Vance Morrison [Thu, 13 Oct 2016 20:10:12 +0000 (13:10 -0700)]
Added Docs for Using Visual Studio to Edit and Debug in CoreCLR Repo (#7582)
Jan Kotas [Thu, 13 Oct 2016 20:09:00 +0000 (13:09 -0700)]
Merge pull request #7600 from jkotas/master-UpdateDependencies
Update CoreClr, CoreFx to beta-24612-03, beta-24612-04, respectively
Pat Gavlin [Thu, 13 Oct 2016 19:38:43 +0000 (12:38 -0700)]
Merge pull request #7593 from pgavlin/DisableGCStressTest
Disable a test that is not compatible with GC stress.
Jan Kotas [Thu, 13 Oct 2016 08:11:47 +0000 (01:11 -0700)]
Update tests to netcoreapp1.1
Pat Gavlin [Wed, 12 Oct 2016 22:13:52 +0000 (15:13 -0700)]
Disable a test that is not compatible with GC stress.
This test fails under GCStress=0xC for all of RyuJIT/x86, legacy backend
x86, and JIT32.
Koundinya Veluri [Thu, 13 Oct 2016 01:39:30 +0000 (18:39 -0700)]
Merge pull request #7510 from kouvel/ApiThread
Expose some members of the Thread class
Jan Kotas [Thu, 13 Oct 2016 00:55:53 +0000 (17:55 -0700)]
Merge pull request #7595 from dotnet-bot/from-tfs
Merge changes from TFS
Sivarv [Wed, 12 Oct 2016 23:45:20 +0000 (16:45 -0700)]
Merge pull request #7584 from sivarv/regressionTest
Adding JIT regression test for the re-morph optimzation that transforms JIT_Stelem_Ref(a, null, i) into a[i]=null.
smile21prc [Wed, 12 Oct 2016 23:23:04 +0000 (16:23 -0700)]
Merge pull request #7590 from smile21prc/coreclr-perf
Permanently fix benchview package version, and add code to produce perf-*-summary.xml on Linux
Carol Eidt [Wed, 12 Oct 2016 22:56:23 +0000 (15:56 -0700)]
Merge pull request #7563 from wateret/fix-7398
[ARM/Linux] Fix HFA unit test regression
Jan Kotas [Wed, 12 Oct 2016 22:26:51 +0000 (15:26 -0700)]
Cleanup files excluded from CoreCLR build
[tfs-changeset: 1633047]
Koundinya Veluri [Wed, 12 Oct 2016 22:06:42 +0000 (15:06 -0700)]
Address feedback
Carol Eidt [Wed, 12 Oct 2016 22:05:51 +0000 (15:05 -0700)]
Merge pull request #7585 from CarolEidt/Fix5699
x86: adjust stack level at throw blocks
Carol Eidt [Wed, 12 Oct 2016 18:24:18 +0000 (11:24 -0700)]
x86: adjust stack level at throw blocks
When we don't have a frame pointer, we need to adjust the frame pointer before calling the throw helper.
I initially had some concern that we had some exposure from instruction reordering in the backend, but we don't do any reordering of the `GT_PUTARG_STK` nodes. I have added a note to sideeffects.h to call attention to the fact that we need to consider stack depth changes to be side effects, for future consideration should we start doing more general code motion.
For the more general issue, I have extracted the code from codegenlegacy.cpp that does the stack adjustment, and am now unconditionally calling it from both codegenlegacy.cpp and codegenxarch.cpp.
Fix #5699
Jan Kotas [Wed, 12 Oct 2016 21:39:43 +0000 (14:39 -0700)]
Return false from IsMarshalByRefImpl/IsContextfulImpl when remoting is not supported (#7583)
Remove references to types have been moved to corefx in the build
Consolidate FEATURE_CORECLR ifdefs
Sean Gillespie [Wed, 12 Oct 2016 21:38:37 +0000 (14:38 -0700)]
Fail fast on a couple flaky tests to get crash dumps (#7588)
John Chen [Wed, 12 Oct 2016 21:02:45 +0000 (14:02 -0700)]
Fix a tailcall test that fails with Ready-to-Run (#7586)
The test case JIT/opt/Tailcall/TailcallVerifyWithPrefix has some methods that
depend on the JIT compiler doing automatic tail call. Since tail is not
supported with Ready-to-Run, the test case is modified to exclude such methods
from Ready-to-Run images.
James Ko [Wed, 12 Oct 2016 21:02:21 +0000 (17:02 -0400)]
Port NullStream.CopyTo override from corert (#7589)
dotnet-bot [Wed, 12 Oct 2016 20:30:02 +0000 (20:30 +0000)]
Update CoreClr, CoreFx to beta-24612-03, beta-24612-04, respectively
sandreenko [Wed, 12 Oct 2016 20:16:55 +0000 (13:16 -0700)]
Merge pull request #7557 from sandreenko/long_decomp_6925
Use the existing operands for long decomposition #6925
Vance Morrison [Wed, 12 Oct 2016 19:37:12 +0000 (12:37 -0700)]
Run VS formatting (no semantic change) (#7587)
Smile Wei [Wed, 28 Sep 2016 18:49:59 +0000 (11:49 -0700)]
Debugging
James Ko [Wed, 12 Oct 2016 19:07:37 +0000 (15:07 -0400)]
Move StreamHelpers from corefx and refactor Stream and MemoryStream to use it (#7579)
Koundinya Veluri [Tue, 11 Oct 2016 22:31:12 +0000 (15:31 -0700)]
Expose members of Thread that hide base class members
Koundinya Veluri [Tue, 11 Oct 2016 04:10:31 +0000 (21:10 -0700)]
Fix build issue
Koundinya Veluri [Mon, 10 Oct 2016 21:02:46 +0000 (14:02 -0700)]
Fix build issue
Koundinya Veluri [Sat, 8 Oct 2016 21:35:02 +0000 (14:35 -0700)]
Address feedback
Koundinya Veluri [Tue, 27 Sep 2016 20:14:24 +0000 (13:14 -0700)]
Expose some members of the Thread class
- Added RuntimeThread class that Thread derives from. An instance of RuntimeThread will be used by the Thread class in CoreFX to forward thread functionality to the runtime.
- Some functionality in the RuntimeThread could potentially be moved to CoreFX. In particular, some of thread creation and apartment state handling could probably be moved with some minimal help from the runtime. Leaving that refactoring for later.
- A few things are left unimplemented (for instance, apartment state handling doesn't work currently). That functionality will be enabled in a future change.
Part of dotnet/corefx#11632, dotnet/corefx#11638, dotnet/corefx#11635
Sergey Andreenko [Fri, 7 Oct 2016 21:19:15 +0000 (14:19 -0700)]
Unnecessary temp loc vars were deleted in DecomposeLong.
Delete temp local variables from DecomposeNeg, DecomposeShift.
Remain local variables have several uses(DecomposeShift) or are call
arguments. Such variables can't be deleted.
sivarv [Wed, 12 Oct 2016 17:49:03 +0000 (10:49 -0700)]
Adding JIT regression test for a re-morph optimization.
Daniel Cazzulino [Wed, 12 Oct 2016 17:37:46 +0000 (19:37 +0200)]
Provide KeyValuePair.Create static factory method (#5755)
Just like Tuple.Create cuts down the noise and annoyance
of having to specify all the Ts when you already have typed
values to provide, a similarly designed KeyValuePair.Create
factory method would make code that creates those pairs
(i.e. public APIs that expose IEnumerable{KeyValuePair}
significantly more concise and readable.
Fixes dotnet/corefx#2127
sandreenko [Wed, 12 Oct 2016 17:35:29 +0000 (10:35 -0700)]
Merge pull request #7576 from sandreenko/check_ready_to_run_for_protojit
set ready to run flag for protojit
Rama krishnan Raghupathy [Wed, 12 Oct 2016 17:34:09 +0000 (10:34 -0700)]
Exposing MarshalByRefObject (#7558)
Sivarv [Wed, 12 Oct 2016 17:21:32 +0000 (10:21 -0700)]
Merge pull request #7575 from sivarv/lsraStat
Dump LSRA stats.
Tarek Mahmoud Sayed [Wed, 12 Oct 2016 16:29:55 +0000 (09:29 -0700)]
Enable TextInfo APIs for netstandard 1.7 on Linux (#7568)
* Enable TextInfo APIs for netstandard 1.7 on Linux
The change here is to enable the newly added netstandard 1.7 TextInfo APIs on Linux.
