Benchmarking: Coremark

Decompress

root@jph1:~# tar zxvf coremark_v1.0
cd coremark_v1.0

Adjust compilation flags

Seems to be in ./linux64/core_portme.mak

• Change PORT_CFLAGS value to -O3
• Add extra flags to the end of the CFLAGS statement

For example (added -funroll-all-loops --param max-inline-insns-auto=200):

CC = gcc
# Flag: CFLAGS
#       Use this flag to define compiler options. Note, you can add compiler options from the command line using XCFLAGS="other flags"
PORT_CFLAGS = -O3
FLAGS_STR = "$(PORT_CFLAGS) $(XCFLAGS) $(XLFLAGS) $(LFLAGS_END)"
CFLAGS = $(PORT_CFLAGS) -I$(PORT_DIR) -I. -DFLAGS_STR=\"$(FLAGS_STR)\" -funroll-all-loops --param max-inline-insns-auto=200

Run benchmark (single thread)

Run benchmark with make:

root@jph1:~/coremark_v1.0# pwd
/root/coremark_v1.0
root@jph1:~/coremark_v1.0# make

Faster systems

If you have a fast system (faster than a RaspberryPi), increase the number of iterations to increase the test run to give a more meaningful result:

root@jph1:~/coremark_v1.0# make ITERATIONS=10000

32-bit ARM systems

Coremark seems to have an issue when not running in a linux64 environment (e.g. 32-bit ARM). Add a flag to the make command:

root@jph1:~/coremark_v1.0# make PORT_DIR=simple

Run benchmark (multi thread)

root@jph1:~/coremark_v1.0# make XCFLAGS="-DMULTITHREAD=8 -DUSE_PTHREAD"

Results

Coremark V1.0 results

CPU	Freq	Cores	Coremark	Coremark/MHz	Coremark/Core	Parallel Execution	Compile Flags
MIPS
Cavium Octeon II CN6880	2.50GHz	10	25671	10.268	2567	10 Threads
Cavium Octeon II CN6880	2.50GHz	10	31172	12.468	3117	10 Threads	-O3 -funroll-all-loops --param max-inline-insns-auto=200
ARM
Calxeda Highbank (Cortex A9)	1.40GHz	4	???	???	2941	1 Thread