diff --git a/sPuReMD/src/GMRES.c b/sPuReMD/src/GMRES.c
index b4387244785fb56c1869d1f5937ba0c247ad6cde..c08019ff0d4f6d6b000782a8a59163d6ee81fea6 100644
--- a/sPuReMD/src/GMRES.c
+++ b/sPuReMD/src/GMRES.c
@@ -74,7 +74,7 @@ static void Sparse_MatVec( const sparse_matrix * const A,
#pragma omp barrier
#endif
- #pragma omp for schedule(guided)
+ #pragma omp for schedule(static)
for ( i = 0; i < n; ++i )
{
si = A->start[i];
@@ -101,7 +101,7 @@ static void Sparse_MatVec( const sparse_matrix * const A,
#endif
}
#ifdef _OPENMP
- #pragma omp for schedule(guided)
+ #pragma omp for schedule(static)
for ( i = 0; i < n; ++i )
{
for ( j = 0; j < omp_get_num_threads(); ++j )
@@ -120,8 +120,8 @@ static void diag_pre_app( const real * const Hdia_inv, const real * const y,
{
unsigned int i;
- #pragma omp parallel for schedule(guided) \
- default(none) private(i)
+ #pragma omp parallel for schedule(static) \
+ default(none) private(i)
for ( i = 0; i < N; ++i )
{
x[i] = y[i] * Hdia_inv[i];
@@ -198,6 +198,7 @@ static void tri_solve_level_sched( const sparse_matrix * const LU, const real *
static int levels_L = 1, levels_U = 1;
static unsigned int *row_levels_L = NULL, *level_rows_L = NULL, *level_rows_cnt_L = NULL;
static unsigned int *row_levels_U = NULL, *level_rows_U = NULL, *level_rows_cnt_U = NULL;
+ static unsigned int *top = NULL;
unsigned int *row_levels, *level_rows, *level_rows_cnt;
if ( tri == LOWER )
@@ -218,8 +219,17 @@ static void tri_solve_level_sched( const sparse_matrix * const LU, const real *
if ( row_levels == NULL || level_rows == NULL || level_rows_cnt == NULL )
{
if ( (row_levels = (unsigned int*) malloc((size_t)LU->n * sizeof(unsigned int))) == NULL
- || (level_rows = (unsigned int*) malloc(MAX_ROWS_PER_LEVEL * (size_t)LU->n * sizeof(unsigned int))) == NULL
- || (level_rows_cnt = (unsigned int*) malloc((size_t)LU->n * sizeof(unsigned int))) == NULL )
+ || (level_rows = (unsigned int*) malloc((size_t)LU->n * sizeof(unsigned int))) == NULL
+ || (level_rows_cnt = (unsigned int*) malloc((size_t)(LU->n + 1) * sizeof(unsigned int))) == NULL )
+ {
+ fprintf( stderr, "Not enough space for triangular solve via level scheduling. Terminating...\n" );
+ exit( INSUFFICIENT_MEMORY );
+ }
+ }
+
+ if ( top == NULL )
+ {
+ if ( (top = (unsigned int*) malloc((size_t)(LU->n + 1) * sizeof(unsigned int))) == NULL )
{
fprintf( stderr, "Not enough space for triangular solve via level scheduling. Terminating...\n" );
exit( INSUFFICIENT_MEMORY );
@@ -229,52 +239,57 @@ static void tri_solve_level_sched( const sparse_matrix * const LU, const real *
/* find levels (row dependencies in substitutions) */
if ( find_levels )
{
- memset( row_levels, 0, LU->n * sizeof( unsigned int) );
- memset( level_rows_cnt, 0, LU->n * sizeof( unsigned int) );
+ memset( row_levels, 0, LU->n * sizeof(unsigned int) );
+ memset( level_rows_cnt, 0, LU->n * sizeof(unsigned int) );
+ memset( top, 0, LU->n * sizeof(unsigned int) );
if ( tri == LOWER )
{
for ( i = 0; i < LU->n; ++i )
{
- local_level = 0;
+ local_level = 1;
for ( pj = LU->start[i]; pj < LU->start[i + 1] - 1; ++pj )
