X-Git-Url: http://info.iut-bm.univ-fcomte.fr/pub/gitweb/simgrid.git/blobdiff_plain/fe9f13ea487593ec11c6af50d8150a6743c84114..9cf75648a5c1af4afba73f782c4e5da71e096d14:/src/smpi/mpi/smpi_datatype.cpp diff --git a/src/smpi/mpi/smpi_datatype.cpp b/src/smpi/mpi/smpi_datatype.cpp index d4257b1ac4..642aa24b22 100644 --- a/src/smpi/mpi/smpi_datatype.cpp +++ b/src/smpi/mpi/smpi_datatype.cpp @@ -1,5 +1,5 @@ /* smpi_datatype.cpp -- MPI primitives to handle datatypes */ -/* Copyright (c) 2009-2019. The SimGrid Team. All rights reserved. */ +/* Copyright (c) 2009-2023. The SimGrid Team. All rights reserved. */ /* This program is free software; you can redistribute it and/or modify it * under the terms of the license (GNU LGPL) which comes with this package. */ @@ -11,95 +11,100 @@ #include "src/instr/instr_private.hpp" #include "src/smpi/include/smpi_actor.hpp" +#include +#include +#include #include +#include XBT_LOG_NEW_DEFAULT_SUBCATEGORY(smpi_datatype, smpi, "Logging specific to SMPI (datatype)"); static std::unordered_map id2type_lookup; -#define CREATE_MPI_DATATYPE(name, id, type) \ - static simgrid::smpi::Datatype _XBT_CONCAT(mpi_, name)((char*)_XBT_STRINGIFY(name), (id), sizeof(type), /* size */ \ +#define CREATE_MPI_DATATYPE(name, id, type, flag) \ + simgrid::smpi::Datatype _XBT_CONCAT(smpi_MPI_, name)((char*)"MPI_"#name, (id), sizeof(type), /* size */ \ 0, /* lb */ \ sizeof(type), /* ub = lb + size */ \ - DT_FLAG_BASIC /* flags */ \ - ); \ - const MPI_Datatype name = &_XBT_CONCAT(mpi_, name); + DT_FLAG_BASIC | flag /* flags */ \ + ); #define CREATE_MPI_DATATYPE_NULL(name, id) \ - static simgrid::smpi::Datatype _XBT_CONCAT(mpi_, name)((char*)_XBT_STRINGIFY(name), (id), 0, /* size */ \ + simgrid::smpi::Datatype _XBT_CONCAT(smpi_MPI_, name)((char*)"MPI_"#name, (id), 0, /* size */ \ 0, /* lb */ \ 0, /* ub = lb + size */ \ DT_FLAG_BASIC /* flags */ \ - ); \ - const MPI_Datatype name = &_XBT_CONCAT(mpi_, name); + ); // Predefined data types -CREATE_MPI_DATATYPE(MPI_CHAR, 2, char); -CREATE_MPI_DATATYPE(MPI_SHORT, 3, short); -CREATE_MPI_DATATYPE(MPI_INT, 1, int); -CREATE_MPI_DATATYPE(MPI_LONG, 4, long); -CREATE_MPI_DATATYPE(MPI_LONG_LONG, 7, long long); -CREATE_MPI_DATATYPE(MPI_SIGNED_CHAR, 8, signed char); -CREATE_MPI_DATATYPE(MPI_UNSIGNED_CHAR, 9, unsigned char); -CREATE_MPI_DATATYPE(MPI_UNSIGNED_SHORT, 10, unsigned short); -CREATE_MPI_DATATYPE(MPI_UNSIGNED, 11, unsigned int); -CREATE_MPI_DATATYPE(MPI_UNSIGNED_LONG, 12, unsigned long); -CREATE_MPI_DATATYPE(MPI_UNSIGNED_LONG_LONG, 13, unsigned long long); -CREATE_MPI_DATATYPE(MPI_FLOAT, 5, float); -CREATE_MPI_DATATYPE(MPI_DOUBLE, 0, double); -CREATE_MPI_DATATYPE(MPI_LONG_DOUBLE, 14, long double); -CREATE_MPI_DATATYPE(MPI_WCHAR, 15, wchar_t); -CREATE_MPI_DATATYPE(MPI_C_BOOL, 16, bool); -CREATE_MPI_DATATYPE(MPI_BYTE, 6, int8_t); -CREATE_MPI_DATATYPE(MPI_INT8_T, 17, int8_t); -CREATE_MPI_DATATYPE(MPI_INT16_T, 18, int16_t); -CREATE_MPI_DATATYPE(MPI_INT32_T, 19, int32_t); -CREATE_MPI_DATATYPE(MPI_INT64_T, 20, int64_t); -CREATE_MPI_DATATYPE(MPI_UINT8_T, 21, uint8_t); -CREATE_MPI_DATATYPE(MPI_UINT16_T, 22, uint16_t); -CREATE_MPI_DATATYPE(MPI_UINT32_T, 23, uint32_t); -CREATE_MPI_DATATYPE(MPI_UINT64_T, 