尝试划分4block但是效果不好，转为研究访存

Rank [  20]: Calc 0.000003 s

Rank [  35]: Calc 0.000003 s

Rank [  10]: Calc 0.000010 s

Rank [  17]: Calc 0.000005 s

Rank [  32]: Calc 0.000003 s，而且rank不固定，一般就是rank 10 和 rank 52；
但尽管有很多，比前者时间还是少很多
对于第二个Surf_Wave 调用 Interp_points，我发现前四个rank时间最长，比较固定，就是下面四个rank

Rank [  27]: Calc 0.331978 s

Rank [  35]: Calc 0.242219 s

Rank [  28]: Calc 0.242132 s

Rank [  36]: Calc 0.197024 s
因此下面surf_wave是核心

AMSS_NCKU_Program.py

@@ -66,7 +66,8 @@ if os.path.exists(File_directory):
     ## Prompt whether to overwrite the existing directory
     while True:
         try:
-            inputvalue = input()
+            ## inputvalue = input()
+            inputvalue = "continue"
             ## If the user agrees to overwrite, proceed and remove the existing directory
             if ( inputvalue == "continue" ):
                 print( " Continue the calculation !!! " )
@@ -270,12 +271,6 @@ if not os.path.exists( ABE_file ):
 ## Copy the executable ABE (or ABEGPU) into the run directory
 shutil.copy2(ABE_file, output_directory)
 
-## Copy interp load balance profile if present (for optimize pass)
-interp_lb_profile = os.path.join(AMSS_NCKU_source_copy, "interp_lb_profile.bin")
-if os.path.exists(interp_lb_profile):
-    shutil.copy2(interp_lb_profile, output_directory)
-    print( " Copied interp_lb_profile.bin to run directory " )
-
 ###########################
 
 ## If the initial-data method is TwoPuncture, copy the TwoPunctureABE executable to the run directory

AMSS_NCKU_Verify_ASC26.py

@@ -1,13 +1,9 @@
 #!/usr/bin/env python3
 """
-AMSS-NCKU GW150914 Simulation Regression Test Script (Comprehensive Version)
+AMSS-NCKU GW150914 Simulation Regression Test Script
 
 Verification Requirements:
-1. RMS errors < 1% for:
+1. XY-plane trajectory RMS error < 1% (Optimized vs. baseline, max of BH1 and BH2)
-   - 3D Vector Total RMS
-   - X Component RMS
-   - Y Component RMS
-   - Z Component RMS
 2. ADM constraint violation < 2 (Grid Level 0)
 
 RMS Calculation Method:
@@ -61,62 +57,79 @@ def load_constraint_data(filepath):
                 data.append([float(x) for x in parts[:8]])
     return np.array(data)
 
-def calculate_all_rms_errors(bh_data_ref, bh_data_target):
+
+def calculate_rms_error(bh_data_ref, bh_data_target):
     """
-    Calculate 3D Vector RMS and component-wise RMS (X, Y, Z) independently.
+    Calculate trajectory-based RMS error on the XY plane between baseline and optimized simulations.
-    Uses r = sqrt(x^2 + y^2) as the denominator for all error normalizations.
+
-    Returns the maximum error between BH1 and BH2 for each category.
+    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"
 
-    results = {}
+    # 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]
 
-    for bh in ['1', '2']:
+    x1_new = bh_data_target['x1'][:M]
-        x_r, y_r, z_r = bh_data_ref[f'x{bh}'][:M], bh_data_ref[f'y{bh}'][:M], bh_data_ref[f'z{bh}'][:M]
+    y1_new = bh_data_target['y1'][:M]
-        x_n, y_n, z_n = bh_data_target[f'x{bh}'][:M], bh_data_target[f'y{bh}'][:M], bh_data_target[f'z{bh}'][:M]
+    x2_new = bh_data_target['x2'][:M]
+    y2_new = bh_data_target['y2'][:M]
 
-        # 核心修改：根据组委会的邮件指示，分母统一使用 r = sqrt(x^2 + y^2)
+    # Calculate RMS for BH1
-        r_ref = np.sqrt(x_r**2 + y_r**2)
+    delta_r1 = np.sqrt((x1_ref - x1_new)**2 + (y1_ref - y1_new)**2)
-        r_new = np.sqrt(x_n**2 + y_n**2)
+    r1_ref = np.sqrt(x1_ref**2 + y1_ref**2)
-        denom_max = np.maximum(r_ref, r_new)
+    r1_new = np.sqrt(x1_new**2 + y1_new**2)
+    r1_max = np.maximum(r1_ref, r1_new)
 
-        valid = denom_max > 1e-15
+    # Calculate RMS for BH2
-        if np.sum(valid) < 10:
+    delta_r2 = np.sqrt((x2_ref - x2_new)**2 + (y2_ref - y2_new)**2)
-            results[f'BH{bh}'] = { '3D_Vector': 0.0, 'X_Component': 0.0, 'Y_Component': 0.0, 'Z_Component': 0.0 }
+    r2_ref = np.sqrt(x2_ref**2 + y2_ref**2)
-            continue
+    r2_new = np.sqrt(x2_new**2 + y2_new**2)
+    r2_max = np.maximum(r2_ref, r2_new)
 
-        def calc_rms(delta):
+    # Avoid division by zero for BH1
-            # 将对应分量的偏差除以统一的轨道半径分母 denom_max
+    valid_mask1 = r1_max > 1e-15
-            return np.sqrt(np.mean((delta[valid] / denom_max[valid])**2)) * 100
+    if np.sum(valid_mask1) < 10:
+        return None, "Insufficient valid data points for BH1"
 
-        # 1. Total 3D Vector RMS
+    terms1 = (delta_r1[valid_mask1] / r1_max[valid_mask1])**2
-        delta_vec = np.sqrt((x_r - x_n)**2 + (y_r - y_n)**2 + (z_r - z_n)**2)
+    rms_bh1 = np.sqrt(np.mean(terms1)) * 100
-        rms_3d = calc_rms(delta_vec)
 
-        # 2. Component-wise RMS (分离计算各轴，但共用半径分母)
+    # Avoid division by zero for BH2
-        rms_x = calc_rms(np.abs(x_r - x_n))
+    valid_mask2 = r2_max > 1e-15
-        rms_y = calc_rms(np.abs(y_r - y_n))
+    if np.sum(valid_mask2) < 10:
-        rms_z = calc_rms(np.abs(z_r - z_n))
+        return None, "Insufficient valid data points for BH2"
 
-        results[f'BH{bh}'] = {
+    terms2 = (delta_r2[valid_mask2] / r2_max[valid_mask2])**2
-            '3D_Vector': rms_3d,
+    rms_bh2 = np.sqrt(np.mean(terms2)) * 100
-            'X_Component': rms_x,
-            'Y_Component': rms_y,
-            'Z_Component': rms_z
-        }
 
-    # 获取 BH1 和 BH2 中的最大误差
+    # Final RMS is the maximum of BH1 and BH2
-    max_rms = {
+    rms_final = max(rms_bh1, rms_bh2)
-        '3D_Vector': max(results['BH1']['3D_Vector'], results['BH2']['3D_Vector']),
+
-        'X_Component': max(results['BH1']['X_Component'], results['BH2']['X_Component']),
+    return rms_final, None
-        'Y_Component': max(results['BH1']['Y_Component'], results['BH2']['Y_Component']),
-        'Z_Component': max(results['BH1']['Z_Component'], results['BH2']['Z_Component'])
-    }
 
-    return max_rms, None
 
 def analyze_constraint_violation(constraint_data, n_levels=9):
     """
@@ -142,32 +155,34 @@ def analyze_constraint_violation(constraint_data, n_levels=9):
 
 
 def print_header():
+    """Print report header"""
     print("\n" + Color.BLUE + Color.BOLD + "=" * 65 + Color.RESET)
-    print(Color.BOLD + "   AMSS-NCKU GW150914 Comprehensive Regression Test" + 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_dict, error, threshold=1.0):
+
-    print(f"\n{Color.BOLD}1. RMS Error Analysis (Maximums of BH1 & BH2){Color.RESET}")
+def print_rms_results(rms_rel, error, threshold=1.0):
-    print("-" * 65)
+    """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
 
-    all_passed = True
+    passed = rms_rel < threshold
-    print(f"   Requirement: < {threshold}%\n")
 
-    for key, val in rms_dict.items():
+    print(f"   RMS relative error: {rms_rel:.4f}%")
-        passed = val < threshold
+    print(f"   Requirement:        < {threshold}%")
-        all_passed = all_passed and passed
+    print(f"   Status:             {get_status_text(passed)}")
-        status = get_status_text(passed)
+
-        print(f"   {key:15}: {val:8.4f}%   |   Status: {status}")
+    return passed
 
-    return all_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("-" * 65)
+    print("-" * 45)
 
     names = ['Ham', 'Px', 'Py', 'Pz', 'Gx', 'Gy', 'Gz']
     for i, name in enumerate(names):
@@ -185,6 +200,7 @@ def print_constraint_results(results, threshold=2.0):
 
 
 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)
@@ -194,7 +210,7 @@ def print_summary(rms_passed, constraint_passed):
     res_rms = get_status_text(rms_passed)
     res_con = get_status_text(constraint_passed)
 
-    print(f"   [1] Comprehensive RMS check:      {res_rms}")
+    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}"
@@ -203,48 +219,61 @@ def print_summary(rms_passed, constraint_passed):
 
     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)
 
-    # Output modified RMS results
+    # Calculate RMS error
-    rms_dict, error = calculate_all_rms_errors(bh_data_ref, bh_data_target)
+    rms_rel, error = calculate_rms_error(bh_data_ref, bh_data_target)
-    rms_passed = print_rms_results(rms_dict, error)
+    rms_passed = print_rms_results(rms_rel, error)
 
-    # Output constraint results
+    # 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/ABE.C

@@ -24,16 +24,18 @@ using namespace std;
 
 #include "misc.h"
 #include "macrodef.h"
-#ifdef USE_GPU
-extern void bssn_cuda_dump_stage_profile();
-#endif
 
 #ifndef ABEtype
 #error "not define ABEtype"
 #endif
 
 #if (ABEtype == 0)
+
+#ifdef USE_GPU
+#include "bssn_gpu_class.h"
+#else
 #include "bssn_class.h"
+#endif
 
 #elif (ABEtype == 1)
 #include "bssnEScalar_class.h"
@@ -472,13 +474,10 @@ int main(int argc, char *argv[])
             cout << endl;
       }
 
-	      ADM->Evolve(Steps);
+      ADM->Evolve(Steps);
-#ifdef USE_GPU
-	      bssn_cuda_dump_stage_profile();
-#endif
 
-	      if (myrank == 0)
+      if (myrank == 0)
-	      {
+      {
             cout << endl;
             cout << " Total Evolve Time: "  << MPI_Wtime() - End_clock   << " seconds!" << endl;
             cout << " Total Running Time: " << MPI_Wtime() - Begin_clock << " seconds!" << endl;

AMSS_NCKU_source/Block.C

@@ -11,10 +11,6 @@ using namespace std;
 
 #include "Block.h"
 #include "misc.h"
-#ifdef USE_GPU
-#include "bssn_gpu.h"
-#include "bssn_cuda_ops.h"
-#endif
 
 Block::Block(int DIM, int *shapei, double *bboxi, int ranki, int ingfsi, int fngfsi, int levi, const int cgpui) : rank(ranki), ingfs(ingfsi), fngfs(fngfsi), lev(levi), cgpu(cgpui)
 {
@@ -105,11 +101,6 @@ Block::~Block()
   MPI_Comm_rank(MPI_COMM_WORLD, &myrank);
   if (myrank == rank)
   {
-#ifdef USE_GPU
-    bssn_gpu_clear_cached_device_buffers();
-    bssn_cuda_release_rk4_caches();
-    bssn_cuda_release_interp_caches();
-#endif
     for (int i = 0; i < dim; i++)
       delete[] X[i];
     for (int i = 0; i < ingfs; i++)

AMSS_NCKU_source/MPatch.C

@@ -7,550 +7,12 @@
 #include <string>
 #include <cmath>
 #include <new>
-#include <map>
-#include <vector>
 using namespace std;
 
 #include "misc.h"
 #include "MPatch.h"
 #include "Parallel.h"
 #include "fmisc.h"
-#include "bssn_cuda_ops.h"
-#ifdef INTERP_LB_PROFILE
-#include "interp_lb_profile.h"
-#endif
-
-#if defined(__GNUC__) || defined(__clang__)
-extern int bssn_cuda_interp_points_batch(const int *ex,
-                                         const double *X, const double *Y, const double *Z,
-                                         const double *const *fields,
-                                         const double *soa_flat,
-                                         int num_var,
-                                         const double *px, const double *py, const double *pz,
-                                         int num_points,
-                                         int ordn,
-                                         int symmetry,
-                                         double *out) __attribute__((weak));
-#endif
-
-namespace
-{
-struct InterpVarDesc
-{
-  int sgfn;
-  double soa[dim];
-};
-
-struct InterpPlanKey
-{
-  const Patch *patch;
-  const double *x;
-  const double *y;
-  const double *z;
-  int NN;
-  int Symmetry;
-  int myrank;
-};
-
-struct InterpPlanKeyLess
-{
-  bool operator()(const InterpPlanKey &lhs, const InterpPlanKey &rhs) const
-  {
-    if (lhs.patch != rhs.patch) return lhs.patch < rhs.patch;
-    if (lhs.x != rhs.x) return lhs.x < rhs.x;
-    if (lhs.y != rhs.y) return lhs.y < rhs.y;
-    if (lhs.z != rhs.z) return lhs.z < rhs.z;
-    if (lhs.NN != rhs.NN) return lhs.NN < rhs.NN;
-    if (lhs.Symmetry != rhs.Symmetry) return lhs.Symmetry < rhs.Symmetry;
-    return lhs.myrank < rhs.myrank;
-  }
-};
-
-struct CachedInterpPlan
-{
-  int nblocks;
-  vector<int> owner_rank;
-  vector<int> owner_block;
-  vector<vector<int> > block_points;
-  vector<vector<double> > block_px;
-  vector<vector<double> > block_py;
-  vector<vector<double> > block_pz;
-
-  CachedInterpPlan() : nblocks(0) {}
-};
-
-struct CachedInterpPlanEntry
-{
-  bool valid;
-  InterpPlanKey key;
-  vector<double> xvals;
-  vector<double> yvals;
-  vector<double> zvals;
-  CachedInterpPlan plan;
-
-  CachedInterpPlanEntry() : valid(false) {}
-};
-
-struct InterpBlockView
-{
-  Block *bp;
-  double llb[dim];
-  double uub[dim];
-};
-
-struct BlockBinIndex
-{
-  int bins[dim];
-  double lo[dim];
-  double inv[dim];
-  vector<InterpBlockView> views;
-  vector<vector<int>> bin_to_blocks;
-  bool valid;
-
-  BlockBinIndex() : valid(false)
-  {
-    for (int i = 0; i < dim; i++)
-    {
-      bins[i] = 1;
-      lo[i] = 0.0;
-      inv[i] = 0.0;
-    }
-  }
-};
-
-inline int clamp_int(int v, int lo, int hi)
-{
-  return (v < lo) ? lo : ((v > hi) ? hi : v);
-}
-
-inline int coord_to_bin(double x, double lo, double inv, int nb)
-{
-  if (nb <= 1 || inv <= 0.0)
-    return 0;
-  int b = int(floor((x - lo) * inv));
-  return clamp_int(b, 0, nb - 1);
-}
-
-inline int bin_loc(const BlockBinIndex &index, int b0, int b1, int b2)
-{
-  return b0 + index.bins[0] * (b1 + index.bins[1] * b2);
-}
-
-inline bool point_in_block_view(const InterpBlockView &view, const double *pox, const double *DH)
-{
-  for (int i = 0; i < dim; i++)
-  {
-    if (pox[i] - view.llb[i] < -DH[i] / 2 || pox[i] - view.uub[i] > DH[i] / 2)
-      return false;
-  }
-  return true;
-}
-
-void build_block_bin_index(Patch *patch, const double *DH, BlockBinIndex &index)
-{
-  index = BlockBinIndex();
-
-  MyList<Block> *Bp = patch->blb;
-  while (Bp)
-  {
-    Block *BP = Bp->data;
-    InterpBlockView view;
-    view.bp = BP;
-    for (int i = 0; i < dim; i++)
-    {
-#ifdef Vertex
-#ifdef Cell
-#error Both Cell and Vertex are defined
-#endif
-      view.llb[i] = (feq(BP->bbox[i], patch->bbox[i], DH[i] / 2)) ? BP->bbox[i] + patch->lli[i] * DH[i] : BP->bbox[i] + (ghost_width - 0.5) * DH[i];
-      view.uub[i] = (feq(BP->bbox[dim + i], patch->bbox[dim + i], DH[i] / 2)) ? BP->bbox[dim + i] - patch->uui[i] * DH[i] : BP->bbox[dim + i] - (ghost_width - 0.5) * DH[i];
-#else
-#ifdef Cell
-      view.llb[i] = (feq(BP->bbox[i], patch->bbox[i], DH[i] / 2)) ? BP->bbox[i] + patch->lli[i] * DH[i] : BP->bbox[i] + ghost_width * DH[i];
-      view.uub[i] = (feq(BP->bbox[dim + i], patch->bbox[dim + i], DH[i] / 2)) ? BP->bbox[dim + i] - patch->uui[i] * DH[i] : BP->bbox[dim + i] - ghost_width * DH[i];
-#else
-#error Not define Vertex nor Cell
-#endif
-#endif
-    }
-    index.views.push_back(view);
-    if (Bp == patch->ble)
-      break;
-    Bp = Bp->next;
-  }
-
-  const int nblocks = int(index.views.size());
-  if (nblocks <= 0)
-    return;
-
-  int bins_1d = int(ceil(pow(double(nblocks), 1.0 / 3.0)));
-  bins_1d = clamp_int(bins_1d, 1, 32);
-  for (int i = 0; i < dim; i++)
-  {
-    index.bins[i] = bins_1d;
-    index.lo[i] = patch->bbox[i] + patch->lli[i] * DH[i];
-    const double hi = patch->bbox[dim + i] - patch->uui[i] * DH[i];
-    if (hi > index.lo[i] && bins_1d > 1)
-      index.inv[i] = bins_1d / (hi - index.lo[i]);
-    else
-      index.inv[i] = 0.0;
-  }
-
-  index.bin_to_blocks.resize(index.bins[0] * index.bins[1] * index.bins[2]);
-
-  for (int bi = 0; bi < nblocks; bi++)
-  {
-    const InterpBlockView &view = index.views[bi];
-    int bmin[dim], bmax[dim];
-    for (int d = 0; d < dim; d++)
-    {
-      const double low = view.llb[d] - DH[d] / 2;
-      const double up = view.uub[d] + DH[d] / 2;
-      bmin[d] = coord_to_bin(low, index.lo[d], index.inv[d], index.bins[d]);
-      bmax[d] = coord_to_bin(up, index.lo[d], index.inv[d], index.bins[d]);
-      if (bmax[d] < bmin[d])
-      {
-        int t = bmin[d];
-        bmin[d] = bmax[d];
-        bmax[d] = t;
-      }
-    }
-
-    for (int bz = bmin[2]; bz <= bmax[2]; bz++)
-      for (int by = bmin[1]; by <= bmax[1]; by++)
-        for (int bx = bmin[0]; bx <= bmax[0]; bx++)
-          index.bin_to_blocks[bin_loc(index, bx, by, bz)].push_back(bi);
-  }
-
-  index.valid = true;
-}
-
-int find_block_index_for_point(const BlockBinIndex &index, const double *pox, const double *DH)
-{
-  if (!index.valid)
-    return -1;
-
-  const int bx = coord_to_bin(pox[0], index.lo[0], index.inv[0], index.bins[0]);
-  const int by = coord_to_bin(pox[1], index.lo[1], index.inv[1], index.bins[1]);
-  const int bz = coord_to_bin(pox[2], index.lo[2], index.inv[2], index.bins[2]);
-  const vector<int> &cand = index.bin_to_blocks[bin_loc(index, bx, by, bz)];
-
-  for (size_t ci = 0; ci < cand.size(); ci++)
-  {
-    const int bi = cand[ci];
-    if (point_in_block_view(index.views[bi], pox, DH))
-      return bi;
-  }
-
-  // Fallback to full scan for numerical edge cases around bin boundaries.
-  for (size_t bi = 0; bi < index.views.size(); bi++)
-    if (point_in_block_view(index.views[bi], pox, DH))
-      return int(bi);
-
-  return -1;
-}
-
-void collect_interp_vars(MyList<var> *VarList, vector<InterpVarDesc> &vars)
-{
-  vars.clear();
-  MyList<var> *varl = VarList;
-  while (varl)
-  {
-    InterpVarDesc desc;
-    desc.sgfn = varl->data->sgfn;
-    for (int d = 0; d < dim; ++d)
-      desc.soa[d] = varl->data->SoA[d];
-    vars.push_back(desc);
-    varl = varl->next;
-  }
-}
-
-bool should_try_cuda_interp(int ordn, int num_points, int num_var)
-{
-#if defined(__GNUC__) || defined(__clang__)
-  if (!bssn_cuda_interp_points_batch)
-    return false;
-#else
-  return false;
-#endif
-  if (ordn != 6)
-    return false;
-  if (num_points < 32)
-    return false;
-  return num_points * num_var >= 256;
-}
-
-CachedInterpPlanEntry &interp_plan_cache_entry()
-{
-  static CachedInterpPlanEntry cache;
-  return cache;
-}
-
-bool same_interp_plan_key(const InterpPlanKey &lhs, const InterpPlanKey &rhs)
-{
-  return lhs.patch == rhs.patch &&
-         lhs.NN == rhs.NN &&
-         lhs.Symmetry == rhs.Symmetry &&
-         lhs.myrank == rhs.myrank;
-}
-
-bool same_interp_plan_points(const CachedInterpPlanEntry &cache, int NN, double **XX)
-{
-  if (static_cast<int>(cache.xvals.size()) != NN ||
-      static_cast<int>(cache.yvals.size()) != NN ||
-      static_cast<int>(cache.zvals.size()) != NN)
-    return false;
-
-  for (int j = 0; j < NN; ++j)
-  {
-    if (cache.xvals[j] != XX[0][j] ||
-        cache.yvals[j] != XX[1][j] ||
-        cache.zvals[j] != XX[2][j])
-      return false;
-  }
-  return true;
-}
-
-CachedInterpPlan &get_cached_interp_plan(Patch *patch,
-                                         int NN, double **XX,
-                                         int Symmetry, int myrank,
-                                         const double *DH,
-                                         const BlockBinIndex &block_index,
-                                         bool report_bounds_here,
-                                         bool allow_missing_points)
-{
-  InterpPlanKey key;
-  key.patch = patch;
-  key.x = XX[0];
-  key.y = XX[1];
-  key.z = XX[2];
-  key.NN = NN;
-  key.Symmetry = Symmetry;
-  key.myrank = myrank;
-
-  CachedInterpPlanEntry &cache = interp_plan_cache_entry();
-  if (cache.valid &&
-      same_interp_plan_key(cache.key, key) &&
-      same_interp_plan_points(cache, NN, XX) &&
-      cache.plan.nblocks == static_cast<int>(block_index.views.size()))
-    return cache.plan;
-
-  cache.valid = true;
-  cache.key = key;
-  cache.xvals.assign(XX[0], XX[0] + NN);
-  cache.yvals.assign(XX[1], XX[1] + NN);
-  cache.zvals.assign(XX[2], XX[2] + NN);
-  cache.plan = CachedInterpPlan();
-  CachedInterpPlan &plan = cache.plan;
-  plan.nblocks = static_cast<int>(block_index.views.size());
-  plan.owner_rank.assign(NN, -1);
-  plan.owner_block.assign(NN, -1);
-  plan.block_points.resize(plan.nblocks);
-  plan.block_px.resize(plan.nblocks);
-  plan.block_py.resize(plan.nblocks);
-  plan.block_pz.resize(plan.nblocks);
-
-  for (int j = 0; j < NN; ++j)
-  {
-    double pox[dim];
-    for (int i = 0; i < dim; ++i)
-    {
-      pox[i] = XX[i][j];
-      if (report_bounds_here &&
-          (XX[i][j] < patch->bbox[i] + patch->lli[i] * DH[i] ||
-           XX[i][j] > patch->bbox[dim + i] - patch->uui[i] * DH[i]))
-      {
-        cout << "Patch::Interp_Points: point (";
-        for (int k = 0; k < dim; ++k)
-        {
-          cout << XX[k][j];
-          if (k < dim - 1)
-            cout << ",";
-          else
-            cout << ") is out of current Patch." << endl;
-        }
-        MPI_Abort(MPI_COMM_WORLD, 1);
-      }
-    }
-
-    const int block_i = find_block_index_for_point(block_index, pox, DH);
-    if (block_i >= 0)
-    {
-      Block *BP = block_index.views[block_i].bp;
-      plan.owner_rank[j] = BP->rank;
-      plan.owner_block[j] = block_i;
-      if (BP->rank == myrank)
-      {
-        plan.block_points[block_i].push_back(j);
-        plan.block_px[block_i].push_back(XX[0][j]);
-        plan.block_py[block_i].push_back(XX[1][j]);
-        plan.block_pz[block_i].push_back(XX[2][j]);
-      }
-    }
-  }
-
-  if (!allow_missing_points && report_bounds_here)
-  {
-    for (int j = 0; j < NN; ++j)
-    {
-      if (plan.owner_rank[j] >= 0)
-        continue;
-      cout << "ERROR: Patch::Interp_Points fails to find point (";
-      for (int d = 0; d < dim; ++d)
-      {
-        cout << XX[d][j];
-        if (d < dim - 1)
-          cout << ",";
-        else
-          cout << ")";
-      }
-      cout << " on Patch (";
-      for (int d = 0; d < dim; ++d)
-      {
-        cout << patch->bbox[d] << "+" << patch->lli[d] * DH[d];
-        if (d < dim - 1)
-          cout << ",";
-        else
-          cout << ")--";
-      }
-      cout << "(";
-      for (int d = 0; d < dim; ++d)
-      {
-        cout << patch->bbox[dim + d] << "-" << patch->uui[d] * DH[d];
-        if (d < dim - 1)
-          cout << ",";
-        else
-          cout << ")" << endl;
-      }
-      MPI_Abort(MPI_COMM_WORLD, 1);
-    }
-  }
-
-  return plan;
-}
-
-void release_interp_plan_cache_internal()
-{
-  CachedInterpPlanEntry &cache = interp_plan_cache_entry();
-  cache.valid = false;
-  cache.xvals.clear();
-  cache.yvals.clear();
-  cache.zvals.clear();
-  cache.plan = CachedInterpPlan();
-}
-
-bool run_cuda_interp_for_block(Block *BP,
-                               const vector<InterpVarDesc> &vars,
-                               const vector<int> &point_ids,
-                               const vector<double> &px,
-                               const vector<double> &py,
-                               const vector<double> &pz,
-                               double *Shellf,
-                               int num_var,
-                               int ordn,
-                               int Symmetry)
-{
-  if (!should_try_cuda_interp(ordn, static_cast<int>(point_ids.size()), num_var))
-    return false;
-
-  vector<const double *> field_ptrs(num_var);
-  vector<double> soa_flat(3 * num_var);
-  for (int v = 0; v < num_var; ++v)
-  {
-    field_ptrs[v] = BP->fgfs[vars[v].sgfn];
-    for (int d = 0; d < dim; ++d)
-      soa_flat[3 * v + d] = vars[v].soa[d];
-  }
-
-  const int npts = static_cast<int>(point_ids.size());
-  vector<double> out(static_cast<size_t>(npts) * static_cast<size_t>(num_var));
-  if (bssn_cuda_interp_points_batch(BP->shape,
-                                    BP->X[0], BP->X[1], BP->X[2],
-                                    field_ptrs.data(),
-                                    soa_flat.data(),
-                                    num_var,
-                                    px.data(), py.data(), pz.data(),
-                                    npts,
-                                    ordn,
-                                    Symmetry,
-                                    out.data()) != 0)
-  {
-    return false;
-  }
-
-  for (int p = 0; p < npts; ++p)
-  {
-    const int j = point_ids[p];
-    memcpy(Shellf + j * num_var, out.data() + p * num_var, sizeof(double) * num_var);
-  }
-  return true;
-}
-
-void run_cpu_interp_for_block(Block *BP,
-                              const vector<InterpVarDesc> &vars,
-                              const vector<int> &point_ids,
-                              const vector<double> &px,
-                              const vector<double> &py,
-                              const vector<double> &pz,
-                              double *Shellf,
-                              int num_var,
-                              int ordn,
-                              int Symmetry)
-{
-  for (size_t p = 0; p < point_ids.size(); ++p)
-  {
-    const int j = point_ids[p];
-    double x = px[p];
-    double y = py[p];
-    double z = pz[p];
-    int ordn_local = ordn;
-    int symmetry_local = Symmetry;
-    for (int v = 0; v < num_var; ++v)
-    {
-      f_global_interp(BP->shape, BP->X[0], BP->X[1], BP->X[2],
-                      BP->fgfs[vars[v].sgfn], Shellf[j * num_var + v],
-                      x, y, z, ordn_local, const_cast<double *>(vars[v].soa), symmetry_local);
-    }
-  }
-}
-
-void interpolate_owned_points(MyList<var> *VarList,
-                              double *Shellf, int Symmetry,
-                              int ordn,
-                              const BlockBinIndex &block_index,
-                              const CachedInterpPlan &plan)
-{
-  vector<InterpVarDesc> vars;
-  collect_interp_vars(VarList, vars);
-  const int num_var = static_cast<int>(vars.size());
-
-  for (size_t bi = 0; bi < plan.block_points.size(); ++bi)
-  {
-    if (plan.block_points[bi].empty())
-      continue;
-
-    Block *BP = block_index.views[bi].bp;
-    bool done = run_cuda_interp_for_block(BP, vars,
-                                          plan.block_points[bi],
-                                          plan.block_px[bi],
-                                          plan.block_py[bi],
-                                          plan.block_pz[bi],
-                                          Shellf, num_var, ordn, Symmetry);
-    if (!done)
-      run_cpu_interp_for_block(BP, vars,
-                               plan.block_points[bi],
-                               plan.block_px[bi],
-                               plan.block_py[bi],
-                               plan.block_pz[bi],
-                               Shellf, num_var, ordn, Symmetry);
-  }
-}
-} // namespace
-
-void patch_release_interp_plan_cache()
-{
-  release_interp_plan_cache_internal();
-}
 
 Patch::Patch(int DIM, int *shapei, double *bboxi, int levi, bool buflog, int Symmetry) : lev(levi)
 {
@@ -895,16 +357,91 @@ void Patch::Interp_Points(MyList<var> *VarList,
 
   memset(Shellf, 0, sizeof(double) * NN * num_var);
 
-  double DH[dim];
+  // owner_rank[j] records which MPI rank owns point j
+  // All ranks traverse the same block list so they all agree on ownership
+  int *owner_rank;
+  owner_rank = new int[NN];
+  for (int j = 0; j < NN; j++)
+    owner_rank[j] = -1;
+
+  double DH[dim], llb[dim], uub[dim];
   for (int i = 0; i < dim; i++)
     DH[i] = getdX(i);
-  BlockBinIndex block_index;
-  build_block_bin_index(this, DH, block_index);
-  CachedInterpPlan &plan = get_cached_interp_plan(this, NN, XX, Symmetry, myrank, DH, block_index, myrank == 0, false);
-  const int *owner_rank = plan.owner_rank.data();
 
-  interpolate_owned_points(VarList, Shellf, Symmetry, ordn, block_index, plan);
+  for (int j = 0; j < NN; j++) // run along points
+  {
+    double pox[dim];
+    for (int i = 0; i < dim; i++)
+    {
+      pox[i] = XX[i][j];
+      if (myrank == 0 && (XX[i][j] < bbox[i] + lli[i] * DH[i] || XX[i][j] > bbox[dim + i] - uui[i] * DH[i]))
+      {
+        cout << "Patch::Interp_Points: point (";
+        for (int k = 0; k < dim; k++)
+        {
+          cout << XX[k][j];
+          if (k < dim - 1)
+            cout << ",";
+          else
+            cout << ") is out of current Patch." << endl;
+        }
+        MPI_Abort(MPI_COMM_WORLD, 1);
+      }
+    }
 
+    MyList<Block> *Bp = blb;
+    bool notfind = true;
+    while (notfind && Bp) // run along Blocks
+    {
+      Block *BP = Bp->data;
+
+      bool flag = true;
+      for (int i = 0; i < dim; i++)
+      {
+#ifdef Vertex
+#ifdef Cell
+#error Both Cell and Vertex are defined
+#endif
+        llb[i] = (feq(BP->bbox[i], bbox[i], DH[i] / 2)) ? BP->bbox[i] + lli[i] * DH[i] : BP->bbox[i] + (ghost_width - 0.5) * DH[i];
+        uub[i] = (feq(BP->bbox[dim + i], bbox[dim + i], DH[i] / 2)) ? BP->bbox[dim + i] - uui[i] * DH[i] : BP->bbox[dim + i] - (ghost_width - 0.5) * DH[i];
+#else
+#ifdef Cell
+        llb[i] = (feq(BP->bbox[i], bbox[i], DH[i] / 2)) ? BP->bbox[i] + lli[i] * DH[i] : BP->bbox[i] + ghost_width * DH[i];
+        uub[i] = (feq(BP->bbox[dim + i], bbox[dim + i], DH[i] / 2)) ? BP->bbox[dim + i] - uui[i] * DH[i] : BP->bbox[dim + i] - ghost_width * DH[i];
+#else
+#error Not define Vertex nor Cell
+#endif
+#endif
+        if (XX[i][j] - llb[i] < -DH[i] / 2 || XX[i][j] - uub[i] > DH[i] / 2)
+        {
+          flag = false;
+          break;
+        }
+      }
+
+      if (flag)
+      {
+        notfind = false;
+        owner_rank[j] = BP->rank;
+        if (myrank == BP->rank)
+        {
+          //---> interpolation
+          varl = VarList;
+          int k = 0;
+          while (varl) // run along variables
+          {
+            f_global_interp(BP->shape, BP->X[0], BP->X[1], BP->X[2], BP->fgfs[varl->data->sgfn], Shellf[j * num_var + k],
+                            pox[0], pox[1], pox[2], ordn, varl->data->SoA, Symmetry);
+            varl = varl->next;
+            k++;
+          }
+        }
+      }
+      if (Bp == ble)
+        break;
+      Bp = Bp->next;
+    }
+  }
   // Replace MPI_Allreduce with per-owner MPI_Bcast:
   // Group consecutive points by owner rank and broadcast each group.
   // Since each point's data is non-zero only on the owner rank,
@@ -959,6 +496,7 @@ void Patch::Interp_Points(MyList<var> *VarList,
     }
   }
 
+  delete[] owner_rank;
 }
 void Patch::Interp_Points(MyList<var> *VarList,
                           int NN, double **XX,
@@ -968,9 +506,9 @@ void Patch::Interp_Points(MyList<var> *VarList,
   // Targeted point-to-point overload: each owner sends each point only to
   // the one rank that needs it for integration (consumer), reducing
   // communication volume by ~nprocs times compared to the Bcast version.
-#ifdef INTERP_LB_PROFILE
+  /*
-  double t_interp_start = MPI_Wtime();
+  double t_calc_end, t_calc_total = 0;
-#endif
+  double t_calc_start = MPI_Wtime();*/
   int myrank, nprocs;
   MPI_Comm_rank(MPI_COMM_WORLD, &myrank);
   MPI_Comm_size(MPI_COMM_WORLD, &nprocs);
@@ -987,20 +525,129 @@ void Patch::Interp_Points(MyList<var> *VarList,
 
   memset(Shellf, 0, sizeof(double) * NN * num_var);
 
-  double DH[dim];
+  // owner_rank[j] records which MPI rank owns point j
+  int *owner_rank;
+  owner_rank = new int[NN];
+  for (int j = 0; j < NN; j++)
+    owner_rank[j] = -1;
+
+  double DH[dim], llb[dim], uub[dim];
   for (int i = 0; i < dim; i++)
     DH[i] = getdX(i);
-  BlockBinIndex block_index;
-  build_block_bin_index(this, DH, block_index);
-  CachedInterpPlan &plan = get_cached_interp_plan(this, NN, XX, Symmetry, myrank, DH, block_index, myrank == 0, false);
-  const int *owner_rank = plan.owner_rank.data();
 
-  interpolate_owned_points(VarList, Shellf, Symmetry, ordn, block_index, plan);
+  // --- Interpolation phase (identical to original) ---
+  for (int j = 0; j < NN; j++)
+  {
+    double pox[dim];
+    for (int i = 0; i < dim; i++)
+    {
+      pox[i] = XX[i][j];
+      if (myrank == 0 && (XX[i][j] < bbox[i] + lli[i] * DH[i] || XX[i][j] > bbox[dim + i] - uui[i] * DH[i]))
+      {
+        cout << "Patch::Interp_Points: point (";
+        for (int k = 0; k < dim; k++)
+        {
+          cout << XX[k][j];
+          if (k < dim - 1)
+            cout << ",";
+          else
+            cout << ") is out of current Patch." << endl;
+        }
+        MPI_Abort(MPI_COMM_WORLD, 1);
+      }
+    }
 
-#ifdef INTERP_LB_PROFILE
+    MyList<Block> *Bp = blb;
-  double t_interp_end = MPI_Wtime();
+    bool notfind = true;
-  double t_interp_local = t_interp_end - t_interp_start;
+    while (notfind && Bp)
+    {
+      Block *BP = Bp->data;
+
+      bool flag = true;
+      for (int i = 0; i < dim; i++)
+      {
+#ifdef Vertex
+#ifdef Cell
+#error Both Cell and Vertex are defined
 #endif
+        llb[i] = (feq(BP->bbox[i], bbox[i], DH[i] / 2)) ? BP->bbox[i] + lli[i] * DH[i] : BP->bbox[i] + (ghost_width - 0.5) * DH[i];
+        uub[i] = (feq(BP->bbox[dim + i], bbox[dim + i], DH[i] / 2)) ? BP->bbox[dim + i] - uui[i] * DH[i] : BP->bbox[dim + i] - (ghost_width - 0.5) * DH[i];
+#else
+#ifdef Cell
+        llb[i] = (feq(BP->bbox[i], bbox[i], DH[i] / 2)) ? BP->bbox[i] + lli[i] * DH[i] : BP->bbox[i] + ghost_width * DH[i];
+        uub[i] = (feq(BP->bbox[dim + i], bbox[dim + i], DH[i] / 2)) ? BP->bbox[dim + i] - uui[i] * DH[i] : BP->bbox[dim + i] - ghost_width * DH[i];
+#else
+#error Not define Vertex nor Cell
+#endif
+#endif
+        if (XX[i][j] - llb[i] < -DH[i] / 2 || XX[i][j] - uub[i] > DH[i] / 2)
+        {
+          flag = false;
+          break;
+        }
+      }
+
+      if (flag)
+      {
+        notfind = false;
+        owner_rank[j] = BP->rank;
+        if (myrank == BP->rank)
+        {
+          varl = VarList;
+          int k = 0;
+          while (varl)
+          {
+            f_global_interp(BP->shape, BP->X[0], BP->X[1], BP->X[2], BP->fgfs[varl->data->sgfn], Shellf[j * num_var + k],
+                            pox[0], pox[1], pox[2], ordn, varl->data->SoA, Symmetry);
+            varl = varl->next;
+            k++;
+          }
+        }
+      }
+      if (Bp == ble)
+        break;
+      Bp = Bp->next;
+    }
+  }
+  /*
+      t_calc_end = MPI_Wtime();
+      t_calc_total = t_calc_end - t_calc_start;*/
+  // --- Error check for unfound points ---
+  for (int j = 0; j < NN; j++)
+  {
+    if (owner_rank[j] < 0 && myrank == 0)
+    {
+      cout<<owner_rank[j-1]<<endl;
+      cout << "ERROR: Patch::Interp_Points fails to find point (";
+      for (int d = 0; d < dim; d++)
+      {
+        cout << XX[d][j];
+        if (d < dim - 1)
+          cout << ",";
+        else
+          cout << ")";
+      }
+      cout << " on Patch (";
+      for (int d = 0; d < dim; d++)
+      {
+        cout << bbox[d] << "+" << lli[d] * DH[d];
+        if (d < dim - 1)
+          cout << ",";
+        else
+          cout << ")--";
+      }
+      cout << "(";
+      for (int d = 0; d < dim; d++)
+      {
+        cout << bbox[dim + d] << "-" << uui[d] * DH[d];
+        if (d < dim - 1)
+          cout << ",";
+        else
+          cout << ")" << endl;
+      }
+      MPI_Abort(MPI_COMM_WORLD, 1);
+    }
+  }
 
   // --- Targeted point-to-point communication phase ---
   // Compute consumer_rank[j] using the same deterministic formula as surface_integral
@@ -1121,31 +768,64 @@ void Patch::Interp_Points(MyList<var> *VarList,
   delete[] send_count;
   delete[] recv_count;
   delete[] consumer_rank;
+  delete[] owner_rank;
+  /*
+  // 4. 汇总并输出真正干活最慢的 Top 4
+  struct RankStats {
+    int rank;
+    double calc_time; // 净计算时间
+  };
 
-#ifdef INTERP_LB_PROFILE
+  // 创建当前进程的统计数据
-  {
+  RankStats local_stat;
-    static bool profile_written = false;
+  local_stat.rank = myrank;
-    if (!profile_written) {
+  local_stat.calc_time = t_calc_total;
-      double *all_times = nullptr;
+
-      if (myrank == 0) all_times = new double[nprocs];
+  // 为所有进程的统计数据分配内存
-      MPI_Gather(&t_interp_local, 1, MPI_DOUBLE,
+  RankStats *all_stats = nullptr;
-                 all_times, 1, MPI_DOUBLE, 0, MPI_COMM_WORLD);
+  if (myrank == 0) {
-      if (myrank == 0) {
+    all_stats = new RankStats[nprocs];
-        int heavy[64];
-        int nh = InterpLBProfile::identify_heavy_ranks(
-            all_times, nprocs, 2.5, heavy, 64);
-        InterpLBProfile::write_profile(
-            "interp_lb_profile.bin", nprocs,
-            all_times, heavy, nh, 2.5);
-        printf("[InterpLB] Profile written: %d heavy ranks\n", nh);
-        for (int i = 0; i < nh; i++)
-          printf("  Heavy rank %d: %.6f s\n", heavy[i], all_times[heavy[i]]);
-        delete[] all_times;
-      }
-      profile_written = true;
-    }
   }
-#endif
+
+  // 使用MPI_Gather收集所有进程的数据到rank 0
+  MPI_Gather(&local_stat, sizeof(RankStats), MPI_BYTE,
+             all_stats, sizeof(RankStats), MPI_BYTE,
+             0, MPI_COMM_WORLD);
+
+  // 准备输出前4个rank的信息（所有rank都参与，确保广播后一致）
+  int top10_ranks[10] = { -1, -1, -1, -1, -1, -1, -1, -1, -1, -1 };
+  double top10_times[10] = { 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0 };
+  int num_top10 = 0;
+
+  if (myrank == 0) {
+    // 按 calc_time（净计算时间）排序
+    std::sort(all_stats, all_stats + nprocs, [](const RankStats& a, const RankStats& b) {
+        return a.calc_time > b.calc_time;
+    });
+
+    // 取前4个
+    num_top10 = (nprocs < 10) ? nprocs : 10;
+    for (int i = 0; i < num_top10; i++) {
+      top10_ranks[i] = all_stats[i].rank;
+      top10_times[i] = all_stats[i].calc_time;
+    }
+
+    printf("\n--- Top %d Ranks by ACTIVE COMPUTATION (CPU Time) ---\n", num_top10);
+    for (int i = 0; i < num_top10; i++) {
+      printf("Rank [%4d]: Calc %.6f s\n", top10_ranks[i], top10_times[i]);
+    }
+
+    // 清理分配的内存
+    delete[] all_stats;
+  }
+
+  // 广播前4个rank的信息给所有进程
+  MPI_Bcast(&num_top10, 1, MPI_INT, 0, MPI_COMM_WORLD);
+  if (num_top10 > 0) {
+    MPI_Bcast(top10_ranks, 10, MPI_INT, 0, MPI_COMM_WORLD);
+    MPI_Bcast(top10_times, 10, MPI_DOUBLE, 0, MPI_COMM_WORLD);
+  }
+*/
 }
 void Patch::Interp_Points(MyList<var> *VarList,
                           int NN, double **XX,
@@ -1168,20 +848,95 @@ void Patch::Interp_Points(MyList<var> *VarList,
 
   memset(Shellf, 0, sizeof(double) * NN * num_var);
 
+  // owner_rank[j] stores the global rank that owns point j
+  int *owner_rank;
+  owner_rank = new int[NN];
+  for (int j = 0; j < NN; j++)
+    owner_rank[j] = -1;
+
   // Build global-to-local rank translation for Comm_here
   MPI_Group world_group, local_group;
   MPI_Comm_group(MPI_COMM_WORLD, &world_group);
   MPI_Comm_group(Comm_here, &local_group);
 
-  double DH[dim];
+  double DH[dim], llb[dim], uub[dim];
   for (int i = 0; i < dim; i++)
     DH[i] = getdX(i);
-  BlockBinIndex block_index;
-  build_block_bin_index(this, DH, block_index);
-  CachedInterpPlan &plan = get_cached_interp_plan(this, NN, XX, Symmetry, myrank, DH, block_index, lmyrank == 0, true);
-  const int *owner_rank = plan.owner_rank.data();
 
-  interpolate_owned_points(VarList, Shellf, Symmetry, ordn, block_index, plan);
+  for (int j = 0; j < NN; j++) // run along points
+  {
+    double pox[dim];
+    for (int i = 0; i < dim; i++)
+    {
+      pox[i] = XX[i][j];
+      if (lmyrank == 0 && (XX[i][j] < bbox[i] + lli[i] * DH[i] || XX[i][j] > bbox[dim + i] - uui[i] * DH[i]))
+      {
+        cout << "Patch::Interp_Points: point (";
+        for (int k = 0; k < dim; k++)
+        {
+          cout << XX[k][j];
+          if (k < dim - 1)
+            cout << ",";
+          else
+            cout << ") is out of current Patch." << endl;
+        }
+        MPI_Abort(MPI_COMM_WORLD, 1);
+      }
+    }
+
+    MyList<Block> *Bp = blb;
+    bool notfind = true;
+    while (notfind && Bp) // run along Blocks
+    {
+      Block *BP = Bp->data;
+
+      bool flag = true;
+      for (int i = 0; i < dim; i++)
+      {
+#ifdef Vertex
+#ifdef Cell
+#error Both Cell and Vertex are defined
+#endif
+        llb[i] = (feq(BP->bbox[i], bbox[i], DH[i] / 2)) ? BP->bbox[i] + lli[i] * DH[i] : BP->bbox[i] + (ghost_width - 0.5) * DH[i];
+        uub[i] = (feq(BP->bbox[dim + i], bbox[dim + i], DH[i] / 2)) ? BP->bbox[dim + i] - uui[i] * DH[i] : BP->bbox[dim + i] - (ghost_width - 0.5) * DH[i];
+#else
+#ifdef Cell
+        llb[i] = (feq(BP->bbox[i], bbox[i], DH[i] / 2)) ? BP->bbox[i] + lli[i] * DH[i] : BP->bbox[i] + ghost_width * DH[i];
+        uub[i] = (feq(BP->bbox[dim + i], bbox[dim + i], DH[i] / 2)) ? BP->bbox[dim + i] - uui[i] * DH[i] : BP->bbox[dim + i] - ghost_width * DH[i];
+#else
+#error Not define Vertex nor Cell
+#endif
+#endif
+        if (XX[i][j] - llb[i] < -DH[i] / 2 || XX[i][j] - uub[i] > DH[i] / 2)
+        {
+          flag = false;
+          break;
+        }
+      }
+
+      if (flag)
+      {
+        notfind = false;
+        owner_rank[j] = BP->rank;
+        if (myrank == BP->rank)
+        {
+          //---> interpolation
+          varl = VarList;
+          int k = 0;
+          while (varl) // run along variables
+          {
+            f_global_interp(BP->shape, BP->X[0], BP->X[1], BP->X[2], BP->fgfs[varl->data->sgfn], Shellf[j * num_var + k],
+                            pox[0], pox[1], pox[2], ordn, varl->data->SoA, Symmetry);
+            varl = varl->next;
+            k++;
+          }
+        }
+      }
+      if (Bp == ble)
+        break;
+      Bp = Bp->next;
+    }
+  }
 
   // Collect unique global owner ranks and translate to local ranks in Comm_here
   // Then broadcast each owner's points via MPI_Bcast on Comm_here
@@ -1211,6 +966,7 @@ void Patch::Interp_Points(MyList<var> *VarList,
 
   MPI_Group_free(&world_group);
   MPI_Group_free(&local_group);
+  delete[] owner_rank;
 }
 void Patch::checkBlock()
 {

AMSS_NCKU_source/MPatch.h

@@ -52,6 +52,4 @@ public:
                      double *Shellf, MPI_Comm Comm_here);
 };
 
-void patch_release_interp_plan_cache();
-
 #endif /* PATCH_H */

AMSS_NCKU_source/NullShellPatch.h

@@ -24,6 +24,7 @@ using namespace std;
 #endif
 
 #include <mpi.h>
+#include <memory.h>
 #include "MyList.h"
 #include "Block.h"
 #include "Parallel.h"

AMSS_NCKU_source/Parallel.h

@@ -11,7 +11,7 @@
 #include <cmath>
 #include <new>
 using namespace std;
-
+#include <memory.h>
 #include "Parallel_bam.h"
 #include "var.h"
 #include "MPatch.h"
@@ -32,16 +32,25 @@ namespace Parallel
   int partition2(int *nxy, int split_size, int *min_width, int cpusize, int *shape); // special for 2 diemnsions
   int partition3(int *nxyz, int split_size, int *min_width, int cpusize, int *shape);
   MyList<Block> *distribute(MyList<Patch> *PatchLIST, int cpusize, int ingfsi, int fngfs, bool periodic, int nodes = 0); // produce corresponding Blocks
-  MyList<Block> *distribute_optimize(MyList<Patch> *PatchLIST, int cpusize, int ingfsi, int fngfs, bool periodic, int nodes = 0);
+
+  MyList<Block> *distribute_hard(MyList<Patch> *PatchLIST, int cpusize, int ingfsi, int fngfs, bool periodic, int nodes = 0); // produce corresponding Blocks
   Block* splitHotspotBlock(MyList<Block>* &BlL, int _dim, 
-                           int ib0_orig, int ib3_orig,
+                                 int ib0_orig, int ib3_orig, 
-                           int jb1_orig, int jb4_orig,
+                                 int jb1_orig, int jb4_orig, 
-                           int kb2_orig, int kb5_orig,
+                                 int kb2_orig, int kb5_orig, 
-                           Patch* PP, int r_left, int r_right,
+                                 Patch* PP, int r_1, int r_2, 
-                           int ingfsi, int fngfsi, bool periodic,
+                                 int ingfsi, int fngfsi, bool periodic,
-                           Block* &split_first_block, Block* &split_last_block);
+                                 Block* &split_first_block, Block* &split_last_block);
+
+Block* splitHotspotBlock(MyList<Block>* &BlL, int _dim, 
+                                 int ib0_orig, int ib3_orig, 
+                                 int jb1_orig, int jb4_orig, 
+                                 int kb2_orig, int kb5_orig, 
+                                 Patch* PP, int r_1, int r_2, int r_3, int r_4,
+                                 int ingfsi, int fngfsi, bool periodic,
+                                 Block* &split_first_block, Block* &split_last_block);
   Block* createMappedBlock(MyList<Block>* &BlL, int _dim, int* shape, double* bbox,
-                           int block_id, int ingfsi, int fngfsi, int lev);
+                        int block_id, int ingfsi, int fngfsi, int lev); 
   void KillBlocks(MyList<Patch> *PatchLIST);
 
   void setfunction(MyList<Block> *BlL, var *vn, double func(double x, double y, double z));
@@ -90,11 +99,8 @@ namespace Parallel
                    MyList<var> *VarList1 /* source */, MyList<var> *VarList2 /*target */,
                    int Symmetry);
   void Sync(Patch *Pat, MyList<var> *VarList, int Symmetry);
-  void Sync(Patch *Pat, MyList<var> *VarList, int Symmetry, const char *context);
   void Sync(MyList<Patch> *PatL, MyList<var> *VarList, int Symmetry);
-  void Sync(MyList<Patch> *PatL, MyList<var> *VarList, int Symmetry, const char *context);
   void Sync_merged(MyList<Patch> *PatL, MyList<var> *VarList, int Symmetry);
-  void Sync_merged(MyList<Patch> *PatL, MyList<var> *VarList, int Symmetry, const char *context);
 
   struct SyncCache {
     bool valid;
@@ -111,10 +117,6 @@ namespace Parallel
     MPI_Status *stats;
     int max_reqs;
     bool lengths_valid;
-    int lengths_var_count;
-    int *tc_req_node;
-    int *tc_req_is_recv;
-    int *tc_completed;
     SyncCache();
     void invalidate();
     void destroy();
@@ -128,17 +130,13 @@ namespace Parallel
   struct AsyncSyncState {
     int req_no;
     bool active;
-    int mpi_tag;
+    AsyncSyncState() : req_no(0), active(false) {}
-    int *req_node;
-    int *req_is_recv;
-    int pending_recv;
-    AsyncSyncState() : req_no(0), active(false), mpi_tag(0), req_node(0), req_is_recv(0), pending_recv(0) {}
   };
 
   void Sync_start(MyList<Patch> *PatL, MyList<var> *VarList, int Symmetry,
                   SyncCache &cache, AsyncSyncState &state);
   void Sync_finish(SyncCache &cache, AsyncSyncState &state,
-                   MyList<var> *VarList, int Symmetry, bool unpack_to_host = true);
+                   MyList<var> *VarList, int Symmetry);
   void OutBdLow2Hi(Patch *Patc, Patch *Patf,
                    MyList<var> *VarList1 /* source */, MyList<var> *VarList2 /* target */,
                    int Symmetry);
@@ -229,6 +227,18 @@ namespace Parallel
 #if (PSTR == 1 || PSTR == 2 || PSTR == 3)
   MyList<Block> *distribute(MyList<Patch> *PatchLIST, int cpusize, int ingfsi, int fngfsi,
                             bool periodic, int start_rank, int end_rank, int nodes = 0);
+  
+  // Redistribute blocks with time statistics for load balancing
+  MyList<Block> *distribute(MyList<Patch> *PatchLIST, MyList<Block> *OldBlockL,
+                            int cpusize, int ingfsi, int fngfsi,
+                            bool periodic, int start_rank, int end_rank, int nodes = 0);
 #endif
-}
+
-#endif /*PARALLEL_H */
+  // Dynamic load balancing: split blocks for heavy ranks
+    void split_heavy_blocks(MyList<Patch> *PatL, int *heavy_ranks, int num_heavy,
+                            int split_factor, int cpusize, int ingfsi, int fngfsi);
+
+    // Check if load balancing is needed based on interpolation times
+    bool check_load_balance_need(double *rank_times, int nprocs, int &num_heavy, int *heavy_ranks);
+  }
+  #endif /*PARALLEL_H */

AMSS_NCKU_source/bssn_class.C

@@ -15,7 +15,6 @@ using namespace std;
 #endif
 
 #include <time.h>
-#include <unistd.h>
 
 #include "macrodef.h"
 #include "misc.h"
@@ -31,10 +30,6 @@ using namespace std;
 #include "getnp4.h"
 #include "shellfunctions.h"
 #include "parameters.h"
-#ifdef USE_GPU
-#include "bssn_macro.h"
-#include "bssn_gpu.h"
-#endif
 
 #ifdef With_AHF
 #include "derivatives.h"
@@ -46,12 +41,6 @@ using namespace std;
 #include "derivatives.h"
 #include "ricci_gamma.h"
 
-// Compile-time switch for per-timestep memory usage collection/printing.
-// Default is OFF to reduce overhead in production runs.
-#ifndef BSSN_ENABLE_MEM_USAGE_LOG
-#define BSSN_ENABLE_MEM_USAGE_LOG 0
-#endif
-
 //================================================================================================
 
 // define bssn_class
@@ -747,9 +736,6 @@ void bssn_class::Initialize()
   sync_cache_cor = new Parallel::SyncCache[GH->levels];
   sync_cache_rp_coarse = new Parallel::SyncCache[GH->levels];
   sync_cache_rp_fine = new Parallel::SyncCache[GH->levels];
-  sync_cache_restrict = new Parallel::SyncCache[GH->levels];
-  sync_cache_outbd = new Parallel::SyncCache[GH->levels];
-  sync_cache_psi4 = new Parallel::SyncCache[GH->levels];
 }
 
 //================================================================================================
@@ -1026,24 +1012,6 @@ bssn_class::~bssn_class()
       sync_cache_rp_fine[i].destroy();
     delete[] sync_cache_rp_fine;
   }
-  if (sync_cache_restrict)
-  {
-    for (int i = 0; i < GH->levels; i++)
-      sync_cache_restrict[i].destroy();
-    delete[] sync_cache_restrict;
-  }
-  if (sync_cache_outbd)
-  {
-    for (int i = 0; i < GH->levels; i++)
-      sync_cache_outbd[i].destroy();
-    delete[] sync_cache_outbd;
-  }
-  if (sync_cache_psi4)
-  {
-    for (int i = 0; i < GH->levels; i++)
-      sync_cache_psi4[i].destroy();
-    delete[] sync_cache_psi4;
-  }
 
   delete GH;
 #ifdef WithShell
@@ -1079,23 +1047,6 @@ bssn_class::~bssn_class()
   delete CheckPoint;
 }
 
-void bssn_class::InvalidateSyncCaches()
-{
-  if (!GH)
-    return;
-
-  for (int il = 0; il < GH->levels; il++)
-  {
-    sync_cache_pre[il].invalidate();
-    sync_cache_cor[il].invalidate();
-    sync_cache_rp_coarse[il].invalidate();
-    sync_cache_rp_fine[il].invalidate();
-    sync_cache_restrict[il].invalidate();
-    sync_cache_outbd[il].invalidate();
-    sync_cache_psi4[il].invalidate();
-  }
-}
-
 //================================================================================================
 
 
@@ -2080,7 +2031,6 @@ void bssn_class::Evolve(int Steps)
   clock_t prev_clock, curr_clock;
   double LastDump = 0.0, LastCheck = 0.0, Last2dDump = 0.0;
   LastAnas = 0;
-  LastConsOut = 0;
 #if 0
 //initial checkpoint for special uasge
      {
@@ -2177,10 +2127,8 @@ void bssn_class::Evolve(int Steps)
   #endif
   */
   
-#if BSSN_ENABLE_MEM_USAGE_LOG
   perf bssn_perf;
   size_t current_min, current_avg, current_max, peak_min, peak_avg, peak_max;
-#endif
 
   for (int lev = 0; lev < GH->levels; lev++)
     GH->Lt[lev] = PhysTime;
@@ -2265,7 +2213,7 @@ void bssn_class::Evolve(int Steps)
     GH->Regrid(Symmetry, BH_num, Porgbr, Porg0,
                SynchList_cor, OldStateList, StateList, SynchList_pre,
                fgt(PhysTime - dT_mon, StartTime, dT_mon / 2), ErrorMonitor);
-    InvalidateSyncCaches();
+    for (int il = 0; il < GH->levels; il++) { sync_cache_pre[il].invalidate(); sync_cache_cor[il].invalidate(); sync_cache_rp_coarse[il].invalidate(); sync_cache_rp_fine[il].invalidate(); }
 #endif
 
 #if (REGLEV == 0 && (PSTR == 1 || PSTR == 2))
@@ -2274,7 +2222,6 @@ void bssn_class::Evolve(int Steps)
 //		fgt(PhysTime-dT_mon,StartTime,dT_mon/2),ErrorMonitor);
 #endif
 
-#if BSSN_ENABLE_MEM_USAGE_LOG
     // Retrieve memory usage information used during computation; master process prints it
     bssn_perf.MemoryUsage(&current_min, &current_avg, &current_max,
                           &peak_min, &peak_avg, &peak_max, nprocs);
@@ -2290,7 +2237,6 @@ void bssn_class::Evolve(int Steps)
              (double)peak_max / (1024.0 * 1024.0));
       cout << endl;
     }
-#endif
     
     // Output puncture positions at each step
     if (myrank == 0)
@@ -2349,9 +2295,6 @@ void bssn_class::Evolve(int Steps)
 #endif
       CheckPoint->write_bssn(LastDump, Last2dDump, LastAnas);
     }
-
-    // Keep output/analysis phases aligned across ranks before the next coarse step.
-    MPI_Barrier(MPI_COMM_WORLD);
   }
   /*
   #ifdef With_AHF
@@ -2483,10 +2426,10 @@ void bssn_class::RecursiveStep(int lev)
 #endif
 
 #if (REGLEV == 0)
-  if (GH->Regrid_Onelevel(lev, Symmetry, BH_num, Porgbr, Porg0,
+  GH->Regrid_Onelevel(lev, Symmetry, BH_num, Porgbr, Porg0,
                       SynchList_cor, OldStateList, StateList, SynchList_pre,
-                      fgt(PhysTime - dT_lev, StartTime, dT_lev / 2), ErrorMonitor))
+                      fgt(PhysTime - dT_lev, StartTime, dT_lev / 2), ErrorMonitor);
-  InvalidateSyncCaches();
+  for (int il = 0; il < GH->levels; il++) { sync_cache_pre[il].invalidate(); sync_cache_cor[il].invalidate(); sync_cache_rp_coarse[il].invalidate(); sync_cache_rp_fine[il].invalidate(); }
 #endif
 }
 
@@ -2662,10 +2605,10 @@ void bssn_class::ParallelStep()
   delete[] tporg;
   delete[] tporgo;
 #if (REGLEV == 0)
-  if (GH->Regrid_Onelevel(GH->mylev, Symmetry, BH_num, Porgbr, Porg0,
+  GH->Regrid_Onelevel(GH->mylev, Symmetry, BH_num, Porgbr, Porg0,
                       SynchList_cor, OldStateList, StateList, SynchList_pre,
-                      fgt(PhysTime - dT_lev, StartTime, dT_lev / 2), ErrorMonitor))
+                      fgt(PhysTime - dT_lev, StartTime, dT_lev / 2), ErrorMonitor);
-  InvalidateSyncCaches();
+  for (int il = 0; il < GH->levels; il++) { sync_cache_pre[il].invalidate(); sync_cache_cor[il].invalidate(); sync_cache_rp_coarse[il].invalidate(); sync_cache_rp_fine[il].invalidate(); }
 #endif
 }
 
@@ -2829,10 +2772,10 @@ void bssn_class::ParallelStep()
       if (lev + 1 >= GH->movls)
       {
         //	       GH->Regrid_Onelevel_aux(lev,Symmetry,BH_num,Porgbr,Porg0,
-        if (GH->Regrid_Onelevel(lev + 1, Symmetry, BH_num, Porgbr, Porg0,
+        GH->Regrid_Onelevel(lev + 1, Symmetry, BH_num, Porgbr, Porg0,
                             SynchList_cor, OldStateList, StateList, SynchList_pre,
-                            fgt(PhysTime - dT_levp1, StartTime, dT_levp1 / 2), ErrorMonitor))
+                            fgt(PhysTime - dT_levp1, StartTime, dT_levp1 / 2), ErrorMonitor);
-        InvalidateSyncCaches();
+        for (int il = 0; il < GH->levels; il++) { sync_cache_pre[il].invalidate(); sync_cache_cor[il].invalidate(); sync_cache_rp_coarse[il].invalidate(); sync_cache_rp_fine[il].invalidate(); }
 
         //               a_stream.clear();
         //               a_stream.str("");
@@ -2844,10 +2787,10 @@ void bssn_class::ParallelStep()
     // for this level
     if (YN == 1)
     {
-      if (GH->Regrid_Onelevel(lev, Symmetry, BH_num, Porgbr, Porg0,
+      GH->Regrid_Onelevel(lev, Symmetry, BH_num, Porgbr, Porg0,
                           SynchList_cor, OldStateList, StateList, SynchList_pre,
-                          fgt(PhysTime - dT_lev, StartTime, dT_lev / 2), ErrorMonitor))
+                          fgt(PhysTime - dT_lev, StartTime, dT_lev / 2), ErrorMonitor);
-      InvalidateSyncCaches();
+      for (int il = 0; il < GH->levels; il++) { sync_cache_pre[il].invalidate(); sync_cache_cor[il].invalidate(); sync_cache_rp_coarse[il].invalidate(); sync_cache_rp_fine[il].invalidate(); }
 
       //               a_stream.clear();
       //               a_stream.str("");
@@ -2863,10 +2806,10 @@ void bssn_class::ParallelStep()
         if (YN == 1)
         {
           //	   GH->Regrid_Onelevel_aux(lev-2,Symmetry,BH_num,Porgbr,Porg0,
-          if (GH->Regrid_Onelevel(lev - 1, Symmetry, BH_num, Porgbr, Porg0,
+          GH->Regrid_Onelevel(lev - 1, Symmetry, BH_num, Porgbr, Porg0,
                               SynchList_cor, OldStateList, StateList, SynchList_pre,
-                              fgt(PhysTime - dT_lev, StartTime, dT_levm1 / 2), ErrorMonitor))
+                              fgt(PhysTime - dT_lev, StartTime, dT_levm1 / 2), ErrorMonitor);
-          InvalidateSyncCaches();
+          for (int il = 0; il < GH->levels; il++) { sync_cache_pre[il].invalidate(); sync_cache_cor[il].invalidate(); sync_cache_rp_coarse[il].invalidate(); sync_cache_rp_fine[il].invalidate(); }
 
           //               a_stream.clear();
           //               a_stream.str("");
@@ -2879,10 +2822,10 @@ void bssn_class::ParallelStep()
         if (i % 4 == 3)
         {
           //	   GH->Regrid_Onelevel_aux(lev-2,Symmetry,BH_num,Porgbr,Porg0,
-          if (GH->Regrid_Onelevel(lev - 1, Symmetry, BH_num, Porgbr, Porg0,
+          GH->Regrid_Onelevel(lev - 1, Symmetry, BH_num, Porgbr, Porg0,
                               SynchList_cor, OldStateList, StateList, SynchList_pre,
-                              fgt(PhysTime - dT_lev, StartTime, dT_levm1 / 2), ErrorMonitor))
+                              fgt(PhysTime - dT_lev, StartTime, dT_levm1 / 2), ErrorMonitor);
-          InvalidateSyncCaches();
+          for (int il = 0; il < GH->levels; il++) { sync_cache_pre[il].invalidate(); sync_cache_cor[il].invalidate(); sync_cache_rp_coarse[il].invalidate(); sync_cache_rp_fine[il].invalidate(); }
 
           //               a_stream.clear();
           //               a_stream.str("");
@@ -3073,11 +3016,6 @@ void bssn_class::RecursiveStep(int lev, int num) // in all 2^(lev+1)-1 steps
 #if 1
 void bssn_class::Step(int lev, int YN)
 {
-#ifdef USE_GPU
-  Step_MainPath_GPU(lev, YN);
-  return;
-#endif
-
   setpbh(BH_num, Porg0, Mass, BH_num_input);
 
   double dT_lev = dT * pow(0.5, Mymax(lev, trfls));
@@ -5858,7 +5796,7 @@ void bssn_class::RestrictProlong(int lev, int YN, bool BB,
 #endif
 
 #if (RPB == 0)
-      Parallel::Restrict_cached(GH->PatL[lev - 1], GH->PatL[lev], SL, SynchList_pre, Symmetry, sync_cache_restrict[lev]);
+      Parallel::Restrict(GH->PatL[lev - 1], GH->PatL[lev], SL, SynchList_pre, Symmetry);
 #elif (RPB == 1)
       //       Parallel::Restrict_bam(GH->PatL[lev-1],GH->PatL[lev],SL,SynchList_pre,Symmetry);
       Parallel::Restrict_bam(GH->PatL[lev - 1], GH->PatL[lev], SL, SynchList_pre, GH->rsul[lev], Symmetry);
@@ -5882,7 +5820,7 @@ void bssn_class::RestrictProlong(int lev, int YN, bool BB,
 
 #if (RPB == 0)
 #if (MIXOUTB == 0)
-      Parallel::OutBdLow2Hi_cached(GH->PatL[lev - 1], GH->PatL[lev], SynchList_pre, SL, Symmetry, sync_cache_outbd[lev]);
+      Parallel::OutBdLow2Hi(GH->PatL[lev - 1], GH->PatL[lev], SynchList_pre, SL, Symmetry);
 #elif (MIXOUTB == 1)
       Parallel::OutBdLow2Himix(GH->PatL[lev - 1], GH->PatL[lev], SynchList_pre, SL, Symmetry);
 #endif
@@ -5909,7 +5847,7 @@ void bssn_class::RestrictProlong(int lev, int YN, bool BB,
 #endif
 
 #if (RPB == 0)
-      Parallel::Restrict_cached(GH->PatL[lev - 1], GH->PatL[lev], SL, SL, Symmetry, sync_cache_restrict[lev]);
+      Parallel::Restrict(GH->PatL[lev - 1], GH->PatL[lev], SL, SL, Symmetry);
 #elif (RPB == 1)
       //       Parallel::Restrict_bam(GH->PatL[lev-1],GH->PatL[lev],SL,SL,Symmetry);
       Parallel::Restrict_bam(GH->PatL[lev - 1], GH->PatL[lev], SL, SL, GH->rsul[lev], Symmetry);
@@ -5933,7 +5871,7 @@ void bssn_class::RestrictProlong(int lev, int YN, bool BB,
 
 #if (RPB == 0)
 #if (MIXOUTB == 0)
-      Parallel::OutBdLow2Hi_cached(GH->PatL[lev - 1], GH->PatL[lev], SL, SL, Symmetry, sync_cache_outbd[lev]);
+      Parallel::OutBdLow2Hi(GH->PatL[lev - 1], GH->PatL[lev], SL, SL, Symmetry);
 #elif (MIXOUTB == 1)
       Parallel::OutBdLow2Himix(GH->PatL[lev - 1], GH->PatL[lev], SL, SL, Symmetry);
 #endif
@@ -6002,7 +5940,7 @@ void bssn_class::RestrictProlong_aux(int lev, int YN, bool BB,
       }
 
 #if (RPB == 0)
-      Parallel::Restrict_cached(GH->PatL[lev - 1], GH->PatL[lev], SL, SynchList_pre, Symmetry, sync_cache_restrict[lev]);
+      Parallel::Restrict(GH->PatL[lev - 1], GH->PatL[lev], SL, SynchList_pre, Symmetry);
 #elif (RPB == 1)
       //       Parallel::Restrict_bam(GH->PatL[lev-1],GH->PatL[lev],SL,SynchList_pre,Symmetry);
       Parallel::Restrict_bam(GH->PatL[lev - 1], GH->PatL[lev], SL, SynchList_pre, GH->rsul[lev], Symmetry);
@@ -6012,7 +5950,7 @@ void bssn_class::RestrictProlong_aux(int lev, int YN, bool BB,
 
 #if (RPB == 0)
 #if (MIXOUTB == 0)
-      Parallel::OutBdLow2Hi_cached(GH->PatL[lev - 1], GH->PatL[lev], SynchList_pre, SL, Symmetry, sync_cache_outbd[lev]);
+      Parallel::OutBdLow2Hi(GH->PatL[lev - 1], GH->PatL[lev], SynchList_pre, SL, Symmetry);
 #elif (MIXOUTB == 1)
       Parallel::OutBdLow2Himix(GH->PatL[lev - 1], GH->PatL[lev], SynchList_pre, SL, Symmetry);
 #endif
@@ -6024,7 +5962,7 @@ void bssn_class::RestrictProlong_aux(int lev, int YN, bool BB,
     else // no time refinement levels and for all same time levels
     {
 #if (RPB == 0)
-      Parallel::Restrict_cached(GH->PatL[lev - 1], GH->PatL[lev], SL, SL, Symmetry, sync_cache_restrict[lev]);
+      Parallel::Restrict(GH->PatL[lev - 1], GH->PatL[lev], SL, SL, Symmetry);
 #elif (RPB == 1)
       //       Parallel::Restrict_bam(GH->PatL[lev-1],GH->PatL[lev],SL,SL,Symmetry);
       Parallel::Restrict_bam(GH->PatL[lev - 1], GH->PatL[lev], SL, SL, GH->rsul[lev], Symmetry);
@@ -6034,7 +5972,7 @@ void bssn_class::RestrictProlong_aux(int lev, int YN, bool BB,
 
 #if (RPB == 0)
 #if (MIXOUTB == 0)
-      Parallel::OutBdLow2Hi_cached(GH->PatL[lev - 1], GH->PatL[lev], SL, SL, Symmetry, sync_cache_outbd[lev]);
+      Parallel::OutBdLow2Hi(GH->PatL[lev - 1], GH->PatL[lev], SL, SL, Symmetry);
 #elif (MIXOUTB == 1)
       Parallel::OutBdLow2Himix(GH->PatL[lev - 1], GH->PatL[lev], SL, SL, Symmetry);
 #endif
@@ -6089,7 +6027,7 @@ void bssn_class::RestrictProlong(int lev, int YN, bool BB)
       }
 
 #if (RPB == 0)
-      Parallel::Restrict_cached(GH->PatL[lev - 1], GH->PatL[lev], SynchList_cor, SynchList_pre, Symmetry, sync_cache_restrict[lev]);
+      Parallel::Restrict(GH->PatL[lev - 1], GH->PatL[lev], SynchList_cor, SynchList_pre, Symmetry);
 #elif (RPB == 1)
       //       Parallel::Restrict_bam(GH->PatL[lev-1],GH->PatL[lev],SynchList_cor,SynchList_pre,Symmetry);
       Parallel::Restrict_bam(GH->PatL[lev - 1], GH->PatL[lev], SynchList_cor, SynchList_pre, GH->rsul[lev], Symmetry);
@@ -6099,7 +6037,7 @@ void bssn_class::RestrictProlong(int lev, int YN, bool BB)
 
 #if (RPB == 0)
 #if (MIXOUTB == 0)
-      Parallel::OutBdLow2Hi_cached(GH->PatL[lev - 1], GH->PatL[lev], SynchList_pre, SynchList_cor, Symmetry, sync_cache_outbd[lev]);
+      Parallel::OutBdLow2Hi(GH->PatL[lev - 1], GH->PatL[lev], SynchList_pre, SynchList_cor, Symmetry);
 #elif (MIXOUTB == 1)
       Parallel::OutBdLow2Himix(GH->PatL[lev - 1], GH->PatL[lev], SynchList_pre, SynchList_cor, Symmetry);
 #endif
@@ -6113,7 +6051,7 @@ void bssn_class::RestrictProlong(int lev, int YN, bool BB)
       if (myrank == 0)
         cout << "===: " << GH->Lt[lev - 1] << "," << GH->Lt[lev] + dT_lev << endl;
 #if (RPB == 0)
-      Parallel::Restrict_cached(GH->PatL[lev - 1], GH->PatL[lev], SynchList_cor, StateList, Symmetry, sync_cache_restrict[lev]);
+      Parallel::Restrict(GH->PatL[lev - 1], GH->PatL[lev], SynchList_cor, StateList, Symmetry);
 #elif (RPB == 1)
       //       Parallel::Restrict_bam(GH->PatL[lev-1],GH->PatL[lev],SynchList_cor,StateList,Symmetry);
       Parallel::Restrict_bam(GH->PatL[lev - 1], GH->PatL[lev], SynchList_cor, StateList, GH->rsul[lev], Symmetry);
@@ -6123,7 +6061,7 @@ void bssn_class::RestrictProlong(int lev, int YN, bool BB)
 
 #if (RPB == 0)
 #if (MIXOUTB == 0)
-      Parallel::OutBdLow2Hi_cached(GH->PatL[lev - 1], GH->PatL[lev], StateList, SynchList_cor, Symmetry, sync_cache_outbd[lev]);
+      Parallel::OutBdLow2Hi(GH->PatL[lev - 1], GH->PatL[lev], StateList, SynchList_cor, Symmetry);
 #elif (MIXOUTB == 1)
       Parallel::OutBdLow2Himix(GH->PatL[lev - 1], GH->PatL[lev], StateList, SynchList_cor, Symmetry);
 #endif
@@ -6164,7 +6102,7 @@ void bssn_class::ProlongRestrict(int lev, int YN, bool BB)
 
 #if (RPB == 0)
 #if (MIXOUTB == 0)
-      Parallel::OutBdLow2Hi_cached(GH->PatL[lev - 1], GH->PatL[lev], SynchList_pre, SynchList_cor, Symmetry, sync_cache_outbd[lev]);
+      Parallel::OutBdLow2Hi(GH->PatL[lev - 1], GH->PatL[lev], SynchList_pre, SynchList_cor, Symmetry);
 #elif (MIXOUTB == 1)
       Parallel::OutBdLow2Himix(GH->PatL[lev - 1], GH->PatL[lev], SynchList_pre, SynchList_cor, Symmetry);
 #endif
@@ -6177,7 +6115,7 @@ void bssn_class::ProlongRestrict(int lev, int YN, bool BB)
     {
 #if (RPB == 0)
 #if (MIXOUTB == 0)
-      Parallel::OutBdLow2Hi_cached(GH->PatL[lev - 1], GH->PatL[lev], StateList, SynchList_cor, Symmetry, sync_cache_outbd[lev]);
+      Parallel::OutBdLow2Hi(GH->PatL[lev - 1], GH->PatL[lev], StateList, SynchList_cor, Symmetry);
 #elif (MIXOUTB == 1)
       Parallel::OutBdLow2Himix(GH->PatL[lev - 1], GH->PatL[lev], StateList, SynchList_cor, Symmetry);
 #endif
@@ -6298,7 +6236,7 @@ for(int ilev = GH->levels-1;ilev>=lev;ilev--)
 for(int ilev=GH->levels-1;ilev>lev;ilev--)
     RestrictProlong(ilev,1,false,DG_List,DG_List,DG_List);
 #else
-  Parallel::Sync_cached(GH->PatL[lev], DG_List, Symmetry, sync_cache_psi4[lev]);
+  Parallel::Sync(GH->PatL[lev], DG_List, Symmetry);
 #endif
 
 #ifdef WithShell
@@ -7325,9 +7263,6 @@ void bssn_class::Constraint_Out()
             Block *cg = BP->data;
             if (myrank == cg->rank)
             {
-#ifdef USE_GPU
-              gpu_rhs(CALLED_BY_CONSTRAINT_CONS_ONLY, myrank, RHS_PARA_CALLED_Constraint_Out);
-#else
               f_compute_rhs_bssn(cg->shape, TRK4, cg->X[0], cg->X[1], cg->X[2],
                                  cg->fgfs[phi0->sgfn], cg->fgfs[trK0->sgfn],
                                  cg->fgfs[gxx0->sgfn], cg->fgfs[gxy0->sgfn], cg->fgfs[gxz0->sgfn], 
@@ -7362,7 +7297,6 @@ void bssn_class::Constraint_Out()
                                  cg->fgfs[Cons_Px->sgfn], cg->fgfs[Cons_Py->sgfn], cg->fgfs[Cons_Pz->sgfn],
                                  cg->fgfs[Cons_Gx->sgfn], cg->fgfs[Cons_Gy->sgfn], cg->fgfs[Cons_Gz->sgfn],
                                  Symmetry, lev, ndeps, pre);
-#endif
             }
             if (BP == Pp->data->ble)
               break;
@@ -7371,7 +7305,7 @@ void bssn_class::Constraint_Out()
           Pp = Pp->next;
         }
       }
-      Parallel::Sync(GH->PatL[lev], ConstraintList, Symmetry, "bssn_class::Constraint_Out[level]");
+      Parallel::Sync(GH->PatL[lev], ConstraintList, Symmetry);
     }
 #ifdef WithShell
     if (0) // if the constrait quantities can be reused from the step rhs calculation
@@ -7593,7 +7527,7 @@ void bssn_class::AH_Prepare_derivatives()
       }
       Pp = Pp->next;
     }
-    Parallel::Sync(GH->PatL[lev], AHDList, Symmetry, "bssn_class::AH_Prepare_derivatives");
+    Parallel::Sync(GH->PatL[lev], AHDList, Symmetry);
   }
 }
 
@@ -7832,9 +7766,6 @@ void bssn_class::Interp_Constraint(bool infg)
             Block *cg = BP->data;
             if (myrank == cg->rank)
             {
-#ifdef USE_GPU
-              gpu_rhs(CALLED_BY_CONSTRAINT_CONS_ONLY, myrank, RHS_PARA_CALLED_Interp_Constraint);
-#else
               f_compute_rhs_bssn(cg->shape, TRK4, cg->X[0], cg->X[1], cg->X[2],
                                  cg->fgfs[phi0->sgfn], cg->fgfs[trK0->sgfn],
                                  cg->fgfs[gxx0->sgfn], cg->fgfs[gxy0->sgfn], cg->fgfs[gxz0->sgfn], 
@@ -7869,7 +7800,6 @@ void bssn_class::Interp_Constraint(bool infg)
                                  cg->fgfs[Cons_Px->sgfn], cg->fgfs[Cons_Py->sgfn], cg->fgfs[Cons_Pz->sgfn],
                                  cg->fgfs[Cons_Gx->sgfn], cg->fgfs[Cons_Gy->sgfn], cg->fgfs[Cons_Gz->sgfn],
                                  Symmetry, lev, ndeps, pre);
-#endif
             }
             if (BP == Pp->data->ble)
               break;
@@ -7878,7 +7808,7 @@ void bssn_class::Interp_Constraint(bool infg)
           Pp = Pp->next;
         }
       }
-      Parallel::Sync(GH->PatL[lev], ConstraintList, Symmetry, "bssn_class::Interp_Constraint[level]");
+      Parallel::Sync(GH->PatL[lev], ConstraintList, Symmetry);
     }
 #ifdef WithShell
     if (0) // if the constrait quantities can be reused from the step rhs calculation
@@ -8136,7 +8066,7 @@ void bssn_class::Compute_Constraint()
         Pp = Pp->next;
       }
     }
-    Parallel::Sync(GH->PatL[lev], ConstraintList, Symmetry, "bssn_class::Compute_Constraint[level]");
+    Parallel::Sync(GH->PatL[lev], ConstraintList, Symmetry);
   }
   // prolong restrict constraint quantities
   for (lev = GH->levels - 1; lev > 0; lev--)
@@ -8451,12 +8381,6 @@ void bssn_class::Enforce_algcon(int lev, int fg)
 
 bool bssn_class::check_Stdin_Abort() 
 {
-    // Non-interactive launches (mpirun via Python/subprocess, batch jobs, redirected stdin)
-    // should not probe stdin. Some MPI runtimes treat stdin as a managed channel and can
-    // fail when rank 0 polls/consumes it.
-    if (!isatty(STDIN_FILENO)) {
-        return false;
-    }
 
     fd_set readfds;
 
@@ -8470,11 +8394,8 @@ bool bssn_class::check_Stdin_Abort()
     timeout.tv_usec = 0;
 
     int activity = select(STDIN_FILENO + 1, &readfds, nullptr, nullptr, &timeout);
-    if (activity <= 0) {
-        return false;
-    }
 
-    if (FD_ISSET(STDIN_FILENO, &readfds)) {
+    if (activity > 0 && FD_ISSET(STDIN_FILENO, &readfds)) {
         string input_abort;
         if (cin >> input_abort) {
             if (input_abort == "stop") {

AMSS_NCKU_source/bssn_class.h

@@ -130,9 +130,6 @@ public:
        Parallel::SyncCache *sync_cache_cor;  // per-level cache for corrector sync
        Parallel::SyncCache *sync_cache_rp_coarse;  // RestrictProlong sync on PatL[lev-1]
        Parallel::SyncCache *sync_cache_rp_fine;    // RestrictProlong sync on PatL[lev]
-       Parallel::SyncCache *sync_cache_restrict;   // cached Restrict in RestrictProlong
-       Parallel::SyncCache *sync_cache_outbd;      // cached OutBdLow2Hi in RestrictProlong
-       Parallel::SyncCache *sync_cache_psi4;       // cached Psi4 sync on PatL[lev]
 
        monitor *ErrorMonitor, *Psi4Monitor, *BHMonitor, *MAPMonitor;
        monitor *ConVMonitor;
@@ -177,12 +174,8 @@ public:
        virtual void Initialize();
        virtual void Read_Ansorg();
        virtual void Read_Pablo() {};
-       void InvalidateSyncCaches();
        virtual void Compute_Psi4(int lev);
        virtual void Step(int lev, int YN);
-#ifdef USE_GPU
-       void Step_MainPath_GPU(int lev, int YN);
-#endif
        virtual void Interp_Constraint(bool infg);
        virtual void Constraint_Out();
        virtual void Compute_Constraint();

AMSS_NCKU_source/bssn_cuda_ops.h

@@ -1,68 +0,0 @@
-#ifndef BSSN_CUDA_OPS_H
-#define BSSN_CUDA_OPS_H
-
-int bssn_cuda_enforce_ga(int *ex,
-                         double *dxx, double *gxy, double *gxz,
-                         double *dyy, double *gyz, double *dzz,
-                         double *Axx, double *Axy, double *Axz,
-                         double *Ayy, double *Ayz, double *Azz);
-
-int bssn_cuda_rk4_boundary_var(int *ex, double dT,
-                               const double *X, const double *Y, const double *Z,
-                               double xmin, double ymin, double zmin,
-                               double xmax, double ymax, double zmax,
-                               const double *state0,
-                               const double *phi_field,
-                               const double *lap_field,
-                               const double *boundary_src,
-                               double *stage_data,
-                               double *rhs_accum,
-                               double propspeed,
-                               const double SoA[3],
-                               int symmetry,
-                               int lev,
-                               int rk_stage,
-                               bool force_host_boundary_fix,
-                               bool download_to_host = true);
-
-int bssn_cuda_rk4_boundary_batch(int *ex, double dT,
-                                 const double *X, const double *Y, const double *Z,
-                                 double xmin, double ymin, double zmin,
-                                 double xmax, double ymax, double zmax,
-                                 int symmetry,
-                                 const double *const *state0_list,
-                                 double *const *stage_data_list,
-                                 double *const *rhs_accum_list,
-                                 int num_var,
-                                 int rk_stage,
-                                 bool download_to_host = false);
-
-int bssn_cuda_lowerbound(int *ex, double *chi, double tinny, bool download_to_host = true);
-int bssn_cuda_download_buffer(int *ex, double *host_ptr);
-void bssn_cuda_release_rk4_caches();
-void bssn_cuda_release_interp_caches();
-
-int bssn_cuda_prolong3_pack(int wei,
-                            const double *llbc, const double *uubc, const int *extc, const double *func,
-                            const double *llbf, const double *uubf, const int *extf, double *funf,
-                            const double *llbp, const double *uubp,
-                            const double *SoA, int symmetry);
-
-int bssn_cuda_restrict3_pack(int wei,
-                             const double *llbc, const double *uubc, const int *extc, double *func,
-                             const double *llbf, const double *uubf, const int *extf, const double *funf,
-                             const double *llbr, const double *uubr,
-                             const double *SoA, int symmetry);
-
-int bssn_cuda_interp_points_batch(const int *ex,
-                                  const double *X, const double *Y, const double *Z,
-                                  const double *const *fields,
-                                  const double *soa_flat,
-                                  int num_var,
-                                  const double *px, const double *py, const double *pz,
-                                  int num_points,
-                                  int ordn,
-                                  int symmetry,
-                                  double *out);
-
-#endif

AMSS_NCKU_source/bssn_cuda_step.C

@@ -1,936 +0,0 @@
-#include "macrodef.h"
-
-#ifdef USE_GPU
-
-#include <algorithm>
-#include <cmath>
-#include <cstring>
-#include <cstdlib>
-#include <iomanip>
-#include <vector>
-
-#include "bssn_class.h"
-#include "bssn_cuda_ops.h"
-#include "bssn_gpu.h"
-#include "bssn_macro.h"
-
-namespace
-{
-enum StageProfileMetric
-{
-  STAGE_PROFILE_TOTAL = 0,
-  STAGE_PROFILE_RHS,
-  STAGE_PROFILE_RUN_STAGE,
-  STAGE_PROFILE_RUN_STAGE_DEVICE,
-  STAGE_PROFILE_RUN_STAGE_HOST_FIX,
-  STAGE_PROFILE_LOWERBOUND,
-  STAGE_PROFILE_ENSURE,
-  STAGE_PROFILE_DOWNLOAD,
-  STAGE_PROFILE_CLEAR_CACHE,
-  STAGE_PROFILE_SYNC_START,
-  STAGE_PROFILE_SYNC_FINISH,
-  STAGE_PROFILE_REFRESH,
-  STAGE_PROFILE_COUNT
-};
-
-static const int kStageProfileMaxLevels = 32;
-
-struct StageProfileStore
-{
-  bool env_checked;
-  bool enabled;
-  int calls[kStageProfileMaxLevels];
-  double metric[kStageProfileMaxLevels][STAGE_PROFILE_COUNT];
-};
-
-StageProfileStore &stage_profile_store()
-{
-  static StageProfileStore store = {};
-  return store;
-}
-
-bool stage_profile_enabled()
-{
-  StageProfileStore &store = stage_profile_store();
-  if (!store.env_checked)
-  {
-    const char *env = getenv("AMSS_GPU_STAGE_TIMING");
-    store.enabled = (env && env[0] && strcmp(env, "0") != 0);
-    store.env_checked = true;
-  }
-  return store.enabled;
-}
-
-void stage_profile_note_call(int lev)
-{
-  if (lev >= 0 && lev < kStageProfileMaxLevels)
-    stage_profile_store().calls[lev]++;
-}
-
-void stage_profile_add(int lev, StageProfileMetric metric, double seconds)
-{
-  if (lev >= 0 && lev < kStageProfileMaxLevels)
-    stage_profile_store().metric[lev][metric] += seconds;
-}
-
-const char *stage_profile_metric_name(StageProfileMetric metric)
-{
-  switch (metric)
-  {
-  case STAGE_PROFILE_TOTAL:
-    return "total";
-  case STAGE_PROFILE_RHS:
-    return "rhs";
-  case STAGE_PROFILE_RUN_STAGE:
-    return "run_stage";
-  case STAGE_PROFILE_RUN_STAGE_DEVICE:
-    return "run_stage_dev";
-  case STAGE_PROFILE_RUN_STAGE_HOST_FIX:
-    return "run_stage_host";
-  case STAGE_PROFILE_LOWERBOUND:
-    return "lower";
-  case STAGE_PROFILE_ENSURE:
-    return "ensure";
-  case STAGE_PROFILE_DOWNLOAD:
-    return "download";
-  case STAGE_PROFILE_CLEAR_CACHE:
-    return "clear_cache";
-  case STAGE_PROFILE_SYNC_START:
-    return "sync_start";
-  case STAGE_PROFILE_SYNC_FINISH:
-    return "sync_finish";
-  case STAGE_PROFILE_REFRESH:
-    return "refresh";
-  default:
-    return "unknown";
-  }
-}
-} // namespace
-
-void bssn_cuda_dump_stage_profile()
-{
-  if (!stage_profile_enabled())
-    return;
-
-  int myrank = 0;
-  MPI_Comm_rank(MPI_COMM_WORLD, &myrank);
-
-  StageProfileStore &store = stage_profile_store();
-  int global_calls_sum[kStageProfileMaxLevels] = {};
-  double global_metric_sum[kStageProfileMaxLevels][STAGE_PROFILE_COUNT] = {};
-  double global_metric_max[kStageProfileMaxLevels][STAGE_PROFILE_COUNT] = {};
-
-  MPI_Reduce(store.calls, global_calls_sum, kStageProfileMaxLevels, MPI_INT, MPI_SUM, 0, MPI_COMM_WORLD);
-  MPI_Reduce(store.metric[0], global_metric_sum[0],
-             kStageProfileMaxLevels * STAGE_PROFILE_COUNT,
-             MPI_DOUBLE, MPI_SUM, 0, MPI_COMM_WORLD);
-  MPI_Reduce(store.metric[0], global_metric_max[0],
-             kStageProfileMaxLevels * STAGE_PROFILE_COUNT,
-             MPI_DOUBLE, MPI_MAX, 0, MPI_COMM_WORLD);
-
-  if (myrank != 0)
-    return;
-
-  cout << endl;
-  cout << " GPU stage timing summary (sum/max over MPI ranks) " << endl;
-  cout << " lev  calls";
-  for (int metric = 0; metric < STAGE_PROFILE_COUNT; ++metric)
-    cout << "  " << setw(22) << stage_profile_metric_name(static_cast<StageProfileMetric>(metric));
-  cout << endl;
-
-  for (int lev = 0; lev < kStageProfileMaxLevels; ++lev)
-  {
-    if (global_calls_sum[lev] == 0)
-      continue;
-
-    cout << setw(4) << lev << "  " << setw(5) << global_calls_sum[lev];
-    for (int metric = 0; metric < STAGE_PROFILE_COUNT; ++metric)
-    {
-      cout << "  "
-           << setw(10) << setprecision(6) << fixed << global_metric_sum[lev][metric]
-           << "/"
-           << setw(10) << setprecision(6) << fixed << global_metric_max[lev][metric];
-    }
-    cout << endl;
-  }
-  cout << endl;
-}
-
-void bssn_class::Step_MainPath_GPU(int lev, int YN)
-{
-#ifdef WithShell
-#error "Step_MainPath_GPU currently supports Patch grids only."
-#endif
-
-  const bool profile_enabled = stage_profile_enabled();
-  const double step_total_begin = profile_enabled ? MPI_Wtime() : 0.0;
-  if (profile_enabled)
-    stage_profile_note_call(lev);
-
-  if (bssn_gpu_bind_process_device(myrank))
-  {
-    cerr << "GPU device bind failure on MPI rank " << myrank << endl;
-    MPI_Abort(MPI_COMM_WORLD, 1);
-  }
-  if (profile_enabled)
-  {
-    const double t0 = MPI_Wtime();
-    bssn_gpu_clear_cached_device_buffers();
-    stage_profile_add(lev, STAGE_PROFILE_CLEAR_CACHE, MPI_Wtime() - t0);
-  }
-  else
-    bssn_gpu_clear_cached_device_buffers();
-
-  setpbh(BH_num, Porg0, Mass, BH_num_input);
-
-  const double dT_lev = dT * pow(0.5, Mymax(lev, trfls));
-
-#if (MAPBH == 1)
-  if (BH_num > 0 && lev == GH->levels - 1)
-  {
-    compute_Porg_rhs(Porg0, Porg_rhs, Sfx0, Sfy0, Sfz0, lev);
-    for (int ithBH = 0; ithBH < BH_num; ithBH++)
-    {
-      for (int ith = 0; ith < 3; ith++)
-        Porg1[ithBH][ith] = Porg0[ithBH][ith] + Porg_rhs[ithBH][ith] * dT_lev;
-      if (Symmetry > 0)
-        Porg1[ithBH][2] = fabs(Porg1[ithBH][2]);
-      if (Symmetry == 2)
-      {
-        Porg1[ithBH][0] = fabs(Porg1[ithBH][0]);
-        Porg1[ithBH][1] = fabs(Porg1[ithBH][1]);
-      }
-    }
-  }
-
-  if (lev == a_lev)
-    AnalysisStuff(lev, dT_lev);
-#endif
-
-#ifdef With_AHF
-  AH_Step_Find(lev, dT_lev);
-#endif
-
-  const bool BB = fgt(PhysTime, StartTime, dT_lev / 2);
-  (void)BB;
-  double ndeps = (lev < GH->movls) ? numepsb : numepss;
-  double TRK4 = PhysTime;
-  int iter_count = 0;
-  int pre = 0, cor = 1;
-  int ERROR = 0;
-  const bool keep_stage_sync_on_device = (RPS == 1) && (MAPBH == 1) && (REGLEV == 0);
-
-  auto run_stage_on_block =
-      [&](Block *cg, Patch *patch, MyList<var> *state0_list,
-          MyList<var> *boundary_src_list, MyList<var> *stage_data_list,
-          MyList<var> *rhs_list, int rk_stage) {
-        MyList<var> *varl0 = state0_list;
-        MyList<var> *varlb = boundary_src_list;
-        MyList<var> *varls = stage_data_list;
-        MyList<var> *varlr = rhs_list;
-        std::vector<const double *> batch_state0;
-        std::vector<double *> batch_stage;
-        std::vector<double *> batch_rhs;
-
-        while (varl0)
-        {
-          const bool force_host_boundary_fix = false;
-          const bool can_batch_device_path = (lev > 0) && !force_host_boundary_fix;
-          if (can_batch_device_path)
-          {
-            batch_state0.push_back(cg->fgfs[varl0->data->sgfn]);
-            batch_stage.push_back(cg->fgfs[varls->data->sgfn]);
-            batch_rhs.push_back(cg->fgfs[varlr->data->sgfn]);
-            varl0 = varl0->next;
-            varlb = varlb->next;
-            varls = varls->next;
-            varlr = varlr->next;
-            continue;
-          }
-
-          const double var_begin = profile_enabled ? MPI_Wtime() : 0.0;
-          if (bssn_cuda_rk4_boundary_var(cg->shape, dT_lev,
-                                         cg->X[0], cg->X[1], cg->X[2],
-                                         patch->bbox[0], patch->bbox[1], patch->bbox[2],
-                                         patch->bbox[3], patch->bbox[4], patch->bbox[5],
-                                         cg->fgfs[varl0->data->sgfn],
-                                         cg->fgfs[phi0->sgfn],
-                                         cg->fgfs[Lap0->sgfn],
-                                         cg->fgfs[varlb->data->sgfn],
-                                         cg->fgfs[varls->data->sgfn],
-                                         cg->fgfs[varlr->data->sgfn],
-                                         varl0->data->propspeed,
-                                         varl0->data->SoA,
-                                         Symmetry, lev, rk_stage,
-                                         force_host_boundary_fix, false))
-          {
-            cerr << "GPU rk4/boundary failure: lev=" << lev
-                 << " rk_stage=" << rk_stage
-                 << " var=" << varl0->data->name
-                 << " bbox=(" << cg->bbox[0] << ":" << cg->bbox[3] << ","
-                 << cg->bbox[1] << ":" << cg->bbox[4] << ","
-                 << cg->bbox[2] << ":" << cg->bbox[5] << ")" << endl;
-            ERROR = 1;
-            break;
-          }
-          if (profile_enabled)
-          {
-            stage_profile_add(lev,
-                              force_host_boundary_fix ? STAGE_PROFILE_RUN_STAGE_HOST_FIX
-                                                      : STAGE_PROFILE_RUN_STAGE_DEVICE,
-                              MPI_Wtime() - var_begin);
-          }
-          varl0 = varl0->next;
-          varlb = varlb->next;
-          varls = varls->next;
-          varlr = varlr->next;
-        }
-
-        if (!ERROR && !batch_state0.empty())
-        {
-          const double batch_begin = profile_enabled ? MPI_Wtime() : 0.0;
-          if (bssn_cuda_rk4_boundary_batch(cg->shape, dT_lev,
-                                           cg->X[0], cg->X[1], cg->X[2],
-                                           patch->bbox[0], patch->bbox[1], patch->bbox[2],
-                                           patch->bbox[3], patch->bbox[4], patch->bbox[5],
-                                           Symmetry,
-                                           &batch_state0[0],
-                                           &batch_stage[0],
-                                           &batch_rhs[0],
-                                           static_cast<int>(batch_state0.size()),
-                                           rk_stage, false))
-          {
-            cerr << "GPU rk4/boundary batch failure: lev=" << lev
-                 << " rk_stage=" << rk_stage
-                 << " vars=" << batch_state0.size()
-                 << " bbox=(" << cg->bbox[0] << ":" << cg->bbox[3] << ","
-                 << cg->bbox[1] << ":" << cg->bbox[4] << ","
-                 << cg->bbox[2] << ":" << cg->bbox[5] << ")" << endl;
-            ERROR = 1;
-          }
-          else if (profile_enabled)
-          {
-            stage_profile_add(lev, STAGE_PROFILE_RUN_STAGE_DEVICE, MPI_Wtime() - batch_begin);
-          }
-        }
-      };
-
-  auto stage_download_var_list =
-      [&](Block *cg, MyList<var> *var_list, bool skip_unmapped) {
-        std::vector<double *> batch_host_ptrs;
-        std::vector<MyList<var> *> batch_vars;
-        while (var_list)
-        {
-          double *host_ptr = cg->fgfs[var_list->data->sgfn];
-          if (skip_unmapped && !bssn_gpu_find_device_buffer(host_ptr))
-          {
-            var_list = var_list->next;
-            continue;
-          }
-          batch_host_ptrs.push_back(host_ptr);
-          batch_vars.push_back(var_list);
-          var_list = var_list->next;
-        }
-        if (!batch_host_ptrs.empty() &&
-            bssn_gpu_download_buffer_batch(cg->shape, &batch_host_ptrs[0],
-                                           static_cast<int>(batch_host_ptrs.size())))
-        {
-          for (size_t i = 0; i < batch_host_ptrs.size(); ++i)
-          {
-            if (bssn_cuda_download_buffer(cg->shape, batch_host_ptrs[i]))
-            {
-              cerr << "GPU stage download failure: lev=" << lev
-                   << " var=" << batch_vars[i]->data->name
-                   << " bbox=(" << cg->bbox[0] << ":" << cg->bbox[3] << ","
-                   << cg->bbox[1] << ":" << cg->bbox[4] << ","
-                   << cg->bbox[2] << ":" << cg->bbox[5] << ")" << endl;
-              ERROR = 1;
-              break;
-            }
-          }
-        }
-      };
-
-  auto stage_download_patch_list =
-      [&](MyList<var> *var_list, bool skip_unmapped) {
-        MyList<Patch> *patch_it = GH->PatL[lev];
-        while (patch_it)
-        {
-          MyList<Block> *block_it = patch_it->data->blb;
-          while (block_it)
-          {
-            Block *cg = block_it->data;
-            if (myrank == cg->rank)
-              stage_download_var_list(cg, var_list, skip_unmapped);
-
-            if (block_it == patch_it->data->ble)
-              break;
-            block_it = block_it->next;
-          }
-          if (ERROR)
-            break;
-          patch_it = patch_it->next;
-        }
-      };
-
-  auto ensure_stage_device_var_list =
-      [&](Block *cg, MyList<var> *var_list) {
-        const int n = cg->shape[0] * cg->shape[1] * cg->shape[2];
-        while (var_list)
-        {
-          double *host_ptr = cg->fgfs[var_list->data->sgfn];
-          if (!bssn_gpu_find_device_buffer(host_ptr) &&
-              bssn_gpu_stage_upload_buffer(host_ptr, n))
-          {
-            cerr << "GPU state ensure failure: lev=" << lev
-                 << " var=" << var_list->data->name
-                 << " bbox=(" << cg->bbox[0] << ":" << cg->bbox[3] << ","
-                 << cg->bbox[1] << ":" << cg->bbox[4] << ","
-                 << cg->bbox[2] << ":" << cg->bbox[5] << ")" << endl;
-            ERROR = 1;
-            break;
-          }
-          var_list = var_list->next;
-        }
-      };
-
-  auto refresh_synced_device_regions =
-      [&](Block *cg, MyList<var> *var_list, Parallel::SyncCache &cache) {
-        std::vector<Parallel::gridseg *> local_segments;
-        for (int node = 0; node < cache.cpusize; ++node)
-        {
-          MyList<Parallel::gridseg> *seg = cache.combined_dst[node];
-          while (seg)
-          {
-            if (seg->data && seg->data->Bg == cg)
-              local_segments.push_back(seg->data);
-            seg = seg->next;
-          }
-        }
-
-        if (local_segments.empty())
-          return;
-
-        const int n = cg->shape[0] * cg->shape[1] * cg->shape[2];
-        while (var_list)
-        {
-          double *host_ptr = cg->fgfs[var_list->data->sgfn];
-          if (!bssn_gpu_find_device_buffer(host_ptr))
-          {
-            if (bssn_gpu_stage_upload_buffer(host_ptr, n))
-            {
-              cerr << "GPU sync refresh upload failure: lev=" << lev
-                   << " var=" << var_list->data->name
-                   << " bbox=(" << cg->bbox[0] << ":" << cg->bbox[3] << ","
-                   << cg->bbox[1] << ":" << cg->bbox[4] << ","
-                   << cg->bbox[2] << ":" << cg->bbox[5] << ")" << endl;
-              ERROR = 1;
-              break;
-            }
-          }
-          else
-          {
-            for (size_t i = 0; i < local_segments.size(); ++i)
-            {
-              Parallel::gridseg *seg = local_segments[i];
-              if (bssn_gpu_stage_upload_region(host_ptr,
-                                               cg->shape,
-                                               cg->bbox,
-                                               cg->bbox + dim,
-                                               seg->shape,
-                                               seg->llb))
-              {
-                cerr << "GPU sync region refresh failure: lev=" << lev
-                     << " var=" << var_list->data->name
-                     << " bbox=(" << cg->bbox[0] << ":" << cg->bbox[3] << ","
-                     << cg->bbox[1] << ":" << cg->bbox[4] << ","
-                     << cg->bbox[2] << ":" << cg->bbox[5] << ")" << endl;
-                ERROR = 1;
-                break;
-              }
-            }
-            if (ERROR)
-              break;
-          }
-          var_list = var_list->next;
-        }
-      };
-
-  auto refresh_stage_device_after_sync =
-      [&](MyList<var> *var_list, Parallel::SyncCache &cache) {
-        MyList<Patch> *patch_it = GH->PatL[lev];
-        while (patch_it)
-        {
-          MyList<Block> *block_it = patch_it->data->blb;
-          while (block_it)
-          {
-            Block *cg = block_it->data;
-            if (myrank == cg->rank)
-              refresh_synced_device_regions(cg, var_list, cache);
-
-            if (block_it == patch_it->data->ble)
-              break;
-            block_it = block_it->next;
-          }
-          if (ERROR)
-            break;
-          patch_it = patch_it->next;
-        }
-      };
-
-  auto refresh_stage_host_before_sync =
-      [&](MyList<var> *var_list, Parallel::SyncCache &cache) -> bool {
-        if (!cache.valid || !cache.combined_src || myrank < 0 || myrank >= cache.cpusize)
-          return false;
-
-        MyList<Patch> *patch_it = GH->PatL[lev];
-        while (patch_it)
-        {
-          MyList<Block> *block_it = patch_it->data->blb;
-          while (block_it)
-          {
-            Block *cg = block_it->data;
-            if (myrank == cg->rank)
-            {
-              std::vector<Parallel::gridseg *> local_segments;
-              MyList<Parallel::gridseg> *seg = cache.combined_src[myrank];
-              while (seg)
-              {
-                if (seg->data && seg->data->Bg == cg)
-                  local_segments.push_back(seg->data);
-                seg = seg->next;
-              }
-
-              if (!local_segments.empty())
-              {
-                MyList<var> *var_it = var_list;
-                while (var_it)
-                {
-                  double *host_ptr = cg->fgfs[var_it->data->sgfn];
-                  for (size_t i = 0; i < local_segments.size(); ++i)
-                  {
-                    Parallel::gridseg *src_seg = local_segments[i];
-                    if (bssn_gpu_stage_download_region(host_ptr,
-                                                       cg->shape,
-                                                       cg->bbox,
-                                                       cg->bbox + dim,
-                                                       src_seg->shape,
-                                                       src_seg->llb))
-                    {
-                      cerr << "GPU sync region download failure: lev=" << lev
-                           << " var=" << var_it->data->name
-                           << " bbox=(" << cg->bbox[0] << ":" << cg->bbox[3] << ","
-                           << cg->bbox[1] << ":" << cg->bbox[4] << ","
-                           << cg->bbox[2] << ":" << cg->bbox[5] << ")" << endl;
-                      ERROR = 1;
-                      return true;
-                    }
-                  }
-                  var_it = var_it->next;
-                }
-              }
-            }
-
-            if (block_it == patch_it->data->ble)
-              break;
-            block_it = block_it->next;
-          }
-          patch_it = patch_it->next;
-        }
-
-        return true;
-      };
-
-  auto can_pack_sync_from_device =
-      [&](MyList<var> *var_list, Parallel::SyncCache &cache) -> bool {
-        if (!cache.valid || !cache.combined_src || myrank < 0 || myrank >= cache.cpusize)
-          return false;
-
-        MyList<Parallel::gridseg> *seg = cache.combined_src[myrank];
-        while (seg)
-        {
-          MyList<var> *var_it = var_list;
-          while (var_it)
-          {
-            if (!bssn_gpu_find_device_buffer(seg->data->Bg->fgfs[var_it->data->sgfn]))
-              return false;
-            var_it = var_it->next;
-          }
-          seg = seg->next;
-        }
-        return true;
-      };
-
-  MyList<Patch> *Pp = GH->PatL[lev];
-  while (Pp)
-  {
-    MyList<Block> *BP = Pp->data->blb;
-    while (BP)
-    {
-      Block *cg = BP->data;
-      if (myrank == cg->rank)
-      {
-        double t0 = 0.0;
-        if (profile_enabled)
-          t0 = MPI_Wtime();
-        if (gpu_rhs(CALLED_BY_STEP, myrank, RHS_PARA_CALLED_FIRST_TIME))
-          ERROR = 1;
-        if (profile_enabled)
-          stage_profile_add(lev, STAGE_PROFILE_RHS, MPI_Wtime() - t0);
-
-        if (profile_enabled)
-          t0 = MPI_Wtime();
-        run_stage_on_block(cg, Pp->data, StateList, StateList, SynchList_pre, RHSList, iter_count);
-        if (profile_enabled)
-          stage_profile_add(lev, STAGE_PROFILE_RUN_STAGE, MPI_Wtime() - t0);
-
-        if (profile_enabled)
-          t0 = MPI_Wtime();
-        if (bssn_cuda_lowerbound(cg->shape, cg->fgfs[phi->sgfn], chitiny, false))
-        {
-          cerr << "GPU lowerbound failure: lev=" << lev
-               << " rk_stage=" << iter_count
-               << " var=" << phi->name
-               << " bbox=(" << cg->bbox[0] << ":" << cg->bbox[3] << ","
-               << cg->bbox[1] << ":" << cg->bbox[4] << ","
-               << cg->bbox[2] << ":" << cg->bbox[5] << ")" << endl;
-          ERROR = 1;
-        }
-        if (profile_enabled)
-          stage_profile_add(lev, STAGE_PROFILE_LOWERBOUND, MPI_Wtime() - t0);
-      }
-      if (BP == Pp->data->ble)
-        break;
-      BP = BP->next;
-    }
-    Pp = Pp->next;
-  }
-
-  if (!ERROR)
-  {
-    if (!keep_stage_sync_on_device)
-    {
-      double t0 = 0.0;
-      if (profile_enabled)
-        t0 = MPI_Wtime();
-      stage_download_patch_list(SynchList_pre, false);
-      if (profile_enabled)
-        stage_profile_add(lev, STAGE_PROFILE_DOWNLOAD, MPI_Wtime() - t0);
-      if (!ERROR)
-      {
-        if (profile_enabled)
-          t0 = MPI_Wtime();
-        bssn_gpu_clear_cached_device_buffers();
-        if (profile_enabled)
-          stage_profile_add(lev, STAGE_PROFILE_CLEAR_CACHE, MPI_Wtime() - t0);
-      }
-    }
-  }
-
-  MPI_Request err_req_pre;
-  {
-    int erh = ERROR;
-    MPI_Iallreduce(&erh, &ERROR, 1, MPI_INT, MPI_SUM, MPI_COMM_WORLD, &err_req_pre);
-  }
-
-  Parallel::AsyncSyncState async_pre;
-  if (profile_enabled)
-  {
-    const double t0 = MPI_Wtime();
-    Parallel::Sync_start(GH->PatL[lev], SynchList_pre, Symmetry, sync_cache_pre[lev], async_pre);
-    stage_profile_add(lev, STAGE_PROFILE_SYNC_START, MPI_Wtime() - t0);
-  }
-  else
-    Parallel::Sync_start(GH->PatL[lev], SynchList_pre, Symmetry, sync_cache_pre[lev], async_pre);
-  if (profile_enabled)
-  {
-    const double t0 = MPI_Wtime();
-    Parallel::Sync_finish(sync_cache_pre[lev], async_pre, SynchList_pre, Symmetry,
-                          !keep_stage_sync_on_device);
-    stage_profile_add(lev, STAGE_PROFILE_SYNC_FINISH, MPI_Wtime() - t0);
-  }
-  else
-    Parallel::Sync_finish(sync_cache_pre[lev], async_pre, SynchList_pre, Symmetry,
-                          !keep_stage_sync_on_device);
-  if (!ERROR && !keep_stage_sync_on_device)
-  {
-    if (profile_enabled)
-    {
-      const double t0 = MPI_Wtime();
-      refresh_stage_device_after_sync(SynchList_pre, sync_cache_pre[lev]);
-      stage_profile_add(lev, STAGE_PROFILE_REFRESH, MPI_Wtime() - t0);
-    }
-    else
-      refresh_stage_device_after_sync(SynchList_pre, sync_cache_pre[lev]);
-  }
-
-  MPI_Wait(&err_req_pre, MPI_STATUS_IGNORE);
-  if (ERROR)
-  {
-    Parallel::Dump_Data(GH->PatL[lev], StateList, 0, PhysTime, dT_lev);
-    if (myrank == 0)
-    {
-      if (ErrorMonitor->outfile)
-        ErrorMonitor->outfile << "find NaN in state variables at t = " << PhysTime
-                              << ", lev = " << lev << endl;
-      MPI_Abort(MPI_COMM_WORLD, 1);
-    }
-  }
-
-#if (MAPBH == 0)
-  if (BH_num > 0 && lev == GH->levels - 1)
-  {
-    compute_Porg_rhs(Porg0, Porg_rhs, Sfx0, Sfy0, Sfz0, lev);
-    for (int ithBH = 0; ithBH < BH_num; ithBH++)
-    {
-      f_rungekutta4_scalar(dT_lev, Porg0[ithBH][0], Porg[ithBH][0], Porg_rhs[ithBH][0], iter_count);
-      f_rungekutta4_scalar(dT_lev, Porg0[ithBH][1], Porg[ithBH][1], Porg_rhs[ithBH][1], iter_count);
-      f_rungekutta4_scalar(dT_lev, Porg0[ithBH][2], Porg[ithBH][2], Porg_rhs[ithBH][2], iter_count);
-      if (Symmetry > 0)
-        Porg[ithBH][2] = fabs(Porg[ithBH][2]);
-      if (Symmetry == 2)
-      {
-        Porg[ithBH][0] = fabs(Porg[ithBH][0]);
-        Porg[ithBH][1] = fabs(Porg[ithBH][1]);
-      }
-    }
-  }
-
-  if (lev == a_lev)
-    AnalysisStuff(lev, dT_lev);
-#endif
-
-  for (iter_count = 1; iter_count < 4; iter_count++)
-  {
-    if (iter_count == 1 || iter_count == 3)
-      TRK4 += dT_lev / 2;
-
-    Pp = GH->PatL[lev];
-    while (Pp)
-    {
-      MyList<Block> *BP = Pp->data->blb;
-      while (BP)
-      {
-        Block *cg = BP->data;
-        if (myrank == cg->rank)
-        {
-          double t0 = 0.0;
-          if (profile_enabled)
-            t0 = MPI_Wtime();
-          ensure_stage_device_var_list(cg, SynchList_pre);
-          if (profile_enabled)
-            stage_profile_add(lev, STAGE_PROFILE_ENSURE, MPI_Wtime() - t0);
-
-          if (profile_enabled)
-            t0 = MPI_Wtime();
-          if (gpu_rhs(CALLED_BY_STEP, myrank, RHS_PARA_CALLED_THEN))
-            ERROR = 1;
-          if (profile_enabled)
-            stage_profile_add(lev, STAGE_PROFILE_RHS, MPI_Wtime() - t0);
-
-          if (profile_enabled)
-            t0 = MPI_Wtime();
-          run_stage_on_block(cg, Pp->data, StateList, SynchList_pre, SynchList_cor, RHSList, iter_count);
-          if (profile_enabled)
-            stage_profile_add(lev, STAGE_PROFILE_RUN_STAGE, MPI_Wtime() - t0);
-
-          if (profile_enabled)
-            t0 = MPI_Wtime();
-          if (bssn_cuda_lowerbound(cg->shape, cg->fgfs[phi1->sgfn], chitiny, false))
-          {
-          cerr << "GPU lowerbound failure: lev=" << lev
-               << " rk_stage=" << iter_count
-               << " var=" << phi1->name
-                 << " bbox=(" << cg->bbox[0] << ":" << cg->bbox[3] << ","
-                 << cg->bbox[1] << ":" << cg->bbox[4] << ","
-                 << cg->bbox[2] << ":" << cg->bbox[5] << ")" << endl;
-            ERROR = 1;
-          }
-          if (profile_enabled)
-            stage_profile_add(lev, STAGE_PROFILE_LOWERBOUND, MPI_Wtime() - t0);
-        }
-
-        if (BP == Pp->data->ble)
-          break;
-        BP = BP->next;
-      }
-      Pp = Pp->next;
-    }
-
-    if (!ERROR)
-    {
-      if (!keep_stage_sync_on_device)
-      {
-        double t0 = 0.0;
-        if (profile_enabled)
-          t0 = MPI_Wtime();
-        stage_download_patch_list(SynchList_cor, false);
-        if (profile_enabled)
-          stage_profile_add(lev, STAGE_PROFILE_DOWNLOAD, MPI_Wtime() - t0);
-        if (!ERROR)
-        {
-          if (profile_enabled)
-            t0 = MPI_Wtime();
-          bssn_gpu_clear_cached_device_buffers();
-          if (profile_enabled)
-            stage_profile_add(lev, STAGE_PROFILE_CLEAR_CACHE, MPI_Wtime() - t0);
-        }
-      }
-    }
-
-    MPI_Request err_req_cor;
-    {
-      int erh = ERROR;
-      MPI_Iallreduce(&erh, &ERROR, 1, MPI_INT, MPI_SUM, MPI_COMM_WORLD, &err_req_cor);
-    }
-
-    Parallel::AsyncSyncState async_cor;
-    if (profile_enabled)
-    {
-      const double t0 = MPI_Wtime();
-      Parallel::Sync_start(GH->PatL[lev], SynchList_cor, Symmetry, sync_cache_cor[lev], async_cor);
-      stage_profile_add(lev, STAGE_PROFILE_SYNC_START, MPI_Wtime() - t0);
-    }
-    else
-      Parallel::Sync_start(GH->PatL[lev], SynchList_cor, Symmetry, sync_cache_cor[lev], async_cor);
-    if (profile_enabled)
-    {
-      const double t0 = MPI_Wtime();
-      Parallel::Sync_finish(sync_cache_cor[lev], async_cor, SynchList_cor, Symmetry,
-                            !keep_stage_sync_on_device);
-      stage_profile_add(lev, STAGE_PROFILE_SYNC_FINISH, MPI_Wtime() - t0);
-    }
-    else
-      Parallel::Sync_finish(sync_cache_cor[lev], async_cor, SynchList_cor, Symmetry,
-                            !keep_stage_sync_on_device);
-    if (!ERROR && !keep_stage_sync_on_device && iter_count < 3)
-    {
-      if (profile_enabled)
-      {
-        const double t0 = MPI_Wtime();
-        refresh_stage_device_after_sync(SynchList_cor, sync_cache_cor[lev]);
-        stage_profile_add(lev, STAGE_PROFILE_REFRESH, MPI_Wtime() - t0);
-      }
-      else
-        refresh_stage_device_after_sync(SynchList_cor, sync_cache_cor[lev]);
-    }
-
-    MPI_Wait(&err_req_cor, MPI_STATUS_IGNORE);
-    if (ERROR)
-    {
-      Parallel::Dump_Data(GH->PatL[lev], SynchList_pre, 0, PhysTime, dT_lev);
-      if (myrank == 0)
-      {
-        if (ErrorMonitor->outfile)
-          ErrorMonitor->outfile << "find NaN in RK4 substep#" << iter_count
-                                << " variables at t = " << PhysTime
-                                << ", lev = " << lev << endl;
-        MPI_Abort(MPI_COMM_WORLD, 1);
-      }
-    }
-
-#if (MAPBH == 0)
-    if (BH_num > 0 && lev == GH->levels - 1)
-    {
-      compute_Porg_rhs(Porg, Porg1, Sfx, Sfy, Sfz, lev);
-      for (int ithBH = 0; ithBH < BH_num; ithBH++)
-      {
-        f_rungekutta4_scalar(dT_lev, Porg0[ithBH][0], Porg1[ithBH][0], Porg_rhs[ithBH][0], iter_count);
-        f_rungekutta4_scalar(dT_lev, Porg0[ithBH][1], Porg1[ithBH][1], Porg_rhs[ithBH][1], iter_count);
-        f_rungekutta4_scalar(dT_lev, Porg0[ithBH][2], Porg1[ithBH][2], Porg_rhs[ithBH][2], iter_count);
-        if (Symmetry > 0)
-          Porg1[ithBH][2] = fabs(Porg1[ithBH][2]);
-        if (Symmetry == 2)
-        {
-          Porg1[ithBH][0] = fabs(Porg1[ithBH][0]);
-          Porg1[ithBH][1] = fabs(Porg1[ithBH][1]);
-        }
-      }
-    }
-#endif
-
-    if (iter_count < 3)
-    {
-      Pp = GH->PatL[lev];
-      while (Pp)
-      {
-        MyList<Block> *BP = Pp->data->blb;
-        while (BP)
-        {
-          BP->data->swapList(SynchList_pre, SynchList_cor, myrank);
-          if (BP == Pp->data->ble)
-            break;
-          BP = BP->next;
-        }
-        Pp = Pp->next;
-      }
-
-#if (MAPBH == 0)
-      if (BH_num > 0 && lev == GH->levels - 1)
-      {
-        for (int ithBH = 0; ithBH < BH_num; ithBH++)
-        {
-          Porg[ithBH][0] = Porg1[ithBH][0];
-          Porg[ithBH][1] = Porg1[ithBH][1];
-          Porg[ithBH][2] = Porg1[ithBH][2];
-        }
-      }
-#endif
-    }
-  }
-
-#if (RPS == 0)
-  RestrictProlong(lev, YN, BB);
-#endif
-
-  Pp = GH->PatL[lev];
-  while (Pp)
-  {
-    MyList<Block> *BP = Pp->data->blb;
-    while (BP)
-    {
-      Block *cg = BP->data;
-      cg->swapList(StateList, SynchList_cor, myrank);
-      cg->swapList(OldStateList, SynchList_cor, myrank);
-      if (BP == Pp->data->ble)
-        break;
-      BP = BP->next;
-    }
-    Pp = Pp->next;
-  }
-
-  if (!ERROR && keep_stage_sync_on_device)
-  {
-    // After the swaps above, only StateList points at arrays updated during this step.
-    // OldStateList/SynchList_cor remain valid on host because their backing arrays were
-    // read-only during the RK step, and SynchList_pre is reused only as scratch later.
-    const double t0 = profile_enabled ? MPI_Wtime() : 0.0;
-    stage_download_patch_list(StateList, true);
-    if (profile_enabled)
-      stage_profile_add(lev, STAGE_PROFILE_DOWNLOAD, MPI_Wtime() - t0);
-  }
-
-  if (profile_enabled)
-  {
-    const double t0 = MPI_Wtime();
-    bssn_gpu_clear_cached_device_buffers();
-    stage_profile_add(lev, STAGE_PROFILE_CLEAR_CACHE, MPI_Wtime() - t0);
-  }
-  else
-    bssn_gpu_clear_cached_device_buffers();
-
-  if (BH_num > 0 && lev == GH->levels - 1)
-  {
-    for (int ithBH = 0; ithBH < BH_num; ithBH++)
-    {
-      Porg0[ithBH][0] = Porg1[ithBH][0];
-      Porg0[ithBH][1] = Porg1[ithBH][1];
-      Porg0[ithBH][2] = Porg1[ithBH][2];
-    }
-  }
-
-  if (profile_enabled)
-    stage_profile_add(lev, STAGE_PROFILE_TOTAL, MPI_Wtime() - step_total_begin);
-}
-
-#endif

AMSS_NCKU_source/bssn_gpu.h

@@ -4,6 +4,8 @@
 #include "bssn_macro.h"
 #include "macrodef.fh"
 
+#define DEVICE_ID 0
+// #define DEVICE_ID_BY_MPI_RANK
 #define GRID_DIM 256
 #define BLOCK_DIM 128
 
@@ -65,42 +67,6 @@ int gpu_rhs(int calledby, int mpi_rank, int *ex, double &T,
 
 int gpu_rhs_ss(RHS_SS_PARA);
 
-int bssn_gpu_bind_process_device(int mpi_rank);
-void bssn_gpu_clear_cached_device_buffers();
-void bssn_gpu_release_pinned_host_buffers();
-const double *bssn_gpu_find_device_buffer(const double *host_ptr);
-void bssn_gpu_register_device_buffer(const double *host_ptr, const double *device_ptr);
-void bssn_gpu_prepare_host_buffer(const double *host_ptr, int count);
-int bssn_gpu_stage_upload_buffer(const double *host_ptr, int count);
-int bssn_gpu_stage_zero_buffer(const double *host_ptr, int count);
-int bssn_gpu_stage_upload_region(const double *host_ptr,
-                                 const int *full_shape,
-                                 const double *full_llb,
-                                 const double *full_uub,
-                                 const int *region_shape,
-                                 const double *region_llb);
-int bssn_gpu_stage_download_region(double *host_ptr,
-                                   const int *full_shape,
-                                   const double *full_llb,
-                                   const double *full_uub,
-                                   const int *region_shape,
-                                   const double *region_llb);
-int bssn_gpu_stage_download_region_to_buffer(const double *host_src_ptr,
-                                             const int *full_shape,
-                                             const double *full_llb,
-                                             const double *full_uub,
-                                             const int *region_shape,
-                                             const double *region_llb,
-                                             double *host_dst_ptr);
-int bssn_gpu_stage_upload_buffer_to_region(const double *host_src_ptr,
-                                           double *host_dst_ptr,
-                                           const int *full_shape,
-                                           const double *full_llb,
-                                           const double *full_uub,
-                                           const int *region_shape,
-                                           const double *region_llb);
-int bssn_gpu_download_buffer_batch(const int *ex, double **host_ptrs, int num_buffers);
-
 /** Init GPU side data in GPUMeta. */
 // void init_fluid_meta_gpu(GPUMeta *gpu_meta);
 

AMSS_NCKU_source/bssn_macro.h

@@ -68,7 +68,6 @@ if(TIME_COUNT_EACH_RANK == 1){\
 //3---------------------GPU---------------------
 #define CALLED_BY_STEP 0
 #define CALLED_BY_CONSTRAINT 1
-#define CALLED_BY_CONSTRAINT_CONS_ONLY 2
 
 #define RHS_PARA_CALLED_FIRST_TIME cg->shape,TRK4,cg->X[0],cg->X[1],cg->X[2],cg->fgfs[phi0->sgfn],cg->fgfs[trK0->sgfn],cg->fgfs[gxx0->sgfn],cg->fgfs[gxy0->sgfn],cg->fgfs[gxz0->sgfn],cg->fgfs[gyy0->sgfn],cg->fgfs[gyz0->sgfn],cg->fgfs[gzz0->sgfn],cg->fgfs[Axx0->sgfn],cg->fgfs[Axy0->sgfn],cg->fgfs[Axz0->sgfn],cg->fgfs[Ayy0->sgfn],cg->fgfs[Ayz0->sgfn],cg->fgfs[Azz0->sgfn],cg->fgfs[Gmx0->sgfn],cg->fgfs[Gmy0->sgfn],cg->fgfs[Gmz0->sgfn],cg->fgfs[Lap0->sgfn],cg->fgfs[Sfx0->sgfn],cg->fgfs[Sfy0->sgfn],cg->fgfs[Sfz0->sgfn],cg->fgfs[dtSfx0->sgfn],cg->fgfs[dtSfy0->sgfn],cg->fgfs[dtSfz0->sgfn],cg->fgfs[phi_rhs->sgfn],cg->fgfs[trK_rhs->sgfn],cg->fgfs[gxx_rhs->sgfn],cg->fgfs[gxy_rhs->sgfn],cg->fgfs[gxz_rhs->sgfn],cg->fgfs[gyy_rhs->sgfn],cg->fgfs[gyz_rhs->sgfn],cg->fgfs[gzz_rhs->sgfn],cg->fgfs[Axx_rhs->sgfn],cg->fgfs[Axy_rhs->sgfn],cg->fgfs[Axz_rhs->sgfn],cg->fgfs[Ayy_rhs->sgfn],cg->fgfs[Ayz_rhs->sgfn],cg->fgfs[Azz_rhs->sgfn],cg->fgfs[Gmx_rhs->sgfn],cg->fgfs[Gmy_rhs->sgfn],cg->fgfs[Gmz_rhs->sgfn],cg->fgfs[Lap_rhs->sgfn],cg->fgfs[Sfx_rhs->sgfn],cg->fgfs[Sfy_rhs->sgfn],cg->fgfs[Sfz_rhs->sgfn],cg->fgfs[dtSfx_rhs->sgfn],cg->fgfs[dtSfy_rhs->sgfn],cg->fgfs[dtSfz_rhs->sgfn],cg->fgfs[rho->sgfn],cg->fgfs[Sx->sgfn],cg->fgfs[Sy->sgfn],cg->fgfs[Sz->sgfn],cg->fgfs[Sxx->sgfn],cg->fgfs[Sxy->sgfn],cg->fgfs[Sxz->sgfn],cg->fgfs[Syy->sgfn],cg->fgfs[Syz->sgfn],cg->fgfs[Szz->sgfn],cg->fgfs[Gamxxx->sgfn],cg->fgfs[Gamxxy->sgfn],cg->fgfs[Gamxxz->sgfn],cg->fgfs[Gamxyy->sgfn],cg->fgfs[Gamxyz->sgfn],cg->fgfs[Gamxzz->sgfn],cg->fgfs[Gamyxx->sgfn],cg->fgfs[Gamyxy->sgfn],cg->fgfs[Gamyxz->sgfn],cg->fgfs[Gamyyy->sgfn],cg->fgfs[Gamyyz->sgfn],cg->fgfs[Gamyzz->sgfn],cg->fgfs[Gamzxx->sgfn],cg->fgfs[Gamzxy->sgfn],cg->fgfs[Gamzxz->sgfn],cg->fgfs[Gamzyy->sgfn],cg->fgfs[Gamzyz->sgfn],cg->fgfs[Gamzzz->sgfn],cg->fgfs[Rxx->sgfn],cg->fgfs[Rxy->sgfn],cg->fgfs[Rxz->sgfn],cg->fgfs[Ryy->sgfn],cg->fgfs[Ryz->sgfn],cg->fgfs[Rzz->sgfn],cg->fgfs[Cons_Ham->sgfn],cg->fgfs[Cons_Px->sgfn],cg->fgfs[Cons_Py->sgfn],cg->fgfs[Cons_Pz->sgfn],cg->fgfs[Cons_Gx->sgfn],cg->fgfs[Cons_Gy->sgfn],cg->fgfs[Cons_Gz->sgfn],Symmetry,lev,ndeps,pre
 

AMSS_NCKU_source/bssn_rhs.f90

@@ -62,7 +62,6 @@
   real*8, dimension(ex(1),ex(2),ex(3)),intent(inout) :: Gmx_Res, Gmy_Res, Gmz_Res
 !  gont = 0: success; gont = 1: something wrong
   integer::gont
-  integer :: i,j,k
 
 !~~~~~~> Other variables:
 
@@ -86,13 +85,6 @@
 
   real*8,dimension(3) ::SSS,AAS,ASA,SAA,ASS,SAS,SSA
   real*8            :: dX, dY, dZ, PI
-  real*8            :: divb_loc,det_loc
-  real*8            :: gupxx_loc,gupxy_loc,gupxz_loc,gupyy_loc,gupyz_loc,gupzz_loc
-  real*8            :: Rxx_loc,Rxy_loc,Rxz_loc,Ryy_loc,Ryz_loc,Rzz_loc
-  real*8            :: fxx_loc,fxy_loc,fxz_loc
-  real*8            :: Gamxa_loc,Gamya_loc,Gamza_loc
-  real*8            :: f_loc,chin_loc
-  real*8            :: l_fxx,l_fxy,l_fxz,l_fyy,l_fyz,l_fzz,S_loc
   real*8, parameter :: ZEO = 0.d0,ONE = 1.D0, TWO = 2.D0, FOUR = 4.D0
   real*8, parameter :: EIGHT = 8.D0, HALF = 0.5D0, THR = 3.d0
   real*8, parameter :: SYM = 1.D0, ANTI= - 1.D0
@@ -105,7 +97,7 @@
 #endif
 
 #if (GAUGE == 6 || GAUGE == 7)
-  integer :: BHN
+  integer :: BHN,i,j,k
   real*8, dimension(9) :: Porg
   real*8, dimension(3) :: Mass
   real*8 :: r1,r2,M,A,w1,w2,C1,C2
@@ -153,24 +145,22 @@
   dY = Y(2) - Y(1)
   dZ = Z(2) - Z(1)
 
-  do k=1,ex(3)
+  alpn1 = Lap + ONE
-  do j=1,ex(2)
+  chin1 = chi + ONE
-  do i=1,ex(1)
+  gxx = dxx + ONE
-    alpn1(i,j,k) = Lap(i,j,k) + ONE
+  gyy = dyy + ONE
-    chin1(i,j,k) = chi(i,j,k) + ONE
+  gzz = dzz + ONE
-    gxx(i,j,k) = dxx(i,j,k) + ONE
-    gyy(i,j,k) = dyy(i,j,k) + ONE
-    gzz(i,j,k) = dzz(i,j,k) + ONE
-  enddo
-  enddo
-  enddo
 
   call fderivs(ex,betax,betaxx,betaxy,betaxz,X,Y,Z,ANTI, SYM, SYM,Symmetry,Lev)
   call fderivs(ex,betay,betayx,betayy,betayz,X,Y,Z, SYM,ANTI, SYM,Symmetry,Lev)
   call fderivs(ex,betaz,betazx,betazy,betazz,X,Y,Z, SYM, SYM,ANTI,Symmetry,Lev)
   
+  div_beta = betaxx + betayy + betazz
+ 
   call fderivs(ex,chi,chix,chiy,chiz,X,Y,Z,SYM,SYM,SYM,symmetry,Lev)
 
+  chi_rhs = F2o3 *chin1*( alpn1 * trK - div_beta ) !rhs for chi
+
   call fderivs(ex,dxx,gxxx,gxxy,gxxz,X,Y,Z,SYM ,SYM ,SYM ,Symmetry,Lev)
   call fderivs(ex,gxy,gxyx,gxyy,gxyz,X,Y,Z,ANTI,ANTI,SYM ,Symmetry,Lev)
   call fderivs(ex,gxz,gxzx,gxzy,gxzz,X,Y,Z,ANTI,SYM ,ANTI,Symmetry,Lev)
@@ -178,179 +168,151 @@
   call fderivs(ex,gyz,gyzx,gyzy,gyzz,X,Y,Z,SYM ,ANTI,ANTI,Symmetry,Lev)
   call fderivs(ex,dzz,gzzx,gzzy,gzzz,X,Y,Z,SYM ,SYM ,SYM ,Symmetry,Lev)
 
-  do k=1,ex(3)
+  gxx_rhs = - TWO * alpn1 * Axx    -  F2o3 * gxx * div_beta          + &
-  do j=1,ex(2)
+              TWO *(  gxx * betaxx +   gxy * betayx +   gxz * betazx)
-  do i=1,ex(1)
-    divb_loc = betaxx(i,j,k) + betayy(i,j,k) + betazz(i,j,k)
-    div_beta(i,j,k) = divb_loc
 
-    chi_rhs(i,j,k) = F2o3 * chin1(i,j,k) * (alpn1(i,j,k) * trK(i,j,k) - divb_loc)
+  gyy_rhs = - TWO * alpn1 * Ayy    -  F2o3 * gyy * div_beta          + &
+              TWO *(  gxy * betaxy +   gyy * betayy +   gyz * betazy)
 
-    gxx_rhs(i,j,k) = - TWO * alpn1(i,j,k) * Axx(i,j,k) - F2o3 * gxx(i,j,k) * divb_loc + &
+  gzz_rhs = - TWO * alpn1 * Azz    -  F2o3 * gzz * div_beta          + &
-         TWO * ( gxx(i,j,k) * betaxx(i,j,k) + gxy(i,j,k) * betayx(i,j,k) + gxz(i,j,k) * betazx(i,j,k) )
+              TWO *(  gxz * betaxz +   gyz * betayz +   gzz * betazz)
 
-    gyy_rhs(i,j,k) = - TWO * alpn1(i,j,k) * Ayy(i,j,k) - F2o3 * gyy(i,j,k) * divb_loc + &
+  gxy_rhs = - TWO * alpn1 * Axy    +  F1o3 * gxy    * div_beta       + &
-         TWO * ( gxy(i,j,k) * betaxy(i,j,k) + gyy(i,j,k) * betayy(i,j,k) + gyz(i,j,k) * betazy(i,j,k) )
+                      gxx * betaxy                  +   gxz * betazy + &
+                                       gyy * betayx +   gyz * betazx   &
+                                                    -   gxy * betazz
 
-    gzz_rhs(i,j,k) = - TWO * alpn1(i,j,k) * Azz(i,j,k) - F2o3 * gzz(i,j,k) * divb_loc + &
+  gyz_rhs = - TWO * alpn1 * Ayz    +  F1o3 * gyz    * div_beta       + &
-         TWO * ( gxz(i,j,k) * betaxz(i,j,k) + gyz(i,j,k) * betayz(i,j,k) + gzz(i,j,k) * betazz(i,j,k) )
+                      gxy * betaxz +   gyy * betayz                  + &
+                      gxz * betaxy                  +   gzz * betazy   &
+                                                    -   gyz * betaxx
  
-    gxy_rhs(i,j,k) = - TWO * alpn1(i,j,k) * Axy(i,j,k) + F1o3 * gxy(i,j,k) * divb_loc + &
+  gxz_rhs = - TWO * alpn1 * Axz    +  F1o3 * gxz    * div_beta       + &
-         gxx(i,j,k) * betaxy(i,j,k) + gxz(i,j,k) * betazy(i,j,k) + gyy(i,j,k) * betayx(i,j,k) + &
+                      gxx * betaxz +   gxy * betayz                  + &
-         gyz(i,j,k) * betazx(i,j,k) - gxy(i,j,k) * betazz(i,j,k)
+                                       gyz * betayx +   gzz * betazx   &
+                                                    -   gxz * betayy     !rhs for gij
 
-    gyz_rhs(i,j,k) = - TWO * alpn1(i,j,k) * Ayz(i,j,k) + F1o3 * gyz(i,j,k) * divb_loc + &
+! invert tilted metric
-         gxy(i,j,k) * betaxz(i,j,k) + gyy(i,j,k) * betayz(i,j,k) + gxz(i,j,k) * betaxy(i,j,k) + &
+  gupzz =  gxx * gyy * gzz + gxy * gyz * gxz + gxz * gxy * gyz - &
-         gzz(i,j,k) * betazy(i,j,k) - gyz(i,j,k) * betaxx(i,j,k)
+           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
 
-    gxz_rhs(i,j,k) = - TWO * alpn1(i,j,k) * Axz(i,j,k) + F1o3 * gxz(i,j,k) * divb_loc + &
+  if(co == 0)then
-         gxx(i,j,k) * betaxz(i,j,k) + gxy(i,j,k) * betayz(i,j,k) + gyz(i,j,k) * betayx(i,j,k) + &
+! Gam^i_Res = Gam^i + gup^ij_,j
-         gzz(i,j,k) * betazx(i,j,k) - gxz(i,j,k) * betayy(i,j,k)
+  Gmx_Res = Gamx - (gupxx*(gupxx*gxxx+gupxy*gxyx+gupxz*gxzx)&
+                   +gupxy*(gupxx*gxyx+gupxy*gyyx+gupxz*gyzx)&
+                   +gupxz*(gupxx*gxzx+gupxy*gyzx+gupxz*gzzx)&
+                   +gupxx*(gupxy*gxxy+gupyy*gxyy+gupyz*gxzy)&
+                   +gupxy*(gupxy*gxyy+gupyy*gyyy+gupyz*gyzy)&
+                   +gupxz*(gupxy*gxzy+gupyy*gyzy+gupyz*gzzy)&
+                   +gupxx*(gupxz*gxxz+gupyz*gxyz+gupzz*gxzz)&
+                   +gupxy*(gupxz*gxyz+gupyz*gyyz+gupzz*gyzz)&
+                   +gupxz*(gupxz*gxzz+gupyz*gyzz+gupzz*gzzz))
+  Gmy_Res = Gamy - (gupxx*(gupxy*gxxx+gupyy*gxyx+gupyz*gxzx)&
+                   +gupxy*(gupxy*gxyx+gupyy*gyyx+gupyz*gyzx)&
+                   +gupxz*(gupxy*gxzx+gupyy*gyzx+gupyz*gzzx)&
+                   +gupxy*(gupxy*gxxy+gupyy*gxyy+gupyz*gxzy)&
+                   +gupyy*(gupxy*gxyy+gupyy*gyyy+gupyz*gyzy)&
+                   +gupyz*(gupxy*gxzy+gupyy*gyzy+gupyz*gzzy)&
+                   +gupxy*(gupxz*gxxz+gupyz*gxyz+gupzz*gxzz)&
+                   +gupyy*(gupxz*gxyz+gupyz*gyyz+gupzz*gyzz)&
+                   +gupyz*(gupxz*gxzz+gupyz*gyzz+gupzz*gzzz))
+  Gmz_Res = Gamz - (gupxx*(gupxz*gxxx+gupyz*gxyx+gupzz*gxzx)&
+                   +gupxy*(gupxz*gxyx+gupyz*gyyx+gupzz*gyzx)&
+                   +gupxz*(gupxz*gxzx+gupyz*gyzx+gupzz*gzzx)&
+                   +gupxy*(gupxz*gxxy+gupyz*gxyy+gupzz*gxzy)&
+                   +gupyy*(gupxz*gxyy+gupyz*gyyy+gupzz*gyzy)&
+                   +gupyz*(gupxz*gxzy+gupyz*gyzy+gupzz*gzzy)&
+                   +gupxz*(gupxz*gxxz+gupyz*gxyz+gupzz*gxzz)&
+                   +gupyz*(gupxz*gxyz+gupyz*gyyz+gupzz*gyzz)&
+                   +gupzz*(gupxz*gxzz+gupyz*gyzz+gupzz*gzzz))
+  endif
 
-    det_loc = gxx(i,j,k) * gyy(i,j,k) * gzz(i,j,k) + gxy(i,j,k) * gyz(i,j,k) * gxz(i,j,k) + &
+! second kind of connection
-         gxz(i,j,k) * gxy(i,j,k) * gyz(i,j,k) - gxz(i,j,k) * gyy(i,j,k) * gxz(i,j,k) - &
+  Gamxxx =HALF*( gupxx*gxxx + gupxy*(TWO*gxyx - gxxy ) + gupxz*(TWO*gxzx - gxxz ))
-         gxy(i,j,k) * gxy(i,j,k) * gzz(i,j,k) - gxx(i,j,k) * gyz(i,j,k) * gyz(i,j,k)
+  Gamyxx =HALF*( gupxy*gxxx + gupyy*(TWO*gxyx - gxxy ) + gupyz*(TWO*gxzx - gxxz ))
-    gupxx_loc = ( gyy(i,j,k) * gzz(i,j,k) - gyz(i,j,k) * gyz(i,j,k) ) / det_loc
+  Gamzxx =HALF*( gupxz*gxxx + gupyz*(TWO*gxyx - gxxy ) + gupzz*(TWO*gxzx - gxxz ))
-    gupxy_loc = - ( gxy(i,j,k) * gzz(i,j,k) - gyz(i,j,k) * gxz(i,j,k) ) / det_loc
-    gupxz_loc = ( gxy(i,j,k) * gyz(i,j,k) - gyy(i,j,k) * gxz(i,j,k) ) / det_loc
-    gupyy_loc = ( gxx(i,j,k) * gzz(i,j,k) - gxz(i,j,k) * gxz(i,j,k) ) / det_loc
-    gupyz_loc = - ( gxx(i,j,k) * gyz(i,j,k) - gxy(i,j,k) * gxz(i,j,k) ) / det_loc
-    gupzz_loc = ( gxx(i,j,k) * gyy(i,j,k) - gxy(i,j,k) * gxy(i,j,k) ) / det_loc
-    gupxx(i,j,k) = gupxx_loc
-    gupxy(i,j,k) = gupxy_loc
-    gupxz(i,j,k) = gupxz_loc
-    gupyy(i,j,k) = gupyy_loc
-    gupyz(i,j,k) = gupyz_loc
-    gupzz(i,j,k) = gupzz_loc
  
-    if(co == 0)then
+  Gamxyy =HALF*( gupxx*(TWO*gxyy - gyyx ) + gupxy*gyyy + gupxz*(TWO*gyzy - gyyz ))
-      Gmx_Res(i,j,k) = Gamx(i,j,k) - ( &
+  Gamyyy =HALF*( gupxy*(TWO*gxyy - gyyx ) + gupyy*gyyy + gupyz*(TWO*gyzy - gyyz ))
-           gupxx_loc*(gupxx_loc*gxxx(i,j,k)+gupxy_loc*gxyx(i,j,k)+gupxz_loc*gxzx(i,j,k)) + &
+  Gamzyy =HALF*( gupxz*(TWO*gxyy - gyyx ) + gupyz*gyyy + gupzz*(TWO*gyzy - gyyz ))
-           gupxy_loc*(gupxx_loc*gxyx(i,j,k)+gupxy_loc*gyyx(i,j,k)+gupxz_loc*gyzx(i,j,k)) + &
-           gupxz_loc*(gupxx_loc*gxzx(i,j,k)+gupxy_loc*gyzx(i,j,k)+gupxz_loc*gzzx(i,j,k)) + &
-           gupxx_loc*(gupxy_loc*gxxy(i,j,k)+gupyy_loc*gxyy(i,j,k)+gupyz_loc*gxzy(i,j,k)) + &
-           gupxy_loc*(gupxy_loc*gxyy(i,j,k)+gupyy_loc*gyyy(i,j,k)+gupyz_loc*gyzy(i,j,k)) + &
-           gupxz_loc*(gupxy_loc*gxzy(i,j,k)+gupyy_loc*gyzy(i,j,k)+gupyz_loc*gzzy(i,j,k)) + &
-           gupxx_loc*(gupxz_loc*gxxz(i,j,k)+gupyz_loc*gxyz(i,j,k)+gupzz_loc*gxzz(i,j,k)) + &
-           gupxy_loc*(gupxz_loc*gxyz(i,j,k)+gupyz_loc*gyyz(i,j,k)+gupzz_loc*gyzz(i,j,k)) + &
-           gupxz_loc*(gupxz_loc*gxzz(i,j,k)+gupyz_loc*gyzz(i,j,k)+gupzz_loc*gzzz(i,j,k)))
-      Gmy_Res(i,j,k) = Gamy(i,j,k) - ( &
-           gupxx_loc*(gupxy_loc*gxxx(i,j,k)+gupyy_loc*gxyx(i,j,k)+gupyz_loc*gxzx(i,j,k)) + &
-           gupxy_loc*(gupxy_loc*gxyx(i,j,k)+gupyy_loc*gyyx(i,j,k)+gupyz_loc*gyzx(i,j,k)) + &
-           gupxz_loc*(gupxy_loc*gxzx(i,j,k)+gupyy_loc*gyzx(i,j,k)+gupyz_loc*gzzx(i,j,k)) + &
-           gupxy_loc*(gupxy_loc*gxxy(i,j,k)+gupyy_loc*gxyy(i,j,k)+gupyz_loc*gxzy(i,j,k)) + &
-           gupyy_loc*(gupxy_loc*gxyy(i,j,k)+gupyy_loc*gyyy(i,j,k)+gupyz_loc*gyzy(i,j,k)) + &
-           gupyz_loc*(gupxy_loc*gxzy(i,j,k)+gupyy_loc*gyzy(i,j,k)+gupyz_loc*gzzy(i,j,k)) + &
-           gupxy_loc*(gupxz_loc*gxxz(i,j,k)+gupyz_loc*gxyz(i,j,k)+gupzz_loc*gxzz(i,j,k)) + &
-           gupyy_loc*(gupxz_loc*gxyz(i,j,k)+gupyz_loc*gyyz(i,j,k)+gupzz_loc*gyzz(i,j,k)) + &
-           gupyz_loc*(gupxz_loc*gxzz(i,j,k)+gupyz_loc*gyzz(i,j,k)+gupzz_loc*gzzz(i,j,k)))
-      Gmz_Res(i,j,k) = Gamz(i,j,k) - ( &
-           gupxx_loc*(gupxz_loc*gxxx(i,j,k)+gupyz_loc*gxyx(i,j,k)+gupzz_loc*gxzx(i,j,k)) + &
-           gupxy_loc*(gupxz_loc*gxyx(i,j,k)+gupyz_loc*gyyx(i,j,k)+gupzz_loc*gyzx(i,j,k)) + &
-           gupxz_loc*(gupxz_loc*gxzx(i,j,k)+gupyz_loc*gyzx(i,j,k)+gupzz_loc*gzzx(i,j,k)) + &
-           gupxy_loc*(gupxz_loc*gxxy(i,j,k)+gupyz_loc*gxyy(i,j,k)+gupzz_loc*gxzy(i,j,k)) + &
-           gupyy_loc*(gupxz_loc*gxyy(i,j,k)+gupyz_loc*gyyy(i,j,k)+gupzz_loc*gyzy(i,j,k)) + &
-           gupyz_loc*(gupxz_loc*gxzy(i,j,k)+gupyz_loc*gyzy(i,j,k)+gupzz_loc*gzzy(i,j,k)) + &
-           gupxz_loc*(gupxz_loc*gxxz(i,j,k)+gupyz_loc*gxyz(i,j,k)+gupzz_loc*gxzz(i,j,k)) + &
-           gupyz_loc*(gupxz_loc*gxyz(i,j,k)+gupyz_loc*gyyz(i,j,k)+gupzz_loc*gyzz(i,j,k)) + &
-           gupzz_loc*(gupxz_loc*gxzz(i,j,k)+gupyz_loc*gyzz(i,j,k)+gupzz_loc*gzzz(i,j,k)))
-    endif
 
-    Gamxxx(i,j,k)=HALF*( gupxx_loc*gxxx(i,j,k) + gupxy_loc*(TWO*gxyx(i,j,k) - gxxy(i,j,k)) + gupxz_loc*(TWO*gxzx(i,j,k) - gxxz(i,j,k)))
+  Gamxzz =HALF*( gupxx*(TWO*gxzz - gzzx ) + gupxy*(TWO*gyzz - gzzy ) + gupxz*gzzz)
-    Gamyxx(i,j,k)=HALF*( gupxy_loc*gxxx(i,j,k) + gupyy_loc*(TWO*gxyx(i,j,k) - gxxy(i,j,k)) + gupyz_loc*(TWO*gxzx(i,j,k) - gxxz(i,j,k)))
+  Gamyzz =HALF*( gupxy*(TWO*gxzz - gzzx ) + gupyy*(TWO*gyzz - gzzy ) + gupyz*gzzz)
-    Gamzxx(i,j,k)=HALF*( gupxz_loc*gxxx(i,j,k) + gupyz_loc*(TWO*gxyx(i,j,k) - gxxy(i,j,k)) + gupzz_loc*(TWO*gxzx(i,j,k) - gxxz(i,j,k)))
+  Gamzzz =HALF*( gupxz*(TWO*gxzz - gzzx ) + gupyz*(TWO*gyzz - gzzy ) + gupzz*gzzz)
 
-    Gamxyy(i,j,k)=HALF*( gupxx_loc*(TWO*gxyy(i,j,k) - gyyx(i,j,k)) + gupxy_loc*gyyy(i,j,k) + gupxz_loc*(TWO*gyzy(i,j,k) - gyyz(i,j,k)))
+  Gamxxy =HALF*( gupxx*gxxy + gupxy*gyyx + gupxz*( gxzy + gyzx - gxyz ) )
-    Gamyyy(i,j,k)=HALF*( gupxy_loc*(TWO*gxyy(i,j,k) - gyyx(i,j,k)) + gupyy_loc*gyyy(i,j,k) + gupyz_loc*(TWO*gyzy(i,j,k) - gyyz(i,j,k)))
+  Gamyxy =HALF*( gupxy*gxxy + gupyy*gyyx + gupyz*( gxzy + gyzx - gxyz ) )
-    Gamzyy(i,j,k)=HALF*( gupxz_loc*(TWO*gxyy(i,j,k) - gyyx(i,j,k)) + gupyz_loc*gyyy(i,j,k) + gupzz_loc*(TWO*gyzy(i,j,k) - gyyz(i,j,k)))
+  Gamzxy =HALF*( gupxz*gxxy + gupyz*gyyx + gupzz*( gxzy + gyzx - gxyz ) )
 
