[Shootout-list] New test in CVS: implicitode

[Sebastien Loisel]

I've added a test in CVS. I don't know if anyone reads the cvs commit
messages (I know I don't) so I'm letting you all know about it. I
won't be copying all CVS commit messages to the mailing list, but this
one is worth seeing.

---------------------------------

New test: implicitode.

This test will challenge floating point performance in the presence
of polymorphism, function parameters and other "realistic" library
situations. Hopefully the code is complex enough to defeat trivial
"unboxing" heuristics in languages where floating points are second
class citizen. This test also requires long doubles. If they are not
available, we will require that "comparable precision" be achieved.

As of now, the c++ and python files aren't the same test. The c++ test
is an emasculated version of the python one. I'm trying to keep the test
at 100 lines of c++. The python test is the test I really wanted to have.
They will be reconciled later.

---------------------------------

It's late and I won't try to see if it breaks everything (I hope
adding directories doesn't break your makefiles, Brent!) Otherwise,
I'll look tomorrow to add it to the sandbox. I'll also reconcile the
python and c++ version this weekend and possibly check in a validator
for the numerical results.

Brent, instead of doing an ascii comparison of program outputs against
a golden file, could we run a custom per-test validating program,
which would return from main() with an exit code of 0 if it's ok, and
nonzero if not? The reason is that if long doubles aren't available in
a target language, they must be emulated in which case exact binary
behavior is very tough to reproduce for floating point.

Cheers,

Sebastien Loisel