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

[Hal Finkel]

----- Original Message -----
> 
> 
> On 8 October 2013 17:18, Eric Christopher < echristo at gmail.com >
> wrote:
> 
> 
> 
> 
> 
> 
> 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?
> 
> 
> 
> Hum, that didn't come out correct. Re-phrasing: It would be a waste
> if you use "make -j $(CPUs)/4" just because LLVM can spawn up to 4
> threads per compilation process.

I think that, in the common case, the 'root' of the compilation tree is where LTO optimizations are done (when building the final executable or shared library), and at that point, there is limited make-level parallelism available. There are obviously counter examples (like our test suite), but so long as we provide an easy way to turn off the extra threading, I think that it will help most people.

 -Hal

> 
> 
> cheers,
> --renato
> _______________________________________________
> llvm-commits mailing list
> llvm-commits at cs.uiuc.edu
> http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits
> 

-- 
Hal Finkel
Assistant Computational Scientist
Leadership Computing Facility
Argonne National Laboratory