FAdaptiveTask.hpp 52.8 KB
Newer Older
Quentin Khan's avatar
Quentin Khan committed
1
2
3
4
5
6
7
8
#ifndef SCALFMM_ADAPTIVE_TASK_ALGO_HPP_
#define SCALFMM_ADAPTIVE_TASK_ALGO_HPP_

#include <algorithm>
#include <numeric>
#include <cmath> // Used to round box differences
#include <functional>
#include <map>
9
#include <vector>
Quentin Khan's avatar
Quentin Khan committed
10
11
12
13
14
15
16
17
18
19
20
21
#include <list>
#include <array>
#include <type_traits>
#include <memory>
#include <sstream>

#include <omp.h>

#include <unistd.h>

#include "Core/FCoreCommon.hpp"
#include "Containers/FTreeCoordinate.hpp"
22
#include "Utils/FAlgorithmTimers.hpp"
Quentin Khan's avatar
Quentin Khan committed
23

24
25
26
#include "Kernels/FKernelConcepts.hpp"

template<class _Tree, class _Kernel,
27
28
29
30
31
32
33
34
35
         typename std::enable_if<scalfmm::sfinae::has_P2M<_Tree, _Kernel>::value>::type* = nullptr,
         typename std::enable_if<scalfmm::sfinae::has_M2M<_Tree, _Kernel>::value>::type* = nullptr,
         typename std::enable_if<scalfmm::sfinae::has_M2L<_Tree, _Kernel>::value>::type* = nullptr,
         typename std::enable_if<scalfmm::sfinae::has_L2L<_Tree, _Kernel>::value>::type* = nullptr,
         typename std::enable_if<scalfmm::sfinae::has_L2P<_Tree, _Kernel>::value>::type* = nullptr,
         typename std::enable_if<scalfmm::sfinae::has_partial_P2P<_Tree, _Kernel>::value>::type* = nullptr,
         typename std::enable_if<scalfmm::sfinae::has_M2P<_Tree, _Kernel>::value>::type* = nullptr,
         typename std::enable_if<scalfmm::sfinae::has_P2L<_Tree, _Kernel>::value>::type* = nullptr
         >
36
class FAdaptiveTask : public FAlgorithmInterface, public FAlgorithmTimers {
Quentin Khan's avatar
Quentin Khan committed
37
38
39
40
41
42
43
44
public:
    using tree_t = _Tree;
    using kernel_t = _Kernel;
    using FReal = typename tree_t::FReal;

    using node_t = typename tree_t::node_t;

protected:
45
    /// Tree
Quentin Khan's avatar
Quentin Khan committed
46
    tree_t&   _tree;
47
    /// Vector of kernels, one per thread
Quentin Khan's avatar
Quentin Khan committed
48
49
    std::vector<std::unique_ptr<kernel_t>> _kernels;

50
51

    /** \brief SFINAE shortcut to check `K.setup(Args...)` existence */
Quentin Khan's avatar
Quentin Khan committed
52
    template<typename K, typename Ret, typename... Args>
53
    using has_setup_member = typename std::enable_if<
Quentin Khan's avatar
Quentin Khan committed
54
55
56
        std::is_same<Ret, decltype(std::declval<K>().setup(std::declval<Args>()...))>::value
        >::type*;

57
58
59
60
    /** \brief Calls K.setup(tree_t) if it exists, otherwise deleted by SFINAE */
    template<typename K, has_setup_member<K, void, tree_t> = nullptr>
    void setup_kernel(K* kernel) {
        kernel->setup(_tree);
61
62
    }

63
64
65
66
    /** \brief Calls K.setup() if it exists, otherwise deleted by SFINAE */
    template<typename K, has_setup_member<K, void> = nullptr>
    void setup_kernel(K* kernel) {
        kernel->setup();
Quentin Khan's avatar
Quentin Khan committed
67
68
    }

69
    /** \brief Fallback if both other overloads are deleted by SFINAE */
Quentin Khan's avatar
Quentin Khan committed
70
71
    void setup_kernel(...) {}

72
73
74
75
76
77
78
79
    /**
     * \brief Data pool for mock dependency
     *
     * Generates addresses that can be used as OpenMP task dependencies. To do
     * so, a list of char arrays (buckets) is created. Each char is a potential
     * dependency. When an array is exhausted, an new one is appended to the
     * list.
     */
Quentin Khan's avatar
Quentin Khan committed
80
    struct mock_dependency {
81
82
83
        /// Size of a bucket
        enum : std::size_t {bucket_size = 512};
        /// List of buckets. A bucket is implented as an array
84
        std::list<std::array<char, bucket_size>> pool;
85
        /// Index of the next dependency to hand out in the current bucket
Quentin Khan's avatar
Quentin Khan committed
86
87
        std::size_t mock_dependency_index = bucket_size;

88
        /// Yield a new dependency
89
        char* next() {
90
91
92
93
94
            // When the bucket is empty, add a new one
            if(bucket_size == mock_dependency_index) {
                pool.emplace_back();
                mock_dependency_index = 0;
            }
95
            return (char*) (&(this->pool.back()[mock_dependency_index++]));
Quentin Khan's avatar
Quentin Khan committed
96
97
98
        }
    } mock_dep;

99
100
101
102
103
104
105
106
107
108
    /**
     * \brief Dependency type names structure
     *
     * Each member of the enum identifies a dependency type name.
     */
    enum class dep_t {
        M,   ///< Multipole
        L,   ///< Local expansion
        P_s, ///< Source particle
        P_t, ///< Target particle
109
110
    };

