[LLVMdev] Rework of Vector/Scalar Classification
dag at cray.com
Wed Dec 9 08:20:24 PST 2009
On Tuesday 08 December 2009 16:01, Dan Gohman wrote:
> > I don't see how a register MachineOperand is obviously a vector unless
> > there's type information somewhere that I'm missing.
> It has a register which has a register class. It's true that LLVM
> currently uses one register class for both scalar and vector uses
> of XMM registers on x86, but you could argue that that's a problem
> which should be fixed for other reasons anyway -- it may allow
> the use of subreg operators to model referencing the first element
> of a vector, which would allow the coalescer to coalesce them.
But until that happens, there is no alternative to the patch I submitted.
> > For now, I just don't want to lose this work. Even just applying this
> > and immediately reverting it is better than nothing because it will
> > at least be preserved in the repository history.
> Please don't abuse the repository like this.
So the answer is...no?
It's a bit frustrating to be told to rework a patch to operate in a
different (and better) fashion only to have it rejected.
More information about the llvm-dev