@@ -71,7 +71,7 @@ static void transpose_chunks(int *sched, int n_pes, int my_pe,
71
71
buf [1 ] = bufs [1 ];
72
72
#endif
73
73
int pe = sched [0 ], pe2 , j = 0 ;
74
- MPI_Status send_status , recv_status ;
74
+ MPI_Request send_request , recv_request ;
75
75
76
76
#ifdef AMD_MPI_TRANSPOSE_LOGS
77
77
printf ("TRANSPOSE-PAIRWISE: n_pes[%d], my_pe[%d], first_pe[%d]\n" , n_pes , my_pe , pe );
@@ -105,13 +105,13 @@ static void transpose_chunks(int *sched, int n_pes, int my_pe,
105
105
#ifdef AMD_MPI_TRANSPOSE_LOGS
106
106
printf ("OVERLAP REGION:(my_pe != pe)=> my_pe[%d], i[%d]:pe[%d], src_offset[%d], dst_offset[%d], size[%d]\n" , my_pe , i , pe2 , sbo [pe2 ], rbo [pe2 ], sbs [pe2 ]);
107
107
#endif
108
- MPI_Isend (buf [j & 0x1 ], (int ) (sbs [pe ]), FFTW_MPI_TYPE , pe , (my_pe * n_pes + pe ) & 0xffff , comm , & send_status );
109
- MPI_Irecv (O + rbo [pe ], (int ) (rbs [pe ]), FFTW_MPI_TYPE , pe , (pe * n_pes + my_pe ) & 0xffff , comm , & recv_status );
108
+ MPI_Isend (buf [j & 0x1 ], (int ) (sbs [pe ]), FFTW_MPI_TYPE , pe , (my_pe * n_pes + pe ) & 0xffff , comm , & send_request );
109
+ MPI_Irecv (O + rbo [pe ], (int ) (rbs [pe ]), FFTW_MPI_TYPE , pe , (pe * n_pes + my_pe ) & 0xffff , comm , & recv_request );
110
110
memcpy (buf [(j + 1 )& 0x1 ], O + sbo [pe2 ], sbs [pe2 ] * sizeof (R ));
111
111
pe = pe2 ;
112
112
//barrier
113
- MPI_Wait (& send_status , MPI_STATUS_IGNORE );
114
- MPI_Wait (& recv_status , MPI_STATUS_IGNORE );
113
+ MPI_Wait (& send_request , MPI_STATUS_IGNORE );
114
+ MPI_Wait (& recv_request , MPI_STATUS_IGNORE );
115
115
}
116
116
j ++ ;
117
117
}
0 commit comments