24, uint64_t); -CREATE_MPI_DATATYPE(MPI_C_FLOAT_COMPLEX, 25, float _Complex); -CREATE_MPI_DATATYPE(MPI_C_DOUBLE_COMPLEX, 26, double _Complex); -CREATE_MPI_DATATYPE(MPI_C_LONG_DOUBLE_COMPLEX, 27, long double _Complex); -CREATE_MPI_DATATYPE(MPI_AINT, 28, MPI_Aint); -CREATE_MPI_DATATYPE(MPI_OFFSET, 29, MPI_Offset); - -CREATE_MPI_DATATYPE(MPI_FLOAT_INT, 30, float_int); -CREATE_MPI_DATATYPE(MPI_LONG_INT, 31, long_int); -CREATE_MPI_DATATYPE(MPI_DOUBLE_INT, 32, double_int); -CREATE_MPI_DATATYPE(MPI_SHORT_INT, 33, short_int); -CREATE_MPI_DATATYPE(MPI_2INT, 34, int_int); -CREATE_MPI_DATATYPE(MPI_2FLOAT, 35, float_float); -CREATE_MPI_DATATYPE(MPI_2DOUBLE, 36, double_double); -CREATE_MPI_DATATYPE(MPI_2LONG, 37, long_long); - -CREATE_MPI_DATATYPE(MPI_REAL, 38, float); -CREATE_MPI_DATATYPE(MPI_REAL4, 39, float); -CREATE_MPI_DATATYPE(MPI_REAL8, 40, double); -CREATE_MPI_DATATYPE(MPI_REAL16, 41, long double); -CREATE_MPI_DATATYPE_NULL(MPI_DATATYPE_NULL, -1); -CREATE_MPI_DATATYPE(MPI_COMPLEX8, 42, float_float); -CREATE_MPI_DATATYPE(MPI_COMPLEX16, 43, double_double); -CREATE_MPI_DATATYPE(MPI_COMPLEX32, 44, double_double); -CREATE_MPI_DATATYPE(MPI_INTEGER1, 45, int); -CREATE_MPI_DATATYPE(MPI_INTEGER2, 46, int16_t); -CREATE_MPI_DATATYPE(MPI_INTEGER4, 47, int32_t); -CREATE_MPI_DATATYPE(MPI_INTEGER8, 48, int64_t); -CREATE_MPI_DATATYPE(MPI_INTEGER16, 49, integer128_t); - -CREATE_MPI_DATATYPE(MPI_LONG_DOUBLE_INT, 50, long_double_int); - -CREATE_MPI_DATATYPE_NULL(MPI_UB, 51); -CREATE_MPI_DATATYPE_NULL(MPI_LB, 52); -CREATE_MPI_DATATYPE(MPI_PACKED, 53, char); +CREATE_MPI_DATATYPE_NULL(DATATYPE_NULL, -1) +CREATE_MPI_DATATYPE(DOUBLE, 0, double, DT_FLAG_FP) +CREATE_MPI_DATATYPE(INT, 1, int, DT_FLAG_C_INTEGER) +CREATE_MPI_DATATYPE(CHAR, 2, char, DT_FLAG_C_INTEGER) +CREATE_MPI_DATATYPE(SHORT, 3, short, DT_FLAG_C_INTEGER) +CREATE_MPI_DATATYPE(LONG, 4, long, DT_FLAG_C_INTEGER) +CREATE_MPI_DATATYPE(FLOAT, 5, float, DT_FLAG_FP) +CREATE_MPI_DATATYPE(BYTE, 6, int8_t, DT_FLAG_BYTE) +CREATE_MPI_DATATYPE(LONG_LONG, 7, long long, DT_FLAG_C_INTEGER) +CREATE_MPI_DATATYPE(SIGNED_CHAR, 8, signed char, DT_FLAG_C_INTEGER) +CREATE_MPI_DATATYPE(UNSIGNED_CHAR, 9, unsigned char, DT_FLAG_C_INTEGER) +CREATE_MPI_DATATYPE(UNSIGNED_SHORT, 10, unsigned short, DT_FLAG_C_INTEGER) +CREATE_MPI_DATATYPE(UNSIGNED, 11, unsigned int, DT_FLAG_C_INTEGER) +CREATE_MPI_DATATYPE(UNSIGNED_LONG, 12, unsigned long, DT_FLAG_C_INTEGER) +CREATE_MPI_DATATYPE(UNSIGNED_LONG_LONG, 13, unsigned long long, DT_FLAG_C_INTEGER) +CREATE_MPI_DATATYPE(LONG_DOUBLE, 14, long double, DT_FLAG_FP) +CREATE_MPI_DATATYPE(WCHAR, 15, wchar_t, DT_FLAG_BASIC) +CREATE_MPI_DATATYPE(C_BOOL, 16, bool, DT_FLAG_LOGICAL) +CREATE_MPI_DATATYPE(INT8_T, 17, int8_t, DT_FLAG_C_INTEGER) +CREATE_MPI_DATATYPE(INT16_T, 18, int16_t, DT_FLAG_C_INTEGER) +CREATE_MPI_DATATYPE(INT32_T, 19, int32_t, DT_FLAG_C_INTEGER) +CREATE_MPI_DATATYPE(INT64_T, 20, int64_t, DT_FLAG_C_INTEGER) +CREATE_MPI_DATATYPE(UINT8_T, 21, uint8_t, DT_FLAG_C_INTEGER) +CREATE_MPI_DATATYPE(UINT16_T, 22, uint16_t, DT_FLAG_C_INTEGER) +CREATE_MPI_DATATYPE(UINT32_T, 23, uint32_t, DT_FLAG_C_INTEGER) +CREATE_MPI_DATATYPE(UINT64_T, 24, uint64_t, DT_FLAG_C_INTEGER) +CREATE_MPI_DATATYPE(C_FLOAT_COMPLEX, 25, float _Complex, DT_FLAG_COMPLEX) +CREATE_MPI_DATATYPE(C_DOUBLE_COMPLEX, 26, double _Complex, DT_FLAG_COMPLEX) +CREATE_MPI_DATATYPE(C_LONG_DOUBLE_COMPLEX, 27, long double _Complex, DT_FLAG_COMPLEX) +CREATE_MPI_DATATYPE(AINT, 28, MPI_Aint, DT_FLAG_MULTILANG) +CREATE_MPI_DATATYPE(OFFSET, 29, MPI_Offset, DT_FLAG_MULTILANG) + +CREATE_MPI_DATATYPE(FLOAT_INT, 30, float_int, DT_FLAG_REDUCTION) +CREATE_MPI_DATATYPE(LONG_INT, 31, long_int, DT_FLAG_REDUCTION) +CREATE_MPI_DATATYPE(DOUBLE_INT, 32, double_int, DT_FLAG_REDUCTION) +CREATE_MPI_DATATYPE(SHORT_INT, 33, short_int, DT_FLAG_REDUCTION) +CREATE_MPI_DATATYPE(2INT, 34, int_int, DT_FLAG_REDUCTION) +CREATE_MPI_DATATYPE(2FLOAT, 35, float_float, DT_FLAG_REDUCTION) +CREATE_MPI_DATATYPE(2DOUBLE, 36, double_double, DT_FLAG_REDUCTION) +CREATE_MPI_DATATYPE(2LONG, 37, long_long, DT_FLAG_REDUCTION) + +CREATE_MPI_DATATYPE(REAL, 38, float, DT_FLAG_FP) +CREATE_MPI_DATATYPE(REAL4, 39, float, DT_FLAG_FP) +CREATE_MPI_DATATYPE(REAL8, 40, double, DT_FLAG_FP) +CREATE_MPI_DATATYPE(REAL16, 41, long double, DT_FLAG_FP) +CREATE_MPI_DATATYPE(COMPLEX8, 42, float_float, DT_FLAG_COMPLEX) +CREATE_MPI_DATATYPE(COMPLEX16, 43, double_double, DT_FLAG_COMPLEX) +CREATE_MPI_DATATYPE(COMPLEX32, 44, double_double, DT_FLAG_COMPLEX) +CREATE_MPI_DATATYPE(INTEGER1, 45, int, DT_FLAG_F_INTEGER) +CREATE_MPI_DATATYPE(INTEGER2, 46, int16_t, DT_FLAG_F_INTEGER) +CREATE_MPI_DATATYPE(INTEGER4, 47, int32_t, DT_FLAG_F_INTEGER) +CREATE_MPI_DATATYPE(INTEGER8, 48, int64_t, DT_FLAG_F_INTEGER) +CREATE_MPI_DATATYPE(INTEGER16, 49, integer128_t, DT_FLAG_F_INTEGER) + +CREATE_MPI_DATATYPE(LONG_DOUBLE_INT, 50, long_double_int, DT_FLAG_REDUCTION) +CREATE_MPI_DATATYPE(CXX_BOOL, 51, bool, DT_FLAG_LOGICAL) +CREATE_MPI_DATATYPE(CXX_FLOAT_COMPLEX, 52, std::complex, DT_FLAG_COMPLEX) +CREATE_MPI_DATATYPE(CXX_DOUBLE_COMPLEX, 53, std::complex, DT_FLAG_COMPLEX) +CREATE_MPI_DATATYPE(CXX_LONG_DOUBLE_COMPLEX, 54, std::complex, DT_FLAG_COMPLEX) + +CREATE_MPI_DATATYPE_NULL(UB, 55) +CREATE_MPI_DATATYPE_NULL(LB, 56) +CREATE_MPI_DATATYPE(PACKED, 57, char, DT_FLAG_PREDEFINED) // Internal use only -CREATE_MPI_DATATYPE(MPI_PTR, 54, void*); -CREATE_MPI_DATATYPE(MPI_COUNT, 55, long long); +CREATE_MPI_DATATYPE(PTR, 58, void*, DT_FLAG_PREDEFINED) +CREATE_MPI_DATATYPE(COUNT, 59, long long, DT_FLAG_MULTILANG) +MPI_Datatype MPI_PTR = &smpi_MPI_PTR; - -namespace simgrid{ -namespace smpi{ +namespace simgrid::smpi { std::unordered_map Datatype::keyvals_; // required by the Keyval class implementation int Datatype::keyval_id_=0; // required by the Keyval class implementation @@ -107,98 +112,110 @@ Datatype::Datatype(int ident, int size, MPI_Aint lb, MPI_Aint ub, int flags) : D { id = std::to_string(ident); } -Datatype::Datatype(int size,MPI_Aint lb, MPI_Aint ub, int flags) : name_(nullptr), size_(size), lb_(lb), ub_(ub), flags_(flags), refcount_(1){ -#if SIMGRID_HAVE_MC - if(MC_is_active()) - MC_ignore(&(refcount_), sizeof(refcount_)); -#endif + +Datatype::Datatype(int size, MPI_Aint lb, MPI_Aint ub, int flags) : size_(size), lb_(lb), ub_(ub), flags_(flags) +{ + this->add_f(); + MC_ignore(&refcount_, sizeof refcount_); } -//for predefined types, so in_use = 