Containing dozens of real-world and synthetic tests, CoreMark®-PRO (2015) is an industry-standard benchmark that measures the multi-processor performance of central processing units (CPU) and embedded microcrontrollers (MCU)
License: Other
Hi,
I have seen the video tutorial of CoreMark-PRO in YouTube, and I know how to transplant one workload to the bare-metal. But I don’t understand how to transplant nine workload to a bare-metal, did I need to download and run the workload one by one? Run 9 times with respective workload?
If so, what should I do to allocate one workload to multicore? Such as Stm32H747, which has two cores, Cortex-M4 and M7, how can I make two cores work simultaneously to calculate the workload? If I can only use one core each time, it’s obvious that I don’t fully use the resources.
Hi,
I am running Coremark-pro on ARM by running individual binary and i believe you are using following
1000 x geomean( ( subtestscore / reference score ) * scale factor)
but i am not able to understand how you are calculating subtestscore and reference score
I am attaching output for cjpeg-rose7-preset.exe
pankaj@pankaj~# ./cjpeg-rose7-preset.exe
Rose256.bmp data supplied by C array
Data Set : Rose256.bmp
Output File : Rose256.jpg
Rose256.bmp data supplied by C array
Data Set : Rose256.bmp
Output File : Rose256.jpg
Rose256.bmp data supplied by C array
Data Set : Rose256.bmp
Output File : Rose256.jpg
Rose256.bmp data supplied by C array
Data Set : Rose256.bmp
Output File : Rose256.jpg
Rose256.bmp data supplied by C array
Data Set : Rose256.bmp
Output File : Rose256.jpg
Rose256.bmp data supplied by C array
Data Set : Rose256.bmp
Output File : Rose256.jpg
Rose256.bmp data supplied by C array
Data Set : Rose256.bmp
Output File : Rose256.jpg
Can you please help me from above result how we can get subtestscore and reference score and also following ???
Thanks
Hi,
Anybody can tell me that how can i build with msvc.
please do the needful.
I use "make TARGET=linux64 build".
The following error occurs:
/usr/bin/gcc -c -g -O2 -DNDEBUG -DHOST_EXAMPLE_CODE=1 -std=gnu99 -DHAVE_SYS_STAT_H=1 -DUSE_NATIVE_PTHREAD=1 -DGCC_INLINE_MACRO=1 -DNO_RESTRICT_QUALIFIER=1 -DEE_SIZEOF_LONG=8 -DEE_SIZEOF_PTR=8 -DEE_PTR_ALIGN=8 -DHAVE_SYS_STAT_H=1 -DUSE_NATIVE_PTHREAD=1 -DGCC_INLINE_MACRO=1 -DNO_RESTRICT_QUALIFIER=1 -DEE_SIZEOF_LONG=8 -DEE_SIZEOF_PTR=8 -DEE_PTR_ALIGN=8 -DEE_SIZEOF_INT=4 -DEE_SIZEOF_LONG=8 -Wall -Wno-long-long -fno-asm -fsigned-char -DUSE_FP32 -I/home/tyliu/project/coremark-pro/mith/include -I/home/tyliu/project/coremark-pro/mith/al/include -I/home/tyliu/project/coremark-pro/mith/al/include /home/tyliu/project/coremark-pro/workloads/linear_alg-mid-100x100-sp/linear_alg-mid-100x100-sp.c -o linear_alg-mid-100x100-sp.o
cd /home/tyliu/project/coremark-pro/builds/linux64/gcc64/obj/bench/fp/linpack/SP && make -f /home/tyliu/project/coremark-pro/benchmarks/fp/linpack/SP/../Makefile build
make[3]: Entering directory '/home/tyliu/project/coremark-pro/builds/linux64/gcc64/obj/bench/fp/linpack/SP'
make[3]: /home/tyliu/project/coremark-pro/benchmarks/fp/linpack/SP/../Makefile: No such file or directory
make[3]: *** No rule to make target '/home/tyliu/project/coremark-pro/benchmarks/fp/linpack/SP/../Makefile'. Stop.
