[PATCH] D26855: New unsafe-fp-math implementation for X86 target

[Gerolf Hoflehner via llvm-commits]

> On Jan 27, 2017, at 1:07 AM, Andrew V. Tischenko via Phabricator <reviews at reviews.llvm.org> wrote:
> 
> avt77 added a comment.
> 
> In https://reviews.llvm.org/D26855#657735, @Gerolf wrote:
> 
>> I think the only issue that needs to be addressed is (finally!) sharing perf data. This has been raised at least 3 times. The possible compile-time implication, the speciality of the application (fast-math) etc are well understood.
>> 
>> Gerolf
> 
> 
> Sorry but I don't understand what means sharing in this case? I put all perf numbers here in comments. Is not it enough for sharing? If not where should I share it? Or maybe my perf numbers are not perf numbers from your point of view? Please, clarify.
> 
> And next question about profiling data. Should I collect it? I've already started the process but now I'm not sure if it's interesting for somebody.

You are working on performance. I expect that you have a set of performance benchmarks that motivated your work and that show gains from your implementation. What you have shown so far is compile-time cost. There is also at least a maintenance trade-off for the new code.  And I took note of your test cases. From my perspective you addressed my concerns for these aspects. What is left is demonstrating actual performance data that show all is worth it.

Thanks
Gerolf
> 
> 
> https://reviews.llvm.org/D26855
> 
> 
>