diff --git a/bfps/_code.py b/bfps/_code.py
index ec1557d2f68ff7bad58abbf629fcdd7a21d8f341..aff52a6e82a5bce403ee0168f89d976d11467d1f 100644
--- a/bfps/_code.py
+++ b/bfps/_code.py
@@ -353,6 +353,7 @@ class _code(_base):
'\n')
script_file.write('echo "Start time is `date`"\n')
script_file.write('cd ' + self.work_dir + '\n')
+ script_file.write('cp -s ../*.h5 ./\n')
script_file.write('poe ' +
os.path.join(
self.work_dir,
@@ -423,6 +424,7 @@ class _code(_base):
'\n')
script_file.write('echo "This is step $LOADL_STEP_ID out of {0}"\n'.format(njobs))
script_file.write('echo "Start time is `date`"\n')
+ script_file.write('cp -s ../*.h5 ./\n')
script_file.write('cd ' + self.work_dir + '\n')
script_file.write('poe ' +
os.path.join(
diff --git a/bfps/cpp/field.cpp b/bfps/cpp/field.cpp
index aef3cb840c86b419cf834c76f4d917b2d87df60e..0d065cede5ba61519b3cfa4ddb854e9bfb3c95ad 100644
--- a/bfps/cpp/field.cpp
+++ b/bfps/cpp/field.cpp
@@ -774,6 +774,8 @@ kspace<be, dt>::kspace(
std::fill_n(nshell_local, this->nshells, 0);
});
+ std::vector<std::unordered_map<int, double>> dealias_filter_threaded(omp_get_max_threads());
+
KSPACE_CLOOP_K2_NXMODES(
this,[&](ptrdiff_t /*cindex*/, hsize_t /*yindex*/, hsize_t /*zindex*/, int nxmodes, hsize_t /*xindex*/, double k2){
if (k2 < this->kM2)
@@ -784,10 +786,16 @@ kspace<be, dt>::kspace(
}
if (dt == TWO_THIRDS){
// Should not be any race condition here it is a "write"
- this->dealias_filter[int(round(k2 / this->dk2))] = exp(-36.0 * pow(k2/this->kM2, 18.));
+ dealias_filter_threaded[omp_get_thread_num()][int(round(k2 / this->dk2))] = exp(-36.0 * pow(k2/this->kM2, 18.));
}
});
+ for(int idxMerge = 0 ; idxMerge < int(dealias_filter_threaded.size()) ; ++idxMerge){
+ for(const auto kv : dealias_filter_threaded[idxMerge]){
+ this->dealias_filter[kv.first] = kv.second;
+ }
+ }
+
nshell_local_threaded.mergeParallel();
kshell_local_threaded.mergeParallel();
diff --git a/bfps/cpp/fluid_solver_base.cpp b/bfps/cpp/fluid_solver_base.cpp
index b50c7c8ce2d1be4949e3307a44566cc4b09c63cd..393f086d02676c0faf3d047cc938b4befea64b93 100644
--- a/bfps/cpp/fluid_solver_base.cpp
+++ b/bfps/cpp/fluid_solver_base.cpp
@@ -513,6 +513,8 @@ fluid_solver_base<rnumber>::fluid_solver_base(
int64_t *nshell_local = new int64_t[this->nshells];
std::fill_n(nshell_local, this->nshells, 0.0);
+ std::vector<std::unordered_map<int, double>> Fourier_filter_threaded(omp_get_max_threads());
+
CLOOP_K2_NXMODES(
this,
@@ -524,9 +526,16 @@ fluid_solver_base<rnumber>::fluid_solver_base(
nshell_local[int(knorm/this->dk)] += nxmodes;
kshell_local[int(knorm/this->dk)] += nxmodes*knorm;
}
- this->Fourier_filter[int(round(k2 / this->dk2))] = exp(-36.0 * pow(k2/this->kM2, 18.));}
+ Fourier_filter_threaded[omp_get_thread_num()][int(round(k2 / this->dk2))] = exp(-36.0 * pow(k2/this->kM2, 18.));}
);
+ // Merge results
+ for(int idxMerge = 0 ; idxMerge < int(Fourier_filter_threaded.size()) ; ++idxMerge){
+ for(const auto kv : Fourier_filter_threaded[idxMerge]){
+ this->Fourier_filter[kv.first] = kv.second;
+ }
+ }
+
MPI_Allreduce(
(void*)(nshell_local),
