Skip to content

I
INMOST
Commit d04675b5 authored by Kirill Terekhov's avatar Kirill Terekhov
Browse files
Options

Various improvements 

Made database file case insensitive.
Tinkered dropping rules in INNER_MLILUC.
Added rule for git to ignore svn presence.
Added google analytics pixel to main page.
parent 01cd1704
Hide whitespace changes
Inline Side-by-side
Showing with 117 additions and 45 deletions
README.md
View file @ d04675b5
......@@ -20,3 +20,5 @@ Refer to https://github.com/INMOST-DEV/INMOST/wiki for general guides.
## Doxygen documentation
Doxygen-generated documentation is available online at http://dodo.inm.ras.ru/inmost/docs/latest/html/
[![Analytics](https://ga-beacon.appspot.com/UA-59408561-4/INMOST/README.md)](https://github.com/igrigorik/ga-beacon)
inmost_solver.h
View file @ d04675b5
......@@ -316,8 +316,8 @@ namespace INMOST
}
//void Reserve(INMOST_DATA_ENUM_TYPE num) { data.reserve(num);}
/// Clear all data of the current row.
void Clear() { data.clear(); }
void Swap(Solver::Row & other) { data.swap(other.data); bool tmp = marker; marker = other.marker; other.marker = tmp; }
void Clear() { data.clear(); }
void Swap(Solver::Row & other) { data.swap(other.data); bool tmp = marker; marker = other.marker; other.marker = tmp; }
/// The size of the sparse row, i.e. the total number of nonzero elements.
INMOST_DATA_ENUM_TYPE Size() const { return static_cast<INMOST_DATA_ENUM_TYPE>(data.size()); }
INMOST_DATA_ENUM_TYPE & GetIndex(INMOST_DATA_ENUM_TYPE k) {assert(k < data.size()); return (data.begin()+k)->first;}
......@@ -325,23 +325,27 @@ namespace INMOST
INMOST_DATA_ENUM_TYPE GetIndex(INMOST_DATA_ENUM_TYPE k) const {assert(k < data.size()); return (data.begin()+k)->first;}
INMOST_DATA_REAL_TYPE GetValue(INMOST_DATA_ENUM_TYPE k) const {assert(k < data.size()); return (data.begin()+k)->second;}
iterator Begin() {return data.begin();}
iterator End() {return data.end();}
const_iterator Begin() const {return data.begin();}
const_iterator End() const {return data.end();}
reverse_iterator rBegin() { return data.rbegin(); }
reverse_iterator rEnd() { return data.rend(); }
const_reverse_iterator rBegin() const { return data.rbegin(); }
const_reverse_iterator rEnd() const { return data.rend(); }
iterator Begin() {return data.begin();}
iterator End() {return data.end();}
const_iterator Begin() const {return data.begin();}
const_iterator End() const {return data.end();}
reverse_iterator rBegin() { return data.rbegin(); }
reverse_iterator rEnd() { return data.rend(); }
const_reverse_iterator rBegin() const { return data.rbegin(); }
const_reverse_iterator rEnd() const { return data.rend(); }
/// Return the scalar product of the current sparse row by a dense Vector.
INMOST_DATA_REAL_TYPE RowVec(Vector & x) const; // returns A(row) * x
void MoveRow(Row & new_pos) {data = new_pos.data;} //here move constructor and std::move may be used in future
void MoveRow(Row & new_pos) {data = new_pos.data;} //here move constructor and std::move may be used in future
/// Set the vector entries by zeroes.
void Zero() {for(iterator it = Begin(); it != End(); ++it) it->second = 0;}
void Zero() {for(iterator it = Begin(); it != End(); ++it) it->second = 0;}
/// Push specified element into sparse row.
/// This function should be used only if the index is not repeated in the row
/// and all indexes are inserted in increasing order.
void Push(INMOST_DATA_ENUM_TYPE ind, INMOST_DATA_REAL_TYPE val) {data.push_back(make_entry(ind,val));}
/// This function should be used only if the index is not repeated in the row.
void Push(INMOST_DATA_ENUM_TYPE ind, INMOST_DATA_REAL_TYPE val) {data.push_back(make_entry(ind,val));}
/// Resize row to specified size.
/// It is intended to be used together with non-const Row::GetIndex and Row::GetValue
/// that allow for the modification of individual entries.
/// @param size New size of the row.
void Resize(INMOST_DATA_ENUM_TYPE size) {data.resize(size);}
};
/// Class to store the distributed sparse matrix by compressed rows.
