/home/runner/work/cccl/cccl/cub/cub/device/device_radix_sort.cuh
File members: /home/runner/work/cccl/cccl/cub/cub/device/device_radix_sort.cuh
/******************************************************************************
* Copyright (c) 2011, Duane Merrill. All rights reserved.
* Copyright (c) 2011-2023, 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.
*
******************************************************************************/
#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/detail/choose_offset.cuh>
#include <cub/detail/nvtx.cuh>
#include <cub/device/dispatch/dispatch_radix_sort.cuh>
#include <cub/util_deprecated.cuh>
#include <cuda/std/type_traits>
CUB_NAMESPACE_BEGIN
struct DeviceRadixSort
{
private:
template <bool IsDescending, typename KeyT, typename ValueT, typename NumItemsT, typename DecomposerT>
CUB_RUNTIME_FUNCTION static cudaError_t custom_radix_sort(
::cuda::std::false_type,
void* d_temp_storage,
size_t& temp_storage_bytes,
bool is_overwrite_okay,
DoubleBuffer<KeyT>& d_keys,
DoubleBuffer<ValueT>& d_values,
NumItemsT num_items,
DecomposerT decomposer,
int begin_bit,
int end_bit,
cudaStream_t stream);
template <bool IsDescending, typename KeyT, typename ValueT, typename OffsetT, typename DecomposerT>
CUB_RUNTIME_FUNCTION static cudaError_t custom_radix_sort(
::cuda::std::true_type,
void* d_temp_storage,
size_t& temp_storage_bytes,
bool is_overwrite_okay,
DoubleBuffer<KeyT>& d_keys,
DoubleBuffer<ValueT>& d_values,
OffsetT num_items,
DecomposerT decomposer,
int begin_bit,
int end_bit,
cudaStream_t stream)
{
return DispatchRadixSort<IsDescending, KeyT, ValueT, OffsetT, DeviceRadixSortPolicy<KeyT, ValueT, OffsetT>, DecomposerT>::
Dispatch(
d_temp_storage,
temp_storage_bytes,
d_keys,
d_values,
static_cast<OffsetT>(num_items),
begin_bit,
end_bit,
is_overwrite_okay,
stream,
decomposer);
}
template <bool IsDescending, typename KeyT, typename ValueT, typename NumItemsT, typename DecomposerT>
CUB_RUNTIME_FUNCTION static cudaError_t custom_radix_sort(
::cuda::std::false_type,
void* d_temp_storage,
size_t& temp_storage_bytes,
bool is_overwrite_okay,
DoubleBuffer<KeyT>& d_keys,
DoubleBuffer<ValueT>& d_values,
NumItemsT num_items,
DecomposerT decomposer,
cudaStream_t stream);
template <bool IsDescending, typename KeyT, typename ValueT, typename OffsetT, typename DecomposerT>
CUB_RUNTIME_FUNCTION static cudaError_t custom_radix_sort(
::cuda::std::true_type,
void* d_temp_storage,
size_t& temp_storage_bytes,
bool is_overwrite_okay,
DoubleBuffer<KeyT>& d_keys,
DoubleBuffer<ValueT>& d_values,
OffsetT num_items,
DecomposerT decomposer,
cudaStream_t stream)
{
constexpr int begin_bit = 0;
const int end_bit = detail::radix::traits_t<KeyT>::default_end_bit(decomposer);
return DeviceRadixSort::custom_radix_sort<IsDescending>(
::cuda::std::true_type{},
d_temp_storage,
temp_storage_bytes,
is_overwrite_okay,
d_keys,
d_values,
num_items,
decomposer,
begin_bit,
end_bit,
stream);
}
// Name reported for NVTX ranges
_CCCL_HOST_DEVICE static constexpr auto GetName() -> const char*
{
return "cub::DeviceRadixSort";
}
public:
template <typename KeyT, typename ValueT, typename NumItemsT>
CUB_RUNTIME_FUNCTION static cudaError_t SortPairs(
void* d_temp_storage,
size_t& temp_storage_bytes,
const KeyT* d_keys_in,
KeyT* d_keys_out,
const ValueT* d_values_in,
ValueT* d_values_out,
NumItemsT num_items,
int begin_bit = 0,
int end_bit = sizeof(KeyT) * 8,
cudaStream_t stream = 0)
{
CUB_DETAIL_NVTX_RANGE_SCOPE_IF(d_temp_storage, GetName());
// Unsigned integer type for global offsets.
using OffsetT = detail::choose_offset_t<NumItemsT>;
// TODO API that doesn't accept decomposer should also contain a static
// assert that the key type is fundamental.
// We cast away const-ness, but will *not* write to these arrays.
// `DispatchRadixSort::Dispatch` will allocate temporary storage and
// create a new double-buffer internally when the `is_overwrite_ok` flag
// is not set.
constexpr bool is_overwrite_okay = false;
DoubleBuffer<KeyT> d_keys(const_cast<KeyT*>(d_keys_in), d_keys_out);
DoubleBuffer<ValueT> d_values(const_cast<ValueT*>(d_values_in), d_values_out);
return DispatchRadixSort<false, KeyT, ValueT, OffsetT>::Dispatch(
d_temp_storage,
temp_storage_bytes,
d_keys,
d_values,
static_cast<OffsetT>(num_items),
begin_bit,
end_bit,
is_overwrite_okay,
stream);
}
#ifndef DOXYGEN_SHOULD_SKIP_THIS // Do not document
template <typename KeyT, typename ValueT, typename NumItemsT>
CUB_DETAIL_RUNTIME_DEBUG_SYNC_IS_NOT_SUPPORTED CUB_RUNTIME_FUNCTION static cudaError_t SortPairs(
void* d_temp_storage,
size_t& temp_storage_bytes,
const KeyT* d_keys_in,
KeyT* d_keys_out,
const ValueT* d_values_in,
ValueT* d_values_out,
NumItemsT num_items,
int begin_bit,
int end_bit,
cudaStream_t stream,
bool debug_synchronous)
{
CUB_DETAIL_RUNTIME_DEBUG_SYNC_USAGE_LOG
return SortPairs<KeyT, ValueT, NumItemsT>(
d_temp_storage,
temp_storage_bytes,
d_keys_in,
d_keys_out,
d_values_in,
d_values_out,
num_items,
begin_bit,
end_bit,
stream);
}
#endif
template <typename KeyT, typename ValueT, typename NumItemsT, typename DecomposerT>
CUB_RUNTIME_FUNCTION static //
typename ::cuda::std::enable_if< //
!::cuda::std::is_convertible<DecomposerT, int>::value, //
cudaError_t>::type
SortPairs(void* d_temp_storage,
size_t& temp_storage_bytes,
const KeyT* d_keys_in,
KeyT* d_keys_out,
const ValueT* d_values_in,
ValueT* d_values_out,
NumItemsT num_items,
DecomposerT decomposer,
int begin_bit,
int end_bit,
cudaStream_t stream = 0)
{
CUB_DETAIL_NVTX_RANGE_SCOPE_IF(d_temp_storage, GetName());
// unsigned integer type for global offsets
using offset_t = detail::choose_offset_t<NumItemsT>;
using decomposer_check_t = detail::radix::decomposer_check_t<KeyT, DecomposerT>;
static_assert(decomposer_check_t::value,
"DecomposerT must be a callable object returning a tuple of references to "
"arithmetic types");
// We cast away const-ness, but will *not* write to these arrays.
// `DispatchRadixSort::Dispatch` will allocate temporary storage and
// create a new double-buffer internally when the `is_overwrite_ok` flag
// is not set.
constexpr bool is_overwrite_okay = false;
constexpr bool is_descending = false;
DoubleBuffer<KeyT> d_keys(const_cast<KeyT*>(d_keys_in), d_keys_out);
DoubleBuffer<ValueT> d_values(const_cast<ValueT*>(d_values_in), d_values_out);
return DeviceRadixSort::custom_radix_sort<is_descending>(
decomposer_check_t{},
d_temp_storage,
temp_storage_bytes,
is_overwrite_okay,
d_keys,
d_values,
static_cast<offset_t>(num_items),
decomposer,
begin_bit,
end_bit,
stream);
}
template <typename KeyT, typename ValueT, typename NumItemsT, typename DecomposerT>
CUB_RUNTIME_FUNCTION static //
typename ::cuda::std::enable_if< //
!::cuda::std::is_convertible<DecomposerT, int>::value, //
cudaError_t>::type
SortPairs(void* d_temp_storage,
size_t& temp_storage_bytes,
const KeyT* d_keys_in,
KeyT* d_keys_out,
const ValueT* d_values_in,
ValueT* d_values_out,
NumItemsT num_items,
DecomposerT decomposer,
cudaStream_t stream = 0)
{
CUB_DETAIL_NVTX_RANGE_SCOPE_IF(d_temp_storage, GetName());
// unsigned integer type for global offsets
using offset_t = detail::choose_offset_t<NumItemsT>;
using decomposer_check_t = detail::radix::decomposer_check_t<KeyT, DecomposerT>;
static_assert(decomposer_check_t::value,
"DecomposerT must be a callable object returning a tuple of references to "
"arithmetic types");
// We cast away const-ness, but will *not* write to these arrays.
// `DispatchRadixSort::Dispatch` will allocate temporary storage and
// create a new double-buffer internally when the `is_overwrite_ok` flag
// is not set.
constexpr bool is_overwrite_okay = false;
constexpr bool is_descending = false;
DoubleBuffer<KeyT> d_keys(const_cast<KeyT*>(d_keys_in), d_keys_out);
DoubleBuffer<ValueT> d_values(const_cast<ValueT*>(d_values_in), d_values_out);
return DeviceRadixSort::custom_radix_sort<is_descending>(
decomposer_check_t{},
d_temp_storage,
temp_storage_bytes,
is_overwrite_okay,
d_keys,
d_values,
static_cast<offset_t>(num_items),
decomposer,
stream);
}
template <typename KeyT, typename ValueT, typename NumItemsT>
CUB_RUNTIME_FUNCTION static cudaError_t SortPairs(
void* d_temp_storage,
size_t& temp_storage_bytes,
DoubleBuffer<KeyT>& d_keys,
DoubleBuffer<ValueT>& d_values,
NumItemsT num_items,
int begin_bit = 0,
int end_bit = sizeof(KeyT) * 8,
cudaStream_t stream = 0)
{
CUB_DETAIL_NVTX_RANGE_SCOPE_IF(d_temp_storage, GetName());
// Unsigned integer type for global offsets.
using OffsetT = detail::choose_offset_t<NumItemsT>;
constexpr bool is_overwrite_okay = true;
return DispatchRadixSort<false, KeyT, ValueT, OffsetT>::Dispatch(
d_temp_storage, temp_storage_bytes, d_keys, d_values, num_items, begin_bit, end_bit, is_overwrite_okay, stream);
}
#ifndef DOXYGEN_SHOULD_SKIP_THIS // Do not document
template <typename KeyT, typename ValueT, typename NumItemsT>
CUB_DETAIL_RUNTIME_DEBUG_SYNC_IS_NOT_SUPPORTED CUB_RUNTIME_FUNCTION static cudaError_t SortPairs(
void* d_temp_storage,
size_t& temp_storage_bytes,
DoubleBuffer<KeyT>& d_keys,
DoubleBuffer<ValueT>& d_values,
NumItemsT num_items,
int begin_bit,
int end_bit,
cudaStream_t stream,
bool debug_synchronous)
{
CUB_DETAIL_RUNTIME_DEBUG_SYNC_USAGE_LOG
return SortPairs<KeyT, ValueT, NumItemsT>(
d_temp_storage, temp_storage_bytes, d_keys, d_values, num_items, begin_bit, end_bit, stream);
}
#endif
template <typename KeyT, typename ValueT, typename NumItemsT, typename DecomposerT>
CUB_RUNTIME_FUNCTION static //
typename ::cuda::std::enable_if< //
!::cuda::std::is_convertible<DecomposerT, int>::value, //
cudaError_t>::type
SortPairs(void* d_temp_storage,
size_t& temp_storage_bytes,
DoubleBuffer<KeyT>& d_keys,
DoubleBuffer<ValueT>& d_values,
NumItemsT num_items,
DecomposerT decomposer,
cudaStream_t stream = 0)
{
CUB_DETAIL_NVTX_RANGE_SCOPE_IF(d_temp_storage, GetName());
// unsigned integer type for global offsets
using offset_t = detail::choose_offset_t<NumItemsT>;
using decomposer_check_t = detail::radix::decomposer_check_t<KeyT, DecomposerT>;
static_assert(decomposer_check_t::value,
"DecomposerT must be a callable object returning a tuple of references to "
"arithmetic types");
constexpr bool is_overwrite_okay = true;
constexpr bool is_descending = false;
return DeviceRadixSort::custom_radix_sort<is_descending>(
decomposer_check_t{},
d_temp_storage,
temp_storage_bytes,
is_overwrite_okay,
d_keys,
d_values,
static_cast<offset_t>(num_items),
decomposer,
stream);
}
template <typename KeyT, typename ValueT, typename NumItemsT, typename DecomposerT>
CUB_RUNTIME_FUNCTION static //
typename ::cuda::std::enable_if< //
!::cuda::std::is_convertible<DecomposerT, int>::value, //
cudaError_t>::type
SortPairs(void* d_temp_storage,
size_t& temp_storage_bytes,
DoubleBuffer<KeyT>& d_keys,
DoubleBuffer<ValueT>& d_values,
NumItemsT num_items,
DecomposerT decomposer,
int begin_bit,
int end_bit,
cudaStream_t stream = 0)
{
CUB_DETAIL_NVTX_RANGE_SCOPE_IF(d_temp_storage, GetName());
// unsigned integer type for global offsets
using offset_t = detail::choose_offset_t<NumItemsT>;
using decomposer_check_t = detail::radix::decomposer_check_t<KeyT, DecomposerT>;
static_assert(decomposer_check_t::value,
"DecomposerT must be a callable object returning a tuple of references to "
"arithmetic types");
constexpr bool is_overwrite_okay = true;
constexpr bool is_descending = false;
return DeviceRadixSort::custom_radix_sort<is_descending>(
decomposer_check_t{},
d_temp_storage,
temp_storage_bytes,
is_overwrite_okay,
d_keys,
d_values,
static_cast<offset_t>(num_items),
decomposer,
begin_bit,
end_bit,
stream);
}
template <typename KeyT, typename ValueT, typename NumItemsT>
CUB_RUNTIME_FUNCTION static cudaError_t SortPairsDescending(
void* d_temp_storage,
size_t& temp_storage_bytes,
const KeyT* d_keys_in,
KeyT* d_keys_out,
const ValueT* d_values_in,
ValueT* d_values_out,
NumItemsT num_items,
int begin_bit = 0,
int end_bit = sizeof(KeyT) * 8,
cudaStream_t stream = 0)
{
CUB_DETAIL_NVTX_RANGE_SCOPE_IF(d_temp_storage, GetName());
// Unsigned integer type for global offsets.
using OffsetT = detail::choose_offset_t<NumItemsT>;
// We cast away const-ness, but will *not* write to these arrays.
// `DispatchRadixSort::Dispatch` will allocate temporary storage and
// create a new double-buffer internally when the `is_overwrite_ok` flag
// is not set.
constexpr bool is_overwrite_okay = false;
DoubleBuffer<KeyT> d_keys(const_cast<KeyT*>(d_keys_in), d_keys_out);
DoubleBuffer<ValueT> d_values(const_cast<ValueT*>(d_values_in), d_values_out);
return DispatchRadixSort<true, KeyT, ValueT, OffsetT>::Dispatch(
d_temp_storage, temp_storage_bytes, d_keys, d_values, num_items, begin_bit, end_bit, is_overwrite_okay, stream);
}
#ifndef DOXYGEN_SHOULD_SKIP_THIS // Do not document
template <typename KeyT, typename ValueT, typename NumItemsT>
CUB_DETAIL_RUNTIME_DEBUG_SYNC_IS_NOT_SUPPORTED CUB_RUNTIME_FUNCTION static cudaError_t SortPairsDescending(
void* d_temp_storage,
size_t& temp_storage_bytes,
const KeyT* d_keys_in,
KeyT* d_keys_out,
const ValueT* d_values_in,
ValueT* d_values_out,
NumItemsT num_items,
int begin_bit,
int end_bit,
cudaStream_t stream,
bool debug_synchronous)
{
CUB_DETAIL_RUNTIME_DEBUG_SYNC_USAGE_LOG
return SortPairsDescending<KeyT, ValueT, NumItemsT>(
d_temp_storage,
temp_storage_bytes,
d_keys_in,
d_keys_out,
d_values_in,
d_values_out,
num_items,
begin_bit,
end_bit,
stream);
}
#endif
template <typename KeyT, typename ValueT, typename NumItemsT, typename DecomposerT>
CUB_RUNTIME_FUNCTION static //
typename ::cuda::std::enable_if< //
!::cuda::std::is_convertible<DecomposerT, int>::value, //
cudaError_t>::type
SortPairsDescending(
void* d_temp_storage,
size_t& temp_storage_bytes,
const KeyT* d_keys_in,
KeyT* d_keys_out,
const ValueT* d_values_in,
ValueT* d_values_out,
NumItemsT num_items,
DecomposerT decomposer,
int begin_bit,
int end_bit,
cudaStream_t stream = 0)
{
CUB_DETAIL_NVTX_RANGE_SCOPE_IF(d_temp_storage, GetName());
// unsigned integer type for global offsets
using offset_t = detail::choose_offset_t<NumItemsT>;
using decomposer_check_t = detail::radix::decomposer_check_t<KeyT, DecomposerT>;
static_assert(decomposer_check_t::value,
"DecomposerT must be a callable object returning a tuple of references to "
"arithmetic types");
// We cast away const-ness, but will *not* write to these arrays.
// `DispatchRadixSort::Dispatch` will allocate temporary storage and
// create a new double-buffer internally when the `is_overwrite_ok` flag
// is not set.
constexpr bool is_overwrite_okay = false;
constexpr bool is_descending = true;
DoubleBuffer<KeyT> d_keys(const_cast<KeyT*>(d_keys_in), d_keys_out);
DoubleBuffer<ValueT> d_values(const_cast<ValueT*>(d_values_in), d_values_out);
return DeviceRadixSort::custom_radix_sort<is_descending>(
decomposer_check_t{},
d_temp_storage,
temp_storage_bytes,
is_overwrite_okay,
d_keys,
d_values,
static_cast<offset_t>(num_items),
decomposer,
begin_bit,
end_bit,
stream);
}
template <typename KeyT, typename ValueT, typename NumItemsT, typename DecomposerT>
CUB_RUNTIME_FUNCTION static //
typename ::cuda::std::enable_if< //
!::cuda::std::is_convertible<DecomposerT, int>::value, //
cudaError_t>::type
SortPairsDescending(
void* d_temp_storage,
size_t& temp_storage_bytes,
const KeyT* d_keys_in,
KeyT* d_keys_out,
const ValueT* d_values_in,
ValueT* d_values_out,
NumItemsT num_items,
DecomposerT decomposer,
cudaStream_t stream = 0)
{
CUB_DETAIL_NVTX_RANGE_SCOPE_IF(d_temp_storage, GetName());
// unsigned integer type for global offsets
using offset_t = detail::choose_offset_t<NumItemsT>;
using decomposer_check_t = detail::radix::decomposer_check_t<KeyT, DecomposerT>;
static_assert(decomposer_check_t::value,
"DecomposerT must be a callable object returning a tuple of references to "
"arithmetic types");
// We cast away const-ness, but will *not* write to these arrays.
// `DispatchRadixSort::Dispatch` will allocate temporary storage and
// create a new double-buffer internally when the `is_overwrite_ok` flag
// is not set.
constexpr bool is_overwrite_okay = false;
constexpr bool is_descending = true;
DoubleBuffer<KeyT> d_keys(const_cast<KeyT*>(d_keys_in), d_keys_out);
DoubleBuffer<ValueT> d_values(const_cast<ValueT*>(d_values_in), d_values_out);
return DeviceRadixSort::custom_radix_sort<is_descending>(
decomposer_check_t{},
d_temp_storage,
temp_storage_bytes,
is_overwrite_okay,
d_keys,
d_values,
static_cast<offset_t>(num_items),
decomposer,
stream);
}
template <typename KeyT, typename ValueT, typename NumItemsT>
CUB_RUNTIME_FUNCTION static cudaError_t SortPairsDescending(
void* d_temp_storage,
size_t& temp_storage_bytes,
DoubleBuffer<KeyT>& d_keys,
DoubleBuffer<ValueT>& d_values,
NumItemsT num_items,
int begin_bit = 0,
int end_bit = sizeof(KeyT) * 8,
cudaStream_t stream = 0)
{
CUB_DETAIL_NVTX_RANGE_SCOPE_IF(d_temp_storage, GetName());
// Unsigned integer type for global offsets.
