[llvm-commits] [llvm] r80245 - in /llvm/trunk/lib/Target/ARM: ARMISelLowering.cpp ARMInstrNEON.td

[Anton Korobeynikov]

Hello, Evan

> It's the right thing to do. If it somehow hurts performance, please
> file a bug and assign it to me.
I added a testcase for this. There it removed extra stack moves.
However, I wasn't be able to see the huge difference in the bigger
file the testcase was derived from.

I've just added the same stuff for integer vectors. It seems that
currently we're ending with redundant stuff, e.g.

vmov.32 r0, d0[0]
vmov.32 d0[0], r0

how this can be eliminated? isMoveInstr() seems to be too weak for this...

-- 
With best regards, Anton Korobeynikov
Faculty of Mathematics and Mechanics, Saint Petersburg State University