10
1
mirror of https://gitlab.com/scemama/qmcchem.git synced 2024-06-18 19:25:18 +02:00

Merge branch 'develop' into feature/mmap

Conflicts:
	src/mo.irp.f
This commit is contained in:
Anthony Scemama 2016-06-24 09:16:22 +02:00
commit 81cd7b858e
8 changed files with 735 additions and 331 deletions

View File

@ -11,9 +11,6 @@ mo_basis
mo_tot_num integer
mo_coef real (ao_basis_ao_num,mo_basis_mo_tot_num)
mo_classif character (mo_basis_mo_tot_num)
mo_closed_num integer =n_count_ch(mo_basis_mo_classif,size(mo_basis_mo_classif),'c')
mo_active_num integer =n_count_ch(mo_basis_mo_classif,size(mo_basis_mo_classif),'a')
mo_virtual_num integer =n_count_ch(mo_basis_mo_classif,size(mo_basis_mo_classif),'v')
mo_energy real (mo_basis_mo_tot_num)
mo_occ real (mo_basis_mo_tot_num)
mo_symmetry character*(8) (mo_basis_mo_tot_num)

View File

@ -48,11 +48,11 @@ subroutine pow_l(r,a,x1,x2,x3)
end function
BEGIN_PROVIDER [ real, ao_axis_block, ((-2*simd_sp+1):ao_block_num_8) ]
&BEGIN_PROVIDER [ real, ao_axis_grad_block_x, ((-2*simd_sp+1):ao_block_num_8) ]
&BEGIN_PROVIDER [ real, ao_axis_grad_block_y, ((-2*simd_sp+1):ao_block_num_8) ]
&BEGIN_PROVIDER [ real, ao_axis_grad_block_z, ((-2*simd_sp+1):ao_block_num_8) ]
&BEGIN_PROVIDER [ real, ao_axis_lapl_block, ((-2*simd_sp+1):ao_block_num_8) ]
BEGIN_PROVIDER [ real, ao_axis_block, (ao_block_num_8) ]
&BEGIN_PROVIDER [ real, ao_axis_grad_block_x, (ao_block_num_8) ]
&BEGIN_PROVIDER [ real, ao_axis_grad_block_y, (ao_block_num_8) ]
&BEGIN_PROVIDER [ real, ao_axis_grad_block_z, (ao_block_num_8) ]
&BEGIN_PROVIDER [ real, ao_axis_lapl_block, (ao_block_num_8) ]
implicit none
include '../types.F'
@ -111,13 +111,13 @@ end function
ao_axis_block(idx) = p023 * p10
p023 = real_of_int(pow1) * p023
ao_axis_grad_block_x(idx) = p023 * p11
ao_axis_grad_block_y(idx) = p013 * p21
ao_axis_grad_block_z(idx) = p012 * p31
ao_axis_lapl_block(idx) = real_of_int(pow1-1) * p023 * p12 &
+ real_of_int(pow2-1) * p013 * p22 &
+ real_of_int(pow3-1) * p012 * p32
ao_axis_grad_block_x(idx) = p023 * p11
ao_axis_grad_block_y(idx) = p013 * p21
ao_axis_grad_block_z(idx) = p012 * p31
enddo

View File

@ -15,7 +15,7 @@ subroutine reconfigure(ipos,w)
tmp = tmp + w(k)
enddo
dwalk_num = dble(walk_num)/tmp
integer :: kp, km
kp=0
km=0

