Quoting James Darnley (2016-01-19 17:54:39) > On 2016-01-19 17:24, Vittorio Giovara wrote: > > On Fri, Jan 15, 2016 at 2:35 PM, James Darnley <[email protected]> > > wrote: > >> The sample factor must be the same for both 8- and 10-bit functions chosen > >> otherwise the output will be incorrect. > > > > Thanks for the set, can you explain in more detail why this > > requirement is necessary? > > > > Before I wrote the 10-bit encoder (and the tests for it) the init > function could have chosen avx for 10-bit and avx2 and sample_factor 2 > for 8-bit. At that time it didn't matter that they were different > because the two depths were on quite different code paths. The 10-bit > branch wasn't yet changed to use sample_factor. > > Now that sample_factor exists and is used on both branches it must be > the same for the two functions chosen. If it isn't then one depth will > end up doing too few pixels, run into the padding, mess up the > width/stride. I think I saw all three of these while working on it. > > In the future someone may add a new simd function on some other platform > for one depth. Neon maybe. Maybe AVX512 will only be worth it on one > depth and not the other. > > I am will admit that my logic may be wrong regarding this but I thought > it worth while to note the limitation (as I saw it) to perhaps save > someone time debugging in the future.
Does it make sense for both 8bit and 10bit functions to be called on one context? IIUC, it only works this way for the checkasm tests. So perhaps it'd be better and more future-proof to initialize the context for a specific bitdepth, and just have two inits in the checkasm test. -- Anton Khirnov _______________________________________________ libav-devel mailing list [email protected] https://lists.libav.org/mailman/listinfo/libav-devel
