MFC
Exascale flow solver
Loading...
Searching...
No Matches
m_amr_transfer.fpp.f90
Go to the documentation of this file.
1# 1 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
2!>
3!!@file
4!!@brief Contains module m_amr_transfer
5
6# 7 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
7# 8 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.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_transfer.fpp" 2
370
371!> @brief Prolongation, restriction and Berger-Colella reflux, including the restrict/reflux/freg waves.
373
374#ifdef MFC_MPI
375 use mpi
376#endif
377
380 use m_mpi_proxy, only: s_mpi_abort
385 use m_amr_state
386 use m_amr_wave
388 use m_amr_store
390 use m_amr_ghost, only: s_amr_slab_add
391 use m_amr_seam
392 use m_amr_frame
393
394 implicit none
395
396 private
399
400contains
401
402 !> True iff rank r refluxes some face of the current block (its unclipped subdomain); the participation map and both exchange
403 !! paths gate freg receives on it.
404 pure logical function f_amr_reflux_participates(r) result(part)
405
406 integer, intent(in) :: r
407 integer :: sidx(3), ext(3)
408 logical :: s_lo(3), s_hi(3)
409
410 call s_amr_rank_decomp(r, sidx, ext)
411 call s_amr_reflux_faces(sidx, ext, .false., s_lo, s_hi)
412 part = any(s_lo .or. s_hi)
413
414 end function f_amr_reflux_participates
415
416 !> Faces of the current block that rank r applies (seam-clipped); sender and receiver derive the identical set from replicated
417 !! data.
418 pure subroutine s_amr_reflux_faces_for(r, s_lo, s_hi)
419
420 integer, intent(in) :: r
421 logical, intent(out) :: s_lo(3), s_hi(3)
422 integer :: sidx(3), ext(3)
423
424 call s_amr_rank_decomp(r, sidx, ext)
425 call s_amr_reflux_faces(sidx, ext, .true., s_lo, s_hi)
426
427 end subroutine s_amr_reflux_faces_for
428
429 !> Per-stage level-1 reflux-face exchange as one wave: faces are packed on the owner's device into one pool, cross as one
430 !! message per peer and unpack on the device. Both ends list a peer's faces by ascending block, dim, low before high, from the
431 !! same predicates, so layouts agree with no handshake. Nothing may post into freg before s_amr_reg_prepare (the apply can
432 !! reallocate the registers).
433 impure subroutine s_amr_reflux_faces_wave()
434
435#ifdef MFC_MPI
436 integer :: k, r, idx, ncand, kk2, t
437 integer :: cand(num_procs), glo(3), ghi(3), msidx(3), mext(3)
438 logical :: s_lo(3), s_hi(3)
439
440 if (num_procs == 1) return
441 call s_amr_reg_prepare()
445 call s_amr_rank_decomp(proc_rank, msidx, mext)
446 do kk2 = 1, amr_n_l1p
447 k = amr_l1p_blk(kk2)
448 if (amr_block_owner(k) == proc_rank) cycle
449 if (.not. f_amr_block_near(k, msidx, mext)) cycle
450 call s_amr_select_slot(k)
451 if (.not. f_amr_reflux_participates(proc_rank)) cycle
452 call s_amr_reflux_faces_for(proc_rank, s_lo, s_hi)
453 call s_amr_face_xfers(amr_wrecv, amr_block_owner(k), k, s_lo, s_hi)
454 end do
456 do kk2 = 1, amr_n_my
457 k = amr_my_blk(kk2)
458 if (amr_block_level(k) /= 1 .or. amr_block_owner(k) /= proc_rank) cycle
459 call s_amr_select_slot(k)
460 glo = merge(amr_region_lo - 1, 0, amr_dim); ghi = merge(amr_region_hi + 1, 0, amr_dim)
461 call s_amr_ranks_overlapping(glo, ghi, cand, ncand)
462 do idx = 1, ncand
463 r = cand(idx)
464 if (r == proc_rank .or. .not. f_amr_reflux_participates(r)) cycle
465 call s_amr_reflux_faces_for(r, s_lo, s_hi)
466 call s_amr_face_xfers(amr_wsend, r, k, s_lo, s_hi)
467 end do
468 end do
472 if (amr_wsend%nx > 0) then
473 call s_amr_freg_pack(amr_wsend, .true.)
474
475# 113 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
476#if defined(MFC_OpenACC)
477# 113 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
478!$acc update host(amr_rf_sq(1:amr_wsend%words))
479# 113 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
480#elif defined(MFC_OpenMP)
481# 113 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
482!$omp target update from(amr_rf_sq(1:amr_wsend%words))
483# 113 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
484#endif
485# 114 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
486 end if
489 if (amr_wrecv%nx > 0) then
490
491# 118 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
492#if defined(MFC_OpenACC)
493# 118 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
494!$acc update device(amr_rf_rq(1:amr_wrecv%words))
495# 118 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
496#elif defined(MFC_OpenMP)
497# 118 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
498!$omp target update to(amr_rf_rq(1:amr_wrecv%words))
499# 118 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
500#endif
501# 119 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
502 call s_amr_freg_pack(amr_wrecv, .false.)
503 end if
504 call s_amr_select_slot(1)
505#endif
506
507 end subroutine s_amr_reflux_faces_wave
508
509 !> Append block k's faces f_lo/f_hi with peer as transfers (bl = dim, side 1 low / 2 high, register slot).
510 impure subroutine s_amr_face_xfers(w, peer, k, f_lo, f_hi)
511
512 type(t_amr_wave_side), intent(inout) :: w
513 integer, intent(in) :: peer, k
514 logical, intent(in) :: f_lo(3), f_hi(3)
515 integer :: d, nw
516
517 do d = 1, num_dims
518 nw = size(freg(d)%lo, 1)*size(freg(d)%lo, 2)*size(freg(d)%lo, 3)
519 if (f_lo(d)) call s_amr_wave_add(w, peer, k, [d, 1, amr_reg_of(k)], [0, 0, 0], nw)
520 if (f_hi(d)) call s_amr_wave_add(w, peer, k, [d, 2, amr_reg_of(k)], [0, 0, 0], nw)
521 end do
522
523 end subroutine s_amr_face_xfers
524
525 !> Copy a side's freg faces into (pack) or out of the device pool, one kernel per face direction. CCE 19 packs garbage when one
526 !! procedure holds several of these loop nests (as in s_amr_apply_reflux), so each direction has its own procedure.
527 impure subroutine s_amr_freg_pack(w, pack)
528
529 type(t_amr_wave_side), intent(in) :: w
530 logical, intent(in) :: pack
531
532# 150 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
533 if (1 <= num_dims) call s_amr_freg_pack_d1(w, pack)
534# 151 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
535# 150 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
536 if (2 <= num_dims) call s_amr_freg_pack_d2(w, pack)
537# 151 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
538# 150 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
539 if (3 <= num_dims) call s_amr_freg_pack_d3(w, pack)
540# 151 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
541# 152 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
542
543 end subroutine s_amr_freg_pack
544
545# 156 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
546 !> s_amr_freg_pack for the faces normal to direction 1, both sides in one kernel.
547# 157 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
548 impure subroutine s_amr_freg_pack_d1(w, pack)
549# 158 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
550
551 type(t_amr_wave_side), intent(in) :: w
552 logical, intent(in) :: pack
553 integer :: tsl(w%nx), toff(w%nx), tsd(w%nx)
554 integer :: t, nt, n1, n2, eq, a, b, j
555
556 nt = 0
557 do t = 1, w%nx
558 if (w%bl(1, t) /= 1) cycle
559# 167 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
560 nt = nt + 1; tsl(nt) = w%bl(3, t); toff(nt) = w%off(t); tsd(nt) = w%bl(2, t)
561 end do
562 if (nt == 0) return
563 n1 = size(freg(1)%lo, 2); n2 = size(freg(1)%lo, 3)
564# 171 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
565
566# 171 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
567
568# 171 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
569#if defined(MFC_OpenACC)
570# 171 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
571!$acc parallel loop collapse(4) gang vector default(present) private(j) copyin(tsl, toff, tsd)
572# 171 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
573#elif defined(MFC_OpenMP)
574# 171 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
575
576# 171 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
577
578# 171 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
579
580# 171 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
581!$omp target teams loop defaultmap(firstprivate:scalar) bind(teams,parallel) collapse(4) defaultmap(tofrom:aggregate) defaultmap(tofrom:allocatable) defaultmap(tofrom:pointer) private(j) &
582# 171 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
583!$omp& map(to:tsl, toff, tsd)
584# 171 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
585#endif
586# 172 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
587 do t = 1, nt
588 do b = 0, n2 - 1
589 do a = 0, n1 - 1
590 do eq = 1, sys_size
591 j = toff(t) + eq + sys_size*(a + n1*b)
592 if (pack) then
593 if (tsd(t) == 1) then
594 amr_rf_sq(j) = freg(1)%lo(eq, a, b, tsl(t))
595# 180 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
596 else
597 amr_rf_sq(j) = freg(1)%hi(eq, a, b, tsl(t))
598# 182 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
599 end if
600 else if (tsd(t) == 1) then
601 freg(1)%lo(eq, a, b, tsl(t)) = amr_rf_rq(j)
602# 185 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
603 else
604 freg(1)%hi(eq, a, b, tsl(t)) = amr_rf_rq(j)
605# 187 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
606 end if
607 end do
608 end do
609 end do
610 end do
611
612# 192 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
613#if defined(MFC_OpenACC)
614# 192 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
615!$acc end parallel loop
616# 192 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
617#elif defined(MFC_OpenMP)
618# 192 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
619
620# 192 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
621!$omp end target teams loop
622# 192 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
623#endif
624# 193 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
625
626 end subroutine s_amr_freg_pack_d1
627# 195 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
628# 156 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
629 !> s_amr_freg_pack for the faces normal to direction 2, both sides in one kernel.
630# 157 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
631 impure subroutine s_amr_freg_pack_d2(w, pack)
632# 158 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
633
634 type(t_amr_wave_side), intent(in) :: w
635 logical, intent(in) :: pack
636 integer :: tsl(w%nx), toff(w%nx), tsd(w%nx)
637 integer :: t, nt, n1, n2, eq, a, b, j
638
639 nt = 0
640 do t = 1, w%nx
641 if (w%bl(1, t) /= 2) cycle
642# 167 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
643 nt = nt + 1; tsl(nt) = w%bl(3, t); toff(nt) = w%off(t); tsd(nt) = w%bl(2, t)
644 end do
645 if (nt == 0) return
646 n1 = size(freg(2)%lo, 2); n2 = size(freg(2)%lo, 3)
647# 171 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
648
649# 171 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
650
651# 171 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
652#if defined(MFC_OpenACC)
653# 171 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
654!$acc parallel loop collapse(4) gang vector default(present) private(j) copyin(tsl, toff, tsd)
655# 171 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
656#elif defined(MFC_OpenMP)
657# 171 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
658
659# 171 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
660
661# 171 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
662
663# 171 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
664!$omp target teams loop defaultmap(firstprivate:scalar) bind(teams,parallel) collapse(4) defaultmap(tofrom:aggregate) defaultmap(tofrom:allocatable) defaultmap(tofrom:pointer) private(j) &
665# 171 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
666!$omp& map(to:tsl, toff, tsd)
667# 171 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
668#endif
669# 172 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
670 do t = 1, nt
671 do b = 0, n2 - 1
672 do a = 0, n1 - 1
673 do eq = 1, sys_size
674 j = toff(t) + eq + sys_size*(a + n1*b)
675 if (pack) then
676 if (tsd(t) == 1) then
677 amr_rf_sq(j) = freg(2)%lo(eq, a, b, tsl(t))
678# 180 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
679 else
680 amr_rf_sq(j) = freg(2)%hi(eq, a, b, tsl(t))
681# 182 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
682 end if
683 else if (tsd(t) == 1) then
684 freg(2)%lo(eq, a, b, tsl(t)) = amr_rf_rq(j)
685# 185 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
686 else
687 freg(2)%hi(eq, a, b, tsl(t)) = amr_rf_rq(j)
688# 187 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
689 end if
690 end do
691 end do
692 end do
693 end do
694
695# 192 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
696#if defined(MFC_OpenACC)
697# 192 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
698!$acc end parallel loop
699# 192 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
700#elif defined(MFC_OpenMP)
701# 192 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
702
703# 192 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
704!$omp end target teams loop
705# 192 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
706#endif
707# 193 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
708
709 end subroutine s_amr_freg_pack_d2
710# 195 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
711# 156 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
712 !> s_amr_freg_pack for the faces normal to direction 3, both sides in one kernel.
