From 0d926a200cfc25e4e8772c93b962425313b65b07 Mon Sep 17 00:00:00 2001
From: Antoine Cyril David Hoffmann <ahoffman@spcpc606.epfl.ch>
Date: Tue, 13 Apr 2021 16:13:48 +0200
Subject: [PATCH] Update intrinsic comput. time profiler
---
matlab/profiler.m | 19 +++++++++++--------
src/basic_mod.F90 | 6 +++---
src/closure_mod.F90 | 12 ++++++++----
src/collision_mod.F90 | 7 +++++++
src/diagnose.F90 | 14 +++++++++-----
src/ghosts_mod.F90 | 8 ++++----
src/poisson.F90 | 13 +++----------
7 files changed, 45 insertions(+), 34 deletions(-)
diff --git a/matlab/profiler.m b/matlab/profiler.m
index 2b5cc433..37d363ba 100644
--- a/matlab/profiler.m
+++ b/matlab/profiler.m
@@ -7,27 +7,30 @@ DT_SIM = h5readatt(filename,'/data/input','dt');
rhs_Tc = h5read(filename,'/profiler/Tc_rhs');
adv_field_Tc = h5read(filename,'/profiler/Tc_adv_field');
+ghost_Tc = h5read(filename,'/profiler/Tc_ghost');
+clos_Tc = h5read(filename,'/profiler/Tc_clos');
+coll_Tc = h5read(filename,'/profiler/Tc_coll');
poisson_Tc = h5read(filename,'/profiler/Tc_poisson');
Sapj_Tc = h5read(filename,'/profiler/Tc_Sapj');
-diag_Tc = h5read(filename,'/profiler/Tc_diag');
-comm_Tc = h5read(filename,'/profiler/Tc_comm');
checkfield_Tc= h5read(filename,'/profiler/Tc_checkfield');
+diag_Tc = h5read(filename,'/profiler/Tc_diag');
step_Tc = h5read(filename,'/profiler/Tc_step');
Ts0D = h5read(filename,'/profiler/time');
-missing_Tc = step_Tc - rhs_Tc - adv_field_Tc - comm_Tc -...
- poisson_Tc - Sapj_Tc -diag_Tc -checkfield_Tc;
+missing_Tc = step_Tc - rhs_Tc - adv_field_Tc - ghost_Tc -clos_Tc ...
+ -coll_Tc -poisson_Tc -Sapj_Tc -checkfield_Tc -diag_Tc;
total_Tc = step_Tc;
-TIME_PER_FCT = [diff(rhs_Tc); diff(adv_field_Tc); diff(poisson_Tc); diff(comm_Tc);...
- diff(Sapj_Tc); diff(checkfield_Tc); diff(diag_Tc); diff(missing_Tc)];
-TIME_PER_FCT = reshape(TIME_PER_FCT,[numel(TIME_PER_FCT)/8,8]);
+TIME_PER_FCT = [diff(rhs_Tc); diff(adv_field_Tc); diff(ghost_Tc);...
+ diff(clos_Tc); diff(coll_Tc); diff(poisson_Tc); diff(Sapj_Tc); ...
+ diff(checkfield_Tc); diff(diag_Tc); diff(missing_Tc)];
+TIME_PER_FCT = reshape(TIME_PER_FCT,[numel(TIME_PER_FCT)/10,10]);
TIME_PER_STEP = sum(TIME_PER_FCT,2);
TIME_PER_CPU = trapz(Ts0D(2:end),TIME_PER_STEP);
%% Plots
-TIME_PER_FCT = [diff(rhs_Tc); diff(adv_field_Tc); diff(poisson_Tc); diff(comm_Tc);...
+TIME_PER_FCT = [diff(rhs_Tc); diff(adv_field_Tc); diff(poisson_Tc); diff(ghost_Tc);...
