User control over splitting granularity while parallel processing

[Peter Levart]

On 11/02/2012 12:59 PM, Aleksey Shipilev wrote:
> Hi Georgiy,
>
> On 11/02/2012 03:45 PM, Georgiy Rakov wrote:
>> It seems to be useful because the computational complexity of supplied
>> functors (predicates, blocks, ... , etc) is known by nobody but user
>> who's created it.
> Exactly. There is some freedom for automatic optimization though.

Either this, or just tune it for the "typical case" and provide a means 
to specify a "granularity factor" hint. For example a factor of 2.0 
would try to make final splits twice as large as by default. The user 
will then use this to perform his own tuning.

Regards, Peter

>
>> I wonder whether such means are going to be introduced. And if they are
>> not why?
> It is being discussed. If you have the concrete suggestions about the
> API you would see useful, please shoot them here.
>
> -Aleksey.
>
>