On 05/23/2014 09:33 AM, Francois Caron wrote:
Hi, While working on tile-based multi-threading, I noticed a bug in the RMD flow. 4x4 blocks were offset an additional 2 samples to the left or down. This resulted in the reconstruction writing garbage in the left and bottom blocks. When running in single-threaded mode, this went unnoticed since the encoding process followed the raster scan order. The garbage was eventually clobbered by good samples. When running in multi-threaded mode, the garbage eventually clobbered valid reconstructed samples and was messing with the rest of the process. I’ve attached the patch for review.
Good catch. Please commit this ASAP, I'm in the middle of fixing more bugs in the analysis.
Laurent -- To unsubscribe visit http://f265.org or send a mail to [email protected].
