Bruce Forstall [Thu, 3 Aug 2017 15:06:10 +0000 (08:06 -0700)]
Merge pull request #13166 from YongseopKim/ryujit/arm32/add_handling_secure_delegate_inv
[RyuJIT/ARM32] Adds code to handle SecureDelegateInvoke
Yongseop Kim [Wed, 2 Aug 2017 12:04:47 +0000 (21:04 +0900)]
[RyuJIT/ARM32] Adds code to handle SecureDelegateInvoke
- Adds code to handle SecureDelegateInvoke on morph phase like legacy's handling.
- Set R4 to proper addr for indirect call of vsd
- This is like legacy's doing from #6500.
- Fix #12993
Stephen Toub [Thu, 3 Aug 2017 01:55:58 +0000 (21:55 -0400)]
Merge pull request #13154 from christianscheuer/fix-readme
Fix reference to System.Threading.Thread class in README.md
Michelle McDaniel [Thu, 3 Aug 2017 00:25:00 +0000 (17:25 -0700)]
Merge pull request #13182 from adiaaida/addMinOptsTP
Add JitMinOpts throughput jobs
Michelle McDaniel [Wed, 2 Aug 2017 23:29:37 +0000 (16:29 -0700)]
Add JitMinOpts throughput jobs
This change adds an opt_level throughput option so that we can run
throughput jobs in minopts mode. This also adds jobs to the lab for min
opts runs
Andy Ayers [Wed, 2 Aug 2017 22:46:56 +0000 (15:46 -0700)]
JIT: refactor logic for removing box upstream side effects (#13169)
Introduce a new method `gtTryRemoveBoxUpstreamEffects` to capture
the logic for cleaning up after a dead value type box into a utility.
Call this from `gtFoldExprSpecial` and now also from the inliner when
it finds a dead box argument.
Also remove the useless `fgExpandInline` field and associated
`fgIsInlining` method, as the jit never set the field to true.
No diffs from the refactoring. The inliner case kicks in in 3 places
in the test tree.
Closes #13136.
José Rivero [Wed, 2 Aug 2017 21:58:46 +0000 (14:58 -0700)]
Add JitBench scenario testing. (#11400)
Noah Falk [Wed, 2 Aug 2017 21:06:21 +0000 (14:06 -0700)]
Merge pull request #13158 from parjong/fix/INTERPERTER_GDBJIT_Conflict
Resolve FEATURE_GDBJIT/FEATURE_INTERPRETER conflict
Carol Eidt [Wed, 2 Aug 2017 18:31:13 +0000 (11:31 -0700)]
Merge pull request #13155 from CarolEidt/UModTreeNodeInfo
Set Register Requirements for Mod/Div
Joseph Tremoulet [Wed, 2 Aug 2017 18:28:40 +0000 (14:28 -0400)]
Merge pull request #13128 from JosephTremoulet/Mulshift
Emit power-of-two constant multiply as shift
Russell C Hadley [Wed, 2 Aug 2017 17:29:19 +0000 (10:29 -0700)]
Merge pull request #11456 from CarolEidt/EHWriteThruDoc
Add an alternate algorithm done during LSRA
Steve MacLean [Wed, 2 Aug 2017 10:48:14 +0000 (06:48 -0400)]
[Arm64] GetLargestOnDieCacheSize (#13071)
Look for cache info in /sys/.../cache/index*/size
If that fails estimate cache size based on processor count
Jonghyun Park [Wed, 2 Aug 2017 09:20:44 +0000 (18:20 +0900)]
Remove unused extern declarations (#13159)
Steve MacLean [Wed, 2 Aug 2017 08:37:37 +0000 (04:37 -0400)]
PAL enable Thread affinity support (#12936)
Jonghyun Park [Wed, 2 Aug 2017 05:11:26 +0000 (14:11 +0900)]
Use 'udf 0xff' instead of 'bkpt 0xbe' as a poison (#13152)
Hyung-Kyu Choi [Wed, 2 Aug 2017 03:40:34 +0000 (12:40 +0900)]
Enable build of crosscomponent for armel (#11066)
* Enable build of crosscomponent for armel
Enable build of crosscomponent for armel in addition to arm
Signed-off-by: Hyung-Kyu Choi <hk0110.choi@samsung.com>
* Fix typo
Signed-off-by: Hyung-Kyu Choi <hk0110.choi@samsung.com>
Jonghyun Park [Wed, 2 Aug 2017 03:05:34 +0000 (12:05 +0900)]
Resolve FEATURE_GDBJIT/FEATURE_INTERPRETER conflict
Carol Eidt [Wed, 2 Aug 2017 01:19:36 +0000 (18:19 -0700)]
Set Register Requirements for Mod/Div
Swaroop Sridhar [Fri, 26 May 2017 02:55:32 +0000 (19:55 -0700)]
LinkBench: Enable Roslyn
This change enables Roslyn runs as part of ILLink.