0. -Datatype::Datatype(char* name, int ident, int size, MPI_Aint lb, MPI_Aint ub, int flags) +// for predefined types, so refcount_ = 0. +Datatype::Datatype(const char* name, int ident, int size, MPI_Aint lb, MPI_Aint ub, int flags) : name_(name), id(std::to_string(ident)), size_(size), lb_(lb), ub_(ub), flags_(flags), refcount_(0) { - id2type_lookup.insert({id, this}); -#if SIMGRID_HAVE_MC - if(MC_is_active()) - MC_ignore(&(refcount_), sizeof(refcount_)); -#endif + id2type_lookup.try_emplace(id, this); + MC_ignore(&refcount_, sizeof refcount_); } -Datatype::Datatype(Datatype *datatype, int* ret) : name_(nullptr), size_(datatype->size_), lb_(datatype->lb_), ub_(datatype->ub_), flags_(datatype->flags_), refcount_(1) +Datatype::Datatype(Datatype* datatype, int* ret) + : size_(datatype->size_), lb_(datatype->lb_), ub_(datatype->ub_), flags_(datatype->flags_), duplicated_datatype_(datatype) { - flags_ &= ~DT_FLAG_PREDEFINED; - *ret = MPI_SUCCESS; - - if (not datatype->attributes()->empty()) { - int flag=0; - void* value_out; - for (auto const& it : *(datatype->attributes())) { - smpi_key_elem elem = keyvals_.at(it.first); - if (elem != nullptr){ - if( elem->copy_fn.type_copy_fn != MPI_NULL_COPY_FN && - elem->copy_fn.type_copy_fn != MPI_TYPE_DUP_FN) - *ret = elem->copy_fn.type_copy_fn(datatype, it.first, elem->extra_state, it.second, &value_out, &flag); - else if ( elem->copy_fn.type_copy_fn_fort != MPI_NULL_COPY_FN && - (*(int*)*elem->copy_fn.type_copy_fn_fort) != 1){ - value_out=(int*)xbt_malloc(sizeof(int)); - elem->copy_fn.type_copy_fn_fort(datatype, it.first, elem->extra_state, it.second, value_out, &flag,ret); - } - if (*ret != MPI_SUCCESS) { - break; - } - if(elem->copy_fn.type_copy_fn == MPI_TYPE_DUP_FN || - ((elem->copy_fn.type_copy_fn_fort != MPI_NULL_COPY_FN) && (*(int*)*elem->copy_fn.type_copy_fn_fort == 1))){ - elem->refcount++; - attributes()->insert({it.first, it.second}); - } else if (flag){ - elem->refcount++; - attributes()->insert({it.first, value_out}); - } - } - } - } + this->add_f(); + datatype->ref(); + *ret = this->copy_attrs(datatype); } -Datatype::~Datatype(){ +Datatype::~Datatype() +{ xbt_assert(refcount_ >= 0); if(flags_ & DT_FLAG_PREDEFINED) return; - + //prevent further usage + flags_ &= ~ DT_FLAG_COMMITED; + if(duplicated_datatype_ != MPI_DATATYPE_NULL) + unref(duplicated_datatype_); + F2C::free_f(this->f2c_id()); //if still used, mark for deletion if(refcount_!=0){ flags_ |=DT_FLAG_DESTROYED; return; } - cleanup_attr(); +} - xbt_free(name_); +int Datatype::copy_attrs(Datatype* datatype){ + flags_ &= ~DT_FLAG_PREDEFINED; + + set_contents(MPI_COMBINER_DUP, 0, nullptr, 0, nullptr, 1, &datatype); + for (auto const& [key, value] : datatype->attributes()) { + auto elem_it = keyvals_.find(key); + xbt_assert(elem_it != keyvals_.end(), "Keyval not found for Datatype: %d", key); + + smpi_key_elem& elem = elem_it->second; + int ret = MPI_SUCCESS; + int flag = 0; + void* value_out = nullptr; + if (elem.copy_fn.type_copy_fn == MPI_TYPE_DUP_FN) { + value_out = value; + flag = 1; + } else if (elem.copy_fn.type_copy_fn != MPI_NULL_COPY_FN) { + ret = elem.copy_fn.type_copy_fn(datatype, key, elem.extra_state, value, &value_out, &flag); + } + if (ret != MPI_SUCCESS) + return ret; + + if (elem.copy_fn.type_copy_fn_fort != MPI_NULL_COPY_FN) { + value_out = xbt_new(int, 1); + if (*(int*)*elem.copy_fn.type_copy_fn_fort == 1) { // MPI_TYPE_DUP_FN + memcpy(value_out, value, sizeof(int)); + flag = 1; + } else { // not null, nor dup + elem.copy_fn.type_copy_fn_fort(datatype, key, elem.extra_state, value, value_out, &flag, &ret); + } + if (ret != MPI_SUCCESS) { + xbt_free(value_out); + return ret; + } + } + if (flag) { + elem.refcount++; + attributes().try_emplace(key, value_out); + } + } + return MPI_SUCCESS; } -void Datatype::ref(){ +int Datatype::clone(MPI_Datatype* type){ + int ret; + *type = new Datatype(this, &ret); + return ret; +} +void Datatype::ref() +{ refcount_++; - -#if SIMGRID_HAVE_MC - if(MC_is_active()) - MC_ignore(&(refcount_), sizeof(refcount_)); -#endif + MC_ignore(&refcount_, sizeof refcount_); } void Datatype::unref(MPI_Datatype datatype) { if (datatype->refcount_ > 0) datatype->refcount_--; + MC_ignore(&datatype->refcount_, sizeof datatype->refcount_); if (datatype->refcount_ == 0 && not(datatype->flags_ & DT_FLAG_PREDEFINED)) delete datatype; - -#if SIMGRID_HAVE_MC - if(MC_is_active()) - MC_ignore(&(datatype->refcount_), sizeof(datatype->refcount_)); -#endif } void Datatype::commit() @@ -206,22 +223,16 @@ void Datatype::commit() flags_ |= DT_FLAG_COMMITED; } -bool Datatype::is_valid(){ +bool Datatype::is_valid() const +{ return (flags_ & DT_FLAG_COMMITED); } -bool Datatype::is_basic() +bool Datatype::is_basic() const { return (flags_ & DT_FLAG_BASIC); } -bool Datatype::is_replayable() -{ - return (simgrid::instr::trace_format == simgrid::instr::TraceFormat::Ti) && - ((this == MPI_BYTE) || (this == MPI_DOUBLE) || (this == MPI_INT) || (this == MPI_CHAR) || - (this == MPI_SHORT) || (this == MPI_LONG) || (this == MPI_FLOAT)); -} - MPI_Datatype Datatype::decode(const std::string& datatype_id) { return id2type_lookup.find(datatype_id)->second; @@ -231,28 +242,28 @@ void Datatype::addflag(int flag){ flags_ &= flag; } -int Datatype::extent(MPI_Aint * lb, MPI_Aint * extent){ +int Datatype::extent(MPI_Aint* lb, MPI_Aint* extent) const +{ *lb = lb_; *extent = ub_ - lb_; return MPI_SUCCESS; } -void Datatype::get_name(char* name, int* length){ - if(name_!=nullptr){ - *length = strlen(name_); - strncpy(name, name_, *length+1); - }else{ - *length = 0; +void Datatype::get_name(char* name, int* length) const +{ + *length = static_cast(name_.length()); + if (not name_.empty()) { + name_.copy(name, *length); + name[*length] = '\0'; } } -void Datatype::set_name(const char* name){ - if(name_!=nullptr && (flags_ & DT_FLAG_PREDEFINED) == 0) - xbt_free(name_); - name_ = xbt_strdup(name); +void Datatype::set_name(const char* name) +{ + name_ = name; } -int Datatype::pack(const void* inbuf, int incount, void* outbuf, int outcount, int* position, MPI_Comm) +int Datatype::pack(const void* inbuf, int incount, void* outbuf, int outcount, int* position, const Comm*) { if (outcount - *position < incount*static_cast(size_)) return MPI_ERR_OTHER; @@ -261,7 +272,7 @@ int Datatype::pack(const void* inbuf, int incount, void* outbuf, int outcount, i return MPI_SUCCESS; } -int