make[3]: Leaving directory '/home/tyliu/project/coremark-pro/builds/linux64/gcc64/obj/bench/fp/linpack/SP'
/home/tyliu/project/coremark-pro/workloads/linear_alg-mid-100x100-sp//Makefile:71: recipe for target '/home/tyliu/project/coremark-pro/builds/linux64/gcc64/obj/bench/fp/linpack/SP/done.build' failed
make[2]: *** [/home/tyliu/project/coremark-pro/builds/linux64/gcc64/obj/bench/fp/linpack/SP/done.build] Error 2
make[2]: Leaving directory '/home/tyliu/project/coremark-pro/builds/linux64/gcc64/obj/workloads/linear_alg-mid-100x100-sp'
hi there,
coremarkPro with cygwin TARGET=linux64 is ok to run,
but when I try to port package to bare-metal env. without file I/O,
cjpeg always fail with "Cannot find file Rose256.bmp".
In "coremark-pro/workloads/cjpeg-rose7-preset/Makefile"
line 5 export SELECT_PRESET_ID=0
in "coremark-pro/benchmarks/consumer_v2/cjpeg/bmark_lite.c"
line 193-204
SELECT_PRESET_ID==1 maps to Rose256
SELECT_PRESET_ID==2 maps to goose
The mismatch cause failure.
(in cygwin, strace shows cjpeg-rose7-reset.exe uses file I/O system call to get data, not from internal preset buffer)
After correcting the Makefile, cjpeg is passed without error.
Hi,
Not able to make build with "gcc-arm-none-eabi-10.3-2021.10"
I changed gcc64.mak and common.mak file to add gcc-arm-none-eabi-10.3-2021.10 path. These files are present in util/make directory.
Changes mainly for below lines:
TOOLS = ./coremark-pro-main/gcc-arm/x86_64/gcc-arm-none-eabi-10.3-2021.10
CC = $(TOOLS)/bin/arm-none-eabi-gcc
AS = $(TOOLS)/bin/arm-none-eabi-as
LD = $(TOOLS)/bin/arm-none-eabi-gcc
AR = $(TOOLS)/bin/arm-none-eabi-ar
INCLUDE = $(TOOLS)/arm-none-eabi/include
ARM gcc-arm-none-eabi-10.3-2021.10 package installed from ARM download link
Command= make TARGET=linux64 build-all
Error Message=
make[3]: Leaving directory ./coremark-pro-main/builds/linux64/gcc64/obj/bench/darkmark/zip' ./coremark-pro-main/gcc-arm/x86_64/gcc-arm-none-eabi-10.3-2021.10/bin/arm-none-eabi-gcc -o./coremark-pro-main/builds/linux64/gcc64/bin/zip-test.exe zip-test.o ./coremark-pro-main/builds/linux64/gcc64/obj/bench/darkmark/zip/*.o ./coremark-pro-main/builds/linux64/gcc64/obj/mith.a -lm -lrt -lpthread -lm -lrt -lpthread ./coremark-pro-main/gcc-arm/x86_64/gcc-arm-none-eabi-10.3-2021.10/bin/../lib/gcc/arm-none-eabi/10.3.1/../../../../arm-none-eabi/bin/ld: cannot find -lrt ./coremark-pro-main/gcc-arm/x86_64/gcc-arm-none-eabi-10.3-2021.10/bin/../lib/gcc/arm-none-eabi/10.3.1/../../../../arm-none-eabi/bin/ld: cannot find -lpthread ./coremark-pro-main/gcc-arm/x86_64/gcc-arm-none-eabi-10.3-2021.10/bin/../lib/gcc/arm-none-eabi/10.3.1/../../../../arm-none-eabi/bin/ld: cannot find -lrt ./coremark-pro-main/gcc-arm/x86_64/gcc-arm-none-eabi-10.3-2021.10/bin/../lib/gcc/arm-none-eabi/10.3.1/../../../../arm-none-eabi/bin/ld: cannot find -lpthread collect2: error: ld returned 1 exit status make[2]: *** [./coremark-pro-main/builds/linux64/gcc64/bin/zip-test.exe] Error 1 make[2]: Leaving directory ./coremark-pro-main/builds/linux64/gcc64/obj/workloads/zip-test'
make[1]: *** [build-all] Error 2
make[1]: Leaving directory `./coremark-pro-main'
make: *** [build] Error 2
Thanks,
SK
I would like to know if I run the benchmarks individually and not for the certify mark, do I still need to add -v0 ? Since from the manual it says
"A value of 0 indicates performance mode, a value of 1 indicates verification mode, which invokes
result checking."