-    Gamxzz(i,j,k)=HALF*( gupxx_loc*(TWO*gxzz(i,j,k) - gzzx(i,j,k)) + gupxy_loc*(TWO*gyzz(i,j,k) - gzzy(i,j,k)) + gupxz_loc*gzzz(i,j,k))
+  Gamxxz =HALF*( gupxx*gxxz + gupxy*( gxyz + gyzx - gxzy ) + gupxz*gzzx )
-    Gamyzz(i,j,k)=HALF*( gupxy_loc*(TWO*gxzz(i,j,k) - gzzx(i,j,k)) + gupyy_loc*(TWO*gyzz(i,j,k) - gzzy(i,j,k)) + gupyz_loc*gzzz(i,j,k))
+  Gamyxz =HALF*( gupxy*gxxz + gupyy*( gxyz + gyzx - gxzy ) + gupyz*gzzx )
-    Gamzzz(i,j,k)=HALF*( gupxz_loc*(TWO*gxzz(i,j,k) - gzzx(i,j,k)) + gupyz_loc*(TWO*gyzz(i,j,k) - gzzy(i,j,k)) + gupzz_loc*gzzz(i,j,k))
+  Gamzxz =HALF*( gupxz*gxxz + gupyz*( gxyz + gyzx - gxzy ) + gupzz*gzzx )
 
-    Gamxxy(i,j,k)=HALF*( gupxx_loc*gxxy(i,j,k) + gupxy_loc*gyyx(i,j,k) + gupxz_loc*(gxzy(i,j,k) + gyzx(i,j,k) - gxyz(i,j,k)) )
+  Gamxyz =HALF*( gupxx*( gxyz + gxzy - gyzx ) + gupxy*gyyz + gupxz*gzzy )
-    Gamyxy(i,j,k)=HALF*( gupxy_loc*gxxy(i,j,k) + gupyy_loc*gyyx(i,j,k) + gupyz_loc*(gxzy(i,j,k) + gyzx(i,j,k) - gxyz(i,j,k)) )
+  Gamyyz =HALF*( gupxy*( gxyz + gxzy - gyzx ) + gupyy*gyyz + gupyz*gzzy )
-    Gamzxy(i,j,k)=HALF*( gupxz_loc*gxxy(i,j,k) + gupyz_loc*gyyx(i,j,k) + gupzz_loc*(gxzy(i,j,k) + gyzx(i,j,k) - gxyz(i,j,k)) )
+  Gamzyz =HALF*( gupxz*( gxyz + gxzy - gyzx ) + gupyz*gyyz + gupzz*gzzy )
-
-    Gamxxz(i,j,k)=HALF*( gupxx_loc*gxxz(i,j,k) + gupxy_loc*(gxyz(i,j,k) + gyzx(i,j,k) - gxzy(i,j,k)) + gupxz_loc*gzzx(i,j,k) )
-    Gamyxz(i,j,k)=HALF*( gupxy_loc*gxxz(i,j,k) + gupyy_loc*(gxyz(i,j,k) + gyzx(i,j,k) - gxzy(i,j,k)) + gupyz_loc*gzzx(i,j,k) )
-    Gamzxz(i,j,k)=HALF*( gupxz_loc*gxxz(i,j,k) + gupyz_loc*(gxyz(i,j,k) + gyzx(i,j,k) - gxzy(i,j,k)) + gupzz_loc*gzzx(i,j,k) )
-
-    Gamxyz(i,j,k)=HALF*( gupxx_loc*(gxyz(i,j,k) + gxzy(i,j,k) - gyzx(i,j,k)) + gupxy_loc*gyyz(i,j,k) + gupxz_loc*gzzy(i,j,k) )
-    Gamyyz(i,j,k)=HALF*( gupxy_loc*(gxyz(i,j,k) + gxzy(i,j,k) - gyzx(i,j,k)) + gupyy_loc*gyyz(i,j,k) + gupyz_loc*gzzy(i,j,k) )
-    Gamzyz(i,j,k)=HALF*( gupxz_loc*(gxyz(i,j,k) + gxzy(i,j,k) - gyzx(i,j,k)) + gupyz_loc*gyyz(i,j,k) + gupzz_loc*gzzy(i,j,k) )
-  enddo
-  enddo
-  enddo
 ! Raise indices of \tilde A_{ij} and store in R_ij
 
+  Rxx =    gupxx * gupxx * Axx + gupxy * gupxy * Ayy + gupxz * gupxz * Azz + &
+      TWO*(gupxx * gupxy * Axy + gupxx * gupxz * Axz + gupxy * gupxz * Ayz)
+
+  Ryy =    gupxy * gupxy * Axx + gupyy * gupyy * Ayy + gupyz * gupyz * Azz + &
+      TWO*(gupxy * gupyy * Axy + gupxy * gupyz * Axz + gupyy * gupyz * Ayz)
+
+  Rzz =    gupxz * gupxz * Axx + gupyz * gupyz * Ayy + gupzz * gupzz * Azz + &
+      TWO*(gupxz * gupyz * Axy + gupxz * gupzz * Axz + gupyz * gupzz * Ayz)
+
+  Rxy =    gupxx * gupxy * Axx + gupxy * gupyy * Ayy + gupxz * gupyz * Azz + &
+          (gupxx * gupyy       + gupxy * gupxy)* Axy                       + &
+          (gupxx * gupyz       + gupxz * gupxy)* Axz                       + &
+          (gupxy * gupyz       + gupxz * gupyy)* Ayz
+
+  Rxz =    gupxx * gupxz * Axx + gupxy * gupyz * Ayy + gupxz * gupzz * Azz + &
+          (gupxx * gupyz       + gupxy * gupxz)* Axy                       + &
+          (gupxx * gupzz       + gupxz * gupxz)* Axz                       + &
+          (gupxy * gupzz       + gupxz * gupyz)* Ayz
+
+  Ryz =    gupxy * gupxz * Axx + gupyy * gupyz * Ayy + gupyz * gupzz * Azz + &
+          (gupxy * gupyz       + gupyy * gupxz)* Axy                       + &
+          (gupxy * gupzz       + gupyz * gupxz)* Axz                       + &
+          (gupyy * gupzz       + gupyz * gupyz)* Ayz
+
 ! Right hand side for Gam^i without shift terms...
   call fderivs(ex,Lap,Lapx,Lapy,Lapz,X,Y,Z,SYM,SYM,SYM,Symmetry,Lev)
   call fderivs(ex,trK,Kx,Ky,Kz,X,Y,Z,SYM,SYM,SYM,symmetry,Lev)
-  do k=1,ex(3)
-  do j=1,ex(2)
-  do i=1,ex(1)
-    gupxx_loc = gupxx(i,j,k)
-    gupxy_loc = gupxy(i,j,k)
-    gupxz_loc = gupxz(i,j,k)
-    gupyy_loc = gupyy(i,j,k)
-    gupyz_loc = gupyz(i,j,k)
-    gupzz_loc = gupzz(i,j,k)
 
-    Rxx_loc = gupxx_loc * gupxx_loc * Axx(i,j,k) + gupxy_loc * gupxy_loc * Ayy(i,j,k) + gupxz_loc * gupxz_loc * Azz(i,j,k) + &
+   Gamx_rhs = - TWO * (   Lapx * Rxx +   Lapy * Rxy +   Lapz * Rxz ) + &
-         TWO * (gupxx_loc * gupxy_loc * Axy(i,j,k) + gupxx_loc * gupxz_loc * Axz(i,j,k) + gupxy_loc * gupxz_loc * Ayz(i,j,k))
+        TWO * alpn1 * (                                                &
-    Ryy_loc = gupxy_loc * gupxy_loc * Axx(i,j,k) + gupyy_loc * gupyy_loc * Ayy(i,j,k) + gupyz_loc * gupyz_loc * Azz(i,j,k) + &
+        -F3o2/chin1 * (   chix * Rxx +   chiy * Rxy +   chiz * Rxz ) - &
-         TWO * (gupxy_loc * gupyy_loc * Axy(i,j,k) + gupxy_loc * gupyz_loc * Axz(i,j,k) + gupyy_loc * gupyz_loc * Ayz(i,j,k))
+              gupxx * (   F2o3 * Kx  +  EIGHT * PI * Sx            ) - &
-    Rzz_loc = gupxz_loc * gupxz_loc * Axx(i,j,k) + gupyz_loc * gupyz_loc * Ayy(i,j,k) + gupzz_loc * gupzz_loc * Azz(i,j,k) + &
+              gupxy * (   F2o3 * Ky  +  EIGHT * PI * Sy            ) - &
-         TWO * (gupxz_loc * gupyz_loc * Axy(i,j,k) + gupxz_loc * gupzz_loc * Axz(i,j,k) + gupyz_loc * gupzz_loc * Ayz(i,j,k))
+              gupxz * (   F2o3 * Kz  +  EIGHT * PI * Sz            ) + &
-    Rxy_loc = gupxx_loc * gupxy_loc * Axx(i,j,k) + gupxy_loc * gupyy_loc * Ayy(i,j,k) + gupxz_loc * gupyz_loc * Azz(i,j,k) + &
+                        Gamxxx * Rxx + Gamxyy * Ryy + Gamxzz * Rzz   + &
-         (gupxx_loc * gupyy_loc + gupxy_loc * gupxy_loc) * Axy(i,j,k) + &
+                TWO * ( Gamxxy * Rxy + Gamxxz * Rxz + Gamxyz * Ryz ) )
-         (gupxx_loc * gupyz_loc + gupxz_loc * gupxy_loc) * Axz(i,j,k) + &
-         (gupxy_loc * gupyz_loc + gupxz_loc * gupyy_loc) * Ayz(i,j,k)
-    Rxz_loc = gupxx_loc * gupxz_loc * Axx(i,j,k) + gupxy_loc * gupyz_loc * Ayy(i,j,k) + gupxz_loc * gupzz_loc * Azz(i,j,k) + &
-         (gupxx_loc * gupyz_loc + gupxy_loc * gupxz_loc) * Axy(i,j,k) + &
-         (gupxx_loc * gupzz_loc + gupxz_loc * gupxz_loc) * Axz(i,j,k) + &
-         (gupxy_loc * gupzz_loc + gupxz_loc * gupyz_loc) * Ayz(i,j,k)
-    Ryz_loc = gupxy_loc * gupxz_loc * Axx(i,j,k) + gupyy_loc * gupyz_loc * Ayy(i,j,k) + gupyz_loc * gupzz_loc * Azz(i,j,k) + &
-         (gupxy_loc * gupyz_loc + gupyy_loc * gupxz_loc) * Axy(i,j,k) + &
-         (gupxy_loc * gupzz_loc + gupyz_loc * gupxz_loc) * Axz(i,j,k) + &
-         (gupyy_loc * gupzz_loc + gupyz_loc * gupyz_loc) * Ayz(i,j,k)
-    Rxx(i,j,k) = Rxx_loc
-    Ryy(i,j,k) = Ryy_loc
-    Rzz(i,j,k) = Rzz_loc
-    Rxy(i,j,k) = Rxy_loc
-    Rxz(i,j,k) = Rxz_loc
-    Ryz(i,j,k) = Ryz_loc
 
-    Gamx_rhs(i,j,k) = - TWO * (Lapx(i,j,k) * Rxx_loc + Lapy(i,j,k) * Rxy_loc + Lapz(i,j,k) * Rxz_loc) + &
+   Gamy_rhs = - TWO * (   Lapx * Rxy +   Lapy * Ryy +   Lapz * Ryz ) + &
-         TWO * alpn1(i,j,k) * ( &
+        TWO * alpn1 * (                                                &
-         -F3o2/chin1(i,j,k) * (chix(i,j,k) * Rxx_loc + chiy(i,j,k) * Rxy_loc + chiz(i,j,k) * Rxz_loc) - &
+        -F3o2/chin1 * (   chix * Rxy +  chiy * Ryy +    chiz * Ryz ) - &
-         gupxx_loc * (F2o3 * Kx(i,j,k) + EIGHT * PI * Sx(i,j,k)) - &
+              gupxy * (   F2o3 * Kx  +  EIGHT * PI * Sx            ) - &
-         gupxy_loc * (F2o3 * Ky(i,j,k) + EIGHT * PI * Sy(i,j,k)) - &
+              gupyy * (   F2o3 * Ky  +  EIGHT * PI * Sy            ) - &
-         gupxz_loc * (F2o3 * Kz(i,j,k) + EIGHT * PI * Sz(i,j,k)) + &
+              gupyz * (   F2o3 * Kz  +  EIGHT * PI * Sz            ) + &
-         Gamxxx(i,j,k) * Rxx_loc + Gamxyy(i,j,k) * Ryy_loc + Gamxzz(i,j,k) * Rzz_loc + &
+                        Gamyxx * Rxx + Gamyyy * Ryy + Gamyzz * Rzz   + &
-         TWO * (Gamxxy(i,j,k) * Rxy_loc + Gamxxz(i,j,k) * Rxz_loc + Gamxyz(i,j,k) * Ryz_loc))
+                TWO * ( Gamyxy * Rxy + Gamyxz * Rxz + Gamyyz * Ryz ) )
 
-    Gamy_rhs(i,j,k) = - TWO * (Lapx(i,j,k) * Rxy_loc + Lapy(i,j,k) * Ryy_loc + Lapz(i,j,k) * Ryz_loc) + &
+   Gamz_rhs = - TWO * (   Lapx * Rxz +   Lapy * Ryz +   Lapz * Rzz ) + &
-         TWO * alpn1(i,j,k) * ( &
+        TWO * alpn1 * (                                                &
-         -F3o2/chin1(i,j,k) * (chix(i,j,k) * Rxy_loc + chiy(i,j,k) * Ryy_loc + chiz(i,j,k) * Ryz_loc) - &
+        -F3o2/chin1 * (   chix * Rxz +  chiy * Ryz +    chiz * Rzz ) - &
-         gupxy_loc * (F2o3 * Kx(i,j,k) + EIGHT * PI * Sx(i,j,k)) - &
+              gupxz * (   F2o3 * Kx  +  EIGHT * PI * Sx            ) - &
-         gupyy_loc * (F2o3 * Ky(i,j,k) + EIGHT * PI * Sy(i,j,k)) - &
+              gupyz * (   F2o3 * Ky  +  EIGHT * PI * Sy            ) - &
-         gupyz_loc * (F2o3 * Kz(i,j,k) + EIGHT * PI * Sz(i,j,k)) + &
+              gupzz * (   F2o3 * Kz  +  EIGHT * PI * Sz            ) + &
-         Gamyxx(i,j,k) * Rxx_loc + Gamyyy(i,j,k) * Ryy_loc + Gamyzz(i,j,k) * Rzz_loc + &
+                        Gamzxx * Rxx + Gamzyy * Ryy + Gamzzz * Rzz   + &
-         TWO * (Gamyxy(i,j,k) * Rxy_loc + Gamyxz(i,j,k) * Rxz_loc + Gamyyz(i,j,k) * Ryz_loc))
+                TWO * ( Gamzxy * Rxy + Gamzxz * Rxz + Gamzyz * Ryz ) )
-
-    Gamz_rhs(i,j,k) = - TWO * (Lapx(i,j,k) * Rxz_loc + Lapy(i,j,k) * Ryz_loc + Lapz(i,j,k) * Rzz_loc) + &
-         TWO * alpn1(i,j,k) * ( &
-         -F3o2/chin1(i,j,k) * (chix(i,j,k) * Rxz_loc + chiy(i,j,k) * Ryz_loc + chiz(i,j,k) * Rzz_loc) - &
-         gupxz_loc * (F2o3 * Kx(i,j,k) + EIGHT * PI * Sx(i,j,k)) - &
-         gupyz_loc * (F2o3 * Ky(i,j,k) + EIGHT * PI * Sy(i,j,k)) - &
-         gupzz_loc * (F2o3 * Kz(i,j,k) + EIGHT * PI * Sz(i,j,k)) + &
-         Gamzxx(i,j,k) * Rxx_loc + Gamzyy(i,j,k) * Ryy_loc + Gamzzz(i,j,k) * Rzz_loc + &
-         TWO * (Gamzxy(i,j,k) * Rxy_loc + Gamzxz(i,j,k) * Rxz_loc + Gamzyz(i,j,k) * Ryz_loc))
-  enddo
-  enddo
-  enddo
 
   call fdderivs(ex,betax,gxxx,gxyx,gxzx,gyyx,gyzx,gzzx,&
                 X,Y,Z,ANTI,SYM, SYM ,Symmetry,Lev)
@@ -359,54 +321,38 @@
   call fdderivs(ex,betaz,gxxz,gxyz,gxzz,gyyz,gyzz,gzzz,&
                 X,Y,Z,SYM ,SYM, ANTI,Symmetry,Lev)
 
+  fxx = gxxx + gxyy + gxzz
+  fxy = gxyx + gyyy + gyzz
+  fxz = gxzx + gyzy + gzzz
+
+  Gamxa =       gupxx * Gamxxx + gupyy * Gamxyy + gupzz * Gamxzz + &
+          TWO*( gupxy * Gamxxy + gupxz * Gamxxz + gupyz * Gamxyz )
+  Gamya =       gupxx * Gamyxx + gupyy * Gamyyy + gupzz * Gamyzz + &
+          TWO*( gupxy * Gamyxy + gupxz * Gamyxz + gupyz * Gamyyz )
+  Gamza =       gupxx * Gamzxx + gupyy * Gamzyy + gupzz * Gamzzz + &
+          TWO*( gupxy * Gamzxy + gupxz * Gamzxz + gupyz * Gamzyz )
+
   call fderivs(ex,Gamx,Gamxx,Gamxy,Gamxz,X,Y,Z,ANTI,SYM ,SYM ,Symmetry,Lev)
   call fderivs(ex,Gamy,Gamyx,Gamyy,Gamyz,X,Y,Z,SYM ,ANTI,SYM ,Symmetry,Lev)
   call fderivs(ex,Gamz,Gamzx,Gamzy,Gamzz,X,Y,Z,SYM ,SYM ,ANTI,Symmetry,Lev)
-  do k=1,ex(3)
-  do j=1,ex(2)
-  do i=1,ex(1)
-    divb_loc = div_beta(i,j,k)
-    fxx_loc = gxxx(i,j,k) + gxyy(i,j,k) + gxzz(i,j,k)
-    fxy_loc = gxyx(i,j,k) + gyyy(i,j,k) + gyzz(i,j,k)
-    fxz_loc = gxzx(i,j,k) + gyzy(i,j,k) + gzzz(i,j,k)
 
-    gupxx_loc = gupxx(i,j,k)
+  Gamx_rhs =               Gamx_rhs +  F2o3 *  Gamxa * div_beta        - &
-    gupxy_loc = gupxy(i,j,k)
+                     Gamxa * betaxx - Gamya * betaxy - Gamza * betaxz  + &
-    gupxz_loc = gupxz(i,j,k)
+             F1o3 * (gupxx * fxx    + gupxy * fxy    + gupxz * fxz    ) + &
-    gupyy_loc = gupyy(i,j,k)
+                     gupxx * gxxx   + gupyy * gyyx   + gupzz * gzzx    + &
-    gupyz_loc = gupyz(i,j,k)
+              TWO * (gupxy * gxyx   + gupxz * gxzx   + gupyz * gyzx  )
-    gupzz_loc = gupzz(i,j,k)
 
-    Gamxa_loc = gupxx_loc * Gamxxx(i,j,k) + gupyy_loc * Gamxyy(i,j,k) + gupzz_loc * Gamxzz(i,j,k) + &
+  Gamy_rhs =               Gamy_rhs +  F2o3 *  Gamya * div_beta        - &
-         TWO * (gupxy_loc * Gamxxy(i,j,k) + gupxz_loc * Gamxxz(i,j,k) + gupyz_loc * Gamxyz(i,j,k))
+                     Gamxa * betayx - Gamya * betayy - Gamza * betayz  + &
-    Gamya_loc = gupxx_loc * Gamyxx(i,j,k) + gupyy_loc * Gamyyy(i,j,k) + gupzz_loc * Gamyzz(i,j,k) + &
+             F1o3 * (gupxy * fxx    + gupyy * fxy    + gupyz * fxz    ) + &
-         TWO * (gupxy_loc * Gamyxy(i,j,k) + gupxz_loc * Gamyxz(i,j,k) + gupyz_loc * Gamyyz(i,j,k))
+                     gupxx * gxxy   + gupyy * gyyy   + gupzz * gzzy    + &
-    Gamza_loc = gupxx_loc * Gamzxx(i,j,k) + gupyy_loc * Gamzyy(i,j,k) + gupzz_loc * Gamzzz(i,j,k) + &
+              TWO * (gupxy * gxyy   + gupxz * gxzy   + gupyz * gyzy  )
-         TWO * (gupxy_loc * Gamzxy(i,j,k) + gupxz_loc * Gamzxz(i,j,k) + gupyz_loc * Gamzyz(i,j,k))
-    Gamxa(i,j,k) = Gamxa_loc
-    Gamya(i,j,k) = Gamya_loc
-    Gamza(i,j,k) = Gamza_loc
 
-    Gamx_rhs(i,j,k) = Gamx_rhs(i,j,k) + F2o3 * Gamxa_loc * divb_loc - &
+  Gamz_rhs =               Gamz_rhs +  F2o3 *  Gamza * div_beta        - &
-         Gamxa_loc * betaxx(i,j,k) - Gamya_loc * betaxy(i,j,k) - Gamza_loc * betaxz(i,j,k) + &
+                     Gamxa * betazx - Gamya * betazy - Gamza * betazz  + &
-         F1o3 * (gupxx_loc * fxx_loc + gupxy_loc * fxy_loc + gupxz_loc * fxz_loc) + &
+             F1o3 * (gupxz * fxx    + gupyz * fxy    + gupzz * fxz    ) + &
-         gupxx_loc * gxxx(i,j,k) + gupyy_loc * gyyx(i,j,k) + gupzz_loc * gzzx(i,j,k) + &
+                     gupxx * gxxz   + gupyy * gyyz   + gupzz * gzzz    + &
-         TWO * (gupxy_loc * gxyx(i,j,k) + gupxz_loc * gxzx(i,j,k) + gupyz_loc * gyzx(i,j,k))
+              TWO * (gupxy * gxyz   + gupxz * gxzz   + gupyz * gyzz  )    !rhs for Gam^i
-
-    Gamy_rhs(i,j,k) = Gamy_rhs(i,j,k) + F2o3 * Gamya_loc * divb_loc - &
-         Gamxa_loc * betayx(i,j,k) - Gamya_loc * betayy(i,j,k) - Gamza_loc * betayz(i,j,k) + &
-         F1o3 * (gupxy_loc * fxx_loc + gupyy_loc * fxy_loc + gupyz_loc * fxz_loc) + &
-         gupxx_loc * gxxy(i,j,k) + gupyy_loc * gyyy(i,j,k) + gupzz_loc * gzzy(i,j,k) + &
-         TWO * (gupxy_loc * gxyy(i,j,k) + gupxz_loc * gxzy(i,j,k) + gupyz_loc * gyzy(i,j,k))
-
-    Gamz_rhs(i,j,k) = Gamz_rhs(i,j,k) + F2o3 * Gamza_loc * divb_loc - &
-         Gamxa_loc * betazx(i,j,k) - Gamya_loc * betazy(i,j,k) - Gamza_loc * betazz(i,j,k) + &
-         F1o3 * (gupxz_loc * fxx_loc + gupyz_loc * fxy_loc + gupzz_loc * fxz_loc) + &
-         gupxx_loc * gxxz(i,j,k) + gupyy_loc * gyyz(i,j,k) + gupzz_loc * gzzz(i,j,k) + &
-         TWO * (gupxy_loc * gxyz(i,j,k) + gupxz_loc * gxzz(i,j,k) + gupyz_loc * gyzz(i,j,k))
-  enddo
-  enddo
-  enddo
 
 !first kind of connection stored in gij,k
   gxxx = gxx * Gamxxx + gxy * Gamyxx + gxz * Gamzxx
@@ -658,187 +604,189 @@
 !covariant second derivative of chi respect to tilted metric
   call fdderivs(ex,chi,fxx,fxy,fxz,fyy,fyz,fzz,X,Y,Z,SYM,SYM,SYM,Symmetry,Lev)
 
-  do k=1,ex(3)
+  fxx = fxx - Gamxxx * chix - Gamyxx * chiy - Gamzxx * chiz
-  do j=1,ex(2)
+  fxy = fxy - Gamxxy * chix - Gamyxy * chiy - Gamzxy * chiz
-  do i=1,ex(1)
+  fxz = fxz - Gamxxz * chix - Gamyxz * chiy - Gamzxz * chiz
-    fxx(i,j,k) = fxx(i,j,k) - Gamxxx(i,j,k) * chix(i,j,k) - Gamyxx(i,j,k) * chiy(i,j,k) - Gamzxx(i,j,k) * chiz(i,j,k)
+  fyy = fyy - Gamxyy * chix - Gamyyy * chiy - Gamzyy * chiz
-    fxy(i,j,k) = fxy(i,j,k) - Gamxxy(i,j,k) * chix(i,j,k) - Gamyxy(i,j,k) * chiy(i,j,k) - Gamzxy(i,j,k) * chiz(i,j,k)
+  fyz = fyz - Gamxyz * chix - Gamyyz * chiy - Gamzyz * chiz
-    fxz(i,j,k) = fxz(i,j,k) - Gamxxz(i,j,k) * chix(i,j,k) - Gamyxz(i,j,k) * chiy(i,j,k) - Gamzxz(i,j,k) * chiz(i,j,k)
+  fzz = fzz - Gamxzz * chix - Gamyzz * chiy - Gamzzz * chiz
-    fyy(i,j,k) = fyy(i,j,k) - Gamxyy(i,j,k) * chix(i,j,k) - Gamyyy(i,j,k) * chiy(i,j,k) - Gamzyy(i,j,k) * chiz(i,j,k)
+! Store D^l D_l chi - 3/(2*chi) D^l chi D_l chi in f
-    fyz(i,j,k) = fyz(i,j,k) - Gamxyz(i,j,k) * chix(i,j,k) - Gamyyz(i,j,k) * chiy(i,j,k) - Gamzyz(i,j,k) * chiz(i,j,k)
-    fzz(i,j,k) = fzz(i,j,k) - Gamxzz(i,j,k) * chix(i,j,k) - Gamyzz(i,j,k) * chiy(i,j,k) - Gamzzz(i,j,k) * chiz(i,j,k)
 
-    chin_loc = chin1(i,j,k)
+  f =        gupxx * ( fxx - F3o2/chin1 * chix * chix ) + &
-    f_loc = gupxx(i,j,k) * (fxx(i,j,k) - F3o2/chin_loc * chix(i,j,k) * chix(i,j,k)) + &
+             gupyy * ( fyy - F3o2/chin1 * chiy * chiy ) + &
-            gupyy(i,j,k) * (fyy(i,j,k) - F3o2/chin_loc * chiy(i,j,k) * chiy(i,j,k)) + &
+             gupzz * ( fzz - F3o2/chin1 * chiz * chiz ) + &
-            gupzz(i,j,k) * (fzz(i,j,k) - F3o2/chin_loc * chiz(i,j,k) * chiz(i,j,k)) + &
+       TWO * gupxy * ( fxy - F3o2/chin1 * chix * chiy ) + &
-            TWO * gupxy(i,j,k) * (fxy(i,j,k) - F3o2/chin_loc * chix(i,j,k) * chiy(i,j,k)) + &
+       TWO * gupxz * ( fxz - F3o2/chin1 * chix * chiz ) + &
-            TWO * gupxz(i,j,k) * (fxz(i,j,k) - F3o2/chin_loc * chix(i,j,k) * chiz(i,j,k)) + &
+       TWO * gupyz * ( fyz - F3o2/chin1 * chiy * chiz ) 
-            TWO * gupyz(i,j,k) * (fyz(i,j,k) - F3o2/chin_loc * chiy(i,j,k) * chiz(i,j,k))
+! Add chi part to Ricci tensor:
-    f(i,j,k) = f_loc
 
-    Rxx(i,j,k) = Rxx(i,j,k) + (fxx(i,j,k) - chix(i,j,k)*chix(i,j,k)/chin_loc/TWO + gxx(i,j,k) * f_loc)/chin_loc/TWO
+  Rxx = Rxx + (fxx - chix*chix/chin1/TWO + gxx * f)/chin1/TWO
-    Ryy(i,j,k) = Ryy(i,j,k) + (fyy(i,j,k) - chiy(i,j,k)*chiy(i,j,k)/chin_loc/TWO + gyy(i,j,k) * f_loc)/chin_loc/TWO
+  Ryy = Ryy + (fyy - chiy*chiy/chin1/TWO + gyy * f)/chin1/TWO
-    Rzz(i,j,k) = Rzz(i,j,k) + (fzz(i,j,k) - chiz(i,j,k)*chiz(i,j,k)/chin_loc/TWO + gzz(i,j,k) * f_loc)/chin_loc/TWO
+  Rzz = Rzz + (fzz - chiz*chiz/chin1/TWO + gzz * f)/chin1/TWO
-    Rxy(i,j,k) = Rxy(i,j,k) + (fxy(i,j,k) - chix(i,j,k)*chiy(i,j,k)/chin_loc/TWO + gxy(i,j,k) * f_loc)/chin_loc/TWO
+  Rxy = Rxy + (fxy - chix*chiy/chin1/TWO + gxy * f)/chin1/TWO
-    Rxz(i,j,k) = Rxz(i,j,k) + (fxz(i,j,k) - chix(i,j,k)*chiz(i,j,k)/chin_loc/TWO + gxz(i,j,k) * f_loc)/chin_loc/TWO
+  Rxz = Rxz + (fxz - chix*chiz/chin1/TWO + gxz * f)/chin1/TWO
-    Ryz(i,j,k) = Ryz(i,j,k) + (fyz(i,j,k) - chiy(i,j,k)*chiz(i,j,k)/chin_loc/TWO + gyz(i,j,k) * f_loc)/chin_loc/TWO
+  Ryz = Ryz + (fyz - chiy*chiz/chin1/TWO + gyz * f)/chin1/TWO
-  enddo
-  enddo
-  enddo
 
 ! covariant second derivatives of the lapse respect to physical metric
   call fdderivs(ex,Lap,fxx,fxy,fxz,fyy,fyz,fzz,X,Y,Z, &
                 SYM,SYM,SYM,symmetry,Lev)
 
-  do k=1,ex(3)
+  gxxx = (gupxx * chix + gupxy * chiy + gupxz * chiz)/chin1
-  do j=1,ex(2)
+  gxxy = (gupxy * chix + gupyy * chiy + gupyz * chiz)/chin1
-  do i=1,ex(1)
+  gxxz = (gupxz * chix + gupyz * chiy + gupzz * chiz)/chin1
-    chin_loc = chin1(i,j,k)
+! now get physical second kind of connection
-    gxxx(i,j,k) = (gupxx(i,j,k) * chix(i,j,k) + gupxy(i,j,k) * chiy(i,j,k) + gupxz(i,j,k) * chiz(i,j,k)) / chin_loc
+  Gamxxx = Gamxxx - ( (chix + chix)/chin1 - gxx * gxxx )*HALF
-    gxxy(i,j,k) = (gupxy(i,j,k) * chix(i,j,k) + gupyy(i,j,k) * chiy(i,j,k) + gupyz(i,j,k) * chiz(i,j,k)) / chin_loc
+  Gamyxx = Gamyxx - (                     - gxx * gxxy )*HALF
-    gxxz(i,j,k) = (gupxz(i,j,k) * chix(i,j,k) + gupyz(i,j,k) * chiy(i,j,k) + gupzz(i,j,k) * chiz(i,j,k)) / chin_loc
+  Gamzxx = Gamzxx - (                     - gxx * gxxz )*HALF
+  Gamxyy = Gamxyy - (                     - gyy * gxxx )*HALF
+  Gamyyy = Gamyyy - ( (chiy + chiy)/chin1 - gyy * gxxy )*HALF
+  Gamzyy = Gamzyy - (                     - gyy * gxxz )*HALF
+  Gamxzz = Gamxzz - (                     - gzz * gxxx )*HALF
+  Gamyzz = Gamyzz - (                     - gzz * gxxy )*HALF
+  Gamzzz = Gamzzz - ( (chiz + chiz)/chin1 - gzz * gxxz )*HALF
+  Gamxxy = Gamxxy - (  chiy        /chin1 - gxy * gxxx )*HALF
+  Gamyxy = Gamyxy - (         chix /chin1 - gxy * gxxy )*HALF
+  Gamzxy = Gamzxy - (                     - gxy * gxxz )*HALF
+  Gamxxz = Gamxxz - (  chiz        /chin1 - gxz * gxxx )*HALF
+  Gamyxz = Gamyxz - (                     - gxz * gxxy )*HALF
+  Gamzxz = Gamzxz - (         chix /chin1 - gxz * gxxz )*HALF
+  Gamxyz = Gamxyz - (                     - gyz * gxxx )*HALF
+  Gamyyz = Gamyyz - (  chiz        /chin1 - gyz * gxxy )*HALF
+  Gamzyz = Gamzyz - (         chiy /chin1 - gyz * gxxz )*HALF
 
-    Gamxxx(i,j,k) = Gamxxx(i,j,k) - ( (chix(i,j,k) + chix(i,j,k))/chin_loc - gxx(i,j,k) * gxxx(i,j,k) )*HALF
+  fxx = fxx - Gamxxx*Lapx - Gamyxx*Lapy - Gamzxx*Lapz
-    Gamyxx(i,j,k) = Gamyxx(i,j,k) - (                                   - gxx(i,j,k) * gxxy(i,j,k) )*HALF
+  fyy = fyy - Gamxyy*Lapx - Gamyyy*Lapy - Gamzyy*Lapz
-    Gamzxx(i,j,k) = Gamzxx(i,j,k) - (                                   - gxx(i,j,k) * gxxz(i,j,k) )*HALF
+  fzz = fzz - Gamxzz*Lapx - Gamyzz*Lapy - Gamzzz*Lapz
-    Gamxyy(i,j,k) = Gamxyy(i,j,k) - (                                   - gyy(i,j,k) * gxxx(i,j,k) )*HALF
+  fxy = fxy - Gamxxy*Lapx - Gamyxy*Lapy - Gamzxy*Lapz
-    Gamyyy(i,j,k) = Gamyyy(i,j,k) - ( (chiy(i,j,k) + chiy(i,j,k))/chin_loc - gyy(i,j,k) * gxxy(i,j,k) )*HALF
+  fxz = fxz - Gamxxz*Lapx - Gamyxz*Lapy - Gamzxz*Lapz
-    Gamzyy(i,j,k) = Gamzyy(i,j,k) - (                                   - gyy(i,j,k) * gxxz(i,j,k) )*HALF
+  fyz = fyz - Gamxyz*Lapx - Gamyyz*Lapy - Gamzyz*Lapz
-    Gamxzz(i,j,k) = Gamxzz(i,j,k) - (                                   - gzz(i,j,k) * gxxx(i,j,k) )*HALF
-    Gamyzz(i,j,k) = Gamyzz(i,j,k) - (                                   - gzz(i,j,k) * gxxy(i,j,k) )*HALF
-    Gamzzz(i,j,k) = Gamzzz(i,j,k) - ( (chiz(i,j,k) + chiz(i,j,k))/chin_loc - gzz(i,j,k) * gxxz(i,j,k) )*HALF
-    Gamxxy(i,j,k) = Gamxxy(i,j,k) - ( chiy(i,j,k) /chin_loc - gxy(i,j,k) * gxxx(i,j,k) )*HALF
-    Gamyxy(i,j,k) = Gamyxy(i,j,k) - ( chix(i,j,k) /chin_loc - gxy(i,j,k) * gxxy(i,j,k) )*HALF
-    Gamzxy(i,j,k) = Gamzxy(i,j,k) - (                     - gxy(i,j,k) * gxxz(i,j,k) )*HALF
-    Gamxxz(i,j,k) = Gamxxz(i,j,k) - ( chiz(i,j,k) /chin_loc - gxz(i,j,k) * gxxx(i,j,k) )*HALF
-    Gamyxz(i,j,k) = Gamyxz(i,j,k) - (                     - gxz(i,j,k) * gxxy(i,j,k) )*HALF
-    Gamzxz(i,j,k) = Gamzxz(i,j,k) - ( chix(i,j,k) /chin_loc - gxz(i,j,k) * gxxz(i,j,k) )*HALF
-    Gamxyz(i,j,k) = Gamxyz(i,j,k) - (                     - gyz(i,j,k) * gxxx(i,j,k) )*HALF
-    Gamyyz(i,j,k) = Gamyyz(i,j,k) - ( chiz(i,j,k) /chin_loc - gyz(i,j,k) * gxxy(i,j,k) )*HALF
-    Gamzyz(i,j,k) = Gamzyz(i,j,k) - ( chiy(i,j,k) /chin_loc - gyz(i,j,k) * gxxz(i,j,k) )*HALF
 
-    fxx(i,j,k) = fxx(i,j,k) - Gamxxx(i,j,k)*Lapx(i,j,k) - Gamyxx(i,j,k)*Lapy(i,j,k) - Gamzxx(i,j,k)*Lapz(i,j,k)
+! store D^i D_i Lap in trK_rhs upto chi
-    fyy(i,j,k) = fyy(i,j,k) - Gamxyy(i,j,k)*Lapx(i,j,k) - Gamyyy(i,j,k)*Lapy(i,j,k) - Gamzyy(i,j,k)*Lapz(i,j,k)
+  trK_rhs =    gupxx * fxx + gupyy * fyy + gupzz * fzz + &
-    fzz(i,j,k) = fzz(i,j,k) - Gamxzz(i,j,k)*Lapx(i,j,k) - Gamyzz(i,j,k)*Lapy(i,j,k) - Gamzzz(i,j,k)*Lapz(i,j,k)
+        TWO* ( gupxy * fxy + gupxz * fxz + gupyz * fyz )
-    fxy(i,j,k) = fxy(i,j,k) - Gamxxy(i,j,k)*Lapx(i,j,k) - Gamyxy(i,j,k)*Lapy(i,j,k) - Gamzxy(i,j,k)*Lapz(i,j,k)
+#if 1        
-    fxz(i,j,k) = fxz(i,j,k) - Gamxxz(i,j,k)*Lapx(i,j,k) - Gamyxz(i,j,k)*Lapy(i,j,k) - Gamzxz(i,j,k)*Lapz(i,j,k)
+!! follow bam code
-    fyz(i,j,k) = fyz(i,j,k) - Gamxyz(i,j,k)*Lapx(i,j,k) - Gamyyz(i,j,k)*Lapy(i,j,k) - Gamzyz(i,j,k)*Lapz(i,j,k)
+  S =  chin1 * ( gupxx * Sxx + gupyy * Syy + gupzz * Szz + &
+     TWO * ( gupxy * Sxy + gupxz * Sxz + gupyz * Syz ) )
+  f = F2o3 * trK * trK -(&
+       gupxx * ( &
+       gupxx * Axx * Axx + gupyy * Axy * Axy + gupzz * Axz * Axz + &
+       TWO * (gupxy * Axx * Axy + gupxz * Axx * Axz + gupyz * Axy * Axz) ) + &
+       gupyy * ( &
+       gupxx * Axy * Axy + gupyy * Ayy * Ayy + gupzz * Ayz * Ayz + &
+       TWO * (gupxy * Axy * Ayy + gupxz * Axy * Ayz + gupyz * Ayy * Ayz) ) + &
+       gupzz * ( &
+       gupxx * Axz * Axz + gupyy * Ayz * Ayz + gupzz * Azz * Azz + &
+       TWO * (gupxy * Axz * Ayz + gupxz * Axz * Azz + gupyz * Ayz * Azz) ) + &
+       TWO * ( &
+       gupxy * ( &
+       gupxx * Axx * Axy + gupyy * Axy * Ayy + gupzz * Axz * Ayz + &
+       gupxy * (Axx * Ayy + Axy * Axy) + &
+       gupxz * (Axx * Ayz + Axz * Axy) + &
+       gupyz * (Axy * Ayz + Axz * Ayy) ) + &
+       gupxz * ( &
+       gupxx * Axx * Axz + gupyy * Axy * Ayz + gupzz * Axz * Azz + &
+       gupxy * (Axx * Ayz + Axy * Axz) + &
+       gupxz * (Axx * Azz + Axz * Axz) + &
+       gupyz * (Axy * Azz + Axz * Ayz) ) + &
+       gupyz * ( &
+       gupxx * Axy * Axz + gupyy * Ayy * Ayz + gupzz * Ayz * Azz + &
+       gupxy * (Axy * Ayz + Ayy * Axz) + &
+       gupxz * (Axy * Azz + Ayz * Axz) + &
+       gupyz * (Ayy * Azz + Ayz * Ayz) ) )) -1.6d1*PI*rho + EIGHT * PI * S
+  f = - F1o3 *(  gupxx * fxx + gupyy * fyy + gupzz * fzz + &
+        TWO* ( gupxy * fxy + gupxz * fxz + gupyz * fyz ) + alpn1/chin1*f)
   
-    trK_rhs(i,j,k) = gupxx(i,j,k) * fxx(i,j,k) + gupyy(i,j,k) * fyy(i,j,k) + gupzz(i,j,k) * fzz(i,j,k) + &
+  fxx = alpn1 * (Rxx - EIGHT * PI * Sxx) - fxx
-                     TWO * (gupxy(i,j,k) * fxy(i,j,k) + gupxz(i,j,k) * fxz(i,j,k) + gupyz(i,j,k) * fyz(i,j,k))
+  fxy = alpn1 * (Rxy - EIGHT * PI * Sxy) - fxy
-  enddo
+  fxz = alpn1 * (Rxz - EIGHT * PI * Sxz) - fxz
-  enddo
+  fyy = alpn1 * (Ryy - EIGHT * PI * Syy) - fyy
-  enddo
+  fyz = alpn1 * (Ryz - EIGHT * PI * Syz) - fyz
-  do k=1,ex(3)
+  fzz = alpn1 * (Rzz - EIGHT * PI * Szz) - fzz
-  do j=1,ex(2)
+#else        
-  do i=1,ex(1)
+! Add lapse and S_ij parts to Ricci tensor:
-    divb_loc = div_beta(i,j,k)
-    chin_loc = chin1(i,j,k)
 
-    S_loc = chin_loc * ( gupxx(i,j,k) * Sxx(i,j,k) + gupyy(i,j,k) * Syy(i,j,k) + gupzz(i,j,k) * Szz(i,j,k) + &
+  fxx = alpn1 * (Rxx - EIGHT * PI * Sxx) - fxx
-           TWO * (gupxy(i,j,k) * Sxy(i,j,k) + gupxz(i,j,k) * Sxz(i,j,k) + gupyz(i,j,k) * Syz(i,j,k)) )
+  fxy = alpn1 * (Rxy - EIGHT * PI * Sxy) - fxy
-    S(i,j,k) = S_loc
+  fxz = alpn1 * (Rxz - EIGHT * PI * Sxz) - fxz
+  fyy = alpn1 * (Ryy - EIGHT * PI * Syy) - fyy
+  fyz = alpn1 * (Ryz - EIGHT * PI * Syz) - fyz
+  fzz = alpn1 * (Rzz - EIGHT * PI * Szz) - fzz
 
-    f_loc = F2o3 * trK(i,j,k) * trK(i,j,k) - ( &
+! Compute trace-free part (note: chi^-1 and chi cancel!):
-            gupxx(i,j,k) * ( gupxx(i,j,k) * Axx(i,j,k) * Axx(i,j,k) + gupyy(i,j,k) * Axy(i,j,k) * Axy(i,j,k) + &
-                             gupzz(i,j,k) * Axz(i,j,k) * Axz(i,j,k) + &
-                             TWO * (gupxy(i,j,k) * Axx(i,j,k) * Axy(i,j,k) + gupxz(i,j,k) * Axx(i,j,k) * Axz(i,j,k) + &
-                                    gupyz(i,j,k) * Axy(i,j,k) * Axz(i,j,k)) ) + &
-            gupyy(i,j,k) * ( gupxx(i,j,k) * Axy(i,j,k) * Axy(i,j,k) + gupyy(i,j,k) * Ayy(i,j,k) * Ayy(i,j,k) + &
-                             gupzz(i,j,k) * Ayz(i,j,k) * Ayz(i,j,k) + &
-                             TWO * (gupxy(i,j,k) * Axy(i,j,k) * Ayy(i,j,k) + gupxz(i,j,k) * Axy(i,j,k) * Ayz(i,j,k) + &
-                                    gupyz(i,j,k) * Ayy(i,j,k) * Ayz(i,j,k)) ) + &
-            gupzz(i,j,k) * ( gupxx(i,j,k) * Axz(i,j,k) * Axz(i,j,k) + gupyy(i,j,k) * Ayz(i,j,k) * Ayz(i,j,k) + &
-                             gupzz(i,j,k) * Azz(i,j,k) * Azz(i,j,k) + &
-                             TWO * (gupxy(i,j,k) * Axz(i,j,k) * Ayz(i,j,k) + gupxz(i,j,k) * Axz(i,j,k) * Azz(i,j,k) + &
-                                    gupyz(i,j,k) * Ayz(i,j,k) * Azz(i,j,k)) ) + &
-            TWO * ( gupxy(i,j,k) * ( gupxx(i,j,k) * Axx(i,j,k) * Axy(i,j,k) + gupyy(i,j,k) * Axy(i,j,k) * Ayy(i,j,k) + &
-                                     gupzz(i,j,k) * Axz(i,j,k) * Ayz(i,j,k) + &
-                                     gupxy(i,j,k) * (Axx(i,j,k) * Ayy(i,j,k) + Axy(i,j,k) * Axy(i,j,k)) + &
-                                     gupxz(i,j,k) * (Axx(i,j,k) * Ayz(i,j,k) + Axz(i,j,k) * Axy(i,j,k)) + &
-                                     gupyz(i,j,k) * (Axy(i,j,k) * Ayz(i,j,k) + Axz(i,j,k) * Ayy(i,j,k)) ) + &
-                    gupxz(i,j,k) * ( gupxx(i,j,k) * Axx(i,j,k) * Axz(i,j,k) + gupyy(i,j,k) * Axy(i,j,k) * Ayz(i,j,k) + &
-                                     gupzz(i,j,k) * Axz(i,j,k) * Azz(i,j,k) + &
-                                     gupxy(i,j,k) * (Axx(i,j,k) * Ayz(i,j,k) + Axy(i,j,k) * Axz(i,j,k)) + &
-                                     gupxz(i,j,k) * (Axx(i,j,k) * Azz(i,j,k) + Axz(i,j,k) * Axz(i,j,k)) + &
-                                     gupyz(i,j,k) * (Axy(i,j,k) * Azz(i,j,k) + Axz(i,j,k) * Ayz(i,j,k)) ) + &
-                    gupyz(i,j,k) * ( gupxx(i,j,k) * Axy(i,j,k) * Axz(i,j,k) + gupyy(i,j,k) * Ayy(i,j,k) * Ayz(i,j,k) + &
-                                     gupzz(i,j,k) * Ayz(i,j,k) * Azz(i,j,k) + &
-                                     gupxy(i,j,k) * (Axy(i,j,k) * Ayz(i,j,k) + Ayy(i,j,k) * Axz(i,j,k)) + &
-                                     gupxz(i,j,k) * (Axy(i,j,k) * Azz(i,j,k) + Ayz(i,j,k) * Axz(i,j,k)) + &
-                                     gupyz(i,j,k) * (Ayy(i,j,k) * Azz(i,j,k) + Ayz(i,j,k) * Ayz(i,j,k)) ) ) ) - &
-            F16 * PI * rho(i,j,k) + EIGHT * PI * S_loc
 
-    f_loc = -F1o3 * ( gupxx(i,j,k) * fxx(i,j,k) + gupyy(i,j,k) * fyy(i,j,k) + gupzz(i,j,k) * fzz(i,j,k) + &
+  f = F1o3 *(  gupxx * fxx + gupyy * fyy + gupzz * fzz + &
-            TWO * (gupxy(i,j,k) * fxy(i,j,k) + gupxz(i,j,k) * fxz(i,j,k) + gupyz(i,j,k) * fyz(i,j,k)) + &
+        TWO* ( gupxy * fxy + gupxz * fxz + gupyz * fyz ) )
-            alpn1(i,j,k)/chin_loc * f_loc )
+#endif
-    f(i,j,k) = f_loc
 
-    l_fxx = alpn1(i,j,k) * (Rxx(i,j,k) - EIGHT * PI * Sxx(i,j,k)) - fxx(i,j,k)
+  Axx_rhs = fxx - gxx * f
-    l_fxy = alpn1(i,j,k) * (Rxy(i,j,k) - EIGHT * PI * Sxy(i,j,k)) - fxy(i,j,k)
+  Ayy_rhs = fyy - gyy * f
-    l_fxz = alpn1(i,j,k) * (Rxz(i,j,k) - EIGHT * PI * Sxz(i,j,k)) - fxz(i,j,k)
+  Azz_rhs = fzz - gzz * f
-    l_fyy = alpn1(i,j,k) * (Ryy(i,j,k) - EIGHT * PI * Syy(i,j,k)) - fyy(i,j,k)
+  Axy_rhs = fxy - gxy * f
-    l_fyz = alpn1(i,j,k) * (Ryz(i,j,k) - EIGHT * PI * Syz(i,j,k)) - fyz(i,j,k)
+  Axz_rhs = fxz - gxz * f
-    l_fzz = alpn1(i,j,k) * (Rzz(i,j,k) - EIGHT * PI * Szz(i,j,k)) - fzz(i,j,k)
+  Ayz_rhs = fyz - gyz * f
 
-    Axx_rhs(i,j,k) = l_fxx - gxx(i,j,k) * f_loc
+! Now: store A_il A^l_j into fij:
-    Ayy_rhs(i,j,k) = l_fyy - gyy(i,j,k) * f_loc
-    Azz_rhs(i,j,k) = l_fzz - gzz(i,j,k) * f_loc
-    Axy_rhs(i,j,k) = l_fxy - gxy(i,j,k) * f_loc
-    Axz_rhs(i,j,k) = l_fxz - gxz(i,j,k) * f_loc
-    Ayz_rhs(i,j,k) = l_fyz - gyz(i,j,k) * f_loc
 
-    fxx(i,j,k) = gupxx(i,j,k) * Axx(i,j,k) * Axx(i,j,k) + gupyy(i,j,k) * Axy(i,j,k) * Axy(i,j,k) + &
+  fxx =       gupxx * Axx * Axx + gupyy * Axy * Axy + gupzz * Axz * Axz + &
-                 gupzz(i,j,k) * Axz(i,j,k) * Axz(i,j,k) + TWO * (gupxy(i,j,k) * Axx(i,j,k) * Axy(i,j,k) + &
+       TWO * (gupxy * Axx * Axy + gupxz * Axx * Axz + gupyz * Axy * Axz)
-                 gupxz(i,j,k) * Axx(i,j,k) * Axz(i,j,k) + gupyz(i,j,k) * Axy(i,j,k) * Axz(i,j,k))
+  fyy =       gupxx * Axy * Axy + gupyy * Ayy * Ayy + gupzz * Ayz * Ayz + &
-    fyy(i,j,k) = gupxx(i,j,k) * Axy(i,j,k) * Axy(i,j,k) + gupyy(i,j,k) * Ayy(i,j,k) * Ayy(i,j,k) + &
+       TWO * (gupxy * Axy * Ayy + gupxz * Axy * Ayz + gupyz * Ayy * Ayz)
-                 gupzz(i,j,k) * Ayz(i,j,k) * Ayz(i,j,k) + TWO * (gupxy(i,j,k) * Axy(i,j,k) * Ayy(i,j,k) + &
+  fzz =       gupxx * Axz * Axz + gupyy * Ayz * Ayz + gupzz * Azz * Azz + &
-                 gupxz(i,j,k) * Axy(i,j,k) * Ayz(i,j,k) + gupyz(i,j,k) * Ayy(i,j,k) * Ayz(i,j,k))
+       TWO * (gupxy * Axz * Ayz + gupxz * Axz * Azz + gupyz * Ayz * Azz)
-    fzz(i,j,k) = gupxx(i,j,k) * Axz(i,j,k) * Axz(i,j,k) + gupyy(i,j,k) * Ayz(i,j,k) * Ayz(i,j,k) + &
+  fxy =       gupxx * Axx * Axy + gupyy * Axy * Ayy + gupzz * Axz * Ayz + &
-                 gupzz(i,j,k) * Azz(i,j,k) * Azz(i,j,k) + TWO * (gupxy(i,j,k) * Axz(i,j,k) * Ayz(i,j,k) + &
+              gupxy *(Axx * Ayy + Axy * Axy)                            + &
-                 gupxz(i,j,k) * Axz(i,j,k) * Azz(i,j,k) + gupyz(i,j,k) * Ayz(i,j,k) * Azz(i,j,k))
+              gupxz *(Axx * Ayz + Axz * Axy)                            + &
-    fxy(i,j,k) = gupxx(i,j,k) * Axx(i,j,k) * Axy(i,j,k) + gupyy(i,j,k) * Axy(i,j,k) * Ayy(i,j,k) + &
+              gupyz *(Axy * Ayz + Axz * Ayy)
-                 gupzz(i,j,k) * Axz(i,j,k) * Ayz(i,j,k) + gupxy(i,j,k) * (Axx(i,j,k) * Ayy(i,j,k) + Axy(i,j,k) * Axy(i,j,k)) + &
+  fxz =       gupxx * Axx * Axz + gupyy * Axy * Ayz + gupzz * Axz * Azz + &
-                 gupxz(i,j,k) * (Axx(i,j,k) * Ayz(i,j,k) + Axz(i,j,k) * Axy(i,j,k)) + &
+              gupxy *(Axx * Ayz + Axy * Axz)                            + &
-                 gupyz(i,j,k) * (Axy(i,j,k) * Ayz(i,j,k) + Axz(i,j,k) * Ayy(i,j,k))
+              gupxz *(Axx * Azz + Axz * Axz)                            + &
-    fxz(i,j,k) = gupxx(i,j,k) * Axx(i,j,k) * Axz(i,j,k) + gupyy(i,j,k) * Axy(i,j,k) * Ayz(i,j,k) + &
+              gupyz *(Axy * Azz + Axz * Ayz)
-                 gupzz(i,j,k) * Axz(i,j,k) * Azz(i,j,k) + gupxy(i,j,k) * (Axx(i,j,k) * Ayz(i,j,k) + Axy(i,j,k) * Axz(i,j,k)) + &
+  fyz =       gupxx * Axy * Axz + gupyy * Ayy * Ayz + gupzz * Ayz * Azz + &
-                 gupxz(i,j,k) * (Axx(i,j,k) * Azz(i,j,k) + Axz(i,j,k) * Axz(i,j,k)) + &
+              gupxy *(Axy * Ayz + Ayy * Axz)                            + &
-                 gupyz(i,j,k) * (Axy(i,j,k) * Azz(i,j,k) + Axz(i,j,k) * Ayz(i,j,k))
+              gupxz *(Axy * Azz + Ayz * Axz)                            + &
-    fyz(i,j,k) = gupxx(i,j,k) * Axy(i,j,k) * Axz(i,j,k) + gupyy(i,j,k) * Ayy(i,j,k) * Ayz(i,j,k) + &
+              gupyz *(Ayy * Azz + Ayz * Ayz)
-                 gupzz(i,j,k) * Ayz(i,j,k) * Azz(i,j,k) + gupxy(i,j,k) * (Axy(i,j,k) * Ayz(i,j,k) + Ayy(i,j,k) * Axz(i,j,k)) + &
-                 gupxz(i,j,k) * (Axy(i,j,k) * Azz(i,j,k) + Ayz(i,j,k) * Axz(i,j,k)) + &
-                 gupyz(i,j,k) * (Ayy(i,j,k) * Azz(i,j,k) + Ayz(i,j,k) * Ayz(i,j,k))
 
-    trK_rhs(i,j,k) = chin_loc * trK_rhs(i,j,k)
+  f = chin1
+! store D^i D_i Lap in trK_rhs
+  trK_rhs = f*trK_rhs
           
-    Axx_rhs(i,j,k) = chin_loc * Axx_rhs(i,j,k) + alpn1(i,j,k) * (trK(i,j,k) * Axx(i,j,k) - TWO * fxx(i,j,k)) + &
+  Axx_rhs =           f * Axx_rhs+ alpn1 * (trK * Axx - TWO * fxx)  + &
-                     TWO * (Axx(i,j,k) * betaxx(i,j,k) + Axy(i,j,k) * betayx(i,j,k) + Axz(i,j,k) * betazx(i,j,k)) - &
+           TWO * (  Axx * betaxx +   Axy * betayx +   Axz * betazx )- &
-                     F2o3 * Axx(i,j,k) * divb_loc
+             F2o3 * Axx * div_beta
-    Ayy_rhs(i,j,k) = chin_loc * Ayy_rhs(i,j,k) + alpn1(i,j,k) * (trK(i,j,k) * Ayy(i,j,k) - TWO * fyy(i,j,k)) + &
-                     TWO * (Axy(i,j,k) * betaxy(i,j,k) + Ayy(i,j,k) * betayy(i,j,k) + Ayz(i,j,k) * betazy(i,j,k)) - &
-                     F2o3 * Ayy(i,j,k) * divb_loc
-    Azz_rhs(i,j,k) = chin_loc * Azz_rhs(i,j,k) + alpn1(i,j,k) * (trK(i,j,k) * Azz(i,j,k) - TWO * fzz(i,j,k)) + &
-                     TWO * (Axz(i,j,k) * betaxz(i,j,k) + Ayz(i,j,k) * betayz(i,j,k) + Azz(i,j,k) * betazz(i,j,k)) - &
-                     F2o3 * Azz(i,j,k) * divb_loc
-    Axy_rhs(i,j,k) = chin_loc * Axy_rhs(i,j,k) + alpn1(i,j,k) * (trK(i,j,k) * Axy(i,j,k) - TWO * fxy(i,j,k)) + &
-                     Axx(i,j,k) * betaxy(i,j,k) + Axz(i,j,k) * betazy(i,j,k) + Ayy(i,j,k) * betayx(i,j,k) + &
-                     Ayz(i,j,k) * betazx(i,j,k) + F1o3 * Axy(i,j,k) * divb_loc - Axy(i,j,k) * betazz(i,j,k)
-    Ayz_rhs(i,j,k) = chin_loc * Ayz_rhs(i,j,k) + alpn1(i,j,k) * (trK(i,j,k) * Ayz(i,j,k) - TWO * fyz(i,j,k)) + &
-                     Axy(i,j,k) * betaxz(i,j,k) + Ayy(i,j,k) * betayz(i,j,k) + Axz(i,j,k) * betaxy(i,j,k) + &
-                     Azz(i,j,k) * betazy(i,j,k) + F1o3 * Ayz(i,j,k) * divb_loc - Ayz(i,j,k) * betaxx(i,j,k)
-    Axz_rhs(i,j,k) = chin_loc * Axz_rhs(i,j,k) + alpn1(i,j,k) * (trK(i,j,k) * Axz(i,j,k) - TWO * fxz(i,j,k)) + &
-                     Axx(i,j,k) * betaxz(i,j,k) + Axy(i,j,k) * betayz(i,j,k) + Ayz(i,j,k) * betayx(i,j,k) + &
-                     Azz(i,j,k) * betazx(i,j,k) + F1o3 * Axz(i,j,k) * divb_loc - Axz(i,j,k) * betayy(i,j,k)
 
-    trK_rhs(i,j,k) = - trK_rhs(i,j,k) + alpn1(i,j,k) * ( F1o3 * trK(i,j,k) * trK(i,j,k) + &
+  Ayy_rhs =           f * Ayy_rhs+ alpn1 * (trK * Ayy - TWO * fyy)  + &
-                    gupxx(i,j,k) * fxx(i,j,k) + gupyy(i,j,k) * fyy(i,j,k) + gupzz(i,j,k) * fzz(i,j,k) + &
+           TWO * (  Axy * betaxy +   Ayy * betayy +   Ayz * betazy )- &
-                    TWO * (gupxy(i,j,k) * fxy(i,j,k) + gupxz(i,j,k) * fxz(i,j,k) + gupyz(i,j,k) * fyz(i,j,k)) + &
+             F2o3 * Ayy * div_beta
-                    FOUR * PI * (rho(i,j,k) + S_loc) )
+
-  enddo
+  Azz_rhs =           f * Azz_rhs+ alpn1 * (trK * Azz - TWO * fzz)  + &
-  enddo
+           TWO * (  Axz * betaxz +   Ayz * betayz +   Azz * betazz )- &
-  enddo
+             F2o3 * Azz * div_beta
+
+  Axy_rhs =           f * Axy_rhs+ alpn1 *( trK * Axy  - TWO * fxy )+ &
+                    Axx * betaxy                  +   Axz * betazy  + &
+                                     Ayy * betayx +   Ayz * betazx  + &
+             F1o3 * Axy * div_beta                -   Axy * betazz
+
+  Ayz_rhs =           f * Ayz_rhs+ alpn1 *( trK * Ayz  - TWO * fyz )+ &
+                    Axy * betaxz +   Ayy * betayz                   + &
+                    Axz * betaxy                  +   Azz * betazy  + &
+             F1o3 * Ayz * div_beta                -   Ayz * betaxx
+ 
+  Axz_rhs =           f * Axz_rhs+ alpn1 *( trK * Axz  - TWO * fxz )+ &
+                    Axx * betaxz +   Axy * betayz                   + &
+                                     Ayz * betayx +   Azz * betazx  + &
+             F1o3 * Axz * div_beta                -   Axz * betayy      !rhs for Aij
+
+! Compute trace of S_ij
+
+  S =  f * ( gupxx * Sxx + gupyy * Syy + gupzz * Szz + &
+     TWO * ( gupxy * Sxy + gupxz * Sxz + gupyz * Syz ) )
+
+  trK_rhs = - trK_rhs + alpn1 *( F1o3 * trK * trK         + &
+                gupxx * fxx + gupyy * fyy + gupzz * fzz   + &
+        TWO * ( gupxy * fxy + gupxz * fxz + gupyz * fyz ) + &
+       FOUR * PI * ( rho + S ))                                !rhs for trK
   
 !!!! gauge variable part
 
@@ -1000,15 +948,15 @@
 !!!!!!!!!advection term + Kreiss-Oliger dissipation (merged for cache efficiency)
 ! lopsided_kodis shares the symmetry_bd buffer between advection and
 ! dissipation, eliminating redundant full-grid copies. For metric variables
-! gxx/gyy/gzz (=dxx/dyy/dzz+1): stencil coefficients sum to zero,
+! gxx/gyy/gzz (=dxx/dyy/dzz+1): kodis stencil coefficients sum to zero,
 ! so the constant offset has no effect on dissipation.
 
-  call lopsided_kodis(ex,X,Y,Z,dxx,gxx_rhs,betax,betay,betaz,Symmetry,SSS,eps)
+  call lopsided_kodis(ex,X,Y,Z,gxx,gxx_rhs,betax,betay,betaz,Symmetry,SSS,eps)
   call lopsided_kodis(ex,X,Y,Z,gxy,gxy_rhs,betax,betay,betaz,Symmetry,AAS,eps)
   call lopsided_kodis(ex,X,Y,Z,gxz,gxz_rhs,betax,betay,betaz,Symmetry,ASA,eps)
-  call lopsided_kodis(ex,X,Y,Z,dyy,gyy_rhs,betax,betay,betaz,Symmetry,SSS,eps)
+  call lopsided_kodis(ex,X,Y,Z,gyy,gyy_rhs,betax,betay,betaz,Symmetry,SSS,eps)
   call lopsided_kodis(ex,X,Y,Z,gyz,gyz_rhs,betax,betay,betaz,Symmetry,SAA,eps)
-  call lopsided_kodis(ex,X,Y,Z,dzz,gzz_rhs,betax,betay,betaz,Symmetry,SSS,eps)
+  call lopsided_kodis(ex,X,Y,Z,gzz,gzz_rhs,betax,betay,betaz,Symmetry,SSS,eps)
 
   call lopsided_kodis(ex,X,Y,Z,Axx,Axx_rhs,betax,betay,betaz,Symmetry,SSS,eps)
   call lopsided_kodis(ex,X,Y,Z,Axy,Axy_rhs,betax,betay,betaz,Symmetry,AAS,eps)

AMSS_NCKU_source/cgh.C

@@ -27,10 +27,6 @@ using namespace std;
 #include "cgh.h"
 #include "Parallel.h"
 #include "parameters.h"
-#ifdef USE_GPU
-#include "bssn_gpu.h"
-#include "bssn_cuda_ops.h"
-#endif
 
 //================================================================================================
 
@@ -47,6 +43,14 @@ cgh::cgh(int ingfsi, int fngfsi, int Symmetry, char *filename, int checkrun,
   end_rank = 0;
 #endif
 
+  // Initialize load balancing variables
+  enable_load_balance = false;
+  load_balance_check_interval = 10;  // Check every 10 time steps
+  current_time_step = 0;
+  rank_interp_times = nullptr;
+  heavy_ranks = nullptr;
+  num_heavy_ranks = 0;
+
   if (!checkrun)
   {
     read_bbox(Symmetry, filename);
@@ -117,6 +121,12 @@ cgh::~cgh()
     delete[] Porgls[lev];
   }
   delete[] Porgls;
+
+  // Clean up load balancing memory
+  if (rank_interp_times)
+    delete[] rank_interp_times;
+  if (heavy_ranks)
+    delete[] heavy_ranks;
 }
 
 //================================================================================================
@@ -134,11 +144,7 @@ void cgh::compose_cgh(int nprocs)
   for (int lev = 0; lev < levels; lev++)
   {
     checkPatchList(PatL[lev], false);
-#ifdef INTERP_LB_OPTIMIZE
+    Parallel::distribute_hard(PatL[lev], nprocs, ingfs, fngfs, false);
-    Parallel::distribute_optimize(PatL[lev], nprocs, ingfs, fngfs, false);
-#else
-    Parallel::distribute(PatL[lev], nprocs, ingfs, fngfs, false);
-#endif
 #if (RPB == 1)
     // we need distributed box of PatL[lev] and PatL[lev-1]
     if (lev > 0)
@@ -889,13 +895,6 @@ void cgh::recompose_cgh(int nprocs, bool *lev_flag,
       bool CC = (lev > trfls);
       Parallel::fill_level_data(tmPat, PatL[lev], PatL[lev - 1], OldList, StateList, FutureList, tmList, Symmetry, BB, CC);
 
-#ifdef USE_GPU
-      bssn_gpu_clear_cached_device_buffers();
-      bssn_gpu_release_pinned_host_buffers();
-      bssn_cuda_release_rk4_caches();
-      bssn_cuda_release_interp_caches();
-      patch_release_interp_plan_cache();
-#endif
       Parallel::KillBlocks(PatL[lev]);
       PatL[lev]->destroyList();
       PatL[lev] = tmPat;
@@ -925,13 +924,6 @@ void cgh::recompose_cgh(int nprocs, bool *lev_flag,
       bool CC = (lev > trfls);
       Parallel::fill_level_data(tmPat, PatL[lev], PatL[lev - 1], OldList, StateList, FutureList, tmList, Symmetry, BB, CC);
 
-#ifdef USE_GPU
-      bssn_gpu_clear_cached_device_buffers();
-      bssn_gpu_release_pinned_host_buffers();
-      bssn_cuda_release_rk4_caches();
-      bssn_cuda_release_interp_caches();
-      patch_release_interp_plan_cache();
-#endif
       Parallel::KillBlocks(PatL[lev]);
       PatL[lev]->destroyList();
       PatL[lev] = tmPat;
@@ -1323,13 +1315,13 @@ bool cgh::Interp_One_Point(MyList<var> *VarList,
 }
 
 
-bool cgh::Regrid_Onelevel(int lev, int Symmetry, int BH_num, double **Porgbr, double **Porg0,
+void cgh::Regrid_Onelevel(int lev, int Symmetry, int BH_num, double **Porgbr, double **Porg0,
                           MyList<var> *OldList, MyList<var> *StateList,
                           MyList<var> *FutureList, MyList<var> *tmList, bool BB,
                           monitor *ErrorMonitor)
 {
   if (lev < movls)
-    return false;
+    return;
 
 #if (0)
   // #if (PSTR == 1 || PSTR == 2)
@@ -1418,7 +1410,7 @@ bool cgh::Regrid_Onelevel(int lev, int Symmetry, int BH_num, double **Porgbr, do
       for (bhi = 0; bhi < BH_num; bhi++)
         delete[] tmpPorg[bhi];
       delete[] tmpPorg;
-      return false;
+      return;
     }
     // x direction
     rr = (Porg0[bhi][0] - handle[lev][grd][0]) / dX;
@@ -1522,7 +1514,6 @@ bool cgh::Regrid_Onelevel(int lev, int Symmetry, int BH_num, double **Porgbr, do
   for (int bhi = 0; bhi < BH_num; bhi++)
     delete[] tmpPorg[bhi];
   delete[] tmpPorg;
-  return tot_flag;
 }
 
 
@@ -1540,13 +1531,6 @@ void cgh::recompose_cgh_Onelevel(int nprocs, int lev,
   bool CC = (lev > trfls);
   Parallel::fill_level_data(tmPat, PatL[lev], PatL[lev - 1], OldList, StateList, FutureList, tmList, Symmetry, BB, CC);
 
-#ifdef USE_GPU
-  bssn_gpu_clear_cached_device_buffers();
-  bssn_gpu_release_pinned_host_buffers();
-  bssn_cuda_release_rk4_caches();
-  bssn_cuda_release_interp_caches();
-  patch_release_interp_plan_cache();
-#endif
   Parallel::KillBlocks(PatL[lev]);
   PatL[lev]->destroyList();
   PatL[lev] = tmPat;
@@ -1570,13 +1554,6 @@ void cgh::recompose_cgh_Onelevel(int nprocs, int lev,
   Parallel::fill_level_data(tmPat, PatL[lev], PatL[lev - 1], OldList, StateList, FutureList, tmList, Symmetry, BB, CC);
   misc::tillherecheck(Commlev[lev], start_rank[lev], "after fill_level_data");
 
-#ifdef USE_GPU
-  bssn_gpu_clear_cached_device_buffers();
-  bssn_gpu_release_pinned_host_buffers();
-  bssn_cuda_release_rk4_caches();
-  bssn_cuda_release_interp_caches();
-  patch_release_interp_plan_cache();
-#endif
   Parallel::KillBlocks(PatL[lev]);
   PatL[lev]->destroyList();
   PatL[lev] = tmPat;
@@ -1742,3 +1719,121 @@ void cgh::settrfls(const int lev)
 {
   trfls = lev;
 }
+
+//================================================================================================
+// Load Balancing Functions
+//================================================================================================
+
+// Initialize load balancing
+void cgh::init_load_balance(int nprocs)
+{
+  if (rank_interp_times)
+    delete[] rank_interp_times;
+  if (heavy_ranks)
+    delete[] heavy_ranks;
+
+  rank_interp_times = new double[nprocs];
+  heavy_ranks = new int[4];  // Maximum 4 heavy ranks
+  num_heavy_ranks = 0;
+
+  for (int i = 0; i < nprocs; i++)
+    rank_interp_times[i] = 0.0;
+}
+
+// Update interpolation time for a rank
+void cgh::update_interp_time(int rank, double time)
+{
+  if (rank_interp_times && rank >= 0)
+  {
+    rank_interp_times[rank] = time;
+  }
+}
+
+// Check and perform load balancing if needed
+bool cgh::check_and_rebalance(int nprocs, int lev,
+                               MyList<var> *OldList, MyList<var> *StateList,
+                               MyList<var> *FutureList, MyList<var> *tmList,
+                               int Symmetry, bool BB)
+{
+  int myrank;
+  MPI_Comm_rank(MPI_COMM_WORLD, &myrank);
+
+  // Only check at specified intervals
+  current_time_step++;
+  if (current_time_step % load_balance_check_interval != 0)
+    return false;
+
+  if (myrank == 0)
+  {
+    cout << "\n=== Checking load balance at time step " << current_time_step << " ===" << endl;
+  }
+
+  // Collect all rank times on rank 0
+  double *all_times = nullptr;
+  if (myrank == 0)
+  {
+    all_times = new double[nprocs];
+  }
+
+  MPI_Gather(rank_interp_times, 1, MPI_DOUBLE, all_times, 1, MPI_DOUBLE, 0, MPI_COMM_WORLD);
+
+  bool need_rebalance = false;
+
+  if (myrank == 0)
+  {
+    // Check if load balancing is needed
+    need_rebalance = Parallel::check_load_balance_need(all_times, nprocs, num_heavy_ranks, heavy_ranks);
+
+    if (need_rebalance)
+    {
+      cout << "=== Load imbalance detected! Need to rebalance ===" << endl;
+      cout << "Top " << num_heavy_ranks << " heavy ranks: ";
+      for (int i = 0; i < num_heavy_ranks; i++)
+      {
+        cout << heavy_ranks[i] << " (" << all_times[heavy_ranks[i]] << " s) ";
+      }
+      cout << endl;
+
+      // Analyze blocks that need to be split
+      Parallel::split_heavy_blocks(PatL[lev], heavy_ranks, num_heavy_ranks, 2, nprocs, ingfs, fngfs);
+
+      // Set lev_flag to trigger recompose_cgh
+      cout << "=== Triggering recompose_cgh for level " << lev << " ===" << endl;
+    }
+    else
+    {
+      cout << "=== Load is balanced, no rebalancing needed ===" << endl;
+    }
+
+    delete[] all_times;
+  }
+
+  // Broadcast the decision to all ranks
+  MPI_Bcast(&need_rebalance, 1, MPI_C_BOOL, 0, MPI_COMM_WORLD);
+
+  if (need_rebalance)
+  {
+    // Broadcast heavy ranks information
+    MPI_Bcast(&num_heavy_ranks, 1, MPI_INT, 0, MPI_COMM_WORLD);
+    MPI_Bcast(heavy_ranks, num_heavy_ranks, MPI_INT, 0, MPI_COMM_WORLD);
+
+    // Perform recompose_cgh on the specified level
+    if (myrank == 0)
+    {
+      cout << "=== Performing recompose_cgh ===" << endl;
+    }
+
+    // Call recompose_cgh_Onelevel for the specified level
+    bool *lev_flag = new bool[1];
+    lev_flag[0] = true;
+    recompose_cgh_Onelevel(nprocs, lev, OldList, StateList, FutureList, tmList, Symmetry, BB);
+    delete[] lev_flag;
+
+    // Reset time counter after rebalancing
+    current_time_step = 0;
+
+    return true;
+  }
+
+  return false;
+}

AMSS_NCKU_source/cgh.h

@@ -74,7 +74,7 @@ public:
                                MyList<var> *OldList, MyList<var> *StateList,
                                MyList<var> *FutureList, MyList<var> *tmList,
                                int Symmetry, bool BB);
-   bool Regrid_Onelevel(int lev, int Symmetry, int BH_num, double **Porgbr, double **Porg0,
+   void Regrid_Onelevel(int lev, int Symmetry, int BH_num, double **Porgbr, double **Porg0,
                         MyList<var> *OldList, MyList<var> *StateList,
                         MyList<var> *FutureList, MyList<var> *tmList, bool BB,
                         monitor *ErrorMonitor);
@@ -87,6 +87,21 @@ public:
 #if (PSTR == 1 || PSTR == 2 || PSTR == 3)
    void construct_mylev(int nprocs);
 #endif
+
+   // Load balancing support
+   bool enable_load_balance;         // Enable load balancing
+   int load_balance_check_interval;  // Check interval (in time steps)
+   int current_time_step;            // Current time step counter
+   double *rank_interp_times;        // Store interpolation times for each rank
+   int *heavy_ranks;                 // Store heavy rank numbers
+   int num_heavy_ranks;              // Number of heavy ranks
+
+   void init_load_balance(int nprocs);
+   void update_interp_time(int rank, double time);
+   bool check_and_rebalance(int nprocs, int lev,
+                           MyList<var> *OldList, MyList<var> *StateList,
+                           MyList<var> *FutureList, MyList<var> *tmList,
+                           int Symmetry, bool BB);
 };
 
 #endif /* CGH_H */

AMSS_NCKU_source/diff_newwb.f90

@@ -1513,7 +1513,6 @@
   real*8,dimension(-1:ex(1),-1:ex(2),-1:ex(3))   :: fh
   real*8, dimension(3) :: SoA
   integer :: imin,jmin,kmin,imax,jmax,kmax,i,j,k
-  integer :: i_core_min,i_core_max,j_core_min,j_core_max,k_core_min,k_core_max
   real*8  :: Sdxdx,Sdydy,Sdzdz,Fdxdx,Fdydy,Fdzdz
   real*8  :: Sdxdy,Sdxdz,Sdydz,Fdxdy,Fdxdz,Fdydz
   integer, parameter :: NO_SYMM = 0, EQ_SYMM = 1, OCTANT = 2
@@ -1566,47 +1565,9 @@
   fxz = ZEO
   fyz = ZEO
 
-  i_core_min = max(1, imin+2)
-  i_core_max = min(ex(1), imax-2)
-  j_core_min = max(1, jmin+2)
-  j_core_max = min(ex(2), jmax-2)
-  k_core_min = max(1, kmin+2)
-  k_core_max = min(ex(3), kmax-2)
-
-  if(i_core_min <= i_core_max .and. j_core_min <= j_core_max .and. k_core_min <= k_core_max)then
-   do k=k_core_min,k_core_max
-   do j=j_core_min,j_core_max
-   do i=i_core_min,i_core_max
-! interior points always use 4th-order stencils without branch checks
-      fxx(i,j,k) = Fdxdx*(-fh(i-2,j,k)+F16*fh(i-1,j,k)-F30*fh(i,j,k) &
-                          -fh(i+2,j,k)+F16*fh(i+1,j,k)              )
-      fyy(i,j,k) = Fdydy*(-fh(i,j-2,k)+F16*fh(i,j-1,k)-F30*fh(i,j,k) &
-                          -fh(i,j+2,k)+F16*fh(i,j+1,k)              )
-      fzz(i,j,k) = Fdzdz*(-fh(i,j,k-2)+F16*fh(i,j,k-1)-F30*fh(i,j,k) &
-                          -fh(i,j,k+2)+F16*fh(i,j,k+1)              )
-      fxy(i,j,k) = Fdxdy*(     (fh(i-2,j-2,k)-F8*fh(i-1,j-2,k)+F8*fh(i+1,j-2,k)-fh(i+2,j-2,k))  &
-                          -F8 *(fh(i-2,j-1,k)-F8*fh(i-1,j-1,k)+F8*fh(i+1,j-1,k)-fh(i+2,j-1,k))  &
-                          +F8 *(fh(i-2,j+1,k)-F8*fh(i-1,j+1,k)+F8*fh(i+1,j+1,k)-fh(i+2,j+1,k))  &
-                          -    (fh(i-2,j+2,k)-F8*fh(i-1,j+2,k)+F8*fh(i+1,j+2,k)-fh(i+2,j+2,k)))
-      fxz(i,j,k) = Fdxdz*(     (fh(i-2,j,k-2)-F8*fh(i-1,j,k-2)+F8*fh(i+1,j,k-2)-fh(i+2,j,k-2))  &
-                          -F8 *(fh(i-2,j,k-1)-F8*fh(i-1,j,k-1)+F8*fh(i+1,j,k-1)-fh(i+2,j,k-1))  &
-                          +F8 *(fh(i-2,j,k+1)-F8*fh(i-1,j,k+1)+F8*fh(i+1,j,k+1)-fh(i+2,j,k+1))  &
-                          -    (fh(i-2,j,k+2)-F8*fh(i-1,j,k+2)+F8*fh(i+1,j,k+2)-fh(i+2,j,k+2)))
-      fyz(i,j,k) = Fdydz*(     (fh(i,j-2,k-2)-F8*fh(i,j-1,k-2)+F8*fh(i,j+1,k-2)-fh(i,j+2,k-2))  &
-                          -F8 *(fh(i,j-2,k-1)-F8*fh(i,j-1,k-1)+F8*fh(i,j+1,k-1)-fh(i,j+2,k-1))  &
-                          +F8 *(fh(i,j-2,k+1)-F8*fh(i,j-1,k+1)+F8*fh(i,j+1,k+1)-fh(i,j+2,k+1))  &
-                          -    (fh(i,j-2,k+2)-F8*fh(i,j-1,k+2)+F8*fh(i,j+1,k+2)-fh(i,j+2,k+2)))
-   enddo
-   enddo
-   enddo
-  endif
-
   do k=1,ex(3)
   do j=1,ex(2)
   do i=1,ex(1)
-      if(i>=i_core_min .and. i<=i_core_max .and. &
-         j>=j_core_min .and. j<=j_core_max .and. &
-         k>=k_core_min .and. k<=k_core_max) cycle
 !~~~~~~ fxx
         if(i+2 <= imax .and. i-2 >= imin)then
 !

AMSS_NCKU_source/fdderivs_c.C

@@ -1,332 +0,0 @@
-#include "tool.h"
-void fdderivs(const int ex[3],
-              const double *f,
-              double *fxx, double *fxy, double *fxz,
-              double *fyy, double *fyz, double *fzz,
-              const double *X, const double *Y, const double *Z,
-              double SYM1, double SYM2, double SYM3,
-              int Symmetry, int onoff)
-{
-    (void)onoff;
-
-    const int NO_SYMM = 0, EQ_SYMM = 1;
-    const double ZEO = 0.0, ONE = 1.0, TWO = 2.0;
-    const double F1o4   = 2.5e-1;          // 1/4
-    const double F8     = 8.0;
-    const double F16    = 16.0;
-    const double F30    = 30.0;
-    const double F1o12  = ONE / 12.0;
-    const double F1o144 = ONE / 144.0;
-
-    const int ex1 = ex[0], ex2 = ex[1], ex3 = ex[2];
-
-    const double dX = X[1] - X[0];
-    const double dY = Y[1] - Y[0];
-    const double dZ = Z[1] - Z[0];
-
-    const int imaxF = ex1;
-    const int jmaxF = ex2;
-    const int kmaxF = ex3;
-
-    int iminF = 1, jminF = 1, kminF = 1;
-    if (Symmetry > NO_SYMM && fabs(Z[0]) < dZ) kminF = -1;
-    if (Symmetry > EQ_SYMM && fabs(X[0]) < dX) iminF = -1;
-    if (Symmetry > EQ_SYMM && fabs(Y[0]) < dY) jminF = -1;
-
-    const double SoA[3] = { SYM1, SYM2, SYM3 };
-
-    /* fh: (ex1+2)*(ex2+2)*(ex3+2) because ord=2 */
-    const size_t nx = (size_t)ex1 + 2;
-    const size_t ny = (size_t)ex2 + 2;
-    const size_t nz = (size_t)ex3 + 2;
-    const size_t fh_size = nx * ny * nz;
-
-    static double *fh = NULL;
-    static size_t cap = 0;
-
-    if (fh_size > cap) {
-        free(fh);
-        fh = (double*)aligned_alloc(64, fh_size * sizeof(double));
-        cap = fh_size;
-    }
-    // double *fh = (double*)malloc(fh_size * sizeof(double));
-    if (!fh) return;
-
-    symmetry_bd(2, ex, f, fh, SoA);
-
-    /* 系数：按 Fortran 原式 */
-    const double Sdxdx = ONE / (dX * dX);
-    const double Sdydy = ONE / (dY * dY);
-    const double Sdzdz = ONE / (dZ * dZ);
-
-    const double Fdxdx = F1o12 / (dX * dX);
-    const double Fdydy = F1o12 / (dY * dY);
-    const double Fdzdz = F1o12 / (dZ * dZ);
-
-    const double Sdxdy = F1o4 / (dX * dY);
-    const double Sdxdz = F1o4 / (dX * dZ);
-    const double Sdydz = F1o4 / (dY * dZ);
-
-    const double Fdxdy = F1o144 / (dX * dY);
-    const double Fdxdz = F1o144 / (dX * dZ);
-    const double Fdydz = F1o144 / (dY * dZ);
-
-    /* 只清零不被主循环覆盖的边界面 */
-    {
-        /* 高边界：k0=ex3-1 */
-        for (int j0 = 0; j0 < ex2; ++j0)
-            for (int i0 = 0; i0 < ex1; ++i0) {
-                const size_t p = idx_ex(i0, j0, ex3 - 1, ex);
-                fxx[p]=ZEO; fyy[p]=ZEO; fzz[p]=ZEO;
-                fxy[p]=ZEO; fxz[p]=ZEO; fyz[p]=ZEO;
-            }
-        /* 高边界：j0=ex2-1 */
-        for (int k0 = 0; k0 < ex3 - 1; ++k0)
-            for (int i0 = 0; i0 < ex1; ++i0) {
-                const size_t p = idx_ex(i0, ex2 - 1, k0, ex);
-                fxx[p]=ZEO; fyy[p]=ZEO; fzz[p]=ZEO;
-                fxy[p]=ZEO; fxz[p]=ZEO; fyz[p]=ZEO;
-            }
-        /* 高边界：i0=ex1-1 */
-        for (int k0 = 0; k0 < ex3 - 1; ++k0)
-            for (int j0 = 0; j0 < ex2 - 1; ++j0) {
-                const size_t p = idx_ex(ex1 - 1, j0, k0, ex);
-                fxx[p]=ZEO; fyy[p]=ZEO; fzz[p]=ZEO;
-                fxy[p]=ZEO; fxz[p]=ZEO; fyz[p]=ZEO;
-            }
-
-        /* 低边界：当二阶模板也不可用时，对应 i0/j0/k0=0 面 */
-        if (kminF == 1) {
-            for (int j0 = 0; j0 < ex2; ++j0)
-                for (int i0 = 0; i0 < ex1; ++i0) {
-                    const size_t p = idx_ex(i0, j0, 0, ex);
-                    fxx[p]=ZEO; fyy[p]=ZEO; fzz[p]=ZEO;
-                    fxy[p]=ZEO; fxz[p]=ZEO; fyz[p]=ZEO;
-                }
-        }
-        if (jminF == 1) {
-            for (int k0 = 0; k0 < ex3; ++k0)
-                for (int i0 = 0; i0 < ex1; ++i0) {
-                    const size_t p = idx_ex(i0, 0, k0, ex);
-                    fxx[p]=ZEO; fyy[p]=ZEO; fzz[p]=ZEO;
-                    fxy[p]=ZEO; fxz[p]=ZEO; fyz[p]=ZEO;
-                }
-        }
-        if (iminF == 1) {
-            for (int k0 = 0; k0 < ex3; ++k0)
-                for (int j0 = 0; j0 < ex2; ++j0) {
-                    const size_t p = idx_ex(0, j0, k0, ex);
-                    fxx[p]=ZEO; fyy[p]=ZEO; fzz[p]=ZEO;
-                    fxy[p]=ZEO; fxz[p]=ZEO; fyz[p]=ZEO;
-                }
-        }
-    }
-
-    /*
-     * 两段式：
-     * 1) 二阶可用区域先计算二阶模板
-     * 2) 高阶可用区域再覆盖四阶模板
-     */
-    const int i2_lo = (iminF > 0) ? iminF : 0;
-    const int j2_lo = (jminF > 0) ? jminF : 0;
-    const int k2_lo = (kminF > 0) ? kminF : 0;
-    const int i2_hi = ex1 - 2;
-    const int j2_hi = ex2 - 2;
-    const int k2_hi = ex3 - 2;
-
-    const int i4_lo = (iminF + 1 > 0) ? (iminF + 1) : 0;
-    const int j4_lo = (jminF + 1 > 0) ? (jminF + 1) : 0;
-    const int k4_lo = (kminF + 1 > 0) ? (kminF + 1) : 0;
-    const int i4_hi = ex1 - 3;
-    const int j4_hi = ex2 - 3;
-    const int k4_hi = ex3 - 3;
-
-    /*
-     * Strategy A:
-     * Avoid redundant work in overlap of 2nd/4th-order regions.
-     * Only compute 2nd-order on shell points that are NOT overwritten by
-     * the 4th-order pass.
-     */
-    const int has4 = (i4_lo <= i4_hi && j4_lo <= j4_hi && k4_lo <= k4_hi);
-
-    if (i2_lo <= i2_hi && j2_lo <= j2_hi && k2_lo <= k2_hi) {
-        for (int k0 = k2_lo; k0 <= k2_hi; ++k0) {
-            const int kF = k0 + 1;
-            for (int j0 = j2_lo; j0 <= j2_hi; ++j0) {
-                const int jF = j0 + 1;
-                for (int i0 = i2_lo; i0 <= i2_hi; ++i0) {
-                    if (has4 &&
-                        i0 >= i4_lo && i0 <= i4_hi &&
-                        j0 >= j4_lo && j0 <= j4_hi &&
-                        k0 >= k4_lo && k0 <= k4_hi) {
-                        continue;
-                    }
-                    const int iF = i0 + 1;
-                    const size_t p = idx_ex(i0, j0, k0, ex);
-
-                    fxx[p] = Sdxdx * (
-                        fh[idx_fh_F_ord2(iF - 1, jF,     kF,     ex)] -
-                        TWO * fh[idx_fh_F_ord2(iF,     jF,     kF,     ex)] +
-                        fh[idx_fh_F_ord2(iF + 1, jF,     kF,     ex)]
-                    );
-
-                    fyy[p] = Sdydy * (
-                        fh[idx_fh_F_ord2(iF,     jF - 1, kF,     ex)] -
-                        TWO * fh[idx_fh_F_ord2(iF,     jF,     kF,     ex)] +
-                        fh[idx_fh_F_ord2(iF,     jF + 1, kF,     ex)]
-                    );
-
-                    fzz[p] = Sdzdz * (
-                        fh[idx_fh_F_ord2(iF,     jF,     kF - 1, ex)] -
-                        TWO * fh[idx_fh_F_ord2(iF,     jF,     kF,     ex)] +
-                        fh[idx_fh_F_ord2(iF,     jF,     kF + 1, ex)]
-                    );
-
-                    fxy[p] = Sdxdy * (
-                        fh[idx_fh_F_ord2(iF - 1, jF - 1, kF, ex)] -
-                        fh[idx_fh_F_ord2(iF + 1, jF - 1, kF, ex)] -
-                        fh[idx_fh_F_ord2(iF - 1, jF + 1, kF, ex)] +
-                        fh[idx_fh_F_ord2(iF + 1, jF + 1, kF, ex)]
-                    );
-
-                    fxz[p] = Sdxdz * (
-                        fh[idx_fh_F_ord2(iF - 1, jF, kF - 1, ex)] -
-                        fh[idx_fh_F_ord2(iF + 1, jF, kF - 1, ex)] -
-                        fh[idx_fh_F_ord2(iF - 1, jF, kF + 1, ex)] +
-                        fh[idx_fh_F_ord2(iF + 1, jF, kF + 1, ex)]
-                    );
-
-                    fyz[p] = Sdydz * (
-                        fh[idx_fh_F_ord2(iF, jF - 1, kF - 1, ex)] -
-                        fh[idx_fh_F_ord2(iF, jF + 1, kF - 1, ex)] -
-                        fh[idx_fh_F_ord2(iF, jF - 1, kF + 1, ex)] +
-                        fh[idx_fh_F_ord2(iF, jF + 1, kF + 1, ex)]
-                    );
-                }
-            }
-        }
-    }
-
-    if (has4) {
-        for (int k0 = k4_lo; k0 <= k4_hi; ++k0) {
-            const int kF = k0 + 1;
-            for (int j0 = j4_lo; j0 <= j4_hi; ++j0) {
-                const int jF = j0 + 1;
-                for (int i0 = i4_lo; i0 <= i4_hi; ++i0) {
-                    const int iF = i0 + 1;
-                    const size_t p = idx_ex(i0, j0, k0, ex);
-
-                    fxx[p] = Fdxdx * (
-                        -fh[idx_fh_F_ord2(iF - 2, jF,     kF,     ex)] +
-                        F16 * fh[idx_fh_F_ord2(iF - 1, jF,     kF,     ex)] -
-                        F30 * fh[idx_fh_F_ord2(iF,     jF,     kF,     ex)] -
-                        fh[idx_fh_F_ord2(iF + 2, jF,     kF,     ex)] +
-                        F16 * fh[idx_fh_F_ord2(iF + 1, jF,     kF,     ex)]
-                    );
-
-                    fyy[p] = Fdydy * (
-                        -fh[idx_fh_F_ord2(iF,     jF - 2, kF,     ex)] +
-                        F16 * fh[idx_fh_F_ord2(iF,     jF - 1, kF,     ex)] -
-                        F30 * fh[idx_fh_F_ord2(iF,     jF,     kF,     ex)] -
-                        fh[idx_fh_F_ord2(iF,     jF + 2, kF,     ex)] +
-                        F16 * fh[idx_fh_F_ord2(iF,     jF + 1, kF,     ex)]
-                    );
-
-                    fzz[p] = Fdzdz * (
-                        -fh[idx_fh_F_ord2(iF,     jF,     kF - 2, ex)] +
-                        F16 * fh[idx_fh_F_ord2(iF,     jF,     kF - 1, ex)] -
-                        F30 * fh[idx_fh_F_ord2(iF,     jF,     kF,     ex)] -
-                        fh[idx_fh_F_ord2(iF,     jF,     kF + 2, ex)] +
-                        F16 * fh[idx_fh_F_ord2(iF,     jF,     kF + 1, ex)]
-                    );
-
-                    {
-                        const double t_jm2 =
-                            ( fh[idx_fh_F_ord2(iF - 2, jF - 2, kF, ex)]
-                             -F8*fh[idx_fh_F_ord2(iF - 1, jF - 2, kF, ex)]
-                             +F8*fh[idx_fh_F_ord2(iF + 1, jF - 2, kF, ex)]
-                             -    fh[idx_fh_F_ord2(iF + 2, jF - 2, kF, ex)] );
-
-                        const double t_jm1 =
-                            ( fh[idx_fh_F_ord2(iF - 2, jF - 1, kF, ex)]
-                             -F8*fh[idx_fh_F_ord2(iF - 1, jF - 1, kF, ex)]
-                             +F8*fh[idx_fh_F_ord2(iF + 1, jF - 1, kF, ex)]
-                             -    fh[idx_fh_F_ord2(iF + 2, jF - 1, kF, ex)] );
-
-                        const double t_jp1 =
-                            ( fh[idx_fh_F_ord2(iF - 2, jF + 1, kF, ex)]
-                             -F8*fh[idx_fh_F_ord2(iF - 1, jF + 1, kF, ex)]
-                             +F8*fh[idx_fh_F_ord2(iF + 1, jF + 1, kF, ex)]
-                             -    fh[idx_fh_F_ord2(iF + 2, jF + 1, kF, ex)] );
-
-                        const double t_jp2 =
-                            ( fh[idx_fh_F_ord2(iF - 2, jF + 2, kF, ex)]
-                             -F8*fh[idx_fh_F_ord2(iF - 1, jF + 2, kF, ex)]
-                             +F8*fh[idx_fh_F_ord2(iF + 1, jF + 2, kF, ex)]
-                             -    fh[idx_fh_F_ord2(iF + 2, jF + 2, kF, ex)] );
-
-                        fxy[p] = Fdxdy * ( t_jm2 - F8 * t_jm1 + F8 * t_jp1 - t_jp2 );
-                    }
-
-                    {
-                        const double t_km2 =
-                            ( fh[idx_fh_F_ord2(iF - 2, jF, kF - 2, ex)]
-                             -F8*fh[idx_fh_F_ord2(iF - 1, jF, kF - 2, ex)]
-                             +F8*fh[idx_fh_F_ord2(iF + 1, jF, kF - 2, ex)]
-                             -    fh[idx_fh_F_ord2(iF + 2, jF, kF - 2, ex)] );
-
-                        const double t_km1 =
-                            ( fh[idx_fh_F_ord2(iF - 2, jF, kF - 1, ex)]
-                             -F8*fh[idx_fh_F_ord2(iF - 1, jF, kF - 1, ex)]
-                             +F8*fh[idx_fh_F_ord2(iF + 1, jF, kF - 1, ex)]
-                             -    fh[idx_fh_F_ord2(iF + 2, jF, kF - 1, ex)] );
-
-                        const double t_kp1 =
-                            ( fh[idx_fh_F_ord2(iF - 2, jF, kF + 1, ex)]
-                             -F8*fh[idx_fh_F_ord2(iF - 1, jF, kF + 1, ex)]
-                             +F8*fh[idx_fh_F_ord2(iF + 1, jF, kF + 1, ex)]
-                             -    fh[idx_fh_F_ord2(iF + 2, jF, kF + 1, ex)] );
-
-                        const double t_kp2 =
-                            ( fh[idx_fh_F_ord2(iF - 2, jF, kF + 2, ex)]
-                             -F8*fh[idx_fh_F_ord2(iF - 1, jF, kF + 2, ex)]
-                             +F8*fh[idx_fh_F_ord2(iF + 1, jF, kF + 2, ex)]
-                             -    fh[idx_fh_F_ord2(iF + 2, jF, kF + 2, ex)] );
-
-                        fxz[p] = Fdxdz * ( t_km2 - F8 * t_km1 + F8 * t_kp1 - t_kp2 );
-                    }
-
-                    {
-                        const double t_km2 =
-                            ( fh[idx_fh_F_ord2(iF, jF - 2, kF - 2, ex)]
-                             -F8*fh[idx_fh_F_ord2(iF, jF - 1, kF - 2, ex)]
-                             +F8*fh[idx_fh_F_ord2(iF, jF + 1, kF - 2, ex)]
-                             -    fh[idx_fh_F_ord2(iF, jF + 2, kF - 2, ex)] );
-
-                        const double t_km1 =
-                            ( fh[idx_fh_F_ord2(iF, jF - 2, kF - 1, ex)]
-                             -F8*fh[idx_fh_F_ord2(iF, jF - 1, kF - 1, ex)]
-                             +F8*fh[idx_fh_F_ord2(iF, jF + 1, kF - 1, ex)]
-                             -    fh[idx_fh_F_ord2(iF, jF + 2, kF - 1, ex)] );
-
-                        const double t_kp1 =
-                            ( fh[idx_fh_F_ord2(iF, jF - 2, kF + 1, ex)]
-                             -F8*fh[idx_fh_F_ord2(iF, jF - 1, kF + 1, ex)]
-                             +F8*fh[idx_fh_F_ord2(iF, jF + 1, kF + 1, ex)]
-                             -    fh[idx_fh_F_ord2(iF, jF + 2, kF + 1, ex)] );
-
-                        const double t_kp2 =
-                            ( fh[idx_fh_F_ord2(iF, jF - 2, kF + 2, ex)]
-                             -F8*fh[idx_fh_F_ord2(iF, jF - 1, kF + 2, ex)]
-                             +F8*fh[idx_fh_F_ord2(iF, jF + 1, kF + 2, ex)]
-                             -    fh[idx_fh_F_ord2(iF, jF + 2, kF + 2, ex)] );
-
-                        fyz[p] = Fdydz * ( t_km2 - F8 * t_km1 + F8 * t_kp1 - t_kp2 );
-                    }
-                }
-            }
-        }
-    }
-
-    // free(fh);
-}

AMSS_NCKU_source/fderivs_c.C

@@ -1,167 +0,0 @@
-#include "tool.h"
-
-/*
- * C 版 fderivs
- *
- * Fortran:
- * subroutine fderivs(ex,f,fx,fy,fz,X,Y,Z,SYM1,SYM2,SYM3,symmetry,onoff)
- *
- * 约定：
- *   f, fx, fy, fz: ex1*ex2*ex3，按 idx_ex 布局
- *   X: ex1, Y: ex2, Z: ex3
- */
-void fderivs(const int ex[3],
-             const double *f,
-             double *fx, double *fy, double *fz,
-             const double *X, const double *Y, const double *Z,
-             double SYM1, double SYM2, double SYM3,
-             int Symmetry, int onoff)
-{
-    (void)onoff; // Fortran 里没用到
-
-    const double ZEO = 0.0, ONE = 1.0;
-    const double TWO = 2.0, EIT = 8.0;
-    const double F12 = 12.0;
-
-    const int NO_SYMM = 0, EQ_SYMM = 1; // OCTANT=2 在本子程序里不直接用
-
-    const int ex1 = ex[0], ex2 = ex[1], ex3 = ex[2];
-
-    // dX = X(2)-X(1) -> C: X[1]-X[0]
-    const double dX = X[1] - X[0];
-    const double dY = Y[1] - Y[0];
-    const double dZ = Z[1] - Z[0];
-
-    // Fortran 1-based bounds
-    const int imaxF = ex1;
-    const int jmaxF = ex2;
-    const int kmaxF = ex3;
-
-    int iminF = 1, jminF = 1, kminF = 1;
-    if (Symmetry > NO_SYMM && fabs(Z[0]) < dZ) kminF = -1;
-    if (Symmetry > EQ_SYMM && fabs(X[0]) < dX) iminF = -1;
-    if (Symmetry > EQ_SYMM && fabs(Y[0]) < dY) jminF = -1;
-
-    // SoA(1:3) = SYM1,SYM2,SYM3
-    const double SoA[3] = { SYM1, SYM2, SYM3 };
-
-    // fh: (ex1+2)*(ex2+2)*(ex3+2) because ord=2
-    const size_t nx = (size_t)ex1 + 2;
-    const size_t ny = (size_t)ex2 + 2;
-    const size_t nz = (size_t)ex3 + 2;
-    const size_t fh_size = nx * ny * nz;
-    static double *fh = NULL;
-    static size_t cap = 0;
-
-    if (fh_size > cap) {
-        free(fh);
-        fh = (double*)aligned_alloc(64, fh_size * sizeof(double));
-        cap = fh_size;
-    }
-    // double *fh = (double*)malloc(fh_size * sizeof(double));
-    if (!fh) return;
-
-    // call symmetry_bd(2,ex,f,fh,SoA)
-    symmetry_bd(2, ex, f, fh, SoA);
-
-    const double d12dx = ONE / F12 / dX;
-    const double d12dy = ONE / F12 / dY;
-    const double d12dz = ONE / F12 / dZ;
-
-    const double d2dx  = ONE / TWO / dX;
-    const double d2dy  = ONE / TWO / dY;
-    const double d2dz  = ONE / TWO / dZ;
-
-    // fx = fy = fz = 0
-    const size_t all = (size_t)ex1 * (size_t)ex2 * (size_t)ex3;
-    for (size_t p = 0; p < all; ++p) {
-        fx[p] = ZEO;
-        fy[p] = ZEO;
-        fz[p] = ZEO;
-    }
-
-    /*
-     * 两段式：
-     * 1) 先在二阶可用区域计算二阶模板
-     * 2) 再在高阶可用区域覆盖为四阶模板
-     *
-     * 与原 if/elseif 逻辑等价，但减少逐点分支判断。
-     */
-    const int i2_lo = (iminF > 0) ? iminF : 0;
-    const int j2_lo = (jminF > 0) ? jminF : 0;
-    const int k2_lo = (kminF > 0) ? kminF : 0;
-    const int i2_hi = ex1 - 2;
-    const int j2_hi = ex2 - 2;
-    const int k2_hi = ex3 - 2;
-
-    const int i4_lo = (iminF + 1 > 0) ? (iminF + 1) : 0;
-    const int j4_lo = (jminF + 1 > 0) ? (jminF + 1) : 0;
-    const int k4_lo = (kminF + 1 > 0) ? (kminF + 1) : 0;
-    const int i4_hi = ex1 - 3;
-    const int j4_hi = ex2 - 3;
-    const int k4_hi = ex3 - 3;
-
-    if (i2_lo <= i2_hi && j2_lo <= j2_hi && k2_lo <= k2_hi) {
-        for (int k0 = k2_lo; k0 <= k2_hi; ++k0) {
-            const int kF = k0 + 1;
-            for (int j0 = j2_lo; j0 <= j2_hi; ++j0) {
-                const int jF = j0 + 1;
-                for (int i0 = i2_lo; i0 <= i2_hi; ++i0) {
-                    const int iF = i0 + 1;
-                    const size_t p = idx_ex(i0, j0, k0, ex);
-
-                    fx[p] = d2dx * (
-                        -fh[idx_fh_F_ord2(iF - 1, jF,     kF,     ex)] +
-                         fh[idx_fh_F_ord2(iF + 1, jF,     kF,     ex)]
-                    );
-
-                    fy[p] = d2dy * (
-                        -fh[idx_fh_F_ord2(iF,     jF - 1, kF,     ex)] +
-                         fh[idx_fh_F_ord2(iF,     jF + 1, kF,     ex)]
-                    );
-
-                    fz[p] = d2dz * (
-                        -fh[idx_fh_F_ord2(iF,     jF,     kF - 1, ex)] +
-                         fh[idx_fh_F_ord2(iF,     jF,     kF + 1, ex)]
-                    );
-                }
-            }
-        }
-    }
-
-    if (i4_lo <= i4_hi && j4_lo <= j4_hi && k4_lo <= k4_hi) {
-        for (int k0 = k4_lo; k0 <= k4_hi; ++k0) {
-            const int kF = k0 + 1;
-            for (int j0 = j4_lo; j0 <= j4_hi; ++j0) {
-                const int jF = j0 + 1;
-                for (int i0 = i4_lo; i0 <= i4_hi; ++i0) {
-                    const int iF = i0 + 1;
-                    const size_t p = idx_ex(i0, j0, k0, ex);
-
-                    fx[p] = d12dx * (
-                        fh[idx_fh_F_ord2(iF - 2, jF,     kF,     ex)] -
-                        EIT * fh[idx_fh_F_ord2(iF - 1, jF,     kF,     ex)] +
-                        EIT * fh[idx_fh_F_ord2(iF + 1, jF,     kF,     ex)] -
-                        fh[idx_fh_F_ord2(iF + 2, jF,     kF,     ex)]
-                    );
-
-                    fy[p] = d12dy * (
-                        fh[idx_fh_F_ord2(iF,     jF - 2, kF,     ex)] -
-                        EIT * fh[idx_fh_F_ord2(iF,     jF - 1, kF,     ex)] +
-                        EIT * fh[idx_fh_F_ord2(iF,     jF + 1, kF,     ex)] -
-                        fh[idx_fh_F_ord2(iF,     jF + 2, kF,     ex)]
-                    );
-
-                    fz[p] = d12dz * (
-                        fh[idx_fh_F_ord2(iF,     jF,     kF - 2, ex)] -
-                        EIT * fh[idx_fh_F_ord2(iF,     jF,     kF - 1, ex)] +
-                        EIT * fh[idx_fh_F_ord2(iF,     jF,     kF + 1, ex)] -
-                        fh[idx_fh_F_ord2(iF,     jF,     kF + 2, ex)]
-                    );
-                }
-            }
-        }
-    }
-
-    // free(fh);
-}

AMSS_NCKU_source/fmisc.f90

@@ -1115,147 +1115,6 @@ end subroutine d2dump
 !------------------------------------------------------------------------------
 ! Lagrangian polynomial interpolation
 !------------------------------------------------------------------------------
-#ifndef POLINT6_USE_BARYCENTRIC
-#define POLINT6_USE_BARYCENTRIC 1
-#endif
-
-!DIR$ ATTRIBUTES FORCEINLINE :: polint6_neville
-  subroutine polint6_neville(xa, ya, x, y, dy)
-  implicit none
-
-  real*8, dimension(6), intent(in) :: xa, ya
-  real*8, intent(in) :: x
-  real*8, intent(out) :: y, dy
-
-  integer :: i, m, ns, n_m
-  real*8, dimension(6) :: c, d, ho
-  real*8 :: dif, dift, hp, h, den_val
-
-  c = ya
-  d = ya
-  ho = xa - x
-
-  ns = 1
-  dif = abs(x - xa(1))
-
-  do i = 2, 6
-    dift = abs(x - xa(i))
-    if (dift < dif) then
-      ns = i
-      dif = dift
-    end if
-  end do
-
-  y = ya(ns)
-  ns = ns - 1
-
-  do m = 1, 5
-    n_m = 6 - m
-    do i = 1, n_m
-      hp = ho(i)
-      h  = ho(i+m)
-      den_val = hp - h
-
-      if (den_val == 0.0d0) then
-        write(*,*) 'failure in polint for point',x
-        write(*,*) 'with input points: ',xa
-        stop
-      end if
-
-      den_val = (c(i+1) - d(i)) / den_val
-
-      d(i) = h * den_val
-      c(i) = hp * den_val
-    end do
-
-    if (2 * ns < n_m) then
-      dy = c(ns + 1)
-    else
-      dy = d(ns)
-      ns = ns - 1
-    end if
-    y = y + dy
-  end do
-
-  return
-  end subroutine polint6_neville
-
-!DIR$ ATTRIBUTES FORCEINLINE :: polint6_barycentric
-  subroutine polint6_barycentric(xa, ya, x, y, dy)
-  implicit none
-
-  real*8, dimension(6), intent(in) :: xa, ya
-  real*8, intent(in) :: x
-  real*8, intent(out) :: y, dy
-
-  integer :: i, j
-  logical :: is_uniform
-  real*8, dimension(6) :: lambda
-  real*8 :: dx, den_i, term, num, den, step, tol
-  real*8, parameter :: c_uniform(6) = (/ -1.d0, 5.d0, -10.d0, 10.d0, -5.d0, 1.d0 /)
-
-  do i = 1, 6
-    if (x == xa(i)) then
-      y = ya(i)
-      dy = 0.d0
-      return
-    end if
-  end do
-
-  step = xa(2) - xa(1)
-  is_uniform = (step /= 0.d0)
-  if (is_uniform) then
-    tol = 64.d0 * epsilon(1.d0) * max(1.d0, abs(step))
-    do i = 3, 6
-      if (abs((xa(i) - xa(i-1)) - step) > tol) then
-        is_uniform = .false.
-        exit
-      end if
-    end do
-  end if
-
-  if (is_uniform) then
-    num = 0.d0
-    den = 0.d0
-    do i = 1, 6
-      term = c_uniform(i) / (x - xa(i))
-      num = num + term * ya(i)
-      den = den + term
-    end do
-    y = num / den
-    dy = 0.d0
-    return
-  end if
-
-  do i = 1, 6
-    den_i = 1.d0
-    do j = 1, 6
-      if (j /= i) then
-        dx = xa(i) - xa(j)
-        if (dx == 0.0d0) then
-          write(*,*) 'failure in polint for point',x
-          write(*,*) 'with input points: ',xa
-          stop
-        end if
-        den_i = den_i * dx
-      end if
-    end do
-    lambda(i) = 1.d0 / den_i
-  end do
-
-  num = 0.d0
-  den = 0.d0
-  do i = 1, 6
-    term = lambda(i) / (x - xa(i))
-    num = num + term * ya(i)
-    den = den + term
-  end do
-
-  y = num / den
-  dy = 0.d0
-
-  return
-  end subroutine polint6_barycentric
 
 !DIR$ ATTRIBUTES FORCEINLINE :: polint
   subroutine polint(xa, ya, x, y, dy, ordn)
@@ -1270,15 +1129,6 @@ end subroutine d2dump
   real*8, dimension(ordn) :: c, d, ho
   real*8 :: dif, dift, hp, h, den_val
 
-  if (ordn == 6) then
-#if POLINT6_USE_BARYCENTRIC
-    call polint6_barycentric(xa, ya, x, y, dy)
-#else
-    call polint6_neville(xa, ya, x, y, dy)
-#endif
-    return
-  end if
-
   c = ya
   d = ya
   ho = xa - x
@@ -1328,41 +1178,6 @@ end subroutine d2dump
   return
   end subroutine polint
 !------------------------------------------------------------------------------
-! Compute Lagrange interpolation basis weights for one target point.
-!------------------------------------------------------------------------------
-!DIR$ ATTRIBUTES FORCEINLINE :: polint_lagrange_weights
-  subroutine polint_lagrange_weights(xa, x, w, ordn)
-  implicit none
-
-  integer, intent(in) :: ordn
-  real*8, dimension(1:ordn), intent(in) :: xa
-  real*8, intent(in) :: x
-  real*8, dimension(1:ordn), intent(out) :: w
-
-  integer :: i, j
-  real*8 :: num, den, dx
-
-  do i = 1, ordn
-    num = 1.d0
-    den = 1.d0
-    do j = 1, ordn
-      if (j /= i) then
-        dx = xa(i) - xa(j)
-        if (dx == 0.0d0) then
-          write(*,*) 'failure in polint for point',x
-          write(*,*) 'with input points: ',xa
-          stop
-        end if
-        num = num * (x - xa(j))
-        den = den * dx
-      end if
-    end do
-    w(i) = num / den
-  end do
-
-  return
-  end subroutine polint_lagrange_weights
-!------------------------------------------------------------------------------
 !
 ! interpolation in 2 dimensions, follow yx order
 !
@@ -1433,26 +1248,19 @@ end subroutine d2dump
   end do
   call polint(x1a,ymtmp,x1,y,dy,ordn)
 #else
-  integer  :: i, j, k
+  integer  :: j, k
-  real*8, dimension(ordn) :: w1, w2
+  real*8, dimension(ordn,ordn) :: yatmp
   real*8, dimension(ordn) :: ymtmp
-  real*8 :: yx_sum, x_sum
+  real*8 :: dy_temp
 
-  call polint_lagrange_weights(x1a, x1, w1, ordn)
+  do k=1,ordn
-  call polint_lagrange_weights(x2a, x2, w2, ordn)
+    do j=1,ordn
-
+      call polint(x1a, ya(:,j,k), x1, yatmp(j,k), dy_temp, ordn)
-  do k = 1, ordn
-    yx_sum = 0.d0
-    do j = 1, ordn
-      x_sum = 0.d0
-      do i = 1, ordn
-        x_sum = x_sum + w1(i) * ya(i,j,k)
-      end do
-      yx_sum = yx_sum + w2(j) * x_sum
     end do
-    ymtmp(k) = yx_sum
   end do
-
+  do k=1,ordn
+    call polint(x2a, yatmp(:,k), x2, ymtmp(k), dy_temp, ordn)
+  end do
   call polint(x3a, ymtmp, x3, y, dy, ordn)
 #endif
 
@@ -1801,11 +1609,8 @@ deallocate(f_flat)
 ! f=3/8*f_1 + 3/4*f_2 - 1/8*f_3
 
   real*8,parameter::C1=3.d0/8.d0,C2=3.d0/4.d0,C3=-1.d0/8.d0
-  integer :: i,j,k
 
-  do concurrent (k=1:ext(3), j=1:ext(2), i=1:ext(1))
+  fout = C1*f1+C2*f2+C3*f3
-    fout(i,j,k) = C1*f1(i,j,k)+C2*f2(i,j,k)+C3*f3(i,j,k)
-  end do
 
   return
 

AMSS_NCKU_source/interp_lb_profile.C

@@ -1,107 +0,0 @@
-#include "interp_lb_profile.h"
-#include <cstdio>
-#include <cstring>
-#include <algorithm>
-
-namespace InterpLBProfile {
-
-bool write_profile(const char *filepath, int nprocs,
-                   const double *rank_times,
-                   const int *heavy_ranks, int num_heavy,
-                   double threshold_ratio)
-{
-    FILE *fp = fopen(filepath, "wb");
-    if (!fp) return false;
-
-    ProfileHeader hdr;
-    hdr.magic = MAGIC;
-    hdr.version = VERSION;
-    hdr.nprocs = nprocs;
-    hdr.num_heavy = num_heavy;
-    hdr.threshold_ratio = threshold_ratio;
-
-    fwrite(&hdr, sizeof(hdr), 1, fp);
-    fwrite(rank_times, sizeof(double), nprocs, fp);
-    fwrite(heavy_ranks, sizeof(int), num_heavy, fp);
-    fclose(fp);
-    return true;
-}
-
-bool read_profile(const char *filepath, int current_nprocs,
-                  int *heavy_ranks, int &num_heavy,
-                  double *rank_times, MPI_Comm comm)
-{
-    int myrank;
-    MPI_Comm_rank(comm, &myrank);
-
-    int valid = 0;
-    ProfileHeader hdr;
-    memset(&hdr, 0, sizeof(hdr));
-
-    if (myrank == 0) {
-        FILE *fp = fopen(filepath, "rb");
-        if (fp) {
-            if (fread(&hdr, sizeof(hdr), 1, fp) == 1 &&
-                hdr.magic == MAGIC && hdr.version == VERSION &&
-                hdr.nprocs == current_nprocs)
-            {
-                if (fread(rank_times, sizeof(double), current_nprocs, fp)
-                    == (size_t)current_nprocs &&
-                    fread(heavy_ranks, sizeof(int), hdr.num_heavy, fp)
-                    == (size_t)hdr.num_heavy)
-                {
-                    num_heavy = hdr.num_heavy;
-                    valid = 1;
-                }
-            } else if (fp) {
-                printf("[InterpLB] Profile rejected: magic=0x%X version=%u "
-                       "nprocs=%d (current=%d)\n",
-                       hdr.magic, hdr.version, hdr.nprocs, current_nprocs);
-            }
-            fclose(fp);
-        }
-    }
-
-    MPI_Bcast(&valid, 1, MPI_INT, 0, comm);
-    if (!valid) return false;
-
-    MPI_Bcast(&num_heavy, 1, MPI_INT, 0, comm);
-    MPI_Bcast(heavy_ranks, num_heavy, MPI_INT, 0, comm);
-    MPI_Bcast(rank_times, current_nprocs, MPI_DOUBLE, 0, comm);
-    return true;
-}
-
-int identify_heavy_ranks(const double *rank_times, int nprocs,
-                         double threshold_ratio,
-                         int *heavy_ranks, int max_heavy)
-{
-    double sum = 0;
-    for (int i = 0; i < nprocs; i++) sum += rank_times[i];
-    double mean = sum / nprocs;
-    double threshold = threshold_ratio * mean;
-
-    // Collect candidates
-    struct RankTime { int rank; double time; };
-    RankTime *candidates = new RankTime[nprocs];
-    int ncand = 0;
-
-    for (int i = 0; i < nprocs; i++) {
-        if (rank_times[i] > threshold)
-            candidates[ncand++] = {i, rank_times[i]};
-    }
-
-    // Sort descending by time
-    std::sort(candidates, candidates + ncand,
-              [](const RankTime &a, const RankTime &b) {
-                  return a.time > b.time;
-              });
-
-    int count = (ncand < max_heavy) ? ncand : max_heavy;
-    for (int i = 0; i < count; i++)
-        heavy_ranks[i] = candidates[i].rank;
-
-    delete[] candidates;
-    return count;
-}
-
-} // namespace InterpLBProfile

AMSS_NCKU_source/interp_lb_profile.h

@@ -1,38 +0,0 @@
-#ifndef INTERP_LB_PROFILE_H
-#define INTERP_LB_PROFILE_H
-
-#include <mpi.h>
-
-namespace InterpLBProfile {
-
-static const unsigned int MAGIC   = 0x494C4250; // "ILBP"
-static const unsigned int VERSION = 1;
-
-struct ProfileHeader {
-    unsigned int magic;
-    unsigned int version;
-    int nprocs;
-    int num_heavy;
-    double threshold_ratio;
-};
-
-// Write profile file (rank 0 only)
-bool write_profile(const char *filepath, int nprocs,
-                   const double *rank_times,
-                   const int *heavy_ranks, int num_heavy,
-                   double threshold_ratio);
-
-// Read profile file (rank 0 reads, then broadcasts to all)
-// Returns true if file found and valid for current nprocs
-bool read_profile(const char *filepath, int current_nprocs,
-                  int *heavy_ranks, int &num_heavy,
-                  double *rank_times, MPI_Comm comm);
-
-// Identify heavy ranks: those with time > threshold_ratio * mean
-int identify_heavy_ranks(const double *rank_times, int nprocs,
-                         double threshold_ratio,
-                         int *heavy_ranks, int max_heavy);
-
-} // namespace InterpLBProfile
-
-#endif /* INTERP_LB_PROFILE_H */

AMSS_NCKU_source/interp_lb_profile_data.h

@@ -1,29 +0,0 @@
-/* 本头文件由自订profile框架自动生成并非人工硬编码针对Case优化 */
-/* 更新：负载均衡问题已经通过优化插值函数解决，此profile静态均衡方案已弃用，本头文件现在未参与编译 */
-/* Auto-generated from interp_lb_profile.bin — do not edit */
-#ifndef INTERP_LB_PROFILE_DATA_H
-#define INTERP_LB_PROFILE_DATA_H
-
-#define INTERP_LB_NPROCS 64
-#define INTERP_LB_NUM_HEAVY 4
-
-static const int interp_lb_heavy_blocks[4] = {27, 35, 28, 36};
-
-/* Split table: {block_id, r_left, r_right} */
-static const int interp_lb_splits[4][3] = {
-    {27, 26, 27},
-    {35, 34, 35},
-    {28, 28, 29},
-    {36, 36, 37},
-};
-
-/* Rank remap for displaced neighbor blocks */
-static const int interp_lb_num_remaps = 4;
-static const int interp_lb_remaps[][2] = {
-    {26, 25},
-    {29, 30},
-    {34, 33},
-    {37, 38},
-};
-
-#endif /* INTERP_LB_PROFILE_DATA_H */

AMSS_NCKU_source/kodiss_c.C

@@ -1,117 +0,0 @@
-#include "tool.h"
-
-/*
- * C 版 kodis
- *
- * Fortran signature:
- * subroutine kodis(ex,X,Y,Z,f,f_rhs,SoA,Symmetry,eps)
- *
- * 约定：
- *   X: ex1, Y: ex2, Z: ex3
- *   f, f_rhs: ex1*ex2*ex3 按 idx_ex 布局
- *   SoA[3]
- *   eps: double
- */
-void kodis(const int ex[3],
-           const double *X, const double *Y, const double *Z,
-           const double *f, double *f_rhs,
-           const double SoA[3],
-           int Symmetry, double eps)
-{
-    const double ONE = 1.0, SIX = 6.0, FIT = 15.0, TWT = 20.0;
-    const double cof = 64.0;             // 2^6
-    const int NO_SYMM = 0, OCTANT = 2;
-
-    const int ex1 = ex[0], ex2 = ex[1], ex3 = ex[2];
-
-    // Fortran: dX = X(2)-X(1) -> C: X[1]-X[0]
-    const double dX = X[1] - X[0];
-    const double dY = Y[1] - Y[0];
-    const double dZ = Z[1] - Z[0];
-    (void)ONE; // ONE 在原 Fortran 里只是参数，这里不一定用得上
-
-    // Fortran: imax=ex(1) 等是 1-based 上界
-    const int imaxF = ex1;
-    const int jmaxF = ex2;
-    const int kmaxF = ex3;
-
-    // Fortran: imin=jmin=kmin=1，某些对称情况变 -2
-    int iminF = 1, jminF = 1, kminF = 1;
-
-    if (Symmetry > NO_SYMM && fabs(Z[0]) < dZ) kminF = -2;
-    if (Symmetry == OCTANT && fabs(X[0]) < dX) iminF = -2;
-    if (Symmetry == OCTANT && fabs(Y[0]) < dY) jminF = -2;
-
-    // 分配 fh：大小 (ex1+3)*(ex2+3)*(ex3+3)，对应 ord=3
-    const size_t nx = (size_t)ex1 + 3;
-    const size_t ny = (size_t)ex2 + 3;
-    const size_t nz = (size_t)ex3 + 3;
-    const size_t fh_size = nx * ny * nz;
-
-    double *fh = (double*)malloc(fh_size * sizeof(double));
-    if (!fh) return;
-
-    // Fortran: call symmetry_bd(3,ex,f,fh,SoA)
-    symmetry_bd(3, ex, f, fh, SoA);
-
-    /*
-     * Fortran loops:
-     * do k=1,ex3
-     * do j=1,ex2
-     * do i=1,ex1
-     *
-     * C: k0=0..ex3-1, j0=0..ex2-1, i0=0..ex1-1
-     * 并定义 Fortran index: iF=i0+1, ...
-     */
-    // 收紧循环范围：只遍历满足 iF±3/jF±3/kF±3 条件的内部点
-    // iF-3 >= iminF => iF >= iminF+3 => i0 >= iminF+2 (因为 iF=i0+1)
-    // iF+3 <= imaxF => iF <= imaxF-3 => i0 <= imaxF-4
-    const int i0_lo = (iminF + 2 > 0) ? iminF + 2 : 0;
-    const int j0_lo = (jminF + 2 > 0) ? jminF + 2 : 0;
-    const int k0_lo = (kminF + 2 > 0) ? kminF + 2 : 0;
-    const int i0_hi = imaxF - 4;  // inclusive
-    const int j0_hi = jmaxF - 4;
-    const int k0_hi = kmaxF - 4;
-
-    if (i0_lo > i0_hi || j0_lo > j0_hi || k0_lo > k0_hi) {
-        free(fh);
-        return;
-    }
-
-    for (int k0 = k0_lo; k0 <= k0_hi; ++k0) {
-        const int kF = k0 + 1;
-        for (int j0 = j0_lo; j0 <= j0_hi; ++j0) {
-            const int jF = j0 + 1;
-            for (int i0 = i0_lo; i0 <= i0_hi; ++i0) {
-                const int iF = i0 + 1;
-
-                    const size_t p = idx_ex(i0, j0, k0, ex);
-
-                    // 三个方向各一份同型的 7 点组合（实际上是对称的 6th-order dissipation/filter 核）
-                    const double Dx_term =
-                        ( (fh[idx_fh_F(iF - 3, jF, kF, ex)] + fh[idx_fh_F(iF + 3, jF, kF, ex)]) -
-                          SIX * (fh[idx_fh_F(iF - 2, jF, kF, ex)] + fh[idx_fh_F(iF + 2, jF, kF, ex)]) +
-                          FIT * (fh[idx_fh_F(iF - 1, jF, kF, ex)] + fh[idx_fh_F(iF + 1, jF, kF, ex)]) -
-                          TWT *  fh[idx_fh_F(iF    , jF, kF, ex)] ) / dX;
-
-                    const double Dy_term =
-                        ( (fh[idx_fh_F(iF, jF - 3, kF, ex)] + fh[idx_fh_F(iF, jF + 3, kF, ex)]) -
-                          SIX * (fh[idx_fh_F(iF, jF - 2, kF, ex)] + fh[idx_fh_F(iF, jF + 2, kF, ex)]) +
-                          FIT * (fh[idx_fh_F(iF, jF - 1, kF, ex)] + fh[idx_fh_F(iF, jF + 1, kF, ex)]) -
-                          TWT *  fh[idx_fh_F(iF, jF    , kF, ex)] ) / dY;
-
-                    const double Dz_term =
-                        ( (fh[idx_fh_F(iF, jF, kF - 3, ex)] + fh[idx_fh_F(iF, jF, kF + 3, ex)]) -
-                          SIX * (fh[idx_fh_F(iF, jF, kF - 2, ex)] + fh[idx_fh_F(iF, jF, kF + 2, ex)]) +
-                          FIT * (fh[idx_fh_F(iF, jF, kF - 1, ex)] + fh[idx_fh_F(iF, jF, kF + 1, ex)]) -
-                          TWT *  fh[idx_fh_F(iF, jF, kF    , ex)] ) / dZ;
-
-                    // Fortran:
-                    // f_rhs(i,j,k) = f_rhs(i,j,k) + eps/cof*(Dx_term + Dy_term + Dz_term)
-                    f_rhs[p] += (eps / cof) * (Dx_term + Dy_term + Dz_term);
-            }
-        }
-    }
-
-    free(fh);
-}

AMSS_NCKU_source/lopsided_c.C

@@ -1,255 +0,0 @@
-#include "tool.h"
-/*
- * 你需要提供 symmetry_bd 的 C 版本（或 Fortran 绑到 C 的接口）。
- * Fortran: call symmetry_bd(3,ex,f,fh,SoA)
- *
- * 约定：
- *   nghost = 3
- *   ex[3]  = {ex1,ex2,ex3}
- *   f      = 原始网格 (ex1*ex2*ex3)
- *   fh     = 扩展网格 ((ex1+3)*(ex2+3)*(ex3+3))，对应 Fortran 的 (-2:ex1, ...)
- *   SoA[3] = 输入参数
- */
-void lopsided(const int ex[3],
-              const double *X, const double *Y, const double *Z,
-              const double *f, double *f_rhs,
-              const double *Sfx, const double *Sfy, const double *Sfz,
-              int Symmetry, const double SoA[3])
-{
-    const double ZEO = 0.0, ONE = 1.0, F3 = 3.0;
-    const double TWO = 2.0, F6 = 6.0, F18 = 18.0;
-    const double F12 = 12.0, F10 = 10.0, EIT = 8.0;
-
-    const int NO_SYMM = 0, EQ_SYMM = 1, OCTANT = 2;
-    (void)OCTANT; // 这里和 Fortran 一样只是定义了不用也没关系
-
-    const int ex1 = ex[0], ex2 = ex[1], ex3 = ex[2];
-
-    // 对应 Fortran: dX = X(2)-X(1)  （Fortran 1-based）
-    // C: X[1]-X[0]
-    const double dX = X[1] - X[0];
-    const double dY = Y[1] - Y[0];
-    const double dZ = Z[1] - Z[0];
-
-    const double d12dx = ONE / F12 / dX;
-    const double d12dy = ONE / F12 / dY;
-    const double d12dz = ONE / F12 / dZ;
-
-    // Fortran 里算了 d2dx/d2dy/d2dz 但本 subroutine 里没用到（保持一致也算出来）
-    const double d2dx  = ONE / TWO / dX;
-    const double d2dy  = ONE / TWO / dY;
-    const double d2dz  = ONE / TWO / dZ;
-    (void)d2dx; (void)d2dy; (void)d2dz;
-
-    // Fortran:
-    // imax = ex(1); jmax = ex(2); kmax = ex(3)
-    const int imaxF = ex1;
-    const int jmaxF = ex2;
-    const int kmaxF = ex3;
-
-    // Fortran:
-    // imin=jmin=kmin=1; 若满足对称条件则设为 -2
-    int iminF = 1, jminF = 1, kminF = 1;
-    if (Symmetry > NO_SYMM && fabs(Z[0]) < dZ) kminF = -2;
-    if (Symmetry > EQ_SYMM && fabs(X[0]) < dX) iminF = -2;
-    if (Symmetry > EQ_SYMM && fabs(Y[0]) < dY) jminF = -2;
-
-    // 分配 fh：大小 (ex1+3)*(ex2+3)*(ex3+3)
-    const size_t nx = (size_t)ex1 + 3;
-    const size_t ny = (size_t)ex2 + 3;
-    const size_t nz = (size_t)ex3 + 3;
-    const size_t fh_size = nx * ny * nz;
-
-    double *fh = (double*)malloc(fh_size * sizeof(double));
-    if (!fh) return; // 内存不足：直接返回（你也可以改成 abort/报错）
-
-    // Fortran: call symmetry_bd(3,ex,f,fh,SoA)
-    symmetry_bd(3, ex, f, fh, SoA);
-
-    /*
-     * Fortran 主循环：
-     * do k=1,ex(3)-1
-     * do j=1,ex(2)-1
-     * do i=1,ex(1)-1
-     *
-     * 转成 C 0-based：
-     * k0 = 0..ex3-2, j0 = 0..ex2-2, i0 = 0..ex1-2
-     *
-     * 并且 Fortran 里的 i/j/k 在 fh 访问时，仍然是 Fortran 索引值：
-     * iF=i0+1, jF=j0+1, kF=k0+1
-     */
-    for (int k0 = 0; k0 <= ex3 - 2; ++k0) {
-        const int kF = k0 + 1;
-        for (int j0 = 0; j0 <= ex2 - 2; ++j0) {
-            const int jF = j0 + 1;
-            for (int i0 = 0; i0 <= ex1 - 2; ++i0) {
-                const int iF = i0 + 1;
-
-                const size_t p = idx_ex(i0, j0, k0, ex);
-
-                // ---------------- x direction ----------------
-                const double sfx = Sfx[p];
-                if (sfx > ZEO) {
-                    // Fortran: if(i+3 <= imax)
-                    // iF+3 <= ex1  <=> i0+4 <= ex1 <=> i0 <= ex1-4
-                    if (i0 <= ex1 - 4) {
-                        f_rhs[p] += sfx * d12dx *
-                            (-F3  * fh[idx_fh_F(iF - 1, jF, kF, ex)]
-                             -F10 * fh[idx_fh_F(iF    , jF, kF, ex)]
-                             +F18 * fh[idx_fh_F(iF + 1, jF, kF, ex)]
-                             -F6  * fh[idx_fh_F(iF + 2, jF, kF, ex)]
-                             +      fh[idx_fh_F(iF + 3, jF, kF, ex)]);
-                    }
-                    // elseif(i+2 <= imax)  <=> i0 <= ex1-3
-                    else if (i0 <= ex1 - 3) {
-                        f_rhs[p] += sfx * d12dx *
-                            ( fh[idx_fh_F(iF - 2, jF, kF, ex)]
-                             -EIT * fh[idx_fh_F(iF - 1, jF, kF, ex)]
-                             +EIT * fh[idx_fh_F(iF + 1, jF, kF, ex)]
-                             -      fh[idx_fh_F(iF + 2, jF, kF, ex)]);
-                    }
-                    // elseif(i+1 <= imax)  <=> i0 <= ex1-2（循环里总成立）
-                    else if (i0 <= ex1 - 2) {
-                        f_rhs[p] -= sfx * d12dx *
-                            (-F3  * fh[idx_fh_F(iF + 1, jF, kF, ex)]
-                             -F10 * fh[idx_fh_F(iF    , jF, kF, ex)]
-                             +F18 * fh[idx_fh_F(iF - 1, jF, kF, ex)]
-                             -F6  * fh[idx_fh_F(iF - 2, jF, kF, ex)]
-                             +      fh[idx_fh_F(iF - 3, jF, kF, ex)]);
-                    }
-                } else if (sfx < ZEO) {
-                    // Fortran: if(i-3 >= imin)
-                    // (iF-3) >= iminF  <=> (i0-2) >= iminF
-                    if ((i0 - 2) >= iminF) {
-                        f_rhs[p] -= sfx * d12dx *
-                            (-F3  * fh[idx_fh_F(iF + 1, jF, kF, ex)]
-                             -F10 * fh[idx_fh_F(iF    , jF, kF, ex)]
-                             +F18 * fh[idx_fh_F(iF - 1, jF, kF, ex)]
-                             -F6  * fh[idx_fh_F(iF - 2, jF, kF, ex)]
-                             +      fh[idx_fh_F(iF - 3, jF, kF, ex)]);
-                    }
-                    // elseif(i-2 >= imin) <=> (i0-1) >= iminF
-                    else if ((i0 - 1) >= iminF) {
-                        f_rhs[p] += sfx * d12dx *
-                            ( fh[idx_fh_F(iF - 2, jF, kF, ex)]
-                             -EIT * fh[idx_fh_F(iF - 1, jF, kF, ex)]
-                             +EIT * fh[idx_fh_F(iF + 1, jF, kF, ex)]
-                             -      fh[idx_fh_F(iF + 2, jF, kF, ex)]);
-                    }
-                    // elseif(i-1 >= imin) <=> i0 >= iminF
-                    else if (i0 >= iminF) {
-                        f_rhs[p] += sfx * d12dx *
-                            (-F3  * fh[idx_fh_F(iF - 1, jF, kF, ex)]
-                             -F10 * fh[idx_fh_F(iF    , jF, kF, ex)]
-                             +F18 * fh[idx_fh_F(iF + 1, jF, kF, ex)]
-                             -F6  * fh[idx_fh_F(iF + 2, jF, kF, ex)]
-                             +      fh[idx_fh_F(iF + 3, jF, kF, ex)]);
-                    }
-                }
-
-                // ---------------- y direction ----------------
-                const double sfy = Sfy[p];
-                if (sfy > ZEO) {
-                    // jF+3 <= ex2 <=> j0+4 <= ex2 <=> j0 <= ex2-4
-                    if (j0 <= ex2 - 4) {
-                        f_rhs[p] += sfy * d12dy *
-                            (-F3  * fh[idx_fh_F(iF, jF - 1, kF, ex)]
-                             -F10 * fh[idx_fh_F(iF, jF    , kF, ex)]
-                             +F18 * fh[idx_fh_F(iF, jF + 1, kF, ex)]
-                             -F6  * fh[idx_fh_F(iF, jF + 2, kF, ex)]
-                             +      fh[idx_fh_F(iF, jF + 3, kF, ex)]);
-                    } else if (j0 <= ex2 - 3) {
-                        f_rhs[p] += sfy * d12dy *
-                            ( fh[idx_fh_F(iF, jF - 2, kF, ex)]
-                             -EIT * fh[idx_fh_F(iF, jF - 1, kF, ex)]
-                             +EIT * fh[idx_fh_F(iF, jF + 1, kF, ex)]
-                             -      fh[idx_fh_F(iF, jF + 2, kF, ex)]);
-                    } else if (j0 <= ex2 - 2) {
-                        f_rhs[p] -= sfy * d12dy *
-                            (-F3  * fh[idx_fh_F(iF, jF + 1, kF, ex)]
-                             -F10 * fh[idx_fh_F(iF, jF    , kF, ex)]
-                             +F18 * fh[idx_fh_F(iF, jF - 1, kF, ex)]
-                             -F6  * fh[idx_fh_F(iF, jF - 2, kF, ex)]
-                             +      fh[idx_fh_F(iF, jF - 3, kF, ex)]);
-                    }
-                } else if (sfy < ZEO) {
-                    if ((j0 - 2) >= jminF) {
-                        f_rhs[p] -= sfy * d12dy *
-                            (-F3  * fh[idx_fh_F(iF, jF + 1, kF, ex)]
-                             -F10 * fh[idx_fh_F(iF, jF    , kF, ex)]
-                             +F18 * fh[idx_fh_F(iF, jF - 1, kF, ex)]
-                             -F6  * fh[idx_fh_F(iF, jF - 2, kF, ex)]
-                             +      fh[idx_fh_F(iF, jF - 3, kF, ex)]);
-                    } else if ((j0 - 1) >= jminF) {
-                        f_rhs[p] += sfy * d12dy *
-                            ( fh[idx_fh_F(iF, jF - 2, kF, ex)]
-                             -EIT * fh[idx_fh_F(iF, jF - 1, kF, ex)]
-                             +EIT * fh[idx_fh_F(iF, jF + 1, kF, ex)]
-                             -      fh[idx_fh_F(iF, jF + 2, kF, ex)]);
-                    } else if (j0 >= jminF) {
-                        f_rhs[p] += sfy * d12dy *
-                            (-F3  * fh[idx_fh_F(iF, jF - 1, kF, ex)]
-                             -F10 * fh[idx_fh_F(iF, jF    , kF, ex)]
-                             +F18 * fh[idx_fh_F(iF, jF + 1, kF, ex)]
-                             -F6  * fh[idx_fh_F(iF, jF + 2, kF, ex)]
-                             +      fh[idx_fh_F(iF, jF + 3, kF, ex)]);
-                    }
-                }
-
-                // ---------------- z direction ----------------
-                const double sfz = Sfz[p];
-                if (sfz > ZEO) {
-                    if (k0 <= ex3 - 4) {
-                        f_rhs[p] += sfz * d12dz *
-                            (-F3  * fh[idx_fh_F(iF, jF, kF - 1, ex)]
-                             -F10 * fh[idx_fh_F(iF, jF, kF    , ex)]
-                             +F18 * fh[idx_fh_F(iF, jF, kF + 1, ex)]
-                             -F6  * fh[idx_fh_F(iF, jF, kF + 2, ex)]
-                             +      fh[idx_fh_F(iF, jF, kF + 3, ex)]);
-                    } else if (k0 <= ex3 - 3) {
-                        f_rhs[p] += sfz * d12dz *
-                            ( fh[idx_fh_F(iF, jF, kF - 2, ex)]
-                             -EIT * fh[idx_fh_F(iF, jF, kF - 1, ex)]
-                             +EIT * fh[idx_fh_F(iF, jF, kF + 1, ex)]
-                             -      fh[idx_fh_F(iF, jF, kF + 2, ex)]);
-                    } else if (k0 <= ex3 - 2) {
-                        f_rhs[p] -= sfz * d12dz *
-                            (-F3  * fh[idx_fh_F(iF, jF, kF + 1, ex)]
-                             -F10 * fh[idx_fh_F(iF, jF, kF    , ex)]
-                             +F18 * fh[idx_fh_F(iF, jF, kF - 1, ex)]
-                             -F6  * fh[idx_fh_F(iF, jF, kF - 2, ex)]
-                             +      fh[idx_fh_F(iF, jF, kF - 3, ex)]);
-                    }
-                } else if (sfz < ZEO) {
-                    if ((k0 - 2) >= kminF) {
-                        f_rhs[p] -= sfz * d12dz *
-                            (-F3  * fh[idx_fh_F(iF, jF, kF + 1, ex)]
-                             -F10 * fh[idx_fh_F(iF, jF, kF    , ex)]
-                             +F18 * fh[idx_fh_F(iF, jF, kF - 1, ex)]
-                             -F6  * fh[idx_fh_F(iF, jF, kF - 2, ex)]
-                             +      fh[idx_fh_F(iF, jF, kF - 3, ex)]);
-                    } else if ((k0 - 1) >= kminF) {
-                        f_rhs[p] += sfz * d12dz *
-                            ( fh[idx_fh_F(iF, jF, kF - 2, ex)]
-                             -EIT * fh[idx_fh_F(iF, jF, kF - 1, ex)]
-                             +EIT * fh[idx_fh_F(iF, jF, kF + 1, ex)]
-                             -      fh[idx_fh_F(iF, jF, kF + 2, ex)]);
-                    } else if (k0 >= kminF) {
-                        f_rhs[p] += sfz * d12dz *
-                            (-F3  * fh[idx_fh_F(iF, jF, kF - 1, ex)]
-                             -F10 * fh[idx_fh_F(iF, jF, kF    , ex)]
-                             +F18 * fh[idx_fh_F(iF, jF, kF + 1, ex)]
-                             -F6  * fh[idx_fh_F(iF, jF, kF + 2, ex)]
-                             +      fh[idx_fh_F(iF, jF, kF + 3, ex)]);
-                    }
-                }
-            }
-        }
-    }
-    free(fh);
-}
-
-
-
-
-

AMSS_NCKU_source/lopsided_kodis_c.C

@@ -1,248 +0,0 @@
-#include "tool.h"
-
-/*
- * Combined advection (lopsided) + KO dissipation (kodis).
- * Uses one shared symmetry_bd buffer per call.
- */
-void lopsided_kodis(const int ex[3],
-                    const double *X, const double *Y, const double *Z,
-                    const double *f, double *f_rhs,
-                    const double *Sfx, const double *Sfy, const double *Sfz,
-                    int Symmetry, const double SoA[3], double eps)
-{
-    const double ZEO = 0.0, ONE = 1.0, F3 = 3.0;
-    const double F6 = 6.0, F18 = 18.0;
-    const double F12 = 12.0, F10 = 10.0, EIT = 8.0;
-    const double SIX = 6.0, FIT = 15.0, TWT = 20.0;
-    const double cof = 64.0; // 2^6
-
-    const int NO_SYMM = 0, EQ_SYMM = 1;
-
-    const int ex1 = ex[0], ex2 = ex[1], ex3 = ex[2];
-
-    const double dX = X[1] - X[0];
-    const double dY = Y[1] - Y[0];
-    const double dZ = Z[1] - Z[0];
-
-    const double d12dx = ONE / F12 / dX;
-    const double d12dy = ONE / F12 / dY;
-    const double d12dz = ONE / F12 / dZ;
-
-    const int imaxF = ex1;
-    const int jmaxF = ex2;
-    const int kmaxF = ex3;
-
-    int iminF = 1, jminF = 1, kminF = 1;
-    if (Symmetry > NO_SYMM && fabs(Z[0]) < dZ) kminF = -2;
-    if (Symmetry > EQ_SYMM && fabs(X[0]) < dX) iminF = -2;
-    if (Symmetry > EQ_SYMM && fabs(Y[0]) < dY) jminF = -2;
-
-    // fh for Fortran-style domain (-2:ex1,-2:ex2,-2:ex3)
-    const size_t nx = (size_t)ex1 + 3;
-    const size_t ny = (size_t)ex2 + 3;
-    const size_t nz = (size_t)ex3 + 3;
-    const size_t fh_size = nx * ny * nz;
-
-    double *fh = (double*)malloc(fh_size * sizeof(double));
-    if (!fh) return;
-
-    symmetry_bd(3, ex, f, fh, SoA);
-
-    // Advection (same stencil logic as lopsided_c.C)
-    for (int k0 = 0; k0 <= ex3 - 2; ++k0) {
-        const int kF = k0 + 1;
-        for (int j0 = 0; j0 <= ex2 - 2; ++j0) {
-            const int jF = j0 + 1;
-            for (int i0 = 0; i0 <= ex1 - 2; ++i0) {
-                const int iF = i0 + 1;
-                const size_t p = idx_ex(i0, j0, k0, ex);
-
-                const double sfx = Sfx[p];
-                if (sfx > ZEO) {
-                    if (i0 <= ex1 - 4) {
-                        f_rhs[p] += sfx * d12dx *
-                            (-F3  * fh[idx_fh_F(iF - 1, jF, kF, ex)]
-                             -F10 * fh[idx_fh_F(iF    , jF, kF, ex)]
-                             +F18 * fh[idx_fh_F(iF + 1, jF, kF, ex)]
-                             -F6  * fh[idx_fh_F(iF + 2, jF, kF, ex)]
-                             +      fh[idx_fh_F(iF + 3, jF, kF, ex)]);
-                    } else if (i0 <= ex1 - 3) {
-                        f_rhs[p] += sfx * d12dx *
-                            ( fh[idx_fh_F(iF - 2, jF, kF, ex)]
-                             -EIT * fh[idx_fh_F(iF - 1, jF, kF, ex)]
-                             +EIT * fh[idx_fh_F(iF + 1, jF, kF, ex)]
-                             -      fh[idx_fh_F(iF + 2, jF, kF, ex)]);
-                    } else if (i0 <= ex1 - 2) {
-                        f_rhs[p] -= sfx * d12dx *
-                            (-F3  * fh[idx_fh_F(iF + 1, jF, kF, ex)]
-                             -F10 * fh[idx_fh_F(iF    , jF, kF, ex)]
-                             +F18 * fh[idx_fh_F(iF - 1, jF, kF, ex)]
-                             -F6  * fh[idx_fh_F(iF - 2, jF, kF, ex)]
-                             +      fh[idx_fh_F(iF - 3, jF, kF, ex)]);
-                    }
-                } else if (sfx < ZEO) {
-                    if ((i0 - 2) >= iminF) {
-                        f_rhs[p] -= sfx * d12dx *
-                            (-F3  * fh[idx_fh_F(iF + 1, jF, kF, ex)]
-                             -F10 * fh[idx_fh_F(iF    , jF, kF, ex)]
-                             +F18 * fh[idx_fh_F(iF - 1, jF, kF, ex)]
-                             -F6  * fh[idx_fh_F(iF - 2, jF, kF, ex)]
-                             +      fh[idx_fh_F(iF - 3, jF, kF, ex)]);
-                    } else if ((i0 - 1) >= iminF) {
-                        f_rhs[p] += sfx * d12dx *
-                            ( fh[idx_fh_F(iF - 2, jF, kF, ex)]
-                             -EIT * fh[idx_fh_F(iF - 1, jF, kF, ex)]
-                             +EIT * fh[idx_fh_F(iF + 1, jF, kF, ex)]
-                             -      fh[idx_fh_F(iF + 2, jF, kF, ex)]);
-                    } else if (i0 >= iminF) {
-                        f_rhs[p] += sfx * d12dx *
-                            (-F3  * fh[idx_fh_F(iF - 1, jF, kF, ex)]
-                             -F10 * fh[idx_fh_F(iF    , jF, kF, ex)]
-                             +F18 * fh[idx_fh_F(iF + 1, jF, kF, ex)]
-                             -F6  * fh[idx_fh_F(iF + 2, jF, kF, ex)]
-                             +      fh[idx_fh_F(iF + 3, jF, kF, ex)]);
-                    }
-                }
-
-                const double sfy = Sfy[p];
-                if (sfy > ZEO) {
-                    if (j0 <= ex2 - 4) {
-                        f_rhs[p] += sfy * d12dy *
-                            (-F3  * fh[idx_fh_F(iF, jF - 1, kF, ex)]
-                             -F10 * fh[idx_fh_F(iF, jF    , kF, ex)]
-                             +F18 * fh[idx_fh_F(iF, jF + 1, kF, ex)]
-                             -F6  * fh[idx_fh_F(iF, jF + 2, kF, ex)]
-                             +      fh[idx_fh_F(iF, jF + 3, kF, ex)]);
-                    } else if (j0 <= ex2 - 3) {
-                        f_rhs[p] += sfy * d12dy *
-                            ( fh[idx_fh_F(iF, jF - 2, kF, ex)]
-                             -EIT * fh[idx_fh_F(iF, jF - 1, kF, ex)]
-                             +EIT * fh[idx_fh_F(iF, jF + 1, kF, ex)]
-                             -      fh[idx_fh_F(iF, jF + 2, kF, ex)]);
-                    } else if (j0 <= ex2 - 2) {
-                        f_rhs[p] -= sfy * d12dy *
-                            (-F3  * fh[idx_fh_F(iF, jF + 1, kF, ex)]
-                             -F10 * fh[idx_fh_F(iF, jF    , kF, ex)]
-                             +F18 * fh[idx_fh_F(iF, jF - 1, kF, ex)]
-                             -F6  * fh[idx_fh_F(iF, jF - 2, kF, ex)]
-                             +      fh[idx_fh_F(iF, jF - 3, kF, ex)]);
-                    }
-                } else if (sfy < ZEO) {
-                    if ((j0 - 2) >= jminF) {
-                        f_rhs[p] -= sfy * d12dy *
-                            (-F3  * fh[idx_fh_F(iF, jF + 1, kF, ex)]
-                             -F10 * fh[idx_fh_F(iF, jF    , kF, ex)]
-                             +F18 * fh[idx_fh_F(iF, jF - 1, kF, ex)]
-                             -F6  * fh[idx_fh_F(iF, jF - 2, kF, ex)]
-                             +      fh[idx_fh_F(iF, jF - 3, kF, ex)]);
-                    } else if ((j0 - 1) >= jminF) {
-                        f_rhs[p] += sfy * d12dy *
-                            ( fh[idx_fh_F(iF, jF - 2, kF, ex)]
-                             -EIT * fh[idx_fh_F(iF, jF - 1, kF, ex)]
-                             +EIT * fh[idx_fh_F(iF, jF + 1, kF, ex)]
-                             -      fh[idx_fh_F(iF, jF + 2, kF, ex)]);
-                    } else if (j0 >= jminF) {
-                        f_rhs[p] += sfy * d12dy *
-                            (-F3  * fh[idx_fh_F(iF, jF - 1, kF, ex)]
-                             -F10 * fh[idx_fh_F(iF, jF    , kF, ex)]
-                             +F18 * fh[idx_fh_F(iF, jF + 1, kF, ex)]
-                             -F6  * fh[idx_fh_F(iF, jF + 2, kF, ex)]
-                             +      fh[idx_fh_F(iF, jF + 3, kF, ex)]);
-                    }
-                }
-
-                const double sfz = Sfz[p];
-                if (sfz > ZEO) {
-                    if (k0 <= ex3 - 4) {
-                        f_rhs[p] += sfz * d12dz *
-                            (-F3  * fh[idx_fh_F(iF, jF, kF - 1, ex)]
-                             -F10 * fh[idx_fh_F(iF, jF, kF    , ex)]
-                             +F18 * fh[idx_fh_F(iF, jF, kF + 1, ex)]
-                             -F6  * fh[idx_fh_F(iF, jF, kF + 2, ex)]
-                             +      fh[idx_fh_F(iF, jF, kF + 3, ex)]);
-                    } else if (k0 <= ex3 - 3) {
-                        f_rhs[p] += sfz * d12dz *
-                            ( fh[idx_fh_F(iF, jF, kF - 2, ex)]
-                             -EIT * fh[idx_fh_F(iF, jF, kF - 1, ex)]
-                             +EIT * fh[idx_fh_F(iF, jF, kF + 1, ex)]
-                             -      fh[idx_fh_F(iF, jF, kF + 2, ex)]);
-                    } else if (k0 <= ex3 - 2) {
-                        f_rhs[p] -= sfz * d12dz *
-                            (-F3  * fh[idx_fh_F(iF, jF, kF + 1, ex)]
-                             -F10 * fh[idx_fh_F(iF, jF, kF    , ex)]
-                             +F18 * fh[idx_fh_F(iF, jF, kF - 1, ex)]
-                             -F6  * fh[idx_fh_F(iF, jF, kF - 2, ex)]
-                             +      fh[idx_fh_F(iF, jF, kF - 3, ex)]);
-                    }
-                } else if (sfz < ZEO) {
-                    if ((k0 - 2) >= kminF) {
-                        f_rhs[p] -= sfz * d12dz *
-                            (-F3  * fh[idx_fh_F(iF, jF, kF + 1, ex)]
-                             -F10 * fh[idx_fh_F(iF, jF, kF    , ex)]
-                             +F18 * fh[idx_fh_F(iF, jF, kF - 1, ex)]
-                             -F6  * fh[idx_fh_F(iF, jF, kF - 2, ex)]
-                             +      fh[idx_fh_F(iF, jF, kF - 3, ex)]);
-                    } else if ((k0 - 1) >= kminF) {
-                        f_rhs[p] += sfz * d12dz *
-                            ( fh[idx_fh_F(iF, jF, kF - 2, ex)]
-                             -EIT * fh[idx_fh_F(iF, jF, kF - 1, ex)]
-                             +EIT * fh[idx_fh_F(iF, jF, kF + 1, ex)]
-                             -      fh[idx_fh_F(iF, jF, kF + 2, ex)]);
-                    } else if (k0 >= kminF) {
-                        f_rhs[p] += sfz * d12dz *
-                            (-F3  * fh[idx_fh_F(iF, jF, kF - 1, ex)]
-                             -F10 * fh[idx_fh_F(iF, jF, kF    , ex)]
-                             +F18 * fh[idx_fh_F(iF, jF, kF + 1, ex)]
-                             -F6  * fh[idx_fh_F(iF, jF, kF + 2, ex)]
-                             +      fh[idx_fh_F(iF, jF, kF + 3, ex)]);
-                    }
-                }
-            }
-        }
-    }
-
-    // KO dissipation (same domain restriction as kodiss_c.C)
-    if (eps > ZEO) {
-        const int i0_lo = (iminF + 2 > 0) ? iminF + 2 : 0;
-        const int j0_lo = (jminF + 2 > 0) ? jminF + 2 : 0;
-        const int k0_lo = (kminF + 2 > 0) ? kminF + 2 : 0;
-        const int i0_hi = imaxF - 4; // inclusive
-        const int j0_hi = jmaxF - 4;
-        const int k0_hi = kmaxF - 4;
-
-        if (!(i0_lo > i0_hi || j0_lo > j0_hi || k0_lo > k0_hi)) {
-            for (int k0 = k0_lo; k0 <= k0_hi; ++k0) {
-                const int kF = k0 + 1;
-                for (int j0 = j0_lo; j0 <= j0_hi; ++j0) {
-                    const int jF = j0 + 1;
-                    for (int i0 = i0_lo; i0 <= i0_hi; ++i0) {
-                        const int iF = i0 + 1;
-                        const size_t p = idx_ex(i0, j0, k0, ex);
-
-                        const double Dx_term =
-                            ((fh[idx_fh_F(iF - 3, jF, kF, ex)] + fh[idx_fh_F(iF + 3, jF, kF, ex)]) -
-                             SIX * (fh[idx_fh_F(iF - 2, jF, kF, ex)] + fh[idx_fh_F(iF + 2, jF, kF, ex)]) +
-                             FIT * (fh[idx_fh_F(iF - 1, jF, kF, ex)] + fh[idx_fh_F(iF + 1, jF, kF, ex)]) -
-                             TWT *  fh[idx_fh_F(iF,     jF, kF, ex)]) / dX;
-
-                        const double Dy_term =
-                            ((fh[idx_fh_F(iF, jF - 3, kF, ex)] + fh[idx_fh_F(iF, jF + 3, kF, ex)]) -
-                             SIX * (fh[idx_fh_F(iF, jF - 2, kF, ex)] + fh[idx_fh_F(iF, jF + 2, kF, ex)]) +
-                             FIT * (fh[idx_fh_F(iF, jF - 1, kF, ex)] + fh[idx_fh_F(iF, jF + 1, kF, ex)]) -
-                             TWT *  fh[idx_fh_F(iF, jF,     kF, ex)]) / dY;
-
-                        const double Dz_term =
-                            ((fh[idx_fh_F(iF, jF, kF - 3, ex)] + fh[idx_fh_F(iF, jF, kF + 3, ex)]) -
-                             SIX * (fh[idx_fh_F(iF, jF, kF - 2, ex)] + fh[idx_fh_F(iF, jF, kF + 2, ex)]) +
-                             FIT * (fh[idx_fh_F(iF, jF, kF - 1, ex)] + fh[idx_fh_F(iF, jF, kF + 1, ex)]) -
-                             TWT *  fh[idx_fh_F(iF, jF, kF,     ex)]) / dZ;
-
-                        f_rhs[p] += (eps / cof) * (Dx_term + Dy_term + Dz_term);
-                    }
-                }
-            }
-        }
-    }
-
-    free(fh);
-}

