[PATCH] blockfreq: Rewrite block frequency analysis

[Duncan P. N. Exon Smith]

On 2014-Apr-17, at 19:08, Duncan P. N. Exon Smith <dexonsmith at apple.com> wrote:

> 
> On 2014-Apr-17, at 14:20, Chandler Carruth <chandlerc at google.com> wrote:
> 
>> 
>> On Fri, Apr 11, 2014 at 5:00 PM, Andrew Trick <atrick at apple.com> wrote:
>>> <block-frequency-20140411.patch>
>> 
>> Thanks. I’m fine if you continue working on this in-tree since you're tracking performance.
>> 
>> At this point, we mainly want to
>> - simplify the code by moving helper code into support or eliminating the need for it (especially PositiveFloat)
>> - add the block bias computation and API
>> - maybe do a little better on irreducible CFGs
>> 
>> LGTM too.
>> 
>> I've looked through the patch, and almost all of the comments are things that really shouldn't be dealt with until the simplifying is done. (stylistic stuff like, using detail namespaces, etc. all of it likely to fall out natrually of the simplifications, no point spending time on it now)
> 
> Awesome.  r206548.
> 
> I'll start the simplifications pronto.

Pulled in r206556, as unit tests are failing [1] that pass for me locally.

I'll look into this tomorrow.

[1]: http://bb.pgr.jp/builders/ninja-x64-msvc-RA-centos6/builds/1816