aboutsummaryrefslogblamecommitdiffstats
path: root/contrib/libs/cxxsupp/openmp/kmp_tasking.cpp
blob: 1622c6aea10d7ce59b8a7baaeb513b4c895442c5 (plain) (tree)
1927
1928
1929
1930
1931
1932
1933
1934
1935
1936
1937
1938
1939
1940
1941
1942
1943
1944
1945
1946
1947
1948
1949
1950
1951
1952
1953
1954
1955
1956
1957
1958
1959
1960
1961
1962
1963
1964
1965
1966
1967
1968
1969
1970
1971
1972
1973
1974
1975
1976
1977
1978
1979
1980
1981
1982
1983
1984
1985
1986
1987
1988
1989
1990
1991
1992
1993
1994
1995
1996
1997
1998
1999
2000
2001
2002
2003
2004
2005
2006
2007
2008
2009
2010
2011
2012
2013
2014
2015
2016
2017
2018
2019
2020
2021
2022
2023
2024
2025
2026
2027
2028
2029
2030
2031
2032
2033
2034
2035
2036
2037
2038
2039
2040
2041
2042
2043
2044
2045
2046
2047
2048
2049
2050
2051
2052
2053
2054
2055
2056
2057
2058
2059
2060
2061
2062
2063
2064
2065
2066
2067
2068
2069
2070
2071
2072
2073
2074
2075
2076
2077
2078
2079
2080
2081
2082
2083
2084
2085
2086
2087
2088
2089
2090
2091
2092
2093
2094
2095
2096
2097
2098
2099
2100
2101
2102
2103
2104
2105
2106
2107
2108
2109
2110
2111
2112
2113
2114
2115
2116
2117
2118
2119
2120
2121
2122
2123
2124
2125
2126
2127
2128
2129
2130
2131
2132
2133
2134
2135
2136
2137
2138
2139
2140
2141
2142
2143
2144
2145
2146
2147
2148
2149
2150
2151
2152
2153
2154
2155
2156
2157
2158
2159
2160
2161
2162
2163
2164
2165
2166
2167
2168
2169
2170
2171
2172
2173
2174
2175
2176
2177
2178
2179
2180
2181
2182
2183
2184
2185
2186
2187
2188
2189
2190
2191
2192
2193
2194
2195
2196
2197
2198
2199
2200
2201
2202
2203
2204
2205
2206
2207
2208
2209
2210
2211
2212
2213
2214
2215
2216
2217
2218
2219
2220
2221
2222
2223
2224
2225
2226
2227
2228
2229
2230
2231
2232
2233
2234
2235
2236
2237
2238
2239
2240
2241
2242
2243
2244
2245
2246
2247
2248
2249
2250
2251
2252
2253
2254
2255
2256
2257
2258
2259
2260
2261
2262
2263
2264
2265
2266
2268
2269
2270
2271
2272
2273
2274
2275
2276
2277
2278
2279
2280
2281
2282
2283
2284
2285
2286
2287
2288
2289
2290
2291
2292
2293
2294
2295
2296
2297
2298
2299
2300
2301
2302
2303
2304
2305
2306
2307
2308
2309
2310
2311
2312
2313
2314
2315
2316
2317
2318
2319
2320
2321
2322
2323
2324
2325
2326
2327
2328
2329
2330
2331
2332
2333
2334
2335
2336
2337
2338
2339
2340
2341
2342
2343
2344
2345
2346
2347
2348
2349
2350
2351
2352
2353
2354
2355
2356
2357
2358
2359
2360
2361
2362
2363
2364
2365
2366
2367
2368
2369
2370
2371
2372
2373
2374
2375
2376
2377
2378
2379
2380
2381
2382
2383
2384
2385
2386
2387
2388
2389
2390
2391
2392
2393
2394
2395
2396
2397
2398
2399
2400
2401
2402
2403
2404
2405
2406
2407
2408
2409
2410
2411
2412
2413
2414
2415
2416
2417
2418
2419
2420
2421
2422
2423
2424
2425
2426
2427
2428
2429
2430
2431
2432
2433
2434
2435
2436
2437
2438
2439
2440
2441
2442
2443
2444
2445
2446
2447
2448
2449
2450
2451
2452
2453
2454
2455
2456
2457
2458
2459
2460
2461
2462
2463
2464
2465
2466
2467
2468
2469
2470
2471
2472
2473
2474
2475
2476
2477
2478
2479
2480
2481
2482
2483
2484
2485
2486
2487
2488
2489
2490
2491
2492
2493
2494
2495
2496
2497
2498
2499
2500
2501
2502
2503
2504
2505
2506
2507
2508
2509
2510
2511
2512
2513
2514
2515
2516
2517
2518
2519
2520
2521
2522
2523
2524
2525
2526
2527
2528
2529
2530
2531
2532
2533
2534
2535
2536
2537
2538
2539
2540
2541
2542
2543
2544
2545
2546
2547
2548
2549
2550
2551
2552
2553
2554
2555
2556
2557
2558
2559
2560
2561
2562
2563
2564
2565
2566
2567
2568
2569
2570
2571
2572
2573
2574
2575
2576
2577
2578
2579
2580
2581
2582
2583
2584
2585
2586
2587
2588
2589
2590
2591
2592
2593
2594
2595
2596
2597
2598
2599
2600
2601
2602
2603
2604
2605
2606
2607
2608
2609
2610
2611
2612
2613
2614
2615
2616
2617
2618
2619
2620
2621
2622
2623
2624
2625
2626
2627
2628
2629
2630
2631
2632
2633
2634
2635
2636
2637
2638
2639
2640
2641
2642
2643
2644
2645
2646
2647
2648
2649
2650
2651
2652
2653
2654
2655
2656
2657
2658
2659
2660
2661
2662
2663
2664
2665
2666
2667
2668
2669
2670
2671
2672
2673
2674
2675
2676
2677
2678
2679
2680
2681
2682
2683
2684
2685
2686
2687
2688
2689
2690
2691
2692
2693
2694
2695
2696
2697
2698
2699
2700
2701
2702
2703
2704
2705
2706
2707
2708
2709
2710
2711
2712
2713
2714
2715
2716
2717
2718
2719
2720
2721
2722
2723
2724
2725
2726
2727
2728
2729
2730
2731
2732
2733
2734
2735
2736
2737
2738
2739
2740
2741
2742
2743
2744
2745
2746
2747
2748
2749
2750
2751
2752
2753
2754
2755
2756
2757
2758
2759
2760
2761
2762
2763
2764
2765
2766
2767
2768
2769
2770
2771
2772
2773
2774
2775
2776
2777
2778
2779
2780
2781
2782
2783
2784
2785
2786
2787
2788
2789
2790
2791
2792
2793
2794
2795
2796
2797
2798
2799
2800
2801
2802
2803
2804
2805
2806
2807
2808
2809
2810
2811
2812
2813
2814
2815
2816
2817
2818
2819
2820
2821
2822
2823
2824
2825
2826
2827
2828
2829
2830
2831
2832
2833
2834
2835
2836
2837
2838
2839
2840
2841
2842
2843
2844
4309
4310
4311
4312
4313
4314
4315
4316
4317
4318
4319
4320
4321
4322
4323
4324
4325
4326
4327
4328
4329
4330
4331
4332
4333
4334
4335
4336
4337
4338
4339
4340
4341
4342
4343
4344
4345
4346
4347
4348
4349
4350
4351
4352
4353
4354
4355
4356
4357
4358
4359
4360
4361
4362
4363
4364
4365
4366
4367
4368
4369
4370
4371
4372
4373
4374
4375
4376
4377
4378
4379
4380
4381
4382
4383
4384
4385
4386
4387
4388
4389
4390
4391
4392
4393
4394
4395
4396
4397
4398
4399
4400
4401
4402
4403
4404
4405
4406
4407
4408
4409
4410
4411
4412
4413
4414
4415
4416
4417
4418
4419
4420
4421
4422
4423
4424
4425
4426
4427
4428
4429
4430
4431
4432
4433
4434
4435
4436
4437
4438
4439
4440
4441
4442
4443
4444
4445
4446
4447
4448
4449
4450
4451
4452
4453
4454
4455
4456
4457
4458
4459
4460
4461
4462
4463
4464
4465
4466
4467
4468
4469
4470
4471
4472
4473
4474
4475
4476
4477
4478
4479
4480
4481
4482
4483
4484
4485
4486
4487
4488
4489
4490
4491
4492
4493
4494
4495
4496
4497
4498
4499
4500
4501
4502
4503
4504
4505
4506
4507
4508
4509
4510
4511
4512
4513
4514
4515
4516
4517
4518
4519
4520
4521
4522
4523
4524
4525
4526
4527
4528
4529
4530
4531
4532
4533
4534
4535
4536
4537
4538
4539
4540
4541
4542
4543
4544
4545
4546
4547
4548
4549
4550
4551
4552
4553
4554
4555
4556
4557
4558
4559
4560
4561
4562
4563
4564
4565
4566
4567
4568
4569
4570
4571
4572
4573
4574
4575
4576
4577
4578
4579
4580
4581
4582
4583
4584
4585
4586
4587
4588
4589
4590
4591
4592
4593
4594
4595
4596
4597
4598
4599
4600
4601
4602
4603
4604
4605
4606
4607
4608
4609
4610
4611
4612
4613
4614
4615
4616
4617
4618
4619
4620
4621
4622
4623
4624
4625
4626
4627
4628
4629
4630
4631
4632
4633
4634
4635
4636
4637
4638
4639
4640
4641
4642
4643
4644
4645
4646
4647
4648
4649
4650
4651
4652
4653
4654
4655
4656
4657
4658
4659
4660
4661
4662
4663
4664
4665
4666
4667
4668
4669
4670
4671
4672
4673
4674
4675
4676
4677
4678
4679
4680
4681
4682
4683
4690
4691
4692
4693
4694
4695
4696
4697
4698
4699
4700
4701
4702
4703
4704
4705
4706
4707
4708
4709
4710
4711
4712
4713
4714
4715
4716
4717
4718
4719
4720
4721
4722
4723
4724
4725
4726
4727
4728
4729
4730
4731
4732
4733
4734
4735
4736
4737
4738
4739
4740
4741
4742
4743
4744
4745
4746
4747
4748
4749
4750
4751
4752
4753
4754
4755
4756
4757
4758
4759
4760
4761
4762
4763
4764
4765
4766
4767
4768
4769
4770
4771
4772
4773
4774
4775
4776
4777
4778
4779
4780
4781
4782
4783
4784
4785
4786
4787
4788
4789
4790
4791
4792
4793
4794
4795
4796
4797
4798
4799
4800
4801
4802
4803
4804
4805
4806
4807
4808
4809
4810
4811
4812
4813
4814
4815
4816
4817
4818
4819
4820
4821
4822
4823
4824
4825
4826
4827
4828
4829
4830
4831
4832
4833
4834
4835
4836
4837
4838
4839
4840
4841
4842
4843
4844
4845
4846
4847
4848
4849
4850
4851
4852
4853
4854
4855
4856
4857
4858
4859
4860
4861
4862
4863
4864
4865
4866
4867
4868
4869
4870
4871
4872
4873
4874
4875
4876
4877
4878
4879
4880
4881
4882
4883
4884
4885
4886
4887
4888
4889
4890
4891
4892
4893
4894
4895
4896
4897
4898
4899
4900
4901
4902
4903
4904
4905
4906
4907
4908
4909
4910
4911
4912
4913
4914
4915
4916
4917
4918
4919
4920
4921
4922
4923
4924
4925
4926
4927
4928
4929
4930
4931
4932
4933
4934
4935
4936
4937
4938
4939
4940
4941
4942
4943
4944
4945
4946
4947
4948
4949
4950
4951
4952
4953
4954
4955
4956
4957
4958
4959
4960
4961
4962
4963
4964
4965
4966
4967
4968
4969
4970
4971
4972
4973
4974
4975
4976
4977
4978
4979
4980
4981
4982
4983
4984
4985
4986
4987
4988
4989
4990
4991
4992
4993
4994
4995
4996
4997
4998
4999
5000
5001
5002
5003
5004
5005
5006
5007
5008
5009
5010
5011
5012
5013
5014
5015
5016
5017
5018
5019
5020
5021
5022
5023
5024
5025
5026
5027
5028
5029
5030
5031
5032
5033
5034
5035
5036
5037
5038
5039
5040
5041
5042
5043
5044
5045
5046
5047
5048
5049
5050
5051
5052
5053
5054
5055
5056
5057
5058
5059
5060
5061
5062
5063
5064
5065
5066
5067
5068
5069
5070
5071
5072
5073
5074
5075
5076
5077
5078
5079
5080
5081
5082
5083
5084
5085
5086
5087
5088
5089
5090
5091
5092
5093
5094
5095
5096
5097
5098
5099
5100
5101
5102
5103
5104
5105
5106
5107
5108
5109
5110
5111
5112
5113
5114
5115
5116
5117
5118































































































































































































































































































































                                                                                








































































































































                                                                                



                                                                    








                                                                               



































                                                                                




                                                                            



































































                                                                                
                                                               



























































































































                                                                                
                                                     































































































































































                                                                                


                                                    













































































                                                                                
















                                                                                














                                                                             
             
                         
      









































































































                                                                                

                                                                             
                                                    


                                                                          



















































































































































































































































                                                                                









                                                           




















                                                                                
                                                                                











                                                                                
                                            




                                                                               

                                                                              
     
                                                         




                                                                                
                                                   

                                                                           
                                                    






















                                                                                
                                                                               
                             
                                                                               
























                                                                               
                                     











                                                                            

                                                            





                                                                             


































                                                                                

                                                                               

















































                                                                                



                                                                     
                                     


















































































































































                                                                                










                                                                             




















                                                                               
                                
                      

                                                              
                            

                                       



















































































































                                                                                












                                                         
   


















































































































































































































































































































































                                                                                
                                                                            































































































































































































































































































































































































































































































































































































                                                                                

































































































                                                                                














































































































































































                                                                                


                                           


                                                                               































































                                                                               


                                                                        
                                                                             
                                                                     



































                                                                                
                                                      


















































































                                                                                


                                               


























                                                                               










                                                                                































                                                                                








                                                                               
























                                                                                






                                                                    




































                                                                              
                      














                                                                                
                                          



























































































































































































































                                                                                
















                                                                  





























                                                                             
                                                               

















                                                                                
                                                                















































                                                                           

                                                   










                                                                             
                  

                        
                         






























                                                                                
                        





                                                                                
                                            






                                                     
                                             














































                                                                                
                                                                      


























































































                                                                                
                                                                              
                                                          
                                                                       
































































































































































                                                                                
             
                         
                                                

                                                                      




























































                                                                                
                                       











                                                          









                                                                           





















































































































































































































































































































































































                                                                                




                                                                           











































































































































































































































































































































































































































                                                                                
                                                                       












                                                                                
/*
 * kmp_tasking.cpp -- OpenMP 3.0 tasking support.
 */

//===----------------------------------------------------------------------===//
//
// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
// See https://llvm.org/LICENSE.txt for license information.
// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
//
//===----------------------------------------------------------------------===//

#include "kmp.h"
#include "kmp_i18n.h"
#include "kmp_itt.h"
#include "kmp_stats.h"
#include "kmp_wait_release.h"
#include "kmp_taskdeps.h"

#if OMPT_SUPPORT
#include "ompt-specific.h"
#endif

/* forward declaration */
static void __kmp_enable_tasking(kmp_task_team_t *task_team,
                                 kmp_info_t *this_thr);
static void __kmp_alloc_task_deque(kmp_info_t *thread,
                                   kmp_thread_data_t *thread_data);
static int __kmp_realloc_task_threads_data(kmp_info_t *thread,
                                           kmp_task_team_t *task_team);
static void __kmp_bottom_half_finish_proxy(kmp_int32 gtid, kmp_task_t *ptask);

#ifdef BUILD_TIED_TASK_STACK

//  __kmp_trace_task_stack: print the tied tasks from the task stack in order
//  from top do bottom
//
//  gtid: global thread identifier for thread containing stack
//  thread_data: thread data for task team thread containing stack
//  threshold: value above which the trace statement triggers
//  location: string identifying call site of this function (for trace)
static void __kmp_trace_task_stack(kmp_int32 gtid,
                                   kmp_thread_data_t *thread_data,
                                   int threshold, char *location) {
  kmp_task_stack_t *task_stack = &thread_data->td.td_susp_tied_tasks;
  kmp_taskdata_t **stack_top = task_stack->ts_top;
  kmp_int32 entries = task_stack->ts_entries;
  kmp_taskdata_t *tied_task;

  KA_TRACE(
      threshold,
      ("__kmp_trace_task_stack(start): location = %s, gtid = %d, entries = %d, "
       "first_block = %p, stack_top = %p \n",
       location, gtid, entries, task_stack->ts_first_block, stack_top));

  KMP_DEBUG_ASSERT(stack_top != NULL);
  KMP_DEBUG_ASSERT(entries > 0);

  while (entries != 0) {
    KMP_DEBUG_ASSERT(stack_top != &task_stack->ts_first_block.sb_block[0]);
    // fix up ts_top if we need to pop from previous block
    if (entries & TASK_STACK_INDEX_MASK == 0) {
      kmp_stack_block_t *stack_block = (kmp_stack_block_t *)(stack_top);

      stack_block = stack_block->sb_prev;
      stack_top = &stack_block->sb_block[TASK_STACK_BLOCK_SIZE];
    }

    // finish bookkeeping
    stack_top--;
    entries--;

    tied_task = *stack_top;

    KMP_DEBUG_ASSERT(tied_task != NULL);
    KMP_DEBUG_ASSERT(tied_task->td_flags.tasktype == TASK_TIED);

    KA_TRACE(threshold,
             ("__kmp_trace_task_stack(%s):             gtid=%d, entry=%d, "
              "stack_top=%p, tied_task=%p\n",
              location, gtid, entries, stack_top, tied_task));
  }
  KMP_DEBUG_ASSERT(stack_top == &task_stack->ts_first_block.sb_block[0]);

  KA_TRACE(threshold,
           ("__kmp_trace_task_stack(exit): location = %s, gtid = %d\n",
            location, gtid));
}

//  __kmp_init_task_stack: initialize the task stack for the first time
//  after a thread_data structure is created.
//  It should not be necessary to do this again (assuming the stack works).
//
//  gtid: global thread identifier of calling thread
//  thread_data: thread data for task team thread containing stack
static void __kmp_init_task_stack(kmp_int32 gtid,
                                  kmp_thread_data_t *thread_data) {
  kmp_task_stack_t *task_stack = &thread_data->td.td_susp_tied_tasks;
  kmp_stack_block_t *first_block;

  // set up the first block of the stack
  first_block = &task_stack->ts_first_block;
  task_stack->ts_top = (kmp_taskdata_t **)first_block;
  memset((void *)first_block, '\0',
         TASK_STACK_BLOCK_SIZE * sizeof(kmp_taskdata_t *));

  // initialize the stack to be empty
  task_stack->ts_entries = TASK_STACK_EMPTY;
  first_block->sb_next = NULL;
  first_block->sb_prev = NULL;
}

//  __kmp_free_task_stack: free the task stack when thread_data is destroyed.
//
//  gtid: global thread identifier for calling thread
//  thread_data: thread info for thread containing stack
static void __kmp_free_task_stack(kmp_int32 gtid,
                                  kmp_thread_data_t *thread_data) {
  kmp_task_stack_t *task_stack = &thread_data->td.td_susp_tied_tasks;
  kmp_stack_block_t *stack_block = &task_stack->ts_first_block;

  KMP_DEBUG_ASSERT(task_stack->ts_entries == TASK_STACK_EMPTY);
  // free from the second block of the stack
  while (stack_block != NULL) {
    kmp_stack_block_t *next_block = (stack_block) ? stack_block->sb_next : NULL;

    stack_block->sb_next = NULL;
    stack_block->sb_prev = NULL;
    if (stack_block != &task_stack->ts_first_block) {
      __kmp_thread_free(thread,
                        stack_block); // free the block, if not the first
    }
    stack_block = next_block;
  }
  // initialize the stack to be empty
  task_stack->ts_entries = 0;
  task_stack->ts_top = NULL;
}

//  __kmp_push_task_stack: Push the tied task onto the task stack.
//     Grow the stack if necessary by allocating another block.
//
//  gtid: global thread identifier for calling thread
//  thread: thread info for thread containing stack
//  tied_task: the task to push on the stack
static void __kmp_push_task_stack(kmp_int32 gtid, kmp_info_t *thread,
                                  kmp_taskdata_t *tied_task) {
  // GEH - need to consider what to do if tt_threads_data not allocated yet
  kmp_thread_data_t *thread_data =
      &thread->th.th_task_team->tt.tt_threads_data[__kmp_tid_from_gtid(gtid)];
  kmp_task_stack_t *task_stack = &thread_data->td.td_susp_tied_tasks;

  if (tied_task->td_flags.team_serial || tied_task->td_flags.tasking_ser) {
    return; // Don't push anything on stack if team or team tasks are serialized
  }

  KMP_DEBUG_ASSERT(tied_task->td_flags.tasktype == TASK_TIED);
  KMP_DEBUG_ASSERT(task_stack->ts_top != NULL);

  KA_TRACE(20,
           ("__kmp_push_task_stack(enter): GTID: %d; THREAD: %p; TASK: %p\n",
            gtid, thread, tied_task));
  // Store entry
  *(task_stack->ts_top) = tied_task;

  // Do bookkeeping for next push
  task_stack->ts_top++;
  task_stack->ts_entries++;

  if (task_stack->ts_entries & TASK_STACK_INDEX_MASK == 0) {
    // Find beginning of this task block
    kmp_stack_block_t *stack_block =
        (kmp_stack_block_t *)(task_stack->ts_top - TASK_STACK_BLOCK_SIZE);

    // Check if we already have a block
    if (stack_block->sb_next !=
        NULL) { // reset ts_top to beginning of next block
      task_stack->ts_top = &stack_block->sb_next->sb_block[0];
    } else { // Alloc new block and link it up
      kmp_stack_block_t *new_block = (kmp_stack_block_t *)__kmp_thread_calloc(
          thread, sizeof(kmp_stack_block_t));

      task_stack->ts_top = &new_block->sb_block[0];
      stack_block->sb_next = new_block;
      new_block->sb_prev = stack_block;
      new_block->sb_next = NULL;

      KA_TRACE(
          30,
          ("__kmp_push_task_stack(): GTID: %d; TASK: %p; Alloc new block: %p\n",
           gtid, tied_task, new_block));
    }
  }
  KA_TRACE(20, ("__kmp_push_task_stack(exit): GTID: %d; TASK: %p\n", gtid,
                tied_task));
}

//  __kmp_pop_task_stack: Pop the tied task from the task stack.  Don't return
//  the task, just check to make sure it matches the ending task passed in.
//
//  gtid: global thread identifier for the calling thread
//  thread: thread info structure containing stack
//  tied_task: the task popped off the stack
//  ending_task: the task that is ending (should match popped task)
static void __kmp_pop_task_stack(kmp_int32 gtid, kmp_info_t *thread,
                                 kmp_taskdata_t *ending_task) {
  // GEH - need to consider what to do if tt_threads_data not allocated yet
  kmp_thread_data_t *thread_data =
      &thread->th.th_task_team->tt_threads_data[__kmp_tid_from_gtid(gtid)];
  kmp_task_stack_t *task_stack = &thread_data->td.td_susp_tied_tasks;
  kmp_taskdata_t *tied_task;

  if (ending_task->td_flags.team_serial || ending_task->td_flags.tasking_ser) {
    // Don't pop anything from stack if team or team tasks are serialized
    return;
  }

  KMP_DEBUG_ASSERT(task_stack->ts_top != NULL);
  KMP_DEBUG_ASSERT(task_stack->ts_entries > 0);

  KA_TRACE(20, ("__kmp_pop_task_stack(enter): GTID: %d; THREAD: %p\n", gtid,
                thread));

  // fix up ts_top if we need to pop from previous block
  if (task_stack->ts_entries & TASK_STACK_INDEX_MASK == 0) {
    kmp_stack_block_t *stack_block = (kmp_stack_block_t *)(task_stack->ts_top);

    stack_block = stack_block->sb_prev;
    task_stack->ts_top = &stack_block->sb_block[TASK_STACK_BLOCK_SIZE];
  }

  // finish bookkeeping
  task_stack->ts_top--;
  task_stack->ts_entries--;

  tied_task = *(task_stack->ts_top);

  KMP_DEBUG_ASSERT(tied_task != NULL);
  KMP_DEBUG_ASSERT(tied_task->td_flags.tasktype == TASK_TIED);
  KMP_DEBUG_ASSERT(tied_task == ending_task); // If we built the stack correctly

  KA_TRACE(20, ("__kmp_pop_task_stack(exit): GTID: %d; TASK: %p\n", gtid,
                tied_task));
  return;
}
#endif /* BUILD_TIED_TASK_STACK */

// returns 1 if new task is allowed to execute, 0 otherwise
// checks Task Scheduling constraint (if requested) and
// mutexinoutset dependencies if any
static bool __kmp_task_is_allowed(int gtid, const kmp_int32 is_constrained,
                                  const kmp_taskdata_t *tasknew,
                                  const kmp_taskdata_t *taskcurr) {
  if (is_constrained && (tasknew->td_flags.tiedness == TASK_TIED)) {
    // Check if the candidate obeys the Task Scheduling Constraints (TSC)
    // only descendant of all deferred tied tasks can be scheduled, checking
    // the last one is enough, as it in turn is the descendant of all others
    kmp_taskdata_t *current = taskcurr->td_last_tied;
    KMP_DEBUG_ASSERT(current != NULL);
    // check if the task is not suspended on barrier
    if (current->td_flags.tasktype == TASK_EXPLICIT ||
        current->td_taskwait_thread > 0) { // <= 0 on barrier
      kmp_int32 level = current->td_level;
      kmp_taskdata_t *parent = tasknew->td_parent;
      while (parent != current && parent->td_level > level) {
        // check generation up to the level of the current task
        parent = parent->td_parent;
        KMP_DEBUG_ASSERT(parent != NULL);
      }
      if (parent != current)
        return false;
    }
  }
  // Check mutexinoutset dependencies, acquire locks
  kmp_depnode_t *node = tasknew->td_depnode;
  if (UNLIKELY(node && (node->dn.mtx_num_locks > 0))) {
    for (int i = 0; i < node->dn.mtx_num_locks; ++i) {
      KMP_DEBUG_ASSERT(node->dn.mtx_locks[i] != NULL);
      if (__kmp_test_lock(node->dn.mtx_locks[i], gtid))
        continue;
      // could not get the lock, release previous locks
      for (int j = i - 1; j >= 0; --j)
        __kmp_release_lock(node->dn.mtx_locks[j], gtid);
      return false;
    }
    // negative num_locks means all locks acquired successfully
    node->dn.mtx_num_locks = -node->dn.mtx_num_locks;
  }
  return true;
}

// __kmp_realloc_task_deque:
// Re-allocates a task deque for a particular thread, copies the content from
// the old deque and adjusts the necessary data structures relating to the
// deque. This operation must be done with the deque_lock being held
static void __kmp_realloc_task_deque(kmp_info_t *thread,
                                     kmp_thread_data_t *thread_data) {
  kmp_int32 size = TASK_DEQUE_SIZE(thread_data->td);
  KMP_DEBUG_ASSERT(TCR_4(thread_data->td.td_deque_ntasks) == size);
  kmp_int32 new_size = 2 * size;

  KE_TRACE(10, ("__kmp_realloc_task_deque: T#%d reallocating deque[from %d to "
                "%d] for thread_data %p\n",
                __kmp_gtid_from_thread(thread), size, new_size, thread_data));

  kmp_taskdata_t **new_deque =
      (kmp_taskdata_t **)__kmp_allocate(new_size * sizeof(kmp_taskdata_t *));

  int i, j;
  for (i = thread_data->td.td_deque_head, j = 0; j < size;
       i = (i + 1) & TASK_DEQUE_MASK(thread_data->td), j++)
    new_deque[j] = thread_data->td.td_deque[i];

  __kmp_free(thread_data->td.td_deque);

  thread_data->td.td_deque_head = 0;
  thread_data->td.td_deque_tail = size;
  thread_data->td.td_deque = new_deque;
  thread_data->td.td_deque_size = new_size;
}

static kmp_task_pri_t *__kmp_alloc_task_pri_list() {
  kmp_task_pri_t *l = (kmp_task_pri_t *)__kmp_allocate(sizeof(kmp_task_pri_t));
  kmp_thread_data_t *thread_data = &l->td;
  __kmp_init_bootstrap_lock(&thread_data->td.td_deque_lock);
  thread_data->td.td_deque_last_stolen = -1;
  KE_TRACE(20, ("__kmp_alloc_task_pri_list: T#%d allocating deque[%d] "
                "for thread_data %p\n",
                __kmp_get_gtid(), INITIAL_TASK_DEQUE_SIZE, thread_data));
  thread_data->td.td_deque = (kmp_taskdata_t **)__kmp_allocate(
      INITIAL_TASK_DEQUE_SIZE * sizeof(kmp_taskdata_t *));
  thread_data->td.td_deque_size = INITIAL_TASK_DEQUE_SIZE;
  return l;
}

// The function finds the deque of priority tasks with given priority, or
// allocates a new deque and put it into sorted (high -> low) list of deques.
// Deques of non-default priority tasks are shared between all threads in team,
// as opposed to per-thread deques of tasks with default priority.
// The function is called under the lock task_team->tt.tt_task_pri_lock.
static kmp_thread_data_t *
__kmp_get_priority_deque_data(kmp_task_team_t *task_team, kmp_int32 pri) {
  kmp_thread_data_t *thread_data;
  kmp_task_pri_t *lst = task_team->tt.tt_task_pri_list;
  if (lst->priority == pri) {
    // Found queue of tasks with given priority.
    thread_data = &lst->td;
  } else if (lst->priority < pri) {
    // All current priority queues contain tasks with lower priority.
    // Allocate new one for given priority tasks.
    kmp_task_pri_t *list = __kmp_alloc_task_pri_list();
    thread_data = &list->td;
    list->priority = pri;
    list->next = lst;
    task_team->tt.tt_task_pri_list = list;
  } else { // task_team->tt.tt_task_pri_list->priority > pri
    kmp_task_pri_t *next_queue = lst->next;
    while (next_queue && next_queue->priority > pri) {
      lst = next_queue;
      next_queue = lst->next;
    }
    // lst->priority > pri && (next == NULL || pri >= next->priority)
    if (next_queue == NULL) {
      // No queue with pri priority, need to allocate new one.
      kmp_task_pri_t *list = __kmp_alloc_task_pri_list();
      thread_data = &list->td;
      list->priority = pri;
      list->next = NULL;
      lst->next = list;
    } else if (next_queue->priority == pri) {
      // Found queue of tasks with given priority.
      thread_data = &next_queue->td;
    } else { // lst->priority > pri > next->priority
      // insert newly allocated between existed queues
      kmp_task_pri_t *list = __kmp_alloc_task_pri_list();
      thread_data = &list->td;
      list->priority = pri;
      list->next = next_queue;
      lst->next = list;
    }
  }
  return thread_data;
}

