[llvm-commits] PATCH: Aligned bundling support in LLVM MC

[Eli Bendersky]

On Fri, Dec 14, 2012 at 4:29 PM, Greg Fitzgerald <garious at gmail.com> wrote:
>> +  // If bundling is enabled:
>> +  // - If we're not in a bundle-locked group, emit the instruction into a
>> data
>> +  //   fragment of its own.
>
> Hi Eli,
>
> This looks expensive.  Do you have any numbers that compare compile time and
> memory usage with and without bundling enabled?

Yes, this is certainly one place that needs optimization when
assembling large objects with bundling enabled (it doesn't affect the
bundling-disabled case). We have some optimizations available (*), but
left them out of this initial patch to keep it simple. Once the
feature gets into LLVM, it will be possible to improve incrementally.

Eli

(*) For example, this is much less needed on ARM and other
architectures where instructions are constant-size and have to be
aligned to some factor of the bundle size anyway.