- 12 7月, 2023 17 次提交
-
-
由 Thays Grazia 提交于
-
由 Pavel Savara 提交于
-
由 Lakshan Fernando 提交于
* Enable finalizer events for EventPipe * FB
-
由 Katelyn Gadd 提交于
* Add browser-bench measurement for int32 and float equals * Add interp intrinsics for Vector128 float and double Equals methods * Implement Vector128 float and double Equals methods in jiterp * Add jiterp validation to make sure we never appendSimd(0) by accident
-
由 SingleAccretion 提交于
https://github.com/dotnet/corert/commit/08d78ae391e3eea9ad51265f867e42ee1649282d The original motivation for this was handling import stubs: ``` Function pointer equality comparison was not handling cross-module pointers correctly when optimizations were enabled (causes target pointers to be wrapped in jump stubs sometimes). The delegate equality comparison was hitting this bug. ``` We do not have import stubs anymore and unwrapping unboxing stubs serves no purpose here. Microbenchmarks of delegate equality show ~3x improvement with this change: ``` Bench_DelegateEquality_Positive_OpenStatic<10000000>() took: 355 ms Bench_DelegateEquality_Positive_ClosedStatic<10000000>() took: 367 ms Bench_DelegateEquality_Positive_ClosedInstance<10000000>() took: 371 ms Bench_DelegateEquality_Positive_OpenStatic<10000000>() took: 121 ms Bench_DelegateEquality_Positive_ClosedStatic<10000000>() took: 120 ms Bench_DelegateEquality_Positive_ClosedInstance<10000000>() took: 122 ms ``` Additionally, there is some desire to upstream changes for a portable RhGetCodeTarget implementation. Not having to deal with it at this relatively low-level layer will make things more robust.
-
由 Ankit Jain 提交于
-
由 Tanner Gooding 提交于
* Expose IUtf8SpanParsable * Have INumberBase implement IUtf8SpanParsable * Deduplicate some floating-point parsing logic * Refactoring the primitive parsing logic to support UTF-8 * Updating the primitive numeric types to include UTF-8 parsing support * Adding tests covering the new UTF-8 parsing support * Ensure that tests don't try to capture a span in lambda * Ensure that MatchChars does the right thing * Account for the switch from string to ROSpan<TChar> for currSymbol * Ensure EqualsIgnoreCaseUtf8_Scalar handles the remaining elements correctly * Allow alloc-free conversion for the UTF8 parsing fallback paths * Remove unnecessary attributes and ensure trailing elements are correctly handled * Fix the handling of UTF8 as per the feedback * Fix a slightly broken merge * Ensure the lengths are properly checked on the ASCII path * Ensure that length is defined for 32-bit * Responding to PR feedback and allow invalid sequences to self match * Fixing a copy/paste error * Fixing a build failure due to not assigning the out parameter
-
由 SingleAccretion 提交于
When you have built a CoreCLR CoreLib, it "poisons" the build s.t. running libraries tests with NAOT becomes impossible. This is because $(IlcFrameworkPath) draws from a shared framework that contains the CoreCLR CoreLib, while it should use the CoreLib from the AOT SDK. This targeted change changes the order in which these two assemblies get passed to ILC and allows one to use "/p:TestNativeAot=true" without additional workarounds.
-
由 Tarek Mahmoud Sayed 提交于
-
由 David Wrighton 提交于
- Move the handling for variance so that it is in the same place as it is for instance methods - Validate the behavior of such dispatch is equivalent to the dispatch behavior involving instance variant default dispatch - Rework the fix for #80350 to avoid violating an invariant of the type system (which is that all MethodDef's of a TypeDef are loaded as MethodDescs when loading the open type). Fixes #78621
-
由 Tanner Gooding 提交于
* Ensure that the CpuId tests set preferredVectorByteLength to a non-zero value * Ensure getPreferredVectorByteLength takes NAOT into account * Don't condition the preferred vector byte length based on stress mode
-
-
由 Alexander Radchenko 提交于
* Read/Write Big/LittleEndian support for Guid * Code review * Jan Kotas code review * Code review MemoryMarshal.Read<Guid>(b); [DoesNotReturn] * Use MemoryMarshal.AsRef * code review try AsBytes(Guid)
-
由 Milos Kotlar 提交于
* Add Native AOT cross-build support for iOS-like platforms * Add -hostos parameter for cross-components build --------- Co-authored-by: NMichal Strehovský <MichalStrehovsky@users.noreply.github.com>
-
由 Jeremy Koritzinsky 提交于
-
由 Eric Erhardt 提交于
* Respect IsDynamicCodeSupported in more places in Linq.Expressions * CanEmitObjectArrayDelegate * CanCreateArbitraryDelegates These properties are all set to `false` when running on NativeAOT, so have them respect RuntimeFeature.IsDynamicCodeSupported. However, CanEmitObjectArrayDelegate needs a work around because DynamicDelegateAugments.CreateObjectArrayDelegate does not exist in CoreClr's System.Private.CoreLib. Allow System.Linq.Expressions to create an object[] delegate using Ref.Emit even though RuntimeFeature.IsDynamicCodeSupported is set to false (ex. using a feature switch). To enable this, add an internal method in CoreLib that temporarily allows the current thread to skip the RuntimeFeature check and allows DynamicMethod instances to be created. When System.Linq.Expressions needs to generate one of these delegates, it calls the internal method through Reflection and continues to use Ref.Emit to generate the delegate. Fix #81803
-
由 Dan Moseley 提交于
-
- 11 7月, 2023 19 次提交
-
-
由 Michal Strehovský 提交于
Fixes #79634. To generate Win32 resources, we'd first use an MSBuild task to dump Win32 resources from the input IL module into a `.res` file. We'd then pass the `.res` file to link.exe. Link.exe would then invoke cvtres.exe to convert the `.res` file to `.obj`. Link.exe then links the `.obj` into the final executable. Skip all of this and instead generate the correct object data in the compiler directly. I'm reusing Win32 resource emission logic from crossgen2. It already mostly does the right thing - we just needed to split the `.rsrc` section into `.rsrc$01` and `.rsrc$02` that link.exe expects. The first part has the data directory. The second part has the actual resource data. Contributes to #73931. The cvtres step was actually non-deterministic because it creates an object file in TEMP and embeds this temporary name into the object.
