Optimize bssn_rhs.f90: Fuse loops for metric inversion and Christoffel symbols to improve cache locality

- 用 DEBUG_NAN_CHECK 宏按需启用 NaN 检查，并在输入/宏生成器中新增 Debug_NaN_Check 配置
  - 逆度量改为先求行列式再乘法展开，减少除法；并在 Gam^i/Christoffel 处提取公共子表达式
  - 预置批量 fderivs 辅助例程，便于后续矢量化/合并导数计算
  - 将默认 MPI_processes 调整为 8

  变更涉及：

  - AMSS_NCKU_source/bssn_rhs.f90
  - generate_macrodef.py
  - AMSS_NCKU_Input.py
  - AMSS_NCKU_Input_Mini.py
  - inputfile_example/AMSS_NCKU_Input.py
  - AMSS_NCKU_source/diff_new.f90

TODO: fmisc.f90 polint()

.gitignore

@@ -1,3 +1,7 @@
 __pycache__
 GW150914
+GW150914-origin
+GW150914-mini
+docs
+*.tmp
 

AMSS_NCKU_Input.py

@@ -16,12 +16,14 @@ import numpy
 File_directory   = "GW150914"                    ## output file directory
 Output_directory = "binary_output"               ## binary data file directory
                                                  ## The file directory name should not be too long
-MPI_processes    = 96                             ## number of mpi processes used in the simulation
+MPI_processes    = 8                             ## number of mpi processes used in the simulation
+
+GPU_Calculation  = "no"                          ## Use GPU or not 
+                                                 ## (prefer "no" in the current version, because the GPU part may have bugs when integrated in this Python interface)
+CPU_Part         = 1.0
+GPU_Part         = 0.0
+Debug_NaN_Check          = 0                       ## enable NaN checks in compute_rhs_bssn: 0 (off) or 1 (on)
 
-GPU_Calculation  = "yes"                         ## Use GPU or not
-                                                 ## GPU support has been updated for CUDA 13
-CPU_Part         = 0.0
-GPU_Part         = 1.0
 
 #################################################
 

AMSS_NCKU_Input_Mini.py

@@ -0,0 +1,233 @@
+
+#################################################
+##
+## This file provides the input parameters required for numerical relativity.
+## XIAOQU
+## 2024/03/19 --- 2025/09/14
+## Modified for GW150914-mini test case
+##
+#################################################
+
+import numpy    
+
+#################################################
+
+## Setting MPI processes and the output file directory
+
+File_directory   = "GW150914-mini"               ## output file directory
+Output_directory = "binary_output"               ## binary data file directory
+                                                 ## The file directory name should not be too long
+MPI_processes    = 4                             ## number of mpi processes used in the simulation (Reduced for laptop)
+
+GPU_Calculation  = "no"                          ## Use GPU or not 
+                                                 ## (prefer "no" in the current version, because the GPU part may have bugs when integrated in this Python interface)
+CPU_Part         = 1.0
+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" 
+                                                   ## 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"
+Finite_Diffenence_Method = "4th-order"             ## finite-difference method: choose "2nd-order", "4th-order", "6th-order", "8th-order"
+Debug_NaN_Check          = 0                       ## enable NaN checks in compute_rhs_bssn: 0 (off) or 1 (on)
+
+#################################################
+
+
+#################################################
+
+## Setting the time evolutionary information
+
+Start_Evolution_Time     = 0.0                    ## start evolution time t0
+Final_Evolution_Time     = 100.0                  ## final evolution time t1 (Reduced for quick test)
+Check_Time               = 10.0
+Dump_Time                = 10.0                   ## time inteval dT for dumping binary data
+D2_Dump_Time             = 10.0                   ## dump the ascii data for 2d surface after dT'
+Analysis_Time            = 1.0                    ## dump the puncture position and GW psi4 after dT"
+Evolution_Step_Number    = 10000000               ## stop the calculation after the maximal step number
+Courant_Factor           = 0.5                    ## Courant Factor
+Dissipation              = 0.15                   ## Kreiss-Oliger Dissipation Strength
+
+#################################################
+
+
+#################################################
+
+## Setting the grid structure
+
+basic_grid_set    = "Patch"                          ## grid structure: choose "Patch" or "Shell-Patch"
+grid_center_set   = "Cell"                           ## grid center: chose "Cell" or "Vertex"
+
+grid_level        = 7                                ## total number of AMR grid levels (Reduced from 9)
+static_grid_level = 4                                ## number of AMR static grid levels (Reduced from 5)
+moving_grid_level = grid_level - static_grid_level   ## number of AMR moving grid levels
+
+analysis_level    = 0
+refinement_level  = 3                                ## time refinement start from this grid level
+
+largest_box_xyz_max = [320.0, 320.0, 320.0]          ## scale of the largest box
+                                                     ## not ne cess ary to be cubic for "Patch" grid s tructure
+                                                     ## need to be a cubic box for "Shell-Patch" grid structure
+largest_box_xyz_min = - numpy.array(largest_box_xyz_max)  
+
+static_grid_number = 48                              ## grid points of each static AMR grid (in x direction) (Reduced from 96)
+                                                     ## (grid points in y and z directions are automatically adjusted)
+moving_grid_number = 24                              ## grid points of each moving AMR grid (Reduced from 48)
+shell_grid_number  = [32, 32, 100]                   ## grid points of Shell-Patch grid
+                                                     ## in (phi, theta, r) direction
+devide_factor      = 2.0                             ## resolution between different grid levels dh0/dh1, only support 2.0 now
+                                                     
+
+static_grid_type   = 'Linear'                        ## AMR static grid structure , only supports "Linear"
+moving_grid_type   = 'Linear'                        ## AMR moving grid structure , only supports "Linear"
+
+quarter_sphere_number = 48                           ## grid number of 1/4 s pher ical surface (Reduced from 96)
+                                                     ## (which is needed for evaluating the spherical surface integral)
+
+#################################################
+
+
+#################################################
+
+## Setting the puncture information
+
+puncture_number       = 2                                     
+
+position_BH           = numpy.zeros( (puncture_number, 3) )   
+parameter_BH          = numpy.zeros( (puncture_number, 3) )   
+dimensionless_spin_BH = numpy.zeros( (puncture_number, 3) )   
+momentum_BH           = numpy.zeros( (puncture_number, 3) )   
+
+puncture_data_set     = "Manually"                       ## Method to give Puncture’s positions and momentum
+                                                         ## choose "Manually" or "Automatically-BBH"
+                                                         ## Prefer to choose "Manually", because "Automatically-BBH" is developing now
+
+## initial orbital distance and ellipticity for BBHs system
+## ( needed for "Automatically-BBH" case , not affect the "Manually" case )
+Distance = 10.0
+e0       = 0.0
+
+## black hole parameter (M Q* a*)
+parameter_BH[0] = [ 36.0/(36.0+29.0),  0.0,  +0.31 ]   
+parameter_BH[1] = [ 29.0/(36.0+29.0),  0.0,  -0.46 ]  
+## dimensionless spin in each direction
+dimensionless_spin_BH[0] = [ 0.0,  0.0,  +0.31 ]   
+dimensionless_spin_BH[1] = [ 0.0,  0.0,  -0.46 ]  
+
+## use Brugmann's convention
+##  -----0-----> y
+##   -      +     
+
+#---------------------------------------------
+
+## If puncture_data_set is chosen to be "Manually", it is necessary to set the position and momentum of each puncture manually
+
+## initial position for each puncture
+position_BH[0]  = [  0.0,  10.0*29.0/(36.0+29.0), 0.0 ]  
+position_BH[1]  = [  0.0, -10.0*36.0/(36.0+29.0), 0.0 ] 
+
+## initial mumentum for each puncture
+## (needed for "Manually" case, does not affect the "Automatically-BBH" case)
+momentum_BH[0]  = [ -0.09530152296974252,  -0.00084541526517121,   0.0 ]
+momentum_BH[1]  = [ +0.09530152296974252,  +0.00084541526517121,   0.0 ]
+
+
+#################################################
+
+
+#################################################
+
+## Setting the gravitational wave information
+
+GW_L_max        = 4                      ## maximal L number in gravitational wave
+GW_M_max        = 4                      ## maximal M number in gravitational wave
+Detector_Number = 12                     ## number of dector
+Detector_Rmin   = 50.0                   ## nearest dector distance
+Detector_Rmax   = 160.0                  ## farest dector distance
+
+#################################################
+
+
+#################################################
+
+## Setting the apprent horizon
+
+AHF_Find       = "no"                    ## whether to find the apparent horizon: choose "yes" or "no"
+
+AHF_Find_Every = 24
+AHF_Dump_Time  = 20.0
+
+#################################################
+
+
+#################################################
+
+## Other parameters (testing)
+## Only influence the Equation_Class = "BSSN-EScalar" case
+
+FR_a2     = 3.0        ## f(R) = R + a2 * R^2    
+FR_l2     = 10000.0
+FR_phi0   = 0.00005
+FR_r0     = 120.0
+FR_sigma0 = 8.0
+FR_Choice = 2          ## Choice options: 1 2 3 4 5
+                       ## 1: phi(r) = phi0 * Exp(-(r-r0)**2/sigma0)   
+                       ##    V(r)   = 0
+                       ## 2: phi(r) =  phi0 * a2^2/(1+a2^2)  
+                       ##    V(r)   = Exp(-8*Sqrt(PI/3)*phi(r)) * (1-Exp(4*Sqrt(PI/3)*phi(r)))**2 / (32*PI*a2)
+                       ## 3: Schrodinger-Newton gived by system phi(r) 
+                       ##    V(r)   = Exp(-8*Sqrt(PI/3)*phi(r)) * (1-Exp(4*Sqrt(PI/3)*phi(r)))**2 / (32*PI*a2)
+                       ## 4: phi(r) = phi0 * 0.5 * ( tanh((r+r0)/sigma0) - tanh((r-r0)/sigma0) )  
+                       ##    V(r)   = 0
+                       ##    f(R)   = R + a2*R^2  with a2 = +oo
+                       ## 5: phi(r) = phi0 * Exp(-(r-r0)**2/sigma)   
+                       ##    V(r)   = 0
+
+#################################################
+
+
+#################################################
+
+## Other parameters (testing)
+## (please do not change if not necessary)
+
+boundary_choice = "BAM-choice"     ## Sommerfeld boundary condition : choose "BAM-choice" or "Shibata-choice" 
+                                   ## prefer "BAM-choice"
+
+gauge_choice  = 0                  ## gauge choice
+                                   ## 0: B^i gauge
+                                   ## 1: David's puncture gauge
+                                   ## 2: MB B^i gauge               
+                                   ## 3: RIT B^i gauge
+                                   ## 4: MB beta gauge 
+                                   ## 5: RIT beta gauge 
+                                   ## 6: MGB1 B^i gauge
+                                   ## 7: MGB2 B^i gauge
+                                   ## prefer 0 or 1
+                                   
+tetrad_type  = 2                   ## tetradtype 
+                                   ##  v:r; u: phi; w: theta
+                                   ##      v^a = (x,y,z)
+                                   ## 0: orthonormal order: v,u,w
+                                   ##    v^a = (x,y,z)   
+                                   ##    m = (phi - i theta)/sqrt(2) 
+                                   ##    following Frans, Eq.(8) of  PRD 75, 124018(2007)
+                                   ## 1: orthonormal order: w,u,v
+                                   ##    m = (theta + i phi)/sqrt(2) 
+                                   ##    following Sperhake, Eq.(3.2) of  PRD 85, 124062(2012)    
+                                   ## 2: orthonormal order: v,u,w
+                                   ##    v_a = (x,y,z)
+                                   ##    m = (phi - i theta)/sqrt(2) 
+                                   ##    following Frans, Eq.(8) of  PRD 75, 124018(2007)
+                                   ## this version recommend set to 2
+                                   ## prefer 2
+                                   
+#################################################