View File

@ -44,8 +44,8 @@ END_SHELL
real, allocatable :: elec_coord_tmp(:,:,:)
integer :: mod_align
double precision :: E_loc_save(walk_num_dmc_max)
double precision :: E_loc_save_tmp(walk_num_dmc_max)
double precision :: E_loc_save(4,walk_num_dmc_max)
double precision :: E_loc_save_tmp(4,walk_num_dmc_max)
double precision :: psi_value_save(walk_num)
double precision :: psi_value_save_tmp(walk_num)
double precision :: srmc_weight(walk_num)
@ -124,7 +124,7 @@ END_SHELL
psi_grad_psi_inv_z(i) = psi_grad_psi_inv_save(i,3,i_walk)
enddo
psi_value = psi_value_save(i_walk)
E_loc = E_loc_save(i_walk)
E_loc = E_loc_save(1,i_walk)
enddo
SOFT_TOUCH elec_coord psi_grad_psi_inv_x psi_grad_psi_inv_y psi_grad_psi_inv_z psi_value E_loc
else
@ -135,7 +135,7 @@ END_SHELL
enddo
TOUCH elec_coord
psi_value_save(i_walk) = psi_value
E_loc_save(i_walk) = E_loc
E_loc_save(:,i_walk) = E_loc
endif
double precision :: p,q
@ -144,19 +144,54 @@ END_SHELL
call brownian_step(p,q,accepted,delta_x)
if ( psi_value * psi_value_save(i_walk) >= 0.d0 ) then
delta = ((E_loc+E_loc_save(i_walk))*0.5d0 - E_ref) * p
if ( delta > thr ) then
srmc_weight(i_walk) = dexp(-dtime_step*thr)
else if ( delta < -thr ) then
srmc_weight(i_walk) = dexp(dtime_step*thr)
else
! delta = (E_loc+E_loc_save(1,i_walk))*0.5d0
! delta = (5.d0 * E_loc + 8.d0 * E_loc_save(1,i_walk) - E_loc_save(2,i_walk))/12.d0
delta = (9.d0*E_loc+19.d0*E_loc_save(1,i_walk)- &
5.d0*E_loc_save(2,i_walk)+E_loc_save(3,i_walk))/24.d0
! delta = -((-251.d0*E_loc)-646.d0*E_loc_save(1,i_walk)+264.d0*E_loc_save(2,i_walk)-&
! 106.d0*E_loc_save(3,i_walk)+19.d0*E_loc_save(4,i_walk))/720.d0
delta = (delta - E_ref)*p
if (delta >= 0.d0) then
srmc_weight(i_walk) = dexp(-dtime_step*delta)
else
srmc_weight(i_walk) = 2.d0-dexp(dtime_step*delta)
endif
! if (accepted) then
! ! Compute correction to past weights
! double precision :: delta_old, delta_new
! delta_old = (9.d0*E_loc_save(1,i_walk)+19.d0*E_loc_save(2,i_walk)-&
! 5.d0*E_loc_save(3,i_walk)+E_loc_save(4,i_walk))/24.d0 - E_ref
!
!
! if (delta_old >= 0.d0) then
! srmc_weight(i_walk) = srmc_weight(i_walk) * dexp(dtime_step*delta_old)
! else
! srmc_weight(i_walk) = srmc_weight(i_walk) * (2.d0-dexp(-dtime_step*delta_old))
! endif
!
! delta_new = (-(E_loc_save_tmp(3,i_walk)-13.d0*E_loc_save_tmp(2,i_walk)&
! -13.d0*E_loc_save_tmp(1,i_walk)+E_loc))/24.d0 - E_ref
!
! if (delta_new >= 0.d0) then
! srmc_weight(i_walk) = srmc_weight(i_walk) * dexp(-dtime_step*delta_new)
! else
! srmc_weight(i_walk) = srmc_weight(i_walk) * (2.d0-dexp(dtime_step*delta_new) )
! endif
!
! endif
elec_coord(elec_num+1,1) += p*time_step
elec_coord(elec_num+1,2) = E_loc
elec_coord(elec_num+1,3) = srmc_weight(i_walk) * pop_weight_mult
do l=1,3
do i=1,elec_num+1
elec_coord_full(i,l,i_walk) = elec_coord(i,l)
enddo
enddo
@ -167,25 +202,30 @@ END_SHELL
enddo
psi_value_save(i_walk) = psi_value
E_loc_save(i_walk) = E_loc
if (accepted) then
E_loc_save(4,i_walk) = E_loc_save(3,i_walk)
E_loc_save(3,i_walk) = E_loc_save(2,i_walk)
E_loc_save(2,i_walk) = E_loc_save(1,i_walk)
E_loc_save(1,i_walk) = E_loc
endif
BEGIN_SHELL [ /usr/bin/python ]
from properties import *
t = """
if (calc_$X) then
! Kahan's summation algorithm to compute these sums reducing the rounding error:
! $X_srmc_block_walk += $X * pop_weight_mult * srmc_weight(i_walk)
! $X_2_srmc_block_walk += $X_2 * pop_weight_mult * srmc_weight(i_walk)
