Stabilize EScalar CUDA fallback path

AMSS_NCKU_Input.py

@@ -31,7 +31,7 @@ GPU_Part         = 0.0
 ## Setting the physical system and numerical method
 
 Symmetry                 = "equatorial-symmetry"   ## Symmetry of System: choose equatorial-symmetry、no-symmetry、octant-symmetry
-Equation_Class           = "BSSN"                  ## Evolution Equation: choose "BSSN", "BSSN-EScalar", "BSSN-EM", "Z4C" 
+Equation_Class           = "BSSN-EScalar"                  ## Evolution Equation: choose "BSSN", "BSSN-EScalar", "BSSN-EM", "Z4C"
                                                    ## If "BSSN-EScalar" is chosen, it is necessary to set other parameters below
 Initial_Data_Method      = "Ansorg-TwoPuncture"    ## initial data method: choose "Ansorg-TwoPuncture", "Lousto-Analytical", "Cao-Analytical", "KerrSchild-Analytical"
 Time_Evolution_Method    = "runge-kutta-45"        ## time evolution method: choose "runge-kutta-45"
@@ -158,7 +158,7 @@ Detector_Rmax   = 160.0                  ## farest dector distance
 
 ## Setting the apprent horizon
 
-AHF_Find       = "yes"                    ## whether to find the apparent horizon: choose "yes" or "no"
+AHF_Find       = "no"                    ## whether to find the apparent horizon: choose "yes" or "no"
 
 AHF_Find_Every = 24
 AHF_Dump_Time  = 20.0

AMSS_NCKU_Program_Plot.py

@@ -1,100 +0,0 @@
-##################################################################
-##
-## AMSS-NCKU Plot-Only Restart Script
-## Author: Xiaoqu / Claude
-## 2026/05/12
-##
-## This script checks for existing output data from AMSS_NCKU_Program.py.
-## If data exists, it skips all computation and goes directly to plotting,
-## saving time when plotting was interrupted.
-## If no data is found, it exits with a message.
-##
-##################################################################
-
-## Guard against re-execution by multiprocessing child processes.
-if __name__ != '__main__':
-    import sys as _sys
-    _sys.exit(0)
-
-
-import os
-import sys
-
-import AMSS_NCKU_Input as input_data
-
-##################################################################
-
-## Construct paths from input configuration
-File_directory         = os.path.join(input_data.File_directory)
-output_directory       = os.path.join(File_directory, "AMSS_NCKU_output")
-binary_results_directory = os.path.join(output_directory, input_data.Output_directory)
-figure_directory       = os.path.join(File_directory, "figure")
-
-##################################################################
-
-## Check whether the required output data files exist
-
-required_files = [
-    os.path.join(binary_results_directory, "bssn_BH.dat"),
-    os.path.join(binary_results_directory, "bssn_ADMQs.dat"),
-    os.path.join(binary_results_directory, "bssn_psi4.dat"),
-    os.path.join(binary_results_directory, "bssn_constraint.dat"),
-]
-
-missing_files = [f for f in required_files if not os.path.exists(f)]
-
-if missing_files:
-    print(" No existing AMSS_NCKU_Program.py output data found. ")
-    print(" The following required files are missing: ")
-    for f in missing_files:
-        print(f"   {f}")
-    print()
-    print(" Please run AMSS_NCKU_Program.py first to generate the simulation data. ")
-    print(" Exiting. ")
-    sys.exit(1)
-
-print(" Found existing AMSS_NCKU_Program.py output data. "          )
-print(" Skipping all computation and going directly to plotting. " )
-print()
-
-## Ensure the figure directory exists (it should, but be safe)
-os.makedirs(figure_directory, exist_ok=True)
-
-##################################################################
-
-## Plot the AMSS-NCKU program results
-
-import plot_xiaoqu
-import plot_GW_strain_amplitude_xiaoqu
-from parallel_plot_helper import run_plot_tasks_parallel
-
-plot_tasks = []
-
-## Plot black hole trajectory
-plot_tasks.append((plot_xiaoqu.generate_puncture_orbit_plot,   (binary_results_directory, figure_directory)))
-plot_tasks.append((plot_xiaoqu.generate_puncture_orbit_plot3D, (binary_results_directory, figure_directory)))
-
-## Plot black hole separation vs. time
-plot_tasks.append((plot_xiaoqu.generate_puncture_distence_plot, (binary_results_directory, figure_directory)))
-
-## Plot gravitational waveforms (psi4 and strain amplitude)
-for i in range(input_data.Detector_Number):
-    plot_tasks.append((plot_xiaoqu.generate_gravitational_wave_psi4_plot, (binary_results_directory, figure_directory, i)))
-    plot_tasks.append((plot_GW_strain_amplitude_xiaoqu.generate_gravitational_wave_amplitude_plot, (binary_results_directory, figure_directory, i)))
-
-## Plot ADM mass evolution
-for i in range(input_data.Detector_Number):
-    plot_tasks.append((plot_xiaoqu.generate_ADMmass_plot, (binary_results_directory, figure_directory, i)))
-
-## Plot Hamiltonian constraint violation over time
-for i in range(input_data.grid_level):
-    plot_tasks.append((plot_xiaoqu.generate_constraint_check_plot, (binary_results_directory, figure_directory, i)))
-
-run_plot_tasks_parallel(plot_tasks)
-
-## Plot stored binary data (runs serially, not in the parallel pool)
-plot_xiaoqu.generate_binary_data_plot(binary_results_directory, figure_directory)
-
-print()
-print(" Plotting completed successfully. ")
-print()

AMSS_NCKU_source/ABE.C

@@ -198,16 +198,16 @@ int main(int argc, char *argv[])
       if (myrank == 0)
       {
             string out_dir;
-            string filename;
-            map<string, string>::iterator iter;
-            iter = parameters::str_par.find("output dir");
-            if (iter != parameters::str_par.end())
-            {
-                  out_dir = iter->second;
-            }
-            filename = out_dir + "/setting.par";
-            ofstream setfile;
-            setfile.open(filename.c_str(), ios::trunc);
+            char filename[50];
+            map<string, string>::iterator iter;
+            iter = parameters::str_par.find("output dir");
+            if (iter != parameters::str_par.end())
+            {
+                  out_dir = iter->second;
+            }
+            sprintf(filename, "%s/setting.par", out_dir.c_str());
+            ofstream setfile;
+            setfile.open(filename, ios::trunc);
 
             if (!setfile.good())
             {
@@ -484,11 +484,7 @@ int main(int argc, char *argv[])
             cout << endl;
       }
 
-      // Let the process teardown reclaim the simulation object.  Some derived
-      // equation classes keep MPI/CUDA-backed state whose destructor ordering
-      // is fragile at program shutdown.
-      if (getenv("AMSS_DELETE_ADM_ON_EXIT"))
-            delete ADM;
+      delete ADM;
 
       //=======================caculation done=============================================================
 

AMSS_NCKU_source/ShellPatch.h

@@ -102,16 +102,6 @@ public:
                  //-1: means no dumy dimension at all; 0: means rho; 1: means sigma
    };
 
-   // Thread-safe search result (no pointers to shared mutable state)
-   struct PointSearchResult
-   {
-      bool found;
-      Block *Bg;
-      double gx, gy, gz; // global Cartesian coordinates
-      double lx, ly, lz; // local coordinates within the found block
-      int ssst;          // source shell-patch type (-1 = Cartesian)
-   };
-
    int myrank;
    int shape[dim];   // for (rho, sigma, R), for rho and sigma means number of points for every pi/2
    double Rrange[2]; // for Rmin and Rmax
@@ -185,12 +175,6 @@ public:
                          MyList<Patch> *Pp, double CDH[dim], MyList<pointstru> *pss);
    bool prolongpointstru(MyList<pointstru> *&psul, bool ssyn, int tsst, MyList<ss_patch> *sPp, double DH[dim],
                          MyList<Patch> *Pp, double CDH[dim], double x, double y, double z, int Symmetry, int rank_in);
-   // Read-only point search — thread-safe (no shared mutable state modified)
-   PointSearchResult prolongpointstru_search(bool ssyn, int tsst, MyList<ss_patch> *sPp, double DH[dim],
-                                             MyList<Patch> *Pp, double CDH[dim], double x, double y, double z,
-                                             int Symmetry, int rank_in);
-   // Append a search result to a linked list — use inside omp critical section
-   void prolongpointstru_append(MyList<pointstru> *&psul, const PointSearchResult &sr, int tsst);
    void setupintintstuff(int cpusize, MyList<Patch> *CPatL, int Symmetry);
    void intertransfer(MyList<pointstru> **src, MyList<pointstru> **dst,
                       MyList<var> *VarList1 /* source */, MyList<var> *VarList2 /*target */,
@@ -211,11 +195,11 @@ public:
    bool Interp_One_Point(MyList<var> *VarList,
                          double *XX, /*input global Cartesian coordinate*/
                          double *Shellf, int Symmetry);
-   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);
-};
+   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);
+};
 
 #endif /* SHELLPATCH_H */

AMSS_NCKU_source/bssnEScalar_class.h

@@ -54,17 +54,21 @@ public:
      void Interp_Constraint();
      void Constraint_Out(); 
 
-protected:
-     var *Sphio, *Spio;
-     var *Sphi0, *Spi0;
+protected:
+     var *Sphio, *Spio;
+     var *Sphi0, *Spi0;
      var *Sphi,  *Spi;
      var *Sphi1, *Spi1;
      var *Sphi_rhs, *Spi_rhs;
-
-     var *Cons_fR;
-
-     monitor *MaxScalar_Monitor;
-};
+
+     var *Cons_fR;
+
+     MyList<var> *BSSNStateList, *BSSNSynchList_pre, *BSSNSynchList_cor;
+     MyList<var> *ScalarSynchList_pre, *ScalarSynchList_cor;
+     Parallel::SyncCache *sync_cache_scalar_pre, *sync_cache_scalar_cor;
+
+     monitor *MaxScalar_Monitor;
+};
 
 #endif /* BSSNESCALAR_CLASS_H */
 

AMSS_NCKU_source/bssnEScalar_rhs.f90

@@ -3,11 +3,143 @@
 !! note that the potential for scalar field in F(R) gravity
 !!       is defined in the file Set_Rho_ADM.f90
 
