MFC
Exascale flow solver
Loading...
Searching...
No Matches
m_amr_seam.fpp.f90
Go to the documentation of this file.
1# 1 "/home/runner/work/MFC/MFC/src/simulation/m_amr_seam.fpp"
2!>
3!!@file
4!!@brief Contains module m_amr_seam
5
6# 7 "/home/runner/work/MFC/MFC/src/simulation/m_amr_seam.fpp"
7# 8 "/home/runner/work/MFC/MFC/src/simulation/m_amr_seam.fpp"
8# 1 "/home/runner/work/MFC/MFC/src/common/include/macros.fpp" 1
9# 1 "/home/runner/work/MFC/MFC/src/common/include/parallel_macros.fpp" 1
10# 1 "/home/runner/work/MFC/MFC/src/common/include/shared_parallel_macros.fpp" 1
11# 2 "/home/runner/work/MFC/MFC/src/common/include/shared_parallel_macros.fpp"
12# 3 "/home/runner/work/MFC/MFC/src/common/include/shared_parallel_macros.fpp"
13# 4 "/home/runner/work/MFC/MFC/src/common/include/shared_parallel_macros.fpp"
14# 5 "/home/runner/work/MFC/MFC/src/common/include/shared_parallel_macros.fpp"
15# 6 "/home/runner/work/MFC/MFC/src/common/include/shared_parallel_macros.fpp"
16
17# 8 "/home/runner/work/MFC/MFC/src/common/include/shared_parallel_macros.fpp"
18# 9 "/home/runner/work/MFC/MFC/src/common/include/shared_parallel_macros.fpp"
19# 10 "/home/runner/work/MFC/MFC/src/common/include/shared_parallel_macros.fpp"
20
21# 12 "/home/runner/work/MFC/MFC/src/common/include/shared_parallel_macros.fpp"
22# 1 "/home/runner/work/MFC/MFC/src/common/include/defaults/thermochem.fpp" 1
23! Default for Fypp runs outside the toolchain (e.g. documentation). Searched last, so the
24! build/include/<target>/thermochem.fpp the toolchain writes from m_thermochem takes precedence.
25# 4 "/home/runner/work/MFC/MFC/src/common/include/defaults/thermochem.fpp"
26# 5 "/home/runner/work/MFC/MFC/src/common/include/defaults/thermochem.fpp"
27# 13 "/home/runner/work/MFC/MFC/src/common/include/shared_parallel_macros.fpp" 2
28
29# 16 "/home/runner/work/MFC/MFC/src/common/include/shared_parallel_macros.fpp"
30# 17 "/home/runner/work/MFC/MFC/src/common/include/shared_parallel_macros.fpp"
31
32# 24 "/home/runner/work/MFC/MFC/src/common/include/shared_parallel_macros.fpp"
33
34# 53 "/home/runner/work/MFC/MFC/src/common/include/shared_parallel_macros.fpp"
35
36# 65 "/home/runner/work/MFC/MFC/src/common/include/shared_parallel_macros.fpp"
37
38# 75 "/home/runner/work/MFC/MFC/src/common/include/shared_parallel_macros.fpp"
39
40# 105 "/home/runner/work/MFC/MFC/src/common/include/shared_parallel_macros.fpp"
41
42# 117 "/home/runner/work/MFC/MFC/src/common/include/shared_parallel_macros.fpp"
43
44# 127 "/home/runner/work/MFC/MFC/src/common/include/shared_parallel_macros.fpp"
45
46# 174 "/home/runner/work/MFC/MFC/src/common/include/shared_parallel_macros.fpp"
47! New line at end of file is required for FYPP
48# 2 "/home/runner/work/MFC/MFC/src/common/include/parallel_macros.fpp" 2
49# 1 "/home/runner/work/MFC/MFC/src/common/include/omp_macros.fpp" 1
50# 1 "/home/runner/work/MFC/MFC/src/common/include/shared_parallel_macros.fpp" 1
51# 2 "/home/runner/work/MFC/MFC/src/common/include/shared_parallel_macros.fpp"
52# 3 "/home/runner/work/MFC/MFC/src/common/include/shared_parallel_macros.fpp"
53# 4 "/home/runner/work/MFC/MFC/src/common/include/shared_parallel_macros.fpp"
54# 5 "/home/runner/work/MFC/MFC/src/common/include/shared_parallel_macros.fpp"
55# 6 "/home/runner/work/MFC/MFC/src/common/include/shared_parallel_macros.fpp"
56
57# 8 "/home/runner/work/MFC/MFC/src/common/include/shared_parallel_macros.fpp"
58# 9 "/home/runner/work/MFC/MFC/src/common/include/shared_parallel_macros.fpp"
59# 10 "/home/runner/work/MFC/MFC/src/common/include/shared_parallel_macros.fpp"
60
61# 12 "/home/runner/work/MFC/MFC/src/common/include/shared_parallel_macros.fpp"
62# 1 "/home/runner/work/MFC/MFC/src/common/include/defaults/thermochem.fpp" 1
63! Default for Fypp runs outside the toolchain (e.g. documentation). Searched last, so the
64! build/include/<target>/thermochem.fpp the toolchain writes from m_thermochem takes precedence.
