bug while building

[Gelisi]

cd /home/spcen/project/FSL/EDA/DREAMPlace/build/thirdparty/OpenTimer && /usr/bin/c++ -D_GLIBCXX_USE_CXX11_ABI=0 -I/home/spcen/project/FSL/EDA/DREAMPlace/thirdparty/flute -I/home/spcen/project/FSL/EDA/DREAMPlace/thirdparty/OpenTimer -fPIC -O3 -DNDEBUG -std=c++17 -MD -MT thirdparty/OpenTimer/CMakeFiles/util.dir/unittest/utility.cpp.o -MF CMakeFiles/util.dir/unittest/utility.cpp.o.d -o CMakeFiles/util.dir/unittest/utility.cpp.o -c /home/spcen/project/FSL/EDA/DREAMPlace/thirdparty/OpenTimer/unittest/utility.cpp
In file included from /home/spcen/project/FSL/EDA/DREAMPlace/thirdparty/OpenTimer/unittest/utility.cpp:3:
/home/spcen/project/FSL/EDA/DREAMPlace/thirdparty/OpenTimer/unittest/doctest.h:4008:47: error: size of array ‘altStackMem’ is not an integral constant-expression
4008 | static char altStackMem[4 * SIGSTKSZ];
| ^
make[2]: *** [thirdparty/OpenTimer/CMakeFiles/util.dir/build.make:79: thirdparty/OpenTimer/CMakeFiles/util.dir/unittest/utility.cpp.o] Error 1
make[2]: Leaving directory '/home/spcen/project/FSL/EDA/DREAMPlace/build'
make[1]: *** [CMakeFiles/Makefile2:3819: thirdparty/OpenTimer/CMakeFiles/util.dir/all] Error 2
make[1]: Leaving directory '/home/spcen/project/FSL/EDA/DREAMPlace/build'
make: *** [Makefile:139: all] Error 2

Please teach me how to fix it. thanks

[limbo018]

You may check this issue for the solution.

#133

[Gelisi]

I have solve this problem, but another one is here.

[100%] Linking CXX executable abacus_unittest
cd /home/spcen/project/FSL/EDA/DREAMPlace/build/unittest/ops/abacus_legalize_unittest && /home/spcen/.local/lib/python3.10/site-packages/cmake/data/bin/cmake -E cmake_link_script CMakeFiles/abacus_unittest.dir/link.txt --verbose=1
/usr/bin/c++ -O3 -DNDEBUG CMakeFiles/abacus_unittest.dir/abacus_unittest.cpp.o -o abacus_unittest -Wl,-rpath,/usr/local/cuda/lib64:/home/spcen/.local/lib/python3.10/site-packages/torch/lib: ../../../dreamplace/ops/utility/libutility_cxx.a /usr/local/cuda/lib64/libcudart.so /home/spcen/.local/lib/python3.10/site-packages/torch/lib/libtorch.so /home/spcen/.local/lib/python3.10/site-packages/torch/lib/libc10.so /home/spcen/.local/lib/python3.10/site-packages/torch/lib/libtorch_cpu.so /home/spcen/.local/lib/python3.10/site-packages/torch/lib/libc10_cuda.so /home/spcen/.local/lib/python3.10/site-packages/torch/lib/libtorch_cuda.so
/usr/bin/ld: /usr/local/MATLAB/MATLAB_Runtime/v94/bin/glnxa64/libnccl.so.2: undefined reference to cudaSetupArgument' /usr/bin/ld: /home/spcen/.local/lib/python3.10/site-packages/torch/lib/libtorch_cuda.so: undefined reference to ncclRecv'
/usr/bin/ld: /home/spcen/.local/lib/python3.10/site-packages/torch/lib/libtorch_cuda.so: undefined reference to ncclCommAbort' /usr/bin/ld: /home/spcen/.local/lib/python3.10/site-packages/torch/lib/libtorch_cuda.so: undefined reference to ncclRedOpDestroy'
/usr/bin/ld: /home/spcen/.local/lib/python3.10/site-packages/torch/lib/libtorch_cuda.so: undefined reference to ncclRedOpCreatePreMulSum' /usr/bin/ld: /home/spcen/.local/lib/python3.10/site-packages/torch/lib/libtorch_cuda.so: undefined reference to ncclGetVersion'
/usr/bin/ld: /home/spcen/.local/lib/python3.10/site-packages/torch/lib/libtorch_cuda.so: undefined reference to ncclSend' /usr/bin/ld: /home/spcen/.local/lib/python3.10/site-packages/torch/lib/libtorch_cuda.so: undefined reference to ncclBroadcast'
/usr/bin/ld: /home/spcen/.local/lib/python3.10/site-packages/torch/lib/libtorch_cuda.so: undefined reference to ncclCommGetAsyncError' /usr/bin/ld: /home/spcen/.local/lib/python3.10/site-packages/torch/lib/libtorch_cuda.so: undefined reference to ncclGetLastError'
/usr/bin/ld: /usr/local/MATLAB/MATLAB_Runtime/v94/bin/glnxa64/libnccl.so.2: undefined reference to `cudaLaunch'
collect2: error: ld returned 1 exit status
make[2]: *** [unittest/ops/abacus_legalize_unittest/CMakeFiles/abacus_unittest.dir/build.make:107: unittest/ops/abacus_legalize_unittest/abacus_unittest] Error 1
make[2]: Leaving directory '/home/spcen/project/FSL/EDA/DREAMPlace/build'
make[1]: *** [CMakeFiles/Makefile2:6111: unittest/ops/abacus_legalize_unittest/CMakeFiles/abacus_unittest.dir/all] Error 2
make[1]: Leaving directory '/home/spcen/project/FSL/EDA/DREAMPlace/build'
make: *** [Makefile:139: all] Error 2

