Merge pull request #5 from Panadestein/as

BLAS for value, gradient and Laplacian

Makefile

@@ -1,5 +1,5 @@
 IRPF90 = irpf90 --codelet=jastrow_full:1000 #-s nelec:10 -s nnuc:2 -s ncord:5 #-a -d
-FC     = ifort -xHost -g -mkl=sequential
+FC     = ifort -xHost -g -mkl=sequential 
 FCFLAGS= -O2 -ffree-line-length-none -I .
 NINJA  = ninja
 AR = ar 

el_nuc_el.irp.f

@@ -51,6 +51,9 @@ BEGIN_PROVIDER [ double precision, factor_een_deriv_e, (4, nelec) ]
   integer                        :: i, j, a, p, k, l, lmax, m, n
   double precision               :: cn, accu, accu2, daccu(1:4), daccu2(1:4)
 
+  factor_een_deriv_e(1:4,1:nelec) = factor_een_deriv_e_blas(1:4,1:nelec)
+  return
+
  factor_een_deriv_e(1:4,1:nelec) = 0.0d0
 
  do p = 2, ncord
@@ -76,14 +79,14 @@ BEGIN_PROVIDER [ double precision, factor_een_deriv_e, (4, nelec) ]
              accu = accu +                                           &
                  rescale_een_e(i,j,k) *                              &
                  rescale_een_n(i,a,m)
-             accu2 = accu2 +                             &
+             accu2 = accu2 +                                         &
                  rescale_een_e(i,j,k) *                              &
                  rescale_een_n(i,a,m+l)
              daccu(1:4) = daccu(1:4) +                               &
-                 rescale_een_e_deriv_e(1:4,j,i,k) *                  &
+                 rescale_een_e_deriv_e_t(1:4,i,j,k) *                &
                  rescale_een_n(i,a,m)
              daccu2(1:4) = daccu2(1:4) +                             &
-                 rescale_een_e_deriv_e(1:4,j,i,k) *                  &
+                 rescale_een_e_deriv_e_t(1:4,i,j,k) *                &
                  rescale_een_n(i,a,m+l)
 
            enddo

el_nuc_el_blas.irp.f

@@ -1,57 +1,5 @@
-BEGIN_PROVIDER [ double precision, factor_een_blas ]
+ BEGIN_PROVIDER [ double precision, factor_een_blas ]
-  implicit none
+&BEGIN_PROVIDER [ double precision, factor_een_deriv_e_blas, (4, nelec) ]
-  BEGIN_DOC
-  ! ElectronE-electron-nuclei contribution to Jastrow factor
-  !
-  ! 4124.84239750000
-  END_DOC
-  integer                        :: i, j, a, p, k, l, lmax, m, n
-  double precision               :: cn(nnuc), accu
-  double precision               :: f(nnuc,0:ncord-2,0:ncord-2)
-  double precision               :: tmp_c(nelec,nnuc,0:ncord,0:ncord-1)
-
- factor_een_blas = 0.0d0
-
- ! r_{ij}^k . R_{ja}^l -> tmp_c_{ia}^{kl}
- do k=0,ncord-1
-   call dgemm('N','N', nelec, nnuc*(ncord+1), nelec, 1.d0,           &
-       rescale_een_e(1,1,k), size(rescale_een_e,1),                  &
-       rescale_een_n(1,1,0), size(rescale_een_n,1), 0.d0,            &
-       tmp_c(1,1,0,k), size(tmp_c,1))
- enddo
-
- do p = 2, ncord
-   do k = 0, p - 1
-     m = p-k
-     if (k > 0) then
-       lmax = m
-     else
-       lmax = m - 2
-     endif
-
-     n = shiftr(m,1)
-     do l = iand(m, 1), lmax, 2
-
-       do a = 1, nnuc
-         cn(a) = cord_vect_lkp(l, k, p, typenuc_arr(a))
-       enddo
-
-       do a = 1, nnuc
-         accu = 0.d0
-         do i=1,nelec
-           accu = accu + rescale_een_n(i,a,n) * tmp_c(i,a,n+l,k)
-         enddo
-         factor_een_blas = factor_een_blas + accu * cn(a)
-       enddo
-       n = n-1
-
-     enddo
-   enddo
- enddo
-
-END_PROVIDER
-
-BEGIN_PROVIDER [ double precision, factor_een_deriv_e_blas, (4, nelec) ]
  implicit none
  BEGIN_DOC
  ! Dimensions 1-3 : dx, dy, dz
@@ -59,12 +7,13 @@ BEGIN_PROVIDER [ double precision, factor_een_deriv_e_blas, (4, nelec) ]
  END_DOC
 
  integer                        :: i, j, a, p, k, l, lmax, m, n
- double precision               :: cn(nnuc), accu(4)
+ double precision               :: cn(ncord), accu
  double precision               :: f(nnuc,0:ncord-2,0:ncord-2)
  double precision               :: tmp_c(nelec,nnuc,0:ncord,0:ncord-1)
  double precision               :: dtmp_c(4,nelec,nnuc,0:ncord,0:ncord-1)
 
