On 9/9/18 9:40 PM, Ilia Mirkin wrote:
On Sun, Sep 9, 2018 at 3:19 PM, Axel Davy <davyax...@gmail.com> wrote:
Tests showed Intel on windows does always clamp
RCP, RSQ and LOG (thus preventing inf/nan generation),
for all shader versions (some vendor behaviours vary
with shader versions).
By the way, this happens because on Intel, the ALU is put into a
special mode where it just doesn't generate NaN's at all under any
conditions. I don't think that other vendors operate this way.
I've found the code source of my tests, and completed them to have a
better picture.
The conclusion is that the clamping can be safely removed all the time
when mul_zero_wins.
I produced a commit in that purpose:
https://github.com/iXit/Mesa-3D/commit/ed82c87da2799a40d8f7b87f8ff99d6f20a9f601
I'll send it to mesa-dev when our testers can check there is no regression.
In the commit I detail the test results.
I still want to get "st/nine: Clamp RCP when 0*inf!=0" merged,
especially for stable.
Yours,
Axel
_______________________________________________
mesa-dev mailing list
mesa-dev@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/mesa-dev
