a/mklinux

Date: Tue, 29 Jun 1999 17:56:22 -0400
To: "Elizabeth O. Coolbaugh" 
From: Larry Kollar 
Subject: More MkLinux news

Hi Liz,

Here's some more MkLinux grist for the distributions mill.

	Larry

===

MkLinux

We should mention that while Apple has stopped further development on MkLinux, the company continues to host the MkLinux mailing lists.

Moving closer to a new release. Fred Bacon has built a first cut at a glibc2-based release. It went pretty smoothly, being built on previous efforts by the LinuxPPC folks. A pre-release FTP-based distribution is already available at the MkLinux FTP site. Once the bugs are shaken out, a CD-based release will be made available.

Big math performance enhancements with libmoto (applies to LinuxPPC and Mac OS users too!). An interesting thread in mklinux-setup discussed poor math performance with both MkLinux and LinuxPPC. Davide Caironi was the first to suggest Motorola's l ibmoto library as a possible fix.

Heinz Nabielek, the original poster, seemed very pleased with the results...

libmoto really does help:
cc Tsqrt.c -O2 -lmoto -lm -o Tsqrt

601
  66MHz 7100,           MkLinux 19700 ns without libmoto
  66MHz 7100,           MkLinux  1535 ns with libmoto, factor 13 faster

603e
 200MHz Performa 6400, linuxPPC  5100 ns without libmoto
 200MHz Performa 6400, linuxPPC   775 ns with libmoto, factor 6.5 faster

Also upgrading to AIX 4.3.2
 AIX POWER2                       430 ns before (don't know which)
 AIX POWER2                       251 ns AIX 4.3.2 option -O4, factor
1.7

So it's well worth the effort to link libmoto into math-intensive applications. Other messages in the thread indicate that glibc2 also provides better performance, but libmoto still helps there too. One drawback: libmoto is not open-source.

The CVS repository is now at mklinux.org. See the MkLinux.org news for further details. The same page also provides a script to automate the changes necessary.

David Gatwood has cracked the NuBus floppy problem. He writes, "After months of pounding, I just noticed that the DMA direction flag was being set backwards, hence the floppy support was trying to write an empty buffer instead of reading data into that buffer." A new development kernel with this support should be available by now.