portelli/Grid
1
0
Fork 0
mirror of https://github.com/paboyle/Grid.git synced 2025-06-17 07:17:06 +01:00
Code Releases Activity

Report only on failing nodes

Browse Source
This commit is contained in:
Guido Cossu
2016-12-09 05:57:30 +00:00
parent ec0c53fa68
commit 90fedbd2af
3 changed files with 53 additions and 38 deletions
Download Patch File Download Diff File Expand all files Collapse all files

69
lib/lattice/Lattice_reduction.h
View File

@ -41,6 +41,7 @@ class ReproducibilityState {
bool do_check;
bool enable_reprocheck;
bool success;
unsigned int interval;
std::vector<sum_type> th_states;
void reset_counter() { n_call = 0; }
@ -52,12 +53,14 @@ class ReproducibilityState {
n_call = 0;
};
ReproducibilityState() { reset(); }
ReproducibilityState():interval(1) {
reset();
}
void check(GridBase* grid, sum_type &sumarray){
/////////////////////// Reproducibility section, not threaded on purpouse
if (enable_reprocheck) {
if (do_check) {
if (do_check && (n_call % interval) == 0) {
for (int thread = 0; thread < sumarray.size(); thread++) {
int words = sizeof(sumarray[thread])/sizeof(unsigned char);
unsigned char xors[words];
@ -65,43 +68,45 @@ class ReproducibilityState {
// OR all words
unsigned char res = 0;
for (int w = 0; w < words; w++) res = res | xors[w];
if ( res ) {
std::cout << GridLogMessage << "Reproducibility failure report" << std::endl;
Grid_unquiesce_nodes();
int rank = 0;
while (rank < grid->_Nprocessors){
if (rank == grid->ThisRank() ){
grid->PrintRankInfo();
std::cout << "Call: "<< n_call << " Thread: " << thread << std::endl;
std::cout << "Size of states: " << th_states.size() << std::endl;
std::cout << std::setprecision(GRID_REAL_DIGITS+1) << std::scientific;
std::cout << "Saved partial sum : " << th_states[n_call][thread] << std::endl;
std::cout << "Current partial sum: " << sumarray[thread] << std::endl;
std::cout << "Saved state " << std::endl; show_binaryrep(th_states[n_call][thread]);
std::cout << "Current state" << std::endl; show_binaryrep(sumarray[thread]);
std::cout << "XOR result" << std::endl; show_binaryrep(xors, words);
Grid_unquiesce_nodes();
int rank = 0;
while (rank < grid->_Nprocessors){
if (rank == grid->ThisRank() ){
if ( res ) {
std::cout << "Reproducibility failure report" << std::endl;
grid->PrintRankInfo();
std::cout << "Call: "<< n_call << " Thread: " << thread << std::endl;
std::cout << "Size of states: " << th_states.size() << std::endl;
std::cout << std::setprecision(GRID_REAL_DIGITS+1) << std::scientific;
std::cout << "Saved partial sum : " << th_states[n_call][thread] << std::endl;
std::cout << "Current partial sum: " << sumarray[thread] << std::endl;
std::cout << "Saved state " << std::endl; show_binaryrep(th_states[n_call][thread]);
std::cout << "Current state" << std::endl; show_binaryrep(sumarray[thread]);
std::cout << "XOR result" << std::endl; show_binaryrep(xors, words);
//std::cout << std::defaultfloat; //not supported by some compilers
std::cout << std::setprecision(6);
success = false;
}
rank++;
grid->Barrier();
std::cout << std::setprecision(6);
success = false;
}
Grid_quiesce_nodes();
}
rank++;
grid->Barrier();
}
n_call++;
} else
{
std::cout << GridLogDebug << "Saving thread state for inner product. Call n. " << n_call << std::endl;
th_states.resize(n_call+1);
th_states[n_call].resize(grid->SumArraySize());
th_states[n_call] = sumarray; // save threads state
n_call++;
Grid_quiesce_nodes();
}
} else if (!do_check)
{
std::cout << GridLogDebug << "Saving thread state for inner product. Call n. "
<< n_call << std::endl;
th_states.resize(n_call+1);
th_states[n_call].resize(grid->SumArraySize());
th_states[n_call] = sumarray; // save threads state
//n_call++;
}
n_call++;
}
}
}
};