bool mat_sqrt(const complex_vec_t& matrix, complex_vec_t& result) {
result.clear();
for(complex_vec_t::const_iterator i = matrix.begin(); i != matrix.end(); ++ i) {
result.push_back(sqrt(*i));
} // for
return true;
} // mat_sqrt()
bool mat_mul(complex_t scalar, const complex_vec_t& matrix, complex_vec_t& result) {
result.clear();
for(complex_vec_t::const_iterator i = matrix.begin(); i != matrix.end(); ++ i) {
result.push_back((*i) * scalar);
} // for
return true;
} // mat_mul()
bool AnalyticFormFactor::mat_sinc(unsigned int x_size, unsigned int y_size, unsigned int z_size,
const complex_vec_t& matrix, complex_vec_t& result) {
result.clear();
for(std::vector<complex_t>::const_iterator i = matrix.begin(); i != matrix.end(); ++ i) {
result.push_back(sinc(*i));
} // for
return true;
} // AnalyticFormFactor::mat_sinc()
bool AnalyticFormFactor::mat_fq_inv(unsigned int x_size, unsigned int y_size, unsigned int z_size,
const complex_vec_t& matrix, real_t y, complex_vec_t& result) {
result.clear();
for(complex_vec_t::const_iterator i = matrix.begin(); i != matrix.end(); ++ i) {
result.push_back(fq_inv(*i, y));
} // for
return true;
} // AnalyticFormFactor::mat_fq_inv()
/**
* computes element-by-element division of two matrices (matrix1 / matrix2) into result
*/
bool mat_dot_div(unsigned int nx1, unsigned int ny1, unsigned int nz1,
const complex_vec_t& matrix1,
unsigned int nx2, unsigned int ny2, unsigned int nz2,
const complex_vec_t& matrix2,
complex_vec_t& result) {
if(nx1 != nx2 || ny1 != ny2 || nz1 != nz2 || matrix1.size() != matrix2.size()) {
std::cerr << "error: matrix sizes are not the same for dot division operation"
<< std::endl;
return false;
} // if
result.clear();
complex_vec_t::const_iterator i1 = matrix1.begin();
complex_vec_t::const_iterator i2 = matrix2.begin();
for(; i1 != matrix1.end(); ++ i1, ++ i2) {
result.push_back((*i1) / (*i2));
} // for
return true;
} // mat_dot_div()
/**
* computes element-by-element product of two matrices into result
*/
bool mat_dot_prod(unsigned int x1_size, unsigned int y1_size, unsigned int z1_size,
const complex_vec_t& matrix1,
unsigned int x2_size, unsigned int y2_size, unsigned int z2_size,
const complex_vec_t& matrix2,
complex_vec_t& result) {
if(x1_size != x2_size || y1_size != y2_size || z1_size != z2_size
|| matrix1.size() != matrix2.size()) {
std::cerr << "error: matrix sizes are not the same for dot product operation" << std::endl;
return false;
} // if
result.clear();
complex_vec_t::const_iterator i1 = matrix1.begin();
complex_vec_t::const_iterator i2 = matrix2.begin();
for(; i1 != matrix1.end(); ++ i1, ++ i2) {
result.push_back((*i1) * (*i2));
} // for
return true;
} // mat_dot_prod()
bool NumericFormFactor::compute2(const char * filename, complex_vec_t &ff,
RotMatrix_t & rot
#ifdef USE_MPI
, woo::MultiNode & world_comm, std::string comm_key
#endif
) {
// initialize
init (rot, ff);
// read file
std::vector<vertex_t> vertices;
std::vector<std::vector<int>> faces;
std::vector<std::vector<int>> dummy;
ObjectShapeReader shape_reader;
if (!shape_reader.load_object (filename, vertices, faces, dummy)) {
std::cerr << "Error: shape reader failed to load triangles" << std::endl;
return false;
}
// create triangles
int num_triangles = faces.size();
triangle_t * triangles = new (std::nothrow) triangle_t [num_triangles];
for (int i = 0; i < num_triangles; i++) {
triangles[i].v1[0] = vertices[faces[i][0]-1].x;
triangles[i].v1[1] = vertices[faces[i][0]-1].y;
triangles[i].v1[2] = vertices[faces[i][0]-1].z;
