LCPQ/qp2
9
1
Fork 0
mirror of https://github.com/QuantumPackage/qp2.git synced 2024-10-06 08:05:58 +02:00
Code Releases Activity

fix nb of blocks autom

Browse Source
This commit is contained in:
Abdallah Ammar 2024-08-02 12:19:05 +02:00
parent f43ee8cf61
commit 76ec02812e
1 changed files with 99 additions and 90 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

189
plugins/local/tc_int/deb_tc_int_cuda.irp.f
View File

@ -107,9 +107,18 @@ subroutine deb_int2_grad1_u12_ao_gpu()
! --- ! ---
integer :: nB
integer :: sB
PROVIDE nxBlocks nyBlocks nzBlocks PROVIDE nxBlocks nyBlocks nzBlocks
PROVIDE blockxSize blockySize blockzSize PROVIDE blockxSize blockySize blockzSize
sB = 32
nB = (n_points_final_grid + sB - 1) / sB
call ezfio_set_tc_int_blockxSize(sB)
call ezfio_set_tc_int_nxBlocks(nB)
allocate(int2_grad1_u12_ao_gpu(ao_num,ao_num,n_points_final_grid,4)) allocate(int2_grad1_u12_ao_gpu(ao_num,ao_num,n_points_final_grid,4))
call wall_time(cuda_time0) call wall_time(cuda_time0)
@ -126,96 +135,96 @@ subroutine deb_int2_grad1_u12_ao_gpu()
! --- ! ---
allocate(int_fct_long_range(n_points_extra_final_grid,ao_num,ao_num)) ! allocate(int_fct_long_range(n_points_extra_final_grid,ao_num,ao_num))
allocate(grad1_u12(n_points_extra_final_grid,n_points_final_grid,4)) ! allocate(grad1_u12(n_points_extra_final_grid,n_points_final_grid,4))
allocate(int2_grad1_u12_ao(ao_num,ao_num,n_points_final_grid,4)) ! allocate(int2_grad1_u12_ao(ao_num,ao_num,n_points_final_grid,4))
!
call wall_time(cpu_time0) ! call wall_time(cpu_time0)
!
call wall_time(cpu_ttime0) ! call wall_time(cpu_ttime0)
!$OMP PARALLEL & ! !$OMP PARALLEL &
!$OMP DEFAULT (NONE) & ! !$OMP DEFAULT (NONE) &
!$OMP PRIVATE (j, i, jpoint) & ! !$OMP PRIVATE (j, i, jpoint) &
!$OMP SHARED (int_fct_long_range, ao_num, n_points_extra_final_grid, final_weight_at_r_vector_extra, aos_in_r_array_extra_transp) ! !$OMP SHARED (int_fct_long_range, ao_num, n_points_extra_final_grid, final_weight_at_r_vector_extra, aos_in_r_array_extra_transp)
!$OMP DO SCHEDULE (static) ! !$OMP DO SCHEDULE (static)
do j = 1, ao_num ! do j = 1, ao_num
do i = 1, ao_num ! do i = 1, ao_num
do jpoint = 1, n_points_extra_final_grid ! do jpoint = 1, n_points_extra_final_grid
int_fct_long_range(jpoint,i,j) = final_weight_at_r_vector_extra(jpoint) * aos_in_r_array_extra_transp(jpoint,i) * aos_in_r_array_extra_transp(jpoint,j) ! int_fct_long_range(jpoint,i,j) = final_weight_at_r_vector_extra(jpoint) * aos_in_r_array_extra_transp(jpoint,i) * aos_in_r_array_extra_transp(jpoint,j)
enddo ! enddo
enddo ! enddo
enddo ! enddo
!$OMP END DO ! !$OMP END DO
!$OMP END PARALLEL ! !$OMP END PARALLEL
call wall_time(cpu_ttime1) ! call wall_time(cpu_ttime1)
write(*,"(A,2X,F15.7)") ' wall time for int_long_range (sec) = ', (cpu_ttime1 - cpu_ttime0) ! write(*,"(A,2X,F15.7)") ' wall time for int_long_range (sec) = ', (cpu_ttime1 - cpu_ttime0)
!
!
call wall_time(cpu_ttime0) ! call wall_time(cpu_ttime0)
!$OMP PARALLEL & ! !$OMP PARALLEL &
!$OMP DEFAULT (NONE) & ! !$OMP DEFAULT (NONE) &
!$OMP PRIVATE (ipoint) & ! !$OMP PRIVATE (ipoint) &
!$OMP SHARED (n_points_final_grid, n_points_extra_final_grid, grad1_u12) ! !$OMP SHARED (n_points_final_grid, n_points_extra_final_grid, grad1_u12)
!$OMP DO ! !$OMP DO
do ipoint = 1, n_points_final_grid ! do ipoint = 1, n_points_final_grid
