- 24 May, 2011 3 commits
-
-
-
Kenneth Moreland authored
-
Kenneth Moreland authored
All the interesting changes on Release were already in master. This merge is basically for safety sake.
-
- 13 May, 2011 7 commits
-
-
Kenneth Moreland authored
-
Kenneth Moreland authored
Use it to correct some background correction problems with blank tiles.
-
Kenneth Moreland authored
-
Kenneth Moreland authored
Makes us more assured that a failure will actually be reported as such.
-
Kenneth Moreland authored
-
Kenneth Moreland authored
-
Kenneth Moreland authored
-
- 12 May, 2011 5 commits
-
-
Kenneth Moreland authored
Previously background correct only happened on display processes after the image was collected. Now we encourage distributed correction before collection. In this case, it was wrong to not correct on non-display processes.
-
Kenneth Moreland authored
This also enables background correction in the sequential and reduce strategies.
-
Kenneth Moreland authored
-
Kenneth Moreland authored
Created new state variables to capture the final background and a flag whether the correction is really necessary. Also moved the background correct function to image.c, but it is not used yet.
-
Kenneth Moreland authored
-
- 11 May, 2011 3 commits
-
-
Kenneth Moreland authored
-
Kenneth Moreland authored
-
Kenneth Moreland authored
-
- 05 May, 2011 6 commits
-
-
Kenneth Moreland authored
-
Kenneth Moreland authored
-
Kenneth Moreland authored
-
Kenneth Moreland authored
-
Kenneth Moreland authored
-
Kenneth Moreland authored
The method did an all gather that collected on each process masks for all processes for what processes it gets messages. This takes n^2 where n is the number of processes. For large numbers of processes, that gets really big. This change uses all to all to collect only which messages the local process gets.
-
- 04 May, 2011 12 commits
-
-
Kenneth Moreland authored
-
Kenneth Moreland authored
-
Kenneth Moreland authored
-
Kenneth Moreland authored
-
Kenneth Moreland authored
-
Kenneth Moreland authored
Previously, debugging logs were shown in tests and had to be shut off. This lead to very verbose output that hardly ever helped much. Now the behavior is the opposite. Debug logs are not shown by default but can be turned on with a flag (assuming compiled in debug mode).
-
Kenneth Moreland authored
Messages that print everywhere should also report from what process they come from.
-
Kenneth Moreland authored
-
Kenneth Moreland authored
-
Kenneth Moreland authored
-
Kenneth Moreland authored
-
Kenneth Moreland authored
-
- 03 May, 2011 2 commits
-
-
Kenneth Moreland authored
-
Kenneth Moreland authored
This was emperically shown to be a better value than 2048 (at least on Intrepid).
-
- 09 Apr, 2011 2 commits
-
-
Kenneth Moreland authored
The point was that if a split did not happen, collect to the image_dest if possible. This way, the pixels would not have to be transferred later. However, in the real use case, there are not that many pixels to transfer. The use case of transfering a large amount of pixels (such as a tree-like composite) does not currently exist.
-
Kenneth Moreland authored
Turned on ordered compositing so that image dest goes to final process. This is causing problems with indexing.
-