Skip to contentSkip to conversation

Algorithm for CPU_X86

DISCUSSION J.A. Magallon 3 participants 3 messages Jun 4, 2002

lore
Share & tools
lore

J.A. Magallon

author
#0 lore raw

Hi all...

Following with the cpu selection changes, setting the flags controlled by the various CPU_X86_xxxx can be a real mess.

But, I have ralized that those CPU_X86 flags can be logically split in two groups: features (TSC,MMX,3DNOW) and bugfixes (PPRO_FENCE) (or perhaps more, it is just a first thought...).

So the algorithm can be easy (i think..)
- Enable all feature flags (say, MMX and 3DNOW)
- Disable all bugfix flags (FENCE)
- For each CPU
if it does not have the feature, kill it
(if VENDOR_INTEL set 3DNOW=n)
if this-cpu-kernel could run on buggy-cpu, enable fix (if GENERIC_586 set FENCE=y)

Right ?

Signature · 4 lines
J.A. Magallon                           #  Let the source be with you...        
mailto:[email protected]
Mandrake Linux release 8.3 (Cooker) for i586
Linux werewolf 2.4.19-pre9-jam1 #1 SMP lun jun 3 19:59:12 CEST 2002 i686
#1 lore raw
Quoted 11 lines · J.A. Magallon

Hi all...
So the algorithm can be easy (i think..)
- Enable all feature flags (say, MMX and 3DNOW)
- Disable all bugfix flags (FENCE)
- For each CPU
if it does not have the feature, kill it
(if VENDOR_INTEL set 3DNOW=n)
if this-cpu-kernel could run on buggy-cpu, enable fix (if GENERIC_586 set FENCE=y)

Right ?

What if the next wizz-bang processor as a great new feature ? You'd have to disable it for *all* the previous CPUs...

ghoz

#2 lore raw
In reply to J.A. Magallon
Followup to:  <[email protected]>
By author:    "J.A. Magallon" <[email protected]>
In newsgroup: linux.dev.kernel
Quoted 8 lines

Following with the cpu selection changes, setting the flags controlled by the various CPU_X86_xxxx can be a real mess.

But, I have ralized that those CPU_X86 flags can be logically split in two groups: features (TSC,MMX,3DNOW) and bugfixes (PPRO_FENCE) (or perhaps more, it is just a first thought...).

Don't forget optimizations. It's a big difference between "optimize for" and "require" -- consider gcc, which have -mcpu= and -march= respectively for the two.

-hpa

Signature · 3 lines
<[email protected]> at work, <[email protected]> in private!
"Unix gives you enough rope to shoot yourself in the foot."
http://www.zytor.com/~hpa/puzzle.txt	<[email protected]>