On Sun, Jun 08, 2025 at 09:05:07AM -0600, Jeff Law wrote:
>
>
> On 6/7/25 4:38 AM, Stafford Horne wrote:
>
> >
> > ## Note on cstoresi4 and PUT_MODE.
> >
> > There was some concern raised by Jeff about the use of PUT_MODE in
> > cstoresi4.
> > This is here to convert a 'ne' to 'ne:SI', for example, allowing the
> > generated
> > RTL to match with the "*scc" insn pattern later.
> >
> > It seems to be working OK so I will leave this as is for now. But if you
> > have
> > any suggestions let me know.
> Well, it's a sign that something is wrong. Patterns shouldn't be changing
> the mode or code of an existing RTL node (say that 5 fast :-)
Understood, just some explaination, in this case we are updating RTL that was
just generated by or1k_expand_compare()..
/* Adjust the operands for use in the caller. */
operands[0] = flag_check_ne ? gen_rtx_NE (VOIDmode, sr_f, const0_rtx)
: gen_rtx_EQ (VOIDmode, sr_f, const0_rtx);
operands[1] = sr_f;
operands[2] = const0_rtx;
}
Depending on the pattern we are expanding we want either:
(ne (reg:BI 34 ?sr_f)(const_int 0 [0]))
Or:
(ne:SI (reg:BI 34 ?sr_f)(const_int 0 [0]))
The PUT_MODE does the final adjustment.
As a cleanup, the PUT_MODE could be moved into or1k_expand_compare.
> I don't mind deferring since apparently it's not playing a role in this
> issue, but that code is definitely very fishy.
OK.
-Stafford
