[patch] Move the materalizer interface to Function

Rafael EspĂ­ndola rafael.espindola at gmail.com
Thu May 8 05:18:26 PDT 2014


On 8 May 2014 01:50, Nick Lewycky <nicholas at mxc.ca> wrote:
> Rafael EspĂ­ndola wrote:
>>
>> A rebased and slightly modified patch is attached. It keeps the
>> interface proper in GlobalValue as a convenience, but the first step
>> is a dyn_cast to Function which allows the actual materializer to be
>> implemented with Function arguments.
>
>
> I realize bitcode only lazy loads functions not globals, but is
> GVMaterializer for non-functions broken?

As far as I know the virtual interface is just so lib/IR doesn't
depend on the bitcode reader. The users of the IR itself have to know
what can and cannot be materialized. The only thing calling
Materialize on something that was not statically know to be a function
was llvm-extract.

Cheers,
Rafael




More information about the llvm-commits mailing list