On Sun, Nov 17, 2002 at 12:51:52PM +0900, Toru Nishimura wrote:

> We have to be serious about implementing 64bit FP insn emulation.
> It should be a full scale implementation even for cores equipped with
> no FP hardware.  Note that no FP hardware is an extreme case of
> FP insn emulation and as long as emulation is smart enough
> there is no necessity to sprinkle SOFTFLOAT/NOFPU.

I keep meaning to try out the softfloat library (in libkern).  64-bit
FP will be easy with this, but I don't know the performance differences
between it and our current FP emulation for 32-bit math.

And yes, the SOFTFLOAT/NOFPU bit needs to be cleaned up at least.  For
example, I'm not exactly sure how to build a kernel right now that
supports say both 5Kc (without FPU) and 20Kc (with FPU) such that the
FPU is used when it's available.  Not that I've looked that closely.

> I once toyed an idea to approach Algorithmics but the company is...

Now that Algorithmics is part of MIPS I dunno if they'd be more or less
interested in releasing their FP emulation under a BSD-style licence...

Simon.
--
Simon Burge                                   <simonb@wasabisystems.com>
NetBSD Development, Support and Service:   http://www.wasabisystems.com/