On 4/13/25 11:03 PM, Alexandre Oliva wrote:
On Apr 12, 2025, Jeff Law <jeffreya...@gmail.com> wrote:
for gcc/ChangeLog
PR target/118182
* config/riscv/vector.md (@pred_broadcast<mode>): Expand to
_zero and _imm variants without vec_duplicate.
I'd said this should tend to wait for gcc-16 since it doesn't fix a
regression.
That works for me.
Technically, the failure now in gcc-14 is not a regression indeed, but
it may seem like one. I have no objections to leaving the patch for
gcc-16 only, and the failure in gcc-14 is nothing but a missed
optimization, but it's a failure nevertheless.
As for the newly-backported failing test in gcc-14... Would you rather
leave it failing-14, adjust it so as to expect the code output by
gcc-14, remove the failing test, or revert the recently-backported patch
that introduced it? I'd be happy to go either way, but my preference
would be for adjusting the test.
No strong opinion. I'd lean towards xfail or twiddling the test since
that's obviously super-save WRT codegen on the gcc-14 release branch.
So just keep it in mind as you're poking around -- what you're finding
likely will show up elsewhere and I'm supportive of moving this stuff
to expansion time.
ACK, thanks for the heads up and support. It's not like I've been
*looking* for such stuff, it's just that FAILs often catch my attention,
especially if they seem like regressions.
Understood. Just wanted to point out something I'd seen which I think
is relatively common. I'd hate to see work duplicated, so better to let
folks know what's in flight than be silent.
jeff