//  __kmp_push_priority_task: Add a task to the team's priority task deque
static kmp_int32 __kmp_push_priority_task(kmp_int32 gtid, kmp_info_t *thread,
                                          kmp_taskdata_t *taskdata,
                                          kmp_task_team_t *task_team,
                                          kmp_int32 pri) {
  kmp_thread_data_t *thread_data = NULL;
  KA_TRACE(20,
           ("__kmp_push_priority_task: T#%d trying to push task %p, pri %d.\n",
            gtid, taskdata, pri));

  // Find task queue specific to priority value
  kmp_task_pri_t *lst = task_team->tt.tt_task_pri_list;
  if (UNLIKELY(lst == NULL)) {
    __kmp_acquire_bootstrap_lock(&task_team->tt.tt_task_pri_lock);
    if (task_team->tt.tt_task_pri_list == NULL) {
      // List of queues is still empty, allocate one.
      kmp_task_pri_t *list = __kmp_alloc_task_pri_list();
      thread_data = &list->td;
      list->priority = pri;
      list->next = NULL;
      task_team->tt.tt_task_pri_list = list;
    } else {
      // Other thread initialized a queue. Check if it fits and get thread_data.
      thread_data = __kmp_get_priority_deque_data(task_team, pri);
    }
    __kmp_release_bootstrap_lock(&task_team->tt.tt_task_pri_lock);
  } else {
    if (lst->priority == pri) {
      // Found queue of tasks with given priority.
      thread_data = &lst->td;
    } else {
      __kmp_acquire_bootstrap_lock(&task_team->tt.tt_task_pri_lock);
      thread_data = __kmp_get_priority_deque_data(task_team, pri);
      __kmp_release_bootstrap_lock(&task_team->tt.tt_task_pri_lock);
    }
  }
  KMP_DEBUG_ASSERT(thread_data);

  __kmp_acquire_bootstrap_lock(&thread_data->td.td_deque_lock);
  // Check if deque is full
  if (TCR_4(thread_data->td.td_deque_ntasks) >=
      TASK_DEQUE_SIZE(thread_data->td)) {
    if (__kmp_enable_task_throttling &&
        __kmp_task_is_allowed(gtid, __kmp_task_stealing_constraint, taskdata,
                              thread->th.th_current_task)) {
      __kmp_release_bootstrap_lock(&thread_data->td.td_deque_lock);
      KA_TRACE(20, ("__kmp_push_priority_task: T#%d deque is full; returning "
                    "TASK_NOT_PUSHED for task %p\n",
                    gtid, taskdata));
      return TASK_NOT_PUSHED;
    } else {
      // expand deque to push the task which is not allowed to execute
      __kmp_realloc_task_deque(thread, thread_data);
    }
  }
  KMP_DEBUG_ASSERT(TCR_4(thread_data->td.td_deque_ntasks) <
                   TASK_DEQUE_SIZE(thread_data->td));
  // Push taskdata.
  thread_data->td.td_deque[thread_data->td.td_deque_tail] = taskdata;
  // Wrap index.
  thread_data->td.td_deque_tail =
      (thread_data->td.td_deque_tail + 1) & TASK_DEQUE_MASK(thread_data->td);
  TCW_4(thread_data->td.td_deque_ntasks,
        TCR_4(thread_data->td.td_deque_ntasks) + 1); // Adjust task count
  KMP_FSYNC_RELEASING(thread->th.th_current_task); // releasing self
  KMP_FSYNC_RELEASING(taskdata); // releasing child
  KA_TRACE(20, ("__kmp_push_priority_task: T#%d returning "
                "TASK_SUCCESSFULLY_PUSHED: task=%p ntasks=%d head=%u tail=%u\n",
                gtid, taskdata, thread_data->td.td_deque_ntasks,
                thread_data->td.td_deque_head, thread_data->td.td_deque_tail));
  __kmp_release_bootstrap_lock(&thread_data->td.td_deque_lock);
  task_team->tt.tt_num_task_pri++; // atomic inc
  return TASK_SUCCESSFULLY_PUSHED;
}

//  __kmp_push_task: Add a task to the thread's deque
static kmp_int32 __kmp_push_task(kmp_int32 gtid, kmp_task_t *task) {
  kmp_info_t *thread = __kmp_threads[gtid];
  kmp_taskdata_t *taskdata = KMP_TASK_TO_TASKDATA(task);

  // If we encounter a hidden helper task, and the current thread is not a
  // hidden helper thread, we have to give the task to any hidden helper thread
  // starting from its shadow one.
  if (UNLIKELY(taskdata->td_flags.hidden_helper &&
               !KMP_HIDDEN_HELPER_THREAD(gtid))) {
    kmp_int32 shadow_gtid = KMP_GTID_TO_SHADOW_GTID(gtid);
    __kmpc_give_task(task, __kmp_tid_from_gtid(shadow_gtid));
    // Signal the hidden helper threads.
    __kmp_hidden_helper_worker_thread_signal();
    return TASK_SUCCESSFULLY_PUSHED;
  }

  kmp_task_team_t *task_team = thread->th.th_task_team;
  kmp_int32 tid = __kmp_tid_from_gtid(gtid);
  kmp_thread_data_t *thread_data;

  KA_TRACE(20,
           ("__kmp_push_task: T#%d trying to push task %p.\n", gtid, taskdata));

  if (UNLIKELY(taskdata->td_flags.tiedness == TASK_UNTIED)) {
    // untied task needs to increment counter so that the task structure is not
    // freed prematurely
    kmp_int32 counter = 1 + KMP_ATOMIC_INC(&taskdata->td_untied_count);
    KMP_DEBUG_USE_VAR(counter);
    KA_TRACE(
        20,
        ("__kmp_push_task: T#%d untied_count (%d) incremented for task %p\n",
         gtid, counter, taskdata));
  }

  // The first check avoids building task_team thread data if serialized
  if (UNLIKELY(taskdata->td_flags.task_serial)) {
    KA_TRACE(20, ("__kmp_push_task: T#%d team serialized; returning "
                  "TASK_NOT_PUSHED for task %p\n",
                  gtid, taskdata));
    return TASK_NOT_PUSHED;
  }

  // Now that serialized tasks have returned, we can assume that we are not in
  // immediate exec mode
  KMP_DEBUG_ASSERT(__kmp_tasking_mode != tskm_immediate_exec);
  if (UNLIKELY(!KMP_TASKING_ENABLED(task_team))) {
    __kmp_enable_tasking(task_team, thread);
  }
  KMP_DEBUG_ASSERT(TCR_4(task_team->tt.tt_found_tasks) == TRUE);
  KMP_DEBUG_ASSERT(TCR_PTR(task_team->tt.tt_threads_data) != NULL);

  if (taskdata->td_flags.priority_specified && task->data2.priority > 0 &&
      __kmp_max_task_priority > 0) {
    int pri = KMP_MIN(task->data2.priority, __kmp_max_task_priority);
    return __kmp_push_priority_task(gtid, thread, taskdata, task_team, pri);
  }

  // Find tasking deque specific to encountering thread
  thread_data = &task_team->tt.tt_threads_data[tid];

  // No lock needed since only owner can allocate. If the task is hidden_helper,
  // we don't need it either because we have initialized the dequeue for hidden
  // helper thread data.
  if (UNLIKELY(thread_data->td.td_deque == NULL)) {
    __kmp_alloc_task_deque(thread, thread_data);
  }

  int locked = 0;
  // Check if deque is full
  if (TCR_4(thread_data->td.td_deque_ntasks) >=
      TASK_DEQUE_SIZE(thread_data->td)) {
    if (__kmp_enable_task_throttling &&
        __kmp_task_is_allowed(gtid, __kmp_task_stealing_constraint, taskdata,
                              thread->th.th_current_task)) {
      KA_TRACE(20, ("__kmp_push_task: T#%d deque is full; returning "
                    "TASK_NOT_PUSHED for task %p\n",
                    gtid, taskdata));
      return TASK_NOT_PUSHED;
    } else {
      __kmp_acquire_bootstrap_lock(&thread_data->td.td_deque_lock);
      locked = 1;
      if (TCR_4(thread_data->td.td_deque_ntasks) >=
          TASK_DEQUE_SIZE(thread_data->td)) {
        // expand deque to push the task which is not allowed to execute
        __kmp_realloc_task_deque(thread, thread_data);
      }
    }
  }
  // Lock the deque for the task push operation
  if (!locked) {
    __kmp_acquire_bootstrap_lock(&thread_data->td.td_deque_lock);
    // Need to recheck as we can get a proxy task from thread outside of OpenMP
    if (TCR_4(thread_data->td.td_deque_ntasks) >=
        TASK_DEQUE_SIZE(thread_data->td)) {
      if (__kmp_enable_task_throttling &&
          __kmp_task_is_allowed(gtid, __kmp_task_stealing_constraint, taskdata,
                                thread->th.th_current_task)) {
        __kmp_release_bootstrap_lock(&thread_data->td.td_deque_lock);
        KA_TRACE(20, ("__kmp_push_task: T#%d deque is full on 2nd check; "
                      "returning TASK_NOT_PUSHED for task %p\n",
                      gtid, taskdata));
        return TASK_NOT_PUSHED;
      } else {
        // expand deque to push the task which is not allowed to execute
        __kmp_realloc_task_deque(thread, thread_data);
      }
    }
  }
  // Must have room since no thread can add tasks but calling thread
  KMP_DEBUG_ASSERT(TCR_4(thread_data->td.td_deque_ntasks) <
                   TASK_DEQUE_SIZE(thread_data->td));

  thread_data->td.td_deque[thread_data->td.td_deque_tail] =
      taskdata; // Push taskdata
  // Wrap index.
  thread_data->td.td_deque_tail =
      (thread_data->td.td_deque_tail + 1) & TASK_DEQUE_MASK(thread_data->td);
  TCW_4(thread_data->td.td_deque_ntasks,
        TCR_4(thread_data->td.td_deque_ntasks) + 1); // Adjust task count
  KMP_FSYNC_RELEASING(thread->th.th_current_task); // releasing self
  KMP_FSYNC_RELEASING(taskdata); // releasing child
  KA_TRACE(20, ("__kmp_push_task: T#%d returning TASK_SUCCESSFULLY_PUSHED: "
                "task=%p ntasks=%d head=%u tail=%u\n",
                gtid, taskdata, thread_data->td.td_deque_ntasks,
                thread_data->td.td_deque_head, thread_data->td.td_deque_tail));

  __kmp_release_bootstrap_lock(&thread_data->td.td_deque_lock);

  return TASK_SUCCESSFULLY_PUSHED;
}

// __kmp_pop_current_task_from_thread: set up current task from called thread
// when team ends
//
// this_thr: thread structure to set current_task in.
void __kmp_pop_current_task_from_thread(kmp_info_t *this_thr) {
  KF_TRACE(10, ("__kmp_pop_current_task_from_thread(enter): T#%d "
                "this_thread=%p, curtask=%p, "
                "curtask_parent=%p\n",
                0, this_thr, this_thr->th.th_current_task,
                this_thr->th.th_current_task->td_parent));

  this_thr->th.th_current_task = this_thr->th.th_current_task->td_parent;

  KF_TRACE(10, ("__kmp_pop_current_task_from_thread(exit): T#%d "
                "this_thread=%p, curtask=%p, "
                "curtask_parent=%p\n",
                0, this_thr, this_thr->th.th_current_task,
                this_thr->th.th_current_task->td_parent));
}

// __kmp_push_current_task_to_thread: set up current task in called thread for a
// new team
//
// this_thr: thread structure to set up
// team: team for implicit task data
// tid: thread within team to set up
void __kmp_push_current_task_to_thread(kmp_info_t *this_thr, kmp_team_t *team,
                                       int tid) {
  // current task of the thread is a parent of the new just created implicit
  // tasks of new team
  KF_TRACE(10, ("__kmp_push_current_task_to_thread(enter): T#%d this_thread=%p "
                "curtask=%p "
                "parent_task=%p\n",
                tid, this_thr, this_thr->th.th_current_task,
                team->t.t_implicit_task_taskdata[tid].td_parent));

  KMP_DEBUG_ASSERT(this_thr != NULL);

  if (tid == 0) {
    if (this_thr->th.th_current_task != &team->t.t_implicit_task_taskdata[0]) {
      team->t.t_implicit_task_taskdata[0].td_parent =
          this_thr->th.th_current_task;
      this_thr->th.th_current_task = &team->t.t_implicit_task_taskdata[0];
    }
  } else {
    team->t.t_implicit_task_taskdata[tid].td_parent =
        team->t.t_implicit_task_taskdata[0].td_parent;
    this_thr->th.th_current_task = &team->t.t_implicit_task_taskdata[tid];
  }

  KF_TRACE(10, ("__kmp_push_current_task_to_thread(exit): T#%d this_thread=%p "
                "curtask=%p "
                "parent_task=%p\n",
                tid, this_thr, this_thr->th.th_current_task,
                team->t.t_implicit_task_taskdata[tid].td_parent));
}

// __kmp_task_start: bookkeeping for a task starting execution
//
// GTID: global thread id of calling thread
// task: task starting execution
// current_task: task suspending
static void __kmp_task_start(kmp_int32 gtid, kmp_task_t *task,
                             kmp_taskdata_t *current_task) {
  kmp_taskdata_t *taskdata = KMP_TASK_TO_TASKDATA(task);
  kmp_info_t *thread = __kmp_threads[gtid];

  KA_TRACE(10,
           ("__kmp_task_start(enter): T#%d starting task %p: current_task=%p\n",
            gtid, taskdata, current_task));

  KMP_DEBUG_ASSERT(taskdata->td_flags.tasktype == TASK_EXPLICIT);

  // mark currently executing task as suspended
  // TODO: GEH - make sure root team implicit task is initialized properly.
  // KMP_DEBUG_ASSERT( current_task -> td_flags.executing == 1 );
  current_task->td_flags.executing = 0;

// Add task to stack if tied
#ifdef BUILD_TIED_TASK_STACK
  if (taskdata->td_flags.tiedness == TASK_TIED) {
    __kmp_push_task_stack(gtid, thread, taskdata);
  }
#endif /* BUILD_TIED_TASK_STACK */

  // mark starting task as executing and as current task
  thread->th.th_current_task = taskdata;

  KMP_DEBUG_ASSERT(taskdata->td_flags.started == 0 ||
                   taskdata->td_flags.tiedness == TASK_UNTIED);
  KMP_DEBUG_ASSERT(taskdata->td_flags.executing == 0 ||
                   taskdata->td_flags.tiedness == TASK_UNTIED);
  taskdata->td_flags.started = 1;
  taskdata->td_flags.executing = 1;
  KMP_DEBUG_ASSERT(taskdata->td_flags.complete == 0);
  KMP_DEBUG_ASSERT(taskdata->td_flags.freed == 0);

  // GEH TODO: shouldn't we pass some sort of location identifier here?
  // APT: yes, we will pass location here.
  // need to store current thread state (in a thread or taskdata structure)
  // before setting work_state, otherwise wrong state is set after end of task

  KA_TRACE(10, ("__kmp_task_start(exit): T#%d task=%p\n", gtid, taskdata));

  return;
}

#if OMPT_SUPPORT
//------------------------------------------------------------------------------
// __ompt_task_init:
//   Initialize OMPT fields maintained by a task. This will only be called after
//   ompt_start_tool, so we already know whether ompt is enabled or not.

static inline void __ompt_task_init(kmp_taskdata_t *task, int tid) {
  // The calls to __ompt_task_init already have the ompt_enabled condition.
  task->ompt_task_info.task_data.value = 0;
  task->ompt_task_info.frame.exit_frame = ompt_data_none;
  task->ompt_task_info.frame.enter_frame = ompt_data_none;
  task->ompt_task_info.frame.exit_frame_flags =
      ompt_frame_runtime | ompt_frame_framepointer;
  task->ompt_task_info.frame.enter_frame_flags =
      ompt_frame_runtime | ompt_frame_framepointer;
  task->ompt_task_info.dispatch_chunk.start = 0;
  task->ompt_task_info.dispatch_chunk.iterations = 0;
}

// __ompt_task_start:
//   Build and trigger task-begin event
static inline void __ompt_task_start(kmp_task_t *task,
                                     kmp_taskdata_t *current_task,
                                     kmp_int32 gtid) {
  kmp_taskdata_t *taskdata = KMP_TASK_TO_TASKDATA(task);
  ompt_task_status_t status = ompt_task_switch;
  if (__kmp_threads[gtid]->th.ompt_thread_info.ompt_task_yielded) {
    status = ompt_task_yield;
    __kmp_threads[gtid]->th.ompt_thread_info.ompt_task_yielded = 0;
  }
  /* let OMPT know that we're about to run this task */
  if (ompt_enabled.ompt_callback_task_schedule) {
    ompt_callbacks.ompt_callback(ompt_callback_task_schedule)(
        &(current_task->ompt_task_info.task_data), status,
        &(taskdata->ompt_task_info.task_data));
  }
  taskdata->ompt_task_info.scheduling_parent = current_task;
}

// __ompt_task_finish:
//   Build and trigger final task-schedule event
static inline void __ompt_task_finish(kmp_task_t *task,
                                      kmp_taskdata_t *resumed_task,
                                      ompt_task_status_t status) {
  if (ompt_enabled.ompt_callback_task_schedule) {
    kmp_taskdata_t *taskdata = KMP_TASK_TO_TASKDATA(task);
    if (__kmp_omp_cancellation && taskdata->td_taskgroup &&
        taskdata->td_taskgroup->cancel_request == cancel_taskgroup) {
      status = ompt_task_cancel;
    }

    /* let OMPT know that we're returning to the callee task */
    ompt_callbacks.ompt_callback(ompt_callback_task_schedule)(
        &(taskdata->ompt_task_info.task_data), status,
        (resumed_task ? &(resumed_task->ompt_task_info.task_data) : NULL));
  }
}
#endif

template <bool ompt>
static void __kmpc_omp_task_begin_if0_template(ident_t *loc_ref, kmp_int32 gtid,
                                               kmp_task_t *task,
                                               void *frame_address,
                                               void *return_address) {
  kmp_taskdata_t *taskdata = KMP_TASK_TO_TASKDATA(task);
  kmp_taskdata_t *current_task = __kmp_threads[gtid]->th.th_current_task;

  KA_TRACE(10, ("__kmpc_omp_task_begin_if0(enter): T#%d loc=%p task=%p "
                "current_task=%p\n",
                gtid, loc_ref, taskdata, current_task));

  if (UNLIKELY(taskdata->td_flags.tiedness == TASK_UNTIED)) {
    // untied task needs to increment counter so that the task structure is not
    // freed prematurely
    kmp_int32 counter = 1 + KMP_ATOMIC_INC(&taskdata->td_untied_count);
    KMP_DEBUG_USE_VAR(counter);
    KA_TRACE(20, ("__kmpc_omp_task_begin_if0: T#%d untied_count (%d) "
                  "incremented for task %p\n",
                  gtid, counter, taskdata));
  }

  taskdata->td_flags.task_serial =
      1; // Execute this task immediately, not deferred.
  __kmp_task_start(gtid, task, current_task);

#if OMPT_SUPPORT
  if (ompt) {
    if (current_task->ompt_task_info.frame.enter_frame.ptr == NULL) {
      current_task->ompt_task_info.frame.enter_frame.ptr =
          taskdata->ompt_task_info.frame.exit_frame.ptr = frame_address;
      current_task->ompt_task_info.frame.enter_frame_flags =
          taskdata->ompt_task_info.frame.exit_frame_flags =
              ompt_frame_application | ompt_frame_framepointer;
    }
    if (ompt_enabled.ompt_callback_task_create) {
      ompt_task_info_t *parent_info = &(current_task->ompt_task_info);
      ompt_callbacks.ompt_callback(ompt_callback_task_create)(
          &(parent_info->task_data), &(parent_info->frame),
          &(taskdata->ompt_task_info.task_data),
          ompt_task_explicit | TASK_TYPE_DETAILS_FORMAT(taskdata), 0,
          return_address);
    }
    __ompt_task_start(task, current_task, gtid);
  }
#endif // OMPT_SUPPORT

  KA_TRACE(10, ("__kmpc_omp_task_begin_if0(exit): T#%d loc=%p task=%p,\n", gtid,
                loc_ref, taskdata));
}

#if OMPT_SUPPORT
OMPT_NOINLINE
static void __kmpc_omp_task_begin_if0_ompt(ident_t *loc_ref, kmp_int32 gtid,
                                           kmp_task_t *task,
                                           void *frame_address,
                                           void *return_address) {
  __kmpc_omp_task_begin_if0_template<true>(loc_ref, gtid, task, frame_address,
                                           return_address);
}
#endif // OMPT_SUPPORT

// __kmpc_omp_task_begin_if0: report that a given serialized task has started
// execution
//
// loc_ref: source location information; points to beginning of task block.
// gtid: global thread number.
// task: task thunk for the started task.
void __kmpc_omp_task_begin_if0(ident_t *loc_ref, kmp_int32 gtid,
                               kmp_task_t *task) {
#if OMPT_SUPPORT
  if (UNLIKELY(ompt_enabled.enabled)) {
    OMPT_STORE_RETURN_ADDRESS(gtid);
    __kmpc_omp_task_begin_if0_ompt(loc_ref, gtid, task,
                                   OMPT_GET_FRAME_ADDRESS(1),
                                   OMPT_LOAD_RETURN_ADDRESS(gtid));
    return;
  }
#endif
  __kmpc_omp_task_begin_if0_template<false>(loc_ref, gtid, task, NULL, NULL);
}

#ifdef TASK_UNUSED
// __kmpc_omp_task_begin: report that a given task has started execution
// NEVER GENERATED BY COMPILER, DEPRECATED!!!
void __kmpc_omp_task_begin(ident_t *loc_ref, kmp_int32 gtid, kmp_task_t *task) {
  kmp_taskdata_t *current_task = __kmp_threads[gtid]->th.th_current_task;

  KA_TRACE(
      10,
      ("__kmpc_omp_task_begin(enter): T#%d loc=%p task=%p current_task=%p\n",
       gtid, loc_ref, KMP_TASK_TO_TASKDATA(task), current_task));

  __kmp_task_start(gtid, task, current_task);

  KA_TRACE(10, ("__kmpc_omp_task_begin(exit): T#%d loc=%p task=%p,\n", gtid,
                loc_ref, KMP_TASK_TO_TASKDATA(task)));
  return;
}
#endif // TASK_UNUSED

// __kmp_free_task: free the current task space and the space for shareds
//
// gtid: Global thread ID of calling thread
// taskdata: task to free
// thread: thread data structure of caller
static void __kmp_free_task(kmp_int32 gtid, kmp_taskdata_t *taskdata,
                            kmp_info_t *thread) {
  KA_TRACE(30, ("__kmp_free_task: T#%d freeing data from task %p\n", gtid,
                taskdata));

  // Check to make sure all flags and counters have the correct values
  KMP_DEBUG_ASSERT(taskdata->td_flags.tasktype == TASK_EXPLICIT);
  KMP_DEBUG_ASSERT(taskdata->td_flags.executing == 0);
  KMP_DEBUG_ASSERT(taskdata->td_flags.complete == 1);
  KMP_DEBUG_ASSERT(taskdata->td_flags.freed == 0);
  KMP_DEBUG_ASSERT(taskdata->td_allocated_child_tasks == 0 ||
                   taskdata->td_flags.task_serial == 1);
  KMP_DEBUG_ASSERT(taskdata->td_incomplete_child_tasks == 0);
  kmp_task_t *task = KMP_TASKDATA_TO_TASK(taskdata);
  // Clear data to not be re-used later by mistake.
  task->data1.destructors = NULL;
  task->data2.priority = 0;

  taskdata->td_flags.freed = 1;
// deallocate the taskdata and shared variable blocks associated with this task
#if USE_FAST_MEMORY
  __kmp_fast_free(thread, taskdata);
#else /* ! USE_FAST_MEMORY */
  __kmp_thread_free(thread, taskdata);
#endif
  KA_TRACE(20, ("__kmp_free_task: T#%d freed task %p\n", gtid, taskdata));
}

// __kmp_free_task_and_ancestors: free the current task and ancestors without
// children
//
// gtid: Global thread ID of calling thread
// taskdata: task to free
// thread: thread data structure of caller
static void __kmp_free_task_and_ancestors(kmp_int32 gtid,
                                          kmp_taskdata_t *taskdata,
                                          kmp_info_t *thread) {
  // Proxy tasks must always be allowed to free their parents
  // because they can be run in background even in serial mode.
  kmp_int32 team_serial =
      (taskdata->td_flags.team_serial || taskdata->td_flags.tasking_ser) &&
      !taskdata->td_flags.proxy;
  KMP_DEBUG_ASSERT(taskdata->td_flags.tasktype == TASK_EXPLICIT);

  kmp_int32 children = KMP_ATOMIC_DEC(&taskdata->td_allocated_child_tasks) - 1;
  KMP_DEBUG_ASSERT(children >= 0);

  // Now, go up the ancestor tree to see if any ancestors can now be freed.
  while (children == 0) {
    kmp_taskdata_t *parent_taskdata = taskdata->td_parent;

    KA_TRACE(20, ("__kmp_free_task_and_ancestors(enter): T#%d task %p complete "
                  "and freeing itself\n",
                  gtid, taskdata));

    // --- Deallocate my ancestor task ---
    __kmp_free_task(gtid, taskdata, thread);

    taskdata = parent_taskdata;

    if (team_serial)
      return;
    // Stop checking ancestors at implicit task instead of walking up ancestor
    // tree to avoid premature deallocation of ancestors.
    if (taskdata->td_flags.tasktype == TASK_IMPLICIT) {
      if (taskdata->td_dephash) { // do we need to cleanup dephash?
        int children = KMP_ATOMIC_LD_ACQ(&taskdata->td_incomplete_child_tasks);
        kmp_tasking_flags_t flags_old = taskdata->td_flags;
        if (children == 0 && flags_old.complete == 1) {
          kmp_tasking_flags_t flags_new = flags_old;
          flags_new.complete = 0;
          if (KMP_COMPARE_AND_STORE_ACQ32(
                  RCAST(kmp_int32 *, &taskdata->td_flags),
                  *RCAST(kmp_int32 *, &flags_old),
                  *RCAST(kmp_int32 *, &flags_new))) {
            KA_TRACE(100, ("__kmp_free_task_and_ancestors: T#%d cleans "
                           "dephash of implicit task %p\n",
                           gtid, taskdata));
            // cleanup dephash of finished implicit task
            __kmp_dephash_free_entries(thread, taskdata->td_dephash);
          }
        }
      }
      return;
    }
    // Predecrement simulated by "- 1" calculation
    children = KMP_ATOMIC_DEC(&taskdata->td_allocated_child_tasks) - 1;
    KMP_DEBUG_ASSERT(children >= 0);
  }

  KA_TRACE(
      20, ("__kmp_free_task_and_ancestors(exit): T#%d task %p has %d children; "
           "not freeing it yet\n",
           gtid, taskdata, children));
}

// Only need to keep track of child task counts if any of the following:
// 1. team parallel and tasking not serialized;
// 2. it is a proxy or detachable or hidden helper task
// 3. the children counter of its parent task is greater than 0.
// The reason for the 3rd one is for serialized team that found detached task,
// hidden helper task, T. In this case, the execution of T is still deferred,
// and it is also possible that a regular task depends on T. In this case, if we
// don't track the children, task synchronization will be broken.
static bool __kmp_track_children_task(kmp_taskdata_t *taskdata) {
  kmp_tasking_flags_t flags = taskdata->td_flags;
  bool ret = !(flags.team_serial || flags.tasking_ser);
  ret = ret || flags.proxy == TASK_PROXY ||
        flags.detachable == TASK_DETACHABLE || flags.hidden_helper;
  ret = ret ||
        KMP_ATOMIC_LD_ACQ(&taskdata->td_parent->td_incomplete_child_tasks) > 0;
  return ret;
}

// __kmp_task_finish: bookkeeping to do when a task finishes execution
//
// gtid: global thread ID for calling thread
// task: task to be finished
// resumed_task: task to be resumed.  (may be NULL if task is serialized)
//
// template<ompt>: effectively ompt_enabled.enabled!=0
// the version with ompt=false is inlined, allowing to optimize away all ompt
// code in this case
template <bool ompt>
static void __kmp_task_finish(kmp_int32 gtid, kmp_task_t *task,
                              kmp_taskdata_t *resumed_task) {
  kmp_taskdata_t *taskdata = KMP_TASK_TO_TASKDATA(task);
  kmp_info_t *thread = __kmp_threads[gtid];
  kmp_task_team_t *task_team =
      thread->th.th_task_team; // might be NULL for serial teams...
#if KMP_DEBUG
  kmp_int32 children = 0;
#endif
  KA_TRACE(10, ("__kmp_task_finish(enter): T#%d finishing task %p and resuming "
                "task %p\n",
                gtid, taskdata, resumed_task));

  KMP_DEBUG_ASSERT(taskdata->td_flags.tasktype == TASK_EXPLICIT);

// Pop task from stack if tied
#ifdef BUILD_TIED_TASK_STACK
  if (taskdata->td_flags.tiedness == TASK_TIED) {
    __kmp_pop_task_stack(gtid, thread, taskdata);
  }
#endif /* BUILD_TIED_TASK_STACK */

  if (UNLIKELY(taskdata->td_flags.tiedness == TASK_UNTIED)) {
    // untied task needs to check the counter so that the task structure is not
    // freed prematurely
    kmp_int32 counter = KMP_ATOMIC_DEC(&taskdata->td_untied_count) - 1;
    KA_TRACE(
        20,
        ("__kmp_task_finish: T#%d untied_count (%d) decremented for task %p\n",
         gtid, counter, taskdata));
    if (counter > 0) {
      // untied task is not done, to be continued possibly by other thread, do
      // not free it now
      if (resumed_task == NULL) {
        KMP_DEBUG_ASSERT(taskdata->td_flags.task_serial);
        resumed_task = taskdata->td_parent; // In a serialized task, the resumed
        // task is the parent
      }
      thread->th.th_current_task = resumed_task; // restore current_task
      resumed_task->td_flags.executing = 1; // resume previous task
      KA_TRACE(10, ("__kmp_task_finish(exit): T#%d partially done task %p, "
                    "resuming task %p\n",
                    gtid, taskdata, resumed_task));
      return;
    }
  }

