Refine GPU runtime controls and input checker

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

AMSS_NCKU_GPUCheck.py

@@ -0,0 +1,559 @@
+#!/usr/bin/env python3
+#
+# Current most stable GPU-branch baseline:
+#   GPU_Calculation="yes"
+#   Equation_Class="BSSN"
+#   Initial_Data_Method="Ansorg-TwoPuncture"
+#   puncture_data_set="Manually"
+#   basic_grid_set="Patch"
+#   grid_center_set="Cell"
+#   Symmetry="equatorial-symmetry"
+#   Time_Evolution_Method="runge-kutta-45"
+#   Finite_Diffenence_Method="4th-order"
+#   boundary_choice="BAM-choice"
+#   gauge_choice=0
+#   tetrad_type=2
+#   AHF_Find="no"
+#   devide_factor=2.0
+#   static_grid_type="Linear"
+#   moving_grid_type="Linear"
+#   AMSS_Z4C_MRBD=0
+#   Do not enable AMSS_CUDA_BH_INTERP_RESIDENT unless a dedicated
+#   CPU/GPU trajectory comparison has been run for that configuration.
+"""
+Check whether AMSS_NCKU_Input.py is suitable for the current GPU branch.
+
+Usage:
+  python3 AMSS_NCKU_GPUCheck.py
+  python3 AMSS_NCKU_GPUCheck.py -f /path/to/AMSS_NCKU_Input.py
+"""
+
+from __future__ import annotations
+
+import argparse
+import importlib.util
+import os
+from dataclasses import dataclass, field
+from pathlib import Path
+from typing import Any, Iterable, List, Sequence
+
+
+SUPPORTED_EQUATIONS = {"BSSN", "BSSN-EScalar", "BSSN-EM", "Z4C"}
+SUPPORTED_INITIAL_DATA = {
+    "Ansorg-TwoPuncture",
+    "Lousto-Analytical",
+    "Cao-Analytical",
+    "KerrSchild-Analytical",
+}
+SUPPORTED_SYMMETRIES = {
+    "no-symmetry",
+    "equatorial-symmetry",
+    "octant-symmetry",
+}
+SUPPORTED_GRIDS = {"Patch", "Shell-Patch"}
+SUPPORTED_CENTERS = {"Cell", "Vertex"}
+SUPPORTED_FD = {"2nd-order", "4th-order", "6th-order", "8th-order"}
+SUPPORTED_GAUGES = {0, 1, 2, 3, 4, 5, 6, 7}
+SUPPORTED_TETRADS = {0, 1, 2}
+SUPPORTED_AHF = {"yes", "no"}
+SUPPORTED_BOUNDARIES = {"BAM-choice", "Shibata-choice"}
+SUPPORTED_PUNCTURE_DATA = {"Manually", "Automatically-BBH"}
+
+STABLE_BASELINE = {
+    "GPU_Calculation": "yes",
+    "Equation_Class": "BSSN",
+    "Initial_Data_Method": "Ansorg-TwoPuncture",
+    "puncture_data_set": "Manually",
+    "basic_grid_set": "Patch",
+    "grid_center_set": "Cell",
+    "Symmetry": "equatorial-symmetry",
+    "Time_Evolution_Method": "runge-kutta-45",
+    "Finite_Diffenence_Method": "4th-order",
+    "boundary_choice": "BAM-choice",
+    "gauge_choice": 0,
+    "tetrad_type": 2,
+    "AHF_Find": "no",
+    "devide_factor": 2.0,
+    "static_grid_type": "Linear",
+    "moving_grid_type": "Linear",
+    "AMSS_Z4C_MRBD": 0,
+}
+
+
+@dataclass
+class CheckResult:
+    ok: bool = True
+    warnings: List[str] = field(default_factory=list)
+    risks: List[str] = field(default_factory=list)
+    notes: List[str] = field(default_factory=list)
+
+    def add_warning(self, msg: str) -> None:
+        self.warnings.append(msg)
+
+    def add_risk(self, msg: str) -> None:
+        self.ok = False
+        self.risks.append(msg)
+
+    def add_note(self, msg: str) -> None:
+        self.notes.append(msg)
+
+    def extend_notes(self, messages: Iterable[str]) -> None:
+        self.notes.extend(messages)
+
+
+def load_input_module(path: Path):
+    spec = importlib.util.spec_from_file_location("amss_ncku_input", str(path))
+    if spec is None or spec.loader is None:
+        raise RuntimeError(f"cannot load input module from {path}")
+    module = importlib.util.module_from_spec(spec)
+    spec.loader.exec_module(module)  # type: ignore[union-attr]
+    return module
+
+
+def get_attr(mod: Any, name: str, default: Any = None) -> Any:
+    return getattr(mod, name, default)
+
+
+def as_text(value: Any) -> str:
+    if isinstance(value, str):
+        return value.strip()
+    return str(value).strip()
+
+
+def as_lower_text(value: Any) -> str:
+    return as_text(value).lower()
+
+
+def as_float(value: Any, default: float | None = None) -> float | None:
+    try:
+        return float(value)
+    except (TypeError, ValueError):
+        return default
+
+
+def as_int(value: Any, default: int | None = None) -> int | None:
+    try:
+        return int(value)
+    except (TypeError, ValueError):
+        return default
+
+
+def sequence_len(value: Any) -> int | None:
+    try:
+        return len(value)
+    except TypeError:
+        return None
+
+
+def sequence_values(value: Any) -> List[float] | None:
+    try:
+        return [float(v) for v in value]
+    except (TypeError, ValueError):
+        return None
+
+
+def approx_equal(a: Any, b: float, tol: float = 1.0e-12) -> bool:
+    value = as_float(a)
+    return value is not None and abs(value - b) <= tol
+
+
+def env_truthy(name: str) -> bool:
+    value = os.environ.get(name)
+    return value is not None and value.strip().lower() in {
+        "1",
+        "yes",
+        "y",
+        "true",
+        "on",
+        "enable",
+        "enabled",
+    }
+
+
+def stable_baseline_differences(mod: Any) -> List[str]:
+    diffs = []
+    for name, expected in STABLE_BASELINE.items():
+        if not hasattr(mod, name):
+            continue
+        actual = get_attr(mod, name, None)
+        if isinstance(expected, float):
+            if not approx_equal(actual, expected):
+                diffs.append(f"{name}={actual!r} (stable baseline: {expected!r})")
+        elif actual != expected:
+            diffs.append(f"{name}={actual!r} (stable baseline: {expected!r})")
+    return diffs
+
+
+def add_membership_check(
+    r: CheckResult,
+    name: str,
+    value: Any,
+    supported: Sequence[Any] | set[Any],
+    *,
+    risk_message: str | None = None,
+    note_message: str | None = None,
+) -> None:
+    if value not in supported:
+        r.add_risk(risk_message or f"Unsupported {name}: {value!r}")
+    elif note_message:
+        r.add_note(note_message)
+
+
+def check_positive_int(r: CheckResult, name: str, value: Any) -> None:
+    parsed = as_int(value)
+    if parsed is None or parsed <= 0:
+        r.add_risk(f"{name} must be a positive integer; got {value!r}")
+
+
+def check_nonnegative_number(r: CheckResult, name: str, value: Any) -> None:
+    parsed = as_float(value)
+    if parsed is None or parsed < 0.0:
+        r.add_risk(f"{name} must be a non-negative number; got {value!r}")
+
+
+def check_grid_geometry(r: CheckResult, mod: Any, grid: str) -> None:
+    grid_level = as_int(get_attr(mod, "grid_level", None))
+    static_grid_level = as_int(get_attr(mod, "static_grid_level", None))
+    moving_grid_level = as_int(get_attr(mod, "moving_grid_level", None))
+    refinement_level = as_int(get_attr(mod, "refinement_level", None))
+    analysis_level = as_int(get_attr(mod, "analysis_level", 0))
+
+    for name in (
+        "grid_level",
+        "static_grid_level",
+        "moving_grid_level",
+        "static_grid_number",
+        "moving_grid_number",
+        "quarter_sphere_number",
+    ):
+        check_positive_int(r, name, get_attr(mod, name, None))
+
+    if grid_level is not None and static_grid_level is not None:
+        if static_grid_level > grid_level:
+            r.add_risk("static_grid_level cannot exceed grid_level.")
+        if moving_grid_level is not None and moving_grid_level != grid_level - static_grid_level:
+            r.add_risk(
+                "moving_grid_level should equal grid_level - static_grid_level; "
+                f"got {moving_grid_level}, expected {grid_level - static_grid_level}."
+            )
+    if grid_level is not None:
+        if refinement_level is None or refinement_level < 0 or refinement_level > grid_level:
+            r.add_risk(f"refinement_level must be in [0, grid_level]; got {refinement_level!r}")
+        if analysis_level is None or analysis_level < 0 or analysis_level >= grid_level:
+            r.add_risk(f"analysis_level must be in [0, grid_level); got {analysis_level!r}")
+
+    largest_max = sequence_values(get_attr(mod, "largest_box_xyz_max", None))
+    largest_min = sequence_values(get_attr(mod, "largest_box_xyz_min", None))
+    if largest_max is None or len(largest_max) != 3:
+        r.add_risk("largest_box_xyz_max must contain three numeric values.")
+    elif any(v <= 0.0 for v in largest_max):
+        r.add_risk(f"largest_box_xyz_max values must be positive; got {largest_max!r}")
+    if largest_min is None or len(largest_min) != 3:
+        r.add_risk("largest_box_xyz_min must contain three numeric values.")
+    elif largest_max is not None and len(largest_max) == 3:
+        for idx, (lo, hi) in enumerate(zip(largest_min, largest_max)):
+            if lo >= hi:
+                r.add_risk(
+                    f"largest_box_xyz_min[{idx}] must be smaller than largest_box_xyz_max[{idx}]."
+                )
+
+    if grid == "Shell-Patch" and largest_max is not None and len(largest_max) == 3:
+        if max(largest_max) - min(largest_max) > 1.0e-12:
+            r.add_risk("Shell-Patch requires a cubic largest_box_xyz_max.")
+
+    if not approx_equal(get_attr(mod, "devide_factor", None), 2.0):
+        r.add_risk("devide_factor must remain 2.0; the AMR code documents only this ratio as supported.")
+    if as_text(get_attr(mod, "static_grid_type", "")) != "Linear":
+        r.add_risk("static_grid_type must remain 'Linear'.")
+    if as_text(get_attr(mod, "moving_grid_type", "")) != "Linear":
+        r.add_risk("moving_grid_type must remain 'Linear'.")
+
+    shell_shape = sequence_values(get_attr(mod, "shell_grid_number", None))
+    if grid == "Shell-Patch":
+        if shell_shape is None or len(shell_shape) != 3:
+            r.add_risk("Shell-Patch requires shell_grid_number with three numeric values.")
+        elif any(int(v) <= 0 for v in shell_shape):
+            r.add_risk(f"shell_grid_number values must be positive; got {shell_shape!r}")
+
+
+def check_punctures(r: CheckResult, mod: Any, init: str, puncture_data: str) -> None:
+    puncture_number = as_int(get_attr(mod, "puncture_number", None))
+    if puncture_number is None or puncture_number <= 0:
+        r.add_risk(f"puncture_number must be a positive integer; got {puncture_number!r}")
+        return
+
+    if init == "Ansorg-TwoPuncture" and puncture_number != 2:
+        r.add_warning(
+            "Ansorg-TwoPuncture is validated on the GPU branch mainly for puncture_number=2."
+        )
+    if puncture_data == "Automatically-BBH":
+        r.add_risk("puncture_data_set='Automatically-BBH' is documented as still developing.")
+
+    for name in ("position_BH", "parameter_BH", "dimensionless_spin_BH", "momentum_BH"):
+        value = get_attr(mod, name, None)
+        outer = sequence_len(value)
+        if outer != puncture_number:
+            r.add_risk(f"{name} must have puncture_number rows; got {outer!r}.")
+            continue
+        for idx in range(puncture_number):
+            if sequence_len(value[idx]) != 3:
+                r.add_risk(f"{name}[{idx}] must contain three values.")
+                break
+
+    if init == "Ansorg-TwoPuncture":
+        for name in ("parameter_BH", "position_BH", "momentum_BH"):
+            if get_attr(mod, name, None) is None:
+                r.add_risk(f"Ansorg-TwoPuncture requires {name}.")
+
+
+def check_output_and_time(r: CheckResult, mod: Any) -> None:
+    for name in (
+        "Final_Evolution_Time",
+        "Check_Time",
+        "Dump_Time",
+        "D2_Dump_Time",
+        "Analysis_Time",
+        "Courant_Factor",
+        "Dissipation",
+    ):
+        check_nonnegative_number(r, name, get_attr(mod, name, None))
+    check_positive_int(r, "Evolution_Step_Number", get_attr(mod, "Evolution_Step_Number", None))
+
+    start_time = as_float(get_attr(mod, "Start_Evolution_Time", None))
+    final_time = as_float(get_attr(mod, "Final_Evolution_Time", None))
+    if start_time is None:
+        r.add_risk("Start_Evolution_Time must be numeric.")
+    elif final_time is not None and final_time <= start_time:
+        r.add_risk("Final_Evolution_Time must be greater than Start_Evolution_Time.")
+
+    for name in ("GW_L_max", "GW_M_max", "Detector_Number"):
+        check_positive_int(r, name, get_attr(mod, name, None))
+    detector_min = as_float(get_attr(mod, "Detector_Rmin", None))
+    detector_max = as_float(get_attr(mod, "Detector_Rmax", None))
+    if detector_min is None or detector_min <= 0.0:
+        r.add_risk(f"Detector_Rmin must be positive; got {detector_min!r}")
+    if detector_max is None or detector_max <= 0.0:
+        r.add_risk(f"Detector_Rmax must be positive; got {detector_max!r}")
+    if detector_min is not None and detector_max is not None and detector_max <= detector_min:
+        r.add_risk("Detector_Rmax must be greater than Detector_Rmin.")
+
+
+def check_equation_specific(r: CheckResult, mod: Any, eq: str, grid: str, fd: str) -> None:
+    if eq == "BSSN":
+        r.add_note("Equation_Class=BSSN is the current validated GPU baseline.")
+    elif eq == "BSSN-EScalar":
+        r.add_warning("BSSN-EScalar has a CUDA path, but it is less broadly validated than BSSN.")
+        fr_choice = as_int(get_attr(mod, "FR_Choice", None))
+        if fr_choice not in {1, 2, 3, 4, 5}:
+            r.add_risk(f"FR_Choice must be one of 1..5 for BSSN-EScalar; got {fr_choice!r}")
+        if approx_equal(get_attr(mod, "FR_a2", None), 0.0):
+            r.add_risk("CUDA BSSN-EScalar requires nonzero FR_a2.")
+        elif not approx_equal(get_attr(mod, "FR_a2", None), 3.0):
+            r.add_warning("CUDA BSSN-EScalar now passes FR_a2 to the kernel, but non-3.0 values need CPU/GPU regression.")
+        for name in ("FR_l2", "FR_phi0", "FR_r0", "FR_sigma0"):
+            check_nonnegative_number(r, name, get_attr(mod, name, None))
+    elif eq == "BSSN-EM":
+        r.add_warning(
+            "BSSN-EM is accepted by the build, but this checker cannot certify its physics/output "
+            "without a CPU/GPU regression run."
+        )
+        if fd == "8th-order":
+            r.add_note("BSSN-EM with 8th-order enables extra CUDA AMR batching defaults.")
+    elif eq == "Z4C":
+        r.add_warning(
+            "Z4C has CUDA support, but the resident path and Shell/CPBC combinations are more constrained."
+        )
+        if grid == "Patch":
+            r.add_warning("Z4C+Patch avoids Shell CPBC, but still needs a dedicated regression test.")
+        else:
+            r.add_warning("Z4C+Shell-Patch uses CPBC/Shell logic and is not the stable BSSN baseline.")
+
+
+def check_runtime_environment(r: CheckResult, mod: Any, eq: str, grid: str, fd: str) -> None:
+    if env_truthy("AMSS_CUDA_BH_INTERP_RESIDENT"):
+        r.add_risk(
+            "AMSS_CUDA_BH_INTERP_RESIDENT is enabled in the environment; this option previously caused "
+            "late-time trajectory drift and should stay off unless explicitly revalidated."
+        )
+    else:
+        r.add_note("AMSS_CUDA_BH_INTERP_RESIDENT is not enabled; this matches the fixed stable default.")
+
+    if eq in {"BSSN", "BSSN-EScalar", "Z4C"}:
+        r.add_note("makefile_and_run.py will default AMSS_CUDA_AMR_RESTRICT_DEVICE=1 for this equation.")
+    if fd in {"2nd-order", "8th-order"}:
+        r.add_warning(
+            f"{fd} disables some interpolation/CUDA-aware MPI fast paths by default; validate performance and output."
+        )
+    if grid == "Shell-Patch":
+        r.add_warning(
+            "Shell-Patch changes runtime defaults and MPI process handling; use at least the script-adjusted 4 MPI ranks."
+        )
+
+    z4c_mrbd = as_int(get_attr(mod, "AMSS_Z4C_MRBD", 0), 0)
+    if z4c_mrbd not in {0, 1, 2}:
+        r.add_risk(f"AMSS_Z4C_MRBD must be 0, 1, or 2; got {z4c_mrbd!r}")
+    elif eq == "Z4C" and z4c_mrbd == 2:
+        r.add_risk("Z4C GPU resident path does not support AMSS_Z4C_MRBD=2.")
+    elif eq == "Z4C" and z4c_mrbd in {0, 1}:
+        r.add_note(f"Z4C will build with AMSS_Z4C_MRBD={z4c_mrbd}.")
+
+
+def check_stable_profile(r: CheckResult, mod: Any) -> None:
+    diffs = stable_baseline_differences(mod)
+    if not diffs:
+        r.add_note("This input matches the documented most stable GPU baseline.")
+        return
+    r.add_warning(
+        "This input differs from the documented most stable GPU baseline: " + "; ".join(diffs)
+    )
+
+
+def check_input(mod: Any) -> CheckResult:
+    r = CheckResult()
+
+    gpu_text = as_lower_text(get_attr(mod, "GPU_Calculation", "no"))
+    gpu = gpu_text == "yes"
+    eq = as_text(get_attr(mod, "Equation_Class", ""))
+    init = as_text(get_attr(mod, "Initial_Data_Method", ""))
+    symmetry = as_text(get_attr(mod, "Symmetry", ""))
+    time_method = as_text(get_attr(mod, "Time_Evolution_Method", ""))
+    grid = as_text(get_attr(mod, "basic_grid_set", ""))
+    center = as_text(get_attr(mod, "grid_center_set", ""))
+    fd = as_text(get_attr(mod, "Finite_Diffenence_Method", ""))
+    gauge = get_attr(mod, "gauge_choice", None)
+    tetrad = get_attr(mod, "tetrad_type", None)
+    ahf = as_text(get_attr(mod, "AHF_Find", "no")).lower()
+    boundary = as_text(get_attr(mod, "boundary_choice", ""))
+    puncture_data = as_text(get_attr(mod, "puncture_data_set", ""))
+    cpu_part = get_attr(mod, "CPU_Part", None)
+    gpu_part = get_attr(mod, "GPU_Part", None)
+
+    if gpu_text not in {"yes", "no"}:
+        r.add_risk(f"GPU_Calculation must be 'yes' or 'no'; got {get_attr(mod, 'GPU_Calculation', None)!r}")
+    if not gpu:
+        r.add_note("GPU_Calculation=no; this check only targets the GPU branch.")
+        return r
+
+    r.add_note("GPU_Calculation=yes detected.")
+
+    add_membership_check(r, "Equation_Class", eq, SUPPORTED_EQUATIONS)
+    add_membership_check(r, "Symmetry", symmetry, SUPPORTED_SYMMETRIES)
+    add_membership_check(r, "Initial_Data_Method", init, SUPPORTED_INITIAL_DATA)
+    add_membership_check(r, "basic_grid_set", grid, SUPPORTED_GRIDS)
+    add_membership_check(r, "grid_center_set", center, SUPPORTED_CENTERS)
+    add_membership_check(r, "Finite_Diffenence_Method", fd, SUPPORTED_FD)
+    add_membership_check(r, "gauge_choice", gauge, SUPPORTED_GAUGES)
+    add_membership_check(r, "tetrad_type", tetrad, SUPPORTED_TETRADS)
+    add_membership_check(r, "AHF_Find", ahf, SUPPORTED_AHF)
+    add_membership_check(r, "boundary_choice", boundary, SUPPORTED_BOUNDARIES)
+    add_membership_check(r, "puncture_data_set", puncture_data, SUPPORTED_PUNCTURE_DATA)
+
+    if init != "Ansorg-TwoPuncture":
+        r.add_risk(
+            f"Initial_Data_Method={init!r} is not validated as safe on this GPU branch; "
+            "the stable path is Ansorg-TwoPuncture."
+        )
+    else:
+        r.add_note("Initial_Data_Method=Ansorg-TwoPuncture is supported.")
+
+    if time_method != "runge-kutta-45":
+        r.add_risk(f"Only Time_Evolution_Method='runge-kutta-45' is supported; got {time_method!r}.")
+    if grid == "Patch":
+        r.add_note("basic_grid_set=Patch is the current stable GPU grid path.")
+    elif grid == "Shell-Patch":
+        r.add_warning("basic_grid_set=Shell-Patch has GPU support but is outside the stable BSSN baseline.")
+    if center == "Vertex":
+        r.add_warning("grid_center_set=Vertex is compiled by macros, but the stable GPU baseline is Cell.")
+    if symmetry != "equatorial-symmetry":
+        r.add_warning("The stable validation case uses equatorial-symmetry; other symmetries need regression tests.")
+    if fd != "4th-order":
+        r.add_warning("The stable validation case uses 4th-order finite differences.")
+    if gauge not in {0, 1}:
+        r.add_warning("Input comments recommend gauge_choice 0 or 1; other gauges need dedicated validation.")
+    if tetrad != 2:
+        r.add_warning("Input comments recommend tetrad_type=2; other tetrads affect wave extraction conventions.")
+
+    if ahf == "yes":
+        r.add_warning("AHF_Find=yes is supported by macros, but it is outside the current stable GPU baseline.")
+
+    if boundary == "Shibata-choice":
+        r.add_risk("Shibata-choice is not faithfully distinguished in the current macro generator; it maps to the BAM branch.")
+    elif boundary == "BAM-choice":
+        r.add_note("boundary_choice=BAM-choice is supported.")
+
+    if cpu_part is not None or gpu_part is not None:
+        r.add_warning("CPU_Part/GPU_Part are printed and propagated, but they do not control a real mixed CPU/GPU split in this branch.")
+
+    check_output_and_time(r, mod)
+    check_grid_geometry(r, mod, grid)
+    check_punctures(r, mod, init, puncture_data)
+    check_equation_specific(r, mod, eq, grid, fd)
+    check_runtime_environment(r, mod, eq, grid, fd)
+    check_stable_profile(r, mod)
+
+    return r
+
+
+def main() -> int:
+    parser = argparse.ArgumentParser()
+    parser.add_argument(
+        "-f",
+        "--file",
+        "--input",
+        dest="input_file",
+        default="AMSS_NCKU_Input.py",
+        help="path to AMSS_NCKU_Input.py",
+    )
+    args = parser.parse_args()
+
+    path = Path(args.input_file).resolve()
+    if not path.exists():
+        print(f"ERROR: input file not found: {path}")
+        return 2
+
+    try:
+        mod = load_input_module(path)
+    except Exception as exc:
+        print(f"ERROR: failed to load input file: {exc}")
+        return 2
+
+    result = check_input(mod)
+
+    print(f"Input: {path}")
+    print(f"GPU_Calculation: {get_attr(mod, 'GPU_Calculation', 'no')}")
+    print(f"Symmetry: {get_attr(mod, 'Symmetry', '')}")
+    print(f"Equation_Class: {get_attr(mod, 'Equation_Class', '')}")
+    print(f"Initial_Data_Method: {get_attr(mod, 'Initial_Data_Method', '')}")
+    print(f"puncture_data_set: {get_attr(mod, 'puncture_data_set', '')}")
+    print(f"basic_grid_set: {get_attr(mod, 'basic_grid_set', '')}")
+    print(f"grid_center_set: {get_attr(mod, 'grid_center_set', '')}")
+    print(f"Finite_Diffenence_Method: {get_attr(mod, 'Finite_Diffenence_Method', '')}")
+    print(f"gauge_choice: {get_attr(mod, 'gauge_choice', '')}")
+    print(f"tetrad_type: {get_attr(mod, 'tetrad_type', '')}")
+    print(f"boundary_choice: {get_attr(mod, 'boundary_choice', '')}")
+    print(f"AHF_Find: {get_attr(mod, 'AHF_Find', '')}")
+    print(f"AMSS_Z4C_MRBD: {get_attr(mod, 'AMSS_Z4C_MRBD', 0)}")
+    print("")
+
+    for msg in result.notes:
+        print(f"NOTE: {msg}")
+    for msg in result.warnings:
+        print(f"WARNING: {msg}")
+    for msg in result.risks:
+        print(f"RISK: {msg}")
+
+    print("")
+    if result.risks:
+        print("Verdict: review the risks above before running.")
+        return 1
+
+    if result.warnings:
+        print("Verdict: runnable on the current GPU branch, but keep the warnings in mind.")
+        return 0
+
+    print("Verdict: OK to run on the current GPU branch.")
+    return 0
+
+
+if __name__ == "__main__":
+    raise SystemExit(main())

AMSS_NCKU_Input.py

@@ -16,9 +16,9 @@ import numpy
 File_directory   = "GW150914"                    ## output file directory
 Output_directory = "binary_output"               ## binary data file directory
                                                  ## The file directory name should not be too long
-MPI_processes    = 64                             ## number of mpi processes used in the simulation
+MPI_processes    = 2                             ## number of mpi processes used in the simulation
 
-GPU_Calculation  = "no"                          ## Use GPU or not 
+GPU_Calculation  = "yes"                          ## Use GPU or not 
                                                  ## (prefer "no" in the current version, because the GPU part may have bugs when integrated in this Python interface)
 CPU_Part         = 1.0
 GPU_Part         = 0.0
@@ -158,7 +158,7 @@ Detector_Rmax   = 160.0                  ## farest dector distance
 
 ## Setting the apprent horizon
 
-AHF_Find       = "no"                    ## whether to find the apparent horizon: choose "yes" or "no"
+AHF_Find       = "no"                     ## whether to find the apparent horizon: choose "yes" or "no"
 
 AHF_Find_Every = 24
 AHF_Dump_Time  = 20.0

AMSS_NCKU_Program.py

@@ -58,31 +58,36 @@ File_directory = os.path.join(input_data.File_directory)
 
 ## If the specified output directory exists, ask the user whether to continue
 if os.path.exists(File_directory):
-    print( " Output dictionary has been existed !!!  "                                                              )
-    print( " If you want to overwrite the existing file directory, please input 'continue' in the terminal !! "     ) 
-    print( " If you want to retain the existing file directory, please input 'stop' in the terminal to stop the "   ) 
-    print( " simulation. Then you can reset the output dictionary in the input script file AMSS_NCKU_Input.py !!! " )
-    print(                                                                                                          )
-    ## Prompt whether to overwrite the existing directory
-    while True:
-        try:
-            inputvalue = input()
-            ## If the user agrees to overwrite, proceed and remove the existing directory
-            if ( inputvalue == "continue" ):
-                print( " Continue the calculation !!! " )
-                print(                                  )
-                break  
-            ## If the user chooses not to overwrite, exit and keep the existing directory
-            elif ( inputvalue == "stop" ):
-                print( " Stop the calculation !!! "    )
-                sys.exit() 
-            ## If the user input is invalid, prompt again
-            else:
+    auto_overwrite = str(getattr(input_data, "Auto_Overwrite_Output", "yes")).strip().lower()
+    if auto_overwrite in ("1", "yes", "y", "true", "on", "continue"):
+        print( " Output dictionary has been existed; Auto_Overwrite_Output=yes, continue the calculation. " )
+        print(                                                                                              )
+    else:
+        print( " Output dictionary has been existed !!!  "                                                              )
+        print( " If you want to overwrite the existing file directory, please input 'continue' in the terminal !! "     ) 
+        print( " If you want to retain the existing file directory, please input 'stop' in the terminal to stop the "   ) 
+        print( " simulation. Then you can reset the output dictionary in the input script file AMSS_NCKU_Input.py !!! " )
+        print(                                                                                                          )
+        ## Prompt whether to overwrite the existing directory
+        while True:
+            try:
+                inputvalue = input()
+                ## If the user agrees to overwrite, proceed and remove the existing directory
+                if ( inputvalue == "continue" ):
+                    print( " Continue the calculation !!! " )
+                    print(                                  )
+                    break  
+                ## If the user chooses not to overwrite, exit and keep the existing directory
+                elif ( inputvalue == "stop" ):
+                    print( " Stop the calculation !!! "    )
+                    sys.exit() 
+                ## If the user input is invalid, prompt again
+                else:
+                    print( " Please input your choice !!! "                   )
+                    print( " Input 'continue' or 'stop' in the terminal !!! " )
+            except ValueError:
                 print( " Please input your choice !!! "                   )
                 print( " Input 'continue' or 'stop' in the terminal !!! " )
-        except ValueError:
-            print( " Please input your choice !!! "                   )
-            print( " Input 'continue' or 'stop' in the terminal !!! " )
         
 ## Remove the existing output directory if present
 shutil.rmtree(File_directory, ignore_errors=True)
@@ -174,14 +179,11 @@ import generate_macrodef
 generate_macrodef.generate_macrodef_h()
 print( " AMSS-NCKU macro file macrodef.h has been generated. " )
      
-generate_macrodef.generate_macrodef_fh()
-print( " AMSS-NCKU macro file macrodef.fh has been generated. " )
-
-generate_macrodef.generate_build_config()
-print( " AMSS-NCKU build config AMSS_NCKU_build.mk has been generated. " )
-
-
-##################################################################
+generate_macrodef.generate_macrodef_fh()
+print( " AMSS-NCKU macro file macrodef.fh has been generated. " )
+
+
+##################################################################
 
 # Compile the AMSS-NCKU program according to user requirements
 
@@ -220,13 +222,11 @@ shutil.copytree(AMSS_NCKU_source_path, AMSS_NCKU_source_copy)
 
 # Copy the generated macro files into the AMSS_NCKU source folder
 
-macrodef_h_path  = os.path.join(File_directory, "macrodef.h")
-macrodef_fh_path = os.path.join(File_directory, "macrodef.fh")
-build_config_path = os.path.join(File_directory, "AMSS_NCKU_build.mk")
-
-shutil.copy2(macrodef_h_path,  AMSS_NCKU_source_copy)
-shutil.copy2(macrodef_fh_path, AMSS_NCKU_source_copy)
-shutil.copy2(build_config_path, AMSS_NCKU_source_copy)
+macrodef_h_path  = os.path.join(File_directory, "macrodef.h") 
+macrodef_fh_path = os.path.join(File_directory, "macrodef.fh") 
+
+shutil.copy2(macrodef_h_path,  AMSS_NCKU_source_copy)
+shutil.copy2(macrodef_fh_path, AMSS_NCKU_source_copy)
 
 # Notes on copying files:
 # shutil.copy2 preserves file metadata such as modification time.
@@ -263,7 +263,7 @@ print()
 if (input_data.GPU_Calculation == "no"):
     ABE_file = os.path.join(AMSS_NCKU_source_copy, "ABE")
 elif (input_data.GPU_Calculation == "yes"):
-    ABE_file = os.path.join(AMSS_NCKU_source_copy, "ABEGPU")
+    ABE_file = os.path.join(AMSS_NCKU_source_copy, "ABE_CUDA")
 
 if not os.path.exists( ABE_file ):
     print(                                                                                                  )

AMSS_NCKU_source/ABE.C

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

AMSS_NCKU_source/Block.C

@@ -6,14 +6,68 @@
 #include <cstdio>
 #include <string>
 #include <cmath>
-#include <new>
-using namespace std;
-
-#include "Block.h"
-#include "misc.h"
-
-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)
-{
+#include <new>
+using namespace std;
+
+#include "Block.h"
+#include "misc.h"
+
+#if USE_CUDA_BSSN || USE_CUDA_Z4C
+#include <cuda_runtime_api.h>
+#endif
+
+namespace {
+
+bool cuda_pin_gridfuncs_enabled()
+{
+  static int enabled = -1;
+  if (enabled < 0)
+  {
+    const char *env = getenv("AMSS_CUDA_PIN_GRIDFUNCS");
+    enabled = (env && atoi(env) != 0) ? 1 : 0;
+  }
+  return enabled != 0;
+}
+
+double *alloc_gridfunc(size_t count, unsigned char &pinned)
+{
+  pinned = 0;
+#if USE_CUDA_BSSN || USE_CUDA_Z4C
+  if (cuda_pin_gridfuncs_enabled())
+  {
+    double *ptr = 0;
+    cudaError_t err = cudaMallocHost((void **)&ptr, count * sizeof(double));
+    if (err == cudaSuccess)
+    {
+      pinned = 1;
+      return ptr;
+    }
+    cudaGetLastError();
+  }
+#endif
+  return (double *)malloc(sizeof(double) * count);
+}
+
+void free_gridfunc(double *ptr, unsigned char pinned)
+{
+  if (!ptr)
+    return;
+#if USE_CUDA_BSSN || USE_CUDA_Z4C
+  if (pinned)
+  {
+    cudaFreeHost(ptr);
+    return;
+  }
+#else
+  (void)pinned;
+#endif
+  free(ptr);
+}
+
+}
+
+Block::Block(int DIM, int *shapei, double *bboxi, int ranki, int ingfsi, int fngfsi, int levi, const int cgpui) : rank(ranki), lev(levi), cgpu(cgpui), ingfs(ingfsi), fngfs(fngfsi), igfs(0), fgfs(0), fgfs_pinned(0)
+{
   for (int i = 0; i < dim; i++)
     X[i] = 0;
 
@@ -68,14 +122,15 @@ Block::Block(int DIM, int *shapei, double *bboxi, int ranki, int ingfsi, int fng
 #endif
     }
 
-    int nn = shape[0] * shape[1] * shape[2];
-    fgfs = new double *[fngfs];
-    for (int i = 0; i < fngfs; i++)
-    {
-      fgfs[i] = (double *)malloc(sizeof(double) * nn);
-      if (!(fgfs[i]))
-      {
-        cout << "on node#" << rank << ", out of memory when constructing Block." << endl;
+    int nn = shape[0] * shape[1] * shape[2];
+    fgfs = new double *[fngfs];
+    fgfs_pinned = new unsigned char[fngfs];
+    for (int i = 0; i < fngfs; i++)
+    {
+      fgfs[i] = alloc_gridfunc((size_t)nn, fgfs_pinned[i]);
+      if (!(fgfs[i]))
+      {
+        cout << "on node#" << rank << ", out of memory when constructing Block." << endl;
         MPI_Abort(MPI_COMM_WORLD, 1);
       }
       memset(fgfs[i], 0, sizeof(double) * nn);
@@ -103,17 +158,19 @@ Block::~Block()
   {
     for (int i = 0; i < dim; i++)
       delete[] X[i];
-    for (int i = 0; i < ingfs; i++)
-      free(igfs[i]);
-    delete[] igfs;
-    for (int i = 0; i < fngfs; i++)
-      free(fgfs[i]);
-    delete[] fgfs;
-    X[0] = X[1] = X[2] = 0;
-    igfs = 0;
-    fgfs = 0;
-  }
-}
+    for (int i = 0; i < ingfs; i++)
+      free(igfs[i]);
+    delete[] igfs;
+    for (int i = 0; i < fngfs; i++)
+      free_gridfunc(fgfs[i], fgfs_pinned ? fgfs_pinned[i] : 0);
+    delete[] fgfs;
+    delete[] fgfs_pinned;
+    X[0] = X[1] = X[2] = 0;
+    igfs = 0;
+    fgfs = 0;
+    fgfs_pinned = 0;
+  }
+}
 void Block::checkBlock()
 {
   int myrank;
@@ -184,12 +241,14 @@ void Block::swapList(MyList<var> *VarList1, MyList<var> *VarList2, int myrank)
   if (rank == myrank)
   {
     MyList<var> *varl1 = VarList1, *varl2 = VarList2;
-    while (varl1 && varl2)
-    {
-      misc::swap<double *>(fgfs[varl1->data->sgfn], fgfs[varl2->data->sgfn]);
-      varl1 = varl1->next;
-      varl2 = varl2->next;
-    }
+    while (varl1 && varl2)
+    {
+      misc::swap<double *>(fgfs[varl1->data->sgfn], fgfs[varl2->data->sgfn]);
+      if (fgfs_pinned)
+        misc::swap<unsigned char>(fgfs_pinned[varl1->data->sgfn], fgfs_pinned[varl2->data->sgfn]);
+      varl1 = varl1->next;
+      varl2 = varl2->next;
+    }
     if (varl1 || varl2)
     {
       cout << "error in Block::swaplist, var lists does not match." << endl;

AMSS_NCKU_source/Block.h

@@ -13,14 +13,15 @@ public:
    int shape[dim];
    double bbox[2 * dim];
    double *X[dim];
-   int rank; // where the real data locate in
-   int lev, cgpu;
-   int ingfs, fngfs;
-   int *(*igfs);
-   double *(*fgfs);
+   int rank; // where the real data locate in
+   int lev, cgpu;
+   int ingfs, fngfs;
+   int *(*igfs);
+   double *(*fgfs);
+   unsigned char *fgfs_pinned;
 
 public:
-   Block() {};
+   Block() : rank(0), lev(0), cgpu(0), ingfs(0), fngfs(0), igfs(0), fgfs(0), fgfs_pinned(0) {};
    Block(int DIM, int *shapei, double *bboxi, int ranki, int ingfsi, int fngfs, int levi, const int cgpui = 0);
 
    ~Block();

AMSS_NCKU_source/MPatch.C

@@ -11,12 +11,15 @@
 using namespace std;
 
 #include "misc.h"
-#include "MPatch.h"
-#include "Parallel.h"
-#include "fmisc.h"
-#ifdef INTERP_LB_PROFILE
-#include "interp_lb_profile.h"
-#endif
+#include "MPatch.h"
+#include "Parallel.h"
+#include "fmisc.h"
+#if USE_CUDA_BSSN
+#include "bssn_rhs_cuda.h"
+#endif
+#ifdef INTERP_LB_PROFILE
+#include "interp_lb_profile.h"
+#endif
 
 namespace
 {
@@ -154,8 +157,8 @@ void build_block_bin_index(Patch *patch, const double *DH, BlockBinIndex &index)
   index.valid = true;
 }
 
-int find_block_index_for_point(const BlockBinIndex &index, const double *pox, const double *DH)
-{
+int find_block_index_for_point(const BlockBinIndex &index, const double *pox, const double *DH)
+{
   if (!index.valid)
     return -1;
 
