MFC
Exascale flow solver
Loading...
Searching...
No Matches
m_amr_wave.fpp.f90
Go to the documentation of this file.
1# 1 "/home/runner/work/MFC/MFC/src/simulation/m_amr_wave.fpp"
2!>
3!!@file
4!!@brief Contains module m_amr_wave
5
6# 1 "/home/runner/work/MFC/MFC/src/common/include/macros.fpp" 1
7# 1 "/home/runner/work/MFC/MFC/src/common/include/parallel_macros.fpp" 1
8# 1 "/home/runner/work/MFC/MFC/src/common/include/shared_parallel_macros.fpp" 1
9# 2 "/home/runner/work/MFC/MFC/src/common/include/shared_parallel_macros.fpp"
10# 3 "/home/runner/work/MFC/MFC/src/common/include/shared_parallel_macros.fpp"
11# 4 "/home/runner/work/MFC/MFC/src/common/include/shared_parallel_macros.fpp"
12# 5 "/home/runner/work/MFC/MFC/src/common/include/shared_parallel_macros.fpp"
13# 6 "/home/runner/work/MFC/MFC/src/common/include/shared_parallel_macros.fpp"
14
15# 8 "/home/runner/work/MFC/MFC/src/common/include/shared_parallel_macros.fpp"
16# 9 "/home/runner/work/MFC/MFC/src/common/include/shared_parallel_macros.fpp"
17# 10 "/home/runner/work/MFC/MFC/src/common/include/shared_parallel_macros.fpp"
18
19# 12 "/home/runner/work/MFC/MFC/src/common/include/shared_parallel_macros.fpp"
20# 1 "/home/runner/work/MFC/MFC/src/common/include/defaults/thermochem.fpp" 1
21! Default for Fypp runs outside the toolchain (e.g. documentation). Searched last, so the
22! build/include/<target>/thermochem.fpp the toolchain writes from m_thermochem takes precedence.
23# 4 "/home/runner/work/MFC/MFC/src/common/include/defaults/thermochem.fpp"
24# 5 "/home/runner/work/MFC/MFC/src/common/include/defaults/thermochem.fpp"
25# 13 "/home/runner/work/MFC/MFC/src/common/include/shared_parallel_macros.fpp" 2
26
27# 16 "/home/runner/work/MFC/MFC/src/common/include/shared_parallel_macros.fpp"
28# 17 "/home/runner/work/MFC/MFC/src/common/include/shared_parallel_macros.fpp"
29
30# 24 "/home/runner/work/MFC/MFC/src/common/include/shared_parallel_macros.fpp"
31
32# 53 "/home/runner/work/MFC/MFC/src/common/include/shared_parallel_macros.fpp"
33
34# 65 "/home/runner/work/MFC/MFC/src/common/include/shared_parallel_macros.fpp"
35
36# 75 "/home/runner/work/MFC/MFC/src/common/include/shared_parallel_macros.fpp"
37
38# 105 "/home/runner/work/MFC/MFC/src/common/include/shared_parallel_macros.fpp"
39
40# 117 "/home/runner/work/MFC/MFC/src/common/include/shared_parallel_macros.fpp"
41
42# 127 "/home/runner/work/MFC/MFC/src/common/include/shared_parallel_macros.fpp"
43
44# 174 "/home/runner/work/MFC/MFC/src/common/include/shared_parallel_macros.fpp"
45! New line at end of file is required for FYPP
46# 2 "/home/runner/work/MFC/MFC/src/common/include/parallel_macros.fpp" 2
47# 1 "/home/runner/work/MFC/MFC/src/common/include/omp_macros.fpp" 1
48# 1 "/home/runner/work/MFC/MFC/src/common/include/shared_parallel_macros.fpp" 1
49# 2 "/home/runner/work/MFC/MFC/src/common/include/shared_parallel_macros.fpp"
50# 3 "/home/runner/work/MFC/MFC/src/common/include/shared_parallel_macros.fpp"
51# 4 "/home/runner/work/MFC/MFC/src/common/include/shared_parallel_macros.fpp"
52# 5 "/home/runner/work/MFC/MFC/src/common/include/shared_parallel_macros.fpp"
53# 6 "/home/runner/work/MFC/MFC/src/common/include/shared_parallel_macros.fpp"
54
55# 8 "/home/runner/work/MFC/MFC/src/common/include/shared_parallel_macros.fpp"
56# 9 "/home/runner/work/MFC/MFC/src/common/include/shared_parallel_macros.fpp"
57# 10 "/home/runner/work/MFC/MFC/src/common/include/shared_parallel_macros.fpp"
58
59# 12 "/home/runner/work/MFC/MFC/src/common/include/shared_parallel_macros.fpp"
60# 1 "/home/runner/work/MFC/MFC/src/common/include/defaults/thermochem.fpp" 1
61! Default for Fypp runs outside the toolchain (e.g. documentation). Searched last, so the
62! build/include/<target>/thermochem.fpp the toolchain writes from m_thermochem takes precedence.
