Tarek Mahmoud Sayed [Fri, 4 Nov 2016 20:15:17 +0000 (13:15 -0700)]
make the returning of DTFI & NFI in CultureInfo atomic (#7998)
we got some cases when more than one thread trying to get DTFI or NFI from the shared current culture object which can result returning different instances.
Although the returning object contents will be exactly but it'll be better to return same instance as apps may assume that. also returning the same instance will help reduce reinitializing more deeper fields when get requested.
Joseph Tremoulet [Fri, 4 Nov 2016 19:11:57 +0000 (15:11 -0400)]
Merge pull request #7911 from JosephTremoulet/CloneExprsForUnroll
Clone exprs for unroll
Joseph Tremoulet [Fri, 14 Oct 2016 17:57:38 +0000 (13:57 -0400)]
Add unroller support to CloneBlockState
This helper is used by the loop cloner to copy a block's attributes and
statements. Expand it so it can also be used by the loop unroller:
- Accept var/val parameters to pass through to `gtCloneExpr` to
perform in-place substitutions
- When presented with an expression that `gtCloneExpr` can't clone,
return false rather than dereferencing null
Pat Gavlin [Fri, 4 Nov 2016 18:20:55 +0000 (11:20 -0700)]
Merge pull request #7983 from pgavlin/VSO284785
Fix int->long cast folding for 32-bit platforms.
Joseph Tremoulet [Wed, 19 Oct 2016 22:14:24 +0000 (18:14 -0400)]
Avoid clobbering FldSeq annotations in cloneExpr
The overload of cloneExpr that takes a variable number and value to
perform replacements is intended to be used by the loop unroller, which
runs between Morph (where FldSeq annotations are generated) and value
numbering (where FldSeq annotations are consumed). Update that codepath
to call `LabelIndex` when replacing a `GTF_VAR_ARR_INDEX` node, and to
avoid calling `gtFoldExpr` (which does not preserve these annotations).
Downstream constant folding should be capable of doing that optimization.
Joseph Tremoulet [Fri, 14 Oct 2016 17:34:45 +0000 (13:34 -0400)]
Do not replace assign LHS in gtCloneExpr
Split the var/val parameters into two sets -- one for the current level of
recursion, the other for deeper levels. Use this to suppress one level of
replacement when recursing to the LHS of an assign. This both avoids
creating nonsensical assign-to-immediate trees and allows replacing uses
of the given var while leaving defs in the code.
Andy Ayers [Fri, 4 Nov 2016 04:28:16 +0000 (21:28 -0700)]
Jit: fix regression in non-AVX SIMD codegen (#7989)
Vector codegen for dot product was inadvertenly altered in non-DEBUG
builds.
Closes #7977.
Alex Ghiondea [Fri, 4 Nov 2016 04:28:01 +0000 (21:28 -0700)]
Add the serialization ctor for CryptographicException to model.xml (#7974)
Joseph Tremoulet [Wed, 2 Nov 2016 15:46:06 +0000 (11:46 -0400)]
Fix check for float args under windows x64 remorph
When re-morphing, `argx` will not have the argument's type if the argument
has been moved to the late args list, so check the arg table entry
instead. The unix x64 codepath already had the correct code, so just
remove the ifdef.
Joseph Tremoulet [Thu, 3 Nov 2016 18:51:17 +0000 (14:51 -0400)]
Clear GTF_VAR_USEDEF when folding <op>= forms
The `GTF_VAR_USEDEF` flag is applied to the def of an expression like "x =
x + y". The `GTF_VAR_USEASG` flag is applied to the op1 of an expression
like "x += y". Update the code to clear the `GTF_VAR_USEDEF` flag (in
addition to setting the `GTF_VAR_USEASG` flag) when folding the former to
the latter.
