File size: 23,540 Bytes
0dc1b04
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
/******************************************************************************
 * Copyright (c) 2011, Duane Merrill.  All rights reserved.
 * Copyright (c) 2011-2018, NVIDIA CORPORATION.  All rights reserved.
 *
 * Redistribution and use in source and binary forms, with or without
 * modification, are permitted provided that the following conditions are met:
 *     * Redistributions of source code must retain the above copyright
 *       notice, this list of conditions and the following disclaimer.
 *     * Redistributions in binary form must reproduce the above copyright
 *       notice, this list of conditions and the following disclaimer in the
 *       documentation and/or other materials provided with the distribution.
 *     * Neither the name of the NVIDIA CORPORATION nor the
 *       names of its contributors may be used to endorse or promote products
 *       derived from this software without specific prior written permission.
 *
 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND
 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
 * WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
 * DISCLAIMED. IN NO EVENT SHALL NVIDIA CORPORATION BE LIABLE FOR ANY
 * DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
 * (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
 * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND
 * ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
 * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 *
 ******************************************************************************/

/**
 * @file
 *   cub::DeviceSelect provides device-wide, parallel operations for selecting items from sequences 
 *   of data items residing within device-accessible memory.
 */

#pragma once

#include <cub/agent/agent_select_if.cuh>
#include <cub/config.cuh>
#include <cub/device/dispatch/dispatch_scan.cuh>
#include <cub/grid/grid_queue.cuh>
#include <cub/thread/thread_operators.cuh>
#include <cub/util_deprecated.cuh>
#include <cub/util_device.cuh>
#include <cub/util_math.cuh>

#include <thrust/system/cuda/detail/core/triple_chevron_launch.h>

#include <nv/target>

#include <cstdio>
#include <iterator>

CUB_NAMESPACE_BEGIN

/******************************************************************************
 * Kernel entry points
 *****************************************************************************/

/**
 * Select kernel entry point (multi-block)
 *
 * Performs functor-based selection if SelectOpT functor type != NullType
 * Otherwise performs flag-based selection if FlagsInputIterator's value type != NullType
 * Otherwise performs discontinuity selection (keep unique)
 *
 * @tparam InputIteratorT 
 *   Random-access input iterator type for reading input items
 *
 * @tparam FlagsInputIteratorT 
 *   Random-access input iterator type for reading selection flags (NullType* if a selection functor 
 *   or discontinuity flagging is to be used for selection)
 *
 * @tparam SelectedOutputIteratorT 
 *   Random-access output iterator type for writing selected items
 *
 * @tparam NumSelectedIteratorT 
 *   Output iterator type for recording the number of items selected
 *
 * @tparam ScanTileStateT 
 *   Tile status interface type
 *
 * @tparam SelectOpT 
 *   Selection operator type (NullType if selection flags or discontinuity flagging is 
 *   to be used for selection)
 *
 * @tparam EqualityOpT 
 *   Equality operator type (NullType if selection functor or selection flags is 
 *   to be used for selection)
 *
 * @tparam OffsetT 
 *   Signed integer type for global offsets
 *
 * @tparam KEEP_REJECTS 
 *   Whether or not we push rejected items to the back of the output
 *
 * @param[in] d_in
 *   Pointer to the input sequence of data items
 * 
 * @param[in] d_flags
 *   Pointer to the input sequence of selection flags (if applicable)
 *
 * @param[out] d_selected_out
 *   Pointer to the output sequence of selected data items
 *
 * @param[out] d_num_selected_out
 *   Pointer to the total number of items selected (i.e., length of \p d_selected_out)
 *
 * @param[in] tile_status 
 *   Tile status interface
 *
 * @param[in] select_op
 *   Selection operator
 * 
 * @param[in] equality_op
 *   Equality operator
 * 
 * @param[in] num_items
 *   Total number of input items (i.e., length of \p d_in)
 * 
 * @param[in] num_tiles
 *   Total number of tiles for the entire problem
 */
template <typename ChainedPolicyT,
          typename InputIteratorT,
          typename FlagsInputIteratorT,
          typename SelectedOutputIteratorT,
          typename NumSelectedIteratorT,
          typename ScanTileStateT,
          typename SelectOpT,
          typename EqualityOpT,
          typename OffsetT,
          bool KEEP_REJECTS>
__launch_bounds__(int(ChainedPolicyT::ActivePolicy::SelectIfPolicyT::BLOCK_THREADS)) __global__
  void DeviceSelectSweepKernel(InputIteratorT d_in,
                               FlagsInputIteratorT d_flags,
                               SelectedOutputIteratorT d_selected_out,
                               NumSelectedIteratorT d_num_selected_out,
                               ScanTileStateT tile_status,
                               SelectOpT select_op,
                               EqualityOpT equality_op,
                               OffsetT num_items,
                               int num_tiles)
{
    using AgentSelectIfPolicyT = typename ChainedPolicyT::ActivePolicy::SelectIfPolicyT;

