[cfe-commits] r57149 - /cfe/trunk/lib/AST/Expr.cpp

[Daniel Dunbar]

On Sun, Oct 5, 2008 at 11:25 PM, Chris Lattner <sabre at nondot.org> wrote:

> Not entirely good and not entirely bad.  :)  It depends on how many
> identifiers you're talking about here.  Currently, the tradeoff is that the
> identifiers need to be populated when the compiler starts up, so the more
> identifiers with magic stuff hanging off them, the more that needs to be set
> up at startup-time.
>
> I supposed the best way to solve this is the push into the "first instance
> of an identifier" codepath.  It would be reasonable and nice to do a quick
> hash lookup from a static hash table to see if the new identifier is
> "known".  If so, the new identifierinfo would be populated with info from
> that table.
>
> Does that make sense?
>

Yes. In my opinion this is an argument for centralizing the mechanism so we
can perform such optimizations more easily (if necessary). In the end there
should be significantly less "known" IDs than X86 builtins. :)

 - Daniel
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/cfe-commits/attachments/20081006/41737f46/attachment.html>