[Mlir-commits] [llvm] [mlir] [MLIR][NVVM] Add NVVMRequiresSM op trait (PR #126886)

[Guray Ozen]

================
@@ -1793,6 +1793,22 @@ void GPUModuleOp::setTargets(ArrayRef<TargetAttrInterface> targets) {
   targetsAttr = ArrayAttr::get(getContext(), targetsVector);
 }
 
+LogicalResult GPUModuleOp::verify() {
+  auto targets = getOperation()->getAttrOfType<ArrayAttr>("targets");
+
+  if (!targets)
+    return success();
+
----------------
grypp wrote:

The LLVM backend already has those checks, so what we’re doing here is useful but redundant.

How about we take a big kernel and see how much compile-time it takes?

I’m not sure about a pass that verifies the operations. But it’s definitely a good idea to have a way to enable or disable it. 


https://github.com/llvm/llvm-project/pull/126886