CMAKE CUDA SOFTWARE
CMAKE CUDA CODE

(INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES

IN NO EVENT SHALL NVIDIA CORPORATION BE LIABLE FOR ANYĭIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE AREĭISCLAIMED.

CMAKE CUDA SOFTWARE

THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS 'AS IS' ANDĪNY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED Names of its contributors may be used to endorse or promote productsĭerived from this software without specific prior written permission. * Neither the name of the NVIDIA CORPORATION nor the Notice, this list of conditions and the following disclaimer in theĭocumentation and/or other materials provided with the distribution. * Redistributions in binary form must reproduce the above copyright Notice, this list of conditions and the following disclaimer.

CMAKE CUDA CODE

* Redistributions of source code must retain the above copyright Modification, are permitted provided that the following conditions are met: Redistribution and use in source and binary forms, with or without All rights reserved.Ĭopyright (c) 2011-2018, NVIDIA CORPORATION. To build CUB as a developer, the followingĬopyright (c) 2010-2011, Duane Merrill. It is recommended to clone ThrustĬUB uses the CMake build system to build unit tests,Įxamples, and header tests. CUB ReleaseĬUB and Thrust depend on each other. See the changelog for details about specific releases. Silenced by defining CUB_IGNORE_DEPRECATED_COMPILER during compilation.ĬUB is distributed with the NVIDIA HPC SDK and the CUDA Toolkit in addition Unsupported versions may emit deprecation warnings, which can be (Alternatively these storage typesĬould be aliased to global memory allocations).ĬUB is regularly tested using the specified versions of the followingĬompilers. Shared memory needed by the thread block. The thread block uses these storage types to statically allocate the union of Once specialized, these classes expose opaque TempStorage member types. Simultaneously access consecutive items) and then transpose the keys intoĪ blocked arrangement of elements across threads. The cub::BlockLoad and cub::BlockStore classes are similarly specialized.įurthermore, to provide coalesced accesses to device memory, these primitives areĬonfigured to access memory using a striped access pattern (where consecutive threads Keys per thread, and implicitly by the targeted compilation architecture. The class is specialized by theĭata type being sorted, by the number of threads per block, by the number of Store(d_out + block_offset, thread_keys) Įach thread block uses cub::BlockRadixSort to collectively sort Store the sorted segment BlockStore(temp_storage. Collectively sort the keys BlockRadixSort(temp_storage. x * ( 128 * 16) // OffsetT for this block's ment // Obtain a segment of 2048 consecutive keys that are blocked across threads int thread_keys īlockLoad(temp_storage. Typename BlockRadixSort::TempStorage sort Using namespace cub // Specialize BlockRadixSort, BlockLoad, and BlockStore for 128 threads // owning 16 integer items each typedef BlockRadixSort BlockRadixSort

Only defined for systems that support being used as the device._global_ void BlockSortKernel( int *d_in, int *d_out).

Attempting to do so will produce a CMake configuration error.

Multiple ::Device targets cannot be combined in the same library/executable.

Configures an interface for using a specific device system.

Only defined for systems that support being used as the host.

Multiple ::Host targets cannot be combined in the same library/executable.

Configures an interface for using a specific host system.

For example, the Thrust::CUDA target is an interface target that combines the interfaces of both Thrust and CUB.Specifies an interface configured to build against all dependencies for this backend (including Thrust::Thrust).It does not locate or create configurations for any dependencies these are lazily loaded on-demand by calls to create_thrust_target, or when explicitly requested via find_package's component mechanism.Īs mentioned, the basic Thrust interface is described by the Thrust::Thrust target.Įach backend system ( CPP, CUDA, TBB, OMP) is described by multiple targets: Internal Targetsīy default, find_package(Thrust) will only create a single Thrust::Thrust target that describes where the actual Thrust headers are located. It should not be necessary for users who just want to use Thrust from their projects. This portion of the file contains descriptions of Thrust's internal CMake target structure for Thrust developers. Math(EXPR version '$ % 100') Thrust Developer Documentation

Cmake cuda