  // bookkeeping for resuming task:
  // GEH - note tasking_ser => task_serial
  KMP_DEBUG_ASSERT(
      (taskdata->td_flags.tasking_ser || taskdata->td_flags.task_serial) ==
      taskdata->td_flags.task_serial);
  if (taskdata->td_flags.task_serial) {
    if (resumed_task == NULL) {
      resumed_task = taskdata->td_parent; // In a serialized task, the resumed
      // task is the parent
    }
  } else {
    KMP_DEBUG_ASSERT(resumed_task !=
                     NULL); // verify that resumed task is passed as argument
  }

  /* If the tasks' destructor thunk flag has been set, we need to invoke the
     destructor thunk that has been generated by the compiler. The code is
     placed here, since at this point other tasks might have been released
     hence overlapping the destructor invocations with some other work in the
     released tasks.  The OpenMP spec is not specific on when the destructors
     are invoked, so we should be free to choose. */
  if (UNLIKELY(taskdata->td_flags.destructors_thunk)) {
    kmp_routine_entry_t destr_thunk = task->data1.destructors;
    KMP_ASSERT(destr_thunk);
    destr_thunk(gtid, task);
  }

  KMP_DEBUG_ASSERT(taskdata->td_flags.complete == 0);
  KMP_DEBUG_ASSERT(taskdata->td_flags.started == 1);
  KMP_DEBUG_ASSERT(taskdata->td_flags.freed == 0);

  bool detach = false;
  if (UNLIKELY(taskdata->td_flags.detachable == TASK_DETACHABLE)) {
    if (taskdata->td_allow_completion_event.type ==
        KMP_EVENT_ALLOW_COMPLETION) {
      // event hasn't been fulfilled yet. Try to detach task.
      __kmp_acquire_tas_lock(&taskdata->td_allow_completion_event.lock, gtid);
      if (taskdata->td_allow_completion_event.type ==
          KMP_EVENT_ALLOW_COMPLETION) {
        // task finished execution
        KMP_DEBUG_ASSERT(taskdata->td_flags.executing == 1);
        taskdata->td_flags.executing = 0; // suspend the finishing task

#if OMPT_SUPPORT
        // For a detached task, which is not completed, we switch back
        // the omp_fulfill_event signals completion
        // locking is necessary to avoid a race with ompt_task_late_fulfill
        if (ompt)
          __ompt_task_finish(task, resumed_task, ompt_task_detach);
#endif

        // no access to taskdata after this point!
        // __kmp_fulfill_event might free taskdata at any time from now

        taskdata->td_flags.proxy = TASK_PROXY; // proxify!
        detach = true;
      }
      __kmp_release_tas_lock(&taskdata->td_allow_completion_event.lock, gtid);
    }
  }

  if (!detach) {
    taskdata->td_flags.complete = 1; // mark the task as completed

#if OMPT_SUPPORT
    // This is not a detached task, we are done here
    if (ompt)
      __ompt_task_finish(task, resumed_task, ompt_task_complete);
#endif
    // TODO: What would be the balance between the conditions in the function
    // and an atomic operation?
    if (__kmp_track_children_task(taskdata)) {
      __kmp_release_deps(gtid, taskdata);
      // Predecrement simulated by "- 1" calculation
#if KMP_DEBUG
      children = -1 +
#endif
          KMP_ATOMIC_DEC(&taskdata->td_parent->td_incomplete_child_tasks);
      KMP_DEBUG_ASSERT(children >= 0);
      if (taskdata->td_taskgroup)
        KMP_ATOMIC_DEC(&taskdata->td_taskgroup->count);
    } else if (task_team && (task_team->tt.tt_found_proxy_tasks ||
                             task_team->tt.tt_hidden_helper_task_encountered)) {
      // if we found proxy or hidden helper tasks there could exist a dependency
      // chain with the proxy task as origin
      __kmp_release_deps(gtid, taskdata);
    }
    // td_flags.executing must be marked as 0 after __kmp_release_deps has been
    // called. Othertwise, if a task is executed immediately from the
    // release_deps code, the flag will be reset to 1 again by this same
    // function
    KMP_DEBUG_ASSERT(taskdata->td_flags.executing == 1);
    taskdata->td_flags.executing = 0; // suspend the finishing task
  }

  KA_TRACE(
      20, ("__kmp_task_finish: T#%d finished task %p, %d incomplete children\n",
           gtid, taskdata, children));

  // Free this task and then ancestor tasks if they have no children.
  // Restore th_current_task first as suggested by John:
  // johnmc: if an asynchronous inquiry peers into the runtime system
  // it doesn't see the freed task as the current task.
  thread->th.th_current_task = resumed_task;
  if (!detach)
    __kmp_free_task_and_ancestors(gtid, taskdata, thread);

  // TODO: GEH - make sure root team implicit task is initialized properly.
  // KMP_DEBUG_ASSERT( resumed_task->td_flags.executing == 0 );
  resumed_task->td_flags.executing = 1; // resume previous task

  KA_TRACE(
      10, ("__kmp_task_finish(exit): T#%d finished task %p, resuming task %p\n",
           gtid, taskdata, resumed_task));

  return;
}

template <bool ompt>
static void __kmpc_omp_task_complete_if0_template(ident_t *loc_ref,
                                                  kmp_int32 gtid,
                                                  kmp_task_t *task) {
  KA_TRACE(10, ("__kmpc_omp_task_complete_if0(enter): T#%d loc=%p task=%p\n",
                gtid, loc_ref, KMP_TASK_TO_TASKDATA(task)));
  KMP_DEBUG_ASSERT(gtid >= 0);
  // this routine will provide task to resume
  __kmp_task_finish<ompt>(gtid, task, NULL);

  KA_TRACE(10, ("__kmpc_omp_task_complete_if0(exit): T#%d loc=%p task=%p\n",
                gtid, loc_ref, KMP_TASK_TO_TASKDATA(task)));

#if OMPT_SUPPORT
  if (ompt) {
    ompt_frame_t *ompt_frame;
    __ompt_get_task_info_internal(0, NULL, NULL, &ompt_frame, NULL, NULL);
    ompt_frame->enter_frame = ompt_data_none;
    ompt_frame->enter_frame_flags =
        ompt_frame_runtime | ompt_frame_framepointer;
  }
#endif

  return;
}

#if OMPT_SUPPORT
OMPT_NOINLINE
void __kmpc_omp_task_complete_if0_ompt(ident_t *loc_ref, kmp_int32 gtid,
                                       kmp_task_t *task) {
  __kmpc_omp_task_complete_if0_template<true>(loc_ref, gtid, task);
}
#endif // OMPT_SUPPORT

// __kmpc_omp_task_complete_if0: report that a task has completed execution
//
// loc_ref: source location information; points to end of task block.
// gtid: global thread number.
// task: task thunk for the completed task.
void __kmpc_omp_task_complete_if0(ident_t *loc_ref, kmp_int32 gtid,
                                  kmp_task_t *task) {
#if OMPT_SUPPORT
  if (UNLIKELY(ompt_enabled.enabled)) {
    __kmpc_omp_task_complete_if0_ompt(loc_ref, gtid, task);
    return;
  }
#endif
  __kmpc_omp_task_complete_if0_template<false>(loc_ref, gtid, task);
}

#ifdef TASK_UNUSED
// __kmpc_omp_task_complete: report that a task has completed execution
// NEVER GENERATED BY COMPILER, DEPRECATED!!!
void __kmpc_omp_task_complete(ident_t *loc_ref, kmp_int32 gtid,
                              kmp_task_t *task) {
  KA_TRACE(10, ("__kmpc_omp_task_complete(enter): T#%d loc=%p task=%p\n", gtid,
                loc_ref, KMP_TASK_TO_TASKDATA(task)));

  __kmp_task_finish<false>(gtid, task,
                           NULL); // Not sure how to find task to resume

  KA_TRACE(10, ("__kmpc_omp_task_complete(exit): T#%d loc=%p task=%p\n", gtid,
                loc_ref, KMP_TASK_TO_TASKDATA(task)));
  return;
}
#endif // TASK_UNUSED

// __kmp_init_implicit_task: Initialize the appropriate fields in the implicit
// task for a given thread
//
// loc_ref:  reference to source location of parallel region
// this_thr:  thread data structure corresponding to implicit task
// team: team for this_thr
// tid: thread id of given thread within team
// set_curr_task: TRUE if need to push current task to thread
// NOTE: Routine does not set up the implicit task ICVS.  This is assumed to
// have already been done elsewhere.
// TODO: Get better loc_ref.  Value passed in may be NULL
void __kmp_init_implicit_task(ident_t *loc_ref, kmp_info_t *this_thr,
                              kmp_team_t *team, int tid, int set_curr_task) {
  kmp_taskdata_t *task = &team->t.t_implicit_task_taskdata[tid];

  KF_TRACE(
      10,
      ("__kmp_init_implicit_task(enter): T#:%d team=%p task=%p, reinit=%s\n",
       tid, team, task, set_curr_task ? "TRUE" : "FALSE"));

  task->td_task_id = KMP_GEN_TASK_ID();
  task->td_team = team;
  //    task->td_parent   = NULL;  // fix for CQ230101 (broken parent task info
  //    in debugger)
  task->td_ident = loc_ref;
  task->td_taskwait_ident = NULL;
  task->td_taskwait_counter = 0;
  task->td_taskwait_thread = 0;

  task->td_flags.tiedness = TASK_TIED;
  task->td_flags.tasktype = TASK_IMPLICIT;
  task->td_flags.proxy = TASK_FULL;

  // All implicit tasks are executed immediately, not deferred
  task->td_flags.task_serial = 1;
  task->td_flags.tasking_ser = (__kmp_tasking_mode == tskm_immediate_exec);
  task->td_flags.team_serial = (team->t.t_serialized) ? 1 : 0;

  task->td_flags.started = 1;
  task->td_flags.executing = 1;
  task->td_flags.complete = 0;
  task->td_flags.freed = 0;

  task->td_depnode = NULL;
  task->td_last_tied = task;
  task->td_allow_completion_event.type = KMP_EVENT_UNINITIALIZED;

  if (set_curr_task) { // only do this init first time thread is created
    KMP_ATOMIC_ST_REL(&task->td_incomplete_child_tasks, 0);
    // Not used: don't need to deallocate implicit task
    KMP_ATOMIC_ST_REL(&task->td_allocated_child_tasks, 0);
    task->td_taskgroup = NULL; // An implicit task does not have taskgroup
    task->td_dephash = NULL;
    __kmp_push_current_task_to_thread(this_thr, team, tid);
  } else {
    KMP_DEBUG_ASSERT(task->td_incomplete_child_tasks == 0);
    KMP_DEBUG_ASSERT(task->td_allocated_child_tasks == 0);
  }

#if OMPT_SUPPORT
  if (UNLIKELY(ompt_enabled.enabled))
    __ompt_task_init(task, tid);
#endif

  KF_TRACE(10, ("__kmp_init_implicit_task(exit): T#:%d team=%p task=%p\n", tid,
                team, task));
}

// __kmp_finish_implicit_task: Release resources associated to implicit tasks
// at the end of parallel regions. Some resources are kept for reuse in the next
// parallel region.
//
// thread:  thread data structure corresponding to implicit task
void __kmp_finish_implicit_task(kmp_info_t *thread) {
  kmp_taskdata_t *task = thread->th.th_current_task;
  if (task->td_dephash) {
    int children;
    task->td_flags.complete = 1;
    children = KMP_ATOMIC_LD_ACQ(&task->td_incomplete_child_tasks);
    kmp_tasking_flags_t flags_old = task->td_flags;
    if (children == 0 && flags_old.complete == 1) {
      kmp_tasking_flags_t flags_new = flags_old;
      flags_new.complete = 0;
      if (KMP_COMPARE_AND_STORE_ACQ32(RCAST(kmp_int32 *, &task->td_flags),
                                      *RCAST(kmp_int32 *, &flags_old),
                                      *RCAST(kmp_int32 *, &flags_new))) {
        KA_TRACE(100, ("__kmp_finish_implicit_task: T#%d cleans "
                       "dephash of implicit task %p\n",
                       thread->th.th_info.ds.ds_gtid, task));
        __kmp_dephash_free_entries(thread, task->td_dephash);
      }
    }
  }
}

// __kmp_free_implicit_task: Release resources associated to implicit tasks
// when these are destroyed regions
//
// thread:  thread data structure corresponding to implicit task
void __kmp_free_implicit_task(kmp_info_t *thread) {
  kmp_taskdata_t *task = thread->th.th_current_task;
  if (task && task->td_dephash) {
    __kmp_dephash_free(thread, task->td_dephash);
    task->td_dephash = NULL;
  }
}

// Round up a size to a power of two specified by val: Used to insert padding
// between structures co-allocated using a single malloc() call
static size_t __kmp_round_up_to_val(size_t size, size_t val) {
  if (size & (val - 1)) {
    size &= ~(val - 1);
    if (size <= KMP_SIZE_T_MAX - val) {
      size += val; // Round up if there is no overflow.
    }
  }
  return size;
} // __kmp_round_up_to_va

// __kmp_task_alloc: Allocate the taskdata and task data structures for a task
//
// loc_ref: source location information
// gtid: global thread number.
// flags: include tiedness & task type (explicit vs. implicit) of the ''new''
// task encountered. Converted from kmp_int32 to kmp_tasking_flags_t in routine.
// sizeof_kmp_task_t:  Size in bytes of kmp_task_t data structure including
// private vars accessed in task.
// sizeof_shareds:  Size in bytes of array of pointers to shared vars accessed
// in task.
// task_entry: Pointer to task code entry point generated by compiler.
// returns: a pointer to the allocated kmp_task_t structure (task).
kmp_task_t *__kmp_task_alloc(ident_t *loc_ref, kmp_int32 gtid,
                             kmp_tasking_flags_t *flags,
                             size_t sizeof_kmp_task_t, size_t sizeof_shareds,
                             kmp_routine_entry_t task_entry) {
  kmp_task_t *task;
  kmp_taskdata_t *taskdata;
  kmp_info_t *thread = __kmp_threads[gtid];
  kmp_team_t *team = thread->th.th_team;
  kmp_taskdata_t *parent_task = thread->th.th_current_task;
  size_t shareds_offset;

  if (UNLIKELY(!TCR_4(__kmp_init_middle)))
    __kmp_middle_initialize();

  if (flags->hidden_helper) {
    if (__kmp_enable_hidden_helper) {
      if (!TCR_4(__kmp_init_hidden_helper))
        __kmp_hidden_helper_initialize();
    } else {
      // If the hidden helper task is not enabled, reset the flag to FALSE.
      flags->hidden_helper = FALSE;
    }
  }

  KA_TRACE(10, ("__kmp_task_alloc(enter): T#%d loc=%p, flags=(0x%x) "
                "sizeof_task=%ld sizeof_shared=%ld entry=%p\n",
                gtid, loc_ref, *((kmp_int32 *)flags), sizeof_kmp_task_t,
                sizeof_shareds, task_entry));

  KMP_DEBUG_ASSERT(parent_task);
  if (parent_task->td_flags.final) {
    if (flags->merged_if0) {
    }
    flags->final = 1;
  }

  if (flags->tiedness == TASK_UNTIED && !team->t.t_serialized) {
    // Untied task encountered causes the TSC algorithm to check entire deque of
    // the victim thread. If no untied task encountered, then checking the head
    // of the deque should be enough.
    KMP_CHECK_UPDATE(thread->th.th_task_team->tt.tt_untied_task_encountered, 1);
  }

  // Detachable tasks are not proxy tasks yet but could be in the future. Doing
  // the tasking setup
  // when that happens is too late.
  if (UNLIKELY(flags->proxy == TASK_PROXY ||
               flags->detachable == TASK_DETACHABLE || flags->hidden_helper)) {
    if (flags->proxy == TASK_PROXY) {
      flags->tiedness = TASK_UNTIED;
      flags->merged_if0 = 1;
    }
    /* are we running in a sequential parallel or tskm_immediate_exec... we need
       tasking support enabled */
    if ((thread->th.th_task_team) == NULL) {
      /* This should only happen if the team is serialized
          setup a task team and propagate it to the thread */
      KMP_DEBUG_ASSERT(team->t.t_serialized);
      KA_TRACE(30,
               ("T#%d creating task team in __kmp_task_alloc for proxy task\n",
                gtid));
      // 1 indicates setup the current team regardless of nthreads
      __kmp_task_team_setup(thread, team, 1);
      thread->th.th_task_team = team->t.t_task_team[thread->th.th_task_state];
    }
    kmp_task_team_t *task_team = thread->th.th_task_team;

    /* tasking must be enabled now as the task might not be pushed */
    if (!KMP_TASKING_ENABLED(task_team)) {
      KA_TRACE(
          30,
          ("T#%d enabling tasking in __kmp_task_alloc for proxy task\n", gtid));
      __kmp_enable_tasking(task_team, thread);
      kmp_int32 tid = thread->th.th_info.ds.ds_tid;
      kmp_thread_data_t *thread_data = &task_team->tt.tt_threads_data[tid];
      // No lock needed since only owner can allocate
      if (thread_data->td.td_deque == NULL) {
        __kmp_alloc_task_deque(thread, thread_data);
      }
    }

    if ((flags->proxy == TASK_PROXY || flags->detachable == TASK_DETACHABLE) &&
        task_team->tt.tt_found_proxy_tasks == FALSE)
      TCW_4(task_team->tt.tt_found_proxy_tasks, TRUE);
    if (flags->hidden_helper &&
        task_team->tt.tt_hidden_helper_task_encountered == FALSE)
      TCW_4(task_team->tt.tt_hidden_helper_task_encountered, TRUE);
  }

  // Calculate shared structure offset including padding after kmp_task_t struct
  // to align pointers in shared struct
  shareds_offset = sizeof(kmp_taskdata_t) + sizeof_kmp_task_t;
  shareds_offset = __kmp_round_up_to_val(shareds_offset, sizeof(void *));

  // Allocate a kmp_taskdata_t block and a kmp_task_t block.
  KA_TRACE(30, ("__kmp_task_alloc: T#%d First malloc size: %ld\n", gtid,
                shareds_offset));
  KA_TRACE(30, ("__kmp_task_alloc: T#%d Second malloc size: %ld\n", gtid,
                sizeof_shareds));

  // Avoid double allocation here by combining shareds with taskdata
#if USE_FAST_MEMORY
  taskdata = (kmp_taskdata_t *)__kmp_fast_allocate(thread, shareds_offset +
                                                               sizeof_shareds);
#else /* ! USE_FAST_MEMORY */
  taskdata = (kmp_taskdata_t *)__kmp_thread_malloc(thread, shareds_offset +
                                                               sizeof_shareds);
#endif /* USE_FAST_MEMORY */

  task = KMP_TASKDATA_TO_TASK(taskdata);

// Make sure task & taskdata are aligned appropriately
#if KMP_ARCH_X86 || KMP_ARCH_PPC64 || !KMP_HAVE_QUAD
  KMP_DEBUG_ASSERT((((kmp_uintptr_t)taskdata) & (sizeof(double) - 1)) == 0);
  KMP_DEBUG_ASSERT((((kmp_uintptr_t)task) & (sizeof(double) - 1)) == 0);
#else
  KMP_DEBUG_ASSERT((((kmp_uintptr_t)taskdata) & (sizeof(_Quad) - 1)) == 0);
  KMP_DEBUG_ASSERT((((kmp_uintptr_t)task) & (sizeof(_Quad) - 1)) == 0);
#endif
  if (sizeof_shareds > 0) {
    // Avoid double allocation here by combining shareds with taskdata
    task->shareds = &((char *)taskdata)[shareds_offset];
    // Make sure shareds struct is aligned to pointer size
    KMP_DEBUG_ASSERT((((kmp_uintptr_t)task->shareds) & (sizeof(void *) - 1)) ==
                     0);
  } else {
    task->shareds = NULL;
  }
  task->routine = task_entry;
  task->part_id = 0; // AC: Always start with 0 part id

  taskdata->td_task_id = KMP_GEN_TASK_ID();
  taskdata->td_team = thread->th.th_team;
  taskdata->td_alloc_thread = thread;
  taskdata->td_parent = parent_task;
  taskdata->td_level = parent_task->td_level + 1; // increment nesting level
  KMP_ATOMIC_ST_RLX(&taskdata->td_untied_count, 0);
  taskdata->td_ident = loc_ref;
  taskdata->td_taskwait_ident = NULL;
  taskdata->td_taskwait_counter = 0;
  taskdata->td_taskwait_thread = 0;
  KMP_DEBUG_ASSERT(taskdata->td_parent != NULL);
  // avoid copying icvs for proxy tasks
  if (flags->proxy == TASK_FULL)
    copy_icvs(&taskdata->td_icvs, &taskdata->td_parent->td_icvs);

  taskdata->td_flags = *flags;
  taskdata->td_task_team = thread->th.th_task_team;
  taskdata->td_size_alloc = shareds_offset + sizeof_shareds;
  taskdata->td_flags.tasktype = TASK_EXPLICIT;
  // If it is hidden helper task, we need to set the team and task team
  // correspondingly.
  if (flags->hidden_helper) {
    kmp_info_t *shadow_thread = __kmp_threads[KMP_GTID_TO_SHADOW_GTID(gtid)];
    taskdata->td_team = shadow_thread->th.th_team;
    taskdata->td_task_team = shadow_thread->th.th_task_team;
  }

  // GEH - TODO: fix this to copy parent task's value of tasking_ser flag
  taskdata->td_flags.tasking_ser = (__kmp_tasking_mode == tskm_immediate_exec);

  // GEH - TODO: fix this to copy parent task's value of team_serial flag
  taskdata->td_flags.team_serial = (team->t.t_serialized) ? 1 : 0;

  // GEH - Note we serialize the task if the team is serialized to make sure
  // implicit parallel region tasks are not left until program termination to
  // execute. Also, it helps locality to execute immediately.

  taskdata->td_flags.task_serial =
      (parent_task->td_flags.final || taskdata->td_flags.team_serial ||
       taskdata->td_flags.tasking_ser || flags->merged_if0);

  taskdata->td_flags.started = 0;
  taskdata->td_flags.executing = 0;
  taskdata->td_flags.complete = 0;
  taskdata->td_flags.freed = 0;

  KMP_ATOMIC_ST_RLX(&taskdata->td_incomplete_child_tasks, 0);
  // start at one because counts current task and children
  KMP_ATOMIC_ST_RLX(&taskdata->td_allocated_child_tasks, 1);
  taskdata->td_taskgroup =
      parent_task->td_taskgroup; // task inherits taskgroup from the parent task
  taskdata->td_dephash = NULL;
  taskdata->td_depnode = NULL;
  if (flags->tiedness == TASK_UNTIED)
    taskdata->td_last_tied = NULL; // will be set when the task is scheduled
  else
    taskdata->td_last_tied = taskdata;
  taskdata->td_allow_completion_event.type = KMP_EVENT_UNINITIALIZED;
#if OMPT_SUPPORT
  if (UNLIKELY(ompt_enabled.enabled))
    __ompt_task_init(taskdata, gtid);
#endif
  // TODO: What would be the balance between the conditions in the function and
  // an atomic operation?
  if (__kmp_track_children_task(taskdata)) {
    KMP_ATOMIC_INC(&parent_task->td_incomplete_child_tasks);
    if (parent_task->td_taskgroup)
      KMP_ATOMIC_INC(&parent_task->td_taskgroup->count);
    // Only need to keep track of allocated child tasks for explicit tasks since
    // implicit not deallocated
    if (taskdata->td_parent->td_flags.tasktype == TASK_EXPLICIT) {
      KMP_ATOMIC_INC(&taskdata->td_parent->td_allocated_child_tasks);
    }
    if (flags->hidden_helper) {
      taskdata->td_flags.task_serial = FALSE;
      // Increment the number of hidden helper tasks to be executed
      KMP_ATOMIC_INC(&__kmp_unexecuted_hidden_helper_tasks);
    }
  }

  KA_TRACE(20, ("__kmp_task_alloc(exit): T#%d created task %p parent=%p\n",
                gtid, taskdata, taskdata->td_parent));

  return task;
}

kmp_task_t *__kmpc_omp_task_alloc(ident_t *loc_ref, kmp_int32 gtid,
                                  kmp_int32 flags, size_t sizeof_kmp_task_t,
                                  size_t sizeof_shareds,
                                  kmp_routine_entry_t task_entry) {
  kmp_task_t *retval;
  kmp_tasking_flags_t *input_flags = (kmp_tasking_flags_t *)&flags;
  __kmp_assert_valid_gtid(gtid);
  input_flags->native = FALSE;
  // __kmp_task_alloc() sets up all other runtime flags
  KA_TRACE(10, ("__kmpc_omp_task_alloc(enter): T#%d loc=%p, flags=(%s %s %s) "
                "sizeof_task=%ld sizeof_shared=%ld entry=%p\n",
                gtid, loc_ref, input_flags->tiedness ? "tied  " : "untied",
                input_flags->proxy ? "proxy" : "",
                input_flags->detachable ? "detachable" : "", sizeof_kmp_task_t,
                sizeof_shareds, task_entry));

  retval = __kmp_task_alloc(loc_ref, gtid, input_flags, sizeof_kmp_task_t,
                            sizeof_shareds, task_entry);

  KA_TRACE(20, ("__kmpc_omp_task_alloc(exit): T#%d retval %p\n", gtid, retval));

  return retval;
}

kmp_task_t *__kmpc_omp_target_task_alloc(ident_t *loc_ref, kmp_int32 gtid,
                                         kmp_int32 flags,
                                         size_t sizeof_kmp_task_t,
                                         size_t sizeof_shareds,
                                         kmp_routine_entry_t task_entry,
                                         kmp_int64 device_id) {
  auto &input_flags = reinterpret_cast<kmp_tasking_flags_t &>(flags);
  // target task is untied defined in the specification
  input_flags.tiedness = TASK_UNTIED;

  if (__kmp_enable_hidden_helper)
    input_flags.hidden_helper = TRUE;

  return __kmpc_omp_task_alloc(loc_ref, gtid, flags, sizeof_kmp_task_t,
                               sizeof_shareds, task_entry);
}

/*!
@ingroup TASKING
@param loc_ref location of the original task directive
@param gtid Global Thread ID of encountering thread
@param new_task task thunk allocated by __kmpc_omp_task_alloc() for the ''new
task''
@param naffins Number of affinity items
@param affin_list List of affinity items
@return Returns non-zero if registering affinity information was not successful.
 Returns 0 if registration was successful
This entry registers the affinity information attached to a task with the task
thunk structure kmp_taskdata_t.
*/
kmp_int32
__kmpc_omp_reg_task_with_affinity(ident_t *loc_ref, kmp_int32 gtid,
                                  kmp_task_t *new_task, kmp_int32 naffins,
                                  kmp_task_affinity_info_t *affin_list) {
  return 0;
}

//  __kmp_invoke_task: invoke the specified task
//
// gtid: global thread ID of caller
// task: the task to invoke
// current_task: the task to resume after task invocation
static void __kmp_invoke_task(kmp_int32 gtid, kmp_task_t *task,
                              kmp_taskdata_t *current_task) {
  kmp_taskdata_t *taskdata = KMP_TASK_TO_TASKDATA(task);
  kmp_info_t *thread;
  int discard = 0 /* false */;
  KA_TRACE(
      30, ("__kmp_invoke_task(enter): T#%d invoking task %p, current_task=%p\n",
           gtid, taskdata, current_task));
  KMP_DEBUG_ASSERT(task);
  if (UNLIKELY(taskdata->td_flags.proxy == TASK_PROXY &&
               taskdata->td_flags.complete == 1)) {
    // This is a proxy task that was already completed but it needs to run
    // its bottom-half finish
    KA_TRACE(
        30,
        ("__kmp_invoke_task: T#%d running bottom finish for proxy task %p\n",
         gtid, taskdata));

    __kmp_bottom_half_finish_proxy(gtid, task);

    KA_TRACE(30, ("__kmp_invoke_task(exit): T#%d completed bottom finish for "
                  "proxy task %p, resuming task %p\n",
                  gtid, taskdata, current_task));

    return;
  }

#if OMPT_SUPPORT
  // For untied tasks, the first task executed only calls __kmpc_omp_task and
  // does not execute code.
  ompt_thread_info_t oldInfo;
  if (UNLIKELY(ompt_enabled.enabled)) {
    // Store the threads states and restore them after the task
    thread = __kmp_threads[gtid];
    oldInfo = thread->th.ompt_thread_info;
    thread->th.ompt_thread_info.wait_id = 0;
    thread->th.ompt_thread_info.state = (thread->th.th_team_serialized)
                                            ? ompt_state_work_serial
                                            : ompt_state_work_parallel;
    taskdata->ompt_task_info.frame.exit_frame.ptr = OMPT_GET_FRAME_ADDRESS(0);
  }
#endif

  // Decreament the counter of hidden helper tasks to be executed
  if (taskdata->td_flags.hidden_helper) {
    // Hidden helper tasks can only be executed by hidden helper threads
    KMP_ASSERT(KMP_HIDDEN_HELPER_THREAD(gtid));
    KMP_ATOMIC_DEC(&__kmp_unexecuted_hidden_helper_tasks);
  }

  // Proxy tasks are not handled by the runtime
  if (taskdata->td_flags.proxy != TASK_PROXY) {
    __kmp_task_start(gtid, task, current_task); // OMPT only if not discarded
  }

  // TODO: cancel tasks if the parallel region has also been cancelled
  // TODO: check if this sequence can be hoisted above __kmp_task_start
  // if cancellation has been enabled for this run ...
  if (UNLIKELY(__kmp_omp_cancellation)) {
    thread = __kmp_threads[gtid];
    kmp_team_t *this_team = thread->th.th_team;
    kmp_taskgroup_t *taskgroup = taskdata->td_taskgroup;
    if ((taskgroup && taskgroup->cancel_request) ||
        (this_team->t.t_cancel_request == cancel_parallel)) {
#if OMPT_SUPPORT && OMPT_OPTIONAL
      ompt_data_t *task_data;
      if (UNLIKELY(ompt_enabled.ompt_callback_cancel)) {
        __ompt_get_task_info_internal(0, NULL, &task_data, NULL, NULL, NULL);
        ompt_callbacks.ompt_callback(ompt_callback_cancel)(
            task_data,
            ((taskgroup && taskgroup->cancel_request) ? ompt_cancel_taskgroup
                                                      : ompt_cancel_parallel) |
                ompt_cancel_discarded_task,
            NULL);
      }
#endif
      KMP_COUNT_BLOCK(TASK_cancelled);
      // this task belongs to a task group and we need to cancel it
      discard = 1 /* true */;
    }
  }

