CUDA tests are broken on .NET 7.0.400

[jgiannuzzi]

The CUDA tests for ILGPU are broken when building with the .NET 7.0.400 SDK, no matter which runtime is targeted. Here is a failure example, which gives the following stack trace:

[xUnit.net 00:00:16.97]     ILGPU.Tests.Cuda.CudaArrayViews_Release.ArrayViewValid(length: 8197) [FAIL]
  Failed ILGPU.Tests.Cuda.CudaArrayViews_Release.ArrayViewValid(length: 8197) [12 ms]
  Error Message:
   ILGPU.InternalCompilerException : An internal compiler error has been detected
---- System.AggregateException : One or more errors occurred. (The function 'Add' does not have an intrinsic implementation for this backend. 'EnableAlgorithms' from the Algorithms library not invoked?)
-------- ILGPU.Backends.NotSupportedIntrinsicException : The function 'Add' does not have an intrinsic implementation for this backend. 'EnableAlgorithms' from the Algorithms library not invoked?
  Stack Trace:
     at ILGPU.Backends.Backend.Compile[TBackendHook](Method kernelMethod, EntryPointDescription& entry, KernelSpecialization& specialization, TBackendHook backendHook) in /_/Src/ILGPU/Backends/Backend.cs:line 725
   at ILGPU.Backends.Backend.Compile[TBackendHook](EntryPointDescription& entry, KernelSpecialization& specialization, TBackendHook backendHook) in /_/Src/ILGPU/Backends/Backend.cs:line 643
   at ILGPU.Backends.Backend.Compile(EntryPointDescription& entry, KernelSpecialization& specialization) in /_/Src/ILGPU/Backends/Backend.cs:line 625
   at ILGPU.Tests.TestBase.Execute[TIndex](MethodInfo kernel, TIndex dimension, Object[] arguments) in /home/runner/runner/_work/ILGPU/ILGPU/Src/ILGPU.Tests/Generic/TestBase.cs:line 150
   at ILGPU.Tests.TestBase.Execute[TIndex](TIndex dimension, Object[] arguments) in /home/runner/runner/_work/ILGPU/ILGPU/Src/ILGPU.Tests/Generic/TestBase.cs:line 189
   at ILGPU.Tests.TestBase.Execute(Int32 dimension, Object[] arguments) in /home/runner/runner/_work/ILGPU/ILGPU/Src/ILGPU.Tests/Generic/TestBase.cs:line 169
   at ILGPU.Tests.ArrayViews.ArrayViewValid(Int32 length) in /home/runner/runner/_work/ILGPU/ILGPU/Src/ILGPU.Tests/ArrayViews.cs:line 46
   at System.RuntimeMethodHandle.InvokeMethod(Object target, Void** arguments, Signature sig, Boolean isConstructor)
   at System.Reflection.MethodInvoker.Invoke(Object obj, IntPtr* args, BindingFlags invokeAttr)
----- Inner Stack Trace -----
   at System.Threading.Tasks.TaskReplicator.Run[TState](ReplicatableUserAction`1 action, ParallelOptions options, Boolean stopOnFirstFailure)
   at System.Threading.Tasks.Parallel.ForWorker[TLocal](Int32 fromInclusive, Int32 toExclusive, ParallelOptions parallelOptions, Action`1 body, Action`2 bodyWithState, Func`4 bodyWithLocal, Func`1 localInit, Action`1 localFinally)
--- End of stack trace from previous location ---
   at System.Threading.Tasks.Parallel.ForWorker[TLocal](Int32 fromInclusive, Int32 toExclusive, ParallelOptions parallelOptions, Action`1 body, Action`2 bodyWithState, Func`4 bodyWithLocal, Func`1 localInit, Action`1 localFinally)
   at System.Threading.Tasks.Parallel.For(Int32 fromInclusive, Int32 toExclusive, Action`1 body)
   at ILGPU.Backends.CodeGeneratorBackend`4.Compile(EntryPoint entryPoint, BackendContext& backendContext, KernelSpecialization& specialization) in /_/Src/ILGPU/Backends/CodeGeneratorBackend.cs:line 104
   at ILGPU.Backends.Backend.Compile[TBackendHook](Method kernelMethod, EntryPointDescription& entry, KernelSpecialization& specialization, TBackendHook backendHook) in /_/Src/ILGPU/Backends/Backend.cs:line 706
----- Inner Stack Trace -----
   at ILGPU.Backends.PTX.PTXInstructions.GetArithmeticOperation(BinaryArithmeticKind kind, ArithmeticBasicValueType type, CudaCapabilityContext capabilities, Boolean fastMath) in /_/Src/ILGPU/Backends/PTX/PTXInstructions.cs:line 148
   at ILGPU.Backends.PTX.PTXCodeGenerator.GenerateCode(BinaryArithmeticValue value) in /_/Src/ILGPU/Backends/PTX/PTXCodeGenerator.Values.cs:line 141
   at ILGPU.Backends.BackendCodeGenerator.BackendValueVisitor`1.Visit(BinaryArithmeticValue value) in /_/Src/ILGPU/Backends/IBackendCodeGenerator.cs:line 393
   at ILGPU.IR.Values.BinaryArithmeticValue.Accept[T](T visitor) in /_/Src/ILGPU/IR/Values/Arithmetic.cs:line 325
   at ILGPU.Backends.BackendCodeGenerator.GenerateCodeFor[TCodeGenerator](TCodeGenerator codeGenerator, Value value) in /_/Src/ILGPU/Backends/IBackendCodeGenerator.cs:line 633
   at ILGPU.Backends.PTX.PTXCodeGenerator.GenerateCodeInternal(Int32 registerOffset) in /_/Src/ILGPU/Backends/PTX/PTXCodeGenerator.cs:line 552
   at ILGPU.Backends.PTX.PTXKernelFunctionGenerator.GenerateCode() in /_/Src/ILGPU/Backends/PTX/PTXKernelFunctionGenerator.cs:line 178
   at ILGPU.Backends.CodeGeneratorBackend`4.<>c__DisplayClass1_0.<Compile>b__0(Int32 i) in /_/Src/ILGPU/Backends/CodeGeneratorBackend.cs:line 104
   at System.Threading.Tasks.Parallel.<>c__DisplayClass19_0`1.<ForWorker>b__1(RangeWorker& currentWorker, Int32 timeout, Boolean& replicationDelegateYieldedBeforeCompletion)
--- End of stack trace from previous location ---
   at System.Threading.Tasks.Parallel.<>c__DisplayClass19_0`1.<ForWorker>b__1(RangeWorker& currentWorker, Int32 timeout, Boolean& replicationDelegateYieldedBeforeCompletion)
   at System.Threading.Tasks.TaskReplicator.Replica.Execute()
  Standard Output Messages:
 Compiling 'ArrayViewValidKernel'

The CUDA tests were working fine under .NET 7.0.306. Here is a successful run from last week.

[m4rs-mt]

Thanks a lot for tracking this separately from the discussion we had in PR #1038. I can confirm that this issue seems to be related to a change in .Net SDK we will address shortly.