{
local_level = MAX( local_level, row_levels[LU->j[pj]] + 1 );
}
- levels = MAX( levels, local_level + 1 );
+ levels = MAX( levels, local_level );
row_levels[i] = local_level;
- level_rows[local_level * MAX_ROWS_PER_LEVEL + level_rows_cnt[local_level]] = i;
++level_rows_cnt[local_level];
- if ( level_rows_cnt[local_level] >= MAX_ROWS_PER_LEVEL )
- {
- fprintf( stderr, "Not enough space for triangular solve via level scheduling" );
- fprintf( stderr, " (MAX_ROWS_PER_LEVEL). Terminating...\n" );
- exit( INSUFFICIENT_MEMORY );
- }
}
+
+ printf("levels(L): %d\n", levels);
+ printf("NNZ(L): %d\n", LU->start[LU->n]);
}
else
{
for ( i = LU->n - 1; i >= 0; --i )
{
- local_level = 0;
+ local_level = 1;
for ( pj = LU->start[i] + 1; pj < LU->start[i + 1]; ++pj )
{
local_level = MAX( local_level, row_levels[LU->j[pj]] + 1 );
}
- levels = MAX( levels, local_level + 1 );
+ levels = MAX( levels, local_level );
row_levels[i] = local_level;
- level_rows[local_level * MAX_ROWS_PER_LEVEL + level_rows_cnt[local_level]] = i;
++level_rows_cnt[local_level];
- if ( level_rows_cnt[local_level] >= MAX_ROWS_PER_LEVEL )
- {
- fprintf( stderr, "Not enough space for triangular solve via level scheduling" );
- fprintf( stderr, " (MAX_ROWS_PER_LEVEL). Terminating...\n" );
- exit( INSUFFICIENT_MEMORY );
- }
}
+
+ printf("levels(U): %d\n", levels);
+ printf("NNZ(U): %d\n", LU->start[LU->n]);
+ }
+
+ for ( i = 1; i < levels + 1; ++i )
+ {
+ level_rows_cnt[i] += level_rows_cnt[i - 1];
+ top[i] = level_rows_cnt[i];
+ }
+
+ for ( i = 0; i < LU->n; ++i )
+ {
+ level_rows[top[row_levels[i] - 1]] = i;
+ ++top[row_levels[i] - 1];
}
}
@@ -286,9 +301,9 @@ static void tri_solve_level_sched( const sparse_matrix * const LU, const real *
for ( i = 0; i < levels; ++i )
{
#pragma omp for schedule(static)
- for ( j = 0; j < level_rows_cnt[i]; ++j )
+ for ( j = level_rows_cnt[i]; j < level_rows_cnt[i + 1]; ++j )
{
- local_row = level_rows[i * MAX_ROWS_PER_LEVEL + j];
+ local_row = level_rows[j];
x[local_row] = y[local_row];
for ( pj = LU->start[local_row]; pj < LU->start[local_row + 1] - 1; ++pj )
{
@@ -304,9 +319,9 @@ static void tri_solve_level_sched( const sparse_matrix * const LU, const real *
for ( i = 0; i < levels; ++i )
{
#pragma omp for schedule(static)
- for ( j = 0; j < level_rows_cnt[i]; ++j )
+ for ( j = level_rows_cnt[i]; j < level_rows_cnt[i + 1]; ++j )
{
- local_row = level_rows[i * MAX_ROWS_PER_LEVEL + j];
+ local_row = level_rows[j];
x[local_row] = y[local_row];
for ( pj = LU->start[local_row] + 1; pj < LU->start[local_row + 1]; ++pj )
{
@@ -411,7 +426,7 @@ static void jacobi_iter( const sparse_matrix * const R, const real * const Dinv,
#pragma omp barrier
/* precompute and cache, as invariant in loop below */
- #pragma omp for schedule(guided)
+ #pragma omp for schedule(static)
for ( i = 0; i < R->n; ++i )
{
Dinv_b[i] = Dinv[i] * b[i];
@@ -432,7 +447,7 @@ static void jacobi_iter( const sparse_matrix * const R, const real * const Dinv,
#pragma omp barrier
- #pragma omp for schedule(guided)
