* RE: [PATCH] fixes for building kernel using Intel compiler (lmben ch data)
@ 2002-10-28 14:47 Nakajima, Jun
2002-10-28 14:51 ` Andi Kleen
0 siblings, 1 reply; 5+ messages in thread
From: Nakajima, Jun @ 2002-10-28 14:47 UTC (permalink / raw)
To: Andi Kleen
Cc: David S. Miller, torvalds, linux-kernel, Mallick, Asit K,
Saxena, Sunil
I don't think people need to use PGO for day-to-day development or
debugging. Rather, it would be used only for systems deployed for actual
use. For example, various kernel binaries optimized for particular use, such
as database, web server, file server, embedded systems, etc, can be
distributed as RPM (with profile feedback data).
For development we should such profile feedback data to optimize the kernel
in source code level (i.e by hand). I don't know the data in gcc has any
clue for that.
Jun
-----Original Message-----
From: Andi Kleen [mailto:ak@suse.de]
Sent: Friday, October 25, 2002 11:27 PM
To: Nakajima, Jun
Cc: Andi Kleen; David S. Miller; torvalds@transmeta.com;
linux-kernel@vger.kernel.org; Mallick, Asit K; Saxena, Sunil
Subject: Re: [PATCH] fixes for building kernel using Intel compiler
(lmben ch data)
On Fri, Oct 25, 2002 at 07:16:49PM -0700, Nakajima, Jun wrote:
> pgoipo -- P70 + PGO (profile-guided optimization) + IPO (interprocedural
> optimization)
That is interesting. I assume you built a custom profiling driver for this.
One problem I see is that profile feedback will make the kernel images much
less
predictible. Currently we need a vmlinux that matches the built kernel
to analyze any oopses etc. When there are small changes (bugfixes etc.)
the functions are still near enough that an oops can be meaningfully
looked at. But with profile based feedback this could completely change -
at least in gcc the profile feedback data has to exactly match the compiled
program. Now when I would change a single line I would need to reprofile
and then the resulting kernel could be totally different with every single
instruction changed, making it impossible to make any sense out of a
ksymoops. This could be a major problem for bug processing on linux-kernel
because the chances are small that I look at a vmlinux that is in any
way related to what the user has.
One way around would be to switch to lkcd crash images which include all
kernel code, but it could be a major problem to send them to a mailing list
because they're so big.
-Andi
^ permalink raw reply [flat|nested] 5+ messages in thread
* Re: [PATCH] fixes for building kernel using Intel compiler (lmben ch data)
2002-10-28 14:47 [PATCH] fixes for building kernel using Intel compiler (lmben ch data) Nakajima, Jun
@ 2002-10-28 14:51 ` Andi Kleen
2002-10-28 15:03 ` Richard B. Johnson
0 siblings, 1 reply; 5+ messages in thread
From: Andi Kleen @ 2002-10-28 14:51 UTC (permalink / raw)
To: Nakajima, Jun
Cc: Andi Kleen, David S. Miller, torvalds, linux-kernel,
Mallick, Asit K, Saxena, Sunil
On Mon, Oct 28, 2002 at 06:47:27AM -0800, Nakajima, Jun wrote:
> I don't think people need to use PGO for day-to-day development or
> debugging. Rather, it would be used only for systems deployed for actual
> use. For example, various kernel binaries optimized for particular use, such
> as database, web server, file server, embedded systems, etc, can be
> distributed as RPM (with profile feedback data).
But unless these kernels are 100% bug free it still leaves the mainteance
issues open.
Also is it really that big a win ?
>
> For development we should such profile feedback data to optimize the kernel
> in source code level (i.e by hand). I don't know the data in gcc has any
> clue for that.
likely/unlikely is the clue. In fact it is already overused.
-Andi
P.S.: Your original mail mentioned that the "P4P kernel was instable
with gcc32". Could you elaborate on the instability? We're using
kernels compiled with gcc 3.2 all the time and they work just fine.