-#include "macrodef.fh"
-
-! rhs for scalar and GR variables
-! here we consider vacuum spacetime only
-  function compute_rhs_bssn_escalar(ex, T,X, Y, Z,                             &
+#include "macrodef.fh"
+
+! scalar RHS and stress-energy only; BSSN RHS can be supplied by CUDA.
+  function compute_rhs_bssn_escalar_matter(ex, T, X, Y, Z,                    &
+               chi    ,   trK    ,                                            &
+               dxx    ,   gxy    ,   gxz    ,   dyy    ,   gyz    ,   dzz,    &
+               Axx    ,   Axy    ,   Axz    ,   Ayy    ,   Ayz    ,   Azz,    &
+               Gamx   ,  Gamy    ,  Gamz    ,                                 &
+               Lap    ,  betax   ,  betay   ,  betaz   ,                      &
+               dtSfx  ,  dtSfy   ,  dtSfz   ,                                 &
+               Sphi   ,   Spi    ,                                            &
+               Sphi_rhs ,   Spi_rhs ,                                         &
+               rho,Sx,Sy,Sz,Sxx,Sxy,Sxz,Syy,Syz,Szz,                          &
+               Symmetry,Lev,eps) result(gont)
+  implicit none
+
+  integer,intent(in ):: ex(1:3), Symmetry,Lev
+  real*8, intent(in ):: T
+  real*8, intent(in ):: X(1:ex(1)),Y(1:ex(2)),Z(1:ex(3))
+  real*8, dimension(ex(1),ex(2),ex(3)),intent(inout) :: chi,dxx,dyy,dzz
+  real*8, dimension(ex(1),ex(2),ex(3)),intent(in ) :: trK
+  real*8, dimension(ex(1),ex(2),ex(3)),intent(in ) :: gxy,gxz,gyz
+  real*8, dimension(ex(1),ex(2),ex(3)),intent(in ) :: Axx,Axy,Axz,Ayy,Ayz,Azz
+  real*8, dimension(ex(1),ex(2),ex(3)),intent(in ) :: Gamx,Gamy,Gamz
+  real*8, dimension(ex(1),ex(2),ex(3)),intent(inout) :: Lap, betax, betay, betaz
+  real*8, dimension(ex(1),ex(2),ex(3)),intent(in ) :: dtSfx,  dtSfy,  dtSfz
+  real*8, dimension(ex(1),ex(2),ex(3)),intent(in ) :: Sphi,Spi
+  real*8, dimension(ex(1),ex(2),ex(3)),intent(out) :: Sphi_rhs,Spi_rhs
+  real*8, dimension(ex(1),ex(2),ex(3)),intent(inout) :: rho,Sx,Sy,Sz
+  real*8, dimension(ex(1),ex(2),ex(3)),intent(inout) :: Sxx,Sxy,Sxz,Syy,Syz,Szz
+  real*8,intent(in) :: eps
+  integer::gont
+
+  real*8, dimension(ex(1),ex(2),ex(3)) :: gxx,gyy,gzz
+  real*8, dimension(ex(1),ex(2),ex(3)) :: chix,chiy,chiz
+  real*8, dimension(ex(1),ex(2),ex(3)) :: Lapx,Lapy,Lapz
+  real*8, dimension(ex(1),ex(2),ex(3)) :: Kx,Ky,Kz,S
+  real*8, dimension(ex(1),ex(2),ex(3)) :: f,fxx,fxy,fxz,fyy,fyz,fzz
+  real*8, dimension(ex(1),ex(2),ex(3)) :: alpn1,chin1
+  real*8, dimension(ex(1),ex(2),ex(3)) :: gupxx,gupxy,gupxz
+  real*8, dimension(ex(1),ex(2),ex(3)) :: gupyy,gupyz,gupzz
+
+  real*8            :: dX
+  real*8, parameter :: ZEO=0.d0, ONE = 1.D0, TWO = 2.D0, HALF = 0.5D0
+  real*8, parameter :: SYM = 1.D0
+
+  dX = sum(chi)+sum(trK)+sum(dxx)+sum(gxy)+sum(gxz)+sum(dyy)+sum(gyz)+sum(dzz) &
+      +sum(Gamx)+sum(Gamy)+sum(Gamz)                                           &
+      +sum(Lap)+sum(Sphi)+sum(Spi)
+  if(dX.ne.dX) then
+     if(sum(chi).ne.sum(chi))write(*,*)"bssn_escalar_matter: find NaN in chi"
+     if(sum(trK).ne.sum(trK))write(*,*)"bssn_escalar_matter: find NaN in trk"
+     if(sum(dxx).ne.sum(dxx))write(*,*)"bssn_escalar_matter: find NaN in dxx"
+     if(sum(gxy).ne.sum(gxy))write(*,*)"bssn_escalar_matter: find NaN in gxy"
+     if(sum(gxz).ne.sum(gxz))write(*,*)"bssn_escalar_matter: find NaN in gxz"
+     if(sum(dyy).ne.sum(dyy))write(*,*)"bssn_escalar_matter: find NaN in dyy"
+     if(sum(gyz).ne.sum(gyz))write(*,*)"bssn_escalar_matter: find NaN in gyz"
+     if(sum(dzz).ne.sum(dzz))write(*,*)"bssn_escalar_matter: find NaN in dzz"
+     if(sum(Gamx).ne.sum(Gamx))write(*,*)"bssn_escalar_matter: find NaN in Gamx"
+     if(sum(Gamy).ne.sum(Gamy))write(*,*)"bssn_escalar_matter: find NaN in Gamy"
+     if(sum(Gamz).ne.sum(Gamz))write(*,*)"bssn_escalar_matter: find NaN in Gamz"
+     if(sum(Lap).ne.sum(Lap))write(*,*)"bssn_escalar_matter: find NaN in Lap"
+     if(sum(Sphi).ne.sum(Sphi))write(*,*)"bssn_escalar_matter: find NaN in Sphi"
+     if(sum(Spi).ne.sum(Spi))write(*,*)"bssn_escalar_matter: find NaN in Spi"
+     gont = 1
+     return
+  endif
+
+  alpn1 = Lap + ONE
+  chin1 = chi + ONE
+  gxx = dxx + ONE
+  gyy = dyy + ONE
+  gzz = dzz + ONE
+
+  call fderivs(ex,chi,chix,chiy,chiz,X,Y,Z,SYM,SYM,SYM,Symmetry,Lev)
+  call fderivs(ex,Lap,Lapx,Lapy,Lapz,X,Y,Z,SYM,SYM,SYM,Symmetry,Lev)
+
+  gupzz =  gxx * gyy * gzz + gxy * gyz * gxz + gxz * gxy * gyz - &
+           gxz * gyy * gxz - gxy * gxy * gzz - gxx * gyz * gyz
+  gupxx =   ( gyy * gzz - gyz * gyz ) / gupzz
+  gupxy = - ( gxy * gzz - gyz * gxz ) / gupzz
+  gupxz =   ( gxy * gyz - gyy * gxz ) / gupzz
+  gupyy =   ( gxx * gzz - gxz * gxz ) / gupzz
+  gupyz = - ( gxx * gyz - gxy * gxz ) / gupzz
+  gupzz =   ( gxx * gyy - gxy * gxy ) / gupzz
+
+#if 1
+  Sphi_rhs = alpn1 * Spi
+  call fderivs(ex,Sphi,Kx,Ky,Kz,X,Y,Z,SYM,SYM,SYM,Symmetry,Lev)
+  call fdderivs(ex,Sphi,fxx,fxy,fxz,fyy,fyz,fzz,X,Y,Z,SYM,SYM,SYM,Symmetry,Lev)
+  Spi_rhs =   gupxx * fxx + gupyy * fyy + gupzz * fzz +           &
+             ( gupxy * fxy + gupxz * fxz + gupyz * fyz ) * TWO -  &
+             ((Gamx+(gupxx*chix+gupxy*chiy+gupxz*chiz)/TWO/chin1)*Kx  &
+            + (Gamy+(gupxy*chix+gupyy*chiy+gupyz*chiz)/TWO/chin1)*Ky  &
+            + (Gamz+(gupxz*chix+gupyz*chiy+gupzz*chiz)/TWO/chin1)*Kz)
+  Spi_rhs = Spi_rhs*alpn1 +                                      &
+            (gupxx*Lapx*Kx + gupxy*Lapx*Ky + gupxz*Lapx*Kz       &
+            +gupxy*Lapy*Kx + gupyy*Lapy*Ky + gupyz*Lapy*Kz       &
+            +gupxz*Lapz*Kx + gupyz*Lapz*Ky + gupzz*Lapz*Kz)
+
+  call frpotential(ex,Sphi,f,S)
+  Spi_rhs = Spi_rhs*chin1 + alpn1*(trK*Spi - S)
+  rho = chin1*((gupxx * Kx * Kx + gupyy * Ky * Ky + gupzz * Kz * Kz)/TWO + &
+                gupxy * Kx * Ky + gupxz * Kx * Kz + gupyz * Ky * Kz )      &
+        + Spi*Spi/TWO+f
+  Sx = -Spi*Kx
+  Sy = -Spi*Ky
+  Sz = -Spi*Kz
+  f = (rho - Spi*Spi)/chin1
+  Sxx = Kx*Kx-f*gxx
+  Sxy = Kx*Ky-f*gxy
+  Sxz = Kx*Kz-f*gxz
+  Syy = Ky*Ky-f*gyy
+  Syz = Ky*Kz-f*gyz
+  Szz = Kz*Kz-f*gzz
+#else
+  Sphi_rhs = ZEO
+  Spi_rhs = ZEO
+  rho = ZEO
+  Sx = ZEO
+  Sy = ZEO
+  Sz = ZEO
+  Sxx = ZEO
+  Sxy = ZEO
+  Sxz = ZEO
+  Syy = ZEO
+  Syz = ZEO
+  Szz = ZEO
+#endif
+
+  gont = 0
+  return
+end function compute_rhs_bssn_escalar_matter
+
+! rhs for scalar and GR variables
+! here we consider vacuum spacetime only
+  function compute_rhs_bssn_escalar(ex, T,X, Y, Z,                             &
                chi    ,   trK    ,                                             &
                dxx    ,   gxy    ,   gxz    ,   dyy    ,   gyz    ,   dzz,     &
                Axx    ,   Axy    ,   Axz    ,   Ayy    ,   Ayz    ,   Azz,     &

AMSS_NCKU_source/bssn_class.h

@@ -144,7 +144,7 @@ public:
        bssn_class(double Couranti, double StartTimei, double TotalTimei, double DumpTimei, double d2DumpTimei, double CheckTimei, double AnasTimei,
                   int Symmetryi, int checkruni, char *checkfilenamei, double numepssi, double numepsbi, double numepshi,
                   int a_levi, int maxli, int decni, double maxrexi, double drexi);
-       virtual ~bssn_class();
+       ~bssn_class();
 
        void Evolve(int Steps);
        void RecursiveStep(int lev);
@@ -178,14 +178,17 @@ public:
        virtual void Initialize();
        virtual void Read_Ansorg();
        virtual void Read_Pablo() {};
-       virtual void Compute_Psi4(int lev);
-       virtual void Step(int lev, int YN);
-       virtual void Interp_Constraint(bool infg);
-       virtual void Constraint_Out();
-       virtual void Compute_Constraint();
-
-#ifdef With_AHF
-protected:
+       virtual void Compute_Psi4(int lev);
+       virtual void Step(int lev, int YN);
+       virtual void Interp_Constraint(bool infg);
+       virtual void Constraint_Out();
+       virtual void Compute_Constraint();
+
+protected:
+       void Initialize_Level_Runtime();
+
+#ifdef With_AHF
+protected:
        MyList<var> *AHList, *AHDList, *GaugeList;
        int AHfindevery;
        double AHdumptime;

AMSS_NCKU_source/bssn_gpu.h

@@ -1,56 +0,0 @@
-
-#ifndef BSSN_GPU_H_
-#define BSSN_GPU_H_
-#include "bssn_macro.h"
-#include "macrodef.h"
-
-#define DEVICE_ID 0
-// #define DEVICE_ID_BY_MPI_RANK
-#define GRID_DIM 256
-#define BLOCK_DIM 128
-
-#define _FH2_(i, j, k) fh[(i) + (j) * _1D_SIZE[2] + (k) * _2D_SIZE[2]]
-#define _FH3_(i, j, k) fh[(i) + (j) * _1D_SIZE[3] + (k) * _2D_SIZE[3]]
-#define pow2(x) ((x) * (x))
-#define TimeBetween(a, b) ((b.tv_sec - a.tv_sec) + (b.tv_usec - a.tv_usec) / 1000000.0f)
-#define M_ metac.
-#define Mh_ meta->
-#define Ms_ metassc.
-#define Msh_ metass->
-
-// #define TIMING
-
-#define RHS_SS_PARA int calledby, int mpi_rank, int *ex, double &T, double *crho, double *sigma, double *R, double *X, double *Y, double *Z, double *drhodx, double *drhody, double *drhodz, double *dsigmadx, double *dsigmady, double *dsigmadz, double *dRdx, double *dRdy, double *dRdz, double *drhodxx, double *drhodxy, double *drhodxz, double *drhodyy, double *drhodyz, double *drhodzz, double *dsigmadxx, double *dsigmadxy, double *dsigmadxz, double *dsigmadyy, double *dsigmadyz, double *dsigmadzz, double *dRdxx, double *dRdxy, double *dRdxz, double *dRdyy, double *dRdyz, double *dRdzz, double *chi, double *trK, double *dxx, double *gxy, double *gxz, double *dyy, double *gyz, double *dzz, double *Axx, double *Axy, double *Axz, double *Ayy, double *Ayz, double *Azz, double *Gamx, double *Gamy, double *Gamz, double *Lap, double *betax, double *betay, double *betaz, double *dtSfx, double *dtSfy, double *dtSfz, double *chi_rhs, double *trK_rhs, double *gxx_rhs, double *gxy_rhs, double *gxz_rhs, double *gyy_rhs, double *gyz_rhs, double *gzz_rhs, double *Axx_rhs, double *Axy_rhs, double *Axz_rhs, double *Ayy_rhs, double *Ayz_rhs, double *Azz_rhs, double *Gamx_rhs, double *Gamy_rhs, double *Gamz_rhs, double *Lap_rhs, double *betax_rhs, double *betay_rhs, double *betaz_rhs, double *dtSfx_rhs, double *dtSfy_rhs, double *dtSfz_rhs, double *rho, double *Sx, double *Sy, double *Sz, double *Sxx, double *Sxy, double *Sxz, double *Syy, double *Syz, double *Szz, double *Gamxxx, double *Gamxxy, double *Gamxxz, double *Gamxyy, double *Gamxyz, double *Gamxzz, double *Gamyxx, double *Gamyxy, double *Gamyxz, double *Gamyyy, double *Gamyyz, double *Gamyzz, double *Gamzxx, double *Gamzxy, double *Gamzxz, double *Gamzyy, double *Gamzyz, double *Gamzzz, double *Rxx, double *Rxy, double *Rxz, double *Ryy, double *Ryz, double *Rzz, double *ham_Res, double *movx_Res, double *movy_Res, double *movz_Res, double *Gmx_Res, double *Gmy_Res, double *Gmz_Res, int &Symmetry, int &Lev, double &eps, int &sst, int &co
-
-/**  main function */
-int gpu_rhs(int calledby, int mpi_rank, int *ex, double &T,
-            double *X, double *Y, double *Z,
-            double *chi, double *trK,
-            double *dxx, double *gxy, double *gxz, double *dyy, double *gyz, double *dzz,
-            double *Axx, double *Axy, double *Axz, double *Ayy, double *Ayz, double *Azz,
-            double *Gamx, double *Gamy, double *Gamz,
-            double *Lap, double *betax, double *betay, double *betaz,
-            double *dtSfx, double *dtSfy, double *dtSfz,
-            double *chi_rhs, double *trK_rhs,
-            double *gxx_rhs, double *gxy_rhs, double *gxz_rhs, double *gyy_rhs, double *gyz_rhs, double *gzz_rhs,
-            double *Axx_rhs, double *Axy_rhs, double *Axz_rhs, double *Ayy_rhs, double *Ayz_rhs, double *Azz_rhs,
-            double *Gamx_rhs, double *Gamy_rhs, double *Gamz_rhs,
-            double *Lap_rhs, double *betax_rhs, double *betay_rhs, double *betaz_rhs,
-            double *dtSfx_rhs, double *dtSfy_rhs, double *dtSfz_rhs,
-            double *rho, double *Sx, double *Sy, double *Sz, double *Sxx,
-            double *Sxy, double *Sxz, double *Syy, double *Syz, double *Szz,
-            double *Gamxxx, double *Gamxxy, double *Gamxxz, double *Gamxyy, double *Gamxyz, double *Gamxzz,
-            double *Gamyxx, double *Gamyxy, double *Gamyxz, double *Gamyyy, double *Gamyyz, double *Gamyzz,
-            double *Gamzxx, double *Gamzxy, double *Gamzxz, double *Gamzyy, double *Gamzyz, double *Gamzzz,
-            double *Rxx, double *Rxy, double *Rxz, double *Ryy, double *Ryz, double *Rzz,
-            double *ham_Res, double *movx_Res, double *movy_Res, double *movz_Res,
-            double *Gmx_Res, double *Gmy_Res, double *Gmz_Res,
-            int &Symmetry, int &Lev, double &eps, int &co);
-
-int gpu_rhs_ss(RHS_SS_PARA);
-
-#define Z4C_SS_PARA int calledby, int mpi_rank, int *ex, double &T, double *crho, double *sigma, double *R, double *X, double *Y, double *Z, double *drhodx, double *drhody, double *drhodz, double *dsigmadx, double *dsigmady, double *dsigmadz, double *dRdx, double *dRdy, double *dRdz, double *drhodxx, double *drhodxy, double *drhodxz, double *drhodyy, double *drhodyz, double *drhodzz, double *dsigmadxx, double *dsigmadxy, double *dsigmadxz, double *dsigmadyy, double *dsigmadyz, double *dsigmadzz, double *dRdxx, double *dRdxy, double *dRdxz, double *dRdyy, double *dRdyz, double *dRdzz, double *chi, double *trK, double *dxx, double *gxy, double *gxz, double *dyy, double *gyz, double *dzz, double *Axx, double *Axy, double *Axz, double *Ayy, double *Ayz, double *Azz, double *Gamx, double *Gamy, double *Gamz, double *Lap, double *betax, double *betay, double *betaz, double *dtSfx, double *dtSfy, double *dtSfz, double *TZ, double *chi_rhs, double *trK_rhs, double *gxx_rhs, double *gxy_rhs, double *gxz_rhs, double *gyy_rhs, double *gyz_rhs, double *gzz_rhs, double *Axx_rhs, double *Axy_rhs, double *Axz_rhs, double *Ayy_rhs, double *Ayz_rhs, double *Azz_rhs, double *Gamx_rhs, double *Gamy_rhs, double *Gamz_rhs, double *Lap_rhs, double *betax_rhs, double *betay_rhs, double *betaz_rhs, double *dtSfx_rhs, double *dtSfy_rhs, double *dtSfz_rhs, double *TZ_rhs, double *rho, double *Sx, double *Sy, double *Sz, double *Sxx, double *Sxy, double *Sxz, double *Syy, double *Syz, double *Szz, double *Gamxxx, double *Gamxxy, double *Gamxxz, double *Gamxyy, double *Gamxyz, double *Gamxzz, double *Gamyxx, double *Gamyxy, double *Gamyxz, double *Gamyyy, double *Gamyyz, double *Gamyzz, double *Gamzxx, double *Gamzxy, double *Gamzxz, double *Gamzyy, double *Gamzyz, double *Gamzzz, double *Rxx, double *Rxy, double *Rxz, double *Ryy, double *Ryz, double *Rzz, double *ham_Res, double *movx_Res, double *movy_Res, double *movz_Res, double *Gmx_Res, double *Gmy_Res, double *Gmz_Res, int &Symmetry, int &Lev, double &eps, int &sst, int &co
-
-int gpu_rhs_z4c_ss(Z4C_SS_PARA);
-
-#endif