@@ -175,10 +178,448 @@ int find_block_index_for_point(const BlockBinIndex &index, const double *pox, co
   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;
-}
-} // namespace
+
+  return -1;
+}
+
+inline int fortran_idint_local(double x)
+{
+  return int(x);
+}
+
+bool interp_fast_enabled()
+{
+  static int enabled = -1;
+  if (enabled < 0)
+  {
+    const char *env = getenv("AMSS_INTERP_FAST");
+    enabled = (!env || atoi(env) != 0) ? 1 : 0;
+  }
+  return enabled != 0;
+}
+
+bool interp_gpu_enabled()
+{
+  static int enabled = -1;
+  if (enabled < 0)
+  {
+    const char *env = getenv("AMSS_INTERP_GPU");
+    enabled = (env && atoi(env) != 0) ? 1 : 0;
+  }
+  return enabled != 0;
+}
+
+bool interp_fast_compare_enabled()
+{
+  static int enabled = -1;
+  if (enabled < 0)
+  {
+    const char *env = getenv("AMSS_INTERP_FAST_COMPARE");
+    enabled = (env && atoi(env) != 0) ? 1 : 0;
+  }
+  return enabled != 0;
+}
+
+double interp_fast_compare_tol()
+{
+  static double tol = -1.0;
+  if (tol < 0.0)
+  {
+    const char *env = getenv("AMSS_INTERP_FAST_COMPARE_TOL");
+    tol = (env && atof(env) > 0.0) ? atof(env) : 1.0e-11;
+  }
+  return tol;
+}
+
+long long interp_fast_compare_limit()
+{
+  static long long limit = -1;
+  if (limit < 0)
+  {
+    const char *env = getenv("AMSS_INTERP_FAST_COMPARE_LIMIT");
+    limit = (env && atoll(env) > 0) ? atoll(env) : 4096;
+  }
+  return limit;
+}
+
+struct FastInterpStencil
+{
+  int cxB[dim];
+  double cx[dim];
+  double wx[8];
+  double wy[8];
+  double wz[8];
+  int nsamples;
+  int loc[512];
+  unsigned char sign_mask[512];
+  double weight[512];
+};
+
+inline void lagrange_unit_weights(double x, int ordn, double *w)
+{
+  for (int i = 0; i < ordn; i++)
+  {
+    double num = 1.0;
+    double den = 1.0;
+    for (int j = 0; j < ordn; j++)
+    {
+      if (j == i)
+        continue;
+      num *= (x - double(j));
+      den *= double(i - j);
+    }
+    w[i] = num / den;
+  }
+}
+
+inline void z_unit_weights(double x, int ordn, double *w)
+{
+  if (ordn == 6)
+  {
+    static const double c_uniform[6] = {-1.0, 5.0, -10.0, 10.0, -5.0, 1.0};
+    for (int i = 0; i < 6; i++)
+    {
+      if (x == double(i))
+      {
+        for (int j = 0; j < 6; j++)
+          w[j] = (j == i) ? 1.0 : 0.0;
+        return;
+      }
+    }
+    double den = 0.0;
+    for (int i = 0; i < 6; i++)
+    {
+      w[i] = c_uniform[i] / (x - double(i));
+      den += w[i];
+    }
+    for (int i = 0; i < 6; i++)
+      w[i] /= den;
+    return;
+  }
+  lagrange_unit_weights(x, ordn, w);
+}
+
+inline bool fast_interp_map_index(int idx, int extent, int d,
+                                  int &mapped, unsigned char &mask)
+{
+  if (idx > 0)
+    mapped = idx;
+  else
+  {
+    mask |= (unsigned char)(1u << d);
+#ifdef Vertex
+#ifdef Cell
+#error Both Cell and Vertex are defined
+#endif
+    mapped = 2 - idx;
+#else
+#ifdef Cell
+    mapped = 1 - idx;
+#else
+#error Not define Vertex nor Cell
+#endif
+#endif
+  }
+  return mapped >= 1 && mapped <= extent;
+}
+
+bool prepare_fast_interp_stencil(Block *BP, const double *pox, int ordn,
+                                 int Symmetry, FastInterpStencil &st)
+{
+  if (!BP || ordn <= 0 || ordn > 8)
+    return false;
+
+  st.nsamples = 0;
+
+  const int NO_SYMM = 0;
+  const int OCTANT = 2;
+  int cmin[dim], cmax[dim], cxT[dim];
+  for (int d = 0; d < dim; d++)
+  {
+    const double *X = BP->X[d];
+    const double dX = X[1] - X[0];
+    const int cxI = fortran_idint_local((pox[d] - X[0]) / dX + 0.4) + 1;
+    st.cxB[d] = cxI - ordn / 2 + 1;
+    cxT[d] = st.cxB[d] + ordn - 1;
+    cmin[d] = 1;
+    cmax[d] = BP->shape[d];
+
+#ifdef Vertex
+#ifdef Cell
+#error Both Cell and Vertex are defined
+#endif
+    if (Symmetry == OCTANT && d < 2 && fabs(X[0]) < dX)
+      cmin[d] = -ordn / 2 + 2;
+    if (Symmetry != NO_SYMM && d == 2 && fabs(X[0]) < dX)
+      cmin[d] = -ordn / 2 + 2;
+#else
+#ifdef Cell
+    if (Symmetry == OCTANT && d < 2 && fabs(X[0]) < dX)
+      cmin[d] = -ordn / 2 + 1;
+    if (Symmetry != NO_SYMM && d == 2 && fabs(X[0]) < dX)
+      cmin[d] = -ordn / 2 + 1;
+#else
+#error Not define Vertex nor Cell
+#endif
+#endif
+
+    if (st.cxB[d] < cmin[d])
+    {
+      st.cxB[d] = cmin[d];
+      cxT[d] = st.cxB[d] + ordn - 1;
+    }
+    if (cxT[d] > cmax[d])
+    {
+      cxT[d] = cmax[d];
+      st.cxB[d] = cxT[d] + 1 - ordn;
+    }
+
+    if (st.cxB[d] > 0)
+      st.cx[d] = (pox[d] - X[st.cxB[d] - 1]) / dX;
+    else
+    {
+#ifdef Vertex
+#ifdef Cell
+#error Both Cell and Vertex are defined
+#endif
+      st.cx[d] = (pox[d] + X[1 - st.cxB[d]]) / dX;
+#else
+#ifdef Cell
+      st.cx[d] = (pox[d] + X[-st.cxB[d]]) / dX;
+#else
+#error Not define Vertex nor Cell
+#endif
+#endif
+    }
+  }
+
+  lagrange_unit_weights(st.cx[0], ordn, st.wx);
+  lagrange_unit_weights(st.cx[1], ordn, st.wy);
+  z_unit_weights(st.cx[2], ordn, st.wz);
+
+  for (int kk = 0; kk < ordn; kk++)
+  {
+    for (int jj = 0; jj < ordn; jj++)
+    {
+      for (int ii = 0; ii < ordn; ii++)
+      {
+        unsigned char mask = 0;
+        int ix, iy, iz;
+        if (!fast_interp_map_index(st.cxB[0] + ii, BP->shape[0], 0, ix, mask) ||
+            !fast_interp_map_index(st.cxB[1] + jj, BP->shape[1], 1, iy, mask) ||
+            !fast_interp_map_index(st.cxB[2] + kk, BP->shape[2], 2, iz, mask))
+          return false;
+        const int s = st.nsamples++;
+        st.loc[s] = (ix - 1) + (iy - 1) * BP->shape[0] +
+                    (iz - 1) * BP->shape[0] * BP->shape[1];
+        st.sign_mask[s] = mask;
+        st.weight[s] = st.wx[ii] * st.wy[jj] * st.wz[kk];
+      }
+    }
+  }
+  return true;
+}
+
+bool interpolate_var_list_with_stencil(Block *BP, MyList<var> *VarList,
+                                       int num_var, const double *pox,
+                                       int ordn, int Symmetry,
+                                       const FastInterpStencil &st,
+                                       double *out)
+{
+  if (num_var <= 0 || num_var > 128)
+    return false;
+
+  double *data_ptrs[128];
+  double *soa_ptrs[128];
+  var *vars[128];
+  MyList<var> *varl = VarList;
+  int k = 0;
+  while (varl)
+  {
+    if (k >= num_var)
+      return false;
+    vars[k] = varl->data;
+    data_ptrs[k] = BP->fgfs[vars[k]->sgfn];
+    soa_ptrs[k] = vars[k]->SoA;
+    out[k] = 0.0;
+    varl = varl->next;
+    k++;
+  }
+
+  if (k != num_var)
+    return false;
+
+  for (int s = 0; s < st.nsamples; s++)
+  {
+    const int loc = st.loc[s];
+    const double w = st.weight[s];
+    const unsigned char mask = st.sign_mask[s];
+    if (mask == 0)
+    {
+      for (int v = 0; v < num_var; v++)
+        out[v] += w * data_ptrs[v][loc];
+    }
+    else
+    {
+      for (int v = 0; v < num_var; v++)
+      {
+        const double *SoA = soa_ptrs[v];
+        double sgn = 1.0;
+        if (mask & 1u)
+          sgn *= SoA[0];
+        if (mask & 2u)
+          sgn *= SoA[1];
+        if (mask & 4u)
+          sgn *= SoA[2];
+        out[v] += w * sgn * data_ptrs[v][loc];
+      }
+    }
+  }
+
+  if (interp_fast_compare_enabled())
+  {
+    static int report_count = 0;
+    static long long compare_calls = 0;
+    if (compare_calls++ >= interp_fast_compare_limit())
+      return true;
+    const double tol = interp_fast_compare_tol();
+    varl = VarList;
+    k = 0;
+    while (varl)
+    {
+      var *vp = vars[k];
+      double ref = 0.0;
+      double x = pox[0], y = pox[1], z = pox[2];
+      f_global_interp(BP->shape, BP->X[0], BP->X[1], BP->X[2],
+                      BP->fgfs[vp->sgfn], ref,
+                      x, y, z, ordn, vp->SoA, Symmetry);
+      const double diff = fabs(ref - out[k]);
+      const double scale = 1.0 + fabs(ref);
+      if (diff > tol * scale && report_count < 32)
+      {
+        int rank = 0;
+        MPI_Comm_rank(MPI_COMM_WORLD, &rank);
+        fprintf(stderr,
+                "[AMSS-INTERP-CMP][rank %d] var=%s diff=%.17e ref=%.17e fast=%.17e p=(%.17e,%.17e,%.17e)\n",
+                rank, vp->name, diff, ref, out[k], pox[0], pox[1], pox[2]);
+        report_count++;
+      }
+      varl = varl->next;
+      k++;
+    }
+  }
+
+  return true;
+}
+
+bool interpolate_var_list_fast(Block *BP, MyList<var> *VarList, int num_var,
+                               const double *pox, int ordn, int Symmetry,
+                               double *out)
+{
+  if (!interp_fast_enabled())
+    return false;
+
+  FastInterpStencil st;
+  if (!prepare_fast_interp_stencil(BP, pox, ordn, Symmetry, st))
+    return false;
+
+  return interpolate_var_list_with_stencil(BP, VarList, num_var, pox,
+                                           ordn, Symmetry, st, out);
+}
+
+struct CachedInterpPoint
+{
+  Block *bp;
+  int owner_rank;
+  FastInterpStencil stencil;
+};
+
+struct SurfaceInterpCache
+{
+  Patch *patch;
+  int NN;
+  int symmetry;
+  double key[9];
+  vector<CachedInterpPoint> points;
+
+  SurfaceInterpCache() : patch(0), NN(0), symmetry(-1) {}
+};
+
+bool surface_cache_key_matches(const SurfaceInterpCache &cache, Patch *patch,
+                               int NN, double **XX, int Symmetry)
+{
+  if (cache.patch != patch || cache.NN != NN || cache.symmetry != Symmetry ||
+      int(cache.points.size()) != NN || NN <= 0)
+    return false;
+  const int mid = NN / 2;
+  const int last = NN - 1;
+  const int ids[3] = {0, mid, last};
+  int p = 0;
+  for (int q = 0; q < 3; q++)
+    for (int d = 0; d < dim; d++)
+      if (cache.key[p++] != XX[d][ids[q]])
+        return false;
+  return true;
+}
+
+SurfaceInterpCache *find_surface_cache(Patch *patch, int NN, double **XX,
+                                       int Symmetry)
+{
+  static vector<SurfaceInterpCache> caches;
+  for (size_t i = 0; i < caches.size(); i++)
+    if (surface_cache_key_matches(caches[i], patch, NN, XX, Symmetry))
+      return &caches[i];
+  if (caches.size() >= 24)
+    caches.erase(caches.begin());
+  caches.push_back(SurfaceInterpCache());
+  return &caches.back();
+}
+
+bool build_surface_cache(SurfaceInterpCache &cache, Patch *patch, int NN,
+                         double **XX, int Symmetry, const double *DH,
+                         const BlockBinIndex &block_index, int ordn)
+{
+  int myrank = 0;
+  MPI_Comm_rank(MPI_COMM_WORLD, &myrank);
+  cache.patch = patch;
+  cache.NN = NN;
+  cache.symmetry = Symmetry;
+  cache.points.clear();
+  cache.points.resize(NN);
+  const int mid = NN / 2;
+  const int last = NN - 1;
+  const int ids[3] = {0, mid, last};
+  int p = 0;
+  for (int q = 0; q < 3; q++)
+    for (int d = 0; d < dim; d++)
+      cache.key[p++] = XX[d][ids[q]];
+
+  for (int j = 0; j < NN; j++)
+  {
+    double pox[dim];
+    for (int d = 0; d < dim; d++)
+      pox[d] = XX[d][j];
+    const int block_i = find_block_index_for_point(block_index, pox, DH);
+    if (block_i < 0)
+    {
+      cache.points[j].bp = 0;
+      cache.points[j].owner_rank = -1;
+      continue;
+    }
+    Block *BP = block_index.views[block_i].bp;
+    cache.points[j].bp = BP;
+    cache.points[j].owner_rank = BP->rank;
+    cache.points[j].stencil.nsamples = 0;
+    if (BP->rank == myrank)
+    {
+      if (!prepare_fast_interp_stencil(BP, pox, ordn, Symmetry,
+                                       cache.points[j].stencil))
+        return false;
+    }
+  }
+  return true;
+}
+} // namespace
 
 Patch::Patch(int DIM, int *shapei, double *bboxi, int levi, bool buflog, int Symmetry) : lev(levi)
 {
@@ -561,22 +1002,26 @@ void Patch::Interp_Points(MyList<var> *VarList,
     if (block_i >= 0)
     {
       Block *BP = block_index.views[block_i].bp;
-      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++;
-        }
-      }
-    }
-  }
+      owner_rank[j] = BP->rank;
+      if (myrank == BP->rank)
+      {
+        //---> interpolation
+        if (!interpolate_var_list_fast(BP, VarList, num_var, pox, ordn,
+                                       Symmetry, Shellf + j * num_var))
+        {
+          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++;
+          }
+        }
+      }
+    }
+  }
 
   // Replace MPI_Allreduce with per-owner MPI_Bcast:
   // Group consecutive points by owner rank and broadcast each group.
@@ -659,10 +1104,8 @@ void Patch::Interp_Points(MyList<var> *VarList,
     varl = varl->next;
   }
 
-  memset(Shellf, 0, sizeof(double) * NN * num_var);
-
-  // owner_rank[j] records which MPI rank owns point j
-  int *owner_rank;
+  // 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;
@@ -670,12 +1113,117 @@ void Patch::Interp_Points(MyList<var> *VarList,
   double DH[dim];
   for (int i = 0; i < dim; i++)
     DH[i] = getdX(i);
-  BlockBinIndex block_index;
-  build_block_bin_index(this, DH, block_index);
-
-  // --- Interpolation phase (identical to original) ---
-  for (int j = 0; j < NN; j++)
-  {
+  BlockBinIndex block_index;
+  build_block_bin_index(this, DH, block_index);
+  SurfaceInterpCache *surface_cache = 0;
+  bool use_surface_cache = false;
+  if (interp_fast_enabled())
+  {
+    surface_cache = find_surface_cache(this, NN, XX, Symmetry);
+    use_surface_cache = surface_cache_key_matches(*surface_cache, this, NN, XX, Symmetry);
+    if (!use_surface_cache)
+      use_surface_cache = build_surface_cache(*surface_cache, this, NN, XX,
+                                              Symmetry, DH, block_index, ordn);
+  }
+
+  // --- Interpolation phase (identical to original) ---
+#if USE_CUDA_BSSN
+  const bool use_gpu_interp = interp_gpu_enabled() && use_surface_cache && num_var == 2 &&
+                              VarList && VarList->next && !VarList->next->next;
+#else
+  const bool use_gpu_interp = false;
+#endif
+  if (use_gpu_interp)
+  {
+#if USE_CUDA_BSSN
+    vector<vector<int> > local_points(block_index.views.size());
+    for (int j = 0; j < NN; j++)
+    {
+      for (int i = 0; i < dim; i++)
+      {
+        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);
+        }
+      }
+
+      CachedInterpPoint &cp = surface_cache->points[j];
+      Block *BP = cp.bp;
+      owner_rank[j] = cp.owner_rank;
+      if (BP && myrank == BP->rank)
+      {
+        for (size_t bi = 0; bi < block_index.views.size(); bi++)
+        {
+          if (block_index.views[bi].bp == BP)
+          {
+            local_points[bi].push_back(j);
+            break;
+          }
+        }
+      }
+    }
+
+    var *v0 = VarList->data;
+    var *v1 = VarList->next->data;
+    double soa6[6] = {
+        v0->SoA[0], v0->SoA[1], v0->SoA[2],
+        v1->SoA[0], v1->SoA[1], v1->SoA[2]};
+
+    for (size_t bi = 0; bi < local_points.size(); bi++)
+    {
+      const int count = int(local_points[bi].size());
+      if (count <= 0)
+        continue;
+
+      Block *BP = block_index.views[bi].bp;
+      vector<double> px(count), py(count), pz(count), out(2 * count);
+      for (int q = 0; q < count; q++)
+      {
+        const int j = local_points[bi][q];
+        px[q] = XX[0][j];
+        py[q] = XX[1][j];
+        pz[q] = XX[2][j];
+      }
+
+      const double dx = BP->X[0][1] - BP->X[0][0];
+      const double dy = BP->X[1][1] - BP->X[1][0];
+      const double dz = BP->X[2][1] - BP->X[2][0];
+      const int ok = bssn_cuda_interp_host_two_fields(
+          BP, BP->shape,
+          BP->fgfs[v0->sgfn], BP->fgfs[v1->sgfn],
+          BP->X[0][0], BP->X[1][0], BP->X[2][0],
+          dx, dy, dz,
+          &px[0], &py[0], &pz[0], count,
+          ordn, Symmetry, soa6, &out[0]);
+      if (ok != 0)
+      {
+        if (myrank == 0)
+          cout << "Patch::Interp_Points: CUDA two-field interpolation failed" << endl;
+        MPI_Abort(MPI_COMM_WORLD, 1);
+      }
+
+      for (int q = 0; q < count; q++)
+      {
+        const int j = local_points[bi][q];
+        Shellf[j * num_var] = out[2 * q];
+        Shellf[j * num_var + 1] = out[2 * q + 1];
+      }
+    }
+#endif
+  }
+  else
+  {
+  for (int j = 0; j < NN; j++)
+  {
     double pox[dim];
     for (int i = 0; i < dim; i++)
     {
@@ -692,28 +1240,59 @@ void Patch::Interp_Points(MyList<var> *VarList,
             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;
-      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 (use_surface_cache)
+    {
+      CachedInterpPoint &cp = surface_cache->points[j];
+      Block *BP = cp.bp;
+      owner_rank[j] = cp.owner_rank;
+      if (BP && myrank == BP->rank)
+      {
+        if (!interpolate_var_list_with_stencil(BP, VarList, num_var, pox,
+                                               ordn, Symmetry, cp.stencil,
+                                               Shellf + j * num_var))
+        {
+          MyList<var> *varl_fallback = VarList;
+          int k = 0;
+          while (varl_fallback)
+          {
+            f_global_interp(BP->shape, BP->X[0], BP->X[1], BP->X[2], BP->fgfs[varl_fallback->data->sgfn], Shellf[j * num_var + k],
+                            pox[0], pox[1], pox[2], ordn, varl_fallback->data->SoA, Symmetry);
+            varl_fallback = varl_fallback->next;
+            k++;
+          }
+        }
+      }
+    }
+    else
+    {
+      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;
+        owner_rank[j] = BP->rank;
+      if (myrank == BP->rank)
+      {
+        if (!interpolate_var_list_fast(BP, VarList, num_var, pox, ordn,
+                                       Symmetry, Shellf + j * num_var))
+        {
+          MyList<var> *varl_fallback = VarList;
+          int k = 0;
+          while (varl_fallback)
+          {
+            f_global_interp(BP->shape, BP->X[0], BP->X[1], BP->X[2], BP->fgfs[varl_fallback->data->sgfn], Shellf[j * num_var + k],
+                            pox[0], pox[1], pox[2], ordn, varl_fallback->data->SoA, Symmetry);
+            varl_fallback = varl_fallback->next;
+            k++;
+          }
+        }
+      }
+      }
+    }
+  }
+  }
 
 #ifdef INTERP_LB_PROFILE
   double t_interp_end = MPI_Wtime();
@@ -965,22 +1544,26 @@ void Patch::Interp_Points(MyList<var> *VarList,
     if (block_i >= 0)
     {
       Block *BP = block_index.views[block_i].bp;
-      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++;
-        }
-      }
-    }
-  }
+      owner_rank[j] = BP->rank;
+      if (myrank == BP->rank)
+      {
+        //---> interpolation
+        if (!interpolate_var_list_fast(BP, VarList, num_var, pox, ordn,
+                                       Symmetry, Shellf + j * num_var))
+        {
+          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++;
+          }
+        }
+      }
+    }
+  }
 
   // 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

AMSS_NCKU_source/Parallel.h

@@ -104,6 +104,14 @@ namespace Parallel
     double **recv_bufs;
     int *send_buf_caps;
     int *recv_buf_caps;
+    unsigned char *send_buf_pinned;
+    unsigned char *recv_buf_pinned;
+    unsigned char *send_buf_is_dev;
+    unsigned char *recv_buf_is_dev;
+    int *send_buf_caps_dev;
+    int *recv_buf_caps_dev;
+    double **send_bufs_dev;
+    double **recv_bufs_dev;
     MPI_Request *reqs;
     MPI_Status *stats;
     int max_reqs;
@@ -111,12 +119,14 @@ namespace Parallel
     int *tc_req_node;
     int *tc_req_is_recv;
     int *tc_completed;
+    bool cuda_aware_mode;
     SyncCache();
     void invalidate();
     void destroy();
   };
 
   void Sync_cached(MyList<Patch> *PatL, MyList<var> *VarList, int Symmetry, SyncCache &cache);
+  void Sync_ensure_cache(MyList<Patch> *PatL, int Symmetry, SyncCache &cache);
   void transfer_cached(MyList<gridseg> **src, MyList<gridseg> **dst,
                        MyList<var> *VarList1, MyList<var> *VarList2,
                        int Symmetry, SyncCache &cache);
@@ -179,13 +189,13 @@ namespace Parallel
   MyList<Parallel::gridseg> *clone_gsl(MyList<Parallel::gridseg> *p, bool first_only);
   MyList<Parallel::gridseg> *build_bulk_gsl(Patch *Pat); // similar to build_owned_gsl0 but does not care rank issue
   MyList<Parallel::gridseg> *build_bulk_gsl(Block *bp, Patch *Pat);
-  void build_PhysBD_gstl(Patch *Pat, MyList<Parallel::gridseg> *srci, MyList<Parallel::gridseg> *dsti,
-                         MyList<Parallel::gridseg> **out_src, MyList<Parallel::gridseg> **out_dst);
-  void PeriodicBD(Patch *Pat, MyList<var> *VarList, int Symmetry);
-  double L2Norm(Patch *Pat, var *vf);
-  void L2Norm7(Patch *Pat, var **vf, double *norms);
-  void checkgsl(MyList<Parallel::gridseg> *pp, bool first_only);
-  void checkvarl(MyList<var> *pp, bool first_only);
+  void build_PhysBD_gstl(Patch *Pat, MyList<Parallel::gridseg> *srci, MyList<Parallel::gridseg> *dsti,
+                         MyList<Parallel::gridseg> **out_src, MyList<Parallel::gridseg> **out_dst);
+  void PeriodicBD(Patch *Pat, MyList<var> *VarList, int Symmetry);
+  double L2Norm(Patch *Pat, var *vf);
+  void L2Norm7(Patch *Pat, var **vf, double *norms);
+  void checkgsl(MyList<Parallel::gridseg> *pp, bool first_only);
+  void checkvarl(MyList<var> *pp, bool first_only);
   MyList<Parallel::gridseg> *divide_gsl(MyList<Parallel::gridseg> *p, Patch *Pat);
   MyList<Parallel::gridseg> *divide_gs(MyList<Parallel::gridseg> *p, Patch *Pat);
   void prepare_inter_time_level(Patch *Pat,
@@ -217,12 +227,12 @@ namespace Parallel
   void aligncheck(double *bbox0, double *bboxl, int lev, double *DH0, int *shape);
   bool point_locat_gsl(double *pox, MyList<Parallel::gridseg> *gsl);
   void checkpatchlist(MyList<Patch> *PatL, bool buflog);
-
-  double L2Norm(Patch *Pat, var *vf, MPI_Comm Comm_here);
-  void L2Norm7(Patch *Pat, var **vf, double *norms, MPI_Comm Comm_here);
-  bool PatList_Interp_Points(MyList<Patch> *PatL, MyList<var> *VarList,
-                             int NN, double **XX,
-                             double *Shellf, int Symmetry, MPI_Comm Comm_here);
+
+  double L2Norm(Patch *Pat, var *vf, MPI_Comm Comm_here);
+  void L2Norm7(Patch *Pat, var **vf, double *norms, MPI_Comm Comm_here);
+  bool PatList_Interp_Points(MyList<Patch> *PatL, MyList<var> *VarList,
+                             int NN, double **XX,
+                             double *Shellf, int Symmetry, MPI_Comm Comm_here);
 #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);

AMSS_NCKU_source/ShellPatch.C

@@ -59,7 +59,7 @@ bool shell_fast_interp_enabled()
   if (enabled < 0)
   {
     const char *env = getenv("AMSS_SHELL_FAST_INTERP");
-    enabled = (!env || atoi(env) != 0) ? 1 : 0;
+    enabled = (env && atoi(env) != 0) ? 1 : 0;
   }
   return enabled != 0;
 }
@@ -70,7 +70,7 @@ bool shell_parallel_interp_enabled()
   if (enabled < 0)
   {
     const char *env = getenv("AMSS_SHELL_PARALLEL_INTERP");
-    enabled = (!env || atoi(env) != 0) ? 1 : 0;
+    enabled = (env && atoi(env) != 0) ? 1 : 0;
   }
   return enabled != 0;
 }

AMSS_NCKU_source/Z4c_class.C

@@ -3,6 +3,7 @@
 #include <sstream>
 #include <cstdio>
 #include <map>
+#include <string>
 using namespace std;
 #else
 #include <stdio.h>
@@ -28,6 +29,20 @@ using namespace std;
 #include "kodiss.h"
 #include "parameters.h"
 
+#ifndef USE_CUDA_Z4C
+#define USE_CUDA_Z4C 0
+#endif
+
+#if USE_CUDA_Z4C && (ABEtype == 2)
+#include "z4c_rhs_cuda.h"
+#endif
+#if USE_CUDA_BSSN
+#include "bssn_rhs_cuda.h"
+#ifdef WithShell
+#include "bssn_gpu.h"
+#endif
+#endif
+
 #ifdef With_AHF
 #include "derivatives.h"
 #include "myglobal.h"
@@ -37,6 +52,81 @@ using namespace std;
 
 // Define Z4c_class
 
+#if USE_CUDA_Z4C && (ABEtype == 2) && defined(WithShell)
+// GPU-accelerated Z4C shell RHS: same parameter signature as f_compute_rhs_Z4c_ss.
+// Internally calls gpu_rhs_z4c_ss which modifies trK→trKd before upload,
+// runs BSSN algebraic kernels, then applies Z4C post-processing (TZ_rhs, damping).
+extern "C" {
+static int cuda_compute_rhs_z4c_ss(
+    int *ex, double &T, double *crho, double *sigma, double *R,
+    double *X, double *Y, double *Z,
+    double *drhodx, double *drhody, double *drhodz,
+    double *dsigmadx, double *dsigmady, double *dsigmadz,
+    double *dRdx, double *dRdy, double *dRdz,
+    double *drhodxx, double *drhodxy, double *drhodxz, double *drhodyy, double *drhodyz, double *drhodzz,
+    double *dsigmadxx, double *dsigmadxy, double *dsigmadxz, double *dsigmadyy, double *dsigmadyz, double *dsigmadzz,
+    double *dRdxx, double *dRdxy, double *dRdxz, double *dRdyy, double *dRdyz, double *dRdzz,
+    double *chi, double *trK,
+    double *gxx, double *gxy, double *gxz, double *gyy, double *gyz, double *gzz,
+    double *Axx, double *Axy, double *Axz, double *Ayy, double *Ayz, double *Azz,
+    double *Gamx, double *Gamy, double *Gamz,
+    double *Lap, double *betax, double *betay, double *betaz,
+    double *dtSfx, double *dtSfy, double *dtSfz,
+    double *TZ,
+    double *chi_rhs, double *trK_rhs,
+    double *gxx_rhs, double *gxy_rhs, double *gxz_rhs, double *gyy_rhs, double *gyz_rhs, double *gzz_rhs,
+    double *Axx_rhs, double *Axy_rhs, double *Axz_rhs, double *Ayy_rhs, double *Ayz_rhs, double *Azz_rhs,
+    double *Gamx_rhs, double *Gamy_rhs, double *Gamz_rhs,
+    double *Lap_rhs, double *betax_rhs, double *betay_rhs, double *betaz_rhs,
+    double *dtSfx_rhs, double *dtSfy_rhs, double *dtSfz_rhs,
+    double *TZ_rhs,
+    double *rho_mat, double *Sx, double *Sy, double *Sz,
+    double *Sxx, double *Sxy, double *Sxz, double *Syy, double *Syz, double *Szz,
+    double *Gamxxx, double *Gamxxy, double *Gamxxz, double *Gamxyy, double *Gamxyz, double *Gamxzz,
+    double *Gamyxx, double *Gamyxy, double *Gamyxz, double *Gamyyy, double *Gamyyz, double *Gamyzz,
+    double *Gamzxx, double *Gamzxy, double *Gamzxz, double *Gamzyy, double *Gamzyz, double *Gamzzz,
+    double *Rxx, double *Rxy, double *Rxz, double *Ryy, double *Ryz, double *Rzz,
+    double *ham_Res, double *movx_Res, double *movy_Res, double *movz_Res,
+    double *Gmx_Res, double *Gmy_Res, double *Gmz_Res,
+    int &Symmetry, int &Lev, double &eps, int &sst, int &co)
+{
+    return gpu_rhs_z4c_ss(0, 0,  // calledby=ABE_main, mpi_rank=device_0
+                     ex, T, crho, sigma, R, X, Y, Z,
+                     drhodx, drhody, drhodz,
+                     dsigmadx, dsigmady, dsigmadz,
+                     dRdx, dRdy, dRdz,
+                     drhodxx, drhodxy, drhodxz, drhodyy, drhodyz, drhodzz,
+                     dsigmadxx, dsigmadxy, dsigmadxz, dsigmadyy, dsigmadyz, dsigmadzz,
+                     dRdxx, dRdxy, dRdxz, dRdyy, dRdyz, dRdzz,
+                     chi, trK,
+                     gxx, gxy, gxz, gyy, gyz, gzz,
+                     Axx, Axy, Axz, Ayy, Ayz, Azz,
+                     Gamx, Gamy, Gamz,
+                     Lap, betax, betay, betaz,
+                     dtSfx, dtSfy, dtSfz,
+                     TZ,
+                     chi_rhs, trK_rhs,
+                     gxx_rhs, gxy_rhs, gxz_rhs, gyy_rhs, gyz_rhs, gzz_rhs,
+                     Axx_rhs, Axy_rhs, Axz_rhs, Ayy_rhs, Ayz_rhs, Azz_rhs,
+                     Gamx_rhs, Gamy_rhs, Gamz_rhs,
+                     Lap_rhs, betax_rhs, betay_rhs, betaz_rhs,
+                     dtSfx_rhs, dtSfy_rhs, dtSfz_rhs,
+                     TZ_rhs,
+                     rho_mat, Sx, Sy, Sz,
+                     Sxx, Sxy, Sxz, Syy, Syz, Szz,
+                     Gamxxx, Gamxxy, Gamxxz, Gamxyy, Gamxyz, Gamxzz,
+                     Gamyxx, Gamyxy, Gamyxz, Gamyyy, Gamyyz, Gamyzz,
+                     Gamzxx, Gamzxy, Gamzxz, Gamzyy, Gamzyz, Gamzzz,
+                     Rxx, Rxy, Rxz, Ryy, Ryz, Rzz,
+                     ham_Res, movx_Res, movy_Res, movz_Res,
+                     Gmx_Res, Gmy_Res, Gmz_Res,
+                     Symmetry, Lev, eps, sst, co);
+}
+}
+// Redirect all Z4C shell RHS calls in Step/SHStep to GPU
+#define f_compute_rhs_Z4c_ss cuda_compute_rhs_z4c_ss
+#endif
+
 // This class inherits some members and methods from the parent `bssn_class` and modifies others.
 // The modified members and methods are defined below (and in the header Z4c_class.h).
 // The remaining members/methods are inherited from `bssn_class` (declared in bssn_class.h).
@@ -132,6 +222,13 @@ void Z4c_class::Initialize()
     PhysTime = StartTime;
     Setup_Black_Hole_position();
   }
+
+  sync_cache_pre = new Parallel::SyncCache[GH->levels];
+  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];
 }
 
 //================================================================================================
@@ -165,13 +262,581 @@ Z4c_class::~Z4c_class()
 
 //================================================================================================
 
-#define MRBD 0 // 0: fix BD for meshrefinement level; 1: sommerfeld_bam for them; 2: sommerfeld_yo for them
+#ifndef AMSS_Z4C_MRBD
+#define AMSS_Z4C_MRBD 0
+#endif
+#define MRBD AMSS_Z4C_MRBD // 0: fix BD for meshrefinement level; 1: sommerfeld_bam for them; 2: sommerfeld_yo for them
 
 #ifndef CPBC
 // for sommerfeld boundary
 
+#if USE_CUDA_Z4C && (ABEtype == 2)
+#if (MRBD == 2)
+#error "USE_CUDA_Z4C resident path does not support MRBD == 2"
+#endif
+
+namespace {
+
+static const int k_z4c_cuda_bh_state_indices[3] = {18, 19, 20};
+
+bool fill_z4c_cuda_views(Block *cg, MyList<var> *vars,
+                         double **host_views,
+                         double *propspeeds = 0,
+                         double *soa_flat = 0)
+{
+  int idx = 0;
+  while (vars && idx < Z4C_CUDA_STATE_COUNT)
+  {
+    host_views[idx] = cg->fgfs[vars->data->sgfn];
+    if (propspeeds)
+      propspeeds[idx] = vars->data->propspeed;
+    if (soa_flat)
+    {
+      soa_flat[3 * idx + 0] = vars->data->SoA[0];
+      soa_flat[3 * idx + 1] = vars->data->SoA[1];
+      soa_flat[3 * idx + 2] = vars->data->SoA[2];
+    }
+    vars = vars->next;
+    ++idx;
+  }
+  return idx == Z4C_CUDA_STATE_COUNT && vars == 0;
+}
+
+void z4c_cuda_download_level_state(MyList<Patch> *PatL, MyList<var> *vars, int myrank, bool release_ctx)
+{
+  MyList<Patch> *Pp = PatL;
+  while (Pp)
+  {
+    MyList<Block> *BP = Pp->data->blb;
+    while (BP)
+    {
+      Block *cg = BP->data;
+      if (myrank == cg->rank && z4c_cuda_has_resident_state(cg))
+      {
+        double *state_out[Z4C_CUDA_STATE_COUNT];
+        if (!fill_z4c_cuda_views(cg, vars, state_out))
+        {
+          cout << "CUDA Z4C state list mismatch on resident state download" << endl;
+          MPI_Abort(MPI_COMM_WORLD, 1);
+        }
+        if (z4c_cuda_download_resident_state(cg, cg->shape, state_out))
+        {
+          cout << "CUDA Z4C resident state download failed" << endl;
+          MPI_Abort(MPI_COMM_WORLD, 1);
+        }
+        if (release_ctx)
+          z4c_cuda_release_step_ctx(cg);
+      }
+      if (BP == Pp->data->ble)
+        break;
+      BP = BP->next;
+    }
+    Pp = Pp->next;
+  }
+}
+
+bool z4c_cuda_patch_contains_point(Patch *patch, const double *point)
+{
+  if (!patch)
+    return false;
+  for (int d = 0; d < dim; d++)
+  {
+    const double h = patch->getdX(d);
+    const double lo = patch->bbox[d] + patch->lli[d] * h;
+    const double hi = patch->bbox[dim + d] - patch->uui[d] * h;
+    if (point[d] < lo || point[d] > hi)
+      return false;
+  }
+  return true;
+}
+
+bool z4c_cuda_point_in_block(Patch *patch, Block *block,
+                             const double *point, const double *DH)
+{
+  if (!patch || !block)
+    return false;
+  for (int d = 0; d < dim; d++)
+  {
+    double llb;
+    double uub;
+#ifdef Vertex
+#ifdef Cell
+#error Both Cell and Vertex are defined
+#endif
+    llb = (feq(block->bbox[d], patch->bbox[d], DH[d] / 2))
+              ? block->bbox[d] + patch->lli[d] * DH[d]
+              : block->bbox[d] + (ghost_width - 0.5) * DH[d];
+    uub = (feq(block->bbox[dim + d], patch->bbox[dim + d], DH[d] / 2))
+              ? block->bbox[dim + d] - patch->uui[d] * DH[d]
+              : block->bbox[dim + d] - (ghost_width - 0.5) * DH[d];
+#else
+#ifdef Cell
+    llb = (feq(block->bbox[d], patch->bbox[d], DH[d] / 2))
+              ? block->bbox[d] + patch->lli[d] * DH[d]
+              : block->bbox[d] + ghost_width * DH[d];
+    uub = (feq(block->bbox[dim + d], patch->bbox[dim + d], DH[d] / 2))
+              ? block->bbox[dim + d] - patch->uui[d] * DH[d]
+              : block->bbox[dim + d] - ghost_width * DH[d];
+#else
+#error Not define Vertex nor Cell
+#endif
+#endif
+    if (point[d] - llb < -DH[d] / 2 || point[d] - uub > DH[d] / 2)
+      return false;
+  }
+  return true;
+}
+
+int z4c_cuda_interp_tile_start(const double *coords, int n, double x, double dx, int ordn)
+{
+  if (!coords || n <= ordn)
+    return 0;
+  int cxi = int((x - coords[0]) / dx + 0.4) + 1;
+  int start = cxi - ordn / 2;
+  if (start < 0)
+    start = 0;
+  const int max_start = n - ordn;
+  if (start > max_start)
+    start = max_start;
+  return start;
+}
+
+bool z4c_cuda_interp_bh_point_resident(MyList<Patch> *PatL,
+                                       int myrank,
+                                       const double *point,
+                                       var *forx, var *fory, var *forz,
+                                       int Symmetry,
+                                       double *shellf)
+{
+  const int ordn = 2 * ghost_width;
+  int owner_rank = -1;
+
+  shellf[0] = shellf[1] = shellf[2] = 0.0;
+
+  MyList<Patch> *PL = PatL;
+  while (PL)
+  {
+    Patch *patch = PL->data;
+    if (!z4c_cuda_patch_contains_point(patch, point))
+    {
+      PL = PL->next;
+      continue;
+    }
+
+    double DH[dim];
+    for (int d = 0; d < dim; d++)
+      DH[d] = patch->getdX(d);
+
+    MyList<Block> *BP = patch->blb;
+    while (BP)
+    {
+      Block *block = BP->data;
+      if (z4c_cuda_point_in_block(patch, block, point, DH))
+      {
+        owner_rank = block->rank;
+        if (myrank == owner_rank)
+        {
+          int interp_ordn = ordn;
+          int interp_sym = Symmetry;
+          double x = point[0];
+          double y = point[1];
+          double z = point[2];
+
+          if (z4c_cuda_has_resident_state(block) &&
+              block->shape[0] >= ordn && block->shape[1] >= ordn && block->shape[2] >= ordn)
+          {
+            const int sx = ordn;
+            const int sy = ordn;
+            const int sz = ordn;
+            const int region_all = sx * sy * sz;
+            const int i0 = z4c_cuda_interp_tile_start(block->X[0], block->shape[0], x, DH[0], ordn);
+            const int j0 = z4c_cuda_interp_tile_start(block->X[1], block->shape[1], y, DH[1], ordn);
+            const int k0 = z4c_cuda_interp_tile_start(block->X[2], block->shape[2], z, DH[2], ordn);
+            double *packed_fields = new double[3 * region_all];
+            var *vars[3] = {forx, fory, forz};
+            for (int f = 0; f < 3; f++)
+            {
+              if (z4c_cuda_pack_state_region_to_host_buffer(block,
+                                                            k_z4c_cuda_bh_state_indices[f],
+                                                            packed_fields + f * region_all,
+                                                            block->shape,
+                                                            i0, j0, k0,
+                                                            sx, sy, sz) != 0)
+              {
+                delete[] packed_fields;
+                cout << "CUDA Z4C BH tile download failed" << endl;
+                MPI_Abort(MPI_COMM_WORLD, 1);
+              }
+              int tile_shape[3] = {sx, sy, sz};
+              f_global_interp(tile_shape,
+                              block->X[0] + i0,
+                              block->X[1] + j0,
+                              block->X[2] + k0,
+                              packed_fields + f * region_all,
+                              shellf[f],
+                              x, y, z,
+                              interp_ordn,
+                              vars[f]->SoA,
+                              interp_sym);
+            }
+            delete[] packed_fields;
+          }
+          else
+          {
+            f_global_interp(block->shape, block->X[0], block->X[1], block->X[2],
+                            block->fgfs[forx->sgfn], shellf[0],
+                            x, y, z, interp_ordn, forx->SoA, interp_sym);
+            f_global_interp(block->shape, block->X[0], block->X[1], block->X[2],
+                            block->fgfs[fory->sgfn], shellf[1],
+                            x, y, z, interp_ordn, fory->SoA, interp_sym);
+            f_global_interp(block->shape, block->X[0], block->X[1], block->X[2],
+                            block->fgfs[forz->sgfn], shellf[2],
+                            x, y, z, interp_ordn, forz->SoA, interp_sym);
+          }
+        }
+        break;
+      }
+      if (BP == patch->ble)
+        break;
+      BP = BP->next;
+    }
+
+    if (owner_rank >= 0)
+      break;
+    PL = PL->next;
+  }
+
+  if (owner_rank < 0)
+    return false;
+
+  MPI_Bcast(shellf, 3, MPI_DOUBLE, owner_rank, MPI_COMM_WORLD);
+  return true;
+}
+
+bool z4c_cuda_compute_porg_rhs_resident(cgh *GH,
+                                        int ilev,
+                                        int myrank,
+                                        int BH_num,
+                                        double **BH_PS,
+                                        double **BH_RHS,
+                                        var *forx, var *fory, var *forz,
+                                        int Symmetry)
+{
+  for (int n = 0; n < BH_num; n++)
+  {
+    double shellf[3] = {0.0, 0.0, 0.0};
+    int lev = ilev;
+    while (lev >= 0 &&
+           !z4c_cuda_interp_bh_point_resident(GH->PatL[lev], myrank, BH_PS[n],
+                                              forx, fory, forz, Symmetry, shellf))
+    {
+      --lev;
+    }
+    if (lev < 0)
+      return false;
+    BH_RHS[n][0] = -shellf[0];
+    BH_RHS[n][1] = -shellf[1];
+    BH_RHS[n][2] = -shellf[2];
+  }
+  return true;
+}
+
+bool z4c_cuda_resident_step_enabled()
+{
+  static int enabled = -1;
+  if (enabled < 0)
+  {
+    const char *env = getenv("AMSS_Z4C_CUDA_RESIDENT");
+    enabled = (env && atoi(env) != 0) ? 1 : 0;
+  }
+  return enabled != 0;
+}
+
+} // namespace
+#endif
+
 void Z4c_class::Step(int lev, int YN)
 {
+#if USE_CUDA_Z4C && (ABEtype == 2)
+  double dT_lev = dT * pow(0.5, Mymax(lev, trfls));
+#ifdef With_AHF
+  AH_Step_Find(lev, dT_lev);
+#endif
+  bool BB = fgt(PhysTime, StartTime, dT_lev / 2);
+  double ndeps = numepss;
+  if (lev < GH->movls)
+    ndeps = numepsb;
+  double TRK4 = PhysTime;
+  int iter_count = 0;
+  int pre = 0, cor = 1;
+  int ERROR = 0;
+
+#ifdef WithShell
+  if (bssn_cuda_use_resident_sync(lev))
+  {
+    for (int dl = 0; dl < GH->levels; dl++)
+      bssn_cuda_download_level_state_if_present(GH->PatL[dl], StateList, myrank);
+  }
+#endif
+
+  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 *state_in[Z4C_CUDA_STATE_COUNT];
+        double *state_out[Z4C_CUDA_STATE_COUNT];
+        double propspeed[Z4C_CUDA_STATE_COUNT];
+        double soa_flat[3 * Z4C_CUDA_STATE_COUNT];
+        if (!fill_z4c_cuda_views(cg, StateList, state_in, propspeed, soa_flat) ||
+            !fill_z4c_cuda_views(cg, SynchList_pre, state_out))
+        {
+          cout << "CUDA Z4C state list mismatch on predictor step" << endl;
+          MPI_Abort(MPI_COMM_WORLD, 1);
+        }
+        int apply_bam_bc = 0;
+#if (MRBD == 0)
+#if (SommerType == 0)
+        apply_bam_bc = (lev == 0) ? 1 : 0;
+#endif
+#elif (MRBD == 1)
+        apply_bam_bc = 1;
+#endif
+        int keep_resident_state = z4c_cuda_resident_step_enabled() ? 1 : 0;
+        int apply_enforce_ga = 0;
+#if (AGM == 0)
+        apply_enforce_ga = 1;
+#endif
+        if (z4c_cuda_rk4_substep(cg,
+                                 cg->shape, cg->X[0], cg->X[1], cg->X[2],
+                                 state_in, state_out,
+                                 propspeed, soa_flat, Pp->data->bbox,
+                                 dT_lev, TRK4, iter_count, apply_bam_bc,
+                                 Symmetry, lev, ndeps, pre,
+                                 keep_resident_state, apply_enforce_ga, chitiny))
+        {
+          cout << "CUDA Z4C predictor substep failed in domain: ("
+               << cg->bbox[0] << ":" << cg->bbox[3] << ","
+               << cg->bbox[1] << ":" << cg->bbox[4] << ","
+               << cg->bbox[2] << ":" << cg->bbox[5] << ")" << endl;
+          ERROR = 1;
+        }
+      }
+      if (BP == Pp->data->ble)
+        break;
+      BP = BP->next;
+    }
+    Pp = Pp->next;
+  }
+
+  {
+    int erh = ERROR;
+    MPI_Allreduce(&erh, &ERROR, 1, MPI_INT, MPI_SUM, MPI_COMM_WORLD);
+  }
+  if (ERROR)
+  {
+    if (myrank == 0 && ErrorMonitor->outfile)
+      ErrorMonitor->outfile << "CUDA Z4C failed in predictor at t = " << PhysTime
+                            << ", lev = " << lev << endl;
+    MPI_Abort(MPI_COMM_WORLD, 1);
+  }
+
+  Parallel::Sync_cached(GH->PatL[lev], SynchList_pre, Symmetry, sync_cache_pre[lev]);
+
+  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) && (LastAnas + dT_lev >= AnasTime))
+    z4c_cuda_download_level_state(GH->PatL[lev], SynchList_pre, myrank, false);
+  if (lev == a_lev)
+    AnalysisStuff(lev, dT_lev);
+
+  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 *state_in[Z4C_CUDA_STATE_COUNT];
+          double *state_out[Z4C_CUDA_STATE_COUNT];
+          double propspeed[Z4C_CUDA_STATE_COUNT];
+          double soa_flat[3 * Z4C_CUDA_STATE_COUNT];
+          if (!fill_z4c_cuda_views(cg, SynchList_pre, state_in, propspeed, soa_flat) ||
+              !fill_z4c_cuda_views(cg, SynchList_cor, state_out))
+          {
+            cout << "CUDA Z4C state list mismatch on corrector step" << endl;
+            MPI_Abort(MPI_COMM_WORLD, 1);
+          }
+          int apply_bam_bc = 0;
+#if (MRBD == 0)
+#if (SommerType == 0)
+          apply_bam_bc = (lev == 0) ? 1 : 0;
+#endif
+#elif (MRBD == 1)
+          apply_bam_bc = 1;
+#endif
+          int keep_resident_state = z4c_cuda_resident_step_enabled() ? 1 : 0;
+          int apply_enforce_ga = 0;
+#if (AGM == 0)
+          apply_enforce_ga = 1;
+#elif (AGM == 1)
+          apply_enforce_ga = (iter_count == 3) ? 1 : 0;
+#endif
+          if (z4c_cuda_rk4_substep(cg,
+                                   cg->shape, cg->X[0], cg->X[1], cg->X[2],
+                                   state_in, state_out,
+                                   propspeed, soa_flat, Pp->data->bbox,
+                                   dT_lev, TRK4, iter_count, apply_bam_bc,
+                                   Symmetry, lev, ndeps, cor,
+                                   keep_resident_state, apply_enforce_ga, chitiny))
+          {
+            cout << "CUDA Z4C corrector substep failed in domain: ("
+                 << cg->bbox[0] << ":" << cg->bbox[3] << ","
+                 << cg->bbox[1] << ":" << cg->bbox[4] << ","
+                 << cg->bbox[2] << ":" << cg->bbox[5] << ")" << endl;
+            ERROR = 1;
+          }
+        }
+        if (BP == Pp->data->ble)
+          break;
+        BP = BP->next;
+      }
+      Pp = Pp->next;
+    }
+
+    {
+      int erh = ERROR;
+      MPI_Allreduce(&erh, &ERROR, 1, MPI_INT, MPI_SUM, MPI_COMM_WORLD);
+    }
+    if (ERROR)
+    {
+      if (myrank == 0 && ErrorMonitor->outfile)
+        ErrorMonitor->outfile << "CUDA Z4C failed in RK4 substep#" << iter_count
+                              << " at t = " << PhysTime
+                              << ", lev = " << lev << endl;
+      MPI_Abort(MPI_COMM_WORLD, 1);
+    }
+
+    Parallel::Sync_cached(GH->PatL[lev], SynchList_cor, Symmetry, sync_cache_cor[lev]);
+
+    if (BH_num > 0 && lev == GH->levels - 1)
+    {
+      if (z4c_cuda_resident_step_enabled())
+      {
+        if (!z4c_cuda_compute_porg_rhs_resident(GH, lev, myrank, BH_num,
+                                                Porg, Porg1,
+                                                Sfx, Sfy, Sfz, Symmetry))
+        {
+          if (myrank == 0 && ErrorMonitor->outfile)
+            ErrorMonitor->outfile << "CUDA Z4C failed to interpolate black-hole shift at t = "
+                                  << PhysTime << endl;
+          MPI_Abort(MPI_COMM_WORLD, 1);
+        }
+      }
+      else
+      {
+        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]);
+        }
+      }
+    }
+
+    if (iter_count < 3)
+    {
+      Pp = GH->PatL[lev];
+      while (Pp)
+      {
+        MyList<Block> *BP = Pp->data->blb;
+        while (BP)
+        {
+          Block *cg = BP->data;
+          cg->swapList(SynchList_pre, SynchList_cor, myrank);
+          if (BP == Pp->data->ble)
+            break;
+          BP = BP->next;
+        }
+        Pp = Pp->next;
+      }
+      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];
+        }
+      }
+    }
+  }
+
+  z4c_cuda_download_level_state(GH->PatL[lev], SynchList_cor, myrank, false);
+
+#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 (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];
+    }
+  }
+#else
   double dT_lev = dT * pow(0.5, Mymax(lev, trfls));
 #ifdef With_AHF
   AH_Step_Find(lev, dT_lev);
@@ -339,6 +1004,13 @@ void Z4c_class::Step(int lev, int YN)
   }
 
