1 /* Copyright (c) 2009-2020. The SimGrid Team. All rights reserved. */
3 /* This program is free software; you can redistribute it and/or modify it
4 * under the terms of the license (GNU LGPL) which comes with this package. */
8 #include "smpi_datatype.hpp"
9 #include "src/smpi/include/smpi_actor.hpp"
11 XBT_LOG_NEW_DEFAULT_SUBCATEGORY(smpi_op, smpi, "Logging specific to SMPI (op)");
13 #define MAX_OP(a, b) (b) = (a) < (b) ? (b) : (a)
14 #define MIN_OP(a, b) (b) = (a) < (b) ? (a) : (b)
15 #define SUM_OP(a, b) (b) += (a)
16 #define SUM_OP_COMPLEX(a, b) \
18 ((b).value) += ((a).value); \
19 ((b).index) += ((a).index); \
21 #define PROD_OP(a, b) (b) *= (a)
22 #define PROD_OP_COMPLEX(a, b) \
24 ((b).value) *= ((a).value); \
25 ((b).index) *= ((a).index); \
27 #define LAND_OP(a, b) (b) = (a) && (b)
28 #define LOR_OP(a, b) (b) = (a) || (b)
29 #define LXOR_OP(a, b) (b) = (not(a) && (b)) || ((a) && not(b))
30 #define BAND_OP(a, b) (b) &= (a)
31 #define BOR_OP(a, b) (b) |= (a)
32 #define BXOR_OP(a, b) (b) ^= (a)
33 #define MAXLOC_OP(a, b) \
34 (b) = ((a).value) < ((b).value) ? (b) : (((a).value) == ((b).value) ? (((a).index) < ((b).index) ? (a) : (b)) : (a))
35 #define MINLOC_OP(a, b) \
36 (b) = ((a).value) < ((b).value) ? (a) : (((a).value) == ((b).value) ? (((a).index) < ((b).index) ? (a) : (b)) : (b))
38 #define APPLY_FUNC(a, b, length, type, func) \
41 type* x = (type*)(a); \
42 type* y = (type*)(b); \
43 for(i = 0; i < *(length); i++) { \
48 #define APPLY_OP_LOOP(dtype, type, op) \
49 if (*datatype == (dtype)) { \
50 APPLY_FUNC(a, b, length, type, op) \
53 #define APPLY_BASIC_OP_LOOP(op)\
54 APPLY_OP_LOOP(MPI_CHAR, char,op)\
55 APPLY_OP_LOOP(MPI_SHORT, short,op)\
56 APPLY_OP_LOOP(MPI_INT, int,op)\
57 APPLY_OP_LOOP(MPI_LONG, long,op)\
58 APPLY_OP_LOOP(MPI_LONG_LONG, long long,op)\
59 APPLY_OP_LOOP(MPI_SIGNED_CHAR, signed char,op)\
60 APPLY_OP_LOOP(MPI_UNSIGNED_CHAR, unsigned char,op)\
61 APPLY_OP_LOOP(MPI_UNSIGNED_SHORT, unsigned short,op)\
62 APPLY_OP_LOOP(MPI_UNSIGNED, unsigned int,op)\
63 APPLY_OP_LOOP(MPI_UNSIGNED_LONG, unsigned long,op)\
64 APPLY_OP_LOOP(MPI_UNSIGNED_LONG_LONG, unsigned long long,op)\
65 APPLY_OP_LOOP(MPI_WCHAR, wchar_t,op)\
66 APPLY_OP_LOOP(MPI_BYTE, int8_t,op)\
67 APPLY_OP_LOOP(MPI_INT8_T, int8_t,op)\
68 APPLY_OP_LOOP(MPI_INT16_T, int16_t,op)\
69 APPLY_OP_LOOP(MPI_INT32_T, int32_t,op)\
70 APPLY_OP_LOOP(MPI_INT64_T, int64_t,op)\
71 APPLY_OP_LOOP(MPI_UINT8_T, uint8_t,op)\
72 APPLY_OP_LOOP(MPI_UINT16_T, uint16_t,op)\
73 APPLY_OP_LOOP(MPI_UINT32_T, uint32_t,op)\
74 APPLY_OP_LOOP(MPI_UINT64_T, uint64_t,op)\
75 APPLY_OP_LOOP(MPI_AINT, MPI_Aint,op)\
76 APPLY_OP_LOOP(MPI_OFFSET, MPI_Offset,op)\
77 APPLY_OP_LOOP(MPI_INTEGER1, int,op)\
78 APPLY_OP_LOOP(MPI_INTEGER2, int16_t,op)\
79 APPLY_OP_LOOP(MPI_INTEGER4, int32_t,op)\
80 APPLY_OP_LOOP(MPI_INTEGER8, int64_t,op)\
81 APPLY_OP_LOOP(MPI_COUNT, long long,op)
