[LLVMbugs] [Bug 15247] [BBVectorizer] Possible regression of vectorization of Linpack on ARM

[bugzilla-daemon at llvm.org]

http://llvm.org/bugs/show_bug.cgi?id=15247

Renato Golin <renato.golin at linaro.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|NEW                         |RESOLVED
         Resolution|---                         |INVALID

--- Comment #5 from Renato Golin <renato.golin at linaro.org> ---
I just remembered that the 64-bit floats are done in the VFP pipeline, so only
access to double registers, thus only doing one 64-bit FP operation per cycle
(at most).

All instructions seem to be VFP, and also generated on the non-vectorized case.
Debugging the loop vectorizer I just learnt that it's not vectorizing anything,
the VFP instructions are comming from the fact that it's hard-fp and VFP.64 is
faster than soft-fp math libraries.

Silly me...

-- 
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/20130213/10f5db25/attachment.html>