^ permalink raw reply [flat|nested] 5+ messages in thread
* Re: [PATCH] fixes for building kernel using Intel compiler (lmben ch data)
2002-10-28 14:51 ` Andi Kleen
@ 2002-10-28 15:03 ` Richard B. Johnson
0 siblings, 0 replies; 5+ messages in thread
From: Richard B. Johnson @ 2002-10-28 15:03 UTC (permalink / raw)
To: Andi Kleen
Cc: Nakajima, Jun, David S. Miller, torvalds, linux-kernel,
Mallick, Asit K, Saxena, Sunil
On Mon, 28 Oct 2002, Andi Kleen wrote:
> On Mon, Oct 28, 2002 at 06:47:27AM -0800, Nakajima, Jun wrote:
> > I don't think people need to use PGO for day-to-day development or
> > debugging. Rather, it would be used only for systems deployed for actual
> > use. For example, various kernel binaries optimized for particular use, such
> > as database, web server, file server, embedded systems, etc, can be
> > distributed as RPM (with profile feedback data).
>
> But unless these kernels are 100% bug free it still leaves the mainteance
> issues open.
Hmmm.... Where can I get a 100% bug-free kernel?
Cheers,
Dick Johnson
Penguin : Linux version 2.4.18 on an i686 machine (797.90 BogoMips).
Bush : The Fourth Reich of America
^ permalink raw reply [flat|nested] 5+ messages in thread
* RE: [PATCH] fixes for building kernel using Intel compiler (lmben ch data)
@ 2002-10-26 2:16 Nakajima, Jun
2002-10-26 6:27 ` Andi Kleen
0 siblings, 1 reply; 5+ messages in thread
From: Nakajima, Jun @ 2002-10-26 2:16 UTC (permalink / raw)
To: Andi Kleen
Cc: David S. Miller, torvalds, linux-kernel, Mallick, Asit K,
Saxena, Sunil
The following are the lmbench results (P3 and P4P) we tried to get. The
kernels built by gcc 3.2 were not stable for P4P, so we used an older
version of gcc.
p70 -- the kernel built by just with the -O2 optimization (Intel compiler
7.0)
pgoipo -- P70 + PGO (profile-guided optimization) + IPO (interprocedural
optimization)
inline -- pgoipo + forced inline
gcc32 -- gcc 3.2
gcc30 -- gcc 3.02
Jun Nakajima
-----Original Message-----
From: Andi Kleen [mailto:ak@suse.de]
Sent: Friday, October 18, 2002 7:29 PM
To: Nakajima, Jun
Cc: Andi Kleen; David S. Miller; torvalds@transmeta.com;
linux-kernel@vger.kernel.org; Mallick, Asit K; Saxena, Sunil
Subject: Re: [PATCH] fixes for building kernel using Intel compiler
BTW do you have any benchmark / code size results to share with
Intel compiler vs gcc 3.2 ? How much does it give ?
-Andi
-------------------------------------------------------------------------
UP P3 ====================================================================
cd results && make summary percent 2>/dev/null | more
make[1]: Entering directory `/home/hd7/yu/LMbench/results'
L M B E N C H 2 . 0 S U M M A R Y
------------------------------------
Basic system parameters
----------------------------------------------------
Host OS Description Mhz
--------- ------------- ----------------------- ----
p70 Linux 2.4.18 i686-pc-linux-gnu 847
p70 Linux 2.4.18 i686-pc-linux-gnu 847