84 #define APPLY_BOOL_OP_LOOP(op)\
85 APPLY_OP_LOOP(MPI_C_BOOL, bool,op)
87 #define APPLY_FLOAT_OP_LOOP(op)\
88 APPLY_OP_LOOP(MPI_FLOAT, float,op)\
89 APPLY_OP_LOOP(MPI_DOUBLE, double,op)\
90 APPLY_OP_LOOP(MPI_LONG_DOUBLE, long double,op)\
91 APPLY_OP_LOOP(MPI_REAL, float,op)\
92 APPLY_OP_LOOP(MPI_REAL4, float,op)\
93 APPLY_OP_LOOP(MPI_REAL8, double,op)\
94 APPLY_OP_LOOP(MPI_REAL16, long double,op)
96 #define APPLY_COMPLEX_OP_LOOP(op)\
97 APPLY_OP_LOOP(MPI_C_FLOAT_COMPLEX, float _Complex,op)\
98 APPLY_OP_LOOP(MPI_C_DOUBLE_COMPLEX, double _Complex,op)\
99 APPLY_OP_LOOP(MPI_C_LONG_DOUBLE_COMPLEX, long double _Complex,op)
101 #define APPLY_PAIR_OP_LOOP(op)\
102 APPLY_OP_LOOP(MPI_FLOAT_INT, float_int,op)\
103 APPLY_OP_LOOP(MPI_LONG_INT, long_int,op)\
104 APPLY_OP_LOOP(MPI_DOUBLE_INT, double_int,op)\
105 APPLY_OP_LOOP(MPI_SHORT_INT, short_int,op)\
106 APPLY_OP_LOOP(MPI_2INT, int_int,op)\
107 APPLY_OP_LOOP(MPI_2FLOAT, float_float,op)\
108 APPLY_OP_LOOP(MPI_2DOUBLE, double_double,op)\
109 APPLY_OP_LOOP(MPI_LONG_DOUBLE_INT, long_double_int,op)\
110 APPLY_OP_LOOP(MPI_2LONG, long_long,op)\
111 APPLY_OP_LOOP(MPI_COMPLEX8, float_float,op)\
112 APPLY_OP_LOOP(MPI_COMPLEX16, double_double,op)\
113 APPLY_OP_LOOP(MPI_COMPLEX32, double_double,op)
115 #define APPLY_END_OP_LOOP(op) \
117 xbt_die("Failed to apply " _XBT_STRINGIFY(op) " to type %s", (*datatype)->name()); \
120 static void max_func(void *a, void *b, int *length, MPI_Datatype * datatype)
122 APPLY_BASIC_OP_LOOP(MAX_OP)
123 APPLY_FLOAT_OP_LOOP(MAX_OP)
124 APPLY_END_OP_LOOP(MAX_OP)
127 static void min_func(void *a, void *b, int *length, MPI_Datatype * datatype)
129 APPLY_BASIC_OP_LOOP(MIN_OP)
130 APPLY_FLOAT_OP_LOOP(MIN_OP)
131 APPLY_END_OP_LOOP(MIN_OP)
134 static void sum_func(void *a, void *b, int *length, MPI_Datatype * datatype)
136 APPLY_BASIC_OP_LOOP(SUM_OP)
137 APPLY_FLOAT_OP_LOOP(SUM_OP)
138 APPLY_COMPLEX_OP_LOOP(SUM_OP)
139 APPLY_PAIR_OP_LOOP(SUM_OP_COMPLEX)
140 APPLY_END_OP_LOOP(SUM_OP)
143 static void prod_func(void *a, void *b, int *length, MPI_Datatype * datatype)
145 APPLY_BASIC_OP_LOOP(PROD_OP)
146 APPLY_FLOAT_OP_LOOP(PROD_OP)
147 APPLY_COMPLEX_OP_LOOP(PROD_OP)
148 APPLY_PAIR_OP_LOOP(PROD_OP_COMPLEX)
149 APPLY_END_OP_LOOP(PROD_OP)
152 static void land_func(void *a, void *b, int *length, MPI_Datatype * datatype)
154 APPLY_BASIC_OP_LOOP(LAND_OP)
155 APPLY_FLOAT_OP_LOOP(LAND_OP)
156 APPLY_BOOL_OP_LOOP(LAND_OP)
157 APPLY_END_OP_LOOP(LAND_OP)
160 static void lor_func(void *a, void *b, int *length, MPI_Datatype * datatype)
162 APPLY_BASIC_OP_LOOP(LOR_OP)
163 APPLY_FLOAT_OP_LOOP(LOR_OP)
164 APPLY_BOOL_OP_LOOP(LOR_OP)
165 APPLY_END_OP_LOOP(LOR_OP)
168 static void lxor_func(void *a, void *b, int *length, MPI_Datatype * datatype)
170 APPLY_BASIC_OP_LOOP(LXOR_OP)
171 APPLY_FLOAT_OP_LOOP(LXOR_OP)
172 APPLY_BOOL_OP_LOOP(LXOR_OP)
