3 months agoOSX: little logical refinement in cmake
Jens Verwiebe [Sat, 04 Feb 2012 11:41:14 +0100] rev 3475
OSX: little logical refinement in cmake

3 months agoRPF: muli-pass bilateral filtering code random_parameter_filtering_test
David Bucciarelli <dade@ngi.it> [Thu, 02 Feb 2012 19:02:05 +0100] rev 3474
RPF: muli-pass bilateral filtering code

3 months agoRPF: first working bilateral filtering code random_parameter_filtering_test
David Bucciarelli <dade@ngi.it> [Thu, 02 Feb 2012 18:42:13 +0100] rev 3473
RPF: first working bilateral filtering code

3 months agoRPF: working of filtering code random_parameter_filtering_test
David Bucciarelli <dade@ngi.it> [Wed, 01 Feb 2012 12:04:19 +0100] rev 3472
RPF: working of filtering code

3 months agoFixed silly error in new tiled contribution code, leading to banding.
Asbjørn Heid <lordcrc@gmail.net> [Sat, 04 Feb 2012 06:12:21 +0100] rev 3471
Fixed silly error in new tiled contribution code, leading to banding.

3 months agoMerged with tiled_contribution_test.
Asbjørn Heid <lordcrc@gmail.net> [Fri, 03 Feb 2012 21:30:42 +0100] rev 3470
Merged with tiled_contribution_test.

The Film class now support the notion of thread-independent "tiles" for adding contributions.
This means that differen threads may add contributions to different tiles in parallel.

For now the code assumes that a contribution overlaps at most two tiles (due to pixel filter
extent), and the tiles are implemented as horizontal slabs. Contributions added to a given tile
are clamped to the tile boundaries.

The contribution pool has been reworked, such that a single Buffer contains contributions for
a single tile. If a contribution straddles tile boundaries, it is added to the Buffer for both
tiles.

In addition the pool will try to splat a full Buffer immediately, instead of waiting until it
has exhausted the list of free Buffers. If another thread is busy splatting on that tile, the
full Buffer is queued and the current thread returns a free Buffer. This reduces the chances
that a thread must wait for another thread to splat.

Outlier rejection is made compatible with this tiled scheme by storing extra rows for the
overlapping area between tiles. It relies on the fact that straddling contributions are added to
both tiles. Thus if the center of a contribution is outside the current tile boundary, the outlier
rejection code will store it in these extra overlap rows.

A parameter "tilecount" has been added to "fleximage" which can be used to request a specific
tile count. Otherwise the number of logical cpus is used as a basis for determining the tile
count.In either case the actual tile count is adjusted to fit certain constraints and may be
less than the requested number.

3 months agoAdded proper handling of outlier rejection with tiled contributions. tiled_contrib_test
Asbjørn Heid <lordcrc@gmail.net> [Wed, 01 Feb 2012 02:50:34 +0100] rev 3469
Added proper handling of outlier rejection with tiled contributions.

3 months agoAdded V1_NORMAL and V1_BSDF debug filter modes random_parameter_filtering_test
David Bucciarelli <dade@ngi.it> [Tue, 31 Jan 2012 17:59:59 +0100] rev 3468
Added V1_NORMAL and V1_BSDF debug filter modes

3 months agoWorking on RPF preprocessing code random_parameter_filtering_test
David Bucciarelli <dade@ngi.it> [Tue, 31 Jan 2012 17:38:09 +0100] rev 3467
Working on RPF preprocessing code

3 months agoBox filter now works (NaNs reported as a bug in plastic and substrate material) random_parameter_filtering_test
David Bucciarelli <dade@ngi.it> [Mon, 30 Jan 2012 19:08:52 +0100] rev 3466
Box filter now works (NaNs reported as a bug in plastic and substrate material)