  // Invoke the task routine and pass in relevant data.
  // Thunks generated by gcc take a different argument list.
  if (!discard) {
    if (taskdata->td_flags.tiedness == TASK_UNTIED) {
      taskdata->td_last_tied = current_task->td_last_tied;
      KMP_DEBUG_ASSERT(taskdata->td_last_tied);
    }
#if KMP_STATS_ENABLED
    KMP_COUNT_BLOCK(TASK_executed);
    switch (KMP_GET_THREAD_STATE()) {
    case FORK_JOIN_BARRIER:
      KMP_PUSH_PARTITIONED_TIMER(OMP_task_join_bar);
      break;
    case PLAIN_BARRIER:
      KMP_PUSH_PARTITIONED_TIMER(OMP_task_plain_bar);
      break;
    case TASKYIELD:
      KMP_PUSH_PARTITIONED_TIMER(OMP_task_taskyield);
      break;
    case TASKWAIT:
      KMP_PUSH_PARTITIONED_TIMER(OMP_task_taskwait);
      break;
    case TASKGROUP:
      KMP_PUSH_PARTITIONED_TIMER(OMP_task_taskgroup);
      break;
    default:
      KMP_PUSH_PARTITIONED_TIMER(OMP_task_immediate);
      break;
    }
#endif // KMP_STATS_ENABLED

// OMPT task begin
#if OMPT_SUPPORT
    if (UNLIKELY(ompt_enabled.enabled))
      __ompt_task_start(task, current_task, gtid);
#endif
#if OMPT_SUPPORT && OMPT_OPTIONAL
    if (UNLIKELY(ompt_enabled.ompt_callback_dispatch &&
                 taskdata->ompt_task_info.dispatch_chunk.iterations > 0)) {
      ompt_data_t instance = ompt_data_none;
      instance.ptr = &(taskdata->ompt_task_info.dispatch_chunk);
      ompt_team_info_t *team_info = __ompt_get_teaminfo(0, NULL);
      ompt_callbacks.ompt_callback(ompt_callback_dispatch)(
          &(team_info->parallel_data), &(taskdata->ompt_task_info.task_data),
          ompt_dispatch_taskloop_chunk, instance);
      taskdata->ompt_task_info.dispatch_chunk = {0, 0};
    }
#endif // OMPT_SUPPORT && OMPT_OPTIONAL

#if OMPD_SUPPORT
    if (ompd_state & OMPD_ENABLE_BP)
      ompd_bp_task_begin();
#endif

#if USE_ITT_BUILD && USE_ITT_NOTIFY
    kmp_uint64 cur_time;
    kmp_int32 kmp_itt_count_task =
        __kmp_forkjoin_frames_mode == 3 && !taskdata->td_flags.task_serial &&
        current_task->td_flags.tasktype == TASK_IMPLICIT;
    if (kmp_itt_count_task) {
      thread = __kmp_threads[gtid];
      // Time outer level explicit task on barrier for adjusting imbalance time
      if (thread->th.th_bar_arrive_time)
        cur_time = __itt_get_timestamp();
      else
        kmp_itt_count_task = 0; // thread is not on a barrier - skip timing
    }
    KMP_FSYNC_ACQUIRED(taskdata); // acquired self (new task)
#endif

    if (task->routine != NULL) {
#ifdef KMP_GOMP_COMPAT
      if (taskdata->td_flags.native) {
        ((void (*)(void *))(*(task->routine)))(task->shareds);
      } else
#endif /* KMP_GOMP_COMPAT */
      {
        (*(task->routine))(gtid, task);
      }
    }
    KMP_POP_PARTITIONED_TIMER();

#if USE_ITT_BUILD && USE_ITT_NOTIFY
    if (kmp_itt_count_task) {
      // Barrier imbalance - adjust arrive time with the task duration
      thread->th.th_bar_arrive_time += (__itt_get_timestamp() - cur_time);
    }
    KMP_FSYNC_CANCEL(taskdata); // destroy self (just executed)
    KMP_FSYNC_RELEASING(taskdata->td_parent); // releasing parent
#endif
  }

#if OMPD_SUPPORT
  if (ompd_state & OMPD_ENABLE_BP)
    ompd_bp_task_end();
#endif

  // Proxy tasks are not handled by the runtime
  if (taskdata->td_flags.proxy != TASK_PROXY) {
#if OMPT_SUPPORT
    if (UNLIKELY(ompt_enabled.enabled)) {
      thread->th.ompt_thread_info = oldInfo;
      if (taskdata->td_flags.tiedness == TASK_TIED) {
        taskdata->ompt_task_info.frame.exit_frame = ompt_data_none;
      }
      __kmp_task_finish<true>(gtid, task, current_task);
    } else
#endif
      __kmp_task_finish<false>(gtid, task, current_task);
  }

  KA_TRACE(
      30,
      ("__kmp_invoke_task(exit): T#%d completed task %p, resuming task %p\n",
       gtid, taskdata, current_task));
  return;
}

// __kmpc_omp_task_parts: Schedule a thread-switchable task for execution
//
// loc_ref: location of original task pragma (ignored)
// gtid: Global Thread ID of encountering thread
// new_task: task thunk allocated by __kmp_omp_task_alloc() for the ''new task''
// Returns:
//    TASK_CURRENT_NOT_QUEUED (0) if did not suspend and queue current task to
//    be resumed later.
//    TASK_CURRENT_QUEUED (1) if suspended and queued the current task to be
//    resumed later.
kmp_int32 __kmpc_omp_task_parts(ident_t *loc_ref, kmp_int32 gtid,
                                kmp_task_t *new_task) {
  kmp_taskdata_t *new_taskdata = KMP_TASK_TO_TASKDATA(new_task);

  KA_TRACE(10, ("__kmpc_omp_task_parts(enter): T#%d loc=%p task=%p\n", gtid,
                loc_ref, new_taskdata));

#if OMPT_SUPPORT
  kmp_taskdata_t *parent;
  if (UNLIKELY(ompt_enabled.enabled)) {
    parent = new_taskdata->td_parent;
    if (ompt_enabled.ompt_callback_task_create) {
      ompt_callbacks.ompt_callback(ompt_callback_task_create)(
          &(parent->ompt_task_info.task_data), &(parent->ompt_task_info.frame),
          &(new_taskdata->ompt_task_info.task_data), ompt_task_explicit, 0,
          OMPT_GET_RETURN_ADDRESS(0));
    }
  }
#endif

  /* Should we execute the new task or queue it? For now, let's just always try
     to queue it.  If the queue fills up, then we'll execute it.  */

  if (__kmp_push_task(gtid, new_task) == TASK_NOT_PUSHED) // if cannot defer
  { // Execute this task immediately
    kmp_taskdata_t *current_task = __kmp_threads[gtid]->th.th_current_task;
    new_taskdata->td_flags.task_serial = 1;
    __kmp_invoke_task(gtid, new_task, current_task);
  }

  KA_TRACE(
      10,
      ("__kmpc_omp_task_parts(exit): T#%d returning TASK_CURRENT_NOT_QUEUED: "
       "loc=%p task=%p, return: TASK_CURRENT_NOT_QUEUED\n",
       gtid, loc_ref, new_taskdata));

#if OMPT_SUPPORT
  if (UNLIKELY(ompt_enabled.enabled)) {
    parent->ompt_task_info.frame.enter_frame = ompt_data_none;
  }
#endif
  return TASK_CURRENT_NOT_QUEUED;
}

// __kmp_omp_task: Schedule a non-thread-switchable task for execution
//
// gtid: Global Thread ID of encountering thread
// new_task:non-thread-switchable task thunk allocated by __kmp_omp_task_alloc()
// serialize_immediate: if TRUE then if the task is executed immediately its
// execution will be serialized
// Returns:
//    TASK_CURRENT_NOT_QUEUED (0) if did not suspend and queue current task to
//    be resumed later.
//    TASK_CURRENT_QUEUED (1) if suspended and queued the current task to be
//    resumed later.
kmp_int32 __kmp_omp_task(kmp_int32 gtid, kmp_task_t *new_task,
                         bool serialize_immediate) {
  kmp_taskdata_t *new_taskdata = KMP_TASK_TO_TASKDATA(new_task);

  /* Should we execute the new task or queue it? For now, let's just always try
     to queue it.  If the queue fills up, then we'll execute it.  */
  if (new_taskdata->td_flags.proxy == TASK_PROXY ||
      __kmp_push_task(gtid, new_task) == TASK_NOT_PUSHED) // if cannot defer
  { // Execute this task immediately
    kmp_taskdata_t *current_task = __kmp_threads[gtid]->th.th_current_task;
    if (serialize_immediate)
      new_taskdata->td_flags.task_serial = 1;
    __kmp_invoke_task(gtid, new_task, current_task);
  } else if (__kmp_dflt_blocktime != KMP_MAX_BLOCKTIME &&
             __kmp_wpolicy_passive) {
    kmp_info_t *this_thr = __kmp_threads[gtid];
    kmp_team_t *team = this_thr->th.th_team;
    kmp_int32 nthreads = this_thr->th.th_team_nproc;
    for (int i = 0; i < nthreads; ++i) {
      kmp_info_t *thread = team->t.t_threads[i];
      if (thread == this_thr)
        continue;
      if (thread->th.th_sleep_loc != NULL) {
        __kmp_null_resume_wrapper(thread);
        break; // awake one thread at a time
      }
    }
  }
  return TASK_CURRENT_NOT_QUEUED;
}

// __kmpc_omp_task: Wrapper around __kmp_omp_task to schedule a
// non-thread-switchable task from the parent thread only!
//
// loc_ref: location of original task pragma (ignored)
// gtid: Global Thread ID of encountering thread
// new_task: non-thread-switchable task thunk allocated by
// __kmp_omp_task_alloc()
// Returns:
//    TASK_CURRENT_NOT_QUEUED (0) if did not suspend and queue current task to
//    be resumed later.
//    TASK_CURRENT_QUEUED (1) if suspended and queued the current task to be
//    resumed later.
kmp_int32 __kmpc_omp_task(ident_t *loc_ref, kmp_int32 gtid,
                          kmp_task_t *new_task) {
  kmp_int32 res;
  KMP_SET_THREAD_STATE_BLOCK(EXPLICIT_TASK);

#if KMP_DEBUG || OMPT_SUPPORT
  kmp_taskdata_t *new_taskdata = KMP_TASK_TO_TASKDATA(new_task);
#endif
  KA_TRACE(10, ("__kmpc_omp_task(enter): T#%d loc=%p task=%p\n", gtid, loc_ref,
                new_taskdata));
  __kmp_assert_valid_gtid(gtid);

#if OMPT_SUPPORT
  kmp_taskdata_t *parent = NULL;
  if (UNLIKELY(ompt_enabled.enabled)) {
    if (!new_taskdata->td_flags.started) {
      OMPT_STORE_RETURN_ADDRESS(gtid);
      parent = new_taskdata->td_parent;
      if (!parent->ompt_task_info.frame.enter_frame.ptr) {
        parent->ompt_task_info.frame.enter_frame.ptr =
            OMPT_GET_FRAME_ADDRESS(0);
      }
      if (ompt_enabled.ompt_callback_task_create) {
        ompt_callbacks.ompt_callback(ompt_callback_task_create)(
            &(parent->ompt_task_info.task_data),
            &(parent->ompt_task_info.frame),
            &(new_taskdata->ompt_task_info.task_data),
            ompt_task_explicit | TASK_TYPE_DETAILS_FORMAT(new_taskdata), 0,
            OMPT_LOAD_RETURN_ADDRESS(gtid));
      }
    } else {
      // We are scheduling the continuation of an UNTIED task.
      // Scheduling back to the parent task.
      __ompt_task_finish(new_task,
                         new_taskdata->ompt_task_info.scheduling_parent,
                         ompt_task_switch);
      new_taskdata->ompt_task_info.frame.exit_frame = ompt_data_none;
    }
  }
#endif

  res = __kmp_omp_task(gtid, new_task, true);

  KA_TRACE(10, ("__kmpc_omp_task(exit): T#%d returning "
                "TASK_CURRENT_NOT_QUEUED: loc=%p task=%p\n",
                gtid, loc_ref, new_taskdata));
#if OMPT_SUPPORT
  if (UNLIKELY(ompt_enabled.enabled && parent != NULL)) {
    parent->ompt_task_info.frame.enter_frame = ompt_data_none;
  }
#endif
  return res;
}

// __kmp_omp_taskloop_task: Wrapper around __kmp_omp_task to schedule
// a taskloop task with the correct OMPT return address
//
// loc_ref: location of original task pragma (ignored)
// gtid: Global Thread ID of encountering thread
// new_task: non-thread-switchable task thunk allocated by
// __kmp_omp_task_alloc()
// codeptr_ra: return address for OMPT callback
// Returns:
//    TASK_CURRENT_NOT_QUEUED (0) if did not suspend and queue current task to
//    be resumed later.
//    TASK_CURRENT_QUEUED (1) if suspended and queued the current task to be
//    resumed later.
kmp_int32 __kmp_omp_taskloop_task(ident_t *loc_ref, kmp_int32 gtid,
                                  kmp_task_t *new_task, void *codeptr_ra) {
  kmp_int32 res;
  KMP_SET_THREAD_STATE_BLOCK(EXPLICIT_TASK);

#if KMP_DEBUG || OMPT_SUPPORT
  kmp_taskdata_t *new_taskdata = KMP_TASK_TO_TASKDATA(new_task);
#endif
  KA_TRACE(10, ("__kmpc_omp_task(enter): T#%d loc=%p task=%p\n", gtid, loc_ref,
                new_taskdata));

#if OMPT_SUPPORT
  kmp_taskdata_t *parent = NULL;
  if (UNLIKELY(ompt_enabled.enabled && !new_taskdata->td_flags.started)) {
    parent = new_taskdata->td_parent;
    if (!parent->ompt_task_info.frame.enter_frame.ptr)
      parent->ompt_task_info.frame.enter_frame.ptr = OMPT_GET_FRAME_ADDRESS(0);
    if (ompt_enabled.ompt_callback_task_create) {
      ompt_callbacks.ompt_callback(ompt_callback_task_create)(
          &(parent->ompt_task_info.task_data), &(parent->ompt_task_info.frame),
          &(new_taskdata->ompt_task_info.task_data),
          ompt_task_explicit | TASK_TYPE_DETAILS_FORMAT(new_taskdata), 0,
          codeptr_ra);
    }
  }
#endif

  res = __kmp_omp_task(gtid, new_task, true);

  KA_TRACE(10, ("__kmpc_omp_task(exit): T#%d returning "
                "TASK_CURRENT_NOT_QUEUED: loc=%p task=%p\n",
                gtid, loc_ref, new_taskdata));
#if OMPT_SUPPORT
  if (UNLIKELY(ompt_enabled.enabled && parent != NULL)) {
    parent->ompt_task_info.frame.enter_frame = ompt_data_none;
  }
#endif
  return res;
}

template <bool ompt>
static kmp_int32 __kmpc_omp_taskwait_template(ident_t *loc_ref, kmp_int32 gtid,
                                              void *frame_address,
                                              void *return_address) {
  kmp_taskdata_t *taskdata = nullptr;
  kmp_info_t *thread;
  int thread_finished = FALSE;
  KMP_SET_THREAD_STATE_BLOCK(TASKWAIT);

  KA_TRACE(10, ("__kmpc_omp_taskwait(enter): T#%d loc=%p\n", gtid, loc_ref));
  KMP_DEBUG_ASSERT(gtid >= 0);

  if (__kmp_tasking_mode != tskm_immediate_exec) {
    thread = __kmp_threads[gtid];
    taskdata = thread->th.th_current_task;

#if OMPT_SUPPORT && OMPT_OPTIONAL
    ompt_data_t *my_task_data;
    ompt_data_t *my_parallel_data;

    if (ompt) {
      my_task_data = &(taskdata->ompt_task_info.task_data);
      my_parallel_data = OMPT_CUR_TEAM_DATA(thread);

      taskdata->ompt_task_info.frame.enter_frame.ptr = frame_address;

      if (ompt_enabled.ompt_callback_sync_region) {
        ompt_callbacks.ompt_callback(ompt_callback_sync_region)(
            ompt_sync_region_taskwait, ompt_scope_begin, my_parallel_data,
            my_task_data, return_address);
      }

      if (ompt_enabled.ompt_callback_sync_region_wait) {
        ompt_callbacks.ompt_callback(ompt_callback_sync_region_wait)(
            ompt_sync_region_taskwait, ompt_scope_begin, my_parallel_data,
            my_task_data, return_address);
      }
    }
#endif // OMPT_SUPPORT && OMPT_OPTIONAL

// Debugger: The taskwait is active. Store location and thread encountered the
// taskwait.
#if USE_ITT_BUILD
// Note: These values are used by ITT events as well.
#endif /* USE_ITT_BUILD */
    taskdata->td_taskwait_counter += 1;
    taskdata->td_taskwait_ident = loc_ref;
    taskdata->td_taskwait_thread = gtid + 1;

#if USE_ITT_BUILD
    void *itt_sync_obj = NULL;
#if USE_ITT_NOTIFY
    KMP_ITT_TASKWAIT_STARTING(itt_sync_obj);
#endif /* USE_ITT_NOTIFY */
#endif /* USE_ITT_BUILD */

    bool must_wait =
        !taskdata->td_flags.team_serial && !taskdata->td_flags.final;

    must_wait = must_wait || (thread->th.th_task_team != NULL &&
                              thread->th.th_task_team->tt.tt_found_proxy_tasks);
    // If hidden helper thread is encountered, we must enable wait here.
    must_wait =
        must_wait ||
        (__kmp_enable_hidden_helper && thread->th.th_task_team != NULL &&
         thread->th.th_task_team->tt.tt_hidden_helper_task_encountered);

    if (must_wait) {
      kmp_flag_32<false, false> flag(
          RCAST(std::atomic<kmp_uint32> *,
                &(taskdata->td_incomplete_child_tasks)),
          0U);
      while (KMP_ATOMIC_LD_ACQ(&taskdata->td_incomplete_child_tasks) != 0) {
        flag.execute_tasks(thread, gtid, FALSE,
                           &thread_finished USE_ITT_BUILD_ARG(itt_sync_obj),
                           __kmp_task_stealing_constraint);
      }
    }
#if USE_ITT_BUILD
    KMP_ITT_TASKWAIT_FINISHED(itt_sync_obj);
    KMP_FSYNC_ACQUIRED(taskdata); // acquire self - sync with children
#endif /* USE_ITT_BUILD */

    // Debugger:  The taskwait is completed. Location remains, but thread is
    // negated.
    taskdata->td_taskwait_thread = -taskdata->td_taskwait_thread;

#if OMPT_SUPPORT && OMPT_OPTIONAL
    if (ompt) {
      if (ompt_enabled.ompt_callback_sync_region_wait) {
        ompt_callbacks.ompt_callback(ompt_callback_sync_region_wait)(
            ompt_sync_region_taskwait, ompt_scope_end, my_parallel_data,
            my_task_data, return_address);
      }
      if (ompt_enabled.ompt_callback_sync_region) {
        ompt_callbacks.ompt_callback(ompt_callback_sync_region)(
            ompt_sync_region_taskwait, ompt_scope_end, my_parallel_data,
            my_task_data, return_address);
      }
      taskdata->ompt_task_info.frame.enter_frame = ompt_data_none;
    }
#endif // OMPT_SUPPORT && OMPT_OPTIONAL

  }

  KA_TRACE(10, ("__kmpc_omp_taskwait(exit): T#%d task %p finished waiting, "
                "returning TASK_CURRENT_NOT_QUEUED\n",
                gtid, taskdata));

  return TASK_CURRENT_NOT_QUEUED;
}

#if OMPT_SUPPORT && OMPT_OPTIONAL
OMPT_NOINLINE
static kmp_int32 __kmpc_omp_taskwait_ompt(ident_t *loc_ref, kmp_int32 gtid,
                                          void *frame_address,
                                          void *return_address) {
  return __kmpc_omp_taskwait_template<true>(loc_ref, gtid, frame_address,
                                            return_address);
}
#endif // OMPT_SUPPORT && OMPT_OPTIONAL

// __kmpc_omp_taskwait: Wait until all tasks generated by the current task are
// complete
kmp_int32 __kmpc_omp_taskwait(ident_t *loc_ref, kmp_int32 gtid) {
#if OMPT_SUPPORT && OMPT_OPTIONAL
  if (UNLIKELY(ompt_enabled.enabled)) {
    OMPT_STORE_RETURN_ADDRESS(gtid);
    return __kmpc_omp_taskwait_ompt(loc_ref, gtid, OMPT_GET_FRAME_ADDRESS(0),
                                    OMPT_LOAD_RETURN_ADDRESS(gtid));
  }
#endif
  return __kmpc_omp_taskwait_template<false>(loc_ref, gtid, NULL, NULL);
}

// __kmpc_omp_taskyield: switch to a different task
kmp_int32 __kmpc_omp_taskyield(ident_t *loc_ref, kmp_int32 gtid, int end_part) {
  kmp_taskdata_t *taskdata = NULL;
  kmp_info_t *thread;
  int thread_finished = FALSE;

  KMP_COUNT_BLOCK(OMP_TASKYIELD);
  KMP_SET_THREAD_STATE_BLOCK(TASKYIELD);

  KA_TRACE(10, ("__kmpc_omp_taskyield(enter): T#%d loc=%p end_part = %d\n",
                gtid, loc_ref, end_part));
  __kmp_assert_valid_gtid(gtid);

  if (__kmp_tasking_mode != tskm_immediate_exec && __kmp_init_parallel) {
    thread = __kmp_threads[gtid];
    taskdata = thread->th.th_current_task;
// Should we model this as a task wait or not?
// Debugger: The taskwait is active. Store location and thread encountered the
// taskwait.
#if USE_ITT_BUILD
// Note: These values are used by ITT events as well.
#endif /* USE_ITT_BUILD */
    taskdata->td_taskwait_counter += 1;
    taskdata->td_taskwait_ident = loc_ref;
    taskdata->td_taskwait_thread = gtid + 1;

#if USE_ITT_BUILD
    void *itt_sync_obj = NULL;
#if USE_ITT_NOTIFY
    KMP_ITT_TASKWAIT_STARTING(itt_sync_obj);
#endif /* USE_ITT_NOTIFY */
#endif /* USE_ITT_BUILD */
    if (!taskdata->td_flags.team_serial) {
      kmp_task_team_t *task_team = thread->th.th_task_team;
      if (task_team != NULL) {
        if (KMP_TASKING_ENABLED(task_team)) {
#if OMPT_SUPPORT
          if (UNLIKELY(ompt_enabled.enabled))
            thread->th.ompt_thread_info.ompt_task_yielded = 1;
#endif
          __kmp_execute_tasks_32(
              thread, gtid, (kmp_flag_32<> *)NULL, FALSE,
              &thread_finished USE_ITT_BUILD_ARG(itt_sync_obj),
              __kmp_task_stealing_constraint);
#if OMPT_SUPPORT
          if (UNLIKELY(ompt_enabled.enabled))
            thread->th.ompt_thread_info.ompt_task_yielded = 0;
#endif
        }
      }
    }
#if USE_ITT_BUILD
    KMP_ITT_TASKWAIT_FINISHED(itt_sync_obj);
#endif /* USE_ITT_BUILD */

    // Debugger:  The taskwait is completed. Location remains, but thread is
    // negated.
    taskdata->td_taskwait_thread = -taskdata->td_taskwait_thread;
  }

  KA_TRACE(10, ("__kmpc_omp_taskyield(exit): T#%d task %p resuming, "
                "returning TASK_CURRENT_NOT_QUEUED\n",
                gtid, taskdata));

  return TASK_CURRENT_NOT_QUEUED;
}

// Task Reduction implementation
//
// Note: initial implementation didn't take into account the possibility
// to specify omp_orig for initializer of the UDR (user defined reduction).
// Corrected implementation takes into account the omp_orig object.
// Compiler is free to use old implementation if omp_orig is not specified.

/*!
@ingroup BASIC_TYPES
@{
*/

/*!
Flags for special info per task reduction item.
*/
typedef struct kmp_taskred_flags {
  /*! 1 - use lazy alloc/init (e.g. big objects, num tasks < num threads) */
  unsigned lazy_priv : 1;
  unsigned reserved31 : 31;
} kmp_taskred_flags_t;

/*!
Internal struct for reduction data item related info set up by compiler.
*/
typedef struct kmp_task_red_input {
  void *reduce_shar; /**< shared between tasks item to reduce into */
  size_t reduce_size; /**< size of data item in bytes */
  // three compiler-generated routines (init, fini are optional):
  void *reduce_init; /**< data initialization routine (single parameter) */
  void *reduce_fini; /**< data finalization routine */
  void *reduce_comb; /**< data combiner routine */
  kmp_taskred_flags_t flags; /**< flags for additional info from compiler */
} kmp_task_red_input_t;

/*!
Internal struct for reduction data item related info saved by the library.
*/
typedef struct kmp_taskred_data {
  void *reduce_shar; /**< shared between tasks item to reduce into */
  size_t reduce_size; /**< size of data item */
  kmp_taskred_flags_t flags; /**< flags for additional info from compiler */
  void *reduce_priv; /**< array of thread specific items */
  void *reduce_pend; /**< end of private data for faster comparison op */
  // three compiler-generated routines (init, fini are optional):
  void *reduce_comb; /**< data combiner routine */
  void *reduce_init; /**< data initialization routine (two parameters) */
  void *reduce_fini; /**< data finalization routine */
  void *reduce_orig; /**< original item (can be used in UDR initializer) */
} kmp_taskred_data_t;

/*!
Internal struct for reduction data item related info set up by compiler.

New interface: added reduce_orig field to provide omp_orig for UDR initializer.
*/
typedef struct kmp_taskred_input {
  void *reduce_shar; /**< shared between tasks item to reduce into */
  void *reduce_orig; /**< original reduction item used for initialization */
  size_t reduce_size; /**< size of data item */
  // three compiler-generated routines (init, fini are optional):
  void *reduce_init; /**< data initialization routine (two parameters) */
  void *reduce_fini; /**< data finalization routine */
  void *reduce_comb; /**< data combiner routine */
  kmp_taskred_flags_t flags; /**< flags for additional info from compiler */
} kmp_taskred_input_t;
/*!
@}
*/

template <typename T> void __kmp_assign_orig(kmp_taskred_data_t &item, T &src);
template <>
void __kmp_assign_orig<kmp_task_red_input_t>(kmp_taskred_data_t &item,
                                             kmp_task_red_input_t &src) {
  item.reduce_orig = NULL;
}
template <>
void __kmp_assign_orig<kmp_taskred_input_t>(kmp_taskred_data_t &item,
                                            kmp_taskred_input_t &src) {
  if (src.reduce_orig != NULL) {
    item.reduce_orig = src.reduce_orig;
  } else {
    item.reduce_orig = src.reduce_shar;
  } // non-NULL reduce_orig means new interface used
}

template <typename T> void __kmp_call_init(kmp_taskred_data_t &item, size_t j);
template <>
void __kmp_call_init<kmp_task_red_input_t>(kmp_taskred_data_t &item,
                                           size_t offset) {
  ((void (*)(void *))item.reduce_init)((char *)(item.reduce_priv) + offset);
}
template <>
void __kmp_call_init<kmp_taskred_input_t>(kmp_taskred_data_t &item,
                                          size_t offset) {
  ((void (*)(void *, void *))item.reduce_init)(
      (char *)(item.reduce_priv) + offset, item.reduce_orig);
}

template <typename T>
void *__kmp_task_reduction_init(int gtid, int num, T *data) {
  __kmp_assert_valid_gtid(gtid);
  kmp_info_t *thread = __kmp_threads[gtid];
  kmp_taskgroup_t *tg = thread->th.th_current_task->td_taskgroup;
  kmp_uint32 nth = thread->th.th_team_nproc;
  kmp_taskred_data_t *arr;

  // check input data just in case
  KMP_ASSERT(tg != NULL);
  KMP_ASSERT(data != NULL);
  KMP_ASSERT(num > 0);
  if (nth == 1) {
    KA_TRACE(10, ("__kmpc_task_reduction_init: T#%d, tg %p, exiting nth=1\n",
                  gtid, tg));
    return (void *)tg;
  }
  KA_TRACE(10, ("__kmpc_task_reduction_init: T#%d, taskgroup %p, #items %d\n",
                gtid, tg, num));
  arr = (kmp_taskred_data_t *)__kmp_thread_malloc(
      thread, num * sizeof(kmp_taskred_data_t));
  for (int i = 0; i < num; ++i) {
    size_t size = data[i].reduce_size - 1;
    // round the size up to cache line per thread-specific item
    size += CACHE_LINE - size % CACHE_LINE;
    KMP_ASSERT(data[i].reduce_comb != NULL); // combiner is mandatory
    arr[i].reduce_shar = data[i].reduce_shar;
    arr[i].reduce_size = size;
    arr[i].flags = data[i].flags;
    arr[i].reduce_comb = data[i].reduce_comb;
    arr[i].reduce_init = data[i].reduce_init;
    arr[i].reduce_fini = data[i].reduce_fini;
    __kmp_assign_orig<T>(arr[i], data[i]);
    if (!arr[i].flags.lazy_priv) {
      // allocate cache-line aligned block and fill it with zeros
      arr[i].reduce_priv = __kmp_allocate(nth * size);
      arr[i].reduce_pend = (char *)(arr[i].reduce_priv) + nth * size;
      if (arr[i].reduce_init != NULL) {
        // initialize all thread-specific items
        for (size_t j = 0; j < nth; ++j) {
          __kmp_call_init<T>(arr[i], j * size);
        }
      }
    } else {
      // only allocate space for pointers now,
      // objects will be lazily allocated/initialized if/when requested
      // note that __kmp_allocate zeroes the allocated memory
      arr[i].reduce_priv = __kmp_allocate(nth * sizeof(void *));
    }
  }
  tg->reduce_data = (void *)arr;
  tg->reduce_num_data = num;
  return (void *)tg;
}

/*!
@ingroup TASKING
@param gtid      Global thread ID
@param num       Number of data items to reduce
@param data      Array of data for reduction
@return The taskgroup identifier

Initialize task reduction for the taskgroup.

Note: this entry supposes the optional compiler-generated initializer routine
has single parameter - pointer to object to be initialized. That means
the reduction either does not use omp_orig object, or the omp_orig is accessible
without help of the runtime library.
*/
void *__kmpc_task_reduction_init(int gtid, int num, void *data) {
  return __kmp_task_reduction_init(gtid, num, (kmp_task_red_input_t *)data);
}

/*!
@ingroup TASKING
@param gtid      Global thread ID
@param num       Number of data items to reduce
@param data      Array of data for reduction
@return The taskgroup identifier

Initialize task reduction for the taskgroup.