......
solver.cpp
View file @ d04675b5
......@@ -1155,20 +1155,21 @@ namespace INMOST
if( str[k] == ':' ) break;
}
if( k == strlen(str) ) continue; //invalid line
for(l = 0; l < k; ++l) str[l] = tolower(str[l]);
l = k+1;
while(l < (int)strlen(str) && isspace(str[l]) ) ++l;
if( l == strlen(str) ) continue; //skip empty entry
if( !strncmp(str,"PETSc",k) )
if( !strncmp(str,"petsc",k) )
petsc_database_file = std::string(str+l);
else if( !strncmp(str,"Trilinos_Ifpack",k) )
else if( !strncmp(str,"trilinos_ifpack",k) )
trilinos_ifpack_database_file = std::string(str+l);
else if( !strncmp(str,"Trilinos_Aztec",k) )
else if( !strncmp(str,"trilinos_aztec",k) )
trilinos_aztec_database_file = std::string(str+l);
else if( !strncmp(str,"Trilinos_ML",k) )
else if( !strncmp(str,"trilinos_ml",k) )
trilinos_ml_database_file = std::string(str+l);
else if( !strncmp(str,"Trilinos_Belos",k) )
else if( !strncmp(str,"trilinos_belos",k) )
trilinos_belos_database_file = std::string(str+l);
else if( !strncmp(str,"ANI",k) )
else if( !strncmp(str,"ani",k) )
ani_database_file = std::string(str+l);
}
}
......@@ -2112,15 +2113,15 @@ namespace INMOST
ret = false;
break;
case AZ_ill_cond:
return_reason = "The Hessenberg matrix within GMRES is illconditioned."
"This could be caused by a number"
"of reasons. For example, the preconditioning"
"matrix could be nearly singular due to an unstable"
"factorization (note: pivoting is not implemented"
"in any of the incomplete factorizations)."
"Ill-conditioned Hessenberg matrices could also"
"arise from a singular application matrix. In this"
"case, GMRES tries to compute a least-squares"
return_reason = "The Hessenberg matrix within GMRES is illconditioned. "
"This could be caused by a number "
"of reasons. For example, the preconditioning "
"matrix could be nearly singular due to an unstable "
"factorization (note: pivoting is not implemented "
"in any of the incomplete factorizations). "
"Ill-conditioned Hessenberg matrices could also "
"arise from a singular application matrix. In this "
"case, GMRES tries to compute a least-squares "
"solution.";
ret = false;
break;
......
solver_ddpqiluc2.hpp
View file @ d04675b5
......@@ -53,10 +53,28 @@ class ILUC_preconditioner : public Method
typedef dynarray<INMOST_DATA_ENUM_TYPE,256> sort_inds;
typedef dynarray<INMOST_DATA_ENUM_TYPE,256> levels_t;
//result of multilevel preconditioner
// |LU F |
// |LDU F |
// A = | |
// | E C |
// LU decomposition is stored in skyline format with reversed direction
//
// For the next step of multilevel factorization we have to compute
// S = C - (E U) D (L F)
//
// U is stored by rows
// L is stored by columns
// F is stored by rows
// E is stored by rows
// C is stored by rows
//
// During LF calculation F is traversed transposed (by columns) and
// each column is solved with L.
// LF is obtained by columns.
