On Monday, January 23, 2017 9:13:38 AM PST Jason Ekstrand wrote: > On Mon, Jan 23, 2017 at 4:59 AM, Ilia Mirkin <imir...@alum.mit.edu> wrote: > > > On Mon, Jan 23, 2017 at 1:42 AM, Kenneth Graunke <kenn...@whitecape.org> > > wrote: > > > + *ymin = MIN2(ndc_gb_ymin, ndc_gb_ymax); > > > > Should this be limited on the negative end of the range? > > > > > + *ymax = MIN2(MAX2(ndc_gb_ymin, ndc_gb_ymax), 16383); > > > > And should this be a different value for gen6? Perhaps based on gb_size? > > > > I don't think so. It's a limit on some hardware for unknown reasons. That > said, Ken, it would be really nice to get an actual PRM citation because I > can't find it anywhere.
I'm going to drop this bonus clamp. It's attempting to implement a workaround (WaGuardbandSize) which is only documented in the BSpec (I believe it was added after the PRMs were published). I believe it was trying to fix some Windows test, but it wasn't terribly clear what the issue was. The documentation doesn't indicate whether it should be done in screen space or NDC space. I believe the Windows driver attempts to do it in screen space, but actually ends up clamping things to [-16385, 16385] because of what looks like an off-by-one bug...so it doesn't match the documentation either. Plus, their viewport transform code is a bit different than Jason's - it just applies the transform, but doesn't consider the framebuffer. It might be that Jason's transform is enough to fix the issue. Unclear. More importantly, implementing the clamping in screenspace like I intended to do (but botched) causes GL45-CTS.gtf21.GL2FixedTests.clip.clip to remain broken, and fixing that was kind of the point of all this. I'll send a v2 of this patch. --Ken
signature.asc
Description: This is a digitally signed message part.
_______________________________________________ mesa-dev mailing list mesa-dev@lists.freedesktop.org https://lists.freedesktop.org/mailman/listinfo/mesa-dev