- moved the locale specific data to its own file Locale.Unic.cs
- added needed data to handle the TextInfo properties (e.g. codepages)
- created some helper methods in LocaleData classto handle accessing the locale data
- added the implementation for added APIs
* Fix some typos
Bruce Forstall [Wed, 12 Oct 2016 15:15:25 +0000 (08:15 -0700)]
Merge pull request #7561 from BruceForstall/Fix7483
Fix assertion regarding byteable reg when ngen'ing desktop mscorlib
Bruce Forstall [Tue, 11 Oct 2016 01:10:16 +0000 (18:10 -0700)]
Fix assertion regarding byteable xor reg,reg when ngen'ing desktop mscorlib
Fixes #7483
The issue is that TreeNodeInfoInitCmp() will, under certain circumstances,
perform a tree transformation of a child node into a TYP_UBYTE type. Since
the TreeNodeInfo walk is bottom-up, its children have already been processed,
and they don't get the "byteable" register processing that is as the end of
TreeNodeInfoInit().
I extracted that processing into a TreeNodeInfoInitCheckByteable() function that
can be called after the TreeNodeInfoInitCmp() transformation.
A better solution might be to extract this, and other, tree transformations into
the prior "lowering" pass, such that TreeNodeInfoInit() and friends will only
do register requirement annotation. But that would be considerably more complicated,
so I opted not to do that for now.
I also fixed up a bunch of comments.
Jan Kotas [Wed, 12 Oct 2016 12:53:28 +0000 (05:53 -0700)]
Fix conflicting identity of System.IO.TextReader (#7578)
System.IO build was switched to use CoreFX copy of System.IO.TextReader and friends. These type have to be removed from CoreLib now to avoid type identity mismatches.
Sergey Andreenko [Wed, 12 Oct 2016 01:16:34 +0000 (18:16 -0700)]
set ready to run flag for protojit
sivarv [Tue, 11 Oct 2016 23:20:39 +0000 (16:20 -0700)]
Dump LSRA stats.
Pat Gavlin [Tue, 11 Oct 2016 23:41:21 +0000 (16:41 -0700)]
Merge pull request #7571 from pgavlin/MkRefAnyFieldTypeX86
Fix the type of the data field of a typed ref on x86.
Rahul Kumar [Tue, 11 Oct 2016 23:05:04 +0000 (16:05 -0700)]
Merge pull request #7574 from DrewScoggins/runtestfix
Fix runtest.cmd return code
Drew Scoggins [Tue, 11 Oct 2016 22:51:49 +0000 (15:51 -0700)]
Fix runtest.cmd return code
When running the GenerateLayoutOnly command of runtest.cmd the script
should return zero instead of one.
Pat Gavlin [Tue, 11 Oct 2016 22:05:03 +0000 (15:05 -0700)]
Fix the type of the data field of a typed ref on x86.
This field should be of type byref; not treating it as such causes GC
holes when passing typed references.
Carol Eidt [Tue, 11 Oct 2016 20:55:33 +0000 (13:55 -0700)]
Merge pull request #7310 from wateret/for-7002-2
[ARM] Generate direct call instructions for recursive calls
Pat Gavlin [Tue, 11 Oct 2016 20:28:43 +0000 (13:28 -0700)]
Merge pull request #7565 from dotnet-bot/from-tfs
Merge changes from TFS
Pat Gavlin [Tue, 11 Oct 2016 18:25:40 +0000 (11:25 -0700)]
Merge pull request #7560 from pgavlin/StoreBlkGCHole
Disable emit{Enable,Disable}GC for x86.
ikopylov [Tue, 11 Oct 2016 17:02:38 +0000 (20:02 +0300)]
SemaphoreSlim performance improvement (#137)
* Waiters notification by the value of releaseCount (reduce the number of false-wakeups).
Vance Morrison [Tue, 11 Oct 2016 16:10:11 +0000 (09:10 -0700)]
Update Readme.md to lead you through basic repo workflow (#7552)
Update README.md to contain much more information about using the repo
Pat Gavlin [Tue, 11 Oct 2016 15:55:09 +0000 (08:55 -0700)]
Fix a build break in the ARM64 JIT.
[tfs-changeset: 1632696]
Pat Gavlin [Tue, 11 Oct 2016 15:26:02 +0000 (08:26 -0700)]
Merge pull request #7562 from pgavlin/x86PreemptiveAssert
Remove an overly fragile assertion.
Hanjoung Lee [Tue, 11 Oct 2016 05:45:46 +0000 (14:45 +0900)]
[ARM/Linux] Fix HFA unit test regression
Fix #7398
Pat Gavlin [Tue, 11 Oct 2016 04:21:23 +0000 (21:21 -0700)]
Remove an overly fragile assertion.
The GC stress infrastructure contained an assertion that was meant to
help detect managed code running in preemptive mode. Unfortunately, this
assertion was derived from the set of instructions emitted by the JIT at
the time at which it was added. As such, it is now out of date.
Because a failure of this assertion does not necessarily indicate a
correctness bug and because it is onerous to keep up-to-date, remove it.
Michal Strehovský [Tue, 11 Oct 2016 01:56:09 +0000 (18:56 -0700)]
Remove AllowPartiallyTrustedCallers from a test (#7555)
Partial trust is not part of .NET Core.
Pat Gavlin [Tue, 11 Oct 2016 00:09:29 +0000 (17:09 -0700)]
Disable emit{Enable,Disable}GC for x86.
The GC info encoding for this platform does not permit marking
arbitrary regions of generated code as non-interruptible.
Michelle McDaniel [Tue, 11 Oct 2016 00:21:07 +0000 (17:21 -0700)]
Merge pull request #7518 from adiaaida/cmpByteable
Force byteable register for CMP if op2 is CNS_INT
Jan Kotas [Tue, 11 Oct 2016 00:06:45 +0000 (17:06 -0700)]
Update adding_new_public_apis.md (#7544)
Pat Gavlin [Tue, 11 Oct 2016 00:00:38 +0000 (17:00 -0700)]
Merge pull request #7537 from pgavlin/PushStructArgsX86
Fix the codegen for by-value struct args on x86.
Michelle McDaniel [Fri, 7 Oct 2016 15:32:17 +0000 (08:32 -0700)]
Force byteable register for CMP if op2 is CNS_INT
On x86 we need to force byteable registers for compares not only when both
ops are typ_byte, but also when the first is of typ_byte and the second is
a constant int. The original condition for cmp only considered the first
case. This change adds the second condition.
Pat Gavlin [Mon, 10 Oct 2016 20:30:00 +0000 (13:30 -0700)]
Address PR feedback.
Rama krishnan Raghupathy [Mon, 10 Oct 2016 21:18:26 +0000 (14:18 -0700)]
Merge pull request #7530 from ramarag/StrongNameKeyPair
Strong name key pair
Rama Krishnan Raghupathy [Fri, 7 Oct 2016 23:36:46 +0000 (16:36 -0700)]
Exposing StrongNameKeyPair
Joseph Tremoulet [Mon, 10 Oct 2016 19:01:43 +0000 (15:01 -0400)]
Merge pull request #7543 from JosephTremoulet/NoRngChecks_Rel
Stop making JitNoRngChecks debug-only
John Chen [Mon, 10 Oct 2016 16:39:08 +0000 (09:39 -0700)]
Fix buffer length usage in UTF-8 to Unicode conversion (#7540)
These changes are ported from desktop CLR.
Hugh Bellamy [Mon, 10 Oct 2016 15:54:41 +0000 (16:54 +0100)]
Fix various CustomAttributeBuilder bugs (#7206)
Fix various CustomAttributeBuilder bugs
Joseph Tremoulet [Mon, 10 Oct 2016 14:48:56 +0000 (10:48 -0400)]
Stop making JitNoRngChecks debug-only
JitNoRngChecks is a flag, only available in special builds, that removes
all array bounds checks from generated code, which exists solely to
facilitate experiments measuring the cost of bounds checks. Commit
0aebfbef0 added support for it to RyuJit, but accidentally did so only for
debug builds; release `FEATURE_ENABLE_NO_RANGE_CHECKS` builds are
currently broken.
This change makes the jit flag's definition available in release builds
with `FEATURE_ENABLE_NO_RANGE_CHECKS` defined as well. The corresponding
code in clrconfigvalues.h already works that way.
Pat Gavlin [Sat, 8 Oct 2016 02:03:10 +0000 (19:03 -0700)]
Fix the codegen for by-value struct args on x86.
Although the existing code generation for by-value struct args containing
GC references on x86 was correct with respect to the layout of the data
being passed, it was not correct with respect to the invariants necessary
for the GC info encoder.
The existing codegen for by-value struct args--whether represented by
`GT_OBJ` or `GT_FIELD_LIST` nodes--takes what is essentially the
following approach:
sub esp, argument_size
mov [esp + offset_0], field_0
mov [esp + offset_1], field_1
...
mov [esp + offset_n], field_n
Although this puts the correct data in the correct locations, the
emitter is unable to recognize that a GC info update is requires if any
of the `mov` instructions write a GC reference to the stack: instead,
on x86, `push` instructions must be used to write GC references on the
stack. Furthermore, the stack must be kept 4-byte aligned.
This change modifies the code generator to use `push` instructions to
write outgoing GC references to the stack while obeying the alignment
invariant required by the GC encoding. For outgoing by-value struct
args represented by `GT_OBJ`, this results in very straightforward code:
push [val_addr]
push [val_addr + 4]
...
push [val_addr + n]
For args represented by `GT_FIELD_LIST`, the picture is a bit more
complicated as the stack must be kept 4-byte aligned. Handling of these
args is split into two cases: if a field list contains no GC references
and is larger than 16 bytes or contains any fields that cannot be
treated as a 4-byte-aligned, int32-sized value, the argument is passed
using the `sub esp, arg_size; mov [esp + offset], field` approach.
Otherwise--if the struct contains any GC references or is smaller than
16 bytes and only contains fields that can be treated as slots--the
argument is passed using the following approach:
- If a field is a GC reference or can be treated as a 4-byte slot, the
field is stored to the stack using a `push` instruction.
- Otherwise, the stack pointer is first adjusted to the next 4-byte
aligned address after the offset of the field and the field is
stored to the appropriate address using a `mov`.
For example, given the following IR:
t55 = lclVar bool V06 tmp4 u:6 esi REG esi RV $40
t138 = dconst float 0.
00000000000000000 REG mm0 $80
/--* t138 float
+--* t55 bool
t139 = * <fldList> void float at offset 4 REG NA $180
/--* t139 void
t152 = * putarg_stk [+0x00] void REG NA
The backend will generate the following code:
sub esp, 8
movss dword ptr [esp+04H], xmm0
mov dword ptr [esp], esi
And given the following IR:
t13 = lclVar ref V04 tmp1 u:2 esi (last use) REG esi RV $1c0
t45 = lclVar int V05 tmp2 u:2 eax (last use) REG eax RV $200
/--* t45 int
+--* t13 ref
t46 = * <fldList> void int at offset 4 REG NA $240
The backend will instead generate the following:
push eax
push esi
These changes also produce beneficial diffs: for example, the native image
for System.Private.CoreLib.dll has diffs in ~1000 methods for an overall
size improvement of 1%.
Gaurav Khanna [Sun, 9 Oct 2016 15:32:34 +0000 (08:32 -0700)]
Merge pull request #7538 from dotnet-bot/master-UpdateDependencies
Update CoreClr to beta-24609-02 (master)
dotnet-bot [Sun, 9 Oct 2016 08:15:37 +0000 (08:15 +0000)]
Update CoreClr to beta-24609-02
Michal Strehovský [Sun, 9 Oct 2016 04:59:20 +0000 (21:59 -0700)]
Add test to validate ByRef type unification (#7534)
Make sure that `ldtoken SomeType&` leads to the same type as `ldtoken
SomeType` followed by `MakeByRefType`.
Stephen Toub [Sun, 9 Oct 2016 04:59:01 +0000 (00:59 -0400)]
Implement CultureNotFoundException id-based ctors (#7536)
Gaurav Khanna [Sun, 9 Oct 2016 02:27:48 +0000 (19:27 -0700)]
Merge pull request #7531 from dotnet-bot/master-UpdateDependencies
Update CoreClr, CoreFx to beta-24609-01, beta-24608-01, respectively (master)
dotnet-bot [Sun, 9 Oct 2016 00:13:50 +0000 (00:13 +0000)]
Update CoreClr, CoreFx to beta-24609-01, beta-24608-01, respectively
Vance Morrison [Sat, 8 Oct 2016 06:08:38 +0000 (23:08 -0700)]
Standardize on _echo name (#7532)
Sergiy Kuryata [Sat, 8 Oct 2016 00:13:06 +0000 (17:13 -0700)]
Merge pull request #7521 from vkvenkat/master
Fixed #7519 - Dotnet crashes while saturating multiple CPU groups on Windows
Sean Gillespie [Fri, 7 Oct 2016 22:11:32 +0000 (15:11 -0700)]
Return null on allocation failure due to OOM instead of throwing a exception from within the GC (#7501)
John Chen (CLR) [Fri, 7 Oct 2016 19:46:16 +0000 (12:46 -0700)]
Disable CER code that is causing access violation
Resolves issue dotnet/corefx#12412.