Note: this entry supposes the optional compiler-generated initializer routine
has two parameters, pointer to object to be initialized and pointer to omp_orig
*/
void *__kmpc_taskred_init(int gtid, int num, void *data) {
  return __kmp_task_reduction_init(gtid, num, (kmp_taskred_input_t *)data);
}

// Copy task reduction data (except for shared pointers).
template <typename T>
void __kmp_task_reduction_init_copy(kmp_info_t *thr, int num, T *data,
                                    kmp_taskgroup_t *tg, void *reduce_data) {
  kmp_taskred_data_t *arr;
  KA_TRACE(20, ("__kmp_task_reduction_init_copy: Th %p, init taskgroup %p,"
                " from data %p\n",
                thr, tg, reduce_data));
  arr = (kmp_taskred_data_t *)__kmp_thread_malloc(
      thr, num * sizeof(kmp_taskred_data_t));
  // threads will share private copies, thunk routines, sizes, flags, etc.:
  KMP_MEMCPY(arr, reduce_data, num * sizeof(kmp_taskred_data_t));
  for (int i = 0; i < num; ++i) {
    arr[i].reduce_shar = data[i].reduce_shar; // init unique shared pointers
  }
  tg->reduce_data = (void *)arr;
  tg->reduce_num_data = num;
}

/*!
@ingroup TASKING
@param gtid    Global thread ID
@param tskgrp  The taskgroup ID (optional)
@param data    Shared location of the item
@return The pointer to per-thread data

Get thread-specific location of data item
*/
void *__kmpc_task_reduction_get_th_data(int gtid, void *tskgrp, void *data) {
  __kmp_assert_valid_gtid(gtid);
  kmp_info_t *thread = __kmp_threads[gtid];
  kmp_int32 nth = thread->th.th_team_nproc;
  if (nth == 1)
    return data; // nothing to do

  kmp_taskgroup_t *tg = (kmp_taskgroup_t *)tskgrp;
  if (tg == NULL)
    tg = thread->th.th_current_task->td_taskgroup;
  KMP_ASSERT(tg != NULL);
  kmp_taskred_data_t *arr = (kmp_taskred_data_t *)(tg->reduce_data);
  kmp_int32 num = tg->reduce_num_data;
  kmp_int32 tid = thread->th.th_info.ds.ds_tid;

  KMP_ASSERT(data != NULL);
  while (tg != NULL) {
    for (int i = 0; i < num; ++i) {
      if (!arr[i].flags.lazy_priv) {
        if (data == arr[i].reduce_shar ||
            (data >= arr[i].reduce_priv && data < arr[i].reduce_pend))
          return (char *)(arr[i].reduce_priv) + tid * arr[i].reduce_size;
      } else {
        // check shared location first
        void **p_priv = (void **)(arr[i].reduce_priv);
        if (data == arr[i].reduce_shar)
          goto found;
        // check if we get some thread specific location as parameter
        for (int j = 0; j < nth; ++j)
          if (data == p_priv[j])
            goto found;
        continue; // not found, continue search
      found:
        if (p_priv[tid] == NULL) {
          // allocate thread specific object lazily
          p_priv[tid] = __kmp_allocate(arr[i].reduce_size);
          if (arr[i].reduce_init != NULL) {
            if (arr[i].reduce_orig != NULL) { // new interface
              ((void (*)(void *, void *))arr[i].reduce_init)(
                  p_priv[tid], arr[i].reduce_orig);
            } else { // old interface (single parameter)
              ((void (*)(void *))arr[i].reduce_init)(p_priv[tid]);
            }
          }
        }
        return p_priv[tid];
      }
    }
    tg = tg->parent;
    arr = (kmp_taskred_data_t *)(tg->reduce_data);
    num = tg->reduce_num_data;
  }
  KMP_ASSERT2(0, "Unknown task reduction item");
  return NULL; // ERROR, this line never executed
}

// Finalize task reduction.
// Called from __kmpc_end_taskgroup()
static void __kmp_task_reduction_fini(kmp_info_t *th, kmp_taskgroup_t *tg) {
  kmp_int32 nth = th->th.th_team_nproc;
  KMP_DEBUG_ASSERT(nth > 1); // should not be called if nth == 1
  kmp_taskred_data_t *arr = (kmp_taskred_data_t *)tg->reduce_data;
  kmp_int32 num = tg->reduce_num_data;
  for (int i = 0; i < num; ++i) {
    void *sh_data = arr[i].reduce_shar;
    void (*f_fini)(void *) = (void (*)(void *))(arr[i].reduce_fini);
    void (*f_comb)(void *, void *) =
        (void (*)(void *, void *))(arr[i].reduce_comb);
    if (!arr[i].flags.lazy_priv) {
      void *pr_data = arr[i].reduce_priv;
      size_t size = arr[i].reduce_size;
      for (int j = 0; j < nth; ++j) {
        void *priv_data = (char *)pr_data + j * size;
        f_comb(sh_data, priv_data); // combine results
        if (f_fini)
          f_fini(priv_data); // finalize if needed
      }
    } else {
      void **pr_data = (void **)(arr[i].reduce_priv);
      for (int j = 0; j < nth; ++j) {
        if (pr_data[j] != NULL) {
          f_comb(sh_data, pr_data[j]); // combine results
          if (f_fini)
            f_fini(pr_data[j]); // finalize if needed
          __kmp_free(pr_data[j]);
        }
      }
    }
    __kmp_free(arr[i].reduce_priv);
  }
  __kmp_thread_free(th, arr);
  tg->reduce_data = NULL;
  tg->reduce_num_data = 0;
}

// Cleanup task reduction data for parallel or worksharing,
// do not touch task private data other threads still working with.
// Called from __kmpc_end_taskgroup()
static void __kmp_task_reduction_clean(kmp_info_t *th, kmp_taskgroup_t *tg) {
  __kmp_thread_free(th, tg->reduce_data);
  tg->reduce_data = NULL;
  tg->reduce_num_data = 0;
}

template <typename T>
void *__kmp_task_reduction_modifier_init(ident_t *loc, int gtid, int is_ws,
                                         int num, T *data) {
  __kmp_assert_valid_gtid(gtid);
  kmp_info_t *thr = __kmp_threads[gtid];
  kmp_int32 nth = thr->th.th_team_nproc;
  __kmpc_taskgroup(loc, gtid); // form new taskgroup first
  if (nth == 1) {
    KA_TRACE(10,
             ("__kmpc_reduction_modifier_init: T#%d, tg %p, exiting nth=1\n",
              gtid, thr->th.th_current_task->td_taskgroup));
    return (void *)thr->th.th_current_task->td_taskgroup;
  }
  kmp_team_t *team = thr->th.th_team;
  void *reduce_data;
  kmp_taskgroup_t *tg;
  reduce_data = KMP_ATOMIC_LD_RLX(&team->t.t_tg_reduce_data[is_ws]);
  if (reduce_data == NULL &&
      __kmp_atomic_compare_store(&team->t.t_tg_reduce_data[is_ws], reduce_data,
                                 (void *)1)) {
    // single thread enters this block to initialize common reduction data
    KMP_DEBUG_ASSERT(reduce_data == NULL);
    // first initialize own data, then make a copy other threads can use
    tg = (kmp_taskgroup_t *)__kmp_task_reduction_init<T>(gtid, num, data);
    reduce_data = __kmp_thread_malloc(thr, num * sizeof(kmp_taskred_data_t));
    KMP_MEMCPY(reduce_data, tg->reduce_data, num * sizeof(kmp_taskred_data_t));
    // fini counters should be 0 at this point
    KMP_DEBUG_ASSERT(KMP_ATOMIC_LD_RLX(&team->t.t_tg_fini_counter[0]) == 0);
    KMP_DEBUG_ASSERT(KMP_ATOMIC_LD_RLX(&team->t.t_tg_fini_counter[1]) == 0);
    KMP_ATOMIC_ST_REL(&team->t.t_tg_reduce_data[is_ws], reduce_data);
  } else {
    while (
        (reduce_data = KMP_ATOMIC_LD_ACQ(&team->t.t_tg_reduce_data[is_ws])) ==
        (void *)1) { // wait for task reduction initialization
      KMP_CPU_PAUSE();
    }
    KMP_DEBUG_ASSERT(reduce_data > (void *)1); // should be valid pointer here
    tg = thr->th.th_current_task->td_taskgroup;
    __kmp_task_reduction_init_copy<T>(thr, num, data, tg, reduce_data);
  }
  return tg;
}

/*!
@ingroup TASKING
@param loc       Source location info
@param gtid      Global thread ID
@param is_ws     Is 1 if the reduction is for worksharing, 0 otherwise
@param num       Number of data items to reduce
@param data      Array of data for reduction
@return The taskgroup identifier

Initialize task reduction for a parallel or worksharing.

Note: this entry supposes the optional compiler-generated initializer routine
has single parameter - pointer to object to be initialized. That means
the reduction either does not use omp_orig object, or the omp_orig is accessible
without help of the runtime library.
*/
void *__kmpc_task_reduction_modifier_init(ident_t *loc, int gtid, int is_ws,
                                          int num, void *data) {
  return __kmp_task_reduction_modifier_init(loc, gtid, is_ws, num,
                                            (kmp_task_red_input_t *)data);
}

/*!
@ingroup TASKING
@param loc       Source location info
@param gtid      Global thread ID
@param is_ws     Is 1 if the reduction is for worksharing, 0 otherwise
@param num       Number of data items to reduce
@param data      Array of data for reduction
@return The taskgroup identifier

Initialize task reduction for a parallel or worksharing.

Note: this entry supposes the optional compiler-generated initializer routine
has two parameters, pointer to object to be initialized and pointer to omp_orig
*/
void *__kmpc_taskred_modifier_init(ident_t *loc, int gtid, int is_ws, int num,
                                   void *data) {
  return __kmp_task_reduction_modifier_init(loc, gtid, is_ws, num,
                                            (kmp_taskred_input_t *)data);
}

/*!
@ingroup TASKING
@param loc       Source location info
@param gtid      Global thread ID
@param is_ws     Is 1 if the reduction is for worksharing, 0 otherwise

Finalize task reduction for a parallel or worksharing.
*/
void __kmpc_task_reduction_modifier_fini(ident_t *loc, int gtid, int is_ws) {
  __kmpc_end_taskgroup(loc, gtid);
}

// __kmpc_taskgroup: Start a new taskgroup
void __kmpc_taskgroup(ident_t *loc, int gtid) {
  __kmp_assert_valid_gtid(gtid);
  kmp_info_t *thread = __kmp_threads[gtid];
  kmp_taskdata_t *taskdata = thread->th.th_current_task;
  kmp_taskgroup_t *tg_new =
      (kmp_taskgroup_t *)__kmp_thread_malloc(thread, sizeof(kmp_taskgroup_t));
  KA_TRACE(10, ("__kmpc_taskgroup: T#%d loc=%p group=%p\n", gtid, loc, tg_new));
  KMP_ATOMIC_ST_RLX(&tg_new->count, 0);
  KMP_ATOMIC_ST_RLX(&tg_new->cancel_request, cancel_noreq);
  tg_new->parent = taskdata->td_taskgroup;
  tg_new->reduce_data = NULL;
  tg_new->reduce_num_data = 0;
  tg_new->gomp_data = NULL;
  taskdata->td_taskgroup = tg_new;

#if OMPT_SUPPORT && OMPT_OPTIONAL
  if (UNLIKELY(ompt_enabled.ompt_callback_sync_region)) {
    void *codeptr = OMPT_LOAD_RETURN_ADDRESS(gtid);
    if (!codeptr)
      codeptr = OMPT_GET_RETURN_ADDRESS(0);
    kmp_team_t *team = thread->th.th_team;
    ompt_data_t my_task_data = taskdata->ompt_task_info.task_data;
    // FIXME: I think this is wrong for lwt!
    ompt_data_t my_parallel_data = team->t.ompt_team_info.parallel_data;

    ompt_callbacks.ompt_callback(ompt_callback_sync_region)(
        ompt_sync_region_taskgroup, ompt_scope_begin, &(my_parallel_data),
        &(my_task_data), codeptr);
  }
#endif
}

// __kmpc_end_taskgroup: Wait until all tasks generated by the current task
//                       and its descendants are complete
void __kmpc_end_taskgroup(ident_t *loc, int gtid) {
  __kmp_assert_valid_gtid(gtid);
  kmp_info_t *thread = __kmp_threads[gtid];
  kmp_taskdata_t *taskdata = thread->th.th_current_task;
  kmp_taskgroup_t *taskgroup = taskdata->td_taskgroup;
  int thread_finished = FALSE;

#if OMPT_SUPPORT && OMPT_OPTIONAL
  kmp_team_t *team;
  ompt_data_t my_task_data;
  ompt_data_t my_parallel_data;
  void *codeptr = nullptr;
  if (UNLIKELY(ompt_enabled.enabled)) {
    team = thread->th.th_team;
    my_task_data = taskdata->ompt_task_info.task_data;
    // FIXME: I think this is wrong for lwt!
    my_parallel_data = team->t.ompt_team_info.parallel_data;
    codeptr = OMPT_LOAD_RETURN_ADDRESS(gtid);
    if (!codeptr)
      codeptr = OMPT_GET_RETURN_ADDRESS(0);
  }
#endif

  KA_TRACE(10, ("__kmpc_end_taskgroup(enter): T#%d loc=%p\n", gtid, loc));
  KMP_DEBUG_ASSERT(taskgroup != NULL);
  KMP_SET_THREAD_STATE_BLOCK(TASKGROUP);

  if (__kmp_tasking_mode != tskm_immediate_exec) {
    // mark task as waiting not on a barrier
    taskdata->td_taskwait_counter += 1;
    taskdata->td_taskwait_ident = loc;
    taskdata->td_taskwait_thread = gtid + 1;
#if USE_ITT_BUILD
    // For ITT the taskgroup wait is similar to taskwait until we need to
    // distinguish them
    void *itt_sync_obj = NULL;
#if USE_ITT_NOTIFY
    KMP_ITT_TASKWAIT_STARTING(itt_sync_obj);
#endif /* USE_ITT_NOTIFY */
#endif /* USE_ITT_BUILD */

#if OMPT_SUPPORT && OMPT_OPTIONAL
    if (UNLIKELY(ompt_enabled.ompt_callback_sync_region_wait)) {
      ompt_callbacks.ompt_callback(ompt_callback_sync_region_wait)(
          ompt_sync_region_taskgroup, ompt_scope_begin, &(my_parallel_data),
          &(my_task_data), codeptr);
    }
#endif

    if (!taskdata->td_flags.team_serial ||
        (thread->th.th_task_team != NULL &&
         (thread->th.th_task_team->tt.tt_found_proxy_tasks ||
          thread->th.th_task_team->tt.tt_hidden_helper_task_encountered))) {
      kmp_flag_32<false, false> flag(
          RCAST(std::atomic<kmp_uint32> *, &(taskgroup->count)), 0U);
      while (KMP_ATOMIC_LD_ACQ(&taskgroup->count) != 0) {
        flag.execute_tasks(thread, gtid, FALSE,
                           &thread_finished USE_ITT_BUILD_ARG(itt_sync_obj),
                           __kmp_task_stealing_constraint);
      }
    }
    taskdata->td_taskwait_thread = -taskdata->td_taskwait_thread; // end waiting

#if OMPT_SUPPORT && OMPT_OPTIONAL
    if (UNLIKELY(ompt_enabled.ompt_callback_sync_region_wait)) {
      ompt_callbacks.ompt_callback(ompt_callback_sync_region_wait)(
          ompt_sync_region_taskgroup, ompt_scope_end, &(my_parallel_data),
          &(my_task_data), codeptr);
    }
#endif

#if USE_ITT_BUILD
    KMP_ITT_TASKWAIT_FINISHED(itt_sync_obj);
    KMP_FSYNC_ACQUIRED(taskdata); // acquire self - sync with descendants
#endif /* USE_ITT_BUILD */
  }
  KMP_DEBUG_ASSERT(taskgroup->count == 0);

  if (taskgroup->reduce_data != NULL &&
      !taskgroup->gomp_data) { // need to reduce?
    int cnt;
    void *reduce_data;
    kmp_team_t *t = thread->th.th_team;
    kmp_taskred_data_t *arr = (kmp_taskred_data_t *)taskgroup->reduce_data;
    // check if <priv> data of the first reduction variable shared for the team
    void *priv0 = arr[0].reduce_priv;
    if ((reduce_data = KMP_ATOMIC_LD_ACQ(&t->t.t_tg_reduce_data[0])) != NULL &&
        ((kmp_taskred_data_t *)reduce_data)[0].reduce_priv == priv0) {
      // finishing task reduction on parallel
      cnt = KMP_ATOMIC_INC(&t->t.t_tg_fini_counter[0]);
      if (cnt == thread->th.th_team_nproc - 1) {
        // we are the last thread passing __kmpc_reduction_modifier_fini()
        // finalize task reduction:
        __kmp_task_reduction_fini(thread, taskgroup);
        // cleanup fields in the team structure:
        // TODO: is relaxed store enough here (whole barrier should follow)?
        __kmp_thread_free(thread, reduce_data);
        KMP_ATOMIC_ST_REL(&t->t.t_tg_reduce_data[0], NULL);
        KMP_ATOMIC_ST_REL(&t->t.t_tg_fini_counter[0], 0);
      } else {
        // we are not the last thread passing __kmpc_reduction_modifier_fini(),
        // so do not finalize reduction, just clean own copy of the data
        __kmp_task_reduction_clean(thread, taskgroup);
      }
    } else if ((reduce_data = KMP_ATOMIC_LD_ACQ(&t->t.t_tg_reduce_data[1])) !=
                   NULL &&
               ((kmp_taskred_data_t *)reduce_data)[0].reduce_priv == priv0) {
      // finishing task reduction on worksharing
      cnt = KMP_ATOMIC_INC(&t->t.t_tg_fini_counter[1]);
      if (cnt == thread->th.th_team_nproc - 1) {
        // we are the last thread passing __kmpc_reduction_modifier_fini()
        __kmp_task_reduction_fini(thread, taskgroup);
        // cleanup fields in team structure:
        // TODO: is relaxed store enough here (whole barrier should follow)?
        __kmp_thread_free(thread, reduce_data);
        KMP_ATOMIC_ST_REL(&t->t.t_tg_reduce_data[1], NULL);
        KMP_ATOMIC_ST_REL(&t->t.t_tg_fini_counter[1], 0);
      } else {
        // we are not the last thread passing __kmpc_reduction_modifier_fini(),
        // so do not finalize reduction, just clean own copy of the data
        __kmp_task_reduction_clean(thread, taskgroup);
      }
    } else {
      // finishing task reduction on taskgroup
      __kmp_task_reduction_fini(thread, taskgroup);
    }
  }
  // Restore parent taskgroup for the current task
  taskdata->td_taskgroup = taskgroup->parent;
  __kmp_thread_free(thread, taskgroup);

  KA_TRACE(10, ("__kmpc_end_taskgroup(exit): T#%d task %p finished waiting\n",
                gtid, taskdata));

#if OMPT_SUPPORT && OMPT_OPTIONAL
  if (UNLIKELY(ompt_enabled.ompt_callback_sync_region)) {
    ompt_callbacks.ompt_callback(ompt_callback_sync_region)(
        ompt_sync_region_taskgroup, ompt_scope_end, &(my_parallel_data),
        &(my_task_data), codeptr);
  }
#endif
}

static kmp_task_t *__kmp_get_priority_task(kmp_int32 gtid,
                                           kmp_task_team_t *task_team,
                                           kmp_int32 is_constrained) {
  kmp_task_t *task = NULL;
  kmp_taskdata_t *taskdata;
  kmp_taskdata_t *current;
  kmp_thread_data_t *thread_data;
  int ntasks = task_team->tt.tt_num_task_pri;
  if (ntasks == 0) {
    KA_TRACE(
        20, ("__kmp_get_priority_task(exit #1): T#%d No tasks to get\n", gtid));
    return NULL;
  }
  do {
    // decrement num_tasks to "reserve" one task to get for execution
    if (__kmp_atomic_compare_store(&task_team->tt.tt_num_task_pri, ntasks,
                                   ntasks - 1))
      break;
  } while (ntasks > 0);
  if (ntasks == 0) {
    KA_TRACE(20, ("__kmp_get_priority_task(exit #2): T#%d No tasks to get\n",
                  __kmp_get_gtid()));
    return NULL;
  }
  // We got a "ticket" to get a "reserved" priority task
  int deque_ntasks;
  kmp_task_pri_t *list = task_team->tt.tt_task_pri_list;
  do {
    KMP_ASSERT(list != NULL);
    thread_data = &list->td;
    __kmp_acquire_bootstrap_lock(&thread_data->td.td_deque_lock);
    deque_ntasks = thread_data->td.td_deque_ntasks;
    if (deque_ntasks == 0) {
      __kmp_release_bootstrap_lock(&thread_data->td.td_deque_lock);
      KA_TRACE(20, ("__kmp_get_priority_task: T#%d No tasks to get from %p\n",
                    __kmp_get_gtid(), thread_data));
      list = list->next;
    }
  } while (deque_ntasks == 0);
  KMP_DEBUG_ASSERT(deque_ntasks);
  int target = thread_data->td.td_deque_head;
  current = __kmp_threads[gtid]->th.th_current_task;
  taskdata = thread_data->td.td_deque[target];
  if (__kmp_task_is_allowed(gtid, is_constrained, taskdata, current)) {
    // Bump head pointer and Wrap.
    thread_data->td.td_deque_head =
        (target + 1) & TASK_DEQUE_MASK(thread_data->td);
  } else {
    if (!task_team->tt.tt_untied_task_encountered) {
      // The TSC does not allow to steal victim task
      __kmp_release_bootstrap_lock(&thread_data->td.td_deque_lock);
      KA_TRACE(20, ("__kmp_get_priority_task(exit #3): T#%d could not get task "
                    "from %p: task_team=%p ntasks=%d head=%u tail=%u\n",
                    gtid, thread_data, task_team, deque_ntasks, target,
                    thread_data->td.td_deque_tail));
      task_team->tt.tt_num_task_pri++; // atomic inc, restore value
      return NULL;
    }
    int i;
    // walk through the deque trying to steal any task
    taskdata = NULL;
    for (i = 1; i < deque_ntasks; ++i) {
      target = (target + 1) & TASK_DEQUE_MASK(thread_data->td);
      taskdata = thread_data->td.td_deque[target];
      if (__kmp_task_is_allowed(gtid, is_constrained, taskdata, current)) {
        break; // found task to execute
      } else {
        taskdata = NULL;
      }
    }
    if (taskdata == NULL) {
      // No appropriate candidate found to execute
      __kmp_release_bootstrap_lock(&thread_data->td.td_deque_lock);
      KA_TRACE(
          10, ("__kmp_get_priority_task(exit #4): T#%d could not get task from "
               "%p: task_team=%p ntasks=%d head=%u tail=%u\n",
               gtid, thread_data, task_team, deque_ntasks,
               thread_data->td.td_deque_head, thread_data->td.td_deque_tail));
      task_team->tt.tt_num_task_pri++; // atomic inc, restore value
      return NULL;
    }
    int prev = target;
    for (i = i + 1; i < deque_ntasks; ++i) {
      // shift remaining tasks in the deque left by 1
      target = (target + 1) & TASK_DEQUE_MASK(thread_data->td);
      thread_data->td.td_deque[prev] = thread_data->td.td_deque[target];
      prev = target;
    }
    KMP_DEBUG_ASSERT(
        thread_data->td.td_deque_tail ==
        (kmp_uint32)((target + 1) & TASK_DEQUE_MASK(thread_data->td)));
    thread_data->td.td_deque_tail = target; // tail -= 1 (wrapped))
  }
  thread_data->td.td_deque_ntasks = deque_ntasks - 1;
  __kmp_release_bootstrap_lock(&thread_data->td.td_deque_lock);
  task = KMP_TASKDATA_TO_TASK(taskdata);
  return task;
}

// __kmp_remove_my_task: remove a task from my own deque
static kmp_task_t *__kmp_remove_my_task(kmp_info_t *thread, kmp_int32 gtid,
                                        kmp_task_team_t *task_team,
                                        kmp_int32 is_constrained) {
  kmp_task_t *task;
  kmp_taskdata_t *taskdata;
  kmp_thread_data_t *thread_data;
  kmp_uint32 tail;

  KMP_DEBUG_ASSERT(__kmp_tasking_mode != tskm_immediate_exec);
  KMP_DEBUG_ASSERT(task_team->tt.tt_threads_data !=
                   NULL); // Caller should check this condition

  thread_data = &task_team->tt.tt_threads_data[__kmp_tid_from_gtid(gtid)];

  KA_TRACE(10, ("__kmp_remove_my_task(enter): T#%d ntasks=%d head=%u tail=%u\n",
                gtid, thread_data->td.td_deque_ntasks,
                thread_data->td.td_deque_head, thread_data->td.td_deque_tail));

  if (TCR_4(thread_data->td.td_deque_ntasks) == 0) {
    KA_TRACE(10,
             ("__kmp_remove_my_task(exit #1): T#%d No tasks to remove: "
              "ntasks=%d head=%u tail=%u\n",
              gtid, thread_data->td.td_deque_ntasks,
              thread_data->td.td_deque_head, thread_data->td.td_deque_tail));
    return NULL;
  }

  __kmp_acquire_bootstrap_lock(&thread_data->td.td_deque_lock);

  if (TCR_4(thread_data->td.td_deque_ntasks) == 0) {
    __kmp_release_bootstrap_lock(&thread_data->td.td_deque_lock);
    KA_TRACE(10,
             ("__kmp_remove_my_task(exit #2): T#%d No tasks to remove: "
              "ntasks=%d head=%u tail=%u\n",
              gtid, thread_data->td.td_deque_ntasks,
              thread_data->td.td_deque_head, thread_data->td.td_deque_tail));
    return NULL;
  }

  tail = (thread_data->td.td_deque_tail - 1) &
         TASK_DEQUE_MASK(thread_data->td); // Wrap index.
  taskdata = thread_data->td.td_deque[tail];

  if (!__kmp_task_is_allowed(gtid, is_constrained, taskdata,
                             thread->th.th_current_task)) {
    // The TSC does not allow to steal victim task
    __kmp_release_bootstrap_lock(&thread_data->td.td_deque_lock);
    KA_TRACE(10,
             ("__kmp_remove_my_task(exit #3): T#%d TSC blocks tail task: "
              "ntasks=%d head=%u tail=%u\n",
              gtid, thread_data->td.td_deque_ntasks,
              thread_data->td.td_deque_head, thread_data->td.td_deque_tail));
    return NULL;
  }

  thread_data->td.td_deque_tail = tail;
  TCW_4(thread_data->td.td_deque_ntasks, thread_data->td.td_deque_ntasks - 1);

  __kmp_release_bootstrap_lock(&thread_data->td.td_deque_lock);

  KA_TRACE(10, ("__kmp_remove_my_task(exit #4): T#%d task %p removed: "
                "ntasks=%d head=%u tail=%u\n",
                gtid, taskdata, thread_data->td.td_deque_ntasks,
                thread_data->td.td_deque_head, thread_data->td.td_deque_tail));

  task = KMP_TASKDATA_TO_TASK(taskdata);
  return task;
}

// __kmp_steal_task: remove a task from another thread's deque
// Assume that calling thread has already checked existence of
// task_team thread_data before calling this routine.
static kmp_task_t *__kmp_steal_task(kmp_info_t *victim_thr, kmp_int32 gtid,
                                    kmp_task_team_t *task_team,
                                    std::atomic<kmp_int32> *unfinished_threads,
                                    int *thread_finished,
                                    kmp_int32 is_constrained) {
  kmp_task_t *task;
  kmp_taskdata_t *taskdata;
  kmp_taskdata_t *current;
  kmp_thread_data_t *victim_td, *threads_data;
  kmp_int32 target;
  kmp_int32 victim_tid;

  KMP_DEBUG_ASSERT(__kmp_tasking_mode != tskm_immediate_exec);

  threads_data = task_team->tt.tt_threads_data;
  KMP_DEBUG_ASSERT(threads_data != NULL); // Caller should check this condition

  victim_tid = victim_thr->th.th_info.ds.ds_tid;
  victim_td = &threads_data[victim_tid];

  KA_TRACE(10, ("__kmp_steal_task(enter): T#%d try to steal from T#%d: "
                "task_team=%p ntasks=%d head=%u tail=%u\n",
                gtid, __kmp_gtid_from_thread(victim_thr), task_team,
                victim_td->td.td_deque_ntasks, victim_td->td.td_deque_head,
                victim_td->td.td_deque_tail));

  if (TCR_4(victim_td->td.td_deque_ntasks) == 0) {
    KA_TRACE(10, ("__kmp_steal_task(exit #1): T#%d could not steal from T#%d: "
                  "task_team=%p ntasks=%d head=%u tail=%u\n",
                  gtid, __kmp_gtid_from_thread(victim_thr), task_team,
                  victim_td->td.td_deque_ntasks, victim_td->td.td_deque_head,
                  victim_td->td.td_deque_tail));
    return NULL;
  }

  __kmp_acquire_bootstrap_lock(&victim_td->td.td_deque_lock);

  int ntasks = TCR_4(victim_td->td.td_deque_ntasks);
  // Check again after we acquire the lock
  if (ntasks == 0) {
    __kmp_release_bootstrap_lock(&victim_td->td.td_deque_lock);
    KA_TRACE(10, ("__kmp_steal_task(exit #2): T#%d could not steal from T#%d: "
                  "task_team=%p ntasks=%d head=%u tail=%u\n",
                  gtid, __kmp_gtid_from_thread(victim_thr), task_team, ntasks,
                  victim_td->td.td_deque_head, victim_td->td.td_deque_tail));
    return NULL;
  }

  KMP_DEBUG_ASSERT(victim_td->td.td_deque != NULL);
  current = __kmp_threads[gtid]->th.th_current_task;
  taskdata = victim_td->td.td_deque[victim_td->td.td_deque_head];
  if (__kmp_task_is_allowed(gtid, is_constrained, taskdata, current)) {
    // Bump head pointer and Wrap.
    victim_td->td.td_deque_head =
        (victim_td->td.td_deque_head + 1) & TASK_DEQUE_MASK(victim_td->td);
  } else {
    if (!task_team->tt.tt_untied_task_encountered) {
      // The TSC does not allow to steal victim task
      __kmp_release_bootstrap_lock(&victim_td->td.td_deque_lock);
      KA_TRACE(10, ("__kmp_steal_task(exit #3): T#%d could not steal from "
                    "T#%d: task_team=%p ntasks=%d head=%u tail=%u\n",
                    gtid, __kmp_gtid_from_thread(victim_thr), task_team, ntasks,
                    victim_td->td.td_deque_head, victim_td->td.td_deque_tail));
      return NULL;
    }
    int i;
    // walk through victim's deque trying to steal any task
    target = victim_td->td.td_deque_head;
    taskdata = NULL;
    for (i = 1; i < ntasks; ++i) {
      target = (target + 1) & TASK_DEQUE_MASK(victim_td->td);
      taskdata = victim_td->td.td_deque[target];
      if (__kmp_task_is_allowed(gtid, is_constrained, taskdata, current)) {
        break; // found victim task
      } else {
        taskdata = NULL;
      }
    }
    if (taskdata == NULL) {
      // No appropriate candidate to steal found
      __kmp_release_bootstrap_lock(&victim_td->td.td_deque_lock);
      KA_TRACE(10, ("__kmp_steal_task(exit #4): T#%d could not steal from "
                    "T#%d: task_team=%p ntasks=%d head=%u tail=%u\n",
                    gtid, __kmp_gtid_from_thread(victim_thr), task_team, ntasks,
                    victim_td->td.td_deque_head, victim_td->td.td_deque_tail));
      return NULL;
    }
    int prev = target;
    for (i = i + 1; i < ntasks; ++i) {
      // shift remaining tasks in the deque left by 1
      target = (target + 1) & TASK_DEQUE_MASK(victim_td->td);
      victim_td->td.td_deque[prev] = victim_td->td.td_deque[target];
      prev = target;
    }
    KMP_DEBUG_ASSERT(
        victim_td->td.td_deque_tail ==
        (kmp_uint32)((target + 1) & TASK_DEQUE_MASK(victim_td->td)));
    victim_td->td.td_deque_tail = target; // tail -= 1 (wrapped))
  }
  if (*thread_finished) {
    // We need to un-mark this victim as a finished victim.  This must be done
    // before releasing the lock, or else other threads (starting with the
    // primary thread victim) might be prematurely released from the barrier!!!
#if KMP_DEBUG
    kmp_int32 count =
#endif
        KMP_ATOMIC_INC(unfinished_threads);
    KA_TRACE(
        20,
        ("__kmp_steal_task: T#%d inc unfinished_threads to %d: task_team=%p\n",
         gtid, count + 1, task_team));
    *thread_finished = FALSE;
  }
  TCW_4(victim_td->td.td_deque_ntasks, ntasks - 1);

  __kmp_release_bootstrap_lock(&victim_td->td.td_deque_lock);

  KMP_COUNT_BLOCK(TASK_stolen);
  KA_TRACE(10,
           ("__kmp_steal_task(exit #5): T#%d stole task %p from T#%d: "
            "task_team=%p ntasks=%d head=%u tail=%u\n",
            gtid, taskdata, __kmp_gtid_from_thread(victim_thr), task_team,
            ntasks, victim_td->td.td_deque_head, victim_td->td.td_deque_tail));

  task = KMP_TASKDATA_TO_TASK(taskdata);
  return task;
}

// __kmp_execute_tasks_template: Choose and execute tasks until either the
// condition is statisfied (return true) or there are none left (return false).
//
// final_spin is TRUE if this is the spin at the release barrier.
// thread_finished indicates whether the thread is finished executing all
// the tasks it has on its deque, and is at the release barrier.
// spinner is the location on which to spin.
// spinner == NULL means only execute a single task and return.
// checker is the value to check to terminate the spin.
template <class C>
static inline int __kmp_execute_tasks_template(
    kmp_info_t *thread, kmp_int32 gtid, C *flag, int final_spin,
    int *thread_finished USE_ITT_BUILD_ARG(void *itt_sync_obj),
    kmp_int32 is_constrained) {
  kmp_task_team_t *task_team = thread->th.th_task_team;
  kmp_thread_data_t *threads_data;
  kmp_task_t *task;
  kmp_info_t *other_thread;
  kmp_taskdata_t *current_task = thread->th.th_current_task;
  std::atomic<kmp_int32> *unfinished_threads;
  kmp_int32 nthreads, victim_tid = -2, use_own_tasks = 1, new_victim = 0,
                      tid = thread->th.th_info.ds.ds_tid;

  KMP_DEBUG_ASSERT(__kmp_tasking_mode != tskm_immediate_exec);
  KMP_DEBUG_ASSERT(thread == __kmp_threads[gtid]);

  if (task_team == NULL || current_task == NULL)
    return FALSE;

  KA_TRACE(15, ("__kmp_execute_tasks_template(enter): T#%d final_spin=%d "
                "*thread_finished=%d\n",
                gtid, final_spin, *thread_finished));

  thread->th.th_reap_state = KMP_NOT_SAFE_TO_REAP;
  threads_data = (kmp_thread_data_t *)TCR_PTR(task_team->tt.tt_threads_data);

  KMP_DEBUG_ASSERT(threads_data != NULL);

  nthreads = task_team->tt.tt_nproc;
  unfinished_threads = &(task_team->tt.tt_unfinished_threads);
  KMP_DEBUG_ASSERT(nthreads > 1 || task_team->tt.tt_found_proxy_tasks ||
                   task_team->tt.tt_hidden_helper_task_encountered);
  KMP_DEBUG_ASSERT(*unfinished_threads >= 0);

  while (1) { // Outer loop keeps trying to find tasks in case of single thread
    // getting tasks from target constructs
    while (1) { // Inner loop to find a task and execute it
      task = NULL;
      if (task_team->tt.tt_num_task_pri) { // get priority task first
        task = __kmp_get_priority_task(gtid, task_team, is_constrained);
      }
      if (task == NULL && use_own_tasks) { // check own queue next
        task = __kmp_remove_my_task(thread, gtid, task_team, is_constrained);
      }
      if ((task == NULL) && (nthreads > 1)) { // Steal a task finally
        int asleep = 1;
        use_own_tasks = 0;
        // Try to steal from the last place I stole from successfully.
        if (victim_tid == -2) { // haven't stolen anything yet
          victim_tid = threads_data[tid].td.td_deque_last_stolen;
          if (victim_tid !=
              -1) // if we have a last stolen from victim, get the thread
            other_thread = threads_data[victim_tid].td.td_thr;
        }
        if (victim_tid != -1) { // found last victim
          asleep = 0;
        } else if (!new_victim) { // no recent steals and we haven't already
          // used a new victim; select a random thread
          do { // Find a different thread to steal work from.
            // Pick a random thread. Initial plan was to cycle through all the
            // threads, and only return if we tried to steal from every thread,
            // and failed.  Arch says that's not such a great idea.
            victim_tid = __kmp_get_random(thread) % (nthreads - 1);
            if (victim_tid >= tid) {
              ++victim_tid; // Adjusts random distribution to exclude self
            }
            // Found a potential victim
            other_thread = threads_data[victim_tid].td.td_thr;
            // There is a slight chance that __kmp_enable_tasking() did not wake
            // up all threads waiting at the barrier.  If victim is sleeping,
            // then wake it up. Since we were going to pay the cache miss
            // penalty for referencing another thread's kmp_info_t struct
            // anyway,
            // the check shouldn't cost too much performance at this point. In
            // extra barrier mode, tasks do not sleep at the separate tasking
            // barrier, so this isn't a problem.
            asleep = 0;
            if ((__kmp_tasking_mode == tskm_task_teams) &&
                (__kmp_dflt_blocktime != KMP_MAX_BLOCKTIME) &&
                (TCR_PTR(CCAST(void *, other_thread->th.th_sleep_loc)) !=
                 NULL)) {
              asleep = 1;
              __kmp_null_resume_wrapper(other_thread);
              // A sleeping thread should not have any tasks on it's queue.
              // There is a slight possibility that it resumes, steals a task
              // from another thread, which spawns more tasks, all in the time
              // that it takes this thread to check => don't write an assertion
              // that the victim's queue is empty.  Try stealing from a
              // different thread.
            }
          } while (asleep);
        }

        if (!asleep) {
          // We have a victim to try to steal from
          task = __kmp_steal_task(other_thread, gtid, task_team,
                                  unfinished_threads, thread_finished,
                                  is_constrained);
        }
        if (task != NULL) { // set last stolen to victim
          if (threads_data[tid].td.td_deque_last_stolen != victim_tid) {
            threads_data[tid].td.td_deque_last_stolen = victim_tid;
            // The pre-refactored code did not try more than 1 successful new
            // vicitm, unless the last one generated more local tasks;
            // new_victim keeps track of this
            new_victim = 1;
          }
        } else { // No tasks found; unset last_stolen
          KMP_CHECK_UPDATE(threads_data[tid].td.td_deque_last_stolen, -1);
          victim_tid = -2; // no successful victim found
        }
      }

      if (task == NULL)
        break; // break out of tasking loop

// Found a task; execute it
#if USE_ITT_BUILD && USE_ITT_NOTIFY
      if (__itt_sync_create_ptr || KMP_ITT_DEBUG) {
        if (itt_sync_obj == NULL) { // we are at fork barrier where we could not
          // get the object reliably
          itt_sync_obj = __kmp_itt_barrier_object(gtid, bs_forkjoin_barrier);
        }
        __kmp_itt_task_starting(itt_sync_obj);
      }
#endif /* USE_ITT_BUILD && USE_ITT_NOTIFY */
      __kmp_invoke_task(gtid, task, current_task);
#if USE_ITT_BUILD
      if (itt_sync_obj != NULL)
        __kmp_itt_task_finished(itt_sync_obj);
#endif /* USE_ITT_BUILD */
      // If this thread is only partway through the barrier and the condition is
      // met, then return now, so that the barrier gather/release pattern can
      // proceed. If this thread is in the last spin loop in the barrier,
      // waiting to be released, we know that the termination condition will not
      // be satisfied, so don't waste any cycles checking it.
      if (flag == NULL || (!final_spin && flag->done_check())) {
        KA_TRACE(
            15,
            ("__kmp_execute_tasks_template: T#%d spin condition satisfied\n",
             gtid));
        return TRUE;
      }
      if (thread->th.th_task_team == NULL) {
        break;
      }
      KMP_YIELD(__kmp_library == library_throughput); // Yield before next task
      // If execution of a stolen task results in more tasks being placed on our
      // run queue, reset use_own_tasks
      if (!use_own_tasks && TCR_4(threads_data[tid].td.td_deque_ntasks) != 0) {
        KA_TRACE(20, ("__kmp_execute_tasks_template: T#%d stolen task spawned "
                      "other tasks, restart\n",
                      gtid));
        use_own_tasks = 1;
        new_victim = 0;
      }
    }

    // The task source has been exhausted. If in final spin loop of barrier,
    // check if termination condition is satisfied. The work queue may be empty
    // but there might be proxy tasks still executing.
    if (final_spin &&
        KMP_ATOMIC_LD_ACQ(&current_task->td_incomplete_child_tasks) == 0) {
      // First, decrement the #unfinished threads, if that has not already been
      // done.  This decrement might be to the spin location, and result in the
      // termination condition being satisfied.
      if (!*thread_finished) {
#if KMP_DEBUG
        kmp_int32 count = -1 +
#endif
            KMP_ATOMIC_DEC(unfinished_threads);
        KA_TRACE(20, ("__kmp_execute_tasks_template: T#%d dec "
                      "unfinished_threads to %d task_team=%p\n",
                      gtid, count, task_team));
        *thread_finished = TRUE;
      }

      // It is now unsafe to reference thread->th.th_team !!!
      // Decrementing task_team->tt.tt_unfinished_threads can allow the primary
      // thread to pass through the barrier, where it might reset each thread's
      // th.th_team field for the next parallel region. If we can steal more
      // work, we know that this has not happened yet.
      if (flag != NULL && flag->done_check()) {
        KA_TRACE(
            15,
            ("__kmp_execute_tasks_template: T#%d spin condition satisfied\n",
             gtid));
        return TRUE;
      }
    }

    // If this thread's task team is NULL, primary thread has recognized that
    // there are no more tasks; bail out
    if (thread->th.th_task_team == NULL) {
      KA_TRACE(15,
               ("__kmp_execute_tasks_template: T#%d no more tasks\n", gtid));
      return FALSE;
    }

    // Check the flag again to see if it has already done in case to be trapped
    // into infinite loop when a if0 task depends on a hidden helper task
    // outside any parallel region. Detached tasks are not impacted in this case
    // because the only thread executing this function has to execute the proxy
    // task so it is in another code path that has the same check.
    if (flag == NULL || (!final_spin && flag->done_check())) {
      KA_TRACE(15,
               ("__kmp_execute_tasks_template: T#%d spin condition satisfied\n",
                gtid));
      return TRUE;
    }

    // We could be getting tasks from target constructs; if this is the only
    // thread, keep trying to execute tasks from own queue
    if (nthreads == 1 &&
        KMP_ATOMIC_LD_ACQ(&current_task->td_incomplete_child_tasks))
      use_own_tasks = 1;
    else {
      KA_TRACE(15,
               ("__kmp_execute_tasks_template: T#%d can't find work\n", gtid));
      return FALSE;
    }
  }
}

template <bool C, bool S>
int __kmp_execute_tasks_32(
    kmp_info_t *thread, kmp_int32 gtid, kmp_flag_32<C, S> *flag, int final_spin,
    int *thread_finished USE_ITT_BUILD_ARG(void *itt_sync_obj),
    kmp_int32 is_constrained) {
  return __kmp_execute_tasks_template(
      thread, gtid, flag, final_spin,
      thread_finished USE_ITT_BUILD_ARG(itt_sync_obj), is_constrained);
}

template <bool C, bool S>
int __kmp_execute_tasks_64(
    kmp_info_t *thread, kmp_int32 gtid, kmp_flag_64<C, S> *flag, int final_spin,
    int *thread_finished USE_ITT_BUILD_ARG(void *itt_sync_obj),
    kmp_int32 is_constrained) {
  return __kmp_execute_tasks_template(
      thread, gtid, flag, final_spin,
      thread_finished USE_ITT_BUILD_ARG(itt_sync_obj), is_constrained);
}

template <bool C, bool S>
int __kmp_atomic_execute_tasks_64(
    kmp_info_t *thread, kmp_int32 gtid, kmp_atomic_flag_64<C, S> *flag,
    int final_spin, int *thread_finished USE_ITT_BUILD_ARG(void *itt_sync_obj),
    kmp_int32 is_constrained) {
  return __kmp_execute_tasks_template(
      thread, gtid, flag, final_spin,
      thread_finished USE_ITT_BUILD_ARG(itt_sync_obj), is_constrained);
}

int __kmp_execute_tasks_oncore(
    kmp_info_t *thread, kmp_int32 gtid, kmp_flag_oncore *flag, int final_spin,
    int *thread_finished USE_ITT_BUILD_ARG(void *itt_sync_obj),
    kmp_int32 is_constrained) {
  return __kmp_execute_tasks_template(
      thread, gtid, flag, final_spin,
      thread_finished USE_ITT_BUILD_ARG(itt_sync_obj), is_constrained);
}

template int
__kmp_execute_tasks_32<false, false>(kmp_info_t *, kmp_int32,
                                     kmp_flag_32<false, false> *, int,
                                     int *USE_ITT_BUILD_ARG(void *), kmp_int32);

template int __kmp_execute_tasks_64<false, true>(kmp_info_t *, kmp_int32,
                                                 kmp_flag_64<false, true> *,
                                                 int,
                                                 int *USE_ITT_BUILD_ARG(void *),
                                                 kmp_int32);

template int __kmp_execute_tasks_64<true, false>(kmp_info_t *, kmp_int32,
                                                 kmp_flag_64<true, false> *,
                                                 int,
                                                 int *USE_ITT_BUILD_ARG(void *),
                                                 kmp_int32);

template int __kmp_atomic_execute_tasks_64<false, true>(
    kmp_info_t *, kmp_int32, kmp_atomic_flag_64<false, true> *, int,
    int *USE_ITT_BUILD_ARG(void *), kmp_int32);

template int __kmp_atomic_execute_tasks_64<true, false>(
    kmp_info_t *, kmp_int32, kmp_atomic_flag_64<true, false> *, int,
    int *USE_ITT_BUILD_ARG(void *), kmp_int32);

// __kmp_enable_tasking: Allocate task team and resume threads sleeping at the
// next barrier so they can assist in executing enqueued tasks.
// First thread in allocates the task team atomically.
static void __kmp_enable_tasking(kmp_task_team_t *task_team,
                                 kmp_info_t *this_thr) {
  kmp_thread_data_t *threads_data;
  int nthreads, i, is_init_thread;

  KA_TRACE(10, ("__kmp_enable_tasking(enter): T#%d\n",
                __kmp_gtid_from_thread(this_thr)));

  KMP_DEBUG_ASSERT(task_team != NULL);
  KMP_DEBUG_ASSERT(this_thr->th.th_team != NULL);

  nthreads = task_team->tt.tt_nproc;
  KMP_DEBUG_ASSERT(nthreads > 0);
  KMP_DEBUG_ASSERT(nthreads == this_thr->th.th_team->t.t_nproc);

  // Allocate or increase the size of threads_data if necessary
  is_init_thread = __kmp_realloc_task_threads_data(this_thr, task_team);

  if (!is_init_thread) {
    // Some other thread already set up the array.
    KA_TRACE(
        20,
        ("__kmp_enable_tasking(exit): T#%d: threads array already set up.\n",
         __kmp_gtid_from_thread(this_thr)));
    return;
  }
  threads_data = (kmp_thread_data_t *)TCR_PTR(task_team->tt.tt_threads_data);
  KMP_DEBUG_ASSERT(threads_data != NULL);

  if (__kmp_tasking_mode == tskm_task_teams &&
      (__kmp_dflt_blocktime != KMP_MAX_BLOCKTIME)) {
    // Release any threads sleeping at the barrier, so that they can steal
    // tasks and execute them.  In extra barrier mode, tasks do not sleep
    // at the separate tasking barrier, so this isn't a problem.
    for (i = 0; i < nthreads; i++) {
      void *sleep_loc;
      kmp_info_t *thread = threads_data[i].td.td_thr;

      if (i == this_thr->th.th_info.ds.ds_tid) {
        continue;
      }
      // Since we haven't locked the thread's suspend mutex lock at this
      // point, there is a small window where a thread might be putting
      // itself to sleep, but hasn't set the th_sleep_loc field yet.
      // To work around this, __kmp_execute_tasks_template() periodically checks
      // see if other threads are sleeping (using the same random mechanism that
      // is used for task stealing) and awakens them if they are.
      if ((sleep_loc = TCR_PTR(CCAST(void *, thread->th.th_sleep_loc))) !=
          NULL) {
        KF_TRACE(50, ("__kmp_enable_tasking: T#%d waking up thread T#%d\n",
                      __kmp_gtid_from_thread(this_thr),
                      __kmp_gtid_from_thread(thread)));
        __kmp_null_resume_wrapper(thread);
      } else {
        KF_TRACE(50, ("__kmp_enable_tasking: T#%d don't wake up thread T#%d\n",
                      __kmp_gtid_from_thread(this_thr),
                      __kmp_gtid_from_thread(thread)));
      }
    }
  }

  KA_TRACE(10, ("__kmp_enable_tasking(exit): T#%d\n",
                __kmp_gtid_from_thread(this_thr)));
}

/* // TODO: Check the comment consistency
 * Utility routines for "task teams".  A task team (kmp_task_t) is kind of
 * like a shadow of the kmp_team_t data struct, with a different lifetime.
 * After a child * thread checks into a barrier and calls __kmp_release() from
 * the particular variant of __kmp_<barrier_kind>_barrier_gather(), it can no
 * longer assume that the kmp_team_t structure is intact (at any moment, the
 * primary thread may exit the barrier code and free the team data structure,
 * and return the threads to the thread pool).
 *
 * This does not work with the tasking code, as the thread is still
 * expected to participate in the execution of any tasks that may have been
 * spawned my a member of the team, and the thread still needs access to all
 * to each thread in the team, so that it can steal work from it.
 *
 * Enter the existence of the kmp_task_team_t struct.  It employs a reference
 * counting mechanism, and is allocated by the primary thread before calling
 * __kmp_<barrier_kind>_release, and then is release by the last thread to
 * exit __kmp_<barrier_kind>_release at the next barrier.  I.e. the lifetimes
 * of the kmp_task_team_t structs for consecutive barriers can overlap
 * (and will, unless the primary thread is the last thread to exit the barrier
 * release phase, which is not typical). The existence of such a struct is
 * useful outside the context of tasking.
 *
 * We currently use the existence of the threads array as an indicator that
 * tasks were spawned since the last barrier.  If the structure is to be
 * useful outside the context of tasking, then this will have to change, but
 * not setting the field minimizes the performance impact of tasking on
 * barriers, when no explicit tasks were spawned (pushed, actually).
 */

static kmp_task_team_t *__kmp_free_task_teams =
    NULL; // Free list for task_team data structures
// Lock for task team data structures
kmp_bootstrap_lock_t __kmp_task_team_lock =
    KMP_BOOTSTRAP_LOCK_INITIALIZER(__kmp_task_team_lock);

// __kmp_alloc_task_deque:
// Allocates a task deque for a particular thread, and initialize the necessary
// data structures relating to the deque.  This only happens once per thread
// per task team since task teams are recycled. No lock is needed during
// allocation since each thread allocates its own deque.
static void __kmp_alloc_task_deque(kmp_info_t *thread,
                                   kmp_thread_data_t *thread_data) {
  __kmp_init_bootstrap_lock(&thread_data->td.td_deque_lock);
  KMP_DEBUG_ASSERT(thread_data->td.td_deque == NULL);

  // Initialize last stolen task field to "none"
  thread_data->td.td_deque_last_stolen = -1;

  KMP_DEBUG_ASSERT(TCR_4(thread_data->td.td_deque_ntasks) == 0);
  KMP_DEBUG_ASSERT(thread_data->td.td_deque_head == 0);
  KMP_DEBUG_ASSERT(thread_data->td.td_deque_tail == 0);

  KE_TRACE(
      10,
      ("__kmp_alloc_task_deque: T#%d allocating deque[%d] for thread_data %p\n",
       __kmp_gtid_from_thread(thread), INITIAL_TASK_DEQUE_SIZE, thread_data));
  // Allocate space for task deque, and zero the deque
  // Cannot use __kmp_thread_calloc() because threads not around for
  // kmp_reap_task_team( ).
  thread_data->td.td_deque = (kmp_taskdata_t **)__kmp_allocate(
      INITIAL_TASK_DEQUE_SIZE * sizeof(kmp_taskdata_t *));
  thread_data->td.td_deque_size = INITIAL_TASK_DEQUE_SIZE;
}

// __kmp_free_task_deque:
// Deallocates a task deque for a particular thread. Happens at library
// deallocation so don't need to reset all thread data fields.
static void __kmp_free_task_deque(kmp_thread_data_t *thread_data) {
  if (thread_data->td.td_deque != NULL) {
    __kmp_acquire_bootstrap_lock(&thread_data->td.td_deque_lock);
    TCW_4(thread_data->td.td_deque_ntasks, 0);
    __kmp_free(thread_data->td.td_deque);
    thread_data->td.td_deque = NULL;
    __kmp_release_bootstrap_lock(&thread_data->td.td_deque_lock);
  }

#ifdef BUILD_TIED_TASK_STACK
  // GEH: Figure out what to do here for td_susp_tied_tasks
  if (thread_data->td.td_susp_tied_tasks.ts_entries != TASK_STACK_EMPTY) {
    __kmp_free_task_stack(__kmp_thread_from_gtid(gtid), thread_data);
  }
#endif // BUILD_TIED_TASK_STACK
}

// __kmp_realloc_task_threads_data:
// Allocates a threads_data array for a task team, either by allocating an
// initial array or enlarging an existing array.  Only the first thread to get
// the lock allocs or enlarges the array and re-initializes the array elements.
// That thread returns "TRUE", the rest return "FALSE".
// Assumes that the new array size is given by task_team -> tt.tt_nproc.
// The current size is given by task_team -> tt.tt_max_threads.
static int __kmp_realloc_task_threads_data(kmp_info_t *thread,
                                           kmp_task_team_t *task_team) {
  kmp_thread_data_t **threads_data_p;
  kmp_int32 nthreads, maxthreads;
  int is_init_thread = FALSE;

  if (TCR_4(task_team->tt.tt_found_tasks)) {
    // Already reallocated and initialized.
    return FALSE;
  }

  threads_data_p = &task_team->tt.tt_threads_data;
  nthreads = task_team->tt.tt_nproc;
  maxthreads = task_team->tt.tt_max_threads;

  // All threads must lock when they encounter the first task of the implicit
  // task region to make sure threads_data fields are (re)initialized before
  // used.
  __kmp_acquire_bootstrap_lock(&task_team->tt.tt_threads_lock);

  if (!TCR_4(task_team->tt.tt_found_tasks)) {
    // first thread to enable tasking
    kmp_team_t *team = thread->th.th_team;
    int i;

    is_init_thread = TRUE;
    if (maxthreads < nthreads) {

      if (*threads_data_p != NULL) {
        kmp_thread_data_t *old_data = *threads_data_p;
        kmp_thread_data_t *new_data = NULL;

        KE_TRACE(
            10,
            ("__kmp_realloc_task_threads_data: T#%d reallocating "
             "threads data for task_team %p, new_size = %d, old_size = %d\n",
             __kmp_gtid_from_thread(thread), task_team, nthreads, maxthreads));
        // Reallocate threads_data to have more elements than current array
        // Cannot use __kmp_thread_realloc() because threads not around for
        // kmp_reap_task_team( ).  Note all new array entries are initialized
        // to zero by __kmp_allocate().
        new_data = (kmp_thread_data_t *)__kmp_allocate(
            nthreads * sizeof(kmp_thread_data_t));
        // copy old data to new data
        KMP_MEMCPY_S((void *)new_data, nthreads * sizeof(kmp_thread_data_t),
                     (void *)old_data, maxthreads * sizeof(kmp_thread_data_t));

#ifdef BUILD_TIED_TASK_STACK
        // GEH: Figure out if this is the right thing to do
        for (i = maxthreads; i < nthreads; i++) {
          kmp_thread_data_t *thread_data = &(*threads_data_p)[i];
          __kmp_init_task_stack(__kmp_gtid_from_thread(thread), thread_data);
        }
#endif // BUILD_TIED_TASK_STACK
       // Install the new data and free the old data
        (*threads_data_p) = new_data;
        __kmp_free(old_data);
      } else {
        KE_TRACE(10, ("__kmp_realloc_task_threads_data: T#%d allocating "
                      "threads data for task_team %p, size = %d\n",
                      __kmp_gtid_from_thread(thread), task_team, nthreads));
        // Make the initial allocate for threads_data array, and zero entries
        // Cannot use __kmp_thread_calloc() because threads not around for
        // kmp_reap_task_team( ).
        *threads_data_p = (kmp_thread_data_t *)__kmp_allocate(
            nthreads * sizeof(kmp_thread_data_t));
#ifdef BUILD_TIED_TASK_STACK
        // GEH: Figure out if this is the right thing to do
        for (i = 0; i < nthreads; i++) {
          kmp_thread_data_t *thread_data = &(*threads_data_p)[i];
          __kmp_init_task_stack(__kmp_gtid_from_thread(thread), thread_data);
        }
#endif // BUILD_TIED_TASK_STACK
      }
      task_team->tt.tt_max_threads = nthreads;
    } else {
      // If array has (more than) enough elements, go ahead and use it
      KMP_DEBUG_ASSERT(*threads_data_p != NULL);
    }

    // initialize threads_data pointers back to thread_info structures
    for (i = 0; i < nthreads; i++) {
      kmp_thread_data_t *thread_data = &(*threads_data_p)[i];
      thread_data->td.td_thr = team->t.t_threads[i];

      if (thread_data->td.td_deque_last_stolen >= nthreads) {
        // The last stolen field survives across teams / barrier, and the number
        // of threads may have changed.  It's possible (likely?) that a new
        // parallel region will exhibit the same behavior as previous region.
        thread_data->td.td_deque_last_stolen = -1;
      }
    }

    KMP_MB();
    TCW_SYNC_4(task_team->tt.tt_found_tasks, TRUE);
  }

  __kmp_release_bootstrap_lock(&task_team->tt.tt_threads_lock);
  return is_init_thread;
}

// __kmp_free_task_threads_data:
// Deallocates a threads_data array for a task team, including any attached
// tasking deques.  Only occurs at library shutdown.
static void __kmp_free_task_threads_data(kmp_task_team_t *task_team) {
  __kmp_acquire_bootstrap_lock(&task_team->tt.tt_threads_lock);
  if (task_team->tt.tt_threads_data != NULL) {
    int i;
    for (i = 0; i < task_team->tt.tt_max_threads; i++) {
      __kmp_free_task_deque(&task_team->tt.tt_threads_data[i]);
    }
    __kmp_free(task_team->tt.tt_threads_data);
    task_team->tt.tt_threads_data = NULL;
  }
  __kmp_release_bootstrap_lock(&task_team->tt.tt_threads_lock);
}

// __kmp_free_task_pri_list:
// Deallocates tasking deques used for priority tasks.
// Only occurs at library shutdown.
static void __kmp_free_task_pri_list(kmp_task_team_t *task_team) {
  __kmp_acquire_bootstrap_lock(&task_team->tt.tt_task_pri_lock);
  if (task_team->tt.tt_task_pri_list != NULL) {
    kmp_task_pri_t *list = task_team->tt.tt_task_pri_list;
    while (list != NULL) {
      kmp_task_pri_t *next = list->next;
      __kmp_free_task_deque(&list->td);
      __kmp_free(list);
      list = next;
    }
    task_team->tt.tt_task_pri_list = NULL;
  }
  __kmp_release_bootstrap_lock(&task_team->tt.tt_task_pri_lock);
}

// __kmp_allocate_task_team:
// Allocates a task team associated with a specific team, taking it from
// the global task team free list if possible.  Also initializes data
// structures.
static kmp_task_team_t *__kmp_allocate_task_team(kmp_info_t *thread,
                                                 kmp_team_t *team) {
  kmp_task_team_t *task_team = NULL;
  int nthreads;

  KA_TRACE(20, ("__kmp_allocate_task_team: T#%d entering; team = %p\n",
                (thread ? __kmp_gtid_from_thread(thread) : -1), team));

  if (TCR_PTR(__kmp_free_task_teams) != NULL) {
    // Take a task team from the task team pool
    __kmp_acquire_bootstrap_lock(&__kmp_task_team_lock);
    if (__kmp_free_task_teams != NULL) {
      task_team = __kmp_free_task_teams;
      TCW_PTR(__kmp_free_task_teams, task_team->tt.tt_next);
      task_team->tt.tt_next = NULL;
    }
    __kmp_release_bootstrap_lock(&__kmp_task_team_lock);
  }

  if (task_team == NULL) {
    KE_TRACE(10, ("__kmp_allocate_task_team: T#%d allocating "
                  "task team for team %p\n",
                  __kmp_gtid_from_thread(thread), team));
    // Allocate a new task team if one is not available. Cannot use
    // __kmp_thread_malloc because threads not around for kmp_reap_task_team.
    task_team = (kmp_task_team_t *)__kmp_allocate(sizeof(kmp_task_team_t));
    __kmp_init_bootstrap_lock(&task_team->tt.tt_threads_lock);
    __kmp_init_bootstrap_lock(&task_team->tt.tt_task_pri_lock);
#if USE_ITT_BUILD && USE_ITT_NOTIFY && KMP_DEBUG
    // suppress race conditions detection on synchronization flags in debug mode
    // this helps to analyze library internals eliminating false positives
    __itt_suppress_mark_range(
        __itt_suppress_range, __itt_suppress_threading_errors,
        &task_team->tt.tt_found_tasks, sizeof(task_team->tt.tt_found_tasks));
    __itt_suppress_mark_range(__itt_suppress_range,
                              __itt_suppress_threading_errors,
                              CCAST(kmp_uint32 *, &task_team->tt.tt_active),
                              sizeof(task_team->tt.tt_active));
#endif /* USE_ITT_BUILD && USE_ITT_NOTIFY && KMP_DEBUG */
    // Note: __kmp_allocate zeroes returned memory, othewise we would need:
    // task_team->tt.tt_threads_data = NULL;
    // task_team->tt.tt_max_threads = 0;
    // task_team->tt.tt_next = NULL;
  }

  TCW_4(task_team->tt.tt_found_tasks, FALSE);
  TCW_4(task_team->tt.tt_found_proxy_tasks, FALSE);
  TCW_4(task_team->tt.tt_hidden_helper_task_encountered, FALSE);
  task_team->tt.tt_nproc = nthreads = team->t.t_nproc;

  KMP_ATOMIC_ST_REL(&task_team->tt.tt_unfinished_threads, nthreads);
  TCW_4(task_team->tt.tt_hidden_helper_task_encountered, FALSE);
  TCW_4(task_team->tt.tt_active, TRUE);

  KA_TRACE(20, ("__kmp_allocate_task_team: T#%d exiting; task_team = %p "
                "unfinished_threads init'd to %d\n",
                (thread ? __kmp_gtid_from_thread(thread) : -1), task_team,
                KMP_ATOMIC_LD_RLX(&task_team->tt.tt_unfinished_threads)));
  return task_team;
}

// __kmp_free_task_team:
// Frees the task team associated with a specific thread, and adds it
// to the global task team free list.
void __kmp_free_task_team(kmp_info_t *thread, kmp_task_team_t *task_team) {
  KA_TRACE(20, ("__kmp_free_task_team: T#%d task_team = %p\n",
                thread ? __kmp_gtid_from_thread(thread) : -1, task_team));

  // Put task team back on free list
  __kmp_acquire_bootstrap_lock(&__kmp_task_team_lock);

  KMP_DEBUG_ASSERT(task_team->tt.tt_next == NULL);
  task_team->tt.tt_next = __kmp_free_task_teams;
  TCW_PTR(__kmp_free_task_teams, task_team);

  __kmp_release_bootstrap_lock(&__kmp_task_team_lock);
}

// __kmp_reap_task_teams:
// Free all the task teams on the task team free list.
// Should only be done during library shutdown.
// Cannot do anything that needs a thread structure or gtid since they are
// already gone.
void __kmp_reap_task_teams(void) {
  kmp_task_team_t *task_team;

  if (TCR_PTR(__kmp_free_task_teams) != NULL) {
    // Free all task_teams on the free list
    __kmp_acquire_bootstrap_lock(&__kmp_task_team_lock);
    while ((task_team = __kmp_free_task_teams) != NULL) {
      __kmp_free_task_teams = task_team->tt.tt_next;
      task_team->tt.tt_next = NULL;

      // Free threads_data if necessary
      if (task_team->tt.tt_threads_data != NULL) {
        __kmp_free_task_threads_data(task_team);
      }
      if (task_team->tt.tt_task_pri_list != NULL) {
        __kmp_free_task_pri_list(task_team);
      }
      __kmp_free(task_team);
    }
    __kmp_release_bootstrap_lock(&__kmp_task_team_lock);
  }
}

// __kmp_wait_to_unref_task_teams:
// Some threads could still be in the fork barrier release code, possibly
// trying to steal tasks.  Wait for each thread to unreference its task team.
void __kmp_wait_to_unref_task_teams(void) {
  kmp_info_t *thread;
  kmp_uint32 spins;
  kmp_uint64 time;
  int done;

  KMP_INIT_YIELD(spins);
  KMP_INIT_BACKOFF(time);

  for (;;) {
    done = TRUE;

    // TODO: GEH - this may be is wrong because some sync would be necessary
    // in case threads are added to the pool during the traversal. Need to
    // verify that lock for thread pool is held when calling this routine.
    for (thread = CCAST(kmp_info_t *, __kmp_thread_pool); thread != NULL;
         thread = thread->th.th_next_pool) {
#if KMP_OS_WINDOWS
      DWORD exit_val;
#endif
      if (TCR_PTR(thread->th.th_task_team) == NULL) {
        KA_TRACE(10, ("__kmp_wait_to_unref_task_team: T#%d task_team == NULL\n",
                      __kmp_gtid_from_thread(thread)));
        continue;
      }
#if KMP_OS_WINDOWS
      // TODO: GEH - add this check for Linux* OS / OS X* as well?
      if (!__kmp_is_thread_alive(thread, &exit_val)) {
        thread->th.th_task_team = NULL;
        continue;
      }
#endif

      done = FALSE; // Because th_task_team pointer is not NULL for this thread

      KA_TRACE(10, ("__kmp_wait_to_unref_task_team: Waiting for T#%d to "
                    "unreference task_team\n",
                    __kmp_gtid_from_thread(thread)));

      if (__kmp_dflt_blocktime != KMP_MAX_BLOCKTIME) {
        void *sleep_loc;
        // If the thread is sleeping, awaken it.
        if ((sleep_loc = TCR_PTR(CCAST(void *, thread->th.th_sleep_loc))) !=
            NULL) {
          KA_TRACE(
              10,
              ("__kmp_wait_to_unref_task_team: T#%d waking up thread T#%d\n",
               __kmp_gtid_from_thread(thread), __kmp_gtid_from_thread(thread)));
          __kmp_null_resume_wrapper(thread);
        }
      }
    }
    if (done) {
      break;
    }

    // If oversubscribed or have waited a bit, yield.
    KMP_YIELD_OVERSUB_ELSE_SPIN(spins, time);
  }
}

// __kmp_task_team_setup:  Create a task_team for the current team, but use
// an already created, unused one if it already exists.
void __kmp_task_team_setup(kmp_info_t *this_thr, kmp_team_t *team, int always) {
  KMP_DEBUG_ASSERT(__kmp_tasking_mode != tskm_immediate_exec);

  // If this task_team hasn't been created yet, allocate it. It will be used in
  // the region after the next.
  // If it exists, it is the current task team and shouldn't be touched yet as
  // it may still be in use.
  if (team->t.t_task_team[this_thr->th.th_task_state] == NULL &&
      (always || team->t.t_nproc > 1)) {
    team->t.t_task_team[this_thr->th.th_task_state] =
        __kmp_allocate_task_team(this_thr, team);
    KA_TRACE(20, ("__kmp_task_team_setup: Primary T#%d created new task_team %p"
                  " for team %d at parity=%d\n",
                  __kmp_gtid_from_thread(this_thr),
                  team->t.t_task_team[this_thr->th.th_task_state], team->t.t_id,
                  this_thr->th.th_task_state));
  }

  // After threads exit the release, they will call sync, and then point to this
  // other task_team; make sure it is allocated and properly initialized. As
  // threads spin in the barrier release phase, they will continue to use the
  // previous task_team struct(above), until they receive the signal to stop
  // checking for tasks (they can't safely reference the kmp_team_t struct,
  // which could be reallocated by the primary thread). No task teams are formed
  // for serialized teams.
  if (team->t.t_nproc > 1) {
    int other_team = 1 - this_thr->th.th_task_state;
    KMP_DEBUG_ASSERT(other_team >= 0 && other_team < 2);
    if (team->t.t_task_team[other_team] == NULL) { // setup other team as well
      team->t.t_task_team[other_team] =
          __kmp_allocate_task_team(this_thr, team);
      KA_TRACE(20, ("__kmp_task_team_setup: Primary T#%d created second new "
                    "task_team %p for team %d at parity=%d\n",
                    __kmp_gtid_from_thread(this_thr),
                    team->t.t_task_team[other_team], team->t.t_id, other_team));
    } else { // Leave the old task team struct in place for the upcoming region;
      // adjust as needed
      kmp_task_team_t *task_team = team->t.t_task_team[other_team];
      if (!task_team->tt.tt_active ||
          team->t.t_nproc != task_team->tt.tt_nproc) {
        TCW_4(task_team->tt.tt_nproc, team->t.t_nproc);
        TCW_4(task_team->tt.tt_found_tasks, FALSE);
        TCW_4(task_team->tt.tt_found_proxy_tasks, FALSE);
        TCW_4(task_team->tt.tt_hidden_helper_task_encountered, FALSE);
        KMP_ATOMIC_ST_REL(&task_team->tt.tt_unfinished_threads,
                          team->t.t_nproc);
        TCW_4(task_team->tt.tt_active, TRUE);
      }
      // if team size has changed, the first thread to enable tasking will
      // realloc threads_data if necessary
      KA_TRACE(20, ("__kmp_task_team_setup: Primary T#%d reset next task_team "
                    "%p for team %d at parity=%d\n",
                    __kmp_gtid_from_thread(this_thr),
                    team->t.t_task_team[other_team], team->t.t_id, other_team));
    }
  }

  // For regular thread, task enabling should be called when the task is going
  // to be pushed to a dequeue. However, for the hidden helper thread, we need
  // it ahead of time so that some operations can be performed without race
  // condition.
  if (this_thr == __kmp_hidden_helper_main_thread) {
    for (int i = 0; i < 2; ++i) {
      kmp_task_team_t *task_team = team->t.t_task_team[i];
      if (KMP_TASKING_ENABLED(task_team)) {
        continue;
      }
      __kmp_enable_tasking(task_team, this_thr);
      for (int j = 0; j < task_team->tt.tt_nproc; ++j) {
        kmp_thread_data_t *thread_data = &task_team->tt.tt_threads_data[j];
        if (thread_data->td.td_deque == NULL) {
          __kmp_alloc_task_deque(__kmp_hidden_helper_threads[j], thread_data);
        }
      }
    }
  }
}

// __kmp_task_team_sync: Propagation of task team data from team to threads
// which happens just after the release phase of a team barrier.  This may be
// called by any thread, but only for teams with # threads > 1.
void __kmp_task_team_sync(kmp_info_t *this_thr, kmp_team_t *team) {
  KMP_DEBUG_ASSERT(__kmp_tasking_mode != tskm_immediate_exec);

  // Toggle the th_task_state field, to switch which task_team this thread
  // refers to
  this_thr->th.th_task_state = (kmp_uint8)(1 - this_thr->th.th_task_state);

  // It is now safe to propagate the task team pointer from the team struct to
  // the current thread.
  TCW_PTR(this_thr->th.th_task_team,
          team->t.t_task_team[this_thr->th.th_task_state]);
  KA_TRACE(20,
           ("__kmp_task_team_sync: Thread T#%d task team switched to task_team "
            "%p from Team #%d (parity=%d)\n",
            __kmp_gtid_from_thread(this_thr), this_thr->th.th_task_team,
            team->t.t_id, this_thr->th.th_task_state));
}

// __kmp_task_team_wait: Primary thread waits for outstanding tasks after the
// barrier gather phase. Only called by primary thread if #threads in team > 1
// or if proxy tasks were created.
//
// wait is a flag that defaults to 1 (see kmp.h), but waiting can be turned off
// by passing in 0 optionally as the last argument. When wait is zero, primary
// thread does not wait for unfinished_threads to reach 0.
void __kmp_task_team_wait(
    kmp_info_t *this_thr,
    kmp_team_t *team USE_ITT_BUILD_ARG(void *itt_sync_obj), int wait) {
  kmp_task_team_t *task_team = team->t.t_task_team[this_thr->th.th_task_state];

  KMP_DEBUG_ASSERT(__kmp_tasking_mode != tskm_immediate_exec);
  KMP_DEBUG_ASSERT(task_team == this_thr->th.th_task_team);

  if ((task_team != NULL) && KMP_TASKING_ENABLED(task_team)) {
    if (wait) {
      KA_TRACE(20, ("__kmp_task_team_wait: Primary T#%d waiting for all tasks "
                    "(for unfinished_threads to reach 0) on task_team = %p\n",
                    __kmp_gtid_from_thread(this_thr), task_team));
      // Worker threads may have dropped through to release phase, but could
      // still be executing tasks. Wait here for tasks to complete. To avoid
      // memory contention, only primary thread checks termination condition.
      kmp_flag_32<false, false> flag(
          RCAST(std::atomic<kmp_uint32> *,
                &task_team->tt.tt_unfinished_threads),
          0U);
      flag.wait(this_thr, TRUE USE_ITT_BUILD_ARG(itt_sync_obj));
    }
    // Deactivate the old task team, so that the worker threads will stop
    // referencing it while spinning.
    KA_TRACE(
        20,
        ("__kmp_task_team_wait: Primary T#%d deactivating task_team %p: "
         "setting active to false, setting local and team's pointer to NULL\n",
         __kmp_gtid_from_thread(this_thr), task_team));
    KMP_DEBUG_ASSERT(task_team->tt.tt_nproc > 1 ||
                     task_team->tt.tt_found_proxy_tasks == TRUE ||
                     task_team->tt.tt_hidden_helper_task_encountered == TRUE);
    TCW_SYNC_4(task_team->tt.tt_found_proxy_tasks, FALSE);
    TCW_SYNC_4(task_team->tt.tt_hidden_helper_task_encountered, FALSE);
    KMP_CHECK_UPDATE(task_team->tt.tt_untied_task_encountered, 0);
    TCW_SYNC_4(task_team->tt.tt_active, FALSE);
    KMP_MB();

    TCW_PTR(this_thr->th.th_task_team, NULL);
  }
}

// __kmp_tasking_barrier:
// This routine is called only when __kmp_tasking_mode == tskm_extra_barrier.
// Internal function to execute all tasks prior to a regular barrier or a join
// barrier. It is a full barrier itself, which unfortunately turns regular
// barriers into double barriers and join barriers into 1 1/2 barriers.
void __kmp_tasking_barrier(kmp_team_t *team, kmp_info_t *thread, int gtid) {
  std::atomic<kmp_uint32> *spin = RCAST(
      std::atomic<kmp_uint32> *,
      &team->t.t_task_team[thread->th.th_task_state]->tt.tt_unfinished_threads);
  int flag = FALSE;
  KMP_DEBUG_ASSERT(__kmp_tasking_mode == tskm_extra_barrier);

#if USE_ITT_BUILD
  KMP_FSYNC_SPIN_INIT(spin, NULL);
#endif /* USE_ITT_BUILD */
  kmp_flag_32<false, false> spin_flag(spin, 0U);
  while (!spin_flag.execute_tasks(thread, gtid, TRUE,
                                  &flag USE_ITT_BUILD_ARG(NULL), 0)) {
#if USE_ITT_BUILD
    // TODO: What about itt_sync_obj??
    KMP_FSYNC_SPIN_PREPARE(RCAST(void *, spin));
#endif /* USE_ITT_BUILD */

    if (TCR_4(__kmp_global.g.g_done)) {
      if (__kmp_global.g.g_abort)
        __kmp_abort_thread();
      break;
    }
    KMP_YIELD(TRUE);
  }
#if USE_ITT_BUILD
  KMP_FSYNC_SPIN_ACQUIRED(RCAST(void *, spin));
#endif /* USE_ITT_BUILD */
}

// __kmp_give_task puts a task into a given thread queue if:
//  - the queue for that thread was created
//  - there's space in that queue
// Because of this, __kmp_push_task needs to check if there's space after
// getting the lock
static bool __kmp_give_task(kmp_info_t *thread, kmp_int32 tid, kmp_task_t *task,
                            kmp_int32 pass) {
  kmp_taskdata_t *taskdata = KMP_TASK_TO_TASKDATA(task);
  kmp_task_team_t *task_team = taskdata->td_task_team;

  KA_TRACE(20, ("__kmp_give_task: trying to give task %p to thread %d.\n",
                taskdata, tid));

  // If task_team is NULL something went really bad...
  KMP_DEBUG_ASSERT(task_team != NULL);

  bool result = false;
  kmp_thread_data_t *thread_data = &task_team->tt.tt_threads_data[tid];

  if (thread_data->td.td_deque == NULL) {
    // There's no queue in this thread, go find another one
    // We're guaranteed that at least one thread has a queue
    KA_TRACE(30,
             ("__kmp_give_task: thread %d has no queue while giving task %p.\n",
              tid, taskdata));
    return result;
  }

  if (TCR_4(thread_data->td.td_deque_ntasks) >=
      TASK_DEQUE_SIZE(thread_data->td)) {
    KA_TRACE(
        30,
        ("__kmp_give_task: queue is full while giving task %p to thread %d.\n",
         taskdata, tid));

    // if this deque is bigger than the pass ratio give a chance to another
    // thread
    if (TASK_DEQUE_SIZE(thread_data->td) / INITIAL_TASK_DEQUE_SIZE >= pass)
      return result;

    __kmp_acquire_bootstrap_lock(&thread_data->td.td_deque_lock);
    if (TCR_4(thread_data->td.td_deque_ntasks) >=
        TASK_DEQUE_SIZE(thread_data->td)) {
      // expand deque to push the task which is not allowed to execute
      __kmp_realloc_task_deque(thread, thread_data);
    }

  } else {

    __kmp_acquire_bootstrap_lock(&thread_data->td.td_deque_lock);

    if (TCR_4(thread_data->td.td_deque_ntasks) >=
        TASK_DEQUE_SIZE(thread_data->td)) {
      KA_TRACE(30, ("__kmp_give_task: queue is full while giving task %p to "
                    "thread %d.\n",
                    taskdata, tid));

      // if this deque is bigger than the pass ratio give a chance to another
      // thread
      if (TASK_DEQUE_SIZE(thread_data->td) / INITIAL_TASK_DEQUE_SIZE >= pass)
        goto release_and_exit;

      __kmp_realloc_task_deque(thread, thread_data);
    }
  }

  // lock is held here, and there is space in the deque

  thread_data->td.td_deque[thread_data->td.td_deque_tail] = taskdata;
  // Wrap index.
  thread_data->td.td_deque_tail =
      (thread_data->td.td_deque_tail + 1) & TASK_DEQUE_MASK(thread_data->td);
  TCW_4(thread_data->td.td_deque_ntasks,
        TCR_4(thread_data->td.td_deque_ntasks) + 1);

  result = true;
  KA_TRACE(30, ("__kmp_give_task: successfully gave task %p to thread %d.\n",
                taskdata, tid));

release_and_exit:
  __kmp_release_bootstrap_lock(&thread_data->td.td_deque_lock);

  return result;
}

#define PROXY_TASK_FLAG 0x40000000
/* The finish of the proxy tasks is divided in two pieces:
    - the top half is the one that can be done from a thread outside the team
    - the bottom half must be run from a thread within the team

   In order to run the bottom half the task gets queued back into one of the
   threads of the team. Once the td_incomplete_child_task counter of the parent
   is decremented the threads can leave the barriers. So, the bottom half needs
   to be queued before the counter is decremented. The top half is therefore
   divided in two parts:
    - things that can be run before queuing the bottom half
    - things that must be run after queuing the bottom half

   This creates a second race as the bottom half can free the task before the
   second top half is executed. To avoid this we use the
   td_incomplete_child_task of the proxy task to synchronize the top and bottom
   half. */
static void __kmp_first_top_half_finish_proxy(kmp_taskdata_t *taskdata) {
  KMP_DEBUG_ASSERT(taskdata->td_flags.tasktype == TASK_EXPLICIT);
  KMP_DEBUG_ASSERT(taskdata->td_flags.proxy == TASK_PROXY);
  KMP_DEBUG_ASSERT(taskdata->td_flags.complete == 0);
  KMP_DEBUG_ASSERT(taskdata->td_flags.freed == 0);

  taskdata->td_flags.complete = 1; // mark the task as completed

  if (taskdata->td_taskgroup)
    KMP_ATOMIC_DEC(&taskdata->td_taskgroup->count);

  // Create an imaginary children for this task so the bottom half cannot
  // release the task before we have completed the second top half
  KMP_ATOMIC_OR(&taskdata->td_incomplete_child_tasks, PROXY_TASK_FLAG);
}

static void __kmp_second_top_half_finish_proxy(kmp_taskdata_t *taskdata) {
#if KMP_DEBUG
  kmp_int32 children = 0;
  // Predecrement simulated by "- 1" calculation
  children = -1 +
#endif
      KMP_ATOMIC_DEC(&taskdata->td_parent->td_incomplete_child_tasks);
  KMP_DEBUG_ASSERT(children >= 0);

  // Remove the imaginary children
  KMP_ATOMIC_AND(&taskdata->td_incomplete_child_tasks, ~PROXY_TASK_FLAG);
}

static void __kmp_bottom_half_finish_proxy(kmp_int32 gtid, kmp_task_t *ptask) {
  kmp_taskdata_t *taskdata = KMP_TASK_TO_TASKDATA(ptask);
  kmp_info_t *thread = __kmp_threads[gtid];

  KMP_DEBUG_ASSERT(taskdata->td_flags.proxy == TASK_PROXY);
  KMP_DEBUG_ASSERT(taskdata->td_flags.complete ==
                   1); // top half must run before bottom half

  // We need to wait to make sure the top half is finished
  // Spinning here should be ok as this should happen quickly
  while ((KMP_ATOMIC_LD_ACQ(&taskdata->td_incomplete_child_tasks) &
          PROXY_TASK_FLAG) > 0)
    ;

  __kmp_release_deps(gtid, taskdata);
  __kmp_free_task_and_ancestors(gtid, taskdata, thread);
}

/*!
@ingroup TASKING
@param gtid Global Thread ID of encountering thread
@param ptask Task which execution is completed

Execute the completion of a proxy task from a thread of that is part of the
team. Run first and bottom halves directly.
*/
void __kmpc_proxy_task_completed(kmp_int32 gtid, kmp_task_t *ptask) {
  KMP_DEBUG_ASSERT(ptask != NULL);
  kmp_taskdata_t *taskdata = KMP_TASK_TO_TASKDATA(ptask);
  KA_TRACE(
      10, ("__kmp_proxy_task_completed(enter): T#%d proxy task %p completing\n",
           gtid, taskdata));
  __kmp_assert_valid_gtid(gtid);
  KMP_DEBUG_ASSERT(taskdata->td_flags.proxy == TASK_PROXY);

  __kmp_first_top_half_finish_proxy(taskdata);
  __kmp_second_top_half_finish_proxy(taskdata);
  __kmp_bottom_half_finish_proxy(gtid, ptask);

  KA_TRACE(10,
           ("__kmp_proxy_task_completed(exit): T#%d proxy task %p completing\n",
            gtid, taskdata));
}

void __kmpc_give_task(kmp_task_t *ptask, kmp_int32 start = 0) {
  KMP_DEBUG_ASSERT(ptask != NULL);
  kmp_taskdata_t *taskdata = KMP_TASK_TO_TASKDATA(ptask);

  // Enqueue task to complete bottom half completion from a thread within the
  // corresponding team
  kmp_team_t *team = taskdata->td_team;
  kmp_int32 nthreads = team->t.t_nproc;
  kmp_info_t *thread;

  // This should be similar to start_k = __kmp_get_random( thread ) % nthreads
  // but we cannot use __kmp_get_random here
  kmp_int32 start_k = start % nthreads;
  kmp_int32 pass = 1;
  kmp_int32 k = start_k;

  do {
    // For now we're just linearly trying to find a thread
    thread = team->t.t_threads[k];
    k = (k + 1) % nthreads;

    // we did a full pass through all the threads
    if (k == start_k)
      pass = pass << 1;

  } while (!__kmp_give_task(thread, k, ptask, pass));

  if (__kmp_dflt_blocktime != KMP_MAX_BLOCKTIME && __kmp_wpolicy_passive) {
    // awake at least one thread to execute given task
    for (int i = 0; i < nthreads; ++i) {
      thread = team->t.t_threads[i];
      if (thread->th.th_sleep_loc != NULL) {
        __kmp_null_resume_wrapper(thread);
        break;
      }
    }
  }
}

/*!
@ingroup TASKING
@param ptask Task which execution is completed

Execute the completion of a proxy task from a thread that could not belong to
the team.
*/
void __kmpc_proxy_task_completed_ooo(kmp_task_t *ptask) {
  KMP_DEBUG_ASSERT(ptask != NULL);
  kmp_taskdata_t *taskdata = KMP_TASK_TO_TASKDATA(ptask);

  KA_TRACE(
      10,
      ("__kmp_proxy_task_completed_ooo(enter): proxy task completing ooo %p\n",
       taskdata));

  KMP_DEBUG_ASSERT(taskdata->td_flags.proxy == TASK_PROXY);

  __kmp_first_top_half_finish_proxy(taskdata);

  __kmpc_give_task(ptask);

  __kmp_second_top_half_finish_proxy(taskdata);

  KA_TRACE(
      10,
      ("__kmp_proxy_task_completed_ooo(exit): proxy task completing ooo %p\n",
       taskdata));
}

kmp_event_t *__kmpc_task_allow_completion_event(ident_t *loc_ref, int gtid,
                                                kmp_task_t *task) {
  kmp_taskdata_t *td = KMP_TASK_TO_TASKDATA(task);
  if (td->td_allow_completion_event.type == KMP_EVENT_UNINITIALIZED) {
    td->td_allow_completion_event.type = KMP_EVENT_ALLOW_COMPLETION;
    td->td_allow_completion_event.ed.task = task;
    __kmp_init_tas_lock(&td->td_allow_completion_event.lock);
  }
  return &td->td_allow_completion_event;
}

void __kmp_fulfill_event(kmp_event_t *event) {
  if (event->type == KMP_EVENT_ALLOW_COMPLETION) {
    kmp_task_t *ptask = event->ed.task;
    kmp_taskdata_t *taskdata = KMP_TASK_TO_TASKDATA(ptask);
    bool detached = false;
    int gtid = __kmp_get_gtid();

    // The associated task might have completed or could be completing at this
    // point.
    // We need to take the lock to avoid races
    __kmp_acquire_tas_lock(&event->lock, gtid);
    if (taskdata->td_flags.proxy == TASK_PROXY) {
      detached = true;
    } else {
#if OMPT_SUPPORT
      // The OMPT event must occur under mutual exclusion,
      // otherwise the tool might access ptask after free
      if (UNLIKELY(ompt_enabled.enabled))
        __ompt_task_finish(ptask, NULL, ompt_task_early_fulfill);
#endif
    }
    event->type = KMP_EVENT_UNINITIALIZED;
    __kmp_release_tas_lock(&event->lock, gtid);

    if (detached) {
#if OMPT_SUPPORT
      // We free ptask afterwards and know the task is finished,
      // so locking is not necessary
      if (UNLIKELY(ompt_enabled.enabled))
        __ompt_task_finish(ptask, NULL, ompt_task_late_fulfill);
#endif
      // If the task detached complete the proxy task
      if (gtid >= 0) {
        kmp_team_t *team = taskdata->td_team;
        kmp_info_t *thread = __kmp_get_thread();
        if (thread->th.th_team == team) {
          __kmpc_proxy_task_completed(gtid, ptask);
          return;
        }
      }

      // fallback
      __kmpc_proxy_task_completed_ooo(ptask);
    }
  }
}

// __kmp_task_dup_alloc: Allocate the taskdata and make a copy of source task
// for taskloop
//
// thread:   allocating thread
// task_src: pointer to source task to be duplicated
// returns:  a pointer to the allocated kmp_task_t structure (task).
kmp_task_t *__kmp_task_dup_alloc(kmp_info_t *thread, kmp_task_t *task_src) {
  kmp_task_t *task;
  kmp_taskdata_t *taskdata;
  kmp_taskdata_t *taskdata_src = KMP_TASK_TO_TASKDATA(task_src);
  kmp_taskdata_t *parent_task = taskdata_src->td_parent; // same parent task
  size_t shareds_offset;
  size_t task_size;

  KA_TRACE(10, ("__kmp_task_dup_alloc(enter): Th %p, source task %p\n", thread,
                task_src));
  KMP_DEBUG_ASSERT(taskdata_src->td_flags.proxy ==
                   TASK_FULL); // it should not be proxy task
  KMP_DEBUG_ASSERT(taskdata_src->td_flags.tasktype == TASK_EXPLICIT);
  task_size = taskdata_src->td_size_alloc;

  // Allocate a kmp_taskdata_t block and a kmp_task_t block.
  KA_TRACE(30, ("__kmp_task_dup_alloc: Th %p, malloc size %ld\n", thread,
                task_size));
#if USE_FAST_MEMORY
  taskdata = (kmp_taskdata_t *)__kmp_fast_allocate(thread, task_size);
#else
  taskdata = (kmp_taskdata_t *)__kmp_thread_malloc(thread, task_size);
#endif /* USE_FAST_MEMORY */
  KMP_MEMCPY(taskdata, taskdata_src, task_size);

  task = KMP_TASKDATA_TO_TASK(taskdata);

  // Initialize new task (only specific fields not affected by memcpy)
  taskdata->td_task_id = KMP_GEN_TASK_ID();
  if (task->shareds != NULL) { // need setup shareds pointer
    shareds_offset = (char *)task_src->shareds - (char *)taskdata_src;
    task->shareds = &((char *)taskdata)[shareds_offset];
    KMP_DEBUG_ASSERT((((kmp_uintptr_t)task->shareds) & (sizeof(void *) - 1)) ==
                     0);
  }
  taskdata->td_alloc_thread = thread;
  taskdata->td_parent = parent_task;
  // task inherits the taskgroup from the parent task
  taskdata->td_taskgroup = parent_task->td_taskgroup;
  // tied task needs to initialize the td_last_tied at creation,
  // untied one does this when it is scheduled for execution
  if (taskdata->td_flags.tiedness == TASK_TIED)
    taskdata->td_last_tied = taskdata;

