On 12/9/25 05:38, Kito Cheng wrote:
> Hi:
>
> My colleague Min is proposing a new syntax for -mtune[1] which makes
> the mtune able to describe few more info and is configurable like
> "-mtune=rocket:short-forward-branch-i-mul".
>
> He is seeking feedback on the proposal, of course I believe we should
> not include this new stuff in GCC 16, but it's still worth discussing
> this :)
>
> I could imagine that this can reduce the number of options to control
> the behavior of processor specific code gen behavior, but in the
> meantime it increases the complexity of  the interface to the user.
>
> Anyway, feedback by reply to this email or just reply on github are
> both welcome :)
>
> [1] https://github.com/riscv-non-isa/riscv-toolchain-conventions/pull/122

We discussed this in the weekly RISC-V patchworks call and overall no one liked
the idea.

While the ask may be the well intentioned, this ends up exposing gcc internals
to the users and almost making them part of the ABI, which as we know becomes
holy and untouchable once set in stone.
Given the maturity (or lack there-of) RISC-V designs in the wild, we have lot of
tune params ATM - but all hidden away from the user which this will now unravel
and moreover prevent gcc developers from removing them for all of eternity.
Lets please not do it.

Thx,
-Vineet

Reply via email to