Roslyn was previously disabled because .Net CLI didnt use the latest
Roslyn compilers. Now that this issue is fixed, we re-enable Roslyn.
The different .Net versions necessary to run the benchmarks are
all downloaded, and the benchmarks are built using them.
Brian Sullivan [Wed, 2 Aug 2017 00:35:50 +0000 (17:35 -0700)]
Merge pull request #13151 from briansull/dump-fixes
Minor fixes for JitDumps
Christian Scheuer [Wed, 2 Aug 2017 00:30:32 +0000 (02:30 +0200)]
Fix reference to System.Threading.Thread class
Steve MacLean [Wed, 2 Aug 2017 00:24:51 +0000 (20:24 -0400)]
[Arm64/Unix] Add OfficialBuildRID linux-arm64 (#12372)
* [Arm64/Unix] Add OfficialBuildRID linux-arm64
* [Arm64/Unix] Fix nuget package build
Steve MacLean [Thu, 6 Jul 2017 23:59:49 +0000 (23:59 +0000)]
GC Prefer CompareExchange
Steve MacLean [Thu, 6 Jul 2017 23:58:36 +0000 (23:58 +0000)]
GC Reduce unnecessary work
Simplify cache alignment
Remove unecessary barriers
Remove unnecessary Interlocked::Decrement
Remove unused join variables
Brian Sullivan [Tue, 1 Aug 2017 22:28:16 +0000 (15:28 -0700)]
Merge pull request #12856 from briansull/method-referenced-ibcdata
Record references to all of the hot methods specified by the IBC profile data
Bruce Forstall [Tue, 1 Aug 2017 21:12:46 +0000 (14:12 -0700)]
Merge pull request #13109 from mikedn/range-alloc
Remove unnecessary allocations from RangeCheck
Bruce Forstall [Tue, 1 Aug 2017 20:39:47 +0000 (13:39 -0700)]
Merge pull request #12929 from mikedn/cast-cmp-test
Use "test reg, reg" instead of "cmp reg, 0" for cast overflow checks
mikedn [Tue, 1 Aug 2017 20:38:12 +0000 (23:38 +0300)]
Refactor emitter::emitInsMov (#12905)
* Split emitter::emitInsMov into separate functions
None of this function's callers need it to figure out what kind of "mov" to emit, the callers already know what they need and they should call the appropiate function.
Copy & paste code, no actual changes.
* Pass the destination register to emitInsLoadInd
This way it can be used for contained indirs where the destination register is set on the parent node.
* Cleanup emitInsLoadInd/emitInsStoreInd/emitInsStoreLcl
Jonghyun Park [Tue, 1 Aug 2017 20:33:50 +0000 (05:33 +0900)]
Fix build errors when TRACK_SYNC is defined (#13122)
* Fix build errors when TRACK_SYNC is defined
* Remove unnecessary default constructor
Bruce Forstall [Tue, 1 Aug 2017 20:32:41 +0000 (13:32 -0700)]
Merge pull request #12926 from mikedn/dead-cast-code
Delete dead code from genIntToIntCast
Joseph Tremoulet [Mon, 31 Jul 2017 20:37:51 +0000 (16:37 -0400)]
Emit power-of-two constant multiply as shift
Most of these get translated to shifts at Morph, but adding a failsafe
here next to the transform for 3/5/9->LEA ensures we won't have these in
the emitted code when they (re-)appear in later phases.
