[RuntimeAsync] Merge from upstream#3079
Merged
jakobbotsch merged 469 commits intodotnet:feature/async2-experimentfrom Apr 10, 2025
Merged
[RuntimeAsync] Merge from upstream#3079jakobbotsch merged 469 commits intodotnet:feature/async2-experimentfrom
jakobbotsch merged 469 commits intodotnet:feature/async2-experimentfrom
Conversation
--------- Co-authored-by: Larry Ewing <lewing@microsoft.com> Co-authored-by: Jan Kotas <jkotas@microsoft.com>
Now that the JIT is able to inline methods with exception handling, we need to disable inlining explicitly to maintain the test invariants. Inlining is no longer disabled as side-effect of exception handling. Fixes #113813
…0.6998.166 (#113802) * Automated bump of chrome version * Update V8 version in BrowserVersions.props --------- Co-authored-by: github-actions[bot] <github-actions[bot]@users.noreply.github.com> Co-authored-by: Ilona Tomkowicz <32700855+ilonatommy@users.noreply.github.com>
* Fix possible null reference exception in SslStream.Protocol * Update src/libraries/System.Net.Security/src/System/Net/Security/SslStream.Protocol.cs Co-authored-by: Marie Píchová <11718369+ManickaP@users.noreply.github.com> * code review feedback * Fix build --------- Co-authored-by: Marie Píchová <11718369+ManickaP@users.noreply.github.com>
* [mono][sgen] Fix DUMP_GRAPH debug option build for tarjan bridge * [mono][sgen] Don't create ScanData* during debug dumping of SCCs It serves no purpose and it would later crash the runtime since we didn't patch the lockword back in place. * [mono][sgen] Fix some null deref crashes in DUMP_GRAPH debug option * [mono][tests] Add bridge tests These are ported from some of the bridge tests we had on mono/mono. In order to test them we compare between the output of the new and the tarjan bridge.
…50317.4 (#113819) Microsoft.DotNet.XHarness.CLI , Microsoft.DotNet.XHarness.TestRunners.Common , Microsoft.DotNet.XHarness.TestRunners.Xunit From Version 10.0.0-prerelease.25164.1 -> To Version 10.0.0-prerelease.25167.4 Co-authored-by: dotnet-maestro[bot] <dotnet-maestro[bot]@users.noreply.github.com>
When a NULL reference exception occurs in a JIT helper or a VSD stub, runtime pretends the exception occured in the managed caller. There is a bug on x86 Windows where the COMPlusThrowCallback considers that frame to be the frame where the exception actually occurred (based on the m_crawl.isFirst). In case the call to the helper is the last instruction in a try region, the exception handler lookup would reject that address and the exception may not get handled at the right place or at all. This change fixes it by ensuring that the m_crawl.isFirst is not set when the frame is not the frame of the failure. Close #113106
Remove explicit GC Poll from GC.SuppressFinalize(). Remove GC Poll from FCall and move it to managed, Buffer.__BulkMoveWithWriteBarrier(). Remove unneccessary explicit FC_GC_POLL() calls. Remove unneccessary explicit FC_GC_POLL_RET() calls. Remove unmanaged GCPoll implementation. Implement JIT intrinsic for GC Poll, Thread.FastGCPoll.
Add ease-of-use overloads for MLKem API where appropriate.
Only the secret stub parameter goes through this path to create an edge in the graph, and it was missing a check for whether it was going into the same register as it was in. This would cause register homing to generate unnecessary moves into a temporary register because it thought there was a conflict. Usually the secret stub parameter has to be allocated to a callee saved variable, so we typically do not see this inefficiency.
…er and DynamicMethod (#113666)
* [RISC-V] Detect Zba & Zbb extensions * [RISC-V] Reflect constants in cpufeatures.h to HardwareIntrinsicHelpers.cs * [RISC-V] Use hwprobe syscall directly * [RISC-V] Check syscall return value and use baseline support on failure * [RISC-V] Regenerate JIT version GUID * [RISC-V] Add Zba Zbb extension configuration variables
* Don't fail altjit compiles when RunAltJitCode=0 With RunAltJitCode=0, the JIT returns CORJIT_SKIPPED. We don't want superpmi to fail these compiles. * Enable SuperPMI replay for APX ISA testing Create a new `runtime-coreclr superpmi-replay-apx` pipeline that does SuperPMI replay an AltJit with various APX enabling configuration options set. This enables some amount of APX testing even without hardware available to run the generated code. Notes: 1. Create a new jit-replay-pipeline.yml template to share the superpmi-replay pipeline implementation with the existing pipeline. 2. The existing pipeline is designated the "standard" replay type and the new pipeline the "apx" replay type. 3. The APX pipeline does replays for Windows x64 and Linux x64. 4. The configurations tested are: ``` a. "RunAltJitCode=0", "EnableAPX=1", "EnableApxNDD=1" b. "RunAltJitCode=0", "EnableAPX=1", "JitStressRex2Encoding=1" c. "RunAltJitCode=0", "EnableAPX=1", "JitStressPromotedEvexEncoding=1" d. "RunAltJitCode=0", "EnableAPX=1", "JitStressRegs=4000" e. "RunAltJitCode=0", "EnableAPX=1", "EnableApxNDD=1", "JitStressRex2Encoding=1", "JitStressPromotedEvexEncoding=1", "JitStressRegs=4000" ```
…ence-packages build 20250320.2 (#113820) Microsoft.SourceBuild.Intermediate.source-build-reference-packages From Version 10.0.616401 -> To Version 10.0.617002 Co-authored-by: dotnet-maestro[bot] <dotnet-maestro[bot]@users.noreply.github.com>
…OT (#113850) CORJIT_FLAG_READYTORUN and CORJIT_FLAG_PREJIT were always both set together. Introduce a new name to be more explicit of a concept that includes both ReadyToRun and NativeAOT (AOT == Ahead Of Time). Introduce new JIT helpers `IsAot` (for either NativeAOT or R2R), `IsNativeAot`, and `IsReadyToRun`. Remove the name "ngen" in a bunch of places in the JIT code.