Jan Kotas [Fri, 4 Nov 2016 03:35:50 +0000 (20:35 -0700)]
Merge pull request #7991 from dotnet/revert-7946-armci_crashed
Revert "ARM-CI : Use archived root-fs and run tests with --sequential option"
Jan Kotas [Fri, 4 Nov 2016 03:35:35 +0000 (20:35 -0700)]
Revert "ARM-CI : Use archived root-fs and run tests with --sequential option"
Jonghyun Park [Fri, 4 Nov 2016 02:43:24 +0000 (11:43 +0900)]
Do not use FP registers when soft-fp ABI is used (#7869)
Soft FP ABI enforces a return value to be passed via integer registers
even though it is of float/double type (which means that 'r0 ~ r4' should
be preseved even for functions that return a float/double value).
The current implementation of 'ArgIteratorTemplate<ARGITERATOR_BASE>::ComputeReturnFlags()' does
not consider this ABI difference, which makes 'CallDescrWorkerInternal'
function to preserve FP registers instead of 'r0 ~ r4' (and spoils
'r0/r1' during preservation), which leads to #7868.
This commit revises this function in order to fix #7868.
Sujin Kim [Fri, 4 Nov 2016 02:31:30 +0000 (11:31 +0900)]
ARM-CI : Use archived root-fs and run tests with --sequential option (#7946)
* ARM-CI : Use archived root-fs and run tests with --sequential option
To make sure the reason of test failure ramdomly( #6298 ),
We checked segmentation faults occurred from mounted rootfs and the multi thread processing.
So I changed root-fs to archived root-fs and run tests with --sequential option.
* change to original clang version
Bruce Forstall [Fri, 4 Nov 2016 00:58:29 +0000 (17:58 -0700)]
Fix JIT build host/target confusion (#7979)
Also, add RyuJIT to Windows ARM32 cross components build (as protojit.dll).
Jan Kotas [Fri, 4 Nov 2016 00:57:48 +0000 (17:57 -0700)]
Merge pull request #7978 from AlexGhiondea/RemoveSecureStringMarshal
The SecureStringMarshal type was an artifact of the layering we had i…
Jan Kotas [Fri, 4 Nov 2016 00:12:04 +0000 (17:12 -0700)]
Merge pull request #7982 from dotnet/revert-7677-StructOpts
Revert "Enable optimization of structs"
Michelle McDaniel [Thu, 3 Nov 2016 23:38:26 +0000 (16:38 -0700)]
Merge pull request #7972 from adiaaida/storeShiftRotate32
Force store of lcl var for shift/rotate by 32
Rama krishnan Raghupathy [Thu, 3 Nov 2016 23:02:20 +0000 (16:02 -0700)]
Merge pull request #7965 from ramarag/GetType
Enabling Type.Getype with overloads for assembly and type resolvers
Rama krishnan Raghupathy [Thu, 3 Nov 2016 23:02:03 +0000 (16:02 -0700)]
Merge pull request #7945 from ramarag/MemoryFailPoint
Disabling checks in MemoryFailPoint for Xplat
Jan Kotas [Thu, 3 Nov 2016 22:33:08 +0000 (15:33 -0700)]
Revert "Enable optimization of structs"
Pat Gavlin [Thu, 3 Nov 2016 22:00:16 +0000 (15:00 -0700)]
Fix int->long cast folding for 32-bit platforms.
The RyuJIT frontend does not allow for the folding of an int->long cast
that is known to overflow into a throw outside of global morph. The
folding, however, was unconditionally changing the operand to the cast
from an integer constant to a long constant, which can change the
behavior of the cast. This rewrite is not necessary, so this change
simply removes it.
Rama Krishnan Raghupathy [Thu, 3 Nov 2016 19:05:56 +0000 (12:05 -0700)]
Enabling Type.GetType with overloads for assembly and type resolvers
Alex Ghiondea [Thu, 3 Nov 2016 21:09:32 +0000 (14:09 -0700)]
The SecureStringMarshal type was an artifact of the layering we had in CoreFx.
This type should not have been added to CoreCLR as part of the SecureString move.
