Subject: Re: CPU instruction set optimization
To: Toshiyasu Morita <tm@netcom.com>
From: David Brownlee <abs@anim.dreamworks.com>
List: port-sun3
Date: 05/05/1998 17:02:35
ecgs (which we are in the rpocess of moving to), has at least
MULL, ULW, DIVW, & LSHIFTRT cost decisions based on CPU type.
There are also quite a few decisions made in m68k.md based on
TARGET_680[46]0. While most of these are related to floating
point, quite a few are not.
However, it does appear that it treats a 68030 exactly as a 68020,
so there is no benefit at all in compiling -m68030 vs -m68020
(Nuking my original suggestion and making this irrelevant to
port-sun :).
This does leave open the option of compiling kernels/userland with
680[46]0... (but it wont affect any sun3 or sun3x users :)
David/absolute
-=- "Promise carved in stone, deeper than the sea" -=-
On Tue, 5 May 1998, Toshiyasu Morita wrote:
> If you look at the m68k machine description then you will see that the only
> difference between 68020/68030/68040/68060 is gcc avoids slow floating-point
> instructions on the 68040 and 68060 which are implemented as software traps.
>
> Please explain why the kernel should be optimized for fast floating-point
> computations.
>
> Toshi
>