    // Thread block type for selecting data from input tiles
    using AgentSelectIfT = AgentSelectIf<AgentSelectIfPolicyT,
                                         InputIteratorT,
                                         FlagsInputIteratorT,
                                         SelectedOutputIteratorT,
                                         SelectOpT,
                                         EqualityOpT,
                                         OffsetT,
                                         KEEP_REJECTS>;

    // Shared memory for AgentSelectIf
    __shared__ typename AgentSelectIfT::TempStorage temp_storage;

    // Process tiles
    AgentSelectIfT(temp_storage, d_in, d_flags, d_selected_out, select_op, equality_op, num_items).ConsumeRange(
        num_tiles,
        tile_status,
        d_num_selected_out);
}


namespace detail 
{

template <class InputT, bool MayAlias>
struct device_select_policy_hub
{
    struct Policy350 : ChainedPolicy<350, Policy350, Policy350>
    {
      static constexpr int NOMINAL_4B_ITEMS_PER_THREAD = 10;

      static constexpr int ITEMS_PER_THREAD =
        CUB_MIN(NOMINAL_4B_ITEMS_PER_THREAD,
                CUB_MAX(1, (NOMINAL_4B_ITEMS_PER_THREAD * 4 / sizeof(InputT))));

      using SelectIfPolicyT = AgentSelectIfPolicy<128,
                                                  ITEMS_PER_THREAD,
                                                  BLOCK_LOAD_DIRECT,
                                                  MayAlias ? LOAD_CA : LOAD_LDG,
                                                  BLOCK_SCAN_WARP_SCANS,
                                                  detail::fixed_delay_constructor_t<350, 450>>;
    };

    using MaxPolicy = Policy350;
};

} // detail


/******************************************************************************
 * Dispatch
 ******************************************************************************/

/**
 * Utility class for dispatching the appropriately-tuned kernels for DeviceSelect
 *
 * @tparam InputIteratorT
 *   Random-access input iterator type for reading input items
 *
 * @tparam FlagsInputIteratorT
 *   Random-access input iterator type for reading selection flags 
 *   (NullType* if a selection functor or discontinuity flagging is to be used for selection)
 *
 * @tparam SelectedOutputIteratorT
 *   Random-access output iterator type for writing selected items
 *
 * @tparam NumSelectedIteratorT
 *   Output iterator type for recording the number of items selected
 *
 * @tparam SelectOpT
 *   Selection operator type (NullType if selection flags or discontinuity flagging is 
 *   to be used for selection)
 *
 * @tparam EqualityOpT
 *   Equality operator type (NullType if selection functor or selection flags is to 
 *   be used for selection)
 *
 * @tparam OffsetT
 *   Signed integer type for global offsets
 *
 * @tparam KEEP_REJECTS
 *   Whether or not we push rejected items to the back of the output
 */
template <typename InputIteratorT,
          typename FlagsInputIteratorT,
          typename SelectedOutputIteratorT,
          typename NumSelectedIteratorT,
          typename SelectOpT,
          typename EqualityOpT,
          typename OffsetT,
          bool KEEP_REJECTS,
          bool MayAlias = false,
          typename SelectedPolicy =
            detail::device_select_policy_hub<cub::detail::value_t<InputIteratorT>, MayAlias>>
struct DispatchSelectIf : SelectedPolicy
{
    /******************************************************************************
     * Types and constants
     ******************************************************************************/