//
// During EU calculating E is traversed by rows and
// each row is solved with U. Then re
//
// For the faster multiplication of
std::vector<Solver::Row::entry> LU_Entries, B_Entries;
interval<INMOST_DATA_ENUM_TYPE, INMOST_DATA_REAL_TYPE> LU_Diag;
interval<INMOST_DATA_ENUM_TYPE, Interval> U_Address, L_Address, B_Address;
......@@ -431,7 +449,7 @@ public:
#if defined(ILUC2)
INMOST_DATA_ENUM_TYPE nzLU2 = 0;
INMOST_DATA_REAL_TYPE tau2 = iluc2_tau;
INMOST_DATA_REAL_TYPE tau2 = iluc2_tau, Anorm;
std::vector<Solver::Row::entry> LU2_Entries;
interval<INMOST_DATA_ENUM_TYPE, Interval> L2_Address(mobeg, moend+1), U2_Address(mobeg, moend+1);
interval<INMOST_DATA_ENUM_TYPE, INMOST_DATA_ENUM_TYPE> U2list(mobeg, moend, UNDEF), L2list(mobeg, moend, UNDEF);
......@@ -603,6 +621,8 @@ public:
// A = | |
// | E C |
//compute B,F
Anorm = 0;
nzA = 0;
for (k = cbeg; k < cend; ++k)
{
F_Address[k].first = static_cast<INMOST_DATA_ENUM_TYPE>(F_Entries.size());
......@@ -613,12 +633,14 @@ public:
u = A_Entries[jt].second;
if (i < cend) B_Entries.push_back(Solver::Row::make_entry(i, u));
else F_Entries.push_back(Solver::Row::make_entry(i,u)); //count nonzero in column of F
Anorm += u*u;
}
F_Address[k].last = static_cast<INMOST_DATA_ENUM_TYPE>(F_Entries.size());
B_Address[k].last = static_cast<INMOST_DATA_ENUM_TYPE>(B_Entries.size());
std::sort(B_Entries.begin() + B_Address[k].first, B_Entries.end());
std::sort(F_Entries.begin() + F_Address[k].first, F_Entries.end());
nzEF += F_Address[k].Size();
nzA += A_Address[k].Size();
}
//Setup column addressing for B,F, in descending order to keep list ordered
for (k = cend; k > cbeg; --k)
......@@ -652,13 +674,16 @@ public:
u = A_Entries[jt].second;
if (i < cend) E_Entries.push_back(Solver::Row::make_entry(i, u)); //put to row of E
else C_Entries.push_back(Solver::Row::make_entry(i, u)); //form new Schur complement
Anorm += u*u;
}
E_Address.back()->at(k).last = static_cast<INMOST_DATA_ENUM_TYPE>(E_Entries.size());
C_Address[k].last = static_cast<INMOST_DATA_ENUM_TYPE>(C_Entries.size());
// sort entries added to E, because it will be very usefull later
std::sort(E_Entries.begin() + E_Address.back()->at(k).first, E_Entries.end());
nzEF += E_Address.back()->at(k).Size();
nzA += A_Address[k].Size();
}
Anorm = sqrt(Anorm/static_cast<INMOST_DATA_REAL_TYPE>(nzA));
tt3 = Timer() - tt3;
ttt = Timer();
applyPQ(wbeg, wend, localP, localQ, invP, invQ);
......@@ -1943,9 +1968,9 @@ swap_algorithm:
{
std::cout << std::fixed << std::setprecision(2) << std::setw(6) << 100.0f*(k - cbeg) / (float)(cend - cbeg) << "%";
std::cout << " nnz U " << std::setprecision(10) << std::setw(10) << nzLU;
std::cout << " nnz LU " << std::setprecision(10) << std::setw(10) << nzLU;
#if defined(ILUC2)
std::cout << " U2 " << std::setw(10) << nzLU2;
std::cout << " LU2 " << std::setw(10) << nzLU2;
#endif
std::cout << std::scientific << std::setprecision(5) << " condition L " << NuL << " D " << NuD << " U " << NuU;
std::cout << " pivot swaps " << swaps;
......@@ -2328,7 +2353,7 @@ swap_algorithm:
Li = LineIndeces[cbeg];
while (Li != EOL)
{
if (fabs(LineValues[Li - 1])*NuL > tau2)
if (fabs(LineValues[Li - 1])*NuL > tau2*LU_Diag[Li-1])
{
curr = Li;
for (INMOST_DATA_ENUM_TYPE ru = U_Address[Li - 1].first; ru != U_Address[Li - 1].last; ++ru)
......@@ -2337,7 +2362,7 @@ swap_algorithm:
j = LU_Entries[ru].first;
if (LineIndeces[j + 1] != UNDEF) // There is an entry in the list
LineValues[j] -= u;
else if (fabs(u)*NuL > tau2)
else if (fabs(u)*NuL > tau2*LU_Diag[Li-1])
{
next = curr;
while (next < j + 1)
......@@ -2357,7 +2382,7 @@ swap_algorithm:
///////////////////////////////////////////////////////////////////////////////////
// perform solve with second-order U part //