AMSS_NCKU_source/bssn_gpu_rhs_ss.cu

@@ -20,14 +20,12 @@ using namespace std;
 
 __device__ volatile unsigned int global_count = 0;
 
-#ifdef RESULT_CHECK
 void compare_result_gpu(int ftag1,double * datac,int data_num){
 	double * data = (double*)malloc(sizeof(double)*data_num);
 	cudaMemcpy(data, datac, data_num * sizeof(double), cudaMemcpyDeviceToHost);
 	compare_result(ftag1,data,data_num);
 	free(data);
 }
-#endif
 
 __global__ void sub_symmetry_bd_ss_partF(int ord, double * func, double *funcc)
 {
@@ -155,11 +153,11 @@ __global__ void sub_symmetry_bd_ss_partJ(int ord,double * func, double * funcc,d
 
 inline void sub_symmetry_bd_ss(int ord,double * func, double * funcc,double * SoA){
 	sub_symmetry_bd_ss_partF<<<GRID_DIM,BLOCK_DIM>>>(ord,func,funcc);
-	cudaDeviceSynchronize();
+	cudaThreadSynchronize();
 	sub_symmetry_bd_ss_partI<<<GRID_DIM,BLOCK_DIM>>>(ord,func,funcc,SoA[0]);
-	cudaDeviceSynchronize();
+	cudaThreadSynchronize();
 	sub_symmetry_bd_ss_partJ<<<GRID_DIM,BLOCK_DIM>>>(ord,func,funcc,SoA[1]);
-	cudaDeviceSynchronize();
+	cudaThreadSynchronize();
 }
 
 __global__ void sub_fderivs_shc_part1(double *fx,double *fy,double *fz){
@@ -249,13 +247,13 @@ inline void sub_fderivs_shc(int& sst,double * f,double * fh,double *fx,double *f
 	//cudaMemset(Msh_ gy,0,h_3D_SIZE[0] * sizeof(double));
 	//cudaMemset(Msh_ gz,0,h_3D_SIZE[0] * sizeof(double));
 	sub_symmetry_bd_ss(2,f,fh,SoA1);
-	cudaDeviceSynchronize();
+	cudaThreadSynchronize();
 			//compare_result_gpu(0,fh,h_3D_SIZE[2]);	
 	sub_fderivs_sh<<<GRID_DIM,BLOCK_DIM>>>(fh,Msh_ gx,Msh_ gy,Msh_ gz);
-	cudaDeviceSynchronize();
+	cudaThreadSynchronize();
 	
 	sub_fderivs_shc_part1<<<GRID_DIM,BLOCK_DIM>>>(fx,fy,fz);
-	cudaDeviceSynchronize();
+	cudaThreadSynchronize();
 			//compare_result_gpu(1,fx,h_3D_SIZE[0]);
 			//compare_result_gpu(2,fy,h_3D_SIZE[0]);
 			//compare_result_gpu(3,fz,h_3D_SIZE[0]);
@@ -453,17 +451,17 @@ inline void sub_fdderivs_shc(int& sst,double * f,double * fh,
 	
 	//fderivs_sh
 	sub_symmetry_bd_ss(2,f,fh,SoA1);
-	cudaDeviceSynchronize();
+	cudaThreadSynchronize();
 			//compare_result_gpu(1,fh,h_3D_SIZE[2]);	
 	sub_fderivs_sh<<<GRID_DIM,BLOCK_DIM>>>(fh,Msh_ gx,Msh_ gy,Msh_ gz);
-	cudaDeviceSynchronize();
+	cudaThreadSynchronize();
 	
 	//fdderivs_sh
 	sub_symmetry_bd_ss(2,f,fh,SoA1);
-	cudaDeviceSynchronize();
+	cudaThreadSynchronize();
 			//compare_result_gpu(21,fh,h_3D_SIZE[2]);
 	sub_fdderivs_sh<<<GRID_DIM,BLOCK_DIM>>>(fh,Msh_ gxx,Msh_ gxy,Msh_ gxz,Msh_ gyy,Msh_ gyz,Msh_ gzz);
-	cudaDeviceSynchronize();
+	cudaThreadSynchronize();
 			/*compare_result_gpu(11,Msh_ gx,h_3D_SIZE[0]);
 			compare_result_gpu(12,Msh_ gy,h_3D_SIZE[0]);
 			compare_result_gpu(13,Msh_ gz,h_3D_SIZE[0]);
@@ -474,7 +472,7 @@ inline void sub_fdderivs_shc(int& sst,double * f,double * fh,
 			compare_result_gpu(5,Msh_ gyz,h_3D_SIZE[0]);
 			compare_result_gpu(6,Msh_ gzz,h_3D_SIZE[0]);*/
 	sub_fdderivs_shc_part1<<<GRID_DIM,BLOCK_DIM>>>(fxx,fxy,fxz,fyy,fyz,fzz);
-	cudaDeviceSynchronize();
+	cudaThreadSynchronize();
 			/*compare_result_gpu(1,fxx,h_3D_SIZE[0]);
 			compare_result_gpu(2,fxy,h_3D_SIZE[0]);
 			compare_result_gpu(3,fxz,h_3D_SIZE[0]);
@@ -498,9 +496,9 @@ __global__ void computeRicci_ss_part1(double * dst)
  inline void computeRicci_ss(int &sst,double * src,double* dst,double * SoA, Meta* meta)
 {
 	sub_fdderivs_shc(sst,src,Mh_ fh,Mh_ fxx,Mh_ fxy,Mh_ fxz,Mh_ fyy,Mh_ fyz,Mh_ fzz,SoA);
-	cudaDeviceSynchronize();
+	cudaThreadSynchronize();
 	computeRicci_ss_part1<<<GRID_DIM,BLOCK_DIM>>>(dst);
-	cudaDeviceSynchronize();
+	cudaThreadSynchronize();
 	
 }
 __global__ void sub_lopsided_ss_part1(double * dst)
@@ -518,9 +516,9 @@ __global__ void sub_lopsided_ss_part1(double * dst)
 inline void sub_lopsided_ss(int& sst,double *src,double* dst,double *SoA)
 {
 		sub_fderivs_shc(sst,src,Mh_ fh,Mh_ fxx,Mh_ fxy,Mh_ fxz,SoA);
-		cudaDeviceSynchronize();
+		cudaThreadSynchronize();
 		sub_lopsided_ss_part1<<<GRID_DIM,BLOCK_DIM>>>(dst);
-		cudaDeviceSynchronize();
+		cudaThreadSynchronize();
 }
 
 __global__ void sub_kodis_sh_part1(double *f,double *fh,double *f_rhs)
@@ -592,11 +590,11 @@ inline void sub_kodis_ss(int &sst,double *f,double *fh,double *f_rhs,double *SoA
 	}
 			//compare_result_gpu(10,f,h_3D_SIZE[0]);
 	sub_symmetry_bd_ss(3,f,fh,SoA1);
-	cudaDeviceSynchronize();
+	cudaThreadSynchronize();
 			//compare_result_gpu(0,fh,h_3D_SIZE[3]);
 			
 	sub_kodis_sh_part1<<<GRID_DIM,BLOCK_DIM>>>(f,fh,f_rhs);
-	cudaDeviceSynchronize();
+	cudaThreadSynchronize();
 			//compare_result_gpu(1,f_rhs,h_3D_SIZE[0]);
 }
 
@@ -1701,7 +1699,7 @@ void destroy_meta(Meta *meta,Metass *metass)
 	if(Msh_ gzz) cudaFree(Msh_ gzz);
 	
 #if (GAUGE == 2 || GAUGE == 3 || GAUGE == 4 || GAUGE == 5 || GAUGE == 6 || GAUGE == 7)
-	if(Mh_ reta) cudaFree(Mh_ reta);
+	if(Mh_ reta) CUDA_SAFE_CALL(cudaFree(Mh_ reta));
 
 #endif	
 	
@@ -1897,7 +1895,7 @@ int gpu_rhs_ss(RHS_SS_PARA)
 
 	//1.2 local Data
 	cudaMalloc((void**)&(Mh_ gxx), matrix_size * sizeof(double));
-	cudaMalloc((void**)&(Mh_ gyy), matrix_size * sizeof(double));
+	CUDA_SAFE_CALL( cudaMalloc((void**)&(Mh_ gyy), matrix_size * sizeof(double)));
 	cudaMalloc((void**)&(Mh_ gzz), matrix_size * sizeof(double));
 	cudaMalloc((void**)&(Mh_ chix), matrix_size * sizeof(double));
 	cudaMalloc((void**)&(Mh_ chiy), matrix_size * sizeof(double));
@@ -2162,7 +2160,7 @@ int gpu_rhs_ss(RHS_SS_PARA)
 	
 	double tmp_con2 = 1/Mass[0] - tmp_con;
 	cudaMemcpyToSymbol(C1, &tmp_con2, sizeof(double));
-	tmp_con2 = 1/Mass[1] - tmp_con;
+	double tmp_con2 = 1/Mass[1] - tmp_con;
 	cudaMemcpyToSymbol(C2, &tmp_con2, sizeof(double));
 		
 
@@ -2235,7 +2233,7 @@ int gpu_rhs_ss(RHS_SS_PARA)
   		if((sst == 2 || sst == 4) && abs[1] < dYh) 
   		{
   			ijkmin_h[1] = -2; 
-  			ijkmin3_h[1] = -3;
+  			ijkmin_h[1] = -3;
   		}
   		if((sst == 3 || sst == 5) && abs_Y_ex2 < dYh)
   		{
@@ -2289,13 +2287,13 @@ int gpu_rhs_ss(RHS_SS_PARA)
 
 
 #ifdef TIMING1
-	cudaDeviceSynchronize();
+	cudaThreadSynchronize();
 	gettimeofday(&tv2, NULL);
    	cout<<"TIME USED"<<TimeBetween(tv1, tv2)<<endl; 
 #endif	
 	//cout<<"GPU meta data ready.\n";
 	
-	cudaDeviceSynchronize();
+	cudaThreadSynchronize();
 
 
 //-------------get device info-------------------------------------
@@ -2308,7 +2306,7 @@ int gpu_rhs_ss(RHS_SS_PARA)
 	//sub_enforce_ga(matrix_size);
 	//4.1-----compute rhs---------
 	compute_rhs_ss_part1<<<GRID_DIM,BLOCK_DIM>>>();
-	cudaDeviceSynchronize();
+	cudaThreadSynchronize();
 
 	sub_fderivs_shc(sst,Mh_ betax,Mh_ fh,Mh_ betaxx,Mh_ betaxy,Mh_ betaxz,ass);
 	sub_fderivs_shc(sst,Mh_ betay,Mh_ fh,Mh_ betayx,Mh_ betayy,Mh_ betayz,sas);
@@ -2324,7 +2322,7 @@ int gpu_rhs_ss(RHS_SS_PARA)
 	sub_fderivs_shc(sst,Mh_ gyz,Mh_ fh,Mh_ gyzx,Mh_ gyzy,Mh_ gyzz, saa);
 	
 	compute_rhs_ss_part2<<<GRID_DIM,BLOCK_DIM>>>();
-	cudaDeviceSynchronize();
+	cudaThreadSynchronize();
 	
 	sub_fdderivs_shc(sst,Mh_ betax,Mh_ fh,Mh_ gxxx,Mh_ gxyx,Mh_ gxzx,Mh_ gyyx,Mh_ gyzx,Mh_ gzzx,ass);
 	sub_fdderivs_shc(sst,Mh_ betay,Mh_ fh,Mh_ gxxy,Mh_ gxyy,Mh_ gxzy,Mh_ gyyy,Mh_ gyzy,Mh_ gzzy,sas);
@@ -2334,7 +2332,7 @@ int gpu_rhs_ss(RHS_SS_PARA)
 	sub_fderivs_shc( sst,Mh_ Gamz, Mh_ fh,Mh_ Gamzx, Mh_ Gamzy, Mh_ Gamzz,ssa);
 	
 	compute_rhs_ss_part3<<<GRID_DIM,BLOCK_DIM>>>();
-	cudaDeviceSynchronize();
+	cudaThreadSynchronize();
 	
 	computeRicci_ss(sst,Mh_ dxx,Mh_ Rxx,sss, meta);
 	computeRicci_ss(sst,Mh_ dyy,Mh_ Ryy,sss, meta);
@@ -2342,25 +2340,25 @@ int gpu_rhs_ss(RHS_SS_PARA)
 	computeRicci_ss(sst,Mh_ gxy,Mh_ Rxy,aas, meta);
 	computeRicci_ss(sst,Mh_ gxz,Mh_ Rxz,asa, meta);
 	computeRicci_ss(sst,Mh_ gyz,Mh_ Ryz,saa, meta);
-	cudaDeviceSynchronize();
+	cudaThreadSynchronize();
 	
 	compute_rhs_ss_part4<<<GRID_DIM,BLOCK_DIM>>>();
-	cudaDeviceSynchronize();
+	cudaThreadSynchronize();
 	
 	sub_fdderivs_shc(sst,Mh_ chi,Mh_ fh,Mh_ fxx,Mh_ fxy,Mh_ fxz,Mh_ fyy,Mh_ fyz,Mh_ fzz,sss);
 	
-	//cudaDeviceSynchronize();
+	//cudaThreadSynchronize();
 	//compare_result_gpu(0,Mh_ chi,h_3D_SIZE[0]);
 	//compare_result_gpu(1,Mh_ chi,h_3D_SIZE[0]);
 	//compare_result_gpu(2,Mh_ fyz,h_3D_SIZE[0]);
 	
 	compute_rhs_ss_part5<<<GRID_DIM,BLOCK_DIM>>>();
-	cudaDeviceSynchronize();
+	cudaThreadSynchronize();
 	
 	sub_fdderivs_shc(sst,Mh_ Lap,Mh_ fh,Mh_ fxx,Mh_ fxy,Mh_ fxz,Mh_ fyy,Mh_ fyz,Mh_ fzz,sss);
 	
 	compute_rhs_ss_part6<<<GRID_DIM,BLOCK_DIM>>>();
-	cudaDeviceSynchronize();
+	cudaThreadSynchronize();
 	
 #if (GAUGE == 2 || GAUGE == 3 || GAUGE == 4 || GAUGE == 5)
 	sub_fderivs_shc(sst,Mh_ chi,Mh_ fh, Mh_ dtSfx_rhs, Mh_ dtSfy_rhs, Mh_ dtSfz_rhs,sss);
@@ -2425,7 +2423,7 @@ int gpu_rhs_ss(RHS_SS_PARA)
 	}
 	if(co == 0){
 		compute_rhs_ss_part7<<<GRID_DIM,BLOCK_DIM>>>();
-		cudaDeviceSynchronize();
+		cudaThreadSynchronize();
 
 		sub_fderivs_shc(sst,Mh_ Axx,Mh_ fh,Mh_ gxxx,Mh_ gxxy,Mh_ gxxz,sss);
 		sub_fderivs_shc(sst,Mh_ Axy,Mh_ fh,Mh_ gxyx,Mh_ gxyy,Mh_ gxyz,aas);
@@ -2434,7 +2432,7 @@ int gpu_rhs_ss(RHS_SS_PARA)
 		sub_fderivs_shc(sst,Mh_ Ayz,Mh_ fh,Mh_ gyzx,Mh_ gyzy,Mh_ gyzz,saa);
 		sub_fderivs_shc(sst,Mh_ Azz,Mh_ fh,Mh_ gzzx,Mh_ gzzy,Mh_ gzzz,sss);
 		compute_rhs_ss_part8<<<GRID_DIM,BLOCK_DIM>>>();
-		cudaDeviceSynchronize();
+		cudaThreadSynchronize();
 	}
 	
