1 /* Copyright (c) 2009-2019. 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) {(b.value) += (a.value);(b.index) += (a.index);}
17 #define PROD_OP(a, b) (b) *= (a)
18 #define PROD_OP_COMPLEX(a, b) {(b.value) *= (a.value);(b.index) *= (a.index);}
19 #define LAND_OP(a, b) (b) = (a) && (b)
20 #define LOR_OP(a, b) (b) = (a) || (b)
21 #define LXOR_OP(a, b) (b) = (not(a) && (b)) || ((a) && not(b))
22 #define BAND_OP(a, b) (b) &= (a)
23 #define BOR_OP(a, b) (b) |= (a)
24 #define BXOR_OP(a, b) (b) ^= (a)
25 #define MAXLOC_OP(a, b) (b) = (a.value) < (b.value) ? (b) : ((a.value) == (b.value) ? ((a.index) < (b.index) ? (a) : (b)) : (a))
26 #define MINLOC_OP(a, b) (b) = (a.value) < (b.value) ? (a) : ((a.value) == (b.value) ? ((a.index) < (b.index) ? (a) : (b)) : (b))
28 #define APPLY_FUNC(a, b, length, type, func) \
31 type* x = (type*)(a); \
32 type* y = (type*)(b); \
33 for(i = 0; i < *(length); i++) { \
38 #define APPLY_OP_LOOP(dtype, type, op) \
39 if (*datatype == dtype) {\
40 APPLY_FUNC(a, b, length, type, op)\
44 #define APPLY_BASIC_OP_LOOP(op)\
45 APPLY_OP_LOOP(MPI_CHAR, char,op)\
46 APPLY_OP_LOOP(MPI_SHORT, short,op)\
47 APPLY_OP_LOOP(MPI_INT, int,op)\
48 APPLY_OP_LOOP(MPI_LONG, long,op)\
49 APPLY_OP_LOOP(MPI_LONG_LONG, long long,op)\
50 APPLY_OP_LOOP(MPI_SIGNED_CHAR, signed char,op)\
51 APPLY_OP_LOOP(MPI_UNSIGNED_CHAR, unsigned char,op)\
52 APPLY_OP_LOOP(MPI_UNSIGNED_SHORT, unsigned short,op)\
53 APPLY_OP_LOOP(MPI_UNSIGNED, unsigned int,op)\
54 APPLY_OP_LOOP(MPI_UNSIGNED_LONG, unsigned long,op)\
55 APPLY_OP_LOOP(MPI_UNSIGNED_LONG_LONG, unsigned long long,op)\
56 APPLY_OP_LOOP(MPI_WCHAR, wchar_t,op)\
57 APPLY_OP_LOOP(MPI_BYTE, int8_t,op)\
58 APPLY_OP_LOOP(MPI_INT8_T, int8_t,op)\
59 APPLY_OP_LOOP(MPI_INT16_T, int16_t,op)\
60 APPLY_OP_LOOP(MPI_INT32_T, int32_t,op)\
61 APPLY_OP_LOOP(MPI_INT64_T, int64_t,op)\
62 APPLY_OP_LOOP(MPI_UINT8_T, uint8_t,op)\
63 APPLY_OP_LOOP(MPI_UINT16_T, uint16_t,op)\
64 APPLY_OP_LOOP(MPI_UINT32_T, uint32_t,op)\
65 APPLY_OP_LOOP(MPI_UINT64_T, uint64_t,op)\
66 APPLY_OP_LOOP(MPI_AINT, MPI_Aint,op)\
67 APPLY_OP_LOOP(MPI_OFFSET, MPI_Offset,op)\
68 APPLY_OP_LOOP(MPI_INTEGER1, int,op)\
69 APPLY_OP_LOOP(MPI_INTEGER2, int16_t,op)\
70 APPLY_OP_LOOP(MPI_INTEGER4, int32_t,op)\
71 APPLY_OP_LOOP(MPI_INTEGER8, int64_t,op)\
72 APPLY_OP_LOOP(MPI_COUNT, long long,op)
75 #define APPLY_BOOL_OP_LOOP(op)\
76 APPLY_OP_LOOP(MPI_C_BOOL, bool,op)
78 #define APPLY_FLOAT_OP_LOOP(op)\
79 APPLY_OP_LOOP(MPI_FLOAT, float,op)\
80 APPLY_OP_LOOP(MPI_DOUBLE, double,op)\
81 APPLY_OP_LOOP(MPI_LONG_DOUBLE, long double,op)\
82 APPLY_OP_LOOP(MPI_REAL, float,op)\
83 APPLY_OP_LOOP(MPI_REAL4, float,op)\
84 APPLY_OP_LOOP(MPI_REAL8, double,op)\
