[Qgis-developer] Re: Benchmarking / optimization

[Radim Blazek]

I have updated the benchmark. The first version was correct, the last
one was wrong. Last time I have accidentally used qgis_bench compiled
with rpath and lib versions so that it was always using the same (last
version) lib.

I have also bisected slowdown caused by 85b55ca5, which was correct
fix for previous 8c1c2add and it is already optimized by 8e2e654c.
Introduction of "relevant fields" is significant improvement for ogr
layers with many attributes, we can hardly dream about more 70%
optimizations. It just took some time to mature - continuous
benchmarking should point to such cases immediately.

There is another slowdown with forest data (complex polygons, no
attributes) between d5a6beb1 and 93405b10, but bisecting is not going
to an obvious commit, slowdown is gradual.

Radim

On Thu, Dec 1, 2011 at 8:54 AM, Radim Blazek <radim.blazek at gmail.com> wrote:
> On Mon, Nov 28, 2011 at 7:49 PM, Radim Blazek <radim.blazek at gmail.com> wrote:
>> You may notice, that something bad has probably happened between
>> 8c1c2add and 1d7d1c62 (if there is no bug in my scripts).
>
> False alarm, I have rerun the bench and the problem disappeared. I
> don't have any explanation yet, the same build, the same project ...
> Now there are mostly positive changes only.
>
> Chart updated http://www.mpasol.it/blazek/qbench/
> but don't take it too seriously for now, I must verify the whole process.
> Sorry for confusion.
>
> Radim