[Piglit] [PATCH v3 2/3] arb_shader_precision: add framework for calculating tolerances for complex functions

Ilia Mirkin imirkin at alum.mit.edu
Mon Mar 9 13:02:46 PDT 2015

On Mon, Mar 9, 2015 at 3:51 PM, Micah Fedke <micah.fedke at collabora.co.uk> wrote:
> Still no idea what to do with a*b+c in complex functions - don't really want
> to evaluate all the permutations if at all possible.

Yeah, that stinks, but the reality is that _something_ needs to be
done. A bunch of tests on nvc0 will fail otherwise since we merge
a*b+c into fma opportunistically. This seems to be perfectly legal per
the spec.

I briefly glanced at your changes, and outside of some trivialities
they seem mostly fine. I'd really prefer it if you changed

+def _analyze_ref_fn(fn, args):
+    interval_results = fn(args)

to fn(*args). Then you avoid all the args[0]/etc insanity since the
arguments just come out as you'd expect.

But the a*b+c disaster needs some sort of resolution. I'm still at a
bit of a loss.


More information about the Piglit mailing list