[PATCH] Allow 0-weight branches in BranchProbabilityInfo.

[Duncan P. N. Exon Smith]

> On 2015 May 6, at 13:39, Diego Novillo <dnovillo at google.com> wrote:
> 
> 
> 
> On 05/06/15 13:37, Hans Wennborg wrote:
>> On Wed, May 6, 2015 at 10:09 AM, Duncan P. N. Exon Smith
>> 
>> Are you saying you don't want to guarantee that anymore, or that
>> forcing each weight <= UINT32_MAX / SI.getNumSuccessors() is
>> unnecessarily strict?
> 
> Yeah, I've wanted to remove this restriction for some time. The only requirement we really need is that the sum of all the weights fits within 32 bits.
> 
> I'm going to be removing this restriction in http://reviews.llvm.org/D9442

You'll have to leave it in place until the downstream consumers stop
flooring the weights at 1, but as long as there's a plan for lifting
the restriction I'm happy.