[llvm] [NVPTX] Annotate CUDA kernel pointer arguments with .ptr .space .align attributes. (PR #79646)

[Artem Belevich via llvm-commits]

================
@@ -1589,11 +1589,17 @@ void NVPTXAsmPrinter::emitFunctionParamList(const Function *F, raw_ostream &O) {
       if (isKernelFunc) {
         if (PTy) {
           // Special handling for pointer arguments to kernel
+          // CUDA kernels assume that pointers are in global address space
+          // See:
+          // https://docs.nvidia.com/cuda/parallel-thread-execution/index.html#parameter-state-space
           O << "\t.param .u" << PTySizeInBits << " ";
 
-          if (static_cast<NVPTXTargetMachine &>(TM).getDrvInterface() !=
+          int addrSpace = PTy->getAddressSpace();
+          if (static_cast<NVPTXTargetMachine &>(TM).getDrvInterface() ==
               NVPTX::CUDA) {
-            int addrSpace = PTy->getAddressSpace();
+            assert(addrSpace == 0 && "Invalid address space");
----------------
Artem-B wrote:

I'd move the comment down into the CUDA branch as it's only relevant there.

Also, considering that we're comparing with *generic* AS, while the comment talks about `global` (we do have  `ADDRESS_SPACE_GENERIC` constant for that, too), I'd rephrase it along the lines of "pointers are expected to be passed in generic AS, but are assumed to be pointing to objects in global AS."

https://github.com/llvm/llvm-project/pull/79646