[Mlir-commits] [mlir] [MLIR][NVVM] Improve inline_ptx, add readwrite support (PR #154358)

[Durgadoss R]

================
@@ -1870,6 +1872,21 @@ llvm::Intrinsic::ID PrefetchOp::getIntrinsicID(NVVM::PrefetchOp &op) {
   }
 }
 
+bool NVVM::InlinePtxOp::getAsmValues(
+    RewriterBase &rewriter,
+    llvm::SmallVectorImpl<std::pair<mlir::Value, mlir::NVVM::PTXRegisterMod>>
+        &asmValues) {
+  for (auto arg : getReadWriteArgs())
+    asmValues.push_back({arg, mlir::NVVM::PTXRegisterMod::ReadWrite});
+  for (auto arg : getResults())
+    asmValues.push_back({arg, mlir::NVVM::PTXRegisterMod::Write});
+  for (auto arg : getReadOnlyArgs())
+    asmValues.push_back({arg, mlir::NVVM::PTXRegisterMod::Read});
+  if (getPredicate())
+    asmValues.push_back({getPredicate(), mlir::NVVM::PTXRegisterMod::Read});
+  return false; // No manual mapping needed
+}
----------------
durga4github wrote:

Now that we have this method, do we still need the `defaultImpl` in the td file?

https://github.com/llvm/llvm-project/pull/154358