[llvm-bugs] [Bug 47437] No vectorization - should use pmaddwd
via llvm-bugs
llvm-bugs at lists.llvm.org
Sun Oct 17 03:21:49 PDT 2021
https://bugs.llvm.org/show_bug.cgi?id=47437
Simon Pilgrim <llvm-dev at redking.me.uk> changed:
What |Removed |Added
----------------------------------------------------------------------------
Status|NEW |RESOLVED
Fixed By Commit(s)| |6ec644e2157d63acccacd877cf0
| |8c68d5836c169
Resolution|--- |FIXED
--- Comment #4 from Simon Pilgrim <llvm-dev at redking.me.uk> ---
(In reply to Simon Pilgrim from comment #3)
> (In reply to David Bolvansky from comment #2)
> > Thanks for improvements, with AVX LLVM 14 produces vpmaddwd, not yet with
> > plain "-O3".
>
> It should in fact work with most SSE41+ targets (although not SLM which has
> some custom costs that push it just over the threshold).
>
> The next step is to improve pre-AVX2 interleaved costs which should allow
> the LV costs analysis to see this as beneficial.
>
> I've started the process at https://reviews.llvm.org/D111822
https://reviews.llvm.org/D111938 should fix this whenever interleaving is
enabled.
-march=atom is the only case where we don't enable interleaving on x86,
although we probably should now that the atom model and costs have been
improved.
--
You are receiving this mail because:
You are on the CC list for the bug.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-bugs/attachments/20211017/cf4649a9/attachment-0001.html>
More information about the llvm-bugs
mailing list