Skip to content

GitLab

Commit 0df899d0 authored by Andreas Marek's avatar Andreas Marek
Browse files

Reproduce old timing functionality of ELPA2 with the new timer objects

parent b2ca1a60
Hide whitespace changes
Inline Side-by-side
Showing with 69 additions and 820 deletions
src/elpa2/elpa2_template.X90
View file @ 0df899d0
......@@ -108,11 +108,11 @@
integer(kind=ik) :: na, nev, lda, ldq, nblk, matrixCols, &
mpi_comm_rows, mpi_comm_cols, mpi_comm_all
call obj%timer%start("solve_evp_&
call obj%timer%start("elpa_solve_evp_&
&MATH_DATATYPE&
&_2stage" // &
&PRECISION_SUFFIX &
)
&_2stage_&
&PRECISION&
&")
na = obj%na
nev = obj%nev
......@@ -243,6 +243,7 @@
do_useGPU_trans_ev_tridi = .true.
endif
endif
call obj%timer%start("bandred")
if (obj%is_set("bandwidth") == 1) then
nbw=obj%get("bandwidth")
......@@ -286,7 +287,6 @@
endif
! Reduction full -> band
call obj%timer%start("bandred")
call bandred_&
&MATH_DATATYPE&
&_&
......@@ -298,9 +298,10 @@
, useQRActual &
#endif
)
call obj%timer%stop("bandred")
if (.not.(success)) return
end if ! matrix not already banded on input
call obj%timer%stop("bandred")
! Reduction band -> tridiagonal
......@@ -401,6 +402,7 @@
stop 1
endif
call obj%timer%start("trans_ev_to_full")
if(obj%is_set("bandwidth") .ne. 1) then
if ( (do_useGPU) .and. .not.(do_useGPU_trans_ev_tridi) ) then
! copy to device if we want to continue on GPU
......@@ -410,7 +412,6 @@
endif
! Backtransform stage 2
call obj%timer%start("trans_ev_to_full")
call trans_ev_band_to_full_&
&MATH_DATATYPE&
......@@ -424,7 +425,7 @@
, useQRActual &
#endif
)
call obj%timer%stop("trans_ev_to_full")
deallocate(tmat, stat=istat, errmsg=errorMessage)
if (istat .ne. 0) then
......@@ -435,12 +436,13 @@
stop 1
endif
endif
call obj%timer%stop("trans_ev_to_full")
call obj%timer%stop("solve_evp_&
call obj%timer%stop("elpa_solve_evp_&
&MATH_DATATYPE&
&_2stage" // &
&PRECISION_SUFFIX &
)
&_2stage_&
&PRECISION&
&")
1 format(a,f10.3)
end function elpa_solve_evp_&
......
src/elpa2/legacy_interface/elpa2_template.X90
View file @ 0df899d0
......@@ -63,11 +63,6 @@
#endif
useGPU) result(success)
#ifdef HAVE_DETAILED_TIMINGS
use timings
#else
use timings_dummy
#endif
use iso_c_binding
use elpa
use elpa_mpi
......@@ -97,12 +92,6 @@
integer(kind=c_int) :: successInternal
class(elpa_t), pointer :: e
call timer%start("solve_evp_&
&MATH_DATATYPE&
&_2stage_&
&PRECISION&
&_legacy_interface")
call mpi_comm_rank(mpi_comm_rows,my_prow,mpierr)
call mpi_comm_rank(mpi_comm_cols,my_pcol,mpierr)
......@@ -204,6 +193,7 @@
success = .false.
