1 // Copyright (c) 2017-2024, 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 gradient in 1D 10 #ifndef CEED_MAGMA_BASIS_GRAD_1D_H 11 #define CEED_MAGMA_BASIS_GRAD_1D_H 12 13 #include "magma-common-tensor.h" 14 15 // macros to abstract access of shared memory and reg. file 16 #define sT(i, j) sT[(j) * P + (i)] 17 18 //////////////////////////////////////////////////////////////////////////////// 19 // grad basis action (1D) 20 template <typename T, int DIM, int NUM_COMP, int P, int Q> 21 static __device__ __inline__ void magma_grad_1d_device(const T *sT, T *sU[NUM_COMP], T *sV[NUM_COMP], const int tx) { 22 // Assumptions 23 // 1. 1D threads of size max(P,Q) 24 // 2. sU[i] is 1xP: in shared memory 25 // 3. sV[i] is 1xQ: in shared memory 26 // 4. P_roduct per component is one row (1xP) times T matrix (PxQ) => one row (1xQ) 27 // 5. Each thread computes one entry in sV[i] 28 // 6. Must sync before and after call 29 // 7. Note that the layout for U and V is different from 2D/3D problem 30 31 if (tx < Q) { 32 for (int comp = 0; comp < NUM_COMP; comp++) { 33 T rv = 0.0; 34 for (int i = 0; i < P; i++) { 35 rv += sU[comp][i] * sT(i, tx); 36 } 37 sV[comp][tx] = rv; 38 } 39 } 40 } 41 42 //////////////////////////////////////////////////////////////////////////////// 43 extern "C" __launch_bounds__(MAGMA_BASIS_BOUNDS(BASIS_MAX_P_Q, MAGMA_MAXTHREADS_1D)) __global__ 44 void magma_gradn_1d_kernel(const CeedScalar *dTinterp, const CeedScalar *dTgrad, const CeedScalar *dU, const int estrdU, const int cstrdU, 45 const int dstrdU, CeedScalar *dV, const int estrdV, const int cstrdV, const int dstrdV, const int nelem) { 46 MAGMA_DEVICE_SHARED(CeedScalar, shared_data) 47 48 const int tx = threadIdx.x; 49 const int ty = threadIdx.y; 50 const int elem_id = (blockIdx.x * blockDim.y) + ty; 51 52 if (elem_id >= nelem) return; 53 54 CeedScalar *sU[BASIS_NUM_COMP]; 55 CeedScalar *sV[BASIS_NUM_COMP]; 56 57 // shift global memory pointers by elem stride 58 dU += elem_id * estrdU; 59 dV += elem_id * estrdV; 60 61 // assign shared memory pointers 62 CeedScalar *sT = (CeedScalar *)shared_data; 63 CeedScalar *sW = sT + BASIS_P * BASIS_Q; 64 sU[0] = sW + ty * BASIS_NUM_COMP * (BASIS_P + BASIS_Q); 65 sV[0] = sU[0] + (BASIS_NUM_COMP * 1 * BASIS_P); 66 for (int comp = 1; comp < BASIS_NUM_COMP; comp++) { 67 sU[comp] = sU[comp - 1] + (1 * BASIS_P); 68 sV[comp] = sV[comp - 1] + (1 * BASIS_Q); 69 } 70 71 // read T 72 if (ty == 0) { 73 read_T_notrans_gm2sm<BASIS_P, BASIS_Q>(tx, dTgrad, sT); 74 } 75 76 // read U 77 read_1d<CeedScalar, BASIS_P, BASIS_NUM_COMP>(dU, cstrdU, sU, tx); 78 79 __syncthreads(); 80 magma_grad_1d_device<CeedScalar, BASIS_DIM, BASIS_NUM_COMP, BASIS_P, BASIS_Q>(sT, sU, sV, tx); 81 __syncthreads(); 82 83 // write V 84 write_1d<CeedScalar, BASIS_Q, BASIS_NUM_COMP>(sV, dV, cstrdV, tx); 85 } 86 87 //////////////////////////////////////////////////////////////////////////////// 88 extern "C" __launch_bounds__(MAGMA_BASIS_BOUNDS(BASIS_MAX_P_Q, MAGMA_MAXTHREADS_1D)) __global__ 89 void magma_gradt_1d_kernel(const CeedScalar *dTinterp, const CeedScalar *dTgrad, const CeedScalar *dU, const int estrdU, const int cstrdU, 90 const int dstrdU, CeedScalar *dV, const int estrdV, const int cstrdV, const int dstrdV, const int nelem) { 91 MAGMA_DEVICE_SHARED(CeedScalar, shared_data) 92 93 const int tx = threadIdx.x; 94 const int ty = threadIdx.y; 95 const int elem_id = (blockIdx.x * blockDim.y) + ty; 96 97 if (elem_id >= nelem) return; 98 99 CeedScalar *sU[BASIS_NUM_COMP]; 100 CeedScalar *sV[BASIS_NUM_COMP]; 101 102 // shift global memory pointers by elem stride 103 dU += elem_id * estrdU; 104 dV += elem_id * estrdV; 105 106 // assign shared memory pointers 107 CeedScalar *sT = (CeedScalar *)shared_data; 108 CeedScalar *sW = sT + BASIS_Q * BASIS_P; 109 sU[0] = sW + ty * BASIS_NUM_COMP * (BASIS_Q + BASIS_P); 110 sV[0] = sU[0] + (BASIS_NUM_COMP * 1 * BASIS_Q); 111 for (int comp = 1; comp < BASIS_NUM_COMP; comp++) { 112 sU[comp] = sU[comp - 1] + (1 * BASIS_Q); 113 sV[comp] = sV[comp - 1] + (1 * BASIS_P); 114 } 115 116 // read T 117 if (ty == 0) { 118 read_T_trans_gm2sm<BASIS_Q, BASIS_P>(tx, dTgrad, sT); 119 } 120 121 // read U 122 read_1d<CeedScalar, BASIS_Q, BASIS_NUM_COMP>(dU, cstrdU, sU, tx); 123 124 __syncthreads(); 125 magma_grad_1d_device<CeedScalar, BASIS_DIM, BASIS_NUM_COMP, BASIS_Q, BASIS_P>(sT, sU, sV, tx); 126 __syncthreads(); 127 128 // write V 129 write_1d<CeedScalar, BASIS_P, BASIS_NUM_COMP>(sV, dV, cstrdV, tx); 130 } 131 132 #endif // CEED_MAGMA_BASIS_GRAD_1D_H 133