Am Montag, den 16.07.2018, 15:10 -0400 schrieb Ilia Mirkin: > > From: Gert Wollny <gert.wol...@collabora.com> [...] > > > > The added function evaluates the array index according to the > > function > > > > idx = floor(ai + 0.5); > > [...] > > Why not use RNDNE? Then you don't have to add the 0.5...
When I starten fixing these failures I used a texture offset "0.5" and a sampler state to do the correct rounding - there "round to next even" was not correct, but it actually does some n.6 rounding (i.e. not to integer), but with cubemap arrays this approach is no longer an option so I translated this to shader code to make it similar. Because of the problems with the samler state rounding I assumed that RNDNE is not what the tests expect for the corner case "1.5", but I didn't really check. I'll recheck whether changing this keeps the tests passing and resubmit a new patch set if this is the case, thanks, Gert _______________________________________________ mesa-dev mailing list mesa-dev@lists.freedesktop.org https://lists.freedesktop.org/mailman/listinfo/mesa-dev