63# 4 "/home/runner/work/MFC/MFC/src/common/include/defaults/thermochem.fpp"
64# 5 "/home/runner/work/MFC/MFC/src/common/include/defaults/thermochem.fpp"
65# 13 "/home/runner/work/MFC/MFC/src/common/include/shared_parallel_macros.fpp" 2
66
67# 16 "/home/runner/work/MFC/MFC/src/common/include/shared_parallel_macros.fpp"
68# 17 "/home/runner/work/MFC/MFC/src/common/include/shared_parallel_macros.fpp"
69
70# 24 "/home/runner/work/MFC/MFC/src/common/include/shared_parallel_macros.fpp"
71
72# 53 "/home/runner/work/MFC/MFC/src/common/include/shared_parallel_macros.fpp"
73
74# 65 "/home/runner/work/MFC/MFC/src/common/include/shared_parallel_macros.fpp"
75
76# 75 "/home/runner/work/MFC/MFC/src/common/include/shared_parallel_macros.fpp"
77
78# 105 "/home/runner/work/MFC/MFC/src/common/include/shared_parallel_macros.fpp"
79
80# 117 "/home/runner/work/MFC/MFC/src/common/include/shared_parallel_macros.fpp"
81
82# 127 "/home/runner/work/MFC/MFC/src/common/include/shared_parallel_macros.fpp"
83
84# 174 "/home/runner/work/MFC/MFC/src/common/include/shared_parallel_macros.fpp"
85! New line at end of file is required for FYPP
86# 2 "/home/runner/work/MFC/MFC/src/common/include/omp_macros.fpp" 2
87
88# 4 "/home/runner/work/MFC/MFC/src/common/include/omp_macros.fpp"
89# 5 "/home/runner/work/MFC/MFC/src/common/include/omp_macros.fpp"
90# 6 "/home/runner/work/MFC/MFC/src/common/include/omp_macros.fpp"
91# 7 "/home/runner/work/MFC/MFC/src/common/include/omp_macros.fpp"
92# 8 "/home/runner/work/MFC/MFC/src/common/include/omp_macros.fpp"
93
94# 20 "/home/runner/work/MFC/MFC/src/common/include/omp_macros.fpp"
95
96# 49 "/home/runner/work/MFC/MFC/src/common/include/omp_macros.fpp"
97
98# 54 "/home/runner/work/MFC/MFC/src/common/include/omp_macros.fpp"
99
100# 59 "/home/runner/work/MFC/MFC/src/common/include/omp_macros.fpp"
101
102# 64 "/home/runner/work/MFC/MFC/src/common/include/omp_macros.fpp"
103
104# 69 "/home/runner/work/MFC/MFC/src/common/include/omp_macros.fpp"
105
106# 74 "/home/runner/work/MFC/MFC/src/common/include/omp_macros.fpp"
107
108# 87 "/home/runner/work/MFC/MFC/src/common/include/omp_macros.fpp"
109
110# 101 "/home/runner/work/MFC/MFC/src/common/include/omp_macros.fpp"
111
112# 106 "/home/runner/work/MFC/MFC/src/common/include/omp_macros.fpp"
113
114# 111 "/home/runner/work/MFC/MFC/src/common/include/omp_macros.fpp"
115
116# 116 "/home/runner/work/MFC/MFC/src/common/include/omp_macros.fpp"
117
118# 121 "/home/runner/work/MFC/MFC/src/common/include/omp_macros.fpp"
119
120# 126 "/home/runner/work/MFC/MFC/src/common/include/omp_macros.fpp"
121
122# 131 "/home/runner/work/MFC/MFC/src/common/include/omp_macros.fpp"
123
124# 136 "/home/runner/work/MFC/MFC/src/common/include/omp_macros.fpp"
125
126# 141 "/home/runner/work/MFC/MFC/src/common/include/omp_macros.fpp"
127
128# 171 "/home/runner/work/MFC/MFC/src/common/include/omp_macros.fpp"
129
130# 231 "/home/runner/work/MFC/MFC/src/common/include/omp_macros.fpp"
131
132# 245 "/home/runner/work/MFC/MFC/src/common/include/omp_macros.fpp"
133
134# 270 "/home/runner/work/MFC/MFC/src/common/include/omp_macros.fpp"
135
136# 281 "/home/runner/work/MFC/MFC/src/common/include/omp_macros.fpp"
137
138# 283 "/home/runner/work/MFC/MFC/src/common/include/omp_macros.fpp"
139# 294 "/home/runner/work/MFC/MFC/src/common/include/omp_macros.fpp"
140
141# 343 "/home/runner/work/MFC/MFC/src/common/include/omp_macros.fpp"
142
143# 353 "/home/runner/work/MFC/MFC/src/common/include/omp_macros.fpp"
144
145# 363 "/home/runner/work/MFC/MFC/src/common/include/omp_macros.fpp"
146
147# 372 "/home/runner/work/MFC/MFC/src/common/include/omp_macros.fpp"
148
149# 389 "/home/runner/work/MFC/MFC/src/common/include/omp_macros.fpp"
150
151# 399 "/home/runner/work/MFC/MFC/src/common/include/omp_macros.fpp"
152
153# 406 "/home/runner/work/MFC/MFC/src/common/include/omp_macros.fpp"
154
155# 412 "/home/runner/work/MFC/MFC/src/common/include/omp_macros.fpp"
156
157# 418 "/home/runner/work/MFC/MFC/src/common/include/omp_macros.fpp"
158
159# 424 "/home/runner/work/MFC/MFC/src/common/include/omp_macros.fpp"
160
161# 430 "/home/runner/work/MFC/MFC/src/common/include/omp_macros.fpp"
162
163# 436 "/home/runner/work/MFC/MFC/src/common/include/omp_macros.fpp"
164! New line at end of file is required for FYPP
165# 3 "/home/runner/work/MFC/MFC/src/common/include/parallel_macros.fpp" 2
166# 1 "/home/runner/work/MFC/MFC/src/common/include/acc_macros.fpp" 1
167# 1 "/home/runner/work/MFC/MFC/src/common/include/shared_parallel_macros.fpp" 1
168# 2 "/home/runner/work/MFC/MFC/src/common/include/shared_parallel_macros.fpp"
169# 3 "/home/runner/work/MFC/MFC/src/common/include/shared_parallel_macros.fpp"
170# 4 "/home/runner/work/MFC/MFC/src/common/include/shared_parallel_macros.fpp"
171# 5 "/home/runner/work/MFC/MFC/src/common/include/shared_parallel_macros.fpp"
172# 6 "/home/runner/work/MFC/MFC/src/common/include/shared_parallel_macros.fpp"
173
174# 8 "/home/runner/work/MFC/MFC/src/common/include/shared_parallel_macros.fpp"
175# 9 "/home/runner/work/MFC/MFC/src/common/include/shared_parallel_macros.fpp"
176# 10 "/home/runner/work/MFC/MFC/src/common/include/shared_parallel_macros.fpp"
177
178# 12 "/home/runner/work/MFC/MFC/src/common/include/shared_parallel_macros.fpp"
179# 1 "/home/runner/work/MFC/MFC/src/common/include/defaults/thermochem.fpp" 1
180! Default for Fypp runs outside the toolchain (e.g. documentation). Searched last, so the
181! build/include/<target>/thermochem.fpp the toolchain writes from m_thermochem takes precedence.
