NativeAOT-LLVM: Merge runtime to 17 May 2023 #2275

yowl · 2023-05-16T15:51:35Z

This PR is the latest merge (to 83f71b5) from runtime.

Set RUNTIME_FLAVOR=CoreCLR for build-native.cmd so that STATIC_ICU does not get set. We need all the function renaming macros such as

#define u_charsToUChars(...) u_charsToUChars_ptr(__VA_ARGS__)

from pal_icushim_internal.h to avoid missing symbols.

Delete the assert from the beginning of SsaBuilder::AddPhiArg

    // For LLVM backed, every predecessor must have its own phi arg
    if (!block->IsLIR())
    {
        // Make sure it isn't already present: we should only add each definition once.
        for (GenTreePhi::Use& use : phi->Uses())
        {
            assert(use.GetNode()->AsPhiArg()->GetSsaNum() != ssaNum);
        }
    }

Not sure that is ok, maybe it is fallout from the ASG refactoring that needs looking at?

Exclude LowLevelLifoSemaphore for Wasi, think its just for threading and we don't have the base LowLevelLifoSemaphoreBase

…uests error. (#85881)

Fixes #72570. Still need to delete workarounds that make things public in corelib but maybe this diff is large enough already? We were gradually getting less and less from reflection blocking: * We stopped blocking things outside corelib (.NET Native blocked all of BCL; we don't). * We trim reflection metadata and that allows us to have method bodies without metadata. With this, we should be able to get type definition-level reflection metadata for any MethodTable there is in the system. This is a ~30 kB size regression for BasicMinimalApis. It is pretty much a wash for Hello World, because all the BCL cruft to handle reflection blocked types costs as much as the benefit of blocking.

…e (#85153) * [mono][interp] Refactor arg offset computation to make code clearer * [mono][interp] Remove redundant assign * [mono][interp] Move mint_type out from header file It makes no sense to inline this big method into callers. * [mono][interp] Fix arg offset computation when doing pinvoke and aot/jit calls from interp The previous code wasn't handling alignment of simd types. * [mono][interp] Use the correct alignment for args when entering interp We stop computing the arguments on the fly since it's getting too expensive doing the alignment checks. We directly use `imethod->arg_offsets` instead which is computed once during the first call, while also ensuring consistency with the calculation. * [mono][interp] Fix arg offset computation when running clauses with interp * [mono][interp] Remove dead code * Temporaril disable jiterp for CI * Update jiterpreter interp_entry wrapper generator and re-enable jiterp * Revert options-def.h changes --------- Co-authored-by: Katelyn Gadd <[email protected]>

When the profile data comes from dynamic PGO, the spanning tree encoded in the schema produced by an earlier tier should exactly match the spanning tree for the current jit attempt, since the JIT and method IL are identical. (This is not the case for static PGO; that schema may have come from a different JIT and/or different version of IL). Note in release modes we won't assert; instead, we will silently throw the PGO data away. Follow-on change to #85805 to catch more issues like #85799.

…al bailout/exit (#85905) * Prune opcodes that follow an unconditional bailout/abort since they are unreachable and we don't need to compile them

…rating a table (#85904) * No longer generate a typescript opcode info table * Fetch opcode info from the interpreter on-demand

Microsoft.DotNet.Arcade.Sdk , Microsoft.DotNet.Build.Tasks.Archives , Microsoft.DotNet.Build.Tasks.Feed , Microsoft.DotNet.Build.Tasks.Installers , Microsoft.DotNet.Build.Tasks.Packaging , Microsoft.DotNet.Build.Tasks.TargetFramework , Microsoft.DotNet.Build.Tasks.Templating , Microsoft.DotNet.Build.Tasks.Workloads , Microsoft.DotNet.CodeAnalysis , Microsoft.DotNet.GenAPI , Microsoft.DotNet.GenFacades , Microsoft.DotNet.Helix.Sdk , Microsoft.DotNet.PackageTesting , Microsoft.DotNet.RemoteExecutor , Microsoft.DotNet.SharedFramework.Sdk , Microsoft.DotNet.VersionTools.Tasks , Microsoft.DotNet.XUnitConsoleRunner , Microsoft.DotNet.XUnitExtensions From Version 8.0.0-beta.23255.1 -> To Version 8.0.0-beta.23255.2 Co-authored-by: dotnet-maestro[bot] <dotnet-maestro[bot]@users.noreply.github.com>