call get_grad1_u12_for_tc(ipoint, n_points_extra_final_grid, grad1_u12(1,ipoint,1) & ! call get_grad1_u12_for_tc(ipoint, n_points_extra_final_grid, grad1_u12(1,ipoint,1) &
, grad1_u12(1,ipoint,2) & ! , grad1_u12(1,ipoint,2) &
, grad1_u12(1,ipoint,3) & ! , grad1_u12(1,ipoint,3) &
, grad1_u12(1,ipoint,4) ) ! , grad1_u12(1,ipoint,4) )
enddo ! enddo
!$OMP END DO ! !$OMP END DO
!$OMP END PARALLEL ! !$OMP END PARALLEL
call wall_time(cpu_ttime1) ! call wall_time(cpu_ttime1)
write(*,"(A,2X,F15.7)") ' wall time for tc_int_bh (sec) = ', (cpu_ttime1 - cpu_ttime0) ! write(*,"(A,2X,F15.7)") ' wall time for tc_int_bh (sec) = ', (cpu_ttime1 - cpu_ttime0)
!
!
call wall_time(cpu_ttime0) ! call wall_time(cpu_ttime0)
do m = 1, 4 ! do m = 1, 4
call dgemm("T", "N", ao_num*ao_num, n_points_final_grid, n_points_extra_final_grid, 1.d0 & ! call dgemm("T", "N", ao_num*ao_num, n_points_final_grid, n_points_extra_final_grid, 1.d0 &
, int_fct_long_range(1,1,1), n_points_extra_final_grid, grad1_u12(1,1,m), n_points_extra_final_grid & ! , int_fct_long_range(1,1,1), n_points_extra_final_grid, grad1_u12(1,1,m), n_points_extra_final_grid &
, 0.d0, int2_grad1_u12_ao(1,1,1,m), ao_num*ao_num) ! , 0.d0, int2_grad1_u12_ao(1,1,1,m), ao_num*ao_num)
enddo ! enddo
call wall_time(cpu_ttime1) ! call wall_time(cpu_ttime1)
write(*,"(A,2X,F15.7)") ' wall time for DGEMM (sec) = ', (cpu_ttime1 - cpu_ttime0) ! write(*,"(A,2X,F15.7)") ' wall time for DGEMM (sec) = ', (cpu_ttime1 - cpu_ttime0)
!
call wall_time(cpu_time1) ! call wall_time(cpu_time1)
write(*,"(A,2X,F15.7)") ' wall time on cpu (sec) = ', (cpu_time1 - cpu_time0) ! write(*,"(A,2X,F15.7)") ' wall time on cpu (sec) = ', (cpu_time1 - cpu_time0)
!
! --- ! ! ---
!
acc_thr = 1d-12 ! acc_thr = 1d-12
err_tot = 0.d0 ! err_tot = 0.d0
nrm_tot = 0.d0 ! nrm_tot = 0.d0
!
do m = 1, 4 ! do m = 1, 4
do ipoint = 1, n_points_final_grid ! do ipoint = 1, n_points_final_grid
do j = 1, ao_num ! do j = 1, ao_num
do i = 1, ao_num ! do i = 1, ao_num
err_loc = dabs(int2_grad1_u12_ao(i,j,ipoint,m) - int2_grad1_u12_ao_gpu(i,j,ipoint,m)) ! err_loc = dabs(int2_grad1_u12_ao(i,j,ipoint,m) - int2_grad1_u12_ao_gpu(i,j,ipoint,m))
if(err_loc > acc_thr) then ! if(err_loc > acc_thr) then
print*, " error on", i, j, ipoint, m ! print*, " error on", i, j, ipoint, m
print*, " CPU res", int2_grad1_u12_ao (i,j,ipoint,m) ! print*, " CPU res", int2_grad1_u12_ao (i,j,ipoint,m)
print*, " GPU res", int2_grad1_u12_ao_gpu(i,j,ipoint,m) ! print*, " GPU res", int2_grad1_u12_ao_gpu(i,j,ipoint,m)
stop ! stop
endif ! endif
err_tot = err_tot + err_loc ! err_tot = err_tot + err_loc
nrm_tot = nrm_tot + dabs(int2_grad1_u12_ao(i,j,ipoint,m)) ! nrm_tot = nrm_tot + dabs(int2_grad1_u12_ao(i,j,ipoint,m))
enddo ! enddo
enddo ! enddo
enddo ! enddo
enddo ! enddo
!
print *, ' absolute accuracy (%) =', 100.d0 * err_tot / nrm_tot ! print *, ' absolute accuracy (%) =', 100.d0 * err_tot / nrm_tot
!
! --- ! ! ---
!
deallocate(r1, r2, rn, aos_data2) ! deallocate(r1, r2, rn, aos_data2)
deallocate(int_fct_long_range, grad1_u12) ! deallocate(int_fct_long_range, grad1_u12)
deallocate(int2_grad1_u12_ao) ! deallocate(int2_grad1_u12_ao)
deallocate(int2_grad1_u12_ao_gpu) ! deallocate(int2_grad1_u12_ao_gpu)
!
call wall_time(time1) ! call wall_time(time1)
write(*,"(A,2X,F15.7)") ' wall time for deb_int2_grad1_u12_ao_gpu (sec) = ', (time1 - time0) ! write(*,"(A,2X,F15.7)") ' wall time for deb_int2_grad1_u12_ao_gpu (sec) = ', (time1 - time0)
return return
end end