source: ThirdParty/mpqc_open/src/lib/chemistry/qc/mbptr12/transform_123inds.cc@ 1513599

//
// transform_123inds.cc
//
// Copyright (C) 2001 Edward Valeev
//
// Author: Edward Valeev <edward.valeev@chemistry.gatech.edu>
// Maintainer: EV
//
// This file is part of the SC Toolkit.
//
// The SC Toolkit is free software; you can redistribute it and/or modify
// it under the terms of the GNU Library General Public License as published by
// the Free Software Foundation; either version 2, or (at your option)
// any later version.
//
// The SC Toolkit is distributed in the hope that it will be useful,
// but WITHOUT ANY WARRANTY; without even the implied warranty of
// MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
// GNU Library General Public License for more details.
//
// You should have received a copy of the GNU Library General Public License
// along with the SC Toolkit; see the file COPYING.LIB.  If not, write to
// the Free Software Foundation, 675 Mass Ave, Cambridge, MA 02139, USA.
//
// The U.S. Government is granted a limited license as per AL 91-7.
//

#ifdef __GNUC__
#pragma implementation
#endif

#include <cmath>
#include <ostream>
#include <string>
#include <stdexcept>

#include <util/misc/formio.h>
#include <util/misc/timer.h>
#include <chemistry/qc/basis/gpetite.h>
#include <chemistry/qc/mbpt/bzerofast.h>
#include <chemistry/qc/mbpt/util.h>
#include <chemistry/qc/basis/distshpair.h>
#include <chemistry/qc/mbptr12/blas.h>
#include <chemistry/qc/mbptr12/transform_123inds.h>

using namespace std;
using namespace sc;

#define PRINT1Q 0
#define PRINT2Q 0
#define PRINT_NUM_TE_TYPES 1

// The FAST_BUT_WRONG flags is useful for exercising the communications
// layer.  It causes the first and second quarter transformation to be
// omitted, but all communication is still performed.  This permits
// problems in communications libraries to be more quickly identified.
#define FAST_BUT_WRONG 0

TwoBodyMOIntsTransform_123Inds::TwoBodyMOIntsTransform_123Inds(
  const Ref<TwoBodyMOIntsTransform>& tform, int mythread, int nthread,
  const Ref<ThreadLock>& lock, const Ref<TwoBodyInt> &tbint, double tol, int debug) :
  tform_(tform), mythread_(mythread), nthread_(nthread), lock_(lock), tbint_(tbint),
  tol_(tol), debug_(debug)
{
  timer_ = new RegionTimer();
  aoint_computed_ = 0;
  ni_ = tform_->batchsize();
  i_offset_ = 0;
}

TwoBodyMOIntsTransform_123Inds::~TwoBodyMOIntsTransform_123Inds()
{
  timer_ = 0;
}

/*
  Distribute work by SR
  
   for all PQ
    compute unique (PQ|RS)
    transform to (RS|IM) where M are all AOs for basis set 2
   end PQ

   use BLAS to transform each rsIM to rsIX
   transform RSIX to IJXS and accumulate to the tasks that holds respective ij-pairs.

  end SR
*/

void
TwoBodyMOIntsTransform_123Inds::run()
{
  Ref<MemoryGrp> mem = tform_->mem();
  Ref<MessageGrp> msg = tform_->msg();
  Ref<R12IntsAcc> ints_acc = tform_->ints_acc();
  const int me = msg->me();
  const int nproc = msg->n();
  Ref<MOIndexSpace> space1 = tform_->space1();
  Ref<MOIndexSpace> space2 = tform_->space2();
  Ref<MOIndexSpace> space3 = tform_->space3();
  Ref<MOIndexSpace> space4 = tform_->space4();

  Ref<GaussianBasisSet> bs1 = space1->basis();
  Ref<GaussianBasisSet> bs2 = space2->basis();
  Ref<GaussianBasisSet> bs3 = space3->basis();
  Ref<GaussianBasisSet> bs4 = space4->basis();
  const bool bs1_eq_bs2 = (bs1 == bs2);
  const bool bs3_eq_bs4 = (bs3 == bs4);

  const bool dynamic = tform_->dynamic();
  const double print_percent = tform_->print_percent();

