https://gcc.gnu.org/bugzilla/show_bug.cgi?id=61319
--- Comment #2 from Uroš Bizjak <ubizjak at gmail dot com> --- (In reply to Andrew Pinski from comment #1) > No they are supposed to generate -nan. If it does not then there is a bug > somewhere else. According to [1], "The sign bit does not matter". As I understand this sentence, the "-" is optional, so the patch from Comment #0 looks correct to me. This test also fails on Centos 5 for the same reason. [1] http://en.wikipedia.org/wiki/NaN#Encoding
