File size: 21,810 Bytes
be11144
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
/******************************************************************************
 * 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.
 *
 ******************************************************************************/

/******************************************************************************
 * Test of DeviceSelect::Unique utilities
 ******************************************************************************/

// Ensure printing of CUDA runtime errors to console
#define CUB_STDERR

#include <stdio.h>
#include <typeinfo>

#include <thrust/device_ptr.h>
#include <thrust/unique.h>

#include <cub/util_allocator.cuh>
#include <cub/iterator/counting_input_iterator.cuh>
#include <cub/device/device_select.cuh>

#include <thrust/device_ptr.h>
#include <thrust/unique.h>

#include "test_util.h"

using namespace cub;


//---------------------------------------------------------------------
// Globals, constants and typedefs
//---------------------------------------------------------------------

bool                    g_verbose               = false;
int                     g_timing_iterations     = 0;
int                     g_repeat                = 0;
float                   g_device_giga_bandwidth;
CachingDeviceAllocator  g_allocator(true);

// Dispatch types
enum Backend
{
    CUB,        // CUB method
    THRUST,     // Thrust method
    CDP,        // GPU-based (dynamic parallelism) dispatch to CUB method
};


//---------------------------------------------------------------------
// Dispatch to different CUB DeviceSelect entrypoints
//---------------------------------------------------------------------


/**
 * Dispatch to unique entrypoint
 */
template <typename InputIteratorT, typename OutputIteratorT, typename NumSelectedIteratorT, typename OffsetT>
CUB_RUNTIME_FUNCTION __forceinline__
cudaError_t Dispatch(
    Int2Type<CUB>               /*dispatch_to*/,
    int                         timing_timing_iterations,
    size_t                      */*d_temp_storage_bytes*/,
    cudaError_t                 */*d_cdp_error*/,

    void*               d_temp_storage,
    size_t                      &temp_storage_bytes,
    InputIteratorT              d_in,
    OutputIteratorT              d_out,
    NumSelectedIteratorT         d_num_selected_out,
    OffsetT                     num_items,
    cudaStream_t                stream,
    bool                        debug_synchronous)
{
    cudaError_t error = cudaSuccess;
    for (int i = 0; i < timing_timing_iterations; ++i)
    {
        error = DeviceSelect::Unique(d_temp_storage, temp_storage_bytes, d_in, d_out, d_num_selected_out, num_items, stream, debug_synchronous);
    }
    return error;
}


//---------------------------------------------------------------------
// Dispatch to different Thrust entrypoints
//---------------------------------------------------------------------


/**
 * Dispatch to unique entrypoint
 */
template <typename InputIteratorT, typename OutputIteratorT, typename NumSelectedIteratorT, typename OffsetT>
__host__ __forceinline__
cudaError_t Dispatch(
    Int2Type<THRUST>            /*dispatch_to*/,
    int                         timing_timing_iterations,
    size_t                      */*d_temp_storage_bytes*/,
    cudaError_t                 */*d_cdp_error*/,

    void                        *d_temp_storage,
    size_t                      &temp_storage_bytes,
    InputIteratorT              d_in,
    OutputIteratorT             d_out,
    NumSelectedIteratorT        d_num_selected_out,
    OffsetT                     num_items,
    cudaStream_t                /*stream*/,
    bool                        /*debug_synchronous*/)
{
    // The input value type
    typedef typename std::iterator_traits<InputIteratorT>::value_type InputT;

    // The output value type
    typedef typename If<(Equals<typename std::iterator_traits<OutputIteratorT>::value_type, void>::VALUE),  // OutputT =  (if output iterator's value type is void) ?
        typename std::iterator_traits<InputIteratorT>::value_type,                                          // ... then the input iterator's value type,
        typename std::iterator_traits<OutputIteratorT>::value_type>::Type OutputT;                          // ... else the output iterator's value type

    if (d_temp_storage == 0)
    {
        temp_storage_bytes = 1;
    }
    else
    {
        thrust::device_ptr<OutputT> d_out_wrapper_end;
        thrust::device_ptr<InputT> d_in_wrapper(d_in);
        thrust::device_ptr<OutputT> d_out_wrapper(d_out);
        for (int i = 0; i < timing_timing_iterations; ++i)
        {
            d_out_wrapper_end = thrust::unique_copy(d_in_wrapper, d_in_wrapper + num_items, d_out_wrapper);
        }