! $X_srmc_block_walk += $X * srmc_pop_weight_mult * srmc_weight(i_walk)
! $X_2_srmc_block_walk += $X_2 * srmc_pop_weight_mult * srmc_weight(i_walk)
! see http://en.wikipedia.org/wiki/Kahan_summation_algorithm
$X_srmc_block_walk_kahan($D2 3) = $X * pop_weight_mult * srmc_weight(i_walk) - $X_srmc_block_walk_kahan($D2 1)
$X_srmc_block_walk_kahan($D2 3) = $X * srmc_pop_weight_mult * srmc_weight(i_walk) - $X_srmc_block_walk_kahan($D2 1)
$X_srmc_block_walk_kahan($D2 2) = $X_srmc_block_walk $D1 + $X_srmc_block_walk_kahan($D2 3)
$X_srmc_block_walk_kahan($D2 1) = ($X_srmc_block_walk_kahan($D2 2) - $X_srmc_block_walk $D1 ) &
- $X_srmc_block_walk_kahan($D2 3)
$X_srmc_block_walk $D1 = $X_srmc_block_walk_kahan($D2 2)
$X_2_srmc_block_walk_kahan($D2 3) = $X_2 * pop_weight_mult * srmc_weight(i_walk) - $X_2_srmc_block_walk_kahan($D2 1)
$X_2_srmc_block_walk_kahan($D2 3) = $X_2 * srmc_pop_weight_mult * srmc_weight(i_walk) - $X_2_srmc_block_walk_kahan($D2 1)
$X_2_srmc_block_walk_kahan($D2 2) = $X_2_srmc_block_walk $D1 + $X_2_srmc_block_walk_kahan($D2 3)
$X_2_srmc_block_walk_kahan($D2 1) = ($X_2_srmc_block_walk_kahan($D2 2) - $X_2_srmc_block_walk $D1 ) &
- $X_2_srmc_block_walk_kahan($D2 3)
@ -203,7 +243,7 @@ for p in properties:
END_SHELL
block_weight += pop_weight_mult * srmc_weight(i_walk)
block_weight += srmc_pop_weight_mult * srmc_weight(i_walk)
else
srmc_weight(i_walk) = 0.d0
@ -221,15 +261,15 @@ END_SHELL
! Eventually, recompute the weight of the population
if (srmc_projection_step == 1) then
pop_weight_mult = 1.d0
srmc_pop_weight_mult = 1.d0
do k=1,srmc_projection
pop_weight_mult *= pop_weight(k)
srmc_pop_weight_mult *= srmc_pop_weight(k)
enddo
endif
! Remove contribution of the old value of the weight at the new
! projection step
pop_weight_mult *= 1.d0/pop_weight(srmc_projection_step)
srmc_pop_weight_mult *= 1.d0/srmc_pop_weight(srmc_projection_step)
! Compute the new weight of the population
double precision :: sum_weight
@ -237,10 +277,10 @@ END_SHELL
do k=1,walk_num
sum_weight += srmc_weight(k)
enddo
pop_weight(srmc_projection_step) = sum_weight/dble(walk_num)
srmc_pop_weight(srmc_projection_step) = sum_weight/dble(walk_num)
! Update the running population weight
pop_weight_mult *= pop_weight(srmc_projection_step)
srmc_pop_weight_mult *= srmc_pop_weight(srmc_projection_step)
! Reconfiguration
integer :: ipos(walk_num)
@ -257,7 +297,7 @@ END_SHELL
enddo
enddo
psi_value_save_tmp(k) = psi_value_save(k)
E_loc_save_tmp(k) = E_loc_save(k)
E_loc_save_tmp(:,k) = E_loc_save(:,k)
enddo
integer :: ipm
@ -272,7 +312,7 @@ END_SHELL
enddo
enddo
psi_value_save(k) = psi_value_save_tmp(ipm)
E_loc_save(k) = E_loc_save_tmp(ipm)
E_loc_save(:,k) = E_loc_save_tmp(:,ipm)
enddo
call system_clock(cpu1, count_rate, count_max)
@ -287,7 +327,7 @@ END_SHELL
cpu2 = cpu1
endif
SOFT_TOUCH elec_coord_full pop_weight_mult
SOFT_TOUCH elec_coord_full srmc_pop_weight_mult
first_loop = .False.
@ -314,7 +354,7 @@ END_SHELL
END_PROVIDER
BEGIN_PROVIDER [ double precision, pop_weight_mult ]
BEGIN_PROVIDER [ double precision, srmc_pop_weight_mult ]
implicit none
BEGIN_DOC
! Population weight of SRMC
@ -335,12 +375,12 @@ END_PROVIDER
srmc_projection_step = 0
END_PROVIDER
BEGIN_PROVIDER [ double precision, pop_weight, (0:srmc_projection+1) ]
BEGIN_PROVIDER [ double precision, srmc_pop_weight, (0:srmc_projection+1) ]
implicit none
BEGIN_DOC
! Population weight of SRMC
END_DOC
pop_weight = 1.d0
srmc_pop_weight = 1.d0
END_PROVIDER