Bruce Forstall [Tue, 1 Aug 2017 20:19:22 +0000 (13:19 -0700)]
Merge pull request #13125 from sdmaclea/PR-POISON
[Arm64] Implement Poison()
Bruce Forstall [Tue, 1 Aug 2017 20:15:19 +0000 (13:15 -0700)]
Merge pull request #13134 from BruceForstall/Fix12911
Handle unspilling part of a multi-use reg with FREE_REG
Brian Sullivan [Tue, 1 Aug 2017 18:45:45 +0000 (11:45 -0700)]
Merge pull request #13147 from jashook/fix_desktop_build_break
Fix Desktop build break in regset.cpp
Brian Sullivan [Tue, 1 Aug 2017 18:23:39 +0000 (11:23 -0700)]
Merge branch 'master' into dump-fixes
Brian Sullivan [Tue, 1 Aug 2017 18:20:53 +0000 (11:20 -0700)]
Minor fixes for JitDumps
- Remove extra ']' in rsMarkRegUsed and rsMarkArgRegUsedByPromotedFieldArg
- removed incorrect newline splits in rsMarkPairUsed
- Check if rsUsedAddr[] is null before calling printTreeID
- Check if spillAddr is null before calling printTreeID
Mark Wilkie [Tue, 1 Aug 2017 18:20:22 +0000 (11:20 -0700)]
Merge pull request #13130 from markwilkie/CROSS-Doc
Added CROSS guidance
Bruce Forstall [Tue, 1 Aug 2017 18:07:57 +0000 (11:07 -0700)]
Merge pull request #13142 from hseok-oh/ryujit/fix_12627
[RyuJIT/ARM32] Tailcall: passing struct argument
Brian Sullivan [Tue, 1 Aug 2017 18:03:25 +0000 (11:03 -0700)]
Merge pull request #13002 from sdmaclea/PR-ARM64-resolveEntryPoint
[Arm64] Fix ResolveStub tokenHash lookup
Bruce Forstall [Tue, 1 Aug 2017 17:54:58 +0000 (10:54 -0700)]
Merge pull request #13138 from hseok-oh/ryujit/nyi_zr
[RyuJIT/ARM32] Remove NYI: using zero register
Mark Wilkie [Tue, 1 Aug 2017 17:45:52 +0000 (10:45 -0700)]
small updates....
Bruce Forstall [Tue, 1 Aug 2017 17:29:04 +0000 (10:29 -0700)]
Merge pull request #13144 from parjong/fix/INTERPRETER_build
Fix FEATURE_INTERPRETER build
Mark Wilkie [Tue, 1 Aug 2017 17:18:16 +0000 (10:18 -0700)]
Updated guidance based on PR feedback
Steve MacLean [Tue, 1 Aug 2017 17:11:55 +0000 (13:11 -0400)]
[Arm64] Adjust alignment for 128 byte cache line (#12801)
jashook [Tue, 1 Aug 2017 16:35:29 +0000 (09:35 -0700)]
Fix Desktop build break in regset.cpp
Pat Gavlin [Tue, 1 Aug 2017 15:53:58 +0000 (08:53 -0700)]
Merge pull request #12947 from pgavlin/GH12919
Lock arg registers when generating a `JMP` on ARM/Legacy.
Jonghyun Park [Tue, 1 Aug 2017 13:04:24 +0000 (22:04 +0900)]
Fix FEATURE_INTERPRETER build error
Hyeongseok Oh [Tue, 1 Aug 2017 08:42:56 +0000 (17:42 +0900)]
[RyuJIT/ARM32] Tailcall: passing struct argument
Fix bug in tailcall(jmp) to pass struct argument
Remove NYI for HFA struct passing
Pat Gavlin [Tue, 1 Aug 2017 04:41:37 +0000 (21:41 -0700)]
Format code.
Hyeongseok Oh [Tue, 1 Aug 2017 01:44:51 +0000 (10:44 +0900)]
[RyuJIT/ARM32] Remove NYI: using zero register
We can remove NYI to using zero register to assign constant zero argument.
There is no zero register on ARM32. (src/jit/registerarm32.h)
So we can change the NYI to unreached() assertion.
Bruce Forstall [Tue, 1 Aug 2017 00:28:04 +0000 (17:28 -0700)]
Handle unspilling part of a multi-use reg with FREE_REG
If rsUnspillOneReg() is called with willKeepNewReg==FREE_REG
(indicating the unspilled value will be free on return),
and the reg to unspill is multi-use, then we were releasing the
underlying temp incorrectly, since the temp still contains the
other multi-use spilled values, which will be subsequently
unspilled.
This occurred in a very large, JitStressRegs=1/JitStress=1 test case.