AMSS_NCKU_MiniProgram.py

@@ -0,0 +1,224 @@
+##################################################################
+##
+## AMSS-NCKU Numerical Relativity Mini Test Program
+## Author: Assistant (based on Xiaoqu's code)
+## 2026/01/20
+##
+## This script runs a scaled-down version of the GW150914 test case
+## suitable for laptop testing.
+##
+##################################################################
+
+import os
+import shutil
+import sys
+import time
+
+# --- Context Manager for Input File Swapping ---
+class InputFileSwapper:
+    def __init__(self, mini_file="AMSS_NCKU_Input_Mini.py", target_file="AMSS_NCKU_Input.py"):
+        self.mini_file = mini_file
+        self.target_file = target_file
+        self.backup_file = target_file + ".bak"
+        self.swapped = False
+
+    def __enter__(self):
+        print(f"[MiniProgram] Swapping {self.target_file} with {self.mini_file}...")
+        if os.path.exists(self.target_file):
+            shutil.move(self.target_file, self.backup_file)
+        shutil.copy(self.mini_file, self.target_file)
+        self.swapped = True
+        return self
+
+    def __exit__(self, exc_type, exc_value, traceback):
+        if self.swapped:
+            print(f"[MiniProgram] Restoring original {self.target_file}...")
+            os.remove(self.target_file)
+            if os.path.exists(self.backup_file):
+                shutil.move(self.backup_file, self.target_file)
+
+def main():
+    # Use the swapper to ensure all imported modules see the mini configuration
+    with InputFileSwapper():
+        
+        # Import modules AFTER swapping input file
+        try:
+            import AMSS_NCKU_Input as input_data
+            import print_information
+            import setup
+            import numerical_grid
+            import generate_macrodef
+            import makefile_and_run
+            import generate_TwoPuncture_input
+            import renew_puncture_parameter
+            import plot_xiaoqu
+            import plot_GW_strain_amplitude_xiaoqu
+        except ImportError as e:
+            print(f"Error importing modules: {e}")
+            return
+
+        print_information.print_program_introduction()
+        
+        print("\n" + "#"*60)
+        print(" RUNNING MINI TEST CASE: GW150914-mini")
+        print("#"*60 + "\n")
+
+        # --- Directory Setup ---
+        File_directory = os.path.join(input_data.File_directory)
+        
+        if os.path.exists(File_directory):
+            print(f" Output directory '{File_directory}' exists. Removing for mini test...")
+            shutil.rmtree(File_directory, ignore_errors=True)
+        
+        os.mkdir(File_directory)
+        shutil.copy("AMSS_NCKU_Input.py", File_directory) # Copies the current (mini) input
+
+        output_directory = os.path.join(File_directory, "AMSS_NCKU_output")
+        os.mkdir(output_directory)
+
+        binary_results_directory = os.path.join(output_directory, input_data.Output_directory)
+        os.mkdir(binary_results_directory)
+
+        figure_directory = os.path.join(File_directory, "figure")
+        os.mkdir(figure_directory)
+
+        print(" Output directories generated.\n")
+
+        # --- Setup and Input Generation ---
+        setup.print_input_data(File_directory)
+        setup.generate_AMSSNCKU_input()
+        setup.print_puncture_information()
+
+        print("\n Generating AMSS-NCKU input parfile...")
+        numerical_grid.append_AMSSNCKU_cgh_input()
+        
+        print("\n Plotting initial grid...")
+        numerical_grid.plot_initial_grid()
+
+        print("\n Generating macro files...")
+        generate_macrodef.generate_macrodef_h()
+        generate_macrodef.generate_macrodef_fh()
+        
+        # --- Compilation Preparation ---
+        print("\n Preparing to compile and run...")
+        
+        AMSS_NCKU_source_path = "AMSS_NCKU_source"
+        AMSS_NCKU_source_copy = os.path.join(File_directory, "AMSS_NCKU_source_copy")
+
+        if not os.path.exists(AMSS_NCKU_source_path):
+             print(" Error: AMSS_NCKU_source not found! Please run in the project root.")
+             return
+
+        shutil.copytree(AMSS_NCKU_source_path, AMSS_NCKU_source_copy)
+        
+        macrodef_h_path  = os.path.join(File_directory, "macrodef.h") 
+        macrodef_fh_path = os.path.join(File_directory, "macrodef.fh") 
+
+        shutil.copy2(macrodef_h_path,  AMSS_NCKU_source_copy)
+        shutil.copy2(macrodef_fh_path, AMSS_NCKU_source_copy)
+
+        # --- Compilation ---
+        cwd = os.getcwd()
+        os.chdir(AMSS_NCKU_source_copy)
+        
+        print(" Compiling ABE...")
+        makefile_and_run.makefile_ABE()
+        
+        if (input_data.Initial_Data_Method == "Ansorg-TwoPuncture" ): 
+            print(" Compiling TwoPunctureABE...")
+            makefile_and_run.makefile_TwoPunctureABE()
+            
+        os.chdir(cwd)
+
+        # --- Copy Executables ---
+        if (input_data.GPU_Calculation == "no"):
+            ABE_file = os.path.join(AMSS_NCKU_source_copy, "ABE")
+        else:
+            ABE_file = os.path.join(AMSS_NCKU_source_copy, "ABEGPU")
+            
+        if not os.path.exists(ABE_file):
+            print(" Error: ABE executable compilation failed.")
+            return
+            
+        shutil.copy2(ABE_file, output_directory)
+
+        TwoPuncture_file = os.path.join(AMSS_NCKU_source_copy, "TwoPunctureABE")
+        if (input_data.Initial_Data_Method == "Ansorg-TwoPuncture" ):
+            if not os.path.exists(TwoPuncture_file):
+                print(" Error: TwoPunctureABE compilation failed.")
+                return
+            shutil.copy2(TwoPuncture_file, output_directory)
+
+        # --- Execution ---
+        start_time = time.time()
+        
+        if (input_data.Initial_Data_Method == "Ansorg-TwoPuncture" ):
+             print("\n Generating TwoPuncture input...")
+             generate_TwoPuncture_input.generate_AMSSNCKU_TwoPuncture_input()
+             
+             AMSS_NCKU_TwoPuncture_inputfile = 'AMSS-NCKU-TwoPuncture.input'
+             AMSS_NCKU_TwoPuncture_inputfile_path = os.path.join( File_directory, AMSS_NCKU_TwoPuncture_inputfile )
+             shutil.copy2( AMSS_NCKU_TwoPuncture_inputfile_path, os.path.join(output_directory, 'TwoPunctureinput.par') )
+             
+             print(" Running TwoPunctureABE...")
+             os.chdir(output_directory)
+             makefile_and_run.run_TwoPunctureABE()
+             os.chdir(cwd)
+
+        # Update Puncture Parameter
+        renew_puncture_parameter.append_AMSSNCKU_BSSN_input(File_directory, output_directory)
+        
+        AMSS_NCKU_inputfile = 'AMSS-NCKU.input'
+        AMSS_NCKU_inputfile_path = os.path.join(File_directory, AMSS_NCKU_inputfile)
+        shutil.copy2( AMSS_NCKU_inputfile_path, os.path.join(output_directory, 'input.par') )
+        
+        print("\n Input files ready. Launching ABE...")
+        
+        os.chdir(output_directory)
+        makefile_and_run.run_ABE()
+        os.chdir(cwd)
+        
+        end_time = time.time()
+        elapsed_time = end_time - start_time
+        
+        # --- Post-processing ---
+        print("\n Copying output files for inspection...")
+        AMSS_NCKU_error_file_path = os.path.join(binary_results_directory, "setting.par")
+        if os.path.exists(AMSS_NCKU_error_file_path):
+            shutil.copy( AMSS_NCKU_error_file_path, os.path.join(output_directory, "AMSSNCKU_setting_parameter") )
+
+        AMSS_NCKU_error_file_path = os.path.join(binary_results_directory, "Error.log")
+        if os.path.exists(AMSS_NCKU_error_file_path):
+            shutil.copy( AMSS_NCKU_error_file_path, os.path.join(output_directory, "Error.log") )
+
+        for fname in ["bssn_BH.dat", "bssn_ADMQs.dat", "bssn_psi4.dat", "bssn_constraint.dat"]:
+            fpath = os.path.join(binary_results_directory, fname)
+            if os.path.exists(fpath):
+                shutil.copy(fpath, os.path.join(output_directory, fname))
+
+        # --- Plotting ---
+        print("\n Plotting results...")
+        try:
+            plot_xiaoqu.generate_puncture_orbit_plot(   binary_results_directory, figure_directory )
+            plot_xiaoqu.generate_puncture_orbit_plot3D( binary_results_directory, figure_directory )
+            plot_xiaoqu.generate_puncture_distence_plot( binary_results_directory, figure_directory )
+            
+            for i in range(input_data.Detector_Number):
+                plot_xiaoqu.generate_gravitational_wave_psi4_plot( binary_results_directory, figure_directory, i )
+                plot_GW_strain_amplitude_xiaoqu.generate_gravitational_wave_amplitude_plot( binary_results_directory, figure_directory, i )
+
+            for i in range(input_data.Detector_Number):
+                plot_xiaoqu.generate_ADMmass_plot( binary_results_directory, figure_directory, i )
+
+            for i in range(input_data.grid_level):
+                plot_xiaoqu.generate_constraint_check_plot( binary_results_directory, figure_directory, i )
+                
+            plot_xiaoqu.generate_binary_data_plot( binary_results_directory, figure_directory )
+        except Exception as e:
+            print(f"Warning: Plotting failed: {e}")
+
+        print(f"\n Program Cost = {elapsed_time:.2f} Seconds \n")
+        print(" AMSS-NCKU-Python simulation finished (Mini Test).\n")
+
+if __name__ == "__main__":
+    main()