182# 4 "/home/runner/work/MFC/MFC/src/common/include/defaults/thermochem.fpp"
183# 5 "/home/runner/work/MFC/MFC/src/common/include/defaults/thermochem.fpp"
184# 13 "/home/runner/work/MFC/MFC/src/common/include/shared_parallel_macros.fpp" 2
185
186# 16 "/home/runner/work/MFC/MFC/src/common/include/shared_parallel_macros.fpp"
187# 17 "/home/runner/work/MFC/MFC/src/common/include/shared_parallel_macros.fpp"
188
189# 24 "/home/runner/work/MFC/MFC/src/common/include/shared_parallel_macros.fpp"
190
191# 53 "/home/runner/work/MFC/MFC/src/common/include/shared_parallel_macros.fpp"
192
193# 65 "/home/runner/work/MFC/MFC/src/common/include/shared_parallel_macros.fpp"
194
195# 75 "/home/runner/work/MFC/MFC/src/common/include/shared_parallel_macros.fpp"
196
197# 105 "/home/runner/work/MFC/MFC/src/common/include/shared_parallel_macros.fpp"
198
199# 117 "/home/runner/work/MFC/MFC/src/common/include/shared_parallel_macros.fpp"
200
201# 127 "/home/runner/work/MFC/MFC/src/common/include/shared_parallel_macros.fpp"
202
203# 174 "/home/runner/work/MFC/MFC/src/common/include/shared_parallel_macros.fpp"
204! New line at end of file is required for FYPP
205# 2 "/home/runner/work/MFC/MFC/src/common/include/acc_macros.fpp" 2
206
207# 7 "/home/runner/work/MFC/MFC/src/common/include/acc_macros.fpp"
208
209# 17 "/home/runner/work/MFC/MFC/src/common/include/acc_macros.fpp"
210
211# 22 "/home/runner/work/MFC/MFC/src/common/include/acc_macros.fpp"
212
213# 27 "/home/runner/work/MFC/MFC/src/common/include/acc_macros.fpp"
214
215# 32 "/home/runner/work/MFC/MFC/src/common/include/acc_macros.fpp"
216
217# 37 "/home/runner/work/MFC/MFC/src/common/include/acc_macros.fpp"
218
219# 42 "/home/runner/work/MFC/MFC/src/common/include/acc_macros.fpp"
220
221# 47 "/home/runner/work/MFC/MFC/src/common/include/acc_macros.fpp"
222
223# 52 "/home/runner/work/MFC/MFC/src/common/include/acc_macros.fpp"
224
225# 57 "/home/runner/work/MFC/MFC/src/common/include/acc_macros.fpp"
226
227# 62 "/home/runner/work/MFC/MFC/src/common/include/acc_macros.fpp"
228
229# 73 "/home/runner/work/MFC/MFC/src/common/include/acc_macros.fpp"
230
231# 78 "/home/runner/work/MFC/MFC/src/common/include/acc_macros.fpp"
232
233# 83 "/home/runner/work/MFC/MFC/src/common/include/acc_macros.fpp"
234
235# 88 "/home/runner/work/MFC/MFC/src/common/include/acc_macros.fpp"
236
237# 103 "/home/runner/work/MFC/MFC/src/common/include/acc_macros.fpp"
238
239# 131 "/home/runner/work/MFC/MFC/src/common/include/acc_macros.fpp"
240
241# 161 "/home/runner/work/MFC/MFC/src/common/include/acc_macros.fpp"
242
243# 176 "/home/runner/work/MFC/MFC/src/common/include/acc_macros.fpp"
244
245# 194 "/home/runner/work/MFC/MFC/src/common/include/acc_macros.fpp"
246
247# 216 "/home/runner/work/MFC/MFC/src/common/include/acc_macros.fpp"
248
249# 245 "/home/runner/work/MFC/MFC/src/common/include/acc_macros.fpp"
250
251# 260 "/home/runner/work/MFC/MFC/src/common/include/acc_macros.fpp"
252
253# 270 "/home/runner/work/MFC/MFC/src/common/include/acc_macros.fpp"
254
255# 279 "/home/runner/work/MFC/MFC/src/common/include/acc_macros.fpp"
256
257# 295 "/home/runner/work/MFC/MFC/src/common/include/acc_macros.fpp"
258
259# 305 "/home/runner/work/MFC/MFC/src/common/include/acc_macros.fpp"
260
261# 312 "/home/runner/work/MFC/MFC/src/common/include/acc_macros.fpp"
262! New line at end of file is required for FYPP
263# 4 "/home/runner/work/MFC/MFC/src/common/include/parallel_macros.fpp" 2
264
265! GPU parallel region (scalar reductions, maxval/minval)
266# 23 "/home/runner/work/MFC/MFC/src/common/include/parallel_macros.fpp"
267
268! GPU parallel loop over threads (most common GPU macro)
269# 44 "/home/runner/work/MFC/MFC/src/common/include/parallel_macros.fpp"
270
271! Required closing for GPU_PARALLEL_LOOP
272# 56 "/home/runner/work/MFC/MFC/src/common/include/parallel_macros.fpp"
273
274! Mark routine for device compilation
275# 113 "/home/runner/work/MFC/MFC/src/common/include/parallel_macros.fpp"
276
277! Declare device-resident data
