On Tue, 10 Mar 1998 05:01:26 -0800 (PST),
SamMaEl <rimsky@teleport.com> writes:
> On Tue, 10 Mar 1998 emacia@solucions.ad wrote:
> 
> > This week end I have d/l the full 1.3 release.
> > I have installed the files on a Performa 630 (no fpu) with 20Ram and 120Mb
> > external IBM scsi disk.
> 
> 	First of all, is your LC040 a buggy one? Have you used it with
> Software FPU, or various benchmarks which would test FPU? The Floating
> Point Emulation isn't quite complete atm, but it definitely won't work on
> a broken FPU. 

The broken masks of LC040s don't do F-line traps correctly.  There's
no workaround for this without sacrificing some things for the other
chips (i.e. 020, 030, 040, and non-broken LC040).  That's why most of
us usually consider the workaround for the broken LC040 not worth
pursuing (I haven't heard anyone state on the contrary, so far).

Having said that, it is possible that his LC040 is in fact broken, but
the probability is not so high.  Afterall, he has successfully booted
into multi-user mode, and you'll be surprised just how many of our
standard commands got some form of FPU instructions (usually fmovem).
The broken f-line trap bug would hit him well before launching X.

> 	Someone suggested recompiling a math library, with patches for a
> mchine with no FPU...

The "patch" must be to the Makefile, not to the source itself.  And
you might want to add -msoft-float to CFLAGS so that the resulting
math library won't use *any* FPU instruction at all.  No, it's not
going to slow things on your machine because any floating point
operation is done by software alone, anyway. (But yes, it will make
things much much slower on a box with FPU...  That's why any binary
distribution or snapshot isn't built that way)

> I'm also reading through some things to educate myself on the FPU
> and the SANE stuff...

Unless you are trying to familiarize yourself with general floating
point issues or planning to dig it up from your ROM and actually call
it from FPE, I think the SANE library has little relevance here.

Although I'm sure there are mathematical bugs in FPE, the biggest
barrier for LC040 users is in FPE's operand handling.  That is, FPE's
ability to fetch and store operands.  Apparently it doesn't work as
well on the LC040 as it does on the 030.  It is my understanding that
FPE somehow sometimes changes contents of address registers in such an
unexpected way that a pointer dereferencing in the user program
results in a seg fault.  So, basically it's more an "M68000 assembly
programming model" kind of issue than a "numerical computing" issue.

Good luck,

Ken