 #ifdef WithShell
+#if USE_CUDA_Z4C
+  if (bssn_cuda_use_resident_sync(lev))
+  {
+    for (int dl = 0; dl < GH->levels; dl++)
+      bssn_cuda_download_level_state_if_present(GH->PatL[dl], StateList, myrank);
+  }
+#endif
   // evolve Shell Patches
   if (lev == 0)
   {
@@ -1042,9 +1714,11 @@ void Z4c_class::Step(int lev, int YN)
       Porg0[ithBH][2] = Porg1[ithBH][2];
     }
   }
+#endif
 }
 #else
 // for constraint preserving boundary (CPBC)
+// Note: CPBC path uses CPU Fortran RHS; GPU resident sync is a no-op here.
 #ifndef WithShell
 #error "CPBC only supports Shell"
 #endif
@@ -1074,6 +1748,14 @@ void Z4c_class::Step(int lev, int YN)
   int pre = 0, cor = 1;
   int ERROR = 0;
 
+#if USE_CUDA_Z4C && defined(WithShell)
+  if (bssn_cuda_use_resident_sync(lev))
+  {
+    for (int dl = 0; dl < GH->levels; dl++)
+      bssn_cuda_download_level_state_if_present(GH->PatL[dl], StateList, myrank);
+  }
+#endif
+
   MyList<ss_patch> *sPp;
   // Predictor
   MyList<Patch> *Pp = GH->PatL[lev];
@@ -2404,6 +3086,11 @@ void Z4c_class::Check_extrop()
 
 //================================================================================================
 
+#if USE_CUDA_Z4C && (ABEtype == 2) && defined(WithShell)
+#undef f_compute_rhs_Z4c_ss
+#define f_compute_rhs_Z4c_ss compute_rhs_z4c_ss_
+#endif
+
 // this member function is used to compute and output constraint violation
 
 //================================================================================================
@@ -2679,11 +3366,12 @@ void Z4c_class::Interp_Constraint()
   }
 
   ofstream outfile;
-  char filename[50];
-  sprintf(filename, "%s/interp_constraint_%05d.dat", ErrorMonitor->out_dir.c_str(), int(PhysTime / dT + 0.5)); 
+  char suffix[64];
+  sprintf(suffix, "/interp_constraint_%05d.dat", int(PhysTime / dT + 0.5));
+  string filename = ErrorMonitor->out_dir + suffix;
   // 0.5 for round off
   
-  outfile.open(filename);
+  outfile.open(filename.c_str());
   outfile << "#  corrdinate, H_Res, Px_Res, Py_Res, Pz_Res, Gx_Res, Gy_Res, Gz_Res, ...." << endl;
   for (int i = 0; i < n; i++)
   {

AMSS_NCKU_source/bssnEScalar_class.h

@@ -51,7 +51,7 @@ public:
      void Compute_Psi4(int lev);
      void Step(int lev, int YN);
      void AnalysisStuff_EScalar(int lev, double dT_lev);
-     void Interp_Constraint(bool infg);
+     void Interp_Constraint();
      void Constraint_Out(); 
 
 protected:

AMSS_NCKU_source/bssn_class.h

@@ -31,19 +31,11 @@ using namespace std;
 #include "surface_integral.h"
 #include "checkpoint.h"
 
-extern void setpbh(int iBHN, double **iPBH, double *iMass, int rBHN);
-
-#ifndef BSSN_USE_TRANSFER_CACHE
-#define BSSN_USE_TRANSFER_CACHE 1
-#endif
-
-#ifndef BSSN_USE_ESCALAR_C_KERNEL
-#define BSSN_USE_ESCALAR_C_KERNEL 1
-#endif
-
-class bssn_class
-{
-public:
+extern void setpbh(int iBHN, double **iPBH, double *iMass, int rBHN);
+
+class bssn_class
+{
+public:
        int ngfs;
        int nprocs, myrank;
        cgh *GH;
@@ -56,6 +48,7 @@ public:
        double StartTime, TotalTime;
        double AnasTime, DumpTime, d2DumpTime, CheckTime;
        double LastAnas, LastConsOut;
+       bool cuda_level0_constraint_cache_valid;
        int *ConstraintRefreshLevels;
        double Courant;
        double numepss, numepsb, numepsh;
@@ -151,7 +144,7 @@ public:
        bssn_class(double Couranti, double StartTimei, double TotalTimei, double DumpTimei, double d2DumpTimei, double CheckTimei, double AnasTimei,
                   int Symmetryi, int checkruni, char *checkfilenamei, double numepssi, double numepsbi, double numepshi,
                   int a_levi, int maxli, int decni, double maxrexi, double drexi);
-       ~bssn_class();
+       virtual ~bssn_class();
 
        void Evolve(int Steps);
        void RecursiveStep(int lev);
@@ -175,25 +168,14 @@ public:
        void Setup_KerrSchild();
        void Enforce_algcon(int lev, int fg);
 
-       void testRestrict();
-       void testOutBd();
-       
-       bool check_Stdin_Abort(); 
-       bool use_transfer_cache() const;
-       void setup_transfer_caches();
-       void invalidate_transfer_caches();
-       void destroy_transfer_caches();
-       void sync_predictor_start(int lev, MyList<var> *VarList, Parallel::AsyncSyncState &async_state);
-       void sync_predictor_finish(int lev, Parallel::AsyncSyncState &async_state, MyList<var> *VarList);
-       void sync_corrector_start(int lev, MyList<var> *VarList, Parallel::AsyncSyncState &async_state);
-       void sync_corrector_finish(int lev, Parallel::AsyncSyncState &async_state, MyList<var> *VarList);
-       void sync_evolution(int lev, MyList<var> *VarList, Parallel::SyncCache *cache_array = 0);
-       void restrict_evolution(int lev, MyList<var> *src_var_list, MyList<var> *dst_var_list);
-       void outbdlow2hi_evolution(int lev, MyList<var> *src_var_list, MyList<var> *dst_var_list);
-
-       virtual void Setup_Initial_Data_Cao();
-       virtual void Setup_Initial_Data_Lousto();
-       virtual void Initialize();
+       void testRestrict();
+       void testOutBd();
+       
+       bool check_Stdin_Abort(); 
+
+       virtual void Setup_Initial_Data_Cao();
+       virtual void Setup_Initial_Data_Lousto();
+       virtual void Initialize();
        virtual void Read_Ansorg();
        virtual void Read_Pablo() {};
        virtual void Compute_Psi4(int lev);

AMSS_NCKU_source/bssn_em_rhs_c.C

@@ -1,323 +0,0 @@
-#include "macrodef.h"
-#include "bssn_rhs.h"
-#include "share_func.h"
-#include "tool.h"
-#include <cstddef>
-
-/*
- * C 版 BSSN-EM RHS kernel — replaces empart.f90 + bssn_rhs.f90 for BSSN+Maxwell.
- *
- * Computes:
- *   1. All metric and EM field derivatives
- *   2. Physical metric, Christoffel-like terms
- *   3. EM field RHS (E, B, Kpsi, Kphi)
- *   4. Stress-energy tensor (rho, Si, Sij)
- *   5. Calls f_compute_rhs_bssn (C BSSN RHS) with stress-energy
- *   6. Advection + KO dissipation for EM fields
- *   7. NaN check
- */
-int f_compute_rhs_bssn_em_c(int *ex, double &T,
-                            double *X, double *Y, double *Z,
-                            double *chi, double *trK,
-                            double *dxx, double *gxy, double *gxz, double *dyy, double *gyz, double *dzz,
-                            double *Axx, double *Axy, double *Axz, double *Ayy, double *Ayz, double *Azz,
-                            double *Gamx, double *Gamy, double *Gamz,
-                            double *Lap, double *betax, double *betay, double *betaz,
-                            double *dtSfx, double *dtSfy, double *dtSfz,
-                            double *Ex,  double *Ey,  double *Ez,
-                            double *Bx,  double *By,  double *Bz,
-                            double *Kpsi, double *Kphi,
-                            double *Jx, double *Jy, double *Jz, double *qchar,
-                            double *chi_rhs, double *trK_rhs,
-                            double *gxx_rhs, double *gxy_rhs, double *gxz_rhs,
-                            double *gyy_rhs, double *gyz_rhs, double *gzz_rhs,
-                            double *Axx_rhs, double *Axy_rhs, double *Axz_rhs,
-                            double *Ayy_rhs, double *Ayz_rhs, double *Azz_rhs,
-                            double *Gamx_rhs, double *Gamy_rhs, double *Gamz_rhs,
-                            double *Lap_rhs, double *betax_rhs, double *betay_rhs, double *betaz_rhs,
-                            double *dtSfx_rhs, double *dtSfy_rhs, double *dtSfz_rhs,
-                            double *Ex_rhs,  double *Ey_rhs,  double *Ez_rhs,
-                            double *Bx_rhs,  double *By_rhs,  double *Bz_rhs,
-                            double *Kpsi_rhs, double *Kphi_rhs,
-                            double *rho, double *Sx, double *Sy, double *Sz,
-                            double *Sxx, double *Sxy, double *Sxz,
-                            double *Syy, double *Syz, double *Szz,
-                            double *Gamxxx, double *Gamxxy, double *Gamxxz,
-                            double *Gamxyy, double *Gamxyz, double *Gamxzz,
-                            double *Gamyxx, double *Gamyxy, double *Gamyxz,
-                            double *Gamyyy, double *Gamyyz, double *Gamyzz,
-                            double *Gamzxx, double *Gamzxy, double *Gamzxz,
-                            double *Gamzyy, double *Gamzyz, double *Gamzzz,
-                            double *Rxx, double *Rxy, double *Rxz,
-                            double *Ryy, double *Ryz, double *Rzz,
-                            double *ham_Res, double *movx_Res, double *movy_Res, double *movz_Res,
-                            double *Gmx_Res, double *Gmy_Res, double *Gmz_Res,
-                            int &Symmetry, int &Lev, double &eps, int &co)
-{
-    (void)T;
-    int gont = 0;
-    const int nx = ex[0], ny = ex[1], nz = ex[2];
-    const int all = nx * ny * nz;
-    const size_t n = (size_t)all;
-
-    const double ZEO = 0.0, ONE = 1.0, TWO = 2.0, FOUR = 4.0, EIT = 8.0;
-    const double HALF = 0.5, THR = 3.0, F3o2 = 1.5, PI = 3.14159265358979323846;
-    const double SYM = 1.0, ANTI = -1.0;
-    const double kappa = 1.0;
-    const double SSS[3]={SYM,SYM,SYM},   AAS[3]={ANTI,ANTI,SYM};
-    const double ASA[3]={ANTI,SYM,ANTI}, SAA[3]={SYM,ANTI,ANTI};
-    const double ASS[3]={ANTI,SYM,SYM},   SAS[3]={SYM,ANTI,SYM};
-    const double SSA[3]={SYM,SYM,ANTI};
-
-    /* ---- allocate temporary arrays ---- */
-    double *chix = (double*)malloc(n*sizeof(double));
-    double *chiy = (double*)malloc(n*sizeof(double));
-    double *chiz = (double*)malloc(n*sizeof(double));
-    double *Exx=(double*)malloc(n*sizeof(double)),*Exy=(double*)malloc(n*sizeof(double)),*Exz=(double*)malloc(n*sizeof(double));
-    double *Eyx=(double*)malloc(n*sizeof(double)),*Eyy=(double*)malloc(n*sizeof(double)),*Eyz=(double*)malloc(n*sizeof(double));
-    double *Ezx=(double*)malloc(n*sizeof(double)),*Ezy=(double*)malloc(n*sizeof(double)),*Ezz=(double*)malloc(n*sizeof(double));
-    double *Bxx=(double*)malloc(n*sizeof(double)),*Bxy=(double*)malloc(n*sizeof(double)),*Bxz=(double*)malloc(n*sizeof(double));
-    double *Byx=(double*)malloc(n*sizeof(double)),*Byy=(double*)malloc(n*sizeof(double)),*Byz=(double*)malloc(n*sizeof(double));
-    double *Bzx=(double*)malloc(n*sizeof(double)),*Bzy=(double*)malloc(n*sizeof(double)),*Bzz=(double*)malloc(n*sizeof(double));
-    double *Kpsix=(double*)malloc(n*sizeof(double)),*Kpsiy=(double*)malloc(n*sizeof(double)),*Kpsiz=(double*)malloc(n*sizeof(double));
-    double *Kphix=(double*)malloc(n*sizeof(double)),*Kphiy=(double*)malloc(n*sizeof(double)),*Kphiz=(double*)malloc(n*sizeof(double));
-    double *Lapx=(double*)malloc(n*sizeof(double)),*Lapy=(double*)malloc(n*sizeof(double)),*Lapz=(double*)malloc(n*sizeof(double));
-    double *betaxx=(double*)malloc(n*sizeof(double)),*betaxy=(double*)malloc(n*sizeof(double)),*betaxz=(double*)malloc(n*sizeof(double));
-    double *betayx=(double*)malloc(n*sizeof(double)),*betayy=(double*)malloc(n*sizeof(double)),*betayz=(double*)malloc(n*sizeof(double));
-    double *betazx=(double*)malloc(n*sizeof(double)),*betazy=(double*)malloc(n*sizeof(double)),*betazz=(double*)malloc(n*sizeof(double));
-    double *gxxx=(double*)malloc(n*sizeof(double)),*gxxy=(double*)malloc(n*sizeof(double)),*gxxz=(double*)malloc(n*sizeof(double));
-    double *gxyx=(double*)malloc(n*sizeof(double)),*gxyy=(double*)malloc(n*sizeof(double)),*gxyz=(double*)malloc(n*sizeof(double));
-    double *gxzx=(double*)malloc(n*sizeof(double)),*gxzy=(double*)malloc(n*sizeof(double)),*gxzz=(double*)malloc(n*sizeof(double));
-    double *gyyx=(double*)malloc(n*sizeof(double)),*gyyy=(double*)malloc(n*sizeof(double)),*gyyz=(double*)malloc(n*sizeof(double));
-    double *gyzx=(double*)malloc(n*sizeof(double)),*gyzy=(double*)malloc(n*sizeof(double)),*gyzz=(double*)malloc(n*sizeof(double));
-    double *gzzx=(double*)malloc(n*sizeof(double)),*gzzy=(double*)malloc(n*sizeof(double)),*gzzz=(double*)malloc(n*sizeof(double));
-    double *gupxx=(double*)malloc(n*sizeof(double)),*gupxy=(double*)malloc(n*sizeof(double)),*gupxz=(double*)malloc(n*sizeof(double));
-    double *gupyy=(double*)malloc(n*sizeof(double)),*gupyz=(double*)malloc(n*sizeof(double)),*gupzz=(double*)malloc(n*sizeof(double));
-
-    if (!chix||!chiy||!chiz||!Exx||!Exy||!Exz||!Eyx||!Eyy||!Eyz||!Ezx||!Ezy||!Ezz||
-        !Bxx||!Bxy||!Bxz||!Byx||!Byy||!Byz||!Bzx||!Bzy||!Bzz||
-        !Kpsix||!Kpsiy||!Kpsiz||!Kphix||!Kphiy||!Kphiz||
-        !Lapx||!Lapy||!Lapz||
-        !betaxx||!betaxy||!betaxz||!betayx||!betayy||!betayz||!betazx||!betazy||!betazz||
-        !gxxx||!gxxy||!gxxz||!gxyx||!gxyy||!gxyz||!gxzx||!gxzy||!gxzz||
-        !gyyx||!gyyy||!gyyz||!gyzx||!gyzy||!gyzz||!gzzx||!gzzy||!gzzz||
-        !gupxx||!gupxy||!gupxz||!gupyy||!gupyz||!gupzz) {
-        gont = 1;
-    }
-
-    /* ==== 1. Compute all derivatives ==== */
-    if (!gont) {
-
-    /* metric derivatives */
-    fderivs(ex, Lap, Lapx, Lapy, Lapz, X, Y, Z, SYM, SYM, SYM, Symmetry, Lev);
-    fderivs(ex, betax, betaxx, betaxy, betaxz, X, Y, Z, ANTI, SYM, SYM, Symmetry, Lev);
-    fderivs(ex, betay, betayx, betayy, betayz, X, Y, Z, SYM, ANTI, SYM, Symmetry, Lev);
-    fderivs(ex, betaz, betazx, betazy, betazz, X, Y, Z, SYM, SYM, ANTI, Symmetry, Lev);
-    fderivs(ex, chi, chix, chiy, chiz, X, Y, Z, SYM, SYM, SYM, Symmetry, Lev);
-    fderivs(ex, dxx, gxxx, gxxy, gxxz, X, Y, Z, SYM, SYM, SYM, Symmetry, Lev);
-    fderivs(ex, gxy, gxyx, gxyy, gxyz, X, Y, Z, ANTI, ANTI, SYM, Symmetry, Lev);
-    fderivs(ex, gxz, gxzx, gxzy, gxzz, X, Y, Z, ANTI, SYM, ANTI, Symmetry, Lev);
-    fderivs(ex, dyy, gyyx, gyyy, gyyz, X, Y, Z, SYM, SYM, SYM, Symmetry, Lev);
-    fderivs(ex, gyz, gyzx, gyzy, gyzz, X, Y, Z, SYM, ANTI, ANTI, Symmetry, Lev);
-    fderivs(ex, dzz, gzzx, gzzy, gzzz, X, Y, Z, SYM, SYM, SYM, Symmetry, Lev);
-
-    /* EM field derivatives */
-    fderivs(ex, Kpsi, Kpsix, Kpsiy, Kpsiz, X, Y, Z, SYM, SYM, SYM, Symmetry, Lev);
-    fderivs(ex, Kphi, Kphix, Kphiy, Kphiz, X, Y, Z, SYM, SYM, SYM, Symmetry, Lev);
-    fderivs(ex, Ex, Exx, Exy, Exz, X, Y, Z, ANTI, SYM, SYM, Symmetry, Lev);
-    fderivs(ex, Ey, Eyx, Eyy, Eyz, X, Y, Z, SYM, ANTI, SYM, Symmetry, Lev);
-    fderivs(ex, Ez, Ezx, Ezy, Ezz, X, Y, Z, SYM, SYM, ANTI, Symmetry, Lev);
-    fderivs(ex, Bx, Bxx, Bxy, Bxz, X, Y, Z, SYM, ANTI, ANTI, Symmetry, Lev);
-    fderivs(ex, By, Byx, Byy, Byz, X, Y, Z, ANTI, SYM, ANTI, Symmetry, Lev);
-    fderivs(ex, Bz, Bzx, Bzy, Bzz, X, Y, Z, ANTI, ANTI, SYM, Symmetry, Lev);
-
-    /* ==== 2. Compute EM RHS and stress-energy ==== */
-    const double F1o4PI = ONE / (FOUR * PI);
-    for (size_t i = 0; i < n; ++i) {
-        const double alpn1 = Lap[i] + ONE;
-        const double chin1 = chi[i] + ONE;
-        const double chi3o2 = sqrt(chin1) * chin1;  // chi^{3/2}
-        const double ichi = ONE / chin1;
-
-        /* physical metric */
-        const double pgxx = (dxx[i] + ONE) * ichi;
-        const double pgyy = (dyy[i] + ONE) * ichi;
-        const double pgzz = (dzz[i] + ONE) * ichi;
-        const double pgxy = gxy[i] * ichi;
-        const double pgxz = gxz[i] * ichi;
-        const double pgyz = gyz[i] * ichi;
-
-        /* inverse physical metric */
-        const double det = pgxx * pgyy * pgzz + pgxy * pgyz * pgxz + pgxz * pgxy * pgyz
-                         - pgxz * pgyy * pgxz - pgxy * pgxy * pgzz - pgxx * pgyz * pgyz;
-        const double idet = ONE / det;
-        const double upxx = (pgyy * pgzz - pgyz * pgyz) * idet;
-        const double upxy = -(pgxy * pgzz - pgyz * pgxz) * idet;
-        const double upxz = (pgxy * pgyz - pgyy * pgxz) * idet;
-        const double upyy = (pgxx * pgzz - pgxz * pgxz) * idet;
-        const double upyz = -(pgxx * pgyz - pgxy * pgxz) * idet;
-        const double upzz = (pgxx * pgyy - pgxy * pgxy) * idet;
-        gupxx[i]=upxx; gupxy[i]=upxy; gupxz[i]=upxz;
-        gupyy[i]=upyy; gupyz[i]=upyz; gupzz[i]=upzz;
-
-        /* E-field RHS */
-        /* curl(B) part: epsilon^{ijk} ∂_j (alpha * B_k) in coordinate basis */
-        /* Using lower-index B fields: B_i_lower = pg_{ij} * B^j */
-        const double BxL = pgxx*Bx[i] + pgxy*By[i] + pgxz*Bz[i];
-        const double ByL = pgxy*Bx[i] + pgyy*By[i] + pgyz*Bz[i];
-        const double BzL = pgxz*Bx[i] + pgyz*By[i] + pgzz*Bz[i];
-
-        /* Physical metric derivatives (chain rule from conformal) */
-        const double pgxx_x = (gxxx[i] - pgxx*chix[i]) * ichi;
-        /* const double pgxx_y = (gxxy[i] - pgxx*chiy[i]) * ichi; */
-        const double pgxy_x = (gxyx[i] - pgxy*chix[i]) * ichi;
-        const double pgxy_y = (gxyy[i] - pgxy*chiy[i]) * ichi;
-        const double pgxz_x = (gxzx[i] - pgxz*chix[i]) * ichi;
-        const double pgxz_z = (gxzz[i] - pgxz*chiz[i]) * ichi;
-        const double pgyy_y = (gyyy[i] - pgyy*chiy[i]) * ichi;
-        const double pgyz_y = (gyzy[i] - pgyz*chiy[i]) * ichi;
-        const double pgyz_z = (gyzz[i] - pgyz*chiz[i]) * ichi;
-        const double pgzz_z = (gzzz[i] - pgzz*chiz[i]) * ichi;
-
-        /* Curl_x(B) = ∂_y (alpha*BzL) - ∂_z (alpha*ByL) */
-        const double aBx = alpn1*BxL, aBy = alpn1*ByL, aBz = alpn1*BzL;
-        const double curlBx = (aBz*Lapy[i] + alpn1*(pgxz*Bxy[i]+pgyz*Byy[i]+pgzz*Bzy[i]) + alpn1*(Bx[i]*gxzy[i]+By[i]*gyzy[i]+Bz[i]*gzzy[i]))
-                            - (aBy*Lapz[i] + alpn1*(pgxy*Bxz[i]+pgyy*Byz[i]+pgyz*Bzz[i]) + alpn1*(Bx[i]*gxyz[i]+By[i]*gyyz[i]+Bz[i]*gyzz[i]));
-        double curlBy = (aBx*Lapz[i] + alpn1*(pgxx*Bxz[i]+pgxy*Byz[i]+pgxz*Bzz[i]) + alpn1*(Bx[i]*gxxz[i]+By[i]*gxyz[i]+Bz[i]*gxzz[i]))
-                      - (aBz*Lapx[i] + alpn1*(pgxz*Bxx[i]+pgyz*Byx[i]+pgzz*Bzx[i]) + alpn1*(Bx[i]*gxzx[i]+By[i]*gyzx[i]+Bz[i]*gzzx[i]));
-        double curlBz = (aBy*Lapx[i] + alpn1*(pgxy*Bxx[i]+pgyy*Byx[i]+pgyz*Bzx[i]) + alpn1*(Bx[i]*gxyx[i]+By[i]*gyyx[i]+Bz[i]*gyzx[i]))
-                      - (aBx*Lapy[i] + alpn1*(pgxx*Bxy[i]+pgxy*Byy[i]+pgxz*Bzy[i]) + alpn1*(Bx[i]*gxxy[i]+By[i]*gxyy[i]+Bz[i]*gxzy[i]));
-
-        /* Advection part: -beta^j * ∂_j E^i */
-        const double advEx = Ex[i]*betaxx[i] + Ey[i]*betaxy[i] + Ez[i]*betaxz[i];
-        const double advEy = Ex[i]*betayx[i] + Ey[i]*betayy[i] + Ez[i]*betayz[i];
-        const double advEz = Ex[i]*betazx[i] + Ey[i]*betazy[i] + Ez[i]*betazz[i];
-
-        /* grad(Kpsi) contracted with inverse metric */
-        const double gupKx = upxx*Kpsix[i] + upxy*Kpsiy[i] + upxz*Kpsiz[i];
-        const double gupKy = upxy*Kpsix[i] + upyy*Kpsiy[i] + upyz*Kpsiz[i];
-        const double gupKz = upxz*Kpsix[i] + upyz*Kpsiy[i] + upzz*Kpsiz[i];
-
-        Ex_rhs[i] = alpn1*trK[i]*Ex[i] - advEx - FOUR*PI*alpn1*Jx[i] - alpn1*gupKx + chi3o2*curlBx;
-        Ey_rhs[i] = alpn1*trK[i]*Ey[i] - advEy - FOUR*PI*alpn1*Jy[i] - alpn1*gupKy + chi3o2*curlBy;
-        Ez_rhs[i] = alpn1*trK[i]*Ez[i] - advEz - FOUR*PI*alpn1*Jz[i] - alpn1*gupKz + chi3o2*curlBz;
-
-        /* B-field RHS: similar but with -chi^{3/2} * curl(E) and grad(Kphi) */
-        const double ExL = pgxx*Ex[i] + pgxy*Ey[i] + pgxz*Ez[i];
-        const double EyL = pgxy*Ex[i] + pgyy*Ey[i] + pgyz*Ez[i];
-        const double EzL = pgxz*Ex[i] + pgyz*Ey[i] + pgzz*Ez[i];
-
-        const double aEx = alpn1*ExL, aEy = alpn1*EyL, aEz = alpn1*EzL;
-        const double curlEx = (aEz*Lapy[i] + alpn1*(pgxz*Exy[i]+pgyz*Eyy[i]+pgzz*Ezy[i]) + alpn1*(Ex[i]*gxzy[i]+Ey[i]*gyzy[i]+Ez[i]*gzzy[i]))
-                            - (aEy*Lapz[i] + alpn1*(pgxy*Exz[i]+pgyy*Eyz[i]+pgyz*Ezz[i]) + alpn1*(Ex[i]*gxyz[i]+Ey[i]*gyyz[i]+Ez[i]*gyzz[i]));
-        double curlEy = (aEx*Lapz[i] + alpn1*(pgxx*Exz[i]+pgxy*Eyz[i]+pgxz*Ezz[i]) + alpn1*(Ex[i]*gxxz[i]+Ey[i]*gxyz[i]+Ez[i]*gxzz[i]))
-                      - (aEz*Lapx[i] + alpn1*(pgxz*Exx[i]+pgyz*Eyx[i]+pgzz*Ezx[i]) + alpn1*(Ex[i]*gxzx[i]+Ey[i]*gyzx[i]+Ez[i]*gzzx[i]));
-        double curlEz = (aEy*Lapx[i] + alpn1*(pgxy*Exx[i]+pgyy*Eyx[i]+pgyz*Ezx[i]) + alpn1*(Ex[i]*gxyx[i]+Ey[i]*gyyx[i]+Ez[i]*gyzx[i]))
-                      - (aEx*Lapy[i] + alpn1*(pgxx*Exy[i]+pgxy*Eyy[i]+pgxz*Ezy[i]) + alpn1*(Ex[i]*gxxy[i]+Ey[i]*gxyy[i]+Ez[i]*gxzy[i]));
-
-        const double advBx = Bx[i]*betaxx[i] + By[i]*betaxy[i] + Bz[i]*betaxz[i];
-        const double advBy = Bx[i]*betayx[i] + By[i]*betayy[i] + Bz[i]*betayz[i];
-        const double advBz = Bx[i]*betazx[i] + By[i]*betazy[i] + Bz[i]*betazz[i];
-
-        const double gupKphix = upxx*Kphix[i] + upxy*Kphiy[i] + upxz*Kphiz[i];
-        const double gupKphiy = upxy*Kphix[i] + upyy*Kphiy[i] + upyz*Kphiz[i];
-        const double gupKphiz = upxz*Kphix[i] + upyz*Kphiy[i] + upzz*Kphiz[i];
-
-        Bx_rhs[i] = alpn1*trK[i]*Bx[i] - advBx - alpn1*gupKphix - chi3o2*curlEx;
-        By_rhs[i] = alpn1*trK[i]*By[i] - advBy - alpn1*gupKphiy - chi3o2*curlEy;
-        Bz_rhs[i] = alpn1*trK[i]*Bz[i] - advBz - alpn1*gupKphiz - chi3o2*curlEz;
-
-        /* Scalar potential RHS */
-        const double divE = Exx[i] + Eyy[i] + Ezz[i];
-        const double divB = Bxx[i] + Byy[i] + Bzz[i];
-        const double chiCont = F3o2 * ichi * (chix[i]*Ex[i] + chiy[i]*Ey[i] + chiz[i]*Ez[i]);
-        Kpsi_rhs[i] = FOUR*PI*alpn1*qchar[i] - alpn1*kappa*Kpsi[i] - alpn1*(divE - chiCont);
-        Kphi_rhs[i] = -alpn1*kappa*Kphi[i] - alpn1*(divB - F3o2*ichi*(chix[i]*Bx[i] + chiy[i]*By[i] + chiz[i]*Bz[i]));
-
-        /* Stress-energy tensor */
-        const double E2 = pgxx*Ex[i]*Ex[i] + pgyy*Ey[i]*Ey[i] + pgzz*Ez[i]*Ez[i]
-                        + TWO*(pgxy*Ex[i]*Ey[i] + pgxz*Ex[i]*Ez[i] + pgyz*Ey[i]*Ez[i]);
-        const double B2 = pgxx*Bx[i]*Bx[i] + pgyy*By[i]*By[i] + pgzz*Bz[i]*Bz[i]
-                        + TWO*(pgxy*Bx[i]*By[i] + pgxz*Bx[i]*Bz[i] + pgyz*By[i]*Bz[i]);
-        rho[i] = (E2 + B2) / (EIT * PI);
-        const double ichi3o2 = ONE / chi3o2;
-        Sx[i] = (Ey[i]*Bz[i] - Ez[i]*By[i]) * F1o4PI * ichi3o2;
-        Sy[i] = (Ez[i]*Bx[i] - Ex[i]*Bz[i]) * F1o4PI * ichi3o2;
-        Sz[i] = (Ex[i]*By[i] - Ey[i]*Bx[i]) * F1o4PI * ichi3o2;
-        const double lExi = pgxx*Ex[i] + pgxy*Ey[i] + pgxz*Ez[i];
-        const double lEyi = pgxy*Ex[i] + pgyy*Ey[i] + pgyz*Ez[i];
-        const double lEzi = pgxz*Ex[i] + pgyz*Ey[i] + pgzz*Ez[i];
-        const double lBxi = pgxx*Bx[i] + pgxy*By[i] + pgxz*Bz[i];
-        const double lByi = pgxy*Bx[i] + pgyy*By[i] + pgyz*Bz[i];
-        const double lBzi = pgxz*Bx[i] + pgyz*By[i] + pgzz*Bz[i];
-        Sxx[i] = rho[i]*pgxx - (lExi*lExi + lBxi*lBxi) * F1o4PI;
-        Sxy[i] = rho[i]*pgxy - (lExi*lEyi + lBxi*lByi) * F1o4PI;
-        Sxz[i] = rho[i]*pgxz - (lExi*lEzi + lBxi*lBzi) * F1o4PI;
-        Syy[i] = rho[i]*pgyy - (lEyi*lEyi + lByi*lByi) * F1o4PI;
-        Syz[i] = rho[i]*pgyz - (lEyi*lEzi + lByi*lBzi) * F1o4PI;
-        Szz[i] = rho[i]*pgzz - (lEzi*lEzi + lBzi*lBzi) * F1o4PI;
-    }
-
-    /* ==== 3. Call BSSN RHS with EM stress-energy ==== */
-    gont = f_compute_rhs_bssn(ex, T, X, Y, Z,
-        chi, trK, dxx, gxy, gxz, dyy, gyz, dzz,
-        Axx, Axy, Axz, Ayy, Ayz, Azz,
-        Gamx, Gamy, Gamz, Lap, betax, betay, betaz, dtSfx, dtSfy, dtSfz,
-        chi_rhs, trK_rhs,
-        gxx_rhs, gxy_rhs, gxz_rhs, gyy_rhs, gyz_rhs, gzz_rhs,
-        Axx_rhs, Axy_rhs, Axz_rhs, Ayy_rhs, Ayz_rhs, Azz_rhs,
-        Gamx_rhs, Gamy_rhs, Gamz_rhs, Lap_rhs, betax_rhs, betay_rhs, betaz_rhs,
-        dtSfx_rhs, dtSfy_rhs, dtSfz_rhs,
-        rho, Sx, Sy, Sz, Sxx, Sxy, Sxz, Syy, Syz, Szz,
-        Gamxxx, Gamxxy, Gamxxz, Gamxyy, Gamxyz, Gamxzz,
-        Gamyxx, Gamyxy, Gamyxz, Gamyyy, Gamyyz, Gamyzz,
-        Gamzxx, Gamzxy, Gamzxz, Gamzyy, Gamzyz, Gamzzz,
-        Rxx, Rxy, Rxz, Ryy, Ryz, Rzz,
-        ham_Res, movx_Res, movy_Res, movz_Res,
-        Gmx_Res, Gmy_Res, Gmz_Res,
-        Symmetry, Lev, eps, co);
-    if (!gont) {
-
-    /* ==== 4. Advection terms for EM fields ==== */
-    lopsided(ex, X, Y, Z, Kpsi, Kpsi_rhs, betax, betay, betaz, Symmetry, SSS);
-    lopsided(ex, X, Y, Z, Kphi, Kphi_rhs, betax, betay, betaz, Symmetry, SSS);
-    lopsided(ex, X, Y, Z, Ex,   Ex_rhs,   betax, betay, betaz, Symmetry, ASS);
-    lopsided(ex, X, Y, Z, Ey,   Ey_rhs,   betax, betay, betaz, Symmetry, SAS);
-    lopsided(ex, X, Y, Z, Ez,   Ez_rhs,   betax, betay, betaz, Symmetry, SSA);
-    lopsided(ex, X, Y, Z, Bx,   Bx_rhs,   betax, betay, betaz, Symmetry, SAA);
-    lopsided(ex, X, Y, Z, By,   By_rhs,   betax, betay, betaz, Symmetry, ASA);
-    lopsided(ex, X, Y, Z, Bz,   Bz_rhs,   betax, betay, betaz, Symmetry, AAS);
-
-    /* ==== 5. KO dissipation for EM fields ==== */
-    if (eps > ZEO) {
-        kodis(ex, X, Y, Z, Kpsi, Kpsi_rhs, SSS, Symmetry, eps);
-        kodis(ex, X, Y, Z, Kphi, Kphi_rhs, SSS, Symmetry, eps);
-        kodis(ex, X, Y, Z, Ex,   Ex_rhs,   ASS, Symmetry, eps);
-        kodis(ex, X, Y, Z, Ey,   Ey_rhs,   SAS, Symmetry, eps);
-        kodis(ex, X, Y, Z, Ez,   Ez_rhs,   SSA, Symmetry, eps);
-        kodis(ex, X, Y, Z, Bx,   Bx_rhs,   SAA, Symmetry, eps);
-        kodis(ex, X, Y, Z, By,   By_rhs,   ASA, Symmetry, eps);
-        kodis(ex, X, Y, Z, Bz,   Bz_rhs,   AAS, Symmetry, eps);
-    }
-
-    /* ==== 6. NaN check ==== */
-        for (int i = 0; i < all; ++i) {
-            if (!isfinite(Ex_rhs[i]+Ey_rhs[i]+Ez_rhs[i]+Bx_rhs[i]+By_rhs[i]+Bz_rhs[i]+Kpsi_rhs[i]+Kphi_rhs[i])) {
-                gont = 1; break;
-            }
-        }
-        } /* inner if (!gont) */
-    } /* outer if (!gont) */
-
-    free(chix);free(chiy);free(chiz);
-    free(Exx);free(Exy);free(Exz);free(Eyx);free(Eyy);free(Eyz);free(Ezx);free(Ezy);free(Ezz);
-    free(Bxx);free(Bxy);free(Bxz);free(Byx);free(Byy);free(Byz);free(Bzx);free(Bzy);free(Bzz);
-    free(Kpsix);free(Kpsiy);free(Kpsiz);
-    free(Kphix);free(Kphiy);free(Kphiz);
-    free(Lapx);free(Lapy);free(Lapz);
-    free(betaxx);free(betaxy);free(betaxz);free(betayx);free(betayy);free(betayz);free(betazx);free(betazy);free(betazz);
-    free(gxxx);free(gxxy);free(gxxz);free(gxyx);free(gxyy);free(gxyz);free(gxzx);free(gxzy);free(gxzz);
-    free(gyyx);free(gyyy);free(gyyz);free(gyzx);free(gyzy);free(gyzz);free(gzzx);free(gzzy);free(gzzz);
-    free(gupxx);free(gupxy);free(gupxz);free(gupyy);free(gupyz);free(gupzz);
-    return gont;
-}

AMSS_NCKU_source/bssn_escalar_rhs_c.C

@@ -1,169 +0,0 @@
-#include "macrodef.h"
-#include "bssn_rhs.h"
-#include "share_func.h"
-#include "tool.h"
-#include <vector>
-
-namespace
-{
-    // Reuse the temporary workspace across block calls to avoid repeated heap churn
-    // in the EScalar wrapper. MPI ranks execute this path sequentially, so a single
-    // process-local buffer is sufficient here.
-    std::vector<double> g_escalar_tmp_store;
-}
-
-#ifdef fortran1
-#define f_frpotential frpotential
-#endif
-#ifdef fortran2
-#define f_frpotential FRPOTENTIAL
-#endif
-#ifdef fortran3
-#define f_frpotential frpotential_
-#endif
-
-extern "C"
-{
-    void f_frpotential(int *, double *, double *, double *);
-}
-
-int f_compute_rhs_bssn_escalar_c(int *ex, double &T,
-                                 double *X, double *Y, double *Z,
-                                 double *chi, double *trK,
-                                 double *dxx, double *gxy, double *gxz, double *dyy, double *gyz, double *dzz,
-                                 double *Axx, double *Axy, double *Axz, double *Ayy, double *Ayz, double *Azz,
-                                 double *Gamx, double *Gamy, double *Gamz,
-                                 double *Lap, double *betax, double *betay, double *betaz,
-                                 double *dtSfx, double *dtSfy, double *dtSfz,
-                                 double *Sphi, double *Spi,
-                                 double *chi_rhs, double *trK_rhs,
-                                 double *gxx_rhs, double *gxy_rhs, double *gxz_rhs, double *gyy_rhs, double *gyz_rhs, double *gzz_rhs,
-                                 double *Axx_rhs, double *Axy_rhs, double *Axz_rhs, double *Ayy_rhs, double *Ayz_rhs, double *Azz_rhs,
-                                 double *Gamx_rhs, double *Gamy_rhs, double *Gamz_rhs,
-                                 double *Lap_rhs, double *betax_rhs, double *betay_rhs, double *betaz_rhs,
-                                 double *dtSfx_rhs, double *dtSfy_rhs, double *dtSfz_rhs,
-                                 double *Sphi_rhs, double *Spi_rhs,
-                                 double *rho, double *Sx, double *Sy, double *Sz,
-                                 double *Sxx, double *Sxy, double *Sxz, double *Syy, double *Syz, double *Szz,
-                                 double *Gamxxx, double *Gamxxy, double *Gamxxz, double *Gamxyy, double *Gamxyz, double *Gamxzz,
-                                 double *Gamyxx, double *Gamyxy, double *Gamyxz, double *Gamyyy, double *Gamyyz, double *Gamyzz,
-                                 double *Gamzxx, double *Gamzxy, double *Gamzxz, double *Gamzyy, double *Gamzyz, double *Gamzzz,
-                                 double *Rxx, double *Rxy, double *Rxz, double *Ryy, double *Ryz, double *Rzz,
-                                 double *ham_Res, double *movx_Res, double *movy_Res, double *movz_Res,
-                                 double *Gmx_Res, double *Gmy_Res, double *Gmz_Res,
-                                 int &Symmetry, int &Lev, double &eps, int &co)
-{
-    const int nx = ex[0], ny = ex[1], nz = ex[2];
-    const int all = nx * ny * nz;
-
-    const size_t workspace_size = size_t(all) * 17;
-    if (g_escalar_tmp_store.size() < workspace_size)
-        g_escalar_tmp_store.resize(workspace_size);
-
-    double *tmp_ptr = g_escalar_tmp_store.data();
-    auto alloc_tmp = [&](int n = 1) -> double *
-    {
-        double *ptr = tmp_ptr;
-        tmp_ptr += size_t(all) * n;
-        return ptr;
-    };
-
-    double *chix = alloc_tmp(), *chiy = alloc_tmp(), *chiz = alloc_tmp();
-    double *Kx = alloc_tmp(), *Ky = alloc_tmp(), *Kz = alloc_tmp();
-    double *fxx = alloc_tmp(), *fxy = alloc_tmp(), *fxz = alloc_tmp();
-    double *fyy = alloc_tmp(), *fyz = alloc_tmp(), *fzz = alloc_tmp();
-    double *Lapx = alloc_tmp(), *Lapy = alloc_tmp(), *Lapz = alloc_tmp();
-    double *V = alloc_tmp(), *dVdSphi = alloc_tmp();
-
-    const double ZEO = 0.0, ONE = 1.0, TWO = 2.0, HALF = 0.5;
-    const double SSS[3] = {1.0, 1.0, 1.0};
-
-    fderivs(ex, chi, chix, chiy, chiz, X, Y, Z, 1.0, 1.0, 1.0, Symmetry, Lev);
-    fderivs(ex, Lap, Lapx, Lapy, Lapz, X, Y, Z, 1.0, 1.0, 1.0, Symmetry, Lev);
-    fderivs(ex, Sphi, Kx, Ky, Kz, X, Y, Z, 1.0, 1.0, 1.0, Symmetry, Lev);
-    fdderivs(ex, Sphi, fxx, fxy, fxz, fyy, fyz, fzz, X, Y, Z, 1.0, 1.0, 1.0, Symmetry, Lev);
-
-    f_frpotential(ex, Sphi, V, dVdSphi);
-
-    for (int i = 0; i < all; ++i)
-    {
-        const double alpn1 = Lap[i] + ONE;
-        const double chin1 = chi[i] + ONE;
-        const double gxx = dxx[i] + ONE;
-        const double gyy = dyy[i] + ONE;
-        const double gzz = dzz[i] + ONE;
-        const double det = gxx * gyy * gzz + gxy[i] * gyz[i] * gxz[i] + gxz[i] * gxy[i] * gyz[i]
-                         - gxz[i] * gyy * gxz[i] - gxy[i] * gxy[i] * gzz - gxx * gyz[i] * gyz[i];
-        const double gupxx = (gyy * gzz - gyz[i] * gyz[i]) / det;
-        const double gupxy = -(gxy[i] * gzz - gyz[i] * gxz[i]) / det;
-        const double gupxz = (gxy[i] * gyz[i] - gyy * gxz[i]) / det;
-        const double gupyy = (gxx * gzz - gxz[i] * gxz[i]) / det;
-        const double gupyz = -(gxx * gyz[i] - gxy[i] * gxz[i]) / det;
-        const double gupzz = (gxx * gyy - gxy[i] * gxy[i]) / det;
-
-        Sphi_rhs[i] = alpn1 * Spi[i];
-
-        Spi_rhs[i] = gupxx * fxx[i] + gupyy * fyy[i] + gupzz * fzz[i]
-                   + TWO * (gupxy * fxy[i] + gupxz * fxz[i] + gupyz * fyz[i])
-                   - ((Gamx[i] + (gupxx * chix[i] + gupxy * chiy[i] + gupxz * chiz[i]) / TWO / chin1) * Kx[i]
-                   +  (Gamy[i] + (gupxy * chix[i] + gupyy * chiy[i] + gupyz * chiz[i]) / TWO / chin1) * Ky[i]
-                   +  (Gamz[i] + (gupxz * chix[i] + gupyz * chiy[i] + gupzz * chiz[i]) / TWO / chin1) * Kz[i]);
-
-        Spi_rhs[i] = Spi_rhs[i] * alpn1
-                   + gupxx * Lapx[i] * Kx[i] + gupxy * Lapx[i] * Ky[i] + gupxz * Lapx[i] * Kz[i]
-                   + gupxy * Lapy[i] * Kx[i] + gupyy * Lapy[i] * Ky[i] + gupyz * Lapy[i] * Kz[i]
-                   + gupxz * Lapz[i] * Kx[i] + gupyz * Lapz[i] * Ky[i] + gupzz * Lapz[i] * Kz[i];
-
-        Spi_rhs[i] = Spi_rhs[i] * chin1 + alpn1 * (trK[i] * Spi[i] - dVdSphi[i]);
-
-        rho[i] = chin1 * ((gupxx * Kx[i] * Kx[i] + gupyy * Ky[i] * Ky[i] + gupzz * Kz[i] * Kz[i]) * HALF
-               + gupxy * Kx[i] * Ky[i] + gupxz * Kx[i] * Kz[i] + gupyz * Ky[i] * Kz[i])
-               + Spi[i] * Spi[i] * HALF + V[i];
-        Sx[i] = -Spi[i] * Kx[i];
-        Sy[i] = -Spi[i] * Ky[i];
-        Sz[i] = -Spi[i] * Kz[i];
-
-        const double pressure = (rho[i] - Spi[i] * Spi[i]) / chin1;
-        Sxx[i] = Kx[i] * Kx[i] - pressure * gxx;
-        Sxy[i] = Kx[i] * Ky[i] - pressure * gxy[i];
-        Sxz[i] = Kx[i] * Kz[i] - pressure * gxz[i];
-        Syy[i] = Ky[i] * Ky[i] - pressure * gyy;
-        Syz[i] = Ky[i] * Kz[i] - pressure * gyz[i];
-        Szz[i] = Kz[i] * Kz[i] - pressure * gzz;
-    }
-
-    if (f_compute_rhs_bssn(ex, T, X, Y, Z,
-                           chi, trK,
-                           dxx, gxy, gxz, dyy, gyz, dzz,
-                           Axx, Axy, Axz, Ayy, Ayz, Azz,
-                           Gamx, Gamy, Gamz,
-                           Lap, betax, betay, betaz,
-                           dtSfx, dtSfy, dtSfz,
-                           chi_rhs, trK_rhs,
-                           gxx_rhs, gxy_rhs, gxz_rhs, gyy_rhs, gyz_rhs, gzz_rhs,
-                           Axx_rhs, Axy_rhs, Axz_rhs, Ayy_rhs, Ayz_rhs, Azz_rhs,
-                           Gamx_rhs, Gamy_rhs, Gamz_rhs,
-                           Lap_rhs, betax_rhs, betay_rhs, betaz_rhs,
-                           dtSfx_rhs, dtSfy_rhs, dtSfz_rhs,
-                           rho, Sx, Sy, Sz,
-                           Sxx, Sxy, Sxz, Syy, Syz, Szz,
-                           Gamxxx, Gamxxy, Gamxxz, Gamxyy, Gamxyz, Gamxzz,
-                           Gamyxx, Gamyxy, Gamyxz, Gamyyy, Gamyyz, Gamyzz,
-                           Gamzxx, Gamzxy, Gamzxz, Gamzyy, Gamzyz, Gamzzz,
-                           Rxx, Rxy, Rxz, Ryy, Ryz, Rzz,
-                           ham_Res, movx_Res, movy_Res, movz_Res,
-                           Gmx_Res, Gmy_Res, Gmz_Res,
-                           Symmetry, Lev, eps, co))
-        return 1;
-
-    lopsided_kodis(ex, X, Y, Z, Sphi, Sphi_rhs, betax, betay, betaz, Symmetry, SSS, eps);
-    lopsided_kodis(ex, X, Y, Z, Spi, Spi_rhs, betax, betay, betaz, Symmetry, SSS, eps);
-
-    for (int i = 0; i < all; ++i)
-    {
-        if (Sphi_rhs[i] != Sphi_rhs[i] || Spi_rhs[i] != Spi_rhs[i] || rho[i] != rho[i])
-            return 1;
-    }
-
-    return 0;
-}

