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
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
|
! Implemented: i1, iv, im, it, r1, rv, rm, rt, r4, c1, cv, cm, ct, c4, c5, r5, r6
#if defined(__CUDA)
PROGRAM test_mp_sum_{vname}_gpu
!
! Simple program to check the functionalities of mp_sum.
!
! NB: mp_sum is allreduce
!
USE cudafor
USE parallel_include
USE util_param, ONLY : DP
USE mp, ONLY : mp_sum
USE mp_world, ONLY : mp_world_start, mp_world_end, mpime, &
root, nproc, world_comm
USE tester
IMPLICIT NONE
!
TYPE(tester_t) :: test
INTEGER :: world_group = 0, valid_sum, rnk
INTEGER, PARAMETER :: datasize = {datasize}
! for comparing with CPU implementation
INTEGER :: i
REAL(DP) :: rnd{size}
! test variable
{type}, DEVICE :: {vname}_d{size}
{type} :: {vname}_h{size}
{type} :: aux_h{size}
!
CALL test%init()
#if defined(__MPI)
world_group = MPI_COMM_WORLD
#endif
CALL mp_world_start(world_group)
{vname}_h = mpime
{vname}_d = {vname}_h
CALL mp_sum({vname}_d, world_comm)
{vname}_h = {vname}_d
!
! The sum of n rank is (zero based)
! sum = (n-1)*n*0.5
!
! For a rank N matrix is 2^N * sum
!
rnk = SIZE(SHAPE({vname}_h))
valid_sum = ({datasize}**rnk) * (nproc-1)*nproc/2
!
CALL test%assert_equal(INT({sumf}({vname}_h )) , valid_sum )
!
! Validate against CPU implementation
!
CALL save_random_seed("test_mp_sum_{vname}_gpu", mpime)
!
CALL RANDOM_NUMBER(rnd)
{vname}_h = {typeconv} ( 10.0 * rnd )
{vname}_d = {vname}_h
aux_h = 0
CALL mp_sum({vname}_d, world_comm)
CALL mp_sum({vname}_h, world_comm)
aux_h = {vname}_d
CALL test%assert_equal( {sumf}({vname}_h) , {sumf}(aux_h) )
!
CALL collect_results(test)
!
CALL mp_world_end()
!
IF (mpime .eq. 0) CALL test%print()
!
END PROGRAM test_mp_sum_{vname}_gpu
#else
PROGRAM test_mp_sum_{vname}_gpu
CALL no_test()
END PROGRAM test_mp_sum_{vname}_gpu
#endif
|
