abacus-develop/source/source_psi/psi_initializer.cpp at develop · pplab/abacus-develop · GitHub

History

231 lines (220 loc) · 8.57 KB

Raw

100

101

102

103

104

105

106

107

108

109

110

111

112

113

114

115

116

117

118

119

120

121

122

123

124

125

126

127

128

129

130

131

132

133

134

135

136

137

138

139

140

141

142

143

144

145

146

147

148

149

150

151

152

153

154

155

156

157

158

159

160

161

162

163

164

165

166

167

168

169

170

171

172

173

174

175

176

177

178

179

180

181

182

183

184

185

186

187

188

189

190

191

192

193

194

195

196

197

198

199

200

201

202

203

204

205

206

207

208

209

210

211

212

213

214

215

216

217

218

219

220

221

222

223

224

225

226

227

228

229

230

231

#include "psi_initializer.h"

#include "source_base/memory.h"

// basic functions support

#include "source_base/timer.h"

#include "source_base/tool_quit.h"

// three global variables definition

#include "source_base/global_variable.h"

#include "source_io/module_parameter/parameter.h"

#ifdef __MPI

#include "source_base/parallel_reduce.h"

#endif

template <typename T>

void psi_initializer<T>::initialize(const Structure_Factor* sf,

const ModulePW::PW_Basis_K* pw_wfc,

const UnitCell* p_ucell,

const K_Vectors* p_kv_in,

const int& random_seed,

const pseudopot_cell_vnl* p_pspot_nl,

const int& rank)

{

this->sf_ = sf;

this->pw_wfc_ = pw_wfc;

this->p_ucell_ = p_ucell;

this->p_kv = p_kv_in;

this->random_seed_ = random_seed;

this->p_pspot_nl_ = p_pspot_nl;

}

template <typename T>

void psi_initializer<T>::random_t(T* psi, const int iw_start, const int iw_end, const int ik, const int mode)

{

ModuleBase::timer::tick("psi_init", "random_t");

assert(mode <= 1);

assert(iw_start >= 0);

const int ng = this->pw_wfc_->npwk[ik];

const int npwk_max = this->pw_wfc_->npwk_max;

const int npol = PARAM.globalv.npol;

// If random seed is specified, then generate random wavefunction satisfying that

// it can generate the same results using different number of processors.

if (this->random_seed_ > 0) // qianrui add 2021-8-13

{

#ifdef __MPI

srand(unsigned(this->random_seed_ + this->p_kv->ik2iktot[ik]));

#else

srand(unsigned(this->random_seed_ + ik));

#endif

const int nxy = this->pw_wfc_->fftnxy;

const int nz = this->pw_wfc_->nz;

const int nstnz = this->pw_wfc_->nst * nz;

std::vector<Real> stickrr(nz);

std::vector<Real> stickarg(nz);

std::vector<Real> tmprr(nstnz);

std::vector<Real> tmparg(nstnz);

for (int iw = iw_start; iw < iw_end; iw++)

{

// get the starting memory address of iw band

T* psi_slice = &(psi[iw * npwk_max * npol]);

for (int ipol = 0; ipol < npol; ++ipol)

{

// loop over all fft (x,y), but actually loop over all sticks

for (int ir = 0; ir < nxy; ir++)

{

// if the stick is not on present processor, then skip

if (this->pw_wfc_->fftixy2ip[ir] < 0)

{

continue;

}

// otherwise

// the following code is very time-consuming, but it can be skipped with pw_seed = 0

if (GlobalV::RANK_IN_POOL == 0)

{

// generate random number for (x,y) and all z, the stick will must

// be filled, because length of stick can be no longer than nz

// with: rr*exp(i*arg) = rr*cos(arg) + i*rr*sin(arg)

for (int iz = 0; iz < nz; iz++)

{

stickrr[iz] = std::rand() / Real(RAND_MAX); // amplitude

stickarg[iz] = std::rand() / Real(RAND_MAX); // phase

}

#ifdef __MPI // the stick-distribution is not used for non-MPI version

// then distribute the data to all processors in the pool

stick_to_pool(stickrr.data(), ir, tmprr.data());

stick_to_pool(stickarg.data(), ir, tmparg.data());

#endif

}

// then for each g-component, initialize the wavefunction value

#ifdef _OPENMP

#pragma omp parallel for schedule(static, 4096 / sizeof(T))

#endif

for (int ig = 0; ig < ng; ig++)

{

// get the correct value of "rr" and "arg" by indexing map "getigl2isz"

const int isz = this->pw_wfc_->getigl2isz(ik, ig);

const double rr = tmprr[isz];

const double arg = ModuleBase::TWO_PI * tmparg[isz];