Michelle McDaniel [Thu, 3 Nov 2016 16:13:06 +0000 (09:13 -0700)]
Force store of lcl var for shift/rotate by 32
In the shift/rotate by 32 case, we move the lo op into the high op. In the
shift case, we set the lo op to 0, in the rotate case, we are swapping lo
and hi, so we also move hi into lo. In the instance that we are reusing
the same lclvars, we need to explicitly store the ops that we are moving
so that we don't overwrite the vars before using them. This change
explicitly replaces the uses of the ops with lclvars so we don't overwrite
vars before finishing the swap.
sandreenko [Thu, 3 Nov 2016 20:58:25 +0000 (13:58 -0700)]
Merge pull request #7908 from sandreenko/fix_tail_call_in_R2R
Correct pattern check for tail calls
Vance Morrison [Thu, 3 Nov 2016 20:37:53 +0000 (13:37 -0700)]
Fix a issue where ETW was losing RCW and CCW in heap dumps. (#7971)
Basically our builk events for RCWs and CCWs were too big, causing ETW to reject them, and thus losing information.
However you have to have at least several hundred befor you have enough to force the problem.
Pat Gavlin [Thu, 3 Nov 2016 19:09:50 +0000 (12:09 -0700)]
Merge pull request #7958 from pgavlin/VSO280123
Make a few VN methods tolerant of NoVN.
John Chen [Thu, 3 Nov 2016 18:59:40 +0000 (11:59 -0700)]
Make CrossGen documentation up to date (#7973)
Sergey Andreenko [Mon, 31 Oct 2016 19:31:10 +0000 (12:31 -0700)]
Correct pattern check
sequense
localVariable = cast.int(call.small_type());
return localVaribale
is correct tail call pattern.
Pat Gavlin [Thu, 3 Nov 2016 17:35:28 +0000 (10:35 -0700)]
Merge pull request #7956 from pgavlin/VSO280110
Properly update block flags during branch opt.
Pat Gavlin [Thu, 3 Nov 2016 17:35:05 +0000 (10:35 -0700)]
Format code.
Pat Gavlin [Thu, 3 Nov 2016 17:29:57 +0000 (10:29 -0700)]
Merge pull request #7959 from pgavlin/VSO280116
Fix an assert condition in lowering.
Pat Gavlin [Thu, 3 Nov 2016 17:29:32 +0000 (10:29 -0700)]
Format code.
Andy Ayers [Thu, 3 Nov 2016 17:11:25 +0000 (10:11 -0700)]
Merge pull request #7968 from AndyAyersMS/CaptureNewObservations
Inliner: capture new observations
Stephen Toub [Thu, 3 Nov 2016 10:06:36 +0000 (06:06 -0400)]
Mark NonRandomizedStringEqualityComparer as serializable (#7966)
Andy Ayers [Tue, 1 Nov 2016 23:22:10 +0000 (16:22 -0700)]
Inliner: capture new observations
Rebase the DiscretionaryPolicy on the EnhancedLegacyPolicy, and
capture and report some of the new observations that have been added
recently.
This change adds a new HAS_GC_STRUCT observation in addition to the
more specific RARE_GC_STRUCT, so that the DiscretionaryPolicy can
be notified of GC struct locals and temps for all candidates, not
just ones invoked from rare call sites.
No changes in codegen.
Jonghyun Park [Thu, 3 Nov 2016 02:46:19 +0000 (11:46 +0900)]
Revises 'JitTimer::PrintCsvHeader' to use 'append' mode (#7073)
* Revises 'JitTimer::PrintCsvHeader' to use 'append' mode
The current implementation of 'JitTimer::PrintCsvHeader' opens a CSV
file with 'read' mode and re-opens it with 'write' mode if it is absent.
The current implementation has two problems:
First, "read" file pointer will be closed even if it is null,
which is reported in #7072.
Second, "write" file pointer will be leaked as there is no corresponding
fclose.
This commit rewrites 'JitTimer::PrintCsvHeader' to fix #7072.
varocarbas [Thu, 3 Nov 2016 02:33:46 +0000 (03:33 +0100)]
ParseNumber & related more efficient (#3163)
David Mason [Thu, 3 Nov 2016 02:23:18 +0000 (19:23 -0700)]
Merge pull request #7936 from vancem/EventSourceDisposeRace.11-01-16
Fix Deadlock associated with ETW commands happening during dispose.
Dmitry-Me [Thu, 3 Nov 2016 01:38:49 +0000 (04:38 +0300)]
Fix race condition (#7928)
chunseoklee [Wed, 2 Nov 2016 22:18:53 +0000 (07:18 +0900)]
Suppress SOS.NETCore.dll error msg in release mode (#6975)
Suppress error msg when running coreclr without
SOS.NETCore.dll in release mode
Pat Gavlin [Wed, 2 Nov 2016 21:52:58 +0000 (14:52 -0700)]
Fix an assert condition in lowering.
Tail calls via the tail call helper must live in GC-safe blocks. A block
is GC-safe if either it or all of its dominators are also GC-safe. If we
are not optimizing, we will not compute dominators, and will therfore
not mark blocks whose dominators are all GC-safe. When deciding whether
or not we need to make a block GC-safe during tail call morphing,
however, we allow a tail call in an unmarked block iff the entry block
is GC-safe, as the entry block trivially dominates every reachable
block. This change fixes an assert that was not allowing for a tail call
in a block that is not marked GC-safe where the entry block is marked
GC-safe.
Jonghyun Park [Wed, 2 Nov 2016 21:44:24 +0000 (06:44 +0900)]
Set pRetVal as NULL on error inside 'VIRTUALCommitMemory' (#7920)
Pat Gavlin [Wed, 2 Nov 2016 21:23:09 +0000 (14:23 -0700)]
Merge pull request #7917 from pgavlin/VSO280120
Fix VSO280120.
Pat Gavlin [Wed, 2 Nov 2016 21:08:41 +0000 (14:08 -0700)]
Make a few VN methods tolerant of NoVN.
There were a few methods on the VN store that were not resilient to
being provided with the "no value number" sentinel value. This lack of
resilience caused assertion failures during assertion propagation when
generating assertions for trees marked with "no value number".
Pat Gavlin [Wed, 2 Nov 2016 19:39:40 +0000 (12:39 -0700)]
Properly update block flags during branch opt.
`fgOptimizeBranch` was updating the flags for the predecessor block
using the flags from the wrong destination block. This was causing an
assert in `AllocateObjects` when verifying that all blocks that contain
`allocObj` nodes are properly marked with `BBF_HAS_NEWOBJ`.
Pat Gavlin [Mon, 31 Oct 2016 22:32:29 +0000 (15:32 -0700)]
Fix VSO280120.
st.lclFld decomposition was splitting a st.lclFld into two of the same
without bumping the destination var's ref count.
Dmitry-Me [Wed, 2 Nov 2016 17:42:41 +0000 (20:42 +0300)]
Check pointer before dereferencing it (#7930)
Jan Kotas [Wed, 2 Nov 2016 16:54:14 +0000 (09:54 -0700)]
Merge pull request #7953 from dotnet-bot/from-tfs
Merge changes from TFS
Sivarv [Wed, 2 Nov 2016 15:01:42 +0000 (08:01 -0700)]
Merge pull request #7943 from sivarv/rotateFix
Exclude GT_ROL/ROR from ==/!= 0 compare peep optimization
Davis Goodin [Wed, 2 Nov 2016 14:37:33 +0000 (09:37 -0500)]
Specify .NETCoreApp,Version=v1.1 for test projects (#7937)
Jan Kotas [Wed, 2 Nov 2016 14:35:30 +0000 (07:35 -0700)]
Fix build break
[tfs-changeset: 1636334]
Jan Kotas [Wed, 2 Nov 2016 14:08:50 +0000 (07:08 -0700)]
Merge pull request #7951 from jkotas/revert-correct-parameter-names
Revert "Correct parameter names."