AMSS_NCKU_source/macrodef.fh

@@ -1,77 +1,83 @@
 
-#define tetradtype 2
-
-#define Cell
-
-#define ghost_width 3
-
-
-
-#define GAUGE 0
-
-#define CPBC_ghost_width  (ghost_width)
-
-#define ABV 0
-
-#define EScalar_CC 2
 
 #if 0
-
+note here
-define tetradtype
+v:r; u: phi; w: theta
-    v:r; u: phi; w: theta
+tetradtype 0
-    tetradtype 0
+v^a = (x,y,z)
-    v^a = (x,y,z)
+orthonormal order: v,u,w
-    orthonormal order: v,u,w
+m = (phi - i theta)/sqrt(2) following Frans, Eq.(8) of  PRD 75, 124018(2007)
-    m = (phi - i theta)/sqrt(2) following Frans, Eq.(8) of  PRD 75, 124018(2007)
+tetradtype 1
-    tetradtype 1
+orthonormal order: w,u,v
-    orthonormal order: w,u,v
+m = (theta + i phi)/sqrt(2) following Sperhake, Eq.(3.2) of  PRD 85, 124062(2012)
-    m = (theta + i phi)/sqrt(2) following Sperhake, Eq.(3.2) of  PRD 85, 124062(2012)
+tetradtype 2
-    tetradtype 2
+v_a = (x,y,z)
-    v_a = (x,y,z)
+orthonormal order: v,u,w
-    orthonormal order: v,u,w
+m = (phi - i theta)/sqrt(2) following Frans, Eq.(8) of  PRD 75, 124018(2007)
-    m = (phi - i theta)/sqrt(2) following Frans, Eq.(8) of  PRD 75, 124018(2007)
-
-define Cell or Vertex
-    Cell center or Vertex center
-
-define ghost_width
-    2nd order: 2
-    4th order: 3
-    6th order: 4
-    8th order: 5
-
-define WithShell
-    use shell or not
-
-define CPBC
-    use constraint preserving boundary condition or not
-    only affect Z4c
-    CPBC only supports WithShell
-
-define GAUGE
-    0: B^i gauge
-    1: David puncture gauge
-    2: MB B^i gauge
-    3: RIT B^i gauge
-    4: MB beta gauge (beta gauge not means Eq.(3) of PRD 84, 124006)
-    5: RIT beta gauge (beta gauge not means Eq.(3) of PRD 84, 124006)
-    6: MGB1 B^i gauge
-    7: MGB2 B^i gauge
-
-define CPBC_ghost_width  (ghost_width)
-    buffer points for CPBC boundary
-
-define ABV
-    0: using BSSN variable for constraint violation and psi4 calculation
-    1: using ADM variable for constraint violation and psi4 calculation
-
-define EScalar_CC
-Type of Potential and Scalar Distribution in F(R) Scalar-Tensor Theory
-    1: Case C of 1112.3928, V=0
-    2: shell with   phi(r) = phi0 * a2^2/(1+a2^2), f(R) = R+a2*R^2 induced V
-    3: ground state of Schrodinger-Newton system,  f(R) = R+a2*R^2 induced V
-    4: a2 = +oo and phi(r) = phi0 * 0.5 * ( tanh((r+r0)/sigma) - tanh((r-r0)/sigma) )
-    5: shell with   phi(r) = phi0 * Exp(-(r-r0)**2/sigma), V = 0
-
 #endif
+#define tetradtype 2
+
+#if 0
+note here
+Cell center or Vertex center
+#endif
+#define Cell
+
+#if 0
+note here
+2nd order: 2
+4th order: 3
+6th order: 4
+8th order: 5
+#endif
+#define ghost_width 3
+
+#if 0
+note here
+use shell or not
+#endif
+#define WithShell
+
+#if 0
+note here
+use constraint preserving boundary condition or not
+only affect Z4c
+#endif
+#define CPBC
+
+#if 0
+note here
+Gauge condition type
+0: B^i gauge
+1: David's puncture gauge
+2: MB B^i gauge
+3: RIT B^i gauge
+4: MB beta gauge (beta gauge not means Eq.(3) of PRD 84, 124006)
+5: RIT beta gauge (beta gauge not means Eq.(3) of PRD 84, 124006)
+6: MGB1 B^i gauge
+7: MGB2 B^i gauge
+#endif
+#define GAUGE 2
+
+#if 0
+buffer points for CPBC boundary
+#endif
+#define CPBC_ghost_width  (ghost_width)
+
+#if 0
+using BSSN variable for constraint violation and psi4 calculation: 0
+using ADM variable for constraint violation and psi4 calculation: 1
+#endif
+#define ABV 0
+
+#if 0
+Type of Potential and Scalar Distribution in F(R) Scalar-Tensor Theory
+1: Case C of 1112.3928, V=0
+2: shell with a2^2*phi0/(1+a2^2), f(R) = R+a2*R^2 induced V
+3: ground state of Schrodinger-Newton system, f(R) = R+a2*R^2 induced V
+4: a2 = oo and phi(r) = phi0 * 0.5 * ( tanh((r+r0)/sigma) - tanh((r-r0)/sigma) )
+5: shell with phi(r) = phi0*Exp(-(r-r0)**2/sigma), V = 0
+#endif
+#define EScalar_CC 2
+
 

AMSS_NCKU_source/macrodef.h

@@ -6,127 +6,95 @@
 
 // application parameters
 
+/// ****
+// sommerfeld boundary type
+// 0: bam, 1: shibata
 #define SommerType 0
 
+/// ****
+// for Using Gauss-Legendre quadrature in theta direction
 #define GaussInt
 
-#define ABEtype 0
+/// ****
+// 0: BSSN vacuum
+// 1: coupled to scalar field
+// 2: Z4c vacuum
+// 3: coupled to Maxwell field
+//
+#define ABEtype 2
 
+/// ****
+// using Apparent Horizon Finder
 //#define With_AHF
+
+/// ****
+// Psi4 calculation method
+// 0: EB method
+// 1: 4-D method
+//
 #define Psi4type 0
 
+/// ****
+// for Using point psi4 or not
 //#define Point_Psi4
 
+/// ****
+// RestrictProlong in Step (0) or after Step (1)
 #define RPS 1
 
+/// ****
+// Enforce algebra constraint
+// for every RK4 sub step: 0
+// only when iter_count == 3: 1
+// after routine Step: 2
 #define AGM 0
 
+/// ****
+// Restrict Prolong using BAM style 1 or old style 0
 #define RPB 0
 
+/// ****
+// 1: move Analysis out ot 4 sub steps and treat PBH with Euler method
 #define MAPBH 1
 
+/// ****
+// parallel structure, 0: level by level, 1: considering all levels, 2: as 1 but reverse the CPU order, 3: Frank's scheme
 #define PSTR 0
 
+/// ****
+// regrid for every level or for all levels at a time
+// 0: for every level; 1: for all
 #define REGLEV 0
 
+/// ****
+// use gpu or not
 //#define USE_GPU
 
+/// ****
+// use checkpoint for every process
 //#define CHECKDETAIL
 
+/// ****
+// use FakeCheckPrepare to write CheckPoint
 //#define FAKECHECK
-
-//
-// define SommerType
-//     sommerfeld boundary type
-//     0: bam
-//     1: shibata
-//
-// define GaussInt
-//     for Using Gauss-Legendre quadrature in theta direction
-//
-// define ABEtype
-//     0: BSSN vacuum
-//     1: coupled to scalar field
-//     2: Z4c vacuum
-//     3: coupled to Maxwell field
-//
-// define With_AHF
-//     using Apparent Horizon Finder
-//
-// define Psi4type
-//     Psi4 calculation method
-//     0: EB method
-//     1: 4-D method
-//
-// define Point_Psi4
-//     for Using point psi4 or not
-//
-// define RPS
-//     RestrictProlong in Step (0) or after Step (1)
-//
-// define AGM
-//     Enforce algebra constraint
-//     for every RK4 sub step: 0
-//     only when iter_count == 3: 1
-//     after routine Step: 2
-//
-// define RPB
-//     Restrict Prolong using BAM style 1 or old style 0
-//
-// define MAPBH
-//     1: move Analysis out ot 4 sub steps and treat PBH with Euler method
-//
-// define PSTR
-//     parallel structure
-//     0: level by level
-//     1: considering all levels
-//     2: as 1 but reverse the CPU order
-//     3: Frank's scheme
-//
-// define REGLEV
-//     regrid for every level or for all levels at a time
-//     0: for every level;
-//     1: for all
-//
-// define USE_GPU
-//     use gpu or not
-//
-// define CHECKDETAIL
-//     use checkpoint for every process
-//
-// define FAKECHECK
-//     use FakeCheckPrepare to write CheckPoint
-//
-
 ////================================================================
 //  some basic parameters for numerical calculation
-////================================================================
-
 #define dim 3
 
-//#define Cell or Vertex in "macrodef.fh" 
+//#define Cell or Vertex in "microdef.fh"
 
+// ******
+// buffer point number for mesh refinement interface
 #define buffer_width 6
 
+// ******
+// buffer point number shell-box interface, on shell
 #define SC_width buffer_width
-
+// buffer point number shell-box interface, on box
 #define CS_width (2*buffer_width)
 
-//
-// define Cell or Vertex in "macrodef.fh" 
-//
-// define buffer_width
-//     buffer point number for mesh refinement interface
-//
-// define SC_width buffer_width
-//     buffer point number shell-box interface, on shell
-//
-// define CS_width
-//     buffer point number shell-box interface, on box
-//
-
 #if(buffer_width < ghost_width)
-#   error we always assume buffer_width>ghost_width
+#error we always assume buffer_width>ghost_width
 #endif
 
 #define PACK 1
@@ -142,4 +110,3 @@
 #define TINY 1e-10
 
 #endif   /* MICRODEF_H */
-

AMSS_NCKU_source/makefile

@@ -2,35 +2,6 @@
 
 include makefile.inc
 
-## polint(ordn=6) kernel selector:
-##   1 (default): barycentric fast path
-##   0          : fallback to Neville path
-POLINT6_USE_BARY ?= 1
-POLINT6_FLAG = -DPOLINT6_USE_BARYCENTRIC=$(POLINT6_USE_BARY)
-
-## ABE build flags selected by PGO_MODE (set in makefile.inc, default: opt)
-##   make                        -> opt  (PGO-guided, maximum performance)
-##   make PGO_MODE=instrument    -> instrument (Phase 1: collect fresh profile data)
-PROFDATA = /home/$(shell whoami)/AMSS-NCKU/pgo_profile/default.profdata
-
-ifeq ($(PGO_MODE),instrument)
-## Phase 1: instrumentation — omit -ipo/-fp-model fast=2 for faster build and numerical stability
-CXXAPPFLAGS = -O3 -xHost -fma -fprofile-instr-generate -ipo \
-              -Dfortran3 -Dnewc -I${MKLROOT}/include $(INTERP_LB_FLAGS)
-f90appflags = -O3 -xHost -fma -fprofile-instr-generate -ipo \
-              -align array64byte -fpp -I${MKLROOT}/include $(POLINT6_FLAG)
-else
-## opt (default): maximum performance with PGO profile data -fprofile-instr-use=$(PROFDATA) \
-## PGO has been turned off, now tested and found to be negative optimization
-## INTERP_LB_FLAGS has been turned off too, now tested and found to be negative optimization
-
-
-CXXAPPFLAGS = -O3 -xHost -fp-model fast=2 -fma -ipo \
-              -Dfortran3 -Dnewc -I${MKLROOT}/include $(INTERP_LB_FLAGS)
-f90appflags = -O3 -xHost -fp-model fast=2 -fma -ipo \
-              -align array64byte -fpp -I${MKLROOT}/include $(POLINT6_FLAG)
-endif
-
 .SUFFIXES: .o .f90 .C .for .cu
 
 .f90.o:
@@ -45,76 +16,31 @@ endif
 .cu.o:
 	$(Cu) $(CUDA_APP_FLAGS) -c $< -o $@ $(CUDA_LIB_PATH)
 
-# C rewrite of BSSN RHS kernel and helpers
-bssn_rhs_c.o: bssn_rhs_c.C
-	${CXX} $(CXXAPPFLAGS) -c $< $(filein) -o $@
-
-fderivs_c.o: fderivs_c.C
-	${CXX} $(CXXAPPFLAGS) -c $< $(filein) -o $@
-
-fdderivs_c.o: fdderivs_c.C
-	${CXX} $(CXXAPPFLAGS) -c $< $(filein) -o $@
-
-kodiss_c.o: kodiss_c.C
-	${CXX} $(CXXAPPFLAGS) -c $< $(filein) -o $@
-
-lopsided_c.o: lopsided_c.C
-	${CXX} $(CXXAPPFLAGS) -c $< $(filein) -o $@
-
-lopsided_kodis_c.o: lopsided_kodis_c.C
-	${CXX} $(CXXAPPFLAGS) -c $< $(filein) -o $@
-
-#interp_lb_profile.o: interp_lb_profile.C interp_lb_profile.h
-#	${CXX} $(CXXAPPFLAGS) -c $< $(filein) -o $@
-
-## TwoPunctureABE uses fixed optimal flags with its own PGO profile, independent of CXXAPPFLAGS
-TP_PROFDATA = /home/$(shell whoami)/AMSS-NCKU/pgo_profile/TwoPunctureABE.profdata
-TP_OPTFLAGS = -O3 -xHost -fp-model fast=2 -fma -ipo \
-              -fprofile-instr-use=$(TP_PROFDATA) \
-              -Dfortran3 -Dnewc -I${MKLROOT}/include
-
 TwoPunctures.o: TwoPunctures.C
-	${CXX} $(TP_OPTFLAGS) -qopenmp -c $< -o $@
+	${CXX} $(CXXAPPFLAGS) -qopenmp -c $< -o $@
 
 TwoPunctureABE.o: TwoPunctureABE.C
-	${CXX} $(TP_OPTFLAGS) -qopenmp -c $< -o $@
+	${CXX} $(CXXAPPFLAGS) -qopenmp -c $< -o $@
 
 # Input files
-
-## Kernel implementation switch (set USE_CXX_KERNELS=0 to fall back to Fortran)
-ifeq ($(USE_CXX_KERNELS),0)
-# Fortran mode: no C rewrite files; bssn_rhs.o is included via F90FILES below
-CFILES =
-else
-# C++ mode (default): C rewrite of bssn_rhs and helper kernels
-CFILES = bssn_rhs_c.o fderivs_c.o fdderivs_c.o kodiss_c.o lopsided_c.o lopsided_kodis_c.o
-endif
-
-## RK4 kernel switch (independent from USE_CXX_KERNELS)
-ifeq ($(USE_CXX_RK4),1)
-CFILES += rungekutta4_rout_c.o
-RK4_F90_OBJ =
-else
-RK4_F90_OBJ = rungekutta4_rout.o
-endif
-
 C++FILES = ABE.o Ansorg.o Block.o misc.o monitor.o Parallel.o MPatch.o var.o\
            cgh.o bssn_class.o surface_integral.o ShellPatch.o\
 	   bssnEScalar_class.o perf.o Z4c_class.o NullShellPatch.o\
 	   bssnEM_class.o cpbc_util.o z4c_rhs_point.o checkpoint.o\
            Parallel_bam.o scalar_class.o transpbh.o NullShellPatch2.o\
-	   NullShellPatch2_Evo.o writefile_f.o interp_lb_profile.o
+	   NullShellPatch2_Evo.o writefile_f.o
 	   
 C++FILES_GPU = ABE.o Ansorg.o Block.o misc.o monitor.o Parallel.o MPatch.o var.o\
-           cgh.o bssn_class.o surface_integral.o ShellPatch.o\
+           cgh.o surface_integral.o ShellPatch.o\
 	   bssnEScalar_class.o perf.o Z4c_class.o NullShellPatch.o\
 	   bssnEM_class.o cpbc_util.o z4c_rhs_point.o checkpoint.o\
            Parallel_bam.o scalar_class.o transpbh.o NullShellPatch2.o\
-	   NullShellPatch2_Evo.o bssn_cuda_step.o writefile_f.o
+	   NullShellPatch2_Evo.o \
+	   bssn_gpu_class.o bssn_step_gpu.o bssn_macro.o writefile_f.o
 
-F90FILES_BASE = enforce_algebra.o fmisc.o initial_puncture.o prolongrestrict.o\
+F90FILES = enforce_algebra.o fmisc.o initial_puncture.o prolongrestrict.o\
 	   prolongrestrict_cell.o prolongrestrict_vertex.o\
-	   $(RK4_F90_OBJ) diff_new.o kodiss.o kodiss_sh.o\
+	   rungekutta4_rout.o bssn_rhs.o diff_new.o kodiss.o kodiss_sh.o\
 	   lopsidediff.o sommerfeld_rout.o getnp4.o diff_new_sh.o\
 	   shellfunctions.o bssn_rhs_ss.o Set_Rho_ADM.o\
            getnp4EScalar.o bssnEScalar_rhs.o bssn_constraint.o ricci_gamma.o\
@@ -125,14 +51,6 @@ F90FILES_BASE = enforce_algebra.o fmisc.o initial_puncture.o prolongrestrict.o\
 	   scalar_rhs.o initial_scalar.o NullEvol2.o initial_null2.o\
 	   NullNews2.o tool_f.o
 
-ifeq ($(USE_CXX_KERNELS),0)
-# Fortran mode: include original bssn_rhs.o
-F90FILES = $(F90FILES_BASE) bssn_rhs.o
-else
-# C++ mode (default): bssn_rhs.o replaced by C++ kernel
-F90FILES = $(F90FILES_BASE)
-endif
-
 F77FILES = zbesh.o
 
 AHFDOBJS = expansion.o expansion_Jacobian.o patch.o coords.o patch_info.o patch_interp.o patch_system.o \
@@ -142,10 +60,10 @@ initial_guess.o Newton.o Jacobian.o ilucg.o IntPnts0.o IntPnts.o
 
 TwoPunctureFILES = TwoPunctureABE.o TwoPunctures.o 
 
-CUDAFILES = bssn_gpu.o bssn_cuda_ops.o
+CUDAFILES = bssn_gpu.o bssn_gpu_rhs_ss.o
 
 # file dependences
-$(C++FILES) $(C++FILES_GPU) $(F90FILES) $(CFILES) $(AHFDOBJS) $(CUDAFILES): macrodef.fh
+$(C++FILES) $(C++FILESGPU) $(F90FILES) $(AHFDOBJS) $(CUDAFILES): macrodef.fh
 
 $(C++FILES): Block.h enforce_algebra.h fmisc.h initial_puncture.h macrodef.h\
 	     misc.h monitor.h MyList.h Parallel.h MPatch.h prolongrestrict.h\
@@ -168,7 +86,7 @@ $(C++FILES_GPU): Block.h enforce_algebra.h fmisc.h initial_puncture.h macrodef.h
              
 $(AHFDOBJS): cctk.h cctk_Config.h cctk_Types.h cctk_Constants.h myglobal.h
 
-$(C++FILES) $(C++FILES_GPU) $(CFILES) $(AHFDOBJS) $(CUDAFILES): macrodef.h
+$(C++FILES) $(C++FILES_GPU) $(AHFDOBJS) $(CUDAFILES): macrodef.h
 
 TwoPunctureFILES: TwoPunctures.h
 
@@ -177,14 +95,14 @@ $(CUDAFILES): bssn_gpu.h gpu_mem.h gpu_rhsSS_mem.h
 misc.o : zbesh.o
 
 # projects
-ABE: $(C++FILES) $(CFILES) $(F90FILES) $(F77FILES) $(AHFDOBJS)
+ABE: $(C++FILES) $(F90FILES) $(F77FILES) $(AHFDOBJS) 
-	$(CLINKER) $(CXXAPPFLAGS) -o $@ $(C++FILES) $(CFILES) $(F90FILES) $(F77FILES) $(AHFDOBJS) $(LDLIBS)
+	$(CLINKER) $(CXXAPPFLAGS) -o $@ $(C++FILES) $(F90FILES) $(F77FILES) $(AHFDOBJS) $(LDLIBS)
 	
-ABEGPU: $(C++FILES_GPU) $(CFILES) $(F90FILES) $(F77FILES) $(AHFDOBJS) $(CUDAFILES)
+ABEGPU: $(C++FILES_GPU) $(F90FILES) $(F77FILES) $(AHFDOBJS) $(CUDAFILES)
-	$(CLINKER) $(CXXAPPFLAGS) -o $@ $(C++FILES_GPU) $(CFILES) $(F90FILES) $(F77FILES) $(AHFDOBJS) $(CUDAFILES) $(LDLIBS)
+	$(CLINKER) $(CXXAPPFLAGS) -o $@ $(C++FILES_GPU) $(F90FILES) $(F77FILES) $(AHFDOBJS) $(CUDAFILES) $(LDLIBS)
 
 TwoPunctureABE: $(TwoPunctureFILES)
-	$(CLINKER) $(TP_OPTFLAGS) -qopenmp -o $@ $(TwoPunctureFILES) $(LDLIBS)
+	$(CLINKER) $(CXXAPPFLAGS) -qopenmp -o $@ $(TwoPunctureFILES) $(LDLIBS)
 
 clean:
 	rm *.o ABE ABEGPU TwoPunctureABE make.log -f

AMSS_NCKU_source/makefile.inc

@@ -8,54 +8,18 @@ filein  = -I/usr/include/ -I${MKLROOT}/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${MKLROOT}/lib -lmkl_intel_lp64 -lmkl_sequential -lmkl_core -lifcore -limf -lpthread -lm -ldl -liomp5
+LDLIBS  = -L${MKLROOT}/lib -lmkl_intel_lp64 -lmkl_sequential -lmkl_core -lifcore -limf -lpthread -lm -ldl
-CUDA_LDLIBS = -L/usr/local/cuda-12.9/targets/x86_64-linux/lib -lcudart
-
-## Memory allocator switch
-##   1 (default) : link Intel oneTBB allocator (libtbbmalloc)
-##   0           : use system default allocator (ptmalloc)
-USE_TBBMALLOC ?= 1
-TBBMALLOC_SO ?= /home/intel/oneapi/2025.3/lib/libtbbmalloc.so
-ifneq ($(wildcard $(TBBMALLOC_SO)),)
-TBBMALLOC_LIBS = -Wl,--no-as-needed $(TBBMALLOC_SO) -Wl,--as-needed
-else
-TBBMALLOC_LIBS = -Wl,--no-as-needed -ltbbmalloc -Wl,--as-needed
-endif
-ifeq ($(USE_TBBMALLOC),1)
-LDLIBS := $(TBBMALLOC_LIBS) $(LDLIBS)
-endif
-
-LDLIBS := $(CUDA_LDLIBS) $(LDLIBS)
-
-## PGO build mode switch (ABE only; TwoPunctureABE always uses opt flags)
-##   opt        : (default) maximum performance with PGO profile-guided optimization
-##   instrument : PGO Phase 1 instrumentation to collect fresh profile data
-PGO_MODE ?= opt
-
-## Interp_Points load balance profiling mode
-##   off        : (default) no load balance instrumentation
-##   profile    : Pass 1 — instrument Interp_Points to collect timing profile
-##   optimize   : Pass 2 — read profile and apply block rebalancing
-INTERP_LB_MODE ?= off
-
-ifeq ($(INTERP_LB_MODE),profile)
-INTERP_LB_FLAGS = -DINTERP_LB_PROFILE
-else ifeq ($(INTERP_LB_MODE),optimize)
-INTERP_LB_FLAGS = -DINTERP_LB_OPTIMIZE
-else
-INTERP_LB_FLAGS =
-endif
-
-## Kernel implementation switch
-##   1 (default) : use C++ rewrite of bssn_rhs and helper kernels (faster)
-##   0           : fall back to original Fortran kernels
-USE_CXX_KERNELS ?= 1
-
-## RK4 kernel implementation switch
-##   1 (default) : use C/C++ rewrite of rungekutta4_rout (for optimization experiments)
-##   0           : use original Fortran rungekutta4_rout.o
-USE_CXX_RK4 ?= 1
 
+## Aggressive optimization flags + PGO Phase 2 (profile-guided optimization)
+## -fprofile-instr-use: use collected profile data to guide optimization decisions
+##   (branch prediction, basic block layout, inlining, loop unrolling)
+PROFDATA     = ../../pgo_profile/default.profdata
+CXXAPPFLAGS  = -O3 -xHost -fp-model fast=2 -fma -ipo \
+               -fprofile-instr-use=$(PROFDATA) \
+               -Dfortran3 -Dnewc -I${MKLROOT}/include
+f90appflags  = -O3 -xHost -fp-model fast=2 -fma -ipo \
+               -fprofile-instr-use=$(PROFDATA) \
+               -align array64byte -fpp -I${MKLROOT}/include
 f90          = ifx
 f77          = ifx
 CXX          = icpx

AMSS_NCKU_source/prolongrestrict_cell.f90

@@ -1934,35 +1934,18 @@
 ! when if=1 -> ic=0, this is different to vertex center grid 
   real*8, dimension(-2:extc(1),-2:extc(2),-2:extc(3))   :: funcc
   integer,dimension(3) :: cxI
-  integer :: i,j,k,ii,jj,kk,px,py,pz
+  integer :: i,j,k,ii,jj,kk
   real*8, dimension(6,6) :: tmp2
   real*8, dimension(6) :: tmp1
-  integer, dimension(extf(1)) :: cix
-  integer, dimension(extf(2)) :: ciy
-  integer, dimension(extf(3)) :: ciz
-  integer, dimension(extf(1)) :: pix
-  integer, dimension(extf(2)) :: piy
-  integer, dimension(extf(3)) :: piz
 
   real*8, parameter :: C1=7.7d1/8.192d3,C2=-6.93d2/8.192d3,C3=3.465d3/4.096d3
   real*8, parameter :: C6=6.3d1/8.192d3,C5=-4.95d2/8.192d3,C4=1.155d3/4.096d3
-  real*8, dimension(6,2), parameter :: WC = reshape((/&
-      C1,C2,C3,C4,C5,C6,&
-      C6,C5,C4,C3,C2,C1/), (/6,2/))
 
   integer::imini,imaxi,jmini,jmaxi,kmini,kmaxi
   integer::imino,imaxo,jmino,jmaxo,kmino,kmaxo
-  integer::maxcx,maxcy,maxcz
 
   real*8,dimension(3) :: CD,FD
-  real*8 :: tmp_yz(extc(1), 6)      ! 存储整条 X 线上 6 个 Y 轴偏置的 Z 向插值结果
+  
-  real*8 :: tmp_xyz_line(-2:extc(1))   ! 包含 X 向 6 点模板访问所需下界
-  real*8 :: v1, v2, v3, v4, v5, v6
-  integer :: ic, jc, kc, ix_offset,ix,iy,iz,jc_min,jc_max,ic_min,ic_max,kc_min,kc_max
-  integer :: i_lo, i_hi, j_lo, j_hi, k_lo, k_hi
-  logical :: need_full_symmetry
-  real*8 :: res_line
-  real*8 :: tmp_z_slab(-2:extc(1), -2:extc(2))  ! 包含 Y/X 向模板访问所需下界
   if(wei.ne.3)then
      write(*,*)"prolongrestrict.f90::prolong3: this routine only surport 3 dimension"
      write(*,*)"dim = ",wei
@@ -2037,140 +2020,145 @@
           return
   endif
 
-  do i = imino,imaxo
+  call symmetry_bd(3,extc,func,funcc,SoA)
-     ii = i + lbf(1) - 1
-     cix(i) = ii/2 - lbc(1) + 1
-     if(ii/2*2 == ii)then
-        pix(i) = 1
-     else
-        pix(i) = 2
-     endif
-  enddo
-  do j = jmino,jmaxo
-     jj = j + lbf(2) - 1
-     ciy(j) = jj/2 - lbc(2) + 1
-     if(jj/2*2 == jj)then
-        piy(j) = 1
-     else
-        piy(j) = 2
-     endif
-  enddo
-  do k = kmino,kmaxo
-     kk = k + lbf(3) - 1
-     ciz(k) = kk/2 - lbc(3) + 1
-     if(kk/2*2 == kk)then
-        piz(k) = 1
-     else
-        piz(k) = 2
-     endif
-  enddo
-
-  ic_min = minval(cix(imino:imaxo))
-  ic_max = maxval(cix(imino:imaxo))
-  jc_min = minval(ciy(jmino:jmaxo))
-  jc_max = maxval(ciy(jmino:jmaxo))
-  kc_min = minval(ciz(kmino:kmaxo))
-  kc_max = maxval(ciz(kmino:kmaxo))
-
-  maxcx = ic_max
-  maxcy = jc_max
-  maxcz = kc_max
-  if(maxcx+3 > extc(1) .or. maxcy+3 > extc(2) .or. maxcz+3 > extc(3))then
-     write(*,*)"error in prolong"
-     return
-  endif
-
-  i_lo = ic_min - 2
-  i_hi = ic_max + 3
-  j_lo = jc_min - 2
-  j_hi = jc_max + 3
-  k_lo = kc_min - 2
-  k_hi = kc_max + 3
-  need_full_symmetry = (i_lo < 1) .or. (j_lo < 1) .or. (k_lo < 1)
-  if(need_full_symmetry)then
-     call symmetry_bd(3,extc,func,funcc,SoA)
-  else
-     funcc(i_lo:i_hi,j_lo:j_hi,k_lo:k_hi) = func(i_lo:i_hi,j_lo:j_hi,k_lo:k_hi)
-  endif
-
-     ! 对每个 k（pz, kc 固定）预计算 Z 向插值的 2D 切片
-
-do k = kmino, kmaxo
-    pz = piz(k); kc = ciz(k)
-    ! --- Pass 1: Z 方向，只算一次 ---
-    do iy = jc_min-2, jc_max+3   ! 仅需的 iy 范围（对应 jc-2:jc+3）
-        do ii = ic_min-2, ic_max+3  ! 仅需的 ii 范围（对应 cix-2:cix+3）
-            tmp_z_slab(ii, iy) = sum(WC(:,pz) * funcc(ii, iy, kc-2:kc+3))
-        end do
-    end do
-
-    do j = jmino, jmaxo
-        py = piy(j); jc = ciy(j)
-        ! --- Pass 2: Y 方向 ---
-        do ii = ic_min-2, ic_max+3
-            tmp_xyz_line(ii) = sum(WC(:,py) * tmp_z_slab(ii, jc-2:jc+3))
-        end do
-        ! --- Pass 3: X 方向 ---
-        do i = imino, imaxo
-            funf(i,j,k) = sum(WC(:,pix(i)) * tmp_xyz_line(cix(i)-2:cix(i)+3))
-        end do
-    end do
-end do
      
 !~~~~~~> prolongation start...
+  do k = kmino,kmaxo
+   do j = jmino,jmaxo
+    do i = imino,imaxo
+       cxI(1) = i
+       cxI(2) = j
+       cxI(3) = k
+! change to coarse level reference
+!|---*--- ---*--- ---*--- ---*--- ---*--- ---*--- ---*--- ---*---| 
+!|=======x===============x===============x===============x=======|
+       cxI = (cxI+lbf-1)/2
+! change to array index      
+       cxI = cxI - lbc + 1
+
+       if(any(cxI+3 > extc)) write(*,*)"error in prolong"
+       ii=i+lbf(1)-1
+       jj=j+lbf(2)-1
+       kk=k+lbf(3)-1
 #if 0
- do k = kmino, kmaxo
+       if(ii/2*2==ii)then
-     pz = piz(k)
+         if(jj/2*2==jj)then
-     kc = ciz(k)
+           if(kk/2*2==kk)then
+             tmp2= C1*funcc(cxI(1)-2:cxI(1)+3,cxI(2)-2:cxI(2)+3,cxI(3)-2)+&
+                   C2*funcc(cxI(1)-2:cxI(1)+3,cxI(2)-2:cxI(2)+3,cxI(3)-1)+&
+                   C3*funcc(cxI(1)-2:cxI(1)+3,cxI(2)-2:cxI(2)+3,cxI(3)  )+&
+                   C4*funcc(cxI(1)-2:cxI(1)+3,cxI(2)-2:cxI(2)+3,cxI(3)+1)+&
+                   C5*funcc(cxI(1)-2:cxI(1)+3,cxI(2)-2:cxI(2)+3,cxI(3)+2)+&
+                   C6*funcc(cxI(1)-2:cxI(1)+3,cxI(2)-2:cxI(2)+3,cxI(3)+3)
+             tmp1= C1*tmp2(:,1)+C2*tmp2(:,2)+C3*tmp2(:,3)+C4*tmp2(:,4)+C5*tmp2(:,5)+C6*tmp2(:,6)
+             funf(i,j,k)= C1*tmp1(1)+C2*tmp1(2)+C3*tmp1(3)+C4*tmp1(4)+C5*tmp1(5)+C6*tmp1(6)
+           else
+             tmp2= C6*funcc(cxI(1)-2:cxI(1)+3,cxI(2)-2:cxI(2)+3,cxI(3)-2)+&
+                   C5*funcc(cxI(1)-2:cxI(1)+3,cxI(2)-2:cxI(2)+3,cxI(3)-1)+&
+                   C4*funcc(cxI(1)-2:cxI(1)+3,cxI(2)-2:cxI(2)+3,cxI(3)  )+&
+                   C3*funcc(cxI(1)-2:cxI(1)+3,cxI(2)-2:cxI(2)+3,cxI(3)+1)+&
+                   C2*funcc(cxI(1)-2:cxI(1)+3,cxI(2)-2:cxI(2)+3,cxI(3)+2)+&
+                   C1*funcc(cxI(1)-2:cxI(1)+3,cxI(2)-2:cxI(2)+3,cxI(3)+3)
+             tmp1= C1*tmp2(:,1)+C2*tmp2(:,2)+C3*tmp2(:,3)+C4*tmp2(:,4)+C5*tmp2(:,5)+C6*tmp2(:,6)
+             funf(i,j,k)=  C1*tmp1(1)+C2*tmp1(2)+C3*tmp1(3)+C4*tmp1(4)+C5*tmp1(5)+C6*tmp1(6)
+           endif
+         else
+           if(kk/2*2==kk)then
+             tmp2= C1*funcc(cxI(1)-2:cxI(1)+3,cxI(2)-2:cxI(2)+3,cxI(3)-2)+&
+                   C2*funcc(cxI(1)-2:cxI(1)+3,cxI(2)-2:cxI(2)+3,cxI(3)-1)+&
+                   C3*funcc(cxI(1)-2:cxI(1)+3,cxI(2)-2:cxI(2)+3,cxI(3)  )+&
+                   C4*funcc(cxI(1)-2:cxI(1)+3,cxI(2)-2:cxI(2)+3,cxI(3)+1)+&
+                   C5*funcc(cxI(1)-2:cxI(1)+3,cxI(2)-2:cxI(2)+3,cxI(3)+2)+&
+                   C6*funcc(cxI(1)-2:cxI(1)+3,cxI(2)-2:cxI(2)+3,cxI(3)+3)
+             tmp1= C6*tmp2(:,1)+C5*tmp2(:,2)+C4*tmp2(:,3)+C3*tmp2(:,4)+C2*tmp2(:,5)+C1*tmp2(:,6)
+             funf(i,j,k)= C1*tmp1(1)+C2*tmp1(2)+C3*tmp1(3)+C4*tmp1(4)+C5*tmp1(5)+C6*tmp1(6)
+           else
+             tmp2= C6*funcc(cxI(1)-2:cxI(1)+3,cxI(2)-2:cxI(2)+3,cxI(3)-2)+&
+                   C5*funcc(cxI(1)-2:cxI(1)+3,cxI(2)-2:cxI(2)+3,cxI(3)-1)+&
+                   C4*funcc(cxI(1)-2:cxI(1)+3,cxI(2)-2:cxI(2)+3,cxI(3)  )+&
+                   C3*funcc(cxI(1)-2:cxI(1)+3,cxI(2)-2:cxI(2)+3,cxI(3)+1)+&
+                   C2*funcc(cxI(1)-2:cxI(1)+3,cxI(2)-2:cxI(2)+3,cxI(3)+2)+&
+                   C1*funcc(cxI(1)-2:cxI(1)+3,cxI(2)-2:cxI(2)+3,cxI(3)+3)
+             tmp1= C6*tmp2(:,1)+C5*tmp2(:,2)+C4*tmp2(:,3)+C3*tmp2(:,4)+C2*tmp2(:,5)+C1*tmp2(:,6)
+             funf(i,j,k)=  C1*tmp1(1)+C2*tmp1(2)+C3*tmp1(3)+C4*tmp1(4)+C5*tmp1(5)+C6*tmp1(6)
+           endif
+         endif
+       else
+         if(jj/2*2==jj)then
+           if(kk/2*2==kk)then               
+             tmp2= C1*funcc(cxI(1)-2:cxI(1)+3,cxI(2)-2:cxI(2)+3,cxI(3)-2)+&
+                   C2*funcc(cxI(1)-2:cxI(1)+3,cxI(2)-2:cxI(2)+3,cxI(3)-1)+&
+                   C3*funcc(cxI(1)-2:cxI(1)+3,cxI(2)-2:cxI(2)+3,cxI(3)  )+&
+                   C4*funcc(cxI(1)-2:cxI(1)+3,cxI(2)-2:cxI(2)+3,cxI(3)+1)+&
+                   C5*funcc(cxI(1)-2:cxI(1)+3,cxI(2)-2:cxI(2)+3,cxI(3)+2)+&
+                   C6*funcc(cxI(1)-2:cxI(1)+3,cxI(2)-2:cxI(2)+3,cxI(3)+3)
+             tmp1= C1*tmp2(:,1)+C2*tmp2(:,2)+C3*tmp2(:,3)+C4*tmp2(:,4)+C5*tmp2(:,5)+C6*tmp2(:,6)
+             funf(i,j,k)= C6*tmp1(1)+C5*tmp1(2)+C4*tmp1(3)+C3*tmp1(4)+C2*tmp1(5)+C1*tmp1(6)
+           else
+             tmp2= C6*funcc(cxI(1)-2:cxI(1)+3,cxI(2)-2:cxI(2)+3,cxI(3)-2)+&
+                   C5*funcc(cxI(1)-2:cxI(1)+3,cxI(2)-2:cxI(2)+3,cxI(3)-1)+&
+                   C4*funcc(cxI(1)-2:cxI(1)+3,cxI(2)-2:cxI(2)+3,cxI(3)  )+&
+                   C3*funcc(cxI(1)-2:cxI(1)+3,cxI(2)-2:cxI(2)+3,cxI(3)+1)+&
+                   C2*funcc(cxI(1)-2:cxI(1)+3,cxI(2)-2:cxI(2)+3,cxI(3)+2)+&
+                   C1*funcc(cxI(1)-2:cxI(1)+3,cxI(2)-2:cxI(2)+3,cxI(3)+3)
+             tmp1= C1*tmp2(:,1)+C2*tmp2(:,2)+C3*tmp2(:,3)+C4*tmp2(:,4)+C5*tmp2(:,5)+C6*tmp2(:,6)
+             funf(i,j,k)=  C6*tmp1(1)+C5*tmp1(2)+C4*tmp1(3)+C3*tmp1(4)+C2*tmp1(5)+C1*tmp1(6)
+           endif
+         else
+           if(kk/2*2==kk)then
+             tmp2= C1*funcc(cxI(1)-2:cxI(1)+3,cxI(2)-2:cxI(2)+3,cxI(3)-2)+&
+                   C2*funcc(cxI(1)-2:cxI(1)+3,cxI(2)-2:cxI(2)+3,cxI(3)-1)+&
+                   C3*funcc(cxI(1)-2:cxI(1)+3,cxI(2)-2:cxI(2)+3,cxI(3)  )+&
+                   C4*funcc(cxI(1)-2:cxI(1)+3,cxI(2)-2:cxI(2)+3,cxI(3)+1)+&
+                   C5*funcc(cxI(1)-2:cxI(1)+3,cxI(2)-2:cxI(2)+3,cxI(3)+2)+&
+                   C6*funcc(cxI(1)-2:cxI(1)+3,cxI(2)-2:cxI(2)+3,cxI(3)+3)
+             tmp1= C6*tmp2(:,1)+C5*tmp2(:,2)+C4*tmp2(:,3)+C3*tmp2(:,4)+C2*tmp2(:,5)+C1*tmp2(:,6)
+             funf(i,j,k)= C6*tmp1(1)+C5*tmp1(2)+C4*tmp1(3)+C3*tmp1(4)+C2*tmp1(5)+C1*tmp1(6)
+           else
+             tmp2= C6*funcc(cxI(1)-2:cxI(1)+3,cxI(2)-2:cxI(2)+3,cxI(3)-2)+&
+                   C5*funcc(cxI(1)-2:cxI(1)+3,cxI(2)-2:cxI(2)+3,cxI(3)-1)+&
+                   C4*funcc(cxI(1)-2:cxI(1)+3,cxI(2)-2:cxI(2)+3,cxI(3)  )+&
+                   C3*funcc(cxI(1)-2:cxI(1)+3,cxI(2)-2:cxI(2)+3,cxI(3)+1)+&
+                   C2*funcc(cxI(1)-2:cxI(1)+3,cxI(2)-2:cxI(2)+3,cxI(3)+2)+&
+                   C1*funcc(cxI(1)-2:cxI(1)+3,cxI(2)-2:cxI(2)+3,cxI(3)+3)
+             tmp1= C6*tmp2(:,1)+C5*tmp2(:,2)+C4*tmp2(:,3)+C3*tmp2(:,4)+C2*tmp2(:,5)+C1*tmp2(:,6)
+             funf(i,j,k)=  C6*tmp1(1)+C5*tmp1(2)+C4*tmp1(3)+C3*tmp1(4)+C2*tmp1(5)+C1*tmp1(6)
+           endif
+         endif
+       endif
+#else 
+       if(kk/2*2==kk)then
+             tmp2= C1*funcc(cxI(1)-2:cxI(1)+3,cxI(2)-2:cxI(2)+3,cxI(3)-2)+&
+                   C2*funcc(cxI(1)-2:cxI(1)+3,cxI(2)-2:cxI(2)+3,cxI(3)-1)+&
+                   C3*funcc(cxI(1)-2:cxI(1)+3,cxI(2)-2:cxI(2)+3,cxI(3)  )+&
+                   C4*funcc(cxI(1)-2:cxI(1)+3,cxI(2)-2:cxI(2)+3,cxI(3)+1)+&
+                   C5*funcc(cxI(1)-2:cxI(1)+3,cxI(2)-2:cxI(2)+3,cxI(3)+2)+&
+                   C6*funcc(cxI(1)-2:cxI(1)+3,cxI(2)-2:cxI(2)+3,cxI(3)+3)
+       else
+             tmp2= C6*funcc(cxI(1)-2:cxI(1)+3,cxI(2)-2:cxI(2)+3,cxI(3)-2)+&
+                   C5*funcc(cxI(1)-2:cxI(1)+3,cxI(2)-2:cxI(2)+3,cxI(3)-1)+&
+                   C4*funcc(cxI(1)-2:cxI(1)+3,cxI(2)-2:cxI(2)+3,cxI(3)  )+&
+                   C3*funcc(cxI(1)-2:cxI(1)+3,cxI(2)-2:cxI(2)+3,cxI(3)+1)+&
+                   C2*funcc(cxI(1)-2:cxI(1)+3,cxI(2)-2:cxI(2)+3,cxI(3)+2)+&
+                   C1*funcc(cxI(1)-2:cxI(1)+3,cxI(2)-2:cxI(2)+3,cxI(3)+3)
+       endif
 
-     do j = jmino, jmaxo
+       if(jj/2*2==jj)then
-        py = piy(j)
+             tmp1= C1*tmp2(:,1)+C2*tmp2(:,2)+C3*tmp2(:,3)+C4*tmp2(:,4)+C5*tmp2(:,5)+C6*tmp2(:,6)
-        jc = ciy(j)
+       else
+             tmp1= C6*tmp2(:,1)+C5*tmp2(:,2)+C4*tmp2(:,3)+C3*tmp2(:,4)+C2*tmp2(:,5)+C1*tmp2(:,6)
+       endif
 
-! --- 步骤 1 & 2 融合：分段处理 X 轴，提升 Cache 命中率 ---
+       if(ii/2*2==ii)then
-        ! 我们将 ii 循环逻辑重组，减少对 funcc 的跨行重复访问
+             funf(i,j,k)= C1*tmp1(1)+C2*tmp1(2)+C3*tmp1(3)+C4*tmp1(4)+C5*tmp1(5)+C6*tmp1(6)
-        do ii = 1, extc(1)
+       else
-           ! 1. 先做 Z 方向的 6 条线插值（针对当前的 ii 和当前的 6 个 iy）
+             funf(i,j,k)= C6*tmp1(1)+C5*tmp1(2)+C4*tmp1(3)+C3*tmp1(4)+C2*tmp1(5)+C1*tmp1(6)
-           ! 我们直接在这里把 Y 方向的加权也做了，省去 tmp_yz 数组
+       endif
-           ! 这样 funcc 的数据读进来后立即完成所有维度的贡献，不再写回内存
-           
-           res_line = 0.0d0
-           do jj = 1, 6
-              iy = jc - 3 + jj
-              ! 这一行代码是核心：一次性完成 Z 插值并加上 Y 的权重
-              ! 编译器会把 WC(jj, py) 存在寄存器里
-              res_line = res_line + WC(jj, py) * ( &
-                         WC(1, pz) * funcc(ii, iy, kc-2) + &
-                         WC(2, pz) * funcc(ii, iy, kc-1) + &
-                         WC(3, pz) * funcc(ii, iy, kc  ) + &
-                         WC(4, pz) * funcc(ii, iy, kc+1) + &
-                         WC(5, pz) * funcc(ii, iy, kc+2) + &
-                         WC(6, pz) * funcc(ii, iy, kc+3) )
-           end do
-           tmp_xyz_line(ii) = res_line
-        end do
-
-
-
-
-        ! 3. 【降维：X 向】最后在最内层只处理 X 方向的 6 点加权
-        ! 此时每个点的计算量从原来的 200+ 次乘法降到了仅 6 次
-        do i = imino, imaxo
-           px = pix(i)
-           ic = cix(i)
-           
-           ! 直接从预计算好的 line 中读取连续的 6 个点
-           ! ic-2 到 ic+3 对应原始 6 点算子
-           funf(i,j,k) = WC(1,px)*tmp_xyz_line(ic-2) + &
-                         WC(2,px)*tmp_xyz_line(ic-1) + &
-                         WC(3,px)*tmp_xyz_line(ic  ) + &
-                         WC(4,px)*tmp_xyz_line(ic+1) + &
-                         WC(5,px)*tmp_xyz_line(ic+2) + &
-                         WC(6,px)*tmp_xyz_line(ic+3)
-        end do
-     end do
-  end do
 #endif
+    enddo
+   enddo
+  enddo
+
   return
 
   end subroutine prolong3
@@ -2370,13 +2358,6 @@ end do
 
   real*8,dimension(3) :: CD,FD
   
-  real*8 :: tmp_xz_plane(-1:extf(1), 6)
-  real*8 :: tmp_x_line(-1:extf(1))
-  integer :: fi, fj, fk, ii, jj, kk
-  integer :: fi_min, fi_max, ii_lo, ii_hi
-  integer :: fj_min, fj_max, fk_min, fk_max, jj_lo, jj_hi, kk_lo, kk_hi
-  logical :: need_full_symmetry
-
   if(wei.ne.3)then
      write(*,*)"prolongrestrict.f90::restrict3: this routine only surport 3 dimension"
      write(*,*)"dim = ",wei
@@ -2455,86 +2436,9 @@ end do
           stop
   endif
 