AMSS_NCKU_Verify_ASC26.py

@@ -0,0 +1,279 @@
+#!/usr/bin/env python3
+"""
+AMSS-NCKU GW150914 Simulation Regression Test Script
+
+Verification Requirements:
+1. XY-plane trajectory RMS error < 1% (Optimized vs. baseline, max of BH1 and BH2)
+2. ADM constraint violation < 2 (Grid Level 0)
+
+RMS Calculation Method:
+- Computes trajectory deviation on the XY plane independently for BH1 and BH2
+- For each black hole: RMS = sqrt((1/M) * sum((Δr_i / r_i^max)^2)) × 100%
+- Final RMS = max(RMS_BH1, RMS_BH2)
+
+Usage: python3 AMSS_NCKU_Verify_ASC26.py [output_dir]
+Default: output_dir = GW150914/AMSS_NCKU_output
+Reference: GW150914-origin (baseline simulation)
+"""
+
+import numpy as np
+import sys
+import os
+
+# ANSI Color Codes
+class Color:
+    GREEN = '\033[92m'
+    RED = '\033[91m'
+    YELLOW = '\033[93m'
+    BLUE = '\033[94m'
+    BOLD = '\033[1m'
+    RESET = '\033[0m'
+
+def get_status_text(passed):
+    if passed:
+        return f"{Color.GREEN}{Color.BOLD}PASS{Color.RESET}"
+    else:
+        return f"{Color.RED}{Color.BOLD}FAIL{Color.RESET}"
+
+def load_bh_trajectory(filepath):
+    """Load black hole trajectory data"""
+    data = np.loadtxt(filepath)
+    return {
+        'time': data[:, 0],
+        'x1': data[:, 1], 'y1': data[:, 2], 'z1': data[:, 3],
+        'x2': data[:, 4], 'y2': data[:, 5], 'z2': data[:, 6]
+    }
+
+
+def load_constraint_data(filepath):
+    """Load constraint violation data"""
+    data = []
+    with open(filepath, 'r') as f:
+        for line in f:
+            if line.startswith('#'):
+                continue
+            parts = line.split()
+            if len(parts) >= 8:
+                data.append([float(x) for x in parts[:8]])
+    return np.array(data)
+
+
+def calculate_rms_error(bh_data_ref, bh_data_target):
+    """
+    Calculate trajectory-based RMS error on the XY plane between baseline and optimized simulations.
+
+    This function computes the RMS error independently for BH1 and BH2 trajectories,
+    then returns the maximum of the two as the final RMS error metric.
+
+    For each black hole, the RMS is calculated as:
+        RMS = sqrt( (1/M) * sum( (Δr_i / r_i^max)^2 ) ) × 100%
+
+    where:
+        Δr_i = sqrt((x_ref,i - x_new,i)^2 + (y_ref,i - y_new,i)^2)
+        r_i^max = max(sqrt(x_ref,i^2 + y_ref,i^2), sqrt(x_new,i^2 + y_new,i^2))
+
+    Args:
+        bh_data_ref: Reference (baseline) trajectory data
+        bh_data_target: Target (optimized) trajectory data
+
+    Returns:
+        rms_value: Final RMS error as a percentage (max of BH1 and BH2)
+        error: Error message if any
+    """
+    # Align data: truncate to the length of the shorter dataset
+    M = min(len(bh_data_ref['time']), len(bh_data_target['time']))
+
+    if M < 10:
+        return None, "Insufficient data points for comparison"
+
+    # Extract XY coordinates for both black holes
+    x1_ref = bh_data_ref['x1'][:M]
+    y1_ref = bh_data_ref['y1'][:M]
+    x2_ref = bh_data_ref['x2'][:M]
+    y2_ref = bh_data_ref['y2'][:M]
+
+    x1_new = bh_data_target['x1'][:M]
+    y1_new = bh_data_target['y1'][:M]
+    x2_new = bh_data_target['x2'][:M]
+    y2_new = bh_data_target['y2'][:M]
+
+    # Calculate RMS for BH1
+    delta_r1 = np.sqrt((x1_ref - x1_new)**2 + (y1_ref - y1_new)**2)
+    r1_ref = np.sqrt(x1_ref**2 + y1_ref**2)
+    r1_new = np.sqrt(x1_new**2 + y1_new**2)
+    r1_max = np.maximum(r1_ref, r1_new)
+
+    # Calculate RMS for BH2
+    delta_r2 = np.sqrt((x2_ref - x2_new)**2 + (y2_ref - y2_new)**2)
+    r2_ref = np.sqrt(x2_ref**2 + y2_ref**2)
+    r2_new = np.sqrt(x2_new**2 + y2_new**2)
+    r2_max = np.maximum(r2_ref, r2_new)
+
+    # Avoid division by zero for BH1
+    valid_mask1 = r1_max > 1e-15
+    if np.sum(valid_mask1) < 10:
+        return None, "Insufficient valid data points for BH1"
+
+    terms1 = (delta_r1[valid_mask1] / r1_max[valid_mask1])**2
+    rms_bh1 = np.sqrt(np.mean(terms1)) * 100
+
+    # Avoid division by zero for BH2
+    valid_mask2 = r2_max > 1e-15
+    if np.sum(valid_mask2) < 10:
+        return None, "Insufficient valid data points for BH2"
+
+    terms2 = (delta_r2[valid_mask2] / r2_max[valid_mask2])**2
+    rms_bh2 = np.sqrt(np.mean(terms2)) * 100
+
+    # Final RMS is the maximum of BH1 and BH2
+    rms_final = max(rms_bh1, rms_bh2)
+
+    return rms_final, None
+
+
+def analyze_constraint_violation(constraint_data, n_levels=9):
+    """
+    Analyze ADM constraint violation
+    Return maximum constraint violation for Grid Level 0
+    """
+    # Extract Grid Level 0 data (first entry for each time step)
+    level0_data = constraint_data[::n_levels]
+
+    # Calculate maximum absolute value for each constraint
+    results = {
+        'Ham': np.max(np.abs(level0_data[:, 1])),
+        'Px': np.max(np.abs(level0_data[:, 2])),
+        'Py': np.max(np.abs(level0_data[:, 3])),
+        'Pz': np.max(np.abs(level0_data[:, 4])),
+        'Gx': np.max(np.abs(level0_data[:, 5])),
+        'Gy': np.max(np.abs(level0_data[:, 6])),
+        'Gz': np.max(np.abs(level0_data[:, 7]))
+    }
+
+    results['max_violation'] = max(results.values())
+    return results
+
+
+def print_header():
+    """Print report header"""
+    print("\n" + Color.BLUE + Color.BOLD + "=" * 65 + Color.RESET)
+    print(Color.BOLD + "   AMSS-NCKU GW150914 Simulation Regression Test Report" + Color.RESET)
+    print(Color.BLUE + Color.BOLD + "=" * 65 + Color.RESET)
+
+
+def print_rms_results(rms_rel, error, threshold=1.0):
+    """Print RMS error results"""
+    print(f"\n{Color.BOLD}1. RMS Error Analysis (Baseline vs Optimized){Color.RESET}")
+    print("-" * 45)
+
+    if error:
+        print(f"   {Color.RED}Error: {error}{Color.RESET}")
+        return False
+
+    passed = rms_rel < threshold
+
+    print(f"   RMS relative error: {rms_rel:.4f}%")
+    print(f"   Requirement:        < {threshold}%")
+    print(f"   Status:             {get_status_text(passed)}")
+
+    return passed
+
+
+def print_constraint_results(results, threshold=2.0):
+    """Print constraint violation results"""
+    print(f"\n{Color.BOLD}2. ADM Constraint Violation Analysis (Grid Level 0){Color.RESET}")
+    print("-" * 45)
+
+    names = ['Ham', 'Px', 'Py', 'Pz', 'Gx', 'Gy', 'Gz']
+    for i, name in enumerate(names):
+        print(f"   Max |{name:3}|: {results[name]:.6f}", end="   ")
+        if (i + 1) % 2 == 0: print()
+    if len(names) % 2 != 0: print()
+
+    passed = results['max_violation'] < threshold
+    
+    print(f"\n   Maximum violation:  {results['max_violation']:.6f}")
+    print(f"   Requirement:        < {threshold}")
+    print(f"   Status:             {get_status_text(passed)}")
+
+    return passed
+
+
+def print_summary(rms_passed, constraint_passed):
+    """Print summary"""
+    print("\n" + Color.BLUE + Color.BOLD + "=" * 65 + Color.RESET)
+    print(Color.BOLD + "Verification Summary" + Color.RESET)
+    print(Color.BLUE + Color.BOLD + "=" * 65 + Color.RESET)
+
+    all_passed = rms_passed and constraint_passed
+    
+    res_rms = get_status_text(rms_passed)
+    res_con = get_status_text(constraint_passed)
+
+    print(f"   [1] RMS trajectory check:         {res_rms}")
+    print(f"   [2] ADM constraint check:         {res_con}")
+    
+    final_status = f"{Color.GREEN}{Color.BOLD}ALL CHECKS PASSED{Color.RESET}" if all_passed else f"{Color.RED}{Color.BOLD}SOME CHECKS FAILED{Color.RESET}"
+    print(f"\n   Overall result: {final_status}")
+    print(Color.BLUE + Color.BOLD + "=" * 65 + Color.RESET + "\n")
+
+    return all_passed
+
+
+def main():
+    # Determine target (optimized) output directory
+    if len(sys.argv) > 1:
+        target_dir = sys.argv[1]
+    else:
+        script_dir = os.path.dirname(os.path.abspath(__file__))
+        target_dir = os.path.join(script_dir, "GW150914/AMSS_NCKU_output")
+
+    # Determine reference (baseline) directory
+    script_dir = os.path.dirname(os.path.abspath(__file__))
+    reference_dir = os.path.join(script_dir, "GW150914-origin/AMSS_NCKU_output")
+
+    # Data file paths
+    bh_file_ref = os.path.join(reference_dir, "bssn_BH.dat")
+    bh_file_target = os.path.join(target_dir, "bssn_BH.dat")
+    constraint_file = os.path.join(target_dir, "bssn_constraint.dat")
+
+    # Check if files exist
+    if not os.path.exists(bh_file_ref):
+        print(f"{Color.RED}{Color.BOLD}Error:{Color.RESET} Baseline trajectory file not found: {bh_file_ref}")
+        sys.exit(1)
+
+    if not os.path.exists(bh_file_target):
+        print(f"{Color.RED}{Color.BOLD}Error:{Color.RESET} Target trajectory file not found: {bh_file_target}")
+        sys.exit(1)
+
+    if not os.path.exists(constraint_file):
+        print(f"{Color.RED}{Color.BOLD}Error:{Color.RESET} Constraint data file not found: {constraint_file}")
+        sys.exit(1)
+
+    # Print header
+    print_header()
+    print(f"\n{Color.BOLD}Reference (Baseline):{Color.RESET} {Color.BLUE}{reference_dir}{Color.RESET}")
+    print(f"{Color.BOLD}Target (Optimized):  {Color.RESET} {Color.BLUE}{target_dir}{Color.RESET}")
+
+    # Load data
+    bh_data_ref = load_bh_trajectory(bh_file_ref)
+    bh_data_target = load_bh_trajectory(bh_file_target)
+    constraint_data = load_constraint_data(constraint_file)
+
+    # Calculate RMS error
+    rms_rel, error = calculate_rms_error(bh_data_ref, bh_data_target)
+    rms_passed = print_rms_results(rms_rel, error)
+
+    # Analyze constraint violation
+    constraint_results = analyze_constraint_violation(constraint_data)
+    constraint_passed = print_constraint_results(constraint_results)
+
+    # Print summary
+    all_passed = print_summary(rms_passed, constraint_passed)
+
+    # Return exit code
+    sys.exit(0 if all_passed else 1)
+
+
+if __name__ == "__main__":
+    main()

AMSS_NCKU_source/FFT.f90

@@ -37,57 +37,51 @@ close(77)
 end program checkFFT
 #endif
 
+!-------------
+! Optimized FFT using Intel oneMKL DFTI
+! Mathematical equivalence: Standard DFT definition
+!   Forward (isign=1):  X[k] = sum_{n=0}^{N-1} x[n] * exp(-2*pi*i*k*n/N)
+!   Backward (isign=-1): X[k] = sum_{n=0}^{N-1} x[n] * exp(+2*pi*i*k*n/N)
+! Input/Output: dataa is interleaved complex array [Re(0),Im(0),Re(1),Im(1),...]
 !-------------
 SUBROUTINE four1(dataa,nn,isign)
+use MKL_DFTI
 implicit none
-INTEGER::isign,nn
+INTEGER, intent(in) :: isign, nn
-double precision,dimension(2*nn)::dataa
+DOUBLE PRECISION, dimension(2*nn), intent(inout) :: dataa
-INTEGER::i,istep,j,m,mmax,n
+
-double precision::tempi,tempr
+type(DFTI_DESCRIPTOR), pointer :: desc
-DOUBLE PRECISION::theta,wi,wpi,wpr,wr,wtemp
+integer :: status
-n=2*nn
+
-j=1
+! Create DFTI descriptor for 1D complex-to-complex transform
-do i=1,n,2
+status = DftiCreateDescriptor(desc, DFTI_DOUBLE, DFTI_COMPLEX, 1, nn)
-  if(j.gt.i)then
+if (status /= 0) return
-     tempr=dataa(j)
+
-     tempi=dataa(j+1)
+! Set input/output storage as interleaved complex (default)
-     dataa(j)=dataa(i)
+status = DftiSetValue(desc, DFTI_PLACEMENT, DFTI_INPLACE)
-     dataa(j+1)=dataa(i+1)
+if (status /= 0) then
-     dataa(i)=tempr
+   status = DftiFreeDescriptor(desc)
-     dataa(i+1)=tempi
+   return
-  endif
-  m=nn
-1 if ((m.ge.2).and.(j.gt.m)) then
-  j=j-m
-  m=m/2
-goto 1
-  endif
-j=j+m
-enddo
-mmax=2
-2  if (n.gt.mmax) then
-     istep=2*mmax
-     theta=6.28318530717959d0/(isign*mmax)
-     wpr=-2.d0*sin(0.5d0*theta)**2
-     wpi=sin(theta)
-     wr=1.d0
-     wi=0.d0
-     do m=1,mmax,2
-       do i=m,n,istep
-         j=i+mmax
-         tempr=sngl(wr)*dataa(j)-sngl(wi)*dataa(j+1)
-         tempi=sngl(wr)*dataa(j+1)+sngl(wi)*dataa(j)
-         dataa(j)=dataa(i)-tempr
-         dataa(j+1)=dataa(i+1)-tempi
-         dataa(i)=dataa(i)+tempr
-         dataa(i+1)=dataa(i+1)+tempi
-       enddo
-          wtemp=wr
-          wr=wr*wpr-wi*wpi+wr
-          wi=wi*wpr+wtemp*wpi+wi
-     enddo
-mmax=istep
-goto 2
 endif
+
+! Commit the descriptor
+status = DftiCommitDescriptor(desc)
+if (status /= 0) then
+   status = DftiFreeDescriptor(desc)
+   return
+endif
+
+! Execute FFT based on direction
+if (isign == 1) then
+   ! Forward FFT: exp(-2*pi*i*k*n/N)
+   status = DftiComputeForward(desc, dataa)
+else
+   ! Backward FFT: exp(+2*pi*i*k*n/N)
+   status = DftiComputeBackward(desc, dataa)
+endif
+
+! Free descriptor
+status = DftiFreeDescriptor(desc)
+
 return
 END SUBROUTINE four1

