[LLVMdev] IR extension proposal: bitset constants

[Peter Collingbourne]

On Tue, Feb 03, 2015 at 04:26:16PM -0800, Sean Silva wrote:
> On Tue, Feb 3, 2015 at 4:13 PM, Peter Collingbourne <peter at pcc.me.uk> wrote:
> 
> > On Tue, Feb 03, 2015 at 04:03:45PM -0800, Sean Silva wrote:
> > > One other thing: if this can be used for control-flow integrity, I assume
> > > it has to have good knowledge of the available indirect call targets.
> > Could
> > > we also use this information for devirtualization?
> >
> > I would expect so. If a bitset contains only one element, we should be able
> > to teach the lowering pass to simply test that the given pointer is equal
> > to that element (i.e. the only valid vptr). If the later IR branches to a
> > trapping block if that condition is false, it should be possible for the
> > optimizer to deduce that the condition is true in any code that is
> > dominated
> > by the branch, and from that do devirtualization.
> >
> 
> 
> Well, I was wondering about for devirtualization without necessarily
> needing any of the CFI stuff. We should probably aim for a design that
> provides a primitive that can satisfy the needs of both CFI and
> devirtualization.

A client that doesn't need CFI could emit IR that looks like this:

%pred = call i1 @llvm.bitset.test(i8* %vptr, metadata !"some_bitset")
call void @llvm.assume(i1 %pred)
; make the call in the normal way

and from that we should be able to lower to the same type of IR for
single-element bitsets and make the same sort of deduction. (The lowering
pass could avoid emitting bitsets unless the call is used by something other
than a call to llvm.assume.)

Thanks,
-- 
Peter