Fix ubench for CUDA provider

[vinser52]

Description

This PR fixes the compilation error during ubench compilation when -DUMF_BUILD_CUDA_PROVIDER=ON -DUMF_BUILD_LEVEL_ZERO_PROVIDER=OFF.

Here is the compilation error:

[ 24%] Built target umf-ubench
In file included from /nfs/pdx/home/svinogra/unified-memory-framework/benchmark/multithread.cpp:10:
/nfs/pdx/home/svinogra/unified-memory-framework/benchmark/multithread.hpp:24:10: fatal error: multithread_helpers.hpp: No such file or directory
   24 | #include "multithread_helpers.hpp"
      |          ^~~~~~~~~~~~~~~~~~~~~~~~~
compilation terminated.
make[2]: *** [benchmark/CMakeFiles/umf-multithreaded.dir/build.make:76: benchmark/CMakeFiles/umf-multithreaded.dir/multithread.cpp.o] Error 1
make[1]: *** [CMakeFiles/Makefile2:3034: benchmark/CMakeFiles/umf-multithreaded.dir/all] Error 2
make[1]: *** Waiting for unfinished jobs....

@lukaszstolarczuk @lplewa Looks like we do not test such a combination of CMake flags

Checklist

• Code compiles without errors locally
• All tests pass locally
• CI workflows execute properly

[lukaszstolarczuk]

@lukaszstolarczuk @lplewa Looks like we do not test such a combination of CMake flags

I guess we'll never test all possible CMake flags combinations...
Since this is "just a benchmark" issue we should rather be good not adding new CI job specifically for this combination.

[vinser52]

@lukaszstolarczuk @lplewa Looks like we do not test such a combination of CMake flags

I guess we'll never test all possible CMake flags combinations... Since this is "just a benchmark" issue we should rather be good not adding new CI job specifically for this combination.

Yeah, but I think in that particular case we just miss the -DUMF_BUILD_BENCHMARKS_MT=ON flag for the CUDA flow. As I can see, currently the CUDA flow uses the following flags:

cmake .. -DCMAKE_BUILD_TYPE=Release -DCMAKE_C_COMPILER=gcc -DCMAKE_CXX_COMPILER=g++ -DUMF_BUILD_SHARED_LIBRARY=ON -DUMF_BUILD_BENCHMARKS=ON -DUMF_BUILD_TESTS=ON -DUMF_BUILD_GPU_TESTS=ON -DUMF_BUILD_GPU_EXAMPLES=ON -DUMF_DEVELOPER_MODE=ON -DUMF_BUILD_LIBUMF_POOL_DISJOINT=ON -DUMF_BUILD_LIBUMF_POOL_JEMALLOC=ON -DUMF_BUILD_CUDA_PROVIDER=OFF -DUMF_BUILD_LEVEL_ZERO_PROVIDER=OFF -DUMF_BUILD_CUDA_PROVIDER=ON -DUMF_TESTS_FAIL_ON_SKIP=ON

[lukaszstolarczuk]

hmm if your talking about GPU flow, then we could enable it, yes - you can add that, please, even in this PR :)
// I'm just not sure why we disabled the mt benchmark execution, BTW (https://github.com/oneapi-src/unified-memory-framework/blob/main/.github/workflows/reusable_gpu.yml#L117)

on the other hand, in the "real" benchmarks workflow we have all flags set
https://github.com/oneapi-src/unified-memory-framework/blob/main/.github/workflows/reusable_benchmarks.yml#L89

[vinser52]

hmm if your talking about GPU flow, then we could enable it, yes - you can add that, please, even in this PR :)

Done. Enabled the -DUMF_BUILD_BENCHMARKS_MT=ON in GPU CIs.

// I'm just not sure why we disabled the mt benchmark execution, BTW (https://github.com/oneapi-src/unified-memory-framework/blob/main/.github/workflows/reusable_gpu.yml#L117)

on the other hand, in the "real" benchmarks workflow we have all flags set https://github.com/oneapi-src/unified-memory-framework/blob/main/.github/workflows/reusable_benchmarks.yml#L89

That is why I decided to ask first to make sure that I did not miss something.