278# 131 "/home/runner/work/MFC/MFC/src/common/include/parallel_macros.fpp"
279
280! Inner loop within a GPU parallel region
281# 146 "/home/runner/work/MFC/MFC/src/common/include/parallel_macros.fpp"
282
283! Scoped GPU data region
284# 165 "/home/runner/work/MFC/MFC/src/common/include/parallel_macros.fpp"
285
286! Host code with device pointers (for MPI with GPU buffers)
287# 191 "/home/runner/work/MFC/MFC/src/common/include/parallel_macros.fpp"
288
289! Allocate device memory (unscoped)
290# 205 "/home/runner/work/MFC/MFC/src/common/include/parallel_macros.fpp"
291
292! Free device memory
293# 217 "/home/runner/work/MFC/MFC/src/common/include/parallel_macros.fpp"
294
295! Atomic operation on device
296# 229 "/home/runner/work/MFC/MFC/src/common/include/parallel_macros.fpp"
297
298! End atomic capture block
299# 240 "/home/runner/work/MFC/MFC/src/common/include/parallel_macros.fpp"
300
301! Copy data between host and device
302# 252 "/home/runner/work/MFC/MFC/src/common/include/parallel_macros.fpp"
303
304! Synchronization barrier
305# 264 "/home/runner/work/MFC/MFC/src/common/include/parallel_macros.fpp"
306
307! Import GPU library module (openacc or omp_lib)
308# 273 "/home/runner/work/MFC/MFC/src/common/include/parallel_macros.fpp"
309
310! Emit code only for AMD compiler
311# 280 "/home/runner/work/MFC/MFC/src/common/include/parallel_macros.fpp"
312
313! Emit code for non-Cray compilers
314# 287 "/home/runner/work/MFC/MFC/src/common/include/parallel_macros.fpp"
315
316! Emit code only for Cray compiler
317# 294 "/home/runner/work/MFC/MFC/src/common/include/parallel_macros.fpp"
318
319! Cray compiler directive, emitted from Fypp so the formatter cannot turn it into a comment
320# 301 "/home/runner/work/MFC/MFC/src/common/include/parallel_macros.fpp"
321
322! Emit code for non-NVIDIA compilers
323# 308 "/home/runner/work/MFC/MFC/src/common/include/parallel_macros.fpp"
324
325# 310 "/home/runner/work/MFC/MFC/src/common/include/parallel_macros.fpp"
326# 311 "/home/runner/work/MFC/MFC/src/common/include/parallel_macros.fpp"
327! New line at end of file is required for FYPP
328# 2 "/home/runner/work/MFC/MFC/src/common/include/macros.fpp" 2
329
330# 14 "/home/runner/work/MFC/MFC/src/common/include/macros.fpp"
331
332! Caution: This macro requires the use of a binding script to set CUDA_VISIBLE_DEVICES, such that we have one GPU device per MPI
333! rank. That's because for both cudaMemAdvise (preferred location) and cudaMemPrefetchAsync we use location = device_id = 0. For an
334! example see misc/nvidia_uvm/bind.sh.
335# 52 "/home/runner/work/MFC/MFC/src/common/include/macros.fpp"
336
337! Allocate and create GPU device memory
338# 72 "/home/runner/work/MFC/MFC/src/common/include/macros.fpp"
339
340! Free GPU device memory and deallocate
341# 80 "/home/runner/work/MFC/MFC/src/common/include/macros.fpp"
342
343! Cray-specific GPU pointer setup for vector fields
344# 104 "/home/runner/work/MFC/MFC/src/common/include/macros.fpp"
345
346! Cray-specific GPU pointer setup for scalar fields
347# 120 "/home/runner/work/MFC/MFC/src/common/include/macros.fpp"
348
349! Cray-specific GPU pointer teardown for scalar fields - exact inverse of ACC_SETUP_SFs. Removes the
350! present-table entries ACC_SETUP_SFs added (the scalar_field descriptor and its %sf copyin). REQUIRED when
351! an ACC_SETUP'd field is freed MID-RUN (e.g. AMR slot free on regrid/restart): Cray 'exit data delete'
352! decrements the reference counter, so the lone @:DEALLOCATE(arg%sf) only undoes the @:ALLOCATE create ref
353! and leaves the descriptor + the ACC_SETUP %sf copyin dangling. Call BEFORE @:DEALLOCATE(arg%sf).
354# 140 "/home/runner/work/MFC/MFC/src/common/include/macros.fpp"
355
356! Cray-specific GPU pointer setup for acoustic source spatials
357# 165 "/home/runner/work/MFC/MFC/src/common/include/macros.fpp"
358
359! Whether stress index idx is a shear component. Spelled out because NVHPC 26.9 at -O2 vectorizes any(idx == shear_indices)
360! inside a loop to always true, which doubled the normal-stress elastic energy.