111

112
113
    /**
     * \brief Get pointer to node ressource handling given dependency type
114
115
116
117
118
119
120
121
122
123
124
125
126
127
     *
     * OpenMP task dependencies do not need to be correlated with real data,
     * they only need to point to the same area in memory. This method maps a
     * dependency type to an address inside the given node object, cast to
     * `const char`.
     *
     * \tparam I Dependency type
     *
     * \param node Node to which the dependency applies
     *
     * \return A pointer to memory inside the node.
     */
    template<dep_t I>
    const char* get_dependency(node_t* node) const {
128
        if(I == dep_t::M) {
129
            return (const char*) node->getData();
130
        } else if(I == dep_t::L) {
131
132
            using type = typename std::decay<decltype((node->getDepth()))>::type;
            return (const char*) const_cast<type*>(&(node->getDepth()));
133
        } else if(I == dep_t::P_s) {
134
            return (const char*) node;
135
        } else if(I == dep_t::P_t) {
136
            return (const char*) node->getParticleContainer();
137
        }
138
        return nullptr;
139
140
141
    }


Quentin Khan's avatar
Quentin Khan committed
142
143
public:

Quentin Khan's avatar
Quentin Khan committed
144
145
146
147
148
149
    /**
     * \brief Constructor
     *
     * \param tree The tree to run over
     * \param kernel The kernel to use
     */
Quentin Khan's avatar
Quentin Khan committed
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
    FAdaptiveTask(tree_t& tree, kernel_t& kernel) :
        _tree(tree)
    {
        // Place kernel objects near their threads in memory to avoid NUMA
        // latency
        #pragma omp parallel
        {
            for(int i = 0; i < omp_get_num_threads(); ++i) {
                if(i == omp_get_thread_num()) {
                    this->_kernels.emplace_back(new kernel_t(kernel));
                }
                #pragma omp barrier
            }
        }
    }

Quentin Khan's avatar
Quentin Khan committed
166
167
168
169
170
171
    /**
     * \brief Constructor
     *
     * \param tree The tree to run over
     * \param kernel The kernel to use
     */
Quentin Khan's avatar
Quentin Khan committed
172
173
174
175
    FAdaptiveTask(tree_t& tree, kernel_t* kernel) :
        FAdaptiveTask(tree, *kernel)
    {}

Quentin Khan's avatar
Quentin Khan committed
176
177
178
179
180
181
    /**
     * \brief Constructor
     *
     * \param tree The tree to run over
     * \param kernel The kernel to use
     */
Quentin Khan's avatar
Quentin Khan committed
182
183
184
185
    FAdaptiveTask(tree_t* tree, kernel_t& kernel) :
        FAdaptiveTask(*tree, kernel)
    {}

Quentin Khan's avatar
Quentin Khan committed
186
187
188
189
190
191
    /**
     * \brief Constructor
     *
     * \param tree The tree to run over
     * \param kernel The kernel to use
     */
Quentin Khan's avatar
Quentin Khan committed
192
193
194
195
196
    FAdaptiveTask(tree_t* tree, kernel_t* kernel) :
        FAdaptiveTask(*tree, *kernel)
    {}


197
198
    std::string name() const override {
        return "Task adaptive algorithm";
Quentin Khan's avatar
Quentin Khan committed
199
200
    }

201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
    std::string description() const override {
        int threads = 1;
        #pragma omp parallel shared(threads)
        {
            #pragma omp single nowait
            {
                threads = omp_get_max_threads();
            }
        }
        return std::string("threads: ") + std::to_string(threads);
    }

    using FAlgorithmInterface::execute;

    /**
     * \brief Run specific steps of the algorithm
     *
     * \param operations Specifies the algorithm operations to run, see
     * FFmmOperations.
     */
    void execute(const unsigned int operations) override {
Quentin Khan's avatar
Quentin Khan committed
222
223
224
        this->run(operations);
    }

225
    void run(int operations) {
Quentin Khan's avatar
Quentin Khan committed
226
227
228

        #pragma omp parallel
        {
229
            this->setup_kernel(this->_kernels[omp_get_thread_num()].get());
230
231
            #pragma omp barrier

Quentin Khan's avatar
Quentin Khan committed
232
233
234
235
            #pragma omp single nowait
            {
                if(operations & FFmmP2M) {
                    // 1. source to up, P2M
236
237
238
239
                    Timers["P2M"].tic();
                    this->source_to_up();;
                    Timers["P2M"].tac();
                    std::cout << "    P2M: " << Timers["P2M"].elapsed() << std::endl;
240
241
242
243
                }

                if(operations & FFmmP2L) {
                    // 3b X-list, P2L
244
245
246
247
                    Timers["P2L"].tic();
                    this->x_list_step();
                    Timers["P2L"].tac();
                    std::cout << "    P2L: " << Timers["P2L"].elapsed() << std::endl;
Quentin Khan's avatar
Quentin Khan committed
248
249
                }

250
251
                if(operations & FFmmP2P) {
                    // A. U-list, P2P
252
253
254
255
                    Timers["P2P"].tic();
                    this->u_list_step();
                    Timers["P2P"].tac();
                    std::cout << "    P2P: " << Timers["P2P"].elapsed() << std::endl;
256
257
                }

Quentin Khan's avatar
Quentin Khan committed
258
259
                if(operations & FFmmM2M) {
                    // 2. up to up, M2M
260
261
262
263
                    Timers["M2M"].tic();
                    this->up_to_up();
                    Timers["M2M"].tac();
                    std::cout << "    M2M: " << Timers["M2M"].elapsed() << std::endl;
Quentin Khan's avatar
Quentin Khan committed
264
265
                }

266
267
                if(operations & FFmmM2P) {
                    // 5a W-list, M2P
268
269
270
271
                    Timers["M2P"].tic();
                    this->w_list_step();
                    Timers["M2P"].tac();
                    std::cout << "    M2P: " << Timers["M2P"].elapsed() << std::endl;
272
273
                }