diff(Sapj_Tc); diff(checkfield_Tc); diff(diag_Tc); diff(missing_Tc)];
TIME_PER_FCT = reshape(TIME_PER_FCT,[numel(TIME_PER_FCT)/8,8]);
fig = figure;
diff --git a/src/basic_mod.F90 b/src/basic_mod.F90
index add5d775..c6a33252 100644
--- a/src/basic_mod.F90
+++ b/src/basic_mod.F90
@@ -43,9 +43,9 @@ MODULE basic
! To measure computation time
real :: start, finish
- real(dp) :: t0_rhs, t0_adv_field, t0_poisson, t0_Sapj, t0_diag, t0_checkfield, t0_step, t0_comm
- real(dp) :: t1_rhs, t1_adv_field, t1_poisson, t1_Sapj, t1_diag, t1_checkfield, t1_step, t1_comm
- real(dp) :: tc_rhs, tc_adv_field, tc_poisson, tc_Sapj, tc_diag, tc_checkfield, tc_step, tc_comm
+ real(dp) :: t0_rhs, t0_adv_field, t0_poisson, t0_Sapj, t0_diag, t0_checkfield, t0_step, t0_clos, t0_ghost, t0_coll
+ real(dp) :: t1_rhs, t1_adv_field, t1_poisson, t1_Sapj, t1_diag, t1_checkfield, t1_step, t1_clos, t1_ghost, t1_coll
+ real(dp) :: tc_rhs, tc_adv_field, tc_poisson, tc_Sapj, tc_diag, tc_checkfield, tc_step, tc_clos, tc_ghost, tc_coll
real(dp):: maxruntime = 1e9 ! Maximum simulation CPU time
INTERFACE allocate_array
diff --git a/src/closure_mod.F90 b/src/closure_mod.F90
index ab6343cd..92363210 100644
--- a/src/closure_mod.F90
+++ b/src/closure_mod.F90
@@ -26,6 +26,8 @@ SUBROUTINE apply_closure_model
sqpp1p_e = SQRT((pmaxe_dp+1)*(pmaxe_dp))
sqpp1p_i = SQRT((pmaxi_dp+1)*(pmaxi_dp))
+ CALL cpu_time(t0_clos)
+
! Negative out of bounds indices are put to zero (analytically correct)
DO ikr = ikrs,ikre
DO ikz = ikzs,ikze
@@ -74,7 +76,7 @@ SUBROUTINE apply_closure_model
moments_i(ipeg_i ,ij,ikr,ikz,:) = 0._dp
ENDDO
kernel_i(ijeg_i,ikr,ikz) = 0._dp
-
+
ENDDO
ENDDO
@@ -140,7 +142,7 @@ SUBROUTINE apply_closure_model
! Copy closure : P+2 <- P, P+1 <- P-1, J+1 <- J
DO ikr = ikrs,ikre
DO ikz = ikzs,ikze
-
+
DO ip = ipsg_e,ipeg_e
! J ghost is J+1, so we put moment J to J+1
moments_e(ip,ijeg_e,ikr,ikz,:) = moments_e(ip,ije_e,ikr,ikz,:)
@@ -158,7 +160,7 @@ SUBROUTINE apply_closure_model
moments_i(ipeg_i-1,ij,ikr,ikz,:) = moments_i(ipe_i-1,ij,ikr,ikz,:)
moments_i(ipeg_i ,ij,ikr,ikz,:) = moments_i(ipe_i ,ij,ikr,ikz,:)
ENDDO
-
+
ENDDO
ENDDO
@@ -166,6 +168,8 @@ SUBROUTINE apply_closure_model
if(my_id .EQ. 0) write(*,*) '! Closure scheme not found !'
ENDIF
-
+
+ CALL cpu_time(t1_clos)
+ tc_clos = tc_clos + (t1_clos - t0_clos)
END SUBROUTINE apply_closure_model
END module closure
diff --git a/src/collision_mod.F90 b/src/collision_mod.F90
index a2f09bb7..84e0b947 100644
--- a/src/collision_mod.F90
+++ b/src/collision_mod.F90
@@ -224,6 +224,10 @@ CONTAINS
COMPLEX(dp), DIMENSION(ips_i:ipe_i) :: TColl_distr_i
COMPLEX(dp) :: TColl
INTEGER :: ikrs_C, ikre_C, ikzs_C, ikze_C
+
+ ! Execution time start
+ CALL cpu_time(t0_coll)
+
IF (ABS(CO) .GE. 2) THEN !compute only if COSOlver matrices are used
DO ikr = ikrs,ikre
@@ -268,6 +272,9 @@ CONTAINS
ENDDO
ENDIF
+ ! Execution time end
+ CALL cpu_time(t1_coll)
+ tc_coll = tc_coll + (t1_coll - t0_coll)
END SUBROUTINE compute_TColl
!******************************************************************************!