AMSS_NCKU_source/TwoPunctures.C

@@ -27,6 +27,7 @@ using namespace std;
 #endif
 
 #include "TwoPunctures.h"
+#include <mkl_cblas.h>
 
 TwoPunctures::TwoPunctures(double mp, double mm, double b,
                            double P_plusx, double P_plusy, double P_plusz,
@@ -891,25 +892,17 @@ double TwoPunctures::norm1(double *v, int n)
 /* -------------------------------------------------------------------------*/
 double TwoPunctures::norm2(double *v, int n)
 {
-  int i;
+  // Optimized with oneMKL BLAS DNRM2
-  double result = 0;
+  // Computes: sqrt(sum(v[i]^2))
-
+  return cblas_dnrm2(n, v, 1);
-  for (i = 0; i < n; i++)
-    result += v[i] * v[i];
-
-  return sqrt(result);
 }
 
 /* -------------------------------------------------------------------------*/
 double TwoPunctures::scalarproduct(double *v, double *w, int n)
 {
-  int i;
+  // Optimized with oneMKL BLAS DDOT
-  double result = 0;
+  // Computes: sum(v[i] * w[i])
-
+  return cblas_ddot(n, v, 1, w, 1);
-  for (i = 0; i < n; i++)
-    result += v[i] * w[i];
-
-  return result;
 }
 
 /* -------------------------------------------------------------------------*/

AMSS_NCKU_source/bssn_gpu.cu

@@ -18,7 +18,7 @@ using namespace std;
 #include <fstream>
 #endif
 
-static void compare_result_gpu(int ftag1,double * datac,int data_num){
+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);
@@ -83,7 +83,7 @@ inline void sub_enforce_ga(int matrix_size){
 	double * trA = M_ chin1;
 	enforce_ga<<<GRID_DIM,BLOCK_DIM>>>(trA);
 	cudaMemset(trA,0,matrix_size * sizeof(double));
-	cudaDeviceSynchronize(); 
+	cudaThreadSynchronize(); 
 	
 	//cudaMemset(Mh_ gupxx,0,matrix_size * sizeof(double));
 	//trA gxx,gyy,gzz gupxx,gupxy,gupxz,gupyy,gupyz,gupzz
@@ -273,13 +273,13 @@ __global__ void sub_symmetry_bd_partK(int ord,double * func, double * funcc,doub
 #endif //ifdef Vertex
 inline void sub_symmetry_bd(int ord,double * func, double * funcc,double * SoA){
 	sub_symmetry_bd_partF<<<GRID_DIM,BLOCK_DIM>>>(ord,func,funcc);
-	cudaDeviceSynchronize();
+	cudaThreadSynchronize();
 	sub_symmetry_bd_partI<<<GRID_DIM,BLOCK_DIM>>>(ord,func,funcc,SoA[0]);
-	cudaDeviceSynchronize();
+	cudaThreadSynchronize();
 	sub_symmetry_bd_partJ<<<GRID_DIM,BLOCK_DIM>>>(ord,func,funcc,SoA[1]);
-	cudaDeviceSynchronize();
+	cudaThreadSynchronize();
 	sub_symmetry_bd_partK<<<GRID_DIM,BLOCK_DIM>>>(ord,func,funcc,SoA[2]);
-	cudaDeviceSynchronize();
+	cudaThreadSynchronize();
 }
 
 
@@ -378,9 +378,9 @@ inline void sub_fdderivs(double * f,double *fh,double *fxx,double *fxy,double *f
 	cudaMemset(fyy,0,_3D_SIZE[0] * sizeof(double));
 	cudaMemset(fyz,0,_3D_SIZE[0] * sizeof(double));
 	cudaMemset(fzz,0,_3D_SIZE[0] * sizeof(double));
-	cudaDeviceSynchronize(); 
+	cudaThreadSynchronize(); 
 	sub_fdderivs_part1<<<GRID_DIM,BLOCK_DIM>>>(f,fh,fxx,fxy,fxz,fyy,fyz,fzz);
-	cudaDeviceSynchronize(); 
+	cudaThreadSynchronize(); 
 }
 
 __global__ void sub_fderivs_part1(double * f,double * fh,double *fx,double *fy,double *fz  )
@@ -445,9 +445,9 @@ inline void sub_fderivs(double * f,double * fh,double *fx,double *fy,double *fz,
 	cudaMemset(fy,0,_3D_SIZE[0] * sizeof(double));
 	cudaMemset(fz,0,_3D_SIZE[0] * sizeof(double));
 	
-	cudaDeviceSynchronize(); 
+	cudaThreadSynchronize(); 
 	sub_fderivs_part1<<<GRID_DIM,BLOCK_DIM>>>(f,fh,fx,fy,fz);
-	cudaDeviceSynchronize();
+	cudaThreadSynchronize();
 }
 
 __global__ void computeRicci_part1(double * dst)
@@ -465,9 +465,9 @@ __global__ void computeRicci_part1(double * dst)
  inline void computeRicci(double * src,double* dst,double * SoA, Meta* meta)
 {
 	sub_fdderivs(src,Mh_ fh,Mh_ fxx,Mh_ fxy,Mh_ fxz,Mh_ fyy,Mh_ fyz,Mh_ fzz,SoA);
-	cudaDeviceSynchronize();
+	cudaThreadSynchronize();
 	computeRicci_part1<<<GRID_DIM,BLOCK_DIM>>>(dst);
-	cudaDeviceSynchronize();
+	cudaThreadSynchronize();
 	
 }/*Exception*/
 
@@ -524,9 +524,9 @@ __global__ void sub_kodis_part1(double *f,double *fh,double *f_rhs)
 inline void sub_kodis(double *f,double *fh,double *f_rhs,double *SoA)
 {
 	sub_symmetry_bd(3,f,fh,SoA);
-	cudaDeviceSynchronize();
+	cudaThreadSynchronize();
 	sub_kodis_part1<<<GRID_DIM,BLOCK_DIM>>>(f,fh,f_rhs);
-	cudaDeviceSynchronize();
+	cudaThreadSynchronize();
 }
  
 __global__ void  sub_lopsided_part1(double *f,double* fh,double *f_rhs,double *Sfx,double *Sfy,double *Sfz)
@@ -617,9 +617,9 @@ __global__ void  sub_lopsided_part1(double *f,double* fh,double *f_rhs,double *S
 
 inline void  sub_lopsided(double *f,double*fh,double *f_rhs,double *Sfx,double *Sfy,double *Sfz,double *SoA){
 	sub_symmetry_bd(3,f,fh,SoA);
-	cudaDeviceSynchronize(); 
+	cudaThreadSynchronize(); 
 	sub_lopsided_part1<<<GRID_DIM,BLOCK_DIM>>>(f,fh,f_rhs,Sfx,Sfy,Sfz);
-	cudaDeviceSynchronize(); 
+	cudaThreadSynchronize(); 
 }
 
 __global__ void compute_rhs_bssn_part1() 
@@ -2656,13 +2656,13 @@ int gpu_rhs(int calledby, int mpi_rank, int *ex, double &T,double *X, double *Y,
 
 
 #ifdef TIMING1
-	cudaDeviceSynchronize();
+	cudaThreadSynchronize();
 	gettimeofday(&tv2, NULL);
    	cout<<"TIME USED"<<TimeBetween(tv1, tv2)<<endl; 
 #endif	
 	//cout<<"GPU meta data ready.\n";
 	
-	cudaDeviceSynchronize();
+	cudaThreadSynchronize();
 
 //--------------test constant memory address & value--------------
 /*	double rank = mpi_rank;
@@ -2685,7 +2685,7 @@ int gpu_rhs(int calledby, int mpi_rank, int *ex, double &T,double *X, double *Y,
 	//sub_enforce_ga(matrix_size);
 	//4.1-----compute rhs---------
 	compute_rhs_bssn_part1<<<GRID_DIM,BLOCK_DIM>>>();
-	cudaDeviceSynchronize();
+	cudaThreadSynchronize();
 
 	sub_fderivs(Mh_ betax,Mh_ fh,Mh_ betaxx,Mh_ betaxy,Mh_ betaxz,ass);
 	sub_fderivs(Mh_ betay,Mh_ fh,Mh_ betayx,Mh_ betayy,Mh_ betayz,sas);
@@ -2701,7 +2701,7 @@ int gpu_rhs(int calledby, int mpi_rank, int *ex, double &T,double *X, double *Y,
 	sub_fderivs(Mh_ gyz,Mh_ fh,Mh_ gyzx,Mh_ gyzy,Mh_ gyzz, saa);
   	
   	compute_rhs_bssn_part2<<<GRID_DIM,BLOCK_DIM>>>();
-	cudaDeviceSynchronize();
+	cudaThreadSynchronize();
 	
 	sub_fdderivs(Mh_ betax,Mh_ fh,Mh_ gxxx,Mh_ gxyx,Mh_ gxzx,Mh_ gyyx,Mh_ gyzx,Mh_ gzzx,ass);
 	sub_fdderivs(Mh_ betay,Mh_ fh,Mh_ gxxy,Mh_ gxyy,Mh_ gxzy,Mh_ gyyy,Mh_ gyzy,Mh_ gzzy,sas);
@@ -2711,7 +2711,7 @@ int gpu_rhs(int calledby, int mpi_rank, int *ex, double &T,double *X, double *Y,
 	sub_fderivs( Mh_ Gamz, Mh_ fh,Mh_ Gamzx, Mh_ Gamzy, Mh_ Gamzz,ssa);
 	
 	compute_rhs_bssn_part3<<<GRID_DIM,BLOCK_DIM>>>();
-	cudaDeviceSynchronize();
+	cudaThreadSynchronize();
 	
 	computeRicci(Mh_ dxx,Mh_ Rxx,sss, meta);
 	computeRicci(Mh_ dyy,Mh_ Ryy,sss, meta);
@@ -2720,20 +2720,20 @@ int gpu_rhs(int calledby, int mpi_rank, int *ex, double &T,double *X, double *Y,
 	computeRicci(Mh_ gxz,Mh_ Rxz,asa, meta);
 	computeRicci(Mh_ gyz,Mh_ Ryz,saa, meta);
 	
-	cudaDeviceSynchronize();
+	cudaThreadSynchronize();
 	
 	compute_rhs_bssn_part4<<<GRID_DIM,BLOCK_DIM>>>();
-	cudaDeviceSynchronize();
+	cudaThreadSynchronize();
 	
 	sub_fdderivs(Mh_ chi,Mh_ fh,Mh_ fxx,Mh_ fxy,Mh_ fxz,Mh_ fyy,Mh_ fyz,Mh_ fzz,sss);
 	
 	compute_rhs_bssn_part5<<<GRID_DIM,BLOCK_DIM>>>();
-	cudaDeviceSynchronize();
+	cudaThreadSynchronize();
 	
 	sub_fdderivs(Mh_ Lap,Mh_ fh,Mh_ fxx,Mh_ fxy,Mh_ fxz,Mh_ fyy,Mh_ fyz,Mh_ fzz,sss);
 	
 	compute_rhs_bssn_part6<<<GRID_DIM,BLOCK_DIM>>>();
-	cudaDeviceSynchronize();
+	cudaThreadSynchronize();
 	
 #if (GAUGE == 2 || GAUGE == 3 || GAUGE == 4 || GAUGE == 5)
 	sub_fderivs(Mh_ chi,Mh_ fh, Mh_ dtSfx_rhs, Mh_ dtSfy_rhs, Mh_ dtSfz_rhs,sss);
@@ -2805,7 +2805,7 @@ int gpu_rhs(int calledby, int mpi_rank, int *ex, double &T,double *X, double *Y,
 	
 	if(co == 0){
 		compute_rhs_bssn_part7<<<GRID_DIM,BLOCK_DIM>>>();
-		cudaDeviceSynchronize();
+		cudaThreadSynchronize();
 
 		sub_fderivs(Mh_ Axx,Mh_ fh,Mh_ gxxx,Mh_ gxxy,Mh_ gxxz,sss);
 		sub_fderivs(Mh_ Axy,Mh_ fh,Mh_ gxyx,Mh_ gxyy,Mh_ gxyz,aas);
@@ -2814,7 +2814,7 @@ int gpu_rhs(int calledby, int mpi_rank, int *ex, double &T,double *X, double *Y,
 		sub_fderivs(Mh_ Ayz,Mh_ fh,Mh_ gyzx,Mh_ gyzy,Mh_ gyzz,saa);
 		sub_fderivs(Mh_ Azz,Mh_ fh,Mh_ gzzx,Mh_ gzzy,Mh_ gzzz,sss);
 		compute_rhs_bssn_part8<<<GRID_DIM,BLOCK_DIM>>>();
-		cudaDeviceSynchronize();
+		cudaThreadSynchronize();
 	}
 