Gaurav Khanna [Fri, 7 Oct 2016 20:57:28 +0000 (13:57 -0700)]
Merge pull request #7522 from dotnet-bot/master-UpdateDependencies
Update CoreClr, CoreFx to beta-24607-03, beta-24607-05, respectively (master)
Pat Gavlin [Fri, 7 Oct 2016 20:08:40 +0000 (13:08 -0700)]
Merge pull request #7520 from pgavlin/PutStructArgStk
Refactor genPutStructArgStk for clarity.
Pat Gavlin [Fri, 7 Oct 2016 18:15:44 +0000 (11:15 -0700)]
Fix formatting.
dotnet-bot [Fri, 7 Oct 2016 18:39:53 +0000 (18:39 +0000)]
Update CoreClr, CoreFx to beta-24607-03, beta-24607-05, respectively
vkvenkat [Fri, 7 Oct 2016 16:53:09 +0000 (09:53 -0700)]
Fixed #7519 - Dotnet crashes while saturating multiple CPU groups on Windows
Pat Gavlin [Fri, 7 Oct 2016 17:38:48 +0000 (10:38 -0700)]
Refactor genPutStructArgStk for clarity.
This also contains a minor CQ improvement: instead of always
incrementing the source and destination pointers when copying a struct
argument to the stack, we now only do so if there are more slots to
copy.
Bruce Forstall [Fri, 7 Oct 2016 16:29:02 +0000 (09:29 -0700)]
Merge pull request #7515 from BruceForstall/FixMerge
Fix bad merge
sandreenko [Fri, 7 Oct 2016 15:46:34 +0000 (08:46 -0700)]
Merge pull request #7509 from sandreenko/master
Set missed default value for coreclr runtest.cmd #7488.
Gaurav Khanna [Fri, 7 Oct 2016 14:58:47 +0000 (07:58 -0700)]
Merge pull request #7514 from dotnet-bot/master-UpdateDependencies
Update CoreClr, CoreFx to beta-24607-02, beta-24607-03, respectively (master)
Michelle McDaniel [Fri, 7 Oct 2016 13:56:45 +0000 (06:56 -0700)]
Merge pull request #7466 from adiaaida/fix7224
Force byteable registers for indir op source
Michelle McDaniel [Mon, 3 Oct 2016 21:49:19 +0000 (14:49 -0700)]
Force byteable registers for indir op source
When we have a GT_STOREIND, we need to not only make sure that the
source and dest of the storeind are marked as requiring byteable
registers, but also the indirect op's source. For example:
```
GT_STOREIND(GT_ADD(TYP_INT, mem of TYP_BYTE, reg of TYP_INT or TYP_BYTE))
```
We need to make sure that the source reg of the GT_ADD is byteable.
This change fixes #7224.
dotnet-bot [Fri, 7 Oct 2016 08:15:03 +0000 (08:15 +0000)]
Update CoreClr, CoreFx to beta-24607-02, beta-24607-03, respectively
Tarek Mahmoud Sayed [Fri, 7 Oct 2016 07:07:04 +0000 (00:07 -0700)]
Enable netstandard1.7 collations APIs (#7502)
* Enable netstandard1.7 collations APIs
This change is adding the netstandard 1.7 APIs to CompareInfo class and adding support to SortKey and SortVersion classes
The change include some support in cultureinfo and culturedata for locale Id's too which is used by CompareInfo class
* optimize the globalization data so we'll have faster initialization
* Remove security attributes
Sean Gillespie [Fri, 7 Oct 2016 01:51:10 +0000 (18:51 -0700)]
Restore some changes lost in a merge conflict (#7512)
Bruce Forstall [Fri, 7 Oct 2016 00:36:59 +0000 (17:36 -0700)]
Fix bad merge
A previous checkin backed-out a change that moved one test out of
the RyuJIT/x86 section. I'm reverting that, and putting the test
exclusion back where it should be.
At the same time, now that there are no RyuJIT/x86 test exclusions,
I'm deleting the RyuJIT/x86 section.
