On 8/8/25 20:39, Richard Sandiford wrote:
Richard Henderson <richard.hender...@linaro.org> writes:
The save/restore_stack_nonlocal patterns passed a DImode rtx
to gen_tbranch_neqi3 for a QImode compare. The tbranch expander
did not do what it said on the tin, that is: emit TBNZ.
It only made it as far as AND+CMP+B.cond.
Yeah, that was done to allow ifcombine to have a go first
(g:17ae956c0fa6baac3d22764019d5dd5ebf5c2b11). But the original
pattern was restricted in g:8e26ac4749c5ddf827e18a846b1010b091f76fa2
to "bit 0 of a QI or HI" that we have now.
If all tests pass without the expander then I suppose we at least need
a new test to justify the expander's existence.
No, there are a few regressions on comparisons vs _Bool. I didn't know where they were
coming from and it didn't occur to me to check if tbranch was an optab. (I hesitate to
say "new", since my gcc lore is at least 10 years out of date.)
Thanks for the links, Richard. And nice commit messages, Tamar.
It doesn't look like any other target has implemented the tbranch optab,
so if we do remove the expander, I suppose we should remove the optab
itself too (separately).
The fix for the save/restore_stack_nonlocal mode mismatch is ok
either way though.
Presumably by retaining the change to CBNZ rather than passing gen_lowpart (QImode, x16)
to tbranch? The branch over the GCS stack manipulation isn't very if-convertable. :-)
r~
