[Mlir-commits] [mlir] [MLIR][NVVM] Update prefetch.tensormap Op (PR #153134)

Guray Ozen llvmlistbot at llvm.org
Sun Aug 17 23:37:34 PDT 2025


================
@@ -1265,6 +1265,20 @@ LogicalResult NVVM::PrefetchOp::verify() {
   return success();
 }
 
+LogicalResult NVVM::PrefetchTensorMapOp::verify() {
+  using MemSpace = NVVM::NVVMMemorySpace;
+  unsigned addressSpace =
+      llvm::cast<LLVM::LLVMPointerType>(getTmaDescriptor().getType())
+          .getAddressSpace();
+
+  if (getInParamSpace()) {
+    if (addressSpace != MemSpace::kGenericMemorySpace)
+      return emitOpError(
+          "in_param_space can only be specified for a generic pointer");
----------------
grypp wrote:

nit: I'd add {} as it's multiline

https://github.com/llvm/llvm-project/pull/153134


More information about the Mlir-commits mailing list