using OffsetT = detail::choose_offset_t<NumItemsT>;
constexpr bool is_overwrite_okay = true;
return DispatchRadixSort<true, KeyT, ValueT, OffsetT>::Dispatch(
d_temp_storage, temp_storage_bytes, d_keys, d_values, num_items, begin_bit, end_bit, is_overwrite_okay, stream);
}
#ifndef DOXYGEN_SHOULD_SKIP_THIS // Do not document
template <typename KeyT, typename ValueT, typename NumItemsT>
CUB_DETAIL_RUNTIME_DEBUG_SYNC_IS_NOT_SUPPORTED CUB_RUNTIME_FUNCTION static cudaError_t SortPairsDescending(
void* d_temp_storage,
size_t& temp_storage_bytes,
DoubleBuffer<KeyT>& d_keys,
DoubleBuffer<ValueT>& d_values,
NumItemsT num_items,
int begin_bit,
int end_bit,
cudaStream_t stream,
bool debug_synchronous)
{
CUB_DETAIL_RUNTIME_DEBUG_SYNC_USAGE_LOG
return SortPairsDescending<KeyT, ValueT, NumItemsT>(
d_temp_storage, temp_storage_bytes, d_keys, d_values, num_items, begin_bit, end_bit, stream);
}
#endif
template <typename KeyT, typename ValueT, typename NumItemsT, typename DecomposerT>
CUB_RUNTIME_FUNCTION static //
typename ::cuda::std::enable_if< //
!::cuda::std::is_convertible<DecomposerT, int>::value, //
cudaError_t>::type
SortPairsDescending(
void* d_temp_storage,
size_t& temp_storage_bytes,
DoubleBuffer<KeyT>& d_keys,
DoubleBuffer<ValueT>& d_values,
NumItemsT num_items,
DecomposerT decomposer,
cudaStream_t stream = 0)
{
CUB_DETAIL_NVTX_RANGE_SCOPE_IF(d_temp_storage, GetName());
// unsigned integer type for global offsets
using offset_t = detail::choose_offset_t<NumItemsT>;
using decomposer_check_t = detail::radix::decomposer_check_t<KeyT, DecomposerT>;
static_assert(decomposer_check_t::value,
"DecomposerT must be a callable object returning a tuple of references to "
"arithmetic types");
constexpr bool is_overwrite_okay = true;
constexpr bool is_descending = true;
return DeviceRadixSort::custom_radix_sort<is_descending>(
decomposer_check_t{},
d_temp_storage,
temp_storage_bytes,
is_overwrite_okay,
d_keys,
d_values,
static_cast<offset_t>(num_items),
decomposer,
stream);
}
template <typename KeyT, typename ValueT, typename NumItemsT, typename DecomposerT>
CUB_RUNTIME_FUNCTION static //
typename ::cuda::std::enable_if< //
!::cuda::std::is_convertible<DecomposerT, int>::value, //
cudaError_t>::type
SortPairsDescending(
void* d_temp_storage,
size_t& temp_storage_bytes,
DoubleBuffer<KeyT>& d_keys,
DoubleBuffer<ValueT>& d_values,
NumItemsT num_items,
DecomposerT decomposer,
int begin_bit,
int end_bit,
cudaStream_t stream = 0)
{
CUB_DETAIL_NVTX_RANGE_SCOPE_IF(d_temp_storage, GetName());
// unsigned integer type for global offsets
using offset_t = detail::choose_offset_t<NumItemsT>;
using decomposer_check_t = detail::radix::decomposer_check_t<KeyT, DecomposerT>;
static_assert(decomposer_check_t::value,
"DecomposerT must be a callable object returning a tuple of references to "
"arithmetic types");
constexpr bool is_overwrite_okay = true;
constexpr bool is_descending = true;
return DeviceRadixSort::custom_radix_sort<is_descending>(
decomposer_check_t{},
d_temp_storage,
temp_storage_bytes,
is_overwrite_okay,
d_keys,
d_values,
static_cast<offset_t>(num_items),
decomposer,
begin_bit,
end_bit,
stream);
}
/******************************************************************/
template <typename KeyT, typename NumItemsT>
CUB_RUNTIME_FUNCTION static cudaError_t SortKeys(
void* d_temp_storage,
size_t& temp_storage_bytes,
const KeyT* d_keys_in,
KeyT* d_keys_out,
NumItemsT num_items,
int begin_bit = 0,
int end_bit = sizeof(KeyT) * 8,
cudaStream_t stream = 0)
{
CUB_DETAIL_NVTX_RANGE_SCOPE_IF(d_temp_storage, GetName());
// Unsigned integer type for global offsets.
using OffsetT = detail::choose_offset_t<NumItemsT>;
// We cast away const-ness, but will *not* write to these arrays.
// `DispatchRadixSort::Dispatch` will allocate temporary storage and
// create a new double-buffer internally when the `is_overwrite_ok` flag
// is not set.
constexpr bool is_overwrite_okay = false;
DoubleBuffer<KeyT> d_keys(const_cast<KeyT*>(d_keys_in), d_keys_out);
// Null value type
DoubleBuffer<NullType> d_values;
return DispatchRadixSort<false, KeyT, NullType, OffsetT>::Dispatch(
d_temp_storage,
temp_storage_bytes,
d_keys,
d_values,
static_cast<OffsetT>(num_items),
begin_bit,
end_bit,
is_overwrite_okay,
stream);
}
template <typename KeyT, typename NumItemsT, typename DecomposerT>
CUB_RUNTIME_FUNCTION static //
typename ::cuda::std::enable_if< //
!::cuda::std::is_convertible<DecomposerT, int>::value, //
cudaError_t>::type
SortKeys(void* d_temp_storage,
size_t& temp_storage_bytes,
const KeyT* d_keys_in,
KeyT* d_keys_out,
NumItemsT num_items,
DecomposerT decomposer,
int begin_bit,
int end_bit,
cudaStream_t stream = 0)
{
CUB_DETAIL_NVTX_RANGE_SCOPE_IF(d_temp_storage, GetName());
// unsigned integer type for global offsets
using offset_t = detail::choose_offset_t<NumItemsT>;
using decomposer_check_t = detail::radix::decomposer_check_t<KeyT, DecomposerT>;
static_assert(decomposer_check_t::value,
"DecomposerT must be a callable object returning a tuple of references to "
"arithmetic types");
// We cast away const-ness, but will *not* write to these arrays.
