> Forgot to write this: currently there is
> '--enable-algebra-optimization' option to configure. As explained in
> INSTALL, if you give:
>
> --enable-algebra-optimization="((speed 3) (safety 0))"
>
> you get fastest code (but unsafe).
Thanks. I knew this, but have not used it. I have bad experiences with
high optimization with the Aldor compiler. I probably don't tell you
anything new, but there were at least 3 errors.
(1) the compiler optimizes too much and produces garbage from correct
sources
(2) the library sources were buggy and just getting a
"Segmentation fault" is not very helpful to find the
source of the problem.
(3) I write buggy user library extensions.
With all-or-noting and no concrete error message a distinction between
(2) and (3) would be hard to make. But this is certainly not a big
problem for me, because you are just speaking about the library code
that is inside the FriCAS repository.
Distinction between (1) and (2) is probably more delicate. But don't we
need a bigger set of unit tests for these low-level domains if they are
going to be super-optimized?
> Currently this is all-or-nothing choice, that is applies to whole
> algebra. I probably will rework it, in particular I think that low
> level domains like U32Vector, U32VectorPolynomialOperations and bunch
> of similar domain should be compiled at high optimization settings,
> while rest of algebra should use current settings. This should give
> us most gain from optimization with only small decrease in safety.
As far as I remember, you have introduced the U32... domains. How can
the optimize level of those influence the spead of the finite field
factorizer?
Nevertheless, don't you think that it would be a good idea to allow
including calls to FLINT in the FriCAS library? Unfortunately, I have no
idea how to do such an interface right, but it would be nice to have one.
Ralf
--
You received this message because you are subscribed to the Google Groups
"FriCAS - computer algebra system" group.
To unsubscribe from this group and stop receiving emails from it, send an email
to [email protected].
To view this discussion on the web visit
https://groups.google.com/d/msgid/fricas-devel/a3dab5ec-f652-c1a0-3298-0e81fe07e332%40hemmecke.org.
