1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
|
/* ************************************************************************
* Copyright 2013 Advanced Micro Devices, Inc.
*
* Licensed under the Apache License, Version 2.0 (the "License");
* you may not use this file except in compliance with the License.
* You may obtain a copy of the License at
*
* http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
* ************************************************************************/
#ifndef SOLVER_H_
#define SOLVER_H_
#include <defbool.h>
#include <cltypes.h>
#include <kerngen.h>
#include <clkern.h>
#include <clBLAS.h>
#include <kernel_extra.h>
struct Kernel;
// OpenCL solver ID
typedef int solver_id_t;
/**
* @internal
* @defgroup SOLVERIF Solver interface
*
* This interface binds the library frontend to the library backend
*/
/*@{*/
/**
* @internal
* @brief Solver flags
*/
typedef enum SolverFlags {
/** supports 1D work space */
SF_WSPACE_1D = 0x01,
/** supports 2D work space */
SF_WSPACE_2D = 0x02,
/** input data blocks at the top level must be square */
SF_TOP_INPUT_SQUARE_BLOCKS = 0x04
} SolverFlags;
typedef enum PatternPerformance{
PPERF_NOT_SUPPORTED = -1,
PPERF_POOR = 0,
PPERF_AVERAGE,
PPERF_GOOD,
PPERF_BEST
} PatternPerformance;
typedef enum CheckCalcPGran{
PGRAN_CHECK = 0,
PGRAN_CALC
} CheckCalcPGran;
/**
* @internal
* @brief type of function generating kernel source for an
* OpenCL based solver
*
* @param[out] buf Pointer to a buffer to store a generated kenrel to
* @param[in] buflen Length of the buffer
* @param[in] subdims Subproblem dimensions to generate an optimal kernel
* @param[in] pgran Data parallelism granularity
* @param[in] extra Generator extra information depending on the
* application fields
*
* If the pointer to the buffer is NULL, the function should just calculate
* needed size of the buffer to fit the code in.
*
* @return size of the generated kernel source on success; negated error code
* otherwise
* - -ENOMEM: enough of memory to allocated internal structures
* - -EOVERFLOW: generated source exceeds the buffer size
* - -EINVAL: invalid argument is passed
*/
typedef ssize_t
(*SolverKgen)(
char *buf,
size_t buflen,
const SubproblemDim *subdims,
const PGranularity *pgran,
void *extra);
/**
* @internal
* @brief Solver operations
*
* The 'args' parameter for 'calcPrepWorkGroups',
* and the second parameter for the 'assignKargs' methods plays the role of pointer
* to a kernel arguments structure depending on the application field.
*/
typedef struct SolverOps {
/** Kernel generator */
SolverKgen genKernel;
/** Assign kernel arguments; the first argument is kernel argument batch
* passed immediately to a kernel */
void (*assignKargs)(KernelArg*, const void* args, const void *extra);
/** Check if available LDS size is enough to fit all needed data at such
* granulation; 'kernelArgs' - kernel arguments depending on the
* application fields */
bool (*isFitToLDS)(
SubproblemDim *dims,
DataType,
cl_ulong ldsSize,
const void *args);
/** Get the pattern`s performance estimation for specified flags,
* arguments and granulation.
* Is used for selecting most suitable pattern current problem */
int (*getPatternPerf)(
unsigned int kflags,
const void *args);
/**
* Inner decomposition axis matching to the fastest moving OpenCL
* work dimension. Used only for those patterns which use 2 dimensional
* decomposition
*/
DecompositionAxis (*innerDecompositionAxis)(const void *args);
/** Calculate number of needed global threads to execute a kernel */
void (*calcThreads)(
size_t threads[2],
const SubproblemDim *subdims,
const PGranularity *pgran,
const void *args,
const void *extra);
/** Set number of lines of the same top level block stored into the image
* together and the direction of blocks storing. A solver that uses images
* and stores data to images by blocks must provide the method */
void (*imgPackMode)(
const void *extra,
const SubproblemDim *subdims,
int dataID,
unsigned int *rate,
clblasOrder *order);
/** Get solver flags */
SolverFlags (*getFlags)(void);
/** Correct problem arguments anr extra kernel parameters
* depending on solver specifics. Basically, a solver should not
* change any arguments that come from the API level to avoid any
* confusing points */
void (*fixupArgs)(void *args, SubproblemDim* pSubDims, void *extra);
/** Function, returning default decomposition for the pattern */
int ( *getDefaultDecomp)(
PGranularity *pgran,
SubproblemDim *subdims,
unsigned int subdimsNum,
void *pArgs);
/** Perform validation of decomposition.
* If "check" flag set to true: validate specified decomposition and
* check, if specified granulation is valid for it.
* If "check" flag set to false: calculate granulation,
* fitting the specified decomposition, if possible */
bool (*checkCalcDecomp)(
PGranularity *pgran,
SubproblemDim *subdims,
unsigned int subdimsNum,
DataType dtype,
int check);
/*
SetBuildOptions
*/
void (*setBuildOptions)( char *buildOptsStr, const void *args);
/*
* selectVectorization
*/
KernelExtraFlags (*selectVectorization)( void *kargs, unsigned int vlen);
} SolverOps;
/*@}*/
#endif /* SOLVER_H_ */
|
