> So even in hardware, it made sense to reserve a special register for the stack pointer. I think it was mostly a question of cleanup / clarity of the interfac

> So even in hardware, it made sense to reserve a special register for the stack pointer. I think it was mostly a question of cleanup / clarity of the interface: on ARM32, using R13/SP as a GPR could lead to odd behaviours and was a bit of a portability concern, to say nothing of R15 (PC/IP, which was only a GPR in name IIRC as some instructions excluded it and it had divergent side-effects). Also unlike every other GPR, R13 (and R14) was banked on all privileged modes (R0-R7 were shared across mode, and R7-R12 were banked in FIQ). Moving SP and PC to dedicated non-GPR makes the instruction set easier to work with (and allowed register 31 to pull double duty as a virtual reference to SP and ZR depending on instruction, and thus ARM to finally have a zero register).