Spaces:
Runtime error
Runtime error
File size: 21,201 Bytes
28958dc |
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 |
/******************************************************************************
* 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.
*
******************************************************************************/
/**
* \file
* cub::DeviceScan provides device-wide, parallel operations for computing a prefix scan across a sequence of data items residing within device-accessible memory.
*/
#pragma once
#include <stdio.h>
#include <iterator>
#include "../../agent/agent_scan.cuh"
#include "../../thread/thread_operators.cuh"
#include "../../grid/grid_queue.cuh"
#include "../../config.cuh"
#include "../../util_debug.cuh"
#include "../../util_device.cuh"
#include <thrust/system/cuda/detail/core/triple_chevron_launch.h>
/// Optional outer namespace(s)
CUB_NS_PREFIX
/// CUB namespace
namespace cub {
/******************************************************************************
* Kernel entry points
*****************************************************************************/
/**
* Initialization kernel for tile status initialization (multi-block)
*/
template <
typename ScanTileStateT> ///< Tile status interface type
__global__ void DeviceScanInitKernel(
ScanTileStateT tile_state, ///< [in] Tile status interface
int num_tiles) ///< [in] Number of tiles
{
// Initialize tile status
tile_state.InitializeStatus(num_tiles);
}
/**
* Initialization kernel for tile status initialization (multi-block)
*/
template <
typename ScanTileStateT, ///< Tile status interface type
typename NumSelectedIteratorT> ///< Output iterator type for recording the number of items selected
__global__ void DeviceCompactInitKernel(
ScanTileStateT tile_state, ///< [in] Tile status interface
int num_tiles, ///< [in] Number of tiles
NumSelectedIteratorT d_num_selected_out) ///< [out] Pointer to the total number of items selected (i.e., length of \p d_selected_out)
{
// Initialize tile status
tile_state.InitializeStatus(num_tiles);
// Initialize d_num_selected_out
if ((blockIdx.x == 0) && (threadIdx.x == 0))
*d_num_selected_out = 0;
}
/**
* Scan kernel entry point (multi-block)
*/
template <
typename ScanPolicyT, ///< Parameterized ScanPolicyT tuning policy type
typename InputIteratorT, ///< Random-access input iterator type for reading scan inputs \iterator
typename OutputIteratorT, ///< Random-access output iterator type for writing scan outputs \iterator
typename ScanTileStateT, ///< Tile status interface type
typename ScanOpT, ///< Binary scan functor type having member <tt>T operator()(const T &a, const T &b)</tt>
typename InitValueT, ///< Initial value to seed the exclusive scan (cub::NullType for inclusive scans)
typename OffsetT> ///< Signed integer type for global offsets
__launch_bounds__ (int(ScanPolicyT::BLOCK_THREADS))
__global__ void DeviceScanKernel(
InputIteratorT d_in, ///< Input data
OutputIteratorT d_out, ///< Output data
ScanTileStateT tile_state, ///< Tile status interface
int start_tile, ///< The starting tile for the current grid
ScanOpT scan_op, ///< Binary scan functor
InitValueT init_value, ///< Initial value to seed the exclusive scan
OffsetT num_items) ///< Total number of scan items for the entire problem
{
// Thread block type for scanning input tiles
typedef AgentScan<
ScanPolicyT,
InputIteratorT,
OutputIteratorT,
ScanOpT,
InitValueT,
OffsetT> AgentScanT;
// Shared memory for AgentScan
__shared__ typename AgentScanT::TempStorage temp_storage;
// Process tiles
AgentScanT(temp_storage, d_in, d_out, scan_op, init_value).ConsumeRange(
num_items,
tile_state,
start_tile);
}
/******************************************************************************
* Policy
******************************************************************************/
template <
typename OutputT> ///< Data type
struct DeviceScanPolicy
{
/// SM10
struct Policy100 : ChainedPolicy<100, Policy100, Policy100>
{
typedef AgentScanPolicy<
64, 9, ///< Threads per block, items per thread
OutputT,
BLOCK_LOAD_WARP_TRANSPOSE,
LOAD_DEFAULT,
BLOCK_STORE_WARP_TRANSPOSE,
BLOCK_SCAN_WARP_SCANS>
ScanPolicyT;
};
/// SM13
struct Policy130 : ChainedPolicy<130, Policy130, Policy100>
{
typedef AgentScanPolicy<
96, 21, ///< Threads per block, items per thread
OutputT,
BLOCK_LOAD_WARP_TRANSPOSE,
LOAD_DEFAULT,
BLOCK_STORE_WARP_TRANSPOSE,
BLOCK_SCAN_RAKING_MEMOIZE>
ScanPolicyT;
};
/// SM20
struct Policy200 : ChainedPolicy<200, Policy200, Policy130>
{
// GTX 580: 20.3B items/s (162.3 GB/s) @ 48M 32-bit T
typedef AgentScanPolicy<
128, 12, ///< Threads per block, items per thread
OutputT,
BLOCK_LOAD_WARP_TRANSPOSE,
LOAD_DEFAULT,
BLOCK_STORE_WARP_TRANSPOSE,
BLOCK_SCAN_WARP_SCANS>
ScanPolicyT;
};
/// SM30
struct Policy300 : ChainedPolicy<300, Policy300, Policy200>
{
typedef AgentScanPolicy<
256, 9, ///< Threads per block, items per thread
OutputT,
BLOCK_LOAD_WARP_TRANSPOSE,
LOAD_DEFAULT,
BLOCK_STORE_WARP_TRANSPOSE,
BLOCK_SCAN_WARP_SCANS>
ScanPolicyT;
};
/// SM35
struct Policy350 : ChainedPolicy<350, Policy350, Policy300>
{
// GTX Titan: 29.5B items/s (232.4 GB/s) @ 48M 32-bit T
typedef AgentScanPolicy<
128, 12, ///< Threads per block, items per thread
OutputT,
BLOCK_LOAD_DIRECT,
LOAD_LDG,
BLOCK_STORE_WARP_TRANSPOSE_TIMESLICED,
BLOCK_SCAN_RAKING>
ScanPolicyT;
};
/// SM520
struct Policy520 : ChainedPolicy<520, Policy520, Policy350>
{
// Titan X: 32.47B items/s @ 48M 32-bit T
typedef AgentScanPolicy<
128, 12, ///< Threads per block, items per thread
OutputT,
BLOCK_LOAD_DIRECT,
LOAD_LDG,
BLOCK_STORE_WARP_TRANSPOSE,
BLOCK_SCAN_WARP_SCANS>
ScanPolicyT;
};
/// SM600
struct Policy600 : ChainedPolicy<600, Policy600, Policy520>
{
typedef AgentScanPolicy<
128, 15, ///< Threads per block, items per thread
OutputT,
BLOCK_LOAD_TRANSPOSE,
LOAD_DEFAULT,
BLOCK_STORE_TRANSPOSE,
BLOCK_SCAN_WARP_SCANS>
ScanPolicyT;
};
/// MaxPolicy
typedef Policy600 MaxPolicy;
};
/******************************************************************************
* Dispatch
******************************************************************************/
/**
* Utility class for dispatching the appropriately-tuned kernels for DeviceScan
*/
template <
typename InputIteratorT, ///< Random-access input iterator type for reading scan inputs \iterator
typename OutputIteratorT, ///< Random-access output iterator type for writing scan outputs \iterator
typename ScanOpT, ///< Binary scan functor type having member <tt>T operator()(const T &a, const T &b)</tt>
typename InitValueT, ///< The init_value element type for ScanOpT (cub::NullType for inclusive scans)
typename OffsetT, ///< Signed integer type for global offsets
typename SelectedPolicy = DeviceScanPolicy<
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> >
struct DispatchScan:
SelectedPolicy
{
//---------------------------------------------------------------------
// Constants and Types
//---------------------------------------------------------------------
enum
{
INIT_KERNEL_THREADS = 128
};
// 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
void* 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.