85 APPLY_OP_LOOP(MPI_REAL16, long double,op)
87 #define APPLY_COMPLEX_OP_LOOP(op)\
88 APPLY_OP_LOOP(MPI_C_FLOAT_COMPLEX, float _Complex,op)\
89 APPLY_OP_LOOP(MPI_C_DOUBLE_COMPLEX, double _Complex,op)\
90 APPLY_OP_LOOP(MPI_C_LONG_DOUBLE_COMPLEX, long double _Complex,op)
92 #define APPLY_PAIR_OP_LOOP(op)\
93 APPLY_OP_LOOP(MPI_FLOAT_INT, float_int,op)\
94 APPLY_OP_LOOP(MPI_LONG_INT, long_int,op)\
95 APPLY_OP_LOOP(MPI_DOUBLE_INT, double_int,op)\
96 APPLY_OP_LOOP(MPI_SHORT_INT, short_int,op)\
97 APPLY_OP_LOOP(MPI_2INT, int_int,op)\
98 APPLY_OP_LOOP(MPI_2FLOAT, float_float,op)\
99 APPLY_OP_LOOP(MPI_2DOUBLE, double_double,op)\
100 APPLY_OP_LOOP(MPI_LONG_DOUBLE_INT, long_double_int,op)\
101 APPLY_OP_LOOP(MPI_2LONG, long_long,op)\
102 APPLY_OP_LOOP(MPI_COMPLEX8, float_float,op)\
103 APPLY_OP_LOOP(MPI_COMPLEX16, double_double,op)\
104 APPLY_OP_LOOP(MPI_COMPLEX32, double_double,op)
106 #define APPLY_END_OP_LOOP(op) \
108 xbt_die("Failed to apply " _XBT_STRINGIFY(op) " to type %s", (*datatype)->name()); \
111 static void max_func(void *a, void *b, int *length, MPI_Datatype * datatype)
113 APPLY_BASIC_OP_LOOP(MAX_OP)
114 APPLY_FLOAT_OP_LOOP(MAX_OP)
115 APPLY_END_OP_LOOP(MAX_OP)
118 static void min_func(void *a, void *b, int *length, MPI_Datatype * datatype)
120 APPLY_BASIC_OP_LOOP(MIN_OP)
121 APPLY_FLOAT_OP_LOOP(MIN_OP)
122 APPLY_END_OP_LOOP(MIN_OP)
125 static void sum_func(void *a, void *b, int *length, MPI_Datatype * datatype)
127 APPLY_BASIC_OP_LOOP(SUM_OP)
128 APPLY_FLOAT_OP_LOOP(SUM_OP)
129 APPLY_COMPLEX_OP_LOOP(SUM_OP)
130 APPLY_PAIR_OP_LOOP(SUM_OP_COMPLEX)
131 APPLY_END_OP_LOOP(SUM_OP)
134 static void prod_func(void *a, void *b, int *length, MPI_Datatype * datatype)
136 APPLY_BASIC_OP_LOOP(PROD_OP)
137 APPLY_FLOAT_OP_LOOP(PROD_OP)
138 APPLY_COMPLEX_OP_LOOP(PROD_OP)
139 APPLY_PAIR_OP_LOOP(PROD_OP_COMPLEX)
140 APPLY_END_OP_LOOP(PROD_OP)
143 static void land_func(void *a, void *b, int *length, MPI_Datatype * datatype)
145 APPLY_BASIC_OP_LOOP(LAND_OP)
146 APPLY_FLOAT_OP_LOOP(LAND_OP)
147 APPLY_BOOL_OP_LOOP(LAND_OP)
148 APPLY_END_OP_LOOP(LAND_OP)
151 static void lor_func(void *a, void *b, int *length, MPI_Datatype * datatype)
153 APPLY_BASIC_OP_LOOP(LOR_OP)
154 APPLY_FLOAT_OP_LOOP(LOR_OP)
155 APPLY_BOOL_OP_LOOP(LOR_OP)
156 APPLY_END_OP_LOOP(LOR_OP)
159 static void lxor_func(void *a, void *b, int *length, MPI_Datatype * datatype)
161 APPLY_BASIC_OP_LOOP(LXOR_OP)
162 APPLY_FLOAT_OP_LOOP(LXOR_OP)
163 APPLY_BOOL_OP_LOOP(LXOR_OP)
164 APPLY_END_OP_LOOP(LXOR_OP)
167 static void band_func(void *a, void *b, int *length, MPI_Datatype * datatype)
169 APPLY_BASIC_OP_LOOP(BAND_OP)
170 APPLY_BOOL_OP_LOOP(BAND_OP)
171 APPLY_END_OP_LOOP(BAND_OP)
