mod_vendor_agnostic_layer.F90 16.8 KB
Newer Older
1
2
#include "config-f90.h"
module elpa_gpu
3
  use precision
4
5
6
7
  use iso_c_binding

  integer(kind=c_int), parameter :: nvidia_gpu = 1
  integer(kind=c_int), parameter :: amd_gpu = 2
8
  integer(kind=c_int), parameter :: intel_gpu = 3
9
10
11
12
13
14
15
16
17
  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

18
19
20
21
22
23
24
25
26
  integer(kind=c_intptr_t), parameter :: size_of_double_real    = 8_rk8
#ifdef WANT_SINGLE_PRECISION_REAL
  integer(kind=c_intptr_t), parameter :: size_of_single_real    = 4_rk4
#endif

  integer(kind=c_intptr_t), parameter :: size_of_double_complex = 16_ck8
#ifdef WANT_SINGLE_PRECISION_COMPLEX
  integer(kind=c_intptr_t), parameter :: size_of_single_complex = 8_ck4
#endif
27
28
29
30
31
32
33
34
35
36
37
38
  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
39
40
41
#endif
#ifdef WITH_INTEL_GPU_VERSION
      vendor = intel_gpu
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
230
231
232
233
234
235
236
237
238
239
240
241
#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
Andreas Marek's avatar
Andreas Marek committed
242

Andreas Marek's avatar
Andreas Marek committed
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269

    function gpu_memcpy2d(dst, dpitch, src, spitch, width, height , 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), intent(in) :: dpitch
      integer(kind=C_intptr_T)           :: src
      integer(kind=c_intptr_t), intent(in) :: spitch
      integer(kind=c_intptr_t), intent(in) :: width
      integer(kind=c_intptr_t), intent(in) :: height
      integer(kind=C_INT), intent(in)    :: dir
      logical                            :: success

      if (use_gpu_vendor == nvidia_gpu) then
        success = cuda_memcpy2d(dst, dpitch, src, spitch, width, height , dir)
      endif

      if (use_gpu_vendor == amd_gpu) then
        success = hip_memcpy2d(dst, dpitch, src, spitch, width, height , dir)
      endif
    end function

Andreas Marek's avatar
Andreas Marek committed
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
    subroutine gpublas_dgemv(cta, m, n, alpha, a, lda, x, incx, beta, y, incy)
      use, intrinsic :: iso_c_binding
      use cuda_functions
      use hip_functions

      implicit none
      character(1,C_CHAR),value       :: cta
      integer(kind=C_INT)             :: m,n
      integer(kind=C_INT), intent(in) :: lda,incx,incy
      real(kind=C_DOUBLE)             :: alpha,beta
      integer(kind=C_intptr_T)        :: a, x, y

      if (use_gpu_vendor == nvidia_gpu) then
        call cublas_dgemv(cta, m, n, alpha, a, lda, x, incx, beta, y, incy)
      endif

      if (use_gpu_vendor == amd_gpu) then
        call rocblas_dgemv(cta, m, n, alpha, a, lda, x, incx, beta, y, incy)
      endif
    end subroutine

    subroutine gpublas_sgemv(cta, m, n, alpha, a, lda, x, incx, beta, y, incy)
      use, intrinsic :: iso_c_binding
      use cuda_functions
      use hip_functions

      implicit none
      character(1,C_CHAR),value       :: cta
      integer(kind=C_INT)             :: m,n
      integer(kind=C_INT), intent(in) :: lda,incx,incy
      real(kind=C_FLOAT)              :: alpha,beta
      integer(kind=C_intptr_T)        :: a, x, y

      if (use_gpu_vendor == nvidia_gpu) then
        call cublas_sgemv(cta, m, n, alpha, a, lda, x, incx, beta, y, incy)
      endif

      if (use_gpu_vendor == amd_gpu) then
        call rocblas_sgemv(cta, m, n, alpha, a, lda, x, incx, beta, y, incy)
      endif

    end subroutine

    subroutine gpublas_zgemv(cta, m, n, alpha, a, lda, x, incx, beta, y, incy)
      use, intrinsic :: iso_c_binding
      use cuda_functions
      use hip_functions

      implicit none
      character(1,C_CHAR),value       :: cta
      integer(kind=C_INT)             :: m,n
      integer(kind=C_INT), intent(in) :: lda,incx,incy
      complex(kind=C_DOUBLE_COMPLEX)             :: alpha,beta
      integer(kind=C_intptr_T)        :: a, x, y