return
endif
call e%set("timings", 1)
call e%solve(a(1:lda,1:matrixCols), ev, q(1:ldq,1:matrixCols), successInternal)
if (successInternal .ne. ELPA_OK) then
......@@ -212,20 +202,65 @@
return
endif
time_evp_fwd = e%get_double("time_evp_fwd")
time_evp_solve = e%get_double("time_evp_solve")
time_evp_back = e%get_double("time_evp_back")
time_evp_fwd = e%get_time("elpa_solve_evp_&
&MATH_DATATYPE&
&_2stage_&
&PRECISION&
&","bandred")
if (my_prow==0 .and. my_pcol==0 .and. elpa_print_times) &
write(error_unit,*) 'Time bandred_real :', time_evp_fwd
time_evp_fwd = time_evp_fwd + e%get_time("elpa_solve_evp_&
&MATH_DATATYPE&
&_2stage_&
&PRECISION&
&","tridiag")
if (my_prow==0 .and. my_pcol==0 .and. elpa_print_times) &
write(error_unit,*) 'Time tridiag_band_real :',e%get_time("elpa_solve_evp_&
&MATH_DATATYPE&
&_2stage_&
&PRECISION&
&","tridiag")
time_evp_solve = e%get_time("elpa_solve_evp_&
&MATH_DATATYPE&
&_2stage_&
&PRECISION&
&","solve")
if (my_prow==0 .and. my_pcol==0 .and. elpa_print_times) &
write(error_unit,*) 'Time solve_tridi :',time_evp_solve
time_evp_back = e%get_time("elpa_solve_evp_&
&MATH_DATATYPE&
&_2stage_&
&PRECISION&
&","trans_ev_to_band")
if (my_prow==0 .and. my_pcol==0 .and. elpa_print_times) &
write(error_unit,*) 'Time trans_ev_tridi_to_band_real:',time_evp_back
time_evp_back = time_evp_back + &
e%get_time("elpa_solve_evp_&
&MATH_DATATYPE&
&_2stage_&
&PRECISION&
&","trans_ev_to_full")
if (my_prow==0 .and. my_pcol==0 .and. elpa_print_times) &
write(error_unit,*) 'Time trans_ev_band_to_full_real :',e%get_time("elpa_solve_evp_&
&MATH_DATATYPE&
&_2stage_&
&PRECISION&
&","trans_ev_to_full")
call elpa_deallocate(e)
call elpa_uninit()
call timer%stop("solve_evp_&
&MATH_DATATYPE&
&_2stage_&
&PRECISION&
&_legacy_interface")
end function
! vim: syntax=fortran
test/Fortran/elpa2/legacy_interface/legacy_complex.F90
View file @ 0df899d0
......@@ -97,9 +97,6 @@ program test_complex2_double_precision
use redirect
#endif
#ifdef HAVE_DETAILED_TIMINGS
use timings
#endif
use output_types
implicit none
......@@ -156,33 +153,6 @@ program test_complex2_double_precision
#include "../../elpa_print_headers.X90"
#ifdef HAVE_DETAILED_TIMINGS
! initialise the timing functionality
#ifdef HAVE_LIBPAPI
call timer%measure_flops(.true.)
#endif
call timer%measure_allocated_memory(.true.)
call timer%measure_virtual_memory(.true.)
call timer%measure_max_allocated_memory(.true.)
call timer%set_print_options(&
#ifdef HAVE_LIBPAPI
print_flop_count=.true., &
print_flop_rate=.true., &
#endif
print_allocated_memory = .true. , &
print_virtual_memory=.true., &
print_max_allocated_memory=.true.)
call timer%enable()
call timer%start("program: test_complex2_double_precision")
#endif
!-------------------------------------------------------------------------------
! Selection of number of processor rows/columns
! We try to set up the grid square-like, i.e. start the search for possible
......@@ -244,9 +214,6 @@ program test_complex2_double_precision
!-------------------------------------------------------------------------------
! Allocate matrices and set up a test matrix for the eigenvalue problem
#ifdef HAVE_DETAILED_TIMINGS
call timer%start("set up matrix")
#endif
allocate(a (na_rows,na_cols))
allocate(z (na_rows,na_cols))
allocate(as(na_rows,na_cols))
......@@ -255,9 +222,6 @@ program test_complex2_double_precision
call prepare_matrix(na, myid, sc_desc, a, z, as)
#ifdef HAVE_DETAILED_TIMINGS
call timer%stop("set up matrix")
#endif
! set print flag in elpa1
elpa_print_times = .true.
