Re: [Starpu-devel] Compiling starpu on Mac OS

[Cedric Augonnet <cedric.augonnet@inria.fr>]

Hi,

These libraries are only used from the examples. Missing FFTW, MKL or 
the MAGMA library only means that the examples using them won't get 
compiled. You actually don't have to switch from CUDA to OpenCL: you can 
use both of them.

To enable OpenCL in case it is not located in a standard location, you 
may however have to explicitely specify where to find the OpenCL 
implementation by using --with-opencl-dir=<path>, 
--with-opencl-include-dir=<path>, or --with-opencl-lib-dir=<path>.

Best,
Cédric

Le 31/03/2011 16:29, tog a écrit :
> Hi
>
> Thanks for the info.
>
> I would also appreciate if you could comment on the need for MAGMA,
> FFTW or any BLAS library as I have noticed that configure is looking
> for the presence of such libraries.
>
> Is there also a way to switch to OpenCL instead of using CUDA ?
>
> Thanks for your help
>
> Best Regards
> Guillaume
>
> On Thu, Mar 31, 2011 at 7:47 PM, Nathalie Furmento
> <nathalie.furmento@labri.fr>  wrote:
>> Hi,
>>
>> We found out what the problem is. On our machines, we are using a linker
>> which automatically adds the dependencies of the specified libraries. Your
>> linker does not seem to do it. This is clearly a problem with the
>> compilation system of StarPU, i am currently working on fixing it, it might
>> take a while (but it will very likely be done by tomorrow evening).
>>
>> Regards,
>>
>> Nathalie
>>
>>
>> On 31/03/2011 13:17, tog wrote:
>>
>> Bonjour
>>
>> Here it is.
>>
>> Guillaume
>>
>> On Thu, Mar 31, 2011 at 4:33 PM, Cedric Augonnet
>> <cedric.augonnet@inria.fr>  wrote:
>>
>>
>> Hello,
>>
>> Thanks a lot for reporting this issue! Perhaps there is some
>> DYLD_LIBRARY_PATH to the CUDA library that is missing in your
>> environment? Could you please send us the config.log file generated
>> during the configure step?
>>
>> Thanks,
>> Cédric
>>
>> Le 31/03/2011 11:47, tog a écrit :
>>
>>
>> Hi,
>>
>> Machine: MacBook Pro running Snow Leopard.
>> nvcc -V says:
>>
>> oggy:starpu alleon$ nvcc -V
>> nvcc: NVIDIA (R) Cuda compiler driver
>> Copyright (c) 2005-2010 NVIDIA Corporation
>> Built on Thu_Nov_11_15:26:50_PST_2010
>> Cuda compilation tools, release 3.2, V0.2.1221
>>
>>
>> I would like to evaluate starpu. I just grab the sources from svn (rev 
>> 3420)
>> autogen.sh&&    ./configure&&    make
>>
>> Compilation is going file until the link which fails.
>> I noticed that I am using CUDA and not OpenCL. Is that OK ?
>>
>> Then what am I doing wrong ?
>>
>> Thanks for your help.