…uild 20230506.1 (#85885) Microsoft.CodeAnalysis.Analyzers , Microsoft.CodeAnalysis.NetAnalyzers From Version 3.3.5-beta1.23254.2 -> To Version 3.3.5-beta1.23256.1 Co-authored-by: dotnet-maestro[bot] <dotnet-maestro[bot]@users.noreply.github.com>

The pre-existing behavior was copying all RVA fields in single-file builds and only selectively copying explicitly referenced fields in composite builds. This has bitten us in WPF where Managed C++ emits an array of early initializers that are marked by an initial and final RVA field. By only selectively copying those two without the remaining entries in-between we end up crashing later on as the app is not fully initialized. Based on David Wrighton's advice I have kept the logic for copying RVA fields in case of array initializers that seem to be hot in terms of performance. For the remaining field access I have ended up constructing a new precode fixup RVA_FIELD_ADDRESS for this purpose. Fixes: #78723 Thanks Tomas Co-authored-by: Jan Kotas <[email protected]>

…tors (#84867)

BasicMinimalApi spends about 1% of samples in this method.

* Ensure the insTupleType mask entries stay correct * Ensure the bitwise result remains uint16_t

…install installing a broken urllib version. (#85935)

* Servicing docs improvements * Apply suggestions from code review * Apply suggestions from code review * Apply suggestions from code review Co-authored-by: Carlos Sánchez López <[email protected]> * Apply suggestions from code review * Update .github/workflows/check-service-labels.yml * Update .github/workflows/check-service-labels.yml --------- Co-authored-by: Carlos Sánchez López <[email protected]> Co-authored-by: Juan Hoyos <[email protected]>

I think this is a leftover from one of the appmodels that we deleted. I don't believe this ever returns non-empty enumeration.

…#85866) * Update dependencies from https://dev.azure.com/dnceng/internal/_git/dotnet-optimization build 20230505.3 optimization.linux-arm64.MIBC.Runtime , optimization.linux-x64.MIBC.Runtime , optimization.windows_nt-arm64.MIBC.Runtime , optimization.windows_nt-x64.MIBC.Runtime , optimization.windows_nt-x86.MIBC.Runtime , optimization.PGO.CoreCLR From Version 1.0.0-prerelease.23254.3 -> To Version 1.0.0-prerelease.23255.3 * Update dependencies from https://dev.azure.com/dnceng/internal/_git/dotnet-optimization build 20230506.3 optimization.linux-arm64.MIBC.Runtime , optimization.linux-x64.MIBC.Runtime , optimization.windows_nt-arm64.MIBC.Runtime , optimization.windows_nt-x64.MIBC.Runtime , optimization.windows_nt-x86.MIBC.Runtime , optimization.PGO.CoreCLR From Version 1.0.0-prerelease.23254.3 -> To Version 1.0.0-prerelease.23256.3 --------- Co-authored-by: dotnet-maestro[bot] <dotnet-maestro[bot]@users.noreply.github.com>

…t (#85880)

Instead of going to great lengths to keep the ref counts correct in physical promotion just stop using them in forward sub. The last use information generalizes what forward sub needs except for when the tracking limit is reached, which shows in the improvements we actually get from consulting the ref counts -- the improvements are very modest and in tests only. This makes early liveness the only consumer of the "regular" ref counts computed by local morph. Weighted ref counts are still used for undoing regular promotion of implicit byref parameters.

Fix a couple of cases where it's possible that the decomposed stores will modify the address being used as part of the operation too early. In those cases we must spill the address to a new local ahead of time. Fix #85874

…ev (#85815)

This test has failed across many configurations, but apparently only on Windows. Tracked by dotnet/runtime#84006, dotnet/runtime#83658.

* Do not generate PerfMap entries for methods we didn't compile dotnet/installer#16318 is currently blocked due to crossgen2 crash. While Jan (dotnet/installer#16318 (comment)) is totally right that the command line to crossgen2 is garbage, we should try not to crash. Based on the stack we already emitted the PE and PDB. The crash seems to be caused by attempting to include methods that didn't successfully compile in the perfmap (`EnumerateCompilerMethods` has an `IsEmpty` check whereas walking all nodes looking for `MethodWithGCInfo` doesn't check that). ``` Internal.TypeSystem.TypeSystemException+TypeLoadException: Failed to load type 'System.Security.Policy.EvidenceBase' from assembly 'System.Security.AccessControl, Version=6.0.0.0, Culture=neutral, PublicKeyToken=b03f5f7f11d50a3a' [/__w/1/s/src/redist/redist.csproj] at Internal.TypeSystem.ThrowHelper.ThrowTypeLoadException(ExceptionStringID, String, String) + 0x39 [/__w/1/s/src/redist/redist.csproj] at Internal.TypeSystem.ResolutionFailure.Throw() + 0x99 [/__w/1/s/src/redist/redist.csproj] at Internal.TypeSystem.Ecma.EcmaModule.GetObject(EntityHandle, NotFoundBehavior) + 0x8f [/__w/1/s/src/redist/redist.csproj] at Internal.TypeSystem.Ecma.EcmaModule.GetType(EntityHandle) + 0x2c [/__w/1/s/src/redist/redist.csproj] at Internal.TypeSystem.Ecma.EcmaType.InitializeBaseType() + 0x81 [/__w/1/s/src/redist/redist.csproj] at Internal.TypeSystem.Ecma.EcmaType.ComputeTypeFlags(TypeFlags) + 0x48 [/__w/1/s/src/redist/redist.csproj] at Internal.TypeSystem.TypeDesc.InitializeTypeFlags(TypeFlags) + 0x1e [/__w/1/s/src/redist/redist.csproj] at Internal.TypeSystem.TypeNameFormatter.AppendName(StringBuilder, TypeDesc) + 0x2d [/__w/1/s/src/redist/redist.csproj] at ILCompiler.PEWriter.TypeString.AppendName(StringBuilder, ArrayType) + 0x1b [/__w/1/s/src/redist/redist.csproj] at Internal.TypeSystem.TypeNameFormatter.FormatName(TypeDesc) + 0x54 [/__w/1/s/src/redist/redist.csproj] at ILCompiler.PEWriter.OutputInfoBuilder.FormatMethodName(MethodDesc, TypeNameFormatter) + 0x174 [/__w/1/s/src/redist/redist.csproj] at ILCompiler.PEWriter.OutputInfoBuilder.<EnumerateMethods>d__16.MoveNext() + 0x15f [/__w/1/s/src/redist/redist.csproj] at ILCompiler.Diagnostics.PerfMapWriter.Write(String, Int32, IEnumerable`1, IEnumerable`1, TargetDetails) + 0x13b [/__w/1/s/src/redist/redist.csproj] at ILCompiler.PEWriter.SymbolFileBuilder.SavePerfMap(String, Int32, String) + 0x1ba [/__w/1/s/src/redist/redist.csproj] at ILCompiler.DependencyAnalysis.ReadyToRunObjectWriter.EmitPortableExecutable() + 0xa59 [/__w/1/s/src/redist/redist.csproj] ``` This is all based on the stack trace. I didn't validate this fixes the issue (or that it compiles at all) but the fix seems legit to me. My local workflows are affected by the installer repo break because the runtime repo had a breaking change in CoreLib that is not compatible with existing installer builds. Nobody seems to be doing anything with the failing integration PR, so this is my attempt...

…ce (#85938) * Allow Base64Decoder to ignore space chars, add IsValid methods and tests * Some cleanup of Base64.IsValid changes This includes making FromBase64Transform significantly faster via SearchValues. * Address PR feedback and some more cleanup --------- Co-authored-by: Heath Baron-Morgan <[email protected]>

Bypass emscripten's generated js wrappers when invoking cwraps, this may improve cross-language inlining Fix signature for a cwrap that was missing an arg

* [JIT] Fix some errors in JIT * Fix errors in unwinding * [JIT] Pass float value with int arg reg * [JIT] Fix register overwriting in RSVD * [JIT] jit-format * Fix an error * [JIT] Add LOONGARCH64

…5959)

Disabled tests that are timing out on node 16.19.0. On the newer version, 19.9.0 it's already fine. Removing timeouts uncovered a bunch of failures that are gone in the newer node as well. Related issues: dotnet/runtime#86317, and dotnet/runtime#86326 * Active issue. * Disable failures uncovered when timeouts are gone.

…s from base methods to generated shadowing methods (#86344) Adds tests to ensure expected diagnostics are produced for interfaces with a Guid and generic interfaces. Copies attributes on methods on base methods to the generated shadowing method definitions. Adds Test type with a CompilationVerifier field of Action to make it easier to make one-off tests that verify compilations without making a new type.

… Build ID 2181920 (#86361)

…ocess. (#85649) * ProcessTests: allow WorkingSet to be zero just after launching the process. The WorkingSet tests fail on Fedora 38+ because a zero working set value is observed just after the process start.

… RISC-V. (#86294) * Fix the regs conflicts within the `addResolution` for LoongArch64/RISC-V. * amend the code for CR. * add terminatorNodeLclVarDsc2 for the second op2.

* Remove NativeAOT test exclusion for issue #83167 Resolves #83167. * Update tests.proj * Update tests.proj

…fied (#86371) - Fix passing startupOptions to resourceLoader creation when loadBootResource is not specified - Fix propagating applicationEnvironment

* Special-case array/list in Dictionary(IEnumerable, ...) ctors It's reasonably common to construct a dictionary from a `KeyValuePair<TKey, TValue>[]` or a `List<KeyValuePair<TKey, TValue>>`, and we can make both cheaper.

…811) * Initial commit to add ToDictionary() overloads for KeyValuePair and ValueTuple kv. * Add xml doc; add tests. * Fix comments: use Dictionary's ctor for IEnumerable<KeyValuePair<>> overload; API signature changed to nullable comparer. * Throw ArgumentNullExp with 'source' parameter name.

Otherwise new stress modes can change what the test is testing. I verified that these stress modes reproduced the original issue back when the test was introduced. Fix #86185

* Haiku: Initial configuration support Initially recognize Haiku as a supported platform in Directory.Build.props and all scripts in eng/**. * Drop some unused configure values Drop some unused configure values from tryrun.cmake, configure.cmake and config.h.in. * Simplify error message in build.sh Simplified an error message for the `--os` option to avoid maintaining a duplicate list of OSes. Co-authored-by: Adeel Mujahid <[email protected]>

…339) Fix DAC syncblk API Only fill in the syncblk data if there are blocks Add SBNumber index validation to API

Remove AddPhiArg assert

…-may-23

…ge-15-may-23 # Conflicts: # src/coreclr/nativeaot/BuildIntegration/Microsoft.NETCore.Native.targets

yowl · 2023-05-18T14:30:20Z

cc @dotnet/nativeaot-llvm

pentp and others added 30 commits May 7, 2023 18:29

Fix JsonNode.ToString allocations (#85886)

b9b5fef

Pin urllib to version 1.26.15 based on github comments to fix ssl req…

5f0d620

…uests error. (#85881)

[wasm] Prune opcodes in the jiterpreter if they follow an uncondition…

f437b2e

…al bailout/exit (#85905) * Prune opcodes that follow an unconditional bailout/abort since they are unreachable and we don't need to compile them

[wasm] Fetch interpreter opcode info from the runtime instead of gene…

de33da0

…rating a table (#85904) * No longer generate a typescript opcode info table * Fetch opcode info from the interpreter on-demand

performance-setup.sh requires urllib3 version <2 (#85915)

b132963

Add test for unnmanaged delegates interop (#85903)

4fb1b76

Use Roslyn Source Generator Testing SDK to test interop source genera…

fc1cb19

…tors (#84867)

Speed up GVMLookupForSlot (#85901)

d0c9457

BasicMinimalApi spends about 1% of samples in this method.

Ensure the insTupleType mask entries stay correct (#85870)

de4378f

* Ensure the insTupleType mask entries stay correct * Ensure the bitwise result remains uint16_t

Previous changes fixed some runs, but some are still broken due to re…

4ea93a6

…install installing a broken urllib version. (#85935)

Delete call to GetMethod in runtime type system (#85902)

231314d

I think this is a leftover from one of the appmodels that we deleted. I don't believe this ever returns non-empty enumeration.

Remove check from FillStringChecked and rename it to CopyStringConten…

7d54c05

…t (#85880)

[wasm] Wasm.Build.Tests - Override KnownWebAssemblySdkPack to 8.0.0-d…

a442971

…ev (#85815)

Disable foreground-shutdown test on Windows (#85936)

eaa7ae1

This test has failed across many configurations, but apparently only on Windows. Tracked by dotnet/runtime#84006, dotnet/runtime#83658.

