Fix: Veff error when MPI_threads larger than nz

Author: dyzheng

source/module_hamilt/ks_pw/veff_pw.cpp

@@ -46,38 +46,48 @@ void Veff<OperatorPW>::act
     {
         if (this->npol == 1)
         {
-            const double* current_veff = &(this->veff[0](current_spin, 0));
             wfcpw->recip2real(tmpsi_in, porter, ik);
-            for (int ir = 0; ir < this->veff->nc; ++ir)
+            // NOTICE: when MPI threads are larger than number of Z grids
+            // veff would contain nothing, and nothing should be do in real space
+            // but the 3DFFT can not be skipped, it will cause stuck
+            if(this->veff->nc != 0)
             {
-                porter[ir] *= current_veff[ir];
+                GlobalV::ofs_running<<__FILE__<<__LINE__<<" "<<this->veff->nr<<" "<<this->veff->nc<<std::endl;
+                const double* current_veff = &(this->veff[0](current_spin, 0));
+                for (int ir = 0; ir < this->veff->nc; ++ir)
+                {
+                    porter[ir] *= current_veff[ir];
+                }
             }
             wfcpw->real2recip(porter, tmhpsi, ik, true);
         }
         else
         {
-            const double* current_veff[4];
-            for(int is=0;is<4;is++)
-            {
-                current_veff[is] = &(this->veff[0](is, 0));
-            }
             std::complex<double> *porter1 = new std::complex<double>[wfcpw->nmaxgr];
             // fft to real space and doing things.
             wfcpw->recip2real(tmpsi_in, porter, ik);
             wfcpw->recip2real(tmpsi_in + this->max_npw, porter1, ik);
             std::complex<double> sup, sdown;
-            for (int ir = 0; ir < this->veff->nc; ir++)
+            if(this->veff->nc != 0)
             {
-                sup = porter[ir] * (current_veff[0][ir] + current_veff[3][ir])
-                      + porter1[ir]
-                            * (current_veff[1][ir]
-                               - std::complex<double>(0.0, 1.0) * current_veff[2][ir]);
-                sdown = porter1[ir] * (current_veff[0][ir] - current_veff[3][ir])
-                        + porter[ir]
-                              * (current_veff[1][ir]
-                                 + std::complex<double>(0.0, 1.0) * current_veff[2][ir]);
-                porter[ir] = sup;
-                porter1[ir] = sdown;
+                const double* current_veff[4];
+                for(int is=0;is<4;is++)
+                {
+                    current_veff[is] = &(this->veff[0](is, 0));
+                }
+                for (int ir = 0; ir < this->veff->nc; ir++)
+                {
+                    sup = porter[ir] * (current_veff[0][ir] + current_veff[3][ir])
+                        + porter1[ir]
+                                * (current_veff[1][ir]
+                                - std::complex<double>(0.0, 1.0) * current_veff[2][ir]);
+                    sdown = porter1[ir] * (current_veff[0][ir] - current_veff[3][ir])
+                            + porter[ir]
+                                * (current_veff[1][ir]
+                                    + std::complex<double>(0.0, 1.0) * current_veff[2][ir]);
+                    porter[ir] = sup;
+                    porter1[ir] = sdown;
+                }
             }
             // (3) fft back to G space.
             wfcpw->real2recip(porter, tmhpsi, this->ik, true);