diff --git a/src/diagnose.F90 b/src/diagnose.F90
index e8def621..51c95bc4 100644
--- a/src/diagnose.F90
+++ b/src/diagnose.F90
@@ -54,11 +54,13 @@ SUBROUTINE diagnose(kstep)
CALL creatg(fidres, "/profiler", "performance analysis")
CALL creatd(fidres, 0, dims, "/profiler/Tc_rhs", "cumulative rhs computation time")
CALL creatd(fidres, 0, dims, "/profiler/Tc_adv_field", "cumulative adv. fields computation time")
- CALL creatd(fidres, 0, dims, "/profiler/Tc_comm", "cumulative communication time")
+ CALL creatd(fidres, 0, dims, "/profiler/Tc_clos", "cumulative closure computation time")
+ CALL creatd(fidres, 0, dims, "/profiler/Tc_ghost", "cumulative communication time")
+ CALL creatd(fidres, 0, dims, "/profiler/Tc_coll", "cumulative collision computation time")
CALL creatd(fidres, 0, dims, "/profiler/Tc_poisson", "cumulative poisson computation time")
CALL creatd(fidres, 0, dims, "/profiler/Tc_Sapj", "cumulative Sapj computation time")
- CALL creatd(fidres, 0, dims, "/profiler/Tc_diag", "cumulative sym computation time")
CALL creatd(fidres, 0, dims, "/profiler/Tc_checkfield", "cumulative checkfield computation time")
+ CALL creatd(fidres, 0, dims, "/profiler/Tc_diag", "cumulative sym computation time")
CALL creatd(fidres, 0, dims, "/profiler/Tc_step", "cumulative total step computation time")
CALL creatd(fidres, 0, dims, "/profiler/time", "current simulation time")
@@ -266,7 +268,7 @@ SUBROUTINE diagnose(kstep)
CALL closef(fidres)
END IF
-
+
CALL cpu_time(t1_diag); tc_diag = tc_diag + (t1_diag - t0_diag)
END SUBROUTINE diagnose
@@ -284,11 +286,13 @@ SUBROUTINE diagnose_0d
! Time measurement data
CALL append(fidres, "/profiler/Tc_rhs", tc_rhs,ionode=0)
CALL append(fidres, "/profiler/Tc_adv_field", tc_adv_field,ionode=0)
+ CALL append(fidres, "/profiler/Tc_clos", tc_clos,ionode=0)
+ CALL append(fidres, "/profiler/Tc_ghost", tc_ghost,ionode=0)
+ CALL append(fidres, "/profiler/Tc_coll", tc_coll,ionode=0)
CALL append(fidres, "/profiler/Tc_poisson", tc_poisson,ionode=0)
CALL append(fidres, "/profiler/Tc_Sapj", tc_Sapj,ionode=0)
- CALL append(fidres, "/profiler/Tc_diag", tc_diag,ionode=0)
CALL append(fidres, "/profiler/Tc_checkfield",tc_checkfield,ionode=0)
- CALL append(fidres, "/profiler/Tc_comm", tc_comm,ionode=0)
+ CALL append(fidres, "/profiler/Tc_diag", tc_diag,ionode=0)
CALL append(fidres, "/profiler/Tc_step", tc_step,ionode=0)
CALL append(fidres, "/profiler/time", time,ionode=0)
! Processing data
diff --git a/src/ghosts_mod.F90 b/src/ghosts_mod.F90
index b3a32d21..416b7a3c 100644
--- a/src/ghosts_mod.F90
+++ b/src/ghosts_mod.F90
@@ -13,7 +13,7 @@ PUBLIC :: update_ghosts
CONTAINS
SUBROUTINE update_ghosts
- CALL cpu_time(t0_comm)
+ CALL cpu_time(t0_ghost)
IF (num_procs_p .GT. 1) THEN ! Do it only if we share the p
CALL MPI_BARRIER(MPI_COMM_WORLD,ierr)
@@ -22,9 +22,9 @@ SUBROUTINE update_ghosts
CALL MPI_BARRIER(MPI_COMM_WORLD,ierr)
CALL update_ghosts_p_i
ENDIF
-
- CALL cpu_time(t1_comm)
- tc_comm = tc_comm + (t1_comm - t0_comm)
+
+ CALL cpu_time(t1_ghost)
+ tc_ghost = tc_ghost + (t1_ghost - t0_ghost)
END SUBROUTINE update_ghosts
diff --git a/src/poisson.F90 b/src/poisson.F90
index c220d812..9aff5dae 100644
--- a/src/poisson.F90
+++ b/src/poisson.F90
@@ -65,14 +65,8 @@ SUBROUTINE poisson
! Cancel origin singularity
IF ((ikr_0 .GE. ikrs) .AND. (ikr_0 .LE. ikre)) phi(ikr_0,ikz_0) = 0
- ! Execution time end
- CALL cpu_time(t1_poisson)
- tc_poisson = tc_poisson + (t1_poisson - t0_poisson)
-
ENDIF
- CALL cpu_time(t0_comm)
-
root_bcast = 0 ! Proc zero computes phi for every p
!!!!! This is a manual way to do MPI_BCAST !!!!!!!!!!!
@@ -105,9 +99,8 @@ SUBROUTINE poisson
ENDIF
ENDIF
- CALL cpu_time(t1_comm)
- tc_comm = tc_comm + (t1_comm - t0_comm)
- !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
-
+ ! Execution time end
+ CALL cpu_time(t1_poisson)
+ tc_poisson = tc_poisson + (t1_poisson - t0_poisson)
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
END SUBROUTINE poisson
--
GitLab