On 05/23/16 15:32, Evandro Menezes wrote:
I'm fine with this patch, as it achieves in part what I intended
before: going beyond the default_case_values_threshold, too
conservative for Exynos M1. My concern is particularly what happens
to in-order targets, like the ubiquitous A53.
I'll get make some figures available soon.
Here's what I noticed using our internal benchmark suite:
* On A53, no noticeable regressions (< -1%) and a handful of minor
improvements (< 3%) here and there.
* On A57, again, no noticeable regressions, a handful of minor
improvements and a some significant improvements (< 5%) here and there.
* On M1, it was more checkered, with some minor regressions, but with
a few minor and significant improvements, resulting in an overall
minor improvement.
I'm still comfortable with the idea of the patch. I agree with Jim that
the logic that it adds begs to be polished up along the lines that he
suggested.
Cheers,
--
Evandro Menezes
