File size: 18,765 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
/******************************************************************************
 * Copyright (c) 2011-2021, 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.
 *
 ******************************************************************************/

#pragma once

#include <cub/agent/agent_three_way_partition.cuh>
#include <cub/config.cuh>
#include <cub/device/dispatch/dispatch_scan.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 <cstdio>
#include <iterator>

#include <nv/target>

CUB_NAMESPACE_BEGIN

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

template <typename ChainedPolicyT,
          typename InputIteratorT,
          typename FirstOutputIteratorT,
          typename SecondOutputIteratorT,
          typename UnselectedOutputIteratorT,
          typename NumSelectedIteratorT,
          typename ScanTileStateT,
          typename SelectFirstPartOp,
          typename SelectSecondPartOp,
          typename OffsetT>
__launch_bounds__(int(ChainedPolicyT::ActivePolicy::ThreeWayPartitionPolicy::BLOCK_THREADS)) __global__
  void DeviceThreeWayPartitionKernel(InputIteratorT d_in,
                                     FirstOutputIteratorT d_first_part_out,
                                     SecondOutputIteratorT d_second_part_out,
                                     UnselectedOutputIteratorT d_unselected_out,
                                     NumSelectedIteratorT d_num_selected_out,
                                     ScanTileStateT tile_status_1,
                                     ScanTileStateT tile_status_2,
                                     SelectFirstPartOp select_first_part_op,
                                     SelectSecondPartOp select_second_part_op,
                                     OffsetT num_items,
                                     int num_tiles)
{
  using AgentThreeWayPartitionPolicyT =
    typename ChainedPolicyT::ActivePolicy::ThreeWayPartitionPolicy;

  // Thread block type for selecting data from input tiles
  using AgentThreeWayPartitionT = AgentThreeWayPartition<AgentThreeWayPartitionPolicyT,
                                                         InputIteratorT,
                                                         FirstOutputIteratorT,
                                                         SecondOutputIteratorT,
                                                         UnselectedOutputIteratorT,
                                                         SelectFirstPartOp,
                                                         SelectSecondPartOp,
                                                         OffsetT>;

  // Shared memory for AgentThreeWayPartition
  __shared__ typename AgentThreeWayPartitionT::TempStorage temp_storage;

  // Process tiles
  AgentThreeWayPartitionT(temp_storage,
                          d_in,
                          d_first_part_out,
                          d_second_part_out,
                          d_unselected_out,
                          select_first_part_op,
                          select_second_part_op,
                          num_items)
    .ConsumeRange(num_tiles, tile_status_1, tile_status_2, d_num_selected_out);
}

/**
 * @brief Initialization kernel for tile status initialization (multi-block)
 *
 * @tparam ScanTileStateT
 *   Tile status interface type
 *
 * @tparam NumSelectedIteratorT
 *   Output iterator type for recording the number of items selected
 *
 * @param[in] tile_state_1
 *   Tile status interface
 *
 * @param[in] tile_state_2
 *   Tile status interface
 *
 * @param[in] num_tiles
 *   Number of tiles
 *
 * @param[out] d_num_selected_out
 *   Pointer to the total number of items selected
 *   (i.e., length of @p d_selected_out)
 */
template <typename ScanTileStateT, typename NumSelectedIteratorT>
__global__ void DeviceThreeWayPartitionInitKernel(ScanTileStateT tile_state_1,
                                                  ScanTileStateT tile_state_2,
                                                  int num_tiles,
                                                  NumSelectedIteratorT d_num_selected_out)
{
  // Initialize tile status
  tile_state_1.InitializeStatus(num_tiles);
  tile_state_2.InitializeStatus(num_tiles);

  // Initialize d_num_selected_out
  if (blockIdx.x == 0)
  {
    if (threadIdx.x < 2)
    {
      d_num_selected_out[threadIdx.x] = 0;
    }
  }
}

namespace detail
{

template <class InputT>
struct device_three_way_partition_policy_hub
{
  /// SM35
  struct Policy350 : ChainedPolicy<350, Policy350, Policy350>
  {
    constexpr static int ITEMS_PER_THREAD = Nominal4BItemsToItems<InputT>(9);

    using ThreeWayPartitionPolicy = cub::AgentThreeWayPartitionPolicy<256,
                                                                      ITEMS_PER_THREAD,
                                                                      cub::BLOCK_LOAD_DIRECT,
                                                                      cub::LOAD_DEFAULT,
                                                                      cub::BLOCK_SCAN_WARP_SCANS>;
  };

  using MaxPolicy = Policy350;
};