713# 157 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
714 impure subroutine s_amr_freg_pack_d3(w, pack)
715# 158 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
716
717 type(t_amr_wave_side), intent(in) :: w
718 logical, intent(in) :: pack
719 integer :: tsl(w%nx), toff(w%nx), tsd(w%nx)
720 integer :: t, nt, n1, n2, eq, a, b, j
721
722 nt = 0
723 do t = 1, w%nx
724 if (w%bl(1, t) /= 3) cycle
725# 167 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
726 nt = nt + 1; tsl(nt) = w%bl(3, t); toff(nt) = w%off(t); tsd(nt) = w%bl(2, t)
727 end do
728 if (nt == 0) return
729 n1 = size(freg(3)%lo, 2); n2 = size(freg(3)%lo, 3)
730# 171 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
731
732# 171 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
733
734# 171 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
735#if defined(MFC_OpenACC)
736# 171 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
737!$acc parallel loop collapse(4) gang vector default(present) private(j) copyin(tsl, toff, tsd)
738# 171 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
739#elif defined(MFC_OpenMP)
740# 171 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
741
742# 171 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
743
744# 171 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
745
746# 171 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
747!$omp target teams loop defaultmap(firstprivate:scalar) bind(teams,parallel) collapse(4) defaultmap(tofrom:aggregate) defaultmap(tofrom:allocatable) defaultmap(tofrom:pointer) private(j) &
748# 171 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
749!$omp& map(to:tsl, toff, tsd)
750# 171 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
751#endif
752# 172 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
753 do t = 1, nt
754 do b = 0, n2 - 1
755 do a = 0, n1 - 1
756 do eq = 1, sys_size
757 j = toff(t) + eq + sys_size*(a + n1*b)
758 if (pack) then
759 if (tsd(t) == 1) then
760 amr_rf_sq(j) = freg(3)%lo(eq, a, b, tsl(t))
761# 180 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
762 else
763 amr_rf_sq(j) = freg(3)%hi(eq, a, b, tsl(t))
764# 182 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
765 end if
766 else if (tsd(t) == 1) then
767 freg(3)%lo(eq, a, b, tsl(t)) = amr_rf_rq(j)
768# 185 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
769 else
770 freg(3)%hi(eq, a, b, tsl(t)) = amr_rf_rq(j)
771# 187 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
772 end if
773 end do
774 end do
775 end do
776 end do
777
778# 192 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
779#if defined(MFC_OpenACC)
780# 192 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
781!$acc end parallel loop
782# 192 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
783#elif defined(MFC_OpenMP)
784# 192 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
785
786# 192 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
787!$omp end target teams loop
788# 192 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
789#endif
790# 193 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
791
792 end subroutine s_amr_freg_pack_d3
793# 195 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
794# 196 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
795
796 !> Split-ownership level>=2 freg exchange as one wave, run once before the reflux fold; zero-copy like the faces wave, on tag
797 !! band 1 (faces use band 0).
798 impure subroutine s_amr_freg_wave()
799
800#ifdef MFC_MPI
801 integer :: k, pblk, cowner, powner, nhr, j, kk2
802 real(wp) :: w_lo(3), w_hi(3)
803
804 if (num_procs == 1) return
805 call s_amr_reg_prepare()
808 nhr = 0
809 ! receive side: freg faces of level>=2 children of my parents that another rank owns, straight into the child's
810 ! register slot; faces no sibling weight selects are poisoned
811 do kk2 = 1, amr_n_fch
812 k = amr_fch_blk(kk2)
813 call s_amr_select_slot(k)
814 pblk = amr_parent_blk(k)
815 cowner = amr_block_owner(k); powner = amr_block_owner(pblk)
816 if (cowner == powner .or. powner /= proc_rank) cycle
817 call s_amr_sibling_face_weights(k, pblk, w_lo, w_hi)
818 nhr = nhr + 1
819 call s_amr_size_int(amr_fw_rblk, nhr)
820 amr_fw_rblk(nhr) = k
821 call s_amr_freg_faces_poison(w_lo > 0._wp, w_hi > 0._wp)
822 call s_amr_freg_faces_req(2, cowner, w_lo > 0._wp, w_hi > 0._wp)
823 end do
824 ! send side: my level>=2 blocks whose parent lives elsewhere ship the faces the sibling weights select
826 do kk2 = 1, amr_n_my
827 k = amr_my_blk(kk2)
828 if (amr_block_level(k) < 2) cycle
829 call s_amr_select_slot(k)
830 pblk = amr_parent_blk(k)
831 cowner = amr_block_owner(k); powner = amr_block_owner(pblk)
832 if (cowner == powner .or. cowner /= proc_rank) cycle
833 call s_amr_sibling_face_weights(k, pblk, w_lo, w_hi)
834 call s_amr_freg_faces_update(.false., w_lo > 0._wp, w_hi > 0._wp)
835 call s_amr_freg_faces_req(1, powner, w_lo > 0._wp, w_hi > 0._wp)
836 end do
838 do j = 1, nhr
839 k = amr_fw_rblk(j)
840 call s_amr_select_slot(k)
841 pblk = amr_parent_blk(k)
842 call s_amr_sibling_face_weights(k, pblk, w_lo, w_hi)
843 call s_amr_freg_faces_update(.true., w_lo > 0._wp, w_hi > 0._wp)
844 end do
845#endif
846
847 end subroutine s_amr_freg_wave
848
849 !> Post one zero-copy request per flagged face register of the current slot (block k): dir 1 sends to / 2 receives from peer, in
850 !! (direction, face) order.
851 impure subroutine s_amr_freg_faces_req(dir, peer, f_lo, f_hi)
852
853 integer, intent(in) :: dir, peer
854 logical, intent(in) :: f_lo(3), f_hi(3)
855 integer :: cnt
856
857# 259 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
858 if (1 <= num_dims) then
859# 260 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
860 cnt = size(freg(1)%lo, 1)*size(freg(1)%lo, 2)*size(freg(1)%lo, 3)
861# 261 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
862 if (f_lo(1)) call s_amr_wave_req_raw(amr_wave, dir, freg(1)%lo(:,:,:,amr_reg_cur), cnt, peer)
863# 262 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
864 if (f_hi(1)) call s_amr_wave_req_raw(amr_wave, dir, freg(1)%hi(:,:,:,amr_reg_cur), cnt, peer)
865# 263 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
866 end if
867# 259 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
868 if (2 <= num_dims) then
869# 260 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
870 cnt = size(freg(2)%lo, 1)*size(freg(2)%lo, 2)*size(freg(2)%lo, 3)
871# 261 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
872 if (f_lo(2)) call s_amr_wave_req_raw(amr_wave, dir, freg(2)%lo(:,:,:,amr_reg_cur), cnt, peer)
873# 262 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
874 if (f_hi(2)) call s_amr_wave_req_raw(amr_wave, dir, freg(2)%hi(:,:,:,amr_reg_cur), cnt, peer)
875# 263 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
876 end if
877# 259 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
878 if (3 <= num_dims) then
879# 260 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
880 cnt = size(freg(3)%lo, 1)*size(freg(3)%lo, 2)*size(freg(3)%lo, 3)
881# 261 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
882 if (f_lo(3)) call s_amr_wave_req_raw(amr_wave, dir, freg(3)%lo(:,:,:,amr_reg_cur), cnt, peer)
883# 262 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
884 if (f_hi(3)) call s_amr_wave_req_raw(amr_wave, dir, freg(3)%hi(:,:,:,amr_reg_cur), cnt, peer)
885# 263 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
886 end if
887# 265 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
888
889 end subroutine s_amr_freg_faces_req
890
891 !> Move the flagged face registers of the current slot to the device (to_device) or to the host.
892 impure subroutine s_amr_freg_faces_update(to_device, f_lo, f_hi)
893
894 logical, intent(in) :: to_device, f_lo(3), f_hi(3)
895
896# 274 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
897# 275 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
898 if (1 <= num_dims .and. f_lo(1)) then
899# 276 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
900 if (to_device) then
901
902# 277 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
903#if defined(MFC_OpenACC)
904# 277 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
905!$acc update device(freg(1)%lo(:, :, :, amr_reg_cur))
906# 277 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
907#elif defined(MFC_OpenMP)
908# 277 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
909!$omp target update to(freg(1)%lo(:, :, :, amr_reg_cur))
910# 277 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
911#endif
912# 278 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
913 else
914
915# 279 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
916#if defined(MFC_OpenACC)
917# 279 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
918!$acc update host(freg(1)%lo(:, :, :, amr_reg_cur))
919# 279 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
920#elif defined(MFC_OpenMP)
921# 279 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
922!$omp target update from(freg(1)%lo(:, :, :, amr_reg_cur))
923# 279 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
924#endif
925# 280 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
926 end if
927 end if
928# 275 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
929 if (1 <= num_dims .and. f_hi(1)) then
930# 276 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
931 if (to_device) then
932
933# 277 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
934#if defined(MFC_OpenACC)
935# 277 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
936!$acc update device(freg(1)%hi(:, :, :, amr_reg_cur))
937# 277 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
938#elif defined(MFC_OpenMP)
939# 277 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
940!$omp target update to(freg(1)%hi(:, :, :, amr_reg_cur))
941# 277 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
942#endif
943# 278 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
944 else
945
946# 279 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
947#if defined(MFC_OpenACC)
948# 279 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
949!$acc update host(freg(1)%hi(:, :, :, amr_reg_cur))
950# 279 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
951#elif defined(MFC_OpenMP)
952# 279 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
953!$omp target update from(freg(1)%hi(:, :, :, amr_reg_cur))
954# 279 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
955#endif
956# 280 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
957 end if
958 end if
959# 283 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
960# 274 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
961# 275 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
962 if (2 <= num_dims .and. f_lo(2)) then
963# 276 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
964 if (to_device) then
965
966# 277 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
967#if defined(MFC_OpenACC)
968# 277 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
969!$acc update device(freg(2)%lo(:, :, :, amr_reg_cur))
970# 277 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
971#elif defined(MFC_OpenMP)
972# 277 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
973!$omp target update to(freg(2)%lo(:, :, :, amr_reg_cur))
974# 277 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
975#endif
976# 278 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
977 else
978
979# 279 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
980#if defined(MFC_OpenACC)
981# 279 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
982!$acc update host(freg(2)%lo(:, :, :, amr_reg_cur))
983# 279 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
984#elif defined(MFC_OpenMP)
985# 279 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
986!$omp target update from(freg(2)%lo(:, :, :, amr_reg_cur))
987# 279 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
988#endif
989# 280 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
990 end if
991 end if
992# 275 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
993 if (2 <= num_dims .and. f_hi(2)) then
994# 276 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
995 if (to_device) then
996
997# 277 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
998#if defined(MFC_OpenACC)
999# 277 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
1000!$acc update device(freg(2)%hi(:, :, :, amr_reg_cur))
1001# 277 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
1002#elif defined(MFC_OpenMP)
1003# 277 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
1004!$omp target update to(freg(2)%hi(:, :, :, amr_reg_cur))
1005# 277 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
1006#endif
1007# 278 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
1008 else
1009
1010# 279 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
1011#if defined(MFC_OpenACC)
1012# 279 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
1013!$acc update host(freg(2)%hi(:, :, :, amr_reg_cur))
1014# 279 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
1015#elif defined(MFC_OpenMP)
1016# 279 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
1017!$omp target update from(freg(2)%hi(:, :, :, amr_reg_cur))
1018# 279 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
1019#endif
1020# 280 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
1021 end if
1022 end if
1023# 283 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
1024# 274 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
1025# 275 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
1026 if (3 <= num_dims .and. f_lo(3)) then
1027# 276 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
1028 if (to_device) then
1029
1030# 277 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
1031#if defined(MFC_OpenACC)
1032# 277 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
1033!$acc update device(freg(3)%lo(:, :, :, amr_reg_cur))
1034# 277 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
1035#elif defined(MFC_OpenMP)
1036# 277 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
1037!$omp target update to(freg(3)%lo(:, :, :, amr_reg_cur))
1038# 277 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
1039#endif
1040# 278 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
1041 else
1042
1043# 279 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
1044#if defined(MFC_OpenACC)
1045# 279 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
1046!$acc update host(freg(3)%lo(:, :, :, amr_reg_cur))
1047# 279 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
1048#elif defined(MFC_OpenMP)
1049# 279 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
1050!$omp target update from(freg(3)%lo(:, :, :, amr_reg_cur))
1051# 279 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
1052#endif
1053# 280 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
1054 end if
1055 end if
1056# 275 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
1057 if (3 <= num_dims .and. f_hi(3)) then
1058# 276 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
1059 if (to_device) then
1060
1061# 277 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
1062#if defined(MFC_OpenACC)
1063# 277 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
1064!$acc update device(freg(3)%hi(:, :, :, amr_reg_cur))
1065# 277 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
1066#elif defined(MFC_OpenMP)
1067# 277 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
1068!$omp target update to(freg(3)%hi(:, :, :, amr_reg_cur))
1069# 277 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
1070#endif
1071# 278 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
1072 else
1073
1074# 279 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
1075#if defined(MFC_OpenACC)
1076# 279 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
1077!$acc update host(freg(3)%hi(:, :, :, amr_reg_cur))
1078# 279 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
1079#elif defined(MFC_OpenMP)
1080# 279 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
1081!$omp target update from(freg(3)%hi(:, :, :, amr_reg_cur))
1082# 279 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
1083#endif
1084# 280 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
1085 end if
1086 end if
1087# 283 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
1088# 284 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
1089
1090 end subroutine s_amr_freg_faces_update
1091
1092 !> Debug builds: NaN-poison the unflagged face registers of the current slot before a receive lands, so a stray read of an
1093 !! unshipped face NaNs.
