compile liddriven 2D with GPU error message

[Pulauwin]

g++ -O3 -Wall -march=native -mtune=native -std=c++17 –generate-code=arch=compute_86,code=[compute_86,sm_86] –extended-lambda –expt-relaxed-constexpr -x cu -Xcudafe “–diag_suppress=implicit_return_from_non_void_function –display_error_number –diag_suppress=20014 –diag_suppress=20011” -DPLATFORM_GPU_CUDA -I../../../src -I../../../external/zlib -I../../../external/tinyxml -c -o cavity2d.o cavity2d.cpp
g++: error: unrecognized command line option ‘–generate-code=arch=compute_86,code=[compute_86,sm_86]’
g++: error: unrecognized command line option ‘–extended-lambda’
g++: error: unrecognized command line option ‘–expt-relaxed-constexpr’
g++: error: unrecognized command line option ‘-Xcudafe’
g++: error: unrecognized command line option ‘–diag_suppress=implicit_return_from_non_void_function –display_error_number –diag_suppress=20014 –diag_suppress=20011’

How to resolve this problem?

[Adrian]

You need to select nvcc as the compiler in config.mk.

Compare e.g. the example config config/gpu_only.mk included in the release.

I hope this helps!

[Pulauwin]

dfsdfs

[Pulauwin]

still has error, it seems my version is too low?
I sue cuda11.8 and gcc7.5.

src/utilities/benchmarkUtil.h(342): error: namespace “std” has no member “for_each_n”

[Adrian]

No, the latest CUDA version 11.8 works in my tests.

The issue here is that the selected C++ standard library headers don’t include the std::for_each_n despite C++17 being requested of the compiler.

You can apply the following change to get it to compile:

diff --git a/src/utilities/benchmarkUtil.h b/src/utilities/benchmarkUtil.h
index 0156bd1ce5..5806c19fa5 100644
--- a/src/utilities/benchmarkUtil.h
+++ b/src/utilities/benchmarkUtil.h
@@ -339,9 +339,9 @@ public:
   }
 
   void reset(T lowerBound, T upperBound, T dt) {
-    std::for_each_n(integrators.begin(),numComponents,[&](auto& integrator){
+    for (auto& integrator : integrators) {
       integrator->reset(lowerBound,upperBound,dt);
-    });
+    }
   }
 
   std::array<T,numComponents> getResult()

As this sadly seems to be a common issue also on our clusters this replacement of std::for_each_n usage will be included in the next release (also std::for_each_n is not actually needed here).

[Pulauwin]

Thanks a lot. it works now.
Can i compile the OpenLB into a shared lib so i can link it in my program.

[Adrian]

Good to hear!

Yes, this is possible (to some degree, depending on how you want to use the library) and will be included as a build option in the next release (enabling e.g. mixing compilers and speeding up the CUDA-enabled compilation).

[Author]

Posts