Jan Kotas [Wed, 2 Nov 2016 14:06:30 +0000 (07:06 -0700)]
Revert "Correct parameter names."
This reverts commit
0cae769f3c60bf2a696d02a88527a46d5b211cd5.
Jan Kotas [Wed, 2 Nov 2016 13:55:43 +0000 (06:55 -0700)]
Merge pull request #7950 from jkotas/revert-parameter-name-changes
Revert "Correct parameter names."
Jan Kotas [Wed, 2 Nov 2016 13:53:53 +0000 (06:53 -0700)]
Revert "Correct parameter names."
This reverts commit
a212082aa4cea8529b4c78844e2a7408ff6f7969.
Hanjoung Lee [Wed, 2 Nov 2016 13:23:57 +0000 (22:23 +0900)]
[ARM] Implement BitPosition with arm intrinsics (#7819)
Faizur Rahman [Wed, 2 Nov 2016 06:53:04 +0000 (23:53 -0700)]
Merge pull request #7904 from shrah/master
Update S.R.InteropServices to ns2.0
John Chen [Wed, 2 Nov 2016 05:50:56 +0000 (22:50 -0700)]
Fix build break 814276: FEATURE_SPAN_OF_T disabled in runtime but enabled in DAC
[tfs-changeset: 1636322]
Jan Kotas [Wed, 2 Nov 2016 01:59:16 +0000 (18:59 -0700)]
Merge pull request #7944 from dotnet-bot/from-tfs
Merge changes from TFS
Ian Hays [Wed, 2 Nov 2016 01:58:47 +0000 (18:58 -0700)]
Expose missing API in models.xml (#7940)
progress towards https://github.com/dotnet/corefx/issues/12812 and https://github.com/dotnet/corefx/issues/12398
Jan Kotas [Wed, 2 Nov 2016 01:57:36 +0000 (18:57 -0700)]
Update GCInfoDecoder with CoreRT changes (#7938)
Porting most of the changes from https://github.com/dotnet/corert/pull/2116 to make GCInfoDecoder compile in CoreRT to keep the different copies in sync.
DDCloud [Wed, 2 Nov 2016 00:48:23 +0000 (17:48 -0700)]
Disabling checks in MemoryFailPoint for Xplat till primitives like VirtualQuery
are implmemented properly
sivarv [Wed, 2 Nov 2016 00:00:32 +0000 (17:00 -0700)]
Rotate Left/Right xarch instructions don't set ZF and ZF flag.
John Chen [Tue, 1 Nov 2016 23:20:08 +0000 (16:20 -0700)]
Fix GitHub build break caused by change #1636179
[tfs-changeset: 1636227]
Tarek Mahmoud Sayed [Tue, 1 Nov 2016 23:17:50 +0000 (16:17 -0700)]
Enable the rest of netstandard1.7 globalization APIs (#7935)
* Enable more netstandard 1.7 Globalization APIs
* Enable Windows support for the new APIs
* Fix spaces and line alignments
Andy Ayers [Tue, 1 Nov 2016 20:18:03 +0000 (13:18 -0700)]
Merge pull request #7923 from AndyAyersMS/FixNoCandidateBug
Jit: spill stack for non-candidate calls
Vance Morrison [Tue, 1 Nov 2016 19:21:34 +0000 (12:21 -0700)]
Fix deadlock when commands race with EventSource Shutdown
John Chen [Tue, 1 Nov 2016 18:34:27 +0000 (11:34 -0700)]
CrossGen: Copy all entries of Debug Directory Table (#7849)
Fixes issue #6482.
dotnet-bot [Tue, 1 Nov 2016 18:26:34 +0000 (11:26 -0700)]
Fix build break 814217 by removing inconsistencies between BCL and runtime
[tfs-changeset: 1636179]
Joseph Tremoulet [Fri, 14 Oct 2016 18:20:49 +0000 (14:20 -0400)]
Handle simd checks in OptimizeRangeCheck
When the index is a constant, this code applies to `GT_SIMD_CHK` equally
well as it applies to `GT_ARR_BOUNDS_CHECK`.
Stephen Toub [Tue, 1 Nov 2016 15:47:18 +0000 (11:47 -0400)]
Use nameof throughout mscorlib (#7929)
Carol Eidt [Tue, 1 Nov 2016 15:42:01 +0000 (08:42 -0700)]
Merge pull request #7677 from CarolEidt/StructOpts
Enable optimization of structs
Jan Kotas [Tue, 1 Nov 2016 14:06:37 +0000 (07:06 -0700)]
Merge pull request #7926 from dotnet-bot/from-tfs
Merge changes from TFS
Michelle McDaniel [Tue, 1 Nov 2016 13:47:54 +0000 (06:47 -0700)]
Merge pull request #7883 from adiaaida/fixDecomposeRotate
Fix DecomposeRotate for > 32 case
Mukul Sabharwal [Tue, 1 Nov 2016 13:28:58 +0000 (14:28 +0100)]
Use resolver to get IL header and size (#7924)
Jan Kotas [Tue, 1 Nov 2016 05:58:49 +0000 (22:58 -0700)]
Fix build break on x86
[tfs-changeset: 1636135]
Andy Ayers [Tue, 1 Nov 2016 05:57:25 +0000 (22:57 -0700)]
Use normal verbosity logging when building tests (#7921)
Test build log file is ~4GB with diag verbosity. With normal it's about 20MB.
This seems to speed up the test build for me considerably.
See #7916.
Jeremy Kuhne [Tue, 1 Nov 2016 05:57:11 +0000 (22:57 -0700)]
Move SR adapter into root (#7925)
CoreFX creates SR in the System namespace. While putting the CoreCLR SR
adapter in the root may be unconventional, it allows us to keep the
shared code identical.
Jan Kotas [Tue, 1 Nov 2016 05:56:58 +0000 (22:56 -0700)]
Merge pull request #7915 from dotnet-bot/from-tfs
Merge changes from TFS
Joseph Tremoulet [Tue, 1 Nov 2016 05:31:02 +0000 (01:31 -0400)]
Merge pull request #7912 from JosephTremoulet/FlagVectorCounts
Flag vector counts
Joseph Tremoulet [Tue, 1 Nov 2016 05:29:57 +0000 (01:29 -0400)]
Merge pull request #7910 from JosephTremoulet/FixInitTest
Fix check for initial induction variable value
Joseph Tremoulet [Thu, 27 Oct 2016 19:04:40 +0000 (15:04 -0400)]
Fix check for initial induction variable value
Checking that `head` dominates `entry` is insufficient, becuase there may
be a codepath from `head` through other nodes that modify the induction
variable to `entry`. Check instead that all preds of `entry` are either
`head` or in the loop.
Resolves #7906.
Joseph Tremoulet [Tue, 1 Nov 2016 05:24:10 +0000 (01:24 -0400)]
Merge pull request #7909 from JosephTremoulet/FixLoopEntryCheck
Skip removed loops in `optBlockIsLoopEntry`
Jan Kotas [Tue, 1 Nov 2016 03:14:32 +0000 (20:14 -0700)]
Fix build break
[tfs-changeset: 1636112]
Andy Ayers [Tue, 1 Nov 2016 00:26:07 +0000 (17:26 -0700)]
Jit: spill stack for non-candidate calls
Closes #7907.
Spill the evaluation stack at non-candidate calls since the call may
modify locals or args that are currently live on the stack.
In the attached test case I can't get the problematic IL sequence to
happen for the `F0` variant, but it is there for the `F1` variant.
No diffs seen in corlib or any of the framework assemblies that are
run via jit-dasm.
Gaurav Khanna [Tue, 1 Nov 2016 00:03:50 +0000 (17:03 -0700)]
Merge pull request #7854 from dotnet-bot/master-UpdateDependencies
Update CoreClr, CoreFx to beta-24631-02, beta-24631-01, respectively (master)
Russ Keldorph [Mon, 31 Oct 2016 23:52:44 +0000 (16:52 -0700)]
Merge pull request #7918 from RussKeldorph/repoint-corefx-1.1.0
Make dotnet/coreclr reference corresponding dotnet/corefx branch
Jan Kotas [Mon, 31 Oct 2016 23:46:32 +0000 (16:46 -0700)]
Merge pull request #7867 from Clockwork-Muse/ArgumentException
Replace hardcoded parameter names to ArgumentException with nameof
Russ Keldorph [Mon, 31 Oct 2016 23:18:11 +0000 (16:18 -0700)]
Make dotnet/coreclr reference corresponding dotnet/corefx branch
We run CoreFx tests under JitStress modes in CoreCLR. Previously, we were
always pulling down CoreFx master, which is bad for CoreCLR release
branches when things are out of synch.
Pat Gavlin [Mon, 31 Oct 2016 22:33:53 +0000 (15:33 -0700)]
Merge pull request #7885 from pgavlin/VSO278376.2
Fix VSO 278376.
Pat Gavlin [Mon, 31 Oct 2016 22:33:29 +0000 (15:33 -0700)]
Merge pull request #7905 from pgavlin/VSO279380
Fix VSO 279380.
Joseph Tremoulet [Fri, 14 Oct 2016 18:11:29 +0000 (14:11 -0400)]
Add LPFLG_SIMD_LIMIT flag
This flag identifies loops whose limit was a Vector<T>.Count expression.
This can be used in heuristics to target more aggressive optimizations at
SIMD code.
Andy Ayers [Mon, 31 Oct 2016 21:47:16 +0000 (14:47 -0700)]
Merge pull request #7852 from AndyAyersMS/InlinerSmallFixes
Inliner: Fix LegacyPolicyAssert; Dump prejit noinline reasons
Michelle McDaniel [Fri, 28 Oct 2016 22:08:42 +0000 (15:08 -0700)]
Fix DecomposeRotate for > 32 case
In DecomposeRotate, we had a bug where we assumed loOp1 was
always gtLong's op1 and hiOp1 is always gtLong's op2 when we
call RepresentOpAsLocalVar, which is not true in the > 32 case, where we
swap hi and lo before generating the shift instructions for the rotate.
This change refactors the code so that we can correctly call
RepresentOpAsLocalVar with which of gtLong's ops hi and lo are.
Alex Ghiondea [Mon, 31 Oct 2016 21:37:00 +0000 (14:37 -0700)]
A couple of changes included:
- Make sure we have a way to disable building with the Roslyn compiler. That should only happen in the internal build.
- Update the FeatureSpanOfT to also condition on using the latest compiler
- Introduce an if/def in string.cs for code that uses the ref locals / ref return.
- Make sure we only include the CryptographicException when building CoreCLR
- Update the PInvokeMap.txt file to include the right set of APIs.
- Remove duplicated APIs from PInvokeMap.txt
[tfs-changeset: 1636041]
AlexGhiondea [Mon, 31 Oct 2016 20:06:26 +0000 (13:06 -0700)]
Merge pull request #7884 from AlexGhiondea/Port_SecureString
Bring to CoreCLR the SecureString implementation from CoreFx.
Joseph Tremoulet [Fri, 14 Oct 2016 18:10:10 +0000 (14:10 -0400)]
Add GTF_ICON_SIMD_COUNT flag
This flag is used to indicate that an integer constant in the IR is the
result of replacing the Vector<T>.Count intrinsic. This can be used to
drive heuristics focusing more aggressive optimizations on SIMD code.
Joseph Tremoulet [Fri, 14 Oct 2016 18:08:48 +0000 (14:08 -0400)]
Skip removed loops in `optBlockIsLoopEntry`
This allows running this utility downstream of loop unrolling.