File diff suppressed because it is too large Load Diff

View File

@ -148,8 +148,8 @@ END_PROVIDER
BEGIN_PROVIDER [ real, elec_dist, (elec_num_8,elec_num) ]
&BEGIN_PROVIDER [ real, elec_dist_vec_x, (elec_num_8,elec_num) ]
&BEGIN_PROVIDER [ real, elec_dist_vec_y, ((-simd_sp+1):elec_num_8,elec_num) ]
&BEGIN_PROVIDER [ real, elec_dist_vec_z, ((-2*simd_sp+1):elec_num_8,elec_num) ]
&BEGIN_PROVIDER [ real, elec_dist_vec_y, (elec_num_8,elec_num) ]
&BEGIN_PROVIDER [ real, elec_dist_vec_z, (elec_num_8,elec_num) ]
implicit none
BEGIN_DOC
! Electron-electron distances
@ -171,19 +171,15 @@ END_PROVIDER
do ie2 = 1,elec_num
real :: x, y, z
real :: x2, y2, z2
x = elec_coord(ie2,1)
y = elec_coord(ie2,2)
z = elec_coord(ie2,3)
!DIR$ VECTOR ALIGNED
!DIR$ LOOP COUNT (200)
do ie1 = 1,elec_num
elec_dist_vec_x(ie1,ie2) = elec_coord(ie1,1) - x
elec_dist_vec_y(ie1,ie2) = elec_coord(ie1,2) - y
elec_dist_vec_z(ie1,ie2) = elec_coord(ie1,3) - z
enddo
!DIR$ VECTOR ALIGNED
!DIR$ LOOP COUNT (200)
do ie1 = 1,elec_num
elec_dist(ie1,ie2) = sqrt( &
elec_dist_vec_x(ie1,ie2)*elec_dist_vec_x(ie1,ie2) + &
elec_dist_vec_y(ie1,ie2)*elec_dist_vec_y(ie1,ie2) + &

View File

@ -51,8 +51,6 @@ data = [ \
data_no_set = [\
("mo_basis_mo_tot_num" , "integer" , ""),
("mo_basis_mo_active_num" , "integer" , ""),
("mo_basis_mo_closed_num" , "integer" , ""),
("pseudo_ao_pseudo_grid" , "double precision" , "(ao_num,pseudo_lmax+pseudo_lmax+1,pseudo_lmax-0+1,nucl_num,pseudo_grid_size)"),
("pseudo_mo_pseudo_grid" , "double precision" , "(ao_num,pseudo_lmax+pseudo_lmax+1,pseudo_lmax-0+1,nucl_num,pseudo_grid_size)"),
("pseudo_pseudo_dz_k" , "double precision" , "(nucl_num,pseudo_klocmax)"),

View File

@ -187,11 +187,11 @@ BEGIN_PROVIDER [ real, mo_coef_transp_present, (num_present_mos_8,ao_num_8) ]
END_PROVIDER
BEGIN_PROVIDER [ real, mo_value_transp, ((-simd_sp+1):mo_num_8,elec_num) ]
&BEGIN_PROVIDER [ real, mo_grad_transp_x, ((-2*simd_sp+1):mo_num_8,elec_num) ]
&BEGIN_PROVIDER [ real, mo_grad_transp_y, ((-3*simd_sp+1):mo_num_8,elec_num) ]
&BEGIN_PROVIDER [ real, mo_grad_transp_z, ((-4*simd_sp+1):mo_num_8,elec_num) ]
&BEGIN_PROVIDER [ real, mo_lapl_transp, ((-5*simd_sp+1):mo_num_8,elec_num) ]
BEGIN_PROVIDER [ real, mo_value_transp, (mo_num_8,elec_num) ]
&BEGIN_PROVIDER [ real, mo_grad_transp_x, (mo_num_8,elec_num) ]
&BEGIN_PROVIDER [ real, mo_grad_transp_y, (mo_num_8,elec_num) ]
&BEGIN_PROVIDER [ real, mo_grad_transp_z, (mo_num_8,elec_num) ]
&BEGIN_PROVIDER [ real, mo_lapl_transp, (mo_num_8,elec_num) ]
implicit none
BEGIN_DOC
@ -267,7 +267,6 @@ END_PROVIDER
cycle
endif
r_inv = nucl_elec_dist_inv(k,i)
!DIR$ LOOP COUNT (500)
do j=1,mo_num
mo_value_transp(j,i) = mo_value_transp(j,i) + nucl_fitcusp_param(1,j,k) +&
r * (nucl_fitcusp_param(2,j,k) + &
@ -299,7 +298,6 @@ END_PROVIDER
! Scale off-diagonal elements
t = prepare_walkers_t
do i=1,mo_num
!DIR$ LOOP COUNT (100)
do j=1,elec_alpha_num
if (i /= j) then
mo_value_transp(i,j) *= t
@ -309,7 +307,6 @@ END_PROVIDER
mo_lapl_transp(i,j) *= t
endif
enddo
!DIR$ LOOP COUNT (100)
do j=1,elec_beta_num
if (i /= j) then
mo_value_transp(i,j+elec_alpha_num) *= t
@ -333,7 +330,6 @@ END_PROVIDER
enddo
END_PROVIDER
BEGIN_PROVIDER [ real, mo_value, (elec_num_8,mo_num) ]
implicit none
BEGIN_DOC
@ -349,7 +345,7 @@ BEGIN_PROVIDER [ real, mo_value, (elec_num_8,mo_num) ]
!DIR$ VECTOR ALIGNED
mo_value = 0.
endif
call transpose(mo_value_transp(1,1),mo_num_8+simd_sp,mo_value,elec_num_8,mo_num,elec_num)