65# 4 "/home/runner/work/MFC/MFC/src/common/include/defaults/thermochem.fpp"
66# 5 "/home/runner/work/MFC/MFC/src/common/include/defaults/thermochem.fpp"
67# 13 "/home/runner/work/MFC/MFC/src/common/include/shared_parallel_macros.fpp" 2
68
69# 16 "/home/runner/work/MFC/MFC/src/common/include/shared_parallel_macros.fpp"
70# 17 "/home/runner/work/MFC/MFC/src/common/include/shared_parallel_macros.fpp"
71
72# 24 "/home/runner/work/MFC/MFC/src/common/include/shared_parallel_macros.fpp"
73
74# 53 "/home/runner/work/MFC/MFC/src/common/include/shared_parallel_macros.fpp"
75
76# 65 "/home/runner/work/MFC/MFC/src/common/include/shared_parallel_macros.fpp"
77
78# 75 "/home/runner/work/MFC/MFC/src/common/include/shared_parallel_macros.fpp"
79
80# 105 "/home/runner/work/MFC/MFC/src/common/include/shared_parallel_macros.fpp"
81
82# 117 "/home/runner/work/MFC/MFC/src/common/include/shared_parallel_macros.fpp"
83
84# 127 "/home/runner/work/MFC/MFC/src/common/include/shared_parallel_macros.fpp"
85
86# 174 "/home/runner/work/MFC/MFC/src/common/include/shared_parallel_macros.fpp"
87! New line at end of file is required for FYPP
88# 2 "/home/runner/work/MFC/MFC/src/common/include/omp_macros.fpp" 2
89
90# 4 "/home/runner/work/MFC/MFC/src/common/include/omp_macros.fpp"
91# 5 "/home/runner/work/MFC/MFC/src/common/include/omp_macros.fpp"
92# 6 "/home/runner/work/MFC/MFC/src/common/include/omp_macros.fpp"
93# 7 "/home/runner/work/MFC/MFC/src/common/include/omp_macros.fpp"
94# 8 "/home/runner/work/MFC/MFC/src/common/include/omp_macros.fpp"
95
96# 20 "/home/runner/work/MFC/MFC/src/common/include/omp_macros.fpp"
97
98# 49 "/home/runner/work/MFC/MFC/src/common/include/omp_macros.fpp"
99
100# 54 "/home/runner/work/MFC/MFC/src/common/include/omp_macros.fpp"
101
102# 59 "/home/runner/work/MFC/MFC/src/common/include/omp_macros.fpp"
103
104# 64 "/home/runner/work/MFC/MFC/src/common/include/omp_macros.fpp"
105
106# 69 "/home/runner/work/MFC/MFC/src/common/include/omp_macros.fpp"
107
108# 74 "/home/runner/work/MFC/MFC/src/common/include/omp_macros.fpp"
109
110# 87 "/home/runner/work/MFC/MFC/src/common/include/omp_macros.fpp"
111
112# 101 "/home/runner/work/MFC/MFC/src/common/include/omp_macros.fpp"
113
114# 106 "/home/runner/work/MFC/MFC/src/common/include/omp_macros.fpp"
115
116# 111 "/home/runner/work/MFC/MFC/src/common/include/omp_macros.fpp"
117
118# 116 "/home/runner/work/MFC/MFC/src/common/include/omp_macros.fpp"
119
120# 121 "/home/runner/work/MFC/MFC/src/common/include/omp_macros.fpp"
121
122# 126 "/home/runner/work/MFC/MFC/src/common/include/omp_macros.fpp"
123
124# 131 "/home/runner/work/MFC/MFC/src/common/include/omp_macros.fpp"
125
126# 136 "/home/runner/work/MFC/MFC/src/common/include/omp_macros.fpp"
127
128# 141 "/home/runner/work/MFC/MFC/src/common/include/omp_macros.fpp"
129
130# 171 "/home/runner/work/MFC/MFC/src/common/include/omp_macros.fpp"
131
132# 231 "/home/runner/work/MFC/MFC/src/common/include/omp_macros.fpp"
133
134# 245 "/home/runner/work/MFC/MFC/src/common/include/omp_macros.fpp"
135
136# 270 "/home/runner/work/MFC/MFC/src/common/include/omp_macros.fpp"
137
138# 281 "/home/runner/work/MFC/MFC/src/common/include/omp_macros.fpp"
139
140# 283 "/home/runner/work/MFC/MFC/src/common/include/omp_macros.fpp"
141# 294 "/home/runner/work/MFC/MFC/src/common/include/omp_macros.fpp"
142
143# 343 "/home/runner/work/MFC/MFC/src/common/include/omp_macros.fpp"
144
145# 353 "/home/runner/work/MFC/MFC/src/common/include/omp_macros.fpp"
146
147# 363 "/home/runner/work/MFC/MFC/src/common/include/omp_macros.fpp"
148
149# 372 "/home/runner/work/MFC/MFC/src/common/include/omp_macros.fpp"
150
151# 389 "/home/runner/work/MFC/MFC/src/common/include/omp_macros.fpp"
152
153# 399 "/home/runner/work/MFC/MFC/src/common/include/omp_macros.fpp"
154
155# 406 "/home/runner/work/MFC/MFC/src/common/include/omp_macros.fpp"
156
157# 412 "/home/runner/work/MFC/MFC/src/common/include/omp_macros.fpp"
158
159# 418 "/home/runner/work/MFC/MFC/src/common/include/omp_macros.fpp"
160
161# 424 "/home/runner/work/MFC/MFC/src/common/include/omp_macros.fpp"
162
163# 430 "/home/runner/work/MFC/MFC/src/common/include/omp_macros.fpp"
164
165# 436 "/home/runner/work/MFC/MFC/src/common/include/omp_macros.fpp"
166! New line at end of file is required for FYPP
167# 3 "/home/runner/work/MFC/MFC/src/common/include/parallel_macros.fpp" 2
168# 1 "/home/runner/work/MFC/MFC/src/common/include/acc_macros.fpp" 1
169# 1 "/home/runner/work/MFC/MFC/src/common/include/shared_parallel_macros.fpp" 1
170# 2 "/home/runner/work/MFC/MFC/src/common/include/shared_parallel_macros.fpp"
171# 3 "/home/runner/work/MFC/MFC/src/common/include/shared_parallel_macros.fpp"
172# 4 "/home/runner/work/MFC/MFC/src/common/include/shared_parallel_macros.fpp"
173# 5 "/home/runner/work/MFC/MFC/src/common/include/shared_parallel_macros.fpp"
174# 6 "/home/runner/work/MFC/MFC/src/common/include/shared_parallel_macros.fpp"
175
176# 8 "/home/runner/work/MFC/MFC/src/common/include/shared_parallel_macros.fpp"
177# 9 "/home/runner/work/MFC/MFC/src/common/include/shared_parallel_macros.fpp"
178# 10 "/home/runner/work/MFC/MFC/src/common/include/shared_parallel_macros.fpp"
179
180# 12 "/home/runner/work/MFC/MFC/src/common/include/shared_parallel_macros.fpp"
181# 1 "/home/runner/work/MFC/MFC/src/common/include/defaults/thermochem.fpp" 1
182! Default for Fypp runs outside the toolchain (e.g. documentation). Searched last, so the
183! build/include/<target>/thermochem.fpp the toolchain writes from m_thermochem takes precedence.