AMSS_NCKU_source/bssn_gpu.h

@@ -1,73 +1,56 @@
-
-#ifndef BSSN_GPU_H_
-#define BSSN_GPU_H_
-#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
-
-#define _FH2_(i, j, k) fh[(i) + (j) * _1D_SIZE[2] + (k) * _2D_SIZE[2]]
-#define _FH3_(i, j, k) fh[(i) + (j) * _1D_SIZE[3] + (k) * _2D_SIZE[3]]
-#define pow2(x) ((x) * (x))
-#define TimeBetween(a, b) ((b.tv_sec - a.tv_sec) + (b.tv_usec - a.tv_usec) / 1000000.0f)
-#define M_ metac.
-#define Mh_ meta->
-#define Ms_ metassc.
-#define Msh_ metass->
-
-// #define TIMING
-
-#define RHS_SS_PARA int calledby, int mpi_rank, int *ex, double &T, double *crho, double *sigma, double *R, double *X, double *Y, double *Z, double *drhodx, double *drhody, double *drhodz, double *dsigmadx, double *dsigmady, double *dsigmadz, double *dRdx, double *dRdy, double *dRdz, double *drhodxx, double *drhodxy, double *drhodxz, double *drhodyy, double *drhodyz, double *drhodzz, double *dsigmadxx, double *dsigmadxy, double *dsigmadxz, double *dsigmadyy, double *dsigmadyz, double *dsigmadzz, double *dRdxx, double *dRdxy, double *dRdxz, double *dRdyy, double *dRdyz, double *dRdzz, double *chi, double *trK, double *dxx, double *gxy, double *gxz, double *dyy, double *gyz, double *dzz, double *Axx, double *Axy, double *Axz, double *Ayy, double *Ayz, double *Azz, double *Gamx, double *Gamy, double *Gamz, double *Lap, double *betax, double *betay, double *betaz, double *dtSfx, double *dtSfy, double *dtSfz, double *chi_rhs, double *trK_rhs, double *gxx_rhs, double *gxy_rhs, double *gxz_rhs, double *gyy_rhs, double *gyz_rhs, double *gzz_rhs, double *Axx_rhs, double *Axy_rhs, double *Axz_rhs, double *Ayy_rhs, double *Ayz_rhs, double *Azz_rhs, double *Gamx_rhs, double *Gamy_rhs, double *Gamz_rhs, double *Lap_rhs, double *betax_rhs, double *betay_rhs, double *betaz_rhs, double *dtSfx_rhs, double *dtSfy_rhs, double *dtSfz_rhs, double *rho, double *Sx, double *Sy, double *Sz, double *Sxx, double *Sxy, double *Sxz, double *Syy, double *Syz, double *Szz, double *Gamxxx, double *Gamxxy, double *Gamxxz, double *Gamxyy, double *Gamxyz, double *Gamxzz, double *Gamyxx, double *Gamyxy, double *Gamyxz, double *Gamyyy, double *Gamyyz, double *Gamyzz, double *Gamzxx, double *Gamzxy, double *Gamzxz, double *Gamzyy, double *Gamzyz, double *Gamzzz, double *Rxx, double *Rxy, double *Rxz, double *Ryy, double *Ryz, double *Rzz, double *ham_Res, double *movx_Res, double *movy_Res, double *movz_Res, double *Gmx_Res, double *Gmy_Res, double *Gmz_Res, int &Symmetry, int &Lev, double &eps, int &sst, int &co
-
-/**  main function */
-int gpu_rhs(int calledby, int mpi_rank, int *ex, double &T,
-            double *X, double *Y, double *Z,
-
-            double *chi, double *trK,
-
-            double *dxx, double *gxy, double *gxz, double *dyy, double *gyz, double *dzz,
-
-            double *Axx, double *Axy, double *Axz, double *Ayy, double *Ayz, double *Azz,
-
-            double *Gamx, double *Gamy, double *Gamz,
-
-            double *Lap, double *betax, double *betay, double *betaz,
-
-            double *dtSfx, double *dtSfy, double *dtSfz,
-
-            double *chi_rhs, double *trK_rhs,
-
-            double *gxx_rhs, double *gxy_rhs, double *gxz_rhs, double *gyy_rhs, double *gyz_rhs, double *gzz_rhs,
-
-            double *Axx_rhs, double *Axy_rhs, double *Axz_rhs, double *Ayy_rhs, double *Ayz_rhs, double *Azz_rhs,
-
-            double *Gamx_rhs, double *Gamy_rhs, double *Gamz_rhs,
-
-            double *Lap_rhs, double *betax_rhs, double *betay_rhs, double *betaz_rhs,
-
-            double *dtSfx_rhs, double *dtSfy_rhs, double *dtSfz_rhs,
-
-            double *rho, double *Sx, double *Sy, double *Sz, double *Sxx,
-            double *Sxy, double *Sxz, double *Syy, double *Syz, double *Szz,
-
-            double *Gamxxx, double *Gamxxy, double *Gamxxz, double *Gamxyy, double *Gamxyz, double *Gamxzz,
-
-            double *Gamyxx, double *Gamyxy, double *Gamyxz, double *Gamyyy, double *Gamyyz, double *Gamyzz,
-
-            double *Gamzxx, double *Gamzxy, double *Gamzxz, double *Gamzyy, double *Gamzyz, double *Gamzzz,
-
-            double *Rxx, double *Rxy, double *Rxz, double *Ryy, double *Ryz, double *Rzz,
-
-            double *ham_Res, double *movx_Res, double *movy_Res, double *movz_Res,
-            double *Gmx_Res, double *Gmy_Res, double *Gmz_Res,
-            int &Symmetry, int &Lev, double &eps, int &co);
-
-int gpu_rhs_ss(RHS_SS_PARA);
-
-/** Init GPU side data in GPUMeta. */
-// void init_fluid_meta_gpu(GPUMeta *gpu_meta);
-
-#endif
+
+#ifndef BSSN_GPU_H_
+#define BSSN_GPU_H_
+#include "bssn_macro.h"
+#include "macrodef.h"
+
+#define DEVICE_ID 0
+// #define DEVICE_ID_BY_MPI_RANK
+#define GRID_DIM 256
+#define BLOCK_DIM 128
+
+#define _FH2_(i, j, k) fh[(i) + (j) * _1D_SIZE[2] + (k) * _2D_SIZE[2]]
+#define _FH3_(i, j, k) fh[(i) + (j) * _1D_SIZE[3] + (k) * _2D_SIZE[3]]
+#define pow2(x) ((x) * (x))
+#define TimeBetween(a, b) ((b.tv_sec - a.tv_sec) + (b.tv_usec - a.tv_usec) / 1000000.0f)
+#define M_ metac.
+#define Mh_ meta->
+#define Ms_ metassc.
+#define Msh_ metass->
+
+// #define TIMING
+
+#define RHS_SS_PARA int calledby, int mpi_rank, int *ex, double &T, double *crho, double *sigma, double *R, double *X, double *Y, double *Z, double *drhodx, double *drhody, double *drhodz, double *dsigmadx, double *dsigmady, double *dsigmadz, double *dRdx, double *dRdy, double *dRdz, double *drhodxx, double *drhodxy, double *drhodxz, double *drhodyy, double *drhodyz, double *drhodzz, double *dsigmadxx, double *dsigmadxy, double *dsigmadxz, double *dsigmadyy, double *dsigmadyz, double *dsigmadzz, double *dRdxx, double *dRdxy, double *dRdxz, double *dRdyy, double *dRdyz, double *dRdzz, double *chi, double *trK, double *dxx, double *gxy, double *gxz, double *dyy, double *gyz, double *dzz, double *Axx, double *Axy, double *Axz, double *Ayy, double *Ayz, double *Azz, double *Gamx, double *Gamy, double *Gamz, double *Lap, double *betax, double *betay, double *betaz, double *dtSfx, double *dtSfy, double *dtSfz, double *chi_rhs, double *trK_rhs, double *gxx_rhs, double *gxy_rhs, double *gxz_rhs, double *gyy_rhs, double *gyz_rhs, double *gzz_rhs, double *Axx_rhs, double *Axy_rhs, double *Axz_rhs, double *Ayy_rhs, double *Ayz_rhs, double *Azz_rhs, double *Gamx_rhs, double *Gamy_rhs, double *Gamz_rhs, double *Lap_rhs, double *betax_rhs, double *betay_rhs, double *betaz_rhs, double *dtSfx_rhs, double *dtSfy_rhs, double *dtSfz_rhs, double *rho, double *Sx, double *Sy, double *Sz, double *Sxx, double *Sxy, double *Sxz, double *Syy, double *Syz, double *Szz, double *Gamxxx, double *Gamxxy, double *Gamxxz, double *Gamxyy, double *Gamxyz, double *Gamxzz, double *Gamyxx, double *Gamyxy, double *Gamyxz, double *Gamyyy, double *Gamyyz, double *Gamyzz, double *Gamzxx, double *Gamzxy, double *Gamzxz, double *Gamzyy, double *Gamzyz, double *Gamzzz, double *Rxx, double *Rxy, double *Rxz, double *Ryy, double *Ryz, double *Rzz, double *ham_Res, double *movx_Res, double *movy_Res, double *movz_Res, double *Gmx_Res, double *Gmy_Res, double *Gmz_Res, int &Symmetry, int &Lev, double &eps, int &sst, int &co
+
+/**  main function */
+int gpu_rhs(int calledby, int mpi_rank, int *ex, double &T,
+            double *X, double *Y, double *Z,
+            double *chi, double *trK,
+            double *dxx, double *gxy, double *gxz, double *dyy, double *gyz, double *dzz,
+            double *Axx, double *Axy, double *Axz, double *Ayy, double *Ayz, double *Azz,
+            double *Gamx, double *Gamy, double *Gamz,
+            double *Lap, double *betax, double *betay, double *betaz,
+            double *dtSfx, double *dtSfy, double *dtSfz,
+            double *chi_rhs, double *trK_rhs,
+            double *gxx_rhs, double *gxy_rhs, double *gxz_rhs, double *gyy_rhs, double *gyz_rhs, double *gzz_rhs,
+            double *Axx_rhs, double *Axy_rhs, double *Axz_rhs, double *Ayy_rhs, double *Ayz_rhs, double *Azz_rhs,
+            double *Gamx_rhs, double *Gamy_rhs, double *Gamz_rhs,
+            double *Lap_rhs, double *betax_rhs, double *betay_rhs, double *betaz_rhs,
+            double *dtSfx_rhs, double *dtSfy_rhs, double *dtSfz_rhs,
+            double *rho, double *Sx, double *Sy, double *Sz, double *Sxx,
+            double *Sxy, double *Sxz, double *Syy, double *Syz, double *Szz,
+            double *Gamxxx, double *Gamxxy, double *Gamxxz, double *Gamxyy, double *Gamxyz, double *Gamxzz,
+            double *Gamyxx, double *Gamyxy, double *Gamyxz, double *Gamyyy, double *Gamyyz, double *Gamyzz,
+            double *Gamzxx, double *Gamzxy, double *Gamzxz, double *Gamzyy, double *Gamzyz, double *Gamzzz,
+            double *Rxx, double *Rxy, double *Rxz, double *Ryy, double *Ryz, double *Rzz,
+            double *ham_Res, double *movx_Res, double *movy_Res, double *movz_Res,
+            double *Gmx_Res, double *Gmy_Res, double *Gmz_Res,
+            int &Symmetry, int &Lev, double &eps, int &co);
+
+int gpu_rhs_ss(RHS_SS_PARA);
+
+#define Z4C_SS_PARA int calledby, int mpi_rank, int *ex, double &T, double *crho, double *sigma, double *R, double *X, double *Y, double *Z, double *drhodx, double *drhody, double *drhodz, double *dsigmadx, double *dsigmady, double *dsigmadz, double *dRdx, double *dRdy, double *dRdz, double *drhodxx, double *drhodxy, double *drhodxz, double *drhodyy, double *drhodyz, double *drhodzz, double *dsigmadxx, double *dsigmadxy, double *dsigmadxz, double *dsigmadyy, double *dsigmadyz, double *dsigmadzz, double *dRdxx, double *dRdxy, double *dRdxz, double *dRdyy, double *dRdyz, double *dRdzz, double *chi, double *trK, double *dxx, double *gxy, double *gxz, double *dyy, double *gyz, double *dzz, double *Axx, double *Axy, double *Axz, double *Ayy, double *Ayz, double *Azz, double *Gamx, double *Gamy, double *Gamz, double *Lap, double *betax, double *betay, double *betaz, double *dtSfx, double *dtSfy, double *dtSfz, double *TZ, double *chi_rhs, double *trK_rhs, double *gxx_rhs, double *gxy_rhs, double *gxz_rhs, double *gyy_rhs, double *gyz_rhs, double *gzz_rhs, double *Axx_rhs, double *Axy_rhs, double *Axz_rhs, double *Ayy_rhs, double *Ayz_rhs, double *Azz_rhs, double *Gamx_rhs, double *Gamy_rhs, double *Gamz_rhs, double *Lap_rhs, double *betax_rhs, double *betay_rhs, double *betaz_rhs, double *dtSfx_rhs, double *dtSfy_rhs, double *dtSfz_rhs, double *TZ_rhs, double *rho, double *Sx, double *Sy, double *Sz, double *Sxx, double *Sxy, double *Sxz, double *Syy, double *Syz, double *Szz, double *Gamxxx, double *Gamxxy, double *Gamxxz, double *Gamxyy, double *Gamxyz, double *Gamxzz, double *Gamyxx, double *Gamyxy, double *Gamyxz, double *Gamyyy, double *Gamyyz, double *Gamyzz, double *Gamzxx, double *Gamzxy, double *Gamzxz, double *Gamzyy, double *Gamzyz, double *Gamzzz, double *Rxx, double *Rxy, double *Rxz, double *Ryy, double *Ryz, double *Rzz, double *ham_Res, double *movx_Res, double *movy_Res, double *movz_Res, double *Gmx_Res, double *Gmy_Res, double *Gmz_Res, int &Symmetry, int &Lev, double &eps, int &sst, int &co
+
+int gpu_rhs_z4c_ss(Z4C_SS_PARA);
+
+#endif

AMSS_NCKU_source/bssn_gpu_class.h

@@ -1,210 +0,0 @@
-
-#ifndef BSSN_GPU_CLASS_H
-#define BSSN_GPU_CLASS_H
-
-#ifdef newc
-#include <iostream>
-#include <iomanip>
-#include <fstream>
-#include <cstdlib>
-#include <string>
-#include <cmath>
-using namespace std;
-#else
-#include <iostream.h>
-#include <iomanip.h>
-#include <fstream.h>
-#include <stdlib.h>
-#include <string.h>
-#include <math.h>
-#endif
-
-#include <mpi.h>
-
-#include "macrodef.h"
-#include "cgh.h"
-#include "ShellPatch.h"
-#include "misc.h"
-#include "var.h"
-#include "MyList.h"
-#include "monitor.h"
-#include "surface_integral.h"
-#include "checkpoint.h"
-
-// added by yangquan
-#include "bssn_macro.h"
-
-extern void setpbh(int iBHN, double **iPBH, double *iMass, int rBHN);
-
-class bssn_class
-{
-public:
-       // added by yangquan
-       //----------------------
-       int gpu_num_mynode;
-       int cpu_core_num_mynode;
-       int mpi_process_num_mynode;
-       int my_sequence_mynode;
-       int mynode_id;
-       int use_gpu;
-
-       virtual void Step_GPU(int lev, int YN);
-       virtual void Get_runtime_envirment();
-       // virtual void Step_OPENMP(int lev,int YN);
-       //----------------------
-
-       int ngfs;
-       int nprocs, myrank;
-       cgh *GH;
-       ShellPatch *SH;
-       double PhysTime;
-
-       int checkrun;
-       char checkfilename[50];
-       int Steps;
-       double StartTime, TotalTime;
-       double AnasTime, DumpTime, d2DumpTime, CheckTime;
-       double LastAnas, LastConsOut;
-       double Courant;
-       double numepss, numepsb, numepsh;
-       int Symmetry;
-       int maxl, decn;
-       double maxrex, drex;
-       int trfls, a_lev;
-
-       double dT;
-       double chitiny;
-
-       double **Porg0, **Porgbr, **Porg, **Porg1, **Porg_rhs;
-       int BH_num, BH_num_input;
-       double *Mass, *Pmom, *Spin;
-       double ADMMass;
-
-       var *phio, *trKo;
-       var *gxxo, *gxyo, *gxzo, *gyyo, *gyzo, *gzzo;
-       var *Axxo, *Axyo, *Axzo, *Ayyo, *Ayzo, *Azzo;
-       var *Gmxo, *Gmyo, *Gmzo;
-       var *Lapo, *Sfxo, *Sfyo, *Sfzo;
-       var *dtSfxo, *dtSfyo, *dtSfzo;
-
-       var *phi0, *trK0;
-       var *gxx0, *gxy0, *gxz0, *gyy0, *gyz0, *gzz0;
-       var *Axx0, *Axy0, *Axz0, *Ayy0, *Ayz0, *Azz0;
-       var *Gmx0, *Gmy0, *Gmz0;
-       var *Lap0, *Sfx0, *Sfy0, *Sfz0;
-       var *dtSfx0, *dtSfy0, *dtSfz0;
-
-       var *phi, *trK;
-       var *gxx, *gxy, *gxz, *gyy, *gyz, *gzz;
-       var *Axx, *Axy, *Axz, *Ayy, *Ayz, *Azz;
-       var *Gmx, *Gmy, *Gmz;
-       var *Lap, *Sfx, *Sfy, *Sfz;
-       var *dtSfx, *dtSfy, *dtSfz;
-
-       var *phi1, *trK1;
-       var *gxx1, *gxy1, *gxz1, *gyy1, *gyz1, *gzz1;
-       var *Axx1, *Axy1, *Axz1, *Ayy1, *Ayz1, *Azz1;
-       var *Gmx1, *Gmy1, *Gmz1;
-       var *Lap1, *Sfx1, *Sfy1, *Sfz1;
-       var *dtSfx1, *dtSfy1, *dtSfz1;
-
-       var *phi_rhs, *trK_rhs;
-       var *gxx_rhs, *gxy_rhs, *gxz_rhs, *gyy_rhs, *gyz_rhs, *gzz_rhs;
-       var *Axx_rhs, *Axy_rhs, *Axz_rhs, *Ayy_rhs, *Ayz_rhs, *Azz_rhs;
-       var *Gmx_rhs, *Gmy_rhs, *Gmz_rhs;
-       var *Lap_rhs, *Sfx_rhs, *Sfy_rhs, *Sfz_rhs;
-       var *dtSfx_rhs, *dtSfy_rhs, *dtSfz_rhs;
-
-       var *rho, *Sx, *Sy, *Sz, *Sxx, *Sxy, *Sxz, *Syy, *Syz, *Szz;
-
-       var *Gamxxx, *Gamxxy, *Gamxxz, *Gamxyy, *Gamxyz, *Gamxzz;
-       var *Gamyxx, *Gamyxy, *Gamyxz, *Gamyyy, *Gamyyz, *Gamyzz;
-       var *Gamzxx, *Gamzxy, *Gamzxz, *Gamzyy, *Gamzyz, *Gamzzz;
-
-       var *Rxx, *Rxy, *Rxz, *Ryy, *Ryz, *Rzz;
-
-       var *Rpsi4, *Ipsi4;
-       var *t1Rpsi4, *t1Ipsi4, *t2Rpsi4, *t2Ipsi4;
-
-       var *Cons_Ham, *Cons_Px, *Cons_Py, *Cons_Pz, *Cons_Gx, *Cons_Gy, *Cons_Gz;
-
-#ifdef Point_Psi4
-       var *phix, *phiy, *phiz;
-       var *trKx, *trKy, *trKz;
-       var *Axxx, *Axxy, *Axxz;
-       var *Axyx, *Axyy, *Axyz;
-       var *Axzx, *Axzy, *Axzz;
-       var *Ayyx, *Ayyy, *Ayyz;
-       var *Ayzx, *Ayzy, *Ayzz;
-       var *Azzx, *Azzy, *Azzz;
-#endif
-       // FIXME: uc = StateList, up = OldStateList, upp = SynchList_cor; so never touch these three data
-       MyList<var> *StateList, *SynchList_pre, *SynchList_cor, *RHSList;
-       MyList<var> *OldStateList, *DumpList;
-       MyList<var> *ConstraintList;
-
-       monitor *ErrorMonitor, *Psi4Monitor, *BHMonitor, *MAPMonitor;
-       monitor *ConVMonitor;
-       surface_integral *Waveshell;
-       checkpoint *CheckPoint;
-
-public:
-       bssn_class(double Couranti, double StartTimei, double TotalTimei, double DumpTimei, double d2DumpTimei, double CheckTimei, double AnasTimei,
-                  int Symmetryi, int checkruni, char *checkfilenamei, double numepssi, double numepsbi, double numepshi,
-                  int a_levi, int maxli, int decni, double maxrexi, double drexi);
-       ~bssn_class();
-
-       void Evolve(int Steps);
-       void RecursiveStep(int lev);
-#if (PSTR == 1)
-       void ParallelStep();
-       void SHStep();
-#endif
-       void RestrictProlong(int lev, int YN, bool BB, MyList<var> *SL, MyList<var> *OL, MyList<var> *corL);
-       void RestrictProlong_aux(int lev, int YN, bool BB, MyList<var> *SL, MyList<var> *OL, MyList<var> *corL);
-       void RestrictProlong(int lev, int YN, bool BB);
-       void ProlongRestrict(int lev, int YN, bool BB);
-       void Setup_Black_Hole_position();
-       void compute_Porg_rhs(double **BH_PS, double **BH_RHS, var *forx, var *fory, var *forz, int lev);
-       bool read_Pablo_file(int *ext, double *datain, char *filename);
-       void write_Pablo_file(int *ext, double xmin, double xmax, double ymin, double ymax, double zmin, double zmax,
-                             char *filename);
-       void AnalysisStuff(int lev, double dT_lev);
-       void Setup_KerrSchild();
-       void Enforce_algcon(int lev, int fg);
-
-       void testRestrict();
-       void testOutBd();
-
-       virtual void Setup_Initial_Data_Lousto();
-       virtual void Setup_Initial_Data_Cao();
-       virtual void Initialize();
-       virtual void Read_Ansorg();
-       virtual void Read_Pablo() {};
-       virtual void Compute_Psi4(int lev);
-       virtual void Step(int lev, int YN);
-       virtual void Interp_Constraint(bool infg);
-       virtual void Constraint_Out();
-       virtual void Compute_Constraint();
-
-#ifdef With_AHF
-protected:
-       MyList<var> *AHList, *AHDList, *GaugeList;
-       int AHfindevery;
-       double AHdumptime;
-       int *lastahdumpid, HN_num; // number of possible horizons
-       int *findeveryl;
-       double *xc, *yc, *zc, *xr, *yr, *zr;
-       bool *trigger;
-       double *dTT;
-       int *dumpid;
-
-public:
-       void AH_Prepare_derivatives();
-       bool AH_Interp_Points(MyList<var> *VarList,
-                             int NN, double **XX,
-                             double *Shellf, int Symmetryi);
-       void AH_Step_Find(int lev, double dT_lev);
-#endif
-};
-#endif /* BSSN_GPU_CLASS_H */

AMSS_NCKU_source/bssn_gpu_rhs_ss.cu

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

AMSS_NCKU_source/bssn_rhs.h

@@ -22,32 +22,32 @@
 #define f_compute_rhs_Z4c_ss COMPUTE_RHS_Z4C_SS
 #define f_compute_constraint_fr COMPUTE_CONSTRAINT_FR
 #endif
-#ifdef fortran3
-#define f_compute_rhs_bssn compute_rhs_bssn_
+#ifdef fortran3
+#define f_compute_rhs_bssn compute_rhs_bssn_
 #define f_compute_rhs_bssn_ss compute_rhs_bssn_ss_
 #define f_compute_rhs_bssn_escalar compute_rhs_bssn_escalar_
 #define f_compute_rhs_bssn_escalar_ss compute_rhs_bssn_escalar_ss_
 #define f_compute_rhs_Z4c compute_rhs_z4c_
 #define f_compute_rhs_Z4cnot compute_rhs_z4cnot_
 #define f_compute_rhs_Z4c_ss compute_rhs_z4c_ss_
-#define f_compute_constraint_fr compute_constraint_fr_
-#endif
-
-#ifdef __cplusplus
-extern "C"
-{
-#endif
-        void f_bssn_rhs_kernel_timing_reset();
-        int f_bssn_rhs_kernel_timing_bucket_count();
-        const double *f_bssn_rhs_kernel_timing_local_seconds();
-        const char *f_bssn_rhs_kernel_timing_label(int);
-#ifdef __cplusplus
-}
-#endif
-
-extern "C"
-{
-        int f_compute_rhs_bssn(int *, double &, double *, double *, double *,                                                      // ex,T,X,Y,Z
+#define f_compute_constraint_fr compute_constraint_fr_
+#endif
+
+#ifdef __cplusplus
+extern "C"
+{
+#endif
+        void f_bssn_rhs_kernel_timing_reset();
+        int f_bssn_rhs_kernel_timing_bucket_count();
+        const double *f_bssn_rhs_kernel_timing_local_seconds();
+        const char *f_bssn_rhs_kernel_timing_label(int);
+#ifdef __cplusplus
+}
+#endif
+
+extern "C"
+{
+        int f_compute_rhs_bssn(int *, double &, double *, double *, double *,                                                      // ex,T,X,Y,Z
                                double *, double *,                                                                                 // chi, trK
                                double *, double *, double *, double *, double *, double *,                                         // gij
                                double *, double *, double *, double *, double *, double *,                                         // Aij
@@ -67,27 +67,6 @@ extern "C"
                                int &, int &, double &, int &);
 }
 
-int f_compute_rhs_bssn_escalar_c(int *, double &, double *, double *, double *,                                                      // ex,T,X,Y,Z
-                                 double *, double *,                                                                                 // chi, trK
-                                 double *, double *, double *, double *, double *, double *,                                         // gij
-                                 double *, double *, double *, double *, double *, double *,                                         // Aij
-                                 double *, double *, double *,                                                                       // Gam
-                                 double *, double *, double *, double *, double *, double *, double *,                               // Gauge
-                                 double *, double *,                                                                                 // Sphi, Spi
-                                 double *, double *,                                                                                 // chi, trK
-                                 double *, double *, double *, double *, double *, double *,                                         // gij
-                                 double *, double *, double *, double *, double *, double *,                                         // Aij
-                                 double *, double *, double *,                                                                       // Gam
-                                 double *, double *, double *, double *, double *, double *, double *,                               // Gauge
-                                 double *, double *,                                                                                 // Sphi, Spi
-                                 double *, double *, double *, double *, double *, double *, double *, double *, double *, double *, // stress-energy
-                                 double *, double *, double *, double *, double *, double *,                                         // Christoffel
-                                 double *, double *, double *, double *, double *, double *,                                         // Christoffel
-                                 double *, double *, double *, double *, double *, double *,                                         // Christoffel
-                                 double *, double *, double *, double *, double *, double *,                                         // Ricci
-                                 double *, double *, double *, double *, double *, double *, double *,                               // constraint violation
-                                 int &, int &, double &, int &);
-
 extern "C"
 {
         int f_compute_rhs_bssn_ss(int *, double &, double *, double *, double *,                                                      // ex,T,rho,sigma,R
@@ -262,31 +241,4 @@ extern "C"
                                      double *);
 } // FR_cons
 
-// BSSN-EM C kernel (replaces empart.f90 + bssn_rhs.f90 for BSSN+Maxwell)
-int f_compute_rhs_bssn_em_c(int *, double &, double *, double *, double *,
-                             double *, double *,
-                             double *, double *, double *, double *, double *, double *,
-                             double *, double *, double *, double *, double *, double *,
-                             double *, double *, double *,
-                             double *, double *, double *, double *, double *, double *, double *,
-                             double *, double *, double *,
-                             double *, double *, double *, double *, double *, double *, double *, double *,
-                             double *, double *, double *,
-                             double *, double *,
-                             double *, double *,
-                             double *, double *, double *, double *, double *, double *,
-                             double *, double *, double *, double *, double *, double *,
-                             double *, double *, double *,
-                             double *, double *, double *, double *, double *, double *, double *,
-                             double *, double *, double *, double *, double *, double *, double *, double *,
-                             double *, double *, double *,
-                             double *, double *, double *, double *,
-                             double *, double *, double *, double *, double *, double *,
-                             double *, double *, double *, double *, double *, double *,
-                             double *, double *, double *, double *, double *, double *,
-                             double *, double *, double *, double *, double *, double *,
-                             double *, double *, double *, double *, double *, double *,
-                             double *, double *, double *, double *, double *, double *,
-                             int &, int &, double &, int &);
-
 #endif /* BSSN_H */

AMSS_NCKU_source/bssn_rhs_c.C

@@ -1075,10 +1075,6 @@ int f_compute_rhs_bssn(int *ex, double &T,
         }
         #endif
 
-        #if (GAUGE == 2 || GAUGE == 3 || GAUGE == 4 || GAUGE == 5)
-        fderivs(ex,chi,dtSfx_rhs,dtSfy_rhs,dtSfz_rhs,X,Y,Z,SYM,SYM,SYM,Symmetry,Lev);
-        #endif
-
         for (int i = 0; i < all; i += 1) {
             #if (GAUGE == 0)
             betax_rhs[i] = FF * dtSfx[i];
@@ -1102,12 +1098,12 @@ int f_compute_rhs_bssn(int *ex, double &T,
             betaz_rhs[i] = FF * dtSfz[i];
 
             reta[i] =
-                gupxx[i] * dtSfx_rhs[i] * dtSfx_rhs[i]
-                + gupyy[i] * dtSfy_rhs[i] * dtSfy_rhs[i]
-                + gupzz[i] * dtSfz_rhs[i] * dtSfz_rhs[i]
-                + TWO * ( gupxy[i] * dtSfx_rhs[i] * dtSfy_rhs[i]
-                        + gupxz[i] * dtSfx_rhs[i] * dtSfz_rhs[i]
-                        + gupyz[i] * dtSfy_rhs[i] * dtSfz_rhs[i] );
+                gupxx[i] * chix[i] * chix[i]
+                + gupyy[i] * chiy[i] * chiy[i]
+                + gupzz[i] * chiz[i] * chiz[i]
+                + TWO * ( gupxy[i] * chix[i] * chiy[i]
+                        + gupxz[i] * chix[i] * chiz[i]
+                        + gupyz[i] * chiy[i] * chiz[i] );
 
             #if (GAUGE == 2)
             reta[i] = 1.31 / 2.0 * sqrt( reta[i] / chin1[i] ) / pow( (ONE - sqrt(chin1[i])), 2.0 );
@@ -1120,12 +1116,12 @@ int f_compute_rhs_bssn(int *ex, double &T,
             dtSfz_rhs[i] = Gamz_rhs[i] - reta[i] * dtSfz[i];
             #elif (GAUGE == 4 || GAUGE == 5)
             reta[i] =
-                gupxx[i] * dtSfx_rhs[i] * dtSfx_rhs[i]
-                + gupyy[i] * dtSfy_rhs[i] * dtSfy_rhs[i]
-                + gupzz[i] * dtSfz_rhs[i] * dtSfz_rhs[i]
-                + TWO * ( gupxy[i] * dtSfx_rhs[i] * dtSfy_rhs[i]
-                        + gupxz[i] * dtSfx_rhs[i] * dtSfz_rhs[i]
-                        + gupyz[i] * dtSfy_rhs[i] * dtSfz_rhs[i] );
+                gupxx[i] * chix[i] * chix[i]
+                + gupyy[i] * chiy[i] * chiy[i]
+                + gupzz[i] * chiz[i] * chiz[i]
+                + TWO * ( gupxy[i] * chix[i] * chiy[i]
+                        + gupxz[i] * chix[i] * chiz[i]
+                        + gupyz[i] * chiy[i] * chiz[i] );
 
             #if (GAUGE == 4)
             reta[i] = 1.31 / 2.0 * sqrt( reta[i] / chin1[i] ) / pow( (ONE - sqrt(chin1[i])), 2.0 );
@@ -1164,17 +1160,11 @@ int f_compute_rhs_bssn(int *ex, double &T,
         lopsided_kodis(ex,X,Y,Z,gyz,gyz_rhs,betax,betay,betaz,Symmetry,SAA,eps);
         lopsided_kodis(ex,X,Y,Z,betaz,betaz_rhs,betax,betay,betaz,Symmetry,SSA,eps);
         lopsided_kodis(ex,X,Y,Z,dzz,gzz_rhs,betax,betay,betaz,Symmetry,SSS,eps);
-        #if (GAUGE == 0 || GAUGE == 2 || GAUGE == 3 || GAUGE == 6 || GAUGE == 7)
         lopsided_kodis(ex,X,Y,Z,dtSfx,dtSfx_rhs,betax,betay,betaz,Symmetry,ASS,eps);
-        #endif
         lopsided_kodis(ex,X,Y,Z,Axx,Axx_rhs,betax,betay,betaz,Symmetry,SSS,eps);
-        #if (GAUGE == 0 || GAUGE == 2 || GAUGE == 3 || GAUGE == 6 || GAUGE == 7)
         lopsided_kodis(ex,X,Y,Z,dtSfy,dtSfy_rhs,betax,betay,betaz,Symmetry,SAS,eps);
-        #endif
         lopsided_kodis(ex,X,Y,Z,Axy,Axy_rhs,betax,betay,betaz,Symmetry,AAS,eps);
-        #if (GAUGE == 0 || GAUGE == 2 || GAUGE == 3 || GAUGE == 6 || GAUGE == 7)
         lopsided_kodis(ex,X,Y,Z,dtSfz,dtSfz_rhs,betax,betay,betaz,Symmetry,SSA,eps);
-        #endif
         lopsided_kodis(ex,X,Y,Z,Axz,Axz_rhs,betax,betay,betaz,Symmetry,ASA,eps);
         lopsided_kodis(ex,X,Y,Z,Ayy,Ayy_rhs,betax,betay,betaz,Symmetry,SSS,eps);
         lopsided_kodis(ex,X,Y,Z,Ayz,Ayz_rhs,betax,betay,betaz,Symmetry,SAA,eps);