274
275
                if(operations & FFmmM2L) {
                    // 3a V-list, M2L
276
277
278
279
                    Timers["M2L"].tic();
                    this->v_list_step();
                    Timers["M2L"].tac();
                    std::cout << "    M2L: " << Timers["M2L"].elapsed() << std::endl;
Quentin Khan's avatar
Quentin Khan committed
280
281
282
283
                }

                if(operations & FFmmL2L) {
                    // 4. down to down, L2L
284
285
286
287
                    Timers["L2L"].tic();
                    this->down_to_down();
                    Timers["L2L"].tac();
                    std::cout << "    L2L: " << Timers["L2L"].elapsed() << std::endl;
Quentin Khan's avatar
Quentin Khan committed
288
289
290
                }

                if(operations & FFmmL2P) {
291
                    // 5b down to target, L2P
292
293
294
295
                    Timers["L2P"].tic();
                    this->down_to_target();
                    Timers["L2P"].tac();
                    std::cout << "    L2P: " << Timers["L2P"].elapsed() << std::endl;
Quentin Khan's avatar
Quentin Khan committed
296
297
                }

298
299


300
                std::cout << "    tasks creation: "
301
302
303
                          << std::accumulate(Timers.begin(), Timers.end(), 0.,
                                             [](double & res, const typename decltype(Timers)::value_type & t)
                                             {return res + t.second.cumulated();})
Quentin Khan's avatar
Quentin Khan committed
304
305
306
307
308
                          << '\n';
            }
        }
    }

309
    /** \brief Walk through leaves and queue P2M tasks */
Quentin Khan's avatar
Quentin Khan committed
310
311
    void source_to_up() {
        for(node_t* leaf : _tree.leaves()) {
312
313
314
315
316
317
318
319
320
321
322
            this->create_P2M_task(leaf);
        }
        // #pragma omp taskwait
    }

    /**
     * \brief Create and queue a P2M task
     *
     * \param leaf The P2M leaf
     */
    void create_P2M_task(node_t* leaf) {
323
324
            const char* ps_dep = get_dependency<dep_t::P_s>(leaf);(void)ps_dep;
            const char* m_dep = get_dependency<dep_t::M>(leaf);(void)m_dep;
325
326
327
328

            #pragma omp task firstprivate(leaf)                        \
                depend(in: ps_dep[:1])                                 \
                depend(inout: m_dep[:1])
Quentin Khan's avatar
Quentin Khan committed
329
330
            {
                const int thread_num = omp_get_thread_num();
331
                _kernels[thread_num]->P2M(leaf->getData(), leaf->getParticleContainer());
Quentin Khan's avatar
Quentin Khan committed
332
            }
333
334
335
336
337
338
339
340
341
    }


    /** \brief Walk through tree and queue M2M tasks */
    void up_to_up() {
        for(node_t& n : _tree.post_order_walk()) {
            if(! n.is_leaf()) {
                create_M2M_task(&n);
            }
Quentin Khan's avatar
Quentin Khan committed
342
        }
343
        // #pragma omp taskwait
Quentin Khan's avatar
Quentin Khan committed
344
345
    }

346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
    /**
     * \brief Create and queue an M2M task
     *
     * \param node An internal node
     *
     * \warning node is assumed not to be a leaf.
     */
    void create_M2M_task(node_t* node) {
        assert(! node->is_leaf());
        // Setup task dependencies
        // children data
        const char* children_dep[node_t::child_count] = {};
        for(node_t* child : node->getChildren()) {
            children_dep[child->getIndex() & (node_t::child_count-1)]
                = get_dependency<dep_t::M>(child);
        }
        // node data
        const char* parent_dep = get_dependency<dep_t::M>(node); (void) parent_dep;

        #pragma omp task                        \
            depend(in:                          \
                   children_dep[0][:1],         \
                   children_dep[1][:1],         \
                   children_dep[2][:1],         \
                   children_dep[3][:1],         \
                   children_dep[4][:1],         \
                   children_dep[5][:1],         \
                   children_dep[6][:1],         \
                   children_dep[7][:1])         \
            depend(out: parent_dep[:1])
        {
            const int thread_num = omp_get_thread_num();
Quentin Khan's avatar
Quentin Khan committed
378

379
380
381
382
383
384
            // Fill array of children data
            std::array<typename node_t::data_t*, node_t::child_count> child_data;
            for(node_t* child : node->getChildren()) {
                // Last bits of child index give the index relative to parent
                child_data[child->getIndex() & (node_t::child_count-1)] = child->getData();
            }
Quentin Khan's avatar
Quentin Khan committed
385

386
387
388
389
390
391
392
393
394
395
396
397
398
            // Call kernel module
            this->_kernels[thread_num]->M2M(node->getData(), child_data.data(),
                                            static_cast<int>(node->getDepth()));
        }
    }



    /** \brief Walk through tree and queue M2L tasks */
    void v_list_step() {
        for(node_t& n : _tree.in_order_walk()) {
            if(n.is_leaf() && n.getParticleContainer()->size() == 0) {
                continue;
Quentin Khan's avatar
Quentin Khan committed
399
            }
400
            create_M2L_task(&n);
Quentin Khan's avatar
Quentin Khan committed
401
402
403
404
405
        }
        // #pragma omp taskwait
    }


406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
    /**
     * \brief Create and queue an M2L task
     *
     * \param node A tree node
     */
    void create_M2L_task(node_t* node) {

        // Generate task dependencies
        // There cannot be more than 7^Dim cells involved in a M2L, in 3D, this is 343
        const char* task_deps[343];
        const char* data_dep = get_dependency<dep_t::L>(node);(void) data_dep;
        std::size_t idx_dep = 0;
        // Add existing dependencies
        for(node_t* v_item : node->V) {
            if(v_item->is_leaf()
               && v_item->getParticleContainer()->size() == 0) {
Quentin Khan's avatar
Quentin Khan committed
422
423
                continue;
            }
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
            task_deps[idx_dep] = get_dependency<dep_t::M>(v_item);
            ++idx_dep;
        }
        // Add mock dependencies, these are generated on the fly and used
        // only once, that way they can never stop a task from starting
        while(idx_dep < 343) {
            task_deps[idx_dep] = this->mock_dep.next();
            ++idx_dep;
        }

        #pragma omp task                                                \
            depend(in:                                                  \
                   task_deps[0][:1],                                    \
                   task_deps[1][:1],                                    \
                   task_deps[2][:1],                                    \
                   task_deps[3][:1],                                    \
                   task_deps[4][:1],                                    \
                   task_deps[5][:1],                                    \
                   task_deps[6][:1],                                    \
                   task_deps[7][:1],                                    \
                   task_deps[8][:1],                                    \
                   task_deps[9][:1],                                    \
                   task_deps[10][:1],                                   \
                   task_deps[11][:1],                                   \
                   task_deps[12][:1],                                   \
                   task_deps[13][:1],                                   \
                   task_deps[14][:1],                                   \
                   task_deps[15][:1],                                   \
                   task_deps[16][:1],                                   \
                   task_deps[17][:1],                                   \
                   task_deps[18][:1],                                   \
                   task_deps[19][:1],                                   \
                   task_deps[20][:1],                                   \
                   task_deps[21][:1],                                   \
                   task_deps[22][:1],                                   \
                   task_deps[23][:1],                                   \
                   task_deps[24][:1],                                   \
                   task_deps[25][:1],                                   \
                   task_deps[26][:1],                                   \
                   task_deps[27][:1],                                   \
                   task_deps[28][:1],                                   \
                   task_deps[29][:1],                                   \
                   task_deps[30][:1],                                   \
                   task_deps[31][:1],                                   \
                   task_deps[32][:1],                                   \
                   task_deps[33][:1],                                   \
                   task_deps[34][:1],                                   \
                   task_deps[35][:1],                                   \
                   task_deps[36][:1],                                   \
                   task_deps[37][:1],                                   \
                   task_deps[38][:1],                                   \
                   task_deps[39][:1],                                   \
                   task_deps[40][:1],                                   \
                   task_deps[41][:1],                                   \
                   task_deps[42][:1],                                   \
                   task_deps[43][:1],                                   \
                   task_deps[44][:1],                                   \
                   task_deps[45][:1],                                   \
                   task_deps[46][:1],                                   \
                   task_deps[47][:1],                                   \
                   task_deps[48][:1],                                   \
                   task_deps[49][:1],                                   \
                   task_deps[50][:1],                                   \
                   task_deps[51][:1],                                   \
                   task_deps[52][:1],                                   \
                   task_deps[53][:1],                                   \
                   task_deps[54][:1],                                   \
                   task_deps[55][:1],                                   \
                   task_deps[56][:1],                                   \
                   task_deps[57][:1],                                   \
                   task_deps[58][:1],                                   \
                   task_deps[59][:1],                                   \
                   task_deps[60][:1],                                   \
                   task_deps[61][:1],                                   \
                   task_deps[62][:1],                                   \
                   task_deps[63][:1],                                   \
                   task_deps[64][:1],                                   \
                   task_deps[65][:1],                                   \
                   task_deps[66][:1],                                   \
                   task_deps[67][:1],                                   \
                   task_deps[68][:1],                                   \
                   task_deps[69][:1],                                   \
                   task_deps[70][:1],                                   \
                   task_deps[71][:1],                                   \
                   task_deps[72][:1],                                   \
                   task_deps[73][:1],                                   \
                   task_deps[74][:1],                                   \
                   task_deps[75][:1],                                   \
                   task_deps[76][:1],                                   \
                   task_deps[77][:1],                                   \
                   task_deps[78][:1],                                   \
                   task_deps[79][:1],                                   \
                   task_deps[80][:1],                                   \
                   task_deps[81][:1],                                   \
                   task_deps[82][:1],                                   \
                   task_deps[83][:1],                                   \
                   task_deps[84][:1],                                   \
                   task_deps[85][:1],                                   \
                   task_deps[86][:1],                                   \
                   task_deps[87][:1],                                   \
                   task_deps[88][:1],                                   \
                   task_deps[89][:1],                                   \
                   task_deps[90][:1],                                   \
                   task_deps[91][:1],                                   \
                   task_deps[92][:1],                                   \
                   task_deps[93][:1],                                   \
                   task_deps[94][:1],                                   \
                   task_deps[95][:1],                                   \
                   task_deps[96][:1],                                   \
                   task_deps[97][:1],                                   \
                   task_deps[98][:1],                                   \
                   task_deps[99][:1],                                   \
                   task_deps[100][:1],                                  \
                   task_deps[101][:1],                                  \
                   task_deps[102][:1],                                  \
                   task_deps[103][:1],                                  \
                   task_deps[104][:1],                                  \
                   task_deps[105][:1],                                  \
                   task_deps[106][:1],                                  \
                   task_deps[107][:1],                                  \
                   task_deps[108][:1],                                  \
                   task_deps[109][:1],                                  \
                   task_deps[110][:1],                                  \
                   task_deps[111][:1],                                  \
                   task_deps[112][:1],                                  \
                   task_deps[113][:1],                                  \
                   task_deps[114][:1],                                  \
                   task_deps[115][:1],                                  \
                   task_deps[116][:1],                                  \
                   task_deps[117][:1],                                  \
                   task_deps[118][:1],                                  \
                   task_deps[119][:1],                                  \
                   task_deps[120][:1],                                  \
                   task_deps[121][:1],                                  \
                   task_deps[122][:1],                                  \
                   task_deps[123][:1],                                  \
                   task_deps[124][:1],                                  \
                   task_deps[125][:1],                                  \
                   task_deps[126][:1],                                  \
                   task_deps[127][:1],                                  \
                   task_deps[128][:1],                                  \
                   task_deps[129][:1],                                  \
                   task_deps[130][:1],                                  \
                   task_deps[131][:1],                                  \
                   task_deps[132][:1],                                  \
                   task_deps[133][:1],                                  \
                   task_deps[134][:1],                                  \
                   task_deps[135][:1],                                  \
                   task_deps[136][:1],                                  \
                   task_deps[137][:1],                                  \
                   task_deps[138][:1],                                  \
                   task_deps[139][:1],                                  \
                   task_deps[140][:1],                                  \
                   task_deps[141][:1],                                  \
                   task_deps[142][:1],                                  \
                   task_deps[143][:1],                                  \
                   task_deps[144][:1],                                  \
                   task_deps[145][:1],                                  \
                   task_deps[146][:1],                                  \
                   task_deps[147][:1],                                  \
                   task_deps[148][:1],                                  \
                   task_deps[149][:1],                                  \
                   task_deps[150][:1],                                  \
                   task_deps[151][:1],                                  \
                   task_deps[152][:1],                                  \
                   task_deps[153][:1],                                  \
                   task_deps[154][:1],                                  \
                   task_deps[155][:1],                                  \
                   task_deps[156][:1],                                  \
                   task_deps[157][:1],                                  \
                   task_deps[158][:1],                                  \
                   task_deps[159][:1],                                  \
                   task_deps[160][:1],                                  \
                   task_deps[161][:1],                                  \
                   task_deps[162][:1],                                  \
                   task_deps[163][:1],                                  \
                   task_deps[164][:1],                                  \
                   task_deps[165][:1],                                  \
                   task_deps[166][:1],                                  \
                   task_deps[167][:1],                                  \
                   task_deps[168][:1],                                  \
                   task_deps[169][:1],                                  \
                   task_deps[170][:1],                                  \