Attempt to bypass emscripten's generated wrappers for cwraps (#85919)

f576c78

Bypass emscripten's generated js wrappers when invoking cwraps, this may improve cross-language inlining Fix signature for a cwrap that was missing an arg

[RISC-V] Fixed errors in jit and unwinding (#85900)

f1030ed

* [JIT] Fix some errors in JIT * Fix errors in unwinding * [JIT] Pass float value with int arg reg * [JIT] Fix register overwriting in RSVD * [JIT] jit-format * Fix an error * [JIT] Add LOONGARCH64

amend the multiply's overflow check and delete some unused codes. (#8…

4677b0d

…5959)

Fix comment for File.Exists (#85648)

8c96a18

ilonatommy and others added 24 commits May 16, 2023 23:01

Localized file check-in by OneLocBuild Task: Build definition ID 679:…

96990fe

… Build ID 2181920 (#86361)

Enable eh runtime tests on Interpreter (#86307)

aeb56e0

Restructure WellKnownOids.cs to reduce binary size (#84921)

976409d

Fix debug compile error in mono mini (#86329)

07819e9

ProcessTests: allow WorkingSet to be zero just after launching the pr…

27d0009

…ocess. (#85649) * ProcessTests: allow WorkingSet to be zero just after launching the process. The WorkingSet tests fail on Fedora 38+ because a zero working set value is observed just after the process start.

fix the regs conflicts within the addResolution for LoongArch64 and…

2b11093

… RISC-V. (#86294) * Fix the regs conflicts within the `addResolution` for LoongArch64/RISC-V. * amend the code for CR. * add terminatorNodeLclVarDsc2 for the second op2.

Baseline Native AOT tests (#86366)

0403ce9

* Remove NativeAOT test exclusion for issue #83167 Resolves #83167. * Update tests.proj * Update tests.proj

[browser] Fix using startupOptions when loadBootResource is not speci…

8454898

…fied (#86371) - Fix passing startupOptions to resourceLoader creation when loadBootResource is not specified - Fix propagating applicationEnvironment

[RISC-V][JIT] Fix floating-point args passing (#86244)

7c72ba5

Include Native AOT pdbs in Helix payload (#86367)

a776b0a

JIT: Stop specifying general JitStress in OSR/largefuncletframe (#86380)

7692f98

Otherwise new stress modes can change what the test is testing. I verified that these stress modes reproduced the original issue back when the test was introduced. Fix #86185

[LoongArch64] amend the crossgen2/r2rdump for LA64. (#85038)

af1171b

Fix DAC syncblk API so the SOS tests stop failing intermittently (#86…

1a598b7

…339) Fix DAC syncblk API Only fill in the syncblk data if there are blocks Add SBNumber index validation to API

Revert RequiredAttribute.DisallowAllDefaultValues (#86378)

bda1f6c

Improve constant folding for some frozen objects (#86318)

83f71b5

Do not set STATIC_ICU for CoreCLR

c873fd6

Remove AddPhiArg assert

Merge commit '83f71b53d7f08700fd059191859f7931cf5712f4' into merge-15…

9ebc89d

…-may-23

Merge remote-tracking branch 'origin/feature/NativeAOT-LLVM' into mer…

e3ced51

…ge-15-may-23 # Conflicts: # src/coreclr/nativeaot/BuildIntegration/Microsoft.NETCore.Native.targets

Exclude LowLevelLifoSemaphore.Unix.cs for wasi

6b25535

yowl changed the title ~~Merge 15 may 23~~ NativeAOT-LLVM: Merge runtime to 17 May 2023 May 18, 2023

yowl marked this pull request as ready for review May 18, 2023 14:30

jkotas added the area-NativeAOT-LLVM LLVM generation for Native AOT compilation (including Web Assembly) label May 18, 2023

yowl mentioned this pull request May 19, 2023

NativeAOT-LLVM: Merge runtime to 17 May 2023 #2283

Merged

yowl closed this May 19, 2023

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

NativeAOT-LLVM: Merge runtime to 17 May 2023 #2275

NativeAOT-LLVM: Merge runtime to 17 May 2023 #2275

Uh oh!

yowl commented May 16, 2023 •

edited

Loading

Uh oh!

yowl commented May 18, 2023

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

77 participants

NativeAOT-LLVM: Merge runtime to 17 May 2023 #2275

NativeAOT-LLVM: Merge runtime to 17 May 2023 #2275

Uh oh!

Conversation

yowl commented May 16, 2023 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

yowl commented May 18, 2023

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

77 participants

yowl commented May 16, 2023 •

edited

Loading