p70 Linux 2.4.18 i686-pc-linux-gnu 847
pgo Linux 2.4.18 i686-pc-linux-gnu 847
pgo Linux 2.4.18 i686-pc-linux-gnu 847
pgo Linux 2.4.18 i686-pc-linux-gnu 847
pgoipo Linux 2.4.18 i686-pc-linux-gnu 847
pgoipo Linux 2.4.18 i686-pc-linux-gnu 847
pgoipo Linux 2.4.18 i686-pc-linux-gnu 847
inline Linux 2.4.18 i686-pc-linux-gnu 847
inline Linux 2.4.18 i686-pc-linux-gnu 847
inline Linux 2.4.18 i686-pc-linux-gnu 847
gcc32 Linux 2.4.18 i686-pc-linux-gnu 847
gcc32 Linux 2.4.18 i686-pc-linux-gnu 847
gcc32 Linux 2.4.18 i686-pc-linux-gnu 847
Processor, Processes - times in microseconds - smaller is better
----------------------------------------------------------------
Host OS Mhz null null open selct sig sig fork exec
sh
call I/O stat clos TCP inst hndl proc proc
proc
--------- ------------- ---- ---- ---- ---- ---- ----- ---- ---- ---- ----
----
p70 Linux 2.4.18 847 0.37 0.54 2.76 3.61 15.4 0.93 2.92 95.4 1144
5418
p70 Linux 2.4.18 847 0.37 0.54 2.80 3.61 15.9 0.93 2.92 96.3 1130
5414
p70 Linux 2.4.18 847 0.37 0.57 2.78 3.68 16.0 0.93 2.92 95.8 1152
5417
pgo Linux 2.4.18 847 0.35 0.52 2.88 3.42 15.5 0.91 2.92 95.8 1122
5342
pgo Linux 2.4.18 847 0.35 0.52 2.73 3.38 16.8 0.91 2.94 97.4 1107
5332
pgo Linux 2.4.18 847 0.35 0.52 2.83 3.41 13.7 0.91 2.92 95.6 1133
5324
pgoipo Linux 2.4.18 847 0.35 0.53 2.32 2.76 13.3 0.92 2.84 91.9 1131
5267
pgoipo Linux 2.4.18 847 0.35 0.52 2.25 2.76 14.5 0.92 2.84 106. 1120
5259
pgoipo Linux 2.4.18 847 0.35 0.51 2.34 2.81 13.2 0.92 2.84 95.8 1121
5270
inline Linux 2.4.18 847 0.35 0.51 2.26 2.77 13.4 0.92 2.84 105. 1173
5294
inline Linux 2.4.18 847 0.35 0.51 2.26 2.89 13.4 0.92 2.84 91.0 1099
5280
inline Linux 2.4.18 847 0.35 0.51 2.32 2.92 13.3 0.92 2.84 90.6 1126
5329
gcc32 Linux 2.4.18 847 0.35 0.52 3.01 3.67 15.0 0.95 3.10 104. 1200
5532
gcc32 Linux 2.4.18 847 0.35 0.51 3.03 3.73 15.6 0.95 3.10 103. 1182
5521
gcc32 Linux 2.4.18 847 0.35 0.51 3.06 3.76 15.0 0.95 3.10 104. 1169
5520
Context switching - times in microseconds - smaller is better
-------------------------------------------------------------
Host OS 2p/0K 2p/16K 2p/64K 8p/16K 8p/64K 16p/16K 16p/64K
ctxsw ctxsw ctxsw ctxsw ctxsw ctxsw ctxsw
--------- ------------- ----- ------ ------ ------ ------ ------- -------
p70 Linux 2.4.18 0.770 3.9900 13.3 4.5000 163.2 33.3 163.8
p70 Linux 2.4.18 0.700 3.9500 13.7 6.1400 163.4 31.4 163.4
p70 Linux 2.4.18 0.730 4.1900 13.0 7.9100 162.7 37.6 162.9
pgo Linux 2.4.18 0.840 4.1100 22.0 8.9200 164.0 36.6 164.4
pgo Linux 2.4.18 0.820 3.9900 13.0 5.6000 162.6 32.8 163.2
pgo Linux 2.4.18 0.930 3.9900 13.1 4.8700 162.9 33.4 163.4
pgoipo Linux 2.4.18 0.820 3.9100 13.0 6.1800 163.5 30.0 163.5
pgoipo Linux 2.4.18 0.860 3.9600 13.1 5.2000 163.7 33.6 163.6
pgoipo Linux 2.4.18 0.820 4.1000 13.5 5.4900 162.4 33.7 163.2
inline Linux 2.4.18 0.900 4.0800 13.1 7.3500 162.8 30.7 162.9
inline Linux 2.4.18 0.830 4.0600 13.3 7.7100 163.6 33.1 163.9
inline Linux 2.4.18 0.810 3.9300 62.7 4.7700 162.2 31.4 163.3
