1 // Copyright (c) 2017-2025, 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 MAGMA tensor basis weight in 3D 10 #include "magma-common-tensor.h" 11 12 //////////////////////////////////////////////////////////////////////////////// 13 // weight basis action -- 3D 14 template <typename T, int DIM, int NUM_COMP, int Q, int i_DIM, int i_COMP> 15 static __device__ __inline__ void magma_weight_3d_device(const T *sTweight, T rV[DIM][NUM_COMP][Q], const int tx) { 16 // Assumptions 17 // 1. 1D thread configuration of size Q^2 18 // 2. rV[][][] matches the storage used in other actions (interp, grad, ... etc) 19 // 3. i_DIM and i_COMP specify which indexes to use in rV, 20 // since the output per thread is a register array of size Q 21 // 4. Sync is recommended after the call (to make sure sTweight can be overwritten) 22 23 if (tx < Q * Q) { 24 // x sTweight[j] for first update 25 // x sTweight[tx%Q] for second update 26 // x sTweight[tx/Q] for third update 27 for (int j = 0; j < Q; j++) { 28 rV[i_DIM][i_COMP][j] = sTweight[j] * sTweight[tx % Q] * sTweight[tx / Q]; 29 } 30 } 31 } 32 33 //////////////////////////////////////////////////////////////////////////////// 34 extern "C" __launch_bounds__(MAGMA_BASIS_BOUNDS(BASIS_Q *BASIS_Q, MAGMA_MAXTHREADS_3D)) __global__ 35 void magma_weight_3d_kernel(const CeedScalar *dqweight1d, CeedScalar *dV, const int v_stride, const int nelem) { 36 MAGMA_DEVICE_SHARED(CeedScalar, shared_data) 37 38 const int tx = threadIdx.x; 39 const int ty = threadIdx.y; 40 const int elem_id = (blockIdx.x * blockDim.y) + ty; 41 42 if (elem_id >= nelem) return; 43 44 CeedScalar rV[1][1][BASIS_Q]; // allocate with BASIS_DIM=BASIS_NUM_COMP=1, but sizes may differ for a fused operator 45 // global memory pointers 46 dV += elem_id * v_stride; 47 48 // shared memory pointers 49 CeedScalar *sTweight = (CeedScalar *)shared_data; 50 51 // read dqweight_1d 52 if (tx < BASIS_Q) { 53 sTweight[tx] = dqweight1d[tx]; 54 } 55 __syncthreads(); 56 57 magma_weight_3d_device<CeedScalar, 1, 1, BASIS_Q, 0, 0>(sTweight, rV, tx); 58 59 // write V 60 if (tx < (BASIS_Q * BASIS_Q)) { 61 for (int j = 0; j < BASIS_Q; j++) { 62 dV[j * (BASIS_Q * BASIS_Q) + tx] = rV[0][0][j]; 63 } 64 } 65 } 66