN64EXCEPTIONS instead of _TARGET_X86_
* Revise FaultingExceptionFrame
This commit revises FaultingExceptionFrame to support WIN64EXCEPTIONS in
x86/Linux port.
* Add RUNTIME_FUNCTION__EndAddress as NYI
* Revise regdisp.h
* Revise eetwain.h
* Comment out exinfo.cpp if WIN64EXCEPTIONS is defined
* Revises excep.cpp
* Fix mistmatch in ThrowControlForThread defintion
* Revises cgenx86.cpp
* Disable SEH-based exception handlers when WIN64EXCEPTIONS is defined
* Revise stackwalk.cpp
* Revise jitinterface.cpp
* Revise readytorun.h
* Revise dbgipcevents.h
* Revise zapcode.cpp
* Revise clrnt.h
* Fix Windows build error
* Mark FaultingExceptionFrame::UpdateRegDisplay as NYI
* Revise per feedback
* Revert #if defined(..) as #ifdef
* Fix style changes
* Fix style changes
* Remove #undef _TARGET_X86_
* 2nd attempt to fix Windows build error
* Revise per feedback
* Revert the chagnes in clrdefinitions.cmake and add BIT32 in CMakeLists.txt
* Use !BIT64 instead of BIT32
* Include exceptionhandling.cpp and gcinfodecoder.cpp in build
This commit includes exceptionhandling.cpp and gcinfodecoder.cpp in
build, and fixes related compile errors.
* Fix COMPlus_EndCatch undefined reference
* Fix build error
* Fix GcInfoDecoder-related undefined references
* Fix AdjustContextForVirtualStub undefined reference
* Fix GetCallerSP undefined reference
* Fix ResetThreadAbortState undefined reference
* Attempt to fix Windows build error
* Fix CLRNoCatchHandler undefined reference
* Another attemp to fix Windows build error
* Fix GetXXXFromRedirectedStubStackFrame undefined references
* Fix Windows Build Error
* Add RtlpGetFunctionEndAddress and RtlVirtualUnwind as NYI
* Fix undefined references on JIT helpers
* Enable Dummy Application Run with WIN64EXCEPTIONS
* Revert "Move GetUnwindInfo and GetNumberOfUnwindInfos into the real code header"
This reverts commit
c2bad85ac1136be3c6fb6ad7eedc5b3814b2ab29.
* Use indirect code header when WIN64EXCEPTIONS is enabled
* Port 'SyncRegDisplayToCurrentContext' and 'FillRegDisplay'
* Revise style 'RUNTIME_FUNCTION__SetUnwindInfoAddress'
* Extract out HandlerData from #ifdef region
* Add UNIXTODO
* Add UNIXTODO
* Port 'GetRegdisplayReturnValue'
* Fix incorrect comment
* Remove messages that mentions WIN32EXCEPTIONS
* Revise AdjustContextForWriteBarrier
* Port 'FaultingExceptionFrame::UpdateRegDisplay'
* Extract out 'AdjustContextForVirtualStub' and 'CLRNoCatchHandler' from #ifdef region
* Merge two #ifdef regions
* Set WIN64EXCEPTIONS as a default for x86/Linux
* Remove unnecessary #ifdef from ThrowControlForThread
* Remove unnecessary stubs
* Add Dependency Check between Compile Flags
* Revise per feedback
Carol Eidt [Wed, 11 Jan 2017 00:20:41 +0000 (16:20 -0800)]
Merge pull request #8881 from CarolEidt/FixPutArgStk
Fix handling of PutArgStk
SaeHie Park [Tue, 10 Jan 2017 23:43:23 +0000 (08:43 +0900)]
[x86/Linux] Fix generic context(hidden arg) position (#8848)
This fixes how to handle generic context position in x86 by num of arguments
SaeHie Park [Tue, 10 Jan 2017 23:42:12 +0000 (08:42 +0900)]
[x86/Linux] Introduce UNIX_X86_ABI definition (#8863)
Add UNIX_X86_ABI definition for Unix/Linux specific ABI parts
First will be for 16 byte stack alignment codes
Carol Eidt [Tue, 10 Jan 2017 22:19:46 +0000 (14:19 -0800)]
Updates based on PR review.
Li Tian [Tue, 10 Jan 2017 22:07:52 +0000 (14:07 -0800)]
fix format error
Pat Gavlin [Tue, 10 Jan 2017 22:04:12 +0000 (14:04 -0800)]
Merge pull request #8876 from pgavlin/LVAThroughput
Do not rerun LVA if only live-out has changed.
Carol Eidt [Tue, 10 Jan 2017 21:55:47 +0000 (13:55 -0800)]
Fix handling of PutArgStk
GT_PUTARG_STK doesn't produce a value, so it should have the GTK_NOVALUE flag set.
Although the dstCount was being set to zero by the parent call, localDefUse was also being set, causing a register to be allocated. Fixing this produces a number of improvements due to reuse of constant registers that were otherwise unnecessarily "overwritten" by the localDefUse.
Also on x86, GT_LONG shouldn't be used to pass a long, since GT_LONG should always be a value-producing node. Instead, use the existing GT_FIELD_LIST approach.
Andy Ayers [Tue, 10 Jan 2017 21:42:32 +0000 (13:42 -0800)]
Merge pull request #8867 from AndyAyersMS/BinaryTreesVariant
JIT: Add BinaryTrees benchmark variant