gcc32 Linux 2.4.18 0.780 4.1100 14.9 8.2400 163.6 33.0 164.3
gcc32 Linux 2.4.18 0.870 4.1900 13.1 4.4300 162.6 30.8 163.0
gcc32 Linux 2.4.18 0.900 4.2000 15.3 6.4100 162.2 30.7 162.5
*Local* Communication latencies in microseconds - smaller is better
-------------------------------------------------------------------
Host OS 2p/0K Pipe AF UDP RPC/ TCP RPC/ TCP
ctxsw UNIX UDP TCP conn
--------- ------------- ----- ----- ---- ----- ----- ----- ----- ----
p70 Linux 2.4.18 0.770 4.111 7.30 13.1 32.8 18.9 44.7 69.1
p70 Linux 2.4.18 0.700 4.405 7.51 13.2 32.8 19.2 45.6 70.1
p70 Linux 2.4.18 0.730 4.392 7.37 13.5 32.8 19.3 44.7 70.8
pgo Linux 2.4.18 0.840 4.187 7.32 11.5 28.6 14.0 36.8 54.4
pgo Linux 2.4.18 0.820 4.354 7.23 11.6 28.4 14.3 36.9 54.8
pgo Linux 2.4.18 0.930 4.386 7.50 11.7 28.6 14.5 37.1 54.9
pgoipo Linux 2.4.18 0.820 4.072 6.83 10.8 27.6 14.0 36.7 53.6
pgoipo Linux 2.4.18 0.860 4.309 7.05 10.9 27.6 14.2 36.3 53.1
pgoipo Linux 2.4.18 0.820 4.168 7.09 11.0 27.7 14.2 37.2 53.3
inline Linux 2.4.18 0.900 4.126 6.96 10.9 27.8 14.1 36.9 53.0
inline Linux 2.4.18 0.830 4.348 6.96 10.8 27.6 14.1 36.7 53.5
inline Linux 2.4.18 0.810 4.363 7.15 10.8 27.3 14.1 36.7 53.7
gcc32 Linux 2.4.18 0.780 4.224 7.33 15.7 35.3 21.6 47.5 72.9
gcc32 Linux 2.4.18 0.870 4.327 7.56 15.7 35.6 21.8 47.1 73.0
gcc32 Linux 2.4.18 0.900 4.377 7.55 15.6 35.5 21.9 48.0 72.8
File & VM system latencies in microseconds - smaller is better
--------------------------------------------------------------
Host OS 0K File 10K File Mmap Prot Page
Create Delete Create Delete Latency Fault Fault
--------- ------------- ------ ------ ------ ------ ------- ----- -----
p70 Linux 2.4.18 17.4 5.0970 101.8 11.6 417.0 0.729 2.00000
p70 Linux 2.4.18 17.5 5.2120 102.3 11.7 420.0 0.729 2.00000
p70 Linux 2.4.18 17.4 5.1150 101.3 11.7 430.0 0.731 2.00000
pgo Linux 2.4.18 18.2 4.9460 102.5 10.9 409.0 0.712 2.00000
pgo Linux 2.4.18 18.1 4.8450 101.6 10.8 422.0 0.700 2.00000
pgo Linux 2.4.18 18.2 4.9180 101.9 10.9 432.0 0.701 2.00000
pgoipo Linux 2.4.18 17.1 4.5070 100.2 10.4 376.0 0.709 2.00000
pgoipo Linux 2.4.18 17.0 4.4400 100.2 10.3 388.0 0.708 2.00000
pgoipo Linux 2.4.18 17.1 4.5340 100.1 10.5 393.0 0.730 2.00000
inline Linux 2.4.18 17.1 4.4790 101.0 10.4 375.0 0.709 2.00000
inline Linux 2.4.18 17.1 4.5270 100.9 10.4 387.0 0.709 2.00000
inline Linux 2.4.18 17.2 4.5390 99.6 10.4 380.0 0.709 2.00000
gcc32 Linux 2.4.18 36.1 5.4180 123.8 12.1 379.0 0.709 2.00000
gcc32 Linux 2.4.18 36.2 5.3890 122.8 12.0 393.0 0.715 2.00000
gcc32 Linux 2.4.18 36.1 5.3990 121.0 12.0 398.0 0.709 2.00000
*Local* Communication bandwidths in MB/s - bigger is better
-----------------------------------------------------------
Host OS Pipe AF TCP File Mmap Bcopy Bcopy Mem Mem
UNIX reread reread (libc) (hand) read
write
--------- ------------- ---- ---- ---- ------ ------ ------ ------ ----
-----
p70 Linux 2.4.18 847. 660. 205. 402.5 372.7 157.5 150.8 372.