361# 171 "/home/runner/work/MFC/MFC/src/common/include/macros.fpp"
362
363# 177 "/home/runner/work/MFC/MFC/src/common/include/macros.fpp"
364
365# 184 "/home/runner/work/MFC/MFC/src/common/include/macros.fpp"
366! New line at end of file is required for FYPP
367# 6 "/home/runner/work/MFC/MFC/src/simulation/m_amr_wave.fpp" 2
368
369!> @brief MPI point-to-point waves for AMR exchange: post all receives, pack, post all sends, wait once. Pooled waves list transfers
370!! on a send/receive side and s_amr_wave_close lays them out contiguously per peer; both ranks enumerate the same transfers in the
371!! same order, so no metadata is exchanged. Pools are caller-owned (device-resident under rdma_mpi).
373
374#ifdef MFC_MPI
375 use mpi
376#endif
379 use m_mpi_proxy, only: s_mpi_abort ! @:ASSERT expands to it
382
383 implicit none
384
385 private
389
390 !> One side (send or receive) of a pooled wave.
392 integer :: nx = 0 !< transfers
393 integer :: np = 0 !< peers
394 integer :: words = 0 !< pool words in use after close
395 !> per transfer (off: pool offset)
396 integer, allocatable :: blk(:), bl(:,:), bh(:,:), cnt(:), peer(:), pi(:), off(:)
397 integer, allocatable :: prank(:), pwords(:), pbase(:), pnx(:) !< per peer: rank, message words, pool base, transfers
398 end type t_amr_wave_side
399
400 !> Fill and restrict waves share amr_wsend/amr_wrecv; the seam wave stays open across the fills, so it has its own pair.
402
403 !> An open wave: tag band and requests; reqw is -1 for sends, the planned word count for receives.
405 integer :: band = 0, nreq = 0
406 integer, allocatable :: req(:), reqw(:)
407 end type t_amr_wave
408 type(t_amr_wave), public, target :: amr_wave, amr_wave_seam
409 !> Rank-indexed scratch for s_amr_wave_close; all-zero between calls.
410 integer, allocatable :: bmap(:), bnx(:), bwords(:)
411
412contains
413
414 !> Open a wave on tag band b.
415 impure subroutine s_amr_wave_open(w, b)
416
417 type(t_amr_wave), intent(inout) :: w
418 integer, intent(in) :: b
419
420 w%band = b; w%nreq = 0
421 call s_amr_tag_wave_open(b)
422
423 end subroutine s_amr_wave_open
424
425 pure integer function f_amr_wave_nreq(w) result(n)
426
427 type(t_amr_wave), intent(in) :: w
428
429 n = w%nreq
430
431 end function f_amr_wave_nreq
432
433 !> Post one request of n words between buf (device-resident when dev) and peer: dir = 1 sends, 2 receives.
434 impure subroutine s_amr_wave_req(w, dir, buf, n, peer, dev)
435
436 type(t_amr_wave), intent(inout) :: w
437 integer, intent(in) :: dir, n, peer
438 real(wp), intent(inout), contiguous :: buf(:)
439 logical, intent(in) :: dev
440 integer :: ierr, tq
441
442 tq = f_amr_wave_tag(w, dir, peer)
443 w%reqw(w%nreq) = merge(n, -1, dir == 2)
444#ifdef MFC_MPI
445 if (dev) then
446
447# 84 "/home/runner/work/MFC/MFC/src/simulation/m_amr_wave.fpp"
448#if defined(MFC_OpenACC)
449# 84 "/home/runner/work/MFC/MFC/src/simulation/m_amr_wave.fpp"
450!$acc host_data use_device(buf)
451# 84 "/home/runner/work/MFC/MFC/src/simulation/m_amr_wave.fpp"
452 if (dir == 2) then
453# 84 "/home/runner/work/MFC/MFC/src/simulation/m_amr_wave.fpp"
454 call mpi_irecv(buf, n, mpi_p, peer, tq, mpi_comm_world, w%req(w%nreq), ierr)
455# 84 "/home/runner/work/MFC/MFC/src/simulation/m_amr_wave.fpp"
456 else
457# 84 "/home/runner/work/MFC/MFC/src/simulation/m_amr_wave.fpp"
458 call mpi_isend(buf, n, mpi_p, peer, tq, mpi_comm_world, w%req(w%nreq), ierr)
459# 84 "/home/runner/work/MFC/MFC/src/simulation/m_amr_wave.fpp"
460 end if
461# 84 "/home/runner/work/MFC/MFC/src/simulation/m_amr_wave.fpp"
462!$acc end host_data
463# 84 "/home/runner/work/MFC/MFC/src/simulation/m_amr_wave.fpp"
464#elif defined(MFC_OpenMP)
465# 84 "/home/runner/work/MFC/MFC/src/simulation/m_amr_wave.fpp"
466!$omp target data use_device_addr(buf)
467# 84 "/home/runner/work/MFC/MFC/src/simulation/m_amr_wave.fpp"
468 if (dir == 2) then
469# 84 "/home/runner/work/MFC/MFC/src/simulation/m_amr_wave.fpp"
470 call mpi_irecv(buf, n, mpi_p, peer, tq, mpi_comm_world, w%req(w%nreq), ierr)
471# 84 "/home/runner/work/MFC/MFC/src/simulation/m_amr_wave.fpp"
472 else
473# 84 "/home/runner/work/MFC/MFC/src/simulation/m_amr_wave.fpp"
474 call mpi_isend(buf, n, mpi_p, peer, tq, mpi_comm_world, w%req(w%nreq), ierr)
475# 84 "/home/runner/work/MFC/MFC/src/simulation/m_amr_wave.fpp"
476 end if
477# 84 "/home/runner/work/MFC/MFC/src/simulation/m_amr_wave.fpp"
478!$omp end target data
479# 84 "/home/runner/work/MFC/MFC/src/simulation/m_amr_wave.fpp"
480#else
481# 84 "/home/runner/work/MFC/MFC/src/simulation/m_amr_wave.fpp"
482 if (dir == 2) then
483# 84 "/home/runner/work/MFC/MFC/src/simulation/m_amr_wave.fpp"
484 call mpi_irecv(buf, n, mpi_p, peer, tq, mpi_comm_world, w%req(w%nreq), ierr)
485# 84 "/home/runner/work/MFC/MFC/src/simulation/m_amr_wave.fpp"
486 else
487# 84 "/home/runner/work/MFC/MFC/src/simulation/m_amr_wave.fpp"
488 call mpi_isend(buf, n, mpi_p, peer, tq, mpi_comm_world, w%req(w%nreq), ierr)
489# 84 "/home/runner/work/MFC/MFC/src/simulation/m_amr_wave.fpp"
490 end if
491# 84 "/home/runner/work/MFC/MFC/src/simulation/m_amr_wave.fpp"
492#endif
493# 91 "/home/runner/work/MFC/MFC/src/simulation/m_amr_wave.fpp"
494 else if (dir == 2) then
495 call mpi_irecv(buf, n, mpi_p, peer, tq, mpi_comm_world, w%req(w%nreq), ierr)
496 else
497 call mpi_isend(buf, n, mpi_p, peer, tq, mpi_comm_world, w%req(w%nreq), ierr)
498 end if
499#endif
500
501 end subroutine s_amr_wave_req
502
503 !> Host-array variant of s_amr_wave_req (assumed-size, any-rank sections, no device address).
