Date: Wed, 29 Nov 2017 06:12:02 +0000 From: Taylor R Campbell <campbell+netbsd-source-change...@mumble.net> Message-ID: <20171129061642.e8dcb60...@jupiter.mumble.net>
| That's pretty interesting! That is what I thought, it was certainly not what I expected. | Can you reproduce it in a small isolated program like the attached | one? I will try later today, but I think I'll try something simpler first. Since we know what bit pattern is in the variables, it should be possible to simply start with values containing those patterns, and just compare, & subtract, and see what happens. The strtod() and setlocale() are most likely just window dressing, and would make determining what is going on more difficult, unless they do turn out to be affecting things. I doubt there is anything very subtle (as in which addresses happen to be used, or whatever) - this test has undergone significant alterations, and it keeps behaving exactly the same way, every time, on i386 (and what I also wasn't sure would happen, on "real" i386 (or as real as Xen approximates, which for floating point, should be quite good) and on qemu (we should see the results from that sometime tomorrow.) kre