[mlir] Remove unnecessary llvm.mlir.cast in AsyncToLLVM lowering.

[ftynse]

Is it always unnecessary or is there a possibility that it still might be (e.g., async pointer)?

[csigg]

In D94400#2489840, @ftynse wrote:

Is it always unnecessary or is there a possibility that it still might be (e.g., async pointer)?

AsyncToLLVM lowering only supports std and llvm types. If the former is lowered to llvm but one of the uses expects a std type, the llvm type converter will already materialize an llvm.mlir.cast if needed.

But a manually created instance can be in the way if the async value type is already an llvm type, because llvm.mlir.cast doesn't currently have a folder that would remove a cast between equal types.

Please take my explanation with a grain of salt, my understanding is limited...

mlir/lib/Conversion/AsyncToLLVM/AsyncToLLVM.cpp

auto storage = rewriter.create<CallOp>(loc, kGetValueStorage,
                                            TypeRange(i8Ptr), operand);
 
     // Cast from i8* to the pointer pointer to LLVM type.
     auto llvmValueType = getTypeConverter()->convertType(valueType);
     auto castedStorage = rewriter.create<LLVM::BitcastOp>(
         loc, LLVM::LLVMPointerType::get(llvmValueType), storage.getResult(0));
 
     // Load from the async value storage.
-    auto loaded = rewriter.create<LLVM::LoadOp>(loc, castedStorage.getResult());
-
-    // Cast from LLVM type to the expected value type if necessary. This cast
-    // will become no-op after lowering to LLVM.
-    if (valueType == loaded.getType())
-      return loaded;
-    return rewriter.create<LLVM::DialectCastOp>(loc, valueType, loaded);
+    return rewriter.create<LLVM::LoadOp>(loc, castedStorage.getResult());
   }
 };
 
 /// Lowering for `async.await_all` operation.
 class AwaitAllOpLowering : public AwaitOpLoweringBase<AwaitAllOp, GroupType> {
   using Base = AwaitOpLoweringBase<AwaitAllOp, GroupType>;
 
 public: