- 06 10月, 2021 2 次提交
-
-
由 SingleAccretion 提交于
* Refactor LONG_MUL recognition Move it to a GenTree method so that it can be reused in lowering. No diffs except for a couple (~5 across all SPMI collection) on ARM due to the new operand swapping behavior. * Fix the definition for MUL_LONG Move it out of the "#if !defined(TARGET_64BIT)" block. Also some miscellaneous moving of comments around. * Support MUL_LONG for ARM64 in codegen and LSRA Note: this commit breaks magic division. * Fix magic division To use the new GT_MUL_LONG support. Plus some tweaks to make it conform to the coding guide. Credit to @pentp for the substantive changes. * Recognize MUL_LONG in lowering for ARM64 * Fix #ifdef's
-
由 Eirik Tsarpalis 提交于
-
- 05 10月, 2021 19 次提交
-
-
由 Alexander Köplinger 提交于
This allows consuming the dotnet host that was just built instead of relying on a downloaded one. We need to move the `host` subset build before the `libs` subset so that the artifacts are available before externals.csproj runs.
-
由 kronic 提交于
* - * - * remove goto * use nameof * feedback * - * - * review feedback * review feedback * fix scope Co-authored-by: Nkronic <kronic@softland.ru>
-
由 Jostein Kjønigsen 提交于
* Enable Crossgen for FreeBSD, now that it is actually be implemented. Support implemented in this commit: https://github.com/dotnet/runtime/commit/3c63559029276fea97633ea6115bcb9acb2cffe2 * FreeBSD: Don't enable crossgen when also doing a crossbuild. It doesn't work, for now at least. So let's leave it be.
-
由 Aleksey Kliger (λgeek) 提交于
For UnmanagedCallersOnly methods that return non-void, pass the address of the return variable to the interp entry method.
-
由 Newell Clark 提交于
* Seal classes in WorkloadBuildTasks * Seal classes in AotCompilerTask * Seal classes in WasmBuildTasks and Crossgen2Tasks * Seal types in AppleAppBuilder
-
由 Jeff Handley 提交于
-
由 Huo Yaoyuan 提交于
* Cleanup StringToDouble. * Remove members used by StringToDouble.
-
由 kronic 提交于
* Refactoring * review feedback * review feedback * review feedback Co-authored-by: Nkronic <kronic@softland.ru>
-
由 Pavel Savara 提交于
just lint
-
由 Adeel Mujahid 提交于
-
由 Pent Ploompuu 提交于
* Optimize division by constant after AND or RSZ * Address PR feedback
-
由 Ethan Crawford 提交于
DependencyContextWriter.WritePortableTarget(): write the duplicate key name when duplicate keys are detected (#57860) * string resource * Respond to PR feedback Co-authored-by: NEthan Crawford <ethancr@microsoft.com> Co-authored-by: NEric Erhardt <eric.erhardt@microsoft.com>
-
由 kronic 提交于
* Refactoring XmlWriterSettings * revert part * review feedback * review feedback Co-authored-by: Nkronic <kronic@softland.ru>
-
由 Jeremy Koritzinsky 提交于
-
由 Egor Bogatov 提交于
Co-authored-by: NTanner Gooding <tagoo@outlook.com>
-
由 Egor Bogatov 提交于
-
由 Vitek Karas 提交于
Show dialog when running .NET 6 GUI single-file on 5.0 hostfxr
-
由 Viktor Hofer 提交于
Fixes https://github.com/dotnet/runtime/issues/54964 net6.0 assets which are part of the Microsoft.NETCore.App shared framework and ship inside a package need to keep the SkipLocalsInit optimization to avoid regressing perf. This would happen when referencing one of the 7.0 packages but consuming either the net6.0 asset in it. The higher assembly version of the net6.0 asset would win over the same asset in the shared framework and without this fix, the SkipLocalsInit optimization would be lost.
-
由 Ilia 提交于
Fixes #59917
-
- 04 10月, 2021 8 次提交
-
-
由 SingleAccretion 提交于
* Stop retyping RHSes to byrefs on type mismatch This is an interesting piece of code. Using it, it is possible to give TYP_BYREF to any node that would ordinarily have TYP_I_IMPL. In particular, I encountered it in my GT_CAST work: we have tests in the SPMI collections with CAST(byref <- long <- int) in them. Ordinarily such a tree would assert in VN, but in that case optimizations were off and so everything went smoothly. It is possible to have TYP_BYREF UMOD/MUL_OVF_UN and similar with this code. Naturally, even though this is legal IL, the compiler does not like it and fails with asserts in multiple places. So, just delete the retyping. The cited assert either no longer exists (though I had a look at the legacy codegen and could not find it there...) or got reworked at some point. There are GC Info diffs with this change. They consist of cases when a constant zero node was retyped to TYP_BYREF and thus its register's liveness was recorded. There are few code size diffs with this change. It does pessimize VN a little because we will now have more cases where a cast VN is added for an assignment, but VN is already very pessimistic for byrefs, so it seems acceptable. * Add a test covering the discovered problem VN-based copy propagation checks the compatibility of types using the type of node at use and the RHS of the new local's def. Thus, when we retype a TYP_I_IMPL RHS to TYP_BYREF, we risk substituting a TYP_BYREF local with a TYP_I_IMPL one, dangerously shortening the lifetime of the managed pointer.