......@@ -314,14 +278,6 @@ program test_complex2_double_precision
deallocate(z)
deallocate(ev)
#ifdef HAVE_DETAILED_TIMINGS
call timer%stop("program: test_complex2_double_precision")
print *," "
print *,"Timings program: test_complex2_double_precision"
call timer%print("program: test_complex2_double_precision")
print *," "
print *,"End timings program: test_complex2_double_precision"
#endif
#ifdef WITH_MPI
call blacs_gridexit(my_blacs_ctxt)
call mpi_finalize(mpierr)
......
test/Fortran/elpa2/legacy_interface/legacy_complex_api.F90
View file @ 0df899d0
......@@ -99,9 +99,6 @@ program test_complex2_choose_kernel_with_api_double_precision
use redirect
#endif
#ifdef HAVE_DETAILED_TIMINGS
use timings
#endif
use output_types
implicit none
......@@ -156,33 +153,6 @@ program test_complex2_choose_kernel_with_api_double_precision
#define COMPLEXCASE
#include "../../elpa_print_headers.X90"
#ifdef HAVE_DETAILED_TIMINGS
! initialise the timing functionality
#ifdef HAVE_LIBPAPI
call timer%measure_flops(.true.)
#endif
call timer%measure_allocated_memory(.true.)
call timer%measure_virtual_memory(.true.)
call timer%measure_max_allocated_memory(.true.)
call timer%set_print_options(&
#ifdef HAVE_LIBPAPI
print_flop_count=.true., &
print_flop_rate=.true., &
#endif
print_allocated_memory = .true. , &
print_virtual_memory=.true., &
print_max_allocated_memory=.true.)
call timer%enable()
call timer%start("program: test_complex2_choose_kernel_with_api_double_precision")
#endif
!-------------------------------------------------------------------------------
! Selection of number of processor rows/columns
! We try to set up the grid square-like, i.e. start the search for possible
......@@ -314,9 +284,6 @@ program test_complex2_choose_kernel_with_api_double_precision
end if
!-------------------------------------------------------------------------------
! Allocate matrices and set up a test matrix for the eigenvalue problem
#ifdef HAVE_DETAILED_TIMINGS
call timer%start("set up matrix")
#endif
allocate(a (na_rows,na_cols))
allocate(z (na_rows,na_cols))
allocate(as(na_rows,na_cols))
......@@ -325,10 +292,6 @@ program test_complex2_choose_kernel_with_api_double_precision
call prepare_matrix(na, myid, sc_desc, a, z, as)
#ifdef HAVE_DETAILED_TIMINGS
call timer%stop("set up matrix")
#endif
! set print flag in elpa1
elpa_print_times = .true.
......@@ -447,14 +410,6 @@ program test_complex2_choose_kernel_with_api_double_precision
deallocate(z)
deallocate(ev)
#ifdef HAVE_DETAILED_TIMINGS
call timer%stop("program: test_complex2_choose_kernel_with_api_double_precision")
print *," "
print *,"Timings program: test_complex2_choose_kernel_with_api_double_precision"
call timer%print("program: test_complex2_choose_kernel_with_api_double_precision")
print *," "
print *,"End timings program: test_complex2_choose_kernel_with_api_double_precision"
#endif
#ifdef WITH_MPI
call blacs_gridexit(my_blacs_ctxt)
call mpi_finalize(mpierr)
......
test/Fortran/elpa2/legacy_interface/legacy_complex_default_kernel.F90
View file @ 0df899d0
......@@ -99,9 +99,6 @@ program test_complex2_default_kernel_double_precision
use redirect
#endif
#ifdef HAVE_DETAILED_TIMINGS
use timings
#endif
use output_types
implicit none
......@@ -156,33 +153,6 @@ program test_complex2_default_kernel_double_precision
#define COMPLEXCASE
#include "../../elpa_print_headers.X90"
#ifdef HAVE_DETAILED_TIMINGS
! initialise the timing functionality
#ifdef HAVE_LIBPAPI
call timer%measure_flops(.true.)
