AW: [U-Boot-Users] [PATCH 4/4] add csb637 (at91rm9200) support

[Martin Krause]

Hi Anders,

Anders Larsen wrote on Montag, 2. Mai 2005 14:13:
> > I've also done an nbench benchmark under linux running in
> > synchronous clock mode. But the results are the same as in fast bus
> > clock mode. Strange. I've no idea why.
> 
> Perhaps your Linux switches clock mode itself? (the at91rm9200 patch

I can't say, but I think not because system boot time decreases
in syncronous bus mode.

> from maxim.org.za does _not_, however)

I use the linuxarm tree from denx.de

> What does /proc/cpuinfo say about this?
> With my patch I get:
> # grep -i bogomips /proc/cpuinfo
> BogoMIPS        : 91.95
> and without it I get appr. 28.

With syncronous mode I get:

root at CMC-TC-PU2:/>grep -i bogomips /proc/cpuinfo
BogoMIPS        : 89.70
root at CMC-TC-PU2:/>

And without it I get:

# grep -i bogomips /proc/cpuinfo
BogoMIPS        : 29.90
#

That's very similar to your results, but not exact the same.
Perhaps because you are using a 2.6 kernel. What CPU clock frequency
do you use? We use 179 MHz.

> For reference here's the output from nbench 2.2.2 on my csb637 board:

Here are my nbench results for synchronous clock mode
(nbench 2.2.1 on cmc-pu2 board):

BYTEmark* Native Mode Benchmark ver. 2 (10/95)
Index-split by Andrew D. Balsa (11/97)
Linux/Unix* port by Uwe F. Mayer (12/96,11/97)

TEST                : Iterations/sec.  : Old Index   : New Index
                    :                  : Pentium 90* : AMD K6/233*
--------------------:------------------:-------------:------------
NUMERIC SORT        :          54.237  :       1.39  :       0.46
STRING SORT         :          3.7939  :       1.70  :       0.26
BITFIELD            :      1.2875e+07  :       2.21  :       0.46
FP EMULATION        :          5.5534  :       2.66  :       0.61
FOURIER             :          5.9549  :       0.01  :       0.00
ASSIGNMENT          :          0.4583  :       1.74  :       0.45
IDEA                :          164.47  :       2.52  :       0.75
HUFFMAN             :            21.4  :       0.59  :       0.19
NEURAL NET          :       0.0068962  :       0.01  :       0.00
LU DECOMPOSITION    :         0.23381  :       0.01  :       0.01
==========================ORIGINAL BYTEMARK RESULTS==========================
INTEGER INDEX       : 1.669
FLOATING-POINT INDEX: 0.010
Baseline (MSDOS*)   : Pentium* 90, 256 KB L2-cache, Watcom* compiler 10.0
==============================LINUX DATA BELOW===============================
CPU                 :
L2 Cache            :
OS                  : Linux 2.4.27-vrs1
C compiler          : arm-linux-gcc
libc                : static
MEMORY INDEX        : 0.380
INTEGER INDEX       : 0.447
FLOATING-POINT INDEX: 0.005
Baseline (LINUX)    : AMD K6/233*, 512 KB L2-cache, gcc 2.7.2.3, libc-5.4.38
* Trademarks are property of their respective holder.

And this are the results for fast bus mode:

BYTEmark* Native Mode Benchmark ver. 2 (10/95)
Index-split by Andrew D. Balsa (11/97)
Linux/Unix* port by Uwe F. Mayer (12/96,11/97)

TEST                : Iterations/sec.  : Old Index   : New Index
                    :                  : Pentium 90* : AMD K6/233*
--------------------:------------------:-------------:------------
NUMERIC SORT        :          20.295  :       0.52  :       0.17
STRING SORT         :          1.2613  :       0.56  :       0.09
BITFIELD            :      4.3108e+06  :       0.74  :       0.15
FP EMULATION        :          1.8553  :       0.89  :       0.21
FOURIER             :          1.9967  :       0.00  :       0.00
ASSIGNMENT          :         0.19771  :       0.75  :       0.20
IDEA                :          54.685  :       0.84  :       0.25
HUFFMAN             :          7.1418  :       0.20  :       0.06
NEURAL NET          :       0.0023085  :       0.00  :       0.00
LU DECOMPOSITION    :        0.078419  :       0.00  :       0.00
==========================ORIGINAL BYTEMARK RESULTS==========================
INTEGER INDEX       : 0.587
FLOATING-POINT INDEX: 0.003
Baseline (MSDOS*)   : Pentium* 90, 256 KB L2-cache, Watcom* compiler 10.0
==============================LINUX DATA BELOW===============================
CPU                 :
L2 Cache            :
OS                  : Linux 2.4.27-vrs1
C compiler          : arm-linux-gcc
libc                : static
MEMORY INDEX        : 0.138
INTEGER INDEX       : 0.153
FLOATING-POINT INDEX: 0.002
Baseline (LINUX)    : AMD K6/233*, 512 KB L2-cache, gcc 2.7.2.3, libc-5.4.38
* Trademarks are property of their respective holder.

On last November I have already done a nbench run for our board. There I 
got the same results like now with synchronous mode. So I guess in earlier
versions of U-boot the synchronous mode had already been set.

I checked this. In U-Boot 1.1.2 the asynchronous mode was used for the
AT91RM9200 cpu. This was changed by a patch by Steven Scholz when doing
the SoC cleanup for the arm920t directory tree (I could not say, which
patch exactly). Steven, did you have a special reason for doing this?

For performance reasons I suggest we should switch back to asynchronous
(or synchronous clock mode, no idea which one is favorable). Any 
suggestions?

Regards,
Martin