elpa_constants.h.in 6.21 KB
Newer Older
1
2
3
4
5
6
7
#pragma once

/* This might seem over-engineered, but helps to re-use this file also on the
 * Fortran side and thus to keep the definitions in this one place here
 */

/* Private helper macros */
8
#define ELPA_ENUM_ENTRY(name, value, ...) \
9
        name = value,
10
#define ELPA_ENUM_SUM(name, value, ...) +1
11
12
13


/* Solver constants */
14
#define ELPA_FOR_ALL_SOLVERS(X) \
15
16
17
18
        X(ELPA_SOLVER_1STAGE, 1) \
        X(ELPA_SOLVER_2STAGE, 2)

enum ELPA_SOLVERS {
19
        ELPA_FOR_ALL_SOLVERS(ELPA_ENUM_ENTRY)
20
21
};

22
#define ELPA_NUMBER_OF_SOLVERS (0 ELPA_FOR_ALL_SOLVERS(ELPA_ENUM_SUM))
23
24

/* Kernel constants */
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
#define ELPA_FOR_ALL_2STAGE_REAL_KERNELS(X, ...) \
        X(ELPA_2STAGE_REAL_GENERIC, 1, @ELPA_2STAGE_REAL_GENERIC_COMPILED@, __VA_ARGS__) \
        X(ELPA_2STAGE_REAL_GENERIC_SIMPLE, 2, @ELPA_2STAGE_REAL_GENERIC_SIMPLE_COMPILED@, __VA_ARGS__) \
        X(ELPA_2STAGE_REAL_BGP, 3, @ELPA_2STAGE_REAL_BGP_COMPILED@, __VA_ARGS__) \
        X(ELPA_2STAGE_REAL_BGQ, 4, @ELPA_2STAGE_REAL_BGQ_COMPILED@, __VA_ARGS__) \
        X(ELPA_2STAGE_REAL_SSE_ASSEMBLY, 5, @ELPA_2STAGE_REAL_SSE_ASSEMBLY_COMPILED@, __VA_ARGS__) \
        X(ELPA_2STAGE_REAL_SSE_BLOCK2, 6, @ELPA_2STAGE_REAL_SSE_BLOCK2_COMPILED@, __VA_ARGS__) \
        X(ELPA_2STAGE_REAL_SSE_BLOCK4, 7, @ELPA_2STAGE_REAL_SSE_BLOCK4_COMPILED@, __VA_ARGS__) \
        X(ELPA_2STAGE_REAL_SSE_BLOCK6, 8, @ELPA_2STAGE_REAL_SSE_BLOCK6_COMPILED@, __VA_ARGS__) \
        X(ELPA_2STAGE_REAL_AVX_BLOCK2, 9, @ELPA_2STAGE_REAL_AVX_BLOCK2_COMPILED@, __VA_ARGS__) \
        X(ELPA_2STAGE_REAL_AVX_BLOCK4, 10, @ELPA_2STAGE_REAL_AVX_BLOCK4_COMPILED@, __VA_ARGS__) \
        X(ELPA_2STAGE_REAL_AVX_BLOCK6, 11, @ELPA_2STAGE_REAL_AVX_BLOCK6_COMPILED@, __VA_ARGS__) \
        X(ELPA_2STAGE_REAL_AVX2_BLOCK2, 12, @ELPA_2STAGE_REAL_AVX2_BLOCK2_COMPILED@, __VA_ARGS__) \
        X(ELPA_2STAGE_REAL_AVX2_BLOCK4, 13, @ELPA_2STAGE_REAL_AVX2_BLOCK4_COMPILED@, __VA_ARGS__) \
        X(ELPA_2STAGE_REAL_AVX2_BLOCK6, 14, @ELPA_2STAGE_REAL_AVX2_BLOCK6_COMPILED@, __VA_ARGS__) \
        X(ELPA_2STAGE_REAL_AVX512_BLOCK2, 15, @ELPA_2STAGE_REAL_AVX512_BLOCK2_COMPILED@, __VA_ARGS__) \
        X(ELPA_2STAGE_REAL_AVX512_BLOCK4, 16, @ELPA_2STAGE_REAL_AVX512_BLOCK4_COMPILED@, __VA_ARGS__) \
        X(ELPA_2STAGE_REAL_AVX512_BLOCK6, 17, @ELPA_2STAGE_REAL_AVX512_BLOCK6_COMPILED@, __VA_ARGS__) \
43
44
45
        X(ELPA_2STAGE_REAL_GPU, 18, @ELPA_2STAGE_REAL_GPU_COMPILED@, __VA_ARGS__) \
        X(ELPA_2STAGE_REAL_SPARC64_BLOCK2, 19, @ELPA_2STAGE_REAL_SPARC64_BLOCK2_COMPILED@, __VA_ARGS__) \
        X(ELPA_2STAGE_REAL_SPARC64_BLOCK4, 20, @ELPA_2STAGE_REAL_SPARC64_BLOCK4_COMPILED@, __VA_ARGS__) \
46
47
48
49
        X(ELPA_2STAGE_REAL_SPARC64_BLOCK6, 21, @ELPA_2STAGE_REAL_SPARC64_BLOCK6_COMPILED@, __VA_ARGS__) \
        X(ELPA_2STAGE_REAL_VSX_BLOCK2, 22, @ELPA_2STAGE_REAL_VSX_BLOCK2_COMPILED@, __VA_ARGS__) \
        X(ELPA_2STAGE_REAL_VSX_BLOCK4, 23, @ELPA_2STAGE_REAL_VSX_BLOCK4_COMPILED@, __VA_ARGS__) \
        X(ELPA_2STAGE_REAL_VSX_BLOCK6, 24, @ELPA_2STAGE_REAL_VSX_BLOCK6_COMPILED@, __VA_ARGS__)
50
51
52
53
54

#define ELPA_FOR_ALL_2STAGE_REAL_KERNELS_AND_DEFAULT(X) \
        ELPA_FOR_ALL_2STAGE_REAL_KERNELS(X) \
        X(ELPA_2STAGE_REAL_INVALID, -1, choke me) \
        X(ELPA_2STAGE_REAL_DEFAULT, @ELPA_2STAGE_REAL_DEFAULT@, choke me)
55
56

enum ELPA_REAL_KERNELS {
57
        ELPA_FOR_ALL_2STAGE_REAL_KERNELS_AND_DEFAULT(ELPA_ENUM_ENTRY)
58
59
60
};


61
62
63
64
65
66
67
68
69
70
71
72
73
74
#define ELPA_FOR_ALL_2STAGE_COMPLEX_KERNELS(X, ...) \
        X(ELPA_2STAGE_COMPLEX_GENERIC, 1, @ELPA_2STAGE_COMPLEX_GENERIC_COMPILED@, __VA_ARGS__) \
        X(ELPA_2STAGE_COMPLEX_GENERIC_SIMPLE, 2, @ELPA_2STAGE_COMPLEX_GENERIC_SIMPLE_COMPILED@, __VA_ARGS__) \
        X(ELPA_2STAGE_COMPLEX_BGP, 3, @ELPA_2STAGE_COMPLEX_BGP_COMPILED@, __VA_ARGS__) \
        X(ELPA_2STAGE_COMPLEX_BGQ, 4, @ELPA_2STAGE_COMPLEX_BGQ_COMPILED@, __VA_ARGS__) \
        X(ELPA_2STAGE_COMPLEX_SSE_ASSEMBLY, 5, @ELPA_2STAGE_COMPLEX_SSE_ASSEMBLY_COMPILED@, __VA_ARGS__) \
        X(ELPA_2STAGE_COMPLEX_SSE_BLOCK1, 6, @ELPA_2STAGE_COMPLEX_SSE_BLOCK1_COMPILED@, __VA_ARGS__) \
        X(ELPA_2STAGE_COMPLEX_SSE_BLOCK2, 7, @ELPA_2STAGE_COMPLEX_SSE_BLOCK2_COMPILED@, __VA_ARGS__) \
        X(ELPA_2STAGE_COMPLEX_AVX_BLOCK1, 8, @ELPA_2STAGE_COMPLEX_AVX_BLOCK1_COMPILED@, __VA_ARGS__) \
        X(ELPA_2STAGE_COMPLEX_AVX_BLOCK2, 9, @ELPA_2STAGE_COMPLEX_AVX_BLOCK2_COMPILED@, __VA_ARGS__) \
        X(ELPA_2STAGE_COMPLEX_AVX2_BLOCK1, 10, @ELPA_2STAGE_COMPLEX_AVX2_BLOCK1_COMPILED@, __VA_ARGS__) \
        X(ELPA_2STAGE_COMPLEX_AVX2_BLOCK2, 11, @ELPA_2STAGE_COMPLEX_AVX2_BLOCK2_COMPILED@, __VA_ARGS__) \
        X(ELPA_2STAGE_COMPLEX_AVX512_BLOCK1, 12, @ELPA_2STAGE_COMPLEX_AVX512_BLOCK1_COMPILED@, __VA_ARGS__) \
        X(ELPA_2STAGE_COMPLEX_AVX512_BLOCK2, 13, @ELPA_2STAGE_COMPLEX_AVX512_BLOCK2_COMPILED@, __VA_ARGS__) \
75
        X(ELPA_2STAGE_COMPLEX_GPU, 14, @ELPA_2STAGE_COMPLEX_GPU_COMPILED@, __VA_ARGS__) 
76
77
78
79
80