size_t& temp_storage_bytes; ///< [in,out] Reference to size in bytes of \p d_temp_storage allocation
InputIteratorT d_in; ///< [in] Pointer to the input sequence of data items
OutputIteratorT d_out; ///< [out] Pointer to the output sequence of data items
ScanOpT scan_op; ///< [in] Binary scan functor
InitValueT init_value; ///< [in] Initial value to seed the exclusive scan
OffsetT num_items; ///< [in] Total number of input items (i.e., the length of \p d_in)
cudaStream_t stream; ///< [in] <b>[optional]</b> CUDA stream to launch kernels within. Default is stream<sub>0</sub>.
bool debug_synchronous;
int ptx_version;
CUB_RUNTIME_FUNCTION __forceinline__
DispatchScan(
void* 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.
size_t& temp_storage_bytes, ///< [in,out] Reference to size in bytes of \p d_temp_storage allocation
InputIteratorT d_in, ///< [in] Pointer to the input sequence of data items
OutputIteratorT d_out, ///< [out] Pointer to the output sequence of data items
OffsetT num_items, ///< [in] Total number of input items (i.e., the length of \p d_in)
ScanOpT scan_op, ///< [in] Binary scan functor
InitValueT init_value, ///< [in] Initial value to seed the exclusive scan
cudaStream_t stream, ///< [in] <b>[optional]</b> CUDA stream to launch kernels within. Default is stream<sub>0</sub>.
bool debug_synchronous,
int ptx_version
):
d_temp_storage(d_temp_storage),
temp_storage_bytes(temp_storage_bytes),
d_in(d_in),
d_out(d_out),
num_items(num_items),
scan_op(scan_op),
init_value(init_value),
stream(stream),
debug_synchronous(debug_synchronous),
ptx_version(ptx_version)
{}
template <typename ActivePolicyT, typename InitKernel, typename ScanKernel>
CUB_RUNTIME_FUNCTION __host__ __forceinline__
cudaError_t Invoke(InitKernel init_kernel, ScanKernel scan_kernel)
{
#ifndef CUB_RUNTIME_ENABLED
(void)init_kernel;
(void)scan_kernel;
// Kernel launch not supported from this device
return CubDebug(cudaErrorNotSupported);
#else
typedef typename ActivePolicyT::ScanPolicyT Policy;
typedef typename cub::ScanTileState<OutputT> ScanTileStateT;
cudaError error = cudaSuccess;
do
{
// Get device ordinal
int device_ordinal;
if (CubDebug(error = cudaGetDevice(&device_ordinal))) break;
// Get SM count
int sm_count;
if (CubDebug(error = cudaDeviceGetAttribute (&sm_count, cudaDevAttrMultiProcessorCount, device_ordinal))) break;
// Number of input tiles
int tile_size = Policy::BLOCK_THREADS * Policy::ITEMS_PER_THREAD;
int num_tiles = (num_items + tile_size - 1) / tile_size;
// Specify temporary storage allocation requirements
size_t allocation_sizes[1];
if (CubDebug(error = ScanTileStateT::AllocationSize(num_tiles, allocation_sizes[0]))) break; // bytes needed for tile status descriptors
// Compute allocation pointers into the single storage blob (or compute the necessary size of the blob)
void* allocations[1] = {};
if (CubDebug(error = AliasTemporaries(d_temp_storage, temp_storage_bytes, allocations, allocation_sizes))) break;
if (d_temp_storage == NULL)
{
// 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_state;
if (CubDebug(error = tile_state.Init(num_tiles, allocations[0], allocation_sizes[0]))) break;
// Log init_kernel configuration
int init_grid_size = (num_tiles + INIT_KERNEL_THREADS - 1) / INIT_KERNEL_THREADS;
if (debug_synchronous) _CubLog("Invoking init_kernel<<<%d, %d, 0, %lld>>>()\n", init_grid_size, INIT_KERNEL_THREADS, (long long) stream);
// Invoke init_kernel to initialize tile descriptors
thrust::cuda_cub::launcher::triple_chevron(
init_grid_size, INIT_KERNEL_THREADS, 0, stream
).doit(init_kernel,
tile_state,
num_tiles);
// Check for failure to launch
if (CubDebug(error = cudaPeekAtLastError())) break;
// Sync the stream if specified to flush runtime errors
if (debug_synchronous && (CubDebug(error = SyncStream(stream)))) break;
// Get SM occupancy for scan_kernel
int scan_sm_occupancy;
if (CubDebug(error = MaxSmOccupancy(
scan_sm_occupancy, // out
scan_kernel,
Policy::BLOCK_THREADS))) break;
// Get max x-dimension of grid
int max_dim_x;
if (CubDebug(error = cudaDeviceGetAttribute(&max_dim_x, cudaDevAttrMaxGridDimX, device_ordinal))) break;;
// Run grids in epochs (in case number of tiles exceeds max x-dimension
int scan_grid_size = CUB_MIN(num_tiles, max_dim_x);
for (int start_tile = 0; start_tile < num_tiles; start_tile += scan_grid_size)
{
// Log scan_kernel configuration
if (debug_synchronous) _CubLog("Invoking %d scan_kernel<<<%d, %d, 0, %lld>>>(), %d items per thread, %d SM occupancy\n",
start_tile, scan_grid_size, Policy::BLOCK_THREADS, (long long) stream, Policy::ITEMS_PER_THREAD, scan_sm_occupancy);
// Invoke scan_kernel
thrust::cuda_cub::launcher::triple_chevron(
scan_grid_size, Policy::BLOCK_THREADS, 0, stream
).doit(scan_kernel,
d_in,
d_out,
tile_state,
start_tile,
scan_op,
init_value,
num_items);
// Check for failure to launch
if (CubDebug(error = cudaPeekAtLastError())) break;
// Sync the stream if specified to flush runtime errors
if (debug_synchronous && (CubDebug(error = SyncStream(stream)))) break;
}
}
while (0);
return error;
#endif // CUB_RUNTIME_ENABLED
}
template <typename ActivePolicyT>
CUB_RUNTIME_FUNCTION __host__ __forceinline__
cudaError_t Invoke()
{
typedef typename ActivePolicyT::ScanPolicyT Policy;
typedef typename cub::ScanTileState<OutputT> ScanTileStateT;
// Ensure kernels are instantiated.
return Invoke<ActivePolicyT>(
DeviceScanInitKernel<ScanTileStateT>,
DeviceScanKernel<Policy, InputIteratorT, OutputIteratorT, ScanTileStateT, ScanOpT, InitValueT, OffsetT>
);
}
/**
* Internal dispatch routine
*/
CUB_RUNTIME_FUNCTION __forceinline__
static cudaError_t Dispatch(
void* 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.
size_t& temp_storage_bytes, ///< [in,out] Reference to size in bytes of \p d_temp_storage allocation
InputIteratorT d_in, ///< [in] Pointer to the input sequence of data items
OutputIteratorT d_out, ///< [out] Pointer to the output sequence of data items
ScanOpT scan_op, ///< [in] Binary scan functor
InitValueT init_value, ///< [in] Initial value to seed the exclusive scan
OffsetT num_items, ///< [in] Total number of input items (i.e., the length of \p d_in)
cudaStream_t stream, ///< [in] <b>[optional]</b> CUDA stream to launch kernels within. Default is stream<sub>0</sub>.
bool debug_synchronous) ///< [in] <b>[optional]</b> Whether or not to synchronize the stream after every kernel launch to check for errors. Also causes launch configurations to be printed to the console. Default is \p false.
{
typedef typename DispatchScan::MaxPolicy MaxPolicyT;
cudaError_t error;
do
{
// Get PTX version
int ptx_version = 0;
if (CubDebug(error = PtxVersion(ptx_version))) break;
// Create dispatch functor
DispatchScan dispatch(
d_temp_storage,
temp_storage_bytes,
d_in,
d_out,
num_items,
scan_op,
init_value,
stream,
debug_synchronous,
ptx_version
);
// Dispatch to chained policy
if (CubDebug(error = MaxPolicyT::Invoke(ptx_version, dispatch))) break;
}
while (0);
return error;
}
};
} // CUB namespace
CUB_NS_POSTFIX // Optional outer namespace(s)
|