call transpose(mo_value_transp(1,1),mo_num_8,mo_value,elec_num_8,mo_num,elec_num)
END_PROVIDER
@ -377,27 +373,11 @@ END_PROVIDER
PROVIDE primitives_reduced
endif
! Transpose x last for cache efficiency
call transpose_to_dp(mo_grad_transp_y(1,1),mo_num_8+3*simd_sp,mo_grad_y(1,1),elec_num_8,mo_num,elec_num)
call transpose_to_dp(mo_grad_transp_z(1,1),mo_num_8+4*simd_sp,mo_grad_z(1,1),elec_num_8,mo_num,elec_num)
call transpose_to_dp(mo_grad_transp_x(1,1),mo_num_8+2*simd_sp,mo_grad_x(1,1),elec_num_8,mo_num,elec_num)
END_PROVIDER
call transpose_to_dp(mo_grad_transp_y(1,1),mo_num_8,mo_grad_y(1,1),elec_num_8,mo_num,elec_num)
call transpose_to_dp(mo_grad_transp_z(1,1),mo_num_8,mo_grad_z(1,1),elec_num_8,mo_num,elec_num)
call transpose_to_dp(mo_grad_transp_x(1,1),mo_num_8,mo_grad_x(1,1),elec_num_8,mo_num,elec_num)
BEGIN_PROVIDER [ double precision, mo_grad_lapl, (4,elec_num,mo_num) ]
implicit none
BEGIN_DOC
! Gradients and laplacian
END_DOC
integer :: i,j
do j=1,mo_num
do i=1,elec_num
mo_grad_lapl(1,i,j) = mo_grad_x(i,j)
mo_grad_lapl(2,i,j) = mo_grad_y(i,j)
mo_grad_lapl(3,i,j) = mo_grad_z(i,j)
mo_grad_lapl(4,i,j) = mo_lapl (i,j)
enddo
enddo
END_PROVIDER
BEGIN_PROVIDER [ double precision, mo_lapl, (elec_num_8,mo_num) ]
@ -414,11 +394,55 @@ BEGIN_PROVIDER [ double precision, mo_lapl, (elec_num_8,mo_num) ]
!DIR$ VECTOR ALIGNED
mo_lapl = 0.d0
endif
call transpose_to_dp(mo_lapl_transp(1,1),mo_num_8+5*simd_sp,mo_lapl,elec_num_8,mo_num,elec_num)
call transpose_to_dp(mo_lapl_transp(1,1),mo_num_8,mo_lapl,elec_num_8,mo_num,elec_num)
END_PROVIDER
BEGIN_PROVIDER [ double precision, mo_grad_lapl_alpha, (4,elec_alpha_num,mo_num) ]
&BEGIN_PROVIDER [ double precision, mo_grad_lapl_beta , (4,elec_alpha_num+1:elec_num,mo_num) ]
implicit none
BEGIN_DOC
! Gradients and laplacian
END_DOC
integer :: i,j
do j=1,mo_num
do i=1,elec_alpha_num
mo_grad_lapl_alpha(1,i,j) = mo_grad_transp_x(j,i)
mo_grad_lapl_alpha(2,i,j) = mo_grad_transp_y(j,i)
mo_grad_lapl_alpha(3,i,j) = mo_grad_transp_z(j,i)
mo_grad_lapl_alpha(4,i,j) = mo_lapl_transp (j,i)
enddo
enddo
do j=1,mo_num
do i=elec_alpha_num+1,elec_num
mo_grad_lapl_beta(1,i,j) = mo_grad_transp_x(j,i)
mo_grad_lapl_beta(2,i,j) = mo_grad_transp_y(j,i)
mo_grad_lapl_beta(3,i,j) = mo_grad_transp_z(j,i)
mo_grad_lapl_beta(4,i,j) = mo_lapl_transp (j,i)
enddo
enddo
END_PROVIDER
BEGIN_PROVIDER [ double precision, mo_grad_lapl_transp, (4,mo_num,elec_num) ]
implicit none
BEGIN_DOC
! Gradients and laplacian
END_DOC
integer :: i,j
do i=1,elec_num
do j=1,mo_num
mo_grad_lapl_transp(1,j,i) = mo_grad_transp_x(j,i)
mo_grad_lapl_transp(2,j,i) = mo_grad_transp_y(j,i)
mo_grad_lapl_transp(3,j,i) = mo_grad_transp_z(j,i)
mo_grad_lapl_transp(4,j,i) = mo_lapl_transp (j,i)
enddo
enddo
END_PROVIDER
BEGIN_PROVIDER [ real, prepare_walkers_t ]
implicit none
BEGIN_DOC
@ -468,7 +492,6 @@ BEGIN_PROVIDER [ double precision , mo_value_at_nucl, (mo_num_8,nucl_num) ]
PROVIDE ao_value_p
!DIR$ LOOP COUNT (2000)
do i=1,ao_num
if (ao_nucl(i) /= k) then
ao_value_at_nucl_no_S(i) = ao_value_p(i)
@ -510,7 +533,6 @@ END_PROVIDER
point(3) = nucl_coord(k,3)+ nucl_fitcusp_radius(k)
TOUCH point
!DIR$ LOOP COUNT (2000)
do j=1,ao_num
ao_value_at_fitcusp_radius(j,k) = ao_value_p(j)
ao_grad_at_fitcusp_radius(j,k) = ao_grad_p(j,3)
@ -664,7 +686,6 @@ BEGIN_PROVIDER [ real, nucl_fitcusp_param, (4,mo_num,nucl_num) ]
cycle
endif
R = nucl_fitcusp_radius(k)
!DIR$ LOOP COUNT (500)
do i=1,mo_num
double precision :: lap_phi, grad_phi, phi, eta
lap_phi = mo_lapl_at_fitcusp_radius(i,k)
@ -730,9 +751,15 @@ subroutine sparse_full_mv(A,LDA, &
! LDC and LDA have to be factors of simd_sp
!DIR$ VECTOR ALIGNED
!DIR$ LOOP COUNT (256)
!$OMP SIMD
IRP_IF NO_PREFETCH
IRP_ELSE
call MM_PREFETCH (A(j,indices(1)),3)
call MM_PREFETCH (A(j,indices(2)),3)
call MM_PREFETCH (A(j,indices(3)),3)
call MM_PREFETCH (A(j,indices(4)),3)
IRP_ENDIF
!DIR$ SIMD
do j=1,LDC
C1(j) = 0.
C2(j) = 0.
@ -740,11 +767,11 @@ subroutine sparse_full_mv(A,LDA, &
C4(j) = 0.
C5(j) = 0.
enddo
!$OMP END SIMD
kmax2 = ishft(indices(0),-2)
kmax2 = ishft(kmax2,2)
kmax3 = indices(0)
!DIR$ LOOP COUNT (200)
do kao=1,kmax2,4
k_vec(1) = indices(kao )
k_vec(2) = indices(kao+1)
@ -761,19 +788,6 @@ subroutine sparse_full_mv(A,LDA, &
d32 = B2(kao+2)
d42 = B2(kao+3)
!DIR$ LOOP COUNT (256)
do k=0,LDA-1,$IRP_ALIGN/4
!DIR$ VECTOR ALIGNED
!$OMP SIMD
do j=1,$IRP_ALIGN/4
C1(j+k) = C1(j+k) + A(j+k,k_vec(1))*d11 + A(j+k,k_vec(2))*d21&
+ A(j+k,k_vec(3))*d31 + A(j+k,k_vec(4))*d41
C2(j+k) = C2(j+k) + A(j+k,k_vec(1))*d12 + A(j+k,k_vec(2))*d22&
+ A(j+k,k_vec(3))*d32 + A(j+k,k_vec(4))*d42
enddo
!$OMP END SIMD
