[Libre-soc-dev] SVP64 parallel map-reduce idea

[Jacob Lifshay]

On Sun, Jun 13, 2021, 10:38 Luke Kenneth Casson Leighton <lkcl at lkcl.net>
wrote:

> ah then we need a prefix sum algorithm.  just a basic one would do.  i
> see this is an active area of research, Raph Levien created a
> memcpy-speed vulkan prefix-sum implementation for example.
>
> the answer when running the example came out to 377.  adding up the
> numbers manually they came out to 307.  i therefore assumed
> (incorrectly) it was doing a prefix sum.
>

The example actually runs the reduce twice, once with a non-trivial
predicate, followed by once with an all-true predicate. you probably
confused the two outputs with each other.

Jacob