>> Guillaume
>>
>> libtool: link: gcc -dynamiclib  -o .libs/libstarpu.0.dylib
>> .libs/libstarpu_la-barrier.o .libs/libstarpu_la-hash.o
>> .libs/libstarpu_la-htable32.o .libs/libstarpu_la-rwlock.o
>> .libs/libstarpu_la-starpu_spinlock.o .libs/libstarpu_la-timing.o
>> .libs/libstarpu_la-fxt.o .libs/libstarpu_la-utils.o
>> .libs/libstarpu_la-jobs.o .libs/libstarpu_la-task.o
>> .libs/libstarpu_la-task_bundle.o .libs/libstarpu_la-workers.o
>> .libs/libstarpu_la-combined_workers.o .libs/libstarpu_la-topology.o
>> .libs/libstarpu_la-debug.o .libs/libstarpu_la-errorcheck.o
>> .libs/libstarpu_la-progress_hook.o .libs/libstarpu_la-cg.o
>> .libs/libstarpu_la-dependencies.o
>> .libs/libstarpu_la-implicit_data_deps.o .libs/libstarpu_la-tags.o
>> .libs/libstarpu_la-task_deps.o .libs/libstarpu_la-htable.o
>> .libs/libstarpu_la-data_concurrency.o
>> .libs/libstarpu_la-perfmodel_history.o
>> .libs/libstarpu_la-perfmodel_bus.o .libs/libstarpu_la-perfmodel.o
>> .libs/libstarpu_la-regression.o .libs/libstarpu_la-sched_policy.o
>> .libs/libstarpu_la-priorities.o .libs/libstarpu_la-parallel_task.o
>> .libs/libstarpu_la-eager_central_policy.o
>> .libs/libstarpu_la-eager_central_priority_policy.o
>> .libs/libstarpu_la-work_stealing_policy.o
>> .libs/libstarpu_la-deque_modeling_policy_data_aware.o
>> .libs/libstarpu_la-heft.o .libs/libstarpu_la-random_policy.o
>> .libs/libstarpu_la-stack_queues.o .libs/libstarpu_la-deque_queues.o
>> .libs/libstarpu_la-fifo_queues.o
>> .libs/libstarpu_la-detect_combined_workers.o
>> .libs/libstarpu_la-parallel_heft.o
>> .libs/libstarpu_la-parallel_greedy.o
>> .libs/libstarpu_la-driver_common.o .libs/libstarpu_la-memory_nodes.o
>> .libs/libstarpu_la-write_back.o .libs/libstarpu_la-coherency.o
>> .libs/libstarpu_la-data_request.o .libs/libstarpu_la-progress.o
>> .libs/libstarpu_la-copy_driver.o .libs/libstarpu_la-filters.o
>> .libs/libstarpu_la-sort_data_handles.o .libs/libstarpu_la-memalloc.o
>> .libs/libstarpu_la-footprint.o .libs/libstarpu_la-datastats.o
>> .libs/libstarpu_la-user_interactions.o .libs/libstarpu_la-reduction.o
>> .libs/libstarpu_la-data_interface.o
>> .libs/libstarpu_la-bcsr_interface.o .libs/libstarpu_la-csr_interface.o
>> .libs/libstarpu_la-matrix_filters.o
>> .libs/libstarpu_la-matrix_interface.o
>> .libs/libstarpu_la-block_filters.o
>> .libs/libstarpu_la-block_interface.o
>> .libs/libstarpu_la-vector_interface.o
>> .libs/libstarpu_la-bcsr_filters.o .libs/libstarpu_la-csr_filters.o
>> .libs/libstarpu_la-vector_filters.o
>> .libs/libstarpu_la-variable_interface.o
>> .libs/libstarpu_la-void_interface.o .libs/libstarpu_la-malloc.o
>> .libs/libstarpu_la-execute_on_all.o
>> .libs/libstarpu_la-starpu_create_sync_task.o
>> .libs/libstarpu_la-starpu_cublas.o .libs/libstarpu_la-file.o
>> .libs/libstarpu_la-starpu_data_cpy.o
>> .libs/libstarpu_la-starpu_insert_task.o
>> .libs/libstarpu_la-starpu_insert_task_utils.o
>> .libs/libstarpu_la-starpu_task_list.o .libs/libstarpu_la-starpu_fxt.o
>> .libs/libstarpu_la-starpu_fxt_mpi.o
>> .libs/libstarpu_la-starpu_fxt_dag.o .libs/libstarpu_la-starpu_paje.o
>> .libs/libstarpu_la-latency.o .libs/libstarpu_la-structures_size.o
>> .libs/libstarpu_la-profiling.o .libs/libstarpu_la-bound.o
>> .libs/libstarpu_la-profiling_helpers.o .libs/libstarpu_la-driver_cpu.o
>> .libs/libstarpu_la-driver_cuda.o   -L/usr/local/cuda/lib -lm -lblas
>> -lcurand -lcublas -lpthread    -install_name
>> /usr/local/lib/libstarpu.0.dylib -compatibility_version 1
>> -current_version 1.0 -Wl,-single_module
>> Undefined symbols:
>>     "_cudaThreadExit", referenced from:
>>         _benchmark_all_gpu_devices in libstarpu_la-perfmodel_bus.o
>>         __starpu_cuda_worker in libstarpu_la-driver_cuda.o
>>     "_cudaMemcpy2D", referenced from:
>>         _copy_ram_to_cuda_async in libstarpu_la-matrix_interface.o
>>         _copy_cuda_to_ram_async in libstarpu_la-matrix_interface.o
>>         _copy_cuda_to_ram in libstarpu_la-matrix_interface.o
>>         _copy_cuda_to_cuda in libstarpu_la-matrix_interface.o
>>         _copy_ram_to_cuda in libstarpu_la-matrix_interface.o
>>         _copy_cuda_async_common in libstarpu_la-block_interface.o
>>         _copy_cuda_async_common in libstarpu_la-block_interface.o
>>         _copy_cuda_common in libstarpu_la-block_interface.o
>>         _copy_cuda_common in libstarpu_la-block_interface.o
>>     "_cudaMemcpy2DAsync", referenced from:
>>         _copy_ram_to_cuda_async in libstarpu_la-matrix_interface.o
>>         _copy_cuda_to_ram_async in libstarpu_la-matrix_interface.o
>>         _copy_cuda_async_common in libstarpu_la-block_interface.o
>>         _copy_cuda_async_common in libstarpu_la-block_interface.o
>>     "_cudaFree", referenced from:
>>         _benchmark_all_gpu_devices in libstarpu_la-perfmodel_bus.o
>>         _benchmark_all_gpu_devices in libstarpu_la-perfmodel_bus.o
>>         _free_bcsr_buffer_on_node in libstarpu_la-bcsr_interface.o
>>         _free_bcsr_buffer_on_node in libstarpu_la-bcsr_interface.o
>>         _free_bcsr_buffer_on_node in libstarpu_la-bcsr_interface.o
>>         _allocate_bcsr_buffer_on_node in libstarpu_la-bcsr_interface.o
>>         _allocate_bcsr_buffer_on_node in libstarpu_la-bcsr_interface.o
>>         _free_csr_buffer_on_node in libstarpu_la-csr_interface.o
>>         _free_csr_buffer_on_node in libstarpu_la-csr_interface.o
>>         _free_csr_buffer_on_node in libstarpu_la-csr_interface.o
>>         _allocate_csr_buffer_on_node in libstarpu_la-csr_interface.o
>>         _allocate_csr_buffer_on_node in libstarpu_la-csr_interface.o
>>         _free_matrix_buffer_on_node in libstarpu_la-matrix_interface.o
>>         _free_block_buffer_on_node in libstarpu_la-block_interface.o
>>         _free_vector_buffer_on_node in libstarpu_la-vector_interface.o
>>         _free_variable_buffer_on_node in libstarpu_la-variable_interface.o
>>         __starpu_cuda_worker in libstarpu_la-driver_cuda.o
>>         __starpu_cuda_worker in libstarpu_la-driver_cuda.o
>>     "_cudaHostAlloc", referenced from:
>>         _benchmark_all_gpu_devices in libstarpu_la-perfmodel_bus.o
>>         _malloc_pinned_cuda_codelet in libstarpu_la-malloc.o
>>     "_cudaEventCreate", referenced from:
>>         __starpu_driver_copy_data_1_to_1 in libstarpu_la-copy_driver.o
>>         __starpu_driver_copy_data_1_to_1 in libstarpu_la-copy_driver.o
>>     "_cudaMemcpy", referenced from:
>>         _benchmark_all_gpu_devices in libstarpu_la-perfmodel_bus.o
>>         _benchmark_all_gpu_devices in libstarpu_la-perfmodel_bus.o
>>         _benchmark_all_gpu_devices in libstarpu_la-perfmodel_bus.o
>>         _benchmark_all_gpu_devices in libstarpu_la-perfmodel_bus.o
>>         _copy_cuda_common in libstarpu_la-bcsr_interface.o
>>         _copy_cuda_common in libstarpu_la-bcsr_interface.o
>>         _copy_cuda_common in libstarpu_la-bcsr_interface.o
>>         _copy_cuda_common in libstarpu_la-csr_interface.o
>>         _copy_cuda_common in libstarpu_la-csr_interface.o
>>         _copy_cuda_common in libstarpu_la-csr_interface.o
>>         _copy_cuda_async_common in libstarpu_la-block_interface.o
>>         _copy_cuda_common in libstarpu_la-block_interface.o
>>         _copy_cuda_async_common in libstarpu_la-vector_interface.o
>>         _copy_cuda_to_ram in libstarpu_la-vector_interface.o
>>         _copy_cuda_to_cuda in libstarpu_la-vector_interface.o
>>         _copy_ram_to_cuda in libstarpu_la-vector_interface.o
>>         _copy_cuda_to_ram_async in libstarpu_la-variable_interface.o
>>         _copy_cuda_to_ram in libstarpu_la-variable_interface.o
>>         _copy_ram_to_cuda_async in libstarpu_la-variable_interface.o
>>         _copy_cuda_to_cuda in libstarpu_la-variable_interface.o
>>         _copy_ram_to_cuda in libstarpu_la-variable_interface.o
>>     "_cudaEventSynchronize", referenced from:
>>         __starpu_driver_wait_request_completion in 
>> libstarpu_la-copy_driver.o
>>     "_cudaThreadSynchronize", referenced from:
>>         _benchmark_all_gpu_devices in libstarpu_la-perfmodel_bus.o
>>         _benchmark_all_gpu_devices in libstarpu_la-perfmodel_bus.o
>>         _benchmark_all_gpu_devices in libstarpu_la-perfmodel_bus.o
>>         _benchmark_all_gpu_devices in libstarpu_la-perfmodel_bus.o
>>     "_cudaEventRecord", referenced from:
>>         __starpu_driver_copy_data_1_to_1 in libstarpu_la-copy_driver.o
>>         __starpu_driver_copy_data_1_to_1 in libstarpu_la-copy_driver.o
>>     "_cudaMemcpyAsync", referenced from:
>>         _copy_cuda_async_common in libstarpu_la-block_interface.o
>>         _copy_cuda_async_common in libstarpu_la-vector_interface.o
>>         _copy_cuda_to_ram_async in libstarpu_la-variable_interface.o
>>         _copy_ram_to_cuda_async in libstarpu_la-variable_interface.o
>>     "_cudaGetErrorString", referenced from:
>>         __starpu_driver_wait_request_completion in 
>> libstarpu_la-copy_driver.o
>>         __starpu_driver_wait_request_completion in 
>> libstarpu_la-copy_driver.o
>>         __starpu_driver_copy_data_1_to_1 in libstarpu_la-copy_driver.o
>>         __starpu_driver_copy_data_1_to_1 in libstarpu_la-copy_driver.o
>>         __starpu_driver_copy_data_1_to_1 in libstarpu_la-copy_driver.o
>>         __starpu_driver_copy_data_1_to_1 in libstarpu_la-copy_driver.o
>>         _copy_cuda_common in libstarpu_la-bcsr_interface.o
>>         _copy_cuda_common in libstarpu_la-bcsr_interface.o
>>         _copy_cuda_common in libstarpu_la-bcsr_interface.o
>>         _copy_cuda_common in libstarpu_la-csr_interface.o
>>         _copy_cuda_common in libstarpu_la-csr_interface.o