 #if (ABV == 1)
@@ -2514,7 +2512,7 @@ int gpu_rhs_ss(RHS_SS_PARA)
 	//test kodis
 	//sub_kodis_sh(sst,Msh_ drhodx,Mh_ fh2,Msh_ drhody,sss);
 #ifdef TIMING
-	cudaDeviceSynchronize();
+	cudaThreadSynchronize();
 	gettimeofday(&tv2, NULL);
    	cout<<"MPI rank is: "<<mpi_rank<<" GPU TIME is"<<TimeBetween(tv1, tv2)<<" (s)."<<endl; 
 #endif
@@ -2524,55 +2522,4 @@ int gpu_rhs_ss(RHS_SS_PARA)
 	return 0;//TODO return
 }
 
-#if (ABEtype == 2)
-// Z4C Shell GPU: calls BSSN gpu_rhs_ss with trKd=trK+2*TZ, then applies
-// TZ_rhs = alpn1*Hcon/2 and constraint damping on CPU.
-int gpu_rhs_z4c_ss(Z4C_SS_PARA)
-{
-    int matrix_size = ex[0] * ex[1] * ex[2];
-    double k1 = 0.02, k2 = 0.0;
-
-    double *trKd_host = new double[matrix_size];
-    for (int _i = 0; _i < matrix_size; _i++)
-        trKd_host[_i] = trK[_i] + 2.0 * TZ[_i];
-
-    int result = gpu_rhs_ss(calledby, mpi_rank,
-        ex, T, crho, sigma, R, X, Y, Z,
-        drhodx, drhody, drhodz, dsigmadx, dsigmady, dsigmadz,
-        dRdx, dRdy, dRdz,
-        drhodxx, drhodxy, drhodxz, drhodyy, drhodyz, drhodzz,
-        dsigmadxx, dsigmadxy, dsigmadxz, dsigmadyy, dsigmadyz, dsigmadzz,
-        dRdxx, dRdxy, dRdxz, dRdyy, dRdyz, dRdzz,
-        chi, trKd_host, dxx, gxy, gxz, dyy, gyz, dzz,
-        Axx, Axy, Axz, Ayy, Ayz, Azz,
-        Gamx, Gamy, Gamz,
-        Lap, betax, betay, betaz,
-        dtSfx, dtSfy, dtSfz,
-        chi_rhs, trK_rhs,
-        gxx_rhs, gxy_rhs, gxz_rhs, gyy_rhs, gyz_rhs, gzz_rhs,
-        Axx_rhs, Axy_rhs, Axz_rhs, Ayy_rhs, Ayz_rhs, Azz_rhs,
-        Gamx_rhs, Gamy_rhs, Gamz_rhs,
-        Lap_rhs, betax_rhs, betay_rhs, betaz_rhs,
-        dtSfx_rhs, dtSfy_rhs, dtSfz_rhs,
-        rho, Sx, Sy, Sz, Sxx, Sxy, Sxz, Syy, Syz, Szz,
-        Gamxxx, Gamxxy, Gamxxz, Gamxyy, Gamxyz, Gamxzz,
-        Gamyxx, Gamyxy, Gamyxz, Gamyyy, Gamyyz, Gamyzz,
-        Gamzxx, Gamzxy, Gamzxz, Gamzyy, Gamzyz, Gamzzz,
-        Rxx, Rxy, Rxz, Ryy, Ryz, Rzz,
-        ham_Res, movx_Res, movy_Res, movz_Res,
-        Gmx_Res, Gmy_Res, Gmz_Res,
-        Symmetry, Lev, eps, sst, co);
-    delete[] trKd_host;
-    if (result != 0) return result;
-
-    for (int _i = 0; _i < matrix_size; _i++) {
-        double alp = Lap[_i] + 1.0;
-        TZ_rhs[_i] = alp * ham_Res[_i] * 0.5;
-        TZ_rhs[_i]  -= alp * (2.0 + k2) * k1 * TZ[_i];
-        trK_rhs[_i] += alp * k1 * (1.0 - k2) * TZ[_i];
-    }
-    return 0;
-}
-#endif // ABEtype == 2
-
 #endif //WithShell

AMSS_NCKU_source/bssn_rhs.h

@@ -5,8 +5,9 @@
 #ifdef fortran1
 #define f_compute_rhs_bssn compute_rhs_bssn
 #define f_compute_rhs_bssn_ss compute_rhs_bssn_ss
-#define f_compute_rhs_bssn_escalar compute_rhs_bssn_escalar
-#define f_compute_rhs_bssn_escalar_ss compute_rhs_bssn_escalar_ss
+#define f_compute_rhs_bssn_escalar compute_rhs_bssn_escalar
+#define f_compute_rhs_bssn_escalar_matter compute_rhs_bssn_escalar_matter
+#define f_compute_rhs_bssn_escalar_ss compute_rhs_bssn_escalar_ss
 #define f_compute_rhs_Z4c compute_rhs_z4c
 #define f_compute_rhs_Z4cnot compute_rhs_z4cnot
 #define f_compute_rhs_Z4c_ss compute_rhs_z4c_ss
@@ -15,8 +16,9 @@
 #ifdef fortran2
 #define f_compute_rhs_bssn COMPUTE_RHS_BSSN
 #define f_compute_rhs_bssn_ss COMPUTE_RHS_BSSN_SS
-#define f_compute_rhs_bssn_escalar COMPUTE_RHS_BSSN_ESCALAR
-#define f_compute_rhs_bssn_escalar_ss COMPUTE_RHS_BSSN_ESCALAR_SS
+#define f_compute_rhs_bssn_escalar COMPUTE_RHS_BSSN_ESCALAR
+#define f_compute_rhs_bssn_escalar_matter COMPUTE_RHS_BSSN_ESCALAR_MATTER
+#define f_compute_rhs_bssn_escalar_ss COMPUTE_RHS_BSSN_ESCALAR_SS
 #define f_compute_rhs_Z4c COMPUTE_RHS_Z4C
 #define f_compute_rhs_Z4cnot COMPUTE_RHS_Z4CNOT
 #define f_compute_rhs_Z4c_ss COMPUTE_RHS_Z4C_SS
@@ -25,8 +27,9 @@
 #ifdef fortran3
 #define f_compute_rhs_bssn compute_rhs_bssn_
 #define f_compute_rhs_bssn_ss compute_rhs_bssn_ss_
-#define f_compute_rhs_bssn_escalar compute_rhs_bssn_escalar_
-#define f_compute_rhs_bssn_escalar_ss compute_rhs_bssn_escalar_ss_
+#define f_compute_rhs_bssn_escalar compute_rhs_bssn_escalar_
+#define f_compute_rhs_bssn_escalar_matter compute_rhs_bssn_escalar_matter_
+#define f_compute_rhs_bssn_escalar_ss compute_rhs_bssn_escalar_ss_
 #define f_compute_rhs_Z4c compute_rhs_z4c_
 #define f_compute_rhs_Z4cnot compute_rhs_z4cnot_
 #define f_compute_rhs_Z4c_ss compute_rhs_z4c_ss_
@@ -96,10 +99,24 @@ extern "C"
                                   int &, int &, double &, int &, int &);
 }
 
-extern "C"
-{
-        int f_compute_rhs_bssn_escalar(int *, double &, double *, double *, double *,                                                      // ex,T,X,Y,Z
-                                       double *, double *,                                                                                 // chi, trK
+extern "C"
+{
+        int f_compute_rhs_bssn_escalar_matter(int *, double &, double *, double *, double *,                                             // ex,T,X,Y,Z
+                                              double *, double *,                                                                        // chi, trK
+                                              double *, double *, double *, double *, double *, double *,                                // gij
+                                              double *, double *, double *, double *, double *, double *,                                // Aij
+                                              double *, double *, double *,                                                              // Gam
+                                              double *, double *, double *, double *, double *, double *, double *,                      // Gauge
+                                              double *, double *,                                                                        // Sphi, Spi
+                                              double *, double *,                                                                        // Sphi, Spi rhs
+                                              double *, double *, double *, double *, double *, double *, double *, double *, double *, double *, // stress-energy
+                                              int &, int &, double &);
+}
+
+extern "C"
+{
+        int f_compute_rhs_bssn_escalar(int *, double &, double *, double *, double *,                                                      // ex,T,X,Y,Z
+                                       double *, double *,                                                                                 // chi, trK
                                        double *, double *, double *, double *, double *, double *,                                         // gij
                                        double *, double *, double *, double *, double *, double *,                                         // Aij
                                        double *, double *, double *,                                                                       // Gam

AMSS_NCKU_source/bssn_rhs_cuda.h

@@ -7,9 +7,6 @@ extern "C" {
 
 enum {
     BSSN_CUDA_STATE_COUNT = 24,
-    BSSN_ESCALAR_CUDA_STATE_COUNT = 26,
-    BSSN_EM_CUDA_STATE_COUNT = 32,
-    BSSN_EM_CUDA_SOURCE_COUNT = 4,
     BSSN_CUDA_MATTER_COUNT = 10
 };
 
@@ -58,53 +55,116 @@ int bssn_cuda_rk4_substep(void *block_tag,
                           int &apply_enforce_ga,
                           double &chitiny);
 
-int bssn_escalar_cuda_rk4_substep(void *block_tag,
-                                  int *ex, double *X, double *Y, double *Z,
-                                  double **state_host_in,
-                                  double **state_host_out,
-                                  const double *propspeed,
-                                  const double *soa_flat,
-                                  const double *bbox,
-                                  double &dT,
-                                  double &T,
-                                  int &RK4,
-                                  int &apply_bam_bc,
-                                  int &Symmetry,
-	                                  int &Lev,
-	                                  double &eps,
-	                                  int &co,
-	                                  int &keep_resident_state,
-	                                  int &apply_enforce_ga,
-	                                  double &chitiny);
+int bssn_cuda_compute_escalar_matter(void *block_tag,
+                                     int *ex, double *X, double *Y, double *Z,
+                                     double **state_host_in,
+                                     double *Sphi_host,
+                                     double *Spi_host,
+                                     double *Sphi_rhs_host,
+                                     double *Spi_rhs_host,
+                                     double a2,
+                                     int &Symmetry,
+                                     int &Lev,
+                                     double &eps,
+                                     int &co,
+                                     int &apply_enforce_ga);
 
-int bssn_escalar_cuda_compute_constraints(int *ex, double *X, double *Y, double *Z,
-                                          double **state_host_in,
-                                          double **constraint_host_out,
-                                          int &Symmetry,
-                                          int &Lev,
-                                          double &eps);
+int bssn_cuda_escalar_finalize_scalar_fields(void *block_tag,
+                                             int *ex, double *X, double *Y, double *Z,
+                                             double *Sphi_out_host,
+                                             double *Spi_out_host,
+                                             const double *propspeed,
+                                             const double *soa_flat,
+                                             const double *bbox,
+                                             double &dT,
+                                             int &RK4,
+                                             int &apply_bam_bc,
+                                             int &Symmetry,
+                                             int &Lev,
+                                             double &eps,
+                                             int &precor);
 
-int bssn_em_cuda_rk4_substep(void *block_tag,
-                             int *ex, double *X, double *Y, double *Z,
-                             double **state_host_in,
-                             double **state_host_out,
-                             double **source_host,
-                             const double *propspeed,
-                             const double *soa_flat,
-                             const double *bbox,
-                             double &dT,
-                             double &T,
-                             int &RK4,
-                             int &apply_bam_bc,
-                             int &Symmetry,
-                             int &Lev,
-                             double &eps,
-                             int &co,
-                             int &keep_resident_state,
-                             int &apply_enforce_ga,
-                             double &chitiny);
+int bssn_cuda_escalar_has_resident_fields(void *block_tag,
+                                          double *Sphi_host,
+                                          double *Spi_host);
 
-int bssn_em_cuda_resident_zero_fast_state(void *block_tag);
+int bssn_cuda_escalar_has_any_resident_fields(void *block_tag);
+
+int bssn_cuda_escalar_download_fields_if_present(void *block_tag,
+                                                 int *ex,
+                                                 double *Sphi_host,
+                                                 double *Spi_host);
+
+int bssn_cuda_pack_escalar_batch_to_host_buffer(void *block_tag,
+                                                double **scalar_host_key,
+                                                double *host_buffer,
+                                                int *ex,
+                                                int i0, int j0, int k0,
+                                                int sx, int sy, int sz);
+
+int bssn_cuda_unpack_escalar_batch_from_host_buffer(void *block_tag,
+                                                    double **scalar_host_key,
+                                                    double *host_buffer,
+                                                    int *ex,
+                                                    int i0, int j0, int k0,
+                                                    int sx, int sy, int sz);
+
+int bssn_cuda_pack_escalar_batch_to_device_buffer(void *block_tag,
+                                                  double **scalar_host_key,
+                                                  double *device_buffer,
+                                                  int *ex,
+                                                  int i0, int j0, int k0,
+                                                  int sx, int sy, int sz);
+
+int bssn_cuda_unpack_escalar_batch_from_device_buffer(void *block_tag,
+                                                      double **scalar_host_key,
+                                                      double *device_buffer,
+                                                      int *ex,
+                                                      int i0, int j0, int k0,
+                                                      int sx, int sy, int sz);
+
+int bssn_cuda_restrict_escalar_batch_to_host_buffer(void *block_tag,
+                                                    double **scalar_host_key,
+                                                    double *host_buffer,
+                                                    int *ex,
+                                                    int sx, int sy, int sz,
+                                                    int fi0, int fj0, int fk0,
+                                                    const double *scalar_soa);
+
+int bssn_cuda_prolong_escalar_batch_to_host_buffer(void *block_tag,
+                                                   double **scalar_host_key,
+                                                   double *host_buffer,
+                                                   int *ex,
+                                                   int sx, int sy, int sz,
+                                                   int ii0, int jj0, int kk0,
+                                                   int lbc_i, int lbc_j, int lbc_k,
+                                                   const double *scalar_soa);
+
+int bssn_cuda_restrict_escalar_batch_to_device_buffer(void *block_tag,
+                                                      double **scalar_host_key,
+                                                      double *device_buffer,
+                                                      int *ex,
+                                                      int sx, int sy, int sz,
+                                                      int fi0, int fj0, int fk0,
+                                                      const double *scalar_soa);
+
+int bssn_cuda_prolong_escalar_batch_to_device_buffer(void *block_tag,
+                                                     double **scalar_host_key,
+                                                     double *device_buffer,
+                                                     int *ex,
+                                                     int sx, int sy, int sz,
+                                                     int ii0, int jj0, int kk0,
+                                                     int lbc_i, int lbc_j, int lbc_k,
+                                                     const double *scalar_soa);
+
+int bssn_cuda_prepare_escalar_inter_time_level(void *block_tag,
+                                               int *ex,
+                                               double **src1_host_key,
+                                               double **src2_host_key,
+                                               double **src3_host_key,
+                                               double **dst_host_key,
+                                               int source_count,
+                                               int tindex);
 
 int bssn_cuda_copy_state_region_to_host(void *block_tag,
                                         int state_index,
@@ -124,37 +184,13 @@ int bssn_cuda_download_resident_state(void *block_tag,
                                       int *ex,
                                       double **state_host_out);
 