1094 impure subroutine s_amr_freg_faces_poison(f_lo, f_hi)
1095
1096#ifdef MFC_DEBUG
1097 use ieee_arithmetic, only: ieee_value, ieee_quiet_nan
1098#endif
1099 logical, intent(in) :: f_lo(3), f_hi(3)
1100#ifdef MFC_DEBUG
1101 real(wp) :: nanv
1102
1103 if (amr_reg_cur <= 0) return
1104 nanv = ieee_value(0._wp, ieee_quiet_nan)
1105# 301 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
1106# 302 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
1107 if (1 <= num_dims .and. .not. f_lo(1)) then
1108# 303 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
1109 freg(1)%lo(:,:,:,amr_reg_cur) = nanv
1110# 304 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
1111
1112# 304 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
1113#if defined(MFC_OpenACC)
1114# 304 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
1115!$acc update device(freg(1)%lo(:, :, :, amr_reg_cur))
1116# 304 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
1117#elif defined(MFC_OpenMP)
1118# 304 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
1119!$omp target update to(freg(1)%lo(:, :, :, amr_reg_cur))
1120# 304 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
1121#endif
1122# 305 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
1123 end if
1124# 302 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
1125 if (1 <= num_dims .and. .not. f_hi(1)) then
1126# 303 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
1127 freg(1)%hi(:,:,:,amr_reg_cur) = nanv
1128# 304 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
1129
1130# 304 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
1131#if defined(MFC_OpenACC)
1132# 304 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
1133!$acc update device(freg(1)%hi(:, :, :, amr_reg_cur))
1134# 304 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
1135#elif defined(MFC_OpenMP)
1136# 304 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
1137!$omp target update to(freg(1)%hi(:, :, :, amr_reg_cur))
1138# 304 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
1139#endif
1140# 305 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
1141 end if
1142# 307 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
1143# 301 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
1144# 302 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
1145 if (2 <= num_dims .and. .not. f_lo(2)) then
1146# 303 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
1147 freg(2)%lo(:,:,:,amr_reg_cur) = nanv
1148# 304 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
1149
1150# 304 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
1151#if defined(MFC_OpenACC)
1152# 304 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
1153!$acc update device(freg(2)%lo(:, :, :, amr_reg_cur))
1154# 304 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
1155#elif defined(MFC_OpenMP)
1156# 304 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
1157!$omp target update to(freg(2)%lo(:, :, :, amr_reg_cur))
1158# 304 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
1159#endif
1160# 305 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
1161 end if
1162# 302 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
1163 if (2 <= num_dims .and. .not. f_hi(2)) then
1164# 303 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
1165 freg(2)%hi(:,:,:,amr_reg_cur) = nanv
1166# 304 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
1167
1168# 304 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
1169#if defined(MFC_OpenACC)
1170# 304 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
1171!$acc update device(freg(2)%hi(:, :, :, amr_reg_cur))
1172# 304 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
1173#elif defined(MFC_OpenMP)
1174# 304 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
1175!$omp target update to(freg(2)%hi(:, :, :, amr_reg_cur))
1176# 304 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
1177#endif
1178# 305 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
1179 end if
1180# 307 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
1181# 301 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
1182# 302 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
1183 if (3 <= num_dims .and. .not. f_lo(3)) then
1184# 303 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
1185 freg(3)%lo(:,:,:,amr_reg_cur) = nanv
1186# 304 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
1187
1188# 304 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
1189#if defined(MFC_OpenACC)
1190# 304 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
1191!$acc update device(freg(3)%lo(:, :, :, amr_reg_cur))
1192# 304 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
1193#elif defined(MFC_OpenMP)
1194# 304 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
1195!$omp target update to(freg(3)%lo(:, :, :, amr_reg_cur))
1196# 304 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
1197#endif
1198# 305 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
1199 end if
1200# 302 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
1201 if (3 <= num_dims .and. .not. f_hi(3)) then
1202# 303 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
1203 freg(3)%hi(:,:,:,amr_reg_cur) = nanv
1204# 304 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
1205
1206# 304 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
1207#if defined(MFC_OpenACC)
1208# 304 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
1209!$acc update device(freg(3)%hi(:, :, :, amr_reg_cur))
1210# 304 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
1211#elif defined(MFC_OpenMP)
1212# 304 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
1213!$omp target update to(freg(3)%hi(:, :, :, amr_reg_cur))
1214# 304 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
1215#endif
1216# 305 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
1217 end if
1218# 307 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
1219# 308 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
1220#endif
1221
1222 end subroutine s_amr_freg_faces_poison
1223
1224 !> Set the fine level's geometry (region, intersection, extents, bounds, coordinates) for the box lo:hi; arrays are
1225 !! preallocated, this updates metadata and coords. Collective. Also refreshes the amr_xchg_coarse_ghosts flag. A level-l block's
1226 !! fine extent is amr_ref_ratio**l * (coarse-region width) - 1, not amr_ref_ratio*width.
1227 impure subroutine s_set_amr_fine_geometry(lo, hi)
1228
1229 integer, intent(in) :: lo(3), hi(3)
1230 integer :: fext(3), nmar, bad_loc, pblk
1231
1232 amr_slots(amr_cur)%region%lo = lo; amr_slots(amr_cur)%region%hi = hi
1233 amr_region_lo = lo; amr_region_hi = hi ! mirror for m_amr_registers (avoids a use-cycle)
1235
1236 ! the owner holds every fine cell of the block, other ranks none; amr_isect_lo/hi is the block's coarse footprint on
1237 ! the owner (empty elsewhere)
1239 pblk = 0
1240 if (amr_rank_owns_block) then
1241 amr_isect_lo = lo; amr_isect_hi = hi
1242 if (amr_block_level(amr_cur) >= 2) then
1243 ! multi-level: footprint in the parent block's fine frame (parent-fine index of L0 cell c is
1244 ! amr_ref_ratio**(parent level)*(c - parent lo)), from replicated metadata: amr_slots(pblk) may be unallocated here
1247 end if
1248 else
1249 amr_isect_lo = 1; amr_isect_hi = 0 ! empty
1250 end if
1253 ! fine extents cover the whole block on the owner, -1 (empty) elsewhere
1254 fext = merge(amr_ref_ratio*max(amr_isect_hi - amr_isect_lo + 1, 0) - 1, 0, amr_dim)
1256 amr_slots(amr_cur)%m = fext(1); amr_slots(amr_cur)%n = fext(2); amr_slots(amr_cur)%p = fext(3)
1257 amr_slots(amr_cur)%idwbuff%beg = merge(-buff_size, 0, amr_dim); amr_slots(amr_cur)%idwbuff%end = merge(fext + buff_size, &
1258 & 0, amr_dim)
1259 ! coords only on ranks with fine cells: replay the ancestor chain from the global L0 boundaries (the owner may hold no
1260 ! coarse coords nor the parent's slot)
1261 if (amr_rank_owns_block) then
1262# 351 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
1263 if (amr_dim(1)) call s_amr_build_block_coords(amr_cur, amr_gxcb, lbound(amr_gxcb, 1), &
1264# 352 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
1265 & amr_slots(amr_cur)%x_cb, amr_slots(amr_cur)%x_cc, amr_slots(amr_cur)%dx, 1)
1266# 353 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
1267# 351 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
1268 if (amr_dim(2)) call s_amr_build_block_coords(amr_cur, amr_gycb, lbound(amr_gycb, 1), &
1269# 352 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
1270 & amr_slots(amr_cur)%y_cb, amr_slots(amr_cur)%y_cc, amr_slots(amr_cur)%dy, 2)
1271# 353 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
1272# 351 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
1273 if (amr_dim(3)) call s_amr_build_block_coords(amr_cur, amr_gzcb, lbound(amr_gzcb, 1), &
1274# 352 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
1275 & amr_slots(amr_cur)%z_cb, amr_slots(amr_cur)%z_cc, amr_slots(amr_cur)%dz, 3)
1276# 353 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
1277# 354 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
1278 end if
1279
1280 ! The ghost prolongation reads up to nmar coarse cells past each face; if that leaves any rank's interior, the coarse
1281 ! cons ghosts must be halo-exchanged before every fill (the solver populates only prim ghosts). All ranks agree on the
1282 ! flag.
1283 nmar = (buff_size + amr_ref_ratio - 1)/amr_ref_ratio + 1
1284 bad_loc = 0
1285 if (amr_rank_owns_block .and. any(amr_dim .and. (amr_isect_lo - amr_sidx < nmar .or. amr_sidx + amr_ext &
1286 & - amr_isect_hi < nmar))) bad_loc = 1
1287 ! Accumulate (OR over blocks), do not reduce: the caller closes the scan with s_amr_reduce_xchg_flag, one collective
1288 ! per scan.
1289 amr_xchg_bad = max(amr_xchg_bad, bad_loc)
1290
1291 end subroutine s_set_amr_fine_geometry
1292
1293 !> Abort unless block k's fine extent is amr_ref_ratio**level*(region width) - 1 and, above level 1, its region lies inside
1294 !! parent pblk.
1295 impure subroutine s_amr_check_block_geometry(k, pblk, fext)
1296
1297 integer, intent(in) :: k, pblk, fext(3)
1298 integer :: want(3)
1299
1300 want = merge(amr_ref_ratio**amr_block_level(k)*(amr_region_hi_all(:,k) - amr_region_lo_all(:,k) + 1) - 1, 0, amr_dim)
1301 if (any(fext /= want)) then
1302 print '(A,I0,A,I0,A,3(1X,I0),A,3(1X,I0))', ' [amr] block ', k, ' level ', amr_block_level(k), ': fine extent', fext, &
1303 & ', expected', want
1304 call s_mpi_abort("AMR: a block's fine extent is not amr_ref_ratio**level*(region width) - 1 (printed above).")
1305 end if
1306 if (amr_block_level(k) < 2) return
1307 if (pblk == 0) call s_mpi_abort("AMR: a block above level 1 has no parent block.")
1308 if (any(amr_dim .and. (amr_region_lo_all(:,k) < amr_region_lo_all(:,pblk) .or. amr_region_hi_all(:, &
1309 & k) > amr_region_hi_all(:,pblk)))) then
1310 print '(A,I0,A,I0)', ' [amr] block ', k, ' is not inside its parent block ', pblk
1311 call s_mpi_abort("AMR: a block above level 1 does not lie inside its parent block (printed above).")
1312 end if
1313
1314 end subroutine s_amr_check_block_geometry
1315
1316 !> Close a geometry scan: one allreduce of the accumulated flag, then reset. Call after every s_set_amr_fine_geometry loop; the
1317 !! flag is read by s_amr_exchange_coarse_cons_halo.
1318 impure subroutine s_amr_reduce_xchg_flag()
1319
1320 integer :: bad_glb
1321
1323 amr_xchg_coarse_ghosts = bad_glb == 1
1324 amr_xchg_bad = 0
1325 call s_amr_build_cfl_map() ! the block set changed
1326
1327 end subroutine s_amr_reduce_xchg_flag
1328
1329 !> amr_cfl_rf from the block table: amr_ref_ratio**level of the deepest block over each of this rank's coarse cells.
1330 impure subroutine s_amr_build_cfl_map()
1331
1332 integer :: k, lo(3), hi(3)
1333
1334 amr_cfl_pow = merge([0._wp, 1._wp, 0.5_wp], [1._wp, 2._wp, 1.5_wp], amr_subcycle)
1335 amr_cfl_rf = 1._wp
1336 do k = 1, amr_num_blocks
1337 lo = max(merge(amr_region_lo_all(:,k) - amr_sidx, 0, amr_dim), 0)
1338 hi = min(merge(amr_region_hi_all(:,k) - amr_sidx, 0, amr_dim), [m, n, p])
1339 if (any(lo > hi)) cycle
1340 amr_cfl_rf(lo(1):hi(1),lo(2):hi(2),lo(3):hi(3)) = max(amr_cfl_rf(lo(1):hi(1),lo(2):hi(2),lo(3):hi(3)), &
1341 & real(amr_ref_ratio**amr_block_level(k), wp))
1342 end do
1343
1344# 419 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
1345#if defined(MFC_OpenACC)
1346# 419 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
1347!$acc update device(amr_cfl_rf)
1348# 419 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
1349#elif defined(MFC_OpenMP)
1350# 419 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
1351!$omp target update to(amr_cfl_rf)
1352# 419 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
1353#endif
1354# 420 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
1355
1356 end subroutine s_amr_build_cfl_map
1357
1358 !> Conservative prolongation of one variable from the patch qc into the fine interior: minmod-limited linear, or the ghost
1359 !! fill's limited quartic with amr_prolong = 2.
1360 impure subroutine s_prolong_one_var(qc, loc, ivar)
1361
1362 type(scalar_field), intent(in) :: qc
1363 integer, intent(in) :: loc, ivar !< flat-store slot and variable of the fine target
1364 integer :: fi, fj, fk, ci, cj, ck, ox, oy, oz, rrat, mm, nn, pp, il1, il2, il3, ky, kz, wy, wz
1365 real(wp) :: u0, xm, xp, ym, yp, zm, zp, xix, xiy, xiz
1366 real(wp) :: q5(-2:2), pl(-2:2), ln(-2:2) !< staged x, y, z
1367 logical :: d2, d3, quart
1368
1369 ! qc is the gathered patch amr_cg: amr_isect_lo + f/rr - amr_cpat_off = nmar + f/rr is the patch-local coarse index.
1370 ! Device kernel reading the patch's device mirror (pushed once per dispatch by s_interpolate_coarse_to_fine).
