scalbln (original) (raw)

This is the mail archive of the gcc-patches@gcc.gnu.orgmailing list for the GCC project.

This patch adds folding of ldexp, scalbn and scalbln. It always transforms the following cases for all three builtins:

f(+-0, y) -> +-0 f(+-Inf, y) -> +-Inf f(NaN, y) -> NaN f(x, 0) -> x

When both arguments are constant, then I evaluate ldexp using real.c:real_ldexp(). If the radix is 2, then I also do that for scalbn/scalbln. I believe at the moment, all supported targets use a radix of 2, but I check for it anyway just in case.

Tested on sparc-sun-solaris2.10, no regressions and the new cases all pass.

Okay for mainline?