1 // Copyright (c) 2017-2022, Lawrence Livermore National Security, LLC and other CEED contributors. 2 // All Rights Reserved. See the top-level LICENSE and NOTICE files for details. 3 // 4 // SPDX-License-Identifier: BSD-2-Clause 5 // 6 // This file is part of CEED: http://github.com/ceed 7 8 /// @file 9 /// Internal header for CUDA operator full assembly 10 #ifndef _ceed_cuda_ref_operator_assemble_h 11 #define _ceed_cuda_ref_operator_assemble_h 12 13 #include <ceed.h> 14 15 #if CEEDSIZE 16 typedef CeedSize IndexType; 17 #else 18 typedef CeedInt IndexType; 19 #endif 20 21 //------------------------------------------------------------------------------ 22 // Matrix assembly kernel for low-order elements (2D thread block) 23 //------------------------------------------------------------------------------ 24 extern "C" __launch_bounds__(BLOCK_SIZE) __global__ 25 void linearAssemble(const CeedScalar *B_in, const CeedScalar *B_out, const CeedScalar *__restrict__ qf_array, 26 CeedScalar *__restrict__ values_array) { 27 // This kernel assumes B_in and B_out have the same number of quadrature points and basis points. 28 // TODO: expand to more general cases 29 const int i = threadIdx.x; // The output row index of each B^TDB operation 30 const int l = threadIdx.y; // The output column index of each B^TDB operation 31 // such that we have (Bout^T)_ij D_jk Bin_kl = C_il 32 33 // Strides for final output ordering, determined by the reference (interface) implementation of the symbolic assembly, slowest --> fastest: element, 34 // comp_in, comp_out, node_row, node_col 35 const IndexType comp_out_stride = NNODES * NNODES; 36 const IndexType comp_in_stride = comp_out_stride * NCOMP; 37 const IndexType e_stride = comp_in_stride * NCOMP; 38 // Strides for QF array, slowest --> fastest: emode_in, comp_in, emode_out, comp_out, elem, qpt 39 const IndexType qe_stride = NQPTS; 40 const IndexType qcomp_out_stride = NELEM * qe_stride; 41 const IndexType qemode_out_stride = qcomp_out_stride * NCOMP; 42 const IndexType qcomp_in_stride = qemode_out_stride * NUMEMODEOUT; 43 const IndexType qemode_in_stride = qcomp_in_stride * NCOMP; 44 45 // Loop over each element (if necessary) 46 for (IndexType e = blockIdx.x * blockDim.z + threadIdx.z; e < NELEM; e += gridDim.x * blockDim.z) { 47 for (IndexType comp_in = 0; comp_in < NCOMP; comp_in++) { 48 for (IndexType comp_out = 0; comp_out < NCOMP; comp_out++) { 49 CeedScalar result = 0.0; 50 IndexType qf_index_comp = qcomp_in_stride * comp_in + qcomp_out_stride * comp_out + qe_stride * e; 51 for (IndexType emode_in = 0; emode_in < NUMEMODEIN; emode_in++) { 52 IndexType b_in_index = emode_in * NQPTS * NNODES; 53 for (IndexType emode_out = 0; emode_out < NUMEMODEOUT; emode_out++) { 54 IndexType b_out_index = emode_out * NQPTS * NNODES; 55 IndexType qf_index = qf_index_comp + qemode_out_stride * emode_out + qemode_in_stride * emode_in; 56 // Perform the B^T D B operation for this 'chunk' of D (the qf_array) 57 for (IndexType j = 0; j < NQPTS; j++) { 58 result += B_out[b_out_index + j * NNODES + i] * qf_array[qf_index + j] * B_in[b_in_index + j * NNODES + l]; 59 } 60 } // end of emode_out 61 } // end of emode_in 62 IndexType val_index = comp_in_stride * comp_in + comp_out_stride * comp_out + e_stride * e + NNODES * i + l; 63 values_array[val_index] = result; 64 } // end of out component 65 } // end of in component 66 } // end of element loop 67 } 68 69 //------------------------------------------------------------------------------ 70 // Fallback kernel for larger orders (1D thread block) 71 //------------------------------------------------------------------------------ 72 extern "C" __launch_bounds__(BLOCK_SIZE) __global__ 73 void linearAssembleFallback(const CeedScalar *B_in, const CeedScalar *B_out, const CeedScalar *__restrict__ qf_array, 74 CeedScalar *__restrict__ values_array) { 75 // This kernel assumes B_in and B_out have the same number of quadrature points and basis points. 76 // TODO: expand to more general cases 77 const int l = threadIdx.x; // The output column index of each B^TDB operation 78 // such that we have (Bout^T)_ij D_jk Bin_kl = C_il 79 80 // Strides for final output ordering, determined by the reference (interface) implementation of the symbolic assembly, slowest --> fastest: element, 81 // comp_in, comp_out, node_row, node_col 82 const IndexType comp_out_stride = NNODES * NNODES; 83 const IndexType comp_in_stride = comp_out_stride * NCOMP; 84 const IndexType e_stride = comp_in_stride * NCOMP; 85 // Strides for QF array, slowest --> fastest: emode_in, comp_in, emode_out, comp_out, elem, qpt 86 const IndexType qe_stride = NQPTS; 87 const IndexType qcomp_out_stride = NELEM * qe_stride; 88 const IndexType qemode_out_stride = qcomp_out_stride * NCOMP; 89 const IndexType qcomp_in_stride = qemode_out_stride * NUMEMODEOUT; 90 const IndexType qemode_in_stride = qcomp_in_stride * NCOMP; 91 92 // Loop over each element (if necessary) 93 for (IndexType e = blockIdx.x * blockDim.z + threadIdx.z; e < NELEM; e += gridDim.x * blockDim.z) { 94 for (IndexType comp_in = 0; comp_in < NCOMP; comp_in++) { 95 for (IndexType comp_out = 0; comp_out < NCOMP; comp_out++) { 96 for (IndexType i = 0; i < NNODES; i++) { 97 CeedScalar result = 0.0; 98 IndexType qf_index_comp = qcomp_in_stride * comp_in + qcomp_out_stride * comp_out + qe_stride * e; 99 for (IndexType emode_in = 0; emode_in < NUMEMODEIN; emode_in++) { 100 IndexType b_in_index = emode_in * NQPTS * NNODES; 101 for (IndexType emode_out = 0; emode_out < NUMEMODEOUT; emode_out++) { 102 IndexType b_out_index = emode_out * NQPTS * NNODES; 103 IndexType qf_index = qf_index_comp + qemode_out_stride * emode_out + qemode_in_stride * emode_in; 104 // Perform the B^T D B operation for this 'chunk' of D (the qf_array) 105 for (IndexType j = 0; j < NQPTS; j++) { 106 result += B_out[b_out_index + j * NNODES + i] * qf_array[qf_index + j] * B_in[b_in_index + j * NNODES + l]; 107 } 108 } // end of emode_out 109 } // end of emode_in 110 IndexType val_index = comp_in_stride * comp_in + comp_out_stride * comp_out + e_stride * e + NNODES * i + l; 111 values_array[val_index] = result; 112 } // end of loop over element node index, i 113 } // end of out component 114 } // end of in component 115 } // end of element loop 116 } 117 118 //------------------------------------------------------------------------------ 119 120 #endif 121