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
|
/******************************************************************************
* Copyright (c) 2023, 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::DeviceCopy provides device-wide, parallel operations for copying data.
*/
#pragma once
#include <cub/config.cuh>
#if defined(_CCCL_IMPLICIT_SYSTEM_HEADER_GCC)
# pragma GCC system_header
#elif defined(_CCCL_IMPLICIT_SYSTEM_HEADER_CLANG)
# pragma clang system_header
#elif defined(_CCCL_IMPLICIT_SYSTEM_HEADER_MSVC)
# pragma system_header
#endif // no system header
#include <cub/device/dispatch/dispatch_batch_memcpy.cuh>
#include <thrust/system/cuda/detail/core/triple_chevron_launch.h>
#include <cstdint>
CUB_NAMESPACE_BEGIN
/**
* @brief cub::DeviceCopy provides device-wide, parallel operations for copying data.
* \ingroup SingleModule
*/
struct DeviceCopy
{
/**
* @brief Copies data from a batch of given source ranges to their corresponding destination
* ranges.
* @note If any input range aliases any output range the behavior is undefined. If
* any output range aliases another output range the behavior is undefined. Input
* ranges can alias one another.
*
* @par Snippet
* The code snippet below illustrates usage of DeviceCopy::Batched to perform a DeviceRunLength
* Decode operation.
* @par
* @code
* struct GetIteratorToRange
* {
* __host__ __device__ __forceinline__ auto operator()(uint32_t index)
* {
* return thrust::make_constant_iterator(d_data_in[index]);
* }
* int32_t *d_data_in;
* };
*
* struct GetPtrToRange
* {
* __host__ __device__ __forceinline__ auto operator()(uint32_t index)
* {
* return d_data_out + d_offsets[index];
* }
* int32_t *d_data_out;
* uint32_t *d_offsets;
* };
*
* struct GetRunLength
* {
* __host__ __device__ __forceinline__ uint32_t operator()(uint32_t index)
* {
* return d_offsets[index + 1] - d_offsets[index];
* }
* uint32_t *d_offsets;
* };
*
* uint32_t num_ranges = 5;
* int32_t *d_data_in; // e.g., [4, 2, 7, 3, 1]
* int32_t *d_data_out; // e.g., [0, ... ]
* uint32_t *d_offsets; // e.g., [0, 2, 5, 6, 9, 14]
*
* // Returns a constant iterator to the element of the i-th run
* thrust::counting_iterator<uint32_t> iota(0);
* auto iterators_in = thrust::make_transform_iterator(iota, GetIteratorToRange{d_data_in});
*
* // Returns the run length of the i-th run
* auto sizes = thrust::make_transform_iterator(iota, GetRunLength{d_offsets});
*
* // Returns pointers to the output range for each run
* auto ptrs_out = thrust::make_transform_iterator(iota, GetPtrToRange{d_data_out, d_offsets});
*
* // Determine temporary device storage requirements
* void *d_temp_storage = nullptr;
* size_t temp_storage_bytes = 0;
* cub::DeviceCopy::Batched(d_temp_storage, temp_storage_bytes, iterators_in, ptrs_out, sizes,
* num_ranges);
*
* // Allocate temporary storage
* cudaMalloc(&d_temp_storage, temp_storage_bytes);
*
* // Run batched copy algorithm (used to perform runlength decoding)
* cub::DeviceCopy::Batched(d_temp_storage, temp_storage_bytes, iterators_in, ptrs_out, sizes,
* num_ranges);
*
* // d_data_out <-- [4, 4, 2, 2, 2, 7, 3, 3, 3, 1, 1, 1, 1, 1]
* @endcode
* @tparam InputIt <b>[inferred]</b> Device-accessible random-access input iterator type
* providing the iterators to the source ranges
* @tparam OutputIt <b>[inferred]</b> Device-accessible random-access input iterator type
* providing the iterators to the destination ranges
* @tparam SizeIteratorT <b>[inferred]</b> Device-accessible random-access input iterator
* type providing the number of items to be copied for each pair of ranges
* @param d_temp_storage [in] Device-accessible allocation of temporary storage. When NULL, the
* required allocation size is written to \p temp_storage_bytes and no work is done.
* @param temp_storage_bytes [in,out] Reference to size in bytes of \p d_temp_storage allocation
* @param input_it [in] Device-accessible iterator providing the iterators to the source
* ranges
* @param output_it [in] Device-accessible iterator providing the iterators to the
* destination ranges
* @param sizes [in] Device-accessible iterator providing the number of elements to be copied
* for each pair of ranges
* @param num_ranges [in] The total number of range pairs
* @param stream [in] <b>[optional]</b> CUDA stream to launch kernels within. Default is
* stream<sub>0</sub>.
*/
template <typename InputIt, typename OutputIt, typename SizeIteratorT>
CUB_RUNTIME_FUNCTION static cudaError_t Batched(void *d_temp_storage,
size_t &temp_storage_bytes,
InputIt input_it,
OutputIt output_it,
SizeIteratorT sizes,
uint32_t num_ranges,
cudaStream_t stream = 0)
{
// Integer type large enough to hold any offset in [0, num_ranges)
using RangeOffsetT = uint32_t;
// Integer type large enough to hold any offset in [0, num_thread_blocks_launched), where a safe
// uppper bound on num_thread_blocks_launched can be assumed to be given by
// IDIV_CEIL(num_ranges, 64)
using BlockOffsetT = uint32_t;
return detail::DispatchBatchMemcpy<InputIt,
OutputIt,
SizeIteratorT,
RangeOffsetT,
BlockOffsetT,
detail::DeviceBatchMemcpyPolicy<RangeOffsetT, BlockOffsetT>,
false>::Dispatch(d_temp_storage,
temp_storage_bytes,
input_it,
output_it,
sizes,
num_ranges,
stream);
}
};
CUB_NAMESPACE_END
|
