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>

2 files changed, 35 insertions, 0 deletions

diff --git a/tests/diagnostics/cuda-kernel-differentiable-params.slang b/tests/diagnostics/cuda-kernel-differentiable-params.slang
new file mode 100644
index 000000000..0e7604b3d
--- /dev/null
+++ b/tests/diagnostics/cuda-kernel-differentiable-params.slang
@@ -0,0 +1,18 @@
+//DIAGNOSTIC_TEST:SIMPLE(filecheck=CHECK):
+
+// Simple check to see if the compiler throws an error if a CUDA kernel is declared with non-void return type.
+
+[CudaKernel]
+[Differentiable]
+void myBadKernel(float x, TensorView<float> t1, TensorView<float> t2)
+{
+    // CHECK:      tests/diagnostics/cuda-kernel-differentiable-params.slang([[@LINE-2]]): error 31214: differentiable kernel entry point cannot have differentiable parameters. Consider using DiffTensorView to pass differentiable data, or marking this parameter with 'no_diff'
+    // CHECK-NEXT:     void myBadKernel(float x, TensorView<float> t1, TensorView<float> t2)
+    // CHECK-NEXT:                            ^
+}
+
+[CudaKernel]
+void myGoodKernel(float x, TensorView<float> t1, TensorView<float> t2)
+{
+
+}
\ No newline at end of file
diff --git a/tests/diagnostics/cuda-kernel-non-void-return.slang b/tests/diagnostics/cuda-kernel-non-void-return.slang
new file mode 100644
index 000000000..75c8bc6d4
--- /dev/null
+++ b/tests/diagnostics/cuda-kernel-non-void-return.slang
@@ -0,0 +1,17 @@
+//DIAGNOSTIC_TEST:SIMPLE(filecheck=CHECK):
+
+// Simple check to see if the compiler throws an error if a CUDA kernel is declared with non-void return type.
+
+[CudaKernel]
+float myBadKernel(TensorView<float> t1, TensorView<float> t2)
+{
+    // CHECK:      tests/diagnostics/cuda-kernel-non-void-return.slang([[@LINE-2]]): error 31213: return type of a CUDA kernel function cannot be non-void.
+    // CHECK-NEXT:     float myBadKernel(TensorView<float> t1, TensorView<float> t2)
+    // CHECK-NEXT:           ^~~~~~~~~~~
+}
+
+[CudaKernel]
+void myGoodKernel(TensorView<float> t1, TensorView<float> t2)
+{
+
+}
\ No newline at end of file