[PPL-devel] patch for merging graphite branch (before tuplification)
Mark Mitchell
mark at codesourcery.com
Fri Aug 8 17:35:44 CEST 2008
Roberto Bagnara wrote:
> I think a little clarification is necessary. It is not strictly true
> that the PPL's interface is such that there is a unique correct output
> for any input.
I will freely admit to not following all the mathematics here, but this
does raise a question for Sebastian.
Sebastian, will the behavior of Graphite be deterministic, in that for a
given compiler and a given input program, we can be assured of the same
output? (I would assume so, but I would like to check.)
Also, given Roberto's comments about floating-point and such, are we
sure that we will get the same results on multiple hosts, if generating
code for the same target? It's an important design goal that this be
the case, and I'm concerned that differences in floating-point (i.e.,
default rounding modes, precision of long double, etc.) could make a
difference in behavior. I'm not sufficiently knowledgeable about
floating-point architectures to know whether explicitly specifying
rounding modes and such is sufficient to get the same answer on all
hosts, or not.
Thanks,
--
Mark Mitchell
CodeSourcery
mark at codesourcery.com
(650) 331-3385 x713
More information about the PPL-devel
mailing list