64-BitBrainstorm_2026/AMSS-NCKU
8
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Batch constraint norm reductions

Browse Source 
(cherry picked from commit 3a58273501)
This commit is contained in:
CGH0S7
2026-04-13 11:48:02 +08:00
committed by ianchb
parent 0396701572
commit b713e5a9be
7 changed files with 275 additions and 95 deletions
Download Patch File Download Diff File Expand all files Collapse all files

70
AMSS_NCKU_source/Parallel.C
View File

@@ -5453,6 +5453,41 @@ double Parallel::L2Norm(Patch *Pat, var *vf)
return tvf;
}
void Parallel::L2Norm7(Patch *Pat, var **vf, double *norms)
{
int myrank;
MPI_Comm_rank(MPI_COMM_WORLD, &myrank);
double tvf[7], dtvf[7];
int BDW = ghost_width;
for (int i = 0; i < 7; i++)
dtvf[i] = 0;
MyList<Block> *BP = Pat->blb;
while (BP)
{
Block *cg = BP->data;
if (myrank == cg->rank)
{
f_l2normhelper7(cg->shape, cg->X[0], cg->X[1], cg->X[2],
Pat->bbox[0], Pat->bbox[1], Pat->bbox[2],
Pat->bbox[3], Pat->bbox[4], Pat->bbox[5],
cg->fgfs[vf[0]->sgfn], cg->fgfs[vf[1]->sgfn], cg->fgfs[vf[2]->sgfn],
cg->fgfs[vf[3]->sgfn], cg->fgfs[vf[4]->sgfn], cg->fgfs[vf[5]->sgfn],
cg->fgfs[vf[6]->sgfn], tvf, BDW);
for (int i = 0; i < 7; i++)
dtvf[i] += tvf[i];
}
if (BP == Pat->ble)
break;
BP = BP->next;
}
MPI_Allreduce(dtvf, tvf, 7, MPI_DOUBLE, MPI_SUM, MPI_COMM_WORLD);
for (int i = 0; i < 7; i++)
norms[i] = sqrt(tvf[i]);
}
double Parallel::L2Norm(Patch *Pat, var *vf, MPI_Comm Comm_here)
{
int myrank;
@@ -5484,6 +5519,41 @@ double Parallel::L2Norm(Patch *Pat, var *vf, MPI_Comm Comm_here)
return tvf;
}
void Parallel::L2Norm7(Patch *Pat, var **vf, double *norms, MPI_Comm Comm_here)
{
int myrank;
MPI_Comm_rank(MPI_COMM_WORLD, &myrank);
double tvf[7], dtvf[7];
int BDW = ghost_width;
for (int i = 0; i < 7; i++)
dtvf[i] = 0;
MyList<Block> *BP = Pat->blb;
while (BP)
{
Block *cg = BP->data;
if (myrank == cg->rank)
{
f_l2normhelper7(cg->shape, cg->X[0], cg->X[1], cg->X[2],
Pat->bbox[0], Pat->bbox[1], Pat->bbox[2],
Pat->bbox[3], Pat->bbox[4], Pat->bbox[5],
cg->fgfs[vf[0]->sgfn], cg->fgfs[vf[1]->sgfn], cg->fgfs[vf[2]->sgfn],
cg->fgfs[vf[3]->sgfn], cg->fgfs[vf[4]->sgfn], cg->fgfs[vf[5]->sgfn],
cg->fgfs[vf[6]->sgfn], tvf, BDW);
for (int i = 0; i < 7; i++)
dtvf[i] += tvf[i];
}
if (BP == Pat->ble)
break;
BP = BP->next;
}
MPI_Allreduce(dtvf, tvf, 7, MPI_DOUBLE, MPI_SUM, Comm_here);
for (int i = 0; i < 7; i++)
norms[i] = sqrt(tvf[i]);
}
void Parallel::checkgsl(MyList<Parallel::gridseg> *pp, bool first_only)
{
int myrank = 0;

