- 22 4月, 2021 18 次提交
-
-
由 Elinor Fung 提交于
-
由 Steve Pfister 提交于
Since AOT'ing each test suite takes between 3-9 min, we need to shift the burden over to helix. ## This is done by: 1. building the test assemblies on the build machine - the wasm part of the build is not executed on the build machine, because it has the AOT build part 2. Zip up the test assembly+friends, *and* any bits required to run the wasm app build for that on helix (eg. emsdk, wasm app targets, cross compiler etc) 3. Send all this to helix, and use a custom `aot-build.proj` - which recreates all the build inputs for the `WasmBuildApp` target using the paths for the assets on helix - then we can run `WasmBuildApp` for the build, resulting in a wasm app bundle. 4. Run the tests! - We already have the bits required for building wasm apps on helix, supported for `Wasm.Build.Tests`, which we can use here too. ## Trimming: - Since, AOT can be so expensive, we use `EnableAggressiveTrimming=true`(EAT), but that means that we could have issues due to trimming. - And it can sometimes be unclear whether the build/test failures are due to trimming or AOT. - Because these builds+test runs are different from other builds, owing to the "build partially on helix" step, a normal EAT build would not be the same as - to help with testing this, we add two lanes to `runtime-staging`: - `*_Mono_AOT`: builds AOT+EAT on helix - `*_Mono_EAT`: builds EAT, on helix - this is required because we want to run almost the same kinda build: 1. build test assembly; *2. send to helix; 3. build wasm app;* 4. run tests - This should effectively mean that we can see which errors might be due to EAT, and which are clearly because of EAT+AOT. Co-authored-by: NSteve Pfister <steve.pfister@microsoft.com> Co-authored-by: NMarek Safar <marek.safar@gmail.com> Co-authored-by: NMitchell Hwang <mitchell.hwang@microsoft.com> Co-authored-by: NAnkit Jain <radical@gmail.com>
-
由 Layomi Akinrinade 提交于
* Add Put/Get methods to HttpClientJsonExtensions that take type metadata * Make JsonContentOfT derive directly from HttpContent * Fix linker error * Rename Helper to JsonHelpers
-
由 Ankit Jain 提交于
* [wasm] Use response files for `emcc` invocations This adds response file for the linker command line, which will be the longest one. And it uses a known name for the response file. In future, once we have dependency checking, we can use such response files for that too. Fixes #51437 . * [wasm] Emit emcc arguments, one per line in the rsp file Having the arguments one per line, also means that we don't have to quote the file paths. And it is more readable. * [wasm] Revert to quote file paths even in response file for emcc Suggested by Pranavkm: ``` .. their format does require whitespaces to be quoted: https://github.com/emscripten-core/emscripten/blob/main/tools/response_file.py#L16-L53. ```
-
由 Anton Lapounov 提交于
-
由 Levi Broderick 提交于
-
由 David Pine 提交于
* Added some API comments to help consumers have a better understanding of the order of operations, added unit tests, fixed redundancy in comment * Update test to call UseEnv more than once directly
-
由 Alex-ABWorld 提交于
-
由 Aaron Robinson 提交于
* Simplify some CallConv logic Reuse the CallConvBuilder class in more places Unify CharSet logic for UnmanagedFunctionPointer and DllImport * Add needed cast. * Explicitly handle the 0 case for the CorNativeLinkType enum. * Comment update. * Review feedback. * Update dllimport.cpp Style feedback.
-
由 Andrew Au 提交于
-
由 Aleksey Kliger (λgeek) 提交于
* [mono] Add HAVE_PTHREAD_JIT_WRITE_PROTECT_NP to cmake config.h.in Otherwise mono_codeman_enable_write is a no-op Should fix jiting on Apple ARM64 * [cmake] Use check_symbol_exists(pthread_jit_write_protect_np) check_function_exists doesn't look at the headers, only at linking, so it can't detect functions marked unavailable in Apple headers. Fixes #49042
-
由 Fan Yang 提交于
-
由 James Newton-King 提交于
-
由 Larry Ewing 提交于
-
由 Tanner Gooding 提交于
* Don't have gtAuxiliaryJitType and gtOtherReg share a union * Applying formatting patch
-
由 Stephen Toub 提交于
-
由 Viktor Hofer 提交于
The by the SDK default included netframework reference assembly are used instead of the old microsoft.netframework.targetingpack. Removing the stale property.