184# 4 "/home/runner/work/MFC/MFC/src/common/include/defaults/thermochem.fpp"
185# 5 "/home/runner/work/MFC/MFC/src/common/include/defaults/thermochem.fpp"
186# 13 "/home/runner/work/MFC/MFC/src/common/include/shared_parallel_macros.fpp" 2
187
188# 16 "/home/runner/work/MFC/MFC/src/common/include/shared_parallel_macros.fpp"
189# 17 "/home/runner/work/MFC/MFC/src/common/include/shared_parallel_macros.fpp"
190
191# 24 "/home/runner/work/MFC/MFC/src/common/include/shared_parallel_macros.fpp"
192
193# 53 "/home/runner/work/MFC/MFC/src/common/include/shared_parallel_macros.fpp"
194
195# 65 "/home/runner/work/MFC/MFC/src/common/include/shared_parallel_macros.fpp"
196
197# 75 "/home/runner/work/MFC/MFC/src/common/include/shared_parallel_macros.fpp"
198
199# 105 "/home/runner/work/MFC/MFC/src/common/include/shared_parallel_macros.fpp"
200
201# 117 "/home/runner/work/MFC/MFC/src/common/include/shared_parallel_macros.fpp"
202
203# 127 "/home/runner/work/MFC/MFC/src/common/include/shared_parallel_macros.fpp"
204
205# 174 "/home/runner/work/MFC/MFC/src/common/include/shared_parallel_macros.fpp"
206! New line at end of file is required for FYPP
207# 2 "/home/runner/work/MFC/MFC/src/common/include/acc_macros.fpp" 2
208
209# 7 "/home/runner/work/MFC/MFC/src/common/include/acc_macros.fpp"
210
211# 17 "/home/runner/work/MFC/MFC/src/common/include/acc_macros.fpp"
212
213# 22 "/home/runner/work/MFC/MFC/src/common/include/acc_macros.fpp"
214
215# 27 "/home/runner/work/MFC/MFC/src/common/include/acc_macros.fpp"
216
217# 32 "/home/runner/work/MFC/MFC/src/common/include/acc_macros.fpp"
218
219# 37 "/home/runner/work/MFC/MFC/src/common/include/acc_macros.fpp"
220
221# 42 "/home/runner/work/MFC/MFC/src/common/include/acc_macros.fpp"
222
223# 47 "/home/runner/work/MFC/MFC/src/common/include/acc_macros.fpp"
224
225# 52 "/home/runner/work/MFC/MFC/src/common/include/acc_macros.fpp"
226
227# 57 "/home/runner/work/MFC/MFC/src/common/include/acc_macros.fpp"
228
229# 62 "/home/runner/work/MFC/MFC/src/common/include/acc_macros.fpp"
230
231# 73 "/home/runner/work/MFC/MFC/src/common/include/acc_macros.fpp"
232
233# 78 "/home/runner/work/MFC/MFC/src/common/include/acc_macros.fpp"
234
235# 83 "/home/runner/work/MFC/MFC/src/common/include/acc_macros.fpp"
236
237# 88 "/home/runner/work/MFC/MFC/src/common/include/acc_macros.fpp"
238
239# 103 "/home/runner/work/MFC/MFC/src/common/include/acc_macros.fpp"
240
241# 131 "/home/runner/work/MFC/MFC/src/common/include/acc_macros.fpp"
242
243# 161 "/home/runner/work/MFC/MFC/src/common/include/acc_macros.fpp"
244
245# 176 "/home/runner/work/MFC/MFC/src/common/include/acc_macros.fpp"
246
247# 194 "/home/runner/work/MFC/MFC/src/common/include/acc_macros.fpp"
248
249# 216 "/home/runner/work/MFC/MFC/src/common/include/acc_macros.fpp"
250
251# 245 "/home/runner/work/MFC/MFC/src/common/include/acc_macros.fpp"
252
253# 260 "/home/runner/work/MFC/MFC/src/common/include/acc_macros.fpp"
254
255# 270 "/home/runner/work/MFC/MFC/src/common/include/acc_macros.fpp"
256
257# 279 "/home/runner/work/MFC/MFC/src/common/include/acc_macros.fpp"
258
259# 295 "/home/runner/work/MFC/MFC/src/common/include/acc_macros.fpp"
260
261# 305 "/home/runner/work/MFC/MFC/src/common/include/acc_macros.fpp"
262
263# 312 "/home/runner/work/MFC/MFC/src/common/include/acc_macros.fpp"
264! New line at end of file is required for FYPP
265# 4 "/home/runner/work/MFC/MFC/src/common/include/parallel_macros.fpp" 2
266
267! GPU parallel region (scalar reductions, maxval/minval)
268# 23 "/home/runner/work/MFC/MFC/src/common/include/parallel_macros.fpp"
269
270! GPU parallel loop over threads (most common GPU macro)
271# 44 "/home/runner/work/MFC/MFC/src/common/include/parallel_macros.fpp"
272
273! Required closing for GPU_PARALLEL_LOOP
274# 56 "/home/runner/work/MFC/MFC/src/common/include/parallel_macros.fpp"
275
276! Mark routine for device compilation
277# 113 "/home/runner/work/MFC/MFC/src/common/include/parallel_macros.fpp"
278
279! Declare device-resident data
280# 131 "/home/runner/work/MFC/MFC/src/common/include/parallel_macros.fpp"
281
282! Inner loop within a GPU parallel region
283# 146 "/home/runner/work/MFC/MFC/src/common/include/parallel_macros.fpp"
284
285! Scoped GPU data region
286# 165 "/home/runner/work/MFC/MFC/src/common/include/parallel_macros.fpp"
287
288! Host code with device pointers (for MPI with GPU buffers)
289# 191 "/home/runner/work/MFC/MFC/src/common/include/parallel_macros.fpp"
290
291! Allocate device memory (unscoped)
292# 205 "/home/runner/work/MFC/MFC/src/common/include/parallel_macros.fpp"
293
294! Free device memory
295# 217 "/home/runner/work/MFC/MFC/src/common/include/parallel_macros.fpp"
296
297! Atomic operation on device
298# 229 "/home/runner/work/MFC/MFC/src/common/include/parallel_macros.fpp"
299
300! End atomic capture block
301# 240 "/home/runner/work/MFC/MFC/src/common/include/parallel_macros.fpp"
302
303! Copy data between host and device
304# 252 "/home/runner/work/MFC/MFC/src/common/include/parallel_macros.fpp"
305
306! Synchronization barrier
307# 264 "/home/runner/work/MFC/MFC/src/common/include/parallel_macros.fpp"
308
309! Import GPU library module (openacc or omp_lib)
310# 273 "/home/runner/work/MFC/MFC/src/common/include/parallel_macros.fpp"
311
312! Emit code only for AMD compiler
313# 280 "/home/runner/work/MFC/MFC/src/common/include/parallel_macros.fpp"
314
315! Emit code for non-Cray compilers
316# 287 "/home/runner/work/MFC/MFC/src/common/include/parallel_macros.fpp"
317
318! Emit code only for Cray compiler
319# 294 "/home/runner/work/MFC/MFC/src/common/include/parallel_macros.fpp"
320
321! Cray compiler directive, emitted from Fypp so the formatter cannot turn it into a comment
322# 301 "/home/runner/work/MFC/MFC/src/common/include/parallel_macros.fpp"
323
324! Emit code for non-NVIDIA compilers
325# 308 "/home/runner/work/MFC/MFC/src/common/include/parallel_macros.fpp"
326
327# 310 "/home/runner/work/MFC/MFC/src/common/include/parallel_macros.fpp"
328# 311 "/home/runner/work/MFC/MFC/src/common/include/parallel_macros.fpp"
329! New line at end of file is required for FYPP
330# 2 "/home/runner/work/MFC/MFC/src/common/include/macros.fpp" 2
331
332# 14 "/home/runner/work/MFC/MFC/src/common/include/macros.fpp"
333
334! Caution: This macro requires the use of a binding script to set CUDA_VISIBLE_DEVICES, such that we have one GPU device per MPI
335! rank. That's because for both cudaMemAdvise (preferred location) and cudaMemPrefetchAsync we use location = device_id = 0. For an
336! example see misc/nvidia_uvm/bind.sh.
337# 52 "/home/runner/work/MFC/MFC/src/common/include/macros.fpp"
338
339! Allocate and create GPU device memory
340# 72 "/home/runner/work/MFC/MFC/src/common/include/macros.fpp"
341
342! Free GPU device memory and deallocate
343# 80 "/home/runner/work/MFC/MFC/src/common/include/macros.fpp"
344
345! Cray-specific GPU pointer setup for vector fields
346# 104 "/home/runner/work/MFC/MFC/src/common/include/macros.fpp"
347
348! Cray-specific GPU pointer setup for scalar fields
349# 120 "/home/runner/work/MFC/MFC/src/common/include/macros.fpp"
350
351! Cray-specific GPU pointer teardown for scalar fields - exact inverse of ACC_SETUP_SFs. Removes the
352! present-table entries ACC_SETUP_SFs added (the scalar_field descriptor and its %sf copyin). REQUIRED when
353! an ACC_SETUP'd field is freed MID-RUN (e.g. AMR slot free on regrid/restart): Cray 'exit data delete'
354! decrements the reference counter, so the lone @:DEALLOCATE(arg%sf) only undoes the @:ALLOCATE create ref
355! and leaves the descriptor + the ACC_SETUP %sf copyin dangling. Call BEFORE @:DEALLOCATE(arg%sf).
356# 140 "/home/runner/work/MFC/MFC/src/common/include/macros.fpp"
357
358! Cray-specific GPU pointer setup for acoustic source spatials
359# 165 "/home/runner/work/MFC/MFC/src/common/include/macros.fpp"
360
361! Whether stress index idx is a shear component. Spelled out because NVHPC 26.9 at -O2 vectorizes any(idx == shear_indices)
362! inside a loop to always true, which doubled the normal-stress elastic energy.
363# 171 "/home/runner/work/MFC/MFC/src/common/include/macros.fpp"
364
365# 177 "/home/runner/work/MFC/MFC/src/common/include/macros.fpp"
366
367# 184 "/home/runner/work/MFC/MFC/src/common/include/macros.fpp"
368! New line at end of file is required for FYPP
369# 9 "/home/runner/work/MFC/MFC/src/simulation/m_amr_seam.fpp" 2
370
371!> @brief Same-level fine-fine seams and the coarse conservative halo.
373
375 use m_box, only: f_morton
377 use m_mpi_proxy, only: s_mpi_abort ! @:ASSERT expands to it
379 use m_amr_state
380 use m_amr_wave
382
383 implicit none
384
385 private
388
389contains
390
391 !> Exchange coarse conservative ghost layers at internal rank boundaries (beg then end per direction); the solver fills only
392 !! prim ghosts. Collective.
393 impure subroutine s_amr_exchange_coarse_cons_halo(q_cons)
394
395 type(scalar_field), dimension(sys_size), intent(inout) :: q_cons
396
397# 37 "/home/runner/work/MFC/MFC/src/simulation/m_amr_seam.fpp"
398 if (amr_dim(1)) then
399# 38 "/home/runner/work/MFC/MFC/src/simulation/m_amr_seam.fpp"
400 if (bc_x%beg >= 0) call s_mpi_sendrecv_variables_buffers(q_cons, 1, -1, sys_size)
401# 39 "/home/runner/work/MFC/MFC/src/simulation/m_amr_seam.fpp"
402 if (bc_x%end >= 0) call s_mpi_sendrecv_variables_buffers(q_cons, 1, 1, sys_size)
403# 40 "/home/runner/work/MFC/MFC/src/simulation/m_amr_seam.fpp"
404 end if
405# 37 "/home/runner/work/MFC/MFC/src/simulation/m_amr_seam.fpp"
406 if (amr_dim(2)) then
407# 38 "/home/runner/work/MFC/MFC/src/simulation/m_amr_seam.fpp"
408 if (bc_y%beg >= 0) call s_mpi_sendrecv_variables_buffers(q_cons, 2, -1, sys_size)
409# 39 "/home/runner/work/MFC/MFC/src/simulation/m_amr_seam.fpp"
410 if (bc_y%end >= 0) call s_mpi_sendrecv_variables_buffers(q_cons, 2, 1, sys_size)
411# 40 "/home/runner/work/MFC/MFC/src/simulation/m_amr_seam.fpp"
412 end if
413# 37 "/home/runner/work/MFC/MFC/src/simulation/m_amr_seam.fpp"
414 if (amr_dim(3)) then
415# 38 "/home/runner/work/MFC/MFC/src/simulation/m_amr_seam.fpp"
416 if (bc_z%beg >= 0) call s_mpi_sendrecv_variables_buffers(q_cons, 3, -1, sys_size)
417# 39 "/home/runner/work/MFC/MFC/src/simulation/m_amr_seam.fpp"
418 if (bc_z%end >= 0) call s_mpi_sendrecv_variables_buffers(q_cons, 3, 1, sys_size)
419# 40 "/home/runner/work/MFC/MFC/src/simulation/m_amr_seam.fpp"
420 end if
421# 42 "/home/runner/work/MFC/MFC/src/simulation/m_amr_seam.fpp"
422
424
425 !> True iff yb sits directly above xb on xb's high face in dim d with identical transverse coarse extents (one ordered pair per
426 !! seam).
427 pure logical function f_amr_seam(xb, yb, d) result(seam)
428
429 integer, intent(in) :: xb, yb, d
430 integer :: t
431
432 seam = amr_region_lo_all(d, yb) == amr_region_hi_all(d, xb) + 1
433 do t = 1, 3
434 if (t /= d) seam = seam .and. amr_region_lo_all(t, xb) == amr_region_lo_all(t, yb) .and. amr_region_hi_all(t, &
435 & xb) == amr_region_hi_all(t, yb)
436 end do
437
438 end function f_amr_seam
439
440 !> Same-rank seam exchange, both directions in one kernel: xb high interior -> yb low ghost and yb low interior -> xb high
441 !! ghost. The four slabs are disjoint, so fusing is safe. Threads span the max pair extent and are masked.
442 impure subroutine s_amr_fine_seam_exchange(npair, plx, ply, pd, pxhi, pfm, ndep)
443
444 integer, intent(in) :: npair
445 integer, intent(in) :: plx(:), ply(:), pd(:), pxhi(:), pfm(:,:) !< per-pair: slots, seam dim, high extent, fine extents
446 integer, intent(in) :: ndep
447 integer :: i, a, b, t, na, nb, pr, g, gmax, lx, ly, d, xhi, cnt
448 integer :: ix1, ix2, ix3, iy1, iy2, iy3
449
450 ! max thread extent over the pairs on this rank (transverse product x seam depth)
451
452 gmax = 0
453 do pr = 1, npair
454 select case (pd(pr))
455 case (1); na = pfm(2, pr) + 1; nb = pfm(3, pr) + 1
456 case (2); na = pfm(1, pr) + 1; nb = pfm(3, pr) + 1
457 case default; na = pfm(1, pr) + 1; nb = pfm(2, pr) + 1
458 end select
459 gmax = max(gmax, na*nb*ndep)
460 end do
461
462
463# 82 "/home/runner/work/MFC/MFC/src/simulation/m_amr_seam.fpp"
464
465# 82 "/home/runner/work/MFC/MFC/src/simulation/m_amr_seam.fpp"
466#if defined(MFC_OpenACC)
467# 82 "/home/runner/work/MFC/MFC/src/simulation/m_amr_seam.fpp"
468!$acc parallel loop collapse(3) gang vector default(present) private(lx, ly, d, xhi, na, nb, cnt, a, b, t, ix1, ix2, ix3, iy1, iy2, iy3) copyin(plx, ply, pd, pxhi, pfm)
469# 82 "/home/runner/work/MFC/MFC/src/simulation/m_amr_seam.fpp"
470#elif defined(MFC_OpenMP)
471# 82 "/home/runner/work/MFC/MFC/src/simulation/m_amr_seam.fpp"
472
473# 82 "/home/runner/work/MFC/MFC/src/simulation/m_amr_seam.fpp"
474
475# 82 "/home/runner/work/MFC/MFC/src/simulation/m_amr_seam.fpp"
476
477# 82 "/home/runner/work/MFC/MFC/src/simulation/m_amr_seam.fpp"
478!$omp target teams loop defaultmap(firstprivate:scalar) bind(teams,parallel) collapse(3) defaultmap(tofrom:aggregate) defaultmap(tofrom:allocatable) defaultmap(tofrom:pointer) &
479# 82 "/home/runner/work/MFC/MFC/src/simulation/m_amr_seam.fpp"
480!$omp& private(lx, ly, d, xhi, na, nb, cnt, a, b, t, ix1, ix2, ix3, iy1, iy2, iy3) map(to:plx, ply, pd, pxhi, pfm)
481# 82 "/home/runner/work/MFC/MFC/src/simulation/m_amr_seam.fpp"
482#endif
483# 84 "/home/runner/work/MFC/MFC/src/simulation/m_amr_seam.fpp"
484 do pr = 1, npair
485 do i = 1, sys_size
486 do g = 0, gmax - 1
487 d = pd(pr)
488 if (d == 1) then
489 na = pfm(2, pr) + 1; nb = pfm(3, pr) + 1
490 else if (d == 2) then
491 na = pfm(1, pr) + 1; nb = pfm(3, pr) + 1
492 else
493 na = pfm(1, pr) + 1; nb = pfm(2, pr) + 1
494 end if
495 cnt = na*nb*ndep
496 if (g < cnt) then
497 lx = plx(pr); ly = ply(pr); xhi = pxhi(pr)
498 a = mod(g, na); b = mod(g/na, nb); t = g/(na*nb)
499 ! (a, b) transverse, t depth into the seam, placed per the pair's seam dim
500 if (d == 1) then
501 ix1 = xhi - ndep + 1 + t; ix2 = a; ix3 = b
502 iy1 = -ndep + t; iy2 = a; iy3 = b
503 else if (d == 2) then
504 ix1 = a; ix2 = xhi - ndep + 1 + t; ix3 = b
505 iy1 = a; iy2 = -ndep + t; iy3 = b
506 else
507 ix1 = a; ix2 = b; ix3 = xhi - ndep + 1 + t
508 iy1 = a; iy2 = b; iy3 = -ndep + t
509 end if
510 amr_cons_st(iy1, iy2, iy3, i, ly) = amr_cons_st(ix1, ix2, ix3, i, lx)
511 if (d == 1) then
512 ix1 = xhi + 1 + t; iy1 = t
513 else if (d == 2) then
514 ix2 = xhi + 1 + t; iy2 = t
515 else
516 ix3 = xhi + 1 + t; iy3 = t
517 end if
518 amr_cons_st(ix1, ix2, ix3, i, lx) = amr_cons_st(iy1, iy2, iy3, i, ly)
519 end if
520 end do
521 end do
522 end do
523
524# 123 "/home/runner/work/MFC/MFC/src/simulation/m_amr_seam.fpp"
525#if defined(MFC_OpenACC)
526# 123 "/home/runner/work/MFC/MFC/src/simulation/m_amr_seam.fpp"
527!$acc end parallel loop
528# 123 "/home/runner/work/MFC/MFC/src/simulation/m_amr_seam.fpp"
529#elif defined(MFC_OpenMP)
530# 123 "/home/runner/work/MFC/MFC/src/simulation/m_amr_seam.fpp"
531
532# 123 "/home/runner/work/MFC/MFC/src/simulation/m_amr_seam.fpp"
533!$omp end target teams loop
534# 123 "/home/runner/work/MFC/MFC/src/simulation/m_amr_seam.fpp"
535#endif
536# 124 "/home/runner/work/MFC/MFC/src/simulation/m_amr_seam.fpp"
537
538 end subroutine s_amr_fine_seam_exchange
539
540 !> Seam dim of the ordered pair (xb, yb): the d in which yb is xb's same-level high-face neighbour, else 0 (adjacent in at most
541 !! one dim).