// `DispatchRadixSort::Dispatch` will allocate temporary storage and
// create a new double-buffer internally when the `is_overwrite_ok` flag
// is not set.
constexpr bool is_overwrite_okay = false;
constexpr bool is_descending = false;
DoubleBuffer<KeyT> d_keys(const_cast<KeyT*>(d_keys_in), d_keys_out);
DoubleBuffer<NullType> d_values;
return DeviceRadixSort::custom_radix_sort<is_descending>(
decomposer_check_t{},
d_temp_storage,
temp_storage_bytes,
is_overwrite_okay,
d_keys,
d_values,
static_cast<offset_t>(num_items),
decomposer,
begin_bit,
end_bit,
stream);
}
template <typename KeyT, typename NumItemsT, typename DecomposerT>
CUB_RUNTIME_FUNCTION static //
typename ::cuda::std::enable_if< //
!::cuda::std::is_convertible<DecomposerT, int>::value, //
cudaError_t>::type
SortKeys(void* d_temp_storage,
size_t& temp_storage_bytes,
const KeyT* d_keys_in,
KeyT* d_keys_out,
NumItemsT num_items,
DecomposerT decomposer,
cudaStream_t stream = 0)
{
CUB_DETAIL_NVTX_RANGE_SCOPE_IF(d_temp_storage, GetName());
// unsigned integer type for global offsets
using offset_t = detail::choose_offset_t<NumItemsT>;
using decomposer_check_t = detail::radix::decomposer_check_t<KeyT, DecomposerT>;
static_assert(decomposer_check_t::value,
"DecomposerT must be a callable object returning a tuple of references to "
"arithmetic types");
// We cast away const-ness, but will *not* write to these arrays.
// `DispatchRadixSort::Dispatch` will allocate temporary storage and
// create a new double-buffer internally when the `is_overwrite_ok` flag
// is not set.
constexpr bool is_overwrite_okay = false;
constexpr bool is_descending = false;
DoubleBuffer<KeyT> d_keys(const_cast<KeyT*>(d_keys_in), d_keys_out);
DoubleBuffer<NullType> d_values;
return DeviceRadixSort::custom_radix_sort<is_descending>(
decomposer_check_t{},
d_temp_storage,
temp_storage_bytes,
is_overwrite_okay,
d_keys,
d_values,
static_cast<offset_t>(num_items),
decomposer,
stream);
}
#ifndef DOXYGEN_SHOULD_SKIP_THIS // Do not document
template <typename KeyT, typename NumItemsT>
CUB_DETAIL_RUNTIME_DEBUG_SYNC_IS_NOT_SUPPORTED CUB_RUNTIME_FUNCTION static cudaError_t SortKeys(
void* d_temp_storage,
size_t& temp_storage_bytes,
const KeyT* d_keys_in,
KeyT* d_keys_out,
NumItemsT num_items,
int begin_bit,
int end_bit,
cudaStream_t stream,
bool debug_synchronous)
{
CUB_DETAIL_RUNTIME_DEBUG_SYNC_USAGE_LOG
return SortKeys<KeyT, NumItemsT>(
d_temp_storage, temp_storage_bytes, d_keys_in, d_keys_out, num_items, begin_bit, end_bit, stream);
}
#endif
template <typename KeyT, typename NumItemsT>
CUB_RUNTIME_FUNCTION static cudaError_t SortKeys(
void* d_temp_storage,
size_t& temp_storage_bytes,
DoubleBuffer<KeyT>& d_keys,
NumItemsT num_items,
int begin_bit = 0,
int end_bit = sizeof(KeyT) * 8,
cudaStream_t stream = 0)
{
CUB_DETAIL_NVTX_RANGE_SCOPE_IF(d_temp_storage, GetName());
// Unsigned integer type for global offsets.
using OffsetT = detail::choose_offset_t<NumItemsT>;
constexpr bool is_overwrite_okay = true;
// Null value type
DoubleBuffer<NullType> d_values;
return DispatchRadixSort<false, KeyT, NullType, OffsetT>::Dispatch(
d_temp_storage, temp_storage_bytes, d_keys, d_values, num_items, begin_bit, end_bit, is_overwrite_okay, stream);
}
#ifndef DOXYGEN_SHOULD_SKIP_THIS // Do not document
template <typename KeyT, typename NumItemsT>
CUB_DETAIL_RUNTIME_DEBUG_SYNC_IS_NOT_SUPPORTED CUB_RUNTIME_FUNCTION static cudaError_t SortKeys(
void* d_temp_storage,
size_t& temp_storage_bytes,
DoubleBuffer<KeyT>& d_keys,
NumItemsT num_items,
int begin_bit,
int end_bit,
cudaStream_t stream,
bool debug_synchronous)
{
CUB_DETAIL_RUNTIME_DEBUG_SYNC_USAGE_LOG
return SortKeys<KeyT, NumItemsT>(d_temp_storage, temp_storage_bytes, d_keys, num_items, begin_bit, end_bit, stream);
}
#endif
template <typename KeyT, typename NumItemsT, typename DecomposerT>
CUB_RUNTIME_FUNCTION static //
typename ::cuda::std::enable_if< //
!::cuda::std::is_convertible<DecomposerT, int>::value, //
cudaError_t>::type
SortKeys(void* d_temp_storage,
size_t& temp_storage_bytes,
DoubleBuffer<KeyT>& d_keys,
NumItemsT num_items,
DecomposerT decomposer,
cudaStream_t stream = 0)
{
CUB_DETAIL_NVTX_RANGE_SCOPE_IF(d_temp_storage, GetName());
// unsigned integer type for global offsets
using offset_t = detail::choose_offset_t<NumItemsT>;
using decomposer_check_t = detail::radix::decomposer_check_t<KeyT, DecomposerT>;
static_assert(decomposer_check_t::value,
"DecomposerT must be a callable object returning a tuple of references to "
"arithmetic types");
constexpr bool is_overwrite_okay = true;
constexpr bool is_descending = false;
DoubleBuffer<NullType> d_values;
return DeviceRadixSort::custom_radix_sort<is_descending>(
decomposer_check_t{},
d_temp_storage,
temp_storage_bytes,
is_overwrite_okay,
d_keys,
d_values,
static_cast<offset_t>(num_items),
decomposer,
stream);
}
template <typename KeyT, typename NumItemsT, typename DecomposerT>
