Performace benchmarks: PARTNER-Jet vs Lauterbach's TRACE32
To demonstrate the superior performance of the PARTNER-Jet, KMC conducted tests using an T-Engine MC9328MX1 (Model TTM101-001 with ARM920T core), one of the most popular development platform for ARM920 series supplied from Yokogawa Digital Computer Corporation. T-Engine is an effective cross-platform measure of system performance because it is widely supported by several ICEs. All tests was done with this board.
KMC ran the benchmark tests using commonly used ICE operations, including file download and upload, step execution, FLASH ROM writing and debug information loading. We measured the time to execute each action or operation and compared the performance of all actions.
Testing Environment:
Processor:
|
Pentium 4 2GHz
|
Memory:
|
512MB
|
Operating system:
|
Window 2000
|
Target boards:
- T-Engine MC9328MX1 (ARM920T core)
Model number: TTM101-001
Yokogawa Digital Computer Corporation
Tested ICEs:
- PARTNER-Jet (USB 2.0 and Ethernet)
Kyoto Microcomputer Co., Ltd.
- TRACE32 (USB 2.0)
Lauterbach Datentecnik GmbH
ICE settings:
Settings for each ICE is shown below. JTAG clock frequency is set manually so that each ICE can debug the target board at the highest speed. I/O system on the target board is initialized after activating debugger.
ICE model
|
PARTNER-Jet
|
TRACE32
|
Supported connection
|
USB 2.0 and Ethernet
|
USB 2.0
|
Tested connection
|
USB 2.0
|
USB 2.0
|
JTAG clock frequency
|
43 MHz
|
47 MHz
|
ICE setting |
CPU CORE 900T ENDIAN LITTLE JMONITOR ADDRESS 0x00300100
|
SYSTEM.CPU ARM920T SYSTEM.JTAGCLOCK CTCK 47MHz SYSTEM.Option WaitReset ON SYSTEM.UP
|
FLASH setting |
FLASH MEMORY 10000000, 10ffffff,BOTH8,32,16,/c
|
FLASH.RESet FLASH.Create 1. 0x10000000-0x10ffffff 0x10000 AM29LV100 Word FLASH.Erase 0x10800000--0x108fffff
|
FLASH setting using working RAM |
FLASH_WORK auto
|
FLASH.RESet FLASH.Create 1. 0x10000000-0x10ffffff 0x10000 TARGET Word FLASH.Target 0x08000000++0xFFF 0x08001000++0xFFFF am29lv100.bin FLASH.Erase 0x10800000--0x108fffff
|
Download command |
rd 4m.bin,08100000
|
Data.LOAD.BINARY 4m.bin 0x08100000
|
Upload command |
wr u4m.bin,08100000,084fffff
|
Data.SAVE.BINARY u4m.bin 0x08100000--0x084fffff
|
FLASH memory writing command |
rd 1m.bin,10800000
|
FLASH.Program ALL DATA.Load.BINARY 1m.bin 0x10800000 FLASH.Program OFF
|
Step command |
t _1000
|
Step.Asm 0x3e8
|
Debug information loading command |
ls vm.out
|
Data.LOAD.Elf vm.out /gnu /nocode
|
- Note
- PARTNER-Jet has a functionality to measure comsumption time of a command and the measured time is as accurate as within 1/100 second.
To demonstrate the superior performance of the PARTNER-Jet, KMC conducted tests by downloading a 4MB binary file into target memory and measured time for the operation. For upload, we transferred 4MB of data on the target memory to host machine. We measured a comsumption time with a stopwatch from a command entry to the next command prompt in the command window. We also ensured that the time was accurate enough to match the one entering command from GUI window.
Benchmark result #1: Downoad and Upload speed
ICE model
|
PARTNER-Jet
|
TRACE32
|
Download time
|
1.67 sec (2452.7 KB/sec)
|
3.1 sec (1290.3 KB/sec)
|
Upload time
|
1.98 sec (2068.7 KB/sec)
|
14.8 sec (270.2 KB/sec)
|
*The best (fastest) value in the test is shown in red.
To demonstrate the superior performance of the PARTNER-Jet, KMC conducted tests by downloading (writing) a 1MB binary file into target FLASH memory and measured time for the operation. The test was done after clearing (zero-ing) the target FLASH memory. Furthermore, if the ICE allows to allocate working area in user RAM space, time using that working area is also measured. The time measurement was done is the same way with downloading a binary file.
Benchmark result #2: FLASH memory writing speed
ICE model
|
PARTNER-Jet
|
TRACE32
|
Writing time (using working RAM)
|
8.0 sec (128 KB/sec)
|
11.8 sec (86.8 KB/sec)
|
Writing time
|
22.1 sec (46.3 KB/sec)
|
798.4 sec (1.3 KB/sec)
|
*The best (fastest) value in the test is shown in red.
To demonstrate the superior performance of the PARTNER-Jet, KMC conducted tests by stepping through small NOP loop and measured time required for 1,000 steps in machine language level. The program code is as follows:
Address
|
Code 
|
Mnemonic
|
08100004
|
E1A00000
|
NOP
|
08100008
|
E1A00000
|
NOP
|
0810000C
|
E1A00000
|
NOP
|
08100010
|
E1A00000
|
NOP
|
08100014
|
E1A00000
|
NOP
|
08100018
|
EAFFFFF8
|
B 08100000
|
The initial program counter value was set to 0x08100000. The time measurement was done is the same way with downloading a binary file.
Benchmark result #3: Step execution speed
ICE model
|
PARTNER-Jet
|
TRACE32
|
Execution time
|
2.94 sec (340.1 steps/sec)
|
5.4 sec (185.1 steps/sec)
|
*The best (fastest) value in the test is shown in red.
KMC conducted tests by downloading a debug information of a Linux kernel (vm.out with debug information, approx. 27MB in size) to debugger and measured time for the operation.
Benchmark result #4: Debug information loading speed
ICE model
|
PARTNER-Jet
|
TRACE32
|
Execution time (27MB) |
2.86 sec
|
1.7 sec
|
Execution time with stabs compression(8MB) |
0.98 sec
|
0.5 sec
|
*The best (fastest) value in the test is shown in red.
*stabs is a standard data compression method.
KMC, the KMC logo, PARTNER, PARTNER-Jet, PARTNER-JetM, exeGCC, JTAG-Ether, and Eclipse for PARTNER Cross DevKit are trademarks of Kyoto Microcomputer Co., Ltd. Product and company names mentioned herein may be trademarks of their respective companies. Product specifications are subject to change without notice. This material is provided for informational purposes only; KMC assumes no liability related to its use.