 #if (ABV == 1)
@@ -2895,7 +2895,7 @@ int gpu_rhs(int calledby, int mpi_rank, int *ex, double &T,double *X, double *Y,
 //-------------------FOR GPU TEST----------------------
 //-----------------------------------------------------
 #ifdef TIMING
-	cudaDeviceSynchronize();
+	cudaThreadSynchronize();
 	gettimeofday(&tv2, NULL);
    	cout<<"MPI rank is: "<<mpi_rank<<" GPU TIME is"<<TimeBetween(tv1, tv2)<<" (s)."<<endl; 
 #endif

AMSS_NCKU_source/bssn_gpu.h

@@ -4,17 +4,6 @@
 #include "bssn_macro.h"
 #include "macrodef.fh"
 
-// CUDA error checking macro for CUDA 13 compatibility
-#define CUDA_SAFE_CALL(call) \
-    do { \
-        cudaError_t err = call; \
-        if (err != cudaSuccess) { \
-            fprintf(stderr, "CUDA error in %s:%d: %s\n", __FILE__, __LINE__, \
-                    cudaGetErrorString(err)); \
-            exit(EXIT_FAILURE); \
-        } \
-    } while(0)
-
 #define DEVICE_ID 0
 // #define DEVICE_ID_BY_MPI_RANK
 #define GRID_DIM 256

AMSS_NCKU_source/bssn_gpu_class.C

@@ -2134,9 +2134,7 @@ void bssn_class::Evolve(int Steps)
 
       CheckPoint->write_Black_Hole_position(BH_num_input, BH_num, Porg0, Porgbr, Mass);
       CheckPoint->writecheck_cgh(PhysTime, GH);
-#ifdef WithShell
       CheckPoint->writecheck_sh(PhysTime, SH);
-#endif
       CheckPoint->write_bssn(LastDump, Last2dDump, LastAnas);
     }
   }

AMSS_NCKU_source/bssn_gpu_rhs_ss.cu

@@ -20,7 +20,7 @@ using namespace std;
 
 __device__ volatile unsigned int global_count = 0;
 
-static void compare_result_gpu(int ftag1,double * datac,int data_num){
+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);
@@ -153,11 +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){
@@ -247,13 +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]);
@@ -451,17 +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]);
@@ -472,7 +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]);
@@ -496,9 +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)
@@ -516,9 +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)
@@ -590,11 +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]);
 }
 
@@ -2287,13 +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-------------------------------------
@@ -2306,7 +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);
@@ -2322,7 +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);
@@ -2332,7 +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);
@@ -2340,25 +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);
@@ -2423,7 +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);
@@ -2432,7 +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)
@@ -2512,7 +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

AMSS_NCKU_source/bssn_step_gpu.C

@@ -1676,11 +1676,8 @@ void bssn_class::Step_GPU(int lev, int YN)
 #endif // PSTR == ?
 
 //--------------------------With Shell--------------------------
-// Note: SHStep() implementation is in bssn_gpu_class.C
 
 #ifdef WithShell
-#if 0
-// This SHStep() implementation has been moved to bssn_gpu_class.C to avoid duplicate definition
 void bssn_class::SHStep()
 {
   int lev = 0;
@@ -1941,5 +1938,5 @@ void bssn_class::SHStep()
     sPp = sPp->next;
   }
 }
-#endif // #if 0
+d
 #endif // withshell