504 impure subroutine s_amr_wave_req_raw(w, dir, buf, n, peer)
505
506 type(t_amr_wave), intent(inout) :: w
507 integer, intent(in) :: dir, n, peer
508 real(wp), intent(inout) :: buf(*)
509 integer :: ierr, tq
510
511 tq = f_amr_wave_tag(w, dir, peer)
512 w%reqw(w%nreq) = merge(n, -1, dir == 2)
513#ifdef MFC_MPI
514 if (dir == 2) then
515 call mpi_irecv(buf, n, mpi_p, peer, tq, mpi_comm_world, w%req(w%nreq), ierr)
516 else
517 call mpi_isend(buf, n, mpi_p, peer, tq, mpi_comm_world, w%req(w%nreq), ierr)
518 end if
519#endif
520
521 end subroutine s_amr_wave_req_raw
522
523 !> Open a keyed-tag wave: clear the touched per-peer seq counters and bump the band's generation. Call at wave entry on every
524 !! rank.
525 impure subroutine s_amr_tag_wave_open(band)
526
527 integer, intent(in) :: band
528 integer :: i, d
529
530 if (.not. allocated(amr_tsq)) then
531 allocate (amr_tsq(0:num_procs - 1,2), amr_tsq_tch(num_procs, 2))
532 amr_tsq = 0; amr_n_tsq = 0
533 end if
534 do d = 1, 2
535 do i = 1, amr_n_tsq(d)
536 amr_tsq(amr_tsq_tch(i, d), d) = 0
537 end do
538 amr_n_tsq(d) = 0
539 end do
540 amr_tag_gen(band) = mod(amr_tag_gen(band) + 1, 16)
541
542 end subroutine s_amr_tag_wave_open
543
544 !> Next seq on the (peer, dir) channel of the open wave. dir: 1 = send, 2 = recv (separate counters).
545 impure integer function f_amr_wave_seq(peer, idir) result(sq)
546
547 integer, intent(in) :: peer, idir
548
549 if (amr_tsq(peer, idir) == 0) then
550 amr_n_tsq(idir) = amr_n_tsq(idir) + 1
551 amr_tsq_tch(amr_n_tsq(idir), idir) = peer
552 end if
553 amr_tsq(peer, idir) = amr_tsq(peer, idir) + 1
554 sq = amr_tsq(peer, idir)
555 if (.not. (sq < 4096)) then
556# 152 "/home/runner/work/MFC/MFC/src/simulation/m_amr_wave.fpp"
557 call s_mpi_abort("m_amr_wave.fpp:152: " // "Assertion failed: sq < 4096. " &
558# 152 "/home/runner/work/MFC/MFC/src/simulation/m_amr_wave.fpp"
559 & // "keyed-tag seq overflows its 12-bit field")
560# 152 "/home/runner/work/MFC/MFC/src/simulation/m_amr_wave.fpp"
561 end if
562# 153 "/home/runner/work/MFC/MFC/src/simulation/m_amr_wave.fpp"
563
564 end function f_amr_wave_seq
565
566 pure integer function f_amr_wave_tag_of(band, sq) result(t)
567
568 integer, intent(in) :: band, sq
569
570 t = amr_tag_base + band*65536 + amr_tag_gen(band)*4096 + sq
571
572 end function f_amr_wave_tag_of
573
574 !> Keyed tag of the next request to peer; grows the request list (w%nreq becomes the new request's index).
575 impure integer function f_amr_wave_tag(w, dir, peer) result(tq)
576
577 type(t_amr_wave), intent(inout) :: w
578 integer, intent(in) :: dir, peer
579 integer :: sq
580
581 sq = f_amr_wave_seq(peer, dir); tq = f_amr_wave_tag_of(w%band, sq)
582 w%nreq = w%nreq + 1
583 call s_amr_size_int(w%req, w%nreq); call s_amr_size_int(w%reqw, w%nreq)
584
585 end function f_amr_wave_tag
586
587 !> Wait for all requests; under MFC_DEBUG each receive length must equal its plan.
