On Mon, 2017-01-16 at 20:03 -0600, Aaron Watry wrote: > erf(11.1) was completely wrong. > > Tested with clover/libclc on PITCAIRN and Beignet on haswell. > > This prevents regressions if/when the piglit ULP calculation is ever fixed.
I think these should use math.erfc and math.erf (1 and 2 respectively), but the numbers look OK now, and they weren't using it before. for patches 1 and 2 Reviewed-by: Jan Vesely <[email protected]> > > Signed-off-by: Aaron Watry <[email protected]> > --- > generated_tests/gen_cl_math_builtins.py | 2 +- > 1 file changed, 1 insertion(+), 1 deletion(-) > > diff --git a/generated_tests/gen_cl_math_builtins.py > b/generated_tests/gen_cl_math_builtins.py > index a14d97b..e9cfebf 100644 > --- a/generated_tests/gen_cl_math_builtins.py > +++ b/generated_tests/gen_cl_math_builtins.py > @@ -243,7 +243,7 @@ tests = { > 'arg_types' : [F, F], > 'function_type': 'ttt', > 'values' : [ > - [0.0, 0.950004, 0.990005, -0.994999475, > 0.27219439321703e-29, 1, -1], # Result > + [0.0, 0.950004, 0.990005, -0.994999475, 1, 1, > -1], # Result > [0.0, 1.960/sqrt(2.0), 2.576/sqrt(2.0), -2.807/sqrt(2.0), 11.1, > float("inf"), float("-inf")] # Arg0 > ], > 'tolerance' : 16
signature.asc
Description: This is a digitally signed message part
_______________________________________________ Piglit mailing list [email protected] https://lists.freedesktop.org/mailman/listinfo/piglit
