[Date Prev][Date Next] [Thread Prev][Thread Next] [Date Index] [Thread Index]

Re: anyone still here?



At Wed, 3 Mar 2004 17:44:52 -0600,
Ryan Underwood wrote:
> This port has been dead for a while.  I'm curious if any of the issues
> regarding SH3/SH4 binary compatibility (FPU emulation, delay slot) have
> had solutions proposed or issued since activity last was going on.

The issue we discussed between SH3 and SH4 are: FPU emulation and
library compatibility.

> It seems that we can build binaries that work on both SH3 and SH4
> by reducing the performance on SH4 slightly and requiring SH3 to emulate
> a FPU by the kernel.  Right?

FPU emulation issue is that SH3 lacks hardware FPU, so we need to add
software FPU emulation code within kernel.  At least I couldn't find
such code in 2.6.3 kernel.  Is it true?

> What else is holding things back?  The
> existing problems of SH3<->SH4 do not seem to be insurmountable,

Please see the below thread which discussed about ABI incompatibility
problem:

	http://lists.debian.org/debian-superh/2001/debian-superh-200109/msg00011.html

Niibe-san, who was ported SH kernel, said that it was hard to define
another SH3 ABI for linux that was compatibile to SH4 at that time.  I
guess they still don't change their ABI.  I don't know SH ABI at all,
but I wonder why they didn't define or switch their own ABI or SH4
ABI.

The another thing which complicates the current situation is that SH-3
is used widely than SH-4 more and more.  Should we make both SH-3 and
SH-4?  FPU issue is not much hard to solve, but ABI issue needs a lot
of considerations.  I would like to know the opinion from SH
developers.

The problem to setup SH buildd is: almost debian SH developers don't
have SH hardware which have a lot of memory (ex: 128MB or over).  I
heard from ISHIKAWA-san and YAEGASHI-san that small memory machine
like DreamCast (32MB RAM) does memory swapping frequently.  This means
that we need to prepare large memory SH machine with IDE disk support
or PCI support (in order to use SCSI).  Such evaluation board is much
expensive for personal users.

> and it
> seems we can ignore big endian because nobody who has such a machine has
> ever spoken up.

Agreed.

Regards,
-- gotom






Reply to: