Now, I haven't used his yet, or any others for that matter...I've been using devkitARM, which hasn't given me any trouble. I've been getting the so called "1:1" results you apparently desire, Para, as evidenced by my ability to write assembly code that, when passed through devkitARM's utilities with the flags I've chosen, outputs a binary that is the exact binary I need, which I then format into text which can be plugged right into an Action Replay DS code manager.
... Which is pretty much what I said.
It is possible to get a 1:1 binary out of GCC, after jumping through a bunch of hoops with compiler arguments and then passing the object through objdump...
He's probably going to be upset to know that there's already an ARM assembler written in py. D: