Age | Commit message (Collapse) | Author | Files | Lines |
|
The case of the COMPlus_ environment variables matters on Linux. Fix instructions to be consistent.
|
|
Just deleted one of the two "for"s in the Setting up your GIT Clone of the CoreCLR Repository.
|
|
Handle cases where a `GetType()` call on a generically typed object
feeds into a type equality comparison.
These calls have constraint prefixes. For value classes the constraint
tells us the type and so we can avoid the box and call and just construct
that type directly. For ref classes the type test can usually reduce to
a method table comparison.
Also, handle cases that arise in prejiited code better, by generalizing
how the jit looks for class handles from type construction trees.
Added test cases.
Closes #14304.
|
|
Undo unnecessary change from DateTime.Parse commit
|
|
- In the lab, the path where the JitBench sources are downloaded is already too long, which eventially causes `dotnet-install.ps1` to fail. Part of the changes are to reduce the PATH where the tests are placed.
- Updated the JitBench version/commit being used. This one is already on 2.1
- Fixing typo on `perf.groovy`
- Increasing the timeout value as JitBench keeps failing on the lab.
|
|
Removed unused opers and code
|
|
* Fix zap and ready to run disabling
While the COMPlus_ZapDisable and COMPlus_ReadyToRun config settings
can be used to disable using crossgened / ready to run images, loading
IL from such images fails.
This change foxes that.
* Reflect PR feedback
Remove ReadyToRunInfo::IsReadyToRunEnabled from Module::Initialize again
and move the same check in the ReadyToRunInfo::Initialize before we
start checking the PE file properties.
|
|
Fix incorrect name of class in RyuJIT overview
|
|
This change shouldn't have been included in my DateTime.Parse span change. It was left-over from some local experimentation. The change doesn't hurt anything functionally, but it does undo a small optimization; I've not measured whether it's impactful or not, so I'm putting it back the way it was.
|
|
* [GDBJIT] Fix DW_AT_comp_dir setting
We should use cuPath to set dirPath in NotifyGdb::EmitDebugInfo instead of
DebugString[1].
* [GDBJIT] Make gdbjit thread-safe
NotifyGdb::MethodPrepared method can be called from multiple threads
simultaneously in this case gdbjit will work incorrectly as it uses
global variable without synchronization.
|
|
|
|
* GT_DIV_HI and GT_MOD_HI are not used anywhere
* genCodeForBinary doesn't handle GT_MUL_LONG
* OperIsHigh is not used anywhere
|
|
respectively (#14378)
|
|
This change fixes a break in build of coreclr on machines with libnuma
installed. The problem was that the numa header contains a couple of
inlined functions and we were using one of them. That made it to
have a hard reference to a function from the numa library that we
need to be soft so that coreclr can work even on machines without
the libnuma installed.
|
|
Fix RyuJIT/arm32 asserts
|
|
Delete some dead code related to Windows Phone and code access security
|
|
Fix handling of continuations for captured context
|
|
My recent delegate optimization for async methods introduced a race condition that breaks posting back to a captured context. If the awaited task completes between the time we check IsCompleted and when we try to store the continuation into the Task, we need to queue the continuation for execution, but my change incorrectly queued it to the thread pool rather than to the captured context.
|
|
Add Span-based methods to DateTime{Offset}
|
|
Fix #14199: propagate `GTF_EXCEPT` bits from the end of `GT_FIELD_LIST`
lists to the beginning, to avoid "Missing flags on tree" asserts.
Fix #14198: for RyuJIT/arm32, `GT_BITCAST` needs to be a MultiRegOp.
This is required when a varargs function, which includes the tailcall
helper, needs to pass a double in integer registers. We can end up
with `GT_PUTARG_REG/long(GT_BITCAST/long(double tree))`.
Fixed various GenTree node flags dumping issues.
|
|
Update BuildTools, CoreClr, CoreFx, PgoData to prerelease-02106-01, preview1-25806-01, preview1-25806-02, master-20171006-0051, respectively (master)
|
|
* undef NULL to fix compilation problems on FreeBSD after 08d39dd.
* remove unneccesary #ifndef NULL
|
|
|
|
* impImportCall: make the failed imports obvious.
Delete unnecessary and confusing dependency on callRetType.
* delete unreached code that calls impImportCall
|
|
Check for native value type has to be first to handle byref-like native value types correctly.
|
|
[Arm64] Implement GT_XADD, GT_XCHG, GT_CMPXCHG ...
|
|
Add Span-based Convert Base64 methods
|
|
|
|
Currently, when coreclr is built on machine that has libnuma installed,
it adds libnuma to its dependencies and so it cannot be used on machines
that don't have libnuma installed. But that is not necessary, the
dependency was changed to soft dependency - PAL loads it only if it is
present and falls back to the non-numa behavior if it is not found.
This change removes the hard libnuma dependency.
|
|
preview1-25806-01, preview1-25806-02, master-20171006-0051, respectively
|
|
The existing implementations just format to a StringBuilder and then get the string from that StringBuilder. For now at least, these TryFormats share the same code, and just copy the StringBuilder to the span instead of using ToString.
|
|
|
|
Workaround regressions from the span switch and also fix a few other issues that popped while profiling.