    // The input value type
    using InputT = cub::detail::value_t<InputIteratorT>;

    // The flag value type
    using FlagT = cub::detail::value_t<FlagsInputIteratorT>;

    // Tile status descriptor interface type
    using ScanTileStateT = ScanTileState<OffsetT>;

    static constexpr int INIT_KERNEL_THREADS = 128;

    /// Device-accessible allocation of temporary storage. 
    /// When `nullptr`, the required allocation size is written to `temp_storage_bytes` 
    /// and no work is done.
    void* d_temp_storage;

    /// Reference to size in bytes of `d_temp_storage` allocation
    size_t& temp_storage_bytes; 

    /// Pointer to the input sequence of data items
    InputIteratorT d_in;

    /// Pointer to the input sequence of selection flags (if applicable)
    FlagsInputIteratorT d_flags;

    /// Pointer to the output sequence of selected data items
    SelectedOutputIteratorT d_selected_out;

    /// Pointer to the total number of items selected (i.e., length of `d_selected_out`)
    NumSelectedIteratorT d_num_selected_out;

    /// Selection operator
    SelectOpT select_op;

    /// Equality operator
    EqualityOpT equality_op;

    /// Total number of input items (i.e., length of `d_in`)
    OffsetT num_items;

    /// CUDA stream to launch kernels within. Default is stream<sub>0</sub>.
    cudaStream_t stream;

    int ptx_version;

    /**
     * @param d_temp_storage
     *   Device-accessible allocation of temporary storage. 
     *   When `nullptr`, the required allocation size is written to `temp_storage_bytes` 
     *   and no work is done.
     * 
     * @param temp_storage_bytes 
     *   Reference to size in bytes of `d_temp_storage` allocation
     *
     * @param d_in
     *   Pointer to the input sequence of data items
     *
     * @param d_flags
     *   Pointer to the input sequence of selection flags (if applicable)
     *
     * @param d_selected_out
     *   Pointer to the output sequence of selected data items
     *
     * @param d_num_selected_out
     *  Pointer to the total number of items selected (i.e., length of `d_selected_out`)
     *
     * @param select_op
     *   Selection operator
     *
     * @param equality_op
     *   Equality operator
     *
     * @param num_items
     *   Total number of input items (i.e., length of `d_in`)
     *
     * @param stream
     *   CUDA stream to launch kernels within. Default is stream<sub>0</sub>.
     */
    CUB_RUNTIME_FUNCTION __forceinline__ DispatchSelectIf(void *d_temp_storage,
                                                          size_t &temp_storage_bytes,
                                                          InputIteratorT d_in,
                                                          FlagsInputIteratorT d_flags,
                                                          SelectedOutputIteratorT d_selected_out,
                                                          NumSelectedIteratorT d_num_selected_out,
                                                          SelectOpT select_op,
                                                          EqualityOpT equality_op,
                                                          OffsetT num_items,
                                                          cudaStream_t stream,
                                                          int ptx_version)
        : d_temp_storage(d_temp_storage)
        , temp_storage_bytes(temp_storage_bytes)
        , d_in(d_in)
        , d_flags(d_flags)
        , d_selected_out(d_selected_out)
        , d_num_selected_out(d_num_selected_out)
        , select_op(select_op)
        , equality_op(equality_op)
        , num_items(num_items)
        , stream(stream)
        , ptx_version(ptx_version)
    {}

    /******************************************************************************
     * Dispatch entrypoints
     ******************************************************************************/

    /**
     * Internal dispatch routine for computing a device-wide selection using the
     * specified kernel functions.
     */
    template <typename ActivePolicyT, typename ScanInitKernelPtrT, typename SelectIfKernelPtrT>
    CUB_RUNTIME_FUNCTION __forceinline__ cudaError_t Invoke(ScanInitKernelPtrT scan_init_kernel,
                                                            SelectIfKernelPtrT select_if_kernel)
    {
        cudaError error = cudaSuccess;

        const int block_threads = ActivePolicyT::SelectIfPolicyT::BLOCK_THREADS;
        const int items_per_thread = ActivePolicyT::SelectIfPolicyT::ITEMS_PER_THREAD;
        const int tile_size = block_threads * items_per_thread;

        do
        {
            // Get device ordinal
            int device_ordinal;
            if (CubDebug(error = cudaGetDevice(&device_ordinal))) 
            {
                break;
            }

            // Number of input tiles
            int num_tiles = static_cast<int>(cub::DivideAndRoundUp(num_items, tile_size));

            // Specify temporary storage allocation requirements
            size_t  allocation_sizes[1];

            // bytes needed for tile status descriptors
            if (CubDebug(error = ScanTileStateT::AllocationSize(num_tiles, allocation_sizes[0]))) 
            {
                break;
            }

            // Compute allocation pointers into the single storage blob (or compute the necessary size of the blob)
            void* allocations[1] = {};
            if (CubDebug(error = AliasTemporaries(d_temp_storage, temp_storage_bytes, allocations, allocation_sizes))) 
            {
                break;
            }

            if (d_temp_storage == nullptr)
            {
                // Return if the caller is simply requesting the size of the storage allocation
                break;
            }

            // Construct the tile status interface
            ScanTileStateT tile_status;
            if (CubDebug(error = tile_status.Init(num_tiles, allocations[0], allocation_sizes[0]))) 
            {
                break;
            }

            // Log scan_init_kernel configuration
            int init_grid_size = CUB_MAX(1, cub::DivideAndRoundUp(num_tiles, INIT_KERNEL_THREADS));

            #ifdef CUB_DETAIL_DEBUG_ENABLE_LOG
            _CubLog("Invoking scan_init_kernel<<<%d, %d, 0, %lld>>>()\n", init_grid_size, INIT_KERNEL_THREADS, (long long) stream);
            #endif

            // Invoke scan_init_kernel to initialize tile descriptors
            THRUST_NS_QUALIFIER::cuda_cub::launcher::triple_chevron(
                init_grid_size, INIT_KERNEL_THREADS, 0, stream
            ).doit(scan_init_kernel,
                tile_status,
                num_tiles,
                d_num_selected_out);

            // Check for failure to launch
            if (CubDebug(error = cudaPeekAtLastError()))
            {
                break;
            }

            // Sync the stream if specified to flush runtime errors
            error = detail::DebugSyncStream(stream);
            if (CubDebug(error))
            {
              break;
            }

            // Return if empty problem
            if (num_items == 0)
            {
                break;
            }

            // Get max x-dimension of grid
            int max_dim_x;
            if (CubDebug(error = cudaDeviceGetAttribute(&max_dim_x, cudaDevAttrMaxGridDimX, device_ordinal))) 
            {
                break;
            }

            // Get grid size for scanning tiles
            dim3 scan_grid_size;
            scan_grid_size.z = 1;
            scan_grid_size.y = cub::DivideAndRoundUp(num_tiles, max_dim_x);
            scan_grid_size.x = CUB_MIN(num_tiles, max_dim_x);

            // Log select_if_kernel configuration
            #ifdef CUB_DETAIL_DEBUG_ENABLE_LOG
            {
              // Get SM occupancy for select_if_kernel
              int range_select_sm_occupancy;
              if (CubDebug(error = MaxSmOccupancy(range_select_sm_occupancy, // out
                                                  select_if_kernel,
                                                  block_threads)))
              {
                break;
              }

              _CubLog("Invoking select_if_kernel<<<{%d,%d,%d}, %d, 0, "
                      "%lld>>>(), %d items per thread, %d SM occupancy\n",
                      scan_grid_size.x,
                      scan_grid_size.y,
                      scan_grid_size.z,
                      block_threads,
                      (long long)stream,
                      items_per_thread,
                      range_select_sm_occupancy);
            }
            #endif

            // Invoke select_if_kernel
            THRUST_NS_QUALIFIER::cuda_cub::launcher::triple_chevron(
                scan_grid_size, block_threads, 0, stream
            ).doit(select_if_kernel,
                d_in,
                d_flags,
                d_selected_out,
                d_num_selected_out,
                tile_status,
                select_op,
                equality_op,
                num_items,
                num_tiles);

            // Check for failure to launch
            if (CubDebug(error = cudaPeekAtLastError()))
            {
                break;
            }

