[llvm] [mlir] [MLIR][NVPTX] Add intrinsics and Ops to read smem-sizes (PR #173089)

[Alex MacLean via llvm-commits]

================
@@ -2406,6 +2406,10 @@ defm int_nvvm_read_ptx_sreg_cluster_nctaid : PTXReadSRegIntrinsicNB_v4i32<MAX_GR
 def int_nvvm_read_ptx_sreg_cluster_ctarank : PTXReadSRegIntrinsicNB_r32;
 def int_nvvm_read_ptx_sreg_cluster_nctarank : PTXReadSRegIntrinsicNB_r32;
 
+def int_nvvm_read_ptx_sreg_total_smem_size : PTXReadSRegIntrinsicNB_r32;
----------------
AlexMaclean wrote:

nit: perhaps we should give these things names like "nvvm.read.prx.sreg.total_smem_size". The underscore is more consistent with PTX while the "." seems to imply that each word is a subfield. 

https://github.com/llvm/llvm-project/pull/173089