[PATCH] [Review Request] Function-based parallel LLVM backend code generation

[Eric Christopher]

On Tue, Oct 8, 2013 at 8:27 AM, Renato Golin <renato.golin at linaro.org> wrote:
> On 8 October 2013 09:34, Wan, Xiaofei <xiaofei.wan at intel.com> wrote:
>>
>> In this patch, function-based parallelism is adopted to accelerate backend
>> code generation. Please review/discuss it.
>
>
> As much as I think this would work, it wouldn't play nice with distributed
> builds (make -jN). Even if you could teach build systems how many threads a
> process could use, it would never use 100% of them all, all the time, being
> a waste of resources on huge source trees.
>

Well, I doubt it'll be a waste, however, the case of codegen after LTO
is a specific and necessary case that this solves. I.e. how do you
generate code quickly after you've linked everything together?

I've not had a chance to look at the patch itself yet, but the choices
here are essentially:

a) some scheme like this,
b) some partitioning scheme like Shuxin's.

-eric