-int bssn_escalar_cuda_download_resident_state(void *block_tag,
-                                              int *ex,
-                                              double **state_host_out);
-
-int bssn_cuda_upload_resident_state_count(void *block_tag,
-                                          int *ex,
-                                          double **state_host_in,
-                                          int state_count);
-
-int bssn_escalar_cuda_upload_resident_state(void *block_tag,
-                                            int *ex,
-                                            double **state_host_in);
-
-int bssn_cuda_keep_only_resident_state_count(void *block_tag,
-                                             int *ex,
-                                             double **state_host_key,
-                                             int state_count);
-
-int bssn_escalar_cuda_keep_only_resident_state(void *block_tag,
-                                               int *ex,
-                                               double **state_host_key);
-
-int bssn_cuda_download_resident_state_count_if_present(void *block_tag,
-                                                       int *ex,
-                                                       double **state_host_out,
-                                                       int state_count);
-
 int bssn_cuda_download_resident_state_if_present(void *block_tag,
                                                 int *ex,
                                                 double **state_host_out);
 
+int bssn_cuda_resident_state_matches(void *block_tag,
+                                     double **state_host_key);
+
 int bssn_cuda_download_constraint_outputs(int *ex,
                                           double **constraint_host_out);
 
@@ -181,7 +217,6 @@ int bssn_cuda_interp_state_point3(void *block_tag,
                                   double pz,
                                   int ordn,
                                   int symmetry,
-                                  double **state_host_key,
                                   const double *soa3,
                                   double *out3);
 
@@ -211,15 +246,6 @@ int bssn_cuda_unpack_state_region_from_host_buffer(void *block_tag,
                                                    int i0, int j0, int k0,
                                                    int sx, int sy, int sz);
 
-int bssn_cuda_unpack_state_region_from_host_buffer_for_host_views(void *block_tag,
-                                                                  double **state_host_key,
-                                                                  int state_count,
-                                                                  int state_index,
-                                                                  double *host_buffer,
-                                                                  int *ex,
-                                                                  int i0, int j0, int k0,
-                                                                  int sx, int sy, int sz);
-
 int bssn_cuda_pack_state_batch_to_host_buffer(void *block_tag,
                                               int state_count,
                                               double *host_buffer,
@@ -250,8 +276,44 @@ int bssn_cuda_unpack_state_batch_from_host_buffer_for_host_views(void *block_tag
                                                                  int i0, int j0, int k0,
                                                                  int sx, int sy, int sz);
 
+int bssn_cuda_restrict_state_batch_to_host_buffer_for_host_views(void *block_tag,
+                                                                 double **state_host_key,
+                                                                 int state_count,
+                                                                 double *host_buffer,
+                                                                 int *ex,
+                                                                 int sx, int sy, int sz,
+                                                                 int fi0, int fj0, int fk0,
+                                                                 const double *state_soa);
+
+int bssn_cuda_restrict_state_batch_to_host_buffer(void *block_tag,
+                                                  int state_count,
+                                                  double *host_buffer,
+                                                  int *ex,
+                                                  int sx, int sy, int sz,
+                                                  int fi0, int fj0, int fk0,
+                                                  const double *state_soa);
+
+int bssn_cuda_prolong_state_batch_to_host_buffer_for_host_views(void *block_tag,
+                                                                double **state_host_key,
+                                                                int state_count,
+                                                                double *host_buffer,
+                                                                int *ex,
+                                                                int sx, int sy, int sz,
+                                                                int ii0, int jj0, int kk0,
+                                                                int lbc_i, int lbc_j, int lbc_k,
+                                                                const double *state_soa);
+
+int bssn_cuda_prolong_state_batch_to_host_buffer(void *block_tag,
+                                                 int state_count,
+                                                 double *host_buffer,
+                                                 int *ex,
+                                                 int sx, int sy, int sz,
+                                                 int ii0, int jj0, int kk0,
+                                                 int lbc_i, int lbc_j, int lbc_k,
+                                                 const double *state_soa);
+
 int bssn_cuda_pack_state_batch_to_device_buffer(void *block_tag,
-                                                int state_count,
+                                                 int state_count,
                                                 double *device_buffer,
                                                 int *ex,
                                                 int i0, int j0, int k0,
@@ -390,7 +452,6 @@ int bssn_cuda_upload_state_subset(void *block_tag,
 
 int bssn_cuda_prepare_inter_time_level(void *block_tag,
                                        int *ex,
-                                       int state_count,
                                        double **src1_host_key,
                                        double **src2_host_key,
                                        double **src3_host_key,
@@ -404,10 +465,6 @@ void bssn_cuda_release_step_ctx(void *block_tag);
 
 #ifdef __cplusplus
 }
-// C++-only helpers declared for derived equation classes (Z4C, etc.)
-// Defined in bssn_class.C. Requires MyList, Patch, var from including TU.
-bool bssn_cuda_use_resident_sync(int lev);
-void bssn_cuda_download_level_state_if_present(MyList<Patch> *PatL, MyList<var> *vars, int myrank);
 #endif
 
 #endif

AMSS_NCKU_source/checkpoint.C

@@ -76,11 +76,8 @@ checkpoint::checkpoint(bool checked, const char fname[], int myrank) : filename(
 
 	I_Print = (myrank == 0);
 
-	size_t filename_len = out_dir.size() + strlen(fname) + 32;
-#ifdef CHECKDETAIL
-	filename_len += 32;
-#endif
-	filename = new char[filename_len];
+	int i = strlen(fname);
+	filename = new char[i+30];
 	// cout << filename << endl;
 	// cout << i << endl;
 
@@ -103,12 +100,12 @@ checkpoint::checkpoint(bool checked, const char fname[], int myrank) : filename(
                 cout << " checkpoint class created " << endl;
         }
 }
-checkpoint::~checkpoint()
-{
-	CheckList->clearList();
-	if (filename)
-		delete[] filename;
-}
+checkpoint::~checkpoint()
+{
+	CheckList->clearList();
+	if (I_Print)
+		delete[] filename;
+}
 
 void checkpoint::addvariable(var *VV)
 {
@@ -139,7 +136,7 @@ void checkpoint::writecheck_cgh(double time, cgh *GH)
 	if (I_Print)
 	{
 		// char fname[50];
-		char fname[4096]; 
+		char fname[50+50]; 
 		sprintf(fname, "%s_cgh.CHK", filename);
 
 		outfile.open(fname, ios::out | ios::trunc);
@@ -198,7 +195,7 @@ void checkpoint::readcheck_cgh(double &time, cgh *GH, int myrank, int nprocs, in
 	int DIM = dim;
 	ifstream infile;
 	// char fname[50];
-	char fname[4096]; 
+	char fname[50+50]; 
 	sprintf(fname, "%s_cgh.CHK", filename);
 
 	infile.open(fname);
@@ -300,7 +297,7 @@ void checkpoint::writecheck_sh(double time, ShellPatch *SH)
 
 	if (I_Print)
 	{
-		char fname[4096];
+		char fname[50];
 		sprintf(fname, "%s_sh.CHK", filename);
 
 		outfile.open(fname, ios::out | ios::trunc);
@@ -338,7 +335,7 @@ void checkpoint::readcheck_sh(ShellPatch *SH, int myrank)
 	int DIM = dim;
 	ifstream infile;
 	// char fname[50];
-	char fname[4096]; 
+	char fname[50+50]; 
 	sprintf(fname, "%s_sh.CHK", filename);
 
 	infile.open(fname);
@@ -393,7 +390,7 @@ void checkpoint::write_Black_Hole_position(int BH_num_input, int BH_num, double
 
 	if (I_Print)
 	{
-		char fname[4096];
+		char fname[50];
 		sprintf(fname, "%s_BHp.CHK", filename);
 
 		outfile.open(fname, ios::out | ios::trunc);
@@ -420,7 +417,7 @@ void checkpoint::read_Black_Hole_position(int &BH_num_input, int &BH_num, double
 {
 	ifstream infile;
 	// char fname[50];
-	char fname[4096]; 
+	char fname[50+50]; 
 	sprintf(fname, "%s_BHp.CHK", filename);
 
 	infile.open(fname);
@@ -464,7 +461,7 @@ void checkpoint::write_bssn(double LastDump, double Last2dDump, double LastAnas)
 
 	if (I_Print)
 	{
-		char fname[4096];
+		char fname[50];
 		sprintf(fname, "%s_bssn.CHK", filename);
 
 		outfile.open(fname, ios::out | ios::trunc);
@@ -484,7 +481,7 @@ void checkpoint::read_bssn(double &LastDump, double &Last2dDump, double &LastAna
 {
 	ifstream infile;
 	// char fname[50];
-	char fname[4096]; 
+	char fname[50+50]; 
 	sprintf(fname, "%s_bssn.CHK", filename);
 
 	infile.open(fname);
@@ -509,7 +506,7 @@ void checkpoint::write_bssn(double LastDump, double Last2dDump, double LastAnas)
 	ofstream outfile;
 
 	// char fname[50];
-	char fname[4096]; 
+	char fname[50+50]; 
 	sprintf(fname, "%s_bssn.CHK", filename);
 
 	outfile.open(fname, ios::out | ios::trunc);
@@ -530,7 +527,7 @@ void checkpoint::read_bssn(double &LastDump, double &Last2dDump, double &LastAna
 {
 	ifstream infile;
 	// char fname[50];
-	char fname[4096]; 
+	char fname[50+50]; 
 	sprintf(fname, "%s_bssn.CHK", filename);
 
 	infile.open(fname);
@@ -554,7 +551,7 @@ void checkpoint::write_Black_Hole_position(int BH_num_input, int BH_num, double
 	ofstream outfile;
 
 	// char fname[50];
-	char fname[4096]; 
+	char fname[50+50]; 
 	sprintf(fname, "%s_BHp.CHK", filename);
 
 	outfile.open(fname, ios::out | ios::trunc);
@@ -584,7 +581,7 @@ void checkpoint::read_Black_Hole_position(int &BH_num_input, int &BH_num, double
 {
 	ifstream infile;
 	// char fname[50];
-	char fname[4096]; 
+	char fname[50+50]; 
 	sprintf(fname, "%s_BHp.CHK", filename);
 
 	infile.open(fname);
@@ -631,7 +628,7 @@ void checkpoint::writecheck_cgh(double time, cgh *GH)
 	ofstream outfile;
 
 	// char fname[50];
-	char fname[4096]; 
+	char fname[50+50]; 
 	sprintf(fname, "%s_cgh.CHK", filename);
 
 	outfile.open(fname, ios::out | ios::trunc);
@@ -741,7 +738,7 @@ void checkpoint::readcheck_cgh(double &time, cgh *GH, int myrank, int nprocs, in
 	int DIM = dim;
 	ifstream infile;
 	// char fname[50];
-	char fname[4096]; 
+	char fname[50+50]; 
 	sprintf(fname, "%s_cgh.CHK", filename);
 
 	infile.open(fname);

