[llvm] [NVPTX] Add NVPTX intrinsics for TMA copies (PR #95289)

[Adam Paszke via llvm-commits]

================
@@ -1448,6 +1448,26 @@ defm int_nvvm_cp_async_ca_shared_global_8 : CP_ASYNC_SHARED_GLOBAL<"8", "ca">;
 defm int_nvvm_cp_async_ca_shared_global_16 : CP_ASYNC_SHARED_GLOBAL<"16", "ca">;
 defm int_nvvm_cp_async_cg_shared_global_16 : CP_ASYNC_SHARED_GLOBAL<"16", "cg">;
 
+// TODO(apaszke): Multicast TMA loads
----------------
apaszke wrote:

If you have a more complete version then early next week sounds great for me

https://github.com/llvm/llvm-project/pull/95289