Fixes #12911
Roman Artemev [Tue, 1 Aug 2017 00:16:37 +0000 (17:16 -0700)]
Merge pull request #13132 from rartemev/legbnd_fix_misprint
Fixed misprint in legacy backend that led to assert failure
Bruce Forstall [Mon, 31 Jul 2017 22:43:05 +0000 (15:43 -0700)]
Merge pull request #13096 from hseok-oh/ryujit/remove_nyi_promote
[RyuJIT/ARM32] Remove NYI: promoted struct argument
Pat Gavlin [Mon, 31 Jul 2017 22:38:54 +0000 (15:38 -0700)]
Lock all ABI arg registers when generating a GS check on ARM/Legacy.
These registers may otherwise be overwritten by code that is generated
between the JMP/tail call and the epilog (e.g. GS cookie checks).
Fixes #12919.
Roman Artemev [Mon, 31 Jul 2017 21:53:57 +0000 (14:53 -0700)]
Fixed misprint in legacy backend that led to assert failure
Pat Gavlin [Mon, 31 Jul 2017 21:44:29 +0000 (14:44 -0700)]
Merge pull request #13103 from pgavlin/GH12910
Only lock unlocked registers in `rsUnspillRegPair`.
Victor "Nate" Graf [Mon, 31 Jul 2017 20:34:53 +0000 (13:34 -0700)]
Add benchmarks mandelbrot and fannkuch (#12933)
* Add benchmarks mandelbrot and fannkuch
Adding mandelbrot and fankuch-redux from Benchmarks Game to
complete the representation of benchmarks from this list
http://benchmarksgame.alioth.debian.org/u64q/csharp.html
* Update source acknolegments and edit return codes
* Add checksums to verify benchmark runs
This commit adds checksums, which are pre-computed for each input
case, to verify that the fannkuch and mandelbrot benchmarks
actualy compute the desired function
* Revert version bump
* Remove XunitPerfHarness from Main
* Remove an unnecessary line
* Add iterations to fannkuch to inscrease runtime
Andy Ayers [Mon, 31 Jul 2017 20:01:01 +0000 (13:01 -0700)]
JIT: Fix value type box optimization (#13016)
Boxing a value type produces a non-null result. If the result of the box is
only used to feed a compare against null, the jit tries to optimize the box
away entirely since the result of the comparison is known. Such idiomatic
expressions arise fairly often in generics instantiated over value types.
In the current implementation the box expands into two parts. The first is
an upstream statement to allocate a boxed object and assign a reference to
the boxed object to a local var known as the "box temp". The second is an
expression tree whose value is the box temp that also contains an an
encapsulated copy from the value being boxed to the payload section of the
boxed object. The box node also contains a pointer back to the first
statement (more on this later).
In the examples being discussed here this second tree is a child of a compare
node whose other child is a null pointer. When the optimization fires, the
upstream allocation statement is located via the pointer in the box node and
removed, and the entire compare is replaced with a constant 0 or 1 as
appropriate. Unfortunately the encapsulated copy in the box subtree may
include side effects that should be preserved, and so this transformation is
unsafe.
Note that the copy subtree as a whole will always contain side effects, since
the copy is storing values into the heap, and that copy now will not happen.
But the side effects that happen when producing the value to box must remain.
In the initial example from #12949 the side effects in question were
introduced by the jit's optimizer to capure a CSE definition. #13016 gives
several other examples where the side effects are present in the initial user
code. For instance the value being boxed might come from an array, in which
case the encapsulated copy in the box expression tree would contain the array
null check and bounds check. So removing the entire tree can alter behavior.
This fix attempts to carefully preserve the important side effects by
reworking how a box is imported. The copy is now moved out from under the box
into a second upstream statement. The box itself is then just a trivial
side-effect-free reference to the box temp. To ensure proper ordering of side
effects the jit spills the evaluation stack before appending the copy
statement.
When the optimization fires the jit removes the upstream heap allocation
as before, as well as the now-trivial compare tree. It analyzes the source
side of the upstream copy. If it is side effect free, the copy is removed
entirely. If not, the jit modifies the copy into a minimal load of the
boxed value, and this load should reproduce the necessary side effects.
The optimization is only performed when the tree shape of the copy matches
expected patterns.
There are some expected cases where the tree won't match, for instance if the
optimization is invoked while the jit is inlining. Because this optimization
runs at several points the jit can catch these cases once inlining completes.
There is one case that is not handled that could be -- if the assignment part
of the copy is itself a subtree of a comma. This doesn't happen often.