2
AMSS_NCKU_source/Parallel.h
View File

@@ -186,6 +186,7 @@ namespace Parallel
MyList<Parallel::gridseg> **out_src, MyList<Parallel::gridseg> **out_dst);
void PeriodicBD(Patch *Pat, MyList<var> *VarList, int Symmetry);
double L2Norm(Patch *Pat, var *vf);
void L2Norm7(Patch *Pat, var **vf, double *norms);
void checkgsl(MyList<Parallel::gridseg> *pp, bool first_only);
void checkvarl(MyList<var> *pp, bool first_only);
MyList<Parallel::gridseg> *divide_gsl(MyList<Parallel::gridseg> *p, Patch *Pat);
@@ -221,6 +222,7 @@ namespace Parallel
void checkpatchlist(MyList<Patch> *PatL, bool buflog);
double L2Norm(Patch *Pat, var *vf, MPI_Comm Comm_here);
void L2Norm7(Patch *Pat, var **vf, double *norms, MPI_Comm Comm_here);
bool PatList_Interp_Points(MyList<Patch> *PatL, MyList<var> *VarList,
int NN, double **XX,
double *Shellf, int Symmetry, MPI_Comm Comm_here);

37
AMSS_NCKU_source/ShellPatch.C
View File

@@ -3472,6 +3472,43 @@ double ShellPatch::L2Norm(var *vf)
return tvf;
}
void ShellPatch::L2Norm7(var **vf, double *norms)
{
double tvf[7], dtvf[7];
int BDW = overghost;
for (int i = 0; i < 7; i++)
dtvf[i] = 0;
MyList<ss_patch> *sPp = PatL;
while (sPp)
{
MyList<Block> *Bp = sPp->data->blb;
while (Bp)
{
Block *cg = Bp->data;
if (myrank == cg->rank)
{
f_l2normhelper7(cg->shape, cg->X[0], cg->X[1], cg->X[2],
sPp->data->bbox[0], sPp->data->bbox[1], sPp->data->bbox[2],
sPp->data->bbox[3], sPp->data->bbox[4], sPp->data->bbox[5],
cg->fgfs[vf[0]->sgfn], cg->fgfs[vf[1]->sgfn], cg->fgfs[vf[2]->sgfn],
cg->fgfs[vf[3]->sgfn], cg->fgfs[vf[4]->sgfn], cg->fgfs[vf[5]->sgfn],
cg->fgfs[vf[6]->sgfn], tvf, BDW);
for (int i = 0; i < 7; i++)
dtvf[i] += tvf[i];
}
if (Bp == sPp->data->ble)
break;
Bp = Bp->next;
}
sPp = sPp->next;
}
MPI_Allreduce(dtvf, tvf, 7, MPI_DOUBLE, MPI_SUM, MPI_COMM_WORLD);
for (int i = 0; i < 7; i++)
norms[i] = sqrt(tvf[i]);
}
// find maximum of abstract value, XX store position for maximum, Shellf store maximum themselvs
void ShellPatch::Find_Maximum(MyList<var> *VarList, double *XX,

1
AMSS_NCKU_source/ShellPatch.h
View File

@@ -198,6 +198,7 @@ public:
void write_Pablo_file_ss(int *ext, double xmin, double xmax, double ymin, double ymax, double zmin, double zmax,
char *filename, int sst);
double L2Norm(var *vf);
void L2Norm7(var **vf, double *norms);
void Find_Maximum(MyList<var> *VarList, double *XX, double *Shellf);
};