  const int ni = ni_;
  const int rank1 = space1->rank();
  const int rank2 = space2->rank();
  const int rank3 = space3->rank();
  const int nfuncmax1 = bs1->max_nfunction_in_shell();
  const int nfuncmax2 = bs2->max_nfunction_in_shell();
  const int nfuncmax3 = bs3->max_nfunction_in_shell();
  const int nfuncmax4 = bs4->max_nfunction_in_shell();
  const int nsh1 = bs1->nshell();
  const int nsh2 = bs2->nshell();
  const int nsh3 = bs3->nshell();
  const int nsh4 = bs4->nshell();
  const int nbasis1 = bs1->nbasis();
  const int nbasis2 = bs2->nbasis();
  const int nbasis3 = bs3->nbasis();
  const int nbasis4 = bs4->nbasis();
  double dtol = pow(2.0,tol_);
  const size_t memgrp_blksize = tform_->memgrp_blksize()/sizeof(double);

  double** vector1 = new double*[nbasis1];
  double** vector2 = new double*[nbasis2];
  double** vector3 = new double*[nbasis3];
  vector1[0] = new double[rank1*nbasis1];
  vector2[0] = new double[rank2*nbasis2];
  vector3[0] = new double[rank3*nbasis3];
  for(int i=1; i<nbasis1; i++) vector1[i] = vector1[i-1] + rank1;
  for(int i=1; i<nbasis2; i++) vector2[i] = vector2[i-1] + rank2;
  for(int i=1; i<nbasis3; i++) vector3[i] = vector3[i-1] + rank3;
  space1->coefs().convert(vector1);
  space2->coefs().convert(vector2);
  space3->coefs().convert(vector3);

  /*-------------------------------------------------------------
    Find integrals buffers to 1/r12, r12, and [r12,T1] integrals
   -------------------------------------------------------------*/
  const int num_te_types = tform_->num_te_types();
  const double *intbuf[TwoBodyInt::num_tbint_types];
  intbuf[TwoBodyInt::eri] = tbint_->buffer(TwoBodyInt::eri);
  intbuf[TwoBodyInt::r12] = tbint_->buffer(TwoBodyInt::r12);
  intbuf[TwoBodyInt::r12t1] = tbint_->buffer(TwoBodyInt::r12t1);
  intbuf[TwoBodyInt::r12t2] = tbint_->buffer(TwoBodyInt::r12t2);

  /*-----------------------------------------------------
    Allocate buffers for partially transformed integrals
   -----------------------------------------------------*/
  double *ijsx_contrib;  // local contributions to integral_ijsx
  double *ijrx_contrib;  // local contributions to integral_ijrx
  double **rsiq_ints = new double*[num_te_types];     // quarter-transformed integrals for each RS pair
  double **rsix_ints = new double*[num_te_types];     // 2 quarter-transformed integrals for each RS pair
  for(int te_type=0;te_type<num_te_types;te_type++) {
    rsiq_ints[te_type] = new double[ni*nbasis2*nfuncmax3*nfuncmax4];
    rsix_ints[te_type] = new double[ni*rank2*nfuncmax3*nfuncmax4];
  }
  ijsx_contrib  = mem->malloc_local_double(rank2*nfuncmax4);
  if (bs3_eq_bs4)
    ijrx_contrib  = mem->malloc_local_double(rank2*nfuncmax4);
  else
    ijrx_contrib  = NULL;
  
  /*-----------------------------
    Initialize work distribution
   -----------------------------*/
  DistShellPair shellpairs(msg,nthread_,mythread_,lock_,bs4,bs3,dynamic,
                           tform_->shell_pair_data());
  shellpairs.set_debug(debug_);
  if (debug_) shellpairs.set_print_percent(print_percent/10.0);
  else shellpairs.set_print_percent(print_percent);
  int work_per_thread = bs3_eq_bs4 ? 
    ((nsh3*(nsh3+1))/2)/(nproc*nthread_) :
    (nsh3*nsh4)/(nproc*nthread_) ;
  int print_interval = work_per_thread/100;
  int time_interval = work_per_thread/10;
  int print_index = 0;
  if (print_interval == 0) print_interval = 1;
  if (time_interval == 0) time_interval = 1;
  if (work_per_thread == 0) work_per_thread = 1;

  if (debug_) {
    lock_->lock();
    ExEnv::outn() << scprintf("%d:%d: starting get_task loop",me,mythread_) << endl;
    lock_->unlock();
  }

  Ref<GenPetite4> p4list
    = construct_gpetite(bs1,bs2,bs3,bs4);

