Added diagnostics & built-in type lowering for `[CUDAKernel]` functions (#4042)

* Added diagnostics & built-in type lowering for `[CUDAKernel]` functions

This PR adds
- Diagnostics for non-void return from a cuda kernel entry point
- Diagnostics for using differentiable types in a differentiable cuda kernel entry point
- Logic for converting built-in types (float3, float3x3, etc..) to portable struct types and unpacks the parameter back into a built-in type on the CUDA side. This is because built-in types have different implementations in CUDA & CPP targets, which causes signature mis-match when linking.

* Fix error codes

* Add ability to lower structs and arrays that contain built-in types.

+ Added tests
+ Fix issue where the host-side was not marshalling data to lowered types.

* Update slang-ir-pytorch-cpp-binding.cpp

---------

Co-authored-by: Yong He <yonghe@outlook.com>

1 files changed, 36 insertions, 0 deletions

diff --git a/source/slang/slang-check-decl.cpp b/source/slang/slang-check-decl.cpp
index 0f9da12c4..921bd38e9 100644
--- a/source/slang/slang-check-decl.cpp
+++ b/source/slang/slang-check-decl.cpp
@@ -7700,6 +7700,16 @@ namespace Slang
                 }
             }
         }
+
+        // If this method is intended to be a CUDA kernel, verify that the return type is void.
+        if (decl->findModifier<CudaKernelAttribute>())
+        {
+            if (decl->returnType.type && !decl->returnType.type->equals(m_astBuilder->getVoidType()))
+            {
+                getSink()->diagnose(decl, Diagnostics::cudaKernelMustReturnVoid);
+            }
+        }
+
         checkVisibility(decl);
     }
 
@@ -9547,6 +9557,30 @@ namespace Slang
         checkDerivativeAttributeImpl(visitor, funcDecl, attr, imaginaryArguments.args, imaginaryArguments.directions);
     }
 
+    static void checkCudaKernelAttribute(SemanticsVisitor* visitor, FunctionDeclBase* funcDecl, CudaKernelAttribute*)
+    {
+        // If the method is also marked differentiable, check that the data types are either non-differentiable
+        // or marked with no_diff.
+        //
+        // Note: This is a temporary restriction until we have a more complete story for differentiability.
+        //
+        if (funcDecl->findModifier<DifferentiableAttribute>())
+        {
+            for (auto paramDecl : funcDecl->getParameters())
+            {
+                auto paramType = paramDecl->type;
+                
+                if (visitor->isTypeDifferentiable(paramType))
+                {
+                    if (!paramDecl->hasModifier<NoDiffModifier>())
+                    {
+                        visitor->getSink()->diagnose(paramDecl, Diagnostics::differentiableKernelEntryPointCannotHaveDifferentiableParams);
+                    }
+                }
+            }
+        }
+    }
+
     template<typename TDerivativeAttr, typename TDerivativeOfAttr>
     bool tryCheckDerivativeOfAttributeImpl(
         SemanticsVisitor* visitor,
@@ -9747,6 +9781,8 @@ namespace Slang
                 checkDerivativeAttribute(this, decl, bwdDerivativeAttr);
             else if (auto primalAttr = as<PrimalSubstituteAttribute>(attr))
                 checkDerivativeAttribute(this, decl, primalAttr);
+            else if (auto cudaKernelAttr = as<CudaKernelAttribute>(attr))
+                checkCudaKernelAttribute(this, decl, cudaKernelAttr);
         }
     }