        OffsetT num_selected = OffsetT(d_out_wrapper_end - d_out_wrapper);
        CubDebugExit(cudaMemcpy(d_num_selected_out, &num_selected, sizeof(OffsetT), cudaMemcpyHostToDevice));

    }

    return cudaSuccess;
}



//---------------------------------------------------------------------
// CUDA Nested Parallelism Test Kernel
//---------------------------------------------------------------------

/**
 * Simple wrapper kernel to invoke DeviceSelect
 */
template <typename InputIteratorT, typename OutputIteratorT, typename NumSelectedIteratorT, typename OffsetT>
__global__ void CnpDispatchKernel(
    int                         timing_timing_iterations,
    size_t                      *d_temp_storage_bytes,
    cudaError_t                 *d_cdp_error,

    void*               d_temp_storage,
    size_t                      temp_storage_bytes,
    InputIteratorT              d_in,
    OutputIteratorT              d_out,
    NumSelectedIteratorT         d_num_selected_out,
    OffsetT                     num_items,
    bool                        debug_synchronous)
{

#ifndef CUB_CDP
    (void)timing_timing_iterations;
    (void)d_temp_storage_bytes;
    (void)d_cdp_error;
    (void)d_temp_storage;
    (void)temp_storage_bytes;
    (void)d_in;
    (void)d_out;
    (void)d_num_selected_out;
    (void)num_items;
    (void)debug_synchronous;
    *d_cdp_error = cudaErrorNotSupported;
#else
    *d_cdp_error = Dispatch(Int2Type<CUB>(), timing_timing_iterations, d_temp_storage_bytes, d_cdp_error,
        d_temp_storage, temp_storage_bytes, d_in, d_out, d_num_selected_out, num_items, 0, debug_synchronous);
    *d_temp_storage_bytes = temp_storage_bytes;
#endif
}


/**
 * Dispatch to CDP kernel
 */
template <typename InputIteratorT, typename OutputIteratorT, typename NumSelectedIteratorT, typename OffsetT>
cudaError_t Dispatch(
    Int2Type<CDP>               dispatch_to,
    int                         timing_timing_iterations,
    size_t                      *d_temp_storage_bytes,
    cudaError_t                 *d_cdp_error,

    void*               d_temp_storage,
    size_t                      &temp_storage_bytes,
    InputIteratorT              d_in,
    OutputIteratorT              d_out,
    NumSelectedIteratorT         d_num_selected_out,
    OffsetT                     num_items,
    cudaStream_t                stream,
    bool                        debug_synchronous)
{
    // Invoke kernel to invoke device-side dispatch
    CnpDispatchKernel<<<1,1>>>(timing_timing_iterations, d_temp_storage_bytes, d_cdp_error,
        d_temp_storage, temp_storage_bytes, d_in, d_out, d_num_selected_out, num_items, debug_synchronous);

    // Copy out temp_storage_bytes
    CubDebugExit(cudaMemcpy(&temp_storage_bytes, d_temp_storage_bytes, sizeof(size_t) * 1, cudaMemcpyDeviceToHost));

    // Copy out error
    cudaError_t retval;
    CubDebugExit(cudaMemcpy(&retval, d_cdp_error, sizeof(cudaError_t) * 1, cudaMemcpyDeviceToHost));
    return retval;
}



//---------------------------------------------------------------------
// Test generation
//---------------------------------------------------------------------


/**
 * Initialize problem
 */
template <typename T>
void Initialize(
    int         entropy_reduction,
    T           *h_in,
    int         num_items,
    int         max_segment)
{
    unsigned int max_int = (unsigned int) -1;

    int key = 0;
    int i = 0;
    while (i < num_items)
    {
        // Select number of repeating occurrences for the current run
        int repeat;
        if (max_segment < 0)
        {
            repeat = num_items;
        }
        else if (max_segment < 2)
        {
            repeat = 1;
        }
        else
        {
            RandomBits(repeat, entropy_reduction);
            repeat = (int) ((double(repeat) * double(max_segment)) / double(max_int));
            repeat = CUB_MAX(1, repeat);
        }

        int j = i;
        while (j < CUB_MIN(i + repeat, num_items))
        {
            InitValue(INTEGER_SEED, h_in[j], key);
            j++;
        }

        i = j;
        key++;
    }

    if (g_verbose)
    {
        printf("Input:\n");
        DisplayResults(h_in, num_items);
        printf("\n\n");
    }
}


