register conversion in triton-opt

include/triton/Conversion/Passes.td

@@ -16,6 +16,12 @@ def ConvertTritonToTritonGPU: Pass<"convert-triton-to-tritongpu", "mlir::ModuleO
                              "mlir::scf::SCFDialect",
                              "mlir::triton::TritonDialect",
                              "mlir::triton::gpu::TritonGPUDialect"];
+
+   let options = [
+       Option<"numWarps", "num-warps",
+              "int32_t", /*default*/"4",
+              "number of warps">
+   ];
 }
 
 #endif

include/triton/Conversion/TritonToTritonGPU/TritonToTritonGPU.h

@@ -11,7 +11,7 @@ template <typename T> class OperationPass;
 namespace triton{
 
 std::unique_ptr<OperationPass<ModuleOp>> 
-createConvertTritonToTritonGPUPass();
+createConvertTritonToTritonGPUPass(int numWarps = 4);
 
 }
 } // namespace mlir

include/triton/Dialect/TritonGPU/IR/TritonGPUOps.td

@@ -53,6 +53,9 @@ def TTG_CopyAsyncOp : TTG_Op<"copy_async",
   let results = (outs TT_Type:$result);
 
   let assemblyFormat = "$ptr`,` $mask`,` $other attr-dict `:` type($ptr) `->` type($result)";
+
+  // result needs to be of shared layout
+  let verifier = [{ return ::verify(*this); }];
 }
 
 // Port Arith_CmpIOp & Arith_CmpFOp to TritonGPU.