+ #pragma omp for schedule(static)
for ( i = 0; i < R->n; ++i )
{
if (tri == LOWER)
@@ -464,7 +479,7 @@ static void jacobi_iter( const sparse_matrix * const R, const real * const Dinv,
#pragma omp barrier
- #pragma omp for schedule(guided)
+ #pragma omp for schedule(static)
for ( i = 0; i < R->n; ++i )
{
#ifdef _OPENMP
diff --git a/sPuReMD/src/QEq.c b/sPuReMD/src/QEq.c
index 5c5b47791b965846fddded0fa24e82b8c09ccb9c..514330bdb342caa65b3446f10cd463cabca75b43 100644
--- a/sPuReMD/src/QEq.c
+++ b/sPuReMD/src/QEq.c
@@ -198,7 +198,7 @@ static void Calculate_Droptol( const sparse_matrix * const A, real * const dropt
#pragma omp barrier
/* calculate sqaure of the norm of each row */
- #pragma omp for schedule(guided)
+ #pragma omp for schedule(static)
for ( i = 0; i < A->n; ++i )
{
for ( k = A->start[i]; k < A->start[i + 1] - 1; ++k )
@@ -226,7 +226,7 @@ static void Calculate_Droptol( const sparse_matrix * const A, real * const dropt
#pragma omp barrier
#ifdef _OPENMP
- #pragma omp for schedule(guided)
+ #pragma omp for schedule(static)
for ( i = 0; i < A->n; ++i )
{
droptol[i] = 0.0;
@@ -241,7 +241,7 @@ static void Calculate_Droptol( const sparse_matrix * const A, real * const dropt
/* calculate local droptol for each row */
//fprintf( stderr, "droptol: " );
- #pragma omp for schedule(guided)
+ #pragma omp for schedule(static)
for ( i = 0; i < A->n; ++i )
{
//fprintf( stderr, "%f-->", droptol[i] );
@@ -261,8 +261,8 @@ static int Estimate_LU_Fill( const sparse_matrix * const A, const real * const d
fillin = 0;
- #pragma omp parallel for schedule(guided) \
- default(none) private(i, j, pj, val) reduction(+: fillin)
+ #pragma omp parallel for schedule(static) \
+ default(none) private(i, j, pj, val) reduction(+: fillin)
for ( i = 0; i < A->n; ++i )
{
for ( pj = A->start[i]; pj < A->start[i + 1] - 1; ++pj )
@@ -587,7 +587,7 @@ static real diag_pre_comp( const reax_system * const system, real * const Hdia_i
start = Get_Time( );
- #pragma omp parallel for schedule(guided) \
+ #pragma omp parallel for schedule(static) \
default(none) private(i)
for ( i = 0; i < system->N; ++i )
{
@@ -811,7 +811,7 @@ static real ICHOL_PAR( const sparse_matrix * const A, const unsigned int sweeps,
for ( i = 0; i < sweeps; ++i )
{
/* for each nonzero */
- #pragma omp parallel for schedule(guided) \
+ #pragma omp parallel for schedule(static) \
default(none) shared(DAD, stderr) private(sum, ei_x, ei_y, k) firstprivate(x, y)
for ( j = 0; j < A->start[A->n]; ++j )
{
@@ -971,7 +971,7 @@ static real ILU_PAR( const sparse_matrix * const A, const unsigned int sweeps,
exit( INSUFFICIENT_MEMORY );
}
- #pragma omp parallel for schedule(guided) \
+ #pragma omp parallel for schedule(static) \
default(none) shared(D, D_inv) private(i)
for ( i = 0; i < A->n; ++i )
{
@@ -982,7 +982,7 @@ static real ILU_PAR( const sparse_matrix * const A, const unsigned int sweeps,
/* to get convergence, A must have unit diagonal, so apply
* transformation DAD, where D = D(1./sqrt(D(A))) */
memcpy( DAD->start, A->start, sizeof(int) * (A->n + 1) );
- #pragma omp parallel for schedule(guided) \
