On Thu, 2 May 2019 at 09:31, Henner Zeller <h.zeller@xxxxxxx> wrote:
On Thu, 2 May 2019 at 09:25, Seth Hillbrand <seth@xxxxxxxxxxxxx>
wrote:
>
> Hmmm... cache will be local to each core, so the we'll probably get
> speed up by dividing the image into quadrants based on the number of
> threads and having a morton sweep per quadrant.
so on my 16 core, 32 'thread' Ryzen Threadripper, I only see about
1-2% slower execution, so it is probably not a big deal even with a
lot of competing cores.
Having said that, I have some ideas for some further optimizations,
but that would go more into the guts of it (e.g. dividing memory areas
by cores etc.) that I didn't want to start it now for a (hopefully)
5.1 acceptable patch.
(uhm, also we are talking in the wrong email-thread here ... This
should've gone to the
"PATCH: Raytracing - a more pleasing way sequencing blocks to render
?"-thread)
