PDA

View Full Version : .i586 packages: not a big deal



RahulSundaram
10th March 2009, 11:45 AM
Will Woods, one of Fedora QA leaders has posted an explanation on what the switch to i586 actually means. In short, it doesn't change much but read

http://qa-rockstar.livejournal.com/7478.html

sideways
10th March 2009, 12:21 PM
That's quite funny after all the fuss that the i586 kernel install bug caused in FC6.

So the real answer back then was, "leave i586 installed it's faster than i686" :D

(Now you tell us :rolleyes: :) )

sideways
10th March 2009, 12:32 PM
atom n270 has pae but no nx flag, does this mean we're gonna get a million netbook users complaining about a i586 kernel install in F11?

RahulSundaram
10th March 2009, 12:34 PM
That's quite funny after all the fuss that the i586 kernel install bug caused in FC6.

So the real answer back then was, "leave i586 installed it's faster than i686" :D

(Now you tell us :rolleyes: :) )

The bug was problematic not because of performance but header mismatch. This change in rawhide doesn't have much to do with performance either but reducing the number of different kernel variants.

RahulSundaram
10th March 2009, 12:35 PM
atom n270 has pae but no nx flag, does this mean we're gonna get a million netbook users complaining about a i586 kernel install in F11?

I doubt many people will care but when some do, this post offers an explanation on why you needn't bother.

sideways
10th March 2009, 01:07 PM
Oh dear. I think you're gonna have to ensure useof cmov in netbooks with n270 atoms (which means i686). Linus' test code (http://ondioline.org/mail/cmov-a-bad-idea-on-out-of-order-cpus) is nearly twice as slow on the "in-order" atom n270 cpu if you don't use cmov


/* t.c */
/* How many iterations? */
#define ITERATIONS (100000000)

/* Which bit of the counter to test? */
#define BIT 1

#ifdef CMOV

#define choose(i, a, b) ({ \
unsigned long result; \
asm("testl %1,%2 ; cmovne %3,%0" \
:"=r" (result) \
:"i" (BIT), \
"g" (i), \
"rm" (a), \
"0" (b)); \
result; })

#else

#define choose(i, a, b) ({ \
unsigned long result; \
asm("testl %1,%2 ; je 1f ; mov %3,%0\n1:" \
:"=r" (result) \
:"i" (BIT), \
"g" (i), \
"g" (a), \
"0" (b)); \
result; })

#endif

#include <stdio.h>

int main(int argc, char **argv)
{
int i;
unsigned long sum = 0;

for (i = 0; i < ITERATIONS; i++) {
unsigned long a = 5, b = 7;
sum += choose(i, a, b);
}
printf("%lu\n", sum);
return 0;
}


[user1@localhost ~]$ cat /proc/cpuinfo | grep -m1 "model name"
model name : Intel(R) Atom(TM) CPU N270 @ 1.60GHz

[user1@localhost ~]$ gcc -DCMOV -Wall -O2 t.c
[user1@localhost ~]$ time ./a.out
600000000

real 0m0.396s
user 0m0.393s
sys 0m0.003s
[user1@localhost ~]$ gcc -Wall -O2 t.c
[user1@localhost ~]$ time ./a.out
600000000

real 0m0.706s
user 0m0.674s
sys 0m0.002s
[user1@localhost ~]$

SlowJet
10th March 2009, 01:16 PM
I'll wait for LT's and WW's data as it is faster on Pent III 800 and I'll bet it is faster on older non ht Pent IV's.
The logic is not even correct for installing (or not installing) the PAE.
NX should not be part of it because it is simulated on the segment of the intel machine.

And by the time anyone collects data it will all be 686 again.

SJ

BTW: What's the real story on the signing server hack? hmmm
Since dev's don't come down here to get real info, don't drop off propaganda either.
Go put it in your lists.

Jongi
25th April 2009, 05:09 PM
Ah I noticed that my packages were i586 and thought here we go again. Glad to hear it's not an issue this time round.