Re: [Starpu-devel] starpu_data_unregister_submit() and MPI

[Benoît Lizé <benoit.lize@gmail.com>]

Oops, I spoke too fast.

I have still an issue, but this time the cause seems rather transparent. It is not possible to copy a void data handle.

Here is the stack trace:

#0  0x0000000000000000 in ?? ()

#1  0x00007ffff5d4582a in _starpu_mpi_copy_cb (arg=0x5eaaef0) at starpu_mpi.c:936

#2  0x00007ffff5fc8841 in _starpu_data_acquire_fetch_data_callback (arg=0x5eaaf20) at datawizard/user_interactions.c:81

#3  0x00007ffff5fb93e2 in _starpu_create_request_to_fetch_data (handle=0x7fffd80044b0, dst_replicate=0x7fffd8004580, mode=STARPU_R, is_prefetch=0, async=1, 

    callback_func=0x7ffff5fc87e0 <_starpu_data_acquire_fetch_data_callback>, callback_arg=0x5eaaf20) at datawizard/coherency.c:418

#4  0x00007ffff5fb998c in _starpu_fetch_data_on_node (handle=0x7fffd80044b0, dst_replicate=0x7fffd8004580, mode=STARPU_R, detached=0, async=1, 

    callback_func=0x7ffff5fc87e0 <_starpu_data_acquire_fetch_data_callback>, callback_arg=0x5eaaf20) at datawizard/coherency.c:565

#5  0x00007ffff5fc88fd in _starpu_data_acquire_continuation_non_blocking (arg=0x5eaaf20) at datawizard/user_interactions.c:99

#6  0x00007ffff5fc897b in starpu_data_acquire_cb_pre_sync_callback (arg=0x5eaaf20) at datawizard/user_interactions.c:118

#7  0x00007ffff5f7cc94 in _starpu_handle_job_termination (j=0x5eab3c0) at core/jobs.c:225

#8  0x00007ffff5fa29c8 in _starpu_push_task (j=0x5eab3c0) at core/sched_policy.c:372

#9  0x00007ffff5f7d8b4 in _starpu_enforce_deps_and_schedule (j=0x5eab3c0) at core/jobs.c:416

#10 0x00007ffff5f7e706 in _starpu_submit_job (j=0x5eab3c0) at core/task.c:288

#11 0x00007ffff5f7f4ae in starpu_task_submit (task=0x5eaafc0) at core/task.c:531

#12 0x00007ffff5f7f526 in _starpu_task_submit_internally (task=0x5eaafc0) at core/task.c:548

#13 0x00007ffff5fc8e65 in starpu_data_acquire_on_node_cb_sequential_consistency (handle=0x7fffd80044b0, node=0, mode=STARPU_R, 

    callback=0x7ffff5d4572e <_starpu_mpi_copy_cb>, arg=0x5eaaef0, sequential_consistency=1) at datawizard/user_interactions.c:170

#14 0x00007ffff5fc8f79 in starpu_data_acquire_on_node_cb (handle=0x7fffd80044b0, node=0, mode=STARPU_R, callback=0x7ffff5d4572e <_starpu_mpi_copy_cb>, 

    arg=0x5eaaef0) at datawizard/user_interactions.c:188

#15 0x00007ffff5fc8fb1 in starpu_data_acquire_cb (handle=0x7fffd80044b0, mode=STARPU_R, callback=0x7ffff5d4572e <_starpu_mpi_copy_cb>, arg=0x5eaaef0)

    at datawizard/user_interactions.c:194

#16 0x00007ffff5d46867 in _starpu_mpi_submit_new_mpi_request (arg=0x5eaa470) at starpu_mpi.c:1048

#17 0x00007ffff5fc8841 in _starpu_data_acquire_fetch_data_callback (arg=0x5eaa630) at datawizard/user_interactions.c:81

#18 0x00007ffff5fb93e2 in _starpu_create_request_to_fetch_data (handle=0x5ea7920, dst_replicate=0x5ea79f0, mode=STARPU_W, is_prefetch=0, async=1, 

    callback_func=0x7ffff5fc87e0 <_starpu_data_acquire_fetch_data_callback>, callback_arg=0x5eaa630) at datawizard/coherency.c:418

#19 0x00007ffff5fb998c in _starpu_fetch_data_on_node (handle=0x5ea7920, dst_replicate=0x5ea79f0, mode=STARPU_W, detached=0, async=1, 

    callback_func=0x7ffff5fc87e0 <_starpu_data_acquire_fetch_data_callback>, callback_arg=0x5eaa630) at datawizard/coherency.c:565

#20 0x00007ffff5fc88fd in _starpu_data_acquire_continuation_non_blocking (arg=0x5eaa630) at datawizard/user_interactions.c:99

