[llvm] r285106 - [AArch64] Adjust the cost model for Exynos M1.

[Evandro Menezes via llvm-commits]

Renato,

Short of going into the details of the indirect branch predictor in M1, 
the change that improved the heuristics creating jump tables 
(https://reviews.llvm.org/D25212) results in a tad more conditional 
branches, which tend to perform better than indirect branches on most 
processors.  This heuristic change called for a revision of the limit of 
the size of jump tables for M1, based on a suite of benchmarks, such as 
SPEC CPU2000 and CPU2006.

Cheers,

-- 
Evandro Menezes

On 10/25/16 15:21, Renato Golin wrote:
> On 25 October 2016 at 21:05, Evandro Menezes via llvm-commits
> <llvm-commits at lists.llvm.org> wrote:
>> Author: evandro
>> Date: Tue Oct 25 15:05:42 2016
>> New Revision: 285106
>>
>> URL: http://llvm.org/viewvc/llvm-project?rev=285106&view=rev
>> Log:
>> [AArch64] Adjust the cost model for Exynos M1.
>>
>> Modify the maximum jump table size.
> Hi Evandro,
>
> Can you expand on what made you take this decision? The commit message
> should really have some more context. An email thread, a review, some
> documentation or just a short description of why magical 12 is a
> better number than magical 8.
>
> cheers,
> --renato
>
>
>