#if FAST_BUT_WRONG
  for(int te_type=0;te_type<num_te_types;te_type++) {
    bzerofast(rsiq_ints[te_type], ni*nbasis2*nfuncmax3*nfuncmax4);
    bzerofast(rsij_ints[te_type], ni*rank2*nfuncmax3*nfuncmax4);
  }
  bzerofast(ijsx_contrib, rank2*nfuncmax4);
  if (bs3_eq_bs4)
    bzerofast(ijrx_contrib, rank2*nfuncmax4);
#endif

  int R = 0;
  int S = 0;
  int RS_count = 0;
  while (shellpairs.get_task(S,R)) {
    // if bs3_eq_bs4 then S >= R always (see sc::DistShellPair)
    int nr = bs3->shell(R).nfunction();
    int r_offset = bs3->shell_to_function(R);
    
    int ns = bs4->shell(S).nfunction();
    int s_offset = bs4->shell_to_function(S);
    
    int nrs = nr*ns;

    if (debug_ > 1 && (print_index++)%print_interval == 0) {
      lock_->lock();
      ExEnv::outn() << scprintf("%d:%d: (PQ|%d %d) %d%%",
                               me,mythread_,R,S,(100*print_index)/work_per_thread)
                   << endl;
      lock_->unlock();
    }
    if (debug_ > 1 && (print_index)%time_interval == 0) {
      lock_->lock();
      ExEnv::outn() << scprintf("timer for %d:%d:",me,mythread_) << endl;
      timer_->print();
      lock_->unlock();
    }

#if !FAST_BUT_WRONG
    // Zero out 1 q.t. storage
    for(int te_type=0;te_type<num_te_types;te_type++)
      bzerofast(rsiq_ints[te_type], nrs*ni*nbasis2);

    for (int P=0; P<nsh1; P++) {
      int np = bs1->shell(P).nfunction();
      int p_offset = bs1->shell_to_function(P);

      int Qmax = (bs1_eq_bs2 ? P : nsh2-1);
      for (int Q=0; Q<=Qmax; Q++) {
        int nq = bs2->shell(Q).nfunction();
        int q_offset = bs2->shell_to_function(Q);
        
        // check if symmetry unique and compute degeneracy
        int deg = p4list->in_p4(P,Q,R,S);
        if (deg == 0)
          continue;
        double symfac = (double) deg;

        if (tbint_->log2_shell_bound(P,Q,R,S) < tol_) {
          continue;  // skip shell quartets less than tol
        }

        aoint_computed_++;

        timer_->enter("AO integrals");
        tbint_->compute_shell(P,Q,R,S);
        timer_->exit("AO integrals");

        timer_->enter("1. q.t.");

        // Begin first quarter transformation;
        // generate (iq|rs) for i active
        // if bs1_eq_bs2 then (ip|rs) are also generated
        // store the integrals as rsiq
        for(int te_type=0; te_type<num_te_types; te_type++) {
          const double *pqrs_ptr = intbuf[te_type];

          for (int bf1 = 0; bf1 < np; bf1++) {
            int p = p_offset + bf1;
            int qmax = (bs1_eq_bs2 && P == Q) ? bf1 : nq-1;

            for (int bf2 = 0; bf2 <= qmax; bf2++) {
              int q = q_offset + bf2;

              for (int bf3 = 0; bf3 < nr; bf3++) {
                int smin = (bs3_eq_bs4 && R == S) ? bf3 : 0;
                pqrs_ptr += smin;

                for (int bf4 = smin; bf4 <ns; bf4++) {

                  // Only transform integrals larger than the threshold
                  if (fabs(*pqrs_ptr) > dtol) {

                    double* rsiq_ptr = &rsiq_ints[te_type][q + nbasis2*(0 + ni*(bf4 + ns*bf3))];
                    const double* c_pi = vector1[p] + i_offset_;

                    double* rsip_ptr;
                    const double* c_qi;
                    if (bs1_eq_bs2) {
                      rsip_ptr = &rsiq_ints[te_type][p + nbasis2*(0 + ni*(bf4 + ns*bf3))];
                      c_qi = vector1[q] + i_offset_;
                    }
                    
                    double rsiq_int_contrib = *pqrs_ptr;
                    // multiply each integral by its symmetry degeneracy factor
                    rsiq_int_contrib *= symfac;
                    
                    if (bs1_eq_bs2) {

                      double rsip_int_contrib = rsiq_int_contrib;
                      if (te_type == TwoBodyInt::r12t1)
                      rsip_int_contrib = -1.0*rsiq_int_contrib;

                      if (p == q) {
                        for (int i=0; i<ni; i++) {
                          *rsiq_ptr += *c_pi++ * rsiq_int_contrib;
                          rsiq_ptr += nbasis2;
                        }
                      }
                      else {
                        // p != q
                        for (int i=0; i<ni; i++) {
                          *rsip_ptr += *c_qi++ * rsip_int_contrib;
                          rsip_ptr += nbasis2;
                          *rsiq_ptr += *c_pi++ * rsiq_int_contrib;
                          rsiq_ptr += nbasis2;
                        }
                      }
                        