- factor_een_deriv_e_blas = 0.0d0
+ factor_een_blas = 0.0d0
+ factor_een_deriv_e_blas(1:4,1:nelec) = 0.0d0
 
  ! r_{ij}^k . R_{ja}^l -> tmp_c_{ia}^{kl}
  do k=0,ncord-1
@@ -82,14 +31,15 @@ BEGIN_PROVIDER [ double precision, factor_een_deriv_e_blas, (4, nelec) ]
        dtmp_c(1,1,1,0,k), 4*size(dtmp_c,2))
  enddo
 
+
  do p = 2, ncord
    do k = 0, p - 1
-     m = p-k
+      m = p-k
-     if (k > 0) then
+      if (k > 0) then
-       lmax = m
+        lmax = m
-     else
+      else
-       lmax = m - 2
+        lmax = m - 2
-     endif
+      endif
 
      n = shiftr(m,1)
      do l = iand(m, 1), lmax, 2
@@ -99,11 +49,30 @@ BEGIN_PROVIDER [ double precision, factor_een_deriv_e_blas, (4, nelec) ]
        enddo
 
        do a = 1, nnuc
-         do i=1,nelec
+         accu = 0.d0
-           accu(1:4) = rescale_een_n(i,a,n) * dtmp_c(1:4,i,a,n+l,k) &
+
-                       + rescale_een_n_deriv_e(1:4,i,a,n) * tmp_c(i,a,n+l,k)
+         do j=1,nelec
-           factor_een_deriv_e_blas(1:4,i) = factor_een_deriv_e_blas(1:4,i) + accu(1:4) * cn(a)
+           accu = accu + rescale_een_n(j,a,n) * tmp_c(j,a,n+l,k)
+
+           factor_een_deriv_e_blas(1:4,j) = factor_een_deriv_e_blas(1:4,j) + (&
+               tmp_c(j,a,n,k) * rescale_een_n_deriv_e(1:4,j,a,n+l) + &
+               dtmp_c(1:4,j,a,n,k) * rescale_een_n(j,a,n+l) +        &
+               dtmp_c(1:4,j,a,n+l,k) * rescale_een_n(j,a,n) +        &
+               tmp_c(j,a,n+l,k)*rescale_een_n_deriv_e(1:4,j,a,n)     &
+               ) * cn(a)
+
+           factor_een_deriv_e_blas(4,j) = factor_een_deriv_e_blas(4,j) + (&
+               dtmp_c(1,j,a,n  ,k) * rescale_een_n_deriv_e(1,j,a,n+l) +&
+               dtmp_c(2,j,a,n  ,k) * rescale_een_n_deriv_e(2,j,a,n+l) +&
+               dtmp_c(3,j,a,n  ,k) * rescale_een_n_deriv_e(3,j,a,n+l) +&
+               dtmp_c(1,j,a,n+l,k) * rescale_een_n_deriv_e(1,j,a,n  ) +&
+               dtmp_c(2,j,a,n+l,k) * rescale_een_n_deriv_e(2,j,a,n  ) +&
+               dtmp_c(3,j,a,n+l,k) * rescale_een_n_deriv_e(3,j,a,n  )&
+               )*cn(a)*2.d0
+
          enddo
+         factor_een_blas = factor_een_blas + accu * cn(a)
+
        enddo
        n = n-1
 

rescale.irp.f

@@ -279,6 +279,21 @@ BEGIN_PROVIDER [double precision, rescale_een_e_deriv_e, (4, nelec, nelec, 0:nco
  enddo
 END_PROVIDER
 
+BEGIN_PROVIDER [double precision, rescale_een_e_deriv_e_t, (4, nelec, nelec, 0:ncord)]
+  implicit none
+  BEGIN_DOC
+! Transposed rescale_een_e_deriv_e
+  END_DOC
+  integer :: i,j,k,l
+  do l=0,ncord
+    do j=1,nelec
+      do i=1,nelec
+        rescale_een_e_deriv_e_t(1:4,j,i,l) = rescale_een_e_deriv_e(1:4,i,j,l)
+      enddo
+    enddo
+  enddo
+END_PROVIDER
+
 
 BEGIN_PROVIDER [double precision, elec_dist_deriv_e, (4, nelec, nelec)]
  BEGIN_DOC