mod_vendor_agnostic_layer.F90 6.77 KB
Newer Older
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
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
#include "config-f90.h"
module elpa_gpu
  use iso_c_binding

  integer(kind=c_int), parameter :: nvidia_gpu = 1
  integer(kind=c_int), parameter :: amd_gpu = 2
  integer(kind=c_int), parameter :: no_gpu = -1
  integer(kind=c_int)            :: use_gpu_vendor
  integer(kind=c_int)            :: gpuHostRegisterDefault    
  integer(kind=c_int)            :: gpuMemcpyHostToDevice    
  integer(kind=c_int)            :: gpuMemcpyDeviceToHost   
  integer(kind=c_int)            :: gpuMemcpyDeviceToDevice
  integer(kind=c_int)            :: gpuHostRegisterMapped
  integer(kind=c_int)            :: gpuHostRegisterPortable

  contains
    function gpu_vendor() result(vendor)
      use precision
      implicit none
      integer(kind=c_int) :: vendor
      ! default
      vendor = no_gpu
#ifdef WITH_NVIDIA_GPU_VERSION
      vendor = nvidia_gpu
#endif
#ifdef WITH_AMD_GPU_VERSION
      vendor = amd_gpu
#endif
      use_gpu_vendor = vendor
      return
    end function

    subroutine set_gpu_parameters
      use cuda_functions
      use hip_functions
      implicit none

      if (use_gpu_vendor == nvidia_gpu) then
        cudaMemcpyHostToDevice   = cuda_memcpyHostToDevice()
        gpuMemcpyHostToDevice    = cudaMemcpyHostToDevice
        cudaMemcpyDeviceToHost   = cuda_memcpyDeviceToHost()
        gpuMemcpyDeviceToHost    = cudaMemcpyDeviceToHost
        cudaMemcpyDeviceToDevice = cuda_memcpyDeviceToDevice()
        gpuMemcpyDeviceToDevice  = cudaMemcpyDeviceToDevice
        cudaHostRegisterPortable = cuda_hostRegisterPortable()
        gpuHostRegisterPortable  = cudaHostRegisterPortable
        cudaHostRegisterMapped   = cuda_hostRegisterMapped()
        gpuHostRegisterMapped    = cudaHostRegisterMapped
        cudaHostRegisterDefault  = cuda_hostRegisterDefault()
        gpuHostRegisterDefault   = cudaHostRegisterDefault
      endif

      if (use_gpu_vendor == amd_gpu) then
        hipMemcpyHostToDevice   = hip_memcpyHostToDevice()
        gpuMemcpyHostToDevice   = hipMemcpyHostToDevice
        hipMemcpyDeviceToHost   = hip_memcpyDeviceToHost()
        gpuMemcpyDeviceToHost   = hipMemcpyDeviceToHost
        hipMemcpyDeviceToDevice = hip_memcpyDeviceToDevice()
        gpuMemcpyDeviceToDevice = hipMemcpyDeviceToDevice
        hipHostRegisterPortable = hip_hostRegisterPortable()
        gpuHostRegisterPortable = hipHostRegisterPortable
        hipHostRegisterMapped   = hip_hostRegisterMapped()
        gpuHostRegisterMapped   = hipHostRegisterMapped
        hipHostRegisterDefault  = hip_hostRegisterDefault()
        gpuHostRegisterDefault  = hipHostRegisterDefault
      endif

    end subroutine

    function gpu_malloc_host(array, elements) result(success)
      use, intrinsic :: iso_c_binding
      use cuda_functions
      use hip_functions
      implicit none
      type(c_ptr)                          :: array
      integer(kind=c_intptr_t), intent(in) :: elements
      logical                              :: success

      if (use_gpu_vendor == nvidia_gpu) then
        success = cuda_malloc_host(array, elements)
      endif

      if (use_gpu_vendor == amd_gpu) then
        success = hip_host_malloc(array, elements)
      endif

    end function

    function gpu_malloc(array, elements) result(success)
      use, intrinsic :: iso_c_binding
      use cuda_functions
      use hip_functions
      implicit none
      integer(kind=C_intptr_T)             :: array
      integer(kind=c_intptr_t), intent(in) :: elements
      logical                              :: success

      if (use_gpu_vendor == nvidia_gpu) then
        success = cuda_malloc(array, elements)
      endif

      if (use_gpu_vendor == amd_gpu) then
        success = hip_malloc(array, elements)
      endif

    end function

    function gpu_host_register(array, elements, flag) result(success)
      use, intrinsic :: iso_c_binding
      use cuda_functions
      use hip_functions
      implicit none
      integer(kind=C_intptr_t)              :: array
      integer(kind=c_intptr_t), intent(in)  :: elements
      integer(kind=C_INT), intent(in)       :: flag
      logical :: success

      if (use_gpu_vendor == nvidia_gpu) then
        success = cuda_host_register(array, elements, flag)
      endif

      if (use_gpu_vendor == amd_gpu) then
        success = hip_host_register(array, elements, flag)
      endif

    end function
    
    function gpu_memcpy(dst, src, size, dir) result(success)
      use, intrinsic :: iso_c_binding
      use cuda_functions
      use hip_functions
      implicit none
      integer(kind=C_intptr_t)              :: dst
      integer(kind=C_intptr_t)              :: src
      integer(kind=c_intptr_t), intent(in)  :: size
      integer(kind=C_INT), intent(in)       :: dir
      logical :: success

      if (use_gpu_vendor == nvidia_gpu) then
        success = cuda_memcpy(dst, src, size, dir)
      endif

      if (use_gpu_vendor == amd_gpu) then
        success = hip_memcpy(dst, src, size, dir)
      endif
    
    end function


    function gpu_memset(a, val, size) result(success)
      use, intrinsic :: iso_c_binding
      use cuda_functions
      use hip_functions
      implicit none
      integer(kind=c_intptr_t)                :: a
      integer(kind=ik)                        :: val
      integer(kind=c_intptr_t), intent(in)      :: size
      integer(kind=C_INT)                     :: istat

      logical :: success

      if (use_gpu_vendor == nvidia_gpu) then
        success = cuda_memset(a, val, size)
      endif

      if (use_gpu_vendor == amd_gpu) then
        success = hip_memset(a, val, size)
      endif

    end function

    function gpu_free(a) result(success)
      use, intrinsic :: iso_c_binding
      use cuda_functions
      use hip_functions
      implicit none
      integer(kind=c_intptr_t)                :: a

      logical :: success

      if (use_gpu_vendor == nvidia_gpu) then
        success = cuda_free(a)
      endif

      if (use_gpu_vendor == amd_gpu) then
        success = hip_free(a)
      endif

    end function

    function gpu_free_host(a) result(success)
      use, intrinsic :: iso_c_binding
      use cuda_functions
      use hip_functions
      implicit none
      type(c_ptr), value          :: a

      logical :: success

      if (use_gpu_vendor == nvidia_gpu) then
        success = cuda_free_host(a)
      endif

      if (use_gpu_vendor == amd_gpu) then
        success = hip_host_free(a)
      endif

    end function

    function gpu_host_unregister(a) result(success)
      use, intrinsic :: iso_c_binding
      use cuda_functions
      use hip_functions
      implicit none
      integer(kind=c_intptr_t)                :: a

      logical :: success

      if (use_gpu_vendor == nvidia_gpu) then
        success = cuda_host_unregister(a)
      endif

      if (use_gpu_vendor == amd_gpu) then
        success = hip_host_unregister(a)
      endif

    end function
    
end module