The optimization is now also extended to handle the case where the comparision
operation is `cgt.un`. This doesn't catch any new cases but causes the
optimization to happen earlier, typically during importation, which should
reduce jit time slightly.
Generally the split of the box into two upstream statements reduces code size,
especially when the box expression is incorporated into a larger tree -- for
example a call. However in some cases where the value being boxed comes from
an array, preserving the array bounds check now causes loop cloning to kick
in and increase code size. Hence the overall size impact on the jit-diff set is
essentially zero.
Added a number of new test cases showing the variety of situations that must
be handled and the need to spill before appending the copy statement.
Fixes #12949.
Joseph Tremoulet [Thu, 20 Jul 2017 17:28:53 +0000 (13:28 -0400)]
Add documents about JIT optimization planning
This change adds two documents:
- JitOptimizerPlanningGuide.md discusses how we can/do/should go about
identifying, prioritizing, and validating optimization improvement
opportunities, as well as several ideas for how we might improve the
process.
- JitOptimizerTodoAssessment.md lists several potential optimization
improvements that always come up in planning discussions, with brief
notes about each, to capture current thinking.
Stephen Toub [Mon, 31 Jul 2017 18:48:55 +0000 (14:48 -0400)]
Merge pull request #13111 from dotnet-maestro-bot/master-UpdateDependencies
Update CoreClr, CoreFx to preview2-25531-02, preview2-25531-01, respectively (master)
Jonghyun Park [Mon, 31 Jul 2017 18:17:47 +0000 (03:17 +0900)]
[ARM/Linux] Set R2R_INDIRECT_PARAM before R2R call (#12952)
* Set R2R_INDIRECT_PARAM before R2R call
* Revise code formatting
* Preserve R2R_INDIRECT_PARAM across R2R call
* Fix format error
* Attempt to fix x86/Windows build error
* 2nd attempt to fix x86/Windows build error
* Revise RA to reserve R2R_INDIRECT_PARAM reg
* Fix format error
[Arm64] Implement Poison()
Fixes #13117
Mark Wilkie [Mon, 31 Jul 2017 16:43:10 +0000 (09:43 -0700)]
Added CROSS guidance
Bruce Forstall [Mon, 31 Jul 2017 16:12:49 +0000 (09:12 -0700)]
Merge pull request #13116 from parjong/fix/x86_stklv_on_temp_block
[x86/Linux] Set proper stack level for temp label BB
Atsushi Kanamori [Mon, 31 Jul 2017 15:28:04 +0000 (08:28 -0700)]
Implement Type.IsByRefLike for CoreCLR (#13123)
Since this already exists in CoreRT and the
virtual method will get propagated to CoreCLR,
we might as well make it work here before an IsByRefLike
check finds its way into shared code and starts
throwing NotSupportedExceptions on CoreCLR.
dotnet-maestro-bot [Mon, 31 Jul 2017 12:49:16 +0000 (05:49 -0700)]
Update CoreClr, CoreFx to preview2-25531-02, preview2-25531-01, respectively
Jan Vorlicek [Mon, 31 Jul 2017 09:51:13 +0000 (11:51 +0200)]
Enable RHEL6 and CentOS 6 RID detection in build (#13101)
This change adds RHEL6 and CentOS 6 RID detection to build.sh.
These distros don't have the /etc/os-release file and so
we need to use another source - the /etc/redhat-release file
Jonghyun Park [Mon, 31 Jul 2017 07:48:16 +0000 (16:48 +0900)]
Fill freed loader heap chunk with non-zero value (#12731)
* Add FEATURE_LOADER_HEAP_GUARD feature
* Invoke memset only for reclaimed regions
* Enable FEATURE_LOADER_HEAP_GUARD by default
* Insert trap inside UMEntryThunk::Terminate
* Make all exectuable heaps not to zero-initialize itself
Use fZeroInit (instead of fMakeRelazed)
* Add comment
* Revert unnecessary changes
* Add and use 'Poison' method to insert a trap
* Do NOT invoke FlushInstructionCache
* Update comment
* Add comment on ARM Poisoning instruction
* Use X86_INSTR_INT3 instead of 0xCC
Jonghyun Park [Mon, 31 Jul 2017 02:27:58 +0000 (11:27 +0900)]
Set proper stack level for temp label BB
Michal Strehovský [Fri, 28 Jul 2017 16:41:38 +0000 (09:41 -0700)]
Merge pull request dotnet/corert#4242 from dotnet/nmirror
Merge master to nmirror
Signed-off-by: dotnet-bot <dotnet-bot@microsoft.com>
Xiang Fan [Sun, 30 Jul 2017 16:18:23 +0000 (09:18 -0700)]
Remove extra qualification in member function declaration (#13104)
Fix issues mentioned in https://github.com/dotnet/coreclr/issues/12799
Bruce Forstall [Sat, 29 Jul 2017 18:46:42 +0000 (11:46 -0700)]
Fix bug converting BBJ_CALLFINALLY block to BBJ_THROW on ARM32 (#13094)
* Fix bug converting BBJ_CALLFINALLY block to BBJ_THROW on ARM32
On ARM32, the target block of a return from a finally (aka, the
"continuation") is marked by a bit, `BBF_FINALLY_TARGET`, that is
used to generate proper code to ensure correct unwinding behavior.