                    }
                    else {

                      for (int i=0; i<ni; i++) {
                        *rsiq_ptr += *c_pi++ * rsiq_int_contrib;
                        rsiq_ptr += nbasis2;
                      }

                    } // endif bs1_eq_bs2
                  }   // endif dtol

                  pqrs_ptr++;
                } // exit bf4 loop
              }   // exit bf3 loop
            }     // exit bf2 loop
            pqrs_ptr += (nq - qmax - 1) * nrs;
          }       // exit bf1 loop
          // end of first quarter transformation
        }
        timer_->exit("1. q.t.");

        }           // exit P loop
      }             // exit Q loop
#endif // !FAST_BUT_WRONG

#if PRINT1Q
    {
      if ( me == 0 ) {
        lock_->lock();
        string filename = tform_->type() + "." + tform_->name() + ".1q.dat";
        ios_base::openmode mode = ios_base::app;
        if (RS_count == 0)
          mode = ios_base::trunc;
        ofstream ints_file(filename.c_str(),mode);

        for(int te_type=0; te_type<PRINT_NUM_TE_TYPES; te_type++) {
          for (int i = 0; i<ni; i++) {
            for (int q = 0; q<nbasis2; q++) {
              for (int r = 0; r<nr; r++) {
                int rr = r+r_offset;
                for (int s = 0; s<ns; s++) {
                  int ss = s+s_offset;
                  double value = rsiq_ints[te_type][q+nbasis2*(i+ni*(s+ns*r))];
                  ints_file << scprintf("1Q: type = %d (%d %d|%d %d) = %12.8f\n",
                                        te_type,i+i_offset_,q,rr,ss,value);
                }
              }
            }
          }
        }
        ints_file.close();
        lock_->unlock();
      }
    }
#endif

    const int nix = ni*rank2;
    const int niq = ni*nbasis2;
    
    timer_->enter("2. q.t.");
    // Begin second quarter transformation;
    // generate (ix|rs) stored as rsix

    for(int te_type=0; te_type<num_te_types; te_type++) {
      const double *rsiq_ptr = rsiq_ints[te_type];
      double *rsix_ptr = rsix_ints[te_type];

      for (int bf3 = 0; bf3 < nr; bf3++) {
        int smin = (bs3_eq_bs4 && R == S) ? bf3 : 0;
        rsiq_ptr += smin*niq;
        rsix_ptr += smin*nix;

        for (int bf4 = smin; bf4 <ns; bf4++) {

          // second quarter transform
          // ix = iq * qx
          const char notransp = 'n';
          const double one = 1.0;
          const double zero = 0.0;
          F77_DGEMM(&notransp,&notransp,&rank2,&ni,&nbasis2,&one,vector2[0],&rank2,
                    rsiq_ptr,&nbasis2,&zero,rsix_ptr,&rank2);

          rsiq_ptr += niq;
          rsix_ptr += nix;

        }
      }
    }
    timer_->exit("2. q.t.");

#if PRINT2Q
    {
      if ( me == 0 ) {
        lock_->lock();
        string filename = tform_->type() + "." + tform_->name() + ".2q.dat";
        ios_base::openmode mode = ios_base::app;
        if (RS_count == 0)
          mode = ios_base::trunc;
        ofstream ints_file(filename.c_str(),mode);

        for(int te_type=0; te_type<PRINT_NUM_TE_TYPES; te_type++) {
          for (int i = 0; i<ni; i++) {
            for (int x = 0; x<rank2; x++) {
              for (int r = 0; r<nr; r++) {
                int rr = r+r_offset;
                for (int s = 0; s<ns; s++) {
                  int ss = s+s_offset;
                  double value = rsix_ints[te_type][x+rank2*(i+ni*(s+ns*r))];
                  ints_file << scprintf("2Q: type = %d (%d %d|%d %d) = %12.8f\n",
                                        te_type,i+i_offset_,x,rr,ss,value);
                }
              }
            }
          }
        }
        ints_file.close();
        lock_->unlock();
      }
    }
#endif    
    
    timer_->enter("3. q.t.");
    // Begin third quarter transformation;
    // generate (ix|js) stored as ijsx (also generate (ix|jr), if needed)

    for(int te_type=0; te_type<num_te_types; te_type++) {
      for (int i=0; i<ni; i++) {
        for (int j=0; j<rank3; j++) {