AMSS_NCKU_source/bssn_rhs_cuda.h

@@ -0,0 +1,413 @@
+#ifndef BSSN_RHS_CUDA_H
+#define BSSN_RHS_CUDA_H
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+enum {
+    BSSN_CUDA_STATE_COUNT = 24,
+    BSSN_ESCALAR_CUDA_STATE_COUNT = 26,
+    BSSN_EM_CUDA_STATE_COUNT = 32,
+    BSSN_EM_CUDA_SOURCE_COUNT = 4,
+    BSSN_CUDA_MATTER_COUNT = 10
+};
+
+int f_compute_rhs_bssn(int *ex, double &T,
+                       double *X, double *Y, double *Z,
+                       double *chi, double *trK,
+                       double *dxx, double *gxy, double *gxz, double *dyy, double *gyz, double *dzz,
+                       double *Axx, double *Axy, double *Axz, double *Ayy, double *Ayz, double *Azz,
+                       double *Gamx, double *Gamy, double *Gamz,
+                       double *Lap, double *betax, double *betay, double *betaz,
+                       double *dtSfx, double *dtSfy, double *dtSfz,
+                       double *chi_rhs, double *trK_rhs,
+                       double *gxx_rhs, double *gxy_rhs, double *gxz_rhs, double *gyy_rhs, double *gyz_rhs, double *gzz_rhs,
+                       double *Axx_rhs, double *Axy_rhs, double *Axz_rhs, double *Ayy_rhs, double *Ayz_rhs, double *Azz_rhs,
+                       double *Gamx_rhs, double *Gamy_rhs, double *Gamz_rhs,
+                       double *Lap_rhs, double *betax_rhs, double *betay_rhs, double *betaz_rhs,
+                       double *dtSfx_rhs, double *dtSfy_rhs, double *dtSfz_rhs,
+                       double *rho, double *Sx, double *Sy, double *Sz,
+                       double *Sxx, double *Sxy, double *Sxz, double *Syy, double *Syz, double *Szz,
+                       double *Gamxxx, double *Gamxxy, double *Gamxxz, double *Gamxyy, double *Gamxyz, double *Gamxzz,
+                       double *Gamyxx, double *Gamyxy, double *Gamyxz, double *Gamyyy, double *Gamyyz, double *Gamyzz,
+                       double *Gamzxx, double *Gamzxy, double *Gamzxz, double *Gamzyy, double *Gamzyz, double *Gamzzz,
+                       double *Rxx, double *Rxy, double *Rxz, double *Ryy, double *Ryz, double *Rzz,
+                       double *ham_Res, double *movx_Res, double *movy_Res, double *movz_Res,
+                       double *Gmx_Res, double *Gmy_Res, double *Gmz_Res,
+                       int &Symmetry, int &Lev, double &eps, int &co);
+
+int bssn_cuda_rk4_substep(void *block_tag,
+                          int *ex, double *X, double *Y, double *Z,
+                          double **state_host_in,
+                          double **state_host_out,
+                          double **matter_host,
+                          const double *propspeed,
+                          const double *soa_flat,
+                          const double *bbox,
+                          double &dT,
+                          double &T,
+                          int &RK4,
+                          int &apply_bam_bc,
+                          int &Symmetry,
+                          int &Lev,
+                          double &eps,
+                          int &co,
+                          int &use_zero_matter,
+                          int &keep_resident_state,
+                          int &apply_enforce_ga,
+                          double &chitiny);
+
+int bssn_escalar_cuda_rk4_substep(void *block_tag,
+                                  int *ex, double *X, double *Y, double *Z,
+                                  double **state_host_in,
+                                  double **state_host_out,
+                                  const double *propspeed,
+                                  const double *soa_flat,
+                                  const double *bbox,
+                                  double &dT,
+                                  double &T,
+                                  int &RK4,
+                                  int &apply_bam_bc,
+                                  int &Symmetry,
+	                                  int &Lev,
+	                                  double &eps,
+	                                  int &co,
+	                                  int &keep_resident_state,
+	                                  int &apply_enforce_ga,
+	                                  double &chitiny);
+
+int bssn_escalar_cuda_compute_constraints(int *ex, double *X, double *Y, double *Z,
+                                          double **state_host_in,
+                                          double **constraint_host_out,
+                                          int &Symmetry,
+                                          int &Lev,
+                                          double &eps);
+
+int bssn_em_cuda_rk4_substep(void *block_tag,
+                             int *ex, double *X, double *Y, double *Z,
+                             double **state_host_in,
+                             double **state_host_out,
+                             double **source_host,
+                             const double *propspeed,
+                             const double *soa_flat,
+                             const double *bbox,
+                             double &dT,
+                             double &T,
+                             int &RK4,
+                             int &apply_bam_bc,
+                             int &Symmetry,
+                             int &Lev,
+                             double &eps,
+                             int &co,
+                             int &keep_resident_state,
+                             int &apply_enforce_ga,
+                             double &chitiny);
+
+int bssn_em_cuda_resident_zero_fast_state(void *block_tag);
+
+int bssn_cuda_copy_state_region_to_host(void *block_tag,
+                                        int state_index,
+                                        double *host_state,
+                                        int *ex,
+                                        int i0, int j0, int k0,
+                                        int sx, int sy, int sz);
+
+int bssn_cuda_copy_state_region_from_host(void *block_tag,
+                                          int state_index,
+                                          double *host_state,
+                                          int *ex,
+                                          int i0, int j0, int k0,
+                                          int sx, int sy, int sz);
+
+int bssn_cuda_download_resident_state(void *block_tag,
+                                      int *ex,
+                                      double **state_host_out);
+
+int bssn_escalar_cuda_download_resident_state(void *block_tag,
+                                              int *ex,
+                                              double **state_host_out);
+
+int bssn_cuda_upload_resident_state_count(void *block_tag,
+                                          int *ex,
+                                          double **state_host_in,
+                                          int state_count);
+
+int bssn_escalar_cuda_upload_resident_state(void *block_tag,
+                                            int *ex,
+                                            double **state_host_in);
+
+int bssn_cuda_keep_only_resident_state_count(void *block_tag,
+                                             int *ex,
+                                             double **state_host_key,
+                                             int state_count);
+
+int bssn_escalar_cuda_keep_only_resident_state(void *block_tag,
+                                               int *ex,
+                                               double **state_host_key);
+
+int bssn_cuda_download_resident_state_count_if_present(void *block_tag,
+                                                       int *ex,
+                                                       double **state_host_out,
+                                                       int state_count);
+
+int bssn_cuda_download_resident_state_if_present(void *block_tag,
+                                                int *ex,
+                                                double **state_host_out);
+
+int bssn_cuda_download_constraint_outputs(int *ex,
+                                          double **constraint_host_out);
+
+int bssn_cuda_pack_state_region_to_host_buffer(void *block_tag,
+                                               int state_index,
+                                               double *host_buffer,
+                                               int *ex,
+                                               int i0, int j0, int k0,
+                                               int sx, int sy, int sz);
+
+int bssn_cuda_interp_state_point3(void *block_tag,
+                                  int *ex,
+                                  int state0,
+                                  int state1,
+                                  int state2,
+                                  double x0,
+                                  double y0,
+                                  double z0,
+                                  double dx,
+                                  double dy,
+                                  double dz,
+                                  double px,
+                                  double py,
+                                  double pz,
+                                  int ordn,
+                                  int symmetry,
+                                  double **state_host_key,
+                                  const double *soa3,
+                                  double *out3);
+
+int bssn_cuda_interp_host_two_fields(void *block_tag,
+                                     int *ex,
+                                     double *field0,
+                                     double *field1,
+                                     double x0,
+                                     double y0,
+                                     double z0,
+                                     double dx,
+                                     double dy,
+                                     double dz,
+                                     const double *px,
+                                     const double *py,
+                                     const double *pz,
+                                     int npoints,
+                                     int ordn,
+                                     int symmetry,
+                                     const double *soa6,
+                                     double *out_interleaved);
+
+int bssn_cuda_unpack_state_region_from_host_buffer(void *block_tag,
+                                                   int state_index,
+                                                   double *host_buffer,
+                                                   int *ex,
+                                                   int i0, int j0, int k0,
+                                                   int sx, int sy, int sz);
+
+int bssn_cuda_unpack_state_region_from_host_buffer_for_host_views(void *block_tag,
+                                                                  double **state_host_key,
+                                                                  int state_count,
+                                                                  int state_index,
+                                                                  double *host_buffer,
+                                                                  int *ex,
+                                                                  int i0, int j0, int k0,
+                                                                  int sx, int sy, int sz);
+
+int bssn_cuda_pack_state_batch_to_host_buffer(void *block_tag,
+                                              int state_count,
+                                              double *host_buffer,
+                                              int *ex,
+                                              int i0, int j0, int k0,
+                                              int sx, int sy, int sz);
+
+int bssn_cuda_pack_state_batch_to_host_buffer_for_host_views(void *block_tag,
+                                                             double **state_host_key,
+                                                             int state_count,
+                                                             double *host_buffer,
+                                                             int *ex,
+                                                             int i0, int j0, int k0,
+                                                             int sx, int sy, int sz);
+
+int bssn_cuda_unpack_state_batch_from_host_buffer(void *block_tag,
+                                                  int state_count,
+                                                  double *host_buffer,
+                                                  int *ex,
+                                                  int i0, int j0, int k0,
+                                                  int sx, int sy, int sz);
+
+int bssn_cuda_unpack_state_batch_from_host_buffer_for_host_views(void *block_tag,
+                                                                 double **state_host_key,
+                                                                 int state_count,
+                                                                 double *host_buffer,
+                                                                 int *ex,
+                                                                 int i0, int j0, int k0,
+                                                                 int sx, int sy, int sz);
+
+int bssn_cuda_pack_state_batch_to_device_buffer(void *block_tag,
+                                                int state_count,
+                                                double *device_buffer,
+                                                int *ex,
+                                                int i0, int j0, int k0,
+                                                int sx, int sy, int sz);
+
+int bssn_cuda_pack_state_batch_to_device_buffer_for_host_views(void *block_tag,
+                                                               double **state_host_key,
+                                                               int state_count,
+                                                               double *device_buffer,
+                                                               int *ex,
+                                                               int i0, int j0, int k0,
+                                                               int sx, int sy, int sz);
+
+int bssn_cuda_unpack_state_batch_from_device_buffer(void *block_tag,
+                                                    int state_count,
+                                                    double *device_buffer,
+                                                    int *ex,
+                                                    int i0, int j0, int k0,
+                                                    int sx, int sy, int sz);
+
+int bssn_cuda_unpack_state_batch_from_device_buffer_for_host_views(void *block_tag,
+                                                                   double **state_host_key,
+                                                                   int state_count,
+                                                                   double *device_buffer,
+                                                                   int *ex,
+                                                                   int i0, int j0, int k0,
+                                                                   int sx, int sy, int sz);
+
+int bssn_cuda_pack_state_segments_to_device_buffer(void *block_tag,
+                                                   int state_count,
+                                                   double *device_buffer,
+                                                   int *ex,
+                                                   int segment_count,
+                                                   const int *segment_meta);
+
+int bssn_cuda_pack_state_segments_to_device_buffer_for_host_views(void *block_tag,
+                                                                  double **state_host_key,
+                                                                  int state_count,
+                                                                  double *device_buffer,
+                                                                  int *ex,
+                                                                  int segment_count,
+                                                                  const int *segment_meta);
+
+int bssn_cuda_unpack_state_segments_from_device_buffer(void *block_tag,
+                                                       int state_count,
+                                                       double *device_buffer,
+                                                       int *ex,
+                                                       int segment_count,
+                                                       const int *segment_meta);
+
+int bssn_cuda_unpack_state_segments_from_device_buffer_for_host_views(void *block_tag,
+                                                                      double **state_host_key,
+                                                                      int state_count,
+                                                                      double *device_buffer,
+                                                                      int *ex,
+                                                                      int segment_count,
+                                                                      const int *segment_meta);
+
+int bssn_cuda_restrict_state_segments_to_device_buffer(void *block_tag,
+                                                       int state_count,
+                                                       double *device_buffer,
+                                                       int *ex,
+                                                       int segment_count,
+                                                       const int *segment_meta);
+
+int bssn_cuda_restrict_state_segments_to_device_buffer_for_host_views(void *block_tag,
+                                                                      double **state_host_key,
+                                                                      int state_count,
+                                                                      double *device_buffer,
+                                                                      int *ex,
+                                                                      int segment_count,
+                                                                      const int *segment_meta,
+                                                                      const double *state_soa);
+
+int bssn_cuda_prolong_state_segments_to_device_buffer(void *block_tag,
+                                                      int state_count,
+                                                      double *device_buffer,
+                                                      int *ex,
+                                                      int segment_count,
+                                                      const int *segment_meta);
+
+int bssn_cuda_prolong_state_segments_to_device_buffer_for_host_views(void *block_tag,
+                                                                     double **state_host_key,
+                                                                     int state_count,
+                                                                     double *device_buffer,
+                                                                     int *ex,
+                                                                     int segment_count,
+                                                                     const int *segment_meta,
+                                                                     const double *state_soa);
+
+int bssn_cuda_restrict_state_batch_to_device_buffer(void *block_tag,
+                                                    int state_count,
+                                                    double *device_buffer,
+                                                    int *ex,
+                                                    int sx, int sy, int sz,
+                                                    int fi0, int fj0, int fk0);
+
+int bssn_cuda_restrict_state_batch_to_device_buffer_for_host_views(void *block_tag,
+                                                                   double **state_host_key,
+                                                                   int state_count,
+                                                                   double *device_buffer,
+                                                                   int *ex,
+                                                                   int sx, int sy, int sz,
+                                                                   int fi0, int fj0, int fk0,
+                                                                   const double *state_soa);
+
+int bssn_cuda_prolong_state_batch_to_device_buffer(void *block_tag,
+                                                   int state_count,
+                                                   double *device_buffer,
+                                                   int *ex,
+                                                   int sx, int sy, int sz,
+                                                   int ii0, int jj0, int kk0,
+                                                   int lbc_i, int lbc_j, int lbc_k);
+
+int bssn_cuda_prolong_state_batch_to_device_buffer_for_host_views(void *block_tag,
+                                                                  double **state_host_key,
+                                                                  int state_count,
+                                                                  double *device_buffer,
+                                                                  int *ex,
+                                                                  int sx, int sy, int sz,
+                                                                  int ii0, int jj0, int kk0,
+                                                                  int lbc_i, int lbc_j, int lbc_k,
+                                                                  const double *state_soa);
+
+int bssn_cuda_download_state_subset(void *block_tag,
+                                    int *ex,
+                                    int subset_count,
+                                    const int *state_indices,
+                                    double **state_host_out);
+
+int bssn_cuda_upload_state_subset(void *block_tag,
+                                  int *ex,
+                                  int subset_count,
+                                  const int *state_indices,
+                                  double **state_host_in);
+
+int bssn_cuda_prepare_inter_time_level(void *block_tag,
+                                       int *ex,
+                                       int state_count,
+                                       double **src1_host_key,
+                                       double **src2_host_key,
+                                       double **src3_host_key,
+                                       double **dst_host_key,
+                                       int source_count,
+                                       int tindex);
+
+int bssn_cuda_has_resident_state(void *block_tag);
+
+void bssn_cuda_release_step_ctx(void *block_tag);
+
+#ifdef __cplusplus
+}
+// C++-only helpers declared for derived equation classes (Z4C, etc.)
+// Defined in bssn_class.C. Requires MyList, Patch, var from including TU.
+bool bssn_cuda_use_resident_sync(int lev);
+void bssn_cuda_download_level_state_if_present(MyList<Patch> *PatL, MyList<var> *vars, int myrank);
+#endif
+
+#endif

AMSS_NCKU_source/checkpoint.C

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

AMSS_NCKU_source/fd_cuda_helpers.cuh

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

AMSS_NCKU_source/fdderivs_sh_c.C

@@ -1,321 +0,0 @@
-#include "macrodef.h"
-#include "share_func.h"
-
-/*
- * fdderivs_sh — second derivatives on shell patch in (rho, sigma, R) coords.
- * Same stencil coefficients as Cartesian fdderivs. Uses symmetry_stbd.
- */
-extern "C" void fdderivs_sh_(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, int sst)
-{
-    (void)SYM3; (void)onoff; (void)sst;
-
-    const int NO_SYMM=0, EQ_SYMM=1, OCTANT=2;
-    const double ZEO=0.0, ONE=1.0, TWO=2.0, F1o4=2.5e-1;
-    const double F8=8.0, F16=16.0, F30=30.0, F1o12=ONE/12.0, F1o144=ONE/144.0;
-    const double F9=9.0, F45=45.0, F60=60.0, F27=27.0, F270=270.0, F490=490.0;
-    const double F1o180=ONE/180.0, F1o3600=ONE/3600.0;
-    const double F32=32.0, F128=128.0, F168=168.0, F672=672.0, F840=840.0;
-    const double F1008=1008.0, F8064=8064.0, F14350=14350.0;
-    const double F1o5040=ONE/5040.0, F1o705600=ONE/705600.0;
-
-    const int ex1=ex[0], ex2=ex[1], ex3=ex[2];
-    const double dX=X[1]-X[0], dY=Y[1]-Y[0], dZ=Z[1]-Z[0];
-    const int imaxF=ex1, jmaxF=ex2, kmaxF=ex3;
-    const double SoA[2]={SYM1,SYM2};
-
-#if (ghost_width == 2)
-    {
-        const int ord=1;
-        int iminF=1,jminF=1,kminF=1;
-        if(Symmetry==OCTANT&&fabs(X[0])<dX)iminF=0;
-        if(Symmetry==OCTANT&&fabs(Y[0])<dY)jminF=0;
-        if((sst==2||sst==4)&&fabs(Y[0])<dY)jminF=0;
-
-        const size_t nx=(size_t)ex1+2*ord,ny=(size_t)ex2+2*ord,nz=(size_t)ex3,fh_size=nx*ny*nz;
-        static double *fh_buf=NULL;static size_t cap=0;
-        if(fh_size>cap){free(fh_buf);fh_buf=(double*)aligned_alloc(64,fh_size*sizeof(double));cap=fh_size;}
-        double *fh=fh_buf;if(!fh)return;
-        symmetry_stbd(ord,ex,f,fh,SoA);
-
-        const double Sdxdx=ONE/(dX*dX),Sdydy=ONE/(dY*dY),Sdzdz=ONE/(dZ*dZ);
-        const double Sdxdy=F1o4/(dX*dY),Sdxdz=F1o4/(dX*dZ),Sdydz=F1o4/(dY*dZ);
-        const size_t all=(size_t)ex1*ex2*ex3;
-        for(size_t p=0;p<all;++p){fxx[p]=fyy[p]=fzz[p]=ZEO;fxy[p]=fxz[p]=fyz[p]=ZEO;}
-
-        const int i2_lo=(iminF>0)?iminF:0,j2_lo=(jminF>0)?jminF:0,k2_lo=1,i2_hi=ex1-2,j2_hi=ex2-2,k2_hi=ex3-2;
-        #define FH(iF,jF,kF) fh[idx_fh_stbd(iF,jF,kF,ord,ex)]
-        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);
-                fxx[p]=Sdxdx*(FH(iF-1,jF,kF)-TWO*FH(iF,jF,kF)+FH(iF+1,jF,kF));
-                fyy[p]=Sdydy*(FH(iF,jF-1,kF)-TWO*FH(iF,jF,kF)+FH(iF,jF+1,kF));
-                fzz[p]=Sdzdz*(FH(iF,jF,kF-1)-TWO*FH(iF,jF,kF)+FH(iF,jF,kF+1));
-                fxy[p]=Sdxdy*(FH(iF-1,jF-1,kF)-FH(iF+1,jF-1,kF)-FH(iF-1,jF+1,kF)+FH(iF+1,jF+1,kF));
-                fxz[p]=Sdxdz*(FH(iF-1,jF,kF-1)-FH(iF+1,jF,kF-1)-FH(iF-1,jF,kF+1)+FH(iF+1,jF,kF+1));
-                fyz[p]=Sdydz*(FH(iF,jF-1,kF-1)-FH(iF,jF+1,kF-1)-FH(iF,jF-1,kF+1)+FH(iF,jF+1,kF+1));
-            }}}
-        }
-        #undef FH
-        return;
-    }
-#elif (ghost_width == 3)
-    {
-        const int ord=2;
-        int iminF=1,jminF=1,kminF=1;
-        if(Symmetry==OCTANT&&fabs(X[0])<dX)iminF=-1;
-        if(Symmetry==OCTANT&&fabs(Y[0])<dY)jminF=-1;
-        if((sst==2||sst==4)&&fabs(Y[0])<dY)jminF=-1;
-
-        const size_t nx=(size_t)ex1+2*ord,ny=(size_t)ex2+2*ord,nz=(size_t)ex3,fh_size=nx*ny*nz;
-        static double *fh_buf=NULL;static size_t cap=0;
-        if(fh_size>cap){free(fh_buf);fh_buf=(double*)aligned_alloc(64,fh_size*sizeof(double));cap=fh_size;}
-        double *fh=fh_buf;if(!fh)return;
-        symmetry_stbd(ord,ex,f,fh,SoA);
-
-        const double Sdxdx=ONE/(dX*dX),Sdydy=ONE/(dY*dY),Sdzdz=ONE/(dZ*dZ);
-        const double Fdxdx=F1o12/(dX*dX),Fdydy=F1o12/(dY*dY),Fdzdz=F1o12/(dZ*dZ);
-        const double Sdxdy=F1o4/(dX*dY),Sdxdz=F1o4/(dX*dZ),Sdydz=F1o4/(dY*dZ);
-        const double Fdxdy=F1o144/(dX*dY),Fdxdz=F1o144/(dX*dZ),Fdydz=F1o144/(dY*dZ);
-        const size_t all=(size_t)ex1*ex2*ex3;
-        for(size_t p=0;p<all;++p){fxx[p]=fyy[p]=fzz[p]=fxy[p]=fxz[p]=fyz[p]=ZEO;}
-
-        const int i2_lo=(iminF>0)?iminF:0,j2_lo=(jminF>0)?jminF:0,k2_lo=1,i2_hi=ex1-2,j2_hi=ex2-2,k2_hi=ex3-2;
-        const int i4_lo=(iminF+1>0)?iminF+1:0,j4_lo=(jminF+1>0)?jminF+1:0,k4_lo=2,i4_hi=ex1-3,j4_hi=ex2-3,k4_hi=ex3-3;
-        const int has4=(i4_lo<=i4_hi&&j4_lo<=j4_hi&&k4_lo<=k4_hi);
-        #define FH(iF,jF,kF) fh[idx_fh_stbd(iF,jF,kF,ord,ex)]
-
-        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(iF-1,jF,kF)-TWO*FH(iF,jF,kF)+FH(iF+1,jF,kF));
-                fyy[p]=Sdydy*(FH(iF,jF-1,kF)-TWO*FH(iF,jF,kF)+FH(iF,jF+1,kF));
-                fzz[p]=Sdzdz*(FH(iF,jF,kF-1)-TWO*FH(iF,jF,kF)+FH(iF,jF,kF+1));
-                fxy[p]=Sdxdy*(FH(iF-1,jF-1,kF)-FH(iF+1,jF-1,kF)-FH(iF-1,jF+1,kF)+FH(iF+1,jF+1,kF));
-                fxz[p]=Sdxdz*(FH(iF-1,jF,kF-1)-FH(iF+1,jF,kF-1)-FH(iF-1,jF,kF+1)+FH(iF+1,jF,kF+1));
-                fyz[p]=Sdydz*(FH(iF,jF-1,kF-1)-FH(iF,jF+1,kF-1)-FH(iF,jF-1,kF+1)+FH(iF,jF+1,kF+1));
-            }}}
-        }
-        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(iF-2,jF,kF)+F16*FH(iF-1,jF,kF)-F30*FH(iF,jF,kF)-FH(iF+2,jF,kF)+F16*FH(iF+1,jF,kF));
-                fyy[p]=Fdydy*(-FH(iF,jF-2,kF)+F16*FH(iF,jF-1,kF)-F30*FH(iF,jF,kF)-FH(iF,jF+2,kF)+F16*FH(iF,jF+1,kF));
-                fzz[p]=Fdzdz*(-FH(iF,jF,kF-2)+F16*FH(iF,jF,kF-1)-F30*FH(iF,jF,kF)-FH(iF,jF,kF+2)+F16*FH(iF,jF,kF+1));
-                {const double t_jm2=(FH(iF-2,jF-2,kF)-F8*FH(iF-1,jF-2,kF)+F8*FH(iF+1,jF-2,kF)-FH(iF+2,jF-2,kF));
-                 const double t_jm1=(FH(iF-2,jF-1,kF)-F8*FH(iF-1,jF-1,kF)+F8*FH(iF+1,jF-1,kF)-FH(iF+2,jF-1,kF));
-                 const double t_jp1=(FH(iF-2,jF+1,kF)-F8*FH(iF-1,jF+1,kF)+F8*FH(iF+1,jF+1,kF)-FH(iF+2,jF+1,kF));
-                 const double t_jp2=(FH(iF-2,jF+2,kF)-F8*FH(iF-1,jF+2,kF)+F8*FH(iF+1,jF+2,kF)-FH(iF+2,jF+2,kF));
-                 fxy[p]=Fdxdy*(t_jm2-F8*t_jm1+F8*t_jp1-t_jp2);}
-                {const double t_km2=(FH(iF-2,jF,kF-2)-F8*FH(iF-1,jF,kF-2)+F8*FH(iF+1,jF,kF-2)-FH(iF+2,jF,kF-2));
-                 const double t_km1=(FH(iF-2,jF,kF-1)-F8*FH(iF-1,jF,kF-1)+F8*FH(iF+1,jF,kF-1)-FH(iF+2,jF,kF-1));
-                 const double t_kp1=(FH(iF-2,jF,kF+1)-F8*FH(iF-1,jF,kF+1)+F8*FH(iF+1,jF,kF+1)-FH(iF+2,jF,kF+1));
-                 const double t_kp2=(FH(iF-2,jF,kF+2)-F8*FH(iF-1,jF,kF+2)+F8*FH(iF+1,jF,kF+2)-FH(iF+2,jF,kF+2));
-                 fxz[p]=Fdxdz*(t_km2-F8*t_km1+F8*t_kp1-t_kp2);}
-                {const double t_km2=(FH(iF,jF-2,kF-2)-F8*FH(iF,jF-1,kF-2)+F8*FH(iF,jF+1,kF-2)-FH(iF,jF+2,kF-2));
-                 const double t_km1=(FH(iF,jF-2,kF-1)-F8*FH(iF,jF-1,kF-1)+F8*FH(iF,jF+1,kF-1)-FH(iF,jF+2,kF-1));
-                 const double t_kp1=(FH(iF,jF-2,kF+1)-F8*FH(iF,jF-1,kF+1)+F8*FH(iF,jF+1,kF+1)-FH(iF,jF+2,kF+1));
-                 const double t_kp2=(FH(iF,jF-2,kF+2)-F8*FH(iF,jF-1,kF+2)+F8*FH(iF,jF+1,kF+2)-FH(iF,jF+2,kF+2));
-                 fyz[p]=Fdydz*(t_km2-F8*t_km1+F8*t_kp1-t_kp2);}
-            }}}
-        }
-        #undef FH
-        return;
-    }
-#elif (ghost_width == 4)
-    {
-        const int ord=3;
-        int iminF=1,jminF=1,kminF=1;
-        if(Symmetry==OCTANT&&fabs(X[0])<dX)iminF=-2;
-        if(Symmetry==OCTANT&&fabs(Y[0])<dY)jminF=-2;
-        if((sst==2||sst==4)&&fabs(Y[0])<dY)jminF=-2;
-
-        const size_t nx=(size_t)ex1+2*ord,ny=(size_t)ex2+2*ord,nz=(size_t)ex3,fh_size=nx*ny*nz;
-        static double *fh_buf=NULL;static size_t cap=0;
-        if(fh_size>cap){free(fh_buf);fh_buf=(double*)aligned_alloc(64,fh_size*sizeof(double));cap=fh_size;}
-        double *fh=fh_buf;if(!fh)return;
-        symmetry_stbd(ord,ex,f,fh,SoA);
-
-        const double Sdxdx=ONE/(dX*dX),Sdydy=ONE/(dY*dY),Sdzdz=ONE/(dZ*dZ);
-        const double Fdxdx=F1o12/(dX*dX),Fdydy=F1o12/(dY*dY),Fdzdz=F1o12/(dZ*dZ);
-        const double Xdxdx=F1o180/(dX*dX),Xdydy=F1o180/(dY*dY),Xdzdz=F1o180/(dZ*dZ);
-        const double Sdxdy=F1o4/(dX*dY),Sdxdz=F1o4/(dX*dZ),Sdydz=F1o4/(dY*dZ);
-        const double Fdxdy=F1o144/(dX*dY),Fdxdz=F1o144/(dX*dZ),Fdydz=F1o144/(dY*dZ);
-        const double Xdxdy=F1o3600/(dX*dY),Xdxdz=F1o3600/(dX*dZ),Xdydz=F1o3600/(dY*dZ);
-        const size_t all=(size_t)ex1*ex2*ex3;
-        for(size_t p=0;p<all;++p){fxx[p]=fyy[p]=fzz[p]=fxy[p]=fxz[p]=fyz[p]=ZEO;}
-
-        const int i2_lo=(iminF>0)?iminF:0,j2_lo=(jminF>0)?jminF:0,k2_lo=1,i2_hi=ex1-2,j2_hi=ex2-2,k2_hi=ex3-2;
-        const int i4_lo=(iminF+1>0)?iminF+1:0,j4_lo=(jminF+1>0)?jminF+1:0,k4_lo=2,i4_hi=ex1-3,j4_hi=ex2-3,k4_hi=ex3-3;
-        const int i6_lo=(iminF+2>0)?iminF+2:0,j6_lo=(jminF+2>0)?jminF+2:0,k6_lo=3,i6_hi=ex1-4,j6_hi=ex2-4,k6_hi=ex3-4;
-        const int has4=(i4_lo<=i4_hi&&j4_lo<=j4_hi&&k4_lo<=k4_hi),has6=(i6_lo<=i6_hi&&j6_lo<=j6_hi&&k6_lo<=k6_hi);
-        #define FH(iF,jF,kF) fh[idx_fh_stbd(iF,jF,kF,ord,ex)]
-
-        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){bool in4=has4&&i0>=i4_lo&&i0<=i4_hi&&j0>=j4_lo&&j0<=j4_hi&&k0>=k4_lo&&k0<=k4_hi;if(in4)continue;
-            const int iF=i0+1;const size_t p=idx_ex(i0,j0,k0,ex);
-            fxx[p]=Sdxdx*(FH(iF-1,jF,kF)-TWO*FH(iF,jF,kF)+FH(iF+1,jF,kF));
-            fyy[p]=Sdydy*(FH(iF,jF-1,kF)-TWO*FH(iF,jF,kF)+FH(iF,jF+1,kF));
-            fzz[p]=Sdzdz*(FH(iF,jF,kF-1)-TWO*FH(iF,jF,kF)+FH(iF,jF,kF+1));
-            fxy[p]=Sdxdy*(FH(iF-1,jF-1,kF)-FH(iF+1,jF-1,kF)-FH(iF-1,jF+1,kF)+FH(iF+1,jF+1,kF));
-            fxz[p]=Sdxdz*(FH(iF-1,jF,kF-1)-FH(iF+1,jF,kF-1)-FH(iF-1,jF,kF+1)+FH(iF+1,jF,kF+1));
-            fyz[p]=Sdydz*(FH(iF,jF-1,kF-1)-FH(iF,jF+1,kF-1)-FH(iF,jF-1,kF+1)+FH(iF,jF+1,kF+1));
-        }}}}
-        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){if(has6&&i0>=i6_lo&&i0<=i6_hi&&j0>=j6_lo&&j0<=j6_hi&&k0>=k6_lo&&k0<=k6_hi)continue;
-            const int iF=i0+1;const size_t p=idx_ex(i0,j0,k0,ex);
-            fxx[p]=Fdxdx*(-FH(iF-2,jF,kF)+F16*FH(iF-1,jF,kF)-F30*FH(iF,jF,kF)-FH(iF+2,jF,kF)+F16*FH(iF+1,jF,kF));
-            fyy[p]=Fdydy*(-FH(iF,jF-2,kF)+F16*FH(iF,jF-1,kF)-F30*FH(iF,jF,kF)-FH(iF,jF+2,kF)+F16*FH(iF,jF+1,kF));
-            fzz[p]=Fdzdz*(-FH(iF,jF,kF-2)+F16*FH(iF,jF,kF-1)-F30*FH(iF,jF,kF)-FH(iF,jF,kF+2)+F16*FH(iF,jF,kF+1));
-            {const double t_jm2=(FH(iF-2,jF-2,kF)-F8*FH(iF-1,jF-2,kF)+F8*FH(iF+1,jF-2,kF)-FH(iF+2,jF-2,kF));
-             const double t_jm1=(FH(iF-2,jF-1,kF)-F8*FH(iF-1,jF-1,kF)+F8*FH(iF+1,jF-1,kF)-FH(iF+2,jF-1,kF));
-             const double t_jp1=(FH(iF-2,jF+1,kF)-F8*FH(iF-1,jF+1,kF)+F8*FH(iF+1,jF+1,kF)-FH(iF+2,jF+1,kF));
-             const double t_jp2=(FH(iF-2,jF+2,kF)-F8*FH(iF-1,jF+2,kF)+F8*FH(iF+1,jF+2,kF)-FH(iF+2,jF+2,kF));
-             fxy[p]=Fdxdy*(t_jm2-F8*t_jm1+F8*t_jp1-t_jp2);}
-            {const double t_km2=(FH(iF-2,jF,kF-2)-F8*FH(iF-1,jF,kF-2)+F8*FH(iF+1,jF,kF-2)-FH(iF+2,jF,kF-2));
-             const double t_km1=(FH(iF-2,jF,kF-1)-F8*FH(iF-1,jF,kF-1)+F8*FH(iF+1,jF,kF-1)-FH(iF+2,jF,kF-1));
-             const double t_kp1=(FH(iF-2,jF,kF+1)-F8*FH(iF-1,jF,kF+1)+F8*FH(iF+1,jF,kF+1)-FH(iF+2,jF,kF+1));
-             const double t_kp2=(FH(iF-2,jF,kF+2)-F8*FH(iF-1,jF,kF+2)+F8*FH(iF+1,jF,kF+2)-FH(iF+2,jF,kF+2));
-             fxz[p]=Fdxdz*(t_km2-F8*t_km1+F8*t_kp1-t_kp2);}
-            {const double t_km2=(FH(iF,jF-2,kF-2)-F8*FH(iF,jF-1,kF-2)+F8*FH(iF,jF+1,kF-2)-FH(iF,jF+2,kF-2));
-             const double t_km1=(FH(iF,jF-2,kF-1)-F8*FH(iF,jF-1,kF-1)+F8*FH(iF,jF+1,kF-1)-FH(iF,jF+2,kF-1));
-             const double t_kp1=(FH(iF,jF-2,kF+1)-F8*FH(iF,jF-1,kF+1)+F8*FH(iF,jF+1,kF+1)-FH(iF,jF+2,kF+1));
-             const double t_kp2=(FH(iF,jF-2,kF+2)-F8*FH(iF,jF-1,kF+2)+F8*FH(iF,jF+1,kF+2)-FH(iF,jF+2,kF+2));
-             fyz[p]=Fdydz*(t_km2-F8*t_km1+F8*t_kp1-t_kp2);}
-        }}}}
-        if(has6){for(int k0=k6_lo;k0<=k6_hi;++k0){const int kF=k0+1;
-        for(int j0=j6_lo;j0<=j6_hi;++j0){const int jF=j0+1;
-        for(int i0=i6_lo;i0<=i6_hi;++i0){const int iF=i0+1;const size_t p=idx_ex(i0,j0,k0,ex);
-            fxx[p]=Xdxdx*(TWO*FH(iF-3,jF,kF)-F27*FH(iF-2,jF,kF)+F270*FH(iF-1,jF,kF)-F490*FH(iF,jF,kF)+F270*FH(iF+1,jF,kF)-F27*FH(iF+2,jF,kF)+TWO*FH(iF+3,jF,kF));
-            fyy[p]=Xdydy*(TWO*FH(iF,jF-3,kF)-F27*FH(iF,jF-2,kF)+F270*FH(iF,jF-1,kF)-F490*FH(iF,jF,kF)+F270*FH(iF,jF+1,kF)-F27*FH(iF,jF+2,kF)+TWO*FH(iF,jF+3,kF));
-            fzz[p]=Xdzdz*(TWO*FH(iF,jF,kF-3)-F27*FH(iF,jF,kF-2)+F270*FH(iF,jF,kF-1)-F490*FH(iF,jF,kF)+F270*FH(iF,jF,kF+1)-F27*FH(iF,jF,kF+2)+TWO*FH(iF,jF,kF+3));
-            #define XS6(JF,KFDUMMY) (-FH(iF-3,JF,KFDUMMY)+F9*FH(iF-2,JF,KFDUMMY)-F45*FH(iF-1,JF,KFDUMMY)+F45*FH(iF+1,JF,KFDUMMY)-F9*FH(iF+2,JF,KFDUMMY)+FH(iF+3,JF,KFDUMMY))
-            fxy[p]=Xdxdy*(-XS6(jF-3,kF)+F9*XS6(jF-2,kF)-F45*XS6(jF-1,kF)+F45*XS6(jF+1,kF)-F9*XS6(jF+2,kF)+XS6(jF+3,kF));
-            fxz[p]=Xdxdz*(-XS6(jF,kF-3)+F9*XS6(jF,kF-2)-F45*XS6(jF,kF-1)+F45*XS6(jF,kF+1)-F9*XS6(jF,kF+2)+XS6(jF,kF+3));
-            #undef XS6
-            #define YS6(JF,KFDUMMY) (-FH(iF,JF-3,KFDUMMY)+F9*FH(iF,JF-2,KFDUMMY)-F45*FH(iF,JF-1,KFDUMMY)+F45*FH(iF,JF+1,KFDUMMY)-F9*FH(iF,JF+2,KFDUMMY)+FH(iF,JF+3,KFDUMMY))
-            fyz[p]=Xdydz*(-YS6(jF,kF-3)+F9*YS6(jF,kF-2)-F45*YS6(jF,kF-1)+F45*YS6(jF,kF+1)-F9*YS6(jF,kF+2)+YS6(jF,kF+3));
-            #undef YS6
-        }}}}
-        #undef FH
-        return;
-    }
-#elif (ghost_width == 5)
-    {
-        /* 8th-order shell second derivatives — inherits 8th-order stencil coeffs from Cartesian */
-        const int ord=4;
-        int iminF=1,jminF=1,kminF=1;
-        if(Symmetry==OCTANT&&fabs(X[0])<dX)iminF=-3;
-        if(Symmetry==OCTANT&&fabs(Y[0])<dY)jminF=-3;
-        if((sst==2||sst==4)&&fabs(Y[0])<dY)jminF=-3;
-
-        const size_t nx=(size_t)ex1+2*ord,ny=(size_t)ex2+2*ord,nz=(size_t)ex3,fh_size=nx*ny*nz;
-        static double *fh_buf=NULL;static size_t cap=0;
-        if(fh_size>cap){free(fh_buf);fh_buf=(double*)aligned_alloc(64,fh_size*sizeof(double));cap=fh_size;}
-        double *fh=fh_buf;if(!fh)return;
-        symmetry_stbd(ord,ex,f,fh,SoA);
-
-        const double Sdxdx=ONE/(dX*dX),Sdydy=ONE/(dY*dY),Sdzdz=ONE/(dZ*dZ);
-        const double Fdxdx=F1o12/(dX*dX),Fdydy=F1o12/(dY*dY),Fdzdz=F1o12/(dZ*dZ);
-        const double Xdxdx=F1o180/(dX*dX),Xdydy=F1o180/(dY*dY),Xdzdz=F1o180/(dZ*dZ);
-        const double Edxdx=F1o5040/(dX*dX),Edydy=F1o5040/(dY*dY),Edzdz=F1o5040/(dZ*dZ);
-        const double Sdxdy=F1o4/(dX*dY),Sdxdz=F1o4/(dX*dZ),Sdydz=F1o4/(dY*dZ);
-        const double Fdxdy=F1o144/(dX*dY),Fdxdz=F1o144/(dX*dZ),Fdydz=F1o144/(dY*dZ);
-        const double Xdxdy=F1o3600/(dX*dY),Xdxdz=F1o3600/(dX*dZ),Xdydz=F1o3600/(dY*dZ);
-        const double Edxdy=F1o705600/(dX*dY),Edxdz=F1o705600/(dX*dZ),Edydz=F1o705600/(dY*dZ);
-        const size_t all=(size_t)ex1*ex2*ex3;
-        for(size_t p=0;p<all;++p){fxx[p]=fyy[p]=fzz[p]=fxy[p]=fxz[p]=fyz[p]=ZEO;}
-
-        const int i2_lo=(iminF>0)?iminF:0,j2_lo=(jminF>0)?jminF:0,k2_lo=1,i2_hi=ex1-2,j2_hi=ex2-2,k2_hi=ex3-2;
-        const int i4_lo=(iminF+1>0)?iminF+1:0,j4_lo=(jminF+1>0)?jminF+1:0,k4_lo=2,i4_hi=ex1-3,j4_hi=ex2-3,k4_hi=ex3-3;
-        const int i6_lo=(iminF+2>0)?iminF+2:0,j6_lo=(jminF+2>0)?jminF+2:0,k6_lo=3,i6_hi=ex1-4,j6_hi=ex2-4,k6_hi=ex3-4;
-        const int i8_lo=(iminF+3>0)?iminF+3:0,j8_lo=(jminF+3>0)?jminF+3:0,k8_lo=4,i8_hi=ex1-5,j8_hi=ex2-5,k8_hi=ex3-5;
-        const int has4=(i4_lo<=i4_hi&&j4_lo<=j4_hi&&k4_lo<=k4_hi),has6=(i6_lo<=i6_hi&&j6_lo<=j6_hi&&k6_lo<=k6_hi),has8=(i8_lo<=i8_hi&&j8_lo<=j8_hi&&k8_lo<=k8_hi);
-        #define FH(iF,jF,kF) fh[idx_fh_stbd(iF,jF,kF,ord,ex)]
-
-        /* 2nd-order pass */
-        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){bool in4=has4&&i0>=i4_lo&&i0<=i4_hi&&j0>=j4_lo&&j0<=j4_hi&&k0>=k4_lo&&k0<=k4_hi;if(in4)continue;
-            const int iF=i0+1;const size_t p=idx_ex(i0,j0,k0,ex);
-            fxx[p]=Sdxdx*(FH(iF-1,jF,kF)-TWO*FH(iF,jF,kF)+FH(iF+1,jF,kF));
-            fyy[p]=Sdydy*(FH(iF,jF-1,kF)-TWO*FH(iF,jF,kF)+FH(iF,jF+1,kF));
-            fzz[p]=Sdzdz*(FH(iF,jF,kF-1)-TWO*FH(iF,jF,kF)+FH(iF,jF,kF+1));
-            fxy[p]=Sdxdy*(FH(iF-1,jF-1,kF)-FH(iF+1,jF-1,kF)-FH(iF-1,jF+1,kF)+FH(iF+1,jF+1,kF));
-            fxz[p]=Sdxdz*(FH(iF-1,jF,kF-1)-FH(iF+1,jF,kF-1)-FH(iF-1,jF,kF+1)+FH(iF+1,jF,kF+1));
-            fyz[p]=Sdydz*(FH(iF,jF-1,kF-1)-FH(iF,jF+1,kF-1)-FH(iF,jF-1,kF+1)+FH(iF,jF+1,kF+1));
-        }}}}
-        /* 4th-order pass */
-        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){bool in6=has6&&i0>=i6_lo&&i0<=i6_hi&&j0>=j6_lo&&j0<=j6_hi&&k0>=k6_lo&&k0<=k6_hi;if(in6)continue;
-            const int iF=i0+1;const size_t p=idx_ex(i0,j0,k0,ex);
-            fxx[p]=Fdxdx*(-FH(iF-2,jF,kF)+F16*FH(iF-1,jF,kF)-F30*FH(iF,jF,kF)-FH(iF+2,jF,kF)+F16*FH(iF+1,jF,kF));
-            fyy[p]=Fdydy*(-FH(iF,jF-2,kF)+F16*FH(iF,jF-1,kF)-F30*FH(iF,jF,kF)-FH(iF,jF+2,kF)+F16*FH(iF,jF+1,kF));
-            fzz[p]=Fdzdz*(-FH(iF,jF,kF-2)+F16*FH(iF,jF,kF-1)-F30*FH(iF,jF,kF)-FH(iF,jF,kF+2)+F16*FH(iF,jF,kF+1));
-            {const double t_jm2=(FH(iF-2,jF-2,kF)-F8*FH(iF-1,jF-2,kF)+F8*FH(iF+1,jF-2,kF)-FH(iF+2,jF-2,kF));
-             const double t_jm1=(FH(iF-2,jF-1,kF)-F8*FH(iF-1,jF-1,kF)+F8*FH(iF+1,jF-1,kF)-FH(iF+2,jF-1,kF));
-             const double t_jp1=(FH(iF-2,jF+1,kF)-F8*FH(iF-1,jF+1,kF)+F8*FH(iF+1,jF+1,kF)-FH(iF+2,jF+1,kF));
-             const double t_jp2=(FH(iF-2,jF+2,kF)-F8*FH(iF-1,jF+2,kF)+F8*FH(iF+1,jF+2,kF)-FH(iF+2,jF+2,kF));
-             fxy[p]=Fdxdy*(t_jm2-F8*t_jm1+F8*t_jp1-t_jp2);}
-            {const double t_km2=(FH(iF-2,jF,kF-2)-F8*FH(iF-1,jF,kF-2)+F8*FH(iF+1,jF,kF-2)-FH(iF+2,jF,kF-2));
-             const double t_km1=(FH(iF-2,jF,kF-1)-F8*FH(iF-1,jF,kF-1)+F8*FH(iF+1,jF,kF-1)-FH(iF+2,jF,kF-1));
-             const double t_kp1=(FH(iF-2,jF,kF+1)-F8*FH(iF-1,jF,kF+1)+F8*FH(iF+1,jF,kF+1)-FH(iF+2,jF,kF+1));
-             const double t_kp2=(FH(iF-2,jF,kF+2)-F8*FH(iF-1,jF,kF+2)+F8*FH(iF+1,jF,kF+2)-FH(iF+2,jF,kF+2));
-             fxz[p]=Fdxdz*(t_km2-F8*t_km1+F8*t_kp1-t_kp2);}
-            {const double t_km2=(FH(iF,jF-2,kF-2)-F8*FH(iF,jF-1,kF-2)+F8*FH(iF,jF+1,kF-2)-FH(iF,jF+2,kF-2));
-             const double t_km1=(FH(iF,jF-2,kF-1)-F8*FH(iF,jF-1,kF-1)+F8*FH(iF,jF+1,kF-1)-FH(iF,jF+2,kF-1));
-             const double t_kp1=(FH(iF,jF-2,kF+1)-F8*FH(iF,jF-1,kF+1)+F8*FH(iF,jF+1,kF+1)-FH(iF,jF+2,kF+1));
-             const double t_kp2=(FH(iF,jF-2,kF+2)-F8*FH(iF,jF-1,kF+2)+F8*FH(iF,jF+1,kF+2)-FH(iF,jF+2,kF+2));
-             fyz[p]=Fdydz*(t_km2-F8*t_km1+F8*t_kp1-t_kp2);}
-        }}}}
-        /* 6th-order pass */
-        if(has6){for(int k0=k6_lo;k0<=k6_hi;++k0){const int kF=k0+1;
-        for(int j0=j6_lo;j0<=j6_hi;++j0){const int jF=j0+1;
-        for(int i0=i6_lo;i0<=i6_hi;++i0){if(has8&&i0>=i8_lo&&i0<=i8_hi&&j0>=j8_lo&&j0<=j8_hi&&k0>=k8_lo&&k0<=k8_hi)continue;
-            const int iF=i0+1;const size_t p=idx_ex(i0,j0,k0,ex);
-            fxx[p]=Xdxdx*(TWO*FH(iF-3,jF,kF)-F27*FH(iF-2,jF,kF)+F270*FH(iF-1,jF,kF)-F490*FH(iF,jF,kF)+F270*FH(iF+1,jF,kF)-F27*FH(iF+2,jF,kF)+TWO*FH(iF+3,jF,kF));
-            fyy[p]=Xdydy*(TWO*FH(iF,jF-3,kF)-F27*FH(iF,jF-2,kF)+F270*FH(iF,jF-1,kF)-F490*FH(iF,jF,kF)+F270*FH(iF,jF+1,kF)-F27*FH(iF,jF+2,kF)+TWO*FH(iF,jF+3,kF));
-            fzz[p]=Xdzdz*(TWO*FH(iF,jF,kF-3)-F27*FH(iF,jF,kF-2)+F270*FH(iF,jF,kF-1)-F490*FH(iF,jF,kF)+F270*FH(iF,jF,kF+1)-F27*FH(iF,jF,kF+2)+TWO*FH(iF,jF,kF+3));
-            #define XS6_8(JF,KFDUMMY) (-FH(iF-3,JF,KFDUMMY)+F9*FH(iF-2,JF,KFDUMMY)-F45*FH(iF-1,JF,KFDUMMY)+F45*FH(iF+1,JF,KFDUMMY)-F9*FH(iF+2,JF,KFDUMMY)+FH(iF+3,JF,KFDUMMY))
-            fxy[p]=Xdxdy*(-XS6_8(jF-3,kF)+F9*XS6_8(jF-2,kF)-F45*XS6_8(jF-1,kF)+F45*XS6_8(jF+1,kF)-F9*XS6_8(jF+2,kF)+XS6_8(jF+3,kF));
-            fxz[p]=Xdxdz*(-XS6_8(jF,kF-3)+F9*XS6_8(jF,kF-2)-F45*XS6_8(jF,kF-1)+F45*XS6_8(jF,kF+1)-F9*XS6_8(jF,kF+2)+XS6_8(jF,kF+3));
-            #undef XS6_8
-            #define YS6_8(JF,KFDUMMY) (-FH(iF,JF-3,KFDUMMY)+F9*FH(iF,JF-2,KFDUMMY)-F45*FH(iF,JF-1,KFDUMMY)+F45*FH(iF,JF+1,KFDUMMY)-F9*FH(iF,JF+2,KFDUMMY)+FH(iF,JF+3,KFDUMMY))
-            fyz[p]=Xdydz*(-YS6_8(jF,kF-3)+F9*YS6_8(jF,kF-2)-F45*YS6_8(jF,kF-1)+F45*YS6_8(jF,kF+1)-F9*YS6_8(jF,kF+2)+YS6_8(jF,kF+3));
-            #undef YS6_8
-        }}}}
-        /* 8th-order pass */
-        if(has8){for(int k0=k8_lo;k0<=k8_hi;++k0){const int kF=k0+1;
-        for(int j0=j8_lo;j0<=j8_hi;++j0){const int jF=j0+1;
-        for(int i0=i8_lo;i0<=i8_hi;++i0){const int iF=i0+1;const size_t p=idx_ex(i0,j0,k0,ex);
-            fxx[p]=Edxdx*(-(double)9*FH(iF-4,jF,kF)+F128*FH(iF-3,jF,kF)-F1008*FH(iF-2,jF,kF)+F8064*FH(iF-1,jF,kF)-F14350*FH(iF,jF,kF)+F8064*FH(iF+1,jF,kF)-F1008*FH(iF+2,jF,kF)+F128*FH(iF+3,jF,kF)-(double)9*FH(iF+4,jF,kF));
-            fyy[p]=Edydy*(-(double)9*FH(iF,jF-4,kF)+F128*FH(iF,jF-3,kF)-F1008*FH(iF,jF-2,kF)+F8064*FH(iF,jF-1,kF)-F14350*FH(iF,jF,kF)+F8064*FH(iF,jF+1,kF)-F1008*FH(iF,jF+2,kF)+F128*FH(iF,jF+3,kF)-(double)9*FH(iF,jF+4,kF));
-            fzz[p]=Edzdz*(-(double)9*FH(iF,jF,kF-4)+F128*FH(iF,jF,kF-3)-F1008*FH(iF,jF,kF-2)+F8064*FH(iF,jF,kF-1)-F14350*FH(iF,jF,kF)+F8064*FH(iF,jF,kF+1)-F1008*FH(iF,jF,kF+2)+F128*FH(iF,jF,kF+3)-(double)9*FH(iF,jF,kF+4));
-            #define XS8(JF,KFDUMMY) (+(double)3*FH(iF-4,JF,KFDUMMY)-F32*FH(iF-3,JF,KFDUMMY)+F168*FH(iF-2,JF,KFDUMMY)-F672*FH(iF-1,JF,KFDUMMY)+F672*FH(iF+1,JF,KFDUMMY)-F168*FH(iF+2,JF,KFDUMMY)+F32*FH(iF+3,JF,KFDUMMY)-(double)3*FH(iF+4,JF,KFDUMMY))
-            fxy[p]=Edxdy*(+(double)3*XS8(jF-4,kF)-F32*XS8(jF-3,kF)+F168*XS8(jF-2,kF)-F672*XS8(jF-1,kF)+F672*XS8(jF+1,kF)-F168*XS8(jF+2,kF)+F32*XS8(jF+3,kF)-(double)3*XS8(jF+4,kF));
-            fxz[p]=Edxdz*(+(double)3*XS8(jF,kF-4)-F32*XS8(jF,kF-3)+F168*XS8(jF,kF-2)-F672*XS8(jF,kF-1)+F672*XS8(jF,kF+1)-F168*XS8(jF,kF+2)+F32*XS8(jF,kF+3)-(double)3*XS8(jF,kF+4));
-            #undef XS8
-            #define YS8(JF,KFDUMMY) (+(double)3*FH(iF,JF-4,KFDUMMY)-F32*FH(iF,JF-3,KFDUMMY)+F168*FH(iF,JF-2,KFDUMMY)-F672*FH(iF,JF-1,KFDUMMY)+F672*FH(iF,JF+1,KFDUMMY)-F168*FH(iF,JF+2,KFDUMMY)+F32*FH(iF,JF+3,KFDUMMY)-(double)3*FH(iF,JF+4,KFDUMMY))
-            fyz[p]=Edydz*(+(double)3*YS8(jF,kF-4)-F32*YS8(jF,kF-3)+F168*YS8(jF,kF-2)-F672*YS8(jF,kF-1)+F672*YS8(jF,kF+1)-F168*YS8(jF,kF+2)+F32*YS8(jF,kF+3)-(double)3*YS8(jF,kF+4));
-            #undef YS8
-        }}}}
-        #undef FH
-        return;
-    }
-#else
-#error "fdderivs_sh_c.C: unsupported ghost_width"
-#endif
-}