Thanks!
Is there a simple sh script that can be leveraged on embedded targets for running cross-compiled executables with necessary flags. This would be really helpful for for embedded development where targets rarely have the development tools the current process requires.
Hi team,
To test NXP imx rt1170 board performance i need to run the coremark-pro benchmark on the board.
To run the coremark on imx rt1170 which tools(SDK) I need to use to flash and is there any reference guide to run the benchmark on rt1170 boards.
Thanks
chaithanya
Hi,
i am able to build coremark-pro with Ubuntu with "gcc-linaro-11.3.1-2022.06-x86_64_arm-linux-gnueabihf".
Could you please suggest recommended cortex-M4 compiler option for coremark-pro.
From time to time, when compiling with "make -j", we have a race condition in Makefile, with messages like:
Fatal error: can't create al/src/th_al.o: No such file or directory
It seems the following patch solves the issue:
diff --git a/mith/Makefile b/mith/Makefile
index faeaea1..71a6f9d 100644
--- a/mith/Makefile
+++ b/mith/Makefile
@@ -54,4 +54,4 @@ print-%:
@echo $* = $($*)
@echo [Defined at $(origin $*)]
-$(TH_OBJS) : $(TH_HDRS)
\ No newline at end of file
+$(TH_OBJS) : $(TH_HDRS) $(MYDIRS)
When trying to run the benchmark on a reference board with the e6500 CPU, the test loops-all-mid-10k-sp.exe fails to run with an 'illegal instruction' error.
Trying to debug this, I find that the error is on line 333 running this command: gdb -ex=r ./builds/linux64/gcc64/bin/loops-all-mid-10k-sp.exe
Logs:
Reading symbols from ./builds/linux64/gcc64/bin/loops-all-mid-10k-sp.exe...
Starting program: /run/media/mmcblk0p2/coremark-pro/builds/linux64/gcc64/bin/loops-all-mid-10k-sp.exe
[Thread debugging using libthread_db enabled]
Using host libthread_db library "/lib64/libthread_db.so.1".
Program received signal SIGILL, Illegal instruction.
