Bug with DFG to MLIR HLS generation

[gcallanan]

I have a bug where the hls mlir file is not generating correctly under some conditions:

I have the following main.mlir file

dfg.process @Temp
	inputs(%In: i32)
	outputs(%Out: i32)
{
	dfg.loop inputs(%In: i32)
	{
		%l_t1__1_d0_0 = dfg.pull %In : i32
		%l_a__3_d0_0 = arith.constant 0 : i32
		dfg.push(%l_t1__1_d0_0) %Out : i32
	}

}

func.func @top(%In: i32) -> (i32)
{
	%queue_from_In, %queue_to_temp_In = dfg.channel(5) : i32
	%queue_from_temp_Out, %queue_to_Out = dfg.channel(5) : i32
	dfg.push(%In) %queue_from_In : i32
	dfg.instantiate @Temp // Instance name: temp
		inputs(%queue_to_temp_In)
		outputs(%queue_from_temp_Out) :
		(i32) -> (i32)
	%Out = dfg.pull %queue_to_Out : i32
	func.return %Out: i32
}

Which I build with the command: dfg-opt main.mlir --convert-std-to-circt --convert-dfg-to-circt --convert-fsm-to-sv --lower-seq-to-sv --export-split-verilog

This generates the following hls_Temp_calc.mlir:

"func.func"() <{function_type = (i32) -> (), sym_name = "hls_Temp_calc"}> ({
^bb0(%arg5: i32):
  %6 = "arith.constant"() <{value = 0 : i32}> : () -> i32
  "func.return"() : () -> ()
}) : () -> ()

and then throws an error: dfg-opt: /home/gareth/software-repos/mlir_related/dfg-mlir-240522/circt/llvm/llvm/include/llvm/ADT/ArrayRef.h:443: T &llvm::MutableArrayRef<mlir::BlockArgument>::operator[](size_t) const [T = mlir::BlockArgument]: Assertion Index < this->size() && "Invalid index!"' failed..

If I comment out the line: %l_a__3_d0_0 = arith.constant 0 : i32 then things seem to work. For context, I am default initializing a variable to 0 and then assigning a token from dfg.pull from that node, so the l_a__3_d0_0 operand ends up never getting used. Its not common but can happen depending on the code compiled. It seems strange that an unused operation would cause such a crash.