/**
 * Solve unique problem
 */
template <
    typename        InputIteratorT,
    typename        T>
int Solve(
    InputIteratorT  h_in,
    T               *h_reference,
    int             num_items)
{
    int num_selected = 0;
    if (num_items > 0)
    {
        h_reference[num_selected] = h_in[0];
        num_selected++;
    }

    for (int i = 1; i < num_items; ++i)
    {
        if (h_in[i] != h_in[i - 1])
        {
            h_reference[num_selected] = h_in[i];
            num_selected++;
        }
    }

    return num_selected;
}



/**
 * Test DeviceSelect for a given problem input
 */
template <
    Backend             BACKEND,
    typename            DeviceInputIteratorT,
    typename            T>
void Test(
    DeviceInputIteratorT d_in,
    T                   *h_reference,
    int                 num_selected,
    int                 num_items)
{
    // Allocate device output array and num selected
    T       *d_out            = NULL;
    int     *d_num_selected_out   = NULL;
    CubDebugExit(g_allocator.DeviceAllocate((void**)&d_out, sizeof(T) * num_items));
    CubDebugExit(g_allocator.DeviceAllocate((void**)&d_num_selected_out, sizeof(int)));

    // Allocate CDP device arrays
    size_t          *d_temp_storage_bytes = NULL;
    cudaError_t     *d_cdp_error = NULL;
    CubDebugExit(g_allocator.DeviceAllocate((void**)&d_temp_storage_bytes,  sizeof(size_t) * 1));
    CubDebugExit(g_allocator.DeviceAllocate((void**)&d_cdp_error,           sizeof(cudaError_t) * 1));

    // Allocate temporary storage
    void            *d_temp_storage = NULL;
    size_t          temp_storage_bytes = 0;
    CubDebugExit(Dispatch(Int2Type<BACKEND>(), 1, d_temp_storage_bytes, d_cdp_error, d_temp_storage, temp_storage_bytes, d_in, d_out, d_num_selected_out, num_items, 0, true));
    CubDebugExit(g_allocator.DeviceAllocate(&d_temp_storage, temp_storage_bytes));

    // Clear device output array
    CubDebugExit(cudaMemset(d_out, 0, sizeof(T) * num_items));
    CubDebugExit(cudaMemset(d_num_selected_out, 0, sizeof(int)));

    // Run warmup/correctness iteration
    CubDebugExit(Dispatch(Int2Type<BACKEND>(), 1, d_temp_storage_bytes, d_cdp_error, d_temp_storage, temp_storage_bytes, d_in, d_out, d_num_selected_out, num_items, 0, true));

    // Check for correctness (and display results, if specified)
    int compare1 = CompareDeviceResults(h_reference, d_out, num_selected, true, g_verbose);
    printf("\t Data %s ", compare1 ? "FAIL" : "PASS");

    int compare2 = CompareDeviceResults(&num_selected, d_num_selected_out, 1, true, g_verbose);
    printf("\t Count %s ", compare2 ? "FAIL" : "PASS");

    // Flush any stdout/stderr
    fflush(stdout);
    fflush(stderr);

    // Performance
    GpuTimer gpu_timer;
    gpu_timer.Start();
    CubDebugExit(Dispatch(Int2Type<BACKEND>(), g_timing_iterations, d_temp_storage_bytes, d_cdp_error, d_temp_storage, temp_storage_bytes, d_in, d_out, d_num_selected_out, num_items, 0, false));
    gpu_timer.Stop();
    float elapsed_millis = gpu_timer.ElapsedMillis();

    // Display performance
    if (g_timing_iterations > 0)
    {
        float avg_millis        = elapsed_millis / g_timing_iterations;
        float giga_rate         = float(num_items) / avg_millis / 1000.0f / 1000.0f;
        float giga_bandwidth    = float((num_items + num_selected) * sizeof(T)) / avg_millis / 1000.0f / 1000.0f;
        printf(", %.3f avg ms, %.3f billion items/s, %.3f logical GB/s, %.1f%% peak", avg_millis, giga_rate, giga_bandwidth, giga_bandwidth / g_device_giga_bandwidth * 100.0);
    }
    printf("\n\n");

