Subject: Re: floating-point bugs in kernel, gcc?
To: None <port-pmax@NetBSD.ORG>
From: Jonathan Stone <jonathan@DSG.Stanford.EDU>
List: port-pmax
Date: 06/22/1995 15:42:39
>I suspect that MIPS floating point kernel support must be especially
>tricky, because a Sony 3860 (a MIPSEB machine running pretty vanilla
>4.3 BSD) would reliably panic if a user ran a C program that contained
>an atan(float) call.  Took a kernel patch to make that go away.
>Probably a completely different problem, but perhaps indicative...

Perhaps not entirely different.

Mips R[2,3]010 FPUs don't implement denormalized numbers in hardware.
FP operations involving a denorm trap, and the FP op is emulated
in software.

I'm claiming there's a bug in the kernel denorm handling -- i.e., in
src/sys/arch/pmax/pmax/fp.S; and that *maybe* that bug is contributing
to other FP anomalies on NetBSD/pmax.

I just don't have time to fix it :(.