AMSS_NCKU_source/fdderivs_shc_c.C

@@ -1,107 +0,0 @@
-#include "macrodef.h"
-#include "share_func.h"
-#include <cstddef>
-
-/* Forward declarations — Fortran-mangled names from shell C kernels */
-extern "C" {
-void fderivs_sh_(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, int sst);
-
-void fdderivs_sh_(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, int sst);
-
-void fdderivs_shc_(int *ex,
-                    double *f,
-                    double *fxx, double *fxy, double *fxz,
-                    double *fyy, double *fyz, double *fzz,
-                    double *crho, double *sigma, double *R,
-                    double &SYM1, double &SYM2, double &SYM3,
-                    int &Symmetry, int &Lev, int &sst,
-                    double *drhodx, double *drhody, double *drhodz,
-                    double *dsigmadx, double *dsigmady, double *dsigmadz,
-                    double *dRdx, double *dRdy, double *dRdz,
-                    double *drhodxx, double *drhodxy, double *drhodxz,
-                    double *drhodyy, double *drhodyz, double *drhodzz,
-                    double *dsigmadxx, double *dsigmadxy, double *dsigmadxz,
-                    double *dsigmadyy, double *dsigmadyz, double *dsigmadzz,
-                    double *dRdxx, double *dRdxy, double *dRdxz,
-                    double *dRdyy, double *dRdyz, double *dRdzz)
-{
-    const int ex3[3] = { ex[0], ex[1], ex[2] };
-    const size_t n = (size_t)ex[0] * (size_t)ex[1] * (size_t)ex[2];
-
-    double *gx  = (double*)malloc(n * sizeof(double));
-    double *gy  = (double*)malloc(n * sizeof(double));
-    double *gz  = (double*)malloc(n * sizeof(double));
-    double *gxx = (double*)malloc(n * sizeof(double));
-    double *gxy = (double*)malloc(n * sizeof(double));
-    double *gxz = (double*)malloc(n * sizeof(double));
-    double *gyy = (double*)malloc(n * sizeof(double));
-    double *gyz = (double*)malloc(n * sizeof(double));
-    double *gzz = (double*)malloc(n * sizeof(double));
-
-    if (!gx||!gy||!gz||!gxx||!gxy||!gxz||!gyy||!gyz||!gzz) {
-        free(gx);free(gy);free(gz);free(gxx);free(gxy);free(gxz);free(gyy);free(gyz);free(gzz);
-        return;
-    }
-
-    fderivs_sh_(ex3, f, gx, gy, gz, crho, sigma, R, SYM1, SYM2, SYM3, Symmetry, Lev, sst);
-    fdderivs_sh_(ex3, f, gxx, gxy, gxz, gyy, gyz, gzz, crho, sigma, R, SYM1, SYM2, SYM3, Symmetry, Lev, sst);
-
-    for (size_t i = 0; i < n; ++i) {
-        const double rx=drhodx[i], ry=drhody[i], rz=drhodz[i];
-        const double sx=dsigmadx[i], sy=dsigmady[i], sz=dsigmadz[i];
-        const double Rx=dRdx[i], Ry=dRdy[i], Rz=dRdz[i];
-        const double rxx=drhodxx[i], rxy=drhodxy[i], rxz=drhodxz[i];
-        const double ryy=drhodyy[i], ryz=drhodyz[i], rzz=drhodzz[i];
-        const double sxx=dsigmadxx[i], sxy=dsigmadxy[i], sxz=dsigmadxz[i];
-        const double syy=dsigmadyy[i], syz=dsigmadyz[i], szz=dsigmadzz[i];
-        const double Rxx=dRdxx[i], Rxy=dRdxy[i], Rxz=dRdxz[i];
-        const double Ryy=dRdyy[i], Ryz=dRdyz[i], Rzz=dRdzz[i];
-
-        const double Gr=gx[i], Gs=gy[i], GR=gz[i];
-        const double Grr=gxx[i], Grs=gxy[i], GrR=gxz[i];
-        const double Gss=gyy[i], GsR=gyz[i], GRR=gzz[i];
-
-        /* fxx */
-        fxx[i] = rx*rx*Grr + sx*sx*Gss + Rx*Rx*GRR
-               + 2.0*(rx*sx*Grs + rx*Rx*GrR + sx*Rx*GsR)
-               + rxx*Gr + sxx*Gs + Rxx*GR;
-
-        /* fxy */
-        fxy[i] = rx*ry*Grr + sx*sy*Gss + Rx*Ry*GRR
-               + rx*sy*Grs + ry*sx*Grs + rx*Ry*GrR + ry*Rx*GrR + sx*Ry*GsR + sy*Rx*GsR
-               + rxy*Gr + sxy*Gs + Rxy*GR;
-
-        /* fxz */
-        fxz[i] = rx*rz*Grr + sx*sz*Gss + Rx*Rz*GRR
-               + rx*sz*Grs + rz*sx*Grs + rx*Rz*GrR + rz*Rx*GrR + sx*Rz*GsR + sz*Rx*GsR
-               + rxz*Gr + sxz*Gs + Rxz*GR;
-
-        /* fyy */
-        fyy[i] = ry*ry*Grr + sy*sy*Gss + Ry*Ry*GRR
-               + 2.0*(ry*sy*Grs + ry*Ry*GrR + sy*Ry*GsR)
-               + ryy*Gr + syy*Gs + Ryy*GR;
-
-        /* fyz */
-        fyz[i] = ry*rz*Grr + sy*sz*Gss + Ry*Rz*GRR
-               + ry*sz*Grs + rz*sy*Grs + ry*Rz*GrR + rz*Ry*GrR + sy*Rz*GsR + sz*Ry*GsR
-               + ryz*Gr + syz*Gs + Ryz*GR;
-
-        /* fzz */
-        fzz[i] = rz*rz*Grr + sz*sz*Gss + Rz*Rz*GRR
-               + 2.0*(rz*sz*Grs + rz*Rz*GrR + sz*Rz*GsR)
-               + rzz*Gr + szz*Gs + Rzz*GR;
-    }
-
-    free(gx);free(gy);free(gz);free(gxx);free(gxy);free(gxz);free(gyy);free(gyz);free(gzz);
-}
-
-} // extern "C"

AMSS_NCKU_source/fderivs_c.C

@@ -1,18 +1,14 @@
-#include "macrodef.h"
 #include "tool.h"
 
 /*
- * C 版 fderivs — first derivatives df/dx, df/dy, df/dz.
+ * C 版 fderivs
  *
- * Finite difference order is selected at compile time via the ghost_width macro
- * (defined in macrodef.fh):
- *   ghost_width = 2 → 2nd-order
- *   ghost_width = 3 → 4th-order
- *   ghost_width = 4 → 6th-order
- *   ghost_width = 5 → 8th-order
+ * Fortran:
+ * subroutine fderivs(ex,f,fx,fy,fz,X,Y,Z,SYM1,SYM2,SYM3,symmetry,onoff)
  *
- * Multi-pass overwrite strategy: compute the widest (lowest-order) stencil first,
- * then overwrite interior regions with progressively higher-order stencils.
+ * 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,
@@ -21,596 +17,151 @@ void fderivs(const int ex[3],
              double SYM1, double SYM2, double SYM3,
              int Symmetry, int onoff)
 {
-    (void)onoff;
+    (void)onoff; // Fortran 里没用到
 
-    const double ZEO = 0.0, ONE = 1.0, TWO = 2.0, EIT = 8.0;
-    const double F9 = 9.0, F12 = 12.0, F45 = 45.0, F60 = 60.0;
-    const double F32 = 32.0, F168 = 168.0, F672 = 672.0, F840 = 840.0;
+    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;
+    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];
 
-    const int imaxF = ex1, jmaxF = ex2, kmaxF = ex3;
+    // Fortran 1-based bounds
+    const int imaxF = ex1;
+    const int jmaxF = ex2;
+    const int kmaxF = ex3;
 
-    const int gw = ghost_width; // compile-time constant
+    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;
 
-#if (ghost_width == 2)
-    /* ---- 2nd-order ------------------------------------------------------ */
-    {
-        const int ord = 1; // symmetry_bd ord = ghost_width - 1
+    // SoA(1:3) = SYM1,SYM2,SYM3
+    const double SoA[3] = { SYM1, SYM2, SYM3 };
 
-        int iminF = 1, jminF = 1, kminF = 1;
-        if (Symmetry > NO_SYMM && fabs(Z[0]) < dZ) kminF = 0;
-        if (Symmetry > EQ_SYMM && fabs(X[0]) < dX) iminF = 0;
-        if (Symmetry > EQ_SYMM && fabs(Y[0]) < dY) jminF = 0;
+    // 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;
 
-        const double SoA[3] = { SYM1, SYM2, SYM3 };
-
-        const size_t nx = (size_t)ex1 + ord;
-        const size_t ny = (size_t)ex2 + ord;
-        const size_t nz = (size_t)ex3 + ord;
-        const size_t fh_size = nx * ny * nz;
-
-        static double *fh_buf = NULL;
-        static size_t cap = 0;
-        if (fh_size > cap) {
-            free(fh_buf);
-            fh_buf = (double*)aligned_alloc(64, fh_size * sizeof(double));
-            cap = fh_size;
-        }
-        double *fh = fh_buf;
-        if (!fh) return;
-
-        symmetry_bd(ord, ex, f, fh, SoA);
-
-        const double d2dx = ONE / TWO / dX;
-        const double d2dy = ONE / TWO / dY;
-        const double d2dz = ONE / TWO / dZ;
-
-        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;
-        }
-
-        /* 2nd-order pass: [-1, 0, +1] / (2*dx) */
-        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;
-
-        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_ord1(iF - 1, jF,     kF,     ex)] +
-                             fh[idx_fh_F_ord1(iF + 1, jF,     kF,     ex)]
-                        );
-
-                        fy[p] = d2dy * (
-                            -fh[idx_fh_F_ord1(iF,     jF - 1, kF,     ex)] +
-                             fh[idx_fh_F_ord1(iF,     jF + 1, kF,     ex)]
-                        );
-
-                        fz[p] = d2dz * (
-                            -fh[idx_fh_F_ord1(iF,     jF,     kF - 1, ex)] +
-                             fh[idx_fh_F_ord1(iF,     jF,     kF + 1, ex)]
-                        );
-                    }
-                }
-            }
-        }
-        return;
+    if (fh_size > cap) {
+        free(fh);
+        fh = (double*)aligned_alloc(64, fh_size * sizeof(double));
+        cap = fh_size;
     }
-#elif (ghost_width == 3)
-    /* ---- 4th-order (original code) ------------------------------------ */
-    {
-        const int ord = 2; // symmetry_bd ord
+    // double *fh = (double*)malloc(fh_size * sizeof(double));
+    if (!fh) return;
 
-        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;
+    // call symmetry_bd(2,ex,f,fh,SoA)
+    symmetry_bd(2, ex, f, fh, SoA);
 
-        const double SoA[3] = { SYM1, SYM2, SYM3 };
+    const double d12dx = ONE / F12 / dX;
+    const double d12dy = ONE / F12 / dY;
+    const double d12dz = ONE / F12 / dZ;
 
-        const size_t nx = (size_t)ex1 + ord;
-        const size_t ny = (size_t)ex2 + ord;
-        const size_t nz = (size_t)ex3 + ord;
-        const size_t fh_size = nx * ny * nz;
+    const double d2dx  = ONE / TWO / dX;
+    const double d2dy  = ONE / TWO / dY;
+    const double d2dz  = ONE / TWO / dZ;
 
-        static double *fh_buf = NULL;
-        static size_t cap = 0;
-        if (fh_size > cap) {
-            free(fh_buf);
-            fh_buf = (double*)aligned_alloc(64, fh_size * sizeof(double));
-            cap = fh_size;
-        }
-        double *fh = fh_buf;
-        if (!fh) return;
-
-        symmetry_bd(ord, 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;
-
-        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;
-        }
-
-        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)]
-                        );
-                    }
-                }
-            }
-        }
-        return;
+    // 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;
     }
-#elif (ghost_width == 4)
-    /* ---- 6th-order ----------------------------------------------------- */
-    {
-        const int ord = 3;
 
-        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;
+    /*
+     * 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 double SoA[3] = { SYM1, SYM2, SYM3 };
+    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;
 
-        const size_t nx = (size_t)ex1 + ord;
-        const size_t ny = (size_t)ex2 + ord;
-        const size_t nz = (size_t)ex3 + ord;
-        const size_t fh_size = nx * ny * nz;
+    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);
 
-        static double *fh_buf = NULL;
-        static size_t cap = 0;
-        if (fh_size > cap) {
-            free(fh_buf);
-            fh_buf = (double*)aligned_alloc(64, fh_size * sizeof(double));
-            cap = fh_size;
-        }
-        double *fh = fh_buf;
-        if (!fh) return;
+                    fx[p] = d2dx * (
+                        -fh[idx_fh_F_ord2(iF - 1, jF,     kF,     ex)] +
+                         fh[idx_fh_F_ord2(iF + 1, jF,     kF,     ex)]
+                    );
 
-        symmetry_bd(ord, ex, f, fh, SoA);
+                    fy[p] = d2dy * (
+                        -fh[idx_fh_F_ord2(iF,     jF - 1, kF,     ex)] +
+                         fh[idx_fh_F_ord2(iF,     jF + 1, kF,     ex)]
+                    );
 
-        /* Denominators */
-        const double d60dx = ONE / F60 / dX;
-        const double d60dy = ONE / F60 / dY;
-        const double d60dz = ONE / F60 / dZ;
-        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;
-
-        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;
-        }
-
-        /* 2nd-order pass: 3pt, widest */
-        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;
-
-        /* 4th-order pass: 5pt */
-        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;
-
-        /* 6th-order pass: 7pt, narrowest interior */
-        const int i6_lo = (iminF + 2 > 0) ? (iminF + 2) : 0;
-        const int j6_lo = (jminF + 2 > 0) ? (jminF + 2) : 0;
-        const int k6_lo = (kminF + 2 > 0) ? (kminF + 2) : 0;
-        const int i6_hi = ex1 - 4;
-        const int j6_hi = ex2 - 4;
-        const int k6_hi = ex3 - 4;
-
-        /* 2nd-order */
-        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(iF - 1, jF,     kF,     ex)] +
-                             fh[idx_fh_F(iF + 1, jF,     kF,     ex)]);
-                        fy[p] = d2dy * (
-                            -fh[idx_fh_F(iF,     jF - 1, kF,     ex)] +
-                             fh[idx_fh_F(iF,     jF + 1, kF,     ex)]);
-                        fz[p] = d2dz * (
-                            -fh[idx_fh_F(iF,     jF,     kF - 1, ex)] +
-                             fh[idx_fh_F(iF,     jF,     kF + 1, ex)]);
-                    }
+                    fz[p] = d2dz * (
+                        -fh[idx_fh_F_ord2(iF,     jF,     kF - 1, ex)] +
+                         fh[idx_fh_F_ord2(iF,     jF,     kF + 1, ex)]
+                    );
                 }
             }
         }
-
-        /* 4th-order overwrite */
-        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(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)]);
-
-                        fy[p] = 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)]);
-
-                        fz[p] = 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)]);
-                    }
-                }
-            }
-        }
-
-        /* 6th-order overwrite: [-1,+9,-45,0,+45,-9,+1] / (60*dx) */
-        if (i6_lo <= i6_hi && j6_lo <= j6_hi && k6_lo <= k6_hi) {
-            for (int k0 = k6_lo; k0 <= k6_hi; ++k0) {
-                const int kF = k0 + 1;
-                for (int j0 = j6_lo; j0 <= j6_hi; ++j0) {
-                    const int jF = j0 + 1;
-                    for (int i0 = i6_lo; i0 <= i6_hi; ++i0) {
-                        const int iF = i0 + 1;
-                        const size_t p = idx_ex(i0, j0, k0, ex);
-
-                        fx[p] = d60dx * (
-                            -fh[idx_fh_F(iF - 3, jF,     kF,     ex)] +
-                            F9 * fh[idx_fh_F(iF - 2, jF,     kF,     ex)] -
-                            F45 * fh[idx_fh_F(iF - 1, jF,     kF,     ex)] +
-                            F45 * fh[idx_fh_F(iF + 1, jF,     kF,     ex)] -
-                            F9 * fh[idx_fh_F(iF + 2, jF,     kF,     ex)] +
-                            fh[idx_fh_F(iF + 3, jF,     kF,     ex)]);
-
-                        fy[p] = d60dy * (
-                            -fh[idx_fh_F(iF,     jF - 3, kF,     ex)] +
-                            F9 * fh[idx_fh_F(iF,     jF - 2, kF,     ex)] -
-                            F45 * fh[idx_fh_F(iF,     jF - 1, kF,     ex)] +
-                            F45 * fh[idx_fh_F(iF,     jF + 1, kF,     ex)] -
-                            F9 * fh[idx_fh_F(iF,     jF + 2, kF,     ex)] +
-                            fh[idx_fh_F(iF,     jF + 3, kF,     ex)]);
-
-                        fz[p] = d60dz * (
-                            -fh[idx_fh_F(iF,     jF,     kF - 3, ex)] +
-                            F9 * fh[idx_fh_F(iF,     jF,     kF - 2, ex)] -
-                            F45 * fh[idx_fh_F(iF,     jF,     kF - 1, ex)] +
-                            F45 * fh[idx_fh_F(iF,     jF,     kF + 1, ex)] -
-                            F9 * fh[idx_fh_F(iF,     jF,     kF + 2, ex)] +
-                            fh[idx_fh_F(iF,     jF,     kF + 3, ex)]);
-                    }
-                }
-            }
-        }
-        return;
     }
-#elif (ghost_width == 5)
-    /* ---- 8th-order ----------------------------------------------------- */
-    {
-        const int ord = 5;
 
-        int iminF = 1, jminF = 1, kminF = 1;
-        if (Symmetry > NO_SYMM && fabs(Z[0]) < dZ) kminF = -3;
-        if (Symmetry > EQ_SYMM && fabs(X[0]) < dX) iminF = -3;
-        if (Symmetry > EQ_SYMM && fabs(Y[0]) < dY) jminF = -3;
+    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);
 
-        const double SoA[3] = { SYM1, SYM2, SYM3 };
+                    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)]
+                    );
 
-        const size_t nx = (size_t)ex1 + ord;
-        const size_t ny = (size_t)ex2 + ord;
-        const size_t nz = (size_t)ex3 + ord;
-        const size_t fh_size = nx * ny * nz;
+                    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)]
+                    );
 
-        static double *fh_buf = NULL;
-        static size_t cap = 0;
-        if (fh_size > cap) {
-            free(fh_buf);
-            fh_buf = (double*)aligned_alloc(64, fh_size * sizeof(double));
-            cap = fh_size;
-        }
-        double *fh = fh_buf;
-        if (!fh) return;
-
-        symmetry_bd(ord, ex, f, fh, SoA);
-
-        const double d840dx = ONE / F840 / dX;
-        const double d840dy = ONE / F840 / dY;
-        const double d840dz = ONE / F840 / dZ;
-        const double d60dx  = ONE / F60  / dX;
-        const double d60dy  = ONE / F60  / dY;
-        const double d60dz  = ONE / F60  / dZ;
-        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;
-
-        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;
-        }
-
-        /* 2nd: 3pt, widest */
-        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;
-
-        /* 4th: 5pt */
-        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;
-
-        /* 6th: 7pt */
-        const int i6_lo = (iminF + 2 > 0) ? (iminF + 2) : 0;
-        const int j6_lo = (jminF + 2 > 0) ? (jminF + 2) : 0;
-        const int k6_lo = (kminF + 2 > 0) ? (kminF + 2) : 0;
-        const int i6_hi = ex1 - 4;
-        const int j6_hi = ex2 - 4;
-        const int k6_hi = ex3 - 4;
-
-        /* 8th: 9pt, narrowest */
-        const int i8_lo = (iminF + 3 > 0) ? (iminF + 3) : 0;
-        const int j8_lo = (jminF + 3 > 0) ? (jminF + 3) : 0;
-        const int k8_lo = (kminF + 3 > 0) ? (kminF + 3) : 0;
-        const int i8_hi = ex1 - 5;
-        const int j8_hi = ex2 - 5;
-        const int k8_hi = ex3 - 5;
-
-        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_ord5(iF - 1, jF,     kF,     ex)] +
-                             fh[idx_fh_F_ord5(iF + 1, jF,     kF,     ex)]);
-                        fy[p] = d2dy * (
-                            -fh[idx_fh_F_ord5(iF,     jF - 1, kF,     ex)] +
-                             fh[idx_fh_F_ord5(iF,     jF + 1, kF,     ex)]);
-                        fz[p] = d2dz * (
-                            -fh[idx_fh_F_ord5(iF,     jF,     kF - 1, ex)] +
-                             fh[idx_fh_F_ord5(iF,     jF,     kF + 1, 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)]
+                    );
                 }
             }
         }
-
-        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_ord5(iF - 2, jF,     kF,     ex)] -
-                            EIT * fh[idx_fh_F_ord5(iF - 1, jF,     kF,     ex)] +
-                            EIT * fh[idx_fh_F_ord5(iF + 1, jF,     kF,     ex)] -
-                            fh[idx_fh_F_ord5(iF + 2, jF,     kF,     ex)]);
-
-                        fy[p] = d12dy * (
-                            fh[idx_fh_F_ord5(iF,     jF - 2, kF,     ex)] -
-                            EIT * fh[idx_fh_F_ord5(iF,     jF - 1, kF,     ex)] +
-                            EIT * fh[idx_fh_F_ord5(iF,     jF + 1, kF,     ex)] -
-                            fh[idx_fh_F_ord5(iF,     jF + 2, kF,     ex)]);
-
-                        fz[p] = d12dz * (
-                            fh[idx_fh_F_ord5(iF,     jF,     kF - 2, ex)] -
-                            EIT * fh[idx_fh_F_ord5(iF,     jF,     kF - 1, ex)] +
-                            EIT * fh[idx_fh_F_ord5(iF,     jF,     kF + 1, ex)] -
-                            fh[idx_fh_F_ord5(iF,     jF,     kF + 2, ex)]);
-                    }
-                }
-            }
-        }
-
-        if (i6_lo <= i6_hi && j6_lo <= j6_hi && k6_lo <= k6_hi) {
-            for (int k0 = k6_lo; k0 <= k6_hi; ++k0) {
-                const int kF = k0 + 1;
-                for (int j0 = j6_lo; j0 <= j6_hi; ++j0) {
-                    const int jF = j0 + 1;
-                    for (int i0 = i6_lo; i0 <= i6_hi; ++i0) {
-                        const int iF = i0 + 1;
-                        const size_t p = idx_ex(i0, j0, k0, ex);
-
-                        fx[p] = d60dx * (
-                            -fh[idx_fh_F_ord5(iF - 3, jF,     kF,     ex)] +
-                            F9 * fh[idx_fh_F_ord5(iF - 2, jF,     kF,     ex)] -
-                            F45 * fh[idx_fh_F_ord5(iF - 1, jF,     kF,     ex)] +
-                            F45 * fh[idx_fh_F_ord5(iF + 1, jF,     kF,     ex)] -
-                            F9 * fh[idx_fh_F_ord5(iF + 2, jF,     kF,     ex)] +
-                            fh[idx_fh_F_ord5(iF + 3, jF,     kF,     ex)]);
-
-                        fy[p] = d60dy * (
-                            -fh[idx_fh_F_ord5(iF,     jF - 3, kF,     ex)] +
-                            F9 * fh[idx_fh_F_ord5(iF,     jF - 2, kF,     ex)] -
-                            F45 * fh[idx_fh_F_ord5(iF,     jF - 1, kF,     ex)] +
-                            F45 * fh[idx_fh_F_ord5(iF,     jF + 1, kF,     ex)] -
-                            F9 * fh[idx_fh_F_ord5(iF,     jF + 2, kF,     ex)] +
-                            fh[idx_fh_F_ord5(iF,     jF + 3, kF,     ex)]);
-
-                        fz[p] = d60dz * (
-                            -fh[idx_fh_F_ord5(iF,     jF,     kF - 3, ex)] +
-                            F9 * fh[idx_fh_F_ord5(iF,     jF,     kF - 2, ex)] -
-                            F45 * fh[idx_fh_F_ord5(iF,     jF,     kF - 1, ex)] +
-                            F45 * fh[idx_fh_F_ord5(iF,     jF,     kF + 1, ex)] -
-                            F9 * fh[idx_fh_F_ord5(iF,     jF,     kF + 2, ex)] +
-                            fh[idx_fh_F_ord5(iF,     jF,     kF + 3, ex)]);
-                    }
-                }
-            }
-        }
-
-        /* 8th-order overwrite: [+3,-32,+168,-672,0,+672,-168,+32,-3] / (840*dx) */
-        if (i8_lo <= i8_hi && j8_lo <= j8_hi && k8_lo <= k8_hi) {
-            for (int k0 = k8_lo; k0 <= k8_hi; ++k0) {
-                const int kF = k0 + 1;
-                for (int j0 = j8_lo; j0 <= j8_hi; ++j0) {
-                    const int jF = j0 + 1;
-                    for (int i0 = i8_lo; i0 <= i8_hi; ++i0) {
-                        const int iF = i0 + 1;
-                        const size_t p = idx_ex(i0, j0, k0, ex);
-
-                        fx[p] = d840dx * (
-                            +(double)3 * fh[idx_fh_F_ord5(iF - 4, jF,     kF,     ex)] -
-                            F32  * fh[idx_fh_F_ord5(iF - 3, jF,     kF,     ex)] +
-                            F168 * fh[idx_fh_F_ord5(iF - 2, jF,     kF,     ex)] -
-                            F672 * fh[idx_fh_F_ord5(iF - 1, jF,     kF,     ex)] +
-                            F672 * fh[idx_fh_F_ord5(iF + 1, jF,     kF,     ex)] -
-                            F168 * fh[idx_fh_F_ord5(iF + 2, jF,     kF,     ex)] +
-                            F32  * fh[idx_fh_F_ord5(iF + 3, jF,     kF,     ex)] -
-                            (double)3 * fh[idx_fh_F_ord5(iF + 4, jF,     kF,     ex)]);
-
-                        fy[p] = d840dy * (
-                            +(double)3 * fh[idx_fh_F_ord5(iF,     jF - 4, kF,     ex)] -
-                            F32  * fh[idx_fh_F_ord5(iF,     jF - 3, kF,     ex)] +
-                            F168 * fh[idx_fh_F_ord5(iF,     jF - 2, kF,     ex)] -
-                            F672 * fh[idx_fh_F_ord5(iF,     jF - 1, kF,     ex)] +
-                            F672 * fh[idx_fh_F_ord5(iF,     jF + 1, kF,     ex)] -
-                            F168 * fh[idx_fh_F_ord5(iF,     jF + 2, kF,     ex)] +
-                            F32  * fh[idx_fh_F_ord5(iF,     jF + 3, kF,     ex)] -
-                            (double)3 * fh[idx_fh_F_ord5(iF,     jF + 4, kF,     ex)]);
-
-                        fz[p] = d840dz * (
-                            +(double)3 * fh[idx_fh_F_ord5(iF,     jF,     kF - 4, ex)] -
-                            F32  * fh[idx_fh_F_ord5(iF,     jF,     kF - 3, ex)] +
-                            F168 * fh[idx_fh_F_ord5(iF,     jF,     kF - 2, ex)] -
-                            F672 * fh[idx_fh_F_ord5(iF,     jF,     kF - 1, ex)] +
-                            F672 * fh[idx_fh_F_ord5(iF,     jF,     kF + 1, ex)] -
-                            F168 * fh[idx_fh_F_ord5(iF,     jF,     kF + 2, ex)] +
-                            F32  * fh[idx_fh_F_ord5(iF,     jF,     kF + 3, ex)] -
-                            (double)3 * fh[idx_fh_F_ord5(iF,     jF,     kF + 4, ex)]);
-                    }
-                }
-            }
-        }
-        return;
     }
-#else
-#error "fderivs_c.C: unsupported ghost_width (must be 2, 3, 4, or 5)"
-#endif
+
+    // free(fh);
 }

AMSS_NCKU_source/fderivs_sh_c.C

@@ -1,234 +0,0 @@
-#include "macrodef.h"
-#include "share_func.h"
-
-/*
- * C 版 fderivs_sh — first derivatives on shell patch in (rho, sigma, R) coords.
- *
- * Same stencil coefficients as Cartesian fderivs, but:
- * - Uses symmetry_stbd (ghost on BOTH sides of x/y, none in z)
- * - fh buffer: (-ord+1:ex+ord) in x/y, (1:ex) in z
- * - SoA is 2-element only (x/y), no z-symmetry
- * - sst parameter (shell surface type, not used in stencil computation)
- */
-extern "C" void fderivs_sh_(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, int sst)
-{
-    (void)SYM3; (void)onoff; (void)sst;
-
-    const double ZEO = 0.0, ONE = 1.0, TWO = 2.0, EIT = 8.0;
-    const double F9 = 9.0, F12 = 12.0, F45 = 45.0, F60 = 60.0;
-    const double F32 = 32.0, F168 = 168.0, F672 = 672.0, F840 = 840.0;
-
-    const int NO_SYMM = 0, EQ_SYMM = 1, OCTANT = 2;
-    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, jmaxF = ex2, kmaxF = ex3;
-    const double SoA[2] = { SYM1, SYM2 };
-
-#if (ghost_width == 2)
-    {
-        const int ord = 1;
-        int iminF = 1, jminF = 1, kminF = 1;
-        if (Symmetry == OCTANT && fabs(X[0]) < dX) iminF = 0;
-        if (Symmetry == OCTANT && fabs(Y[0]) < dY) jminF = 0;
-        if ((sst==2||sst==4) && fabs(Y[0]) < dY) jminF = 0; // EQ reflection
-
-        const size_t nx = (size_t)ex1 + 2 * ord;
-        const size_t ny = (size_t)ex2 + 2 * ord;
-        const size_t nz = (size_t)ex3;
-        const size_t fh_size = nx * ny * nz;
-        static double *fh_buf = NULL; static size_t cap = 0;
-        if (fh_size > cap) { free(fh_buf); fh_buf = (double*)aligned_alloc(64, fh_size*sizeof(double)); cap = fh_size; }
-        double *fh = fh_buf; if (!fh) return;
-        symmetry_stbd(ord, ex, f, fh, SoA);
-
-        const double d2dx = ONE/TWO/dX, d2dy = ONE/TWO/dY, d2dz = ONE/TWO/dZ;
-        const size_t all = (size_t)ex1*ex2*ex3;
-        for (size_t p=0;p<all;++p) { fx[p]=ZEO; fy[p]=ZEO; fz[p]=ZEO; }
-
-        const int i2_lo=(iminF>0)?iminF:0, j2_lo=(jminF>0)?jminF:0, k2_lo=1;
-        const int i2_hi=ex1-2, j2_hi=ex2-2, k2_hi=ex3-2;
-        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_stbd(iF-1,jF,kF,ord,ex)]+fh[idx_fh_stbd(iF+1,jF,kF,ord,ex)]);
-                fy[p]=d2dy*(-fh[idx_fh_stbd(iF,jF-1,kF,ord,ex)]+fh[idx_fh_stbd(iF,jF+1,kF,ord,ex)]);
-                fz[p]=d2dz*(-fh[idx_fh_stbd(iF,jF,kF-1,ord,ex)]+fh[idx_fh_stbd(iF,jF,kF+1,ord,ex)]);
-            }}}
-        }
-        return;
-    }
-#elif (ghost_width == 3)
-    {
-        const int ord = 2;
-        int iminF = 1, jminF = 1, kminF = 1;
-        if (Symmetry == OCTANT && fabs(X[0]) < dX) iminF = -1;
-        if (Symmetry == OCTANT && fabs(Y[0]) < dY) jminF = -1;
-        if ((sst==2||sst==4) && fabs(Y[0]) < dY) jminF = -1;
-
-        const size_t nx=(size_t)ex1+2*ord, ny=(size_t)ex2+2*ord, nz=(size_t)ex3;
-        const size_t fh_size=nx*ny*nz;
-        static double *fh_buf=NULL; static size_t cap=0;
-        if (fh_size>cap){free(fh_buf);fh_buf=(double*)aligned_alloc(64,fh_size*sizeof(double));cap=fh_size;}
-        double *fh=fh_buf; if(!fh)return;
-        symmetry_stbd(ord,ex,f,fh,SoA);
-
-        const double d12dx=ONE/F12/dX, d12dy=ONE/F12/dY, d12dz=ONE/F12/dZ;
-        const double d2dx=ONE/TWO/dX, d2dy=ONE/TWO/dY, d2dz=ONE/TWO/dZ;
-        const size_t all=(size_t)ex1*ex2*ex3;
-        for(size_t p=0;p<all;++p){fx[p]=ZEO;fy[p]=ZEO;fz[p]=ZEO;}
-
-        const int i2_lo=(iminF>0)?iminF:0, j2_lo=(jminF>0)?jminF:0, k2_lo=1;
-        const int i2_hi=ex1-2, j2_hi=ex2-2, k2_hi=ex3-2;
-        const int i4_lo=(iminF+1>0)?iminF+1:0, j4_lo=(jminF+1>0)?jminF+1:0, k4_lo=2;
-        const int i4_hi=ex1-3, j4_hi=ex2-3, 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_stbd(iF-1,jF,kF,ord,ex)]+fh[idx_fh_stbd(iF+1,jF,kF,ord,ex)]);
-                fy[p]=d2dy*(-fh[idx_fh_stbd(iF,jF-1,kF,ord,ex)]+fh[idx_fh_stbd(iF,jF+1,kF,ord,ex)]);
-                fz[p]=d2dz*(-fh[idx_fh_stbd(iF,jF,kF-1,ord,ex)]+fh[idx_fh_stbd(iF,jF,kF+1,ord,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_stbd(iF-2,jF,kF,ord,ex)]-EIT*fh[idx_fh_stbd(iF-1,jF,kF,ord,ex)]+EIT*fh[idx_fh_stbd(iF+1,jF,kF,ord,ex)]-fh[idx_fh_stbd(iF+2,jF,kF,ord,ex)]);
-                fy[p]=d12dy*(fh[idx_fh_stbd(iF,jF-2,kF,ord,ex)]-EIT*fh[idx_fh_stbd(iF,jF-1,kF,ord,ex)]+EIT*fh[idx_fh_stbd(iF,jF+1,kF,ord,ex)]-fh[idx_fh_stbd(iF,jF+2,kF,ord,ex)]);
-                fz[p]=d12dz*(fh[idx_fh_stbd(iF,jF,kF-2,ord,ex)]-EIT*fh[idx_fh_stbd(iF,jF,kF-1,ord,ex)]+EIT*fh[idx_fh_stbd(iF,jF,kF+1,ord,ex)]-fh[idx_fh_stbd(iF,jF,kF+2,ord,ex)]);
-            }}}
-        }
-        return;
-    }
-#elif (ghost_width == 4)
-    {
-        const int ord = 3;
-        int iminF=1,jminF=1,kminF=1;
-        if(Symmetry==OCTANT&&fabs(X[0])<dX)iminF=-2;
-        if(Symmetry==OCTANT&&fabs(Y[0])<dY)jminF=-2;
-        if((sst==2||sst==4)&&fabs(Y[0])<dY)jminF=-2;
-
-        const size_t nx=(size_t)ex1+2*ord,ny=(size_t)ex2+2*ord,nz=(size_t)ex3;
-        const size_t fh_size=nx*ny*nz;
-        static double *fh_buf=NULL;static size_t cap=0;
-        if(fh_size>cap){free(fh_buf);fh_buf=(double*)aligned_alloc(64,fh_size*sizeof(double));cap=fh_size;}
-        double *fh=fh_buf;if(!fh)return;
-        symmetry_stbd(ord,ex,f,fh,SoA);
-
-        const double d60dx=ONE/F60/dX,d60dy=ONE/F60/dY,d60dz=ONE/F60/dZ;
-        const double d12dx=ONE/F12/dX,d12dy=ONE/F12/dY,d12dz=ONE/F12/dZ;
-        const double d2dx=ONE/TWO/dX,d2dy=ONE/TWO/dY,d2dz=ONE/TWO/dZ;
-        const size_t all=(size_t)ex1*ex2*ex3;
-        for(size_t p=0;p<all;++p){fx[p]=ZEO;fy[p]=ZEO;fz[p]=ZEO;}
-
-        const int i2_lo=(iminF>0)?iminF:0,j2_lo=(jminF>0)?jminF:0,k2_lo=1,i2_hi=ex1-2,j2_hi=ex2-2,k2_hi=ex3-2;
-        const int i4_lo=(iminF+1>0)?iminF+1:0,j4_lo=(jminF+1>0)?jminF+1:0,k4_lo=2,i4_hi=ex1-3,j4_hi=ex2-3,k4_hi=ex3-3;
-        const int i6_lo=(iminF+2>0)?iminF+2:0,j6_lo=(jminF+2>0)?jminF+2:0,k6_lo=3,i6_hi=ex1-4,j6_hi=ex2-4,k6_hi=ex3-4;
-
-        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_stbd(iF-1,jF,kF,ord,ex)]+fh[idx_fh_stbd(iF+1,jF,kF,ord,ex)]);
-                fy[p]=d2dy*(-fh[idx_fh_stbd(iF,jF-1,kF,ord,ex)]+fh[idx_fh_stbd(iF,jF+1,kF,ord,ex)]);
-                fz[p]=d2dz*(-fh[idx_fh_stbd(iF,jF,kF-1,ord,ex)]+fh[idx_fh_stbd(iF,jF,kF+1,ord,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_stbd(iF-2,jF,kF,ord,ex)]-EIT*fh[idx_fh_stbd(iF-1,jF,kF,ord,ex)]+EIT*fh[idx_fh_stbd(iF+1,jF,kF,ord,ex)]-fh[idx_fh_stbd(iF+2,jF,kF,ord,ex)]);
-                fy[p]=d12dy*(fh[idx_fh_stbd(iF,jF-2,kF,ord,ex)]-EIT*fh[idx_fh_stbd(iF,jF-1,kF,ord,ex)]+EIT*fh[idx_fh_stbd(iF,jF+1,kF,ord,ex)]-fh[idx_fh_stbd(iF,jF+2,kF,ord,ex)]);
-                fz[p]=d12dz*(fh[idx_fh_stbd(iF,jF,kF-2,ord,ex)]-EIT*fh[idx_fh_stbd(iF,jF,kF-1,ord,ex)]+EIT*fh[idx_fh_stbd(iF,jF,kF+1,ord,ex)]-fh[idx_fh_stbd(iF,jF,kF+2,ord,ex)]);
-            }}}
-        }
-        if(i6_lo<=i6_hi&&j6_lo<=j6_hi&&k6_lo<=k6_hi){
-            for(int k0=k6_lo;k0<=k6_hi;++k0){const int kF=k0+1;
-            for(int j0=j6_lo;j0<=j6_hi;++j0){const int jF=j0+1;
-            for(int i0=i6_lo;i0<=i6_hi;++i0){const int iF=i0+1;
-                const size_t p=idx_ex(i0,j0,k0,ex);
-                fx[p]=d60dx*(-fh[idx_fh_stbd(iF-3,jF,kF,ord,ex)]+F9*fh[idx_fh_stbd(iF-2,jF,kF,ord,ex)]-F45*fh[idx_fh_stbd(iF-1,jF,kF,ord,ex)]+F45*fh[idx_fh_stbd(iF+1,jF,kF,ord,ex)]-F9*fh[idx_fh_stbd(iF+2,jF,kF,ord,ex)]+fh[idx_fh_stbd(iF+3,jF,kF,ord,ex)]);
-                fy[p]=d60dy*(-fh[idx_fh_stbd(iF,jF-3,kF,ord,ex)]+F9*fh[idx_fh_stbd(iF,jF-2,kF,ord,ex)]-F45*fh[idx_fh_stbd(iF,jF-1,kF,ord,ex)]+F45*fh[idx_fh_stbd(iF,jF+1,kF,ord,ex)]-F9*fh[idx_fh_stbd(iF,jF+2,kF,ord,ex)]+fh[idx_fh_stbd(iF,jF+3,kF,ord,ex)]);
-                fz[p]=d60dz*(-fh[idx_fh_stbd(iF,jF,kF-3,ord,ex)]+F9*fh[idx_fh_stbd(iF,jF,kF-2,ord,ex)]-F45*fh[idx_fh_stbd(iF,jF,kF-1,ord,ex)]+F45*fh[idx_fh_stbd(iF,jF,kF+1,ord,ex)]-F9*fh[idx_fh_stbd(iF,jF,kF+2,ord,ex)]+fh[idx_fh_stbd(iF,jF,kF+3,ord,ex)]);
-            }}}
-        }
-        return;
-    }
-#elif (ghost_width == 5)
-    {
-        const int ord = 4;
-        int iminF=1,jminF=1,kminF=1;
-        if(Symmetry==OCTANT&&fabs(X[0])<dX)iminF=-3;
-        if(Symmetry==OCTANT&&fabs(Y[0])<dY)jminF=-3;
-        if((sst==2||sst==4)&&fabs(Y[0])<dY)jminF=-3;
-
-        const size_t nx=(size_t)ex1+2*ord,ny=(size_t)ex2+2*ord,nz=(size_t)ex3;
-        const size_t fh_size=nx*ny*nz;
-        static double *fh_buf=NULL;static size_t cap=0;
-        if(fh_size>cap){free(fh_buf);fh_buf=(double*)aligned_alloc(64,fh_size*sizeof(double));cap=fh_size;}
-        double *fh=fh_buf;if(!fh)return;
-        symmetry_stbd(ord,ex,f,fh,SoA);
-
-        const double d840dx=ONE/F840/dX,d840dy=ONE/F840/dY,d840dz=ONE/F840/dZ;
-        const double d60dx=ONE/F60/dX,d60dy=ONE/F60/dY,d60dz=ONE/F60/dZ;
-        const double d12dx=ONE/F12/dX,d12dy=ONE/F12/dY,d12dz=ONE/F12/dZ;
-        const double d2dx=ONE/TWO/dX,d2dy=ONE/TWO/dY,d2dz=ONE/TWO/dZ;
-        const size_t all=(size_t)ex1*ex2*ex3;
-        for(size_t p=0;p<all;++p){fx[p]=ZEO;fy[p]=ZEO;fz[p]=ZEO;}
-
-        const int i2_lo=(iminF>0)?iminF:0,j2_lo=(jminF>0)?jminF:0,k2_lo=1,i2_hi=ex1-2,j2_hi=ex2-2,k2_hi=ex3-2;
-        const int i4_lo=(iminF+1>0)?iminF+1:0,j4_lo=(jminF+1>0)?jminF+1:0,k4_lo=2,i4_hi=ex1-3,j4_hi=ex2-3,k4_hi=ex3-3;
-        const int i6_lo=(iminF+2>0)?iminF+2:0,j6_lo=(jminF+2>0)?jminF+2:0,k6_lo=3,i6_hi=ex1-4,j6_hi=ex2-4,k6_hi=ex3-4;
-        const int i8_lo=(iminF+3>0)?iminF+3:0,j8_lo=(jminF+3>0)?jminF+3:0,k8_lo=4,i8_hi=ex1-5,j8_hi=ex2-5,k8_hi=ex3-5;
-
-        #define FH_S(iF,jF,kF) fh[idx_fh_stbd(iF,jF,kF,ord,ex)]
-        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_S(iF-1,jF,kF)+FH_S(iF+1,jF,kF));
-            fy[p]=d2dy*(-FH_S(iF,jF-1,kF)+FH_S(iF,jF+1,kF));
-            fz[p]=d2dz*(-FH_S(iF,jF,kF-1)+FH_S(iF,jF,kF+1));}}}}
-
-        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_S(iF-2,jF,kF)-EIT*FH_S(iF-1,jF,kF)+EIT*FH_S(iF+1,jF,kF)-FH_S(iF+2,jF,kF));
-            fy[p]=d12dy*(FH_S(iF,jF-2,kF)-EIT*FH_S(iF,jF-1,kF)+EIT*FH_S(iF,jF+1,kF)-FH_S(iF,jF+2,kF));
-            fz[p]=d12dz*(FH_S(iF,jF,kF-2)-EIT*FH_S(iF,jF,kF-1)+EIT*FH_S(iF,jF,kF+1)-FH_S(iF,jF,kF+2));}}}}
-
-        if(i6_lo<=i6_hi&&j6_lo<=j6_hi&&k6_lo<=k6_hi){for(int k0=k6_lo;k0<=k6_hi;++k0){const int kF=k0+1;
-        for(int j0=j6_lo;j0<=j6_hi;++j0){const int jF=j0+1;
-        for(int i0=i6_lo;i0<=i6_hi;++i0){const int iF=i0+1;const size_t p=idx_ex(i0,j0,k0,ex);
-            fx[p]=d60dx*(-FH_S(iF-3,jF,kF)+F9*FH_S(iF-2,jF,kF)-F45*FH_S(iF-1,jF,kF)+F45*FH_S(iF+1,jF,kF)-F9*FH_S(iF+2,jF,kF)+FH_S(iF+3,jF,kF));
-            fy[p]=d60dy*(-FH_S(iF,jF-3,kF)+F9*FH_S(iF,jF-2,kF)-F45*FH_S(iF,jF-1,kF)+F45*FH_S(iF,jF+1,kF)-F9*FH_S(iF,jF+2,kF)+FH_S(iF,jF+3,kF));
-            fz[p]=d60dz*(-FH_S(iF,jF,kF-3)+F9*FH_S(iF,jF,kF-2)-F45*FH_S(iF,jF,kF-1)+F45*FH_S(iF,jF,kF+1)-F9*FH_S(iF,jF,kF+2)+FH_S(iF,jF,kF+3));}}}}
-
-        if(i8_lo<=i8_hi&&j8_lo<=j8_hi&&k8_lo<=k8_hi){for(int k0=k8_lo;k0<=k8_hi;++k0){const int kF=k0+1;
-        for(int j0=j8_lo;j0<=j8_hi;++j0){const int jF=j0+1;
-        for(int i0=i8_lo;i0<=i8_hi;++i0){const int iF=i0+1;const size_t p=idx_ex(i0,j0,k0,ex);
-            fx[p]=d840dx*(+(double)3*FH_S(iF-4,jF,kF)-F32*FH_S(iF-3,jF,kF)+F168*FH_S(iF-2,jF,kF)-F672*FH_S(iF-1,jF,kF)+F672*FH_S(iF+1,jF,kF)-F168*FH_S(iF+2,jF,kF)+F32*FH_S(iF+3,jF,kF)-(double)3*FH_S(iF+4,jF,kF));
-            fy[p]=d840dy*(+(double)3*FH_S(iF,jF-4,kF)-F32*FH_S(iF,jF-3,kF)+F168*FH_S(iF,jF-2,kF)-F672*FH_S(iF,jF-1,kF)+F672*FH_S(iF,jF+1,kF)-F168*FH_S(iF,jF+2,kF)+F32*FH_S(iF,jF+3,kF)-(double)3*FH_S(iF,jF+4,kF));
-            fz[p]=d840dz*(+(double)3*FH_S(iF,jF,kF-4)-F32*FH_S(iF,jF,kF-3)+F168*FH_S(iF,jF,kF-2)-F672*FH_S(iF,jF,kF-1)+F672*FH_S(iF,jF,kF+1)-F168*FH_S(iF,jF,kF+2)+F32*FH_S(iF,jF,kF+3)-(double)3*FH_S(iF,jF,kF+4));}}}}
-        #undef FH_S
-        return;
-    }
-#else
-#error "fderivs_sh_c.C: unsupported ghost_width"
-#endif
-}

