Compiling the 2020-12 release for cuda

[dosarudaniel]

Hi,

I've been trying to compile the current latest version of this repository but I encountered some problems with the support for this flag: -fsycl-targets in adding multiple targets in the list. So I turned my focus on the latest release of this repository because I would like to use a stable version of clang++ to compile my sycl code which I will like to run on NVIDIA back end.

I have followed these steps:

python ~/sycl_workspace/llvm/buildbot/configure.py --cuda # see here the full output
python ~/sycl_workspace/llvm/buildbot/compile.py # see here the full output

But the compilation failed at this point:

[2435/3069] Building CXX object tools/clang/lib/StaticAnalyzer/Frontend/CMakeFiles/obj.clangStaticAnalyzerFrontend.dir/AnalysisConsumer.cpp.o
ninja: build stopped: subcommand failed.
Traceback (most recent call last):
  File "/home/user123/sycl_workspace/llvm/buildbot/compile.py", line 67, in <module>
    ret = main()
  File "/home/user123/sycl_workspace/llvm/buildbot/compile.py", line 63, in main
    return do_compile(args)
  File "/home/user123/sycl_workspace/llvm/buildbot/compile.py", line 40, in do_compile
    subprocess.check_call(cmake_cmd, cwd=abs_obj_dir)
  File "/usr/lib/python2.7/subprocess.py", line 190, in check_call
    raise CalledProcessError(retcode, cmd)

Could you please point me to a fairly stable version of this repository which will compile and support the latest features for SYCL ?

Thanks!

[AlexeySachkov]

Hi @dosarudaniel,

Could you please point me to a fairly stable version of this repository which will compile and support the latest features for SYCL ?

Basically, almost each version should be stable enough as we have a number of pre-commit checks including builds and testing with CUDA (see any merged PR) as well as some post-commit checks (here: https://github.com/intel/llvm/actions/workflows/linux_post_commit.yml)

From logs you attached:

[2404/3069] Performing build step for 'ocl-icd'
FAILED: tools/sycl/ocl-icd-prefix/src/ocl-icd-stamp/ocl-icd-build lib/libOpenCL.so
cd /home/user123/sycl_workspace/llvm/build/tools/sycl/icd_build && /usr/local/bin/cmake -P /home/user123/sycl_workspace/llvm/build/tools/sycl/ocl-icd-prefix/src/ocl-icd-stamp/ocl-icd-build-Release.cmake && /usr/local/bin/cmake -E touch /home/user123/sycl_workspace/llvm/build/tools/sycl/ocl-icd-prefix/src/ocl-icd-stamp/ocl-icd-build
CMake Error at /home/user123/sycl_workspace/llvm/build/tools/sycl/ocl-icd-prefix/src/ocl-icd-stamp/ocl-icd-build-Release.cmake:49 (message):
Command failed: 1

'/usr/local/bin/cmake' '--build' '.'

See also

/home/user123/sycl_workspace/llvm/build/tools/sycl/ocl-icd-prefix/src/ocl-icd-stamp/ocl-icd-build-*.log

It looks like OpenCL ICD build failed, could you please also share those ocl-icd-build-*.log files so we can try to analyze what went wrong?

[dosarudaniel]

Thank you, @AlexeySachkov !

I've tried that, I am compiling this version of llvm repo: see my fork here. I used this value: 7edca72746c9fc5de5db0acd1fc245cb8ef1b29d for the GIT_TAG for OpenCL headers. (Most recent version before December 25th 2020). From the compiling output it seemed that it passed previous point, but now I have another problem related to SPIRV.:

/home/user123/sycl_workspace/llvm/llvm-spirv/lib/SPIRV/libSPIRV/SPIRVEntry.h: In member function ‘SPIRV::SPIRVContinuedInstINTELBase<OC>::EnableIfStruct<OPC, SPIRV::SPIRVType*> SPIRV::SPIRVContinuedInstINTELBase<OC>::getMemberType(size_t) const’:
/home/user123/sycl_workspace/llvm/llvm-spirv/lib/SPIRV/libSPIRV/SPIRVEntry.h:897:70: error: invalid static_cast from type ‘SPIRV::SPIRVEntry*’ to type ‘SPIRV::SPIRVType*’
     return static_cast<SPIRVType *>(SPIRVEntry::getEntry(Elements[I]));
                                                                      ^
[2471/3069] Building CXX object tools/sycl/source/CMakeFiles/sycl_object.dir/detail/builtins_math.cpp.o
ninja: build stopped: subcommand failed.
Traceback (most recent call last):
  File "/home/user123/sycl_workspace/llvm/buildbot/compile.py", line 67, in <module>
    ret = main()
  File "/home/user123/sycl_workspace/llvm/buildbot/compile.py", line 63, in main
    return do_compile(args)
  File "/home/user123/sycl_workspace/llvm/buildbot/compile.py", line 40, in do_compile
    subprocess.check_call(cmake_cmd, cwd=abs_obj_dir)
  File "/usr/lib/python2.7/subprocess.py", line 190, in check_call
    raise CalledProcessError(retcode, cmd)
subprocess.CalledProcessError: Command '['cmake', '--build', '/home/user123/sycl_workspace/llvm/build', '--', 'deploy-sycl-toolchain', 'deploy-opencl-aot', '-j', '32']' returned non-zero exit status 1

Again, I am following the same steps:
python ~/sycl_workspace/llvm/buildbot/configure.py --cuda # see here the full output
python ~/sycl_workspace/llvm/buildbot/compile.py # see here the full output

Do you have tips on this issue ? Let me know if you need more info.

