[llvm-dev] [RFC] Clean up the way we store optional Function data
David Majnemer via llvm-dev
llvm-dev at lists.llvm.org
Mon Oct 12 11:44:24 PDT 2015
On Mon, Oct 12, 2015 at 11:00 AM, Duncan P. N. Exon Smith via llvm-dev <
llvm-dev at lists.llvm.org> wrote:
> > On 2015-Oct-12, at 10:41, Sanjoy Das <sanjoy at playingwithpointers.com>
> > Vedant Kumar wrote:
> >>>> That's a neat idea. To summarize: make Function have 3 optional
> operands. (For context -- Function currently has 1 optional operand, and my
> proposal is to move to 0.)
> >>>> Could someone else chime in on what they'd like to see?
> >>> Sanjoy's idea makes sense to me, but only if we never need to add
> >>> prefix/prologue data after functions are created. Are there any places
> >>> where we need/want to add them after the fact?
> >> I think so. I see:
> >> LinkModules.cpp: Dst.setPrefixData(MapValue(Src.getPrefixData(),
> >> BitcodeReader.cpp:
> >> InlineFunction.cpp: Caller->setPersonalityFn(CalledPersonality);
> >> Some of these sites could be refactored so that the Functions are
> created with the prefix/prologue data they need. I don't think that's
> possible for personality functions (see my third example).
> >> Would we inhibit any future patches which add prefix/prologue data to
> Functions on the fly by taking this approach?
> > You should always be able to create a new `llvm::Function` instance (and
> RAUW it in) if you want to add prefix/prologue data to functions after
> they've been created; just like you have to do today for any other
> `llvm::User`s that do not have hung off uses.
> It's possible, but a lot more involved with `Function`s. Besides
> RAUW, you need to transfer over all the basic blocks.
> This seems kind of wrong to me, if we expect it to happen.
> > Which brings me to -- can you use hung off uses for this? These use
> lists can be resized on the fly, so you should be able to add and remove
> prologue data on the fly. If you're using hung off uses, you'll probably
> still need a descriptor to remember whether / which operands are prologue
> data etc.
> Sure, this is another option. It might be simplest. I'd be
> tempted to start with a 0/3 choice (if we allocate any hung-off
> uses, allocate enough for all three operands) to simplify the
> logic. Although I can't remember right now whether that's
> legal (having nullptr operands followed by real ones)...
> >>>>>> Personalities are stored as ``optional`` Function operands. We
> actually always
> >>>>>> allocate the space for this ``optional`` operand: there's a FIXME
> in the
> >>>>>> destructor for Function about this.
> Makes me wonder, why didn't we use hung off uses to begin with?
> Do functions "usually" have personality functions, for some
> definition of?
Depends. In C++? It's pretty common to have objects which have non-trivial
destructors on the stack which means calling a function will be an invoke
which will require the function to have a personality. In C? It's pretty
rare. You'd need something like __attribute__((cleanup)) to do it, the most
common source of this will be something like pthread_cleanup_push. If I
recall correctly, Julia sets the personality on functions regardless of
whether or not there are any invokes, they need the AsmPrinter to scribble
something down. I can't say for other languages (Rust, etc.). From what I
understand, Swift doesn't use landingpad for EH so they wouldn't need the
> LLVM Developers mailing list
> llvm-dev at lists.llvm.org
-------------- next part --------------
An HTML attachment was scrubbed...
More information about the llvm-dev