AMSS_NCKU_source/fd_cuda_helpers.cuh

@@ -1,412 +0,0 @@
-#ifndef AMSS_NCKU_FD_CUDA_HELPERS_CUH
-#define AMSS_NCKU_FD_CUDA_HELPERS_CUH
-
-#ifndef ghost_width
-#error "ghost_width must be defined before including fd_cuda_helpers.cuh"
-#endif
-
-#if ghost_width < 2 || ghost_width > 5
-#error "CUDA finite-difference helpers support ghost_width 2..5"
-#endif
-
-#define AMSS_FD_CENTER_RADIUS (ghost_width - 1)
-#define AMSS_FD_LK_RADIUS     (ghost_width)
-
-__device__ __forceinline__ int fd_axis_radius(int qF, int qminF, int qmaxF)
-{
-#if AMSS_FD_CENTER_RADIUS >= 4
-    if (qF - 4 >= qminF && qF + 4 <= qmaxF) return 4;
-#endif
-#if AMSS_FD_CENTER_RADIUS >= 3
-    if (qF - 3 >= qminF && qF + 3 <= qmaxF) return 3;
-#endif
-#if AMSS_FD_CENTER_RADIUS >= 2
-    if (qF - 2 >= qminF && qF + 2 <= qmaxF) return 2;
-#endif
-    if (qF - 1 >= qminF && qF + 1 <= qmaxF) return 1;
-    return 0;
-}
-
-__device__ __forceinline__ int fd_common_radius(int iF, int jF, int kF,
-                                                int iminF, int jminF, int kminF,
-                                                int imaxF, int jmaxF, int kmaxF)
-{
-    int r = fd_axis_radius(iF, iminF, imaxF);
-    const int ry = fd_axis_radius(jF, jminF, jmaxF);
-    const int rz = fd_axis_radius(kF, kminF, kmaxF);
-    if (ry < r) r = ry;
-    if (rz < r) r = rz;
-    return r;
-}
-
-__device__ __forceinline__ double fd_first_coef(int r, int off)
-{
-    switch (r) {
-    case 1:
-        if (off == -1) return -1.0;
-        if (off ==  1) return  1.0;
-        return 0.0;
-    case 2:
-        if (off == -2) return  1.0;
-        if (off == -1) return -8.0;
-        if (off ==  1) return  8.0;
-        if (off ==  2) return -1.0;
-        return 0.0;
-    case 3:
-        if (off == -3) return  -1.0;
-        if (off == -2) return   9.0;
-        if (off == -1) return -45.0;
-        if (off ==  1) return  45.0;
-        if (off ==  2) return  -9.0;
-        if (off ==  3) return   1.0;
-        return 0.0;
-    case 4:
-        if (off == -4) return    3.0;
-        if (off == -3) return  -32.0;
-        if (off == -2) return  168.0;
-        if (off == -1) return -672.0;
-        if (off ==  1) return  672.0;
-        if (off ==  2) return -168.0;
-        if (off ==  3) return   32.0;
-        if (off ==  4) return   -3.0;
-        return 0.0;
-    default:
-        return 0.0;
-    }
-}
-
-__device__ __forceinline__ double fd_second_coef(int r, int off)
-{
-    switch (r) {
-    case 1:
-        if (off == -1) return  1.0;
-        if (off ==  0) return -2.0;
-        if (off ==  1) return  1.0;
-        return 0.0;
-    case 2:
-        if (off == -2) return  -1.0;
-        if (off == -1) return  16.0;
-        if (off ==  0) return -30.0;
-        if (off ==  1) return  16.0;
-        if (off ==  2) return  -1.0;
-        return 0.0;
-    case 3:
-        if (off == -3) return    2.0;
-        if (off == -2) return  -27.0;
-        if (off == -1) return  270.0;
-        if (off ==  0) return -490.0;
-        if (off ==  1) return  270.0;
-        if (off ==  2) return  -27.0;
-        if (off ==  3) return    2.0;
-        return 0.0;
-    case 4:
-        if (off == -4) return     -9.0;
-        if (off == -3) return    128.0;
-        if (off == -2) return  -1008.0;
-        if (off == -1) return   8064.0;
-        if (off ==  0) return -14350.0;
-        if (off ==  1) return   8064.0;
-        if (off ==  2) return  -1008.0;
-        if (off ==  3) return    128.0;
-        if (off ==  4) return     -9.0;
-        return 0.0;
-    default:
-        return 0.0;
-    }
-}
-
-__device__ __forceinline__ double fd_first_denom(int r)
-{
-    return (r == 4) ? 840.0 : ((r == 3) ? 60.0 : ((r == 2) ? 12.0 : 2.0));
-}
-
-__device__ __forceinline__ double fd_second_denom(int r)
-{
-    return (r == 4) ? 5040.0 : ((r == 3) ? 180.0 : ((r == 2) ? 12.0 : 1.0));
-}
-
-__device__ __forceinline__ double fd_fetch_axis(const double *src,
-                                                int iF, int jF, int kF,
-                                                int axis, int off,
-                                                int SoA0, int SoA1, int SoA2)
-{
-    if (axis == 0) iF += off;
-    else if (axis == 1) jF += off;
-    else kF += off;
-    return fetch_sym_ord2_direct(src, iF, jF, kF, SoA0, SoA1, SoA2);
-}
-
-__device__ __forceinline__ double fd_fetch_axis2(const double *src,
-                                                 int iF, int jF, int kF,
-                                                 int axis_a, int off_a,
-                                                 int axis_b, int off_b,
-                                                 int SoA0, int SoA1, int SoA2)
-{
-    if (axis_a == 0) iF += off_a;
-    else if (axis_a == 1) jF += off_a;
-    else kF += off_a;
-    if (axis_b == 0) iF += off_b;
-    else if (axis_b == 1) jF += off_b;
-    else kF += off_b;
-    return fetch_sym_ord2_direct(src, iF, jF, kF, SoA0, SoA1, SoA2);
-}
-
-__device__ __forceinline__ double fd_first_axis_radius(const double *src,
-                                                       int iF, int jF, int kF,
-                                                       int axis, int r, double h,
-                                                       int SoA0, int SoA1, int SoA2)
-{
-    if (r <= 0) return 0.0;
-    double s = 0.0;
-#pragma unroll
-    for (int off = -4; off <= 4; ++off) {
-        const double c = fd_first_coef(r, off);
-        if (c != 0.0) {
-            s += c * fd_fetch_axis(src, iF, jF, kF, axis, off, SoA0, SoA1, SoA2);
-        }
-    }
-    return s / (fd_first_denom(r) * h);
-}
-
-__device__ __forceinline__ double fd_second_axis_radius(const double *src,
-                                                        int iF, int jF, int kF,
-                                                        int axis, int r, double h,
-                                                        int SoA0, int SoA1, int SoA2)
-{
-    if (r <= 0) return 0.0;
-    double s = 0.0;
-#pragma unroll
-    for (int off = -4; off <= 4; ++off) {
-        const double c = fd_second_coef(r, off);
-        if (c != 0.0) {
-            s += c * fd_fetch_axis(src, iF, jF, kF, axis, off, SoA0, SoA1, SoA2);
-        }
-    }
-    return s / (fd_second_denom(r) * h * h);
-}
-
-__device__ __forceinline__ double fd_mixed_axis_radius(const double *src,
-                                                       int iF, int jF, int kF,
-                                                       int axis_a, int r_a, double h_a,
-                                                       int axis_b, int r_b, double h_b,
-                                                       int SoA0, int SoA1, int SoA2)
-{
-    if (r_a <= 0 || r_b <= 0) return 0.0;
-    double s = 0.0;
-#pragma unroll
-    for (int off_a = -4; off_a <= 4; ++off_a) {
-        const double ca = fd_first_coef(r_a, off_a);
-        if (ca == 0.0) continue;
-#pragma unroll
-        for (int off_b = -4; off_b <= 4; ++off_b) {
-            const double cb = fd_first_coef(r_b, off_b);
-            if (cb != 0.0) {
-                s += ca * cb * fd_fetch_axis2(src, iF, jF, kF, axis_a, off_a,
-                                              axis_b, off_b, SoA0, SoA1, SoA2);
-            }
-        }
-    }
-    return s / (fd_first_denom(r_a) * fd_first_denom(r_b) * h_a * h_b);
-}
-
-__device__ __forceinline__ void fd_compute_first3(const double *src,
-                                                  int iF, int jF, int kF,
-                                                  int iminF, int jminF, int kminF,
-                                                  int imaxF, int jmaxF, int kmaxF,
-                                                  int SoA0, int SoA1, int SoA2,
-                                                  double &fx, double &fy, double &fz)
-{
-#if ghost_width == 3
-    const int r = fd_common_radius(iF, jF, kF, iminF, jminF, kminF, imaxF, jmaxF, kmaxF);
-    fx = fd_first_axis_radius(src, iF, jF, kF, 0, r, d_gp.dX, SoA0, SoA1, SoA2);
-    fy = fd_first_axis_radius(src, iF, jF, kF, 1, r, d_gp.dY, SoA0, SoA1, SoA2);
-    fz = fd_first_axis_radius(src, iF, jF, kF, 2, r, d_gp.dZ, SoA0, SoA1, SoA2);
-#else
-    fx = fd_first_axis_radius(src, iF, jF, kF, 0, fd_axis_radius(iF, iminF, imaxF),
-                              d_gp.dX, SoA0, SoA1, SoA2);
-    fy = fd_first_axis_radius(src, iF, jF, kF, 1, fd_axis_radius(jF, jminF, jmaxF),
-                              d_gp.dY, SoA0, SoA1, SoA2);
-    fz = fd_first_axis_radius(src, iF, jF, kF, 2, fd_axis_radius(kF, kminF, kmaxF),
-                              d_gp.dZ, SoA0, SoA1, SoA2);
-#endif
-}
-
-__device__ __forceinline__ void fd_compute_second6(const double *src,
-                                                   int iF, int jF, int kF,
-                                                   int iminF, int jminF, int kminF,
-                                                   int imaxF, int jmaxF, int kmaxF,
-                                                   int SoA0, int SoA1, int SoA2,
-                                                   double &fxx, double &fxy, double &fxz,
-                                                   double &fyy, double &fyz, double &fzz)
-{
-#if ghost_width == 3
-    const int r = fd_common_radius(iF, jF, kF, iminF, jminF, kminF, imaxF, jmaxF, kmaxF);
-    const int rx = r, ry = r, rz = r;
-#else
-    const int rx = fd_axis_radius(iF, iminF, imaxF);
-    const int ry = fd_axis_radius(jF, jminF, jmaxF);
-    const int rz = fd_axis_radius(kF, kminF, kmaxF);
-#endif
-    fxx = fd_second_axis_radius(src, iF, jF, kF, 0, rx, d_gp.dX, SoA0, SoA1, SoA2);
-    fyy = fd_second_axis_radius(src, iF, jF, kF, 1, ry, d_gp.dY, SoA0, SoA1, SoA2);
-    fzz = fd_second_axis_radius(src, iF, jF, kF, 2, rz, d_gp.dZ, SoA0, SoA1, SoA2);
-    fxy = fd_mixed_axis_radius(src, iF, jF, kF, 0, rx, d_gp.dX, 1, ry, d_gp.dY, SoA0, SoA1, SoA2);
-    fxz = fd_mixed_axis_radius(src, iF, jF, kF, 0, rx, d_gp.dX, 2, rz, d_gp.dZ, SoA0, SoA1, SoA2);
-    fyz = fd_mixed_axis_radius(src, iF, jF, kF, 1, ry, d_gp.dY, 2, rz, d_gp.dZ, SoA0, SoA1, SoA2);
-}
-
-__device__ __forceinline__ bool fd_lop_fits(int qF, int qminF, int qmaxF,
-                                            int dir, int lo, int hi)
-{
-    for (int off = lo; off <= hi; ++off) {
-        const int q = qF + dir * off;
-        if (q < qminF || q > qmaxF) return false;
-    }
-    return true;
-}
-
-__device__ __forceinline__ double fd_lop_fetch_sum(const double *src,
-                                                   int iF, int jF, int kF,
-                                                   int axis, int dir,
-                                                   const double *coef,
-                                                   int lo, int hi,
-                                                   int SoA0, int SoA1, int SoA2)
-{
-    double s = 0.0;
-    for (int off = lo; off <= hi; ++off) {
-        const double c = coef[off - lo];
-        if (c != 0.0) {
-            s += c * fd_fetch_axis(src, iF, jF, kF, axis, dir * off, SoA0, SoA1, SoA2);
-        }
-    }
-    return s;
-}
-
-__device__ __forceinline__ double fd_lopsided_axis(const double *src,
-                                                   int iF, int jF, int kF,
-                                                   int axis, double speed,
-                                                   int qF, int qminF, int qmaxF,
-                                                   double h,
-                                                   int SoA0, int SoA1, int SoA2)
-{
-    if (speed == 0.0) return 0.0;
-    const int dir = (speed > 0.0) ? 1 : -1;
-    const double mag = (speed > 0.0) ? speed : -speed;
-
-#if ghost_width == 2
-    if (fd_lop_fits(qF, qminF, qmaxF, dir, 0, 2)) {
-        const double c[] = {-3.0, 4.0, -1.0};
-        return mag * fd_lop_fetch_sum(src, iF, jF, kF, axis, dir, c, 0, 2, SoA0, SoA1, SoA2) / (2.0 * h);
-    }
-    if (fd_lop_fits(qF, qminF, qmaxF, dir, 0, 1)) {
-        const double c[] = {-1.0, 1.0};
-        return mag * fd_lop_fetch_sum(src, iF, jF, kF, axis, dir, c, 0, 1, SoA0, SoA1, SoA2) / (2.0 * h);
-    }
-    return 0.0;
-#elif ghost_width == 3
-    if (fd_lop_fits(qF, qminF, qmaxF, dir, -1, 3)) {
-        const double c[] = {-3.0, -10.0, 18.0, -6.0, 1.0};
-        return mag * fd_lop_fetch_sum(src, iF, jF, kF, axis, dir, c, -1, 3, SoA0, SoA1, SoA2) / (12.0 * h);
-    }
-    const int r = fd_axis_radius(qF, qminF, qmaxF);
-    return speed * fd_first_axis_radius(src, iF, jF, kF, axis, r, h, SoA0, SoA1, SoA2);
-#elif ghost_width == 4
-    if (fd_lop_fits(qF, qminF, qmaxF, dir, -2, 4)) {
-        const double c[] = {2.0, -24.0, -35.0, 80.0, -30.0, 8.0, -1.0};
-        return mag * fd_lop_fetch_sum(src, iF, jF, kF, axis, dir, c, -2, 4, SoA0, SoA1, SoA2) / (60.0 * h);
-    }
-    if (fd_lop_fits(qF, qminF, qmaxF, dir, -1, 5)) {
-        const double c[] = {-10.0, -77.0, 150.0, -100.0, 50.0, -15.0, 2.0};
-        return mag * fd_lop_fetch_sum(src, iF, jF, kF, axis, dir, c, -1, 5, SoA0, SoA1, SoA2) / (60.0 * h);
-    }
-    const int r = fd_axis_radius(qF, qminF, qmaxF);
-    return speed * fd_first_axis_radius(src, iF, jF, kF, axis, r, h, SoA0, SoA1, SoA2);
-#else
-    if (fd_lop_fits(qF, qminF, qmaxF, dir, -3, 5)) {
-        const double c[] = {-5.0, 60.0, -420.0, -378.0, 1050.0, -420.0, 140.0, -30.0, 3.0};
-        return mag * fd_lop_fetch_sum(src, iF, jF, kF, axis, dir, c, -3, 5, SoA0, SoA1, SoA2) / (840.0 * h);
-    }
-    const int r = fd_axis_radius(qF, qminF, qmaxF);
-    return speed * fd_first_axis_radius(src, iF, jF, kF, axis, r, h, SoA0, SoA1, SoA2);
-#endif
-}
-
-__device__ __forceinline__ double fd_ko_coef(int r, int off)
-{
-    const int a = off < 0 ? -off : off;
-    if (r == 2) {
-        if (a == 0) return 6.0;
-        if (a == 1) return -4.0;
-        if (a == 2) return 1.0;
-    } else if (r == 3) {
-        if (a == 0) return -20.0;
-        if (a == 1) return 15.0;
-        if (a == 2) return -6.0;
-        if (a == 3) return 1.0;
-    } else if (r == 4) {
-        if (a == 0) return 70.0;
-        if (a == 1) return -56.0;
-        if (a == 2) return 28.0;
-        if (a == 3) return -8.0;
-        if (a == 4) return 1.0;
-    } else if (r == 5) {
-        if (a == 0) return -252.0;
-        if (a == 1) return 210.0;
-        if (a == 2) return -120.0;
-        if (a == 3) return 45.0;
-        if (a == 4) return -10.0;
-        if (a == 5) return 1.0;
-    }
-    return 0.0;
-}
-
-__device__ __forceinline__ double fd_ko_axis(const double *src,
-                                             int iF, int jF, int kF,
-                                             int axis, int r,
-                                             int SoA0, int SoA1, int SoA2)
-{
-    double s = 0.0;
-#pragma unroll
-    for (int off = -5; off <= 5; ++off) {
-        if (off < -r || off > r) continue;
-        const double c = fd_ko_coef(r, off);
-        if (c != 0.0) {
-            s += c * fd_fetch_axis(src, iF, jF, kF, axis, off, SoA0, SoA1, SoA2);
-        }
-    }
-    return s;
-}
-
-__device__ __forceinline__ double fd_ko_term(const double *src,
-                                             int iF, int jF, int kF,
-                                             int iminF, int jminF, int kminF,
-                                             int imaxF, int jmaxF, int kmaxF,
-                                             double eps_val,
-                                             int SoA0, int SoA1, int SoA2)
-{
-    const int r = AMSS_FD_LK_RADIUS;
-    if (eps_val <= 0.0) return 0.0;
-#if ghost_width >= 4
-    if (iF - r <= iminF || iF + r >= imaxF ||
-        jF - r <= jminF || jF + r >= jmaxF ||
-        kF - r <= kminF || kF + r >= kmaxF) {
-        return 0.0;
-    }
-#else
-    if (iF - r < iminF || iF + r > imaxF ||
-        jF - r < jminF || jF + r > jmaxF ||
-        kF - r < kminF || kF + r > kmaxF) {
-        return 0.0;
-    }
-#endif
-    double cof = 1.0;
-#pragma unroll
-    for (int n = 0; n < 2 * r; ++n) cof *= 2.0;
-    const double sign = (r & 1) ? 1.0 : -1.0;
-    const double dx = fd_ko_axis(src, iF, jF, kF, 0, r, SoA0, SoA1, SoA2);
-    const double dy = fd_ko_axis(src, iF, jF, kF, 1, r, SoA0, SoA1, SoA2);
-    const double dz = fd_ko_axis(src, iF, jF, kF, 2, r, SoA0, SoA1, SoA2);
-    return sign * eps_val * (dx / d_gp.dX + dy / d_gp.dY + dz / d_gp.dZ) / cof;
-}
-
-#endif

