Can't build with CUDA OpenCL

[HoaSoup]

Dear Developers,

I am compiling ROSETTA source code release 371 with the following command:
./scons.py -j20 mode=release extras=opencl bin

I stumbled this error:
g++ -o build/src/release/linux/6.8/64/x86/gcc/13/opencl/protocols/indel/IndelOptimizationMover.os -c -std=c++0x -ffor-scope -isystem external/boost_submod/ -isystem external/ -isystem external/include/ -isystem external/dbio/ -isystem external/libxml2/include -isystem external/rdkit -pipe -Wall -Wextra -pedantic -Wno-long-long -Wno-strict-aliasing -march=core2 -mtune=generic -O3 -ffast-math -fno-finite-math-only -funroll-loops -finline-functions -finline-limit=20000 -s -Wno-unused-variable -Wno-unused-parameter -fPIC -DBOOST_ERROR_CODE_HEADER_ONLY -DBOOST_SYSTEM_NO_DEPRECATED -DBOOST_MATH_NO_LONG_DOUBLE_MATH_FUNCTIONS -DBOOST_DISABLE_THREADS -DPTR_STD -DNDEBUG -DUSEOPENCL -Isrc -Iexternal/include -Isrc/platform/linux/64/gcc/13 -Isrc/platform/linux/64/gcc -Isrc/platform/linux/64 -Isrc/platform/linux -I/opt/AMDAPP/include -I/opt/AMDAPP/include/CL -I/opt/AMDAPP/lib/x86 -I/opt/AMDAPP/lib/x86_64 src/protocols/indel/IndelOptimizationMover.cc g++: warning: switch '-ffor-scope' is no longer supported In file included from src/core/scoring/sc/MolecularSurfaceCalculator.hh:47, from src/protocols/hotspot_hashing/SurfaceSearchPattern.cc:26: src/basic/gpu/GPU.hh:28:10: fatal error: CL/cl_platform.h: No such file or directory 28 | #include <CL/cl_platform.h> | ^~~~~~~~~~~~~~~~~~ compilation terminated.

I then added the path to the CL include directory of CUDA 12.6 into the environment variable CPLUS_INCLUDE_PATH. As a test, I compiled some code that used CL/cl_platform.h. It went successfully, showing that g++ can find the header file. Yet, the compilation error of ROSETTA still persisted.

Kind regards,
HS

[vmullig]

Note that almost nothing in Rosetta benefits from CUDA or OpenCL. The simplest solution is just to do a default build.

…

On Sat, Dec 21, 2024, 8:48 PM HoaSoup ***@***.***> wrote: Dear Developers, I am compiling ROSETTA source code release 371 with the following command: ./scons.py -j20 mode=release extras=opencl bin I stumbled this error: g++ -o build/src/release/linux/6.8/64/x86/gcc/13/opencl/protocols/indel/IndelOptimizationMover.os -c -std=c++0x -ffor-scope -isystem external/boost_submod/ -isystem external/ -isystem external/include/ -isystem external/dbio/ -isystem external/libxml2/include -isystem external/rdkit -pipe -Wall -Wextra -pedantic -Wno-long-long -Wno-strict-aliasing -march=core2 -mtune=generic -O3 -ffast-math -fno-finite-math-only -funroll-loops -finline-functions -finline-limit=20000 -s -Wno-unused-variable -Wno-unused-parameter -fPIC -DBOOST_ERROR_CODE_HEADER_ONLY -DBOOST_SYSTEM_NO_DEPRECATED -DBOOST_MATH_NO_LONG_DOUBLE_MATH_FUNCTIONS -DBOOST_DISABLE_THREADS -DPTR_STD -DNDEBUG -DUSEOPENCL -Isrc -Iexternal/include -Isrc/platform/linux/64/gcc/13 -Isrc/platform/linux/64/gcc -Isrc/platform/linux/64 -Isrc/platform/linux -I/opt/AMDAPP/include -I/opt/AMDAPP/include/CL -I/opt/AMDAPP/lib/x86 -I/opt/AMDAPP/lib/x86_64 src/protocols/indel/IndelOptimizationMover.cc g++: warning: switch '-ffor-scope' is no longer supported In file included from src/core/scoring/sc/MolecularSurfaceCalculator.hh:47, from src/protocols/hotspot_hashing/SurfaceSearchPattern.cc:26: src/basic/gpu/GPU.hh:28:10: fatal error: CL/cl_platform.h: No such file or directory 28 | #include <CL/cl_platform.h> | ^~~~~~~~~~~~~~~~~~ compilation terminated. I then added the path to the CL include directory of CUDA 12.6 into the environment variable CPLUS_INCLUDE_PATH. As a test, I compiled some code that used CL/cl_platform.h. It went successfully, showing that g++ can find the header file. Yet, the compilation error of ROSETTA still persisted. Could you help me? Kind regards, HS — Reply to this email directly, view it on GitHub <#283>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/ABACZUG2WNGFY6QUUVL5DUD2GYK5PAVCNFSM6AAAAABUBEY2ZWVHI2DSMVQWIX3LMV43ERDJONRXK43TNFXW4OZXG4ZDKOJSHE> . You are receiving this because you are subscribed to this thread.Message ID: ***@***.***>

[HoaSoup]

Thank you for your answer! Default build worked. No more issue for me.

[vmullig]

I should have added: the most multithreading support comes from extras=cxx11thread, but even that is limited. (Interaction graph precomputation is efficiently multithreaded, as is peptide conformational sampling with simple_cycpep_predict and a few other things, but that is about it.) Mostly, we have relied either on launching many parallel instances of Rosetta or on MPI processes for parallelism. Efforts to make Rosetta compatible with GPUs have largely moved on to successor projects (Tmol and Masala). Oh, there is limited support for a few Tensorflow models (which can be executed on GPU) with extras=tensorflow or extras=tensorflow_gpu.

…

On Sun, Dec 22, 2024, 10:18 PM Hoa Thanh Le ***@***.***> wrote: Thank you for your answer! Default build worked. No more issue for me. — Reply to this email directly, view it on GitHub <#283 (reply in thread)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/ABACZUCTHR4HVGRLOAM5XYT2G56J3AVCNFSM6AAAAABUBEY2ZWVHI2DSMVQWIX3LMV43URDJONRXK43TNFXW4Q3PNVWWK3TUHMYTCNRUGY2DCMI> . You are receiving this because you commented.Message ID: ***@***.***>