[chbot] Programming GPU's

[Charles Manning]

Then there are also SIMD vector processors (eg. ARM NEON and whatever the
Intel thing is called), and people writing C-like code for FPGAs.

Bottom line far all these techs is that they are not magic. Unless you
write the code to be friendly to the underlying architecture you will never
achieve optimal results. Yes, the compilers will try hard, but there is a
limit to what they can achieve.

In my experience with ARM NEON, we ended up having to write the tightest
loops in intrinsic functions (essentially C-like assembler) to get the
results we needed.

The same goes for optimal GPU code etc.



On Sat, Jun 13, 2020 at 9:06 AM Bevin Brett <bevin_brett at hotmail.com> wrote:

> Haven't done any directly myself, but have followed discussions about them
> by the Intel h/w people over the years
>
> There is a variety of tools available to make it a lot simpler, and
> competing "standards".
>
> Yes, memory traffic has to be managed very carefully - this was my
> specialty for my last 2 years at Intel.  Whether a GPU is going to be
> faster or slower than a multi-core or single-core cpu will depend on this,
> and on whether the code is really capable of being massively parallel
>
> /Bevin
> _______________________________________________
> Chchrobotics mailing list Chchrobotics at lists.ourshack.com
> https://lists.ourshack.com/mailman/listinfo/chchrobotics
> Mail Archives: http://lists.ourshack.com/pipermail/chchrobotics/
> Meetings usually 3rd Monday each month. See http://kiwibots.org for
> venue, directions and dates.
> When replying, please edit your Subject line to reflect new subjects.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.ourshack.com/pipermail/chchrobotics/attachments/20200613/797442dd/attachment-0001.html>