Re: [Mono-dev] Re. PR 1501: Mark hot functions in the runtime.

[Miguel de Icaza]

Hello Jonathan,

There seems to have been a couple of regressions on the test suite, but it
is hard to read from the graph.

Do we know what could have caused this?

On Mon, Jan 26, 2015 at 8:45 PM, Jonathan Purdy 
wrote:

> In this pull request[1] I add GCC/Clang’s “__attribute__((hot))” to
> various hot GC functions. This increases code size a small amount, and
> makes a very small improvement in performance—see the attached graph. I
> think actually marking some slow-path/error-handling functions as
> “__attribute__((cold))” would do more to improve branch prediction in cases
> where we haven’t marked things as “G_UNLIKELY”.
>
> [1]: https://github.com/mono/mono/pull/1501
>
>
> ___
> Mono-devel-list mailing list
> Mono-devel-list@lists.ximian.com
> http://lists.ximian.com/mailman/listinfo/mono-devel-list
>
>
___
Mono-devel-list mailing list
Mono-devel-list@lists.ximian.com
http://lists.ximian.com/mailman/listinfo/mono-devel-list

[Mono-dev] Re. PR 1501: Mark hot functions in the runtime.

[Jonathan Purdy]

In this pull request[1] I add GCC/Clang’s “__attribute__((hot))” to various
hot GC functions. This increases code size a small amount, and makes a very
small improvement in performance—see the attached graph. I think actually
marking some slow-path/error-handling functions as “__attribute__((cold))”
would do more to improve branch prediction in cases where we haven’t marked
things as “G_UNLIKELY”.

[1]: https://github.com/mono/mono/pull/1501
___
Mono-devel-list mailing list
Mono-devel-list@lists.ximian.com
http://lists.ximian.com/mailman/listinfo/mono-devel-list