} // namespace detail

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

template <typename InputIteratorT,
          typename FirstOutputIteratorT,
          typename SecondOutputIteratorT,
          typename UnselectedOutputIteratorT,
          typename NumSelectedIteratorT,
          typename SelectFirstPartOp,
          typename SelectSecondPartOp,
          typename OffsetT,
          typename SelectedPolicy =
            detail::device_three_way_partition_policy_hub<cub::detail::value_t<InputIteratorT>>>
struct DispatchThreeWayPartitionIf
{
  /*****************************************************************************
   * Types and constants
   ****************************************************************************/

  using ScanTileStateT = cub::ScanTileState<OffsetT>;

  constexpr static int INIT_KERNEL_THREADS = 256;

  void *d_temp_storage;
  std::size_t &temp_storage_bytes;
  InputIteratorT d_in;
  FirstOutputIteratorT d_first_part_out;
  SecondOutputIteratorT d_second_part_out;
  UnselectedOutputIteratorT d_unselected_out;
  NumSelectedIteratorT d_num_selected_out;
  SelectFirstPartOp select_first_part_op;
  SelectSecondPartOp select_second_part_op;
  OffsetT num_items;
  cudaStream_t stream;

  CUB_RUNTIME_FUNCTION __forceinline__
  DispatchThreeWayPartitionIf(void *d_temp_storage,
                              std::size_t &temp_storage_bytes,
                              InputIteratorT d_in,
                              FirstOutputIteratorT d_first_part_out,
                              SecondOutputIteratorT d_second_part_out,
                              UnselectedOutputIteratorT d_unselected_out,
                              NumSelectedIteratorT d_num_selected_out,
                              SelectFirstPartOp select_first_part_op,
                              SelectSecondPartOp select_second_part_op,
                              OffsetT num_items,
                              cudaStream_t stream)
      : d_temp_storage(d_temp_storage)
      , temp_storage_bytes(temp_storage_bytes)
      , d_in(d_in)
      , d_first_part_out(d_first_part_out)
      , d_second_part_out(d_second_part_out)
      , d_unselected_out(d_unselected_out)
      , d_num_selected_out(d_num_selected_out)
      , select_first_part_op(select_first_part_op)
      , select_second_part_op(select_second_part_op)
      , num_items(num_items)
      , stream(stream)
  {}

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

  template <typename ActivePolicyT, typename ScanInitKernelPtrT, typename SelectIfKernelPtrT>
  CUB_RUNTIME_FUNCTION __forceinline__ cudaError_t
  Invoke(ScanInitKernelPtrT three_way_partition_init_kernel,
         SelectIfKernelPtrT three_way_partition_kernel)
  {
    cudaError error = cudaSuccess;

    const int block_threads = ActivePolicyT::ThreeWayPartitionPolicy::BLOCK_THREADS;
    const int items_per_thread = ActivePolicyT::ThreeWayPartitionPolicy::ITEMS_PER_THREAD;

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

      // Number of input tiles
      int tile_size = block_threads * items_per_thread;
      int num_tiles = static_cast<int>(DivideAndRoundUp(num_items, tile_size));

      // Specify temporary storage allocation requirements
      size_t allocation_sizes[2]; // bytes needed for tile status descriptors

      if (CubDebug(error = ScanTileStateT::AllocationSize(num_tiles, allocation_sizes[0])))
      {
        break;
      }

      allocation_sizes[1] = allocation_sizes[0];

      // Compute allocation pointers into the single storage blob (or compute
      // the necessary size of the blob)
      void *allocations[2] = {};
      if (CubDebug(error = cub::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;
      }