    // Flush any stdout/stderr
    fflush(stdout);
    fflush(stderr);

    // Cleanup
    if (d_out) CubDebugExit(g_allocator.DeviceFree(d_out));
    if (d_num_selected_out) CubDebugExit(g_allocator.DeviceFree(d_num_selected_out));
    if (d_temp_storage_bytes) CubDebugExit(g_allocator.DeviceFree(d_temp_storage_bytes));
    if (d_cdp_error) CubDebugExit(g_allocator.DeviceFree(d_cdp_error));
    if (d_temp_storage) CubDebugExit(g_allocator.DeviceFree(d_temp_storage));

    // Correctness asserts
    AssertEquals(0, compare1 | compare2);
}


/**
 * Test DeviceSelect on pointer type
 */
template <
    Backend         BACKEND,
    typename        T>
void TestPointer(
    int             num_items,
    int             entropy_reduction,
    int             max_segment)
{
    // Allocate host arrays
    T*  h_in        = new T[num_items];
    T*  h_reference = new T[num_items];

    // Initialize problem and solution
    Initialize(entropy_reduction, h_in, num_items, max_segment);
    int num_selected = Solve(h_in, h_reference, num_items);

    printf("\nPointer %s cub::DeviceSelect::Unique %d items, %d selected (avg run length %.3f), %s %d-byte elements, entropy_reduction %d\n",
        (BACKEND == CDP) ? "CDP CUB" : (BACKEND == THRUST) ? "Thrust" : "CUB",
        num_items, num_selected, float(num_items) / num_selected,
        typeid(T).name(),
        (int) sizeof(T),
        entropy_reduction);
    fflush(stdout);

    // Allocate problem device arrays
    T *d_in = NULL;
    CubDebugExit(g_allocator.DeviceAllocate((void**)&d_in, sizeof(T) * num_items));

    // Initialize device input
    CubDebugExit(cudaMemcpy(d_in, h_in, sizeof(T) * num_items, cudaMemcpyHostToDevice));

    // Run Test
    Test<BACKEND>(d_in, h_reference, num_selected, num_items);

    // Cleanup
    if (h_in) delete[] h_in;
    if (h_reference) delete[] h_reference;
    if (d_in) CubDebugExit(g_allocator.DeviceFree(d_in));
}


/**
 * Test DeviceSelect on iterator type
 */
template <
    Backend         BACKEND,
    typename        T>
void TestIterator(
    int             num_items)
{
    // Use a counting iterator as the input
    CountingInputIterator<T, int> h_in(0);

    // Allocate host arrays
    T*  h_reference = new T[num_items];

    // Initialize problem and solution
    int num_selected = Solve(h_in, h_reference, num_items);

    printf("\nIterator %s cub::DeviceSelect::Unique %d items, %d selected (avg run length %.3f), %s %d-byte elements\n",
        (BACKEND == CDP) ? "CDP CUB" : (BACKEND == THRUST) ? "Thrust" : "CUB",
        num_items, num_selected, float(num_items) / num_selected,
        typeid(T).name(),
        (int) sizeof(T));
    fflush(stdout);

    // Run Test
    Test<BACKEND>(h_in, h_reference, num_selected, num_items);

    // Cleanup
    if (h_reference) delete[] h_reference;
}


/**
 * Test different gen modes
 */
template <
    Backend         BACKEND,
    typename        T>
void Test(
    int             num_items)
{
    for (int max_segment = 1; ((max_segment > 0) && (max_segment < num_items)); max_segment *= 11)
    {
        TestPointer<BACKEND, T>(num_items, 0, max_segment);
        TestPointer<BACKEND, T>(num_items, 2, max_segment);
        TestPointer<BACKEND, T>(num_items, 7, max_segment);
    }
}


/**
 * Test different dispatch
 */
template <
    typename        T>
void TestOp(
    int             num_items)
{
    Test<CUB, T>(num_items);
#ifdef CUB_CDP
    Test<CDP, T>(num_items);
#endif
}