CUB_RUNTIME_FUNCTION static //
typename ::cuda::std::enable_if< //
!::cuda::std::is_convertible<DecomposerT, int>::value, //
cudaError_t>::type
SortKeys(void* d_temp_storage,
size_t& temp_storage_bytes,
DoubleBuffer<KeyT>& d_keys,
NumItemsT num_items,
DecomposerT decomposer,
int begin_bit,
int end_bit,
cudaStream_t stream = 0)
{
CUB_DETAIL_NVTX_RANGE_SCOPE_IF(d_temp_storage, GetName());
// unsigned integer type for global offsets
using offset_t = detail::choose_offset_t<NumItemsT>;
using decomposer_check_t = detail::radix::decomposer_check_t<KeyT, DecomposerT>;
static_assert(decomposer_check_t::value,
"DecomposerT must be a callable object returning a tuple of references to "
"arithmetic types");
constexpr bool is_overwrite_okay = true;
constexpr bool is_descending = false;
DoubleBuffer<NullType> d_values;
return DeviceRadixSort::custom_radix_sort<is_descending>(
decomposer_check_t{},
d_temp_storage,
temp_storage_bytes,
is_overwrite_okay,
d_keys,
d_values,
static_cast<offset_t>(num_items),
decomposer,
begin_bit,
end_bit,
stream);
}
template <typename KeyT, typename NumItemsT>
CUB_RUNTIME_FUNCTION static cudaError_t SortKeysDescending(
void* d_temp_storage,
size_t& temp_storage_bytes,
const KeyT* d_keys_in,
KeyT* d_keys_out,
NumItemsT num_items,
int begin_bit = 0,
int end_bit = sizeof(KeyT) * 8,
cudaStream_t stream = 0)
{
CUB_DETAIL_NVTX_RANGE_SCOPE_IF(d_temp_storage, GetName());
// Unsigned integer type for global offsets.
using OffsetT = detail::choose_offset_t<NumItemsT>;
// We cast away const-ness, but will *not* write to these arrays.
// `DispatchRadixSort::Dispatch` will allocate temporary storage and
// create a new double-buffer internally when the `is_overwrite_ok` flag
// is not set.
constexpr bool is_overwrite_okay = false;
DoubleBuffer<KeyT> d_keys(const_cast<KeyT*>(d_keys_in), d_keys_out);
DoubleBuffer<NullType> d_values;
return DispatchRadixSort<true, KeyT, NullType, OffsetT>::Dispatch(
d_temp_storage, temp_storage_bytes, d_keys, d_values, num_items, begin_bit, end_bit, is_overwrite_okay, stream);
}
#ifndef DOXYGEN_SHOULD_SKIP_THIS // Do not document
template <typename KeyT, typename NumItemsT>
CUB_DETAIL_RUNTIME_DEBUG_SYNC_IS_NOT_SUPPORTED CUB_RUNTIME_FUNCTION static cudaError_t SortKeysDescending(
void* d_temp_storage,
size_t& temp_storage_bytes,
const KeyT* d_keys_in,
KeyT* d_keys_out,
NumItemsT num_items,
int begin_bit,
int end_bit,
cudaStream_t stream,
bool debug_synchronous)
{
CUB_DETAIL_RUNTIME_DEBUG_SYNC_USAGE_LOG
return SortKeysDescending<KeyT, NumItemsT>(
d_temp_storage, temp_storage_bytes, d_keys_in, d_keys_out, num_items, begin_bit, end_bit, stream);
}
#endif
template <typename KeyT, typename NumItemsT, typename DecomposerT>
CUB_RUNTIME_FUNCTION static //
typename ::cuda::std::enable_if< //
!::cuda::std::is_convertible<DecomposerT, int>::value, //
cudaError_t>::type
SortKeysDescending(
void* d_temp_storage,
size_t& temp_storage_bytes,
const KeyT* d_keys_in,
KeyT* d_keys_out,
NumItemsT num_items,
DecomposerT decomposer,
int begin_bit,
int end_bit,
cudaStream_t stream = 0)
{
CUB_DETAIL_NVTX_RANGE_SCOPE_IF(d_temp_storage, GetName());
// unsigned integer type for global offsets
using offset_t = detail::choose_offset_t<NumItemsT>;
using decomposer_check_t = detail::radix::decomposer_check_t<KeyT, DecomposerT>;
static_assert(decomposer_check_t::value,
"DecomposerT must be a callable object returning a tuple of references to "
"arithmetic types");
// We cast away const-ness, but will *not* write to these arrays.
// `DispatchRadixSort::Dispatch` will allocate temporary storage and
// create a new double-buffer internally when the `is_overwrite_ok` flag
// is not set.
constexpr bool is_overwrite_okay = false;
constexpr bool is_descending = true;
DoubleBuffer<KeyT> d_keys(const_cast<KeyT*>(d_keys_in), d_keys_out);
DoubleBuffer<NullType> d_values;
return DeviceRadixSort::custom_radix_sort<is_descending>(
decomposer_check_t{},
d_temp_storage,
temp_storage_bytes,
is_overwrite_okay,
d_keys,
d_values,
static_cast<offset_t>(num_items),
decomposer,
begin_bit,
end_bit,
stream);
}
template <typename KeyT, typename NumItemsT, typename DecomposerT>
CUB_RUNTIME_FUNCTION static //
typename ::cuda::std::enable_if< //
!::cuda::std::is_convertible<DecomposerT, int>::value, //
cudaError_t>::type
SortKeysDescending(
void* d_temp_storage,
size_t& temp_storage_bytes,
const KeyT* d_keys_in,
KeyT* d_keys_out,
NumItemsT num_items,
DecomposerT decomposer,
cudaStream_t stream = 0)
{
CUB_DETAIL_NVTX_RANGE_SCOPE_IF(d_temp_storage, GetName());
// unsigned integer type for global offsets
using offset_t = detail::choose_offset_t<NumItemsT>;
using decomposer_check_t = detail::radix::decomposer_check_t<KeyT, DecomposerT>;
static_assert(decomposer_check_t::value,
"DecomposerT must be a callable object returning a tuple of references to "
"arithmetic types");
// We cast away const-ness, but will *not* write to these arrays.
// `DispatchRadixSort::Dispatch` will allocate temporary storage and
// create a new double-buffer internally when the `is_overwrite_ok` flag
// is not set.
constexpr bool is_overwrite_okay = false;
constexpr bool is_descending = true;
DoubleBuffer<KeyT> d_keys(const_cast<KeyT*>(d_keys_in), d_keys_out);
DoubleBuffer<NullType> d_values;
return DeviceRadixSort::custom_radix_sort<is_descending>(
decomposer_check_t{},
d_temp_storage,
temp_storage_bytes,
is_overwrite_okay,
d_keys,
d_values,
static_cast<offset_t>(num_items),
decomposer,
stream);
}
template <typename KeyT, typename NumItemsT>
CUB_RUNTIME_FUNCTION static cudaError_t SortKeysDescending(
void* d_temp_storage,
size_t& temp_storage_bytes,
DoubleBuffer<KeyT>& d_keys,
NumItemsT num_items,
int begin_bit = 0,
int end_bit = sizeof(KeyT) * 8,
cudaStream_t stream = 0)
{
CUB_DETAIL_NVTX_RANGE_SCOPE_IF(d_temp_storage, GetName());
// Unsigned integer type for global offsets.
using OffsetT = detail::choose_offset_t<NumItemsT>;
constexpr bool is_overwrite_okay = true;
// Null value type
DoubleBuffer<NullType> d_values;
return DispatchRadixSort<true, KeyT, NullType, OffsetT>::Dispatch(
d_temp_storage, temp_storage_bytes, d_keys, d_values, num_items, begin_bit, end_bit, is_overwrite_okay, stream);
}
#ifndef DOXYGEN_SHOULD_SKIP_THIS // Do not document
template <typename KeyT, typename NumItemsT>
CUB_DETAIL_RUNTIME_DEBUG_SYNC_IS_NOT_SUPPORTED CUB_RUNTIME_FUNCTION static cudaError_t SortKeysDescending(
void* d_temp_storage,
size_t& temp_storage_bytes,
DoubleBuffer<KeyT>& d_keys,
NumItemsT num_items,
int begin_bit,
int end_bit,
cudaStream_t stream,
bool debug_synchronous)
{
CUB_DETAIL_RUNTIME_DEBUG_SYNC_USAGE_LOG
return SortKeysDescending<KeyT, NumItemsT>(
d_temp_storage, temp_storage_bytes, d_keys, num_items, begin_bit, end_bit, stream);
}
#endif
template <typename KeyT, typename NumItemsT, typename DecomposerT>
CUB_RUNTIME_FUNCTION static //
typename ::cuda::std::enable_if< //
!::cuda::std::is_convertible<DecomposerT, int>::value, //
cudaError_t>::type
SortKeysDescending(
void* d_temp_storage,
size_t& temp_storage_bytes,
DoubleBuffer<KeyT>& d_keys,
NumItemsT num_items,
DecomposerT decomposer,
cudaStream_t stream = 0)
{
CUB_DETAIL_NVTX_RANGE_SCOPE_IF(d_temp_storage, GetName());
// unsigned integer type for global offsets
using offset_t = detail::choose_offset_t<NumItemsT>;
using decomposer_check_t = detail::radix::decomposer_check_t<KeyT, DecomposerT>;
static_assert(decomposer_check_t::value,
"DecomposerT must be a callable object returning a tuple of references to "
"arithmetic types");
constexpr bool is_overwrite_okay = true;
constexpr bool is_descending = true;
DoubleBuffer<NullType> d_values;
return DeviceRadixSort::custom_radix_sort<is_descending>(
decomposer_check_t{},
d_temp_storage,
temp_storage_bytes,
is_overwrite_okay,
d_keys,
d_values,
static_cast<offset_t>(num_items),
decomposer,
stream);
}
template <typename KeyT, typename NumItemsT, typename DecomposerT>
CUB_RUNTIME_FUNCTION static //
typename ::cuda::std::enable_if< //
!::cuda::std::is_convertible<DecomposerT, int>::value, //
cudaError_t>::type
SortKeysDescending(
void* d_temp_storage,
size_t& temp_storage_bytes,
DoubleBuffer<KeyT>& d_keys,
NumItemsT num_items,
DecomposerT decomposer,
int begin_bit,
int end_bit,
cudaStream_t stream = 0)
{
CUB_DETAIL_NVTX_RANGE_SCOPE_IF(d_temp_storage, GetName());
// unsigned integer type for global offsets
using offset_t = detail::choose_offset_t<NumItemsT>;
using decomposer_check_t = detail::radix::decomposer_check_t<KeyT, DecomposerT>;
static_assert(decomposer_check_t::value,
"DecomposerT must be a callable object returning a tuple of references to "
"arithmetic types");
constexpr bool is_overwrite_okay = true;
constexpr bool is_descending = true;
DoubleBuffer<NullType> d_values;
return DeviceRadixSort::custom_radix_sort<is_descending>(
decomposer_check_t{},
d_temp_storage,
temp_storage_bytes,
is_overwrite_okay,
d_keys,
d_values,
static_cast<offset_t>(num_items),
decomposer,
begin_bit,
end_bit,
stream);
}
};
CUB_NAMESPACE_END