This bit is set in the importer, and maintained through to codegen.
We assert if we attempt to delete a block with this bit. Thus, various
code that deletes dead code needs to call fgClearFinallyTargetBit()
to keep the bit updated.
This bug is a case where very early in morph, a BBJ_CALLFINALLY
block is converted to a BBJ_THROW. We should be able to just call
fgClearFinallyTargetBit(), but that function depends on the predecessor
lists, which in this case, haven't yet been built. So, instead, clear
the bits everywhere, and then recompute them at the end of morph.
This is similar to what is done with the various try/finally
optimizations done at the beginning of morph.
(It's an open question about whether we could avoid setting the bits
at all until just before codegen.)
Fixes VSO468731
No diffs in desktop altjit asm diffs.
* Formatting
Stephen Toub [Sat, 29 Jul 2017 16:49:02 +0000 (12:49 -0400)]
Merge pull request #12890 from dotnet-maestro-bot/master-UpdateDependencies
Update CoreClr, CoreFx to preview2-25529-02, preview2-25529-02, respectively (master)
dotnet-maestro-bot [Sat, 29 Jul 2017 13:27:03 +0000 (06:27 -0700)]
Update CoreClr, CoreFx to preview2-25529-02, preview2-25529-02, respectively
Mike Danes [Sat, 29 Jul 2017 10:36:25 +0000 (13:36 +0300)]
Remove unnecessary allocations from RangeCheck
There's no need to collect PHI arg ranges in a list and then merge them all. Merging can be done as arg ranges are computed.
James Singleton [Sat, 29 Jul 2017 05:47:24 +0000 (06:47 +0100)]
Fix typo in comment (#13089)
integeter -> integer
Pat Gavlin [Sat, 29 Jul 2017 03:10:14 +0000 (20:10 -0700)]
Format code.
Steve MacLean [Sat, 29 Jul 2017 02:08:01 +0000 (22:08 -0400)]
Optimize for idle WorkStealingQueueList (#13007)
When WorkStealingQueue is empty there is an O(N^2)
penalty for searching the empty queues. Avoid by
stashing global emptiness
Carol Eidt [Fri, 28 Jul 2017 23:32:54 +0000 (16:32 -0700)]
Merge pull request #13079 from CarolEidt/RyuJITTutorial
RyuJIT Tutorial
Carol Eidt [Thu, 27 Jul 2017 06:48:10 +0000 (23:48 -0700)]
RyuJIT Tutorial
This is an initial translation of the slides presented at the PLDI 2016
RyuJIT Tutorial into markdown.
The IR dumps need to be updated to reflect IR changes since the tutorial was given.
Pat Gavlin [Fri, 28 Jul 2017 18:42:16 +0000 (11:42 -0700)]
Only lock unlocked registers in `rsUnspillRegPair`.
`rsUnspillRegPair` needs to lock the hi/lo part of the reg pair when
unspilling the lo/hi part in order to ensure that the two parts of the
pair do not end up in the same register. However, the register to be
locked may already be locked if it is a multi-use register. In this case
`rsUnspillRegPair` does not need to re-lock the register.
Fixes #12910.
Stephen Toub [Fri, 28 Jul 2017 22:10:13 +0000 (18:10 -0400)]
Merge pull request #13075 from stephentoub/string_span_ctor
Add new String(ReadOnlySpan<char>) ctor
Pat Gavlin [Fri, 28 Jul 2017 18:32:17 +0000 (11:32 -0700)]
Merge pull request #13095 from pgavlin/VSO469217
Rerun reg predict on ARM if codegen temps increase the frame size.