- Specialize Compare{String} methods to take one span and one string
- Ensure frequently called methods get inlined
- Avoid unnecessary writes to a ref
- Avoid unnecessary Math.Pow calls
|
|
|
|
The base implementation makes several virtual calls that the derived implementation can avoid.
|
|
|
|
So that we can reuse the same implementation for the public span-based APIs.
|
|
|
|
With the update to llvm 3.9, the pgo data is incompatible since it was
generated with llvm 3.6. Currently, both the pgo data generation builds
and the automatic updates of counts are currently down, so pgo data is
not being updated. Until these problems are fixed, disable pgo
optimization for release builds on linux.
|
|
* enable lttng for FreeBSD. it is supported for user space
* unify Linux and FreeBSD branches for lttng. At lest for now, they are identical
|
|
[RyuJIT/armel] Fix regression
|
|
Partial port of https://github.com/dotnet/coreclr/pull/10264
Fixes #4640
Signed-off-by: dotnet-bot <dotnet-bot@microsoft.com>
|
|
The images now contains clang 3.9 with PGO support
|
|
|
|
(#14337)
Fix new deadlock possibility in ReaderWriterLockSlim from a recent change
Fixed the following scenario that would lead to deadlock (rare, but frequently reproducible if code is changed):
- Thread T0 signals the write waiter event or the upgradeable read waiter event to wake a waiter
- There are no threads waiting on the event, but T1 is in WaitOnEvent() after exiting the spin lock and before actually waiting on the event (that is, it's recorded that there is one waiter for the event). It remains in this region for a while, in the repro case it typically gets context-switched out.
- T2 acquires the RW lock in some fashion that blocks T0 or T3 from acquiring the RW lock
- T0 or T3 fails to acquire the RW lock enough times for it to enter WaitOnEvent for the same event as T1
- T0 or T3 resets the event
- T2 releases the RW lock and does not wake a waiter because the reset at the previous step lost a signal but _waiterStates was not updated to reflect that
- T1 and other threads begin waiting on the event, but there's no longer any thread that would wake them
|
|
* JIT: Wrap some runtime lookups in new node type
Work based on the plan outlined in #14305.
Introduce a new unary node type GT_RUNTIMELOOKUP that wraps existing
runtime lookup trees created in `impLookupToTree` and holds onto the
handle that inspired the lookup.
Note there are other importer paths that create lookups directly that
we might also want to wrap, though wrapping is not a requirement for
correctness.
Keep this node type around through morph, then unwrap and just use
the wrapped node after that.
* JIT: More enhancements to type equality testing
The jit is now able to optimize some type equality checks in shared
method instances where one or both of the types require runtime lookup,
for instance comparing `T` to a value type, or comparing two different
value types `V1<T>` and `V2<T>`.
Add two new jit interface methods, one for testing for type equality and
the other for type casting.
These return Must/MustNot/Maybe results depending on whether or not
the equality or cast can be resolved at jit time.
Implement the equality check. Use this to enhance the type equality opts
in the jit for both direct comparison and the checking done by unbox.any.
|
|
* fix gtDispTree
It is legal situation when we pass ILNone (for example for local stores:
gtDispChild(tree->gtOp.gtOp1, indentStack, IINone);). In this code we
poped value and didn't push anything back, it ended with the stack
underflow.
|
|
Fix calling a function passing double-type stack arguments.
Fix #14251
|
|
GetObjFieldOffset function expects that cdaObj and cdaMT arguments
have CLRDATA_ADDRESS type (sign-extended) but in some cases we
use TADDR instead of CLRDATA_ADDRESS that isn't sign-extended.
Due to it some SOS commands (for example DumpRuntimeTypes) don't
work on x86 and ARM32.
|
|
* [x86/Linux][SOS] Add support for x86 in GetContextFromFrame
It's need for 'clrstack -f' command of SOS plugin on x86.
* [x86/Linux] Fix RtlpGetFunctionEndAddress function
We should use PTR_UNWIND_INFO instead of PUNWIND_INFO for pointer to
UNWIND_INFO structure because it's pointer from other process and
we need to use DAC to read data using it.
* [x86/Linux][SOS] Define DEBUG_STACK_CONTEXT for x86
It's needed for 'clrstack -f' command in libsosplugin.
* [x86/Linux] Fix undefined references in libmscordbi.so on x86
Asm block like following:
__asm fnsave currentFPUState
where currentFPUState is structure works with MSVC but leads to
undefined reference currentFPUState in the binary with other
compilers. So rewrite such asm blocks for them.
This patch fixes error "Unable to load 'libmscordbi.so'" during
execution of 'clrstack -f' command of SOS plugin on x86.
* [x86/Linux] Fix calling convention inconsistency
WINAPI and STDAPI are defined as __cdecl but in some cases functions
with these attributes are called using stdcall calling convention.
It leads to crashes during execution of 'clrstack -i' command of
SOS plugin on x86.
|