[PATCH 1/2] Make strtok benchtest competitive

[Siddhesh Poyarekar]

On 27 February 2014 18:02, Ondřej Bílka <neleai@seznam.cz> wrote:
> Hi,
>
> One purpose of benchtests is look for regressions. For that you need to
> test implementations that have chance to be fast.
>
> A baseline is a string/function implementation. It mostly reduces
> problem to other functions that could be better optimized than
> benchmarked one and in that case we know that we need to fix it.
>
> This is a case with strtok on sse4_2 capable machines as strpbrk is
> optimized here but strtok is not. As result a current assembly strtok is
> 2-4 times slower than generic one, results on ivy_bridge below.
>
> OK to commit this benchmark?

Looks good to me.

Siddhesh