[Mlir-commits] [mlir] [MLIR][NVVM] Extend TMA Bulk Copy Op (PR #140232)

Guray Ozen llvmlistbot at llvm.org
Fri May 16 22:47:57 PDT 2025


================
@@ -1253,6 +1253,35 @@ CpAsyncOp::getIntrinsicIDAndArgs(Operation &op, LLVM::ModuleTranslation &mt,
   return id;
 }
 
+std::pair<llvm::Intrinsic::ID, llvm::SmallVector<llvm::Value *>>
+CpAsyncBulkSharedCTAToGlobalOp::getIntrinsicIDAndArgs(
+    Operation &op, LLVM::ModuleTranslation &mt, llvm::IRBuilderBase &builder) {
+  auto thisOp = cast<NVVM::CpAsyncBulkSharedCTAToGlobalOp>(op);
+  llvm::SmallVector<llvm::Value *> args;
+  llvm::Intrinsic::ID id =
+      llvm::Intrinsic::nvvm_cp_async_bulk_shared_cta_to_global;
+
+  // Fill the Intrinsic Args
+  args.push_back(mt.lookupValue(thisOp.getDstMem()));
+  args.push_back(mt.lookupValue(thisOp.getSrcMem()));
+  args.push_back(mt.lookupValue(thisOp.getSize()));
+
+  auto cacheHint = thisOp.getL2CacheHint();
----------------
grypp wrote:

LLVM doesn't use auto when type isn't obvious. (It is okay to use for the auto a = cast<type>)

https://github.com/llvm/llvm-project/pull/140232


More information about the Mlir-commits mailing list