  // Only need to keep track of child task counts if team parallel and tasking
  // not serialized
  if (!(taskdata->td_flags.team_serial || taskdata->td_flags.tasking_ser)) {
    KMP_ATOMIC_INC(&parent_task->td_incomplete_child_tasks);
    if (parent_task->td_taskgroup)
      KMP_ATOMIC_INC(&parent_task->td_taskgroup->count);
    // Only need to keep track of allocated child tasks for explicit tasks since
    // implicit not deallocated
    if (taskdata->td_parent->td_flags.tasktype == TASK_EXPLICIT)
      KMP_ATOMIC_INC(&taskdata->td_parent->td_allocated_child_tasks);
  }

  KA_TRACE(20,
           ("__kmp_task_dup_alloc(exit): Th %p, created task %p, parent=%p\n",
            thread, taskdata, taskdata->td_parent));
#if OMPT_SUPPORT
  if (UNLIKELY(ompt_enabled.enabled))
    __ompt_task_init(taskdata, thread->th.th_info.ds.ds_gtid);
#endif
  return task;
}

// Routine optionally generated by the compiler for setting the lastprivate flag
// and calling needed constructors for private/firstprivate objects
// (used to form taskloop tasks from pattern task)
// Parameters: dest task, src task, lastprivate flag.
typedef void (*p_task_dup_t)(kmp_task_t *, kmp_task_t *, kmp_int32);

KMP_BUILD_ASSERT(sizeof(long) == 4 || sizeof(long) == 8);

// class to encapsulate manipulating loop bounds in a taskloop task.
// this abstracts away the Intel vs GOMP taskloop interface for setting/getting
// the loop bound variables.
class kmp_taskloop_bounds_t {
  kmp_task_t *task;
  const kmp_taskdata_t *taskdata;
  size_t lower_offset;
  size_t upper_offset;

public:
  kmp_taskloop_bounds_t(kmp_task_t *_task, kmp_uint64 *lb, kmp_uint64 *ub)
      : task(_task), taskdata(KMP_TASK_TO_TASKDATA(task)),
        lower_offset((char *)lb - (char *)task),
        upper_offset((char *)ub - (char *)task) {
    KMP_DEBUG_ASSERT((char *)lb > (char *)_task);
    KMP_DEBUG_ASSERT((char *)ub > (char *)_task);
  }
  kmp_taskloop_bounds_t(kmp_task_t *_task, const kmp_taskloop_bounds_t &bounds)
      : task(_task), taskdata(KMP_TASK_TO_TASKDATA(_task)),
        lower_offset(bounds.lower_offset), upper_offset(bounds.upper_offset) {}
  size_t get_lower_offset() const { return lower_offset; }
  size_t get_upper_offset() const { return upper_offset; }
  kmp_uint64 get_lb() const {
    kmp_int64 retval;
#if defined(KMP_GOMP_COMPAT)
    // Intel task just returns the lower bound normally
    if (!taskdata->td_flags.native) {
      retval = *(kmp_int64 *)((char *)task + lower_offset);
    } else {
      // GOMP task has to take into account the sizeof(long)
      if (taskdata->td_size_loop_bounds == 4) {
        kmp_int32 *lb = RCAST(kmp_int32 *, task->shareds);
        retval = (kmp_int64)*lb;
      } else {
        kmp_int64 *lb = RCAST(kmp_int64 *, task->shareds);
        retval = (kmp_int64)*lb;
      }
    }
#else
    (void)taskdata;
    retval = *(kmp_int64 *)((char *)task + lower_offset);
#endif // defined(KMP_GOMP_COMPAT)
    return retval;
  }
  kmp_uint64 get_ub() const {
    kmp_int64 retval;
#if defined(KMP_GOMP_COMPAT)
    // Intel task just returns the upper bound normally
    if (!taskdata->td_flags.native) {
      retval = *(kmp_int64 *)((char *)task + upper_offset);
    } else {
      // GOMP task has to take into account the sizeof(long)
      if (taskdata->td_size_loop_bounds == 4) {
        kmp_int32 *ub = RCAST(kmp_int32 *, task->shareds) + 1;
        retval = (kmp_int64)*ub;
      } else {
        kmp_int64 *ub = RCAST(kmp_int64 *, task->shareds) + 1;
        retval = (kmp_int64)*ub;
      }
    }
#else
    retval = *(kmp_int64 *)((char *)task + upper_offset);
#endif // defined(KMP_GOMP_COMPAT)
    return retval;
  }
  void set_lb(kmp_uint64 lb) {
#if defined(KMP_GOMP_COMPAT)
    // Intel task just sets the lower bound normally
    if (!taskdata->td_flags.native) {
      *(kmp_uint64 *)((char *)task + lower_offset) = lb;
    } else {
      // GOMP task has to take into account the sizeof(long)
      if (taskdata->td_size_loop_bounds == 4) {
        kmp_uint32 *lower = RCAST(kmp_uint32 *, task->shareds);
        *lower = (kmp_uint32)lb;
      } else {
        kmp_uint64 *lower = RCAST(kmp_uint64 *, task->shareds);
        *lower = (kmp_uint64)lb;
      }
    }
#else
    *(kmp_uint64 *)((char *)task + lower_offset) = lb;
#endif // defined(KMP_GOMP_COMPAT)
  }
  void set_ub(kmp_uint64 ub) {
#if defined(KMP_GOMP_COMPAT)
    // Intel task just sets the upper bound normally
    if (!taskdata->td_flags.native) {
      *(kmp_uint64 *)((char *)task + upper_offset) = ub;
    } else {
      // GOMP task has to take into account the sizeof(long)
      if (taskdata->td_size_loop_bounds == 4) {
        kmp_uint32 *upper = RCAST(kmp_uint32 *, task->shareds) + 1;
        *upper = (kmp_uint32)ub;
      } else {
        kmp_uint64 *upper = RCAST(kmp_uint64 *, task->shareds) + 1;
        *upper = (kmp_uint64)ub;
      }
    }
#else
    *(kmp_uint64 *)((char *)task + upper_offset) = ub;
#endif // defined(KMP_GOMP_COMPAT)
  }
};

// __kmp_taskloop_linear: Start tasks of the taskloop linearly
//
// loc        Source location information
// gtid       Global thread ID
// task       Pattern task, exposes the loop iteration range
// lb         Pointer to loop lower bound in task structure
// ub         Pointer to loop upper bound in task structure
// st         Loop stride
// ub_glob    Global upper bound (used for lastprivate check)
// num_tasks  Number of tasks to execute
// grainsize  Number of loop iterations per task
// extras     Number of chunks with grainsize+1 iterations
// last_chunk Reduction of grainsize for last task
// tc         Iterations count
// task_dup   Tasks duplication routine
// codeptr_ra Return address for OMPT events
void __kmp_taskloop_linear(ident_t *loc, int gtid, kmp_task_t *task,
                           kmp_uint64 *lb, kmp_uint64 *ub, kmp_int64 st,
                           kmp_uint64 ub_glob, kmp_uint64 num_tasks,
                           kmp_uint64 grainsize, kmp_uint64 extras,
                           kmp_int64 last_chunk, kmp_uint64 tc,
#if OMPT_SUPPORT
                           void *codeptr_ra,
#endif
                           void *task_dup) {
  KMP_COUNT_BLOCK(OMP_TASKLOOP);
  KMP_TIME_PARTITIONED_BLOCK(OMP_taskloop_scheduling);
  p_task_dup_t ptask_dup = (p_task_dup_t)task_dup;
  // compiler provides global bounds here
  kmp_taskloop_bounds_t task_bounds(task, lb, ub);
  kmp_uint64 lower = task_bounds.get_lb();
  kmp_uint64 upper = task_bounds.get_ub();
  kmp_uint64 i;
  kmp_info_t *thread = __kmp_threads[gtid];
  kmp_taskdata_t *current_task = thread->th.th_current_task;
  kmp_task_t *next_task;
  kmp_int32 lastpriv = 0;

  KMP_DEBUG_ASSERT(tc == num_tasks * grainsize +
                             (last_chunk < 0 ? last_chunk : extras));
  KMP_DEBUG_ASSERT(num_tasks > extras);
  KMP_DEBUG_ASSERT(num_tasks > 0);
  KA_TRACE(20, ("__kmp_taskloop_linear: T#%d: %lld tasks, grainsize %lld, "
                "extras %lld, last_chunk %lld, i=%lld,%lld(%d)%lld, dup %p\n",
                gtid, num_tasks, grainsize, extras, last_chunk, lower, upper,
                ub_glob, st, task_dup));

  // Launch num_tasks tasks, assign grainsize iterations each task
  for (i = 0; i < num_tasks; ++i) {
    kmp_uint64 chunk_minus_1;
    if (extras == 0) {
      chunk_minus_1 = grainsize - 1;
    } else {
      chunk_minus_1 = grainsize;
      --extras; // first extras iterations get bigger chunk (grainsize+1)
    }
    upper = lower + st * chunk_minus_1;
    if (upper > *ub) {
      upper = *ub;
    }
    if (i == num_tasks - 1) {
      // schedule the last task, set lastprivate flag if needed
      if (st == 1) { // most common case
        KMP_DEBUG_ASSERT(upper == *ub);
        if (upper == ub_glob)
          lastpriv = 1;
      } else if (st > 0) { // positive loop stride
        KMP_DEBUG_ASSERT((kmp_uint64)st > *ub - upper);
        if ((kmp_uint64)st > ub_glob - upper)
          lastpriv = 1;
      } else { // negative loop stride
        KMP_DEBUG_ASSERT(upper + st < *ub);
        if (upper - ub_glob < (kmp_uint64)(-st))
          lastpriv = 1;
      }
    }
    next_task = __kmp_task_dup_alloc(thread, task); // allocate new task
    kmp_taskdata_t *next_taskdata = KMP_TASK_TO_TASKDATA(next_task);
    kmp_taskloop_bounds_t next_task_bounds =
        kmp_taskloop_bounds_t(next_task, task_bounds);

    // adjust task-specific bounds
    next_task_bounds.set_lb(lower);
    if (next_taskdata->td_flags.native) {
      next_task_bounds.set_ub(upper + (st > 0 ? 1 : -1));
    } else {
      next_task_bounds.set_ub(upper);
    }
    if (ptask_dup != NULL) // set lastprivate flag, construct firstprivates,
                           // etc.
      ptask_dup(next_task, task, lastpriv);
    KA_TRACE(40,
             ("__kmp_taskloop_linear: T#%d; task #%llu: task %p: lower %lld, "
              "upper %lld stride %lld, (offsets %p %p)\n",
              gtid, i, next_task, lower, upper, st,
              next_task_bounds.get_lower_offset(),
              next_task_bounds.get_upper_offset()));
#if OMPT_SUPPORT
    __kmp_omp_taskloop_task(NULL, gtid, next_task,
                            codeptr_ra); // schedule new task
#if OMPT_OPTIONAL
    if (ompt_enabled.ompt_callback_dispatch) {
      OMPT_GET_DISPATCH_CHUNK(next_taskdata->ompt_task_info.dispatch_chunk,
                              lower, upper, st);
    }
#endif // OMPT_OPTIONAL
#else
    __kmp_omp_task(gtid, next_task, true); // schedule new task
#endif
    lower = upper + st; // adjust lower bound for the next iteration
  }
  // free the pattern task and exit
  __kmp_task_start(gtid, task, current_task); // make internal bookkeeping
  // do not execute the pattern task, just do internal bookkeeping
  __kmp_task_finish<false>(gtid, task, current_task);
}

// Structure to keep taskloop parameters for auxiliary task
// kept in the shareds of the task structure.
typedef struct __taskloop_params {
  kmp_task_t *task;
  kmp_uint64 *lb;
  kmp_uint64 *ub;
  void *task_dup;
  kmp_int64 st;
  kmp_uint64 ub_glob;
  kmp_uint64 num_tasks;
  kmp_uint64 grainsize;
  kmp_uint64 extras;
  kmp_int64 last_chunk;
  kmp_uint64 tc;
  kmp_uint64 num_t_min;
#if OMPT_SUPPORT
  void *codeptr_ra;
#endif
} __taskloop_params_t;

void __kmp_taskloop_recur(ident_t *, int, kmp_task_t *, kmp_uint64 *,
                          kmp_uint64 *, kmp_int64, kmp_uint64, kmp_uint64,
                          kmp_uint64, kmp_uint64, kmp_int64, kmp_uint64,
                          kmp_uint64,
#if OMPT_SUPPORT
                          void *,
#endif
                          void *);

// Execute part of the taskloop submitted as a task.
int __kmp_taskloop_task(int gtid, void *ptask) {
  __taskloop_params_t *p =
      (__taskloop_params_t *)((kmp_task_t *)ptask)->shareds;
  kmp_task_t *task = p->task;
  kmp_uint64 *lb = p->lb;
  kmp_uint64 *ub = p->ub;
  void *task_dup = p->task_dup;
  //  p_task_dup_t ptask_dup = (p_task_dup_t)task_dup;
  kmp_int64 st = p->st;
  kmp_uint64 ub_glob = p->ub_glob;
  kmp_uint64 num_tasks = p->num_tasks;
  kmp_uint64 grainsize = p->grainsize;
  kmp_uint64 extras = p->extras;
  kmp_int64 last_chunk = p->last_chunk;
  kmp_uint64 tc = p->tc;
  kmp_uint64 num_t_min = p->num_t_min;
#if OMPT_SUPPORT
  void *codeptr_ra = p->codeptr_ra;
#endif
#if KMP_DEBUG
  kmp_taskdata_t *taskdata = KMP_TASK_TO_TASKDATA(task);
  KMP_DEBUG_ASSERT(task != NULL);
  KA_TRACE(20,
           ("__kmp_taskloop_task: T#%d, task %p: %lld tasks, grainsize"
            " %lld, extras %lld, last_chunk %lld, i=%lld,%lld(%d), dup %p\n",
            gtid, taskdata, num_tasks, grainsize, extras, last_chunk, *lb, *ub,
            st, task_dup));
#endif
  KMP_DEBUG_ASSERT(num_tasks * 2 + 1 > num_t_min);
  if (num_tasks > num_t_min)
    __kmp_taskloop_recur(NULL, gtid, task, lb, ub, st, ub_glob, num_tasks,
                         grainsize, extras, last_chunk, tc, num_t_min,
#if OMPT_SUPPORT
                         codeptr_ra,
#endif
                         task_dup);
  else
    __kmp_taskloop_linear(NULL, gtid, task, lb, ub, st, ub_glob, num_tasks,
                          grainsize, extras, last_chunk, tc,
#if OMPT_SUPPORT
                          codeptr_ra,
#endif
                          task_dup);

  KA_TRACE(40, ("__kmp_taskloop_task(exit): T#%d\n", gtid));
  return 0;
}

// Schedule part of the taskloop as a task,
// execute the rest of the taskloop.
//
// loc        Source location information
// gtid       Global thread ID
// task       Pattern task, exposes the loop iteration range
// lb         Pointer to loop lower bound in task structure
// ub         Pointer to loop upper bound in task structure
// st         Loop stride
// ub_glob    Global upper bound (used for lastprivate check)
// num_tasks  Number of tasks to execute
// grainsize  Number of loop iterations per task
// extras     Number of chunks with grainsize+1 iterations
// last_chunk Reduction of grainsize for last task
// tc         Iterations count
// num_t_min  Threshold to launch tasks recursively
// task_dup   Tasks duplication routine
// codeptr_ra Return address for OMPT events
void __kmp_taskloop_recur(ident_t *loc, int gtid, kmp_task_t *task,
                          kmp_uint64 *lb, kmp_uint64 *ub, kmp_int64 st,
                          kmp_uint64 ub_glob, kmp_uint64 num_tasks,
                          kmp_uint64 grainsize, kmp_uint64 extras,
                          kmp_int64 last_chunk, kmp_uint64 tc,
                          kmp_uint64 num_t_min,
#if OMPT_SUPPORT
                          void *codeptr_ra,
#endif
                          void *task_dup) {
  kmp_taskdata_t *taskdata = KMP_TASK_TO_TASKDATA(task);
  KMP_DEBUG_ASSERT(task != NULL);
  KMP_DEBUG_ASSERT(num_tasks > num_t_min);
  KA_TRACE(20,
           ("__kmp_taskloop_recur: T#%d, task %p: %lld tasks, grainsize"
            " %lld, extras %lld, last_chunk %lld, i=%lld,%lld(%d), dup %p\n",
            gtid, taskdata, num_tasks, grainsize, extras, last_chunk, *lb, *ub,
            st, task_dup));
  p_task_dup_t ptask_dup = (p_task_dup_t)task_dup;
  kmp_uint64 lower = *lb;
  kmp_info_t *thread = __kmp_threads[gtid];
  //  kmp_taskdata_t *current_task = thread->th.th_current_task;
  kmp_task_t *next_task;
  size_t lower_offset =
      (char *)lb - (char *)task; // remember offset of lb in the task structure
  size_t upper_offset =
      (char *)ub - (char *)task; // remember offset of ub in the task structure

  KMP_DEBUG_ASSERT(tc == num_tasks * grainsize +
                             (last_chunk < 0 ? last_chunk : extras));
  KMP_DEBUG_ASSERT(num_tasks > extras);
  KMP_DEBUG_ASSERT(num_tasks > 0);

  // split the loop in two halves
  kmp_uint64 lb1, ub0, tc0, tc1, ext0, ext1;
  kmp_int64 last_chunk0 = 0, last_chunk1 = 0;
  kmp_uint64 gr_size0 = grainsize;
  kmp_uint64 n_tsk0 = num_tasks >> 1; // num_tasks/2 to execute
  kmp_uint64 n_tsk1 = num_tasks - n_tsk0; // to schedule as a task
  if (last_chunk < 0) {
    ext0 = ext1 = 0;
    last_chunk1 = last_chunk;
    tc0 = grainsize * n_tsk0;
    tc1 = tc - tc0;
  } else if (n_tsk0 <= extras) {
    gr_size0++; // integrate extras into grainsize
    ext0 = 0; // no extra iters in 1st half
    ext1 = extras - n_tsk0; // remaining extras
    tc0 = gr_size0 * n_tsk0;
    tc1 = tc - tc0;
  } else { // n_tsk0 > extras
    ext1 = 0; // no extra iters in 2nd half
    ext0 = extras;
    tc1 = grainsize * n_tsk1;
    tc0 = tc - tc1;
  }
  ub0 = lower + st * (tc0 - 1);
  lb1 = ub0 + st;

  // create pattern task for 2nd half of the loop
  next_task = __kmp_task_dup_alloc(thread, task); // duplicate the task
  // adjust lower bound (upper bound is not changed) for the 2nd half
  *(kmp_uint64 *)((char *)next_task + lower_offset) = lb1;
  if (ptask_dup != NULL) // construct firstprivates, etc.
    ptask_dup(next_task, task, 0);
  *ub = ub0; // adjust upper bound for the 1st half

  // create auxiliary task for 2nd half of the loop
  // make sure new task has same parent task as the pattern task
  kmp_taskdata_t *current_task = thread->th.th_current_task;
  thread->th.th_current_task = taskdata->td_parent;
  kmp_task_t *new_task =
      __kmpc_omp_task_alloc(loc, gtid, 1, 3 * sizeof(void *),
                            sizeof(__taskloop_params_t), &__kmp_taskloop_task);
  // restore current task
  thread->th.th_current_task = current_task;
  __taskloop_params_t *p = (__taskloop_params_t *)new_task->shareds;
  p->task = next_task;
  p->lb = (kmp_uint64 *)((char *)next_task + lower_offset);
  p->ub = (kmp_uint64 *)((char *)next_task + upper_offset);
  p->task_dup = task_dup;
  p->st = st;
  p->ub_glob = ub_glob;
  p->num_tasks = n_tsk1;
  p->grainsize = grainsize;
  p->extras = ext1;
  p->last_chunk = last_chunk1;
  p->tc = tc1;
  p->num_t_min = num_t_min;
#if OMPT_SUPPORT
  p->codeptr_ra = codeptr_ra;
#endif

#if OMPT_SUPPORT
  // schedule new task with correct return address for OMPT events
  __kmp_omp_taskloop_task(NULL, gtid, new_task, codeptr_ra);
#else
  __kmp_omp_task(gtid, new_task, true); // schedule new task
#endif

  // execute the 1st half of current subrange
  if (n_tsk0 > num_t_min)
    __kmp_taskloop_recur(loc, gtid, task, lb, ub, st, ub_glob, n_tsk0, gr_size0,
                         ext0, last_chunk0, tc0, num_t_min,
#if OMPT_SUPPORT
                         codeptr_ra,
#endif
                         task_dup);
  else
    __kmp_taskloop_linear(loc, gtid, task, lb, ub, st, ub_glob, n_tsk0,
                          gr_size0, ext0, last_chunk0, tc0,
#if OMPT_SUPPORT
                          codeptr_ra,
#endif
                          task_dup);

  KA_TRACE(40, ("__kmp_taskloop_recur(exit): T#%d\n", gtid));
}

static void __kmp_taskloop(ident_t *loc, int gtid, kmp_task_t *task, int if_val,
                           kmp_uint64 *lb, kmp_uint64 *ub, kmp_int64 st,
                           int nogroup, int sched, kmp_uint64 grainsize,
                           int modifier, void *task_dup) {
  kmp_taskdata_t *taskdata = KMP_TASK_TO_TASKDATA(task);
  KMP_DEBUG_ASSERT(task != NULL);
  if (nogroup == 0) {
#if OMPT_SUPPORT && OMPT_OPTIONAL
    OMPT_STORE_RETURN_ADDRESS(gtid);
#endif
    __kmpc_taskgroup(loc, gtid);
  }

  // =========================================================================
  // calculate loop parameters
  kmp_taskloop_bounds_t task_bounds(task, lb, ub);
  kmp_uint64 tc;
  // compiler provides global bounds here
  kmp_uint64 lower = task_bounds.get_lb();
  kmp_uint64 upper = task_bounds.get_ub();
  kmp_uint64 ub_glob = upper; // global upper used to calc lastprivate flag
  kmp_uint64 num_tasks = 0, extras = 0;
  kmp_int64 last_chunk =
      0; // reduce grainsize of last task by last_chunk in strict mode
  kmp_uint64 num_tasks_min = __kmp_taskloop_min_tasks;
  kmp_info_t *thread = __kmp_threads[gtid];
  kmp_taskdata_t *current_task = thread->th.th_current_task;

  KA_TRACE(20, ("__kmp_taskloop: T#%d, task %p, lb %lld, ub %lld, st %lld, "
                "grain %llu(%d, %d), dup %p\n",
                gtid, taskdata, lower, upper, st, grainsize, sched, modifier,
                task_dup));

  // compute trip count
  if (st == 1) { // most common case
    tc = upper - lower + 1;
  } else if (st < 0) {
    tc = (lower - upper) / (-st) + 1;
  } else { // st > 0
    tc = (upper - lower) / st + 1;
  }
  if (tc == 0) {
    KA_TRACE(20, ("__kmp_taskloop(exit): T#%d zero-trip loop\n", gtid));
    // free the pattern task and exit
    __kmp_task_start(gtid, task, current_task);
    // do not execute anything for zero-trip loop
    __kmp_task_finish<false>(gtid, task, current_task);
    return;
  }

#if OMPT_SUPPORT && OMPT_OPTIONAL
  ompt_team_info_t *team_info = __ompt_get_teaminfo(0, NULL);
  ompt_task_info_t *task_info = __ompt_get_task_info_object(0);
  if (ompt_enabled.ompt_callback_work) {
    ompt_callbacks.ompt_callback(ompt_callback_work)(
        ompt_work_taskloop, ompt_scope_begin, &(team_info->parallel_data),
        &(task_info->task_data), tc, OMPT_GET_RETURN_ADDRESS(0));
  }
#endif

  if (num_tasks_min == 0)
    // TODO: can we choose better default heuristic?
    num_tasks_min =
        KMP_MIN(thread->th.th_team_nproc * 10, INITIAL_TASK_DEQUE_SIZE);

  // compute num_tasks/grainsize based on the input provided
  switch (sched) {
  case 0: // no schedule clause specified, we can choose the default
    // let's try to schedule (team_size*10) tasks
    grainsize = thread->th.th_team_nproc * 10;
    KMP_FALLTHROUGH();
  case 2: // num_tasks provided
    if (grainsize > tc) {
      num_tasks = tc; // too big num_tasks requested, adjust values
      grainsize = 1;
      extras = 0;
    } else {
      num_tasks = grainsize;
      grainsize = tc / num_tasks;
      extras = tc % num_tasks;
    }
    break;
  case 1: // grainsize provided
    if (grainsize > tc) {
      num_tasks = 1;
      grainsize = tc; // too big grainsize requested, adjust values
      extras = 0;
    } else {
      if (modifier) {
        num_tasks = (tc + grainsize - 1) / grainsize;
        last_chunk = tc - (num_tasks * grainsize);
        extras = 0;
      } else {
        num_tasks = tc / grainsize;
        // adjust grainsize for balanced distribution of iterations
        grainsize = tc / num_tasks;
        extras = tc % num_tasks;
      }
    }
    break;
  default:
    KMP_ASSERT2(0, "unknown scheduling of taskloop");
  }

  KMP_DEBUG_ASSERT(tc == num_tasks * grainsize +
                             (last_chunk < 0 ? last_chunk : extras));
  KMP_DEBUG_ASSERT(num_tasks > extras);
  KMP_DEBUG_ASSERT(num_tasks > 0);
  // =========================================================================

  // check if clause value first
  // Also require GOMP_taskloop to reduce to linear (taskdata->td_flags.native)
  if (if_val == 0) { // if(0) specified, mark task as serial
    taskdata->td_flags.task_serial = 1;
    taskdata->td_flags.tiedness = TASK_TIED; // AC: serial task cannot be untied
    // always start serial tasks linearly
    __kmp_taskloop_linear(loc, gtid, task, lb, ub, st, ub_glob, num_tasks,
                          grainsize, extras, last_chunk, tc,
#if OMPT_SUPPORT
                          OMPT_GET_RETURN_ADDRESS(0),
#endif
                          task_dup);
    // !taskdata->td_flags.native => currently force linear spawning of tasks
    // for GOMP_taskloop
  } else if (num_tasks > num_tasks_min && !taskdata->td_flags.native) {
    KA_TRACE(20, ("__kmp_taskloop: T#%d, go recursive: tc %llu, #tasks %llu"
                  "(%lld), grain %llu, extras %llu, last_chunk %lld\n",
                  gtid, tc, num_tasks, num_tasks_min, grainsize, extras,
                  last_chunk));
    __kmp_taskloop_recur(loc, gtid, task, lb, ub, st, ub_glob, num_tasks,
                         grainsize, extras, last_chunk, tc, num_tasks_min,
#if OMPT_SUPPORT
                         OMPT_GET_RETURN_ADDRESS(0),
#endif
                         task_dup);
  } else {
    KA_TRACE(20, ("__kmp_taskloop: T#%d, go linear: tc %llu, #tasks %llu"
                  "(%lld), grain %llu, extras %llu, last_chunk %lld\n",
                  gtid, tc, num_tasks, num_tasks_min, grainsize, extras,
                  last_chunk));
    __kmp_taskloop_linear(loc, gtid, task, lb, ub, st, ub_glob, num_tasks,
                          grainsize, extras, last_chunk, tc,
#if OMPT_SUPPORT
                          OMPT_GET_RETURN_ADDRESS(0),
#endif
                          task_dup);
  }

#if OMPT_SUPPORT && OMPT_OPTIONAL
  if (ompt_enabled.ompt_callback_work) {
    ompt_callbacks.ompt_callback(ompt_callback_work)(
        ompt_work_taskloop, ompt_scope_end, &(team_info->parallel_data),
        &(task_info->task_data), tc, OMPT_GET_RETURN_ADDRESS(0));
  }
#endif

  if (nogroup == 0) {
#if OMPT_SUPPORT && OMPT_OPTIONAL
    OMPT_STORE_RETURN_ADDRESS(gtid);
#endif
    __kmpc_end_taskgroup(loc, gtid);
  }
  KA_TRACE(20, ("__kmp_taskloop(exit): T#%d\n", gtid));
}

/*!
@ingroup TASKING
@param loc       Source location information
@param gtid      Global thread ID
@param task      Task structure
@param if_val    Value of the if clause
@param lb        Pointer to loop lower bound in task structure
@param ub        Pointer to loop upper bound in task structure
@param st        Loop stride
@param nogroup   Flag, 1 if nogroup clause specified, 0 otherwise
@param sched     Schedule specified 0/1/2 for none/grainsize/num_tasks
@param grainsize Schedule value if specified
@param task_dup  Tasks duplication routine

Execute the taskloop construct.
*/
void __kmpc_taskloop(ident_t *loc, int gtid, kmp_task_t *task, int if_val,
                     kmp_uint64 *lb, kmp_uint64 *ub, kmp_int64 st, int nogroup,
                     int sched, kmp_uint64 grainsize, void *task_dup) {
  __kmp_assert_valid_gtid(gtid);
  KA_TRACE(20, ("__kmpc_taskloop(enter): T#%d\n", gtid));
  __kmp_taskloop(loc, gtid, task, if_val, lb, ub, st, nogroup, sched, grainsize,
                 0, task_dup);
  KA_TRACE(20, ("__kmpc_taskloop(exit): T#%d\n", gtid));
}

/*!
@ingroup TASKING
@param loc       Source location information
@param gtid      Global thread ID
@param task      Task structure
@param if_val    Value of the if clause
@param lb        Pointer to loop lower bound in task structure
@param ub        Pointer to loop upper bound in task structure
@param st        Loop stride
@param nogroup   Flag, 1 if nogroup clause specified, 0 otherwise
@param sched     Schedule specified 0/1/2 for none/grainsize/num_tasks
@param grainsize Schedule value if specified
@param modifier  Modifier 'strict' for sched, 1 if present, 0 otherwise
@param task_dup  Tasks duplication routine

Execute the taskloop construct.
*/
void __kmpc_taskloop_5(ident_t *loc, int gtid, kmp_task_t *task, int if_val,
                       kmp_uint64 *lb, kmp_uint64 *ub, kmp_int64 st,
                       int nogroup, int sched, kmp_uint64 grainsize,
                       int modifier, void *task_dup) {
  __kmp_assert_valid_gtid(gtid);
  KA_TRACE(20, ("__kmpc_taskloop_5(enter): T#%d\n", gtid));
  __kmp_taskloop(loc, gtid, task, if_val, lb, ub, st, nogroup, sched, grainsize,
                 modifier, task_dup);
  KA_TRACE(20, ("__kmpc_taskloop_5(exit): T#%d\n", gtid));
}