                   task_deps[171][:1],                                  \
                   task_deps[172][:1],                                  \
                   task_deps[173][:1],                                  \
                   task_deps[174][:1],                                  \
                   task_deps[175][:1],                                  \
                   task_deps[176][:1],                                  \
                   task_deps[177][:1],                                  \
                   task_deps[178][:1],                                  \
                   task_deps[179][:1],                                  \
                   task_deps[180][:1],                                  \
                   task_deps[181][:1],                                  \
                   task_deps[182][:1],                                  \
                   task_deps[183][:1],                                  \
                   task_deps[184][:1],                                  \
                   task_deps[185][:1],                                  \
                   task_deps[186][:1],                                  \
                   task_deps[187][:1],                                  \
                   task_deps[188][:1],                                  \
                   task_deps[189][:1],                                  \
                   task_deps[190][:1],                                  \
                   task_deps[191][:1],                                  \
                   task_deps[192][:1],                                  \
                   task_deps[193][:1],                                  \
                   task_deps[194][:1],                                  \
                   task_deps[195][:1],                                  \
                   task_deps[196][:1],                                  \
                   task_deps[197][:1],                                  \
                   task_deps[198][:1],                                  \
                   task_deps[199][:1],                                  \
                   task_deps[200][:1],                                  \
                   task_deps[201][:1],                                  \
                   task_deps[202][:1],                                  \
                   task_deps[203][:1],                                  \
                   task_deps[204][:1],                                  \
                   task_deps[205][:1],                                  \
                   task_deps[206][:1],                                  \
                   task_deps[207][:1],                                  \
                   task_deps[208][:1],                                  \
                   task_deps[209][:1],                                  \
                   task_deps[210][:1],                                  \
                   task_deps[211][:1],                                  \
                   task_deps[212][:1],                                  \
                   task_deps[213][:1],                                  \
                   task_deps[214][:1],                                  \
                   task_deps[215][:1],                                  \
                   task_deps[216][:1],                                  \
                   task_deps[217][:1],                                  \
                   task_deps[218][:1],                                  \
                   task_deps[219][:1],                                  \
                   task_deps[220][:1],                                  \
                   task_deps[221][:1],                                  \
                   task_deps[222][:1],                                  \
                   task_deps[223][:1],                                  \
                   task_deps[224][:1],                                  \
                   task_deps[225][:1],                                  \
                   task_deps[226][:1],                                  \
                   task_deps[227][:1],                                  \
                   task_deps[228][:1],                                  \
                   task_deps[229][:1],                                  \
                   task_deps[230][:1],                                  \
                   task_deps[231][:1],                                  \
                   task_deps[232][:1],                                  \
                   task_deps[233][:1],                                  \
                   task_deps[234][:1],                                  \
                   task_deps[235][:1],                                  \
                   task_deps[236][:1],                                  \
                   task_deps[237][:1],                                  \
                   task_deps[238][:1],                                  \
                   task_deps[239][:1],                                  \
                   task_deps[240][:1],                                  \
                   task_deps[241][:1],                                  \
                   task_deps[242][:1],                                  \
                   task_deps[243][:1],                                  \
                   task_deps[244][:1],                                  \
                   task_deps[245][:1],                                  \
                   task_deps[246][:1],                                  \
                   task_deps[247][:1],                                  \
                   task_deps[248][:1],                                  \
                   task_deps[249][:1],                                  \
                   task_deps[250][:1],                                  \
                   task_deps[251][:1],                                  \
                   task_deps[252][:1],                                  \
                   task_deps[253][:1],                                  \
                   task_deps[254][:1],                                  \
                   task_deps[255][:1],                                  \
                   task_deps[256][:1],                                  \
                   task_deps[257][:1],                                  \
                   task_deps[258][:1],                                  \
                   task_deps[259][:1],                                  \
                   task_deps[260][:1],                                  \
                   task_deps[261][:1],                                  \
                   task_deps[262][:1],                                  \
                   task_deps[263][:1],                                  \
                   task_deps[264][:1],                                  \
                   task_deps[265][:1],                                  \
                   task_deps[266][:1],                                  \
                   task_deps[267][:1],                                  \
                   task_deps[268][:1],                                  \
                   task_deps[269][:1],                                  \
                   task_deps[270][:1],                                  \
                   task_deps[271][:1],                                  \
                   task_deps[272][:1],                                  \
                   task_deps[273][:1],                                  \
                   task_deps[274][:1],                                  \
                   task_deps[275][:1],                                  \
                   task_deps[276][:1],                                  \
                   task_deps[277][:1],                                  \
                   task_deps[278][:1],                                  \
                   task_deps[279][:1],                                  \
                   task_deps[280][:1],                                  \
                   task_deps[281][:1],                                  \
                   task_deps[282][:1],                                  \
                   task_deps[283][:1],                                  \
                   task_deps[284][:1],                                  \
                   task_deps[285][:1],                                  \
                   task_deps[286][:1],                                  \
                   task_deps[287][:1],                                  \
                   task_deps[288][:1],                                  \
                   task_deps[289][:1],                                  \
                   task_deps[290][:1],                                  \
                   task_deps[291][:1],                                  \
                   task_deps[292][:1],                                  \
                   task_deps[293][:1],                                  \
                   task_deps[294][:1],                                  \
                   task_deps[295][:1],                                  \
                   task_deps[296][:1],                                  \
                   task_deps[297][:1],                                  \
                   task_deps[298][:1],                                  \
                   task_deps[299][:1],                                  \
                   task_deps[300][:1],                                  \
                   task_deps[301][:1],                                  \
                   task_deps[302][:1],                                  \
                   task_deps[303][:1],                                  \
                   task_deps[304][:1],                                  \
                   task_deps[305][:1],                                  \
                   task_deps[306][:1],                                  \
                   task_deps[307][:1],                                  \
                   task_deps[308][:1],                                  \
                   task_deps[309][:1],                                  \
                   task_deps[310][:1],                                  \
                   task_deps[311][:1],                                  \
                   task_deps[312][:1],                                  \
                   task_deps[313][:1],                                  \
                   task_deps[314][:1],                                  \
                   task_deps[315][:1],                                  \
                   task_deps[316][:1],                                  \
                   task_deps[317][:1],                                  \
                   task_deps[318][:1],                                  \
                   task_deps[319][:1],                                  \
                   task_deps[320][:1],                                  \
                   task_deps[321][:1],                                  \
                   task_deps[322][:1],                                  \
                   task_deps[323][:1],                                  \
                   task_deps[324][:1],                                  \
                   task_deps[325][:1],                                  \
                   task_deps[326][:1],                                  \
                   task_deps[327][:1],                                  \
                   task_deps[328][:1],                                  \
                   task_deps[329][:1],                                  \
                   task_deps[330][:1],                                  \
                   task_deps[331][:1],                                  \
                   task_deps[332][:1],                                  \
                   task_deps[333][:1],                                  \
                   task_deps[334][:1],                                  \
                   task_deps[335][:1],                                  \
                   task_deps[336][:1],                                  \
                   task_deps[337][:1],                                  \
                   task_deps[338][:1],                                  \
                   task_deps[339][:1],                                  \
                   task_deps[340][:1],                                  \
                   task_deps[341][:1],                                  \
                   task_deps[342][:1]                                   \
                ) depend(inout: data_dep[:1]) firstprivate(node)
        {
Quentin Khan's avatar
Quentin Khan committed
781

782
            const int thread_num = omp_get_thread_num();
Quentin Khan's avatar
Quentin Khan committed
783

784
785
786
            std::vector<decltype(std::declval<const node_t>().getData())> v_item_data_list;
            std::vector<int> v_item_indices;
            // Needed to compute offset between boxes
Quentin Khan's avatar
Quentin Khan committed
787
788
789
790
791
            for(node_t* v_item : node->V) {
                if(v_item->is_leaf()
                   && v_item->getParticleContainer()->size() == 0) {
                    continue;
                }
792
793
                v_item_indices.push_back(compute_box_offset_index(node, v_item, 3));
                v_item_data_list.push_back(v_item->getData());
Quentin Khan's avatar
Quentin Khan committed
794
795
            }

796
797
            // Call kernel M2L operator
            this->_kernels[thread_num]->M2L(node->getData(), v_item_data_list.data(), v_item_indices.data(), static_cast<int>(v_item_data_list.size()), static_cast<int>(node->getDepth()));
Quentin Khan's avatar
Quentin Khan committed
798
799
800
        }
    }

801
    /** \brief Walk through leaves and queue P2L tasks */
Quentin Khan's avatar
Quentin Khan committed
802
803
804
805
806
807
808
809
    void x_list_step() {
        /* NOTE: the X list and W list are complementary: if A is in X(B) then B
         * is in W(A).
         *
         * We loop over the leaves first to detect the empty ones early on.
         */
        for(node_t* leaf : _tree.leaves()) {
            if(leaf->getParticleContainer()->size() > 0) {
810
811
812
813
814
                this->create_P2L_task(leaf);
            }
        }
        // #pragma omp taskwait
    }
815
816


817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
    /**
     * \brief Create and queue a P2L task
     *
     * \param leaf A tree leaf
     */
    void create_P2L_task(node_t* leaf) {
        for(node_t* w_item : leaf->W) {
            if(w_item->is_leaf() && w_item->getParticleContainer()->size() == 0) {
                continue;
            }

            const char* w_dep = get_dependency<dep_t::L>(w_item);(void)w_dep;
            const char* ps_dep = get_dependency<dep_t::P_s>(leaf);(void)ps_dep;

            #pragma omp task firstprivate(leaf, w_item) depend(in: ps_dep[:1]) depend(inout: w_dep[:1])
            {
                const int thread_num = omp_get_thread_num();
                this->_kernels[thread_num]->P2L(w_item->getData(), leaf->getParticleContainer());
Quentin Khan's avatar
Quentin Khan committed
835
836
837
838
            }
        }
    }

839
    /** Walk through tree and queue L2L tasks */
Quentin Khan's avatar
Quentin Khan committed
840
841
    void down_to_down() {
        for(node_t& n : _tree.pre_order_walk()) {
842
843
            if(! n.is_leaf()) {
                this->create_L2L_task(&n);
Quentin Khan's avatar
Quentin Khan committed
844
845
846
847
848
            }
        }
        // #pragma omp taskwait
    }

