// ============================================================================= // === GPUQREngine/Include/Kernel/uberKernel.cu ================================ // ============================================================================= // GPUQREngine, Copyright (c) 2013, Timothy A Davis, Sencer Nuri Yeralan, // and Sanjay Ranka. All Rights Reserved. // SPDX-License-Identifier: GPL-2.0+ //------------------------------------------------------------------------------ #include "GPUQREngine_Common.hpp" #include "GPUQREngine_SEntry.hpp" #include "GPUQREngine_TaskDescriptor.hpp" /*** Shared Memory Allocation ************************************************/ #include "sharedMemory.hpp" /*** Assembly Device Functions ***********************************************/ #include "Kernel/Assemble/sAssemble.cu" #include "Kernel/Assemble/packAssemble.cu" /*** Apply Device Functions **************************************************/ #include "Kernel/Apply/params_apply.hpp" #include "Kernel/Apply/block_apply_3.cu" #include "Kernel/Apply/block_apply_2.cu" #include "Kernel/Apply/block_apply_1.cu" #ifdef GPUQRENGINE_PIPELINING #include "Kernel/Apply/block_apply_3_by_1.cu" #include "Kernel/Apply/block_apply_2_by_1.cu" #endif /*** Factorize Device Functions **********************************************/ #include "Kernel/Factorize/factorize_vt_3_by_1.cu" #include "Kernel/Factorize/factorize_vt_2_by_1.cu" #include "Kernel/Factorize/factorize_vt_1_by_1.cu" #include "Kernel/Factorize/factorize_vt_3_by_1_edge.cu" #include "Kernel/Factorize/factorize_vt_2_by_1_edge.cu" #include "Kernel/Factorize/factorize_vt_1_by_1_edge.cu" #include "Kernel/Factorize/factorize_3_by_1.cu" /*** Main Uberkernel Global Function *****************************************/ #include "Kernel/qrKernel.cu"