>>         _copy_cuda_common in libstarpu_la-csr_interface.o
>>         _copy_ram_to_cuda_async in libstarpu_la-matrix_interface.o
>>         _copy_cuda_to_ram_async in libstarpu_la-matrix_interface.o
>>         _copy_cuda_to_ram in libstarpu_la-matrix_interface.o
>>         _free_matrix_buffer_on_node in libstarpu_la-matrix_interface.o
>>         _allocate_matrix_buffer_on_node in libstarpu_la-matrix_interface.o
>>         _copy_cuda_to_cuda in libstarpu_la-matrix_interface.o
>>         _copy_ram_to_cuda in libstarpu_la-matrix_interface.o
>>         _copy_cuda_async_common in libstarpu_la-block_interface.o
>>         _copy_cuda_async_common in libstarpu_la-block_interface.o
>>         _copy_cuda_async_common in libstarpu_la-block_interface.o
>>         _copy_cuda_common in libstarpu_la-block_interface.o
>>         _copy_cuda_common in libstarpu_la-block_interface.o
>>         _copy_cuda_common in libstarpu_la-block_interface.o
>>         _free_block_buffer_on_node in libstarpu_la-block_interface.o
>>         _allocate_block_buffer_on_node in libstarpu_la-block_interface.o
>>         _copy_cuda_async_common in libstarpu_la-vector_interface.o
>>         _copy_cuda_to_ram in libstarpu_la-vector_interface.o
>>         _allocate_vector_buffer_on_node in libstarpu_la-vector_interface.o
>>         _copy_cuda_to_cuda in libstarpu_la-vector_interface.o
>>         _copy_ram_to_cuda in libstarpu_la-vector_interface.o
>>         _copy_cuda_to_ram_async in libstarpu_la-variable_interface.o
>>         _copy_cuda_to_ram in libstarpu_la-variable_interface.o
>>         _allocate_variable_buffer_on_node in
>> libstarpu_la-variable_interface.o
>>         _copy_ram_to_cuda_async in libstarpu_la-variable_interface.o
>>         _copy_cuda_to_cuda in libstarpu_la-variable_interface.o
>>         _copy_ram_to_cuda in libstarpu_la-variable_interface.o
>>         _free_pinned_cuda_codelet in libstarpu_la-malloc.o
>>         _malloc_pinned_cuda_codelet in libstarpu_la-malloc.o
>>         __starpu_cuda_worker in libstarpu_la-driver_cuda.o
>>         __starpu_cuda_worker in libstarpu_la-driver_cuda.o
>>         __starpu_cuda_worker in libstarpu_la-driver_cuda.o
>>         __starpu_cuda_worker in libstarpu_la-driver_cuda.o
>>         __starpu_cuda_worker in libstarpu_la-driver_cuda.o
>>         __starpu_cuda_worker in libstarpu_la-driver_cuda.o
>>         __starpu_cuda_worker in libstarpu_la-driver_cuda.o
>>         __starpu_cuda_worker in libstarpu_la-driver_cuda.o
>>     "_cudaEventDestroy", referenced from:
>>         __starpu_driver_test_request_completion in 
>> libstarpu_la-copy_driver.o
>>         __starpu_driver_wait_request_completion in 
>> libstarpu_la-copy_driver.o
>>     "_cudaMemset", referenced from:
>>         _benchmark_all_gpu_devices in libstarpu_la-perfmodel_bus.o
>>     "_cudaEventQuery", referenced from:
>>         __starpu_driver_test_request_completion in 
>> libstarpu_la-copy_driver.o
>>     "_cudaGetDeviceCount", referenced from:
>>         _benchmark_all_gpu_devices in libstarpu_la-perfmodel_bus.o
>>         __starpu_load_bus_performance_files in libstarpu_la-perfmodel_bus.o
