[PATCH] D28742: [InstCombine] Don't DSE across readnone functions that may throw

[Daniel Berlin via llvm-commits]

On Sun, Jan 15, 2017 at 12:02 PM, Sanjoy Das <sanjoy at playingwithpointers.com
> wrote:

> Hi,
>
> On Sun, Jan 15, 2017 at 11:46 AM, Daniel Berlin <dberlin at dberlin.org>
> wrote:
> > FWIW: I agree.
> > But i'd go farther.
> > I think at this point, ->hasSideEffects (or ->sideEffects) may just want
> to
> > return a mask, so that the above just becomes
> >
> > ->sideEffects & (MayWrite | MayRead | MayThrow)
> >
> > or whatever nice version we want.
>
> Yes, that's a good idea.  I suspect we'll still want to keep things
> like mayRead() and mayThrow(), since it would be tedious to have to
> write
>
>  -> sideEffects() & Instruction::MayThrow
>

Sure. I'm mainly concerned with the places we are starting to test like 30
things :)


>
> when we care about just one single thing, but I like the idea.
>
Yup.


>
> I'm not sure about the name though -- generally, most people would not
> say MayRead is a side-effect.  I was thinking of something like
> "implicitBehavior()", since this really is an enum of things that an
> instruction can do that are not explicitly specified in SSA / CFG.
>


Yes. I think "side effects" is already way too overloaded :)


> One thing though: let's not stall this set of patches on solving that
> other problem. :)
>

Sure
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20170115/248adb48/attachment.html>