#21 0x00007ffff5fc897b in starpu_data_acquire_cb_pre_sync_callback (arg=0x5eaa630) at datawizard/user_interactions.c:118

#22 0x00007ffff5f7cc94 in _starpu_handle_job_termination (j=0x5eaaad0) at core/jobs.c:225

#23 0x00007ffff5fa29c8 in _starpu_push_task (j=0x5eaaad0) at core/sched_policy.c:372

#24 0x00007ffff5f7d8b4 in _starpu_enforce_deps_and_schedule (j=0x5eaaad0) at core/jobs.c:416

#25 0x00007ffff5f7e706 in _starpu_submit_job (j=0x5eaaad0) at core/task.c:288

#26 0x00007ffff5f7f4ae in starpu_task_submit (task=0x5eaa6d0) at core/task.c:531

#27 0x00007ffff5f7f526 in _starpu_task_submit_internally (task=0x5eaa6d0) at core/task.c:548

#28 0x00007ffff5fc8e65 in starpu_data_acquire_on_node_cb_sequential_consistency (handle=0x5ea7920, node=0, mode=STARPU_W, 

    callback=0x7ffff5d45992 <_starpu_mpi_submit_new_mpi_request>, arg=0x5eaa470, sequential_consistency=1) at datawizard/user_interactions.c:170

#29 0x00007ffff5fc8ff3 in starpu_data_acquire_cb_sequential_consistency (handle=0x5ea7920, mode=STARPU_W, 

    callback=0x7ffff5d45992 <_starpu_mpi_submit_new_mpi_request>, arg=0x5eaa470, sequential_consistency=1) at datawizard/user_interactions.c:200

#30 0x00007ffff5d4129c in _starpu_mpi_isend_irecv_common (data_handle=0x5ea7920, srcdst=1, mpi_tag=5209, comm=0x63df20 <ompi_mpi_comm_world>, detached=1, 

    callback=0x0, arg=0x0, request_type=RECV_REQ, func=0x7ffff5d42036 <_starpu_mpi_irecv_data_func>, mode=STARPU_W, sequential_consistency=1, 

    is_internal_req=0, count=0) at starpu_mpi.c:338

#31 0x00007ffff5d425e6 in _starpu_mpi_irecv_common (data_handle=0x5ea7920, source=1, mpi_tag=5209, comm=0x63df20 <ompi_mpi_comm_world>, detached=1, 

    callback=0x0, arg=0x0, sequential_consistency=1, is_internal_req=0, count=0) at starpu_mpi.c:509

#32 0x00007ffff5d4282d in starpu_mpi_irecv_detached (data_handle=0x5ea7920, source=1, mpi_tag=5209, comm=0x63df20 <ompi_mpi_comm_world>, callback=0x0, arg=0x0)

    at starpu_mpi.c:545

#33 0x00007ffff5d4c8c2 in _starpu_mpi_exchange_data_before_execution (data="0x5ea7920," mode=STARPU_R, me=0, dest=0, do_execute=1, 

    comm=0x63df20 <ompi_mpi_comm_world>) at starpu_mpi_task_insert.c:117

#34 0x00007ffff5d4de18 in _starpu_mpi_task_build_v (comm=0x63df20 <ompi_mpi_comm_world>, 

    codelet=0x7ffff6a92080 <StarPuCodelets<std::complex<float> >::deleteCodelet>, task=0x7fffffffbc38, xrank_p=0x7fffffffbc04, dest_p=0x7fffffffbc08, 

    varg_list=0x7fffffffbca8) at starpu_mpi_task_insert.c:380

#35 0x00007ffff5d4f329 in _starpu_mpi_task_insert_v (comm=0x63df20 <ompi_mpi_comm_world>, 

    codelet=0x7ffff6a92080 <StarPuCodelets<std::complex<float> >::deleteCodelet>, varg_list=0x7fffffffbca8) at starpu_mpi_task_insert.c:582

And the culprit:

(gdb) f 1

#1  0x00007ffff5d4582a in _starpu_mpi_copy_cb (arg=0x5eaaef0) at starpu_mpi.c:936

936                     itf_dst->unpack_data(args->data_handle, 0, args->buffer, itf_src->get_size(args->copy_handle));

(gdb) info locals

itf_src = 0x7ffff62198e0 <starpu_interface_vector_ops>

itf_dst = 0x7ffff62199e0 <starpu_interface_void_ops>

args = 0x5eaaef0

However I don't understand why itf_src is a vector interface.

On Fri, Mar 14, 2014 at 3:48 PM, Benoît Lizé <benoit.lize@gmail.com> wrote:

Hello,

Thank you for the quick fix !

My code now works with r12407.

-- 

Benoit Lize