+ #pragma omp parallel for schedule(static) \
default(none) shared(DAD, D) private(i, pj)
for ( i = 0; i < A->n; ++i )
{
@@ -1008,8 +1008,7 @@ static real ILU_PAR( const sparse_matrix * const A, const unsigned int sweeps,
memcpy( U->val, DAD->val, sizeof(real) * (DAD->start[DAD->n]) );
/* L has unit diagonal, by convention */
- #pragma omp parallel for schedule(guided) \
- default(none) private(i)
+ #pragma omp parallel for schedule(static) default(none) private(i)
for ( i = 0; i < A->n; ++i )
{
L->val[L->start[i + 1] - 1] = 1.0;
@@ -1018,8 +1017,8 @@ static real ILU_PAR( const sparse_matrix * const A, const unsigned int sweeps,
for ( i = 0; i < sweeps; ++i )
{
/* for each nonzero in L */
- #pragma omp parallel for schedule(guided) \
- default(none) shared(DAD) private(j, k, x, y, ei_x, ei_y, sum)
+ #pragma omp parallel for schedule(static) \
+ default(none) shared(DAD) private(j, k, x, y, ei_x, ei_y, sum)
for ( j = 0; j < DAD->start[DAD->n]; ++j )
{
sum = ZERO;
@@ -1068,7 +1067,7 @@ static real ILU_PAR( const sparse_matrix * const A, const unsigned int sweeps,
}
}
- #pragma omp parallel for schedule(guided) \
+ #pragma omp parallel for schedule(static) \
default(none) shared(DAD) private(j, k, x, y, ei_x, ei_y, sum)
for ( j = 0; j < DAD->start[DAD->n]; ++j )
{
@@ -1119,7 +1118,7 @@ static real ILU_PAR( const sparse_matrix * const A, const unsigned int sweeps,
/* apply inverse transformation:
* since DAD \approx LU, then
* D^{-1}DADD^{-1} = A \approx D^{-1}LUD^{-1} */
- #pragma omp parallel for schedule(guided) \
+ #pragma omp parallel for schedule(static) \
default(none) shared(DAD, D_inv) private(i, pj)
for ( i = 0; i < DAD->n; ++i )
{
@@ -1181,7 +1180,7 @@ static real ILUT_PAR( const sparse_matrix * const A, const real * droptol,
exit( INSUFFICIENT_MEMORY );
}
- #pragma omp parallel for schedule(guided) \
+ #pragma omp parallel for schedule(static) \
default(none) shared(D, D_inv) private(i)
for ( i = 0; i < A->n; ++i )
{
@@ -1192,7 +1191,7 @@ static real ILUT_PAR( const sparse_matrix * const A, const real * droptol,
/* to get convergence, A must have unit diagonal, so apply
* transformation DAD, where D = D(1./sqrt(D(A))) */
memcpy( DAD->start, A->start, sizeof(int) * (A->n + 1) );
- #pragma omp parallel for schedule(guided) \
+ #pragma omp parallel for schedule(static) \
default(none) shared(DAD, D) private(i, pj)
for ( i = 0; i < A->n; ++i )
{
@@ -1218,7 +1217,7 @@ static real ILUT_PAR( const sparse_matrix * const A, const real * droptol,
memcpy( U_temp->val, DAD->val, sizeof(real) * (DAD->start[DAD->n]) );
/* L has unit diagonal, by convention */
- #pragma omp parallel for schedule(guided) \
+ #pragma omp parallel for schedule(static) \
default(none) private(i) shared(L_temp)
for ( i = 0; i < A->n; ++i )
{
@@ -1228,7 +1227,7 @@ static real ILUT_PAR( const sparse_matrix * const A, const real * droptol,
for ( i = 0; i < sweeps; ++i )
{
/* for each nonzero in L */
- #pragma omp parallel for schedule(guided) \
+ #pragma omp parallel for schedule(static) \
default(none) shared(DAD, L_temp, U_temp) private(j, k, x, y, ei_x, ei_y, sum)
for ( j = 0; j < DAD->start[DAD->n]; ++j )
{
@@ -1278,7 +1277,7 @@ static real ILUT_PAR( const sparse_matrix * const A, const real * droptol,
}
}
- #pragma omp parallel for schedule(guided) \
+ #pragma omp parallel for schedule(static) \
default(none) shared(DAD, L_temp, U_temp) private(j, k, x, y, ei_x, ei_y, sum)
for ( j = 0; j < DAD->start[DAD->n]; ++j )
{
@@ -1329,7 +1328,7 @@ static real ILUT_PAR( const sparse_matrix * const A, const real * droptol,
/* apply inverse transformation:
* since DAD \approx LU, then
* D^{-1}DADD^{-1} = A \approx D^{-1}LUD^{-1} */
- #pragma omp parallel for schedule(guided) \
+ #pragma omp parallel for schedule(static) \
default(none) shared(DAD, L_temp, U_temp, D_inv) private(i, pj)
for ( i = 0; i < DAD->n; ++i )
{
diff --git a/sPuReMD/src/init_md.c b/sPuReMD/src/init_md.c
index 9d2aea4cf107a48d00113ebe1383f615f637f75a..7821462a05ac2c7224a1a86a96db26d72f15e920 100644
--- a/sPuReMD/src/init_md.c
+++ b/sPuReMD/src/init_md.c
@@ -538,7 +538,7 @@ void Init_Out_Controls(reax_system *system, control_params *control,
strcpy( temp, control->sim_name );
strcat( temp, ".log" );
out_control->log = fopen( temp, "w" );
- fprintf( out_control->log, "%-6s%10s%10s%10s%10s%10s%10s%10s%10s%10s%10s%10s%10s%10s%10s\n",
+ fprintf( out_control->log, "%-6s %10s %10s %10s %10s %10s %10s %10s %10s %10s %10s %10s %10s %10s %10s\n",
"step", "total", "neighbors", "init", "bonded",
"nonbonded", "QEq", "QEq Sort", "S iters", "Pre Comp", "Pre App",
"S spmv", "S vec ops", "S orthog", "S tsolve" );
diff --git a/sPuReMD/src/mytypes.h b/sPuReMD/src/mytypes.h
index b225c508a54f14d86c59a64122f52c2f77babbda..9be7cf757d54471195b5afa892a08d2d6bb8ed53 100644
--- a/sPuReMD/src/mytypes.h
+++ b/sPuReMD/src/mytypes.h
@@ -81,17 +81,17 @@
#define C_ele 332.06371
//#define K_B 503.398008 // kcal/mol/K
#define K_B 0.831687 // amu A^2 / ps^2 / K
-#define F_CONV 1e6 / 48.88821291 / 48.88821291 // --> amu A / ps^2
+#define F_CONV (1e6 / 48.88821291 / 48.88821291) // --> amu A / ps^2
#define E_CONV 0.002391 // amu A^2 / ps^2 --> kcal/mol
#define EV_to_KCALpMOL 14.400000 // ElectronVolt --> KCAL per MOLe
#define KCALpMOL_to_EV 23.060549 // 23.020000//KCAL per MOLe --> ElectronVolt
#define ECxA_to_DEBYE 4.803204 // elem. charge * angstrom -> debye conv
#define CAL_to_JOULES 4.184000 // CALories --> JOULES
-#define JOULES_to_CAL 1/4.184000 // JOULES --> CALories
+#define JOULES_to_CAL (1/4.184000) // JOULES --> CALories
#define AMU_to_GRAM 1.6605e-24
#define ANG_to_CM 1.0e-8
#define AVOGNR 6.0221367e23
-#define P_CONV 1.0e-24 * AVOGNR * JOULES_to_CAL
+#define P_CONV (1.0e-24 * AVOGNR * JOULES_to_CAL)
#define MAX_STR 1024
#define MAX_LINE 1024
@@ -115,7 +115,6 @@
#define MAX_ITR 10
#define RESTART 50
-#define MAX_ROWS_PER_LEVEL 10000 /* triangular solve using level scheduling */
#define ZERO 0.000000000000000e+00
#define ALMOST_ZERO 1e-10
diff --git a/sPuReMD/src/print_utils.c b/sPuReMD/src/print_utils.c
index 647b3701a391e75f8623a08fcfe1e6359e86f3ce..579ba6290a5c9812cd8b1342f96f959f597d6709 100644
