@@ -58,9 +58,9 @@ def do_iteration(self, message_size, num_iters, done_future):
58
58
59
59
if self .gpu_direct and USE_ADDRESS_OPTIMIZATION :
60
60
d_data_recv_addr = array .array ('L' , [0 ])
61
- d_data_recv_size = array .array ('L ' , [0 ])
61
+ d_data_recv_size = array .array ('i ' , [0 ])
62
62
d_data_send_addr = array .array ('L' , [0 ])
63
- d_data_send_size = array .array ('L ' , [0 ])
63
+ d_data_send_size = array .array ('i ' , [0 ])
64
64
65
65
d_data_recv_addr [0 ] = d_data_recv .__cuda_array_interface__ ['data' ][0 ]
66
66
d_data_recv_size [0 ] = d_data_recv .nbytes
@@ -85,9 +85,9 @@ def do_iteration(self, message_size, num_iters, done_future):
85
85
charm .lib .CudaStreamSynchronize (stream_address )
86
86
else :
87
87
if USE_ADDRESS_OPTIMIZATION :
88
- partner_channel .send (gpu_src_ptrs = d_data_send_addr , gpu_src_sizes = d_data_send_size )
89
- partner_channel .recv (post_buf_addresses = d_data_recv_addr ,
90
- post_buf_sizes = d_data_recv_size
88
+ partner_channel .send (src_ptrs = d_data_send_addr , src_sizes = d_data_send_size )
89
+ partner_channel .recv (post_addresses = d_data_recv_addr ,
90
+ post_sizes = d_data_recv_size
91
91
)
92
92
else :
93
93
partner_channel .send (d_data_send )
@@ -103,10 +103,10 @@ def do_iteration(self, message_size, num_iters, done_future):
103
103
partner_channel .send (h_data_send )
104
104
else :
105
105
if USE_ADDRESS_OPTIMIZATION :
106
- partner_channel .recv (post_buf_addresses = d_data_recv_addr ,
107
- post_buf_sizes = d_data_recv_size
106
+ partner_channel .recv (post_addresses = d_data_recv_addr ,
107
+ post_sizes = d_data_recv_size
108
108
)
109
- partner_channel .send (gpu_src_ptrs = d_data_send_addr , gpu_src_sizes = d_data_send_size )
109
+ partner_channel .send (src_ptrs = d_data_send_addr , src_sizes = d_data_send_size )
110
110
else :
111
111
partner_channel .recv (d_data_recv )
112
112
partner_channel .send (d_data_send )
0 commit comments