1371
1372 ox = amr_cpat_off(1); oy = amr_cpat_off(2); oz = amr_cpat_off(3)
1373 rrat = amr_slots(amr_cur)%amr_ref_ratio
1374 mm = amr_slots(amr_cur)%m; nn = amr_slots(amr_cur)%n; pp = amr_slots(amr_cur)%p
1375 il1 = amr_isect_lo(1); il2 = amr_isect_lo(2); il3 = amr_isect_lo(3)
1376 d2 = n_glb > 0; d3 = p_glb > 0
1377 quart = amr_prolong == 2; wy = merge(2, 0, d2); wz = merge(2, 0, d3)
1378
1379# 443 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
1380
1381# 443 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
1382#if defined(MFC_OpenACC)
1383# 443 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
1384!$acc parallel loop collapse(3) gang vector default(present) private(ci, cj, ck, xix, xiy, xiz, u0, xm, xp, ym, yp, zm, zp, ky, kz, q5, pl, ln)
1385# 443 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
1386#elif defined(MFC_OpenMP)
1387# 443 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
1388
1389# 443 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
1390
1391# 443 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
1392
1393# 443 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
1394!$omp target teams loop defaultmap(firstprivate:scalar) bind(teams,parallel) collapse(3) defaultmap(tofrom:aggregate) defaultmap(tofrom:allocatable) defaultmap(tofrom:pointer) &
1395# 443 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
1396!$omp& private(ci, cj, ck, xix, xiy, xiz, u0, xm, xp, ym, yp, zm, zp, ky, kz, q5, pl, ln)
1397# 443 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
1398#endif
1399# 444 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
1400 do fk = 0, pp
1401 do fj = 0, nn
1402 do fi = 0, mm
1403 ck = il3 + fk/rrat - oz; if (.not. d3) ck = 0
1404 xiz = 0._wp; if (d3) xiz = (real(mod(fk, rrat), wp) - real(rrat - 1, wp)*0.5_wp)/real(rrat, wp)
1405 cj = il2 + fj/rrat - oy; if (.not. d2) cj = 0
1406 xiy = 0._wp; if (d2) xiy = (real(mod(fj, rrat), wp) - real(rrat - 1, wp)*0.5_wp)/real(rrat, wp)
1407 ci = il1 + fi/rrat - ox
1408 xix = (real(mod(fi, rrat), wp) - real(rrat - 1, wp)*0.5_wp)/real(rrat, wp)
1409 if (quart) then
1410 do kz = -wz, wz
1411 do ky = -wy, wy
1412 q5 = real(qc%sf(ci - 2:ci + 2,cj + ky, ck + kz), wp)
1413 pl(ky) = f_amr_quartic_half(q5, xix)
1414 end do
1415 ln(kz) = pl(0)
1416 if (d2) ln(kz) = f_amr_quartic_half(pl, xiy)
1417 end do
1418 u0 = ln(0)
1419 if (d3) u0 = f_amr_quartic_half(ln, xiz)
1420 amr_cons_st(fi, fj, fk, ivar, loc) = u0
1421 else
1422 u0 = real(qc%sf(ci, cj, ck), wp)
1423 xm = real(qc%sf(ci - 1, cj, ck), wp); xp = real(qc%sf(ci + 1, cj, ck), wp)
1424 ym = u0; yp = u0
1425 if (d2) then
1426 ym = real(qc%sf(ci, cj - 1, ck), wp); yp = real(qc%sf(ci, cj + 1, ck), wp)
1427 end if
1428 zm = u0; zp = u0
1429 if (d3) then
1430 zm = real(qc%sf(ci, cj, ck - 1), wp); zp = real(qc%sf(ci, cj, ck + 1), wp)
1431 end if
1432 amr_cons_st(fi, fj, fk, ivar, loc) = f_amr_lin(u0, xm, xp, ym, yp, zm, zp, xix, xiy, xiz)
1433 end if
1434 end do
1435 end do
1436 end do
1437
1438# 481 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
1439#if defined(MFC_OpenACC)
1440# 481 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
1441!$acc end parallel loop
1442# 481 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
1443#elif defined(MFC_OpenMP)
1444# 481 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
1445
1446# 481 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
1447!$omp end target teams loop
1448# 481 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
1449#endif
1450# 482 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
1451
1452 end subroutine s_prolong_one_var
1453
1454 !> Quartic safety net (amr_prolong = 2): a parent whose children leave the admissible state (non-positive density or internal
1455 !! energy) is re-prolonged minmod-limited linear in every variable.
1456 impure subroutine s_prolong_guard(qc, loc)
1457
1458 type(scalar_field), dimension(sys_size), intent(in) :: qc
1459 integer, intent(in) :: loc
1460 integer :: i, pi, pj, pk, a, b, c, fi, fj, fk, ci, cj, ck, rrat, ry, rz
1461 integer :: ox, oy, oz, il1, il2, il3, nx, ny, nz, advb, adve, spb, spe
1462 real(wp) :: u0, xm, xp, ym, yp, zm, zp, xix, xiy, xiz
1463 logical :: d2, d3, bad, skip_alpha
1464
1465 ox = amr_cpat_off(1); oy = amr_cpat_off(2); oz = amr_cpat_off(3)
1466 rrat = amr_slots(amr_cur)%amr_ref_ratio
1467 il1 = amr_isect_lo(1); il2 = amr_isect_lo(2); il3 = amr_isect_lo(3)
1468 d2 = n_glb > 0; d3 = p_glb > 0
1469 ry = merge(rrat, 1, d2); rz = merge(rrat, 1, d3)
1470 nx = (amr_slots(amr_cur)%m + 1)/rrat; ny = (amr_slots(amr_cur)%n + 1)/ry; nz = (amr_slots(amr_cur)%p + 1)/rz
1471 skip_alpha = num_fluids > 1 .and. (.not. bubbles_lagrange)
1472 advb = eqn_idx%adv%beg; adve = eqn_idx%adv%end; spb = eqn_idx%species%beg; spe = eqn_idx%species%end
1473
1474# 504 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
1475
1476# 504 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
1477#if defined(MFC_OpenACC)
1478# 504 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
1479!$acc parallel loop collapse(3) gang vector default(present) private(i, a, b, c, fi, fj, fk, ci, cj, ck, xix, xiy, xiz, u0, xm, xp, ym, yp, zm, zp, bad)
1480# 504 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
1481#elif defined(MFC_OpenMP)
1482# 504 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
1483
1484# 504 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
1485
1486# 504 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
1487
1488# 504 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
1489!$omp target teams loop defaultmap(firstprivate:scalar) bind(teams,parallel) collapse(3) defaultmap(tofrom:aggregate) defaultmap(tofrom:allocatable) defaultmap(tofrom:pointer) &
1490# 504 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
1491!$omp& private(i, a, b, c, fi, fj, fk, ci, cj, ck, xix, xiy, xiz, u0, xm, xp, ym, yp, zm, zp, bad)
1492# 504 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
1493#endif
1494# 506 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
1495 do pk = 0, nz - 1
1496 do pj = 0, ny - 1
1497 do pi = 0, nx - 1
1498 bad = .false.
1499 do c = 0, rz - 1
1500 do b = 0, ry - 1
1501 do a = 0, rrat - 1
1502 fi = pi*rrat + a; fj = pj*ry + b; fk = pk*rz + c
1503 if (.not. f_amr_cons_ok(fi, fj, fk, loc)) bad = .true.
1504 end do
1505 end do
1506 end do
1507 if (bad) then
1508 ci = il1 + pi - ox; cj = 0; ck = 0
1509 if (d2) cj = il2 + pj - oy
1510 if (d3) ck = il3 + pk - oz
1511 do c = 0, rz - 1
1512 do b = 0, ry - 1
1513 do a = 0, rrat - 1
1514 fi = pi*rrat + a; fj = pj*ry + b; fk = pk*rz + c
1515 xix = (real(a, wp) - real(rrat - 1, wp)*0.5_wp)/real(rrat, wp)
1516 xiy = 0._wp; if (d2) xiy = (real(b, wp) - real(rrat - 1, wp)*0.5_wp)/real(rrat, wp)
1517 xiz = 0._wp; if (d3) xiz = (real(c, wp) - real(rrat - 1, wp)*0.5_wp)/real(rrat, wp)
1518
1519# 529 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
1520#if defined(MFC_OpenACC)
1521# 529 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
1522!$acc loop seq
1523# 529 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
1524#elif defined(MFC_OpenMP)
1525# 529 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
1526
1527# 529 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
1528#endif
1529# 530 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
1530 do i = 1, sys_size
1531 if (skip_alpha .and. i >= advb .and. i <= adve) cycle
1532 if (chemistry .and. i >= spb .and. i <= spe) cycle
1533 u0 = real(qc(i)%sf(ci, cj, ck), wp)
1534 xm = real(qc(i)%sf(ci - 1, cj, ck), wp); xp = real(qc(i)%sf(ci + 1, cj, ck), wp)
1535 ym = u0; yp = u0
1536 if (d2) then
1537 ym = real(qc(i)%sf(ci, cj - 1, ck), wp); yp = real(qc(i)%sf(ci, cj + 1, ck), wp)
1538 end if
1539 zm = u0; zp = u0
1540 if (d3) then
1541 zm = real(qc(i)%sf(ci, cj, ck - 1), wp); zp = real(qc(i)%sf(ci, cj, ck + 1), wp)
1542 end if
1543 amr_cons_st(fi, fj, fk, i, loc) = f_amr_lin(u0, xm, xp, ym, yp, zm, zp, xix, xiy, xiz)
1544 end do
1545 end do
1546 end do
1547 end do
1548 end if
1549 end do
1550 end do
1551 end do
1552
1553# 552 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
1554#if defined(MFC_OpenACC)
1555# 552 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
1556!$acc end parallel loop
1557# 552 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
1558#elif defined(MFC_OpenMP)
1559# 552 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
1560
1561# 552 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
1562!$omp end target teams loop
1563# 552 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
1564#endif
1565# 553 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
1566
1567 end subroutine s_prolong_guard
1568
1569 !> Conservative prolongation: fill amr_fine interior from coarse (minmod-limited linear, or quartic with amr_prolong = 2);
1570 !! symmetric child offsets make the children average to the coarse value. Multi-fluid volume fractions take the sum- preserving
1571 !! closure path.
1573
1574 integer :: i
1575
1576 ! the prolong kernels read the gathered patch on the device
1577
1578 integer :: loc_cur !< plain scalar for device routines (NVHPC -Minline)
1579
1580 do i = 1, sys_size
1581 ! Lagrangian bubbles: alphas sum to beta, not 1, so each alpha prolongs plainly
1582 if (num_fluids > 1 .and. (.not. bubbles_lagrange) .and. i >= eqn_idx%adv%beg .and. i <= eqn_idx%adv%end) cycle
1583 if (chemistry .and. i >= eqn_idx%species%beg .and. i <= eqn_idx%species%end) cycle ! closure below
1584 loc_cur = amr_loc_of(amr_cur)
1585 call s_prolong_one_var(amr_cg(i), loc_cur, i)
1586 end do
1587 loc_cur = amr_loc_of(amr_cur)
1588 if (num_fluids > 1 .and. (.not. bubbles_lagrange)) call s_prolong_alphas_closure(amr_cg, loc_cur)
1589 loc_cur = amr_loc_of(amr_cur)
1590 if (chemistry) call s_prolong_species_closure(amr_cg, loc_cur)
1591 if (amr_prolong == 2) call s_prolong_guard(amr_cg, loc_cur)
1592
1593 end subroutine s_interpolate_coarse_to_fine
1594
1595 !> Sum-preserving volume-fraction prolongation (num_fluids > 1): fluids adv%beg..adv%end-1 are interpolated with minmod slopes
1596 !! under a shared per-cell limiter switch and clamped to [0,1]; the last fluid closes alpha_n = 1 - sum(others). For >2 fluids
1597 !! residual closure undershoot is handled by mpp_lim. Index mapping as s_prolong_one_var.
1598 impure subroutine s_prolong_alphas_closure(qc, loc)
1599
1600 type(scalar_field), dimension(sys_size), intent(in) :: qc
1601 integer, intent(in) :: loc
1602 integer :: fi, fj, fk, ci, cj, ck, ox, oy, oz, i
1603 integer :: rrat, mm, nn, pp, il1, il2, il3, advb, adve
1604 real(wp) :: xix, xiy, xiz, u0, xm, xp, ym, yp, zm, zp, av, asum
1605 logical :: shx, shy, shz, d2, d3
1606
1607 ! qc is the gathered patch amr_cg (patch-frame offset). Device kernel; the shared limiter switch is inlined.
1608
1609 ox = amr_cpat_off(1); oy = amr_cpat_off(2); oz = amr_cpat_off(3)
1610 rrat = amr_slots(amr_cur)%amr_ref_ratio
1611 mm = amr_slots(amr_cur)%m; nn = amr_slots(amr_cur)%n; pp = amr_slots(amr_cur)%p
1612 il1 = amr_isect_lo(1); il2 = amr_isect_lo(2); il3 = amr_isect_lo(3)
1613 d2 = n_glb > 0; d3 = p_glb > 0
1614 advb = eqn_idx%adv%beg; adve = eqn_idx%adv%end
1615
1616# 602 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
1617
1618# 602 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
1619#if defined(MFC_OpenACC)
1620# 602 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
1621!$acc parallel loop collapse(3) gang vector default(present) private(ci, cj, ck, xix, xiy, xiz, u0, xm, xp, ym, yp, zm, zp, av, asum, shx, shy, shz, i)
1622# 602 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
1623#elif defined(MFC_OpenMP)
1624# 602 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
1625
1626# 602 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
1627
1628# 602 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
1629
1630# 602 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
1631!$omp target teams loop defaultmap(firstprivate:scalar) bind(teams,parallel) collapse(3) defaultmap(tofrom:aggregate) defaultmap(tofrom:allocatable) defaultmap(tofrom:pointer) &
1632# 602 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
1633!$omp& private(ci, cj, ck, xix, xiy, xiz, u0, xm, xp, ym, yp, zm, zp, av, asum, shx, shy, shz, i)
1634# 602 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
1635#endif
1636# 604 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
1637 do fk = 0, pp
1638 do fj = 0, nn
1639 do fi = 0, mm
1640 ck = il3 + fk/rrat - oz; if (.not. d3) ck = 0
1641 xiz = 0._wp; if (d3) xiz = (real(mod(fk, rrat), wp) - real(rrat - 1, wp)*0.5_wp)/real(rrat, wp)
1642 cj = il2 + fj/rrat - oy; if (.not. d2) cj = 0
1643 xiy = 0._wp; if (d2) xiy = (real(mod(fj, rrat), wp) - real(rrat - 1, wp)*0.5_wp)/real(rrat, wp)
1644 ci = il1 + fi/rrat - ox
1645 xix = (real(mod(fi, rrat), wp) - real(rrat - 1, wp)*0.5_wp)/real(rrat, wp)
1646 ! shared per-cell limiter switch: per dim, slopes stay on only if no fluid's centered differences change
1647 ! sign there
1648 shx = .true.; shy = d2; shz = d3
1649 do i = advb, adve
1650 u0 = real(qc(i)%sf(ci, cj, ck), wp)
1651 if ((real(qc(i)%sf(ci + 1, cj, ck), wp) - u0)*(u0 - real(qc(i)%sf(ci - 1, cj, ck), &
1652 & wp)) <= 0._wp) shx = .false.
