[llvm-bugs] [Bug 34041] [X86] IR shufflevector instructions with undef values lowered to sub-optimal sequences

[via llvm-bugs]

https://bugs.llvm.org/show_bug.cgi?id=34041

Simon Pilgrim <llvm-dev at redking.me.uk> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|NEW                         |RESOLVED
 Fixed By Commit(s)|                            |r354340,r356043,r356064,r35
                   |                            |6067
         Resolution|---                         |FIXED

--- Comment #3 from Simon Pilgrim <llvm-dev at redking.me.uk> ---
Resolving this, we now handle all the scalar/subvector broadcast cases raised
in this bug.

We currently prefer to use VBLENDPS etc. for constant blends where possible
over the AVX512 masks due to the cost of transferring from gpr to k-regs. If
that is a problem it should be raised as another bug.

This shows the above cases with non-constant masks showing the broadcasts are
still used:
https://godbolt.org/z/yKEKuW

-- 
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/20190315/77ed25aa/attachment.html>