On Wednesday, March 11, 2015 10:29:02 AM Ian Romanick wrote: > On 03/09/2015 01:58 AM, Kenneth Graunke wrote: > > We can't safely call nir_optimize() with register present, since several > > passes called in the loop can't handle registers, and will fail asserts. > > Doesn't that mean this patch should go before patch 2?
Probably. I only provoked the bug by using this code path for vertex shaders, which aren't enabled until the end of the series. It may be possible to provoke it via fragment shaders (and piglit simply doesn't), but if so, it would have already been broken - we already have one nir_optimize() before lower_to_registers. So I don't feel too strongly about it.
signature.asc
Description: This is a digitally signed message part.
_______________________________________________ mesa-dev mailing list mesa-dev@lists.freedesktop.org http://lists.freedesktop.org/mailman/listinfo/mesa-dev