--- a/sPuReMD/src/print_utils.c
+++ b/sPuReMD/src/print_utils.c
@@ -616,7 +616,7 @@ void Output_Results( reax_system *system, control_params *control,
f_update = 1;
else f_update = out_control->energy_update_freq;
- fprintf( out_control->log, "%6d%10.2f%10.2f%10.2f%10.2f%10.2f%10.6f%10.6f%10.2f%10.6f%10.6f%10.6f%10.6f%10.6f%10.6f\n",
+ fprintf( out_control->log, "%6d %10.2f %10.2f %10.2f %10.2f %10.2f %10.4f %10.4f %10.2f %10.4f %10.4f %10.4f %10.4f %10.4f %10.4f\n",
data->step, t_elapsed / f_update,
data->timing.nbrs / f_update,
data->timing.init_forces / f_update,
diff --git a/sPuReMD/src/vector.c b/sPuReMD/src/vector.c
index f784ee11e73d4e2ce0b59fd00e67af6d025f903c..1525d94d45428b6eb19e4bf4b5e4ca1e6d97cfee 100644
--- a/sPuReMD/src/vector.c
+++ b/sPuReMD/src/vector.c
@@ -26,7 +26,7 @@ inline int Vector_isZero( const real * const v, const unsigned int k )
{
unsigned int i, ret = TRUE;
- #pragma omp parallel for default(none) private(i) reduction(&&: ret) schedule(guided)
+ #pragma omp parallel for default(none) private(i) reduction(&&: ret) schedule(static)
for ( i = 0; i < k; ++i )
{
if ( fabs( v[i] ) > ALMOST_ZERO )
@@ -65,7 +65,7 @@ inline void Vector_Scale( real * const dest, const real c, const real * const v,
{
unsigned int i;
- #pragma omp parallel for default(none) private(i) schedule(guided)
+ #pragma omp parallel for default(none) private(i) schedule(static)
for ( i = 0; i < k; ++i )
{
dest[i] = c * v[i];
@@ -78,7 +78,7 @@ inline void Vector_Sum( real * const dest, const real c, const real * const v, c
{
unsigned int i;
- #pragma omp parallel for default(none) private(i) schedule(guided)
+ #pragma omp parallel for default(none) private(i) schedule(static)
for ( i = 0; i < k; ++i )
{
dest[i] = c * v[i] + d * y[i];
@@ -90,7 +90,7 @@ inline void Vector_Add( real * const dest, const real c, const real * const v, c
{
unsigned int i;
- #pragma omp parallel for default(none) private(i) schedule(guided)
+ #pragma omp parallel for default(none) private(i) schedule(static)
for ( i = 0; i < k; ++i )
{
dest[i] += c * v[i];
@@ -117,7 +117,7 @@ inline real Dot( const real * const v1, const real * const v2, const unsigned in
real ret = ZERO;
unsigned int i;
- #pragma omp parallel for default(none) private(i) reduction(+: ret) schedule(guided)
+ #pragma omp parallel for default(none) private(i) reduction(+: ret) schedule(static)
for ( i = 0; i < k; ++i )
{
ret += v1[i] * v2[i];
@@ -132,7 +132,7 @@ inline real Norm( const real * const v1, const unsigned int k )
real ret = ZERO;
unsigned int i;
- #pragma omp parallel for default(none) private(i) reduction(+: ret) schedule(guided)
+ #pragma omp parallel for default(none) private(i) reduction(+: ret) schedule(static)
for ( i = 0; i < k; ++i )
{
ret += SQR( v1[i] );
diff --git a/tools/run_sim.py b/tools/run_sim.py
index 842beacfbf08a38c74065d97b0d638f87fcae19e..8bcf19bd89e14865d51da49dfdaf316e61e8e1c0 100644
--- a/tools/run_sim.py
+++ b/tools/run_sim.py
@@ -115,10 +115,7 @@ class TestCase():
print(stderr)
else:
- #TODO: fix
- start = 0.
