[lkml]   [1997]   [Nov]   [6]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
Messages in this thread
SubjectRe: Cyrix patch : Proposal?
Andre Derrick Balsa wrote:
> Hello,
> Teunis, I would *not* want to see the (fixed) 2.1.39 Cyrix patch go into
> the kernel source, as it stands now.
> I have a simple technical reason, and a more "philosophical" reason for
> this.
> Technical: no performance gains
> ==========
> The main performance improvement that can be had by adapting Linux
> operation to a 6x86 processor is the correct setup of the ARR registers.
> This can bring a 30-50% CPU/video memory bandwidth improvement.
> The 2.1.39 patch does *not* set the ARRs, because this would imply the
> detection of the linear frame buffer address, which changes from system
> to system.
> The other performance features bring a *measured* performance
> improvement of less than 0.5%. This is insignificant IMHO.
> Philosophical: kernel pollution vs. user-space utility
> ==============
> Everything (nearly) the patch does can be done in user-space using the
> set6x86 GPLed utility written by Koen Gadeyne. You don't have to
> recompile your kernel everytime a 6x86 feature needs to be tested or
> changed. And set6x86 allows setting the ARRs.
> The only thing set6x86 cannot do is implement the 6x86 VSPM feature,
> which only works in some 6x86 revisions, has been dropped in the new
> 6x86MX and does not provide any measurable performance improvement.

Another thing set6x86 cannot do is make /proc/cpuinfo show the correct
information instead of just reporting an unknown 486. This has been by
biggest peeve with cyrix processors (cpuid is disabled on bootup).
Granted this is mostly cosmetic, but it still is the right thing to do.

If the patch were stripped down to just do the cpu detection, and leave
the configuration up to set6x86 (except for enabling cpuid), would you
object to that?


Brian Gerst

 \ /
  Last update: 2005-03-22 13:40    [W:0.026 / U:6.664 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site