mpl/gpu: fix conversions/comparisons between runtime/driver API enums

[nmnobre]

Pull Request Description

Fixes conversions and comparisons between different CUDA runtime and driver API enum types:

  CC       src/gpu/mpl_gpu_cuda.lo
../../../mpich/src/mpl/src/gpu/mpl_gpu_cuda.c: In function 'MPL_gpu_ipc_handle_create':
../../../mpich/src/mpl/src/gpu/mpl_gpu_cuda.c:146:9: warning: implicit conversion from 'CUresult' {aka 'enum cudaError_enum'} to 'cudaError_t' {aka 'enum cudaError'} [-Wenum-conversion]
  146 |     ret = cuPointerGetAttribute(&ipc_handle->id, CU_POINTER_ATTRIBUTE_BUFFER_ID, (CUdeviceptr) ptr);
      |         ^
In file included from ../../../mpich/src/mpl/src/gpu/mpl_gpu_cuda.c:9:
../../../mpich/src/mpl/src/gpu/mpl_gpu_cuda.c: In function 'MPL_gpu_get_buffer_id':
../../../mpich/src/mpl/src/gpu/mpl_gpu_cuda.c:200:16: warning: comparison between 'CUresult' {aka 'enum cudaError_enum'} and 'enum cudaError' [-Wenum-compare]
  200 |     assert(ret == cudaSuccess);
      |                ^~
../../../mpich/src/mpl/src/gpu/mpl_gpu_cuda.c: In function 'MPL_gpu_ipc_handle_is_valid':
../../../mpich/src/mpl/src/gpu/mpl_gpu_cuda.c:211:16: warning: comparison between 'CUresult' {aka 'enum cudaError_enum'} and 'enum cudaError' [-Wenum-compare]
  211 |     assert(ret == cudaSuccess);
      |                ^~

Also removes HI_ERR_CHECK(), currently only used to check the return value of hipMemGetAddressRange(), which is of type hipError_t, so should compare with hipSuccess instead:

In file included from ../../../mpich/src/mpl/include/mpl.h:9:
../../../mpich/src/mpl/src/gpu/mpl_gpu_hip.c: In function 'MPL_gpu_get_buffer_bounds':
../../../mpich/src/mpl/src/gpu/mpl_gpu_hip.c:12:46: warning: comparison between 'hipError_t' and 'enum <anonymous>' [-Wenum-compare]
   12 | #define HI_ERR_CHECK(ret) if (unlikely((ret) != HIP_SUCCESS)) goto fn_fail
      |                                              ^~
../../../mpich/src/mpl/include/mpl_base.h:81:42: note: in definition of macro 'unlikely'
   81 | #define unlikely(x_) __builtin_expect(!!(x_),0)
      |                                          ^~
../../../mpich/src/mpl/src/gpu/mpl_gpu_hip.c:448:5: note: in expansion of macro 'HI_ERR_CHECK'
  448 |     HI_ERR_CHECK(hiret);
      |     ^~~~~~~~~~~~

Plus an unused variable:

../../../mpich/src/mpl/src/gpu/mpl_gpu_hip.c: In function 'MPL_gpu_init':
../../../mpich/src/mpl/src/gpu/mpl_gpu_hip.c:358:17: warning: unused variable 'global_dev_id' [-Wunused-variable]
  358 |             int global_dev_id;
      |                 ^~~~~~~~~~~~~

Plus unused code when no ZE:

../mpich/src/mpid/ch4/shm/ipc/gpu/gpu_post.c: In function 'MPIDI_GPU_ipc_local_mmap':
../mpich/src/mpid/ch4/shm/ipc/gpu/gpu_post.c:837:3: warning: label 'fn_fail' defined but not used [-Wunused-label]
  837 |   fn_fail:
      |   ^~~~~~~
../mpich/src/mpid/ch4/shm/ipc/gpu/gpu_post.c: At top level:
../mpich/src/mpid/ch4/shm/ipc/gpu/gpu_post.c:245:13: warning: 'ipc_track_cache_can_insert' defined but not used [-Wunused-function]
  245 | static bool ipc_track_cache_can_insert(void)
      |             ^~~~~~~~~~~~~~~~~~~~~~~~~~