542 pure integer function f_amr_seam_dim(xb, yb) result(d)
543
544 integer, intent(in) :: xb, yb
545
546 d = 0
547 if (xb == yb) return
548 if (amr_block_level(xb) /= amr_block_level(yb)) return
549 if (f_amr_seam(xb, yb, 1)) d = 1
550 if (n_glb > 0) then; if (f_amr_seam(xb, yb, 2)) d = 2; end if
551 if (p_glb > 0) then; if (f_amr_seam(xb, yb, 3)) d = 3; end if
552
553 end function f_amr_seam_dim
554
555 !> Binary-search the Morton-sorted lo corners for the block at clo, verify the seam predicate against xb and record it in (mb,
556 !! md, nm); a block already recorded is raised to the higher d.
557 pure subroutine s_amr_seam_probe(xb, d, clo, mkey, ord, nblk, mb, md, nm)
558
559 integer, intent(in) :: xb, d, clo(3), nblk
560 integer(kind=8), intent(in) :: mkey(:)
561 integer, intent(in) :: ord(:)
562 integer, intent(inout) :: mb(3), md(3), nm
563 integer :: yb, t, i, lo_s, hi_s, mid_s
564 integer(kind=8) :: ck
565 logical :: ok
566
567 ck = f_morton(clo(1), clo(2), clo(3))
568 lo_s = 1; hi_s = nblk
569 do while (lo_s <= hi_s)
570 mid_s = (lo_s + hi_s)/2
571 if (mkey(ord(mid_s)) < ck) then
572 lo_s = mid_s + 1
573 else
574 hi_s = mid_s - 1
575 end if
576 end do
577 ! scan the equal-key run: f_morton keeps 21 bits/dim, so the lo check rejects aliased corners
578 do i = lo_s, nblk
579 if (mkey(ord(i)) /= ck) exit
580 yb = ord(i)
581 if (yb == xb) cycle
582 if (amr_block_level(yb) /= amr_block_level(xb)) cycle
583 ok = .true.
584 do t = 1, 3
585 if (amr_region_lo_all(t, yb) /= clo(t)) ok = .false.
586 if (t /= d) then
587 if (amr_region_hi_all(t, yb) /= amr_region_hi_all(t, xb)) ok = .false.
588 end if
589 end do
590 if (.not. ok) cycle
591 do t = 1, nm
592 if (mb(t) == yb) then
593 md(t) = max(md(t), d)
594 return
595 end if
596 end do
597 nm = nm + 1
598 mb(nm) = yb; md(nm) = d
599 return
600 end do
601
602 end subroutine s_amr_seam_probe
603
604 !> Rebuild the same-level seam-pair list once per regrid/restart in identical (xb, yb) order on all ranks, and the per-block
605 !! overlap-rank lists (amr_ovl_gather/scatter).
606 impure subroutine s_amr_build_seam_pairs()
607
608 integer :: xb, d, np, k, mx, pass, nm, im, jm, tb, td, nb
609 integer :: plo(3), phi(3), rlo(3), rhi(3)
610 integer :: mb(3), md(3), clo(3)
611 integer(kind=8), allocatable :: mkey(:)
612 integer, allocatable :: ord(:)
613
614 if (allocated(amr_seam_pairs)) deallocate (amr_seam_pairs)
615
616 ! Blocks are disjoint and the seam fixes the neighbour's lo corner, so each (block, dim) needs one Morton binary
617 ! search. Emission order (xb, then yb ascending) must match on all ranks or sends and recvs mismatch and deadlock.
618 nb = max(amr_num_blocks, 1)
619 allocate (mkey(nb), ord(nb))
620 do k = 1, amr_num_blocks
622 end do
623
624 call s_amr_sort_by_key(mkey, amr_num_blocks, ord)
625
626 ! pass 1 counts, pass 2 fills: keeps amr_seam_pairs exactly sized
627 do pass = 1, 2
628 np = 0
629 do xb = 1, amr_num_blocks
630 nm = 0
631 do d = 1, num_dims
632 clo = amr_region_lo_all(:,xb)
633 clo(d) = amr_region_hi_all(d, xb) + 1
634 call s_amr_seam_probe(xb, d, clo, mkey, ord, amr_num_blocks, mb, md, nm)
635 end do
636 ! ascending yb within xb (nm <= 3)
637 do im = 2, nm
638 tb = mb(im); td = md(im); jm = im - 1
639 do while (jm >= 1)
640 if (mb(jm) <= tb) exit
641 mb(jm + 1) = mb(jm); md(jm + 1) = md(jm); jm = jm - 1
642 end do
643 mb(jm + 1) = tb; md(jm + 1) = td
644 end do
645 do im = 1, nm
646 np = np + 1
647 if (pass == 2) then
648 amr_seam_pairs(1, np) = xb; amr_seam_pairs(2, np) = mb(im); amr_seam_pairs(3, np) = md(im)
649 end if
650 end do
651 end do
652 if (pass == 1) then
654 allocate (amr_seam_pairs(3, max(np, 1)))
655 end if
656 end do
657 deallocate (mkey, ord)
658 ! per-block overlap-rank lists, rank-ascending (same MPI order as a 0..num_procs-1 scan); the first dim is the max
659 ! overlap, reallocated each build. Consumers run behind a build_seam_pairs guard.
660 mx = 1
661 do k = 1, amr_num_blocks
662 call s_amr_patch_box(k, plo, phi); call s_amr_region_box(k, rlo, rhi)
663 mx = max(mx, f_amr_overlap_count(plo, phi), f_amr_overlap_count(rlo, rhi))
664 end do
665 if (allocated(amr_ovl_gather)) deallocate (amr_ovl_gather)
666 if (allocated(amr_ovl_scatter)) deallocate (amr_ovl_scatter)
667 allocate (amr_ovl_gather(mx, amr_max_blocks), amr_ovl_scatter(mx, amr_max_blocks))
668 do k = 1, amr_num_blocks
669 call s_amr_patch_box(k, plo, phi); call s_amr_region_box(k, rlo, rhi)
672 end do
674 amr_seam_pairs_dirty = .false.
675
676 end subroutine s_amr_build_seam_pairs
677
678 !> Fine-fine halo between tiled sub-blocks: overwrite seam ghosts with the neighbour's stage-entry interior (cross-rank via the
679 !! seam wave, same-rank by local copy). Wire is wp, cast to stp on unpack.
680 impure subroutine s_amr_fine_fine_post()
681
682 integer :: xb, yb, d, rx, ry, cnt, xm(3), tsz, fmul, idx, lo, hi
683 integer :: r, sblk, sdlo, sdhi, ublk, udlo, udhi, eblk, edlo, edhi
684
685 amr_sw_nsame = 0
686 if (.not. amr) return
687 if (amr_num_blocks < 2) return
688
690 if (allocated(amr_sw_plx)) deallocate (amr_sw_plx, amr_sw_ply, amr_sw_pd, amr_sw_pxhi, amr_sw_pfm)
693 ! Seam wave: one aggregated message per peer and direction; both owners walk the same replicated pair list, so the
694 ! layout agrees. The wave stays open across the fill waves (own handle and sides).
696 ! transfer records: blk = the slot the data belongs to, bl = (d, dlo, dhi) its slab, bh = (peer slot, its dlo, dhi)
698 do idx = 1, amr_num_seam_pairs
699 xb = amr_seam_pairs(1, idx); yb = amr_seam_pairs(2, idx); d = amr_seam_pairs(3, idx)
700 rx = amr_block_owner(xb); ry = amr_block_owner(yb)
701 if (proc_rank /= rx .and. proc_rank /= ry) cycle
702 fmul = amr_ref_ratio**amr_block_level(xb)
703 xm(1) = fmul*(amr_region_hi_all(1, xb) - amr_region_lo_all(1, xb) + 1) - 1
704 xm(2) = merge(fmul*(amr_region_hi_all(2, xb) - amr_region_lo_all(2, xb) + 1) - 1, 0, n_glb > 0)
705 xm(3) = merge(fmul*(amr_region_hi_all(3, xb) - amr_region_lo_all(3, xb) + 1) - 1, 0, p_glb > 0)
706 tsz = 1
707 if (d /= 1) tsz = tsz*(xm(1) + 1)
708 if (d /= 2 .and. n_glb > 0) tsz = tsz*(xm(2) + 1)
709 if (d /= 3 .and. p_glb > 0) tsz = tsz*(xm(3) + 1)
710 cnt = sys_size*buff_size*tsz
711 if (rx == ry) then ! same rank owns both: deferred to the batched kernel at drain
715 cycle
716 end if
717 if (proc_rank == rx) then
718 r = ry; sblk = xb; sdlo = xm(d) - buff_size + 1; sdhi = xm(d)
719 ublk = xb; udlo = xm(d) + 1; udhi = xm(d) + buff_size
720 eblk = yb; edlo = 0; edhi = buff_size - 1
721 else
722 r = rx; sblk = yb; sdlo = 0; sdhi = buff_size - 1
723 ublk = yb; udlo = -buff_size; udhi = -1
724 eblk = xb; edlo = xm(d) - buff_size + 1; edhi = xm(d)
725 end if
726 call s_amr_wave_add(amr_wseam_snd, r, sblk, [d, sdlo, sdhi], [cnt, 0, 0], cnt)
727 call s_amr_wave_add(amr_wseam_rcv, r, ublk, [d, udlo, udhi], [eblk, edlo, edhi], cnt)
728 end do
732 do idx = 1, amr_wseam_snd%nx
733 call s_amr_wave_slice(amr_wseam_snd, idx, lo, hi)
734 call s_amr_fine_slice(amr_wseam_snd%blk(idx), amr_wseam_snd%bl(1, idx), amr_wseam_snd%bl(2, idx), amr_wseam_snd%bl(3, &
735 & idx), amr_sw_sq(lo:hi), 1)
736 end do
738
739 end subroutine s_amr_fine_fine_post
740
741 impure subroutine s_amr_fine_fine_drain()
742
743 integer :: idx, lo, hi
744
745 if (f_amr_wave_nreq(amr_wave_seam) == 0 .and. amr_sw_nsame == 0) return
747 do idx = 1, amr_wseam_rcv%nx
748 call s_amr_wave_slice(amr_wseam_rcv, idx, lo, hi)
749 call s_amr_fine_slice(amr_wseam_rcv%blk(idx), amr_wseam_rcv%bl(1, idx), amr_wseam_rcv%bl(2, idx), amr_wseam_rcv%bl(3, &
750 & idx), amr_sw_rq(lo:hi), -1)
751 end do
754 call s_amr_select_slot(1)
755
756 end subroutine s_amr_fine_fine_drain
757
758end module m_amr_seam
integer, intent(in) k
Rank decomposition, box arithmetic and block ownership (SFC cut, owner lookups, owned-block lists).
pure integer function, public f_amr_overlap_count(blo, bhi)
Overlap count only (allocation sizing), = product of the per-dim coord-range widths.
pure subroutine, public s_amr_sort_by_key(keys, n, ord)
ord(1:n) = item indices in ascending key order. Stable iterative bottom-up merge sort,...
pure subroutine, public s_amr_ranks_overlapping(blo, bhi, ranks, nr)
Ascending rank list overlapping coarse box [blo:bhi]: the coord brick is enumerated cx->cy->cz so r =...
Same-level fine-fine seams and the coarse conservative halo.
impure subroutine, public s_amr_fine_fine_drain()
pure subroutine s_amr_seam_probe(xb, d, clo, mkey, ord, nblk, mb, md, nm)
Binary-search the Morton-sorted lo corners for the block at clo, verify the seam predicate against xb...
impure subroutine s_amr_fine_seam_exchange(npair, plx, ply, pd, pxhi, pfm, ndep)
Same-rank seam exchange, both directions in one kernel: xb high interior -> yb low ghost and yb low i...
impure subroutine, public s_amr_exchange_coarse_cons_halo(q_cons)
Exchange coarse conservative ghost layers at internal rank boundaries (beg then end per direction); t...
pure logical function, public f_amr_seam(xb, yb, d)
True iff yb sits directly above xb on xb's high face in dim d with identical transverse coarse extent...
pure integer function, public f_amr_seam_dim(xb, yb)
Seam dim of the ordered pair (xb, yb): the d in which yb is xb's same-level high-face neighbour,...
impure subroutine, public s_amr_fine_fine_post()
Fine-fine halo between tiled sub-blocks: overwrite seam ghosts with the neighbour's stage-entry inter...
impure subroutine, public s_amr_build_seam_pairs()
Rebuild the same-level seam-pair list once per regrid/restart in identical (xb, yb) order on all rank...
Module-scope AMR state (block pool, flat field store, wave buffers) and the leaf helpers every AMR mo...
integer, dimension(:), allocatable amr_ovl_scatter_n
per-block list lengths
integer, dimension(:,:), allocatable amr_sw_pfm
real(wp), dimension(:), allocatable amr_sw_rq
integer amr_num_seam_pairs
pure subroutine s_amr_patch_box(k, plo, phi)
Block k's padded coarse patch box [plo, phi] (region +/- amr_cpat_mar; collapsed dims 0:0).
impure subroutine s_amr_fine_slice(slot, d, dlo, dhi, buf, dir)
Pack (dir=+1) / unpack (dir=-1) the fine cells of slot's q_cons over [dlo:dhi] in dim d (interior tra...
integer, dimension(:,:), allocatable amr_seam_pairs
cached same-level seam list (xb, yb, seam-dim) so s_amr_fine_fine_post iterates O(seams); rebuilt whe...
integer, dimension(:), allocatable amr_ovl_gather_n
integer, parameter amr_band_seam
integer amr_seam_pairs_nblk
integer, dimension(:), allocatable amr_loc_of
Dense local index for live slots: amr_slots is sparse over 1:amr_max_blocks, the field store is dense...
real(wp), dimension(:), allocatable amr_sw_sq
pure subroutine s_amr_region_box(k, rlo, rhi)
Block k's region box in global coarse cells; collapsed dimensions read 0:0.
integer, dimension(:), allocatable amr_sw_ply
integer, dimension(:), allocatable amr_sw_pd
integer, dimension(:), allocatable amr_sw_pxhi
integer amr_sw_nsame
seam pairs both of whose blocks this rank owns: exchanged by one batched kernel at drain,...
logical amr_seam_pairs_dirty
real(stp), dimension(:,:,:,:,:), allocatable amr_cons_st
Flat per-block field store (x, y, z, var, local slot) so one kernel can run over every live block; al...
integer, dimension(:,:), allocatable amr_ovl_scatter
integer, dimension(:), allocatable amr_sw_plx
integer, dimension(:,:), allocatable amr_ovl_gather
cached per-block P2P overlap-rank lists, rebuilt with the seam list: amr_ovl_gather(:,...
MPI point-to-point waves for AMR exchange: post all receives, pack, post all sends,...
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 subroutine, public s_amr_wave_post(w, s, pool, dev)
One receive per peer into the side's pool runs.
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.
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)
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.
Box type, per-rank partition arithmetic (one box per rank) and the Morton key.
pure integer(kind=8) function, public f_morton(ix, iy, iz)
3D Morton (Z-order) key of (ix, iy, iz), 21 bits/dim; negative coords clamp to 0.
Shared derived types for field data, patch geometry, bubble dynamics, and MPI I/O structures.
Global parameters for the computational domain, fluid properties, and simulation algorithm configurat...
integer, dimension(:,:), allocatable amr_region_lo_all
Per-slot mirrors (allocated 1:amr_max_blocks by m_amr): region box, rank intersection and ownership f...
integer proc_rank
Rank of the local processor.
integer, dimension(:,:), allocatable amr_region_hi_all
integer, dimension(:), allocatable amr_block_level
Refinement level of each active block (1..amr_max_level) and the deepest level populated....
integer buff_size
Number of ghost cells for boundary condition storage.
logical, dimension(3) amr_dim
The active dimensions, and this rank's coarse subdomain origin and extent padded to 3-vectors (0 in c...
integer, dimension(:), allocatable amr_block_owner
Fine-level distribution map: SFC/work-balanced single-owner rank per active block,...
integer amr_num_blocks
Number of active AMR fine-block slots (>= 1) and the working slot index selecting which one the per-b...
subroutine s_amr_select_slot(islot)
Make block slot islot the working slot: set amr_cur and copy its stored mirrors (region,...
MPI communication layer: domain decomposition, halo exchange, reductions, and parallel I/O setup.
subroutine s_mpi_sendrecv_variables_buffers(q_comm, mpi_dir, pbc_loc, nvar, pb_in, mv_in, q_t_sf)
The goal of this procedure is to populate the buffers of the cell-average conservative variables by c...
MPI halo exchange, domain decomposition, and buffer packing/unpacking for the simulation solver.
Derived type annexing a scalar field (SF).