196.1
p70 Linux 2.4.18 841. 409. 233. 410.2 372.8 153.8 147.5 372.
197.1
p70 Linux 2.4.18 848. 712. 235. 414.5 372.8 156.1 151.5 372.
202.0
pgo Linux 2.4.18 863. 571. 280. 415.1 372.8 152.8 146.8 372.
193.2
pgo Linux 2.4.18 860. 555. 285. 415.1 372.8 151.9 145.3 372.
195.0
pgo Linux 2.4.18 847. 750. 168. 415.4 372.6 152.7 146.6 372.
195.7
pgoipo Linux 2.4.18 869. 704. 237. 417.4 372.9 152.7 146.1 372.
193.1
pgoipo Linux 2.4.18 868. 683. 308. 414.2 372.9 151.5 145.3 372.
194.9
pgoipo Linux 2.4.18 866. 393. 220. 413.9 372.9 152.8 146.4 372.
195.8
inline Linux 2.4.18 880. 550. 223. 406.1 372.6 152.9 146.7 372.
191.8
inline Linux 2.4.18 803. 345. 212. 416.7 372.9 151.5 145.3 372.
195.0
inline Linux 2.4.18 872. 704. 236. 413.6 372.9 152.7 146.2 372.
195.9
gcc32 Linux 2.4.18 872. 612. 190. 414.0 372.8 158.5 151.2 372.
194.9
gcc32 Linux 2.4.18 868. 384. 230. 416.8 372.5 155.2 150.1 372.
199.4
gcc32 Linux 2.4.18 856. 675. 196. 416.6 372.8 157.1 151.8 372.
202.8
Memory latencies in nanoseconds - smaller is better
(WARNING - may not be correct, check graphs)
---------------------------------------------------
Host OS Mhz L1 $ L2 $ Main mem Guesses
--------- ------------- ---- ----- ------ -------- -------
p70 Linux 2.4.18 847 3.540 8.2610 131.8
p70 Linux 2.4.18 847 3.540 8.2630 131.8
p70 Linux 2.4.18 847 3.541 8.2610 131.8
pgo Linux 2.4.18 847 3.541 8.2600 131.8
pgo Linux 2.4.18 847 3.541 8.2600 131.8
pgo Linux 2.4.18 847 3.541 8.2610 131.8
pgoipo Linux 2.4.18 847 3.540 8.2710 131.8
pgoipo Linux 2.4.18 847 3.540 8.2700 131.8
pgoipo Linux 2.4.18 847 3.541 8.2610 131.8
inline Linux 2.4.18 847 3.541 8.2610 131.8
inline Linux 2.4.18 847 3.540 8.2610 131.8
inline Linux 2.4.18 847 3.541 8.2700 131.8
gcc32 Linux 2.4.18 847 3.541 46.8 131.8
gcc32 Linux 2.4.18 847 3.541 8.2610 131.8
gcc32 Linux 2.4.18 847 3.540 8.2610 131.8
make[1]: Leaving directory `/home/hd7/yu/LMbench/results'
UP P4P
========================================================================
cd results && make summary percent 2>/dev/null | more
make[1]: Entering directory `/home/benjamin/LMbench/results'
L M B E N C H 2 . 0 S U M M A R Y
------------------------------------
Basic system parameters
----------------------------------------------------
Host OS Description Mhz
--------- ------------- ----------------------- ----
p70 Linux 2.4.18 i686-pc-linux-gnu 1900
p70 Linux 2.4.18 i686-pc-linux-gnu 1900
p70 Linux 2.4.18 i686-pc-linux-gnu 1900
pgoipo Linux 2.4.18 i686-pc-linux-gnu 1900
pgoipo Linux 2.4.18 i686-pc-linux-gnu 1900
pgoipo Linux 2.4.18 i686-pc-linux-gnu 1900
inline Linux 2.4.18 i686-pc-linux-gnu 1900
inline Linux 2.4.18 i686-pc-linux-gnu 1900
inline Linux 2.4.18 i686-pc-linux-gnu 1900
gcc30 Linux 2.4.18 i686-pc-linux-gnu 1900
gcc30 Linux 2.4.18 i686-pc-linux-gnu 1900
gcc30 Linux 2.4.18 i686-pc-linux-gnu 1900
Processor, Processes - times in microseconds - smaller is better
----------------------------------------------------------------
Host OS Mhz null null open selct sig sig fork exec
sh
call I/O stat clos TCP inst hndl proc proc
proc
--------- ------------- ---- ---- ---- ---- ---- ----- ---- ---- ---- ----
----
p70 Linux 2.4.18 1900 0.84 0.92 2.19 3.31 7.107 1.16 3.34 139. 731.
3328
p70 Linux 2.4.18 1900 0.84 0.93 2.19 3.37 7.102 1.17 3.38 128. 707.
3280
p70 Linux 2.4.18 1900 0.84 0.94 2.23 3.39 7.026 1.17 3.38 132. 709.
3286
pgoipo Linux 2.4.18 1900 0.84 0.92 1.98 2.99 6.471 1.15 3.17 124. 701.
3240
pgoipo Linux 2.4.18 1900 0.84 0.92 2.03 3.00 6.471 1.15 3.16 131. 697.
3223
pgoipo Linux 2.4.18 1900 0.84 0.92 2.02 2.93 6.492 1.15 3.15 126. 697.
3223
inline Linux 2.4.18 1900 0.84 0.91 2.04 3.00 5.852 1.12 3.15 119. 691.
3209
inline Linux 2.4.18 1900 0.84 0.93 2.09 3.00 5.858 1.14 3.13 124. 692.
3219
inline Linux 2.4.18 1900 0.84 0.91 2.05 3.01 6.552 1.14 3.12 130. 699.
3216
gcc30 Linux 2.4.18 1900 0.84 0.93 2.75 3.77 7.908 1.16 3.34 131. 707.
3303
gcc30 Linux 2.4.18 1900 0.84 0.93 2.71 3.76 7.947 1.16 3.35 136. 707.
3322
gcc30 Linux 2.4.18 1900 0.84 0.93 2.72 3.72 7.913 1.16 3.35 130. 710.
3326
Context switching - times in microseconds - smaller is better
-------------------------------------------------------------
Host OS 2p/0K 2p/16K 2p/64K 8p/16K 8p/64K 16p/16K 16p/64K
ctxsw ctxsw ctxsw ctxsw ctxsw ctxsw ctxsw
--------- ------------- ----- ------ ------ ------ ------ ------- -------
p70 Linux 2.4.18 1.000 2.1900 5.7400 6.4100 41.5 12.2 41.5
p70 Linux 2.4.18 1.070 2.2400 9.6600 4.7900 41.4 12.2 41.7
p70 Linux 2.4.18 1.020 2.4400 10.0 8.0700 41.4 12.4 41.4
pgoipo Linux 2.4.18 0.960 2.2300 7.3700 5.6800 41.4 12.8 41.5
pgoipo Linux 2.4.18 1.160 2.1900 5.8900 5.6300 41.5 12.5 41.6
pgoipo Linux 2.4.18 0.960 2.2200 5.9400 5.6000 41.4 12.8 41.4
inline Linux 2.4.18 0.930 2.2100 14.8 6.5200 41.6 12.5 41.6
inline Linux 2.4.18 1.130 2.1400 6.3400 5.5700 41.4 12.4 41.4
inline Linux 2.4.18 0.790 2.1800 6.2600 5.3600 41.3 12.2 41.5
gcc30 Linux 2.4.18 1.040 2.9000 6.6000 5.8500 41.4 12.2 41.6
gcc30 Linux 2.4.18 1.020 2.3000 11.8 4.8200 41.6 12.1 41.5
gcc30 Linux 2.4.18 1.040 2.4500 7.9700 5.4100 41.0 12.5 41.5
*Local* Communication latencies in microseconds - smaller is better
-------------------------------------------------------------------
Host OS 2p/0K Pipe AF UDP RPC/ TCP RPC/ TCP
ctxsw UNIX UDP TCP conn
--------- ------------- ----- ----- ---- ----- ----- ----- ----- ----
p70 Linux 2.4.18 1.000 6.086 8.91 12.3 22.9 14.1 29.3 41.5
p70 Linux 2.4.18 1.070 6.087 8.79 12.2 23.5 14.1 29.7 41.1
p70 Linux 2.4.18 1.020 6.515 9.01 12.3 24.1 14.2 30.6 41.5
pgoipo Linux 2.4.18 0.960 6.133 8.38 11.2 23.6 12.5 26.8 38.2
pgoipo Linux 2.4.18 1.160 5.961 8.73 11.3 24.1 13.0 27.8 37.4
pgoipo Linux 2.4.18 0.960 6.037 8.48 11.3 21.3 12.7 26.6 38.1
inline Linux 2.4.18 0.930 5.972 8.48 11.2 22.0 12.5 26.3 38.0
inline Linux 2.4.18 1.130 6.213 8.69 12.2 22.4 12.8 26.6 37.5
inline Linux 2.4.18 0.790 5.881 9.21 11.2 22.5 12.4 27.1 39.5
gcc30 Linux 2.4.18 1.040 6.259 8.95 19.1 23.8 14.7 28.9 43.1