-
由 Alexander Köplinger 提交于
The Helix workitem timeout still looked for the old `Outerloop` property but we're using `TestScope` instead now.
-
由 Michal Strehovský 提交于
ILLinker allows specifying substitutions on the command line, same as descriptors. This can be useful for various workarounds. Add support for this for ILC as well.
-
由 Ankit Jain 提交于
-
由 Thays Grazia 提交于
-
由 Mike Lorbetske 提交于
* Fixes #82871 Add an additional constructor for StackTrace * Update new signature as prescribed Update ref Add tests * Updated documentation * Use the constructor for List<T> that takes in the set of objects it should contain
-
由 Dan Moseley 提交于
-
由 Zoltan Varga 提交于
* [mono] Precompute the CallInfo structure used by the mono_arch_..native_call_context_.. functions. * Fix indentation. * Fix the arm build.
-
由 Zoltan Varga 提交于
[mono][aot] Avoid prefer-ing gsharedvt versions of AsyncMethodBuilderCore etc. methods in profile-only mode. (#87990) The gsharedvt versions might not be generated if they are not in the profile, leading to crashes at runtime. Fixes https://github.com/dotnet/runtime/issues/87856.
-
由 kkeirstead 提交于
Add Support For Multiplexing `System.Diagnostics.Metrics` for `Dotnet Monitor` and `Dotnet Counters` (#86504)
-
由 Jo Shields 提交于
* Work around for Android builds using GCC for cross compiler native tools build * Allow true cross-compiled cross-compiler during Android repo builds
-
由 Ankit Jain 提交于
Issue: https://github.com/dotnet/runtime/issues/88528
-
由 David Fowler 提交于
- Logging got overhauled to use CompositeFormat in .NET 8, this allows for non boxing string formatting via string.Format. There's one piece of code that boxing the generic argument before handing off to string.Format so remove it for primitive types and enums.
-
由 dotnet-maestro[bot] 提交于
* Update dependencies from https://github.com/dotnet/roslyn-analyzers build 20230618.1 * Suppress RS1038 analyzer * Fix new failures found with CA1859 analyzer updates * Disable CA2261 on test projects, fix more CA1859 findings --------- Co-authored-by: Ndotnet-maestro[bot] <dotnet-maestro[bot]@users.noreply.github.com> Co-authored-by: NBuyaa Namnan <bunamnan@microsoft.com>
-
由 Justin Anderson 提交于
-
由 Radek Doulik 提交于
Implements https://github.com/dotnet/runtime/issues/88092
-
由 Vlad Brezae 提交于
When a valuetype is passed by ref on the stack
-
由 Tomas Weinfurt 提交于
-
由 ashaurtaev 提交于
-
- 10 7月, 2023 4 次提交
-
-
由 Tomas Weinfurt 提交于
-
由 Giridhar Trivedi 提交于
While working on netcore debugger for mono,the crash was happening. This crash was due to an invalid pointer to the free(). Valgrind was showing the traces of mismatch between the new() and delete(). After debugging the issue, even though hidden flag was set on shared library the objdump on the shared library showed that the overloaded new() operator was taken from a local scoped library but overloaded delete() operator was overidden by defaults setting of underlying system libraries in which delete() was hosted. Changed the fvisibility flag to ensure that clang picks up the local new() and delete() overloaded operators. Co-authored-by: NGiridhar Trivedi <giridhar.trivedi@ibm.com>
-
由 Giridhar Trivedi 提交于
* [NCDB]Fix crash caused due to unavailability of symbols. While working on netcore debugger for mono below symbols were missing from the binary. - mdHasSemantic - m_zeros - CBlobPoolHash - StgBlobPool - RecordPool - SplitPath - PEDecoder::CheckCorHeader - GetOsPageSize - PAL_Random These symbols were part of mdruntime-dbi and mdruntimerw-dbi library. These library was properly linked to mscodbi. Though these were linked properly compiler was unableto find these symbols. Further investigations revealed that the order of linking these libraries plays an important role. After modifying the order of linkage the crash is resolved. * Incorporating review comments Removed the commented code --------- Co-authored-by: NGiridhar Trivedi <giridhar.trivedi@ibm.com>
-
由 Mitchell Hwang 提交于
[libs][Unix][perf] Lazily initialize TimeZoneInfo names and order GetSystemTimeZones by Ids (#88368) * [libs] Remove unnecessary assignment TZifHead * [libs][perf] Add lazy initialization for TimeZoneInfo names * Reduce comparisons for UTC alias and remove static array allocation * [libs] Lazy init display names for utc aliases * [libs][perf] Order system time zones by id * Directly compare numerical value * Fix internal field naming * Remove TryPopulateTimeZoneDisplayNamesFromGlobalizationData * Make lazy initialization methods static on windows * Revert "[libs][perf] Order system time zones by id" This reverts commit 580a765775693a2e8f5de81a46591c8a94890088. * Fix lazy initialization for Minimal Globalization Data * Avoid lazy initialization where internal display name fields are set to null * Fix CreateLocal not preserving lazy initialized names * Prevent unintended lazy initialization in CreateCustomTimeZone * Make UICulture a property * Substitute null name properties with empty string in default constructor * Assert not reached in Invariant mode
-