173 APPLY_END_OP_LOOP(LXOR_OP)
176 static void band_func(void *a, void *b, int *length, MPI_Datatype * datatype)
178 APPLY_BASIC_OP_LOOP(BAND_OP)
179 APPLY_BOOL_OP_LOOP(BAND_OP)
180 APPLY_END_OP_LOOP(BAND_OP)
183 static void bor_func(void *a, void *b, int *length, MPI_Datatype * datatype)
185 APPLY_BASIC_OP_LOOP(BOR_OP)
186 APPLY_BOOL_OP_LOOP(BOR_OP)
187 APPLY_END_OP_LOOP(BOR_OP)
190 static void bxor_func(void *a, void *b, int *length, MPI_Datatype * datatype)
192 APPLY_BASIC_OP_LOOP(BXOR_OP)
193 APPLY_BOOL_OP_LOOP(BXOR_OP)
194 APPLY_END_OP_LOOP(BXOR_OP)
197 static void minloc_func(void *a, void *b, int *length, MPI_Datatype * datatype)
199 APPLY_PAIR_OP_LOOP(MINLOC_OP)
200 APPLY_END_OP_LOOP(MINLOC_OP)
203 static void maxloc_func(void *a, void *b, int *length, MPI_Datatype * datatype)
205 APPLY_PAIR_OP_LOOP(MAXLOC_OP)
206 APPLY_END_OP_LOOP(MAXLOC_OP)
209 static void replace_func(void *a, void *b, int *length, MPI_Datatype * datatype)
211 memcpy(b, a, *length * (*datatype)->size());
214 static void no_func(void*, void*, int*, MPI_Datatype*)
216 /* obviously a no-op */
219 #define CREATE_MPI_OP(name, func) \
220 static SMPI_Op _XBT_CONCAT(mpi_, name)(&(func) /* func */, true, true); \
221 MPI_Op name = &_XBT_CONCAT(mpi_, name);
223 CREATE_MPI_OP(MPI_MAX, max_func)
224 CREATE_MPI_OP(MPI_MIN, min_func)
225 CREATE_MPI_OP(MPI_SUM, sum_func)
226 CREATE_MPI_OP(MPI_PROD, prod_func)
227 CREATE_MPI_OP(MPI_LAND, land_func)
228 CREATE_MPI_OP(MPI_LOR, lor_func)
229 CREATE_MPI_OP(MPI_LXOR, lxor_func)
230 CREATE_MPI_OP(MPI_BAND, band_func)
231 CREATE_MPI_OP(MPI_BOR, bor_func)
232 CREATE_MPI_OP(MPI_BXOR, bxor_func)
233 CREATE_MPI_OP(MPI_MAXLOC, maxloc_func)
234 CREATE_MPI_OP(MPI_MINLOC, minloc_func)
235 CREATE_MPI_OP(MPI_REPLACE, replace_func)
236 CREATE_MPI_OP(MPI_NO_OP, no_func)
241 void Op::apply(const void* invec, void* inoutvec, const int* len, MPI_Datatype datatype)
243 if (smpi_cfg_privatization() == SmpiPrivStrategies::MMAP) {
244 // we need to switch as the called function may silently touch global variables
245 XBT_DEBUG("Applying operation, switch to the right data frame ");
246 smpi_switch_data_segment(simgrid::s4u::Actor::self());
249 if (not smpi_process()->replaying() && *len > 0) {
250 if (not is_fortran_op_)
251 this->func_(const_cast<void*>(invec), inoutvec, const_cast<int*>(len), &datatype);
253 XBT_DEBUG("Applying operation of length %d from %p and from/to %p", *len, invec, inoutvec);
254 int tmp = datatype->c2f();
255 /* Unfortunately, the C and Fortran version of the MPI standard do not agree on the type here,
256 thus the reinterpret_cast. */
257 this->func_(const_cast<void*>(invec), inoutvec, const_cast<int*>(len), reinterpret_cast<MPI_Datatype*>(&tmp));
263 return static_cast<Op*>(F2C::f2c(id));
270 void Op::unref(MPI_Op* op){
271 if((*op)!=MPI_OP_NULL){
273 if((*op)->refcount_==0 && (*op)->predefined_==false)