25
AMSS_NCKU_source/bssn_class.C
View File

@@ -7930,35 +7930,18 @@ void bssn_class::Constraint_Out()
#if (PSTR == 1 || PSTR == 2)
double ConV_h[7];
#endif
var *ConstraintVars[7] = {Cons_Ham, Cons_Px, Cons_Py, Cons_Pz, Cons_Gx, Cons_Gy, Cons_Gz};
#ifdef WithShell
ConV[0] = SH->L2Norm(Cons_Ham);
ConV[1] = SH->L2Norm(Cons_Px);
ConV[2] = SH->L2Norm(Cons_Py);
ConV[3] = SH->L2Norm(Cons_Pz);
ConV[4] = SH->L2Norm(Cons_Gx);
ConV[5] = SH->L2Norm(Cons_Gy);
ConV[6] = SH->L2Norm(Cons_Gz);
SH->L2Norm7(ConstraintVars, ConV);
ConVMonitor->writefile(PhysTime, 7, ConV);
#endif
for (int levi = 0; levi < GH->levels; levi++)
{
#if (PSTR == 0)
ConV[0] = Parallel::L2Norm(GH->PatL[levi]->data, Cons_Ham);
ConV[1] = Parallel::L2Norm(GH->PatL[levi]->data, Cons_Px);
ConV[2] = Parallel::L2Norm(GH->PatL[levi]->data, Cons_Py);
ConV[3] = Parallel::L2Norm(GH->PatL[levi]->data, Cons_Pz);
ConV[4] = Parallel::L2Norm(GH->PatL[levi]->data, Cons_Gx);
ConV[5] = Parallel::L2Norm(GH->PatL[levi]->data, Cons_Gy);
ConV[6] = Parallel::L2Norm(GH->PatL[levi]->data, Cons_Gz);
Parallel::L2Norm7(GH->PatL[levi]->data, ConstraintVars, ConV);
#elif (PSTR == 1 || PSTR == 2)
ConV[0] = Parallel::L2Norm(GH->PatL[levi]->data, Cons_Ham, GH->Commlev[levi]);
ConV[1] = Parallel::L2Norm(GH->PatL[levi]->data, Cons_Px, GH->Commlev[levi]);
ConV[2] = Parallel::L2Norm(GH->PatL[levi]->data, Cons_Py, GH->Commlev[levi]);
ConV[3] = Parallel::L2Norm(GH->PatL[levi]->data, Cons_Pz, GH->Commlev[levi]);
ConV[4] = Parallel::L2Norm(GH->PatL[levi]->data, Cons_Gx, GH->Commlev[levi]);
ConV[5] = Parallel::L2Norm(GH->PatL[levi]->data, Cons_Gy, GH->Commlev[levi]);
ConV[6] = Parallel::L2Norm(GH->PatL[levi]->data, Cons_Gz, GH->Commlev[levi]);
Parallel::L2Norm7(GH->PatL[levi]->data, ConstraintVars, ConV, GH->Commlev[levi]);
// misc::tillherecheck("before collect data to cpu0");
// MPI_ALLREDUCE( sendbuf, recvbuf, count, datatype, op, comm), sendbu and recvbuf must be different
if (levi > 0)

