Re: [PATCH 6/6, OpenACC, libgomp] Async re-work, nvptx changes

[Tom de Vries <tdevries@suse.de>]

On 9/25/18 3:11 PM, Chung-Lin Tang wrote:
> Hi Tom,
> this patch removes large portions of plugin/plugin-nvptx.c, since a lot
> of it is
> now in oacc-async.c now.

Yay!

> The new code is essentially a
> NVPTX/CUDA-specific implementation
> of the new-style goacc_asyncqueues.
> 
> Also, some needed functions in cuda-lib.def are added. The cuda.h
> function has also
> been updated to build independently without a CUDA installation.
> 

I see these formatting issues:
...
$ check_GNU_style.sh async-06.nvptx.patch

There should be exactly one space between function name and parenthesis.
35:+CUresult cuStreamAddCallback(CUstream, CUstreamCallback, void *,
unsigned int);

Trailing operator.
1320:+      struct nvptx_thread *nvthd =
...

Otherwise, OK.

Thanks,
- Tom


> Thanks,
> Chung-Lin
> 
>     * plugin/plugin-nvptx.c (struct cuda_map): Remove.
>     (struct ptx_stream): Remove.
>     (struct nvptx_thread): Remove current_stream field.
>     (cuda_map_create): Remove.
>     (cuda_map_destroy): Remove.
>     (map_init): Remove.
>     (map_fini): Remove.
>     (map_pop): Remove.
>     (map_push): Remove.
>     (struct goacc_asyncqueue): Define.
>     (struct nvptx_callback): Define.
>     (struct ptx_free_block): Define.
>     (struct ptx_device): Remove null_stream, active_streams, async_streams,
>     stream_lock, and next fields.
>     (enum ptx_event_type): Remove.
>     (struct ptx_event): Remove.
>     (ptx_event_lock): Remove.
>     (ptx_events): Remove.
>     (init_streams_for_device): Remove.
>     (fini_streams_for_device): Remove.
>     (select_stream_for_async): Remove.
>     (nvptx_init): Remove ptx_events and ptx_event_lock references.
>     (nvptx_attach_host_thread_to_device): Remove CUDA_ERROR_NOT_PERMITTED
>     case.
>     (nvptx_open_device): Add free_blocks initialization, remove
>     init_streams_for_device call.
>     (nvptx_close_device): Remove fini_streams_for_device call, add
>     free_blocks destruct code.
>     (event_gc): Remove.
>     (event_add): Remove.
>     (nvptx_exec): Adjust parameters and code.
>     (nvptx_free): Likewise.
>     (nvptx_host2dev): Remove.
>     (nvptx_dev2host): Remove.
>     (nvptx_set_async): Remove.
>     (nvptx_async_test): Remove.
>     (nvptx_async_test_all): Remove.
>     (nvptx_wait): Remove.
>     (nvptx_wait_async): Remove.
>     (nvptx_wait_all): Remove.
>     (nvptx_wait_all_async): Remove.
>     (nvptx_get_cuda_stream): Remove.
>     (nvptx_set_cuda_stream): Remove.
>     (GOMP_OFFLOAD_alloc): Adjust code.
>     (GOMP_OFFLOAD_free): Likewise.
>     (GOMP_OFFLOAD_openacc_register_async_cleanup): Remove.
>     (GOMP_OFFLOAD_openacc_exec): Adjust parameters and code.
>     (GOMP_OFFLOAD_openacc_async_test_all): Remove.
>     (GOMP_OFFLOAD_openacc_async_wait): Remove.
>     (GOMP_OFFLOAD_openacc_async_wait_async): Remove.
>     (GOMP_OFFLOAD_openacc_async_wait_all): Remove.
>     (GOMP_OFFLOAD_openacc_async_wait_all_async): Remove.
>     (GOMP_OFFLOAD_openacc_async_set_async): Remove.
>     (cuda_free_argmem): New function.
>     (GOMP_OFFLOAD_openacc_async_exec): New plugin hook function.
>     (GOMP_OFFLOAD_openacc_create_thread_data): Adjust code.
>     (GOMP_OFFLOAD_openacc_cuda_get_stream): Adjust code.
>     (GOMP_OFFLOAD_openacc_cuda_set_stream): Adjust code.
>     (GOMP_OFFLOAD_openacc_async_construct): New plugin hook function.
>     (GOMP_OFFLOAD_openacc_async_destruct): New plugin hook function.
>     (GOMP_OFFLOAD_openacc_async_test): Remove and re-implement.
>     (GOMP_OFFLOAD_openacc_async_synchronize): New plugin hook function.
>     (GOMP_OFFLOAD_openacc_async_serialize): New plugin hook function.
>     (GOMP_OFFLOAD_openacc_async_queue_callback): New plugin hook function.
>     (cuda_callback_wrapper): New function.
>     (cuda_memcpy_sanity_check): New function.
>     (GOMP_OFFLOAD_host2dev): Remove and re-implement.
>     (GOMP_OFFLOAD_dev2host): Remove and re-implement.
>     (GOMP_OFFLOAD_openacc_async_host2dev): New plugin hook function.
>     (GOMP_OFFLOAD_openacc_async_dev2host): New plugin hook function.