// initialize the wavefunction value with rr * exp(i*arg)

psi_slice[ig] = this->template cast_to_T<T>(std::complex<double>(rr * cos(arg), rr * sin(arg)));

}

#ifdef _OPENMP

#pragma omp parallel for schedule(static, 4096 / sizeof(T))

#endif

for (int ig = ng; ig < npwk_max; ++ig)

{

psi_slice[ig] = static_cast<T>(0.0);

}

psi_slice += npwk_max; // move to the next polarization

}

// If random seed is not specified, then generate random wavefunction directly

// It does not guarantee the same results using different number of processors.

else

{

for (int iw = iw_start; iw < iw_end; iw++)

{

T* psi_slice = &(psi[iw * npwk_max * npol]); // get the memory to write directly. For nspin 4, nbasis*2

// donot use openmp here, because the random number generator is not thread-safe

for (int ig = 0; ig < ng; ig++)

{

const double rr = std::rand() / double(RAND_MAX);

const double arg = ModuleBase::TWO_PI * std::rand() / double(RAND_MAX);

const double gk2 = this->pw_wfc_->getgk2(ik, ig);

psi_slice[ig] = this->template cast_to_T<T>(

std::complex<double>(rr * cos(arg) / (gk2 + 1.0), rr * sin(arg) / (gk2 + 1.0)));

}

if (npol == 2)

{

for (int ig = npwk_max; ig < npwk_max + ng; ig++)

{

const double rr = std::rand() / double(RAND_MAX);

const double arg = ModuleBase::TWO_PI * std::rand() / double(RAND_MAX);

const double gk2 = this->pw_wfc_->getgk2(ik, ig - npwk_max);

psi_slice[ig] = this->template cast_to_T<T>(

std::complex<double>(rr * cos(arg) / (gk2 + 1.0), rr * sin(arg) / (gk2 + 1.0)));

}

if (mode == 1)

{

for (int iw = iw_start; iw < iw_end; iw++)

{

T* psi_slice = &(psi[iw * npwk_max * npol]);

for (int ipol = 0; ipol < npol; ipol++)

{

#ifdef _OPENMP

#pragma omp parallel for schedule(static, 4096 / sizeof(T))

#endif

for (int ig = 0; ig < ng; ig++)

{

const double gk2 = this->pw_wfc_->getgk2(ik, ig);

const Real inv_gk2 = 1.0 / (gk2 + 1.0);

psi_slice[ig] *= inv_gk2;

}

psi_slice += npwk_max;

}

ModuleBase::timer::tick("psi_init", "random_t");

}

#ifdef __MPI

template <typename T>

void psi_initializer<T>::stick_to_pool(Real* stick, const int& ir, Real* out) const

{

ModuleBase::timer::tick("psi_init", "stick_to_pool");

MPI_Status ierror;

const int is = this->ixy2is_[ir];

const int ip = this->pw_wfc_->fftixy2ip[ir];

const int nz = this->pw_wfc_->nz;

if (ip == 0 && GlobalV::RANK_IN_POOL == 0)

{

for (int iz = 0; iz < nz; iz++)

{

out[is * nz + iz] = stick[iz];

}

else if (ip == GlobalV::RANK_IN_POOL)

{

if (std::is_same<Real, double>::value)

{

MPI_Recv(stick, nz, MPI_DOUBLE, 0, ir, POOL_WORLD, &ierror);

}

else if (std::is_same<Real, float>::value)

{

MPI_Recv(stick, nz, MPI_FLOAT, 0, ir, POOL_WORLD, &ierror);

}

else

{

ModuleBase::WARNING_QUIT("psi_initializer", "stick_to_pool: Real type not supported");

}

for (int iz = 0; iz < nz; iz++)

{

out[is * nz + iz] = stick[iz];

}

else if (GlobalV::RANK_IN_POOL == 0)

{

if (std::is_same<Real, double>::value)

{

MPI_Send(stick, nz, MPI_DOUBLE, ip, ir, POOL_WORLD);

}

else if (std::is_same<Real, float>::value)

{

MPI_Send(stick, nz, MPI_FLOAT, ip, ir, POOL_WORLD);

}

else

{

ModuleBase::WARNING_QUIT("psi_initializer", "stick_to_pool: Real type not supported");

}

ModuleBase::timer::tick("psi_init", "stick_to_pool");

return;

}

#endif

// explicit instantiation

template class psi_initializer<std::complex<double>>;

template class psi_initializer<std::complex<float>>;

// gamma point calculation

template class psi_initializer<double>;

template class psi_initializer<float>;

abacus-develop/source/source_psi/psi_initializer.cpp at develop · pplab/abacus-develop · GitHub

Expand file tree

Breadcrumbs

psi_initializer.cpp

Latest commit

History

Breadcrumbs

psi_initializer.cpp

File metadata and controls