gcc30 Linux 2.4.18 1.020 6.216 9.09 12.6 25.4 14.7 30.1 48.4
gcc30 Linux 2.4.18 1.040 6.253 9.00 12.6 24.1 15.3 28.7 600K
File & VM system latencies in microseconds - smaller is better
--------------------------------------------------------------
Host OS 0K File 10K File Mmap Prot Page
Create Delete Create Delete Latency Fault Fault
--------- ------------- ------ ------ ------ ------ ------- ----- -----
p70 Linux 2.4.18 9.1580 3.3540 33.3 6.4340 737.0 1.598 3.00000
p70 Linux 2.4.18 11.6 4.5730 35.4 7.9510 746.0 1.542 3.00000
p70 Linux 2.4.18 9.2350 3.4030 31.5 6.5050 744.0 1.567 3.00000
pgoipo Linux 2.4.18 9.5180 2.9890 32.6 5.9670 732.0 1.579 3.00000
pgoipo Linux 2.4.18 9.5060 2.9900 31.9 5.8690 724.0 1.579 3.00000
pgoipo Linux 2.4.18 9.7190 2.9990 31.5 5.9370 723.0 1.578 3.00000
inline Linux 2.4.18 9.2640 2.9400 33.9 5.9020 721.0 1.566 3.00000
inline Linux 2.4.18 9.2830 2.9270 32.9 5.8440 721.0 1.573 3.00000
inline Linux 2.4.18 9.2660 2.9700 32.4 5.9030 798.0 1.572 3.00000
gcc30 Linux 2.4.18 24.6 3.3240 48.5 6.8920 751.0 1.580 3.00000
gcc30 Linux 2.4.18 24.3 3.3140 47.7 6.8220 749.0 1.536 3.00000
gcc30 Linux 2.4.18 24.6 3.3320 46.9 6.9150 751.0 1.555 3.00000
*Local* Communication bandwidths in MB/s - bigger is better
-----------------------------------------------------------
Host OS Pipe AF TCP File Mmap Bcopy Bcopy Mem Mem
UNIX reread reread (libc) (hand) read
write
--------- ------------- ---- ---- ---- ------ ------ ------ ------ ----
-----
p70 Linux 2.4.18 1188 1470 714. 1471.7 1459.6 623.7 615.5 1459
730.5
p70 Linux 2.4.18 984. 1490 394. 1520.3 1459.7 609.0 621.2 1459
771.0
p70 Linux 2.4.18 1054 1053 380. 1507.1 1459.5 624.2 626.8 1459
798.7
pgoipo Linux 2.4.18 1130 1432 408. 1529.6 1459.8 623.9 616.3 1459
731.7
pgoipo Linux 2.4.18 905. 1734 337. 1527.2 1459.6 616.1 624.9 1459
767.3
pgoipo Linux 2.4.18 1100 1494 396. 1493.3 1459.6 630.9 628.9 1459
796.7
inline Linux 2.4.18 1151 1629 380. 1530.8 1459.5 612.5 610.3 1459
732.6
inline Linux 2.4.18 1186 1128 413. 1132.0 1459.5 608.4 623.8 1459
768.0
inline Linux 2.4.18 937. 1105 345. 1522.3 1459.5 628.1 632.2 1459
800.1
gcc30 Linux 2.4.18 1056 1098 349. 1449.5 1460.0 613.3 608.7 1459
734.2
gcc30 Linux 2.4.18 1231 1343 368. 1472.3 1460.1 616.7 627.0 1459
773.9
gcc30 Linux 2.4.18 1134 1182 372. 1503.9 1459.9 626.1 632.1 1460
801.0
Memory latencies in nanoseconds - smaller is better
(WARNING - may not be correct, check graphs)
---------------------------------------------------
Host OS Mhz L1 $ L2 $ Main mem Guesses
--------- ------------- ---- ----- ------ -------- -------
p70 Linux 2.4.18 1900 1.056 9.7150 166.6
p70 Linux 2.4.18 1900 1.056 20.1 166.4
p70 Linux 2.4.18 1900 1.056 9.7120 166.4