1653 if (d2) then
1654 if ((real(qc(i)%sf(ci, cj + 1, ck), wp) - u0)*(u0 - real(qc(i)%sf(ci, cj - 1, ck), &
1655 & wp)) <= 0._wp) shy = .false.
1656 end if
1657 if (d3) then
1658 if ((real(qc(i)%sf(ci, cj, ck + 1), wp) - u0)*(u0 - real(qc(i)%sf(ci, cj, ck - 1), &
1659 & wp)) <= 0._wp) shz = .false.
1660 end if
1661 end do
1662 asum = 0._wp
1663 do i = advb, adve - 1
1664 u0 = real(qc(i)%sf(ci, cj, ck), wp)
1665 xm = u0; xp = u0
1666 if (shx) then
1667 xm = real(qc(i)%sf(ci - 1, cj, ck), wp); xp = real(qc(i)%sf(ci + 1, cj, ck), wp)
1668 end if
1669 ym = u0; yp = u0
1670 if (shy) then
1671 ym = real(qc(i)%sf(ci, cj - 1, ck), wp); yp = real(qc(i)%sf(ci, cj + 1, ck), wp)
1672 end if
1673 zm = u0; zp = u0
1674 if (shz) then
1675 zm = real(qc(i)%sf(ci, cj, ck - 1), wp); zp = real(qc(i)%sf(ci, cj, ck + 1), wp)
1676 end if
1677 av = min(max(f_amr_lin(u0, xm, xp, ym, yp, zm, zp, xix, xiy, xiz), 0._wp), 1._wp)
1678 amr_cons_st(fi, fj, fk, i, loc) = av
1679 asum = asum + av
1680 end do
1681 amr_cons_st(fi, fj, fk, adve, loc) = 1._wp - asum
1682 end do
1683 end do
1684 end do
1685
1686# 652 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
1687#if defined(MFC_OpenACC)
1688# 652 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
1689!$acc end parallel loop
1690# 652 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
1691#elif defined(MFC_OpenMP)
1692# 652 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
1693
1694# 652 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
1695!$omp end target teams loop
1696# 652 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
1697#endif
1698# 653 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
1699
1700 end subroutine s_prolong_alphas_closure
1701
1702 !> Species prolongation closure (chemistry): each rho*Y_k is minmod-prolonged and clamped non-negative, then rescaled so
1703 !! sum_k(rho*Y_k) equals the prolonged continuity density. Index mapping as s_prolong_one_var; cont is prolonged before this
1704 !! runs.
1705 impure subroutine s_prolong_species_closure(qc, loc)
1706
1707 type(scalar_field), dimension(sys_size), intent(in) :: qc
1708 integer, intent(in) :: loc
1709 integer :: fi, fj, fk, ci, cj, ck, ox, oy, oz, i
1710 integer :: rrat, mm, nn, pp, il1, il2, il3, spb, spe, cte
1711 real(wp) :: xix, xiy, xiz, u0, xm, xp, ym, yp, zm, zp, av, rsum, rscale
1712 logical :: d2, d3
1713
1714 ! qc is the gathered patch amr_cg (patch-frame offset). Device kernel; the rescale reads only the thread's own cell.
1715
1716 ox = amr_cpat_off(1); oy = amr_cpat_off(2); oz = amr_cpat_off(3)
1717 rrat = amr_slots(amr_cur)%amr_ref_ratio
1718 mm = amr_slots(amr_cur)%m; nn = amr_slots(amr_cur)%n; pp = amr_slots(amr_cur)%p
1719 il1 = amr_isect_lo(1); il2 = amr_isect_lo(2); il3 = amr_isect_lo(3)
1720 d2 = n_glb > 0; d3 = p_glb > 0
1721 spb = eqn_idx%species%beg; spe = eqn_idx%species%end; cte = eqn_idx%cont%end
1722
1723# 676 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
1724
1725# 676 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
1726#if defined(MFC_OpenACC)
1727# 676 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
1728!$acc parallel loop collapse(3) gang vector default(present) private(ci, cj, ck, xix, xiy, xiz, u0, xm, xp, ym, yp, zm, zp, av, rsum, rscale, i)
1729# 676 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
1730#elif defined(MFC_OpenMP)
1731# 676 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
1732
1733# 676 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
1734
1735# 676 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
1736
1737# 676 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
1738!$omp target teams loop defaultmap(firstprivate:scalar) bind(teams,parallel) collapse(3) defaultmap(tofrom:aggregate) defaultmap(tofrom:allocatable) defaultmap(tofrom:pointer) &
1739# 676 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
1740!$omp& private(ci, cj, ck, xix, xiy, xiz, u0, xm, xp, ym, yp, zm, zp, av, rsum, rscale, i)
1741# 676 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
1742#endif
1743# 677 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
1744 do fk = 0, pp
1745 do fj = 0, nn
1746 do fi = 0, mm
1747 ck = il3 + fk/rrat - oz; if (.not. d3) ck = 0
1748 xiz = 0._wp; if (d3) xiz = (real(mod(fk, rrat), wp) - real(rrat - 1, wp)*0.5_wp)/real(rrat, wp)
1749 cj = il2 + fj/rrat - oy; if (.not. d2) cj = 0
1750 xiy = 0._wp; if (d2) xiy = (real(mod(fj, rrat), wp) - real(rrat - 1, wp)*0.5_wp)/real(rrat, wp)
1751 ci = il1 + fi/rrat - ox
1752 xix = (real(mod(fi, rrat), wp) - real(rrat - 1, wp)*0.5_wp)/real(rrat, wp)
1753 rsum = 0._wp
1754 do i = spb, spe
1755 u0 = real(qc(i)%sf(ci, cj, ck), wp)
1756 xm = real(qc(i)%sf(ci - 1, cj, ck), wp); xp = real(qc(i)%sf(ci + 1, cj, ck), wp)
1757 ym = u0; yp = u0
1758 if (d2) then
1759 ym = real(qc(i)%sf(ci, cj - 1, ck), wp); yp = real(qc(i)%sf(ci, cj + 1, ck), wp)
1760 end if
1761 zm = u0; zp = u0
1762 if (d3) then
1763 zm = real(qc(i)%sf(ci, cj, ck - 1), wp); zp = real(qc(i)%sf(ci, cj, ck + 1), wp)
1764 end if
1765 av = max(f_amr_lin(u0, xm, xp, ym, yp, zm, zp, xix, xiy, xiz), 0._wp)
1766 amr_cons_st(fi, fj, fk, i, loc) = av
1767 rsum = rsum + av
1768 end do
1769 rscale = real(amr_cons_st(fi, fj, fk, cte, loc), wp)/max(rsum, 1.e-30_wp)
1770 do i = spb, spe
1771 amr_cons_st(fi, fj, fk, i, loc) = real(amr_cons_st(fi, fj, fk, i, loc), wp)*rscale
1772 end do
1773 end do
1774 end do
1775 end do
1776
1777# 709 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
1778#if defined(MFC_OpenACC)
1779# 709 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
1780!$acc end parallel loop
1781# 709 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
1782#elif defined(MFC_OpenMP)
1783# 709 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
1784
1785# 709 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
1786!$omp end target teams loop
1787# 709 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
1788#endif
1789# 710 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
1790
1791 end subroutine s_prolong_species_closure
1792
1793 !> Disblock prolongation. Guard: no-op unless amr.
1794 impure subroutine s_populate_amr_fine(q_cons_base)
1795
1796 type(scalar_field), dimension(sys_size), intent(inout) :: q_cons_base
1797 integer :: islot, kk, i
1798
1799 if (.not. amr) return
1800 ! Prolong every block from its coarse patch, assembled by the level-1 fill wave (no coarse-ghost halo needed). Runs
1801 ! before s_initialize_gpu_vars and the wave packs from the device, so push the ICs first. All ranks enter the wave;
1802 ! only owners prolong.
1803 do i = 1, sys_size
1804
1805# 724 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
1806#if defined(MFC_OpenACC)
1807# 724 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
1808!$acc update device(q_cons_base(i)%sf)
1809# 724 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
1810#elif defined(MFC_OpenMP)
1811# 724 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
1812!$omp target update to(q_cons_base(i)%sf)
1813# 724 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
1814#endif
1815# 725 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
1816 end do
1817 call s_amr_l1_fill_exchange(q_cons_base, .true.)
1819 do kk = 1, amr_n_my
1820 islot = amr_my_blk(kk)
1821 if (amr_block_level(islot) /= 1) cycle
1822 call s_amr_select_slot(islot)
1823 call s_amr_l1_fill_consume(q_cons_base, islot, .true.)
1825 end do
1827 if (amr_max_level >= 2) call s_amr_build_static_multilevel()
1828 call s_amr_select_slot(1)
1829
1830 end subroutine s_populate_amr_fine
1831
1832 !> Build the static multi-level hierarchy (amr_regrid_int = 0): nest one level-2 block inside level-1 block 1 by a fixed
1833 !! geometric inset, prolong the parent state into it and keep it persistent.
1835
1836 integer :: l2, n1, par, inset(3)
1837
1838 if (amr_max_level < 2) return
1839 n1 = amr_num_blocks
1840 if (n1 < 1) return
1841 ! the static hierarchy nests one level-2 block; without pool room it would silently refine only to level 1, so abort
1842 ! here (n1 is unknown at checker time). Replicated inputs: collective-safe.
1843 if (n1 + 1 > amr_max_blocks) call s_mpi_abort('amr static multi-level (amr_max_level > 1, ' &
1844 & // 'amr_regrid_int = 0): amr_max_blocks is too small to nest the level-2 block (need >= level-1 block count + 1); ' &
1845 & // 'increase amr_max_blocks')
1846 l2 = n1 + 1
1847 par = 1
1848 inset = merge(max((amr_region_hi_all(:,par) - amr_region_lo_all(:,par) + 1)/4, amr_cpat_mar), 0, amr_dim)
1849 amr_region_lo_all(:,l2) = amr_region_lo_all(:,par) + inset
1850 amr_region_hi_all(:,l2) = amr_region_hi_all(:,par) - inset
1851 ! Guard the fixed-inset box against configs this builder cannot represent: (a) a level-1 block smaller than 2*inset
1852 ! inverts the box; (b) a level-2 L0 extent > amr_maxc_fit/2 overruns the creg register (0:amr_maxc_fit-1) in the L2->L1
1853 ! reflux capture. Replicated inputs: collective-safe.
1854 if (any(amr_dim .and. amr_region_lo_all(:,l2) > amr_region_hi_all(:, &
1855 & l2))) call s_mpi_abort('amr static multi-level: level-1 block 1 is too small to nest a level-2 block (the fixed ' &
1856 & // 'inset inverts the box); enlarge the base amr block or reduce amr_cpat_mar')
1857 if (any(amr_dim .and. amr_ref_ratio*(amr_region_hi_all(:,l2) - amr_region_lo_all(:, &
1858 & l2) + 1) > amr_maxc_fit)) &
1859 & call s_mpi_abort('amr static multi-level: the nested level-2 block exceeds the per-rank scratch cap ' &
1860 & // '(2*L0-extent > amr_maxc_fit); static multi-level does not tile the level-2 block - use a smaller base amr ' &
1861 & // 'block or the dynamic regrid path (amr_regrid_int > 0)')
1862 amr_block_level(l2) = 2
1866 amr_cur = l2
1869 ! the parent-fill wave assembles the L2 patch from the parent's fine array (all ranks enter; the owner prolongs in place)
1870 call s_amr_parent_fill_exchange(2, .true.)
1871 call s_amr_select_slot(l2)
1872 if (amr_rank_owns_block) then
1873 call s_amr_parent_fill_consume(l2, .true.)
1875 end if
1877 call s_amr_select_slot(1)
1878
1879 end subroutine s_amr_build_static_multilevel
1880
1881 !> Sync level lev >= 2 into its parents: restrict its blocks into the covered parent cells, then state-reflux each into its
1882 !! parent (step-integral fluxes scaled by dt_reflux; 1 when registers carry dt).
1883 impure subroutine s_amr_sync_level(lev, dt_reflux)
1884
1885 integer, intent(in) :: lev
1886 real(wp), intent(in) :: dt_reflux
1887 integer :: k, io, ifc, ko, kf
1888
1889 call s_phase_tic(ph_restr)
1891 call s_phase_toc(ph_restr)
1892 ! s_amr_reflux_to_parent acts only on the parent's owner. The two block lists overlap and a duplicate visit
1893 ! would reflux twice, so walk their union with a two-cursor merge from the tails (descending block order: parent
1894 ! registers accumulate in that order).