enddo
d13 = B3(kao )
d23 = B3(kao+1)
d33 = B3(kao+2)
@ -784,38 +798,47 @@ subroutine sparse_full_mv(A,LDA, &
d34 = B4(kao+2)
d44 = B4(kao+3)
!DIR$ LOOP COUNT (256)
do k=0,LDA-1,$IRP_ALIGN/4
!DIR$ VECTOR ALIGNED
!$OMP SIMD
do j=1,$IRP_ALIGN/4
C3(j+k) = C3(j+k) + A(j+k,k_vec(1))*d13 + A(j+k,k_vec(2))*d23&
+ A(j+k,k_vec(3))*d33 + A(j+k,k_vec(4))*d43
C4(j+k) = C4(j+k) + A(j+k,k_vec(1))*d14 + A(j+k,k_vec(2))*d24&
+ A(j+k,k_vec(3))*d34 + A(j+k,k_vec(4))*d44
enddo
!$OMP END SIMD
enddo
d15 = B5(kao )
d25 = B5(kao+1)
d35 = B5(kao+2)
d45 = B5(kao+3)
!DIR$ LOOP COUNT (256)
do k=0,LDA-1,$IRP_ALIGN/4
!DIR$ VECTOR ALIGNED
!$OMP SIMD
!DIR$ SIMD FIRSTPRIVATE(d11,d21,d31,d41)
do j=1,$IRP_ALIGN/4
IRP_IF NO_PREFETCH
IRP_ELSE
call MM_PREFETCH (A(j+k,indices(kao+4)),3)
call MM_PREFETCH (A(j+k,indices(kao+5)),3)
call MM_PREFETCH (A(j+k,indices(kao+6)),3)
call MM_PREFETCH (A(j+k,indices(kao+7)),3)
IRP_ENDIF
C1(j+k) = C1(j+k) + A(j+k,k_vec(1))*d11 + A(j+k,k_vec(2))*d21&
+ A(j+k,k_vec(3))*d31 + A(j+k,k_vec(4))*d41
enddo
!DIR$ VECTOR ALIGNED
!DIR$ SIMD FIRSTPRIVATE(d12,d22,d32,d42,d13,d23,d33,d43)
do j=1,$IRP_ALIGN/4
C2(j+k) = C2(j+k) + A(j+k,k_vec(1))*d12 + A(j+k,k_vec(2))*d22&
+ A(j+k,k_vec(3))*d32 + A(j+k,k_vec(4))*d42
C3(j+k) = C3(j+k) + A(j+k,k_vec(1))*d13 + A(j+k,k_vec(2))*d23&
+ A(j+k,k_vec(3))*d33 + A(j+k,k_vec(4))*d43
enddo
!DIR$ VECTOR ALIGNED
!DIR$ SIMD FIRSTPRIVATE(d14,d24,d34,d44,d15,d25,d35,d45)
do j=1,$IRP_ALIGN/4
C4(j+k) = C4(j+k) + A(j+k,k_vec(1))*d14 + A(j+k,k_vec(2))*d24&
+ A(j+k,k_vec(3))*d34 + A(j+k,k_vec(4))*d44
C5(j+k) = C5(j+k) + A(j+k,k_vec(1))*d15 + A(j+k,k_vec(2))*d25&
+ A(j+k,k_vec(3))*d35 + A(j+k,k_vec(4))*d45
enddo
!$OMP END SIMD
enddo
enddo
!DIR$ LOOP COUNT (200)
do kao = kmax2+1, kmax3
k_vec(1) = indices(kao)
d11 = B1(kao)
@ -824,10 +847,9 @@ subroutine sparse_full_mv(A,LDA, &
d14 = B4(kao)
d15 = B5(kao)
!DIR$ VECTOR ALIGNED
!DIR$ LOOP COUNT (256)
do k=0,LDA-1,simd_sp
do k=0,LDA-1,$IRP_ALIGN/4
!DIR$ VECTOR ALIGNED
!$OMP SIMD
!DIR$ SIMD FIRSTPRIVATE(d11,d12,d13,d14,d15)
do j=1,$IRP_ALIGN/4
C1(j+k) = C1(j+k) + A(j+k,k_vec(1))*d11
C2(j+k) = C2(j+k) + A(j+k,k_vec(1))*d12
@ -835,7 +857,6 @@ subroutine sparse_full_mv(A,LDA, &
C4(j+k) = C4(j+k) + A(j+k,k_vec(1))*d14
C5(j+k) = C5(j+k) + A(j+k,k_vec(1))*d15
enddo
!$OMP END SIMD
enddo
enddo