pgoipo Linux 2.4.18 1900 1.056 54.7 166.6
pgoipo Linux 2.4.18 1900 1.056 9.7130 166.5
pgoipo Linux 2.4.18 1900 1.056 9.7170 166.4
inline Linux 2.4.18 1900 1.056 9.7110 166.3
inline Linux 2.4.18 1900 1.056 9.7240 166.8
inline Linux 2.4.18 1900 1.057 9.7120 166.7
gcc30 Linux 2.4.18 1900 1.146 9.7200 166.3
gcc30 Linux 2.4.18 1900 1.059 9.7120 166.6
gcc30 Linux 2.4.18 1900 1.058 9.7130 166.6
make[1]: Leaving directory `/home/benjamin/LMbench/results'
^ permalink raw reply [flat|nested] 5+ messages in thread* Re: [PATCH] fixes for building kernel using Intel compiler (lmben ch data)
2002-10-26 2:16 Nakajima, Jun
@ 2002-10-26 6:27 ` Andi Kleen
0 siblings, 0 replies; 5+ messages in thread
From: Andi Kleen @ 2002-10-26 6:27 UTC (permalink / raw)
To: Nakajima, Jun
Cc: Andi Kleen, David S. Miller, torvalds, linux-kernel,
Mallick, Asit K, Saxena, Sunil
On Fri, Oct 25, 2002 at 07:16:49PM -0700, Nakajima, Jun wrote:
> pgoipo -- P70 + PGO (profile-guided optimization) + IPO (interprocedural
> optimization)
That is interesting. I assume you built a custom profiling driver for this.
One problem I see is that profile feedback will make the kernel images much less
predictible. Currently we need a vmlinux that matches the built kernel
to analyze any oopses etc. When there are small changes (bugfixes etc.)
the functions are still near enough that an oops can be meaningfully
looked at. But with profile based feedback this could completely change -
at least in gcc the profile feedback data has to exactly match the compiled
program. Now when I would change a single line I would need to reprofile
and then the resulting kernel could be totally different with every single
instruction changed, making it impossible to make any sense out of a
ksymoops. This could be a major problem for bug processing on linux-kernel
because the chances are small that I look at a vmlinux that is in any
way related to what the user has.
One way around would be to switch to lkcd crash images which include all
kernel code, but it could be a major problem to send them to a mailing list
because they're so big.
-Andi
^ permalink raw reply [flat|nested] 5+ messages in thread
end of thread, other threads:[~2002-10-28 14:55 UTC | newest]
Thread overview: 5+ messages (download: mbox.gz follow: Atom feed
-- links below jump to the message on this page --
2002-10-28 14:47 [PATCH] fixes for building kernel using Intel compiler (lmben ch data) Nakajima, Jun
2002-10-28 14:51 ` Andi Kleen
2002-10-28 15:03 ` Richard B. Johnson
-- strict thread matches above, loose matches on Subject: below --
2002-10-26 2:16 Nakajima, Jun
2002-10-26 6:27 ` Andi Kleen
This is an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.