/**
 * Test different input sizes
 */
template <typename T>
void Test(
    int             num_items)
{
    if (num_items < 0)
    {
        TestOp<T>(0);
        TestOp<T>(1);
        TestOp<T>(100);
        TestOp<T>(10000);
        TestOp<T>(1000000);
    }
    else
    {
        TestOp<T>(num_items);
    }
}



//---------------------------------------------------------------------
// Main
//---------------------------------------------------------------------

/**
 * Main
 */
int main(int argc, char** argv)
{
    int num_items           = -1;
    int entropy_reduction   = 0;
    int maxseg              = 1000;

    // Initialize command line
    CommandLineArgs args(argc, argv);
    g_verbose = args.CheckCmdLineFlag("v");
    args.GetCmdLineArgument("n", num_items);
    args.GetCmdLineArgument("i", g_timing_iterations);
    args.GetCmdLineArgument("repeat", g_repeat);
    args.GetCmdLineArgument("maxseg", maxseg);
    args.GetCmdLineArgument("entropy", entropy_reduction);

    // Print usage
    if (args.CheckCmdLineFlag("help"))
    {
        printf("%s "
            "[--n=<input items> "
            "[--i=<timing iterations> "
            "[--device=<device-id>] "
            "[--maxseg=<max segment length>]"
            "[--entropy=<segment length bit entropy reduction rounds>]"
            "[--repeat=<repetitions of entire test suite>]"
            "[--v] "
            "[--cdp]"
            "\n", argv[0]);
        exit(0);
    }

    // Initialize device
    CubDebugExit(args.DeviceInit());
    g_device_giga_bandwidth = args.device_giga_bandwidth;
    printf("\n");

#ifdef QUICKER_TEST

    // Compile/run basic CUB test
    if (num_items < 0) num_items = 32000000;
    TestPointer<CUB, int>(         num_items,                                 entropy_reduction, maxseg);

#elif defined(QUICK_TEST)

    // Get device ordinal
    int device_ordinal;
    CubDebugExit(cudaGetDevice(&device_ordinal));

    // Get device SM version
    int sm_version;
    CubDebugExit(SmVersion(sm_version, device_ordinal));

    // Compile/run quick tests
    if (num_items < 0) num_items = 32000000;

    printf("-- Iterator ----------------------------\n");
    TestIterator<CUB, int>(        num_items);

    printf("----------------------------\n");
    TestPointer<CUB, char>(        num_items * ((sm_version <= 130) ? 1 : 4), entropy_reduction, maxseg);
    TestPointer<THRUST, char>(     num_items * ((sm_version <= 130) ? 1 : 4), entropy_reduction, maxseg);

    printf("----------------------------\n");
    TestPointer<CUB, short>(       num_items * ((sm_version <= 130) ? 1 : 2), entropy_reduction, maxseg);
    TestPointer<THRUST, short>(    num_items * ((sm_version <= 130) ? 1 : 2), entropy_reduction, maxseg);

    printf("----------------------------\n");
    TestPointer<CUB, int>(         num_items,                                 entropy_reduction, maxseg);
    TestPointer<THRUST, int>(      num_items,                                 entropy_reduction, maxseg);

    printf("----------------------------\n");
    TestPointer<CUB, long long>(   num_items / 2,                             entropy_reduction, maxseg);
    TestPointer<THRUST, long long>(num_items / 2,                             entropy_reduction, maxseg);

    printf("----------------------------\n");
    TestPointer<CUB, TestFoo>(     num_items / 4,                             entropy_reduction, maxseg);
    TestPointer<THRUST, TestFoo>(  num_items / 4,                             entropy_reduction, maxseg);

#else

    // Compile/run thorough tests
    for (int i = 0; i <= g_repeat; ++i)
    {
        // Test different input types
        Test<unsigned char>(num_items);
        Test<unsigned short>(num_items);
        Test<unsigned int>(num_items);
        Test<unsigned long long>(num_items);

        Test<uchar2>(num_items);
        Test<ushort2>(num_items);
        Test<uint2>(num_items);
        Test<ulonglong2>(num_items);

        Test<uchar4>(num_items);
        Test<ushort4>(num_items);
        Test<uint4>(num_items);
        Test<ulonglong4>(num_items);

        Test<TestFoo>(num_items);
        Test<TestBar>(num_items);
    }

#endif

    return 0;
}