#if !FAST_BUT_WRONG
          bzerofast(ijsx_contrib, rank2*ns);
          if (bs3_eq_bs4)
            bzerofast(ijrx_contrib, rank2*nr);

          int ij_proc =  (i*rank3 + j)%nproc;
          int ij_index = (i*rank3 + j)/nproc;
          const size_t ijsx_start = (size_t)(num_te_types*ij_index + te_type) * memgrp_blksize;
          const double *rsix_ptr = rsix_ints[te_type] + i*rank2;

          if (bs3_eq_bs4) {

            for (int bf3 = 0; bf3 < nr; bf3++) {
              int r = r_offset + bf3;
              int smin = (bs3_eq_bs4 && R == S) ? bf3 : 0;
              rsix_ptr += smin*nix;

              for (int bf4 = smin; bf4 <ns; bf4++, rsix_ptr+=nix) {
                int s = s_offset + bf4;

                // third quarter transform
                // rs = js
                // rs = jr
                
                double* ijsx_ptr = ijsx_contrib + bf4*rank2;
                double* ijrx_ptr = ijrx_contrib + bf3*rank2;
                const double* i_ptr = rsix_ptr;

                const double c_rj = vector3[r][j];
                const double c_sj = vector3[s][j];

                if (r != s) {
                  for (int x=0; x<rank2; x++) {

                    double value = *i_ptr++;
                    *ijsx_ptr++ += c_rj * value;
                    *ijrx_ptr++ += c_sj * value;

                  }
                }
                else {
                  for (int x=0; x<rank2; x++) {

                    double value = *i_ptr++;
                    *ijsx_ptr++ += c_rj * value;

                  }
                }
              }
            }
          }
          else {

            for (int bf3 = 0; bf3 < nr; bf3++) {
              int r= r_offset + bf3;
              for (int bf4 = 0; bf4 <ns; bf4++, rsix_ptr+=nix) {

                // third quarter transform
                // rs = js
                double* ijsx_ptr = ijsx_contrib + bf4*rank2;
                const double* i_ptr = rsix_ptr;

                const double c_rj = vector3[r][j];

                for (int x=0; x<rank2; x++) {

                  double value = *i_ptr++;
                  *ijsx_ptr++ += c_rj * value;

                }
              }
            }
          }

          // We now have contributions to ijxs (and ijxr) for one pair i,j,
          // all x, and s in S (r in R); send ijxs (and ijxr) to the node
          // (ij_proc) which is going to have this ij pair
#endif // !FAST_BUT_WRONG

          // Sum the ijxs_contrib to the appropriate place
          size_t ij_offset = (size_t)rank2*s_offset + ijsx_start;
          mem->sum_reduction_on_node(ijsx_contrib,
                                     ij_offset, ns*rank2, ij_proc);

          if (bs3_eq_bs4) {
            size_t ij_offset = (size_t)rank2*r_offset + ijsx_start;
            mem->sum_reduction_on_node(ijrx_contrib,
                                       ij_offset, nr*rank2, ij_proc);
          }

        } // endif j
      } // endif i
    }  // endif te_type
    timer_->exit("3. q.t.");
          
    ++RS_count;  
  }         // exit while get_task

  if (debug_) {
    lock_->lock();
    ExEnv::outn() << scprintf("%d:%d: done with get_task loop",me,mythread_) << endl;
    lock_->unlock();
  }

  for(int te_type=0; te_type<num_te_types; te_type++) {
    delete[] rsiq_ints[te_type];
    delete[] rsix_ints[te_type];
  }
  delete[] rsiq_ints;
  delete[] rsix_ints;
  mem->free_local_double(ijsx_contrib);
  if (bs3_eq_bs4)
    mem->free_local_double(ijrx_contrib);
  delete[] vector1[0]; delete[] vector1;
  delete[] vector2[0]; delete[] vector2;
  delete[] vector3[0]; delete[] vector3;
}

////////////////////////////////////////////////////////////////////////////

// Local Variables:
// mode: c++
// c-file-style: "CLJ-CONDENSED"
// End: