xref: /libCEED/rust/libceed-sys/c-src/backends/cuda-ref/ceed-cuda-ref-qfunction.c (revision 0d0321e0e600f17fbb9528732fcb5c1d5c63fc0f)
1 // Copyright (c) 2017-2018, Lawrence Livermore National Security, LLC.
2 // Produced at the Lawrence Livermore National Laboratory. LLNL-CODE-734707.
3 // All Rights 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 #include <ceed/ceed.h>
18 #include <ceed/backend.h>
19 #include <cuda.h>
20 #include <stdio.h>
21 #include <string.h>
22 #include "ceed-cuda-ref.h"
23 #include "ceed-cuda-ref-qfunction-load.h"
24 #include "../cuda/ceed-cuda-compile.h"
25 
26 //------------------------------------------------------------------------------
27 // Apply QFunction
28 //------------------------------------------------------------------------------
29 static int CeedQFunctionApply_Cuda(CeedQFunction qf, CeedInt Q,
30                                    CeedVector *U, CeedVector *V) {
31   int ierr;
32   Ceed ceed;
33   ierr = CeedQFunctionGetCeed(qf, &ceed); CeedChkBackend(ierr);
34 
35   // Build and compile kernel, if not done
36   ierr = CeedCudaBuildQFunction(qf); CeedChkBackend(ierr);
37 
38   CeedQFunction_Cuda *data;
39   ierr = CeedQFunctionGetData(qf, &data); CeedChkBackend(ierr);
40   Ceed_Cuda *ceed_Cuda;
41   ierr = CeedGetData(ceed, &ceed_Cuda); CeedChkBackend(ierr);
42   CeedInt numinputfields, numoutputfields;
43   ierr = CeedQFunctionGetNumArgs(qf, &numinputfields, &numoutputfields);
44   CeedChkBackend(ierr);
45 
46   // Read vectors
47   for (CeedInt i = 0; i < numinputfields; i++) {
48     ierr = CeedVectorGetArrayRead(U[i], CEED_MEM_DEVICE, &data->fields.inputs[i]);
49     CeedChkBackend(ierr);
50   }
51   for (CeedInt i = 0; i < numoutputfields; i++) {
52     ierr = CeedVectorGetArrayWrite(V[i], CEED_MEM_DEVICE, &data->fields.outputs[i]);
53     CeedChkBackend(ierr);
54   }
55 
56   // Get context data
57   CeedQFunctionContext ctx;
58   ierr = CeedQFunctionGetInnerContext(qf, &ctx); CeedChkBackend(ierr);
59   if (ctx) {
60     ierr = CeedQFunctionContextGetData(ctx, CEED_MEM_DEVICE, &data->d_c);
61     CeedChkBackend(ierr);
62   }
63 
64   // Run kernel
65   void *args[] = {&data->d_c, (void *) &Q, &data->fields};
66   ierr = CeedRunKernelAutoblockCuda(ceed, data->qFunction, Q, args);
67   CeedChkBackend(ierr);
68 
69   // Restore vectors
70   for (CeedInt i = 0; i < numinputfields; i++) {
71     ierr = CeedVectorRestoreArrayRead(U[i], &data->fields.inputs[i]);
72     CeedChkBackend(ierr);
73   }
74   for (CeedInt i = 0; i < numoutputfields; i++) {
75     ierr = CeedVectorRestoreArray(V[i], &data->fields.outputs[i]);
76     CeedChkBackend(ierr);
77   }
78 
79   // Restore context
80   if (ctx) {
81     ierr = CeedQFunctionContextRestoreData(ctx, &data->d_c);
82     CeedChkBackend(ierr);
83   }
84   return CEED_ERROR_SUCCESS;
85 }
86 
87 //------------------------------------------------------------------------------
88 // Destroy QFunction
89 //------------------------------------------------------------------------------
90 static int CeedQFunctionDestroy_Cuda(CeedQFunction qf) {
91   int ierr;
92   CeedQFunction_Cuda *data;
93   ierr = CeedQFunctionGetData(qf, &data); CeedChkBackend(ierr);
94   Ceed ceed;
95   ierr = CeedQFunctionGetCeed(qf, &ceed); CeedChkBackend(ierr);
96   if  (data->module)
97     CeedChk_Cu(ceed, cuModuleUnload(data->module));
98   ierr = CeedFree(&data); CeedChkBackend(ierr);
99   return CEED_ERROR_SUCCESS;
100 }
101 
102 //------------------------------------------------------------------------------
103 // Set User QFunction
104 //------------------------------------------------------------------------------
105 static int CeedQFunctionSetCUDAUserFunction_Cuda(CeedQFunction qf,
106     CUfunction f) {
107   int ierr;
108   CeedQFunction_Cuda *data;
109   ierr = CeedQFunctionGetData(qf, &data); CeedChkBackend(ierr);
110   data->qFunction = f;
111   return CEED_ERROR_SUCCESS;
112 }
113 
114 //------------------------------------------------------------------------------
115 // Create QFunction
116 //------------------------------------------------------------------------------
117 int CeedQFunctionCreate_Cuda(CeedQFunction qf) {
118   int ierr;
119   Ceed ceed;
120   CeedQFunctionGetCeed(qf, &ceed);
121   CeedQFunction_Cuda *data;
122   ierr = CeedCalloc(1, &data); CeedChkBackend(ierr);
123   ierr = CeedQFunctionSetData(qf, data); CeedChkBackend(ierr);
124 
125   // Read QFunction source
126   ierr = CeedQFunctionGetKernelName(qf, &data->qFunctionName);
127   CeedChkBackend(ierr);
128   ierr = CeedQFunctionLoadSourceToBuffer(qf, &data->qFunctionSource);
129   CeedChkBackend(ierr);
130 
131   // Register backend functions
132   ierr = CeedSetBackendFunction(ceed, "QFunction", qf, "Apply",
133                                 CeedQFunctionApply_Cuda); CeedChkBackend(ierr);
134   ierr = CeedSetBackendFunction(ceed, "QFunction", qf, "Destroy",
135                                 CeedQFunctionDestroy_Cuda); CeedChkBackend(ierr);
136   ierr = CeedSetBackendFunction(ceed, "QFunction", qf, "SetCUDAUserFunction",
137                                 CeedQFunctionSetCUDAUserFunction_Cuda);
138   CeedChkBackend(ierr);
139   return CEED_ERROR_SUCCESS;
140 }
141 //------------------------------------------------------------------------------
142