Hi,
On 12.05.2016 20:11, Matt Turner wrote:
On Thu, May 12, 2016 at 4:49 AM, Eero Tamminen
<eero.t.tammi...@intel.com> wrote:
Multiply is faster instruction than sin/cos, and depending on what else the
shader does, good scheduler may be able to hide latency of additional
multiply, so I don't think it's quite that bad. :-)
I think you misunderstood what was happening in the program that was
hurt by this. It wasn't scheduling that was able to hide the
additional latency. Every sin/cos in the program was multiplied by a
constant, so Ken moved the workaround to NIR which allowed the two
constants to be combined, removing the additional multiply.
My comment was for the shaders where that is NOT the case.
- Eero
_______________________________________________
mesa-dev mailing list
mesa-dev@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/mesa-dev