[AlexeySachkov]

Do you have tips on this issue ?

@dosarudaniel, yes, it is a known issue which was fixed a week after that release tag was created, you need to apply the following patch to llvm-spiv directory: KhronosGroup/SPIRV-LLVM-Translator@cbeeb9b

[dosarudaniel]

Thank you @AlexeySachkov , I applied the patch on my fork, but now I have another issue:

[21/26] Building C object test/loader_test/CMakeFiles/icd_loader_test.dir/test_create_calls.c.o
In file included from /home/user123/sycl_workspace/llvm/build/tools/opencl-aot/OpenCL/icd/test/loader_test/param_struct.h:6,
                 from /home/user123/sycl_workspace/llvm/build/tools/opencl-aot/OpenCL/icd/test/loader_test/test_create_calls.c:9:
/home/user123/sycl_workspace/llvm/build/tools/opencl-aot/OpenCL/inc/CL/cl_gl_ext.h:18:9: note: #pragma message: All OpenGL-related extensions have been moved into cl_gl.h.  Please include cl_gl.h directly.
 #pragma message("All OpenGL-related extensions have been moved into cl_gl.h.  Please include cl_gl.h directly.")
         ^~~~~~~
ninja: build stopped: subcommand failed.
[2545/3069] Building CXX object tools/llvm-spirv/lib/SPIRV/CMakeFiles/LLVMSPIRVLib.dir/libSPIRV/SPIRVType.cpp.o
/home/user123/sycl_workspace/llvm/llvm-spirv/lib/SPIRV/libSPIRV/SPIRVType.cpp: In member function ‘SPIRV::SPIRVWord SPIRV::SPIRVType::getMatrixColumnCount() const’:
/home/user123/sycl_workspace/llvm/llvm-spirv/lib/SPIRV/libSPIRV/SPIRVType.cpp:119:56: warning: type qualifiers ignored on cast result type [-Wignored-qualifiers]
   return static_cast<const SPIRVTypeMatrix *const>(this)->getColumnCount();
                                                        ^
/home/user123/sycl_workspace/llvm/llvm-spirv/lib/SPIRV/libSPIRV/SPIRVType.cpp: In member function ‘SPIRV::SPIRVType* SPIRV::SPIRVType::getMatrixColumnType() const’:
/home/user123/sycl_workspace/llvm/llvm-spirv/lib/SPIRV/libSPIRV/SPIRVType.cpp:124:56: warning: type qualifiers ignored on cast result type [-Wignored-qualifiers]
   return static_cast<const SPIRVTypeMatrix *const>(this)->getColumnType();
                                                        ^
[2557/3069] Building CXX object tools/sycl/source/CMakeFiles/sycl_object.dir/detail/builtins_math.cpp.o
ninja: build stopped: subcommand failed.
Traceback (most recent call last):
  File "/home/user123/sycl_workspace/llvm/buildbot/compile.py", line 67, in <module>
    ret = main()
  File "/home/user123/sycl_workspace/llvm/buildbot/compile.py", line 63, in main
    return do_compile(args)
  File "/home/user123/sycl_workspace/llvm/buildbot/compile.py", line 40, in do_compile
    subprocess.check_call(cmake_cmd, cwd=abs_obj_dir)
  File "/usr/lib/python2.7/subprocess.py", line 190, in check_call
    raise CalledProcessError(retcode, cmd)
subprocess.CalledProcessError: Command '['cmake', '--build', '/home/user123/sycl_workspace/llvm/build', '--', 'deploy-sycl-toolchain', 'deploy-opencl-aot', '-j', '32']' returned non-zero exit status 1

python ~/sycl_workspace/llvm/buildbot/compile.py # see here the full output for 3rd trial.

Is this also a known issue?

[AlexeySachkov]

Is this also a known issue?

It is again an issue with OpenCL Loader: two of our sub-components use it and I guess that you need to update GIT_TAG in one more cmake file here:

llvm/opencl-aot/CMakeLists.txt

Line 20 in 5eebd1e

GIT_TAG origin/master

P.S. this situation should be improved in #3485, i.e. we will have a single place where versions of those dependencies are specified

[dosarudaniel]

Thank you, @AlexeySachkov, for your help!
I managed to compile it now.
I used this version of my llvm fork.

[AlexeySachkov]

So I turned my focus on the latest release of this repository because I would like to use a stable version of clang++ to compile my sycl code which I will like to run on NVIDIA back end.

BTW, @dosarudaniel, the new release is coming and commit hash is already known, see PR with release notes: https://github.com/intel/llvm/pull/3590/files#diff-34d07c751527855529b6a40917a72f3e46983c6abda27765486594794dd0e19dR3, it is 6a49170 - I would assume that newer commit already has all required fixes for issues you have encountered (but it might introduce new ones :))