mod2pi broken with USE_SYSTEM_LIBM

[nolta]

Currently mod2pi calls __ieee754_rem_pio2, which is an openlibm specific function. This causes make test to fail if julia is compiled w/ USE_SYSTEM_LIBM=1.

Not really a big deal right now, but w/ ARM & PowerPC ports on the horizon, it would be nice to fix, since openlibm is likely to remain Intel-only.

[ViralBShah]

We can bring the arm and powerpc bits into openlibm from FreeBSD msun at the right time.

[JeffBezanson]

We should try to make USE_SYSTEM_LIBM work. Maybe __ieee754_rem_pio2 should be copied into openspecfun or just into the runtime system to ensure availability.

[ViralBShah]

Let's just have a Julia implementation of it.

[andrioni]

This could also be a good time to add a BigFloat version of it.

[ViralBShah]

I am going to copy this into openspecfun for now.

[nalimilan]

This makes me think it would be nice to list the minimum openspecfun version required by Julia in the docs (and bump the version in openspecfun). A packager not aware of this issue wouldn't have noticed this requirement.

[ViralBShah]

I did bump the openspecfun and openlibm versions.

[ViralBShah]

Cc: @staticfloat @sebastien-villemot

[tkelman]

Interesting, now that the issue with __ieee754_rem_pio2 has been resolved, there's an O(ε) mismatch between system libm's (at least for Ubuntu 14.04) gamma(Float64[1:25]) and gamma(1:25) which does a lookup into the factorial table for n ≤ 20.

[jiahao]

Perhaps they were computed in a different rounding mode?

[nalimilan]

Should I make releases for openlibm and openspecfun 0.4? I need them to build packages for Julia RC1.

[ViralBShah]

I think we should fix the SONAME stuff ideally before we do that. What do you think?

[nalimilan]

Right. Should be pretty easy.