1895 io = amr_n_own; ifc = amr_n_fch
1896 do
1897 do while (io >= 1)
1898 if (amr_block_level(amr_own_blk(io)) == lev) exit
1899 io = io - 1
1900 end do
1901 do while (ifc >= 1)
1902 if (amr_block_level(amr_fch_blk(ifc)) == lev) exit
1903 ifc = ifc - 1
1904 end do
1905 ko = 0; kf = 0
1906 if (io >= 1) ko = amr_own_blk(io)
1907 if (ifc >= 1) kf = amr_fch_blk(ifc)
1908 k = max(ko, kf)
1909 if (k == 0) exit
1910 if (ko == k) io = io - 1
1911 if (kf == k) ifc = ifc - 1
1912 call s_amr_select_slot(k)
1913 call s_phase_tic(ph_restr)
1914 call s_amr_reflux_to_parent(dt_reflux)
1915 call s_phase_toc(ph_restr)
1916 end do
1917
1918 end subroutine s_amr_sync_level
1919
1920 !> Lock-step restrict fold as plan-based waves : per level, finest first, the split-ownership child->parent folds
1921 !! (s_amr_restrict_parent_wave) then the per-box reflux-to-parent applies (freg already exchanged by s_amr_freg_wave); then the
1922 !! level-1 -> L0 scatter (s_amr_restrict_l1_wave). Both sides derive identical transfer lists from replicated metadata; one rank
1923 !! has no peers. Within a level the covered targets are disjoint.
1924 impure subroutine s_amr_restrict_wave(coarse_tgt, dt_reflux)
1925
1926 type(scalar_field), dimension(sys_size), intent(inout) :: coarse_tgt
1927 real(wp), intent(in) :: dt_reflux
1928 integer :: lev
1929
1930 call s_amr_refresh_lists()
1931 ! under subcycling the levels >= 2 were synced after each of their parent's substeps
1932 if (.not. amr_subcycle) then
1933 do lev = amr_max_level, 2, -1
1934 call s_amr_sync_level(lev, dt_reflux)
1935 end do
1936 end if
1937 call s_phase_tic(ph_restr)
1938 call s_amr_restrict_l1_wave(coarse_tgt)
1939 call s_phase_toc(ph_restr)
1940 call s_amr_select_slot(1)
1941
1942 end subroutine s_amr_restrict_wave
1943
1944 !> Restrict every level into its parent, finest first, with no reflux: for a cell-local update of the fine blocks made after the
1945 !! step's fold (phase-change relaxation).
1946 impure subroutine s_amr_restrict_fold(coarse_tgt)
1947
1948 type(scalar_field), dimension(sys_size), intent(inout) :: coarse_tgt
1949 integer :: lev
1950
1951 call s_amr_refresh_lists()
1952 call s_phase_tic(ph_restr)
1953 do lev = amr_max_level, 2, -1
1955 end do
1956 call s_amr_restrict_l1_wave(coarse_tgt)
1957 call s_phase_toc(ph_restr)
1958 call s_amr_select_slot(1)
1959
1960 end subroutine s_amr_restrict_fold
1961
1962 !> One per-level wave of the split-ownership level>=2 child->parent restrict folds. Co-located folds run inline on the child
1963 !! owner; cross-rank folds ship the parent-frame covered box, one message per (child-owner, parent-owner) peer. Both sides walk
1964 !! the same block list ascending with per-peer running offsets.
1965 impure subroutine s_amr_restrict_parent_wave(lev)
1966
1967 integer, intent(in) :: lev
1968 integer :: h_bl_1, h_bl_2, h_bl_3, h_ploc, h_cloc !< plain scalars for device routines (NVHPC -Minline)
1969 integer :: k, pblk, cowner, powner, rr, idx, lo, hi, kk
1970 integer :: plo(3), phi(3), bl(3), bh(3)
1971
1972 rr = amr_ref_ratio
1974 ! send side: owned level-lev blocks whose parent lives elsewhere ship their whole footprint to the parent's owner; a
1975 ! co-located parent is folded in place
1978 call s_amr_refresh_lists()
1979 do kk = 1, amr_n_my
1980 k = amr_my_blk(kk)
1981 if (amr_block_level(k) /= lev) cycle
1982 cowner = amr_block_owner(k)
1983 pblk = amr_parent_blk(k)
1984 powner = amr_block_owner(pblk)
1985 call s_amr_parent_foot(k, pblk, plo, phi)
1986 if (any(plo > phi)) cycle
1987 if (cowner == powner) then
1988 h_ploc = amr_loc_of(pblk); h_cloc = amr_loc_of(k)
1989 call s_amr_restrict_device_st(h_ploc, h_cloc, plo, phi, plo, rr, 0, 0, 0)
1990 cycle
1991 end if
1992 call s_amr_wave_add(amr_wsend, powner, k, plo, phi, sys_size*product(phi - plo + 1))
1993 end do
1995 ! receive side: level-lev children of my parents that another rank owns
1997 do kk = 1, amr_n_fch
1998 k = amr_fch_blk(kk)
1999 if (amr_block_level(k) /= lev) cycle
2000 pblk = amr_parent_blk(k)
2001 cowner = amr_block_owner(k); powner = amr_block_owner(pblk)
2002 if (cowner == powner .or. proc_rank /= powner) cycle
2003 call s_amr_parent_foot(k, pblk, plo, phi)
2004 if (any(plo > phi)) cycle
2005 call s_amr_wave_add(amr_wrecv, cowner, k, plo, phi, sys_size*product(phi - plo + 1))
2006 end do
2008 if (amr_wsend%np + amr_wrecv%np == 0) return
2010 do idx = 1, amr_wsend%nx
2011 call s_amr_wave_slice(amr_wsend, idx, lo, hi)
2012 bl = amr_wsend%bl(:,idx); bh = amr_wsend%bh(:,idx); h_cloc = amr_loc_of(amr_wsend%blk(idx))
2013 call s_amr_restrict_device_wire(amr_fw_sq(lo:hi), h_cloc, bl, bh, bl, rr)
2014 end do
2017 do idx = 1, amr_wrecv%nx
2018 call s_amr_wave_slice(amr_wrecv, idx, lo, hi)
2019 bl = amr_wrecv%bl(:,idx); bh = amr_wrecv%bh(:,idx)
2020 h_bl_1 = bl(1)
2021 h_bl_2 = bl(2)
2022 h_bl_3 = bl(3)
2023 h_ploc = amr_loc_of(amr_parent_blk(amr_wrecv%blk(idx)))
2024 call s_amr_pack_unpack_block_st(h_ploc, h_bl_1, h_bl_2, h_bl_3, bh(1) - h_bl_1, bh(2) - h_bl_2, bh(3) - h_bl_3, &
2025 & amr_fw_rq(lo:hi), .false.)
2026 end do
2027
2028 end subroutine s_amr_restrict_parent_wave
2029
2030 !> Level-1 -> L0 covered-cell scatter as one wave: every owned level-1 block's covered slabs ship to each listed coarse owner in
2031 !! one message per peer; the owner-local overwrite stays per block during the pack walk. The receiver plan (my interior x
2032 !! region(k) over level-1 blocks I do not own) equals the sender's list membership.
2033 impure subroutine s_amr_restrict_l1_wave(coarse_tgt)
2034
2035 type(scalar_field), dimension(sys_size), intent(inout) :: coarse_tgt
2036 integer :: k, owner, rr, idx, r, lo, hi, o1, o2, o3, cur, kk
2037 integer :: rlo(3), rhi(3), ilo(3), ihi(3), milo(3), mihi(3), bl(3), bh(3)
2038
2040 o1 = amr_sidx(1); o2 = amr_sidx(2); o3 = amr_sidx(3)
2041 ! block set changed: rebuild the cached overlap-rank lists (local, replicated)
2043 call s_amr_rank_interior(proc_rank, milo, mihi)
2044 ! send side: for every owned level-1 block, its covered slab inside each listed coarse owner's interior
2047 do kk = 1, amr_n_my
2048 k = amr_my_blk(kk)
2049 if (amr_block_level(k) /= 1) cycle
2050 call s_amr_region_box(k, rlo, rhi)
2051 do idx = 1, amr_ovl_scatter_n(k)
2052 r = amr_ovl_scatter(idx, k)
2053 if (r == proc_rank) cycle
2054 call s_amr_rank_interior(r, ilo, ihi)
2055 call s_amr_box_isect(rlo, rhi, ilo, ihi, bl, bh)
2056 if (any(bl > bh)) cycle
2057 call s_amr_wave_add(amr_wsend, r, k, bl, bh, sys_size*product(bh - bl + 1))
2058 end do
2059 end do
2061 ! receive side: every level-1 block I do not own that covers my interior, from its owner
2063 call s_amr_refresh_lists()
2064 do kk = 1, amr_n_l1r
2065 k = amr_l1r_blk(kk)
2066 owner = amr_block_owner(k)
2067 call s_amr_region_box(k, rlo, rhi)
2068 call s_amr_box_isect(rlo, rhi, milo, mihi, bl, bh)
2069 call s_amr_wave_add(amr_wrecv, owner, k, bl, bh, sys_size*product(bh - bl + 1))
2070 end do
2073 call s_amr_restrict_l1_batched(coarse_tgt, milo, mihi)
2074
2075 end subroutine s_amr_restrict_l1_wave
2076
2077 !> Rest of the level-1 restrict wave (receives posted): owned-block overwrite, send packs and received unpacks as one launch
2078 !! each over slab rows.
2079 impure subroutine s_amr_restrict_l1_batched(coarse_tgt, milo, mihi)
2080
2081 type(scalar_field), dimension(sys_size), intent(inout) :: coarse_tgt
2082 integer, intent(in) :: milo(3), mihi(3)
2083 integer :: k, kk, t, rlo(3), rhi(3), bl(3), bh(3)
2084
2086 do kk = 1, amr_n_my
2087 k = amr_my_blk(kk)
2088 if (amr_block_level(k) /= 1) cycle
2089 call s_amr_region_box(k, rlo, rhi)
2090 call s_amr_box_isect(rlo, rhi, milo, mihi, bl, bh)
2091 if (all(bl <= bh)) call s_amr_slab_add(bl, bh, amr_loc_of(k), rlo, [0, 0, 0])
2092 end do
2093 call s_amr_restrict_rows_sf(coarse_tgt, amr_sidx(1), amr_sidx(2), amr_sidx(3))
2094 do t = 1, amr_wsend%nx
2095 k = amr_wsend%blk(t)
2096 call s_amr_region_box(k, rlo, rhi)
2097 call s_amr_slab_add(amr_wsend%bl(:,t), amr_wsend%bh(:,t), amr_loc_of(k), rlo, [amr_wsend%off(t), 0, 0])
2098 end do
2099 if (amr_wsend%nx > 0) call s_amr_restrict_rows_wire(amr_fw_sq(1:amr_wsend%words))
2102 do t = 1, amr_wrecv%nx
2103 call s_amr_slab_add(amr_wrecv%bl(:,t), amr_wrecv%bh(:,t), amr_wrecv%off(t), [0, 0, 0], amr_sidx)
2104 end do
2105 if (amr_wrecv%nx > 0) call s_amr_unpack_rows(amr_fw_rq(1:amr_wrecv%words), coarse_tgt)
2106
2107 end subroutine s_amr_restrict_l1_batched
2108
2109 !> Sibling-seam face weights for level>=2 block kb under parent pblk: 0 on a face shared with a same-parent sibling tile
2110 !! (fine-fine; refluxing there double-writes), 1 otherwise. Uses replicated metadata only, and both the reflux apply and the
2111 !! freg wire-skip call it, so shipped and consumed faces agree.
2112 impure subroutine s_amr_sibling_face_weights(kb, pblk, w_lo, w_hi)
2113
2114 integer, intent(in) :: kb, pblk
2115 real(wp), intent(out) :: w_lo(3), w_hi(3)
2116 integer :: c, y, d
2117
2118 ! iterate pblk's cached children (same level by construction); do not refresh the lists here, a caller may be iterating them
2119
2120 w_lo = 1._wp; w_hi = 1._wp
2121 if (pblk <= 0) return ! orphan parent: no siblings; the CSR would index ptr(-1)
2122 do c = amr_child_ptr(pblk - 1) + 1, amr_child_ptr(pblk)
2123 y = amr_child_idx(c)
2124 if (y == kb) cycle
2125 do d = 1, num_dims
2126 if (f_amr_seam(kb, y, d)) w_hi(d) = 0._wp ! sibling above: shared high face
2127 if (f_amr_seam(y, kb, d)) w_lo(d) = 0._wp ! sibling below: shared low face
2128 end do
2129 end do
2130
2131 end subroutine s_amr_sibling_face_weights
2132
2133 !> Multi-level reflux: apply the Berger-Colella C/F flux correction from the current level>=2 block into its parent block's
2134 !! cells just outside the block footprint, in the parent-fine frame (mirror of the L0 s_amr_apply_reflux targeted at the parent;
2135 !! "the coarse" is level l-1). State form: q_parent(outside) += dt*(F_coarse - Fbar_fine)/dxf on the low face and +=
2136 !! dt*(Fbar_fine - F_coarse)/dxf on the high face, where Fbar_fine is the child-averaged fine register. creg/freg key off this
2137 !! block's slot. Per-face parent-fine dx (stretched-grid safe).
2138 !!
2139 !! The parent's owner applies: it holds the parent field and the parent-side creg (captured over its own advance). Only freg
2140 !! crosses the wire, and only when the two owners differ. Both participants must reach this routine or the P2P pair deadlocks
2141 !! (cf. the restrict).
2142 impure subroutine s_amr_reflux_to_parent(dt_reflux)
2143
2144 real(wp), intent(in) :: dt_reflux
2145 integer :: pblk, d, olo(3), ohi(3), glo(3), ghi(3), woff(3), plo(3), phi(3)
2146 real(wp) :: w_lo(3), w_hi(3), mlo(3), mhi(3)
2147 integer :: loc_pblk !< plain scalar for device routines (NVHPC -Minline)
2148
2149 call s_amr_refresh_lists() ! caches the parent (f_amr_parent_block is a global scan)
2150 pblk = amr_parent_blk(amr_cur)
2151 if (pblk <= 0) return ! orphan: no parent to reflux into
2152 if (amr_block_owner(pblk) /= proc_rank) return ! the parent's owner applies the correction
2153 ! A face shared with an adjacent same-parent sibling tile is fine-fine, not c/f: its outside parent cell is covered by
2154 ! the sibling's restrict, so skip it (weight 0). No siblings: all weights 1.
