[FFmpeg-devel] [PATCH] mlpdec: Move quant_step_size out of substream context.
Wed Aug 13 00:38:11 CEST 2008
On Tue, Aug 12, 2008 at 7:15 PM, Michael Niedermayer <michaelni at gmx.at> wrote:
> On Tue, Aug 12, 2008 at 06:24:14PM -0300, Ramiro Polla wrote:
>> Since I don't have any specs for MLP, it's hard to know if this change
>> will introduce any regression. Probably not, since the current code
>> already has a similar limitation.
>> quant_step_size is currently defined for every channel for every
>> substream. I guess that all variables related to audio channels (not
>> matrix channels) should be the same for all substreams. So we could
>> save by having the channel information on a separate struct.
>> I have tested this with all my samples. Whenever two different streams
>> write quant_step_size to the same channel, it has always been the same
>> value. Furthermore, the code already has only enough space for each
>> channel (and not channel * substream) for sign_huff_offset, which is
>> updated whenever quant_step_size is read.
> Is this useful or needed for your encoder? if not then iam slightly
> against it.
Well, I'm trying to cleanup the context and have a few structs shared
between encoder and decoder. But I guess that can be left for later.
I don't think it'll introduce any regressions, but I'll wait for Ian's opinion.
More information about the ffmpeg-devel