849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891

    /**
     * \brief Create and queue an L2L task
     *
     * \param node An internal tree node
     *
     * \node node is assumed not to be a leaf
     */
    void create_L2L_task(node_t* node) {
        assert(! node->is_leaf());

        const char* parent_dep = get_dependency<dep_t::L>(node);(void)parent_dep;
        const char* children_dep[node_t::child_count];
        for(std::size_t i = 0; i < node_t::child_count; ++i) {
            children_dep[i] = get_dependency<dep_t::L>(node->getChild(i));
        }

        #pragma omp task                        \
            depend(in: parent_dep[:1])          \
            depend(inout:                       \
                   children_dep[0][:1],         \
                   children_dep[1][:1],         \
                   children_dep[2][:1],         \
                   children_dep[3][:1],         \
                   children_dep[4][:1],         \
                   children_dep[5][:1],         \
                   children_dep[6][:1],         \
                   children_dep[7][:1]          \
                )
        {
            const int thread_num = omp_get_thread_num();

            typename node_t::data_t* child_data[node_t::child_count];
            for(std::size_t i = 0; i < node_t::child_count; ++i) {
                child_data[i] = node->getChild(i)->getData();
            }
            this->_kernels[thread_num]->L2L(node->getData(), child_data, static_cast<int>(node->getDepth()));
        }
    }



    /** \brief Walk through the leaves an queue M2P tasks */
Quentin Khan's avatar
Quentin Khan committed
892
893
894
    void w_list_step() {
        for(node_t* leaf : _tree.leaves()) {
            if(leaf->getParticleContainer()->size() > 0) {
895
896
897
898
899
                create_M2P_task(leaf);
            }
        }
        // #pragma omp taskwait
    }
900

901
902
903
904
905
906
907
908
909
910
    /**
     * \brief Create and queue an M2P task
     *
     * \param leaf A leaf
     */
    void create_M2P_task(node_t* leaf) {
        for(node_t* w_item : leaf->W) {
            if(w_item->is_leaf() && w_item->getParticleContainer()->size() == 0) {
                continue;
            }
911

912
913
914
915
916
917
918
            const char* w_dep = get_dependency<dep_t::M>(w_item); (void)w_dep;
            const char* pt_dep = get_dependency<dep_t::P_t>(leaf); (void)pt_dep;

            #pragma omp task depend(inout: pt_dep[:1]) depend(in: w_dep[:1])
            {
                const int thread_num = omp_get_thread_num();
                this->_kernels[thread_num]->M2P(w_item->getData(), leaf->getParticleContainer());
Quentin Khan's avatar
Quentin Khan committed
919
920
921
922
            }
        }
    }

923
924

    /** \brief Walk through the leaves and queue L2P tasks */
Quentin Khan's avatar
Quentin Khan committed
925
926
    void down_to_target() {
        for(node_t* leaf : _tree.leaves()) {
927
            if(leaf->getParticleContainer()->size() != 0) {
928
                this->create_L2P_task(leaf);
Quentin Khan's avatar
Quentin Khan committed
929
930
931
932
933
            }
        }
        // #pragma omp taskwait
    }

934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
    /**
     * \brief Create and queue an L2P task
     *
     * \param leaf A leaf
     */
    void create_L2P_task(node_t* leaf) {
        const char* data_dep = get_dependency<dep_t::L>(leaf); (void)data_dep;
        const char* pt_dep = get_dependency<dep_t::P_t>(leaf); (void)pt_dep;
        #pragma omp task depend(inout: pt_dep[:1]) depend(in: data_dep[:1])
        {
            const int thread_num = omp_get_thread_num();
            this->_kernels[thread_num]->L2P(leaf->getData(), leaf->getParticleContainer());
        }
    }


    /** \brief Walk through the leaves and queue P2P tasks */
    void u_list_step() {
        for(node_t* leaf : _tree.leaves()) {
            this->create_P2P_task(leaf);
        }
        // #pragma omp taskwait
    }

Quentin Khan's avatar
Quentin Khan committed
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984

    /** \brief Direct computation step (P2P)
     *
     * For each tree leaf, a direct computation is done with the adjacent
     * leaves.
     *
     * The symmetric computation kernels expect to receive the list of adjacent
     * leaves, stored in a leaf's U list, sorted by offset index.
     *
     * The offset index is of a node compared to another follows the
     * numerotation of the adjacent nodes, on each axis. In 3D, this woud give,
     * with L the reference leaf:
     *
     *          x          x            x
     *        0 1 2  |   9 10 11  |  18 19 20
     *     y  3 4 5  |  12  L 14  |  21 22 23
     *        6 7 8  |  15 16 17  |  24 25 26
     *         z=0        z=1          z=2
     *
     * When two node that are to be compared are not on the same level, the
     * lowest one's ancestor that is at the highest one's level is used to
     * compute the index.
     *
     * To statisfy this condition, the leaves have to be sorted before being
     * given to the kernel P2P method.
     *
     */
985
    void create_P2P_task(node_t* leaf) {
Quentin Khan's avatar
Quentin Khan committed
986
987
        using container_t = typename node_t::particle_container_t;

988
989
990
991
        container_t* const leaf_source_particle_container =
            leaf->getParticleContainer();
        container_t* const leaf_target_particle_container =
            leaf->getParticleContainer();
Quentin Khan's avatar
Quentin Khan committed
992

993
994
995
996
997
        // Skip empty leaves
        if( leaf_source_particle_container->size() == 0
            && leaf_target_particle_container->size() == 0) {
            return;
        }
Quentin Khan's avatar
Quentin Khan committed
998

999
1000
        auto it = leaf->U.begin();
        bool do_inner = true;
Quentin Khan's avatar
Quentin Khan committed
1001

1002
1003
1004
1005
        while(it != leaf->U.end()) {
            constexpr const std::size_t max_task_size = 27;
            std::size_t i = 0;
            auto first = it;
Quentin Khan's avatar
Quentin Khan committed
1006

