On 01/06/2012 01:26 PM, Ian Romanick wrote: > On 01/06/2012 09:04 AM, Dave Airlie wrote: >> Hi guys, >> >> Just a quick note, I've just spent a week or so trying to see where >> gallium and softpipe were w.r.t GL3.0 support. >> >> I've pushed a branch to my repo called softpipe-gl3. It contains >> patches in various state of usefulness but it brings the piglit >> results to 220 failures in 7623 tests, which isn't bad. >> >> Outstanding known problems (stuff I've dug into). >> >> smooth interpolation is broken in softpipe, worth about 70-100 fixes >> at a quick guess. >> >> integer abs - we have no TGSI representation for this, should we lower >> it to something? > > Or just generate some TGSI instructions to implement it. You should > be able to fake it with a CMP-like instruction. I think that's how > i915 does it in hardware.
Depends on whether there's any hardware with a native integer abs instruciton. If there is, we should just add a new IABS instruction to TGSI and let drivers implement it how they want. Otherwise, your suggestion should work. > >> integer SSG (set sign) - no TGSI for this, lower it? > > Where is SSG being generated? I thought ir_to_mesa was the only thing > that generated it, and Gallium shouldn't hit that path. glsl_to_tgsi still generates the TGSI equivalent; that part hasn't been changed from ir_to_mesa. Bryan _______________________________________________ mesa-dev mailing list mesa-dev@lists.freedesktop.org http://lists.freedesktop.org/mailman/listinfo/mesa-dev