GenomeWorks fails to compile with CUDA 12

[quim0]

I'm trying to compile GenomeWorks using the lastest CUDA version, and there seem to be some breaking changes in thrust that generates some compilation errors, here are the errors reported when doing make -j install:

In file included from /home/qaguado/GenomeWorks/cudaaligner/src/batched_device_matrices.cuh:25,                                                                                                            
                 from /home/qaguado/GenomeWorks/cudaaligner/src/ukkonen_gpu.cu:18:                                                                                                                         
/home/qaguado/GenomeWorks/common/base/include/claraparabricks/genomeworks/utils/pinned_host_vector.hpp:23:10: fatal error: thrust/system/cuda/experimental/pinned_allocator.h: No such file or directory   
   23 | #include <thrust/system/cuda/experimental/pinned_allocator.h>                                                                                                                                      
      |          ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~                                                                                                                                      
compilation terminated.                                                                                                                                                                                    
In file included from /home/qaguado/GenomeWorks/cudaaligner/src/batched_device_matrices.cuh:25,                                                                                                            
                 from /home/qaguado/GenomeWorks/cudaaligner/src/hirschberg_myers_gpu.cuh:20,                                                                                                               
                 from /home/qaguado/GenomeWorks/cudaaligner/src/hirschberg_myers_gpu.cu:17:                                                                                                                
/home/qaguado/GenomeWorks/common/base/include/claraparabricks/genomeworks/utils/pinned_host_vector.hpp:23:10: fatal error: thrust/system/cuda/experimental/pinned_allocator.h: No such file or directory   
   23 | #include <thrust/system/cuda/experimental/pinned_allocator.h>                                                                                                                                      
      |          ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~                                                                                                                                      
compilation terminated.                                                                                                                                                                                    
CMake Error at cudaaligner_generated_ukkonen_gpu.cu.o.Release.cmake:220 (message):                                                                                                                         
  Error generating                                                                                                                                                                                         
  /home/qaguado/GenomeWorks/build/cudaaligner/CMakeFiles/cudaaligner.dir/src/./cudaaligner_generated_ukkonen_gpu.cu.o

I tried to fix the error by doing the following changes in /common/base/include/claraparabricks/genomeworks/utils/pinned_host_vector.hpp (from this NVIDIA forum answer):

// ...

//#include <thrust/system/cuda/experimental/pinned_allocator.h>
#include <thrust/system/cuda/memory_resource.h>
#include <thrust/mr/allocator.h>
#include <thrust/system/cpp/memory.h>

// ...

//using pinned_host_vector = std::vector<T, thrust::system::cuda::experimental::pinned_allocator<T>>;
template <typename T>
using pinned_host_vector = std::vector<T, thrust::mr::stateless_resource_allocator<T, thrust::cuda::universal_host_pinned_memory_resource>>;

But it still fails with the following error:

/usr/include/c++/9/bits/alloc_traits.h:556:25: error: no matching function for call to ‘__do_alloc_on_move(thrust::mr::stateless_resource_allocator<claraparabricks::genomeworks::cudaaligner::batched_devi
ce_matrices<unsigned int>::device_interface, thrust::system::cuda::detail::cuda_memory_resource<cudaMallocHost, cudaFreeHost, thrust::pointer<void, thrust::cuda_cub::tag, void, thrust::use_default> > >&, thrust::mr::stateless_resource_allocator<claraparabricks::genomeworks::cudaaligner::batched_device_matrices<unsigned int>::device_interface, thrust::system::cuda::detail::cuda_memory_resource<cudaMallocHost, cudaFreeHost, thrust::pointer<void, thrust::cuda_cub::tag, void, thrust::use_default> > >&, __pocma)’

Are there any plans to make GenomeWorks compatible with CUDA 12?

Thanks

[tonyjie]

I encountered the same issue with CUDA 12

[ioeddk]

I encounter the same issue in CUDA 12. I can't find the pinned_allocator.h library in the CUDA 12.2 include folder. I guess CUDA 12 has this part refactored and the GenomeWorks needs some patch to work on CUDA 12.