From mboxrd@z Thu Jan 1 00:00:00 1970 From: "John David Anglin" To: dave@hiauly1.hia.nrc.ca (John David Anglin) Cc: gcc-patches@gcc.gnu.org Subject: Re: PATCH: Upgrade floating comparisons on PA to support unordered operands Date: Mon, 16 Oct 2000 13:21:00 -0000 Message-id: <200010162021.QAA09573@hiauly1.hia.nrc.ca> References: X-SW-Source: 2000-10/msg00467.html > In testing this patch, I examined two other ieee tests which fail on the > PA port. A fix for the hugeval tests under 10.20 will be sent separately. > The mzero2 tests still fail because the insns negdf2 and negsf2 are > implemented as a subtract from +0, since no fneg instruction is available > on the PA. The problem is that, under the ieee standard, (+0.) - (+0.) = +0. Actually, PA2.0 machines do have fneg. However, it behaves identically to fsub: fneg (+0.) yields +0. Thus, it works as a subtract from +0. rather than xor'ing the sign bit. Look's like a little bug in the HP floating point implementation. Dave -- J. David Anglin dave.anglin@nrc.ca National Research Council of Canada (613) 990-0752 (FAX: 952-6605)