174 static void bor_func(void *a, void *b, int *length, MPI_Datatype * datatype)
176 APPLY_BASIC_OP_LOOP(BOR_OP)
177 APPLY_BOOL_OP_LOOP(BOR_OP)
178 APPLY_END_OP_LOOP(BOR_OP)
181 static void bxor_func(void *a, void *b, int *length, MPI_Datatype * datatype)
183 APPLY_BASIC_OP_LOOP(BXOR_OP)
184 APPLY_BOOL_OP_LOOP(BXOR_OP)
185 APPLY_END_OP_LOOP(BXOR_OP)
188 static void minloc_func(void *a, void *b, int *length, MPI_Datatype * datatype)
190 APPLY_PAIR_OP_LOOP(MINLOC_OP)
191 APPLY_END_OP_LOOP(MINLOC_OP)
194 static void maxloc_func(void *a, void *b, int *length, MPI_Datatype * datatype)
196 APPLY_PAIR_OP_LOOP(MAXLOC_OP)
197 APPLY_END_OP_LOOP(MAXLOC_OP)
200 static void replace_func(void *a, void *b, int *length, MPI_Datatype * datatype)
202 memcpy(b, a, *length * (*datatype)->size());
205 static void no_func(void*, void*, int*, MPI_Datatype*)
207 /* obviously a no-op */
210 #define CREATE_MPI_OP(name, func) \
211 static SMPI_Op _XBT_CONCAT(mpi_, name)(&(func) /* func */, true, true); \
212 MPI_Op name = &_XBT_CONCAT(mpi_, name);
214 CREATE_MPI_OP(MPI_MAX, max_func);
215 CREATE_MPI_OP(MPI_MIN, min_func);
216 CREATE_MPI_OP(MPI_SUM, sum_func);
217 CREATE_MPI_OP(MPI_PROD, prod_func);
218 CREATE_MPI_OP(MPI_LAND, land_func);
219 CREATE_MPI_OP(MPI_LOR, lor_func);
220 CREATE_MPI_OP(MPI_LXOR, lxor_func);
221 CREATE_MPI_OP(MPI_BAND, band_func);
222 CREATE_MPI_OP(MPI_BOR, bor_func);
223 CREATE_MPI_OP(MPI_BXOR, bxor_func);
224 CREATE_MPI_OP(MPI_MAXLOC, maxloc_func);
225 CREATE_MPI_OP(MPI_MINLOC, minloc_func);
226 CREATE_MPI_OP(MPI_REPLACE, replace_func);
227 CREATE_MPI_OP(MPI_NO_OP, no_func);
232 void Op::apply(const void* invec, void* inoutvec, const int* len, MPI_Datatype datatype)
234 if (smpi_privatize_global_variables == SmpiPrivStrategies::MMAP) {
235 // we need to switch as the called function may silently touch global variables
236 XBT_DEBUG("Applying operation, switch to the right data frame ");
237 smpi_switch_data_segment(simgrid::s4u::Actor::self());
240 if (not smpi_process()->replaying() && *len > 0) {
241 if (not is_fortran_op_)
242 this->func_(const_cast<void*>(invec), inoutvec, const_cast<int*>(len), &datatype);
244 XBT_DEBUG("Applying operation of length %d from %p and from/to %p", *len, invec, inoutvec);
245 int tmp = datatype->c2f();
246 /* Unfortunately, the C and Fortran version of the MPI standard do not agree on the type here,
247 thus the reinterpret_cast. */
248 this->func_(const_cast<void*>(invec), inoutvec, const_cast<int*>(len), reinterpret_cast<MPI_Datatype*>(&tmp));
254 return static_cast<Op*>(F2C::f2c(id));
261 void Op::unref(MPI_Op* op){
262 if((*op)!=MPI_OP_NULL){
264 if((*op)->refcount_==0 && (*op)->predefined_==false)