AMSS_NCKU_source/fderivs_shc_c.C

@@ -1,54 +0,0 @@
-#include "macrodef.h"
-#include "share_func.h"
-#include <cstddef>
-
-/*
- * fderivs_shc — shell first derivatives converted to Cartesian via chain rule.
- *
- * Calls fderivs_sh internally, then:
- *   fx = drhodx * df/drho + dsigmadx * df/dsigma + dRdx * df/dR
- *   fy = drhody * df/drho + dsigmady * df/dsigma + dRdy * df/dR
- *   fz = drhodz * df/drho + dsigmadz * df/dsigma + dRdz * df/dR
- */
-
-// Forward declaration (defined in fderivs_sh_c.C with extern "C" name fderivs_sh_)
-extern "C" {
-void fderivs_sh_(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, int sst);
-
-void fderivs_shc_(int *ex,
-                   double *f,
-                   double *fx, double *fy, double *fz,
-                   double *crho, double *sigma, double *R,
-                   double &SYM1, double &SYM2, double &SYM3,
-                   int &Symmetry, int &Lev, int &sst,
-                   double *drhodx, double *drhody, double *drhodz,
-                   double *dsigmadx, double *dsigmady, double *dsigmadz,
-                   double *dRdx, double *dRdy, double *dRdz)
-{
-    const int ex3[3] = { ex[0], ex[1], ex[2] };
-    const size_t n = (size_t)ex[0] * (size_t)ex[1] * (size_t)ex[2];
-
-    // Temporary shell-coordinate derivatives
-    double *gx = (double*)malloc(n * sizeof(double));
-    double *gy = (double*)malloc(n * sizeof(double));
-    double *gz = (double*)malloc(n * sizeof(double));
-    if (!gx || !gy || !gz) { free(gx); free(gy); free(gz); return; }
-
-    // Compute shell-coordinate derivatives
-    fderivs_sh_(ex3, f, gx, gy, gz, crho, sigma, R, SYM1, SYM2, SYM3, Symmetry, Lev, sst);
-
-    // Chain rule to Cartesian
-    for (size_t i = 0; i < n; ++i) {
-        fx[i] = drhodx[i]   * gx[i] + dsigmadx[i] * gy[i] + dRdx[i] * gz[i];
-        fy[i] = drhody[i]   * gx[i] + dsigmady[i] * gy[i] + dRdy[i] * gz[i];
-        fz[i] = drhodz[i]   * gx[i] + dsigmadz[i] * gy[i] + dRdz[i] * gz[i];
-    }
-
-    free(gx); free(gy); free(gz);
-}
-
-} // extern "C"

AMSS_NCKU_source/gpu_rhsSS_mem.h

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

AMSS_NCKU_source/kodiss_c.C

@@ -1,16 +1,16 @@
-#include "macrodef.h"
 #include "tool.h"
 
 /*
- * C 版 kodis — Kreiss-Oliger numerical dissipation (Cartesian patches).
+ * C 版 kodis
  *
- * The KO operator is (D₊D₋)^r applied to f_rhs with alternating sign (-1)^(r-1).
+ * Fortran signature:
+ * subroutine kodis(ex,X,Y,Z,f,f_rhs,SoA,Symmetry,eps)
  *
- * FD order → r → cof=2^(2r) mapping:
- *   ghost_width=2 (2nd) → r=2, cof=16,  sign=-
- *   ghost_width=3 (4th) → r=3, cof=64,  sign=+
- *   ghost_width=4 (6th) → r=4, cof=256, sign=-
- *   ghost_width=5 (8th) → r=5, cof=1024,sign=+
+ * 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,
@@ -18,304 +18,100 @@ void kodis(const int ex[3],
            const double SoA[3],
            int Symmetry, double eps)
 {
-    const double ZEO = 0.0;
+    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
 
-    const int imaxF = ex1, jmaxF = ex2, kmaxF = ex3;
+    // Fortran: imax=ex(1) 等是 1-based 上界
+    const int imaxF = ex1;
+    const int jmaxF = ex2;
+    const int kmaxF = ex3;
 
-#if (ghost_width == 2)
-    /* ---- r=2, cof=16, sign=-, 5pt stencil ----------------------------- */
-    {
-        const int ord = 2;
-        const int r = 2;
-        const double cof = 16.0;
-        const double F4 = 4.0, F6 = 6.0;
-        const int NO_SYMM = 0, EQ_SYMM = 1;
+    // Fortran: imin=jmin=kmin=1，某些对称情况变 -2
+    int iminF = 1, jminF = 1, kminF = 1;
 
-        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;
+    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;
 
-        const size_t nx = (size_t)ex1 + ord;
-        const size_t ny = (size_t)ex2 + ord;
-        const size_t nz = (size_t)ex3 + ord;
-        const size_t fh_size = nx * ny * nz;
+    // 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;
+    double *fh = (double*)malloc(fh_size * sizeof(double));
+    if (!fh) return;
 
-        symmetry_bd(ord, ex, f, fh, SoA);
+    // Fortran: call symmetry_bd(3,ex,f,fh,SoA)
+    symmetry_bd(3, ex, f, fh, SoA);
 
-        /* i±2 must be valid: i-2 >= iminF && i+2 <= imaxF
-           C 0-based: i0 >= iminF+1, i0 <= ex1-3 */
-        const int i0_lo = (iminF + 1 > 0) ? (iminF + 1) : 0;
-        const int j0_lo = (jminF + 1 > 0) ? (jminF + 1) : 0;
-        const int k0_lo = (kminF + 1 > 0) ? (kminF + 1) : 0;
-        const int i0_hi = imaxF - 3;
-        const int j0_hi = jmaxF - 3;
-        const int k0_hi = kmaxF - 3;
+    /*
+     * 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)) {
-            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 = (
-                            (fh[idx_fh_F_ord2(iF - 2, jF, kF, ex)] + fh[idx_fh_F_ord2(iF + 2, jF, kF, ex)]) -
-                            F4 * (fh[idx_fh_F_ord2(iF - 1, jF, kF, ex)] + fh[idx_fh_F_ord2(iF + 1, jF, kF, ex)]) +
-                            F6 *  fh[idx_fh_F_ord2(iF,     jF, kF, ex)]
-                        ) / dX;
-
-                        const double Dy = (
-                            (fh[idx_fh_F_ord2(iF, jF - 2, kF, ex)] + fh[idx_fh_F_ord2(iF, jF + 2, kF, ex)]) -
-                            F4 * (fh[idx_fh_F_ord2(iF, jF - 1, kF, ex)] + fh[idx_fh_F_ord2(iF, jF + 1, kF, ex)]) +
-                            F6 *  fh[idx_fh_F_ord2(iF, jF,     kF, ex)]
-                        ) / dY;
-
-                        const double Dz = (
-                            (fh[idx_fh_F_ord2(iF, jF, kF - 2, ex)] + fh[idx_fh_F_ord2(iF, jF, kF + 2, ex)]) -
-                            F4 * (fh[idx_fh_F_ord2(iF, jF, kF - 1, ex)] + fh[idx_fh_F_ord2(iF, jF, kF + 1, ex)]) +
-                            F6 *  fh[idx_fh_F_ord2(iF, jF, kF,     ex)]
-                        ) / dZ;
-
-                        f_rhs[p] -= (eps / cof) * (Dx + Dy + Dz); /* sign=- */
-                    }
-                }
-            }
-        }
+    if (i0_lo > i0_hi || j0_lo > j0_hi || k0_lo > k0_hi) {
         free(fh);
         return;
     }
-#elif (ghost_width == 3)
-    /* ---- r=3, cof=64, sign=+, 7pt stencil (current default) ---------- */
-    {
-        const int ord = 3;
-        const int r = 3;
-        const double cof = 64.0;
-        const double SIX = 6.0, FIT = 15.0, TWT = 20.0;
-        const int NO_SYMM = 0, OCTANT = 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;
+    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 nx = (size_t)ex1 + ord;
-        const size_t ny = (size_t)ex2 + ord;
-        const size_t nz = (size_t)ex3 + ord;
-        const size_t fh_size = nx * ny * nz;
+                    const size_t p = idx_ex(i0, j0, k0, ex);
 
-        double *fh = (double*)malloc(fh_size * sizeof(double));
-        if (!fh) return;
+                    // 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;
 
-        symmetry_bd(ord, ex, f, fh, SoA);
+                    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 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;
-        const int j0_hi = jmaxF - 4;
-        const int k0_hi = kmaxF - 4;
+                    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;
 
-        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 = (
-                            (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 = (
-                            (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 = (
-                            (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 + Dy + Dz); /* sign=+ */
-                    }
-                }
+                    // 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);
-        return;
     }
-#elif (ghost_width == 4)
-    /* ---- r=4, cof=256, sign=-, 9pt stencil ---------------------------- */
-    {
-        const int ord = 4;
-        const int r = 4;
-        const double cof = 256.0;
-        const double F8 = 8.0, F28 = 28.0, F56 = 56.0, F70 = 70.0;
-        const int NO_SYMM = 0, EQ_SYMM = 1;
 
-        int iminF = 1, jminF = 1, kminF = 1;
-        if (Symmetry > NO_SYMM && fabs(Z[0]) < dZ) kminF = -3;
-        if (Symmetry > EQ_SYMM && fabs(X[0]) < dX) iminF = -3;
-        if (Symmetry > EQ_SYMM && fabs(Y[0]) < dY) jminF = -3;
-
-        const size_t nx = (size_t)ex1 + ord;
-        const size_t ny = (size_t)ex2 + ord;
-        const size_t nz = (size_t)ex3 + ord;
-        const size_t fh_size = nx * ny * nz;
-
-        double *fh = (double*)malloc(fh_size * sizeof(double));
-        if (!fh) return;
-
-        symmetry_bd(ord, ex, f, fh, SoA);
-
-        /* i±4 valid: i-4>=iminF → i0>=iminF+3, i+4<=imaxF → i0<=ex1-5 */
-        const int i0_lo = (iminF + 3 > 0) ? iminF + 3 : 0;
-        const int j0_lo = (jminF + 3 > 0) ? jminF + 3 : 0;
-        const int k0_lo = (kminF + 3 > 0) ? kminF + 3 : 0;
-        const int i0_hi = imaxF - 5;
-        const int j0_hi = jmaxF - 5;
-        const int k0_hi = kmaxF - 5;
-
-        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);
-
-                        /* Stencil: [1,-8,28,-56,70,-56,28,-8,1] */
-                        const double Dx = (
-                            (fh[idx_fh_F_ord4(iF - 4, jF, kF, ex)] + fh[idx_fh_F_ord4(iF + 4, jF, kF, ex)]) -
-                            F8 * (fh[idx_fh_F_ord4(iF - 3, jF, kF, ex)] + fh[idx_fh_F_ord4(iF + 3, jF, kF, ex)]) +
-                            F28* (fh[idx_fh_F_ord4(iF - 2, jF, kF, ex)] + fh[idx_fh_F_ord4(iF + 2, jF, kF, ex)]) -
-                            F56* (fh[idx_fh_F_ord4(iF - 1, jF, kF, ex)] + fh[idx_fh_F_ord4(iF + 1, jF, kF, ex)]) +
-                            F70*  fh[idx_fh_F_ord4(iF,     jF, kF, ex)]
-                        ) / dX;
-
-                        const double Dy = (
-                            (fh[idx_fh_F_ord4(iF, jF - 4, kF, ex)] + fh[idx_fh_F_ord4(iF, jF + 4, kF, ex)]) -
-                            F8 * (fh[idx_fh_F_ord4(iF, jF - 3, kF, ex)] + fh[idx_fh_F_ord4(iF, jF + 3, kF, ex)]) +
-                            F28* (fh[idx_fh_F_ord4(iF, jF - 2, kF, ex)] + fh[idx_fh_F_ord4(iF, jF + 2, kF, ex)]) -
-                            F56* (fh[idx_fh_F_ord4(iF, jF - 1, kF, ex)] + fh[idx_fh_F_ord4(iF, jF + 1, kF, ex)]) +
-                            F70*  fh[idx_fh_F_ord4(iF, jF,     kF, ex)]
-                        ) / dY;
-
-                        const double Dz = (
-                            (fh[idx_fh_F_ord4(iF, jF, kF - 4, ex)] + fh[idx_fh_F_ord4(iF, jF, kF + 4, ex)]) -
-                            F8 * (fh[idx_fh_F_ord4(iF, jF, kF - 3, ex)] + fh[idx_fh_F_ord4(iF, jF, kF + 3, ex)]) +
-                            F28* (fh[idx_fh_F_ord4(iF, jF, kF - 2, ex)] + fh[idx_fh_F_ord4(iF, jF, kF + 2, ex)]) -
-                            F56* (fh[idx_fh_F_ord4(iF, jF, kF - 1, ex)] + fh[idx_fh_F_ord4(iF, jF, kF + 1, ex)]) +
-                            F70*  fh[idx_fh_F_ord4(iF, jF, kF,     ex)]
-                        ) / dZ;
-
-                        f_rhs[p] -= (eps / cof) * (Dx + Dy + Dz); /* sign=- */
-                    }
-                }
-            }
-        }
-        free(fh);
-        return;
-    }
-#elif (ghost_width == 5)
-    /* ---- r=5, cof=1024, sign=+, 11pt stencil ------------------------- */
-    {
-        const int ord = 5;
-        const int r = 5;
-        const double cof = 1024.0;
-        const double F10 = 10.0, F45 = 45.0, F120 = 120.0;
-        const double F210 = 210.0, F252 = 252.0;
-        const int NO_SYMM = 0, EQ_SYMM = 1;
-
-        int iminF = 1, jminF = 1, kminF = 1;
-        if (Symmetry > NO_SYMM && fabs(Z[0]) < dZ) kminF = -4;
-        if (Symmetry > EQ_SYMM && fabs(X[0]) < dX) iminF = -4;
-        if (Symmetry > EQ_SYMM && fabs(Y[0]) < dY) jminF = -4;
-
-        const size_t nx = (size_t)ex1 + ord;
-        const size_t ny = (size_t)ex2 + ord;
-        const size_t nz = (size_t)ex3 + ord;
-        const size_t fh_size = nx * ny * nz;
-
-        double *fh = (double*)malloc(fh_size * sizeof(double));
-        if (!fh) return;
-
-        symmetry_bd(ord, ex, f, fh, SoA);
-
-        /* i±5 valid: i0>=iminF+4, i0<=ex1-6 */
-        const int i0_lo = (iminF + 4 > 0) ? iminF + 4 : 0;
-        const int j0_lo = (jminF + 4 > 0) ? jminF + 4 : 0;
-        const int k0_lo = (kminF + 4 > 0) ? kminF + 4 : 0;
-        const int i0_hi = imaxF - 6;
-        const int j0_hi = jmaxF - 6;
-        const int k0_hi = kmaxF - 6;
-
-        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);
-
-                        /* Stencil: [1,-10,45,-120,210,-252,210,-120,45,-10,1] */
-                        const double Dx = (
-                            (fh[idx_fh_F_ord5(iF - 5, jF, kF, ex)] + fh[idx_fh_F_ord5(iF + 5, jF, kF, ex)]) -
-                            F10 * (fh[idx_fh_F_ord5(iF - 4, jF, kF, ex)] + fh[idx_fh_F_ord5(iF + 4, jF, kF, ex)]) +
-                            F45 * (fh[idx_fh_F_ord5(iF - 3, jF, kF, ex)] + fh[idx_fh_F_ord5(iF + 3, jF, kF, ex)]) -
-                            F120* (fh[idx_fh_F_ord5(iF - 2, jF, kF, ex)] + fh[idx_fh_F_ord5(iF + 2, jF, kF, ex)]) +
-                            F210* (fh[idx_fh_F_ord5(iF - 1, jF, kF, ex)] + fh[idx_fh_F_ord5(iF + 1, jF, kF, ex)]) -
-                            F252*  fh[idx_fh_F_ord5(iF,     jF, kF, ex)]
-                        ) / dX;
-
-                        const double Dy = (
-                            (fh[idx_fh_F_ord5(iF, jF - 5, kF, ex)] + fh[idx_fh_F_ord5(iF, jF + 5, kF, ex)]) -
-                            F10 * (fh[idx_fh_F_ord5(iF, jF - 4, kF, ex)] + fh[idx_fh_F_ord5(iF, jF + 4, kF, ex)]) +
-                            F45 * (fh[idx_fh_F_ord5(iF, jF - 3, kF, ex)] + fh[idx_fh_F_ord5(iF, jF + 3, kF, ex)]) -
-                            F120* (fh[idx_fh_F_ord5(iF, jF - 2, kF, ex)] + fh[idx_fh_F_ord5(iF, jF + 2, kF, ex)]) +
-                            F210* (fh[idx_fh_F_ord5(iF, jF - 1, kF, ex)] + fh[idx_fh_F_ord5(iF, jF + 1, kF, ex)]) -
-                            F252*  fh[idx_fh_F_ord5(iF, jF,     kF, ex)]
-                        ) / dY;
-
-                        const double Dz = (
-                            (fh[idx_fh_F_ord5(iF, jF, kF - 5, ex)] + fh[idx_fh_F_ord5(iF, jF, kF + 5, ex)]) -
-                            F10 * (fh[idx_fh_F_ord5(iF, jF, kF - 4, ex)] + fh[idx_fh_F_ord5(iF, jF, kF + 4, ex)]) +
-                            F45 * (fh[idx_fh_F_ord5(iF, jF, kF - 3, ex)] + fh[idx_fh_F_ord5(iF, jF, kF + 3, ex)]) -
-                            F120* (fh[idx_fh_F_ord5(iF, jF, kF - 2, ex)] + fh[idx_fh_F_ord5(iF, jF, kF + 2, ex)]) +
-                            F210* (fh[idx_fh_F_ord5(iF, jF, kF - 1, ex)] + fh[idx_fh_F_ord5(iF, jF, kF + 1, ex)]) -
-                            F252*  fh[idx_fh_F_ord5(iF, jF, kF,     ex)]
-                        ) / dZ;
-
-                        f_rhs[p] += (eps / cof) * (Dx + Dy + Dz); /* sign=+ */
-                    }
-                }
-            }
-        }
-        free(fh);
-        return;
-    }
-#else
-#error "kodiss_c.C: unsupported ghost_width (must be 2, 3, 4, or 5)"
-#endif
-}
+    free(fh);
+}

AMSS_NCKU_source/kodiss_sh_c.C

@@ -1,136 +0,0 @@
-#include "macrodef.h"
-#include "share_func.h"
-
-/*
- * kodis_sh — Kreiss-Oliger dissipation on shell patches.
- * Same stencil coefficients as Cartesian kodis. Uses symmetry_stbd.
- */
-extern "C" void kodis_sh_(const int ex[3],
-              const double *X, const double *Y, const double *Z,
-              const double *f, double *f_rhs,
-              const double SoAi[2],
-              int Symmetry, double eps, int sst)
-{
-    (void)sst;
-    const double ZEO=0.0;
-    const int ex1=ex[0], ex2=ex[1], ex3=ex[2];
-    const double dX=X[1]-X[0], dY=Y[1]-Y[0], dZ=Z[1]-Z[0];
-    const int imaxF=ex1, jmaxF=ex2, kmaxF=ex3;
-    const double SoA[2]={SoAi[0],SoAi[1]};
-
-#if (ghost_width == 2)
-    {
-        const int ord=2, r=2;
-        const double cof=16.0, F4=4.0, F6=6.0;
-        const int NO_SYMM=0, OCTANT=2;
-        int iminF=1,jminF=1,kminF=1;
-        if(Symmetry==OCTANT&&fabs(X[0])<dX)iminF=-1;
-        if(Symmetry==OCTANT&&fabs(Y[0])<dY)jminF=-1;
-        if((sst==2||sst==4)&&fabs(Y[0])<dY)jminF=-1;
-
-        const size_t nx=(size_t)ex1+2*ord,ny=(size_t)ex2+2*ord,nz=(size_t)ex3,fh_size=nx*ny*nz;
-        double *fh=(double*)malloc(fh_size*sizeof(double));if(!fh)return;
-        symmetry_stbd(ord,ex,f,fh,SoA);
-
-        const int i0_lo=(iminF+1>0)?iminF+1:0,j0_lo=(jminF+1>0)?jminF+1:0,k0_lo=2;
-        const int i0_hi=imaxF-3,j0_hi=jmaxF-3,k0_hi=kmaxF-3;
-        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=((fh[idx_fh_stbd(iF-2,jF,kF,ord,ex)]+fh[idx_fh_stbd(iF+2,jF,kF,ord,ex)])-F4*(fh[idx_fh_stbd(iF-1,jF,kF,ord,ex)]+fh[idx_fh_stbd(iF+1,jF,kF,ord,ex)])+F6*fh[idx_fh_stbd(iF,jF,kF,ord,ex)])/dX;
-                const double Dy=((fh[idx_fh_stbd(iF,jF-2,kF,ord,ex)]+fh[idx_fh_stbd(iF,jF+2,kF,ord,ex)])-F4*(fh[idx_fh_stbd(iF,jF-1,kF,ord,ex)]+fh[idx_fh_stbd(iF,jF+1,kF,ord,ex)])+F6*fh[idx_fh_stbd(iF,jF,kF,ord,ex)])/dY;
-                const double Dz=((fh[idx_fh_stbd(iF,jF,kF-2,ord,ex)]+fh[idx_fh_stbd(iF,jF,kF+2,ord,ex)])-F4*(fh[idx_fh_stbd(iF,jF,kF-1,ord,ex)]+fh[idx_fh_stbd(iF,jF,kF+1,ord,ex)])+F6*fh[idx_fh_stbd(iF,jF,kF,ord,ex)])/dZ;
-                f_rhs[p]-=(eps/cof)*(Dx+Dy+Dz);
-            }}}
-        }
-        free(fh);return;
-    }
-#elif (ghost_width == 3)
-    {
-        const int ord=3, r=3;
-        const double cof=64.0,SIX=6.0,FIT=15.0,TWT=20.0;
-        const int NO_SYMM=0,OCTANT=2;
-        int iminF=1,jminF=1,kminF=1;
-        if(Symmetry==OCTANT&&fabs(X[0])<dX)iminF=-2;
-        if(Symmetry==OCTANT&&fabs(Y[0])<dY)jminF=-2;
-        if((sst==2||sst==4)&&fabs(Y[0])<dY)jminF=-2;
-
-        const size_t nx=(size_t)ex1+2*ord,ny=(size_t)ex2+2*ord,nz=(size_t)ex3,fh_size=nx*ny*nz;
-        double *fh=(double*)malloc(fh_size*sizeof(double));if(!fh)return;
-        symmetry_stbd(ord,ex,f,fh,SoA);
-
-        const int i0_lo=(iminF+2>0)?iminF+2:0,j0_lo=(jminF+2>0)?jminF+2:0,k0_lo=3;
-        const int i0_hi=imaxF-4,j0_hi=jmaxF-4,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=((fh[idx_fh_stbd(iF-3,jF,kF,ord,ex)]+fh[idx_fh_stbd(iF+3,jF,kF,ord,ex)])-SIX*(fh[idx_fh_stbd(iF-2,jF,kF,ord,ex)]+fh[idx_fh_stbd(iF+2,jF,kF,ord,ex)])+FIT*(fh[idx_fh_stbd(iF-1,jF,kF,ord,ex)]+fh[idx_fh_stbd(iF+1,jF,kF,ord,ex)])-TWT*fh[idx_fh_stbd(iF,jF,kF,ord,ex)])/dX;
-                const double Dy=((fh[idx_fh_stbd(iF,jF-3,kF,ord,ex)]+fh[idx_fh_stbd(iF,jF+3,kF,ord,ex)])-SIX*(fh[idx_fh_stbd(iF,jF-2,kF,ord,ex)]+fh[idx_fh_stbd(iF,jF+2,kF,ord,ex)])+FIT*(fh[idx_fh_stbd(iF,jF-1,kF,ord,ex)]+fh[idx_fh_stbd(iF,jF+1,kF,ord,ex)])-TWT*fh[idx_fh_stbd(iF,jF,kF,ord,ex)])/dY;
-                const double Dz=((fh[idx_fh_stbd(iF,jF,kF-3,ord,ex)]+fh[idx_fh_stbd(iF,jF,kF+3,ord,ex)])-SIX*(fh[idx_fh_stbd(iF,jF,kF-2,ord,ex)]+fh[idx_fh_stbd(iF,jF,kF+2,ord,ex)])+FIT*(fh[idx_fh_stbd(iF,jF,kF-1,ord,ex)]+fh[idx_fh_stbd(iF,jF,kF+1,ord,ex)])-TWT*fh[idx_fh_stbd(iF,jF,kF,ord,ex)])/dZ;
-                f_rhs[p]+=(eps/cof)*(Dx+Dy+Dz);
-            }}}
-        }
-        free(fh);return;
-    }
-#elif (ghost_width == 4)
-    {
-        const int ord=4, r=4;
-        const double cof=256.0,F8=8.0,F28=28.0,F56=56.0,F70=70.0;
-        const int NO_SYMM=0,OCTANT=2;
-        int iminF=1,jminF=1,kminF=1;
-        if(Symmetry==OCTANT&&fabs(X[0])<dX)iminF=-3;
-        if(Symmetry==OCTANT&&fabs(Y[0])<dY)jminF=-3;
-        if((sst==2||sst==4)&&fabs(Y[0])<dY)jminF=-3;
-
-        const size_t nx=(size_t)ex1+2*ord,ny=(size_t)ex2+2*ord,nz=(size_t)ex3,fh_size=nx*ny*nz;
-        double *fh=(double*)malloc(fh_size*sizeof(double));if(!fh)return;
-        symmetry_stbd(ord,ex,f,fh,SoA);
-
-        const int i0_lo=(iminF+3>0)?iminF+3:0,j0_lo=(jminF+3>0)?jminF+3:0,k0_lo=4;
-        const int i0_hi=imaxF-5,j0_hi=jmaxF-5,k0_hi=kmaxF-5;
-        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=((fh[idx_fh_stbd(iF-4,jF,kF,ord,ex)]+fh[idx_fh_stbd(iF+4,jF,kF,ord,ex)])-F8*(fh[idx_fh_stbd(iF-3,jF,kF,ord,ex)]+fh[idx_fh_stbd(iF+3,jF,kF,ord,ex)])+F28*(fh[idx_fh_stbd(iF-2,jF,kF,ord,ex)]+fh[idx_fh_stbd(iF+2,jF,kF,ord,ex)])-F56*(fh[idx_fh_stbd(iF-1,jF,kF,ord,ex)]+fh[idx_fh_stbd(iF+1,jF,kF,ord,ex)])+F70*fh[idx_fh_stbd(iF,jF,kF,ord,ex)])/dX;
-                const double Dy=((fh[idx_fh_stbd(iF,jF-4,kF,ord,ex)]+fh[idx_fh_stbd(iF,jF+4,kF,ord,ex)])-F8*(fh[idx_fh_stbd(iF,jF-3,kF,ord,ex)]+fh[idx_fh_stbd(iF,jF+3,kF,ord,ex)])+F28*(fh[idx_fh_stbd(iF,jF-2,kF,ord,ex)]+fh[idx_fh_stbd(iF,jF+2,kF,ord,ex)])-F56*(fh[idx_fh_stbd(iF,jF-1,kF,ord,ex)]+fh[idx_fh_stbd(iF,jF+1,kF,ord,ex)])+F70*fh[idx_fh_stbd(iF,jF,kF,ord,ex)])/dY;
-                const double Dz=((fh[idx_fh_stbd(iF,jF,kF-4,ord,ex)]+fh[idx_fh_stbd(iF,jF,kF+4,ord,ex)])-F8*(fh[idx_fh_stbd(iF,jF,kF-3,ord,ex)]+fh[idx_fh_stbd(iF,jF,kF+3,ord,ex)])+F28*(fh[idx_fh_stbd(iF,jF,kF-2,ord,ex)]+fh[idx_fh_stbd(iF,jF,kF+2,ord,ex)])-F56*(fh[idx_fh_stbd(iF,jF,kF-1,ord,ex)]+fh[idx_fh_stbd(iF,jF,kF+1,ord,ex)])+F70*fh[idx_fh_stbd(iF,jF,kF,ord,ex)])/dZ;
-                f_rhs[p]-=(eps/cof)*(Dx+Dy+Dz);
-            }}}
-        }
-        free(fh);return;
-    }
-#elif (ghost_width == 5)
-    {
-        const int ord=5, r=5;
-        const double cof=1024.0,F10=10.0,F45k=45.0,F120=120.0,F210=210.0,F252=252.0;
-        const int NO_SYMM=0,OCTANT=2;
-        int iminF=1,jminF=1,kminF=1;
-        if(Symmetry==OCTANT&&fabs(X[0])<dX)iminF=-4;
-        if(Symmetry==OCTANT&&fabs(Y[0])<dY)jminF=-4;
-        if((sst==2||sst==4)&&fabs(Y[0])<dY)jminF=-4;
-
-        const size_t nx=(size_t)ex1+2*ord,ny=(size_t)ex2+2*ord,nz=(size_t)ex3,fh_size=nx*ny*nz;
-        double *fh=(double*)malloc(fh_size*sizeof(double));if(!fh)return;
-        symmetry_stbd(ord,ex,f,fh,SoA);
-
-        const int i0_lo=(iminF+4>0)?iminF+4:0,j0_lo=(jminF+4>0)?jminF+4:0,k0_lo=5;
-        const int i0_hi=imaxF-6,j0_hi=jmaxF-6,k0_hi=kmaxF-6;
-        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=((fh[idx_fh_stbd(iF-5,jF,kF,ord,ex)]+fh[idx_fh_stbd(iF+5,jF,kF,ord,ex)])-F10*(fh[idx_fh_stbd(iF-4,jF,kF,ord,ex)]+fh[idx_fh_stbd(iF+4,jF,kF,ord,ex)])+F45k*(fh[idx_fh_stbd(iF-3,jF,kF,ord,ex)]+fh[idx_fh_stbd(iF+3,jF,kF,ord,ex)])-F120*(fh[idx_fh_stbd(iF-2,jF,kF,ord,ex)]+fh[idx_fh_stbd(iF+2,jF,kF,ord,ex)])+F210*(fh[idx_fh_stbd(iF-1,jF,kF,ord,ex)]+fh[idx_fh_stbd(iF+1,jF,kF,ord,ex)])-F252*fh[idx_fh_stbd(iF,jF,kF,ord,ex)])/dX;
-                const double Dy=((fh[idx_fh_stbd(iF,jF-5,kF,ord,ex)]+fh[idx_fh_stbd(iF,jF+5,kF,ord,ex)])-F10*(fh[idx_fh_stbd(iF,jF-4,kF,ord,ex)]+fh[idx_fh_stbd(iF,jF+4,kF,ord,ex)])+F45k*(fh[idx_fh_stbd(iF,jF-3,kF,ord,ex)]+fh[idx_fh_stbd(iF,jF+3,kF,ord,ex)])-F120*(fh[idx_fh_stbd(iF,jF-2,kF,ord,ex)]+fh[idx_fh_stbd(iF,jF+2,kF,ord,ex)])+F210*(fh[idx_fh_stbd(iF,jF-1,kF,ord,ex)]+fh[idx_fh_stbd(iF,jF+1,kF,ord,ex)])-F252*fh[idx_fh_stbd(iF,jF,kF,ord,ex)])/dY;
-                const double Dz=((fh[idx_fh_stbd(iF,jF,kF-5,ord,ex)]+fh[idx_fh_stbd(iF,jF,kF+5,ord,ex)])-F10*(fh[idx_fh_stbd(iF,jF,kF-4,ord,ex)]+fh[idx_fh_stbd(iF,jF,kF+4,ord,ex)])+F45k*(fh[idx_fh_stbd(iF,jF,kF-3,ord,ex)]+fh[idx_fh_stbd(iF,jF,kF+3,ord,ex)])-F120*(fh[idx_fh_stbd(iF,jF,kF-2,ord,ex)]+fh[idx_fh_stbd(iF,jF,kF+2,ord,ex)])+F210*(fh[idx_fh_stbd(iF,jF,kF-1,ord,ex)]+fh[idx_fh_stbd(iF,jF,kF+1,ord,ex)])-F252*fh[idx_fh_stbd(iF,jF,kF,ord,ex)])/dZ;
-                f_rhs[p]+=(eps/cof)*(Dx+Dy+Dz);
-            }}}
-        }
-        free(fh);return;
-    }
-#else
-#error "kodiss_sh_c.C: unsupported ghost_width"
-#endif
-}

AMSS_NCKU_source/lopsided_c.C

@@ -1,13 +1,14 @@
-#include "macrodef.h"
 #include "tool.h"
-
 /*
- * C 版 lopsided — upwind (lopsided) advection derivatives.
+ * 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)
  *
- * Adds advection terms to f_rhs for all three spatial directions.
- * Uses sign-biased (one-sided) stencils with centered fallbacks.
- *
- * For lopsided, symmetry_bd ord = ghost_width (same as kodiss).
+ * 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,
@@ -15,577 +16,240 @@ void lopsided(const int ex[3],
               const double *Sfx, const double *Sfy, const double *Sfz,
               int Symmetry, const double SoA[3])
 {
-    const double ZEO = 0.0, ONE = 1.0;
-    const double TWO = 2.0, F6 = 6.0, EIT = 8.0;
-    const double F3 = 3.0, F4 = 4.0, F5 = 5.0, F10 = 10.0, F12 = 12.0, F18 = 18.0;
-    const double F9 = 9.0, F45 = 45.0, F60 = 60.0;
-    const double F2 = 2.0, F15 = 15.0, F24 = 24.0, F30 = 30.0, F35 = 35.0;
-    const double F50 = 50.0, F77 = 77.0, F80 = 80.0, F100 = 100.0, F150 = 150.0;
-    const double F32 = 32.0, F168 = 168.0, F672 = 672.0, F840 = 840.0;
-    const double F140=140.0, F378=378.0, F420=420.0, F1050=1050.0;
+    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 NO_SYMM = 0, EQ_SYMM = 1;
     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];
 
-#if (ghost_width == 2)
-    /* ---- 2nd-order lopsided --------------------------------------------- */
-    {
-        const int ord = 2;
-        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 d12dx = ONE / F12 / dX;
+    const double d12dy = ONE / F12 / dY;
+    const double d12dz = ONE / F12 / dZ;
 
-        const size_t nx = (size_t)ex1 + ord;
-        const size_t ny = (size_t)ex2 + ord;
-        const size_t nz = (size_t)ex3 + ord;
-        const size_t fh_size = nx * ny * nz;
+    // 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;
 
-        double *fh = (double*)malloc(fh_size * sizeof(double));
-        if (!fh) return;
-        symmetry_bd(ord, ex, f, fh, SoA);
+    // Fortran:
+    // imax = ex(1); jmax = ex(2); kmax = ex(3)
+    const int imaxF = ex1;
+    const int jmaxF = ex2;
+    const int kmaxF = ex3;
 
-        const double d2dx  = ONE / TWO / dX;
-        const double d2dy  = ONE / TWO / dY;
-        const double d2dz  = ONE / TWO / dZ;
+    // 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;
 
-        const int imaxF = ex1, jmaxF = ex2, kmaxF = ex3;
+    // 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;
 
-        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);
+    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）
 
-                    /* x-direction */
-                    const double sfx = Sfx[p];
-                    if (sfx > ZEO) {
-                        if (i0 <= ex1 - 3) // i+2 <= imax
-                            f_rhs[p] += sfx * d2dx * (
-                                -F3*fh[idx_fh_F_ord2(iF,   jF, kF, ex)] +
-                                 F4*fh[idx_fh_F_ord2(iF+1, jF, kF, ex)] -
-                                    fh[idx_fh_F_ord2(iF+2, jF, kF, ex)]);
-                        else if (i0 <= ex1 - 2) // i+1 <= imax
-                            f_rhs[p] += sfx * d2dx * (
-                                -fh[idx_fh_F_ord2(iF,   jF, kF, ex)] +
-                                 fh[idx_fh_F_ord2(iF+1, jF, kF, ex)]);
-                    } else if (sfx < ZEO) {
-                        if ((i0 - 1) >= iminF) // i-2 >= imin
-                            f_rhs[p] -= sfx * d2dx * (
-                                -F3*fh[idx_fh_F_ord2(iF,   jF, kF, ex)] +
-                                 F4*fh[idx_fh_F_ord2(iF-1, jF, kF, ex)] -
-                                    fh[idx_fh_F_ord2(iF-2, jF, kF, ex)]);
-                        else if (i0 >= iminF) // i-1 >= imin
-                            f_rhs[p] -= sfx * d2dx * (
-                                -fh[idx_fh_F_ord2(iF,   jF, kF, ex)] +
-                                 fh[idx_fh_F_ord2(iF-1, jF, kF, ex)]);
+    // 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)]);
                     }
-
-                    /* y-direction */
-                    const double sfy = Sfy[p];
-                    if (sfy > ZEO) {
-                        if (j0 <= ex2-3)
-                            f_rhs[p] += sfy * d2dy * (
-                                -F3*fh[idx_fh_F_ord2(iF, jF,   kF, ex)] +
-                                 F4*fh[idx_fh_F_ord2(iF, jF+1, kF, ex)] -
-                                    fh[idx_fh_F_ord2(iF, jF+2, kF, ex)]);
-                        else if (j0 <= ex2-2)
-                            f_rhs[p] += sfy * d2dy * (
-                                -fh[idx_fh_F_ord2(iF, jF,   kF, ex)] +
-                                 fh[idx_fh_F_ord2(iF, jF+1, kF, ex)]);
-                    } else if (sfy < ZEO) {
-                        if ((j0-1) >= jminF)
-                            f_rhs[p] -= sfy * d2dy * (
-                                -F3*fh[idx_fh_F_ord2(iF, jF,   kF, ex)] +
-                                 F4*fh[idx_fh_F_ord2(iF, jF-1, kF, ex)] -
-                                    fh[idx_fh_F_ord2(iF, jF-2, kF, ex)]);
-                        else if (j0 >= jminF)
-                            f_rhs[p] -= sfy * d2dy * (
-                                -fh[idx_fh_F_ord2(iF, jF,   kF, ex)] +
-                                 fh[idx_fh_F_ord2(iF, jF-1, 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)]);
+                    }
+                }
 