      if (use_gpu_vendor == nvidia_gpu) then
        call cublas_zgemv(cta, m, n, alpha, a, lda, x, incx, beta, y, incy)
      endif

      if (use_gpu_vendor == amd_gpu) then
        call rocblas_zgemv(cta, m, n, alpha, a, lda, x, incx, beta, y, incy)
      endif

    end subroutine

    subroutine gpublas_cgemv(cta, m, n, alpha, a, lda, x, incx, beta, y, incy)
      use, intrinsic :: iso_c_binding
      use cuda_functions
      use hip_functions

      implicit none
      character(1,C_CHAR),value       :: cta
      integer(kind=C_INT)             :: m,n
      integer(kind=C_INT), intent(in) :: lda,incx,incy
      complex(kind=C_FLOAT_COMPLEX)              :: alpha,beta
      integer(kind=C_intptr_T)        :: a, x, y

      if (use_gpu_vendor == nvidia_gpu) then
        call cublas_cgemv(cta, m, n, alpha, a, lda, x, incx, beta, y, incy)
      endif

      if (use_gpu_vendor == amd_gpu) then
        call rocblas_cgemv(cta, m, n, alpha, a, lda, x, incx, beta, y, incy)
      endif

    end subroutine

    subroutine gpublas_dgemm(cta, ctb, m, n, k, alpha, a, lda, b, ldb, beta, c, ldc)
      use, intrinsic :: iso_c_binding
      use cuda_functions
      use hip_functions

      implicit none
      character(1,C_CHAR),value       :: cta, ctb
      integer(kind=C_INT)             :: m,n,k
      integer(kind=C_INT), intent(in) :: lda,ldb,ldc
      real(kind=C_DOUBLE)             :: alpha,beta
      integer(kind=C_intptr_T)        :: a, b, c

      if (use_gpu_vendor == nvidia_gpu) then
        call cublas_dgemm(cta, ctb, m, n, k, alpha, a, lda, b, ldb, beta, c,ldc)
      endif

      if (use_gpu_vendor == amd_gpu) then
        call rocblas_dgemm(cta, ctb, m, n, k, alpha, a, lda, b, ldb, beta, c,ldc)
      endif

    end subroutine 


    subroutine gpublas_sgemm(cta, ctb, m, n, k, alpha, a, lda, b, ldb, beta, c, ldc)
      use, intrinsic :: iso_c_binding
      use cuda_functions
      use hip_functions

      implicit none
      character(1,C_CHAR),value       :: cta, ctb
      integer(kind=C_INT)             :: m,n,k
      integer(kind=C_INT), intent(in) :: lda,ldb,ldc
      real(kind=C_FLOAT)              :: alpha,beta
      integer(kind=C_intptr_T)        :: a, b, c

      if (use_gpu_vendor == nvidia_gpu) then
        call cublas_sgemm(cta, ctb, m, n, k, alpha, a, lda, b, ldb, beta, c,ldc)
      endif

      if (use_gpu_vendor == amd_gpu) then
        call rocblas_sgemm(cta, ctb, m, n, k, alpha, a, lda, b, ldb, beta, c,ldc)
      endif


    end subroutine

    subroutine gpublas_zgemm(cta, ctb, m, n, k, alpha, a, lda, b, ldb, beta, c,ldc)

      use, intrinsic :: iso_c_binding
      use cuda_functions
      use hip_functions

      implicit none
      character(1,C_CHAR),value       :: cta, ctb
      integer(kind=C_INT)             :: m,n,k
      integer(kind=C_INT), intent(in) :: lda,ldb,ldc
      complex(kind=C_DOUBLE_COMPLEX)          :: alpha,beta
      integer(kind=C_intptr_T)        :: a, b, c

      if (use_gpu_vendor == nvidia_gpu) then
        call cublas_zgemm(cta, ctb, m, n, k, alpha, a, lda, b, ldb, beta, c,ldc)
      endif

      if (use_gpu_vendor == amd_gpu) then
        call rocblas_zgemm(cta, ctb, m, n, k, alpha, a, lda, b, ldb, beta, c,ldc)
      endif

    end subroutine

    subroutine gpublas_cgemm(cta, ctb, m, n, k, alpha, a, lda, b, ldb, beta, c,ldc)