///////////////////////////////////////////////////////////////////////////////////
if (fabs(LineValues[Li - 1])*NuL > tau)
if (fabs(LineValues[Li - 1])*NuL > tau*LU_Diag[Li-1])
{
curr = Li;
for (INMOST_DATA_ENUM_TYPE ru = U2_Address[Li - 1].first; ru != U2_Address[Li - 1].last; ++ru)
......@@ -2366,7 +2391,7 @@ swap_algorithm:
j = LU2_Entries[ru].first;
if (LineIndeces[j + 1] != UNDEF) // There is an entry in the list
LineValues[j] -= u;
else if (fabs(u)*NuL > tau2 )
else if (fabs(u)*NuL > tau2*LU_Diag[Li-1] )
{
next = curr;
while (next < j + 1)
......@@ -2398,7 +2423,7 @@ swap_algorithm:
while (Li != EOL)
{
j = LineIndeces[Li];
if (fabs(LineValues[Li - 1])*NuL < tau2 )//tau2)
if (fabs(LineValues[Li - 1])*NuL < tau2*LU_Diag[Li-1] )//tau2)
{
LineIndeces[Ui] = j;
LineIndeces[Li] = UNDEF;
......@@ -2439,16 +2464,18 @@ swap_algorithm:
while (Li != EOL)
{
//iterate over corresponding row of LF, add multiplication to list
l = LineValues[Li - 1];
for (INMOST_DATA_ENUM_TYPE r = LFt_Address[Li - 1].first; r < LFt_Address[Li - 1].last; ++r)
{
j = LFt_Entries[r].first;
u = LFt_Entries[r].second;
l = LineValues[Li - 1];
v = -l*u;
if (Ulist[j] != UNDEF)
temp[j] -= u*l;
else if( fabs(u)*NuU > tau || fabs(l)*NuL > tau )
temp[j] += v;
//else if( fabs(u)*NuU > tau || fabs(l)*NuL > tau )
else if( fabs(v) > tau2 )
{
temp[j] = -u*l;
temp[j] = v;
Ulist[j] = Ubeg[cbeg];
Ubeg[cbeg] = j;
}
......@@ -2477,7 +2504,7 @@ swap_algorithm:
///////////////////////////////////////////////////////////////////////////////////
// put calculated row to Schur complement //
///////////////////////////////////////////////////////////////////////////////////
tol_schur = tau2 * Smax / std::max(NuU,NuL);//tau2 * Smax / std::max(NuU,NuL);
tol_schur = tau2 * Smax * (Smax / Snorm );//tau2 * Smax / std::max(NuU,NuL);
Ui = Ubeg[cbeg];
S_Address[i].first = static_cast<INMOST_DATA_ENUM_TYPE>(S_Entries.size());
while (Ui != EOL)
......@@ -2518,7 +2545,7 @@ swap_algorithm:
if (i % 10 == 0)
{
//carrige return
printf("%6.2f%% nzS %d Smax %g Snorm %g tol %g\r",
printf("%6.2f%% nzS %8d Smax %20g Snorm %20g tol %20g\r",
100.f*(i - cend) / (1.f*(wend - cend)), nzS,Smax,Snorm,tol_schur);
fflush(stdout);
}
......@@ -2903,6 +2930,43 @@ swap_algorithm:
output[k] += input[B_Entries[m].first] * B_Entries[m].second;
}
}
void ApplyLU(int level, Solver::Vector & input,Solver::Vector & output)
{
INMOST_DATA_ENUM_TYPE k, m, cbeg, cend;
//current B block interval
cbeg = level_interval[level].first;
cend = level_interval[level].last;
for (k = cbeg; k < cend; ++k) output[k] = input[k];
//Solve with L first
for (k = cbeg; k < cend; ++k) //iterator over columns of L
{
for (INMOST_DATA_ENUM_TYPE r = L_Address[k].first; r != L_Address[k].last; ++r)
output[LU_Entries[r].first] -= output[k] * LU_Entries[r].second;
}
//Solve with diagonal
for (k = cbeg; k < cend; ++k) temp[k] /= LU_Diag[k];
//Solve with U
for (k = cend; k > cbeg; --k) //iterator over rows of U
{
for (INMOST_DATA_ENUM_TYPE r = U_Address[k - 1].first; r != U_Address[k - 1].last; ++r)
output[k - 1] -= output[LU_Entries[r].first] * LU_Entries[r].second;
}
}
void ApplyB(int level, double alpha, Solver::Vector & x, double beta, Solver::Vector & y) // y = alpha A x + beta y
{
INMOST_DATA_ENUM_TYPE k, m, cbeg, cend;
INMOST_DATA_REAL_TYPE temp;
for(k = cbeg; k < cend; ++k)
{
temp = 0.0;
for(m = B_Address[k].first; m != B_Address[k].last; ++m)
{
temp += B_Entries[m].second*x[B_Entries[m].first];
}
y[k] = beta*y[k] + alpha * temp;
}
}
int Descend(int level, Solver::Vector & input, Solver::Vector & output)
{
INMOST_DATA_ENUM_TYPE k, m, cbeg, cend;
......
Markdown is supported
0% or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment