1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
|
#!/usr/bin/python3
import numpy as np
import os
import argparse
print ("++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++")
print (' ')
print ('Usage example: python3 rccl_bw_test.py --np 2 --host_ip 10.67.10.104,10.67.10.154 --gpus_per_node 8')
print (' --test_iteration 100 --test_exe /home/tmp1/rccl-tests/build/all_reduce_perf')
print (' ')
print ('Note: This script currently is only compatible with MPICH mpirun syntax')
print (' ')
print ("++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++")
parser = argparse.ArgumentParser()
parser.add_argument('--np', type=int, default=1)
parser.add_argument('--host_ip', type=str, default=None)
parser.add_argument('--gpus_per_node', type=int, default=8)
parser.add_argument('--test_iteration', type=int, default=100)
parser.add_argument('--test_exe', type=str, default='/home/tmp1/rccl-tests/build/all_reduce_perf')
args = parser.parse_args()
nodes = str(args.np) + ' ' # number of mNodes
if (args.host_ip == None):
host_ip = ''
else :
host_ip = '-host ' + str(args.host_ip) + ' '
env1 = '-env HSA_FORCE_FINE_GRAIN_PCIE 1 '
env2 = '-env NCCL_DEBUG_INFO '
env3 = '-env NCCL_SOCKET_IFNAME enp98s0f0 '
env4 = '-env HIP_VISIBLE_DEVICES '
test_dir = str(args.test_exe) + ' '
test_param1 = '-b 8 ' # min size
test_param2 = '-e 128M ' # max size
test_param3 = '-f 2 '
test_param4 = '-g ' + str(args.gpus_per_node) + ' ' # number of GPUs
for i in range (1, args.test_iteration):
arr = np.arange(8)
np.random.shuffle(arr)
hip_dev = ','.join(str(i) for i in arr)
cmd = 'mpirun ' + '-np ' + nodes + host_ip + env1 + env3 + env4 + hip_dev + ' ' + test_dir + test_param1 + test_param2 + test_param3 + test_param4
print(cmd)
res = os.popen(cmd)
output = res.read()
print(output)
|