      use, intrinsic :: iso_c_binding
      use cuda_functions
      use hip_functions

      implicit none
      character(1,C_CHAR),value       :: cta, ctb
      integer(kind=C_INT)             :: m,n,k
      integer(kind=C_INT), intent(in) :: lda,ldb,ldc
      complex(kind=C_FLOAT_COMPLEX)           :: alpha,beta
      integer(kind=C_intptr_T)        :: a, b, c

      if (use_gpu_vendor == nvidia_gpu) then
        call cublas_cgemm(cta, ctb, m, n, k, alpha, a, lda, b, ldb, beta, c,ldc)
      endif

      if (use_gpu_vendor == amd_gpu) then
        call rocblas_cgemm(cta, ctb, m, n, k, alpha, a, lda, b, ldb, beta, c,ldc)
      endif
    end subroutine


449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
    subroutine gpublas_dtrmm(side, uplo, trans, diag, m, n, alpha, a, lda, b, ldb)

      use, intrinsic :: iso_c_binding
      use cuda_functions
      use hip_functions

      implicit none
      character(1,C_CHAR),value       :: side, uplo, trans, diag
      integer(kind=C_INT)             :: m,n
      integer(kind=C_INT), intent(in) :: lda,ldb
      real(kind=C_DOUBLE)             :: alpha
      integer(kind=C_intptr_T)        :: a, b

      if (use_gpu_vendor == nvidia_gpu) then
        call cublas_dtrmm(side, uplo, trans, diag, m, n, alpha, a, lda, b, ldb)
      endif

      if (use_gpu_vendor == amd_gpu) then
        call rocblas_dtrmm(side, uplo, trans, diag, m, n, alpha, a, lda, b, ldb)
      endif

    end subroutine


    subroutine gpublas_strmm(side, uplo, trans, diag, m, n, alpha, a, lda, b, ldb)

      use, intrinsic :: iso_c_binding
      use cuda_functions
      use hip_functions

      implicit none
      character(1,C_CHAR),value       :: side, uplo, trans, diag
      integer(kind=C_INT)             :: m,n
      integer(kind=C_INT), intent(in) :: lda,ldb
      real(kind=C_FLOAT)              :: alpha
      integer(kind=C_intptr_T)        :: a, b

      if (use_gpu_vendor == nvidia_gpu) then
        call cublas_strmm(side, uplo, trans, diag, m, n, alpha, a, lda, b, ldb)
      endif

      if (use_gpu_vendor == amd_gpu) then
        call rocblas_strmm(side, uplo, trans, diag, m, n, alpha, a, lda, b, ldb)
      endif
    end subroutine



    subroutine gpublas_ztrmm(side, uplo, trans, diag, m, n, alpha, a, lda, b, ldb)

      use, intrinsic :: iso_c_binding
      use cuda_functions
      use hip_functions

      implicit none
      character(1,C_CHAR),value       :: side, uplo, trans, diag
      integer(kind=C_INT)             :: m,n
      integer(kind=C_INT), intent(in) :: lda,ldb
      complex(kind=C_DOUBLE_COMPLEX)          :: alpha
      integer(kind=C_intptr_T)        :: a, b

      if (use_gpu_vendor == nvidia_gpu) then
        call cublas_ztrmm(side, uplo, trans, diag, m, n, alpha, a, lda, b, ldb)
      endif

      if (use_gpu_vendor == amd_gpu) then
        call rocblas_ztrmm(side, uplo, trans, diag, m, n, alpha, a, lda, b, ldb)
      endif
    end subroutine


    subroutine gpublas_ctrmm(side, uplo, trans, diag, m, n, alpha, a, lda, b, ldb)

      use, intrinsic :: iso_c_binding
      use cuda_functions
      use hip_functions

      implicit none
      character(1,C_CHAR),value       :: side, uplo, trans, diag
      integer(kind=C_INT)             :: m,n
      integer(kind=C_INT), intent(in) :: lda,ldb
      complex(kind=C_FLOAT_COMPLEX)           :: alpha
      integer(kind=C_intptr_T)        :: a, b

      if (use_gpu_vendor == nvidia_gpu) then
        call cublas_ctrmm(side, uplo, trans, diag, m, n, alpha, a, lda, b, ldb)
      endif

      if (use_gpu_vendor == amd_gpu) then
        call rocblas_ctrmm(side, uplo, trans, diag, m, n, alpha, a, lda, b, ldb)
      endif
    end subroutine


543
end module