(Solved) Serious GCC regression

[jzakiya]

I have a Toshiba Satelite laptop, 2.8 Ghz, Intel P4, 1GB ram.
Dualboot with XP, 2 Linux partitions, '/' and '/home' (from previous)

I installed KDE4 2010 last week and have experienced a severe
regression in the speed of everything. It appears the gcc compiles
apps with serious performance regressions from 2009.1.

I have recompiled from source various languages I use:
Ruby (various versions), Python (2.6.5, 3.1.1) and Gforth 0.7.0.
All the benchmarks I have run are at least 4-6 times slower now than before.
The performance still remains the same when run under XP.

This is really disappointing and bewildering since every review I've read
talked about the speed/snappiness of 2010.

If I can't get to at least my previous performance level, I'll have to wipe
2010 and go back to 2009.1, or find something else in the mean time.

[muungwana]

thats a serious charge you are making against gcc mr. Do you have any proof to back them up? what makes you think gcc is at fault for the slowness you experience?

most slowness experienced comes from a bad video driver, video card combination. What video card do you have, what video driver is in use?

[menotu]

In addition to muungwana post, are you using your old /home (2009x) partition?

[Neal ManBear]

Other than compiling languages from source, what other alterations / installations have you done?

[jzakiya]

It was a standard stock install. The same thing I've done with
every other ISO update.  I just let the installer use my old partitions
'/' and '/home' and do its thing.

The reason I highly suspect it's a gcc issue is because the same gcc is used to
compile all the code put on the system.

The language benchmarks were MUCH SLOWER when I ran than after the initial
install (these languages were compiled by the 2009.1 gcc, and reside in my /home,
so weren't touched by the new HD Install).  When I recompiled then with the new
2010 gcc they ran faster, but were still 4-6 times slower than before.

I know from reading about the 4.4 gcc series there have been regression problems
brought up. My laptop has an Intel P4, which apparently the 4.4.1 settings isn't producing optimized code for (or even just normally compiled code), because my
Ruby environment on XP provided the same performance (even a little bit faster)
than my 2009.1 did.  The ONLY difference I see that affects this is the gcc version
and the settings used to compile the sources.

To test this, I ran the same benchmarks as Root user, which should not be touching
my /home partition, and thus, the apps compiled there, and it exhibited the same
slower performance.

And in general, ALL the apps, and general feel of the OS now, is unusably slow
and lethargic. Reminds me of my 1999 AMD K6 600Mhz desktop.

To be clear, I am making a technical diagnostic analysis here, and adore PCLOS.
I just want it to be at least just as fast as it was before. I have waited 5 days, and searched various forums, b4 posting this, to get a clue to the origins of this problem, and a possible fix (other than wiping the install and starting over).

[muungwana]

it will be helpful if somebody else can reproduce  your problem, what benchmark are you running, can somebody else have access to it?

Can you find any online reference  where there is a discussion about problems with our particular version of gcc or options used.

can you build a new version of gcc and run the benchmark against it?

4-6 times slowness is pretty big, it doesnt make sense why nobody else has picked up such reduction in speed after all this time

[menotu]

To be clear, I am making a technical diagnostic analysis here, and adore PCLOS.

I didn't think you were "knocking" PCLinuxOS - if you have issues you should post them as you've done.

My knowledge of compiling is very weak therefore I can't help there. 

I' would suggest - if possible - doing another install, but use a new partition setup, including a brand new formatted /home.

If you have data you need from your /home you could re-introduce it after you have the OS setup as you like. At least that would take that out of the equation.

[jzakiya]

I had posted something else with an attachment, but it seems to have gotten lost.

I've attached a Ruby file that contains some benchmark code. All you have to do is
run it under Ruby. The repos has Ruby 1.8.7, so just install it, and from a terminal:

$ ruby <wherever you put file>/<filename>

and it will display output will times and some other stuff.
The attached file is EratSievesY.txt
(Rename it to *.rb from *.txt.  Attachment wouldn't take *.rb file extension)

Of course, the results you get will be based on your cpu type/speed, but if you post the results I can probably assess if they look reasonable to what I had been getting.

[muungwana]

run the test with pentium 4, 2.0 Ghz processor with 1GB of main memory and this is what i got:

Code: [Select]

[ink@kdi downloads]$ ruby EratSievesY.rb
EratSievesY.rb:1877: warning: statement not reached
                    user     system      total        real
primes up to 10000001:  19.360000   2.270000  21.630000 ( 26.083136)
primes up to 10000001:  29.000000   4.720000  33.720000 ( 40.611239)
primes up to 10000001:  10.100000   1.320000  11.420000 ( 13.752512)
primes up to 10000001:   0.000000  14.980000  14.980000 ( 18.022460)
primes up to 10000001:   0.000000  11.300000  11.300000 ( 13.963937)
primes up to 10000001:   0.000000   9.640000   9.640000 ( 11.623490)
primes up to 10000001:   0.000000  10.800000  10.800000 ( 12.981930)
primes up to 10000001:   0.000000  15.120000  15.120000 ( 18.194831)
primes up to 10000001:   0.000000  11.600000  11.600000 ( 14.058229)
primes up to 10000001:   0.000000   9.650000   9.650000 ( 11.590924)
primes up to 10000001:   0.000000  11.430000  11.430000 ( 14.096823)

[9999889, 9999901, 9999907, 9999929, 9999931, 9999937, 9999943, 9999971, 9999973, 9999991]
664579

[9999889, 9999901, 9999907, 9999929, 9999931, 9999937, 9999943, 9999971, 9999973, 9999991]
664579

[9999889, 9999901, 9999907, 9999929, 9999931, 9999937, 9999943, 9999971, 9999973, 9999991]
664579

[9999889, 9999901, 9999907, 9999929, 9999931, 9999937, 9999943, 9999971, 9999973, 9999991]
664579

[9999889, 9999901, 9999907, 9999929, 9999931, 9999937, 9999943, 9999971, 9999973, 9999991]
664579

[9999889, 9999901, 9999907, 9999929, 9999931, 9999937, 9999943, 9999971, 9999973, 9999991]
664579

[9999889, 9999901, 9999907, 9999929, 9999931, 9999937, 9999943, 9999971, 9999973, 9999991]
664579

[9999889, 9999901, 9999907, 9999929, 9999931, 9999937, 9999943, 9999971, 9999973, 9999991]
664579

[9999889, 9999901, 9999907, 9999929, 9999931, 9999937, 9999943, 9999971, 9999973, 9999991]
664579

[9999889, 9999901, 9999907, 9999929, 9999931, 9999937, 9999943, 9999971, 9999973, 9999991]
664579

[9999889, 9999901, 9999907, 9999929, 9999931, 9999937, 9999943, 9999971, 9999973, 9999991]
664579


[mellon]

And here is another run.
HP DV7 1150ed laptop with T5800 dual core 2.0 Ghz and 4GB memory on PAE 2.6.33.4 kernel

(I planned to do a run on a 2009 version, but I'm afraid I was only offered a manual fsck which marks the end of pclos 2009 on my machine).

Code: [Select]

ruby EratSievesY.rb
EratSievesY.rb:1877: warning: statement not reached
                    user     system      total        real
primes up to 10000001:  10.710000   2.630000  13.340000 ( 13.368470)
primes up to 10000001:  15.500000   5.440000  20.940000 ( 21.084126)
primes up to 10000001:   5.810000   1.400000   7.210000 (  7.249503)
primes up to 10000001:   7.800000   1.710000   9.510000 (  9.637418)
primes up to 10000001:   5.910000   1.340000   7.250000 (  7.365571)
primes up to 10000001:   5.030000   1.060000   6.090000 (  6.112546)
primes up to 10000001:   5.490000   1.250000   6.740000 (  6.748306)
primes up to 10000001:   8.000000   1.600000   9.600000 (  9.617613)
primes up to 10000001:   5.800000   1.320000   7.120000 (  7.157527)
primes up to 10000001:   5.120000   0.970000   6.090000 (  6.091731)
primes up to 10000001:   5.880000   1.100000   6.980000 (  7.004667)

[9999889, 9999901, 9999907, 9999929, 9999931, 9999937, 9999943, 9999971,
9999973, 9999991]
664579

[9999889, 9999901, 9999907, 9999929, 9999931, 9999937, 9999943, 9999971,
9999973, 9999991]
664579

[9999889, 9999901, 9999907, 9999929, 9999931, 9999937, 9999943, 9999971,
9999973, 9999991]
664579

[9999889, 9999901, 9999907, 9999929, 9999931, 9999937, 9999943, 9999971,
9999973, 9999991]
664579

[9999889, 9999901, 9999907, 9999929, 9999931, 9999937, 9999943, 9999971,
9999973, 9999991]
664579

[9999889, 9999901, 9999907, 9999929, 9999931, 9999937, 9999943, 9999971,
9999973, 9999991]
664579

[9999889, 9999901, 9999907, 9999929, 9999931, 9999937, 9999943, 9999971,
9999973, 9999991]
664579

[9999889, 9999901, 9999907, 9999929, 9999931, 9999937, 9999943, 9999971,
9999973, 9999991]
664579

[9999889, 9999901, 9999907, 9999929, 9999931, 9999937, 9999943, 9999971,
9999973, 9999991]
664579

[9999889, 9999901, 9999907, 9999929, 9999931, 9999937, 9999943, 9999971,
9999973, 9999991]
664579

[9999889, 9999901, 9999907, 9999929, 9999931, 9999937, 9999943, 9999971,
9999973, 9999991]
664579

mellon

[kjpetrie]

GCC has been rebuilt but remains at the same version as in 2009 - 4.4.1.

[Neal ManBear]

GCC has been rebuilt but remains at the same version as in 2009 - 4.4.1.

GCC 4.4.1 was built for 2010 in 2009. Hey, we had to start somewhere.

[gseaman]

I haven't built any programs since installing 2010, but unless there is unusual hardware involved, I am confident gcc is not the problem for you. It will be difficult to help solve it until you are willing to consider the problem lies elsewhere. If apps compiled by gcc for pclinuxos were slowed down, even a little bit, others would be screaming, myself included, that it be fixed right away.

I agree that you should re-install with newly formatted partitions.

Galen

[Neal ManBear]

I haven't built any programs since installing 2010, but unless there is unusual hardware involved, I am confident gcc is not the problem for you. It will be difficult to help solve it until you are willing to consider the problem lies elsewhere. If apps compiled by gcc for pclinuxos were slowed down, even a little bit, others would be screaming, myself included, that it be fixed right away.

I agree that you should re-install with newly formatted partitions.

Galen

+1

[jaydot]

I just let the installer use my old partitions '/' and '/home' and do its thing.

a clean install, formatting both / and /home is required for 2010.