triangles[i].v2[0] = vertices[faces[i][1]-1].x;
triangles[i].v2[1] = vertices[faces[i][1]-1].y;
triangles[i].v2[2] = vertices[faces[i][1]-1].z;
triangles[i].v3[0] = vertices[faces[i][2]-1].x;
triangles[i].v3[1] = vertices[faces[i][2]-1].y;
triangles[i].v3[2] = vertices[faces[i][2]-1].z;
}
//#ifndef __SSE3__
real_vec_t shape_def;
//#else
//#ifdef USE_GPU
// real_vec_t shape_def;
//#else
// real_t * shape_def = NULL;
//#endif
//#endif
//unsigned int num_triangles = read_shapes_file(filename, shape_def);
#ifdef USE_MPI
int num_procs = world_comm.size(comm_key);
int rank = world_comm.rank(comm_key);
bool master = world_comm.is_master(comm_key);
#else
bool master = true;
#endif
if(master) {
std::cout << "-- Numerical form factor computation ..." << std::endl
<< "** Using input shape file: " << filename << std::endl
<< "** Number of input triangles: " << num_triangles << std::endl
<< "** Q-grid resolution (q-points): " << nqy_ << std::endl
#ifdef USE_MPI
<< "** Number of processes requested: " << num_procs << std::endl
#endif
<< std::flush;
} // if
// copy q-points
real_t * qx = new (std::nothrow) real_t [nqy_];
if (qx == NULL) {
std::cerr << "Error: failure in allocation memeroy." << std::endl;
return false;
}
for (int i = 0; i < nqy_; i++ ) qx[i] = QGrid::instance().qx(i);
real_t * qy = new (std::nothrow) real_t [nqy_];
if (qy == NULL) {
std::cerr << "Error: failure in allocation memeroy." << std::endl;
return false;
}
for (int i = 0; i < nqy_; i++) qy[i] = QGrid::instance().qy(i);
#ifdef FF_NUM_GPU
cucomplex_t * qz = new (std::nothrow) cucomplex_t [nqz_];
if (qz == NULL) {
std::cerr << "Error: failure in memeroy allocation." << std::endl;
return 0;
}
for (int i = 0; i < nqz_; i++) {
qz[i].x = QGrid::instance().qz_extended(i).real();
qz[i].y = QGrid::instance().qz_extended(i).imag();
}
#else
complex_t * qz = new (std::nothrow) complex_t [nqz_];
if (qz == NULL) {
std::cerr << "Error: failure in memeroy allocation." << std::endl;
return 0;
}
for (int i = 0; i < nqz_; i++) qz[i] = QGrid::instance().qz_extended(i);
#endif
real_t compute_time = 0.;
#ifdef FF_NUM_GPU
cucomplex_t * p_ff = NULL;
// call kernel
if (num_triangles != gff_.compute_exact_triangle(triangles, num_triangles,
p_ff, nqy_, qx, qy,
nqz_, qz, rot_, compute_time)) {
std::cerr << "Calculation of numerical form-factor failed" << std::endl;
return false;
}
for (int i = 0; i < nqz_; i++) ff.push_back (complex_t(p_ff[i].x, p_ff[i].y));
std::cout << "** FF GPU compute time: " << compute_time << " ms." << std::endl;
#else
complex_t * p_ff = new (std::nothrow) complex_t[nqz_];
if (p_ff == NULL){
std::cerr << "Error: failed to allocate memory of size: "
<< nqz_ * sizeof(complex_t) << std::endl;
return false;
}
if (num_triangles != cff_.compute_exact_triangle(triangles, num_triangles,
p_ff, nqy_, qx, qy,
nqz_, qz, rot_, compute_time)) {
std::cerr << "Calculation of numerical form-factor failed" << std::endl;
return false;
}
for (int i = 0; i < nqz_; i++) ff.push_back(p_ff[i]);
std::cout << "** FF CPU compute time: " << compute_time << " ms." << std::endl;
#endif
delete [] qx;
delete [] qy;
delete [] qz;
delete [] triangles;
if (p_ff != NULL) delete [] p_ff;
return true;
}
bool NumericFormFactor::compute(const char * filename, complex_vec_t & ff,
RotMatrix_t & rot
#ifdef USE_MPI
, woo::MultiNode &world_comm, std::string comm_key
#endif
){
real_t comp_time = 0.0;
// initialize
init (rot, ff);
unsigned int nqy = QGrid::instance().nqy();
unsigned int nqz = QGrid::instance().nqz_extended();
// warning: all procs read the shape file!!!!
