Date: Sun, 19 Mar 2017 18:06:08 +0100 From: Szabolcs Nagy <nsz@...t70.net> To: musl@...ts.openwall.com Subject: Re: [PATCH] fix underflow exception in fma and fmal * Rich Felker <dalias@...c.org> [2017-03-19 10:53:40 -0400]: > On Sun, Mar 19, 2017 at 03:39:53PM +0100, Szabolcs Nagy wrote: > > this case is for nearest rounding mode when the > > result is in the subnormal range, at this point the > > result is represented as hi,lo,scale but the final > > returned value is computed as scalbn(hi,scale) > > (the last bits of hi are adjusted if required for > > correct rounding), however scalbn fails to raise > > underflow if lo!=0 and all lost bits of hi are 0. > > > > the example is such a case: 0x1p-1022 - 0x1.000001p-1074 > > then hi=1-eps,lo=-0x1p-76,scale=-1022 or maybe with > > shifted scale and exponents, but in the end only one > > bit is lost from hi which is zero, alternatively i > > could do scalbn(lo,scale) too to raise underflow. > > That makes sense. I tend to prefer the scalbn(lo,scale) approach if > there aren't good reasons (performance?) against it, simply because > it's more self-documenting and less special-cased. But whichever you > like is fine. BTW we should probably check that scalbn raises inexact > in all cases it should; I'm not sure what it (especially asm versions) > does in cases where the scale is smaller than the min exponent. ok i can do it with scalbn. generic scalbn is correct and i386 should be correct too: it does a mul at the end which should raise the flags (i added tests for this now). i think we should fix the fma underflow before- vs after-rounding issue, the current code is only correct for after-rounding archs (x86, mips, sh) even with the fix, but doing it correctly is tricky.
Powered by blists - more mailing lists
Powered by Openwall GNU/*/Linux - Powered by OpenVZ