-
由 Pavel Savara 提交于
* lint auto fix
-
由 Thays Grazia 提交于
-
由 Pavel Savara 提交于
* - only one emscripten library file library-dotnet.js - flattened typescript folders - binding_support.js split into multiple TS files - clarified exports - binding js methods to linker is generated during emcc module loading - types: Co-authored-by: NDaniel Genkin <genkind@gmail.com> * feedback from @kg
-
由 dotnet-maestro[bot] 提交于
[main] Update dependencies from dnceng/internal/dotnet-optimization dotnet/xharness dotnet/roslyn-analyzers dotnet/msquic (#59905) Co-authored-by: Ndotnet-maestro[bot] <dotnet-maestro[bot]@users.noreply.github.com> Co-authored-by: NPremek Vysoky <premek.vysoky@microsoft.com>
-
由 Jan Vorlicek 提交于
When I've moved the heap metadata out of the actual code heaps some time ago, I've forgotten to account for the personality routine slot allocated at the beginning of the heaps. This was exposed by an assert when executing under the JIT stress mode 2. This change fixes it by adding accounting for those. Close #59794
-
由 Sarah Oslund 提交于
Co-authored-by: NWilliam Li <wul@microsoft.com>
-
由 Michal Strehovský 提交于
Fixes #53242. Saddened this didn't make .NET 6.0. Stack traces in trimmed apps look ugly without this fix: ``` Unhandled exception. System.ArgumentNullException: Value cannot be null. (Parameter 'id') at System.TimeZoneInfo.ValidateTimeZoneInfo(String , TimeSpan , AdjustmentRule[] , Boolean& ) at System.TimeZoneInfo..ctor(String , TimeSpan , String , String , String , AdjustmentRule[] , Boolean , Boolean ) at System.TimeZoneInfo.CreateCustomTimeZone(String , TimeSpan , String , String ) at Program.<Main>$(String[] args) in C:\stacktrace\Program.cs:line 3 ```
-
- 03 10月, 2021 2 次提交
-
-
由 Stephen Toub 提交于
-
由 Stephen Toub 提交于
* Clean up code emitted for single-character loops We can emit a simple while loop rather than a while true loop with gotos and (now unnecessary) labels. * Clean up inverted loop in FindFirstChar * Clean up minimum length check to be more idiomatic * Clean up MatchCharacterClass condition to be more idiomatic * Address PR feedback
-
- 02 10月, 2021 9 次提交
-
-
https://github.com/dotnet/linker由 dotnet-maestro[bot] 提交于
Microsoft.NET.ILLink.Tasks From Version 7.0.100-1.21479.1 -> To Version 7.0.100-1.21501.1 Co-authored-by: Ndotnet-maestro[bot] <dotnet-maestro[bot]@users.noreply.github.com>
-
由 Jakob Botsch Nielsen 提交于
This is a better fix for #58373 that changes the handling of this to happen in morph for all cases. We sometimes missed the insertion of necessary casts because we forgot to remove a GTF_DONT_CSE flag when folding an indirection. Fixing this leads to some new GT_CNS_DBL cases in lowering that hit an assert, but those cases should be correctly handled by the default case so just remove the assert. To get rid of some of the regressions I have allowed generating assertions when assigning struct fields from casts. It was unclear why this was not allowed in the first place.
-
由 Egor Bogatov 提交于
-
由 Egor Bogatov 提交于
-
* Move runtime rolling builds to run twice a day on main * Update eng/pipelines/runtime.yml
-
由 Michal Strehovský 提交于
The move was pretty straightforward and I didn't really need to touch anything. The only changes I made: * crossgen2command.txt is now command.txt * crossgen2repro.rsp is now repro.rsp * Instead of "CrossGen2 version" the TXT file now says "Compiler version". What compiler will be obvious from the file name and version strings.
-
由 Jeremy Koritzinsky 提交于
* Make libraries tests use the same config as the matching libraries build. This change is useful in two ways: 1. It simplifies our CI build dependency graph so that we don't have a windowx x86 job depending on a windows x64 job to finish first. 2. It unblocks #59579 which introduces an arch-dependent native component into the libraries tests with DNNE. * Mark one of the win x86 runs as release. * Remove testBuildPlatforms. * Build debug Linux_arm/Linux_musl_arm/Linux_musl_arm64 libraries builds on PR. * Update to only new libraries Debug builds to run on a coreclr change in PR. * Upload test artifacts with os subgroup name * Use a "preprocessor if" instead of a condition for isFullMatrix. * Fix conditions based on feedback. * Fix incorrect condition. * Don't run tests locally when we're running tests using Helix.
-
由 Adeel Mujahid 提交于
* Use ProductVersion in how-to-debug-dump
-
由 Jeremy Koritzinsky 提交于
-