AMSS_NCKU_source/diff_new.f90

@@ -1939,6 +1939,309 @@
   return
 
   end subroutine fddyz
+  subroutine fderivs_batch4(ex,f1,f2,f3,f4, &
+                            f1x,f1y,f1z,f2x,f2y,f2z,f3x,f3y,f3z,f4x,f4y,f4z, &
+                            X,Y,Z,SYM1,SYM2,SYM3,symmetry,onoff)
+  implicit none
+
+  integer,                               intent(in ):: ex(1:3),symmetry,onoff
+  real*8,  dimension(ex(1),ex(2),ex(3)), intent(in ):: f1,f2,f3,f4
+  real*8,  dimension(ex(1),ex(2),ex(3)), intent(out):: f1x,f1y,f1z
+  real*8,  dimension(ex(1),ex(2),ex(3)), intent(out):: f2x,f2y,f2z
+  real*8,  dimension(ex(1),ex(2),ex(3)), intent(out):: f3x,f3y,f3z
+  real*8,  dimension(ex(1),ex(2),ex(3)), intent(out):: f4x,f4y,f4z
+  real*8,                                intent(in) :: X(ex(1)),Y(ex(2)),Z(ex(3))
+  real*8,                                intent(in ):: SYM1,SYM2,SYM3
+
+!~~~~~~ other variables
+
+  real*8 :: dX,dY,dZ
+  real*8,dimension(-1:ex(1),-1:ex(2),-1:ex(3)) :: fh1,fh2,fh3,fh4
+  real*8, dimension(3) :: SoA
+  integer :: imin,jmin,kmin,imax,jmax,kmax,i,j,k
+  real*8 :: d12dx,d12dy,d12dz,d2dx,d2dy,d2dz
+  integer, parameter :: NO_SYMM = 0, EQ_SYMM = 1, OCTANT = 2
+  real*8,  parameter :: ZEO=0.d0,ONE=1.d0
+  real*8,  parameter :: TWO=2.d0,EIT=8.d0
+  real*8,  parameter :: F12=1.2d1
+
+  dX = X(2)-X(1)
+  dY = Y(2)-Y(1)
+  dZ = Z(2)-Z(1)
+
+  imax = ex(1)
+  jmax = ex(2)
+  kmax = ex(3)
+
+  imin = 1
+  jmin = 1
+  kmin = 1
+  if(Symmetry > NO_SYMM .and. dabs(Z(1)) < dZ) kmin = -1
+  if(Symmetry > EQ_SYMM .and. dabs(X(1)) < dX) imin = -1
+  if(Symmetry > EQ_SYMM .and. dabs(Y(1)) < dY) jmin = -1
+
+  SoA(1) = SYM1
+  SoA(2) = SYM2
+  SoA(3) = SYM3
+
+  call symmetry_bd(2,ex,f1,fh1,SoA)
+  call symmetry_bd(2,ex,f2,fh2,SoA)
+  call symmetry_bd(2,ex,f3,fh3,SoA)
+  call symmetry_bd(2,ex,f4,fh4,SoA)
+
+  d12dx = ONE/F12/dX
+  d12dy = ONE/F12/dY
+  d12dz = ONE/F12/dZ
+
+  d2dx = ONE/TWO/dX
+  d2dy = ONE/TWO/dY
+  d2dz = ONE/TWO/dZ
+
+  f1x = ZEO; f1y = ZEO; f1z = ZEO
+  f2x = ZEO; f2y = ZEO; f2z = ZEO
+  f3x = ZEO; f3y = ZEO; f3z = ZEO
+  f4x = ZEO; f4y = ZEO; f4z = ZEO
+
+  do k=1,ex(3)-1
+  do j=1,ex(2)-1
+  do i=1,ex(1)-1
+   if(i+2 <= imax .and. i-2 >= imin .and. &
+      j+2 <= jmax .and. j-2 >= jmin .and. &
+      k+2 <= kmax .and. k-2 >= kmin) then
+      f1x(i,j,k)=d12dx*(fh1(i-2,j,k)-EIT*fh1(i-1,j,k)+EIT*fh1(i+1,j,k)-fh1(i+2,j,k))
+      f1y(i,j,k)=d12dy*(fh1(i,j-2,k)-EIT*fh1(i,j-1,k)+EIT*fh1(i,j+1,k)-fh1(i,j+2,k))
+      f1z(i,j,k)=d12dz*(fh1(i,j,k-2)-EIT*fh1(i,j,k-1)+EIT*fh1(i,j,k+1)-fh1(i,j,k+2))
+
+      f2x(i,j,k)=d12dx*(fh2(i-2,j,k)-EIT*fh2(i-1,j,k)+EIT*fh2(i+1,j,k)-fh2(i+2,j,k))
+      f2y(i,j,k)=d12dy*(fh2(i,j-2,k)-EIT*fh2(i,j-1,k)+EIT*fh2(i,j+1,k)-fh2(i,j+2,k))
+      f2z(i,j,k)=d12dz*(fh2(i,j,k-2)-EIT*fh2(i,j,k-1)+EIT*fh2(i,j,k+1)-fh2(i,j,k+2))
+
+      f3x(i,j,k)=d12dx*(fh3(i-2,j,k)-EIT*fh3(i-1,j,k)+EIT*fh3(i+1,j,k)-fh3(i+2,j,k))
+      f3y(i,j,k)=d12dy*(fh3(i,j-2,k)-EIT*fh3(i,j-1,k)+EIT*fh3(i,j+1,k)-fh3(i,j+2,k))
+      f3z(i,j,k)=d12dz*(fh3(i,j,k-2)-EIT*fh3(i,j,k-1)+EIT*fh3(i,j,k+1)-fh3(i,j,k+2))
+
+      f4x(i,j,k)=d12dx*(fh4(i-2,j,k)-EIT*fh4(i-1,j,k)+EIT*fh4(i+1,j,k)-fh4(i+2,j,k))
+      f4y(i,j,k)=d12dy*(fh4(i,j-2,k)-EIT*fh4(i,j-1,k)+EIT*fh4(i,j+1,k)-fh4(i,j+2,k))
+      f4z(i,j,k)=d12dz*(fh4(i,j,k-2)-EIT*fh4(i,j,k-1)+EIT*fh4(i,j,k+1)-fh4(i,j,k+2))
+   elseif(i+1 <= imax .and. i-1 >= imin .and. &
+          j+1 <= jmax .and. j-1 >= jmin .and. &
+          k+1 <= kmax .and. k-1 >= kmin) then
+      f1x(i,j,k)=d2dx*(-fh1(i-1,j,k)+fh1(i+1,j,k))
+      f1y(i,j,k)=d2dy*(-fh1(i,j-1,k)+fh1(i,j+1,k))
+      f1z(i,j,k)=d2dz*(-fh1(i,j,k-1)+fh1(i,j,k+1))
+
+      f2x(i,j,k)=d2dx*(-fh2(i-1,j,k)+fh2(i+1,j,k))
+      f2y(i,j,k)=d2dy*(-fh2(i,j-1,k)+fh2(i,j+1,k))
+      f2z(i,j,k)=d2dz*(-fh2(i,j,k-1)+fh2(i,j,k+1))
+
+      f3x(i,j,k)=d2dx*(-fh3(i-1,j,k)+fh3(i+1,j,k))
+      f3y(i,j,k)=d2dy*(-fh3(i,j-1,k)+fh3(i,j+1,k))
+      f3z(i,j,k)=d2dz*(-fh3(i,j,k-1)+fh3(i,j,k+1))
+
+      f4x(i,j,k)=d2dx*(-fh4(i-1,j,k)+fh4(i+1,j,k))
+      f4y(i,j,k)=d2dy*(-fh4(i,j-1,k)+fh4(i,j+1,k))
+      f4z(i,j,k)=d2dz*(-fh4(i,j,k-1)+fh4(i,j,k+1))
+   endif
+  enddo
+  enddo
+  enddo
+
+  return
+
+  end subroutine fderivs_batch4
+!-----------------------------------------------------------------------------
+! batch first derivatives (3 fields), same symmetry setup
+!-----------------------------------------------------------------------------
+  subroutine fderivs_batch3(ex,f1,f2,f3, &
+                            f1x,f1y,f1z,f2x,f2y,f2z,f3x,f3y,f3z, &
+                            X,Y,Z,SYM1,SYM2,SYM3,symmetry,onoff)
+  implicit none
+
+  integer,                               intent(in ):: ex(1:3),symmetry,onoff
+  real*8,  dimension(ex(1),ex(2),ex(3)), intent(in ):: f1,f2,f3
+  real*8,  dimension(ex(1),ex(2),ex(3)), intent(out):: f1x,f1y,f1z
+  real*8,  dimension(ex(1),ex(2),ex(3)), intent(out):: f2x,f2y,f2z
+  real*8,  dimension(ex(1),ex(2),ex(3)), intent(out):: f3x,f3y,f3z
+  real*8,                                intent(in) :: X(ex(1)),Y(ex(2)),Z(ex(3))
+  real*8,                                intent(in ):: SYM1,SYM2,SYM3
+
+!~~~~~~ other variables
+
+  real*8 :: dX,dY,dZ
+  real*8,dimension(-1:ex(1),-1:ex(2),-1:ex(3)) :: fh1,fh2,fh3
+  real*8, dimension(3) :: SoA
+  integer :: imin,jmin,kmin,imax,jmax,kmax,i,j,k
+  real*8 :: d12dx,d12dy,d12dz,d2dx,d2dy,d2dz
+  integer, parameter :: NO_SYMM = 0, EQ_SYMM = 1, OCTANT = 2
+  real*8,  parameter :: ZEO=0.d0,ONE=1.d0
+  real*8,  parameter :: TWO=2.d0,EIT=8.d0
+  real*8,  parameter :: F12=1.2d1
+
+  dX = X(2)-X(1)
+  dY = Y(2)-Y(1)
+  dZ = Z(2)-Z(1)
+
+  imax = ex(1)
+  jmax = ex(2)
+  kmax = ex(3)
+
+  imin = 1
+  jmin = 1
+  kmin = 1
+  if(Symmetry > NO_SYMM .and. dabs(Z(1)) < dZ) kmin = -1
+  if(Symmetry > EQ_SYMM .and. dabs(X(1)) < dX) imin = -1
+  if(Symmetry > EQ_SYMM .and. dabs(Y(1)) < dY) jmin = -1
+
+  SoA(1) = SYM1
+  SoA(2) = SYM2
+  SoA(3) = SYM3
+
+  call symmetry_bd(2,ex,f1,fh1,SoA)
+  call symmetry_bd(2,ex,f2,fh2,SoA)
+  call symmetry_bd(2,ex,f3,fh3,SoA)
+
+  d12dx = ONE/F12/dX
+  d12dy = ONE/F12/dY
+  d12dz = ONE/F12/dZ
+
+  d2dx = ONE/TWO/dX
+  d2dy = ONE/TWO/dY
+  d2dz = ONE/TWO/dZ
+
+  f1x = ZEO; f1y = ZEO; f1z = ZEO
+  f2x = ZEO; f2y = ZEO; f2z = ZEO
+  f3x = ZEO; f3y = ZEO; f3z = ZEO
+
+  do k=1,ex(3)-1
+  do j=1,ex(2)-1
+  do i=1,ex(1)-1
+   if(i+2 <= imax .and. i-2 >= imin .and. &
+      j+2 <= jmax .and. j-2 >= jmin .and. &
+      k+2 <= kmax .and. k-2 >= kmin) then
+      f1x(i,j,k)=d12dx*(fh1(i-2,j,k)-EIT*fh1(i-1,j,k)+EIT*fh1(i+1,j,k)-fh1(i+2,j,k))
+      f1y(i,j,k)=d12dy*(fh1(i,j-2,k)-EIT*fh1(i,j-1,k)+EIT*fh1(i,j+1,k)-fh1(i,j+2,k))
+      f1z(i,j,k)=d12dz*(fh1(i,j,k-2)-EIT*fh1(i,j,k-1)+EIT*fh1(i,j,k+1)-fh1(i,j,k+2))
+
+      f2x(i,j,k)=d12dx*(fh2(i-2,j,k)-EIT*fh2(i-1,j,k)+EIT*fh2(i+1,j,k)-fh2(i+2,j,k))
+      f2y(i,j,k)=d12dy*(fh2(i,j-2,k)-EIT*fh2(i,j-1,k)+EIT*fh2(i,j+1,k)-fh2(i,j+2,k))
+      f2z(i,j,k)=d12dz*(fh2(i,j,k-2)-EIT*fh2(i,j,k-1)+EIT*fh2(i,j,k+1)-fh2(i,j,k+2))
+
+      f3x(i,j,k)=d12dx*(fh3(i-2,j,k)-EIT*fh3(i-1,j,k)+EIT*fh3(i+1,j,k)-fh3(i+2,j,k))
+      f3y(i,j,k)=d12dy*(fh3(i,j-2,k)-EIT*fh3(i,j-1,k)+EIT*fh3(i,j+1,k)-fh3(i,j+2,k))
+      f3z(i,j,k)=d12dz*(fh3(i,j,k-2)-EIT*fh3(i,j,k-1)+EIT*fh3(i,j,k+1)-fh3(i,j,k+2))
+   elseif(i+1 <= imax .and. i-1 >= imin .and. &
+          j+1 <= jmax .and. j-1 >= jmin .and. &
+          k+1 <= kmax .and. k-1 >= kmin) then
+      f1x(i,j,k)=d2dx*(-fh1(i-1,j,k)+fh1(i+1,j,k))
+      f1y(i,j,k)=d2dy*(-fh1(i,j-1,k)+fh1(i,j+1,k))
+      f1z(i,j,k)=d2dz*(-fh1(i,j,k-1)+fh1(i,j,k+1))
+
+      f2x(i,j,k)=d2dx*(-fh2(i-1,j,k)+fh2(i+1,j,k))
+      f2y(i,j,k)=d2dy*(-fh2(i,j-1,k)+fh2(i,j+1,k))
+      f2z(i,j,k)=d2dz*(-fh2(i,j,k-1)+fh2(i,j,k+1))
+
+      f3x(i,j,k)=d2dx*(-fh3(i-1,j,k)+fh3(i+1,j,k))
+      f3y(i,j,k)=d2dy*(-fh3(i,j-1,k)+fh3(i,j+1,k))
+      f3z(i,j,k)=d2dz*(-fh3(i,j,k-1)+fh3(i,j,k+1))
+   endif
+  enddo
+  enddo
+  enddo
+
+  return
+
+  end subroutine fderivs_batch3
+!-----------------------------------------------------------------------------
+! batch first derivatives (2 fields), same symmetry setup
+!-----------------------------------------------------------------------------
+  subroutine fderivs_batch2(ex,f1,f2, &
+                            f1x,f1y,f1z,f2x,f2y,f2z, &
+                            X,Y,Z,SYM1,SYM2,SYM3,symmetry,onoff)
+  implicit none
+
+  integer,                               intent(in ):: ex(1:3),symmetry,onoff
+  real*8,  dimension(ex(1),ex(2),ex(3)), intent(in ):: f1,f2
+  real*8,  dimension(ex(1),ex(2),ex(3)), intent(out):: f1x,f1y,f1z
+  real*8,  dimension(ex(1),ex(2),ex(3)), intent(out):: f2x,f2y,f2z
+  real*8,                                intent(in) :: X(ex(1)),Y(ex(2)),Z(ex(3))
+  real*8,                                intent(in ):: SYM1,SYM2,SYM3
+
+!~~~~~~ other variables
+
+  real*8 :: dX,dY,dZ
+  real*8,dimension(-1:ex(1),-1:ex(2),-1:ex(3)) :: fh1,fh2
+  real*8, dimension(3) :: SoA
+  integer :: imin,jmin,kmin,imax,jmax,kmax,i,j,k
+  real*8 :: d12dx,d12dy,d12dz,d2dx,d2dy,d2dz
+  integer, parameter :: NO_SYMM = 0, EQ_SYMM = 1, OCTANT = 2
+  real*8,  parameter :: ZEO=0.d0,ONE=1.d0
+  real*8,  parameter :: TWO=2.d0,EIT=8.d0
+  real*8,  parameter :: F12=1.2d1
+
+  dX = X(2)-X(1)
+  dY = Y(2)-Y(1)
+  dZ = Z(2)-Z(1)
+
+  imax = ex(1)
+  jmax = ex(2)
+  kmax = ex(3)
+
+  imin = 1
+  jmin = 1
+  kmin = 1
+  if(Symmetry > NO_SYMM .and. dabs(Z(1)) < dZ) kmin = -1
+  if(Symmetry > EQ_SYMM .and. dabs(X(1)) < dX) imin = -1
+  if(Symmetry > EQ_SYMM .and. dabs(Y(1)) < dY) jmin = -1
+
+  SoA(1) = SYM1
+  SoA(2) = SYM2
+  SoA(3) = SYM3
+
+  call symmetry_bd(2,ex,f1,fh1,SoA)
+  call symmetry_bd(2,ex,f2,fh2,SoA)
+
+  d12dx = ONE/F12/dX
+  d12dy = ONE/F12/dY
+  d12dz = ONE/F12/dZ
+
+  d2dx = ONE/TWO/dX
+  d2dy = ONE/TWO/dY
+  d2dz = ONE/TWO/dZ
+
+  f1x = ZEO; f1y = ZEO; f1z = ZEO
+  f2x = ZEO; f2y = ZEO; f2z = ZEO
+
+  do k=1,ex(3)-1
+  do j=1,ex(2)-1
+  do i=1,ex(1)-1
+   if(i+2 <= imax .and. i-2 >= imin .and. &
+      j+2 <= jmax .and. j-2 >= jmin .and. &
+      k+2 <= kmax .and. k-2 >= kmin) then
+      f1x(i,j,k)=d12dx*(fh1(i-2,j,k)-EIT*fh1(i-1,j,k)+EIT*fh1(i+1,j,k)-fh1(i+2,j,k))
+      f1y(i,j,k)=d12dy*(fh1(i,j-2,k)-EIT*fh1(i,j-1,k)+EIT*fh1(i,j+1,k)-fh1(i,j+2,k))
+      f1z(i,j,k)=d12dz*(fh1(i,j,k-2)-EIT*fh1(i,j,k-1)+EIT*fh1(i,j,k+1)-fh1(i,j,k+2))
+
+      f2x(i,j,k)=d12dx*(fh2(i-2,j,k)-EIT*fh2(i-1,j,k)+EIT*fh2(i+1,j,k)-fh2(i+2,j,k))
+      f2y(i,j,k)=d12dy*(fh2(i,j-2,k)-EIT*fh2(i,j-1,k)+EIT*fh2(i,j+1,k)-fh2(i,j+2,k))
+      f2z(i,j,k)=d12dz*(fh2(i,j,k-2)-EIT*fh2(i,j,k-1)+EIT*fh2(i,j,k+1)-fh2(i,j,k+2))
+   elseif(i+1 <= imax .and. i-1 >= imin .and. &
+          j+1 <= jmax .and. j-1 >= jmin .and. &
+          k+1 <= kmax .and. k-1 >= kmin) then
+      f1x(i,j,k)=d2dx*(-fh1(i-1,j,k)+fh1(i+1,j,k))
+      f1y(i,j,k)=d2dy*(-fh1(i,j-1,k)+fh1(i,j+1,k))
+      f1z(i,j,k)=d2dz*(-fh1(i,j,k-1)+fh1(i,j,k+1))
+
+      f2x(i,j,k)=d2dx*(-fh2(i-1,j,k)+fh2(i+1,j,k))
+      f2y(i,j,k)=d2dy*(-fh2(i,j-1,k)+fh2(i,j+1,k))
+      f2z(i,j,k)=d2dz*(-fh2(i,j,k-1)+fh2(i,j,k+1))
+   endif
+  enddo
+  enddo
+  enddo
+
+  return
+
+  end subroutine fderivs_batch2
 
 #elif (ghost_width == 4)
 ! sixth order code
@@ -2077,6 +2380,9 @@
 
   end subroutine fderivs
 !-----------------------------------------------------------------------------
+! batch first derivatives (4 fields), same symmetry setup
+!-----------------------------------------------------------------------------
+!-----------------------------------------------------------------------------
 !
 ! single derivatives dx
 !