- stop = 0.
- self._process_result(fout, stop - start, param_dict)
+ self._process_result(fout, param_dict)
fout.close()
if path.exists(temp_file):
@@ -126,7 +123,8 @@ class TestCase():
if path.exists(temp_dir):
rmdir(temp_dir)
- def _process_result(self, fout, time, param):
+ def _process_result(self, fout, param):
+ time = 0.
qeq = 0.
iters = 0.
pre_comp = 0.
@@ -143,25 +141,35 @@ class TestCase():
line = line.split()
try:
qeq = qeq + float(line[6])
- iters = iters + float(line[7])
- pre_comp = pre_comp + float(line[8])
- pre_app = pre_app + float(line[9])
- spmv = spmv + float(line[10])
+ iters = iters + float(line[8])
+ pre_comp = pre_comp + float(line[9])
+ pre_app = pre_app + float(line[10])
+ spmv = spmv + float(line[11])
cnt = cnt + 1
+ pass
except Exception:
pass
+ if line[0] == 'total:':
+ try:
+ time = float(line[1])
+ except Exception:
+ pass
cnt = cnt - 1
- qeq = qeq / cnt
- iters = iters / cnt
- pre_comp = pre_comp / cnt
- pre_app = pre_app / cnt
- spmv = spmv / cnt
+ if cnt > 0:
+ qeq = qeq / cnt
+ iters = iters / cnt
+ pre_comp = pre_comp / cnt
+ pre_app = pre_app / cnt
+ spmv = spmv / cnt
- fout.write(self.__result_body_fmt.format(path.basename(self.__geo_file).split('.')[0],
- param['nsteps'], param['qeq_solver_type'], param['qeq_solver_q_err'],
- param['pre_comp_type'], param['pre_comp_droptol'], param['pre_comp_sweeps'],
- param['pre_app_type'], param['pre_app_jacobi_iters'], pre_comp, pre_app, iters, spmv,
- qeq, param['threads'], time))
+ if cnt == int(param['nsteps']):
+ fout.write(self.__result_body_fmt.format(path.basename(self.__geo_file).split('.')[0],
+ param['nsteps'], param['qeq_solver_type'], param['qeq_solver_q_err'],
+ param['pre_comp_type'], param['pre_comp_droptol'], param['pre_comp_sweeps'],
+ param['pre_app_type'], param['pre_app_jacobi_iters'], pre_comp, pre_app, iters, spmv,
+ qeq, param['threads'], time))
+ else:
+ print('**WARNING: nsteps not correct in file {0}...'.format(log_file))
fout.flush()
@@ -193,7 +201,7 @@ if __name__ == '__main__':
data_dir = path.join(base_dir, 'data/benchmarks')
header_fmt_str = '{:15}|{:5}|{:5}|{:5}|{:5}|{:5}|{:5}|{:5}|{:5}|{:10}|{:10}|{:10}|{:10}|{:10}|{:3}|{:10}\n'
- header_str = ['Data Set', 'Steps', 'Q Tol', 'QType', 'PreCT', 'PreCD', 'PreCS', 'PreAT', 'PreAJ', 'Pre Comp',
+ header_str = ['Data Set', 'Steps', 'QType', 'Q Tol', 'PreCT', 'PreCD', 'PreCS', 'PreAT', 'PreAJ', 'Pre Comp',
'Pre App', 'Iters', 'SpMV', 'QEq', 'Thd', 'Time (s)']
body_fmt_str = '{:15} {:5} {:5} {:5} {:5} {:5} {:5} {:5} {:5} {:10.3f} {:10.3f} {:10.3f} {:10.3f} {:10.3f} {:3} {:10.3f}\n'