- Convert to the FCALL fast path/QCALL slow path approach - Make EnterHelperResult/LeaveHelperAction into enum class so that they can safely be silently marshaled between native and managed - Move the lockTaken flag handling to managed code so we can share more helpers Benchmarking indicates this may actually be faster than the previous approach. I'm a bit skeptical, but its not out of the question.
…formation (#112825) * Fix an edge case in the Tarjan SCC that lead to losing xref information In the Tarjan SCC bridge processing there's a color graph used to find out connections between SCCs. There was a rare case which only manifested when a cycle in the object graph points to another cycle that points to a bridge object. We only recognized direct bridge pointers but not pointers to other non-bridge SCCs that in turn point to bridges and where we already calculated the xrefs. These xrefs were then lost. * Add test case to sgen-bridge-pathologies and add an assert to catch the original bug * Add review --------- Co-authored-by: Vlad Brezae <brezaevlad@gmail.com>
…e-file bundle probing" (#113776) Undo the revert in dotnet/runtime#113738 Fix on top of the original: Make a copy of the host contract information on initialization instead of just operating on a pointer to the struct. This should address intermittent failures on shutdown where a thread is still running after coreclr_shutdown has completed and the actual contract struct provided by a host may be cleaned up. The assumption is that any function pointers provided on the host contract at initialization should remain valid for the process lifetime.
* Fix SslStreamDisposeTest.Dispose_ParallelWithHandshake_ThrowsODE test failures * Fix build
The dot.net redirector is currently having issues, we should be using the script directly from the CDN anyway.
* fix cast asserts * fix containment of CreateScalar * add tests
When we hit a generic recursion over arrays, we detect it, but there was no cutoff logic to cut off the expansion. The result is a compile time OOM. This adds a cutoff, using the same cutoff number we use to look for deep generics.
* Intrinsify LeadingZeroCount * TrailingZeroCount * PopCount * Build fix * Remove assertion from default case * cosmetics * Separate bit counting from System.Math intrinsics * Fix assert * Revert unnecessary diffs
* Convert to [Fact] * rename class * Sve.IsSupported
Alpine is having problems with not enough file handles.
The public API surface looks exactly like RequiresUnreferencedCode, but the private implementation details vary slightly. I don't think there's a meaningful difference in the annotations for RDC vs RUC.
Fix #113458 -- failures on Azure Linux 3 due to running out of file handles.
prepare-signed-artifacts.yml tries to download artifacts from all items in the buildStages array, but since we disabled the workloads job it fails to find an artifact. Comment out the entry in yml instead.
Describe current stack buffer overrun detection capabilities.
…set to a directory (#114362) Currently, host tracing can output to a file specified by the `COREHOST_TRACEFILE` environment variable. With multiple concurrent processes, the tracing gets interleaved in the resulting log. With this change, if `COREHOST_TRACEFILE` is set to a directory that exists, the host will trace to a file named `<exe_name>.<pid>.log` in that directory. If the specified path does not exist or is not a directory, the host uses the existing behavior of tracing to the file specified.
Implements Vector Round intrinsic in terms of PackedSimd.
* slightly better calculation of stack frame for SVE ldr/str * add comments * review feedback
Co-authored-by: github-actions[bot] <github-actions[bot]@users.noreply.github.com> Co-authored-by: Ilona Tomkowicz <32700855+ilonatommy@users.noreply.github.com>
This PR changes the patchpoint key to be a pair <function version, IL offset> instead of by IP of the return address. This is in preparation for runtime async changes where suspension inside OSR methods have to be resumed via resumption stub -> tier 0 -> OSR method. The latter step needs to retransition into the right OSR method from the tier 0 resumption code. With the current key this is problematic because the return address is not going to be the same as the original patchpoint that induced the OSR transition.
Member
|
x64 builds and passes runtime tests. For x86 I see build errors. I think I have fixes for this in my other PR. |
Member
|
x86 builds, but a few tests fail - for example I think I have a fix for that too |
Member
|
now only R2RDump fails on x86 ( I think it is ok to merge and take a look at that later. |
Member
Author
|
Thanks for fixing that! |
jakobbotsch
merged commit bbb9741
into
dotnet:feature/async2-experiment
1 of 7 checks passed
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
20 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Fixes:
BuildResumptionStubCalliSignaturewas building a signature withHASTHIS | EXPLICITTHIScalling convention always, even for signatures without any 'this'. With Support HasThis and ExplicitThis calling conventions in AssemblyBuilder and DynamicMethod runtime#113666 this broke as we started reordering the first argument with ret buffers.getEHinfoto handle getting EH clause information when trying to inline methods with transient IL, like async thunksg_isNewExceptionHandlingEnabledcc @VSadov