AMSS_NCKU_source/gpu_rhsSS_mem.h

@@ -1,6 +1,6 @@
 #ifndef GPU_MEM_H_
 #define GPU_MEM_H_
-#include "macrodef.h"
+#include "macrodef.fh"
 
 #ifdef WithShell
 struct Metass
@@ -48,8 +48,6 @@ struct Meta
 	double * Gamx_rhs,*Gamy_rhs,*Gamz_rhs;//out
 	double * Lap_rhs, *betax_rhs, *betay_rhs, *betaz_rhs;//out
 	double * dtSfx_rhs,*dtSfy_rhs,*dtSfz_rhs;//out
-	double * TZ;                                  //in  (Z4C)
-	double * TZ_rhs;                              //out (Z4C)
 	double * rho,*Sx,*Sy,*Sz			;	//in
 	double * Sxx,*Sxy,*Sxz,*Syy,*Syz,*Szz;	//in
 	
@@ -134,8 +132,6 @@ __constant__ double SYM = 1.0;
 __constant__ double ANTI = -1.0; 
 __constant__ double FF = 0.75;
 __constant__ double eta = 2.0;
-__constant__ double kappa1_c = 0.02;
-__constant__ double kappa2_c = 0.0;
 __constant__ double F1o3;
 __constant__ double F2o3;
 __constant__ double F3o2 = 1.5;

AMSS_NCKU_source/macrodef.fh

@@ -13,7 +13,7 @@
 
 #define ABV 0
 
-#define EScalar_CC 2
+#define EScalar_CC  2
 
 #if 0
 

AMSS_NCKU_source/macrodef.h

@@ -10,7 +10,7 @@
 
 #define GaussInt
 
-#define ABEtype 0
+#define ABEtype 1
 
 //#define With_AHF
 #define Psi4type 0
@@ -167,3 +167,4 @@
 #define TINY 1e-10
 
 #endif   /* MICRODEF_H */
+

AMSS_NCKU_source/makefile

@@ -35,6 +35,7 @@ f90appflags = -O3 -xHost -fp-model fast=2 -fma -ipo \
 endif
 
 TP_OPTFLAGS = -O3 -xHost -fp-model fast=2 -fma -ipo \
+              -fprofile-instr-use=$(TP_PROFDATA) \
               -Dfortran3 -Dnewc $(MKL_INC)
 else
 ## NVHPC defaults: mpicc/mpicxx/mpifort wrappers
@@ -56,26 +57,18 @@ endif
 .C.o:
 	${CXX} $(CXXAPPFLAGS) -c $< $(filein) -o $@
 
-# ShellPatch.C uses OpenMP for setupintintstuff search loops
-ShellPatch.o: ShellPatch.C
-	${CXX} $(CXXAPPFLAGS) $(OMP_FLAG) -c $< $(filein) -o $@
-
 .for.o:
 	$(f77) -c $< -o $@
 
-.cu.o:
-	$(Cu) $(CUDA_APP_FLAGS) -c $< -o $@ $(CUDA_LIB_PATH)
-
-# CUDA rewrite of BSSN RHS (drop-in replacement for bssn_rhs_c + stencil helpers)
-bssn_rhs_cuda.o: bssn_rhs_cuda.cu bssn_rhs.h macrodef.h fd_cuda_helpers.cuh
-	$(Cu) $(CUDA_APP_FLAGS) -c $< -o $@ $(CUDA_LIB_PATH)
-
-# CUDA rewrite of BSSN Shell-Patch RHS (drop-in replacement for bssn_rhs_ss)
-bssn_gpu_rhs_ss.o: bssn_gpu_rhs_ss.cu bssn_gpu.h gpu_rhsSS_mem.h bssn_macro.h macrodef.fh
-	$(Cu) $(CUDA_APP_FLAGS) -c $< -o $@ $(CUDA_LIB_PATH)
+.cu.o:
+	$(Cu) $(CUDA_APP_FLAGS) -c $< -o $@ $(CUDA_LIB_PATH)
+
+# CUDA rewrite of BSSN RHS (drop-in replacement for bssn_rhs_c + stencil helpers)
+bssn_rhs_cuda.o: bssn_rhs_cuda.cu bssn_rhs.h macrodef.h
+	$(Cu) $(CUDA_APP_FLAGS) -c $< -o $@ $(CUDA_LIB_PATH)
 
 # CUDA rewrite of Z4C Cartesian RHS
-z4c_rhs_cuda.o: z4c_rhs_cuda.cu z4c_rhs_cuda.h bssn_rhs.h macrodef.h ricci_gamma.h fd_cuda_helpers.cuh
+z4c_rhs_cuda.o: z4c_rhs_cuda.cu z4c_rhs_cuda.h bssn_rhs.h macrodef.h ricci_gamma.h
 	$(Cu) $(CUDA_APP_FLAGS) -c $< -o $@ $(CUDA_LIB_PATH)
 
 # C rewrite of BSSN RHS kernel and helpers
@@ -131,7 +124,7 @@ else
 CFILES_CPU = bssn_rhs_c.o fderivs_c.o fdderivs_c.o kodiss_c.o lopsided_c.o lopsided_kodis_c.o
 endif
 
-CFILES_CUDA_BSSN = bssn_rhs_cuda.o bssn_gpu_rhs_ss.o
+CFILES_CUDA_BSSN = bssn_rhs_cuda.o
 
 ifeq ($(USE_CUDA_BSSN),1)
 CFILES = $(CFILES_CUDA_BSSN)

AMSS_NCKU_source/makefile.inc

@@ -1,7 +1,7 @@
 ## Toolchain selection
 ##   nvhpc : NVIDIA HPC SDK + CUDA-aware MPI (default)
 ##   intel : Intel oneAPI toolchain (legacy path)
-TOOLCHAIN ?= intel
+TOOLCHAIN ?= nvhpc
 
 ## PGO build mode switch (ABE only; TwoPunctureABE always uses opt flags)
 ##   opt        : (default) maximum performance with PGO profile-guided optimization
@@ -88,4 +88,4 @@ endif
 
 Cu = $(NVHPC_ROOT)/compilers/bin/nvcc
 CUDA_LIB_PATH = -L$(CUDA_HOME)/lib64 -I$(CUDA_HOME)/include
-CUDA_APP_FLAGS = -c -g -O3 --ptxas-options=-v -Dfortran3 -Dnewc -arch=$(CUDA_ARCH)
+CUDA_APP_FLAGS = -c -g -O3 --ptxas-options=-v -Dfortran3 -Dnewc -arch=$(CUDA_ARCH)

AMSS_NCKU_source/monitor.C

@@ -1,8 +1,7 @@
 
-#ifdef newc
-#include <cstdio>
-#include <sstream>
-using namespace std;
+#ifdef newc
+#include <cstdio>
+using namespace std;
 #else
 #include <stdio.h>
 #endif
@@ -78,17 +77,16 @@ monitor::monitor(const char fname[], int myrank, string head)
       parameters::str_par.insert(map<string, string>::value_type("output dir", out_dir));
     }
     // considering checkpoint run
-    string filename = out_dir + "/" + fname;
-    int i = 1;
-    while ((access(filename.c_str(), F_OK)) != -1)
-    {
-      stringstream ss;
-      ss << out_dir << "/" << i << "_" << fname;
-      filename = ss.str();
-      i++;
-    }
-
-    outfile.open(filename.c_str(), ios::trunc);
+    char filename[50];
+    sprintf(filename, "%s/%s", out_dir.c_str(), fname);
+    int i = 1;
+    while ((access(filename, F_OK)) != -1)
+    {
+      sprintf(filename, "%s/%d_%s", out_dir.c_str(), i, fname);
+      i++;
+    }
+
+    outfile.open(filename, ios::trunc);
 
     time_t tnow;
     time(&tnow);
@@ -109,17 +107,16 @@ monitor::monitor(const char fname[], int myrank, const int out_rank, string head
   if (I_Print)
   {
     // considering checkpoint run
-    string filename = out_dir + "/" + fname;
-    int i = 1;
-    while ((access(filename.c_str(), F_OK)) != -1)
-    {
-      stringstream ss;
-      ss << out_dir << "/" << i << "_" << fname;
-      filename = ss.str();
-      i++;
-    }
-
-    outfile.open(filename.c_str(), ios::trunc);
+    char filename[50];
+    sprintf(filename, "%s/%s", out_dir.c_str(), fname);
+    int i = 1;
+    while ((access(filename, F_OK)) != -1)
+    {
+      sprintf(filename, "%s/%d_%s", out_dir.c_str(), i, fname);
+      i++;
+    }
+
+    outfile.open(filename, ios::trunc);
 
     time_t tnow;
     time(&tnow);

AMSS_NCKU_source/z4c_rhs_cuda.h

@@ -53,6 +53,14 @@ int z4c_cuda_pack_state_batch_to_host_buffer(void *block_tag,
                                              int i0, int j0, int k0,
                                              int sx, int sy, int sz);
 
+int z4c_cuda_pack_state_batch_to_host_buffer_for_host_views(void *block_tag,
+                                                            double **state_host_key,
+                                                            int state_count,
+                                                            double *host_buffer,
+                                                            int *ex,
+                                                            int i0, int j0, int k0,
+                                                            int sx, int sy, int sz);
+
 int z4c_cuda_unpack_state_batch_from_host_buffer(void *block_tag,
                                                  int state_count,
                                                  double *host_buffer,
@@ -60,6 +68,14 @@ int z4c_cuda_unpack_state_batch_from_host_buffer(void *block_tag,
                                                  int i0, int j0, int k0,
                                                  int sx, int sy, int sz);
 
+int z4c_cuda_unpack_state_batch_from_host_buffer_for_host_views(void *block_tag,
+                                                                double **state_host_key,
+                                                                int state_count,
+                                                                double *host_buffer,
+                                                                int *ex,
+                                                                int i0, int j0, int k0,
+                                                                int sx, int sy, int sz);
+
 int z4c_cuda_pack_state_batch_to_device_buffer(void *block_tag,
                                                int state_count,
                                                double *device_buffer,
@@ -67,6 +83,14 @@ int z4c_cuda_pack_state_batch_to_device_buffer(void *block_tag,
                                                int i0, int j0, int k0,
                                                int sx, int sy, int sz);
 
+int z4c_cuda_pack_state_batch_to_device_buffer_for_host_views(void *block_tag,
+                                                              double **state_host_key,
+                                                              int state_count,
+                                                              double *device_buffer,
+                                                              int *ex,
+                                                              int i0, int j0, int k0,
+                                                              int sx, int sy, int sz);
+
 int z4c_cuda_unpack_state_batch_from_device_buffer(void *block_tag,
                                                    int state_count,
                                                    double *device_buffer,
@@ -74,6 +98,78 @@ int z4c_cuda_unpack_state_batch_from_device_buffer(void *block_tag,
                                                    int i0, int j0, int k0,
                                                    int sx, int sy, int sz);
 
+int z4c_cuda_unpack_state_batch_from_device_buffer_for_host_views(void *block_tag,
+                                                                  double **state_host_key,
+                                                                  int state_count,
+                                                                  double *device_buffer,
+                                                                  int *ex,
+                                                                  int i0, int j0, int k0,
+                                                                  int sx, int sy, int sz);
+
+int z4c_cuda_pack_state_segments_to_device_buffer(void *block_tag,
+                                                  int state_count,
+                                                  double *device_buffer,
+                                                  int *ex,
+                                                  int segment_count,
+                                                  const int *segment_meta);
+
+int z4c_cuda_pack_state_segments_to_device_buffer_for_host_views(void *block_tag,
+                                                                 double **state_host_key,
+                                                                 int state_count,
+                                                                 double *device_buffer,
+                                                                 int *ex,
+                                                                 int segment_count,
+                                                                 const int *segment_meta);
+
+int z4c_cuda_unpack_state_segments_from_device_buffer(void *block_tag,
+                                                      int state_count,
+                                                      double *device_buffer,
+                                                      int *ex,
+                                                      int segment_count,
+                                                      const int *segment_meta);
+
+int z4c_cuda_unpack_state_segments_from_device_buffer_for_host_views(void *block_tag,
+                                                                     double **state_host_key,
+                                                                     int state_count,
+                                                                     double *device_buffer,
+                                                                     int *ex,
+                                                                     int segment_count,
+                                                                     const int *segment_meta);
+
+int z4c_cuda_restrict_state_segments_to_device_buffer(void *block_tag,
+                                                      int state_count,
+                                                      double *device_buffer,
+                                                      int *ex,
+                                                      int segment_count,
+                                                      const int *segment_meta,
+                                                      const double *state_soa);
+
+int z4c_cuda_restrict_state_segments_to_device_buffer_for_host_views(void *block_tag,
+                                                                     double **state_host_key,
+                                                                     int state_count,
+                                                                     double *device_buffer,
+                                                                     int *ex,
+                                                                     int segment_count,
+                                                                     const int *segment_meta,
+                                                                     const double *state_soa);
+
+int z4c_cuda_prolong_state_segments_to_device_buffer(void *block_tag,
+                                                     int state_count,
+                                                     double *device_buffer,
+                                                     int *ex,
+                                                     int segment_count,
+                                                     const int *segment_meta,
+                                                     const double *state_soa);
+
+int z4c_cuda_prolong_state_segments_to_device_buffer_for_host_views(void *block_tag,
+                                                                    double **state_host_key,
+                                                                    int state_count,
+                                                                    double *device_buffer,
+                                                                    int *ex,
+                                                                    int segment_count,
+                                                                    const int *segment_meta,
+                                                                    const double *state_soa);
+
 int z4c_cuda_restrict_state_batch_to_device_buffer(void *block_tag,
                                                    int state_count,
                                                    double *device_buffer,
@@ -82,6 +178,15 @@ int z4c_cuda_restrict_state_batch_to_device_buffer(void *block_tag,
                                                    int fi0, int fj0, int fk0,
                                                    const double *state_soa);
 
+int z4c_cuda_restrict_state_batch_to_device_buffer_for_host_views(void *block_tag,
+                                                                  double **state_host_key,
+                                                                  int state_count,
+                                                                  double *device_buffer,
+                                                                  int *ex,
+                                                                  int sx, int sy, int sz,
+                                                                  int fi0, int fj0, int fk0,
+                                                                  const double *state_soa);
+
 int z4c_cuda_prolong_state_batch_to_device_buffer(void *block_tag,
                                                   int state_count,
                                                   double *device_buffer,
@@ -91,6 +196,16 @@ int z4c_cuda_prolong_state_batch_to_device_buffer(void *block_tag,
                                                   int lbc_i, int lbc_j, int lbc_k,
                                                   const double *state_soa);
 
+int z4c_cuda_prolong_state_batch_to_device_buffer_for_host_views(void *block_tag,
+                                                                 double **state_host_key,
+                                                                 int state_count,
+                                                                 double *device_buffer,
+                                                                 int *ex,
+                                                                 int sx, int sy, int sz,
+                                                                 int ii0, int jj0, int kk0,
+                                                                 int lbc_i, int lbc_j, int lbc_k,
+                                                                 const double *state_soa);
+
 int z4c_cuda_download_state_subset(void *block_tag,
                                    int *ex,
                                    int subset_count,
@@ -103,7 +218,36 @@ int z4c_cuda_upload_state_subset(void *block_tag,
                                  const int *state_indices,
                                  double **state_host_in);
 
+int z4c_cuda_compute_constraints_resident(void *block_tag,
+                                          int *ex, double *X, double *Y, double *Z,
+                                          int Symmetry, double eps, int co,
+                                          double **constraint_host_out);
+
+int z4c_cuda_interp_state_point3(void *block_tag,
+                                 int *ex,
+                                 int state0,
+                                 int state1,
+                                 int state2,
+                                 double x0,
+                                 double y0,
+                                 double z0,
+                                 double dx,
+                                 double dy,
+                                 double dz,
+                                 double px,
+                                 double py,
+                                 double pz,
+                                 int ordn,
+                                 int symmetry,
+                                 const double *soa3,
+                                 double *out3);
+
+int z4c_cuda_download_constraint_outputs(int *ex,
+                                         double **constraint_host_out);
+
 int z4c_cuda_has_resident_state(void *block_tag);
+int z4c_cuda_resident_state_matches(void *block_tag,
+                                    double **state_host_key);
 
 void z4c_cuda_release_step_ctx(void *block_tag);
 

makefile_and_run.py

@@ -138,58 +138,25 @@ def _stop_cuda_mps(runtime_env):
 
 def _gpu_runtime_env():
     runtime_env = os.environ.copy()
-    finite_difference = str(getattr(input_data, "Finite_Diffenence_Method", "4th-order")).strip()
 
     defaults = {
-        "AMSS_EVOLVE_TIMING": "1",
-        "AMSS_ESCALAR_STEP_TIMING": "0",
         "AMSS_INTERP_FAST": "1",
         "AMSS_INTERP_GPU": "1",
         "AMSS_ANALYSIS_MAP_EVERY": "1000000",
         "AMSS_CUDA_AWARE_MPI": "1",
         "AMSS_CUDA_KEEP_RESIDENT_AFTER_STEP": "1",
+        "AMSS_CUDA_Z4C_KEEP_RESIDENT_AFTER_STEP": "1",
         "AMSS_CUDA_KEEP_ALL_LEVELS": "1",
-        "AMSS_CUDA_ESCALAR_KEEP_RESIDENT_AFTER_STEP": "1",
-        "AMSS_CUDA_ESCALAR_KEEP_ALL_LEVELS": "1",
-        "AMSS_CUDA_EM_CACHE_SOURCES": "1",
-        "AMSS_CUDA_EM_ZERO_FASTPATH": "1",
-        "AMSS_EM_ZERO_ANALYSIS_FASTPATH": "1",
-        "AMSS_EM_ZERO_RESIDENT_DOWNLOAD_FASTPATH": "1",
-        "AMSS_CUDA_AMR_HOST_STAGED": "1",
-        "AMSS_CUDA_AMR_RESTRICT_DEVICE": "0",
+        "AMSS_CUDA_Z4C_AMR_DEVICE": "0",
+        "AMSS_CUDA_AMR_RESTRICT_DEVICE": "1",
         "AMSS_CUDA_AMR_RESTRICT_BATCH": "0",
         "AMSS_CUDA_DEVICE_SEGMENT_BATCH": "0",
-        "AMSS_CUDA_UNCACHED_DEVICE_BUFFERS": "1",
-        "AMSS_SHELL_FAST_INTERP": "0",
-        "AMSS_SHELL_PARALLEL_INTERP": "0",
-        "AMSS_SHELL_CUDA_INTERP": "0",
+        "AMSS_CUDA_PIN_ESCALAR_TRANSFERS": "0",
+        "AMSS_ESCALAR_GPU_RK": "0",
     }
-    if finite_difference in ("2nd-order", "8th-order"):
-        defaults.update({
-            "AMSS_INTERP_FAST": "0",
-            "AMSS_INTERP_GPU": "0",
-            "AMSS_CUDA_AWARE_MPI": "0",
-        })
-    if finite_difference == "8th-order" and getattr(input_data, "Equation_Class", "") == "BSSN-EM":
-        defaults.update({
-            "AMSS_CUDA_AMR_RESTRICT_DEVICE": "1",
-            "AMSS_CUDA_AMR_RESTRICT_BATCH": "1",
-            "AMSS_CUDA_DEVICE_SEGMENT_BATCH": "1",
-        })
-    if getattr(input_data, "basic_grid_set", "") == "Shell-Patch":
-        defaults.update({
-            "AMSS_CUDA_AWARE_MPI": "0",
-            "AMSS_SHELL_FAST_INTERP": "1",
-            "AMSS_SHELL_PARALLEL_INTERP": "1",
-            "AMSS_SHELL_INTERP_THREADS": "16",
-        })
-    if getattr(input_data, "Equation_Class", "") in ("BSSN", "BSSN-EScalar", "Z4C"):
-        defaults["AMSS_CUDA_AMR_RESTRICT_DEVICE"] = "1"
     if getattr(input_data, "Equation_Class", "") == "Z4C":
-        defaults.update({
-            "AMSS_Z4C_CUDA_RESIDENT": "1",
-            "AMSS_CONSTRAINT_OUT_EVERY": "1000000",
-        })
+        defaults["AMSS_CUDA_Z4C_KEEP_RESIDENT_AFTER_STEP"] = "0"
+        defaults["AMSS_CUDA_KEEP_ALL_LEVELS"] = "0"
     for key, value in defaults.items():
         runtime_env.setdefault(key, value)
 
@@ -301,58 +268,29 @@ def run_ABE():
     mpi_env = None
     started_mps = False
     
-    mpi_processes = int(input_data.MPI_processes)
-    if (input_data.GPU_Calculation == "yes" and
-        getattr(input_data, "Equation_Class", "") == "Z4C"):
-        z4c_env_np = os.environ.get("AMSS_Z4C_GPU_MPI_PROCESSES")
-        if z4c_env_np and int(z4c_env_np) > 0:
-            mpi_processes = int(z4c_env_np)
-        elif mpi_processes < 4:
-            mpi_processes = 4
-    if (input_data.GPU_Calculation == "yes" and
-        getattr(input_data, "basic_grid_set", "") == "Shell-Patch"):
-        shell_env_np = os.environ.get("AMSS_SHELL_GPU_MPI_PROCESSES")
-        if shell_env_np and int(shell_env_np) > 0:
-            mpi_processes = int(shell_env_np)
-        elif mpi_processes < 4:
-            mpi_processes = 4
-
     if (input_data.GPU_Calculation == "no"):
-        mpi_command         = NUMACTL_CPU_BIND + " mpirun -np " + str(mpi_processes) + " ./ABE"
+        mpi_command         = NUMACTL_CPU_BIND + " mpirun -np " + str(input_data.MPI_processes) + " ./ABE"
         #mpi_command         = " mpirun -np " + str(input_data.MPI_processes) + " ./ABE"
         mpi_command_outfile = "ABE_out.log"
     elif (input_data.GPU_Calculation == "yes"):
-        mpi_command         = NUMACTL_CPU_BIND + " I_MPI_OFFLOAD=1 I_MPI_OFFLOAD_IPC=0 mpirun -np " + str(mpi_processes) + " ./ABE_CUDA"
+        mpi_command         = NUMACTL_CPU_BIND + " mpirun -np " + str(input_data.MPI_processes) + " ./ABE_CUDA"
         mpi_command_outfile = "ABEGPU_out.log"
         mpi_env = _gpu_runtime_env()
         started_mps = _start_cuda_mps_if_requested(mpi_env)
         print(" GPU optimized runtime switches:")
-        print(f"   MPI processes={mpi_processes}")
         print(f"   AMSS_INTERP_FAST={mpi_env.get('AMSS_INTERP_FAST', '')}")
         print(f"   AMSS_INTERP_GPU={mpi_env.get('AMSS_INTERP_GPU', '')}")
         print(f"   AMSS_ANALYSIS_MAP_EVERY={mpi_env.get('AMSS_ANALYSIS_MAP_EVERY', '')}")
-        print(f"   AMSS_EVOLVE_TIMING={mpi_env.get('AMSS_EVOLVE_TIMING', '')}")
-        print(f"   AMSS_ESCALAR_STEP_TIMING={mpi_env.get('AMSS_ESCALAR_STEP_TIMING', '')}")
         print(f"   AMSS_CUDA_AWARE_MPI={mpi_env.get('AMSS_CUDA_AWARE_MPI', '')}")
         print(f"   AMSS_CUDA_KEEP_RESIDENT_AFTER_STEP={mpi_env.get('AMSS_CUDA_KEEP_RESIDENT_AFTER_STEP', '')}")
+        print(f"   AMSS_CUDA_Z4C_KEEP_RESIDENT_AFTER_STEP={mpi_env.get('AMSS_CUDA_Z4C_KEEP_RESIDENT_AFTER_STEP', '')}")
         print(f"   AMSS_CUDA_KEEP_ALL_LEVELS={mpi_env.get('AMSS_CUDA_KEEP_ALL_LEVELS', '')}")
-        print(f"   AMSS_CUDA_ESCALAR_KEEP_RESIDENT_AFTER_STEP={mpi_env.get('AMSS_CUDA_ESCALAR_KEEP_RESIDENT_AFTER_STEP', '')}")
-        print(f"   AMSS_CUDA_ESCALAR_KEEP_ALL_LEVELS={mpi_env.get('AMSS_CUDA_ESCALAR_KEEP_ALL_LEVELS', '')}")
-        print(f"   AMSS_CUDA_EM_CACHE_SOURCES={mpi_env.get('AMSS_CUDA_EM_CACHE_SOURCES', '')}")
-        print(f"   AMSS_CUDA_EM_ZERO_FASTPATH={mpi_env.get('AMSS_CUDA_EM_ZERO_FASTPATH', '')}")
-        print(f"   AMSS_EM_ZERO_ANALYSIS_FASTPATH={mpi_env.get('AMSS_EM_ZERO_ANALYSIS_FASTPATH', '')}")
-        print(f"   AMSS_EM_ZERO_RESIDENT_DOWNLOAD_FASTPATH={mpi_env.get('AMSS_EM_ZERO_RESIDENT_DOWNLOAD_FASTPATH', '')}")
-        print(f"   AMSS_CUDA_AMR_HOST_STAGED={mpi_env.get('AMSS_CUDA_AMR_HOST_STAGED', '')}")
+        print(f"   AMSS_CUDA_Z4C_AMR_DEVICE={mpi_env.get('AMSS_CUDA_Z4C_AMR_DEVICE', '')}")
         print(f"   AMSS_CUDA_AMR_RESTRICT_DEVICE={mpi_env.get('AMSS_CUDA_AMR_RESTRICT_DEVICE', '')}")
         print(f"   AMSS_CUDA_AMR_RESTRICT_BATCH={mpi_env.get('AMSS_CUDA_AMR_RESTRICT_BATCH', '')}")
         print(f"   AMSS_CUDA_DEVICE_SEGMENT_BATCH={mpi_env.get('AMSS_CUDA_DEVICE_SEGMENT_BATCH', '')}")
-        print(f"   AMSS_CUDA_UNCACHED_DEVICE_BUFFERS={mpi_env.get('AMSS_CUDA_UNCACHED_DEVICE_BUFFERS', '')}")
-        print(f"   AMSS_SHELL_FAST_INTERP={mpi_env.get('AMSS_SHELL_FAST_INTERP', '')}")
-        print(f"   AMSS_SHELL_PARALLEL_INTERP={mpi_env.get('AMSS_SHELL_PARALLEL_INTERP', '')}")
-        print(f"   AMSS_SHELL_CUDA_INTERP={mpi_env.get('AMSS_SHELL_CUDA_INTERP', '')}")
-        print(f"   AMSS_SHELL_INTERP_THREADS={mpi_env.get('AMSS_SHELL_INTERP_THREADS', '')}")
-        print(f"   AMSS_Z4C_CUDA_RESIDENT={mpi_env.get('AMSS_Z4C_CUDA_RESIDENT', '')}")
-        print(f"   AMSS_CONSTRAINT_OUT_EVERY={mpi_env.get('AMSS_CONSTRAINT_OUT_EVERY', '')}")
+        print(f"   AMSS_CUDA_PIN_ESCALAR_TRANSFERS={mpi_env.get('AMSS_CUDA_PIN_ESCALAR_TRANSFERS', '')}")
+        print(f"   AMSS_ESCALAR_GPU_RK={mpi_env.get('AMSS_ESCALAR_GPU_RK', '')}")
         if "CUDA_MPS_PIPE_DIRECTORY" in mpi_env:
             print(f"   CUDA_MPS_PIPE_DIRECTORY={mpi_env['CUDA_MPS_PIPE_DIRECTORY']}")
  

plot_xiaoqu.py

@@ -808,10 +808,10 @@ def generate_ADMmass_plot( outdir, figure_outdir, detector_number_i ):
 
 ## Plot constraint violation for each grid level
 
-def generate_constraint_check_plot( outdir, figure_outdir, input_level_number ):
-
-    # path to data file
-    file0 = os.path.join(outdir, "bssn_constraint.dat")
+def generate_constraint_check_plot( outdir, figure_outdir, input_level_number ):
+
+    # path to data file
+    file0 = os.path.join(outdir, "bssn_constraint.dat")
 
     if ( input_level_number == 0 ):
         print(                                                   )
@@ -819,26 +819,13 @@ def generate_constraint_check_plot( outdir, figure_outdir, input_level_number ):
         print(                                                          )
         print( " corresponding data file = ", file0 )
         print(                                      )
-
-    print( " Begin the constraint violation plot for grid level number =  ", input_level_number )
-
-    if (not os.path.exists(file0)) or os.path.getsize(file0) == 0:
-        if ( input_level_number == 0 ):
-            print( " Constraint data file is empty; skip constraint violation plots" )
-            print(                                                            )
-        return
-
-    # load the full data file (assumed whitespace-separated floats)
-    data = numpy.loadtxt(file0)
-    data = numpy.atleast_2d(data)
-
-    if data.shape[1] < 8:
-        if ( input_level_number == 0 ):
-            print( " Constraint data file has insufficient columns; skip constraint violation plots" )
-            print(                                                                                )
-        return
-
-    # extract columns from the constraint data file
+
+    print( " Begin the constraint violation plot for grid level number =  ", input_level_number )
+    
+    # load the full data file (assumed whitespace-separated floats)
+    data = numpy.loadtxt(file0)
+    
+    # extract columns from the constraint data file
     time          = data[:,0]
     Constraint_H  = data[:,1]
     Constraint_Px = data[:,2]