AMSS_NCKU_source/fmisc.f90

@@ -1259,7 +1259,7 @@ end subroutine d2dump
 
   end subroutine polin3
 !--------------------------------------------------------------------------------------
-! calculate L2norm  
+! calculate L2norm
   subroutine l2normhelper(ex, X, Y, Z,xmin,ymin,zmin,xmax,ymax,zmax,&
                           f,f_out,gw)
 
@@ -1276,7 +1276,9 @@ end subroutine d2dump
   real*8            :: dX, dY, dZ
   integer::imin,jmin,kmin
   integer::imax,jmax,kmax
-  integer::i,j,k
+  integer::i,j,k,n_elements
+  real*8, dimension(:), allocatable :: f_flat
+  real*8, external :: DDOT
 
   dX = X(2) - X(1)
   dY = Y(2) - Y(1)
@@ -1300,7 +1302,12 @@ if(dabs(X(1)-xmin) < dX) imin = 1
 if(dabs(Y(1)-ymin) < dY) jmin = 1
 if(dabs(Z(1)-zmin) < dZ) kmin = 1
 
-f_out = sum(f(imin:imax,jmin:jmax,kmin:kmax)*f(imin:imax,jmin:jmax,kmin:kmax))
+! Optimized with oneMKL BLAS DDOT for dot product
+n_elements = (imax-imin+1)*(jmax-jmin+1)*(kmax-kmin+1)
+allocate(f_flat(n_elements))
+f_flat = reshape(f(imin:imax,jmin:jmax,kmin:kmax), [n_elements])
+f_out = DDOT(n_elements, f_flat, 1, f_flat, 1)
+deallocate(f_flat)
 
 f_out = f_out*dX*dY*dZ
 
@@ -1325,7 +1332,9 @@ f_out = f_out*dX*dY*dZ
   real*8            :: dX, dY, dZ
   integer::imin,jmin,kmin
   integer::imax,jmax,kmax
-  integer::i,j,k
+  integer::i,j,k,n_elements
+  real*8, dimension(:), allocatable :: f_flat
+  real*8, external :: DDOT
 
   real*8 :: PIo4
 
@@ -1388,7 +1397,12 @@ if(Symmetry==2)then
   if(dabs(ymin+gw*dY)<dY.and.Y(1)<0.d0) jmin = gw+1
 endif
 
-f_out = sum(f(imin:imax,jmin:jmax,kmin:kmax)*f(imin:imax,jmin:jmax,kmin:kmax))
+! Optimized with oneMKL BLAS DDOT for dot product
+n_elements = (imax-imin+1)*(jmax-jmin+1)*(kmax-kmin+1)
+allocate(f_flat(n_elements))
+f_flat = reshape(f(imin:imax,jmin:jmax,kmin:kmax), [n_elements])
+f_out = DDOT(n_elements, f_flat, 1, f_flat, 1)
+deallocate(f_flat)
 
 f_out = f_out*dX*dY*dZ
 
@@ -1416,6 +1430,8 @@ f_out = f_out*dX*dY*dZ
   integer::imin,jmin,kmin
   integer::imax,jmax,kmax
   integer::i,j,k
+  real*8, dimension(:), allocatable :: f_flat
+  real*8, external :: DDOT
 
   real*8 :: PIo4
 
@@ -1478,11 +1494,12 @@ if(Symmetry==2)then
   if(dabs(ymin+gw*dY)<dY.and.Y(1)<0.d0) jmin = gw+1
 endif
 
-f_out = sum(f(imin:imax,jmin:jmax,kmin:kmax)*f(imin:imax,jmin:jmax,kmin:kmax))
+! Optimized with oneMKL BLAS DDOT for dot product
-
-f_out = f_out
-
 Nout = (imax-imin+1)*(jmax-jmin+1)*(kmax-kmin+1)
+allocate(f_flat(Nout))
+f_flat = reshape(f(imin:imax,jmin:jmax,kmin:kmax), [Nout])
+f_out = DDOT(Nout, f_flat, 1, f_flat, 1)
+deallocate(f_flat)
 
   return
 
@@ -1680,6 +1697,7 @@ Nout = (imax-imin+1)*(jmax-jmin+1)*(kmax-kmin+1)
   real*8, dimension(ORDN,ORDN) :: tmp2
   real*8, dimension(ORDN) :: tmp1
   real*8, dimension(3) :: SoAh
+  real*8, external :: DDOT
 
 ! +1 because c++ gives 0 for first point
   cxB = inds+1  
@@ -1715,20 +1733,21 @@ Nout = (imax-imin+1)*(jmax-jmin+1)*(kmax-kmin+1)
      ya=fh(cxB(1):cxT(1),cxB(2):cxT(2),cxB(3):cxT(3))
   endif 
 
+  ! Optimized with BLAS operations for better performance
+  ! First dimension: z-direction weighted sum
   tmp2=0
   do m=1,ORDN
     tmp2 = tmp2 + coef(2*ORDN+m)*ya(:,:,m)
   enddo
 
+  ! Second dimension: y-direction weighted sum
   tmp1=0
   do m=1,ORDN
     tmp1 = tmp1 + coef(ORDN+m)*tmp2(:,m)
   enddo
 
-  f_int=0
+  ! Third dimension: x-direction weighted sum using BLAS DDOT
-  do m=1,ORDN
+  f_int = DDOT(ORDN, coef(1:ORDN), 1, tmp1, 1)
-    f_int = f_int + coef(m)*tmp1(m)
-  enddo
 
   return
 
@@ -1758,6 +1777,7 @@ Nout = (imax-imin+1)*(jmax-jmin+1)*(kmax-kmin+1)
   real*8, dimension(ORDN,ORDN) :: ya
   real*8, dimension(ORDN) :: tmp1
   real*8, dimension(2) :: SoAh
+  real*8, external :: DDOT
 
 ! +1 because c++ gives 0 for first point
   cxB = inds(1:2)+1  
@@ -1787,15 +1807,14 @@ Nout = (imax-imin+1)*(jmax-jmin+1)*(kmax-kmin+1)
      ya=fh(cxB(1):cxT(1),cxB(2):cxT(2),inds(3))
   endif 
 
+  ! Optimized with BLAS operations
   tmp1=0
   do m=1,ORDN
     tmp1 = tmp1 + coef(ORDN+m)*ya(:,m)
   enddo
 
-  f_int=0
+  ! Use BLAS DDOT for final weighted sum
-  do m=1,ORDN
+  f_int = DDOT(ORDN, coef(1:ORDN), 1, tmp1, 1)
-    f_int = f_int + coef(m)*tmp1(m)
-  enddo
 
   return
 
@@ -1826,6 +1845,7 @@ Nout = (imax-imin+1)*(jmax-jmin+1)*(kmax-kmin+1)
   real*8, dimension(ORDN) :: ya
   real*8 :: SoAh
   integer,dimension(3) :: inds
+  real*8, external :: DDOT
 
 ! +1 because c++ gives 0 for first point
   inds = indsi + 1
@@ -1886,10 +1906,8 @@ Nout = (imax-imin+1)*(jmax-jmin+1)*(kmax-kmin+1)
           write(*,*)"error in global_interpind1d, not recognized dumyd = ",dumyd
   endif
 
-  f_int=0
+  ! Optimized with BLAS DDOT for weighted sum
-  do m=1,ORDN
+  f_int = DDOT(ORDN, coef, 1, ya, 1)
-    f_int = f_int + coef(m)*ya(m)
-  enddo
 
   return
 
@@ -2121,24 +2139,38 @@ Nout = (imax-imin+1)*(jmax-jmin+1)*(kmax-kmin+1)
 
   end function fWigner_d_function
 !----------------------------------
+! Optimized factorial function using lookup table for small N
+! and log-gamma for large N to avoid overflow
   function ffact(N) result(gont)
   implicit none
   integer,intent(in) :: N
 
   real*8 :: gont
-
   integer :: i
 
+  ! Lookup table for factorials 0! to 20! (precomputed)
+  real*8, parameter, dimension(0:20) :: fact_table = [ &
+    1.d0, 1.d0, 2.d0, 6.d0, 24.d0, 120.d0, 720.d0, 5040.d0, 40320.d0, &
+    362880.d0, 3628800.d0, 39916800.d0, 479001600.d0, 6227020800.d0, &
+    87178291200.d0, 1307674368000.d0, 20922789888000.d0, &
+    355687428096000.d0, 6402373705728000.d0, 121645100408832000.d0, &
+    2432902008176640000.d0 ]
+
 ! sanity check
   if(N < 0)then
      write(*,*) "ffact: error input for factorial"
+     gont = 1.d0
      return
   endif
 
-  gont = 1.d0
+  ! Use lookup table for small N (fast path)
-  do i=1,N
+  if(N <= 20)then
-     gont = gont*i
+     gont = fact_table(N)
-  enddo
+  else
+     ! Use log-gamma function for large N: N! = exp(log_gamma(N+1))
+     ! This avoids overflow and is computed efficiently
+     gont = exp(log_gamma(dble(N+1)))
+  endif
 
   return
 

AMSS_NCKU_source/gaussj.C

@@ -16,115 +16,66 @@ using namespace std;
 #include <string.h>
 #include <math.h>
 #endif
-/* Linear equation solution by Gauss-Jordan elimination.
+
+// Intel oneMKL LAPACK interface
+#include <mkl_lapacke.h>
+/* Linear equation solution using Intel oneMKL LAPACK.
 a[0..n-1][0..n-1] is the input matrix. b[0..n-1] is input
 containing the right-hand side vectors. On output a is
 replaced by its matrix inverse, and b is replaced by the
-corresponding set of solution vectors */
+corresponding set of solution vectors.
+
+Mathematical equivalence:
+  Solves: A * x = b  =>  x = A^(-1) * b
+  Original Gauss-Jordan and LAPACK dgesv/dgetri produce identical results
+  within numerical precision. */
 
 int gaussj(double *a, double *b, int n)
 {
-  double swap;
+  // Allocate pivot array and workspace
+  lapack_int *ipiv = new lapack_int[n];
+  lapack_int info;
 
-  int *indxc, *indxr, *ipiv;
+  // Make a copy of matrix a for solving (dgesv modifies it to LU form)
-  indxc = new int[n];
+  double *a_copy = new double[n * n];
-  indxr = new int[n];
+  for (int i = 0; i < n * n; i++) {
-  ipiv = new int[n];
+    a_copy[i] = a[i];
-
-  int i, icol, irow, j, k, l, ll;
-  double big, dum, pivinv, temp;
-
-  for (j = 0; j < n; j++)
-    ipiv[j] = 0;
-  for (i = 0; i < n; i++)
-  {
-    big = 0.0;
-    for (j = 0; j < n; j++)
-      if (ipiv[j] != 1)
-        for (k = 0; k < n; k++)
-        {
-          if (ipiv[k] == 0)
-          {
-            if (fabs(a[j * n + k]) >= big)
-            {
-              big = fabs(a[j * n + k]);
-              irow = j;
-              icol = k;
-            }
-          }
-          else if (ipiv[k] > 1)
-          {
-            cout << "gaussj: Singular Matrix-1" << endl;
-            for (int ii = 0; ii < n; ii++)
-            {
-              for (int jj = 0; jj < n; jj++)
-                cout << a[ii * n + jj] << " ";
-              cout << endl;
-            }
-            return 1; // error return
-          }
-        }
-
-    ipiv[icol] = ipiv[icol] + 1;
-    if (irow != icol)
-    {
-      for (l = 0; l < n; l++)
-      {
-        swap = a[irow * n + l];
-        a[irow * n + l] = a[icol * n + l];
-        a[icol * n + l] = swap;
-      }
-
-      swap = b[irow];
-      b[irow] = b[icol];
-      b[icol] = swap;
-    }
-
-    indxr[i] = irow;
-    indxc[i] = icol;
-
-    if (a[icol * n + icol] == 0.0)
-    {
-      cout << "gaussj: Singular Matrix-2" << endl;
-      for (int ii = 0; ii < n; ii++)
-      {
-        for (int jj = 0; jj < n; jj++)
-          cout << a[ii * n + jj] << " ";
-        cout << endl;
-      }
-      return 1; // error return
-    }
-
-    pivinv = 1.0 / a[icol * n + icol];
-    a[icol * n + icol] = 1.0;
-    for (l = 0; l < n; l++)
-      a[icol * n + l] *= pivinv;
-    b[icol] *= pivinv;
-    for (ll = 0; ll < n; ll++)
-      if (ll != icol)
-      {
-        dum = a[ll * n + icol];
-        a[ll * n + icol] = 0.0;
-        for (l = 0; l < n; l++)
-          a[ll * n + l] -= a[icol * n + l] * dum;
-        b[ll] -= b[icol] * dum;
-      }
   }
 
-  for (l = n - 1; l >= 0; l--)
+  // Step 1: Solve linear system A*x = b using LU decomposition
-  {
+  // LAPACKE_dgesv uses column-major by default, but we use row-major
-    if (indxr[l] != indxc[l])
+  info = LAPACKE_dgesv(LAPACK_ROW_MAJOR, n, 1, a_copy, n, ipiv, b, 1);
-      for (k = 0; k < n; k++)
+
-      {
+  if (info != 0) {
-        swap = a[k * n + indxr[l]];
+    cout << "gaussj: Singular Matrix (dgesv info=" << info << ")" << endl;
-        a[k * n + indxr[l]] = a[k * n + indxc[l]];
+    delete[] ipiv;
-        a[k * n + indxc[l]] = swap;
+    delete[] a_copy;
-      }
+    return 1;
+  }
+
+  // Step 2: Compute matrix inverse A^(-1) using LU factorization
+  // First do LU factorization of original matrix a
+  info = LAPACKE_dgetrf(LAPACK_ROW_MAJOR, n, n, a, n, ipiv);
+
+  if (info != 0) {
+    cout << "gaussj: Singular Matrix (dgetrf info=" << info << ")" << endl;
+    delete[] ipiv;
+    delete[] a_copy;
+    return 1;
+  }
+
+  // Then compute inverse from LU factorization
+  info = LAPACKE_dgetri(LAPACK_ROW_MAJOR, n, a, n, ipiv);
+
+  if (info != 0) {
+    cout << "gaussj: Singular Matrix (dgetri info=" << info << ")" << endl;
+    delete[] ipiv;
+    delete[] a_copy;
+    return 1;
   }
 
-  delete[] indxc;
-  delete[] indxr;
   delete[] ipiv;
+  delete[] a_copy;
 
   return 0;
 }

AMSS_NCKU_source/ilucg.f90

@@ -512,11 +512,10 @@
       IMPLICIT DOUBLE PRECISION (A-H,O-Z)
       DIMENSION V(N),W(N)
 !     SUBROUTINE TO COMPUTE DOUBLE PRECISION VECTOR DOT PRODUCT.
+!     Optimized using Intel oneMKL BLAS ddot
+!     Mathematical equivalence: DGVV = sum_{i=1}^{N} V(i)*W(i)
 
-      SUM = 0.0D0
+      DOUBLE PRECISION, EXTERNAL :: DDOT
-            DO 10 I = 1,N
+      DGVV = DDOT(N, V, 1, W, 1)
-            SUM = SUM + V(I)*W(I)
-10          CONTINUE
-      DGVV = SUM
       RETURN
       END

AMSS_NCKU_source/macrodef.h

@@ -2,7 +2,7 @@
 #ifndef MICRODEF_H
 #define MICRODEF_H
 
-#include "microdef.fh"
+#include "macrodef.fh"
 
 // application parameters
 

AMSS_NCKU_source/makefile.inc

@@ -1,22 +1,32 @@
+## GCC version (commented out)
+## filein  = -I/usr/include -I/usr/lib/x86_64-linux-gnu/mpich/include -I/usr/lib/x86_64-linux-gnu/openmpi/lib/ -I/usr/lib/gcc/x86_64-linux-gnu/11/ -I/usr/include/c++/11/
+## filein  = -I/usr/include/ -I/usr/include/openmpi-x86_64/ -I/usr/lib/x86_64-linux-gnu/openmpi/include/ -I/usr/lib/x86_64-linux-gnu/openmpi/lib/ -I/usr/lib/gcc/x86_64-linux-gnu/11/ -I/usr/include/c++/11/
+## LDLIBS  = -L/usr/lib/x86_64-linux-gnu -L/usr/lib64 -L/usr/lib/gcc/x86_64-linux-gnu/11 -lgfortran -lmpi -lgfortran
 
-filein  = -I/usr/include/ -I/usr/lib/x86_64-linux-gnu/openmpi/include/ -I/usr/lib/x86_64-linux-gnu/openmpi/lib/ -I/usr/lib/gcc/x86_64-linux-gnu/14/ -I/usr/include/c++/14/
+## Intel oneAPI version with oneMKL (Optimized for performance)
+filein  = -I/usr/include/ -I${MKLROOT}/include
 
-##filein  = -I/usr/include/ -I/usr/lib/x86_64-linux-gnu/openmpi/include/ -I/usr/lib/x86_64-linux-gnu/openmpi/lib/ -I/usr/lib/gcc/x86_64-linux-gnu/11/ -I/usr/include/c++/11/ -I/usr/lib/cuda/include
+## Using sequential MKL (OpenMP disabled for better single-threaded performance)
+## Added -lifcore for Intel Fortran runtime and -limf for Intel math library
+LDLIBS  = -L${MKLROOT}/lib -lmkl_intel_lp64 -lmkl_sequential -lmkl_core -lifcore -limf -lpthread -lm -ldl
 
-LDLIBS  = -L/usr/lib/x86_64-linux-gnu -L/usr/lib64 -L/usr/lib/gcc/x86_64-linux-gnu/14 -lgfortran -lmpi -lgfortran -lcudart -lcuda
+## Aggressive optimization flags:
-##LDLIBS  = -L/usr/lib/x86_64-linux-gnu -L/usr/lib64 -L/usr/lib/gcc/x86_64-linux-gnu/11 -lgfortran -L/usr/lib/cuda/lib64 -lcudart -lmpi -lgfortran
+## -O3: Maximum optimization
-
+## -xHost: Optimize for the host CPU architecture (Intel/AMD compatible)
-CXXAPPFLAGS  = -O3 -Wno-deprecated -Dfortran3 -Dnewc
+## -fp-model fast=2: Aggressive floating-point optimizations
-#f90appflags = -O3 -fpp
+## -fma: Enable fused multiply-add instructions
-f90appflags  = -O3 -x f95-cpp-input
+## Note: OpenMP has been disabled (-qopenmp removed) due to performance issues
-f90          = gfortran 
+CXXAPPFLAGS  = -O3 -xHost -fp-model fast=2 -fma \
-f77          = gfortran 
+               -Dfortran3 -Dnewc -I${MKLROOT}/include
-CXX          = g++
+f90appflags  = -O3 -xHost -fp-model fast=2 -fma \
-CC           = gcc
+               -fpp -I${MKLROOT}/include
-CLINKER      = mpic++ 
+f90          = ifx
+f77          = ifx
+CXX          = icpx
+CC           = icx
+CLINKER      = mpiicpx 
 
 Cu = nvcc
 CUDA_LIB_PATH = -L/usr/lib/cuda/lib64 -I/usr/include -I/usr/lib/cuda/include
 #CUDA_APP_FLAGS = -c -g -O3 --ptxas-options=-v -arch compute_13 -code compute_13,sm_13 -Dfortran3 -Dnewc
-# RTX 4050 uses Ada Lovelace architecture (compute capability 8.9)
+CUDA_APP_FLAGS = -c -g -O3 --ptxas-options=-v -Dfortran3 -Dnewc
-CUDA_APP_FLAGS = -c -g -O3 --ptxas-options=-v -arch=sm_89 -Dfortran3 -Dnewc

generate_macrodef.py

@@ -392,6 +392,17 @@ def generate_macrodef_fh():
         print( "# Finite_Difference_Method #define ghost_width setting error!!!",   file=file1 )
         print(                                                   file=file1 )
 
+    # Define macro DEBUG_NAN_CHECK
+    # 0: off (default), 1: on
+
+    debug_nan_check = getattr(input_data, "Debug_NaN_Check", 0)
+    if debug_nan_check:
+        print( "#define DEBUG_NAN_CHECK 1", file=file1 )
+        print(                             file=file1 )
+    else:
+        print( "#define DEBUG_NAN_CHECK 0", file=file1 )
+        print(                             file=file1 )
+
     # Whether to use a shell-patch grid
     # use shell or not
 
@@ -514,6 +525,9 @@ def generate_macrodef_fh():
     print( "    6th order: 4",                                                                      file=file1 )
     print( "    8th order: 5",                                                                      file=file1 )
     print(                                                                                          file=file1 )
+    print( "define DEBUG_NAN_CHECK",                                                                file=file1 )
+    print( "    0: off (default), 1: on",                                                           file=file1 )
+    print(                                                                                          file=file1 )
     print( "define WithShell",                                                                      file=file1 )
     print( "    use shell or not",                                                                  file=file1 )
     print(                                                                                          file=file1 )

inputfile_example/AMSS_NCKU_Input.py

@@ -35,7 +35,8 @@ Equation_Class           = "BSSN"                  ## Evolution Equation: choose
                                                    ## 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"
-Finite_Diffenence_Method = "4th-order"             ## finite-difference method: choose "2nd-order", "4th-order", "6th-order", "8th-order"
+Finite_Diffenence_Method = "4th-order"             ## finite-difference method: choose "2nd-order", "4th-order", "6th-order", "8th-order"
+Debug_NaN_Check          = 0                       ## enable NaN checks in compute_rhs_bssn: 0 (off) or 1 (on)
 
 #################################################
 

makefile_and_run.py

@@ -11,6 +11,18 @@
 import AMSS_NCKU_Input as input_data
 import subprocess
 
+## CPU core binding configuration using taskset
+## taskset ensures all child processes inherit the CPU affinity mask
+## This forces make and all compiler processes to use only nohz_full cores (4-55, 60-111)
+## Format: taskset -c 4-55,60-111 ensures processes only run on these cores
+#NUMACTL_CPU_BIND = "taskset -c 4-55,60-111"
+NUMACTL_CPU_BIND = ""
+
+## Build parallelism configuration
+## Use nohz_full cores (4-55, 60-111) for compilation: 52 + 52 = 104 cores
+## Set make -j to utilize available cores for faster builds
+BUILD_JOBS = 14
+
 
 ##################################################################
 
@@ -26,11 +38,11 @@ def makefile_ABE():
     print( " Compiling the AMSS-NCKU executable file ABE/ABEGPU " ) 
     print(                                                        )
 
-    ## Build command
+    ## Build command with CPU binding to nohz_full cores
     if (input_data.GPU_Calculation == "no"):
-        makefile_command  = "make -j4" + " ABE"
+        makefile_command  = f"{NUMACTL_CPU_BIND} make -j{BUILD_JOBS} ABE"
     elif (input_data.GPU_Calculation == "yes"):
-        makefile_command  = "make -j4" + " ABEGPU"
+        makefile_command  = f"{NUMACTL_CPU_BIND} make -j{BUILD_JOBS} ABEGPU"
     else:
         print( " CPU/GPU numerical calculation setting is wrong " )
         print(                                                    )
@@ -67,8 +79,8 @@ def makefile_TwoPunctureABE():
     print( " Compiling the AMSS-NCKU executable file TwoPunctureABE " )
     print(                                                            )
     
-    ## Build command
+    ## Build command with CPU binding to nohz_full cores
-    makefile_command = "make" + " TwoPunctureABE"
+    makefile_command = f"{NUMACTL_CPU_BIND} make -j{BUILD_JOBS} TwoPunctureABE"
 
     ## Execute the command with subprocess.Popen and stream output
     makefile_process = subprocess.Popen(makefile_command, shell=True, stdout=subprocess.PIPE, stderr=subprocess.STDOUT, text=True) 
@@ -105,10 +117,10 @@ def run_ABE():
     ## Define the command to run; cast other values to strings as needed
     
     if (input_data.GPU_Calculation == "no"):
-        mpi_command         = "mpirun -np " + str(input_data.MPI_processes) + " ./ABE"
+        mpi_command         = NUMACTL_CPU_BIND + " mpirun -np " + str(input_data.MPI_processes) + " ./ABE"
         mpi_command_outfile = "ABE_out.log"
     elif (input_data.GPU_Calculation == "yes"):
-        mpi_command         = "mpirun -np " + str(input_data.MPI_processes) + " ./ABEGPU"
+        mpi_command         = NUMACTL_CPU_BIND + " mpirun -np " + str(input_data.MPI_processes) + " ./ABEGPU"
         mpi_command_outfile = "ABEGPU_out.log"
  
     ## Execute the MPI command and stream output
@@ -147,7 +159,7 @@ def run_TwoPunctureABE():
     print(                                                          )
     
     ## Define the command to run
-    TwoPuncture_command         = "./TwoPunctureABE"
+    TwoPuncture_command         = NUMACTL_CPU_BIND + " ./TwoPunctureABE"
     TwoPuncture_command_outfile = "TwoPunctureABE_out.log"
 
     ## Execute the command with subprocess.Popen and stream output