Re: Patch: gcj optimizations

[Tom Tromey <tromey at redhat dot com>]

Bryce> Cool. It would be interesting to see how much of an improvement
Bryce> we get on the FFT benchmarks in either jbyte or scimark, since
Bryce> they rely on these functions.

I'd like to do that at some point, but I probably won't for a while.

Bryce> Does the __builtin_cos etc with -funsafe-math-optimizations
Bryce> handle the NaN and sign-of-argument cases specified in the docs
Bryce> correctly?

Good question.  I think this isn't guaranteed by gcc.

Bryce> But using __builtin_cos is still a big win since it will at
Bryce> least use a hardware optimized implementation (and bypass at
Bryce> least one level of call).

If the call isn't inlined it will call `cos', not `__builtin_cos'.
Same difference though -- we supply `cos' in libgcj.

Tom