Serious bug

[Brad Lucier]

It's a problem in that you're disagreeing with how egcs sets up the
floating-point arithmetic on the *87.  It isn't "wrong", it's just different.

You have some control over this in linux.  Look at the options in the
include file fpu_control.h.  If you want to do all computations rounded
to double-precision, then you could say, for example,

#define FPU_CW \
(_FPU_MASK_IM+_FPU_MASK_ZM+_FPU_MASK_OM+_FPU_MASK_UM+_FPU_MASK_PM+ \
_FPU_MASK_DM+_FPU_DOUBLE+_FPU_RC_NEAREST)

  __setfpucw (FPU_CW);

somewhere in the setup of your program.

Brad Lucier    lucier@math.purdue.edu