>>         __starpu_load_bus_performance_files in libstarpu_la-perfmodel_bus.o
>>         __starpu_get_cuda_device_count in libstarpu_la-driver_cuda.o
>>         __starpu_init_cuda in libstarpu_la-driver_cuda.o
>>     "_cudaFreeHost", referenced from:
>>         _benchmark_all_gpu_devices in libstarpu_la-perfmodel_bus.o
>>         _free_pinned_cuda_codelet in libstarpu_la-malloc.o
>>     "_cudaGetDeviceProperties", referenced from:
>>         __starpu_cuda_worker in libstarpu_la-driver_cuda.o
>>         __starpu_cuda_worker in libstarpu_la-driver_cuda.o
>>     "_cudaMalloc", referenced from:
>>         _benchmark_all_gpu_devices in libstarpu_la-perfmodel_bus.o
>>         _allocate_bcsr_buffer_on_node in libstarpu_la-bcsr_interface.o
>>         _allocate_bcsr_buffer_on_node in libstarpu_la-bcsr_interface.o
>>         _allocate_bcsr_buffer_on_node in libstarpu_la-bcsr_interface.o
>>         _allocate_csr_buffer_on_node in libstarpu_la-csr_interface.o
>>         _allocate_csr_buffer_on_node in libstarpu_la-csr_interface.o
>>         _allocate_csr_buffer_on_node in libstarpu_la-csr_interface.o
>>         _allocate_matrix_buffer_on_node in libstarpu_la-matrix_interface.o
>>         _allocate_block_buffer_on_node in libstarpu_la-block_interface.o
>>         _allocate_vector_buffer_on_node in libstarpu_la-vector_interface.o
>>         _allocate_variable_buffer_on_node in
>> libstarpu_la-variable_interface.o
>>         __starpu_cuda_worker in libstarpu_la-driver_cuda.o
>>     "_cudaStreamDestroy", referenced from:
>>         __starpu_cuda_worker in libstarpu_la-driver_cuda.o
>>         __starpu_cuda_worker in libstarpu_la-driver_cuda.o
>>     "_cudaStreamSynchronize", referenced from:
>>         __starpu_cuda_worker in libstarpu_la-driver_cuda.o
>>     "_cudaSetDevice", referenced from:
>>         _benchmark_all_gpu_devices in libstarpu_la-perfmodel_bus.o
>>         __starpu_cuda_worker in libstarpu_la-driver_cuda.o
>>     "_cudaStreamCreate", referenced from:
>>         __starpu_cuda_worker in libstarpu_la-driver_cuda.o
>>         __starpu_cuda_worker in libstarpu_la-driver_cuda.o
>> ld: symbol(s) not found
>> collect2: ld returned 1 exit status
>> make[2]: *** [libstarpu.la] Error 1
>> make[1]: *** [all-recursive] Error 1
>> make: *** [all-recursive] Error 1
>>
>>
>>
>>
>> --
>> Cédric AUGONNET
>> PhD student -- Runtime Team
>> INRIA Bordeaux - LaBRI
>> Tel: [33] (0)5 24 57 40 74
>>
>>
>>
>>
>>
>> _______________________________________________
>> Starpu-devel mailing list
>> Starpu-devel@lists.gforge.inria.fr
>> http://lists.gforge.inria.fr/cgi-bin/mailman/listinfo/starpu-devel
>>
>>
>> --
>>
>> ----------------------------------------------------------------------------
>> Dr Nathalie Furmento                         CNRS R&D Engineer
>> Centre de Recherche INRIA Bordeaux Sud-Ouest http://www.labri.fr/~furmento/
>> 351 cours de la Libération, Bât A29          Tel: +33 (0)5 24 57 41 20
>> 33405 Talence cedex, FRANCE                  Fax: +33 (0)5 24 57 40 41
>> ----------------------------------------------------------------------------
>>
>
>
>


-- 
Cédric AUGONNET
PhD student -- Runtime Team
INRIA Bordeaux - LaBRI
Tel: [33] (0)5 24 57 40 74