Plus (solved by initialising the ptr):

../../../../../../mpich/src/mpi/datatype/typerep/yaksa/src/backend/src/yaksur_pup.c: In function 'ipup':
../../../../../../mpich/src/mpi/datatype/typerep/yaksa/src/backend/src/yaksur_pup.c:37:33: warning: 'infopriv' may be used uninitialized [-Wmaybe-uninitialized]
   37 |             if (info && infopriv->gpudriver_id != YAKSURI_GPUDRIVER_ID__UNSET &&
      |                         ~~~~~~~~^~~~~~~~~~~~~~
../../../../../../mpich/src/mpi/datatype/typerep/yaksa/src/backend/src/yaksur_pup.c:22:21: note: 'infopriv' was declared here
   22 |     yaksuri_info_s *infopriv;
      |                     ^~~~~~~~

Plus (solved by adding error check):

In function 'MPIDI_SHM_am_send_hdr',
    inlined from 'reply_ipc_write' at ../mpich/src/mpid/ch4/shm/ipc/src/ipc_control.c:229:5:
../mpich/src/mpid/ch4/shm/src/shm_am.h:21:11: warning: 'hdr_sz' may be used uninitialized [-Wmaybe-uninitialized]
   21 |     ret = MPIDI_POSIX_am_send_hdr(rank, comm, handler_id, am_hdr, am_hdr_sz, src_vci, dst_vci);
      |           ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
../mpich/src/mpid/ch4/shm/ipc/src/ipc_control.c: In function 'reply_ipc_write':
../mpich/src/mpid/ch4/shm/ipc/src/ipc_control.c:219:14: note: 'hdr_sz' was declared here
  219 |     MPI_Aint hdr_sz;
      |              ^~~~~~

[hzhou]

LGTM

[hzhou]

test:mpich/ch4/gpu

[nmnobre]

Rebased and patched a few more warnings.

Only one other left:

../../../../../../mpich/src/mpi/datatype/typerep/yaksa/src/backend/cuda/hooks/yaksuri_cuda_init_hooks.c: In function 'yaksuri_cuda_init_hook':
../../../../../../mpich/src/mpi/datatype/typerep/yaksa/src/backend/cuda/hooks/yaksuri_cuda_init_hooks.c:212:36: warning: argument 1 range [18446744071562067968, 18446744073709551615] exceeds maximum object size 9223372036854775807 [-Walloc-size-larger-than=]
  212 |     yaksuri_cudai_global.streams = calloc(yaksuri_cudai_global.ndevices, sizeof(cudai_stream));
      |                                    ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
In file included from ../../../../../../mpich/src/mpi/datatype/typerep/yaksa/src/external/yuthash.h:31,
                 from ../../../../../../mpich/src/mpi/datatype/typerep/yaksa/src/frontend/include/yaksi.h:12,
                 from ../../../../../../mpich/src/mpi/datatype/typerep/yaksa/src/backend/cuda/hooks/yaksuri_cuda_init_hooks.c:6:
/usr/include/stdlib.h:675:14: note: in a call to allocation function 'calloc' declared here
  675 | extern void *calloc (size_t __nmemb, size_t __size)
      |              ^~~~~~

The least bad way I found of getting rid of it is by adding a check just before the calloc(). yaksuri_cudai_global.ndevices is a global and the compiler can't guarantee that it's positive due to the opaque cudaDeviceGetAttribute() in between... something like:

    if (yaksuri_cudai_global.ndevices <= 0) {
        /* no devices to initialize; also lets the compiler know ndevices is non-negative below */
        goto fn_exit;
    }

[hzhou]

test:mpich/ch4/ofi
test:mpich/ch4/ucx