[LLVMdev] DebugInfo and Metadata Store

[Nick Lewycky]

2009/11/30 Devang Patel <devang.patel at gmail.com>

> On Mon, Nov 30, 2009 at 10:14 AM, Nick Lewycky <nlewycky at google.com>
> wrote:
> > What's the memory usage penalty of that on a
> > large program?
>
> I have not measured that yet.
>

I think that's the reason we chose the "on the side" design that we did.
Please measure it that so we know what tradeoff we'd be making.

 > If we're willing to pay this memory cost, even for programs that aren't
> > using debug info, why not just store pointer to a linked list of metadata
> > info in the Instruction and remove the MetadataContext? If it's a 10%
> > performance win for debug info we should be able to get the exact same
> win
> > across the board for all metadata with no cost beyond what your patch
> > already proposes.
>
> It all depends on how metadata is used. I am myself not very thrilled
> about this patch (hence requesting feedback first), but I'd try every
> possible opportunity to speedup -O0 -g compile time :)
>

Sure, I was just thinking it could be made more general. Sparse metadata
with the MetadataContext that we have now, dense metadata with a linked-list
on the Instruction. If we did this, we could require MDKinds to pick sparse
or dense so we don't have to play mix-and-match all the time.

Nick
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-dev/attachments/20091130/3a7afe36/attachment.html>