Datatype::unpack(const void* inbuf, int insize, int* position, void* outbuf, int outcount, MPI_Comm) +int Datatype::unpack(const void* inbuf, int insize, int* position, void* outbuf, int outcount, const Comm*) { if (outcount*static_cast(size_)> insize) return MPI_ERR_OTHER; @@ -270,34 +281,80 @@ int Datatype::unpack(const void* inbuf, int insize, int* position, void* outbuf, return MPI_SUCCESS; } -int Datatype::copy(const void *sendbuf, int sendcount, MPI_Datatype sendtype, - void *recvbuf, int recvcount, MPI_Datatype recvtype){ - -// FIXME Handle the case of a partial shared malloc. +int Datatype::get_contents(int max_integers, int max_addresses, int max_datatypes, int* array_of_integers, + MPI_Aint* array_of_addresses, MPI_Datatype* array_of_datatypes) const +{ + if(contents_==nullptr) + return MPI_ERR_ARG; + if (static_cast(max_integers) < contents_->integers_.size()) + return MPI_ERR_COUNT; + std::copy(begin(contents_->integers_), end(contents_->integers_), array_of_integers); + if (static_cast(max_addresses) < contents_->addresses_.size()) + return MPI_ERR_COUNT; + std::copy(begin(contents_->addresses_), end(contents_->addresses_), array_of_addresses); + if (static_cast(max_datatypes) < contents_->datatypes_.size()) + return MPI_ERR_COUNT; + std::copy(begin(contents_->datatypes_), end(contents_->datatypes_), array_of_datatypes); + for (auto& datatype : contents_->datatypes_) + datatype->ref(); + return MPI_SUCCESS; +} - if (smpi_privatize_global_variables == SmpiPrivStrategies::MMAP) { - smpi_switch_data_segment(simgrid::s4u::Actor::self()); +int Datatype::get_envelope(int* num_integers, int* num_addresses, int* num_datatypes, int* combiner) const +{ + if(contents_==nullptr){ + *num_integers = 0; + *num_addresses = 0; + *num_datatypes = 0; + *combiner = MPI_COMBINER_NAMED; + }else{ + *num_integers = contents_->integers_.size(); + *num_addresses = contents_->addresses_.size(); + *num_datatypes = contents_->datatypes_.size(); + *combiner = contents_->combiner_; } + return MPI_SUCCESS; +} + +int Datatype::copy(const void* sendbuf, int sendcount, MPI_Datatype sendtype, void* recvbuf, int recvcount, + MPI_Datatype recvtype) +{ + // FIXME Handle the case of a partial shared malloc. + + smpi_switch_data_segment(simgrid::s4u::Actor::self()); + /* First check if we really have something to do */ + size_t offset = 0; + std::vector> private_blocks; + if(smpi_is_shared(sendbuf,private_blocks,&offset) + && (private_blocks.size()==1 + && (private_blocks[0].second - private_blocks[0].first)==(unsigned long)(sendcount * sendtype->get_extent()))){ + XBT_VERB("sendbuf is shared. Ignoring copies"); + return 0; + } + if(smpi_is_shared(recvbuf,private_blocks,&offset) + && (private_blocks.size()==1 + && (private_blocks[0].second - private_blocks[0].first)==(unsigned long)(recvcount * recvtype->get_extent()))){ + XBT_VERB("recvbuf is shared. Ignoring copies"); + return 0; + } + if (recvcount > 0 && recvbuf != sendbuf) { sendcount *= sendtype->size(); recvcount *= recvtype->size(); int count = sendcount < recvcount ? sendcount : recvcount; XBT_DEBUG("Copying %d bytes from %p to %p", count, sendbuf, recvbuf); if (not(sendtype->flags() & DT_FLAG_DERIVED) && not(recvtype->flags() & DT_FLAG_DERIVED)) { - if (not smpi_process()->replaying()) + if (not smpi_process()->replaying() && count > 0) memcpy(recvbuf, sendbuf, count); } else if (not(sendtype->flags() & DT_FLAG_DERIVED)) { recvtype->unserialize(sendbuf, recvbuf, count / recvtype->size(), MPI_REPLACE); } else if (not(recvtype->flags() & DT_FLAG_DERIVED)) { sendtype->serialize(sendbuf, recvbuf, count / sendtype->size()); - }else{ - + } else if(sendtype->size() != 0 && recvtype->size() != 0){ void * buf_tmp = xbt_malloc(count); - sendtype->serialize( sendbuf, buf_tmp,count/sendtype->size()); recvtype->unserialize( buf_tmp, recvbuf,count/recvtype->size(), MPI_REPLACE); - xbt_free(buf_tmp); } } @@ -308,14 +365,14 @@ int Datatype::copy(const void *sendbuf, int sendcount, MPI_Datatype sendtype, //Default serialization method : memcpy. void Datatype::serialize(const void* noncontiguous_buf, void* contiguous_buf, int count) { - char* contiguous_buf_char = static_cast(contiguous_buf); - const char* noncontiguous_buf_char = static_cast(noncontiguous_buf)+lb_; + auto* contiguous_buf_char = static_cast(contiguous_buf); + const auto* noncontiguous_buf_char = static_cast(noncontiguous_buf) + lb_; memcpy(contiguous_buf_char, noncontiguous_buf_char, count*size_); } void Datatype::unserialize(const void* contiguous_buf, void *noncontiguous_buf, int count, MPI_Op op){ - const char* contiguous_buf_char = static_cast(contiguous_buf); - char* noncontiguous_buf_char = static_cast(noncontiguous_buf)+lb_; + const auto* contiguous_buf_char = static_cast(contiguous_buf); + auto* noncontiguous_buf_char = static_cast(noncontiguous_buf) + lb_; int n=count; if(op!=MPI_OP_NULL) op->apply( contiguous_buf_char, noncontiguous_buf_char, &n, this); @@ -346,13 +403,16 @@ int Datatype::create_vector(int count, int block_length, int stride, MPI_Datatyp ub=((count-1)*stride+block_length-1)*old_type->get_extent()+old_type->ub(); } if(old_type->flags() & DT_FLAG_DERIVED || stride != block_length){ - *new_type = new Type_Vector(count * (block_length) * old_type->size(), lb, ub, - DT_FLAG_DERIVED, count, block_length, stride, old_type); + *new_type = new Type_Vector(old_type->size() * block_length * count, lb, ub, DT_FLAG_DERIVED, count, block_length, + stride, old_type); retval=MPI_SUCCESS; }else{ /* in this situation the data are contiguous thus it's not required to serialize and unserialize it*/ - *new_type = new Datatype(count * block_length * old_type->size(), 0, ((count -1) * stride + block_length)* - old_type->size(), DT_FLAG_CONTIGUOUS); + *new_type = + new Datatype(old_type->size() * block_length * count, 0, + old_type->size() * ((count - 1) * stride + block_length), DT_FLAG_CONTIGUOUS | DT_FLAG_DERIVED); + const std::array ints = {{count, block_length, stride}}; + (*new_type)->set_contents(MPI_COMBINER_VECTOR, 3, ints.data(), 0, nullptr, 1, &old_type); retval=MPI_SUCCESS; } return retval; @@ -371,12 +431,15 @@ int Datatype::create_hvector(int count, int block_length, MPI_Aint stride, MPI_D ub=((count-1)*stride)+(block_length-1)*old_type->get_extent()+old_type->ub(); } if(old_type->flags() & DT_FLAG_DERIVED || stride != block_length*old_type->get_extent()){ - *new_type = new Type_Hvector(count * (block_length) * old_type->size(), lb, ub, - DT_FLAG_DERIVED, count, block_length, stride, old_type); + *new_type = new Type_Hvector(old_type->size() * block_length * count, lb, ub, DT_FLAG_DERIVED, count, block_length, + stride, old_type); retval=MPI_SUCCESS; }else{ /* in this situation the data are contiguous thus it's not required to serialize and unserialize it*/ - *new_type = new Datatype(count * block_length * old_type->size(), 0, count * block_length * old_type->size(), DT_FLAG_CONTIGUOUS); + *new_type = new Datatype(old_type->size() * block_length * count, 0, old_type->size() * block_length * count, + DT_FLAG_CONTIGUOUS | DT_FLAG_DERIVED); + const std::array ints = {{count, block_length}}; + (*new_type)->set_contents(MPI_COMBINER_HVECTOR, 2, ints.data(), 1, &stride, 1, &old_type); retval=MPI_SUCCESS; } return retval; @@ -436,7 +499,7 @@ int Datatype::create_hindexed(int count, const int* block_lengths, const MPI_Ain if(indices[i]+block_lengths[i]*old_type->ub()>ub) ub = indices[i]+block_lengths[i]*old_type->ub(); - if ( (i< count -1) && (indices[i]+block_lengths[i]*(static_cast(old_type->size())) != indices[i+1]) ) + if ((i < count - 1) && (indices[i] + static_cast(old_type->size()) * block_lengths[i] != indices[i + 1])) contiguous=false; } if (old_type->flags_ & DT_FLAG_DERIVED || lb!=0) @@ -484,7 +547,8 @@ int Datatype::create_struct(int count, const int* block_lengths, const MPI_Aint* if (not forced_ub && indices[i] + block_lengths[i] * old_types[i]->ub() > ub) ub = indices[i]+block_lengths[i]*old_types[i]->ub(); - if ( (i< count -1) && (indices[i]+block_lengths[i]*static_cast(old_types[i]->size()) != indices[i+1]) ) + if ((i < count - 1) && + (indices[i] + static_cast(old_types[i]->size() * block_lengths[i]) != indices[i + 1])) contiguous=false; } if (not contiguous) { @@ -544,10 +608,10 @@ int Datatype::create_subarray(int ndims, const int* array_of_sizes, tmp = *newtype; } - MPI_Aint lbs[1] = {lb * extent}; - int sizes [1]={1}; + const MPI_Aint lbs = lb * extent; + const int sizes = 1; //handle LB and UB with a resized call - create_hindexed( 1, sizes, lbs, tmp, newtype); + create_hindexed(1, &sizes, &lbs, tmp, newtype); unref(tmp); tmp = *newtype; @@ -558,21 +622,20 @@ int Datatype::create_subarray(int ndims, const int* array_of_sizes, } int Datatype::create_resized(MPI_Datatype oldtype,MPI_Aint lb, MPI_Aint extent, MPI_Datatype *newtype){ - int blocks[3] = {1, 1, 1}; - MPI_Aint disps[3] = {lb, 0, lb + extent}; - MPI_Datatype types[3] = {MPI_LB, oldtype, MPI_UB}; + const std::array blocks = {{1, 1, 1}}; + const std::array disps = {{lb, 0, lb + extent}}; + const std::array types = {{MPI_LB, oldtype, MPI_UB}}; - *newtype = new simgrid::smpi::Type_Struct(oldtype->size(), lb, lb + extent, DT_FLAG_DERIVED, 3, blocks, disps, types); + *newtype = new simgrid::smpi::Type_Struct(oldtype->size(), lb, lb + extent, DT_FLAG_DERIVED, 3, blocks.data(), + disps.data(), types.data()); (*newtype)->addflag(~DT_FLAG_COMMITED); return MPI_SUCCESS; } -Datatype* Datatype::f2c(int id){ +Datatype* Datatype::f2c(int id) +{ return static_cast(F2C::f2c(id)); } - -} -} - +} // namespace simgrid::smpi