https://gcc.gnu.org/bugzilla/show_bug.cgi?id=98612
--- Comment #7 from Hongtao.liu <crazylht at gmail dot com> --- (In reply to Guillaume Piolat from comment #5) > My reasoning for reporting (while it is minor) is the following: > - it might not be a conscious choice from GCC developers > - this is the only intrinsics I've found in mmx/sse/sse2 for which GCC has > different semantics than ICC and clang. > - clang wiki define intel intrinsics as the "portable API" between compilers > - people will copy/paste intel intrinsics code between compilers. So fixing > it is breaking, but leaving it is also breaking > - in any case it is surprising I asked my colleagues within intel to revise the descriptions in the intrinsics guide to make it more explicit about NAN operands. I'll fix this issue after the update of intrinsics guide.