            // Sync the stream if specified to flush runtime errors
            error = detail::DebugSyncStream(stream);
            if (CubDebug(error))
            {
              break;
            }
        }
        while (0);

        return error;
    }

    template <typename ActivePolicyT>
    CUB_RUNTIME_FUNCTION __forceinline__ cudaError_t Invoke()
    {
        using MaxPolicyT = typename SelectedPolicy::MaxPolicy;

        return Invoke<ActivePolicyT>(DeviceCompactInitKernel<ScanTileStateT, NumSelectedIteratorT>,
                                     DeviceSelectSweepKernel<MaxPolicyT,
                                                             InputIteratorT,
                                                             FlagsInputIteratorT,
                                                             SelectedOutputIteratorT,
                                                             NumSelectedIteratorT,
                                                             ScanTileStateT,
                                                             SelectOpT,
                                                             EqualityOpT,
                                                             OffsetT,
                                                             KEEP_REJECTS>);
    }

    /**
     * Internal dispatch routine
     *
     * @param d_temp_storage
     *   Device-accessible allocation of temporary storage. 
     *   When `nullptr`, the required allocation size is written to `temp_storage_bytes` 
     *   and no work is done.
     * 
     * @param temp_storage_bytes 
     *   Reference to size in bytes of `d_temp_storage` allocation
     *
     * @param d_in
     *   Pointer to the input sequence of data items
     *
     * @param d_flags
     *   Pointer to the input sequence of selection flags (if applicable)
     *
     * @param d_selected_out
     *   Pointer to the output sequence of selected data items
     *
     * @param d_num_selected_out
     *  Pointer to the total number of items selected (i.e., length of `d_selected_out`)
     *
     * @param select_op
     *   Selection operator
     *
     * @param equality_op
     *   Equality operator
     *
     * @param num_items
     *   Total number of input items (i.e., length of `d_in`)
     *
     * @param stream
     *   CUDA stream to launch kernels within. Default is stream<sub>0</sub>.
     */
    CUB_RUNTIME_FUNCTION __forceinline__ static cudaError_t
    Dispatch(void *d_temp_storage,
             size_t &temp_storage_bytes,
             InputIteratorT d_in,
             FlagsInputIteratorT d_flags,
             SelectedOutputIteratorT d_selected_out,
             NumSelectedIteratorT d_num_selected_out,
             SelectOpT select_op,
             EqualityOpT equality_op,
             OffsetT num_items,
             cudaStream_t stream)
    {
        using MaxPolicyT = typename SelectedPolicy::MaxPolicy;

        int ptx_version = 0;
        if (cudaError_t error = CubDebug(PtxVersion(ptx_version))) 
        {
            return error;
        }

        DispatchSelectIf dispatch(d_temp_storage,
                                    temp_storage_bytes,
                                    d_in,
                                    d_flags,
                                    d_selected_out,
                                    d_num_selected_out,
                                    select_op,
                                    equality_op,
                                    num_items,
                                    stream,
                                    ptx_version);
        
        return CubDebug(MaxPolicyT::Invoke(ptx_version, dispatch));
    }

    CUB_DETAIL_RUNTIME_DEBUG_SYNC_IS_NOT_SUPPORTED
    CUB_RUNTIME_FUNCTION __forceinline__
    static cudaError_t Dispatch(
        void*                       d_temp_storage,          
        size_t&                     temp_storage_bytes,       
        InputIteratorT              d_in,                      
        FlagsInputIteratorT         d_flags,                    
        SelectedOutputIteratorT     d_selected_out,              
        NumSelectedIteratorT        d_num_selected_out,           
        SelectOpT                   select_op,                     
        EqualityOpT                 equality_op,                    
        OffsetT                     num_items,            
        cudaStream_t                stream,                
        bool                        debug_synchronous)
    {
      CUB_DETAIL_RUNTIME_DEBUG_SYNC_USAGE_LOG

      return Dispatch(d_temp_storage,
                      temp_storage_bytes,
                      d_in,
                      d_flags,
                      d_selected_out,
                      d_num_selected_out,
                      select_op,
                      equality_op,
                      num_items,
                      stream);
    }
};


CUB_NAMESPACE_END