Bruce Forstall [Fri, 7 Oct 2016 00:24:32 +0000 (17:24 -0700)]
Merge pull request #7490 from BruceForstall/EnableLegacy
Enable COMPlus_useLegacyJit=1 for CoreCLR for x86
Gaurav Khanna [Thu, 6 Oct 2016 20:15:03 +0000 (13:15 -0700)]
Merge pull request #7507 from dotnet-bot/master-UpdateDependencies
Update CoreClr, CoreFx to beta-24606-03, beta-24606-02, respectively (master)
Pat Gavlin [Thu, 6 Oct 2016 19:37:58 +0000 (12:37 -0700)]
Merge pull request #7504 from pgavlin/NormalizeSpills
Always normalize stores when spilling lclVars.
Sergey Andreenko [Thu, 6 Oct 2016 18:56:40 +0000 (11:56 -0700)]
Set missed default value for coreclr runtest.cmd.
__MSBuildBuildArch is set to x64 by default like __BuildArch. Fix 7488.
Bruce Forstall [Tue, 4 Oct 2016 22:12:46 +0000 (15:12 -0700)]
Enable legacy JIT fallback for CoreCLR on Windows x86
This is to support the eventual case where RyuJIT/x86 becomes the default
x86 JIT, but we support fallback to the older, legacy JIT32/x86 (which
will be renamed compatjit.dll). This uses the same mechanism that was
built for .NET 4.6 when RyuJIT/x64 was shipped, and allowed fallback
to JIT64 as compatjit.dll.
However, we use a different configuration name to avoid conflicting
with the .NET 4.6 COMPlus_useLegacy name (and unintentionally causing
users to fall back to JIT64 with their .NET 4.6 apps).
The COMPlus variable is COMPlus_UseWindowsX86CoreLegacyJit=1. For the
dotnet.exe host, you can set `"System.JIT.UseWindowsX86CoreLegacyJit": true`
in the "configProperties" section of the app.runtimeconfig.json file.
There is a new COMPlus_RequireLegacyJit=1 option to aid testing JIT
fallback.
Pat Gavlin [Thu, 6 Oct 2016 17:43:00 +0000 (10:43 -0700)]
Fix a typo and a formatting issue.
Pat Gavlin [Wed, 5 Oct 2016 20:43:10 +0000 (13:43 -0700)]
Always normalize stores when spilling lclVars.
In order to support the semantics of the MSIL evaluation stack, RyuJIT
normalizes all "small" lclVars (i.e. lclVars with a non-struct
type that is smaller than an int32) from the actual type of the lclVar
to int32. This normalization may be performed in one of two ways:
- the lclVar's may be normalized when the lclVar is used (load
normalization)
- at any point at which the value is defined, the value being stored to
the lclVar is normalized (store normalization)
Store normalization must not be used if the lclVar is aliasable, since
the contents of the upper bytes for that lclVar's slot may be modified
by some other alias. Load normalization must be used if the lclVar is
a method parameter, as there is no guarantee that the caller has
properly nomralized the arguments it has passed. As such, any lclVar
that is either aliasable or is a method parameter is load-normalized,
and any other lclVar is store-normalized. lclVars that are
store-normalized are assumed to be represented as properly zero- or
sign-extended 4-byte values at all times, and so can simply be loaded
as int32s.
This picture becomes somewhat more subtle after register allocation,
however: once a lclVar has been enregistered, the contents of its
register are guaranteed to be properly zero- or sign-extended for the
purposes of store normalization, and in order for the lclVar to have
been enregistered in the first place, it must not have been aliasable.
As such, all enregistered variables may be store-normalized (as well as
load-normalized, if they represent method parameters). This has a
particularly significant impact on spills, as it implies that even if
the lclVar being spilled is not normally considered store-normalized, it
may be considered so for the purposes of the spill. This is advantageous
for architectures such as x86, as it allows a small lclVar to be
successfully spilled from a register that is not addressable at the size
of the lclVar's type. This change switches both LSRA and the code
generator to always use the actual type of a spilled lclVar for the
corresponding store rather than making this behavior conditional on
whether or not the lclVar is store-normalized.
Fixes #7236.
dotnet-bot [Thu, 6 Oct 2016 16:18:36 +0000 (16:18 +0000)]
Update CoreClr, CoreFx to beta-24606-03, beta-24606-02, respectively