(void*)(this->nshell),
diff --git a/bfps/cpp/scope_timer.hpp b/bfps/cpp/scope_timer.hpp
index 5fd6578c9e9f593191a44049c4f425448a53b21a..76fc69f16f1772aca3242756fcbaf58acf0eed29 100644
--- a/bfps/cpp/scope_timer.hpp
+++ b/bfps/cpp/scope_timer.hpp
@@ -37,7 +37,7 @@
#include <mpi.h>
#include <cstring>
#include <stdexcept>
-
+#include <fstream>
#include "base.hpp"
#include "bfps_timer.hpp"
@@ -185,10 +185,10 @@ public:
retMpi = MPI_Comm_size( inComm, &nbProcess);
assert(retMpi == MPI_SUCCESS);
- if((&outputStream == &std::cout || &outputStream == &std::clog) && myrank != nbProcess-1){
+ if((&outputStream == &std::cout || &outputStream == &std::clog) && myRank != nbProcess-1){
// Print in reverse order
char tmp;
- retMpi = MPI_Recv(&tmp, 1, MPI_BYTE, myrank+1, 99, inComm, MPI_STATUS_IGNORE);
+ retMpi = MPI_Recv(&tmp, 1, MPI_BYTE, myRank+1, 99, inComm, MPI_STATUS_IGNORE);
assert(retMpi == MPI_SUCCESS);
}
outputStream.flush();
@@ -230,10 +230,10 @@ public:
}
outputStream.flush();
- if((&outputStream == &std::cout || &outputStream == &std::clog) && myrank != 0){
+ if((&outputStream == &std::cout || &outputStream == &std::clog) && myRank != 0){
// Print in reverse order
char tmp;
- retMpi = MPI_Send(&tmp, 1, MPI_BYTE, myrank-1, 99, inComm);
+ retMpi = MPI_Send(&tmp, 1, MPI_BYTE, myRank-1, 99, inComm);
assert(retMpi == MPI_SUCCESS);
}
}
@@ -283,7 +283,7 @@ public:
}
}
- if(myrank != 0){
+ if(myRank != 0){
const std::string strOutput = myResults.str();
int sizeOutput = strOutput.length();
retMpi = MPI_Send(&sizeOutput, 1, MPI_INT, 0, 99, inComm);
@@ -444,6 +444,153 @@ public:
outputStream.flush();
}
+ void showHtml(const MPI_Comm inComm) const {
+ int myRank, nbProcess;
+ int retMpi = MPI_Comm_rank( inComm, &myRank);
+ assert(retMpi == MPI_SUCCESS);
+ retMpi = MPI_Comm_size( inComm, &nbProcess);
+ assert(retMpi == MPI_SUCCESS);
+
+ std::stringstream myResults;
+
+ std::stack<std::pair<int, const std::shared_ptr<CoreEvent>>> events;
+
+ for (int idx = static_cast<int>(root->getChildren().size()) - 1; idx >= 0; --idx) {
+ events.push({0, root->getChildren()[idx]});
+ }
+
+ myResults << "<h1>Process : " << myRank << "</h1>\n";
+
+ double totalDuration = 0;
+ for (int idx =
+ static_cast<int>(root->getChildren().size()) - 1;
+ idx >= 0; --idx) {
+ totalDuration += root->getChildren()[idx]->getDuration();
+ }
+
+ myResults << "<h2> " << root->getName() << " (" << totalDuration << "s)</h2>\n";
+ myResults << "<ul>\n";
+ int idxBox = myRank*100000;
+
+ while (events.size()) {
+ const std::pair<int, const std::shared_ptr<CoreEvent>> eventToShow =
+ events.top();
+ events.pop();
+
+ if(eventToShow.first == -1){
+ myResults << "</ul>\n";
+ myResults << "</li>\n";
+ }
+ else if(eventToShow.second->getChildren().size() == 0){
+ myResults << "<li>● <span title=\"";
+ if (eventToShow.second->getOccurrence() != 1) {
+ myResults << "Min = " << eventToShow.second->getMin() << "s ; Max = " << eventToShow.second->getMax()
+ << "s ; Average = " << eventToShow.second->getAverage() << "s ; Occurrence = "
+ << eventToShow.second->getOccurrence();