0x00000000404a7448 in define_params_loops (idx=<optimized out>, name=<optimized out>, dataset=<optimized out>) at /<redacted>/benchmarks/fp/loops/loops.c:333
333 int m2=(int)th_sqrt((e_fp)(params->N*4));
Taking a closer look at this line after decompiling the frame, I can see a number of instructions:
gdb -batch -ex "disassemble/rs define_params_loops" ./builds/linux64/gcc64/bin/loops-all-mid-10k-sp.exe | less
logs:
333 int m2=(int)th_sqrt((e_fp)(params->N*4));
0x000000000000b420 <+688>: 80 fe 00 14 lwz r7,20(r30)
0x000000000000b424 <+692>: 3d 22 ff fd addis r9,r2,-3
0x000000000000b428 <+696>: c0 09 ec 08 lfs f0,-5112(r9)
0x000000000000b42c <+700>: 78 e9 10 28 rldic r9,r7,2,32
0x000000000000b430 <+704>: f9 21 00 88 std r9,136(r1)
0x000000000000b434 <+708>: 60 42 00 00 ori r2,r2,0
0x000000000000b438 <+712>: c8 21 00 88 lfd f1,136(r1)
0x000000000000b43c <+716>: fd 80 0e 9c fcfid f12,f1
0x000000000000b440 <+720>: fc 20 60 18 frsp f1,f12
0x000000000000b444 <+724>: fc 01 00 00 fcmpu cr0,f1,f0
0x000000000000b448 <+728>: ef e0 08 2c fsqrts f31,f1
0x000000000000b44c <+732>: 41 80 03 44 blt 0xb790 <define_params_loops+1568>
0x000000000000b450 <+736>: ff e0 f8 1e fctiwz f31,f31
0x000000000000b454 <+740>: 39 01 00 80 addi r8,r1,128
0x000000000000b458 <+744>: 38 c0 00 20 li r6,32
0x000000000000b45c <+748>: 7c c9 03 a6 mtctr r6
0x000000000000b460 <+752>: 7c ea 07 b4 extsw r10,r7
0x000000000000b464 <+756>: 39 3e 00 18 addi r9,r30,24
0x000000000000b468 <+760>: 7f e0 47 ae stfiwx f31,0,r8
0x000000000000b46c <+764>: e9 1e 01 1e lwa r8,284(r30)
0x000000000000b470 <+768>: 80 a1 00 80 lwz r5,128(r1)
I have a feeling the issue might be the blt call, but I'm not 100% certain. Any thoughts or help would be useful.
In the main() of workload "cjpeg-rose7-preset", I see a lot of calls to helper_cjpegrose7preset() where repeats_override is given 1. Then at this line, leading to these lines, we know tcdef->iterations is thus set to 1.
When this workload is run, kernel t_run_test_cjpeg() is executed.
According to this line, override is 1.
Therefore tcdef->rec_iterations will not be used (this line).
tcdef->rec_iterations is configured here and is supposed to be the "recommended" iteration count according to this and this.
I wonder, why is repeats_override not given 0 to have the recommended value used?
(BTW, the kernel of other workloads does not use tcdef->iterations.)
According to the documentation the coremark pro score should be calculated with:
100 x geomean( ( subtestscore / reference score ) * scale factor)
but it's in fact the produced score seems to be:
1000 x geomean( ( subtestscore / reference score ) * scale factor)
The produced score seems to be 10x what it should be
Line 132 in 4832cc6
Hello,
I'm trying to cross compile the benchmark for ARM AARCH64 .. I'm using: gcc-7.3
I basically changed the following about util/make/gcc64.mk:
CC = $(COMPILER_HOME)/bin/aarch64-buildroot-linux-gnu-gcc
AS = $(COMPILER_HOME)/bin/aarch64-buildroot-linux-gnu-as
LD = $(COMPILER_HOME)/bin/aarch64-buildroot-linux-gnu-ld
AR = $(COMPILER_HOME)/bin/aarch64-buildroot-linux-gnu-ar
Also:
INCLUDE = $(COMPILER_HOME)/include/
However, the following error occurs when I use make TARGET=linux64 build:
coremark-pro/coremark-pro/builds/linux64/gcc64/obj/mith.a(th_lib.o): undefined reference to symbol 'vsnprintf@@GLIBC_2.17' /project/codelink/tools/toolchains/mentor/aarch64/gcc-7.3.0/bin/../aarch64-buildroot-linux-gnu/sysroot/lib64/libc.so.6: error adding symbols: DSO missing from command line
Any idea how to resolve this?
Thank you in advance.
I'm doing some research related to LLVM and need the LLVM IR of each individual benchmark. This means I have to bypass the Makefile and use clang manually to generate IR from the relevant C files. I've managed to do this but when I then compile the IR to binary and run the benchmark workload, the workload prints that there's been an error. If I compile with make then the benchmarks all run fine.