near 100%

[limbo018]

It looks like your libtorch_cuda wrongly links to the libnccl.so under MATLAB rather than cuda. Can you avoid /usr/local/MATLAB directory to be searched in $PATH?

[Gelisi]

Morning professor! Sincerely thank you for your help. I have run the demo and I get the placement of the demo adaptec1.log. (1) Is My dreamplace 4.1.0 ? But the project mentioned the HPWL is much lower than mine. Waiting for your help. It's my first time to do this topic. Sorry to bother you
adaptec1.log

[limbo018]

The log looks OK. You can use git status to see which commit the code is. The tool does not report the version number at runtime.

Which baseline are you comparing too? If you are comparing with some numbers in published papers, they may use other tools to report HPWL, like NTUplace3 or NTUplace4dr.

[Gelisi]

This result in your readme? I just git the code days before. I think it's the newest one?

[Gelisi]

But the result in log is about [INFO ] K-reorder time: 888.724 ms
[INFO ] Legality check takes 48.2037 ms
[INFO ] DREAMPlace - K-Reorder legal flag = 1
[INFO ] DREAMPlace - detailed placement takes 4.658 seconds
[INFO ] DREAMPlace - iteration 611, wHPWL 7.277018E+07, time 0.908ms
[INFO ] DREAMPlace - non-linear placement takes 17.55 seconds
[INFO ] DREAMPlace - writing to results/adaptec1/adaptec1.gp.pl
[INFO ] writing placement to results/adaptec1/adaptec1.gp.pl
[INFO ] DREAMPlace - write SolutionFileFormat.BOOKSHELF takes 0.327 seconds
[INFO ] DREAMPlace - placement takes 33.508 seconds

[limbo018]

I think you need to modify the benchmarks by removing the FIXED and FIXED_NI keywords in each case to make fixed terminals and IO pads movable.

@Constwelve Please double check if this is the correct way.

[Gelisi]

so download in this way is not correct?
How can I download the modified ISPD2005 benchmarks with all fixed macros and IO pads regarded as movable.

[limbo018]

See #195 (comment)

You need to modify the case a little bit.

[Gelisi]

where can I modify the benchmarks by removing the FIXED and FIXED_NI keywords…

[limbo018]

Open the input .pl, .nodes files with any text editors.

[Gelisi]

Here, do we need to remove the keyword "fixed" from all files ending with ".pl"? And in the "nodes" section, only the keyword "terminal" is present, so it doesn't need to be touched, right?

[Gelisi]

there are many .pl files

[Gelisi]

[INFO ] Kernel time: 968.246 ms
[INFO ] K-reorder time: 1211.18 ms
[INFO ] DREAMPlace - K-Reorder legal flag = 1
[INFO ] DREAMPlace - detailed placement takes 11.330 seconds
[INFO ] Legality check takes 48.5227 ms
[INFO ] DREAMPlace - iteration 637, wHPWL 1.011426E+08, time 1.338ms
[INFO ] writing placement to results/adaptec1/plot/iter0638.png
[WARNING] filename = results/adaptec1/plot/iter0638.png, width = 800, height = 799, file format = 3 not used, as DRAWPLACE not enabled
[I] plotting to results/adaptec1/plot/iter0638.png takes 10.976 seconds
[INFO ] DREAMPlace - plotting to results/adaptec1/plot/iter0638.png takes 10.991 seconds
[INFO ] DREAMPlace - non-linear placement takes 154.05 seconds
[INFO ] DREAMPlace - writing to results/adaptec1/adaptec1.gp.pl
[INFO ] writing placement to results/adaptec1/adaptec1.gp.pl
[INFO ] DREAMPlace - write SolutionFileFormat.BOOKSHELF takes 0.340 seconds
[INFO ] DREAMPlace - placement takes 169.947 seconds

Professor I have get the result, it seems that my result is the same as Dreamplace 4.0