588 impure subroutine s_amr_wave_wait(w)
589
590 type(t_amr_wave), intent(inout) :: w
591 integer :: ierr
592
593 if (w%nreq == 0) return
594#ifdef MFC_MPI
595#ifdef MFC_DEBUG
596 block
597 integer :: st(mpi_status_size, w%nreq), gotw, q
598 call mpi_waitall(w%nreq, w%req, st, ierr)
599 do q = 1, w%nreq
600 if (w%reqw(q) < 0) cycle
601 call mpi_get_count(st(:,q), mpi_p, gotw, ierr)
602 if (.not. (gotw == w%reqw(q))) then
603# 192 "/home/runner/work/MFC/MFC/src/simulation/m_amr_wave.fpp"
604 call s_mpi_abort("m_amr_wave.fpp:192: " // "Assertion failed: gotw == w%reqw(q). " &
605# 192 "/home/runner/work/MFC/MFC/src/simulation/m_amr_wave.fpp"
606 & // "amr wave: a received message length differs from the plan")
607# 192 "/home/runner/work/MFC/MFC/src/simulation/m_amr_wave.fpp"
608 end if
609# 193 "/home/runner/work/MFC/MFC/src/simulation/m_amr_wave.fpp"
610 end do
611 end block
612#else
613 call mpi_waitall(w%nreq, w%req, mpi_statuses_ignore, ierr)
614#endif
615#endif
616 w%nreq = 0
617
618 end subroutine s_amr_wave_wait
619
620 !> Start enumerating a side's transfers.
621 impure subroutine s_amr_wave_reset(s)
622
623 type(t_amr_wave_side), intent(inout) :: s
624
625 s%nx = 0; s%np = 0; s%words = 0
626
627 end subroutine s_amr_wave_reset
628
629 !> Append one transfer of `words` payload words for block blk over box [bl, bh] with peer.
630 impure subroutine s_amr_wave_add(s, peer, blk, bl, bh, words)
631
632 type(t_amr_wave_side), intent(inout) :: s
633 integer, intent(in) :: peer, blk, bl(3), bh(3), words
634
635 s%nx = s%nx + 1
636 call s_amr_size_int(s%blk, s%nx); call s_amr_size_int3(s%bl, s%nx); call s_amr_size_int3(s%bh, s%nx)
637 call s_amr_size_int(s%cnt, s%nx); call s_amr_size_int(s%peer, s%nx); call s_amr_size_int(s%pi, s%nx)
638 call s_amr_size_int(s%off, s%nx)
639 s%blk(s%nx) = blk; s%bl(:,s%nx) = bl; s%bh(:,s%nx) = bh; s%cnt(s%nx) = words; s%peer(s%nx) = peer
640
641 end subroutine s_amr_wave_add
642
643 !> Add the msl slabs [tb:te] of block blk as transfers with peer.
644 impure subroutine s_amr_wave_add_slabs(s, peer, blk, msl, tb, te)
645
646 type(t_amr_wave_side), intent(inout) :: s
647 integer, intent(in) :: peer, blk, msl, tb(3, 6), te(3, 6)
648 integer :: isl
649
650 do isl = 1, msl
651 call s_amr_wave_add(s, peer, blk, tb(:,isl), te(:,isl), sys_size*product(te(:,isl) - tb(:,isl) + 1))
652 end do
653
654 end subroutine s_amr_wave_add_slabs
655
656 !> Lay the side out in its pool (peers in first-appearance order, transfers contiguous per peer); grows pool to the words in
657 !! use.
658 impure subroutine s_amr_wave_close(s, pool, dev)
659
660 type(t_amr_wave_side), intent(inout) :: s
661 real(wp), allocatable, intent(inout) :: pool(:)
662 logical, intent(in) :: dev
663 integer :: t, r, ip, base
664
665 if (.not. allocated(bmap)) then
666 allocate (bmap(0:num_procs - 1), bnx(0:num_procs - 1), bwords(0:num_procs - 1))
667 bmap = 0; bnx = 0; bwords = 0
668 end if
669 s%np = 0
670 do t = 1, s%nx
671 r = s%peer(t)
672 if (bmap(r) == 0) then
673 s%np = s%np + 1
674 call s_amr_size_int(s%prank, s%np); call s_amr_size_int(s%pwords, s%np)
675 call s_amr_size_int(s%pbase, s%np); call s_amr_size_int(s%pnx, s%np)
676 bmap(r) = s%np; s%prank(s%np) = r
677 end if
678 s%pi(t) = bmap(r)
679 s%off(t) = bwords(r) ! relative to the peer's base
680 bwords(r) = bwords(r) + s%cnt(t)
681 bnx(r) = bnx(r) + 1
682 end do
683 base = 0
684 do ip = 1, s%np
685 r = s%prank(ip)
686 s%pnx(ip) = bnx(r)
687 s%pwords(ip) = bwords(r)
688 s%pbase(ip) = base; base = base + s%pwords(ip)
689 bmap(r) = 0; bnx(r) = 0; bwords(r) = 0
690 end do
691 do t = 1, s%nx
692 s%off(t) = s%pbase(s%pi(t)) + s%off(t)
693 end do
694 s%words = base
695 call s_amr_size_real(pool, base, dev)
696
697 end subroutine s_amr_wave_close
698
699 !> Pool bounds [lo, hi] of transfer t's payload.
700 pure subroutine s_amr_wave_slice(s, t, lo, hi)
701
702 type(t_amr_wave_side), intent(in) :: s
703 integer, intent(in) :: t
704 integer, intent(out) :: lo, hi
705
706 lo = s%off(t) + 1; hi = s%off(t) + s%cnt(t)
707
708 end subroutine s_amr_wave_slice
709
710 !> One message per peer (dir = 1 sends, 2 receives) over the side's pool runs.
711 impure subroutine s_amr_wave_peers(w, dir, s, pool, dev)
712
713 type(t_amr_wave), intent(inout) :: w
714 integer, intent(in) :: dir
715 type(t_amr_wave_side), intent(in) :: s
716 real(wp), intent(inout), contiguous :: pool(:)
717 logical, intent(in) :: dev
718 integer :: ip
719
720 do ip = 1, s%np
721 call s_amr_wave_req(w, dir, pool(s%pbase(ip) + 1:s%pbase(ip) + s%pwords(ip)), s%pwords(ip), s%prank(ip), dev)
722 end do
723
724 end subroutine s_amr_wave_peers
725
726 !> One receive per peer into the side's pool runs.
727 impure subroutine s_amr_wave_post(w, s, pool, dev)
728
729 type(t_amr_wave), intent(inout) :: w
730 type(t_amr_wave_side), intent(in) :: s
731 real(wp), intent(inout), contiguous :: pool(:)
732 logical, intent(in) :: dev
733
734 call s_amr_wave_peers(w, 2, s, pool, dev)
735
736 end subroutine s_amr_wave_post
737
738 !> One send per peer (the caller has packed every slice).
739 impure subroutine s_amr_wave_send(w, s, pool, dev)
740
741 type(t_amr_wave), intent(inout) :: w
742 type(t_amr_wave_side), intent(in) :: s
743 real(wp), intent(inout), contiguous :: pool(:)
744 logical, intent(in) :: dev
745
746 call s_amr_wave_peers(w, 1, s, pool, dev)
747
748 end subroutine s_amr_wave_send
749
750end module m_amr_wave
Module-scope AMR state (block pool, flat field store, wave buffers) and the leaf helpers every AMR mo...
integer, dimension(0:amr_n_bands - 1) amr_tag_gen
integer, dimension(:,:), allocatable amr_tsq_tch
touched peers per dir, so the reset is O(active peers), not O(P)
impure subroutine s_amr_size_real(a, n, dev)
Wire pools, preserving on grow. dev keeps the pool device-resident across (re)allocation: the old ima...
impure subroutine s_amr_size_int3(a, n)
impure subroutine s_amr_size_int(a, n)
Grow-on-demand sizing for append-only tables and pools: a grow preserves the entries already appended...
integer, dimension(:,:), allocatable amr_tsq
(0:np-1, dir) in-wave per-peer seq counters; touched-reset
integer, dimension(2) amr_n_tsq
MPI point-to-point waves for AMR exchange: post all receives, pack, post all sends,...
impure integer function f_amr_wave_seq(peer, idir)
Next seq on the (peer, dir) channel of the open wave. dir: 1 = send, 2 = recv (separate counters).
impure subroutine, public s_amr_wave_send(w, s, pool, dev)
One send per peer (the caller has packed every slice).
type(t_amr_wave), target, public amr_wave_seam
impure integer function f_amr_wave_tag(w, dir, peer)
Keyed tag of the next request to peer; grows the request list (wnreq becomes the new request's index)...
impure subroutine, public s_amr_wave_req_raw(w, dir, buf, n, peer)
Host-array variant of s_amr_wave_req (assumed-size, any-rank sections, no device address).
impure subroutine, public s_amr_wave_post(w, s, pool, dev)
One receive per peer into the side's pool runs.
integer, dimension(:), allocatable bwords
impure subroutine s_amr_tag_wave_open(band)
Open a keyed-tag wave: clear the touched per-peer seq counters and bump the band's generation....
type(t_amr_wave_side), public amr_wsend
Fill and restrict waves share amr_wsend/amr_wrecv; the seam wave stays open across the fills,...
impure subroutine, public s_amr_wave_add_slabs(s, peer, blk, msl, tb, te)
Add the msl slabs [tb:te] of block blk as transfers with peer.
impure subroutine, public s_amr_wave_add(s, peer, blk, bl, bh, words)
Append one transfer of words payload words for block blk over box [bl, bh] with peer.
impure subroutine, public s_amr_wave_close(s, pool, dev)
Lay the side out in its pool (peers in first-appearance order, transfers contiguous per peer); grows ...
type(t_amr_wave_side), public amr_wseam_snd
impure subroutine, public s_amr_wave_wait(w)
Wait for all requests; under MFC_DEBUG each receive length must equal its plan.
integer, dimension(:), allocatable bnx
impure subroutine, public s_amr_wave_req(w, dir, buf, n, peer, dev)
Post one request of n words between buf (device-resident when dev) and peer: dir = 1 sends,...
type(t_amr_wave_side), public amr_wseam_rcv
impure subroutine, public s_amr_wave_open(w, b)
Open a wave on tag band b.
pure integer function, public f_amr_wave_nreq(w)
type(t_amr_wave), target, public amr_wave
integer, dimension(:), allocatable bmap
Rank-indexed scratch for s_amr_wave_close; all-zero between calls.
impure subroutine s_amr_wave_peers(w, dir, s, pool, dev)
One message per peer (dir = 1 sends, 2 receives) over the side's pool runs.
pure subroutine, public s_amr_wave_slice(s, t, lo, hi)
Pool bounds [lo, hi] of transfer t's payload.
impure subroutine, public s_amr_wave_reset(s)
Start enumerating a side's transfers.
type(t_amr_wave_side), public amr_wrecv
pure integer function f_amr_wave_tag_of(band, sq)
Global parameters for the computational domain, fluid properties, and simulation algorithm configurat...
integer num_procs
Number of processors.
MPI halo exchange, domain decomposition, and buffer packing/unpacking for the simulation solver.
Working-precision kind selection (half/single/double) and corresponding MPI datatype parameters.
integer, parameter mpi_p
Default value when MPI is not used.
integer, parameter wp
Change to single_precision if needed.
One side (send or receive) of a pooled wave.
An open wave: tag band and requests; reqw is -1 for sends, the planned word count for receives.