For example, for the loop kernels benchmark, in the directory coremark-pro/workloads/loops-all-mid-10k-sp, I run the following:
clang -O3 -S -emit-llvm -I ../../mirth/include -I ../../mirth/al/include -I ../../benchmarks/fp/loops -o loops-all-mid-10k-sp.ll
I also defined USE_FP32=1.
I run the same clang command for all the .c files in /mirth/src, /mirth/al/src and /benchmarks/fp/loops before combining all the .ll files into one using llvm-link. I then compile the single IR file to a binary and execute, and get the following:
$ ./a.out -v0
- Info: Starting Run...
-- Workload:loops-all-mid-10k-sp=1814569103
-- loops-all-mid-10k-sp:time(ns)=7000
-- loops-all-mid-10k-sp:ERRORS=50
-- loops-all-mid-10k-sp:contexts=1
-- loops-all-mid-10k-sp:iterations=50
-- loops-all-mid-10k-sp:time(secs)= 7
-- loops-all-mid-10k-sp:secs/workload= 0.14
-- loops-all-mid-10k-sp:workloads/sec= 7.14286
-- accbits:min=0
-- accbits:max=0
-- accbits:avg=0
-- Done:loops-all-mid-10k-sp=1814569103
What am I missing in my build process to make this work? I've tried going through coremark's own make files but can't figure it out. Any help would be appreciated, thanks.
I run CoreMark Pro on Cygwin in Windows 11.
When testing parser-125k multicore, I see that it didn't use all CPU threads like the other testing.
As you suggest "if you had used arm-linux-eabi then it would have compiled. This is unrelated to the benchmark. Here is a good explanation of the three components in the toolchain name."
I have downloaded GNU tool chain version 11.3 “https://snapshots.linaro.org/gnu-toolchain/ but getting the below error.
./linaro/gcc-linaro-11.3.1-2022.06-x86_64_arm-linux-gnueabihf/bin/../lib/gcc/arm-linux-gnueabihf/11.3.1/../../../../arm-linux-gnueabihf/bin/ld: /lib64/libc.so.6: version `GLIBC_2.27' not found (required by ./linaro/gcc-linaro-11.3.1-2022.06-x86_64_arm-linux-gnueabihf/bin/../lib/gcc/arm-linux-gnueabihf/11.3.1/../../../../arm-linux-gnueabihf/bin/ld)
I have downloaded “sysroot-glibc-linaro-2.34-2022.06-arm-linux-gnueabihf” from the gnu-tool chain package and changed the path but still getting GLIB_2.27 not found error.
please help me on this.
Hi,
iam running coremark-pro in arm based hardware which is 64-bit and 32-bit supported board.
when i ran in specifying the TARGET=linux64 it is computing the score but when i changed it to TARGET=linux
the overall score is not generating.
React
A declarative, efficient, and flexible JavaScript library for building user interfaces.
🖖 Vue.js is a progressive, incrementally-adoptable JavaScript framework for building UI on the web.
Typescript
TypeScript is a superset of JavaScript that compiles to clean JavaScript output.
An Open Source Machine Learning Framework for Everyone
Django
The Web framework for perfectionists with deadlines.
Laravel
A PHP framework for web artisans
Bring data to life with SVG, Canvas and HTML. 📊📈🎉
JavaScript (JS) is a lightweight interpreted programming language with first-class functions.
Some thing interesting about web. New door for the world.
A server is a program made to process requests and deliver data to clients.
Machine learning is a way of modeling and interpreting data that allows a piece of software to respond intelligently.
Some thing interesting about visualization, use data art
Some thing interesting about game, make everyone happy.
Facebook
We are working to build community through open source technology. NB: members must have two-factor auth.
Microsoft
Open source projects and samples from Microsoft.
Google
Google ❤️ Open Source for everyone.
Alibaba
Alibaba Open Source for everyone
D3
Data-Driven Documents codes.
Tencent
China tencent open source team.