cub/block/block_discontinuity.cuh
File members: cub/block/block_discontinuity.cuh
/******************************************************************************
* 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.
*
******************************************************************************/
#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/util_ptx.cuh>
#include <cub/util_type.cuh>
CUB_NAMESPACE_BEGIN
template <typename T, int BLOCK_DIM_X, int BLOCK_DIM_Y = 1, int BLOCK_DIM_Z = 1, int LEGACY_PTX_ARCH = 0>
class BlockDiscontinuity
{
private:
enum
{
BLOCK_THREADS = BLOCK_DIM_X * BLOCK_DIM_Y * BLOCK_DIM_Z,
};
struct _TempStorage
{
T first_items[BLOCK_THREADS];
T last_items[BLOCK_THREADS];
};
_CCCL_DEVICE _CCCL_FORCEINLINE _TempStorage& PrivateStorage()
{
__shared__ _TempStorage private_storage;
return private_storage;
}
template <typename FlagOp, bool HAS_PARAM = BinaryOpHasIdxParam<T, FlagOp>::value>
struct ApplyOp
{
// Apply flag operator
static _CCCL_DEVICE _CCCL_FORCEINLINE bool FlagT(FlagOp flag_op, const T& a, const T& b, int idx)
{
return flag_op(a, b, idx);
}
};
template <typename FlagOp>
struct ApplyOp<FlagOp, false>
{
// Apply flag operator
static _CCCL_DEVICE _CCCL_FORCEINLINE bool FlagT(FlagOp flag_op, const T& a, const T& b, int /*idx*/)
{
return flag_op(a, b);
}
};
struct Iterate
{
template <int ITEMS_PER_THREAD, typename FlagT, typename FlagOp>
static _CCCL_DEVICE _CCCL_FORCEINLINE void FlagHeads(
int linear_tid,
FlagT (&flags)[ITEMS_PER_THREAD],
T (&input)[ITEMS_PER_THREAD],
T (&preds)[ITEMS_PER_THREAD],
FlagOp flag_op)
{
#pragma unroll
for (int i = 1; i < ITEMS_PER_THREAD; ++i)
{
preds[i] = input[i - 1];
flags[i] = ApplyOp<FlagOp>::FlagT(flag_op, preds[i], input[i], (linear_tid * ITEMS_PER_THREAD) + i);
}
}
template <int ITEMS_PER_THREAD, typename FlagT, typename FlagOp>
static _CCCL_DEVICE _CCCL_FORCEINLINE void
FlagTails(int linear_tid, FlagT (&flags)[ITEMS_PER_THREAD], T (&input)[ITEMS_PER_THREAD], FlagOp flag_op)
{
#pragma unroll
for (int i = 0; i < ITEMS_PER_THREAD - 1; ++i)
{
flags[i] = ApplyOp<FlagOp>::FlagT(flag_op, input[i], input[i + 1], (linear_tid * ITEMS_PER_THREAD) + i + 1);
}
}
};
/******************************************************************************
* Thread fields
******************************************************************************/
_TempStorage& temp_storage;
unsigned int linear_tid;
public:
struct TempStorage : Uninitialized<_TempStorage>
{};
_CCCL_DEVICE _CCCL_FORCEINLINE BlockDiscontinuity()
: temp_storage(PrivateStorage())
, linear_tid(RowMajorTid(BLOCK_DIM_X, BLOCK_DIM_Y, BLOCK_DIM_Z))
{}
_CCCL_DEVICE _CCCL_FORCEINLINE BlockDiscontinuity(TempStorage& temp_storage)
: temp_storage(temp_storage.Alias())
, linear_tid(RowMajorTid(BLOCK_DIM_X, BLOCK_DIM_Y, BLOCK_DIM_Z))
{}
#ifndef _CCCL_DOXYGEN_INVOKED // Do not document
template <int ITEMS_PER_THREAD, typename FlagT, typename FlagOp>
_CCCL_DEVICE _CCCL_FORCEINLINE void FlagHeads(
FlagT (&head_flags)[ITEMS_PER_THREAD], T (&input)[ITEMS_PER_THREAD], T (&preds)[ITEMS_PER_THREAD], FlagOp flag_op)
{
// Share last item
temp_storage.last_items[linear_tid] = input[ITEMS_PER_THREAD - 1];
CTA_SYNC();
if (linear_tid == 0)
{
// Set flag for first thread-item (preds[0] is undefined)
head_flags[0] = 1;
}
else
{
preds[0] = temp_storage.last_items[linear_tid - 1];
head_flags[0] = ApplyOp<FlagOp>::FlagT(flag_op, preds[0], input[0], linear_tid * ITEMS_PER_THREAD);
}
// Set head_flags for remaining items
Iterate::FlagHeads(linear_tid, head_flags, input, preds, flag_op);
}
template <int ITEMS_PER_THREAD, typename FlagT, typename FlagOp>
_CCCL_DEVICE _CCCL_FORCEINLINE void FlagHeads(
FlagT (&head_flags)[ITEMS_PER_THREAD],
T (&input)[ITEMS_PER_THREAD],
T (&preds)[ITEMS_PER_THREAD],
FlagOp flag_op,
T tile_predecessor_item)
{
// Share last item
temp_storage.last_items[linear_tid] = input[ITEMS_PER_THREAD - 1];
CTA_SYNC();
// Set flag for first thread-item
preds[0] = (linear_tid == 0) ? tile_predecessor_item : // First thread
temp_storage.last_items[linear_tid - 1];
head_flags[0] = ApplyOp<FlagOp>::FlagT(flag_op, preds[0], input[0], linear_tid * ITEMS_PER_THREAD);
// Set head_flags for remaining items
Iterate::FlagHeads(linear_tid, head_flags, input, preds, flag_op);
}
#endif // _CCCL_DOXYGEN_INVOKED
template <int ITEMS_PER_THREAD, typename FlagT, typename FlagOp>
_CCCL_DEVICE _CCCL_FORCEINLINE void
FlagHeads(FlagT (&head_flags)[ITEMS_PER_THREAD], T (&input)[ITEMS_PER_THREAD], FlagOp flag_op)
{
T preds[ITEMS_PER_THREAD];
FlagHeads(head_flags, input, preds, flag_op);
}
template <int ITEMS_PER_THREAD, typename FlagT, typename FlagOp>
_CCCL_DEVICE _CCCL_FORCEINLINE void FlagHeads(
FlagT (&head_flags)[ITEMS_PER_THREAD], T (&input)[ITEMS_PER_THREAD], FlagOp flag_op, T tile_predecessor_item)
{
T preds[ITEMS_PER_THREAD];
FlagHeads(head_flags, input, preds, flag_op, tile_predecessor_item);
}
template <int ITEMS_PER_THREAD, typename FlagT, typename FlagOp>
_CCCL_DEVICE _CCCL_FORCEINLINE void
FlagTails(FlagT (&tail_flags)[ITEMS_PER_THREAD], T (&input)[ITEMS_PER_THREAD], FlagOp flag_op)
{
// Share first item
temp_storage.first_items[linear_tid] = input[0];
CTA_SYNC();
// Set flag for last thread-item
tail_flags[ITEMS_PER_THREAD - 1] =
(linear_tid == BLOCK_THREADS - 1) ? 1 : // Last thread
ApplyOp<FlagOp>::FlagT(
flag_op,
input[ITEMS_PER_THREAD - 1],
temp_storage.first_items[linear_tid + 1],
(linear_tid * ITEMS_PER_THREAD) + ITEMS_PER_THREAD);
// Set tail_flags for remaining items
Iterate::FlagTails(linear_tid, tail_flags, input, flag_op);
}
template <int ITEMS_PER_THREAD, typename FlagT, typename FlagOp>
_CCCL_DEVICE _CCCL_FORCEINLINE void
FlagTails(FlagT (&tail_flags)[ITEMS_PER_THREAD], T (&input)[ITEMS_PER_THREAD], FlagOp flag_op, T tile_successor_item)
{
// Share first item
temp_storage.first_items[linear_tid] = input[0];
CTA_SYNC();
// Set flag for last thread-item
T successor_item = (linear_tid == BLOCK_THREADS - 1) ? tile_successor_item : // Last thread
temp_storage.first_items[linear_tid + 1];
tail_flags[ITEMS_PER_THREAD - 1] = ApplyOp<FlagOp>::FlagT(
flag_op, input[ITEMS_PER_THREAD - 1], successor_item, (linear_tid * ITEMS_PER_THREAD) + ITEMS_PER_THREAD);
// Set tail_flags for remaining items
Iterate::FlagTails(linear_tid, tail_flags, input, flag_op);
}
template <int ITEMS_PER_THREAD, typename FlagT, typename FlagOp>
_CCCL_DEVICE _CCCL_FORCEINLINE void FlagHeadsAndTails(
FlagT (&head_flags)[ITEMS_PER_THREAD],
FlagT (&tail_flags)[ITEMS_PER_THREAD],
T (&input)[ITEMS_PER_THREAD],
FlagOp flag_op)
{
// Share first and last items
temp_storage.first_items[linear_tid] = input[0];
temp_storage.last_items[linear_tid] = input[ITEMS_PER_THREAD - 1];
CTA_SYNC();
T preds[ITEMS_PER_THREAD];
// Set flag for first thread-item
if (linear_tid == 0)
{
head_flags[0] = 1;
}
else
{
preds[0] = temp_storage.last_items[linear_tid - 1];
head_flags[0] = ApplyOp<FlagOp>::FlagT(flag_op, preds[0], input[0], linear_tid * ITEMS_PER_THREAD);
}
// Set flag for last thread-item
tail_flags[ITEMS_PER_THREAD - 1] =
(linear_tid == BLOCK_THREADS - 1) ? 1 : // Last thread
ApplyOp<FlagOp>::FlagT(
flag_op,
input[ITEMS_PER_THREAD - 1],
temp_storage.first_items[linear_tid + 1],
(linear_tid * ITEMS_PER_THREAD) + ITEMS_PER_THREAD);
// Set head_flags for remaining items
Iterate::FlagHeads(linear_tid, head_flags, input, preds, flag_op);
// Set tail_flags for remaining items
Iterate::FlagTails(linear_tid, tail_flags, input, flag_op);
}
template <int ITEMS_PER_THREAD, typename FlagT, typename FlagOp>
_CCCL_DEVICE _CCCL_FORCEINLINE void FlagHeadsAndTails(
FlagT (&head_flags)[ITEMS_PER_THREAD],
FlagT (&tail_flags)[ITEMS_PER_THREAD],
T tile_successor_item,
T (&input)[ITEMS_PER_THREAD],
FlagOp flag_op)
{
// Share first and last items
temp_storage.first_items[linear_tid] = input[0];
temp_storage.last_items[linear_tid] = input[ITEMS_PER_THREAD - 1];
CTA_SYNC();
T preds[ITEMS_PER_THREAD];
// Set flag for first thread-item
if (linear_tid == 0)
{
head_flags[0] = 1;
}
else
{
preds[0] = temp_storage.last_items[linear_tid - 1];
head_flags[0] = ApplyOp<FlagOp>::FlagT(flag_op, preds[0], input[0], linear_tid * ITEMS_PER_THREAD);
}
// Set flag for last thread-item
T successor_item = (linear_tid == BLOCK_THREADS - 1) ? tile_successor_item : // Last thread
temp_storage.first_items[linear_tid + 1];
tail_flags[ITEMS_PER_THREAD - 1] = ApplyOp<FlagOp>::FlagT(
flag_op, input[ITEMS_PER_THREAD - 1], successor_item, (linear_tid * ITEMS_PER_THREAD) + ITEMS_PER_THREAD);
// Set head_flags for remaining items
Iterate::FlagHeads(linear_tid, head_flags, input, preds, flag_op);
// Set tail_flags for remaining items
Iterate::FlagTails(linear_tid, tail_flags, input, flag_op);
}
template <int ITEMS_PER_THREAD, typename FlagT, typename FlagOp>
_CCCL_DEVICE _CCCL_FORCEINLINE void FlagHeadsAndTails(
FlagT (&head_flags)[ITEMS_PER_THREAD],
T tile_predecessor_item,
FlagT (&tail_flags)[ITEMS_PER_THREAD],
T (&input)[ITEMS_PER_THREAD],
FlagOp flag_op)
{
// Share first and last items
temp_storage.first_items[linear_tid] = input[0];
temp_storage.last_items[linear_tid] = input[ITEMS_PER_THREAD - 1];
CTA_SYNC();
T preds[ITEMS_PER_THREAD];
// Set flag for first thread-item
preds[0] = (linear_tid == 0) ? tile_predecessor_item : // First thread
temp_storage.last_items[linear_tid - 1];
head_flags[0] = ApplyOp<FlagOp>::FlagT(flag_op, preds[0], input[0], linear_tid * ITEMS_PER_THREAD);
// Set flag for last thread-item
tail_flags[ITEMS_PER_THREAD - 1] =
(linear_tid == BLOCK_THREADS - 1) ? 1 : // Last thread
ApplyOp<FlagOp>::FlagT(
flag_op,
input[ITEMS_PER_THREAD - 1],
temp_storage.first_items[linear_tid + 1],
(linear_tid * ITEMS_PER_THREAD) + ITEMS_PER_THREAD);
// Set head_flags for remaining items
Iterate::FlagHeads(linear_tid, head_flags, input, preds, flag_op);
// Set tail_flags for remaining items
Iterate::FlagTails(linear_tid, tail_flags, input, flag_op);
}
template <int ITEMS_PER_THREAD, typename FlagT, typename FlagOp>
_CCCL_DEVICE _CCCL_FORCEINLINE void FlagHeadsAndTails(
FlagT (&head_flags)[ITEMS_PER_THREAD],
T tile_predecessor_item,
FlagT (&tail_flags)[ITEMS_PER_THREAD],
T tile_successor_item,
T (&input)[ITEMS_PER_THREAD],
FlagOp flag_op)
{
// Share first and last items
temp_storage.first_items[linear_tid] = input[0];
temp_storage.last_items[linear_tid] = input[ITEMS_PER_THREAD - 1];
CTA_SYNC();
T preds[ITEMS_PER_THREAD];
// Set flag for first thread-item
preds[0] = (linear_tid == 0) ? tile_predecessor_item : // First thread
temp_storage.last_items[linear_tid - 1];
head_flags[0] = ApplyOp<FlagOp>::FlagT(flag_op, preds[0], input[0], linear_tid * ITEMS_PER_THREAD);
// Set flag for last thread-item
T successor_item = (linear_tid == BLOCK_THREADS - 1) ? tile_successor_item : // Last thread
temp_storage.first_items[linear_tid + 1];
tail_flags[ITEMS_PER_THREAD - 1] = ApplyOp<FlagOp>::FlagT(
flag_op, input[ITEMS_PER_THREAD - 1], successor_item, (linear_tid * ITEMS_PER_THREAD) + ITEMS_PER_THREAD);
// Set head_flags for remaining items
Iterate::FlagHeads(linear_tid, head_flags, input, preds, flag_op);
// Set tail_flags for remaining items
Iterate::FlagTails(linear_tid, tail_flags, input, flag_op);
}
};
CUB_NAMESPACE_END