Pat Gavlin [Fri, 28 Jul 2017 16:06:41 +0000 (09:06 -0700)]
PR feedback.
Stephen Toub [Fri, 28 Jul 2017 16:00:09 +0000 (12:00 -0400)]
Add String(ReadOnlySpan<char>) ctor
Jan Kotas [Fri, 28 Jul 2017 15:53:01 +0000 (11:53 -0400)]
Add generic encoding support to metasig
Noah Falk [Fri, 28 Jul 2017 09:26:42 +0000 (02:26 -0700)]
Fix tier0 flag on default code versions (#13098)
Fixes github issue 13019.
Hyeongseok Oh [Fri, 28 Jul 2017 01:41:59 +0000 (10:41 +0900)]
[RyuJIT/ARM32] Remove NYI: promoted struct argument
We can remove NYI for promoted struct argument on codegen phase.
On morphing phase, argument assignment for promoted struct will be changed to assignment on tmpvar struct.
And this tmpvar is not promoted. So we change this NYI to assertion check.
Pat Gavlin [Fri, 28 Jul 2017 00:41:56 +0000 (17:41 -0700)]
Merge pull request #13087 from pgavlin/FixGetTargetOfCallARM
Fix `getTargetOfCall` for ARM.
Pat Gavlin [Thu, 27 Jul 2017 22:48:48 +0000 (15:48 -0700)]
Rerun reg predict on ARM if codegen temps increase the frame size.
When compiling for ARM, we need to reserve an additional register if the
frame size exceeds a particular watermark. We currently evaluate whether
or not we need to reserve this register in two places: first before
register prediction and then again during register prediction but after
allocating registers for lclVars. The latter evalutation is used to
check whether or not we can remove the register reservation if the frame
size shrinks due to lclVar register allocation. Unfortunately, its
frame size estimate does not include codegen temps, so we can end up
undoing the register reservation in a frame that requires the reserved
register once the number of temps is estimated.
This change adds an additional check to the register predictor after
temp estimation has occurred and reruns prediction if the frame size has
increased and now requires a register reservation that it did not
beforehand.
Fixes VSO 469217.
Jan Vorlicek [Thu, 27 Jul 2017 22:18:31 +0000 (00:18 +0200)]
Fix Alpine build (#13052)
This change fixes CoreCLR build on Alpine Linux
Jarret Shook [Thu, 27 Jul 2017 19:48:40 +0000 (12:48 -0700)]
Merge pull request #13032 from jashook/add_arm_stress_scenarios
Add arm legacy backend stress jobs
David Mason [Thu, 27 Jul 2017 19:46:56 +0000 (12:46 -0700)]
Merge pull request #13043 from davmason/profiler_r2r
don't use r2r images when the profiler requests that ngen images are disabled
Pat Gavlin [Thu, 27 Jul 2017 19:32:07 +0000 (12:32 -0700)]
Fix `getTargetOfCall` for ARM.
If the instruction we're decoding is not a call, don't attempt to decode
it as if it were an x86 instruction. Instead, just return 0 as on ARM64.
Maoni Stephens [Thu, 27 Jul 2017 19:26:15 +0000 (12:26 -0700)]
gcconcurrent config should be obtained from g_pConfig which takes env vars/reg values into consideration as well (#13078)
Stephen Toub [Thu, 27 Jul 2017 17:20:48 +0000 (13:20 -0400)]
Add new Span-based virtual sync Read/Write Stream methods (#13058)
* Add virtual Stream.Read/Write Span-based APIs
* Override Span-based Read/Write on MemoryStream
* Override Span-based Read/Write on UnmanagedMemoryStream
* Address PR feedback
jashook [Tue, 25 Jul 2017 21:40:46 +0000 (14:40 -0700)]
Add arm legacy backend stress jobs
Bruce Forstall [Thu, 27 Jul 2017 15:56:37 +0000 (08:56 -0700)]
Merge pull request #13077 from wateret/fix-runtest
Fix print message in runtest.sh
Hanjoung Lee [Thu, 27 Jul 2017 05:04:43 +0000 (14:04 +0900)]
Fix print message in runtest.sh
Pat Gavlin [Thu, 27 Jul 2017 01:35:26 +0000 (18:35 -0700)]
Merge pull request #13072 from pgavlin/VSO468730
Disable folding in genCreateAddrMode under LB-specific circumstances.