2155 call s_amr_sibling_face_weights(amr_cur, pblk, w_lo, w_hi)
2156 ! parent-fine frame for the shared reflux kernel: outside cell = isect boundary +/-1, creg-local loop range 0:extent,
2157 ! transverse write at the isect origin, per-face parent-fine dx. Footprint from s_amr_parent_foot (replicated), not
2158 ! amr_isect_lo/hi, which is empty on the parent's owner; rr is the global amr_ref_ratio since amr_slots(amr_cur) may be
2159 ! unallocated.
2160 call s_amr_parent_foot(amr_cur, pblk, plo, phi)
2161 olo = 0; ohi = 0; glo = 0; ghi = 0; woff = 0; mlo = 1._wp; mhi = 1._wp
2162 do d = 1, num_dims
2163 olo(d) = plo(d) - 1; ohi(d) = phi(d) + 1
2164 ghi(d) = phi(d) - plo(d)
2165 woff(d) = plo(d)
2166 end do
2167# 1077 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
2168 if (amr_dim(1)) then; mlo(1) = amr_slots(pblk)%dx(olo(1)); mhi(1) &
2169# 1078 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
2170 & = amr_slots(pblk)%dx(ohi(1)); end if
2171# 1079 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
2172# 1077 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
2173 if (amr_dim(2)) then; mlo(2) = amr_slots(pblk)%dy(olo(2)); mhi(2) &
2174# 1078 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
2175 & = amr_slots(pblk)%dy(ohi(2)); end if
2176# 1079 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
2177# 1077 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
2178 if (amr_dim(3)) then; mlo(3) = amr_slots(pblk)%dz(olo(3)); mhi(3) &
2179# 1078 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
2180 & = amr_slots(pblk)%dz(ohi(3)); end if
2181# 1079 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
2182# 1080 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
2183 loc_pblk = amr_loc_of(pblk)
2184 call s_amr_br_load_faces(loc_pblk, olo, ohi, glo, ghi, woff, w_lo, w_hi)
2185 call s_amr_reflux_apply_faces(amr_cons_br, amr_reg_cur, amr_ref_ratio, dt_reflux, olo, ohi, glo, ghi, woff, w_lo, w_hi, &
2186 & mlo, mhi)
2187 loc_pblk = amr_loc_of(pblk)
2188 call s_amr_br_store_faces(loc_pblk, olo, ohi, glo, ghi, woff, w_lo, w_hi)
2189
2190 end subroutine s_amr_reflux_to_parent
2191
2192 !> Device-native restriction of the fine block (flat store) to coarse averages over the covered cells [bl:bh] (global; block
2193 !! origin rlo, ratio rr); child-sum order ddk, ddj, ddi, /nchild. One body for two destinations so results match bit-for-bit: a
2194 !! parent block in the flat store (`_st`, local origin o) or the wire buffer buf (`_wire`; ci fastest, then cj, ck, i; wp).
2195 !! Never push the whole coarse array to the device.
2196# 1094 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
2197# 1095 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
2198 impure subroutine s_amr_restrict_device_st(ctloc, loc, bl, bh, rlo, rr, o1, o2, o3)
2199# 1096 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
2200
2201# 1098 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
2202 integer, intent(in) :: ctloc
2203# 1102 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
2204 integer, intent(in) :: loc, bl(3), bh(3), rlo(3), rr, o1, o2, o3
2205# 1103 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
2206 integer :: i, ci, cj, ck, fi0, fj0, fk0, ddi, ddj, ddk, dj_hi, dk_hi, nchild
2207 integer :: bl1, bl2, bl3, bh1, bh2, bh3, rl1, rl2, rl3, n1, n2, n3
2208 real(wp) :: acc
2209
2210 bl1 = bl(1); bl2 = bl(2); bl3 = bl(3); bh1 = bh(1); bh2 = bh(2); bh3 = bh(3)
2211 rl1 = rlo(1); rl2 = rlo(2); rl3 = rlo(3)
2212 n1 = bh1 - bl1 + 1; n2 = bh2 - bl2 + 1; n3 = bh3 - bl3 + 1
2213 dj_hi = merge(rr - 1, 0, amr_dim(2)); dk_hi = merge(rr - 1, 0, amr_dim(3)); nchild = rr**num_dims
2214# 1114 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
2215
2216# 1114 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
2217
2218# 1114 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
2219#if defined(MFC_OpenACC)
2220# 1114 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
2221!$acc parallel loop collapse(4) gang vector default(present) private(fi0, fj0, fk0, ddi, ddj, ddk, acc)
2222# 1114 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
2223#elif defined(MFC_OpenMP)
2224# 1114 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
2225
2226# 1114 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
2227
2228# 1114 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
2229
2230# 1114 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
2231!$omp target teams loop defaultmap(firstprivate:scalar) bind(teams,parallel) collapse(4) defaultmap(tofrom:aggregate) defaultmap(tofrom:allocatable) defaultmap(tofrom:pointer) &
2232# 1114 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
2233!$omp& private(fi0, fj0, fk0, ddi, ddj, ddk, acc)
2234# 1114 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
2235#endif
2236# 1115 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
2237# 1116 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
2238 do i = 1, sys_size
2239 do ck = bl3, bh3
2240 do cj = bl2, bh2
2241 do ci = bl1, bh1
2242 fi0 = (ci - rl1)*rr; fj0 = (cj - rl2)*rr; fk0 = (ck - rl3)*rr
2243 acc = 0._wp
2244 do ddk = 0, dk_hi
2245 do ddj = 0, dj_hi
2246 do ddi = 0, rr - 1
2247 acc = acc + real(amr_cons_st(fi0 + ddi, fj0 + ddj, fk0 + ddk, i, loc), wp)
2248 end do
2249 end do
2250 end do
2251# 1130 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
2252 amr_cons_st(ci - o1, cj - o2, ck - o3, i, ctloc) = real(acc/real(nchild, wp), stp)
2253# 1134 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
2254 end do
2255 end do
2256 end do
2257 end do
2258
2259# 1138 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
2260#if defined(MFC_OpenACC)
2261# 1138 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
2262!$acc end parallel loop
2263# 1138 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
2264#elif defined(MFC_OpenMP)
2265# 1138 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
2266
2267# 1138 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
2268!$omp end target teams loop
2269# 1138 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
2270#endif
2271# 1139 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
2272
2273 end subroutine s_amr_restrict_device_st
2274# 1141 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
2275# 1094 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
2276# 1095 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
2277 impure subroutine s_amr_restrict_device_wire(buf, loc, bl, bh, rlo, rr)
2278# 1096 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
2279
2280# 1100 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
2281 real(wp), intent(inout), contiguous :: buf(:)
2282# 1102 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
2283 integer, intent(in) :: loc, bl(3), bh(3), rlo(3), rr
2284# 1103 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
2285 integer :: i, ci, cj, ck, fi0, fj0, fk0, ddi, ddj, ddk, dj_hi, dk_hi, nchild
2286 integer :: bl1, bl2, bl3, bh1, bh2, bh3, rl1, rl2, rl3, n1, n2, n3
2287 real(wp) :: acc
2288
2289 bl1 = bl(1); bl2 = bl(2); bl3 = bl(3); bh1 = bh(1); bh2 = bh(2); bh3 = bh(3)
2290 rl1 = rlo(1); rl2 = rlo(2); rl3 = rlo(3)
2291 n1 = bh1 - bl1 + 1; n2 = bh2 - bl2 + 1; n3 = bh3 - bl3 + 1
2292 dj_hi = merge(rr - 1, 0, amr_dim(2)); dk_hi = merge(rr - 1, 0, amr_dim(3)); nchild = rr**num_dims
2293# 1112 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
2294
2295# 1112 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
2296
2297# 1112 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
2298#if defined(MFC_OpenACC)
2299# 1112 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
2300!$acc parallel loop collapse(4) gang vector default(present) private(fi0, fj0, fk0, ddi, ddj, ddk, acc) copyout(buf)
2301# 1112 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
2302#elif defined(MFC_OpenMP)
2303# 1112 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
2304
2305# 1112 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
2306
2307# 1112 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
2308
2309# 1112 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
2310!$omp target teams loop defaultmap(firstprivate:scalar) bind(teams,parallel) collapse(4) defaultmap(tofrom:aggregate) defaultmap(tofrom:allocatable) defaultmap(tofrom:pointer) &
2311# 1112 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
2312!$omp& private(fi0, fj0, fk0, ddi, ddj, ddk, acc) map(from:buf)
2313# 1112 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
2314#endif
2315# 1113 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
2316# 1116 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
2317 do i = 1, sys_size
2318 do ck = bl3, bh3
2319 do cj = bl2, bh2
2320 do ci = bl1, bh1
2321 fi0 = (ci - rl1)*rr; fj0 = (cj - rl2)*rr; fk0 = (ck - rl3)*rr
2322 acc = 0._wp
2323 do ddk = 0, dk_hi
2324 do ddj = 0, dj_hi
2325 do ddi = 0, rr - 1
2326 acc = acc + real(amr_cons_st(fi0 + ddi, fj0 + ddj, fk0 + ddk, i, loc), wp)
2327 end do
2328 end do
2329 end do
2330# 1132 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
2331 buf(1 + (ci - bl1) + n1*((cj - bl2) + n2*((ck - bl3) + n3*(i - 1)))) = acc/real(nchild, wp)
2332# 1134 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
2333 end do
2334 end do
2335 end do
2336 end do
2337
2338# 1138 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
2339#if defined(MFC_OpenACC)
2340# 1138 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
2341!$acc end parallel loop
2342# 1138 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
2343#elif defined(MFC_OpenMP)
2344# 1138 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
2345
2346# 1138 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
2347!$omp end target teams loop
2348# 1138 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
2349#endif
2350# 1139 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
2351
2352 end subroutine s_amr_restrict_device_wire
2353# 1141 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
2354# 1142 "/home/runner/work/MFC/MFC/src/simulation/m_amr_transfer.fpp"
2355end module m_amr_transfer
integer, intent(in) k
integer, intent(in) j
Rank decomposition, box arithmetic and block ownership (SFC cut, owner lookups, owned-block lists).
impure subroutine, public s_amr_refresh_my_blocks()
Refresh the owned-block list if any owner write has happened since the last refresh....
impure subroutine, public s_amr_refresh_lists()
Rebuild the epoch-keyed block lists when the mesh epoch has moved (one O(global blocks) walk per regr...
pure subroutine, public s_amr_rank_decomp(r, sidx, ext)
Rank r's coarse-grid decomposition (start_idx + local extent m/n/p), computed O(1) from its cartesian...
pure subroutine, public s_amr_rank_interior(r, ilo, ihi)
Rank r's coarse interior box (global, no ghosts) from s_amr_rank_decomp; restriction targets are iden...
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 =...
pure subroutine, public s_amr_box_isect(alo, ahi, blo, bhi, olo, ohi)
Per-dim intersection of two global boxes [alo:ahi] and [blo:bhi] -> [olo:ohi] (empty when olo > ohi i...
Parent/child data exchange: gather plans, pack/unpack and the fill waves.
impure subroutine, public s_amr_l1_fill_consume(q_cons_coarse, k, full)
Level-1 fill wave, consume for owned box k: set the patch frame, copy the own slice,...
impure subroutine, public s_amr_restrict_rows_sf(coarse_tgt, o1, o2, o3)
Restriction over the queued slab rows (row 9 store column, 10:12 region lo, 13 wire payload offset): ...
impure subroutine, public s_amr_pack_unpack_block_st(loc, o1, o2, o3, fm1, fm2, fm3, buf, to_buf)
Device pack (to_buf=T) / unpack (F) of the interior block [o+0:o+fm] of the flat store at column loc ...
impure subroutine, public s_amr_unpack_rows(buf, cg)
Unpack the queued slab rows (row 9: payload offset in buf) into the pooled amr_cg patches: a whole fi...
impure subroutine, public s_amr_parent_fill_exchange(lev, full)
Level-lev parent-fill wave, exchange phase: each block's parent patch from its parent's fine array (p...
impure subroutine, public s_amr_l1_fill_exchange(q_cons_coarse, full)
Level-1 fill wave, exchange phase: each level-1 box's padded coarse patch (region +/- amr_cpat_mar) i...
impure subroutine, public s_amr_restrict_rows_wire(buf)
impure subroutine, public s_amr_fill_wave_done()
Every receive transfer of the open fill wave has been consumed.
impure subroutine, public s_amr_parent_fill_consume(k, full)
The parent-fill wave, consume phase for owned box k (the current slot): set the patch frame,...
Block-frame swap (fine grid state in/out of the shared solver).
Coarse-to-fine ghost fill: slab table, limited-linear or quartic fill, quartic positivity guard,...
impure subroutine, public s_amr_slab_add(lo, hi, loc, ilo, coff)
Append slab [lo, hi] to amr_slab_tab (rows: 1:3 lo, 4:6 hi, 7 flat offset, 8 cells,...