// TODO: improve to parallel IO, or one proc reading and sending to all ...
// #ifndef __SSE3__
real_vec_t shape_def;
// #else
// #ifdef USE_GPU
// real_vec_t shape_def;
// #else
// real_t* shape_def = NULL;
// #endif
// #endif
// use the new file reader instead ...
unsigned int num_triangles = read_shapes_file(filename, shape_def);
// TODO ... <--- sadly all procs read this! IMPROVE!!!
#ifdef USE_MPI
int num_procs = world_comm.size(comm_key);
int rank = world_comm.rank(comm_key);
bool master = world_comm.is_master(comm_key);
#else
bool master = true;
#endif
if(master) {
std::cout << "-- Numerical form factor computation ..." << std::endl
<< "** Using input shape file: " << filename << std::endl
<< "** Number of input triangles: " << num_triangles << std::endl
<< "** Q-grid resolution (q-points): " << nqz << std::endl
#ifdef USE_MPI
<< "** Number of processes requested: " << num_procs << std::endl
#endif
<< std::flush;
} // if
if(num_triangles < 1) {
std::cerr << "error: no triangles found in specified definition file" << std::endl;
return false;
} // if
// FIXME: this is a yucky temporary fix ... fix properly ...
real_t* qx = new (std::nothrow) real_t[nqy]();
real_t* qy = new (std::nothrow) real_t[nqy]();
#ifdef FF_NUM_GPU
cucomplex_t* qz = new (std::nothrow) cucomplex_t[nqz]();
#else
complex_t* qz = new (std::nothrow) complex_t[nqz]();
#endif
// create qy_and qz using qgrid instance
for(unsigned int i = 0; i < nqy; ++ i) qx[i] = QGrid::instance().qx(i);
for(unsigned int i = 0; i < nqy; ++ i) qy[i] = QGrid::instance().qy(i);
for(unsigned int i = 0; i < nqz; ++ i) {
#ifdef FF_NUM_GPU
qz[i].x = QGrid::instance().qz_extended(i).real();
qz[i].y = QGrid::instance().qz_extended(i).imag();
#else
qz[i] = QGrid::instance().qz_extended(i);
#endif
} // for
#ifdef FF_NUM_GPU
cucomplex_t *p_ff = NULL;
#else
complex_t *p_ff = NULL;
#endif
real_t kernel_time = 0.;
unsigned int ret_numtriangles = 0;
#ifdef FF_NUM_GPU // use GPU
ret_numtriangles = gff_.compute_approx_triangle(shape_def,
p_ff, nqy, qx, qy, nqz, qz, rot_, kernel_time);
for (int i = 0; i < nqz; i++) ff.push_back(complex_t(p_ff[i].x, p_ff[i].y));
std::cout << "** FF GPU compute time: " << kernel_time << " ms." << std::endl;
#else // use only CPU
ret_numtriangles = cff_.compute_approx_triangle(shape_def,
p_ff, nqy, qx, qy, nqz, qz, rot_, kernel_time);
for (int i = 0; i < nqz; i++) ff.push_back(p_ff[i]);
std::cout << "** FF CPU compute time: " << kernel_time << " ms." << std::endl;
#endif
if(p_ff != NULL) delete[] p_ff;
delete[] qz;
delete[] qy;
delete[] qx;
}
bool mat_exp(complex_vec_t& matrix, complex_vec_t& result) {
result.clear();
for(complex_vec_t::iterator i = matrix.begin(); i != matrix.end(); ++ i)
result.push_back(exp(*i));
} // mat_exp()