xref: /libCEED/backends/magma/ceed-magma.h (revision 014ec18d8ea934e4fc0f69a8c06ffce36a53ff2a)
1 // Copyright (c) 2017, Lawrence Livermore National Security, LLC. Produced at
2 // the Lawrence Livermore National Laboratory. LLNL-CODE-734707. All Rights
3 // reserved. See files LICENSE and NOTICE for details.
4 //
5 // This file is part of CEED, a collection of benchmarks, miniapps, software
6 // libraries and APIs for efficient high-order finite element and spectral
7 // element discretizations for exascale applications. For more information and
8 // source code availability see http://github.com/ceed.
9 //
10 // The CEED research is supported by the Exascale Computing Project 17-SC-20-SC,
11 // a collaborative effort of two U.S. Department of Energy organizations (Office
12 // of Science and the National Nuclear Security Administration) responsible for
13 // the planning and preparation of a capable exascale ecosystem, including
14 // software, applications, hardware, advanced system engineering and early
15 // testbed platforms, in support of the nation's exascale computing imperative.
16 
17 // magma functions specific to ceed
18 
19 #include <string.h>
20 #include <ceed-backend.h>
21 #include "magma.h"
22 
23 typedef struct {
24   CeedScalar *dqref1d;
25   CeedScalar *dinterp1d;
26   CeedScalar *dgrad1d;
27   CeedScalar *dqweight1d;
28 } CeedBasis_Magma;
29 
30 typedef struct {
31   CeedScalar *dqref;
32   CeedScalar *dinterp;
33   CeedScalar *dgrad;
34   CeedScalar *dqweight;
35 } CeedBasisNonTensor_Magma;
36 
37 typedef struct {
38   CeedInt *indices;
39   CeedInt *dindices;
40   int  own_;
41   int down_;            // cover a case where we own Device memory
42 } CeedElemRestriction_Magma;
43 
44 typedef struct {
45   const CeedScalar **inputs;
46   CeedScalar **outputs;
47   bool setupdone;
48 } CeedQFunction_Magma;
49 
50 #define USE_MAGMA_BATCH
51 #define USE_MAGMA_BATCH2
52 #define USE_MAGMA_BATCH3
53 #define USE_MAGMA_BATCH4
54 
55 #ifdef __cplusplus
56 CEED_INTERN {
57 #endif
58   void magmablas_dbasis_apply_batched_eval_interp(magma_int_t P, magma_int_t Q,
59       magma_int_t dim, magma_int_t ncomp,
60       const double *dT, CeedTransposeMode tmode,
61       const double *dU, magma_int_t u_elemstride,
62       magma_int_t u_compstride,
63       double *dV, magma_int_t v_elemstride,
64       magma_int_t v_compstride,
65       magma_int_t nelem);
66 
67   void magmablas_dbasis_apply_batched_eval_grad(magma_int_t P, magma_int_t Q,
68       magma_int_t dim, magma_int_t ncomp,
69       magma_int_t nqpt, const double* dinterp1d,
70       const double *dgrad1d, CeedTransposeMode tmode,
71       const double *dU, magma_int_t u_elemstride,
72       magma_int_t u_compstride, magma_int_t u_dimstride,
73       double *dV, magma_int_t v_elemstride,
74       magma_int_t v_compstride, magma_int_t v_dimstride,
75       magma_int_t dim_id, magma_int_t nelem);
76 
77   void magmablas_dbasis_apply_batched_eval_weight(magma_int_t Q, magma_int_t dim,
78       const double *dqweight1d, double *dV,
79       magma_int_t v_elemstride,
80       magma_int_t nelem);
81 
82   void magma_weight(magma_int_t grid, magma_int_t threads, magma_int_t nelem,
83                     magma_int_t Q,
84                     double *dqweight, double *dv);
85 
86   void magma_readDofsOffset(const magma_int_t NCOMP,
87                             const magma_int_t compstride,
88                             const magma_int_t esize, const magma_int_t nelem,
89                             magma_int_t *indices, const double *du, double *dv);
90 
91   void magma_readDofsStrided(const magma_int_t NCOMP, const magma_int_t esize,
92                              const magma_int_t nelem, magma_int_t *strides,
93                              const double *du, double *dv);
94 
95   void magma_writeDofsOffset(const magma_int_t NCOMP,
96                              const magma_int_t compstride,
97                              const magma_int_t esize, const magma_int_t nelem,
98                              magma_int_t *indices,const double *du, double *dv);
99 
100   void magma_writeDofsStrided(const magma_int_t NCOMP, const magma_int_t esize,
101                               const magma_int_t nelem, magma_int_t *strides,
102                               const double *du, double *dv);
103 
104   magma_int_t
105   magma_isdevptr(const void *A);
106 
107   int CeedBasisCreateTensorH1_Magma(CeedInt dim, CeedInt P1d,
108                                     CeedInt Q1d,
109                                     const CeedScalar *interp1d,
110                                     const CeedScalar *grad1d,
111                                     const CeedScalar *qref1d,
112                                     const CeedScalar *qweight1d,
113                                     CeedBasis basis);
114 
115   int CeedBasisCreateH1_Magma(CeedElemTopology topo, CeedInt dim,
116                               CeedInt ndof, CeedInt nqpts,
117                               const CeedScalar *interp,
118                               const CeedScalar *grad,
119                               const CeedScalar *qref,
120                               const CeedScalar *qweight,
121                               CeedBasis basis);
122 
123   int CeedElemRestrictionCreate_Magma(CeedMemType mtype,
124                                       CeedCopyMode cmode,
125                                       const CeedInt *indices,
126                                       CeedElemRestriction r);
127 
128   int CeedElemRestrictionCreateBlocked_Magma(const CeedMemType mtype,
129       const CeedCopyMode cmode,
130       const CeedInt *indices,
131       const CeedElemRestriction res);
132   #ifdef __cplusplus
133 }
134   #endif
135 
136 #define CeedDebug(...)
137 //#define CeedDebug(format, ...) fprintf(stderr, format, ## __VA_ARGS__)
138 
139 // comment the line below to use the default magma_is_devptr function
140 #define magma_is_devptr magma_isdevptr
141 
142 // batch stride, override using -DMAGMA_BATCH_STRIDE=<desired-value>
143 #ifndef MAGMA_BATCH_STRIDE
144 #define MAGMA_BATCH_STRIDE (1000)
145 #endif
146