#endif
call timer%measure_allocated_memory(.true.)
call timer%measure_virtual_memory(.true.)
call timer%measure_max_allocated_memory(.true.)
call timer%set_print_options(&
#ifdef HAVE_LIBPAPI
print_flop_count=.true., &
print_flop_rate=.true., &
#endif
print_allocated_memory = .true. , &
print_virtual_memory=.true., &
print_max_allocated_memory=.true.)
call timer%enable()
call timer%start("program: test_complex2_default_kernel_double_precision")
#endif
!-------------------------------------------------------------------------------
! Selection of number of processor rows/columns
! We try to set up the grid square-like, i.e. start the search for possible
......@@ -260,9 +230,6 @@ program test_complex2_default_kernel_double_precision
end if
!-------------------------------------------------------------------------------
! Allocate matrices and set up a test matrix for the eigenvalue problem
#ifdef HAVE_DETAILED_TIMINGS
call timer%start("set up matrix")
#endif
allocate(a (na_rows,na_cols))
allocate(z (na_rows,na_cols))
allocate(as(na_rows,na_cols))
......@@ -271,9 +238,6 @@ program test_complex2_default_kernel_double_precision
call prepare_matrix(na, myid, sc_desc, a, z, as)
#ifdef HAVE_DETAILED_TIMINGS
call timer%stop("set up matrix")
#endif
! set print flag in elpa1
elpa_print_times = .true.
......@@ -342,14 +306,6 @@ program test_complex2_default_kernel_double_precision
deallocate(z)
deallocate(ev)
#ifdef HAVE_DETAILED_TIMINGS
call timer%stop("program: test_complex2_default_kernel_double_precision")
print *," "
print *,"Timings program: test_complex2_default_kernel_double_precision"
call timer%print("program: test_complex2_default_kernel_double_precision")
print *," "
print *,"End timings program: test_complex2_default_kernel_double_precision"
#endif
#ifdef WITH_MPI
call blacs_gridexit(my_blacs_ctxt)
call mpi_finalize(mpierr)
......
test/Fortran/elpa2/legacy_interface/legacy_complex_gpu.F90
View file @ 0df899d0
......@@ -99,9 +99,6 @@ program test_complex2_gpu_version_double_precision
use redirect
#endif
#ifdef HAVE_DETAILED_TIMINGS
use timings
#endif
use output_types
implicit none
......@@ -157,32 +154,6 @@ program test_complex2_gpu_version_double_precision
#define COMPLEXCASE
#include "../../elpa_print_headers.X90"
#ifdef HAVE_DETAILED_TIMINGS
! initialise the timing functionality
#ifdef HAVE_LIBPAPI
call timer%measure_flops(.true.)
#endif
call timer%measure_allocated_memory(.true.)
call timer%measure_virtual_memory(.true.)
call timer%measure_max_allocated_memory(.true.)
call timer%set_print_options(&
#ifdef HAVE_LIBPAPI
print_flop_count=.true., &
print_flop_rate=.true., &
#endif
print_allocated_memory = .true. , &
print_virtual_memory=.true., &
print_max_allocated_memory=.true.)
call timer%enable()
call timer%start("program: test_complex2_gpu_version_double_precision")
#endif
!-------------------------------------------------------------------------------
! Selection of number of processor rows/columns
......@@ -256,9 +227,6 @@ program test_complex2_gpu_version_double_precision
end if
!-------------------------------------------------------------------------------
! Allocate matrices and set up a test matrix for the eigenvalue problem
#ifdef HAVE_DETAILED_TIMINGS
call timer%start("set up matrix")
#endif
allocate(a (na_rows,na_cols))
allocate(z (na_rows,na_cols))
allocate(as(na_rows,na_cols))
......@@ -267,9 +235,6 @@ program test_complex2_gpu_version_double_precision
call prepare_matrix(na, myid, sc_desc, a, z, as)
#ifdef HAVE_DETAILED_TIMINGS
call timer%stop("set up matrix")
#endif
! set print flag in elpa1
elpa_print_times = .true.
......@@ -339,14 +304,6 @@ program test_complex2_gpu_version_double_precision
deallocate(z)
deallocate(ev)
#ifdef HAVE_DETAILED_TIMINGS
call timer%stop("program: test_complex2_gpu_version_double_precision")
print *," "
print *,"Timings program: test_complex2_gpu_version_double_precision"
call timer%print("program: test_complex2_gpu_version_double_precision")
print *," "
print *,"End timings program: test_complex2_gpu_version_double_precision"
#endif
#ifdef WITH_MPI
call blacs_gridexit(my_blacs_ctxt)
call mpi_finalize(mpierr)
......
test/Fortran/elpa2/legacy_interface/legacy_real.F90
View file @ 0df899d0
......@@ -96,9 +96,6 @@ program test_real2_double_precision
use elpa_mpi
#ifdef HAVE_REDIRECT
use redirect
#endif
#ifdef HAVE_DETAILED_TIMINGS
use timings
#endif
use output_types
implicit none
......@@ -151,33 +148,6 @@ program test_real2_double_precision
#define REALCASE
#include "../../elpa_print_headers.X90"
#ifdef HAVE_DETAILED_TIMINGS
! initialise the timing functionality
#ifdef HAVE_LIBPAPI
call timer%measure_flops(.true.)
#endif
call timer%measure_allocated_memory(.true.)
call timer%measure_virtual_memory(.true.)
call timer%measure_max_allocated_memory(.true.)
call timer%set_print_options(&
#ifdef HAVE_LIBPAPI
print_flop_count=.true., &
print_flop_rate=.true., &
#endif
print_allocated_memory = .true. , &
print_virtual_memory=.true., &
print_max_allocated_memory=.true.)
call timer%enable()
call timer%start("program: test_real2_double_precision")
#endif
!-------------------------------------------------------------------------------
! Selection of number of processor rows/columns
! We try to set up the grid square-like, i.e. start the search for possible
......@@ -236,9 +206,6 @@ program test_real2_double_precision
!-------------------------------------------------------------------------------
! Allocate matrices and set up a test matrix for the eigenvalue problem
#ifdef HAVE_DETAILED_TIMINGS
call timer%start("set up matrix")
#endif
allocate(a (na_rows,na_cols))
allocate(z (na_rows,na_cols))
allocate(as(na_rows,na_cols))
......@@ -247,9 +214,6 @@ program test_real2_double_precision
call prepare_matrix(na, myid, sc_desc, a, z, as)
#ifdef HAVE_DETAILED_TIMINGS
call timer%stop("set up matrix")
#endif
! set print flag in elpa1
elpa_print_times = .true.
......@@ -319,14 +283,6 @@ program test_real2_double_precision
deallocate(z)
deallocate(ev)
#ifdef HAVE_DETAILED_TIMINGS
call timer%stop("program: test_real2_double_precision")
print *," "
print *,"Timings program: test_real2_double_precision"
call timer%print("program: test_real2_double_precision")
print *," "
print *,"End timings program: test_real2_double_precision"
#endif
#ifdef WITH_MPI
call blacs_gridexit(my_blacs_ctxt)
call mpi_finalize(mpierr)
......
test/Fortran/elpa2/legacy_interface/legacy_real_api.F90
View file @ 0df899d0
......@@ -100,9 +100,6 @@ program test_real2_choose_kernel_with_api_double_precision
use redirect
#endif
#ifdef HAVE_DETAILED_TIMINGS
use timings
#endif
use output_types
implicit none
......@@ -154,32 +151,6 @@ program test_real2_choose_kernel_with_api_double_precision
#define REALCASE
#include "../../elpa_print_headers.X90"
#ifdef HAVE_DETAILED_TIMINGS
! initialise the timing functionality
#ifdef HAVE_LIBPAPI
call timer%measure_flops(.true.)