AMR flux registers: per-RK-stage refluxing at the coarse/fine block boundary. Depends only on m_deriv...
impure subroutine, public s_amr_reg_prepare()
Build the dense participation-local register index (amr_reg_of/amr_reg_n) and size and zero the regis...
pure logical function, public f_amr_block_near(k, sidx, ext)
False only if block k's region widened by its outside face layer misses the coarse subdomain [sidx,...
pure subroutine, public s_amr_reflux_faces(sidx, ext, clip, s_lo, s_hi)
Faces of the current block whose outside coarse layer lies in the coarse subdomain [sidx,...
impure subroutine, public s_amr_reflux_apply_faces(q, islot, rr, dtl, olo, ohi, glo, ghi, woff, w_lo, w_hi, mlo, mhi)
Berger-Colella state reflux: q(outside) += w*dtl*(F_coarse - Fbar_fine)/m on the low face and the neg...
type(t_face_reg), dimension(3), public freg
fine flux at covering fine faces (0-based fine transverse)
pure subroutine, public s_amr_parent_foot(k, pblk, plo, phi)
Parent-fine footprint of block k inside its parent pblk from replicated metadata, so every rank gets ...
Same-level fine-fine seams and the coarse conservative halo.
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...
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 amr_cpat_mar
coarse-cell stencil reach (f_amr_cpat_mar)
integer, dimension(:), allocatable amr_ovl_scatter_n
per-block list lengths
integer, parameter amr_band_freg
logical amr_myblk_dirty
Set wherever amr_block_owner is written (assigner, inheritance, restart/migration),...
impure subroutine s_amr_build_block_coords(k, gcb, glb, fcb, fcc, fdx, d)
Fine cell coordinates of block k in dimension d, rebuilt from the global coarse boundaries gcb by rep...
real(wp), dimension(:), allocatable amr_gycb
integer, dimension(:), allocatable amr_child_ptr
children of p = amr_child_idx(amr_child_ptr(p-1)+1 : amr_child_ptr(p)), ascending
real(wp), dimension(:), allocatable amr_fw_sq
Wire pools of the pooled waves (m_amr_wave lays them out; contents never survive a wave)....
integer, parameter amr_band_restrict_l1
integer, parameter amr_band_restrict_parent
integer, parameter amr_band_reflux
Keyed wave tags: amr_tag_base + band*65536 + gen*4096 + seq, checked against MPI_TAG_UB....
integer, dimension(:), allocatable amr_child_idx
(amr_child_ptr(0) = 0)
type(t_level), dimension(:), allocatable amr_slots
Fixed pool of refined-block slots; the working slot amr_cur selects which one per-block routines oper...
integer, dimension(:), allocatable amr_fch_blk
foreign children of my owned parents (level >= 2, powner == me, cowner /= me)
real(wp), dimension(:), allocatable amr_rf_rq
device-resident pools of the packed reflux-faces wave
integer, dimension(:), allocatable amr_parent_blk
cached f_amr_parent_block (0 for level <= 1), avoiding an O(global blocks) scan per block per stage
type(scalar_field), dimension(:), allocatable amr_cons_br
Copy bridge to the shared solver, whose routines take type(scalar_field) and serve the monolithic pat...
pure integer function f_amr_parent_block(k)
Index of the level-(level(k)-1) block that block k refines, or 0 for level 1 (parent is the L0 grid)....
pure real(wp) function f_amr_quartic_half(q, xi)
Average over one half (xi < 0 left, > 0 right; |xi| = 1/4) of the middle of five cell averages q: the...
integer amr_seam_pairs_nblk
type(scalar_field), dimension(:), allocatable amr_cg
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...
pure subroutine s_amr_region_box(k, rlo, rhi)
Block k's region box in global coarse cells; collapsed dimensions read 0:0.
logical amr_fw_dev
Device-resident wire pools: with rdma_mpi MPI sends/receives them by device address,...
pure real(wp) function f_amr_lin(u0, xm, xp, ym, yp, zm, zp, xix, xiy, xiz)
Minmod-limited linear value at fine offset (xix, xiy, xiz) in coarse cell u0, from its face neighbour...
real(wp), dimension(:), allocatable amr_gxcb
Global coarse cell-boundary arrays (-1:X_glb), assembled once at init; the fine-distribution owner re...
real(wp), dimension(:), allocatable amr_rf_sq
integer, dimension(:), allocatable amr_fw_rblk
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
pure logical function f_amr_cons_ok(fi, fj, fk, loc)
Admissible fine cell: positive density and (except model 4) positive internal energy; a NaN state is ...
integer, dimension(:), allocatable amr_own_blk
blocks this rank intersects (amr_owns_all, the multi-owner notion, all levels)
real(wp), dimension(:), allocatable amr_fw_rq
integer, dimension(:), allocatable amr_l1r_blk
Level-1 blocks this rank receives restriction for (not its own, overlapping its interior),...
real(wp), dimension(:), allocatable amr_gzcb
integer, dimension(:), allocatable amr_l1p_blk
padded variant (region +/- amr_cpat_mar vs my coarse range): stage-fill gather
integer, dimension(:), allocatable amr_my_blk
Block indices this rank owns, ascending (the paired MPI_SENDRECVs depend on the order); built in s_am...
integer, dimension(3) amr_maxc_fit
Regrid box size cap per dim (fixed for the run, identical on all ranks; 1 in collapsed dims): a box o...
integer, dimension(3) amr_cpat_off
global coarse index of amr_cg local cell 0 (region_lo - amr_cpat_mar)
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...
logical amr_xchg_coarse_ghosts
True (identically on all ranks) iff some rank's fine ghost fill reads its coarse ghost cells: the sol...
integer amr_xchg_bad
local accumulator behind amr_xchg_coarse_ghosts; s_amr_reduce_xchg_flag does one allreduce per scan
Flat per-block field store: dense slot indexing, reservation, alloc/free and the prim/bridge loaders.
impure subroutine, public s_amr_reconcile_slots()
Reconcile allocated slots to current ownership: allocate every active block this rank owns,...
impure subroutine, public s_amr_br_store_faces(loc, olo, ohi, glo, ghi, woff, w_lo, w_hi)
impure subroutine, public s_amr_br_load_faces(loc, olo, ohi, glo, ghi, woff, w_lo, w_hi)
Face-bounded bridge move for the level>=2 reflux apply: moves only the six one-cell planes (outside c...
Prolongation, restriction and Berger-Colella reflux, including the restrict/reflux/freg waves.
impure subroutine, public s_interpolate_coarse_to_fine()
Conservative prolongation: fill amr_fine interior from coarse (minmod-limited linear,...
pure subroutine s_amr_reflux_faces_for(r, s_lo, s_hi)
Faces of the current block that rank r applies (seam-clipped); sender and receiver derive the identic...
impure subroutine, public s_amr_reflux_to_parent(dt_reflux)
Multi-level reflux: apply the Berger-Colella C/F flux correction from the current level>=2 block into...
impure subroutine, public s_set_amr_fine_geometry(lo, hi)
Set the fine level's geometry (region, intersection, extents, bounds, coordinates) for the box lo:hi;...
impure subroutine s_amr_freg_pack_d3(w, pack)
s_amr_freg_pack for the faces normal to direction 3, both sides in one kernel.
impure subroutine s_amr_freg_faces_poison(f_lo, f_hi)
Debug builds: NaN-poison the unflagged face registers of the current slot before a receive lands,...
impure subroutine, public s_amr_freg_wave()
Split-ownership level>=2 freg exchange as one wave, run once before the reflux fold; zero-copy like t...
impure subroutine s_prolong_guard(qc, loc)
Quartic safety net (amr_prolong = 2): a parent whose children leave the admissible state (non-positiv...
impure subroutine, public s_amr_restrict_wave(coarse_tgt, dt_reflux)
Lock-step restrict fold as plan-based waves : per level, finest first, the split-ownership child->par...
impure subroutine s_amr_face_xfers(w, peer, k, f_lo, f_hi)
Append block k's faces f_lo/f_hi with peer as transfers (bl = dim, side 1 low / 2 high,...
impure subroutine, public s_amr_reduce_xchg_flag()
Close a geometry scan: one allreduce of the accumulated flag, then reset. Call after every s_set_amr_...
impure subroutine s_prolong_alphas_closure(qc, loc)
Sum-preserving volume-fraction prolongation (num_fluids > 1): fluids advbeg..advend-1 are interpolate...
impure subroutine, public s_populate_amr_fine(q_cons_base)
Disblock prolongation. Guard: no-op unless amr.
impure subroutine s_amr_restrict_l1_batched(coarse_tgt, milo, mihi)
Rest of the level-1 restrict wave (receives posted): owned-block overwrite, send packs and received u...
impure subroutine s_prolong_species_closure(qc, loc)
Species prolongation closure (chemistry): each rho*Y_k is minmod-prolonged and clamped non-negative,...
impure subroutine s_amr_restrict_parent_wave(lev)
One per-level wave of the split-ownership level>=2 child->parent restrict folds. Co-located folds run...
impure subroutine s_amr_restrict_device_wire(buf, loc, bl, bh, rlo, rr)
impure subroutine s_amr_freg_pack_d1(w, pack)
s_amr_freg_pack for the faces normal to direction 1, both sides in one kernel.
impure subroutine s_amr_restrict_l1_wave(coarse_tgt)
Level-1 -> L0 covered-cell scatter as one wave: every owned level-1 block's covered slabs ship to eac...
impure subroutine s_prolong_one_var(qc, loc, ivar)
Conservative prolongation of one variable from the patch qc into the fine interior: minmod-limited li...
impure subroutine s_amr_build_cfl_map()
amr_cfl_rf from the block table: amr_ref_ratio**level of the deepest block over each of this rank's c...
impure subroutine s_amr_freg_pack(w, pack)
Copy a side's freg faces into (pack) or out of the device pool, one kernel per face direction....
impure subroutine s_amr_freg_pack_d2(w, pack)
s_amr_freg_pack for the faces normal to direction 2, both sides in one kernel.
impure subroutine s_amr_freg_faces_req(dir, peer, f_lo, f_hi)
Post one zero-copy request per flagged face register of the current slot (block k): dir 1 sends to / ...
impure subroutine s_amr_build_static_multilevel()
Build the static multi-level hierarchy (amr_regrid_int = 0): nest one level-2 block inside level-1 bl...
impure subroutine s_amr_freg_faces_update(to_device, f_lo, f_hi)
Move the flagged face registers of the current slot to the device (to_device) or to the host.
impure subroutine s_amr_sibling_face_weights(kb, pblk, w_lo, w_hi)
Sibling-seam face weights for level>=2 block kb under parent pblk: 0 on a face shared with a same-par...
impure subroutine s_amr_restrict_device_st(ctloc, loc, bl, bh, rlo, rr, o1, o2, o3)
Device-native restriction of the fine block (flat store) to coarse averages over the covered cells [b...
impure subroutine, public s_amr_sync_level(lev, dt_reflux)
Sync level lev >= 2 into its parents: restrict its blocks into the covered parent cells,...
impure subroutine s_amr_check_block_geometry(k, pblk, fext)
Abort unless block k's fine extent is amr_ref_ratio**level*(region width) - 1 and,...
pure logical function f_amr_reflux_participates(r)
True iff rank r refluxes some face of the current block (its unclipped subdomain); the participation ...
impure subroutine, public s_amr_restrict_fold(coarse_tgt)
Restrict every level into its parent, finest first, with no reflux: for a cell-local update of the fi...
impure subroutine, public s_amr_reflux_faces_wave()
Per-stage level-1 reflux-face exchange as one wave: faces are packed on the owner's device into one p...
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).
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.
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(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 ...
impure subroutine, public s_amr_wave_wait(w)
Wait for all requests; under MFC_DEBUG each receive length must equal its plan.
impure subroutine, public s_amr_wave_open(w, b)
Open a wave on tag band b.
type(t_amr_wave), target, public amr_wave
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
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_isect_lo_all
integer, dimension(3) amr_sidx
integer, dimension(3) amr_region_lo
Current AMR fine-block box in level-0 cell indices (kept by s_set_amr_fine_geometry); lives here so m...
real(wp), dimension(:,:,:), allocatable amr_cfl_rf
Spacing ratio of the finest cell over each coarse cell (1 where uncovered), and the powers of it by w...
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_reg_of
Participation-local flux-register index (m_amr_registers): global block slot -> dense register slot,...
integer, dimension(:,:), allocatable amr_isect_hi_all
integer, dimension(3) amr_isect_lo
The block's coarse footprint driving the coarse<->fine gather/scatter, per dim (collapsed dims 0:0): ...
integer, dimension(:,:), allocatable amr_region_hi_all
logical, dimension(:), allocatable amr_owns_all
integer, dimension(3) amr_region_hi
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(3) amr_isect_hi
integer num_procs
Number of processors.
integer, dimension(:), allocatable amr_block_owner
Fine-level distribution map: SFC/work-balanced single-owner rank per active block,...
real(wp), dimension(3) amr_cfl_pow
integer amr_num_blocks
Number of active AMR fine-block slots (>= 1) and the working slot index selecting which one the per-b...
integer, dimension(3) amr_ext
integer amr_reg_cur
Dense register slot of the working block (amr_reg_of(amr_cur), 0 if unmapped); kept by s_amr_select_s...
logical amr_rank_owns_block
true on the current block's single owner rank: amr_block_owner(amr_cur) == proc_rank; kept by s_set_a...
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.
impure subroutine s_mpi_abort(prnt, code)
The subroutine terminates the MPI execution environment.
impure subroutine s_mpi_allreduce_integer_max(var_loc, var_glb)
Reduce a local integer value to its global maximum across all MPI ranks.
MPI halo exchange, domain decomposition, and buffer packing/unpacking for the simulation solver.
Per-phase wall-time budget of the AMR step, reported when rank_time_wrt is on. Brackets are disjoint ...
integer, parameter, public ph_restr
post-stage restrict + reflux-to-parent fold
real(wp), dimension(ph_n) acc
impure subroutine, public s_phase_tic(id)
impure subroutine, public s_phase_toc(id)
One side (send or receive) of a pooled wave.
Derived type annexing a scalar field (SF).