-
由 Natalia Kondratyeva 提交于
Fixes #2262
-
- 21 4月, 2021 22 次提交
-
-
由 Zoltan Varga 提交于
-
由 Johan Lorensson 提交于
Lazy init of ThreadWaitInfo in Thread.Mono.cs to prevent crash on external attached threads. (#51483) * Init ThreadWaitInfo to prevent crash on external attached threads. Commit https://github.com/dotnet/runtime/commit/457f58cd30082c9de0be9ca62e64fa179c37dab3 added a new managed object into Mono's managed thread object, _waitInfo. This new object was initialized in Thread managed constructor, problem with that is for all threads that gets their thread object created in native code, create_thread_object, that code did the same thing as constructor before commit, since managed constructor called down into native code calling the same native method as create_thread_object meaning a thread object created from managed or native would be initialized identical. After above commit this is no longer true, meaning that all managed thread objects created in native code (like attached external threads) won't get _waitInfo initialized and that leads to a crash when OnThreadExiting gets called from Finalizer thread. Fix is to make sure _waitInfo get initialize regardless of thread object gets created from managed or native code. This fix implements a lazy init as part of property access, I tested a native version of the fix, calling back into managed to init managed parts of thread when tread gets created in native, but turns out that such a fix gets a little complicated, since some of the threads are created early when it's not possible to run managed code, so that in turn needed additional changes into the init order, that in turn has bigger impact and could cause other unwanted side effects. Fixing it in managed using lazy/on demand init is simpler/cleaner approach but might come with a small performance hit for callers of WaitInfo property (will add volatile read and conditional branch), but looking at callers of the property, they appear in functions with rather high "cost", like Sleep, NewMutex, Wait, SignalAndWait, Interrupt and RelinquishOwnership.
-
由 Levi Broderick 提交于
-
由 Andrew Au 提交于
Nuget package for ILCompiler.Reflection.ReadyToRun Co-authored-by: NMichal Strehovský <MichalStrehovsky@users.noreply.github.com>
-
由 Jan Kotas 提交于
* Simplify non-generic ArrayEnumerator * Implement GetFlattenedIndex for Mono
-
由 Andy Ayers 提交于
Rework the inlinee profile scale computations so that all scaling happens during the profile incorporation phase, rather than sometimes deferring the scaling until inlining. Because of this we no longer need to record the scale on the inline info. Toss out profile data if all counts are zero. Update the edge profile solver to handle a special case where no return block was executed, but edges within the method had counts. In such cases the entry block count can end up zero and blocking proper scaling computations. For this case, try and deduce a plausible count in this case from nearby blocks and edges. Fix the edge weight computations to tolerate inconsistent data rather than to assert.
-
由 Jan Kotas 提交于
Disable the tests for values are not guaranteed to roundtrip depending on the GC settings. We have more extensive test coverage for the latency modes under runtime tests.
-
由 Kevin Jones 提交于
The Apple API offers significant performance improvements (an order of a magnitude) over reading from /dev/urandom.
-
由 Andy Ayers 提交于
Records of this type are created when class profile histograms in dynamic PGO data are summarized by the static PGO tooling. These records can appear in both prejit and jit schemas when the static PGO data is passed back to the jit.
-
由 David Pine 提交于
There appeared to be two issues, where the config.GetReloadToken().RegisterChangeCallback was registered after the updates were made.
-
由 Zoltan Varga 提交于
This forces the llvm backend to allocate a stack location for them, so they get GC tracking.
-
由 Ankit Jain 提交于
* [wasm] Set runtime pack directory to the nuget path, if not set * Update src/mono/wasm/build/WasmApp.targets Co-authored-by: NLarry Ewing <lewing@microsoft.com> Co-authored-by: NLarry Ewing <lewing@microsoft.com>
-
由 Layomi Akinrinade 提交于
[JSON source gen 3/3] Add new methods to JsonSerializer and System.Net.Http.Json APIs that take type metadata (#51528) * Add new methods to JsonSerializer and System.Net.Http.Json APIs that take type metadata * Address review feedback
-
由 Jeff Handley 提交于
-
由 Levi Broderick 提交于
-
由 dotnet-maestro[bot] 提交于
[main] Update dependencies from dotnet/runtime dotnet/arcade dotnet/icu dotnet/xharness dotnet/llvm-project dotnet/runtime-assets (#51488) [main] Update dependencies from dotnet/runtime dotnet/arcade dotnet/icu dotnet/xharness dotnet/llvm-project dotnet/runtime-assets
-
由 Eric StJohn 提交于
* Disable source generators from binplacing * Ensure APICompat doesn't run for non-Source projects
-
由 Konrad Kokosa 提交于
It's 14 hours of lectures about how .NET GC is implemented.
-
由 David Wrighton 提交于
-
由 Jeremy Koritzinsky 提交于
Use adjusted parent layout size to handle a non-zero-sized class inheriting from a zero-sized class that inherits from another zero-sized class. (#50861)
-
由 Stephen Toub 提交于
* Remove pinning in BufferedFileStreamStrategy * Simplify code/synchronization in ValueTaskSource
-
由 Stephen Toub 提交于
* Add metadata update handler to System.ComponentModel.TypeConverter * Update src/libraries/System.ComponentModel.TypeConverter/src/System/ComponentModel/ReflectTypeDescriptionProvider.cs
-