75
AMSS_NCKU_source/fmisc.f90
View File

@@ -1514,6 +1514,81 @@ f_out = f_out*dX*dY*dZ
return
end subroutine l2normhelper
!--------------------------------------------------------------------------------------
subroutine l2normhelper7(ex, X, Y, Z,xmin,ymin,zmin,xmax,ymax,zmax,&
f1,f2,f3,f4,f5,f6,f7,f_out,gw)
implicit none
!~~~~~~> Input parameters:
integer,intent(in ):: ex(1:3)
real*8, intent(in ):: X(1:ex(1)),Y(1:ex(2)),Z(1:ex(3)),xmin,ymin,zmin,xmax,ymax,zmax
integer,intent(in)::gw
real*8, dimension(ex(1),ex(2),ex(3)),intent(in) :: f1,f2,f3,f4,f5,f6,f7
real*8, intent(out) :: f_out(7)
!~~~~~~> Other variables:
real*8 :: dX, dY, dZ
integer::imin,jmin,kmin
integer::imax,jmax,kmax
integer::i,j,k
real*8 :: s1,s2,s3,s4,s5,s6,s7
dX = X(2) - X(1)
dY = Y(2) - Y(1)
dZ = Z(2) - Z(1)
! for ghost zone
imin = gw+1
jmin = gw+1
kmin = gw+1
imax = ex(1) - gw
jmax = ex(2) - gw
kmax = ex(3) - gw
!for patch boundary (i.e., not ghost boundary)
if(dabs(X(ex(1))-xmax) < dX) imax = ex(1)
if(dabs(Y(ex(2))-ymax) < dY) jmax = ex(2)
if(dabs(Z(ex(3))-zmax) < dZ) kmax = ex(3)
if(dabs(X(1)-xmin) < dX) imin = 1
if(dabs(Y(1)-ymin) < dY) jmin = 1
if(dabs(Z(1)-zmin) < dZ) kmin = 1
s1 = 0.d0
s2 = 0.d0
s3 = 0.d0
s4 = 0.d0
s5 = 0.d0
s6 = 0.d0
s7 = 0.d0
do k=kmin,kmax
do j=jmin,jmax
!DIR$ SIMD REDUCTION(+:s1,s2,s3,s4,s5,s6,s7)
do i=imin,imax
s1 = s1 + f1(i,j,k)*f1(i,j,k)
s2 = s2 + f2(i,j,k)*f2(i,j,k)
s3 = s3 + f3(i,j,k)*f3(i,j,k)
s4 = s4 + f4(i,j,k)*f4(i,j,k)
s5 = s5 + f5(i,j,k)*f5(i,j,k)
s6 = s6 + f6(i,j,k)*f6(i,j,k)
s7 = s7 + f7(i,j,k)*f7(i,j,k)
enddo
enddo
enddo
f_out(1) = s1*dX*dY*dZ
f_out(2) = s2*dX*dY*dZ
f_out(3) = s3*dX*dY*dZ
f_out(4) = s4*dX*dY*dZ
f_out(5) = s5*dX*dY*dZ
f_out(6) = s6*dX*dY*dZ
f_out(7) = s7*dX*dY*dZ
return
end subroutine l2normhelper7
!--------------------------------------------------------------------------------------
! calculate L2norm especially for shell Blocks
subroutine l2normhelper_sh(ex, X, Y, Z,xmin,ymin,zmin,xmax,ymax,zmax,&

12
AMSS_NCKU_source/fmisc.h
View File

@@ -13,6 +13,7 @@
#define f_global_interpind2d global_interpind2d
#define f_global_interpind1d global_interpind1d
#define f_l2normhelper l2normhelper
#define f_l2normhelper7 l2normhelper7
#define f_l2normhelper_sh l2normhelper_sh
#define f_l2normhelper_sh_rms l2normhelper_sh_rms
#define f_average average
@@ -42,6 +43,7 @@
#define f_global_interpind2d GLOBAL_INTERPIND2D
#define f_global_interpind1d GLOBAL_INTERPIND1D
#define f_l2normhelper L2NORMHELPER
#define f_l2normhelper7 L2NORMHELPER7
#define f_l2normhelper_sh L2NORMHELPER_SH
#define f_l2normhelper_sh_rms L2NORMHELPER_SH_RMS
#define f_average AVERAGE
@@ -71,6 +73,7 @@
#define f_global_interpind2d global_interpind2d_
#define f_global_interpind1d global_interpind1d_
#define f_l2normhelper l2normhelper_
#define f_l2normhelper7 l2normhelper7_
#define f_l2normhelper_sh l2normhelper_sh_
#define f_l2normhelper_sh_rms l2normhelper_sh_rms_
#define f_average average_
@@ -164,6 +167,15 @@ extern "C"
double *, double &, int &);
}
extern "C"
{
void f_l2normhelper7(int *, double *, double *, double *,
double &, double &, double &,
double &, double &, double &,
double *, double *, double *, double *,
double *, double *, double *, double *, int &);
}
extern "C"
{
void f_l2normhelper_sh(int *, double *, double *, double *,