1007
1008
1009
            const char* task_deps[max_task_size];
            const char* leaf_s_dep = get_dependency<dep_t::P_s>(leaf);(void)leaf_s_dep;
            const char* leaf_t_dep = get_dependency<dep_t::P_t>(leaf);(void)leaf_t_dep;
Quentin Khan's avatar
Quentin Khan committed
1010

1011
1012
1013
1014
1015
            while((it != leaf->U.end()) && (i < max_task_size)) {
                if((*it) == leaf || (*it)->getParticleContainer()->size() == 0) {
                    ++it;
                    continue;
                }
Quentin Khan's avatar
Quentin Khan committed
1016

1017
1018
1019
1020
                task_deps[i] = get_dependency<dep_t::P_t>(*it);
                ++i;
                ++it;
            }
1021

1022
1023
1024
1025
            while(i < max_task_size) {
                task_deps[i] = this->mock_dep.next();
                ++i;
            }
1026

1027
1028
1029
1030
1031
1032
1033
1034
1035
1036
1037
1038
1039
1040
1041
1042
1043
1044
1045
1046
1047
1048
1049
1050
1051
1052
1053
1054
1055
1056
1057
1058
1059
1060
1061
1062
1063
1064
            #pragma omp task                                            \
                firstprivate(leaf, leaf_source_particle_container, leaf_target_particle_container, first, it, do_inner) \
                depend(inout:                                           \
                       leaf_s_dep[:1],                                  \
                       leaf_t_dep[:1],                                  \
                       task_deps[0][:1],                                \
                       task_deps[1][:1],                                \
                       task_deps[2][:1],                                \
                       task_deps[3][:1],                                \
                       task_deps[4][:1],                                \
                       task_deps[5][:1],                                \
                       task_deps[6][:1],                                \
                       task_deps[7][:1],                                \
                       task_deps[8][:1],                                \
                       task_deps[9][:1],                                \
                       task_deps[10][:1],                               \
                       task_deps[11][:1],                               \
                       task_deps[12][:1],                               \
                       task_deps[13][:1],                               \
                       task_deps[14][:1],                               \
                       task_deps[15][:1],                               \
                       task_deps[16][:1],                               \
                       task_deps[17][:1],                               \
                       task_deps[18][:1],                               \
                       task_deps[19][:1],                               \
                       task_deps[20][:1],                               \
                       task_deps[21][:1],                               \
                       task_deps[22][:1],                               \
                       task_deps[23][:1],                               \
                       task_deps[24][:1],                               \
                       task_deps[25][:1],                               \
                       task_deps[26][:1]                                \
                    )
            {
                const int thread_num = omp_get_thread_num();
                // Vectors to be filled after sort and passed to kernel P2P
                std::vector<container_t*> u_item_source_particle_containers;
                std::vector<int> u_item_indices;
Quentin Khan's avatar
Quentin Khan committed
1065

1066
                auto last = it;
Quentin Khan's avatar
Quentin Khan committed
1067

1068
1069
1070
                while(first != last) {
                    // Skip empty u_items
                    if((*first) == leaf || (*first)->getParticleContainer()->size() == 0) {
Quentin Khan's avatar
Quentin Khan committed
1071
                        ++first;
1072
                        continue;
Quentin Khan's avatar
Quentin Khan committed
1073
                    }
1074
1075
1076
                    u_item_source_particle_containers.push_back((*first)->getParticleContainer());
                    u_item_indices.push_back(compute_box_offset_index(leaf, *first, 1));
                    ++first;
Quentin Khan's avatar
Quentin Khan committed
1077
                }
1078
1079
1080
1081
1082
1083
1084
1085
1086
1087
1088
                // Call P2P on vectors data
                this->_kernels[thread_num]
                    ->P2P(
                        FTreeCoordinate(MortonIndex(leaf->getIndex())),
                        leaf_target_particle_container,
                        leaf_source_particle_container,
                        u_item_source_particle_containers.data(),
                        u_item_indices.data(),
                        static_cast<int>(u_item_source_particle_containers.size()),
                        do_inner
                        );
Quentin Khan's avatar
Quentin Khan committed
1089
            }
1090
1091
            do_inner = false;

Quentin Khan's avatar
Quentin Khan committed
1092
1093
1094
1095
1096
1097
1098
1099
1100
1101
1102
1103
1104
1105
1106
1107
1108
1109
1110
1111
1112
1113
1114
1115
1116
1117
1118
1119
1120
1121
1122
1123
1124
1125
1126
1127
1128
1129
1130
1131
1132
1133
1134
1135
1136
1137
1138
1139
1140
1141
        }
    }


    /** Compute resulting index in pre-computation array.
     *
     * Kernels precompute some operators according to node offsets. This method
     * returns the `other_node` index in the precomputation array.
     *
     * The index has the following form, with d the space dimension:
     *
     * x+n * 7^d + y+n * 7^(d-1) + ... + z+n
     *
     * Because an `other_node` is at most `n` boxes away in every direction from a
     * `node`, in order to get a positive index, we add `n` to the offset in every
     * direction.
     *
     * \param node The target node
     * \param other_node The interacting node
     * \param n The longest possible distance between node and other_node in
     * terms of boxes on an axis
     *
     * \warning If the nodes are not at the same level, the lowest one's
     * ancestor at the highest one's level is used.
     */
    int compute_box_offset_index(node_t* node, node_t* other_node, const std::size_t n) {
        while(node->getDepth() > other_node->getDepth()) {
            node = node->getParent();
        }
        while(node->getDepth() < other_node->getDepth()) {
            other_node = other_node->getParent();
        }

        typename node_t::FReal box_width = (node->getBox().c2 - node->getBox().c1)[0];

        auto center_offset = other_node->getBox().center() - node->getBox().center();
        std::size_t other_node_index = 0;
        for(std::size_t i = 0; i < node_t::Dim; ++i) {
            other_node_index = other_node_index * (2 * n + 1)
                + static_cast<unsigned long>(std::lround(center_offset[i] / box_width))
                + n;
        }

        return static_cast<int>(other_node_index);
    }


};

#endif