#define ELPA_FOR_ALL_2STAGE_COMPLEX_KERNELS_AND_DEFAULT(X) \
        ELPA_FOR_ALL_2STAGE_COMPLEX_KERNELS(X) \
        X(ELPA_2STAGE_COMPLEX_INVALID, -1, choke me) \
        X(ELPA_2STAGE_COMPLEX_DEFAULT, @ELPA_2STAGE_COMPLEX_DEFAULT@, choke me)
81
82

enum ELPA_COMPLEX_KERNELS {
83
        ELPA_FOR_ALL_2STAGE_COMPLEX_KERNELS_AND_DEFAULT(ELPA_ENUM_ENTRY)
84
85
86
87
88
};



/* General constants */
89
90
#define ELPA_FOR_ALL_ERRORS(X) \
        X(ELPA_OK, 0) \
91
        X(ELPA_ERROR, -1) \
92
        X(ELPA_ERROR_ENTRY_NOT_FOUND, -2) \
93
94
95
96
        X(ELPA_ERROR_ENTRY_INVALID_VALUE, -3) \
        X(ELPA_ERROR_ENTRY_ALREADY_SET, -4) \
        X(ELPA_ERROR_ENTRY_NO_STRING_REPRESENTATION, -5) \
        X(ELPA_ERROR_ENTRY_READONLY, -6)
97

98
99
100
enum ELPA_ERRORS {
        ELPA_FOR_ALL_ERRORS(ELPA_ENUM_ENTRY)
};
101

102
103
104
enum ELPA_CONSTANTS {
        ELPA_2STAGE_NUMBER_OF_COMPLEX_KERNELS = (0 ELPA_FOR_ALL_2STAGE_COMPLEX_KERNELS(ELPA_ENUM_SUM)),
        ELPA_2STAGE_NUMBER_OF_REAL_KERNELS = (0 ELPA_FOR_ALL_2STAGE_REAL_KERNELS(ELPA_ENUM_SUM)),
105
106
};

107
108
109
#define ELPA_FOR_ALL_AUTOTUNE_LEVELS(X, ...) \
        X(ELPA_AUTOTUNE_NOT_TUNABLE, 0) \
        X(ELPA_AUTOTUNE_FAST, 1) \
110
111
        X(ELPA_AUTOTUNE_MEDIUM, 2) \
	X(ELPA_AUTOTUNE_EXTENSIVE, 3)
112
113
114
115
116
117
118
119
120
121
122
123
124
125

enum ELPA_AUTOTUNE_LEVELS {
        ELPA_FOR_ALL_AUTOTUNE_LEVELS(ELPA_ENUM_ENTRY)
};


#define ELPA_FOR_ALL_AUTOTUNE_DOMAINS(X, ...) \
        X(ELPA_AUTOTUNE_DOMAIN_REAL, 1) \
        X(ELPA_AUTOTUNE_DOMAIN_COMPLEX, 2) \
        X(ELPA_AUTOTUNE_DOMAIN_ANY, 3)

enum ELPA_AUTOTUNE_DOMAINS {
        ELPA_FOR_ALL_AUTOTUNE_DOMAINS(ELPA_ENUM_ENTRY)
};