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

      // Construct the tile status interface
      ScanTileStateT tile_status_1;
      ScanTileStateT tile_status_2;

      if (CubDebug(error = tile_status_1.Init(num_tiles, allocations[0], allocation_sizes[0])))
      {
        break;
      }

      if (CubDebug(error = tile_status_2.Init(num_tiles, allocations[1], allocation_sizes[1])))
      {
        break;
      }

      // Log three_way_partition_init_kernel configuration
      int init_grid_size = CUB_MAX(1, DivideAndRoundUp(num_tiles, INIT_KERNEL_THREADS));

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

      // Invoke three_way_partition_init_kernel to initialize tile descriptors
      THRUST_NS_QUALIFIER::cuda_cub::launcher::triple_chevron(init_grid_size,
                                                              INIT_KERNEL_THREADS,
                                                              0,
                                                              stream)
        .doit(three_way_partition_init_kernel,
              tile_status_1,
              tile_status_2,
              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;
      }

      // 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 = 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
                                            three_way_partition_kernel,
                                            block_threads)))
        {
          break;
        }

        _CubLog("Invoking three_way_partition_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,
                reinterpret_cast<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(three_way_partition_kernel,
              d_in,
              d_first_part_out,
              d_second_part_out,
              d_unselected_out,
              d_num_selected_out,
              tile_status_1,
              tile_status_2,
              select_first_part_op,
              select_second_part_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>(
      DeviceThreeWayPartitionInitKernel<ScanTileStateT, NumSelectedIteratorT>,
      DeviceThreeWayPartitionKernel<MaxPolicyT,
                                    InputIteratorT,
                                    FirstOutputIteratorT,
                                    SecondOutputIteratorT,
                                    UnselectedOutputIteratorT,
                                    NumSelectedIteratorT,
                                    ScanTileStateT,
                                    SelectFirstPartOp,
                                    SelectSecondPartOp,
                                    OffsetT>);
  }

  /**
   * Internal dispatch routine
   */
  CUB_RUNTIME_FUNCTION __forceinline__ static cudaError_t
  Dispatch(void *d_temp_storage,
           std::size_t &temp_storage_bytes,
           InputIteratorT d_in,
           FirstOutputIteratorT d_first_part_out,
           SecondOutputIteratorT d_second_part_out,
           UnselectedOutputIteratorT d_unselected_out,
           NumSelectedIteratorT d_num_selected_out,
           SelectFirstPartOp select_first_part_op,
           SelectSecondPartOp select_second_part_op,
           OffsetT num_items,
           cudaStream_t stream)
  {
    using MaxPolicyT = typename SelectedPolicy::MaxPolicy;

    cudaError error = cudaSuccess;

    do
    {
      // Get PTX version
      int ptx_version = 0;
      if (CubDebug(error = cub::PtxVersion(ptx_version)))
      {
        break;
      }

      DispatchThreeWayPartitionIf dispatch(d_temp_storage,
                                           temp_storage_bytes,
                                           d_in,
                                           d_first_part_out,
                                           d_second_part_out,
                                           d_unselected_out,
                                           d_num_selected_out,
                                           select_first_part_op,
                                           select_second_part_op,
                                           num_items,
                                           stream);

      // Dispatch
      if (CubDebug(error = MaxPolicyT::Invoke(ptx_version, dispatch)))
      {
        break;
      }
    } while (0);

    return error;
  }

  CUB_DETAIL_RUNTIME_DEBUG_SYNC_IS_NOT_SUPPORTED
  CUB_RUNTIME_FUNCTION __forceinline__ static cudaError_t
  Dispatch(void *d_temp_storage,
           std::size_t &temp_storage_bytes,
           InputIteratorT d_in,
           FirstOutputIteratorT d_first_part_out,
           SecondOutputIteratorT d_second_part_out,
           UnselectedOutputIteratorT d_unselected_out,
           NumSelectedIteratorT d_num_selected_out,
           SelectFirstPartOp select_first_part_op,
           SelectSecondPartOp select_second_part_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_first_part_out,
                    d_second_part_out,
                    d_unselected_out,
                    d_num_selected_out,
                    select_first_part_op,
                    select_second_part_op,
                    num_items,
                    stream);
  }
};

CUB_NAMESPACE_END