[Libre-soc-dev] daily kan-ban update 20aug2020

[Luke Kenneth Casson Leighton]

On Thu, Aug 20, 2020 at 7:44 PM Alain D D Williams <addw at phcomp.co.uk> wrote:
>
> On Thu, Aug 20, 2020 at 07:35:00PM +0100, Luke Kenneth Casson Leighton wrote:
> > On Thu, Aug 20, 2020 at 7:25 PM Jacob Lifshay <programmerjake at gmail.com> wrote:
> >
> > > my assumption has been video processing unit, since we're adding
> > > instructions to accelerate video decode (and maybe encode too).
> >
> > yep VPU.  we're including audio in that as well, because lauri pointed
> > out that audio processing can be really quite CPU-intensive.
>
> So how does our VPU differ from a GPU ?

there is some crossover.

for example: both a GPU and a VPU might have certain types of pixel
conversion instructions (RGB24 to RGB32 conversion instruction,
YUV2RGB conversion instruction, and so on).

what a GPU will *not* typically have is: "a special instruction
exclusively dedicated to DCT (discrete cosine transformation) in a
format exclusively targetted at video processing and not at all really
suited to 3D graphics processing".

if you  *only* have GPU tasks, you absolutely do not bother to design
or put in hardware instructions that accelerate the decoding and
encoding of H.265, or VP9.

you "normally" leave that to a *dedicated* VPU.  or a hard macro that
is incapable of anything other than actual H.265.  or actual MP4.

we are not doing that.

we are *specifically* augmenting a *general-purpose* ISA so that it
has the capacity to perform reasonable video decoding *without*
requiring a completely separate VPU hard macro.

l.