-  ! 仅计算 X 向最终写回所需的窗口：
+  call symmetry_bd(2,extf,funf,funff,SoA)
-  ! func(i,j,k) 只访问 tmp_x_line(fi-2:fi+3)
-  fi_min = 2*(imino + lbc(1) - 1) - 1 - lbf(1) + 1
-  fi_max = 2*(imaxo + lbc(1) - 1) - 1 - lbf(1) + 1
-  fj_min = 2*(jmino + lbc(2) - 1) - 1 - lbf(2) + 1
-  fj_max = 2*(jmaxo + lbc(2) - 1) - 1 - lbf(2) + 1
-  fk_min = 2*(kmino + lbc(3) - 1) - 1 - lbf(3) + 1
-  fk_max = 2*(kmaxo + lbc(3) - 1) - 1 - lbf(3) + 1
-  ii_lo = fi_min - 2
-  ii_hi = fi_max + 3
-  jj_lo = fj_min - 2
-  jj_hi = fj_max + 3
-  kk_lo = fk_min - 2
-  kk_hi = fk_max + 3
-  if(ii_lo < -1 .or. ii_hi > extf(1) .or. &
-     jj_lo < -1 .or. jj_hi > extf(2) .or. &
-     kk_lo < -1 .or. kk_hi > extf(3))then
-      write(*,*)"restrict3: invalid stencil window"
-      write(*,*)"ii=",ii_lo,ii_hi," jj=",jj_lo,jj_hi," kk=",kk_lo,kk_hi
-      write(*,*)"extf=",extf
-      stop
-  endif
-  need_full_symmetry = (ii_lo < 1) .or. (jj_lo < 1) .or. (kk_lo < 1)
-  if(need_full_symmetry)then
-      call symmetry_bd(2,extf,funf,funff,SoA)
-  else
-      funff(ii_lo:ii_hi,jj_lo:jj_hi,kk_lo:kk_hi) = funf(ii_lo:ii_hi,jj_lo:jj_hi,kk_lo:kk_hi)
-  endif
 
 !~~~~~~> restriction start...
-do k = kmino, kmaxo
-    fk = 2*(k + lbc(3) - 1) - 1 - lbf(3) + 1
-
-    do j = jmino, jmaxo
-        fj = 2*(j + lbc(2) - 1) - 1 - lbf(2) + 1
-        
-        ! 优化点 1: 显式展开 Z 方向计算，减少循环开销
-        ! 确保 ii 循环是最内层且连续访问
-        !DIR$ VECTOR ALWAYS
-        do ii = ii_lo, ii_hi
-            ! 预计算当前 j 对应的 6 行在 Z 方向的压缩结果
-            ! 这里直接硬编码 jj 的偏移，彻底消除一层循环
-            tmp_xz_plane(ii, 1) = C1*(funff(ii,fj-2,fk-2)+funff(ii,fj-2,fk+3)) + &
-                                  C2*(funff(ii,fj-2,fk-1)+funff(ii,fj-2,fk+2)) + &
-                                  C3*(funff(ii,fj-2,fk  )+funff(ii,fj-2,fk+1))
-            tmp_xz_plane(ii, 2) = C1*(funff(ii,fj-1,fk-2)+funff(ii,fj-1,fk+3)) + &
-                                  C2*(funff(ii,fj-1,fk-1)+funff(ii,fj-1,fk+2)) + &
-                                  C3*(funff(ii,fj-1,fk  )+funff(ii,fj-1,fk+1))
-            tmp_xz_plane(ii, 3) = C1*(funff(ii,fj  ,fk-2)+funff(ii,fj  ,fk+3)) + &
-                                  C2*(funff(ii,fj  ,fk-1)+funff(ii,fj  ,fk+2)) + &
-                                  C3*(funff(ii,fj  ,fk  )+funff(ii,fj  ,fk+1))
-            tmp_xz_plane(ii, 4) = C1*(funff(ii,fj+1,fk-2)+funff(ii,fj+1,fk+3)) + &
-                                  C2*(funff(ii,fj+1,fk-1)+funff(ii,fj+1,fk+2)) + &
-                                  C3*(funff(ii,fj+1,fk  )+funff(ii,fj+1,fk+1))
-            tmp_xz_plane(ii, 5) = C1*(funff(ii,fj+2,fk-2)+funff(ii,fj+2,fk+3)) + &
-                                  C2*(funff(ii,fj+2,fk-1)+funff(ii,fj+2,fk+2)) + &
-                                  C3*(funff(ii,fj+2,fk  )+funff(ii,fj+2,fk+1))
-            tmp_xz_plane(ii, 6) = C1*(funff(ii,fj+3,fk-2)+funff(ii,fj+3,fk+3)) + &
-                                  C2*(funff(ii,fj+3,fk-1)+funff(ii,fj+3,fk+2)) + &
-                                  C3*(funff(ii,fj+3,fk  )+funff(ii,fj+3,fk+1))
-        end do
-
-        ! 优化点 2: 同样向量化 Y 方向压缩
-        !DIR$ VECTOR ALWAYS
-        do ii = ii_lo, ii_hi
-            tmp_x_line(ii) = C1*(tmp_xz_plane(ii, 1) + tmp_xz_plane(ii, 6)) + &
-                            C2*(tmp_xz_plane(ii, 2) + tmp_xz_plane(ii, 5)) + &
-                            C3*(tmp_xz_plane(ii, 3) + tmp_xz_plane(ii, 4))
-        end do
-
-        ! 优化点 3: 最终写入，利用已经缓存在 tmp_x_line 的数据
-        do i = imino, imaxo
-            fi = 2*(i + lbc(1) - 1) - 1 - lbf(1) + 1
-            func(i, j, k) = C1*(tmp_x_line(fi-2) + tmp_x_line(fi+3)) + &
-                            C2*(tmp_x_line(fi-1) + tmp_x_line(fi+2)) + &
-                            C3*(tmp_x_line(fi  ) + tmp_x_line(fi+1))
-        end do
-    end do
-end do
-#if 0
   do k = kmino,kmaxo
    do j = jmino,jmaxo
     do i = imino,imaxo
@@ -2558,7 +2462,7 @@ end do
     enddo
    enddo
   enddo
-#endif
+  
   return
 
   end subroutine restrict3

AMSS_NCKU_source/rungekutta4_rout_c.C

@@ -1,212 +0,0 @@
-#include "rungekutta4_rout.h"
-#include <cstdio>
-#include <cstdlib>
-#include <cstddef>
-#include <complex>
-#include <immintrin.h>
-
-namespace {
-
-inline void rk4_stage0(std::size_t n,
-                       const double *__restrict f0,
-                       const double *__restrict frhs,
-                       double *__restrict f1,
-                       double c) {
-    std::size_t i = 0;
-#if defined(__AVX512F__)
-    const __m512d vc = _mm512_set1_pd(c);
-    for (; i + 7 < n; i += 8) {
-        const __m512d v0 = _mm512_loadu_pd(f0 + i);
-        const __m512d vr = _mm512_loadu_pd(frhs + i);
-        _mm512_storeu_pd(f1 + i, _mm512_fmadd_pd(vc, vr, v0));
-    }
-#elif defined(__AVX2__)
-    const __m256d vc = _mm256_set1_pd(c);
-    for (; i + 3 < n; i += 4) {
-        const __m256d v0 = _mm256_loadu_pd(f0 + i);
-        const __m256d vr = _mm256_loadu_pd(frhs + i);
-        _mm256_storeu_pd(f1 + i, _mm256_fmadd_pd(vc, vr, v0));
-    }
-#endif
-#pragma ivdep
-    for (; i < n; ++i) {
-        f1[i] = f0[i] + c * frhs[i];
-    }
-}
-
-inline void rk4_rhs_accum(std::size_t n,
-                          const double *__restrict f1,
-                          double *__restrict frhs) {
-    std::size_t i = 0;
-#if defined(__AVX512F__)
-    const __m512d v2 = _mm512_set1_pd(2.0);
-    for (; i + 7 < n; i += 8) {
-        const __m512d v1 = _mm512_loadu_pd(f1 + i);
-        const __m512d vrhs = _mm512_loadu_pd(frhs + i);
-        _mm512_storeu_pd(frhs + i, _mm512_fmadd_pd(v2, v1, vrhs));
-    }
-#elif defined(__AVX2__)
-    const __m256d v2 = _mm256_set1_pd(2.0);
-    for (; i + 3 < n; i += 4) {
-        const __m256d v1 = _mm256_loadu_pd(f1 + i);
-        const __m256d vrhs = _mm256_loadu_pd(frhs + i);
-        _mm256_storeu_pd(frhs + i, _mm256_fmadd_pd(v2, v1, vrhs));
-    }
-#endif
-#pragma ivdep
-    for (; i < n; ++i) {
-        frhs[i] = frhs[i] + 2.0 * f1[i];
-    }
-}
-
-inline void rk4_f1_from_f0_f1(std::size_t n,
-                              const double *__restrict f0,
-                              double *__restrict f1,
-                              double c) {
-    std::size_t i = 0;
-#if defined(__AVX512F__)
-    const __m512d vc = _mm512_set1_pd(c);
-    for (; i + 7 < n; i += 8) {
-        const __m512d v0 = _mm512_loadu_pd(f0 + i);
-        const __m512d v1 = _mm512_loadu_pd(f1 + i);
-        _mm512_storeu_pd(f1 + i, _mm512_fmadd_pd(vc, v1, v0));
-    }
-#elif defined(__AVX2__)
-    const __m256d vc = _mm256_set1_pd(c);
-    for (; i + 3 < n; i += 4) {
-        const __m256d v0 = _mm256_loadu_pd(f0 + i);
-        const __m256d v1 = _mm256_loadu_pd(f1 + i);
-        _mm256_storeu_pd(f1 + i, _mm256_fmadd_pd(vc, v1, v0));
-    }
-#endif
-#pragma ivdep
-    for (; i < n; ++i) {
-        f1[i] = f0[i] + c * f1[i];
-    }
-}
-
-inline void rk4_stage3(std::size_t n,
-                       const double *__restrict f0,
-                       double *__restrict f1,
-                       const double *__restrict frhs,
-                       double c) {
-    std::size_t i = 0;
-#if defined(__AVX512F__)
-    const __m512d vc = _mm512_set1_pd(c);
-    for (; i + 7 < n; i += 8) {
-        const __m512d v0 = _mm512_loadu_pd(f0 + i);
-        const __m512d v1 = _mm512_loadu_pd(f1 + i);
-        const __m512d vr = _mm512_loadu_pd(frhs + i);
-        _mm512_storeu_pd(f1 + i, _mm512_fmadd_pd(vc, _mm512_add_pd(v1, vr), v0));
-    }
-#elif defined(__AVX2__)
-    const __m256d vc = _mm256_set1_pd(c);
-    for (; i + 3 < n; i += 4) {
-        const __m256d v0 = _mm256_loadu_pd(f0 + i);
-        const __m256d v1 = _mm256_loadu_pd(f1 + i);
-        const __m256d vr = _mm256_loadu_pd(frhs + i);
-        _mm256_storeu_pd(f1 + i, _mm256_fmadd_pd(vc, _mm256_add_pd(v1, vr), v0));
-    }
-#endif
-#pragma ivdep
-    for (; i < n; ++i) {
-        f1[i] = f0[i] + c * (f1[i] + frhs[i]);
-    }
-}
-
-} // namespace
-
-extern "C" {
-
-void f_rungekutta4_scalar(double &dT, double &f0, double &f1, double &f_rhs, int &RK4) {
-    constexpr double F1o6 = 1.0 / 6.0;
-    constexpr double HLF = 0.5;
-    constexpr double TWO = 2.0;
-
-    switch (RK4) {
-    case 0:
-        f1 = f0 + HLF * dT * f_rhs;
-        break;
-    case 1:
-        f_rhs = f_rhs + TWO * f1;
-        f1 = f0 + HLF * dT * f1;
-        break;
-    case 2:
-        f_rhs = f_rhs + TWO * f1;
-        f1 = f0 + dT * f1;
-        break;
-    case 3:
-        f1 = f0 + F1o6 * dT * (f1 + f_rhs);
-        break;
-    default:
-        std::fprintf(stderr, "rungekutta4_scalar_c: invalid RK4 stage %d\n", RK4);
-        std::abort();
-    }
-}
-
-void rungekutta4_cplxscalar_(double &dT,
-                             std::complex<double> &f0,
-                             std::complex<double> &f1,
-                             std::complex<double> &f_rhs,
-                             int &RK4) {
-    constexpr double F1o6 = 1.0 / 6.0;
-    constexpr double HLF = 0.5;
-    constexpr double TWO = 2.0;
-
-    switch (RK4) {
-    case 0:
-        f1 = f0 + HLF * dT * f_rhs;
-        break;
-    case 1:
-        f_rhs = f_rhs + TWO * f1;
-        f1 = f0 + HLF * dT * f1;
-        break;
-    case 2:
-        f_rhs = f_rhs + TWO * f1;
-        f1 = f0 + dT * f1;
-        break;
-    case 3:
-        f1 = f0 + F1o6 * dT * (f1 + f_rhs);
-        break;
-    default:
-        std::fprintf(stderr, "rungekutta4_cplxscalar_c: invalid RK4 stage %d\n", RK4);
-        std::abort();
-    }
-}
-
-int f_rungekutta4_rout(int *ex, double &dT,
-                       double *f0, double *f1, double *f_rhs,
-                       int &RK4) {
-    const std::size_t n = static_cast<std::size_t>(ex[0]) *
-                          static_cast<std::size_t>(ex[1]) *
-                          static_cast<std::size_t>(ex[2]);
-    const double *const __restrict f0r = f0;
-    double *const __restrict f1r = f1;
-    double *const __restrict frhs = f_rhs;
-
-    if (__builtin_expect(static_cast<unsigned>(RK4) > 3u, 0)) {
-        std::fprintf(stderr, "rungekutta4_rout_c: invalid RK4 stage %d\n", RK4);
-        std::abort();
-    }
-
-    switch (RK4) {
-    case 0:
-        rk4_stage0(n, f0r, frhs, f1r, 0.5 * dT);
-        break;
-    case 1:
-        rk4_rhs_accum(n, f1r, frhs);
-        rk4_f1_from_f0_f1(n, f0r, f1r, 0.5 * dT);
-        break;
-    case 2:
-        rk4_rhs_accum(n, f1r, frhs);
-        rk4_f1_from_f0_f1(n, f0r, f1r, dT);
-        break;
-    default:
-        rk4_stage3(n, f0r, f1r, frhs, (1.0 / 6.0) * dT);
-        break;
-    }
-
-    return 0;
-}
-
-} // extern "C"

AMSS_NCKU_source/share_func.h

@@ -1,246 +0,0 @@
-#ifndef SHARE_FUNC_H
-#define SHARE_FUNC_H
-
-#include <stdlib.h>
-#include <stddef.h>
-#include <math.h>
-#include <stdio.h>
-#include <string.h>
-/* 主网格：0-based -> 1D */
-static inline size_t idx_ex(int i0, int j0, int k0, const int ex[3]) {
-    const int ex1 = ex[0], ex2 = ex[1];
-    return (size_t)i0 + (size_t)j0 * (size_t)ex1 + (size_t)k0 * (size_t)ex1 * (size_t)ex2;
-}
-
-/*
- * fh 对应 Fortran: fh(-1:ex1, -1:ex2, -1:ex3)
- * ord=2 => shift=1
- * iF/jF/kF 为 Fortran 索引（可为 -1,0,1..ex）
- */
-static inline size_t idx_fh_F_ord2(int iF, int jF, int kF, const int ex[3]) {
-    const int shift = 1;
-    const int nx = ex[0] + 2;      // ex1 + ord
-    const int ny = ex[1] + 2;
-
-    const int ii = iF + shift;     // 0..ex1+1
-    const int jj = jF + shift;     // 0..ex2+1
-    const int kk = kF + shift;     // 0..ex3+1
-
-    return (size_t)ii + (size_t)jj * (size_t)nx + (size_t)kk * (size_t)nx * (size_t)ny;
-}
-
-/*
- * fh 对应 Fortran: fh(-2:ex1, -2:ex2, -2:ex3)
- * ord=3 => shift=2
- * iF/jF/kF 是 Fortran 索引（可为负）
- */
-static inline size_t idx_fh_F(int iF, int jF, int kF, const int ex[3]) {
-    const int shift = 2;                 // ord=3 -> -2..ex
-    const int nx = ex[0] + 3;            // ex1 + ord
-    const int ny = ex[1] + 3;
-
-    const int ii = iF + shift;           // 0..ex1+2
-    const int jj = jF + shift;           // 0..ex2+2
-    const int kk = kF + shift;           // 0..ex3+2
-
-    return (size_t)ii + (size_t)jj * (size_t)nx + (size_t)kk * (size_t)nx * (size_t)ny;
-}
-
-/*
- * func:  (1..extc1, 1..extc2, 1..extc3)   1-based in Fortran
- * funcc: (-ord+1..extc1, -ord+1..extc2, -ord+1..extc3) in Fortran
- *
- * C 里我们把：
- *   func  视为 0-based: i0=0..extc1-1, j0=0..extc2-1, k0=0..extc3-1
- *   funcc 用“平移下标”存为一维数组：
- *     iF in [-ord+1..extc1]  -> ii = iF + (ord-1)  in [0..extc1+ord-1]
- *     总长度 nx = extc1 + ord
- *     同理 ny = extc2 + ord, nz = extc3 + ord
- */
-
-static inline size_t idx_func0(int i0, int j0, int k0, const int extc[3]) {
-    const int nx = extc[0], ny = extc[1];
-    return (size_t)i0 + (size_t)j0 * (size_t)nx + (size_t)k0 * (size_t)nx * (size_t)ny;
-}
-
-static inline size_t idx_funcc_F(int iF, int jF, int kF, int ord, const int extc[3]) {
-    const int shift = ord - 1;          // iF = -shift .. extc1
-    const int nx = extc[0] + ord;       // [-shift..extc1] 共 extc1+ord 个
-    const int ny = extc[1] + ord;
-
-    const int ii = iF + shift;          // 0..extc1+shift
-    const int jj = jF + shift;          // 0..extc2+shift
-    const int kk = kF + shift;          // 0..extc3+shift
-
-    return (size_t)ii + (size_t)jj * (size_t)nx + (size_t)kk * (size_t)nx * (size_t)ny;
-}
-
-/*
- * 等价于 Fortran:
- * funcc(1:extc1,1:extc2,1:extc3)=func
- * do i=0,ord-1
- *   funcc(-i,1:extc2,1:extc3) = funcc(i+1,1:extc2,1:extc3)*SoA(1)
- * enddo
- * do i=0,ord-1
- *   funcc(:,-i,1:extc3) = funcc(:,i+1,1:extc3)*SoA(2)
- * enddo
- * do i=0,ord-1
- *   funcc(:,:,-i) = funcc(:,:,i+1)*SoA(3)
- * enddo
- */
-static inline void symmetry_bd_impl(int ord,
-                 int shift,
-                 const int extc[3],
-                 const double *__restrict func,
-                 double *__restrict funcc,
-                 const double SoA[3])
-{
-    const int extc1 = extc[0], extc2 = extc[1], extc3 = extc[2];
-    const int nx = extc1 + ord;
-    const int ny = extc2 + ord;
-
-    const size_t snx = (size_t)nx;
-    const size_t splane = (size_t)nx * (size_t)ny;
-    const size_t interior_i = (size_t)shift + 1u;          /* iF = 1 */
-    const size_t interior_j = ((size_t)shift + 1u) * snx;  /* jF = 1 */
-    const size_t interior_k = ((size_t)shift + 1u) * splane; /* kF = 1 */
-    const size_t interior0 = interior_k + interior_j + interior_i;
-
-    /* 1) funcc(1:extc1,1:extc2,1:extc3) = func */
-    for (int k0 = 0; k0 < extc3; ++k0) {
-        const double *src_k = func + (size_t)k0 * (size_t)extc2 * (size_t)extc1;
-        const size_t dst_k0 = interior0 + (size_t)k0 * splane;
-        for (int j0 = 0; j0 < extc2; ++j0) {
-            const double *src = src_k + (size_t)j0 * (size_t)extc1;
-            double *dst = funcc + dst_k0 + (size_t)j0 * snx;
-            memcpy(dst, src, (size_t)extc1 * sizeof(double));
-        }
-    }
-
-    /* 2) funcc(-i,1:extc2,1:extc3) = funcc(i+1,1:extc2,1:extc3)*SoA(1) */
-    const double s1 = SoA[0];
-    if (s1 == 1.0) {
-        for (int ii = 0; ii < ord; ++ii) {
-            const size_t dst_i = (size_t)(shift - ii);
-            const size_t src_i = (size_t)(shift + ii + 1);
-            for (int k0 = 0; k0 < extc3; ++k0) {
-                const size_t kbase = interior_k + (size_t)k0 * splane + interior_j;
-                for (int j0 = 0; j0 < extc2; ++j0) {
-                    const size_t off = kbase + (size_t)j0 * snx;
-                    funcc[off + dst_i] = funcc[off + src_i];
-                }
-            }
-        }
-    } else if (s1 == -1.0) {
-        for (int ii = 0; ii < ord; ++ii) {
-            const size_t dst_i = (size_t)(shift - ii);
-            const size_t src_i = (size_t)(shift + ii + 1);
-            for (int k0 = 0; k0 < extc3; ++k0) {
-                const size_t kbase = interior_k + (size_t)k0 * splane + interior_j;
-                for (int j0 = 0; j0 < extc2; ++j0) {
-                    const size_t off = kbase + (size_t)j0 * snx;
-                    funcc[off + dst_i] = -funcc[off + src_i];
-                }
-            }
-        }
-    } else {
-        for (int ii = 0; ii < ord; ++ii) {
-            const size_t dst_i = (size_t)(shift - ii);
-            const size_t src_i = (size_t)(shift + ii + 1);
-            for (int k0 = 0; k0 < extc3; ++k0) {
-                const size_t kbase = interior_k + (size_t)k0 * splane + interior_j;
-                for (int j0 = 0; j0 < extc2; ++j0) {
-                    const size_t off = kbase + (size_t)j0 * snx;
-                    funcc[off + dst_i] = funcc[off + src_i] * s1;
-                }
-            }
-        }
-    }
-
-    /* 3) funcc(:,-j,1:extc3) = funcc(:,j+1,1:extc3)*SoA(2) */
-    const double s2 = SoA[1];
-    if (s2 == 1.0) {
-        for (int jj = 0; jj < ord; ++jj) {
-            const size_t dst_j = (size_t)(shift - jj) * snx;
-            const size_t src_j = (size_t)(shift + jj + 1) * snx;
-            for (int k0 = 0; k0 < extc3; ++k0) {
-                const size_t kbase = interior_k + (size_t)k0 * splane;
-                double *dst = funcc + kbase + dst_j;
-                const double *src = funcc + kbase + src_j;
-                for (int i = 0; i < nx; ++i) dst[i] = src[i];
-            }
-        }
-    } else if (s2 == -1.0) {
-        for (int jj = 0; jj < ord; ++jj) {
-            const size_t dst_j = (size_t)(shift - jj) * snx;
-            const size_t src_j = (size_t)(shift + jj + 1) * snx;
-            for (int k0 = 0; k0 < extc3; ++k0) {
-                const size_t kbase = interior_k + (size_t)k0 * splane;
-                double *dst = funcc + kbase + dst_j;
-                const double *src = funcc + kbase + src_j;
-                for (int i = 0; i < nx; ++i) dst[i] = -src[i];
-            }
-        }
-    } else {
-        for (int jj = 0; jj < ord; ++jj) {
-            const size_t dst_j = (size_t)(shift - jj) * snx;
-            const size_t src_j = (size_t)(shift + jj + 1) * snx;
-            for (int k0 = 0; k0 < extc3; ++k0) {
-                const size_t kbase = interior_k + (size_t)k0 * splane;
-                double *dst = funcc + kbase + dst_j;
-                const double *src = funcc + kbase + src_j;
-                for (int i = 0; i < nx; ++i) dst[i] = src[i] * s2;
-            }
-        }
-    }
-
-    /* 4) funcc(:,:,-k) = funcc(:,:,k+1)*SoA(3) */
-    const double s3 = SoA[2];
-    if (s3 == 1.0) {
-        for (int kk = 0; kk < ord; ++kk) {
-            const size_t dst_k = (size_t)(shift - kk) * splane;
-            const size_t src_k = (size_t)(shift + kk + 1) * splane;
-            double *dst = funcc + dst_k;
-            const double *src = funcc + src_k;
-            for (size_t p = 0; p < splane; ++p) dst[p] = src[p];
-        }
-    } else if (s3 == -1.0) {
-        for (int kk = 0; kk < ord; ++kk) {
-            const size_t dst_k = (size_t)(shift - kk) * splane;
-            const size_t src_k = (size_t)(shift + kk + 1) * splane;
-            double *dst = funcc + dst_k;
-            const double *src = funcc + src_k;
-            for (size_t p = 0; p < splane; ++p) dst[p] = -src[p];
-        }
-    } else {
-        for (int kk = 0; kk < ord; ++kk) {
-            const size_t dst_k = (size_t)(shift - kk) * splane;
-            const size_t src_k = (size_t)(shift + kk + 1) * splane;
-            double *dst = funcc + dst_k;
-            const double *src = funcc + src_k;
-            for (size_t p = 0; p < splane; ++p) dst[p] = src[p] * s3;
-        }
-    }
-}
-
-static inline void symmetry_bd(int ord,
-                 const int extc[3],
-                 const double *func,
-                 double *funcc,
-                 const double SoA[3])
-{
-    if (ord <= 0) return;
-
-    /* Fast paths used by current C kernels: ord=2 (derivs), ord=3 (lopsided/KO). */
-    if (ord == 2) {
-        symmetry_bd_impl(2, 1, extc, func, funcc, SoA);
-        return;
-    }
-    if (ord == 3) {
-        symmetry_bd_impl(3, 2, extc, func, funcc, SoA);
-        return;
-    }
-
-    symmetry_bd_impl(ord, ord - 1, extc, func, funcc, SoA);
-}
-#endif

AMSS_NCKU_source/surface_integral.C

@@ -11,6 +11,8 @@
 #include <strstream>
 #include <cmath>
 #include <map>
+#include <vector>
+#include <algorithm>
 using namespace std;
 #else
 #include <iostream.h>
@@ -182,7 +184,6 @@ surface_integral::surface_integral(int iSymmetry) : Symmetry(iSymmetry)
 //|============================================================================
 surface_integral::~surface_integral()
 {
-  release_cached_buffers();
   delete[] nx_g;
   delete[] ny_g;
   delete[] nz_g;
@@ -191,50 +192,6 @@ surface_integral::~surface_integral()
   delete[] wtcostheta;
 #endif
 }
-
-void surface_integral::get_surface_points(double rex, double **pox)
-{
-  SpherePointCache &cache = sphere_point_cache[rex];
-  if (!cache.pox[0])
-  {
-    for (int i = 0; i < 3; ++i)
-      cache.pox[i] = new double[n_tot];
-    for (int n = 0; n < n_tot; ++n)
-    {
-      cache.pox[0][n] = rex * nx_g[n];
-      cache.pox[1][n] = rex * ny_g[n];
-      cache.pox[2][n] = rex * nz_g[n];
-    }
-  }
-
-  pox[0] = cache.pox[0];
-  pox[1] = cache.pox[1];
-  pox[2] = cache.pox[2];
-}
-
-double *surface_integral::get_shellf_buffer(int num_var)
-{
-  double *&buffer = shellf_cache[num_var];
-  if (!buffer)
-    buffer = new double[n_tot * num_var];
-  return buffer;
-}
-
-void surface_integral::release_cached_buffers()
-{
-  for (map<double, SpherePointCache>::iterator it = sphere_point_cache.begin(); it != sphere_point_cache.end(); ++it)
-  {
-    delete[] it->second.pox[0];
-    delete[] it->second.pox[1];
-    delete[] it->second.pox[2];
-    it->second.pox[0] = it->second.pox[1] = it->second.pox[2] = 0;
-  }
-  sphere_point_cache.clear();
-
-  for (map<int, double *>::iterator it = shellf_cache.begin(); it != shellf_cache.end(); ++it)
-    delete[] it->second;
-  shellf_cache.clear();
-}
 //|----------------------------------------------------------------
 //  spin weighted spinw component of psi4, general routine
 //  l takes from spinw to maxl; m takes from -l to l
@@ -256,7 +213,14 @@ void surface_integral::surf_Wave(double rex, int lev, cgh *GH, var *Rpsi4, var *
 
   int n;
   double *pox[3];
-  get_surface_points(rex, pox);
+  for (int i = 0; i < 3; i++)
+    pox[i] = new double[n_tot];
+  for (n = 0; n < n_tot; n++)
+  {
+    pox[0][n] = rex * nx_g[n];
+    pox[1][n] = rex * ny_g[n];
+    pox[2][n] = rex * nz_g[n];
+  }
 
   int mp, Lp, Nmin, Nmax;
   mp = n_tot / cpusize;
@@ -272,12 +236,10 @@ void surface_integral::surf_Wave(double rex, int lev, cgh *GH, var *Rpsi4, var *
     Nmax = Nmin + mp - 1;
   }
 
-  double *shellf = get_shellf_buffer(InList);
+  double *shellf;
+  shellf = new double[n_tot * InList];
 
   GH->PatL[lev]->data->Interp_Points(DG_List, n_tot, pox, shellf, Symmetry, Nmin, Nmax);
-
-  //|~~~~~> Integrate the dot product of Dphi with the surface normal.
-
   double *RP_out, *IP_out;
   RP_out = new double[NN];
   IP_out = new double[NN];
@@ -412,6 +374,10 @@ void surface_integral::surf_Wave(double rex, int lev, cgh *GH, var *Rpsi4, var *
 
   //|------= Free memory.
 
+  delete[] pox[0];
+  delete[] pox[1];
+  delete[] pox[2];
+  delete[] shellf;
   delete[] RP_out;
   delete[] IP_out;
   DG_List->clearList();
@@ -437,9 +403,17 @@ void surface_integral::surf_Wave(double rex, int lev, cgh *GH, var *Rpsi4, var *
 
   int n;
   double *pox[3];
-  get_surface_points(rex, pox);
+  for (int i = 0; i < 3; i++)
+    pox[i] = new double[n_tot];
+  for (n = 0; n < n_tot; n++)
+  {
+    pox[0][n] = rex * nx_g[n];
+    pox[1][n] = rex * ny_g[n];
+    pox[2][n] = rex * nz_g[n];
+  }
 
-  double *shellf = get_shellf_buffer(InList);
+  double *shellf;
+  shellf = new double[n_tot * InList];
 
   //    misc::tillherecheck(GH->Commlev[lev],GH->start_rank[lev],"before Interp_Points");
 
@@ -602,6 +576,10 @@ void surface_integral::surf_Wave(double rex, int lev, cgh *GH, var *Rpsi4, var *
 
   //|------= Free memory.
 
+  delete[] pox[0];
+  delete[] pox[1];
+  delete[] pox[2];
+  delete[] shellf;
   delete[] RP_out;
   delete[] IP_out;
   DG_List->clearList();
@@ -620,9 +598,17 @@ void surface_integral::surf_Wave(double rex, int lev, ShellPatch *GH, var *Rpsi4
 
   int n;
   double *pox[3];
-  get_surface_points(rex, pox);
+  for (int i = 0; i < 3; i++)
+    pox[i] = new double[n_tot];
+  for (n = 0; n < n_tot; n++)
+  {
+    pox[0][n] = rex * nx_g[n];
+    pox[1][n] = rex * ny_g[n];
+    pox[2][n] = rex * nz_g[n];
+  }
 
-  double *shellf = get_shellf_buffer(InList);
+  double *shellf;
+  shellf = new double[n_tot * InList];
 
   GH->Interp_Points(DG_List, n_tot, pox, shellf, Symmetry);
 
@@ -2583,6 +2569,10 @@ void surface_integral::surf_MassPAng(double rex, int lev, cgh *GH, var *chi, var
   Rout[5] = sy;
   Rout[6] = sz;
 
+  delete[] pox[0];
+  delete[] pox[1];
+  delete[] pox[2];
+  delete[] shellf;
   DG_List->clearList();
 }
 void surface_integral::surf_MassPAng(double rex, int lev, cgh *GH, var *chi, var *trK,
@@ -2648,9 +2638,17 @@ void surface_integral::surf_MassPAng(double rex, int lev, cgh *GH, var *chi, var
 
   int n;
   double *pox[3];
-  get_surface_points(rex, pox);
+  for (int i = 0; i < 3; i++)
+    pox[i] = new double[n_tot];
+  for (n = 0; n < n_tot; n++)
+  {
+    pox[0][n] = rex * nx_g[n];
+    pox[1][n] = rex * ny_g[n];
+    pox[2][n] = rex * nz_g[n];
+  }
 
-  double *shellf = get_shellf_buffer(InList);
+  double *shellf;
+  shellf = new double[n_tot * InList];
 
   // we have assumed there is only one box on this level,
   // so we do not need loop boxes
@@ -2840,6 +2838,10 @@ void surface_integral::surf_MassPAng(double rex, int lev, cgh *GH, var *chi, var
   Rout[5] = sy;
   Rout[6] = sz;
 
+  delete[] pox[0];
+  delete[] pox[1];
+  delete[] pox[2];
+  delete[] shellf;
   DG_List->clearList();
 }
 //|----------------------------------------------------------------

AMSS_NCKU_source/surface_integral.h

@@ -23,21 +23,11 @@ using namespace std;
 #include "NullShellPatch2.h"
 #include "var.h"
 #include "monitor.h"
-#include <map>
 
 class surface_integral
 {
 
 private:
-	struct SpherePointCache
-	{
-		double *pox[3];
-		SpherePointCache()
-		{
-			pox[0] = pox[1] = pox[2] = 0;
-		}
-	};
-
 	int Symmetry, factor;
 	int N_theta, N_phi; // Number of points in Theta & Phi directions
 	double dphi, dcostheta;
@@ -46,12 +36,6 @@ private:
 
 	double *nx_g, *ny_g, *nz_g; // global list of unit normals
 	int myrank, cpusize;
-	map<double, SpherePointCache> sphere_point_cache;
-	map<int, double *> shellf_cache;
-
-	void get_surface_points(double rex, double **pox);
-	double *get_shellf_buffer(int num_var);
-	void release_cached_buffers();
 
 public:
 	surface_integral(int iSymmetry);

AMSS_NCKU_source/tool.h

@@ -1,33 +0,0 @@
-#include "share_func.h"
-void fdderivs(const int ex[3],
-              const double *f,
-              double *fxx, double *fxy, double *fxz,
-              double *fyy, double *fyz, double *fzz,
-              const double *X, const double *Y, const double *Z,
-              double SYM1, double SYM2, double SYM3,
-              int Symmetry, int onoff);
-
-void fderivs(const int ex[3],
-             const double *f,
-             double *fx, double *fy, double *fz,
-             const double *X, const double *Y, const double *Z,
-             double SYM1, double SYM2, double SYM3,
-             int Symmetry, int onoff);
-
-void kodis(const int ex[3],
-           const double *X, const double *Y, const double *Z,
-           const double *f, double *f_rhs,
-           const double SoA[3],
-           int Symmetry, double eps);
-
-void lopsided(const int ex[3],
-              const double *X, const double *Y, const double *Z,
-              const double *f, double *f_rhs,
-              const double *Sfx, const double *Sfy, const double *Sfz,
-              int Symmetry, const double SoA[3]);
-
-void lopsided_kodis(const int ex[3],
-                    const double *X, const double *Y, const double *Z,
-                    const double *f, double *f_rhs,
-                    const double *Sfx, const double *Sfy, const double *Sfz,
-                    int Symmetry, const double SoA[3], double eps);

generate_interp_lb_header.py

@@ -1,72 +0,0 @@
-#!/usr/bin/env python3
-"""Convert interp_lb_profile.bin to a C header for compile-time embedding."""
-import struct, sys
-
-if len(sys.argv) < 3:
-    print(f"Usage: {sys.argv[0]} <profile.bin> <output.h>")
-    sys.exit(1)
-
-with open(sys.argv[1], 'rb') as f:
-    magic, version, nprocs, num_heavy = struct.unpack('IIii', f.read(16))
-    threshold = struct.unpack('d', f.read(8))[0]
-    times = list(struct.unpack(f'{nprocs}d', f.read(nprocs * 8)))
-    heavy = list(struct.unpack(f'{num_heavy}i', f.read(num_heavy * 4)))
-
-# For each heavy rank, compute split: left half -> lighter neighbor, right half -> heavy rank
-# (or vice versa depending on which neighbor is lighter)
-splits = []
-for hr in heavy:
-    prev_t = times[hr - 1] if hr > 0 else 1e30
-    next_t = times[hr + 1] if hr < nprocs - 1 else 1e30
-    if prev_t <= next_t:
-        splits.append((hr, hr - 1, hr))  # (block_id, r_left, r_right)
-    else:
-        splits.append((hr, hr, hr + 1))
-
-# Also remap the displaced neighbor blocks
-remaps = {}
-for hr, r_l, r_r in splits:
-    if r_l != hr:
-        # We took r_l's slot, so remap block r_l to its other neighbor
-        displaced = r_l
-        if displaced > 0 and displaced - 1 not in [s[0] for s in splits]:
-            remaps[displaced] = displaced - 1
-        elif displaced < nprocs - 1:
-            remaps[displaced] = displaced + 1
-    else:
-        displaced = r_r
-        if displaced < nprocs - 1 and displaced + 1 not in [s[0] for s in splits]:
-            remaps[displaced] = displaced + 1
-        elif displaced > 0:
-            remaps[displaced] = displaced - 1
-
-with open(sys.argv[2], 'w') as out:
-    out.write("/* Auto-generated from interp_lb_profile.bin — do not edit */\n")
-    out.write("#ifndef INTERP_LB_PROFILE_DATA_H\n")
-    out.write("#define INTERP_LB_PROFILE_DATA_H\n\n")
-    out.write(f"#define INTERP_LB_NPROCS {nprocs}\n")
-    out.write(f"#define INTERP_LB_NUM_HEAVY {num_heavy}\n\n")
-    out.write(f"static const int interp_lb_heavy_blocks[{num_heavy}] = {{")
-    out.write(", ".join(str(h) for h in heavy))
-    out.write("};\n\n")
-    out.write("/* Split table: {block_id, r_left, r_right} */\n")
-    out.write(f"static const int interp_lb_splits[{num_heavy}][3] = {{\n")
-    for bid, rl, rr in splits:
-        out.write(f"    {{{bid}, {rl}, {rr}}},\n")
-    out.write("};\n\n")
-    out.write("/* Rank remap for displaced neighbor blocks */\n")
-    out.write(f"static const int interp_lb_num_remaps = {len(remaps)};\n")
-    out.write(f"static const int interp_lb_remaps[][2] = {{\n")
-    for src, dst in sorted(remaps.items()):
-        out.write(f"    {{{src}, {dst}}},\n")
-    if not remaps:
-        out.write("    {-1, -1},\n")
-    out.write("};\n\n")
-    out.write("#endif /* INTERP_LB_PROFILE_DATA_H */\n")
-
-print(f"Generated {sys.argv[2]}:")
-print(f"  {num_heavy} heavy blocks to split: {heavy}")
-for bid, rl, rr in splits:
-    print(f"    block {bid}: split -> rank {rl} (left), rank {rr} (right)")
-for src, dst in sorted(remaps.items()):
-    print(f"    block {src}: remap -> rank {dst}")

makefile_and_run.py

@@ -9,93 +9,20 @@
 
 
 import AMSS_NCKU_Input as input_data
-import os
 import subprocess
 import time
+## 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 0-111"
+NUMACTL_CPU_BIND = "taskset -c 16-47,64-95"
 
+## 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 = 96
 
-def get_last_n_cores_per_socket(n=32):
-    """
-    Read CPU topology via lscpu and return a taskset -c string
-    selecting the last `n` cores of each NUMA node (socket).
-
-    Example: 2 sockets x 56 cores each, n=32 -> node0: 24-55, node1: 80-111
-    -> "taskset -c 24-55,80-111"
-    """
-    result = subprocess.run(["lscpu", "--parse=NODE,CPU"], capture_output=True, text=True)
-
-    # Build a dict: node_id -> sorted list of CPU ids
-    node_cpus = {}
-    for line in result.stdout.splitlines():
-        if line.startswith("#") or not line.strip():
-            continue
-        parts = line.split(",")
-        if len(parts) < 2:
-            continue
-        node_id, cpu_id = int(parts[0]), int(parts[1])
-        node_cpus.setdefault(node_id, []).append(cpu_id)
-
-    segments = []
-    for node_id in sorted(node_cpus):
-        cpus = sorted(node_cpus[node_id])
-        selected = cpus[-n:]          # last n cores of this socket
-        segments.append(f"{selected[0]}-{selected[-1]}")
-
-    cpu_str = ",".join(segments)
-    total = len(segments) * n
-    print(f" CPU binding: taskset -c {cpu_str}  ({total} cores, last {n} per socket)")
-    #return f"taskset -c {cpu_str}"
-    return f""
-
-
-## CPU core binding: dynamically select the last 32 cores of each socket (64 cores total)
-NUMACTL_CPU_BIND = get_last_n_cores_per_socket(n=32)
-
-## Build parallelism: match the number of bound cores
-BUILD_JOBS = 64
-
-
-##################################################################
-
-
-##################################################################
-
-def prepare_gpu_runtime_env():
-    """
-    Create a user-private CUDA MPS environment for GPU runs.
-
-    On shared machines another user's daemon may already occupy the default
-    /tmp/nvidia-mps pipe directory, which makes plain cudaSetDevice/cudaMalloc
-    fail with cudaErrorMpsConnectionFailed.  Binding AMSS-NCKU to a private
-    pipe directory avoids cross-user interference.
-    """
-    env = os.environ.copy()
-
-    pipe_dir = env.get("CUDA_MPS_PIPE_DIRECTORY", f"/tmp/amss-ncku-mps-{os.getuid()}")
-    log_dir  = env.get("CUDA_MPS_LOG_DIRECTORY",  f"/tmp/amss-ncku-mps-log-{os.getuid()}")
-
-    os.makedirs(pipe_dir, exist_ok=True)
-    os.makedirs(log_dir, exist_ok=True)
-
-    env["CUDA_MPS_PIPE_DIRECTORY"] = pipe_dir
-    env["CUDA_MPS_LOG_DIRECTORY"] = log_dir
-
-    control_socket = os.path.join(pipe_dir, "control")
-    if not os.path.exists(control_socket):
-        start = subprocess.run(
-            ["nvidia-cuda-mps-control", "-d"],
-            env=env,
-            stdout=subprocess.DEVNULL,
-            stderr=subprocess.DEVNULL,
-        )
-        if start.returncode != 0:
-            print(f" Warning: failed to start private CUDA MPS daemon in {pipe_dir}")
-        else:
-            print(f" Using private CUDA MPS pipe directory: {pipe_dir}")
-    else:
-        print(f" Using existing private CUDA MPS pipe directory: {pipe_dir}")
-
-    return env
 
 ##################################################################
 
@@ -113,7 +40,7 @@ def makefile_ABE():
 
     ## Build command with CPU binding to nohz_full cores
     if (input_data.GPU_Calculation == "no"):
-        makefile_command  = f"{NUMACTL_CPU_BIND} make -j{BUILD_JOBS} INTERP_LB_MODE=off ABE"
+        makefile_command  = f"{NUMACTL_CPU_BIND} make -j{BUILD_JOBS} ABE"
     elif (input_data.GPU_Calculation == "yes"):
         makefile_command  = f"{NUMACTL_CPU_BIND} make -j{BUILD_JOBS} ABEGPU"
     else:
@@ -189,29 +116,16 @@ def run_ABE():
 
     ## Define the command to run; cast other values to strings as needed
     
-    run_env = None
-
     if (input_data.GPU_Calculation == "no"):
         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"):
-        run_env = prepare_gpu_runtime_env()
+        mpi_command         = NUMACTL_CPU_BIND + " mpirun -np " + str(input_data.MPI_processes) + " ./ABEGPU"
-        if int(input_data.MPI_processes) == 1:
-            mpi_command = "./ABEGPU"
-        else:
-            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
-    mpi_process = subprocess.Popen(
+    mpi_process = subprocess.Popen(mpi_command, shell=True, stdout=subprocess.PIPE, stderr=subprocess.STDOUT, text=True)
-        mpi_command,
-        shell=True,
-        stdout=subprocess.PIPE,
-        stderr=subprocess.STDOUT,
-        text=True,
-        env=run_env,
-    )
 
     ## Write ABE run output to file while printing to stdout
     with open(mpi_command_outfile, 'w') as file0:  

pgo_profile/PGO_Profile_Analysis.md

@@ -0,0 +1,97 @@
+# AMSS-NCKU PGO Profile Analysis Report
+
+## 1. Profiling Environment
+
+| Item | Value |
+|------|-------|
+| Compiler | Intel oneAPI DPC++/C++ 2025.3.0 (icpx/ifx) |
+| Instrumentation Flag | `-fprofile-instr-generate` |
+| Optimization Level (instrumented) | `-O2 -xHost -fma` |
+| MPI Processes | 1 (single process to avoid MPI+instrumentation deadlock) |
+| Profile File | `default_9725750769337483397_0.profraw` (327 KB) |
+| Merged Profile | `default.profdata` (394 KB) |
+| llvm-profdata | `/home/intel/oneapi/compiler/2025.3/bin/compiler/llvm-profdata` |
+
+## 2. Reduced Simulation Parameters (for profiling run)
+
+| Parameter | Production Value | Profiling Value |
+|-----------|-----------------|-----------------|
+| MPI_processes | 64 | 1 |
+| grid_level | 9 | 4 |
+| static_grid_level | 5 | 3 |
+| static_grid_number | 96 | 24 |
+| moving_grid_number | 48 | 16 |
+| largest_box_xyz_max | 320^3 | 160^3 |
+| Final_Evolution_Time | 1000.0 | 10.0 |
+| Evolution_Step_Number | 10,000,000 | 1,000 |
+| Detector_Number | 12 | 2 |
+
+## 3. Profile Summary
+
+| Metric | Value |
+|--------|-------|
+| Total instrumented functions | 1,392 |
+| Functions with non-zero counts | 117 (8.4%) |
+| Functions with zero counts | 1,275 (91.6%) |
+| Maximum function entry count | 386,459,248 |
+| Maximum internal block count | 370,477,680 |
+| Total block count | 4,198,023,118 |
+
+## 4. Top 20 Hotspot Functions
+
+| Rank | Total Count | Max Block Count | Function | Category |
+|------|------------|-----------------|----------|----------|
+| 1 | 1,241,601,732 | 370,477,680 | `polint_` | Interpolation |
+| 2 | 755,994,435 | 230,156,640 | `prolong3_` | Grid prolongation |
+| 3 | 667,964,095 | 3,697,792 | `compute_rhs_bssn_` | BSSN RHS evolution |
+| 4 | 539,736,051 | 386,459,248 | `symmetry_bd_` | Symmetry boundary |
+| 5 | 277,310,808 | 53,170,728 | `lopsided_` | Lopsided FD stencil |
+| 6 | 155,534,488 | 94,535,040 | `decide3d_` | 3D grid decision |
+| 7 | 119,267,712 | 19,266,048 | `rungekutta4_rout_` | RK4 time integrator |
+| 8 | 91,574,616 | 48,824,160 | `kodis_` | Kreiss-Oliger dissipation |
+| 9 | 67,555,389 | 43,243,680 | `fderivs_` | Finite differences |
+| 10 | 55,296,000 | 42,246,144 | `misc::fact(int)` | Factorial utility |
+| 11 | 43,191,071 | 27,663,328 | `fdderivs_` | 2nd-order FD derivatives |
+| 12 | 36,233,965 | 22,429,440 | `restrict3_` | Grid restriction |
+| 13 | 24,698,512 | 17,231,520 | `polin3_` | Polynomial interpolation |
+| 14 | 22,962,942 | 20,968,768 | `copy_` | Data copy |
+| 15 | 20,135,696 | 17,259,168 | `Ansorg::barycentric(...)` | Spectral interpolation |
+| 16 | 14,650,224 | 7,224,768 | `Ansorg::barycentric_omega(...)` | Spectral weights |
+| 17 | 13,242,296 | 2,871,920 | `global_interp_` | Global interpolation |
+| 18 | 12,672,000 | 7,734,528 | `sommerfeld_rout_` | Sommerfeld boundary |
+| 19 | 6,872,832 | 1,880,064 | `sommerfeld_routbam_` | Sommerfeld boundary (BAM) |
+| 20 | 5,709,900 | 2,809,632 | `l2normhelper_` | L2 norm computation |
+
+## 5. Hotspot Category Breakdown
+
+Top 20 functions account for ~98% of total execution counts:
+
+| Category | Functions | Combined Count | Share |
+|----------|-----------|---------------|-------|
+| Interpolation / Prolongation / Restriction | polint_, prolong3_, restrict3_, polin3_, global_interp_, Ansorg::* | ~2,093M | ~50% |
+| BSSN RHS + FD stencils | compute_rhs_bssn_, lopsided_, fderivs_, fdderivs_ | ~1,056M | ~25% |
+| Boundary conditions | symmetry_bd_, sommerfeld_rout_, sommerfeld_routbam_ | ~559M | ~13% |
+| Time integration | rungekutta4_rout_ | ~119M | ~3% |
+| Dissipation | kodis_ | ~92M | ~2% |
+| Utilities | misc::fact, decide3d_, copy_, l2normhelper_ | ~256M | ~6% |
+
+## 6. Conclusions
+
+1. **Profile data is valid**: 1,392 functions instrumented, 117 exercised with ~4.2 billion total counts.
+2. **Hotspot concentration is high**: Top 5 functions alone account for ~76% of all counts, which is ideal for PGO — the compiler has strong branch/layout optimization targets.
+3. **Fortran numerical kernels dominate**: `polint_`, `prolong3_`, `compute_rhs_bssn_`, `symmetry_bd_`, `lopsided_` are all Fortran routines in the inner evolution loop. PGO will optimize their branch prediction and basic block layout.
+4. **91.6% of functions have zero counts**: These are code paths for unused features (GPU, BSSN-EScalar, BSSN-EM, Z4C, etc.). PGO will deprioritize them, improving instruction cache utilization.
+5. **Profile is representative**: Despite the reduced grid size, the code path coverage matches production — the same kernels (RHS, prolongation, restriction, boundary) are exercised. PGO branch probabilities from this profile will transfer well to full-scale runs.
+
+## 7. PGO Phase 2 Usage
+
+To apply the profile, use the following flags in `makefile.inc`:
+
+```makefile
+CXXAPPFLAGS = -O3 -xHost -fp-model fast=2 -fma -ipo \
+              -fprofile-instr-use=/home/amss/AMSS-NCKU/pgo_profile/default.profdata \
+              -Dfortran3 -Dnewc -I${MKLROOT}/include
+f90appflags = -O3 -xHost -fp-model fast=2 -fma -ipo \
+              -fprofile-instr-use=/home/amss/AMSS-NCKU/pgo_profile/default.profdata \
+              -align array64byte -fpp -I${MKLROOT}/include
+```