Quoting James Almer (2020-03-26 15:18:53) > On 3/26/2020 11:06 AM, Anton Khirnov wrote: > > Quoting James Almer (2020-03-25 18:56:44) > >> Increase it in a linear way instead. > >> Helps reduce memory usage, especially on long, non fragmented output. > >> > >> Signed-off-by: James Almer <jamr...@gmail.com> > >> --- > >> An alternative is using av_fast_realloc(), in a similar fashion as > >> ff_add_index_entry(), which will keep the memory usage more closely tied > >> to the > >> amount of entries needed, but the amount of reallocations will be much > >> higher, > >> so i don't know if the tradeof is beneficial. > > > > Wait, wouldn't fast_realloc() do pretty much the same as your patch? Why > > should the number of reallocations be higher? > > I meant mimicking the behavior of ff_add_index_entry(), doing something like > > cluster = av_fast_realloc(trk->cluster, > &trk->cluster_capacity, > (trk->entry + 1) * sizeof(*trk->cluster)); > trk->cluster = cluster; > > Which will get rid of the fixed MOV_INDEX_CLUSTER_SIZE increase per > realloc, and resulting in more, smaller size increment reallocs (but of > course not after every single new entry, which is the entire point of > av_fast_realloc()).
Ah, got it. I'd say your patch is better, but it probably doesn't matter much. -- Anton Khirnov _______________________________________________ ffmpeg-devel mailing list ffmpeg-devel@ffmpeg.org https://ffmpeg.org/mailman/listinfo/ffmpeg-devel To unsubscribe, visit link above, or email ffmpeg-devel-requ...@ffmpeg.org with subject "unsubscribe".