+ }
+ myResults << "\">" << eventToShow.second->getName();
+ myResults << " (" << 100*eventToShow.second->getDuration()/totalDuration << "% -- " ;
+ myResults << eventToShow.second->getDuration() <<"s)</span></li>\n";
+ }
+ else{
+ myResults << "<li><input type=\"checkbox\" id=\"c" << idxBox << "\" />\n";
+ myResults << " <i class=\"fa fa-angle-double-right\">→ </i>\n";
+ myResults << " <i class=\"fa fa-angle-double-down\">↓ </i>\n";
+ myResults << " <label for=\"c" << idxBox++ << "\"><span title=\"";
+ if (eventToShow.second->getOccurrence() != 1) {
+ myResults << "Min = " << eventToShow.second->getMin() << "s ; Max = " << eventToShow.second->getMax()
+ << "s ; Average = " << eventToShow.second->getAverage() << "s ; Occurrence = "
+ << eventToShow.second->getOccurrence();
+ }
+ myResults << "\">" << eventToShow.second->getName();
+ myResults << " (" << 100*eventToShow.second->getDuration()/totalDuration << "% -- " ;
+ myResults << eventToShow.second->getDuration() <<"s)</span></label>\n";
+ myResults << "<ul>\n";
+ events.push({-1, std::shared_ptr<CoreEvent>()});
+
+ for (int idx =
+ static_cast<int>(eventToShow.second->getChildren().size()) - 1;
+ idx >= 0; --idx) {
+ events.push(
+ {eventToShow.first + 1, eventToShow.second->getChildren()[idx]});
+ }
+ }
+ }
+
+ myResults << "</ul>\n";
+
+ if(myRank != 0){
+ const std::string strOutput = myResults.str();
+ int sizeOutput = strOutput.length();
+ retMpi = MPI_Send(&sizeOutput, 1, MPI_INT, 0, 99, inComm);
+ assert(retMpi == MPI_SUCCESS);
+ retMpi = MPI_Send((void*)strOutput.data(), sizeOutput, MPI_CHAR, 0, 100, inComm);
+ assert(retMpi == MPI_SUCCESS);
+ }
+ else{
+ const std::string htmlOutput = (getenv("HTMLOUTPUT")?getenv("HTMLOUTPUT"):"timings.html");
+
+ std::cout << "Timing output html set to : " << htmlOutput << std::endl;
+
+ std::ofstream htmlfile(htmlOutput);
+
+ htmlfile << "<html>\
+ <head>\
+ <style>\
+ input {\
+ display: none;\
+ }\
+ input ~ ul {\
+ display: none;\
+ }\
+ input:checked ~ ul {\
+ display: block;\
+ }\
+ input ~ .fa-angle-double-down {\
+ display: none;\
+ }\
+ input:checked ~ .fa-angle-double-right {\
+ display: none;\
+ }\
+ input:checked ~ .fa-angle-double-down {\
+ display: inline;\
+ }\
+ li {\
+ display: block;\
+ font-family: 'Arial';\
+ font-size: 15px;\
+ padding: 0.2em;\
+ border: 1px solid transparent;\
+ }\
+ li:hover {\
+ border: 1px solid grey;\
+ border-radius: 3px;\
+ background-color: lightgrey;\
+ }\
+ span:hover {\
+ color: blue;\
+ }\
+ </style>\
+ </head>\
+ <body>";
+
+ std::vector<char> buffer;
+ for(int idxProc = nbProcess-1 ; idxProc > 0 ; --idxProc){
+ int sizeRecv;
+ retMpi = MPI_Recv(&sizeRecv, 1, MPI_INT, idxProc, 99, inComm, MPI_STATUS_IGNORE);
+ assert(retMpi == MPI_SUCCESS);
+ buffer.resize(sizeRecv+1);
+ retMpi = MPI_Recv(buffer.data(), sizeRecv, MPI_CHAR, idxProc, 100, inComm, MPI_STATUS_IGNORE);
+ assert(retMpi == MPI_SUCCESS);
+ buffer[sizeRecv]='\0';
+ htmlfile << buffer.data();
+ }
+ htmlfile << myResults.str();
+ htmlfile << "</body>\
+ </html>";
+ }
+ }
+
friend scope_timer;
};