-                    /* z-direction */
-                    const double sfz = Sfz[p];
-                    if (sfz > ZEO) {
-                        if (k0 <= ex3-3)
-                            f_rhs[p] += sfz * d2dz * (
-                                -F3*fh[idx_fh_F_ord2(iF, jF, kF,   ex)] +
-                                 F4*fh[idx_fh_F_ord2(iF, jF, kF+1, ex)] -
-                                    fh[idx_fh_F_ord2(iF, jF, kF+2, ex)]);
-                        else if (k0 <= ex3-2)
-                            f_rhs[p] += sfz * d2dz * (
-                                -fh[idx_fh_F_ord2(iF, jF, kF,   ex)] +
-                                 fh[idx_fh_F_ord2(iF, jF, kF+1, ex)]);
-                    } else if (sfz < ZEO) {
-                        if ((k0-1) >= kminF)
-                            f_rhs[p] -= sfz * d2dz * (
-                                -F3*fh[idx_fh_F_ord2(iF, jF, kF,   ex)] +
-                                 F4*fh[idx_fh_F_ord2(iF, jF, kF-1, ex)] -
-                                    fh[idx_fh_F_ord2(iF, jF, kF-2, ex)]);
-                        else if (k0 >= kminF)
-                            f_rhs[p] -= sfz * d2dz * (
-                                -fh[idx_fh_F_ord2(iF, jF, kF,   ex)] +
-                                 fh[idx_fh_F_ord2(iF, jF, kF-1, 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);
-        return;
     }
-#elif (ghost_width == 3)
-    /* ---- 4th-order lopsided (original code) ---------------------------- */
-    {
-        const int ord = 3;
-        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;
-
-        const size_t nx = (size_t)ex1 + ord;
-        const size_t ny = (size_t)ex2 + ord;
-        const size_t nz = (size_t)ex3 + ord;
-        const size_t fh_size = nx * ny * nz;
-
-        double *fh = (double*)malloc(fh_size * sizeof(double));
-        if (!fh) return;
-        symmetry_bd(ord, ex, f, fh, SoA);
-
-        const double d12dx = ONE / F12 / dX;
-        const double d12dy = ONE / F12 / dY;
-        const double d12dz = ONE / F12 / dZ;
-
-        const int imaxF = ex1, jmaxF = ex2, kmaxF = ex3;
-
-        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) // i+3 <= imax
-                            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) // i+2 <= imax
-                            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) // i+1 <= imax → mirrored
-                            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) // i-3 >= imin
-                            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) // i-2 >= imin
-                            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) // i-1 >= imin → mirrored
-                            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)]);
-                    }
-                }
-            }
-        }
-        free(fh);
-        return;
-    }
-#elif (ghost_width == 4)
-    /* ---- 6th-order lopsided --------------------------------------------- */
-    {
-        const int ord = 4;
-        int iminF = 1, jminF = 1, kminF = 1;
-        if (Symmetry > NO_SYMM && fabs(Z[0]) < dZ) kminF = -3;
-        if (Symmetry > EQ_SYMM && fabs(X[0]) < dX) iminF = -3;
-        if (Symmetry > EQ_SYMM && fabs(Y[0]) < dY) jminF = -3;
-
-        const size_t nx = (size_t)ex1 + ord;
-        const size_t ny = (size_t)ex2 + ord;
-        const size_t nz = (size_t)ex3 + ord;
-        const size_t fh_size = nx * ny * nz;
-
-        double *fh = (double*)malloc(fh_size * sizeof(double));
-        if (!fh) return;
-        symmetry_bd(ord, ex, f, fh, SoA);
-
-        const double d60dx = ONE / F60 / dX;
-        const double d60dy = ONE / F60 / dY;
-        const double d60dz = ONE / F60 / dZ;
-        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;
-
-        const int imaxF = ex1, jmaxF = ex2, kmaxF = ex3;
-
-        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) {
-                        /* Primary biased: 2*f(i-2)-24*f(i-1)-35*f(i)+80*f(i+1)-30*f(i+2)+8*f(i+3)-f(i+4) */
-                        if (i0 <= ex1-5 && (i0-1)>=iminF) // i+4<=imax && i-2>=imin
-                            f_rhs[p] += sfx * d60dx * (
-                                +F2*fh[idx_fh_F_ord4(iF-2,jF,kF,ex)]-F24*fh[idx_fh_F_ord4(iF-1,jF,kF,ex)]
-                                -F35*fh[idx_fh_F_ord4(iF,jF,kF,ex)]+F80*fh[idx_fh_F_ord4(iF+1,jF,kF,ex)]
-                                -F30*fh[idx_fh_F_ord4(iF+2,jF,kF,ex)]+EIT*fh[idx_fh_F_ord4(iF+3,jF,kF,ex)]
-                                -fh[idx_fh_F_ord4(iF+4,jF,kF,ex)]);
-                        /* Boundary-adapted: -10*f(i-1)-77*f(i)+150*f(i+1)-100*f(i+2)+50*f(i+3)-15*f(i+4)+2*f(i+5) */
-                        else if (i0 <= ex1-6 && i0 >= iminF) // i+5<=imax && i-1>=imin
-                            f_rhs[p] += sfx * d60dx * (
-                                -F10*fh[idx_fh_F_ord4(iF-1,jF,kF,ex)]-F77*fh[idx_fh_F_ord4(iF,jF,kF,ex)]
-                                +F150*fh[idx_fh_F_ord4(iF+1,jF,kF,ex)]-F100*fh[idx_fh_F_ord4(iF+2,jF,kF,ex)]
-                                +F50*fh[idx_fh_F_ord4(iF+3,jF,kF,ex)]-F15*fh[idx_fh_F_ord4(iF+4,jF,kF,ex)]
-                                +F2*fh[idx_fh_F_ord4(iF+5,jF,kF,ex)]);
-                        /* Centered fallbacks */
-                        else if (i0 <= ex1-4 && (i0-2)>=iminF) // 6th: i+3<=imax && i-3>=imin
-                            f_rhs[p] += sfx * d60dx * (
-                                -fh[idx_fh_F_ord4(iF-3,jF,kF,ex)]+F9*fh[idx_fh_F_ord4(iF-2,jF,kF,ex)]
-                                -F45*fh[idx_fh_F_ord4(iF-1,jF,kF,ex)]+F45*fh[idx_fh_F_ord4(iF+1,jF,kF,ex)]
-                                -F9*fh[idx_fh_F_ord4(iF+2,jF,kF,ex)]+fh[idx_fh_F_ord4(iF+3,jF,kF,ex)]);
-                        else if (i0 <= ex1-3 && (i0-1)>=iminF) // 4th
-                            f_rhs[p] += sfx * d12dx * (
-                                fh[idx_fh_F_ord4(iF-2,jF,kF,ex)]-EIT*fh[idx_fh_F_ord4(iF-1,jF,kF,ex)]
-                                +EIT*fh[idx_fh_F_ord4(iF+1,jF,kF,ex)]-fh[idx_fh_F_ord4(iF+2,jF,kF,ex)]);
-                        else if (i0 <= ex1-2 && i0>=iminF) // 2nd
-                            f_rhs[p] += sfx * d2dx * (
-                                -fh[idx_fh_F_ord4(iF-1,jF,kF,ex)]+fh[idx_fh_F_ord4(iF+1,jF,kF,ex)]);
-                    } else if (sfx < ZEO) {
-                        if ((i0-3)>=iminF && i0<=ex1-3) // i-4>=imin && i+2<=imax
-                            f_rhs[p] -= sfx * d60dx * (
-                                +F2*fh[idx_fh_F_ord4(iF+2,jF,kF,ex)]-F24*fh[idx_fh_F_ord4(iF+1,jF,kF,ex)]
-                                -F35*fh[idx_fh_F_ord4(iF,jF,kF,ex)]+F80*fh[idx_fh_F_ord4(iF-1,jF,kF,ex)]
-                                -F30*fh[idx_fh_F_ord4(iF-2,jF,kF,ex)]+EIT*fh[idx_fh_F_ord4(iF-3,jF,kF,ex)]
-                                -fh[idx_fh_F_ord4(iF-4,jF,kF,ex)]);
-                        else if ((i0-4)>=iminF && i0<=ex1-2) // i-5>=imin && i+1<=imax
-                            f_rhs[p] -= sfx * d60dx * (
-                                -F10*fh[idx_fh_F_ord4(iF+1,jF,kF,ex)]-F77*fh[idx_fh_F_ord4(iF,jF,kF,ex)]
-                                +F150*fh[idx_fh_F_ord4(iF-1,jF,kF,ex)]-F100*fh[idx_fh_F_ord4(iF-2,jF,kF,ex)]
-                                +F50*fh[idx_fh_F_ord4(iF-3,jF,kF,ex)]-F15*fh[idx_fh_F_ord4(iF-4,jF,kF,ex)]
-                                +F2*fh[idx_fh_F_ord4(iF-5,jF,kF,ex)]);
-                        else if ((i0-2)>=iminF && i0<=ex1-4) // 6th centered
-                            f_rhs[p] += sfx * d60dx * (
-                                -fh[idx_fh_F_ord4(iF-3,jF,kF,ex)]+F9*fh[idx_fh_F_ord4(iF-2,jF,kF,ex)]
-                                -F45*fh[idx_fh_F_ord4(iF-1,jF,kF,ex)]+F45*fh[idx_fh_F_ord4(iF+1,jF,kF,ex)]
-                                -F9*fh[idx_fh_F_ord4(iF+2,jF,kF,ex)]+fh[idx_fh_F_ord4(iF+3,jF,kF,ex)]);
-                        else if ((i0-1)>=iminF && i0<=ex1-3) // 4th
-                            f_rhs[p] += sfx * d12dx * (
-                                fh[idx_fh_F_ord4(iF-2,jF,kF,ex)]-EIT*fh[idx_fh_F_ord4(iF-1,jF,kF,ex)]
-                                +EIT*fh[idx_fh_F_ord4(iF+1,jF,kF,ex)]-fh[idx_fh_F_ord4(iF+2,jF,kF,ex)]);
-                        else if (i0>=iminF && i0<=ex1-2) // 2nd
-                            f_rhs[p] += sfx * d2dx * (
-                                -fh[idx_fh_F_ord4(iF-1,jF,kF,ex)]+fh[idx_fh_F_ord4(iF+1,jF,kF,ex)]);
-                    }
-
-                    /* ---- y-direction ---- */
-                    const double sfy = Sfy[p];
-                    if (sfy > ZEO) {
-                        if (j0<=ex2-5 && (j0-1)>=jminF)
-                            f_rhs[p] += sfy * d60dy*(F2*fh[idx_fh_F_ord4(iF,jF-2,kF,ex)]-F24*fh[idx_fh_F_ord4(iF,jF-1,kF,ex)]-F35*fh[idx_fh_F_ord4(iF,jF,kF,ex)]+F80*fh[idx_fh_F_ord4(iF,jF+1,kF,ex)]-F30*fh[idx_fh_F_ord4(iF,jF+2,kF,ex)]+EIT*fh[idx_fh_F_ord4(iF,jF+3,kF,ex)]-fh[idx_fh_F_ord4(iF,jF+4,kF,ex)]);
-                        else if (j0<=ex2-6 && j0>=jminF)
-                            f_rhs[p] += sfy * d60dy*(-F10*fh[idx_fh_F_ord4(iF,jF-1,kF,ex)]-F77*fh[idx_fh_F_ord4(iF,jF,kF,ex)]+F150*fh[idx_fh_F_ord4(iF,jF+1,kF,ex)]-F100*fh[idx_fh_F_ord4(iF,jF+2,kF,ex)]+F50*fh[idx_fh_F_ord4(iF,jF+3,kF,ex)]-F15*fh[idx_fh_F_ord4(iF,jF+4,kF,ex)]+F2*fh[idx_fh_F_ord4(iF,jF+5,kF,ex)]);
-                        else if (j0<=ex2-4 && (j0-2)>=jminF)
-                            f_rhs[p] += sfy * d60dy*(-fh[idx_fh_F_ord4(iF,jF-3,kF,ex)]+F9*fh[idx_fh_F_ord4(iF,jF-2,kF,ex)]-F45*fh[idx_fh_F_ord4(iF,jF-1,kF,ex)]+F45*fh[idx_fh_F_ord4(iF,jF+1,kF,ex)]-F9*fh[idx_fh_F_ord4(iF,jF+2,kF,ex)]+fh[idx_fh_F_ord4(iF,jF+3,kF,ex)]);
-                        else if (j0<=ex2-3 && (j0-1)>=jminF)
-                            f_rhs[p] += sfy * d12dy*(fh[idx_fh_F_ord4(iF,jF-2,kF,ex)]-EIT*fh[idx_fh_F_ord4(iF,jF-1,kF,ex)]+EIT*fh[idx_fh_F_ord4(iF,jF+1,kF,ex)]-fh[idx_fh_F_ord4(iF,jF+2,kF,ex)]);
-                        else if (j0<=ex2-2 && j0>=jminF)
-                            f_rhs[p] += sfy * d2dy*(-fh[idx_fh_F_ord4(iF,jF-1,kF,ex)]+fh[idx_fh_F_ord4(iF,jF+1,kF,ex)]);
-                    } else if (sfy < ZEO) {
-                        if ((j0-3)>=jminF && j0<=ex2-3)
-                            f_rhs[p] -= sfy * d60dy*(F2*fh[idx_fh_F_ord4(iF,jF+2,kF,ex)]-F24*fh[idx_fh_F_ord4(iF,jF+1,kF,ex)]-F35*fh[idx_fh_F_ord4(iF,jF,kF,ex)]+F80*fh[idx_fh_F_ord4(iF,jF-1,kF,ex)]-F30*fh[idx_fh_F_ord4(iF,jF-2,kF,ex)]+EIT*fh[idx_fh_F_ord4(iF,jF-3,kF,ex)]-fh[idx_fh_F_ord4(iF,jF-4,kF,ex)]);
-                        else if ((j0-4)>=jminF && j0<=ex2-2)
-                            f_rhs[p] -= sfy * d60dy*(-F10*fh[idx_fh_F_ord4(iF,jF+1,kF,ex)]-F77*fh[idx_fh_F_ord4(iF,jF,kF,ex)]+F150*fh[idx_fh_F_ord4(iF,jF-1,kF,ex)]-F100*fh[idx_fh_F_ord4(iF,jF-2,kF,ex)]+F50*fh[idx_fh_F_ord4(iF,jF-3,kF,ex)]-F15*fh[idx_fh_F_ord4(iF,jF-4,kF,ex)]+F2*fh[idx_fh_F_ord4(iF,jF-5,kF,ex)]);
-                        else if ((j0-2)>=jminF && j0<=ex2-4)
-                            f_rhs[p] += sfy * d60dy*(-fh[idx_fh_F_ord4(iF,jF-3,kF,ex)]+F9*fh[idx_fh_F_ord4(iF,jF-2,kF,ex)]-F45*fh[idx_fh_F_ord4(iF,jF-1,kF,ex)]+F45*fh[idx_fh_F_ord4(iF,jF+1,kF,ex)]-F9*fh[idx_fh_F_ord4(iF,jF+2,kF,ex)]+fh[idx_fh_F_ord4(iF,jF+3,kF,ex)]);
-                        else if ((j0-1)>=jminF && j0<=ex2-3)
-                            f_rhs[p] += sfy * d12dy*(fh[idx_fh_F_ord4(iF,jF-2,kF,ex)]-EIT*fh[idx_fh_F_ord4(iF,jF-1,kF,ex)]+EIT*fh[idx_fh_F_ord4(iF,jF+1,kF,ex)]-fh[idx_fh_F_ord4(iF,jF+2,kF,ex)]);
-                        else if (j0>=jminF && j0<=ex2-2)
-                            f_rhs[p] += sfy * d2dy*(-fh[idx_fh_F_ord4(iF,jF-1,kF,ex)]+fh[idx_fh_F_ord4(iF,jF+1,kF,ex)]);
-                    }
-
-                    /* ---- z-direction ---- */
-                    const double sfz = Sfz[p];
-                    if (sfz > ZEO) {
-                        if (k0<=ex3-5 && (k0-1)>=kminF)
-                            f_rhs[p] += sfz * d60dz*(F2*fh[idx_fh_F_ord4(iF,jF,kF-2,ex)]-F24*fh[idx_fh_F_ord4(iF,jF,kF-1,ex)]-F35*fh[idx_fh_F_ord4(iF,jF,kF,ex)]+F80*fh[idx_fh_F_ord4(iF,jF,kF+1,ex)]-F30*fh[idx_fh_F_ord4(iF,jF,kF+2,ex)]+EIT*fh[idx_fh_F_ord4(iF,jF,kF+3,ex)]-fh[idx_fh_F_ord4(iF,jF,kF+4,ex)]);
-                        else if (k0<=ex3-6 && k0>=kminF)
-                            f_rhs[p] += sfz * d60dz*(-F10*fh[idx_fh_F_ord4(iF,jF,kF-1,ex)]-F77*fh[idx_fh_F_ord4(iF,jF,kF,ex)]+F150*fh[idx_fh_F_ord4(iF,jF,kF+1,ex)]-F100*fh[idx_fh_F_ord4(iF,jF,kF+2,ex)]+F50*fh[idx_fh_F_ord4(iF,jF,kF+3,ex)]-F15*fh[idx_fh_F_ord4(iF,jF,kF+4,ex)]+F2*fh[idx_fh_F_ord4(iF,jF,kF+5,ex)]);
-                        else if (k0<=ex3-4 && (k0-2)>=kminF)
-                            f_rhs[p] += sfz * d60dz*(-fh[idx_fh_F_ord4(iF,jF,kF-3,ex)]+F9*fh[idx_fh_F_ord4(iF,jF,kF-2,ex)]-F45*fh[idx_fh_F_ord4(iF,jF,kF-1,ex)]+F45*fh[idx_fh_F_ord4(iF,jF,kF+1,ex)]-F9*fh[idx_fh_F_ord4(iF,jF,kF+2,ex)]+fh[idx_fh_F_ord4(iF,jF,kF+3,ex)]);
-                        else if (k0<=ex3-3 && (k0-1)>=kminF)
-                            f_rhs[p] += sfz * d12dz*(fh[idx_fh_F_ord4(iF,jF,kF-2,ex)]-EIT*fh[idx_fh_F_ord4(iF,jF,kF-1,ex)]+EIT*fh[idx_fh_F_ord4(iF,jF,kF+1,ex)]-fh[idx_fh_F_ord4(iF,jF,kF+2,ex)]);
-                        else if (k0<=ex3-2 && k0>=kminF)
-                            f_rhs[p] += sfz * d2dz*(-fh[idx_fh_F_ord4(iF,jF,kF-1,ex)]+fh[idx_fh_F_ord4(iF,jF,kF+1,ex)]);
-                    } else if (sfz < ZEO) {
-                        if ((k0-3)>=kminF && k0<=ex3-3)
-                            f_rhs[p] -= sfz * d60dz*(F2*fh[idx_fh_F_ord4(iF,jF,kF+2,ex)]-F24*fh[idx_fh_F_ord4(iF,jF,kF+1,ex)]-F35*fh[idx_fh_F_ord4(iF,jF,kF,ex)]+F80*fh[idx_fh_F_ord4(iF,jF,kF-1,ex)]-F30*fh[idx_fh_F_ord4(iF,jF,kF-2,ex)]+EIT*fh[idx_fh_F_ord4(iF,jF,kF-3,ex)]-fh[idx_fh_F_ord4(iF,jF,kF-4,ex)]);
-                        else if ((k0-4)>=kminF && k0<=ex3-2)
-                            f_rhs[p] -= sfz * d60dz*(-F10*fh[idx_fh_F_ord4(iF,jF,kF+1,ex)]-F77*fh[idx_fh_F_ord4(iF,jF,kF,ex)]+F150*fh[idx_fh_F_ord4(iF,jF,kF-1,ex)]-F100*fh[idx_fh_F_ord4(iF,jF,kF-2,ex)]+F50*fh[idx_fh_F_ord4(iF,jF,kF-3,ex)]-F15*fh[idx_fh_F_ord4(iF,jF,kF-4,ex)]+F2*fh[idx_fh_F_ord4(iF,jF,kF-5,ex)]);
-                        else if ((k0-2)>=kminF && k0<=ex3-4)
-                            f_rhs[p] += sfz * d60dz*(-fh[idx_fh_F_ord4(iF,jF,kF-3,ex)]+F9*fh[idx_fh_F_ord4(iF,jF,kF-2,ex)]-F45*fh[idx_fh_F_ord4(iF,jF,kF-1,ex)]+F45*fh[idx_fh_F_ord4(iF,jF,kF+1,ex)]-F9*fh[idx_fh_F_ord4(iF,jF,kF+2,ex)]+fh[idx_fh_F_ord4(iF,jF,kF+3,ex)]);
-                        else if ((k0-1)>=kminF && k0<=ex3-3)
-                            f_rhs[p] += sfz * d12dz*(fh[idx_fh_F_ord4(iF,jF,kF-2,ex)]-EIT*fh[idx_fh_F_ord4(iF,jF,kF-1,ex)]+EIT*fh[idx_fh_F_ord4(iF,jF,kF+1,ex)]-fh[idx_fh_F_ord4(iF,jF,kF+2,ex)]);
-                        else if (k0>=kminF && k0<=ex3-2)
-                            f_rhs[p] += sfz * d2dz*(-fh[idx_fh_F_ord4(iF,jF,kF-1,ex)]+fh[idx_fh_F_ord4(iF,jF,kF+1,ex)]);
-                    }
-                }
-            }
-        }
-        free(fh);
-        return;
-    }
-#elif (ghost_width == 5)
-    /* ---- 8th-order lopsided --------------------------------------------- */
-    {
-        const int ord = 5;
-        int iminF = 1, jminF = 1, kminF = 1;
-        if (Symmetry > NO_SYMM && fabs(Z[0]) < dZ) kminF = -4;
-        if (Symmetry > EQ_SYMM && fabs(X[0]) < dX) iminF = -4;
-        if (Symmetry > EQ_SYMM && fabs(Y[0]) < dY) jminF = -4;
-
-        const size_t nx = (size_t)ex1 + ord;
-        const size_t ny = (size_t)ex2 + ord;
-        const size_t nz = (size_t)ex3 + ord;
-        const size_t fh_size = nx * ny * nz;
-
-        double *fh = (double*)malloc(fh_size * sizeof(double));
-        if (!fh) return;
-        symmetry_bd(ord, ex, f, fh, SoA);
-
-        const double d840dx = ONE / F840 / dX;
-        const double d840dy = ONE / F840 / dY;
-        const double d840dz = ONE / F840 / dZ;
-        const double d60dx = ONE / F60 / dX;
-        const double d60dy = ONE / F60 / dY;
-        const double d60dz = ONE / F60 / dZ;
-        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;
-
-        const int imaxF = ex1, jmaxF = ex2, kmaxF = ex3;
-
-        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) {
-                        /* 8th biased: -5*f(i-3)+60*f(i-2)-420*f(i-1)-378*f(i)+1050*f(i+1)-420*f(i+2)+140*f(i+3)-30*f(i+4)+3*f(i+5) */
-                        if (i0 <= ex1-6 && (i0-2)>=iminF) // i+5<=imax && i-3>=imin
-                            f_rhs[p] += sfx * d840dx * (
-                                -F5*fh[idx_fh_F_ord5(iF-3,jF,kF,ex)]+F60*fh[idx_fh_F_ord5(iF-2,jF,kF,ex)]
-                                -F420*fh[idx_fh_F_ord5(iF-1,jF,kF,ex)]-F378*fh[idx_fh_F_ord5(iF,jF,kF,ex)]
-                                +F1050*fh[idx_fh_F_ord5(iF+1,jF,kF,ex)]-F420*fh[idx_fh_F_ord5(iF+2,jF,kF,ex)]
-                                +F140*fh[idx_fh_F_ord5(iF+3,jF,kF,ex)]-F30*fh[idx_fh_F_ord5(iF+4,jF,kF,ex)]
-                                +F3*fh[idx_fh_F_ord5(iF+5,jF,kF,ex)]);
-                        /* 8th centered: +3*f(i-4)-32*f(i-3)+168*f(i-2)-672*f(i-1)+672*f(i+1)-168*f(i+2)+32*f(i+3)-3*f(i+4) */
-                        else if (i0 <= ex1-5 && (i0-3)>=iminF)
-                            f_rhs[p] += sfx * d840dx * (
-                                +F3*fh[idx_fh_F_ord5(iF-4,jF,kF,ex)]-F32*fh[idx_fh_F_ord5(iF-3,jF,kF,ex)]
-                                +F168*fh[idx_fh_F_ord5(iF-2,jF,kF,ex)]-F672*fh[idx_fh_F_ord5(iF-1,jF,kF,ex)]
-                                +F672*fh[idx_fh_F_ord5(iF+1,jF,kF,ex)]-F168*fh[idx_fh_F_ord5(iF+2,jF,kF,ex)]
-                                +F32*fh[idx_fh_F_ord5(iF+3,jF,kF,ex)]-F3*fh[idx_fh_F_ord5(iF+4,jF,kF,ex)]);
-                        else if (i0 <= ex1-4 && (i0-2)>=iminF) // 6th centered
-                            f_rhs[p] += sfx * d60dx * (
-                                -fh[idx_fh_F_ord5(iF-3,jF,kF,ex)]+F9*fh[idx_fh_F_ord5(iF-2,jF,kF,ex)]
-                                -F45*fh[idx_fh_F_ord5(iF-1,jF,kF,ex)]+F45*fh[idx_fh_F_ord5(iF+1,jF,kF,ex)]
-                                -F9*fh[idx_fh_F_ord5(iF+2,jF,kF,ex)]+fh[idx_fh_F_ord5(iF+3,jF,kF,ex)]);
-                        else if (i0 <= ex1-3 && (i0-1)>=iminF) // 4th centered
-                            f_rhs[p] += sfx * d12dx * (
-                                fh[idx_fh_F_ord5(iF-2,jF,kF,ex)]-EIT*fh[idx_fh_F_ord5(iF-1,jF,kF,ex)]
-                                +EIT*fh[idx_fh_F_ord5(iF+1,jF,kF,ex)]-fh[idx_fh_F_ord5(iF+2,jF,kF,ex)]);
-                        else if (i0 <= ex1-2 && i0>=iminF) // 2nd centered
-                            f_rhs[p] += sfx * d2dx * (
-                                -fh[idx_fh_F_ord5(iF-1,jF,kF,ex)]+fh[idx_fh_F_ord5(iF+1,jF,kF,ex)]);
-                    } else if (sfx < ZEO) {
-                        if ((i0-4)>=iminF && i0<=ex1-4)
-                            f_rhs[p] -= sfx * d840dx * (
-                                -F5*fh[idx_fh_F_ord5(iF+3,jF,kF,ex)]+F60*fh[idx_fh_F_ord5(iF+2,jF,kF,ex)]
-                                -F420*fh[idx_fh_F_ord5(iF+1,jF,kF,ex)]-F378*fh[idx_fh_F_ord5(iF,jF,kF,ex)]
-                                +F1050*fh[idx_fh_F_ord5(iF-1,jF,kF,ex)]-F420*fh[idx_fh_F_ord5(iF-2,jF,kF,ex)]
-                                +F140*fh[idx_fh_F_ord5(iF-3,jF,kF,ex)]-F30*fh[idx_fh_F_ord5(iF-4,jF,kF,ex)]
-                                +F3*fh[idx_fh_F_ord5(iF-5,jF,kF,ex)]);
-                        else if ((i0-3)>=iminF && i0<=ex1-5) // 8th centered
-                            f_rhs[p] += sfx * d840dx * (
-                                +F3*fh[idx_fh_F_ord5(iF-4,jF,kF,ex)]-F32*fh[idx_fh_F_ord5(iF-3,jF,kF,ex)]
-                                +F168*fh[idx_fh_F_ord5(iF-2,jF,kF,ex)]-F672*fh[idx_fh_F_ord5(iF-1,jF,kF,ex)]
-                                +F672*fh[idx_fh_F_ord5(iF+1,jF,kF,ex)]-F168*fh[idx_fh_F_ord5(iF+2,jF,kF,ex)]
-                                +F32*fh[idx_fh_F_ord5(iF+3,jF,kF,ex)]-F3*fh[idx_fh_F_ord5(iF+4,jF,kF,ex)]);
-                        else if ((i0-2)>=iminF && i0<=ex1-4) // 6th centered
-                            f_rhs[p] += sfx * d60dx * (
-                                -fh[idx_fh_F_ord5(iF-3,jF,kF,ex)]+F9*fh[idx_fh_F_ord5(iF-2,jF,kF,ex)]
-                                -F45*fh[idx_fh_F_ord5(iF-1,jF,kF,ex)]+F45*fh[idx_fh_F_ord5(iF+1,jF,kF,ex)]
-                                -F9*fh[idx_fh_F_ord5(iF+2,jF,kF,ex)]+fh[idx_fh_F_ord5(iF+3,jF,kF,ex)]);
-                        else if ((i0-1)>=iminF && i0<=ex1-3) // 4th centered
-                            f_rhs[p] += sfx * d12dx * (
-                                fh[idx_fh_F_ord5(iF-2,jF,kF,ex)]-EIT*fh[idx_fh_F_ord5(iF-1,jF,kF,ex)]
-                                +EIT*fh[idx_fh_F_ord5(iF+1,jF,kF,ex)]-fh[idx_fh_F_ord5(iF+2,jF,kF,ex)]);
-                        else if (i0>=iminF && i0<=ex1-2) // 2nd centered
-                            f_rhs[p] += sfx * d2dx * (
-                                -fh[idx_fh_F_ord5(iF-1,jF,kF,ex)]+fh[idx_fh_F_ord5(iF+1,jF,kF,ex)]);
-                    }
-
-                    const double sfy = Sfy[p];
-                    if (sfy > ZEO) {
-                        if (j0<=ex2-6 && (j0-2)>=jminF)
-                            f_rhs[p] += sfy * d840dy*(-F5*fh[idx_fh_F_ord5(iF,jF-3,kF,ex)]+F60*fh[idx_fh_F_ord5(iF,jF-2,kF,ex)]-F420*fh[idx_fh_F_ord5(iF,jF-1,kF,ex)]-F378*fh[idx_fh_F_ord5(iF,jF,kF,ex)]+F1050*fh[idx_fh_F_ord5(iF,jF+1,kF,ex)]-F420*fh[idx_fh_F_ord5(iF,jF+2,kF,ex)]+F140*fh[idx_fh_F_ord5(iF,jF+3,kF,ex)]-F30*fh[idx_fh_F_ord5(iF,jF+4,kF,ex)]+F3*fh[idx_fh_F_ord5(iF,jF+5,kF,ex)]);
-                        else if (j0<=ex2-5 && (j0-3)>=jminF)
-                            f_rhs[p] += sfy * d840dy*(+F3*fh[idx_fh_F_ord5(iF,jF-4,kF,ex)]-F32*fh[idx_fh_F_ord5(iF,jF-3,kF,ex)]+F168*fh[idx_fh_F_ord5(iF,jF-2,kF,ex)]-F672*fh[idx_fh_F_ord5(iF,jF-1,kF,ex)]+F672*fh[idx_fh_F_ord5(iF,jF+1,kF,ex)]-F168*fh[idx_fh_F_ord5(iF,jF+2,kF,ex)]+F32*fh[idx_fh_F_ord5(iF,jF+3,kF,ex)]-F3*fh[idx_fh_F_ord5(iF,jF+4,kF,ex)]);
-                        else if (j0<=ex2-4 && (j0-2)>=jminF)
-                            f_rhs[p] += sfy * d60dy*(-fh[idx_fh_F_ord5(iF,jF-3,kF,ex)]+F9*fh[idx_fh_F_ord5(iF,jF-2,kF,ex)]-F45*fh[idx_fh_F_ord5(iF,jF-1,kF,ex)]+F45*fh[idx_fh_F_ord5(iF,jF+1,kF,ex)]-F9*fh[idx_fh_F_ord5(iF,jF+2,kF,ex)]+fh[idx_fh_F_ord5(iF,jF+3,kF,ex)]);
-                        else if (j0<=ex2-3 && (j0-1)>=jminF)
-                            f_rhs[p] += sfy * d12dy*(fh[idx_fh_F_ord5(iF,jF-2,kF,ex)]-EIT*fh[idx_fh_F_ord5(iF,jF-1,kF,ex)]+EIT*fh[idx_fh_F_ord5(iF,jF+1,kF,ex)]-fh[idx_fh_F_ord5(iF,jF+2,kF,ex)]);
-                        else if (j0<=ex2-2 && j0>=jminF)
-                            f_rhs[p] += sfy * d2dy*(-fh[idx_fh_F_ord5(iF,jF-1,kF,ex)]+fh[idx_fh_F_ord5(iF,jF+1,kF,ex)]);
-                    } else if (sfy < ZEO) {
-                        if ((j0-4)>=jminF && j0<=ex2-4)
-                            f_rhs[p] -= sfy * d840dy*(-F5*fh[idx_fh_F_ord5(iF,jF+3,kF,ex)]+F60*fh[idx_fh_F_ord5(iF,jF+2,kF,ex)]-F420*fh[idx_fh_F_ord5(iF,jF+1,kF,ex)]-F378*fh[idx_fh_F_ord5(iF,jF,kF,ex)]+F1050*fh[idx_fh_F_ord5(iF,jF-1,kF,ex)]-F420*fh[idx_fh_F_ord5(iF,jF-2,kF,ex)]+F140*fh[idx_fh_F_ord5(iF,jF-3,kF,ex)]-F30*fh[idx_fh_F_ord5(iF,jF-4,kF,ex)]+F3*fh[idx_fh_F_ord5(iF,jF-5,kF,ex)]);
-                        else if ((j0-3)>=jminF && j0<=ex2-5)
-                            f_rhs[p] += sfy * d840dy*(+F3*fh[idx_fh_F_ord5(iF,jF-4,kF,ex)]-F32*fh[idx_fh_F_ord5(iF,jF-3,kF,ex)]+F168*fh[idx_fh_F_ord5(iF,jF-2,kF,ex)]-F672*fh[idx_fh_F_ord5(iF,jF-1,kF,ex)]+F672*fh[idx_fh_F_ord5(iF,jF+1,kF,ex)]-F168*fh[idx_fh_F_ord5(iF,jF+2,kF,ex)]+F32*fh[idx_fh_F_ord5(iF,jF+3,kF,ex)]-F3*fh[idx_fh_F_ord5(iF,jF+4,kF,ex)]);
-                        else if ((j0-2)>=jminF && j0<=ex2-4)
-                            f_rhs[p] += sfy * d60dy*(-fh[idx_fh_F_ord5(iF,jF-3,kF,ex)]+F9*fh[idx_fh_F_ord5(iF,jF-2,kF,ex)]-F45*fh[idx_fh_F_ord5(iF,jF-1,kF,ex)]+F45*fh[idx_fh_F_ord5(iF,jF+1,kF,ex)]-F9*fh[idx_fh_F_ord5(iF,jF+2,kF,ex)]+fh[idx_fh_F_ord5(iF,jF+3,kF,ex)]);
-                        else if ((j0-1)>=jminF && j0<=ex2-3)
-                            f_rhs[p] += sfy * d12dy*(fh[idx_fh_F_ord5(iF,jF-2,kF,ex)]-EIT*fh[idx_fh_F_ord5(iF,jF-1,kF,ex)]+EIT*fh[idx_fh_F_ord5(iF,jF+1,kF,ex)]-fh[idx_fh_F_ord5(iF,jF+2,kF,ex)]);
-                        else if (j0>=jminF && j0<=ex2-2)
-                            f_rhs[p] += sfy * d2dy*(-fh[idx_fh_F_ord5(iF,jF-1,kF,ex)]+fh[idx_fh_F_ord5(iF,jF+1,kF,ex)]);
-                    }
-
-                    const double sfz = Sfz[p];
-                    if (sfz > ZEO) {
-                        if (k0<=ex3-6 && (k0-2)>=kminF)
-                            f_rhs[p] += sfz * d840dz*(-F5*fh[idx_fh_F_ord5(iF,jF,kF-3,ex)]+F60*fh[idx_fh_F_ord5(iF,jF,kF-2,ex)]-F420*fh[idx_fh_F_ord5(iF,jF,kF-1,ex)]-F378*fh[idx_fh_F_ord5(iF,jF,kF,ex)]+F1050*fh[idx_fh_F_ord5(iF,jF,kF+1,ex)]-F420*fh[idx_fh_F_ord5(iF,jF,kF+2,ex)]+F140*fh[idx_fh_F_ord5(iF,jF,kF+3,ex)]-F30*fh[idx_fh_F_ord5(iF,jF,kF+4,ex)]+F3*fh[idx_fh_F_ord5(iF,jF,kF+5,ex)]);
-                        else if (k0<=ex3-5 && (k0-3)>=kminF)
-                            f_rhs[p] += sfz * d840dz*(+F3*fh[idx_fh_F_ord5(iF,jF,kF-4,ex)]-F32*fh[idx_fh_F_ord5(iF,jF,kF-3,ex)]+F168*fh[idx_fh_F_ord5(iF,jF,kF-2,ex)]-F672*fh[idx_fh_F_ord5(iF,jF,kF-1,ex)]+F672*fh[idx_fh_F_ord5(iF,jF,kF+1,ex)]-F168*fh[idx_fh_F_ord5(iF,jF,kF+2,ex)]+F32*fh[idx_fh_F_ord5(iF,jF,kF+3,ex)]-F3*fh[idx_fh_F_ord5(iF,jF,kF+4,ex)]);
-                        else if (k0<=ex3-4 && (k0-2)>=kminF)
-                            f_rhs[p] += sfz * d60dz*(-fh[idx_fh_F_ord5(iF,jF,kF-3,ex)]+F9*fh[idx_fh_F_ord5(iF,jF,kF-2,ex)]-F45*fh[idx_fh_F_ord5(iF,jF,kF-1,ex)]+F45*fh[idx_fh_F_ord5(iF,jF,kF+1,ex)]-F9*fh[idx_fh_F_ord5(iF,jF,kF+2,ex)]+fh[idx_fh_F_ord5(iF,jF,kF+3,ex)]);
-                        else if (k0<=ex3-3 && (k0-1)>=kminF)
-                            f_rhs[p] += sfz * d12dz*(fh[idx_fh_F_ord5(iF,jF,kF-2,ex)]-EIT*fh[idx_fh_F_ord5(iF,jF,kF-1,ex)]+EIT*fh[idx_fh_F_ord5(iF,jF,kF+1,ex)]-fh[idx_fh_F_ord5(iF,jF,kF+2,ex)]);
-                        else if (k0<=ex3-2 && k0>=kminF)
-                            f_rhs[p] += sfz * d2dz*(-fh[idx_fh_F_ord5(iF,jF,kF-1,ex)]+fh[idx_fh_F_ord5(iF,jF,kF+1,ex)]);
-                    } else if (sfz < ZEO) {
-                        if ((k0-4)>=kminF && k0<=ex3-4)
-                            f_rhs[p] -= sfz * d840dz*(-F5*fh[idx_fh_F_ord5(iF,jF,kF+3,ex)]+F60*fh[idx_fh_F_ord5(iF,jF,kF+2,ex)]-F420*fh[idx_fh_F_ord5(iF,jF,kF+1,ex)]-F378*fh[idx_fh_F_ord5(iF,jF,kF,ex)]+F1050*fh[idx_fh_F_ord5(iF,jF,kF-1,ex)]-F420*fh[idx_fh_F_ord5(iF,jF,kF-2,ex)]+F140*fh[idx_fh_F_ord5(iF,jF,kF-3,ex)]-F30*fh[idx_fh_F_ord5(iF,jF,kF-4,ex)]+F3*fh[idx_fh_F_ord5(iF,jF,kF-5,ex)]);
-                        else if ((k0-3)>=kminF && k0<=ex3-5)
-                            f_rhs[p] += sfz * d840dz*(+F3*fh[idx_fh_F_ord5(iF,jF,kF-4,ex)]-F32*fh[idx_fh_F_ord5(iF,jF,kF-3,ex)]+F168*fh[idx_fh_F_ord5(iF,jF,kF-2,ex)]-F672*fh[idx_fh_F_ord5(iF,jF,kF-1,ex)]+F672*fh[idx_fh_F_ord5(iF,jF,kF+1,ex)]-F168*fh[idx_fh_F_ord5(iF,jF,kF+2,ex)]+F32*fh[idx_fh_F_ord5(iF,jF,kF+3,ex)]-F3*fh[idx_fh_F_ord5(iF,jF,kF+4,ex)]);
-                        else if ((k0-2)>=kminF && k0<=ex3-4)
-                            f_rhs[p] += sfz * d60dz*(-fh[idx_fh_F_ord5(iF,jF,kF-3,ex)]+F9*fh[idx_fh_F_ord5(iF,jF,kF-2,ex)]-F45*fh[idx_fh_F_ord5(iF,jF,kF-1,ex)]+F45*fh[idx_fh_F_ord5(iF,jF,kF+1,ex)]-F9*fh[idx_fh_F_ord5(iF,jF,kF+2,ex)]+fh[idx_fh_F_ord5(iF,jF,kF+3,ex)]);
-                        else if ((k0-1)>=kminF && k0<=ex3-3)
-                            f_rhs[p] += sfz * d12dz*(fh[idx_fh_F_ord5(iF,jF,kF-2,ex)]-EIT*fh[idx_fh_F_ord5(iF,jF,kF-1,ex)]+EIT*fh[idx_fh_F_ord5(iF,jF,kF+1,ex)]-fh[idx_fh_F_ord5(iF,jF,kF+2,ex)]);
-                        else if (k0>=kminF && k0<=ex3-2)
-                            f_rhs[p] += sfz * d2dz*(-fh[idx_fh_F_ord5(iF,jF,kF-1,ex)]+fh[idx_fh_F_ord5(iF,jF,kF+1,ex)]);
-                    }
-                }
-            }
-        }
-        free(fh);
-        return;
-    }
-#else
-#error "lopsided_c.C: unsupported ghost_width (must be 2, 3, 4, or 5)"
-#endif
+    free(fh);
 }
+
+
+
+
+

AMSS_NCKU_source/lopsided_kodis_c.C

@@ -1,17 +1,8 @@
-#include "macrodef.h"
 #include "tool.h"
 
 /*
- * C 版 lopsided_kodis — combined upwind advection + KO dissipation.
- * Uses one shared symmetry_bd buffer (ord = ghost_width for both components)
- * where a stable merged stencil is available. The 8th-order path delegates to
- * the separate lopsided + kodis kernels, matching the original Fortran flow.
- *
- * FD order selection via ghost_width:
- *   2 → 2nd-order advection + r=2 KO (cof=16, sign=-)
- *   3 → 4th-order advection + r=3 KO (cof=64, sign=+)
- *   4 → 6th-order advection + r=4 KO (cof=256, sign=-)
- *   5 → 8th-order advection + r=5 KO (cof=1024, sign=+)
+ * 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,
@@ -19,286 +10,239 @@ void lopsided_kodis(const int ex[3],
                     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;
-    const double TWO = 2.0, F6 = 6.0, EIT = 8.0;
-    const double F3 = 3.0, F4 = 4.0, F5 = 5.0, F10 = 10.0, F12 = 12.0, F18 = 18.0;
-    const double F9 = 9.0, F45 = 45.0, F60 = 60.0;
-    const double F2 = 2.0, F15 = 15.0, F24 = 24.0, F30 = 30.0, F35 = 35.0;
-    const double F50 = 50.0, F77 = 77.0, F80 = 80.0, F100 = 100.0, F150 = 150.0;
-    const double F32 = 32.0, F168 = 168.0, F672 = 672.0, F840 = 840.0;
-    const double F140=140.0, F378=378.0, F420=420.0, F1050=1050.0;
+    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 int imaxF = ex1, jmaxF = ex2, kmaxF = ex3;
+    const double d12dx = ONE / F12 / dX;
+    const double d12dy = ONE / F12 / dY;
+    const double d12dz = ONE / F12 / dZ;
 
-#if (ghost_width == 2)
-    {
-        const int ord = 2;
-        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 int imaxF = ex1;
+    const int jmaxF = ex2;
+    const int kmaxF = ex3;
 
-        const size_t nx = (size_t)ex1 + ord;
-        const size_t ny = (size_t)ex2 + ord;
-        const size_t nz = (size_t)ex3 + ord;
-        double *fh = (double*)malloc(nx*ny*nz*sizeof(double));
-        if (!fh) return;
-        symmetry_bd(ord, ex, f, fh, SoA);
+    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;
 
-        const double d2dx = ONE/TWO/dX, d2dy = ONE/TWO/dY, d2dz = ONE/TWO/dZ;
+    // 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;
 
-        /* ---- advection (2nd-order) ---- */
-        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);
+    double *fh = (double*)malloc(fh_size * sizeof(double));
+    if (!fh) return;
 
-                    const double sfx = Sfx[p];
-                    if (sfx > ZEO) {
-                        if (i0<=ex1-3) f_rhs[p] += sfx*d2dx*(-F3*fh[idx_fh_F_ord2(iF,jF,kF,ex)]+F4*fh[idx_fh_F_ord2(iF+1,jF,kF,ex)]-fh[idx_fh_F_ord2(iF+2,jF,kF,ex)]);
-                        else if (i0<=ex1-2) f_rhs[p] += sfx*d2dx*(-fh[idx_fh_F_ord2(iF,jF,kF,ex)]+fh[idx_fh_F_ord2(iF+1,jF,kF,ex)]);
-                    } else if (sfx < ZEO) {
-                        if ((i0-1)>=iminF) f_rhs[p] -= sfx*d2dx*(-F3*fh[idx_fh_F_ord2(iF,jF,kF,ex)]+F4*fh[idx_fh_F_ord2(iF-1,jF,kF,ex)]-fh[idx_fh_F_ord2(iF-2,jF,kF,ex)]);
-                        else if (i0>=iminF) f_rhs[p] -= sfx*d2dx*(-fh[idx_fh_F_ord2(iF,jF,kF,ex)]+fh[idx_fh_F_ord2(iF-1,jF,kF,ex)]);
+    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)]);
                     }
-                    const double sfy = Sfy[p];
-                    if (sfy > ZEO) {
-                        if (j0<=ex2-3) f_rhs[p] += sfy*d2dy*(-F3*fh[idx_fh_F_ord2(iF,jF,kF,ex)]+F4*fh[idx_fh_F_ord2(iF,jF+1,kF,ex)]-fh[idx_fh_F_ord2(iF,jF+2,kF,ex)]);
-                        else if (j0<=ex2-2) f_rhs[p] += sfy*d2dy*(-fh[idx_fh_F_ord2(iF,jF,kF,ex)]+fh[idx_fh_F_ord2(iF,jF+1,kF,ex)]);
-                    } else if (sfy < ZEO) {
-                        if ((j0-1)>=jminF) f_rhs[p] -= sfy*d2dy*(-F3*fh[idx_fh_F_ord2(iF,jF,kF,ex)]+F4*fh[idx_fh_F_ord2(iF,jF-1,kF,ex)]-fh[idx_fh_F_ord2(iF,jF-2,kF,ex)]);
-                        else if (j0>=jminF) f_rhs[p] -= sfy*d2dy*(-fh[idx_fh_F_ord2(iF,jF,kF,ex)]+fh[idx_fh_F_ord2(iF,jF-1,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 sfz = Sfz[p];
-                    if (sfz > ZEO) {
-                        if (k0<=ex3-3) f_rhs[p] += sfz*d2dz*(-F3*fh[idx_fh_F_ord2(iF,jF,kF,ex)]+F4*fh[idx_fh_F_ord2(iF,jF,kF+1,ex)]-fh[idx_fh_F_ord2(iF,jF,kF+2,ex)]);
-                        else if (k0<=ex3-2) f_rhs[p] += sfz*d2dz*(-fh[idx_fh_F_ord2(iF,jF,kF,ex)]+fh[idx_fh_F_ord2(iF,jF,kF+1,ex)]);
-                    } else if (sfz < ZEO) {
-                        if ((k0-1)>=kminF) f_rhs[p] -= sfz*d2dz*(-F3*fh[idx_fh_F_ord2(iF,jF,kF,ex)]+F4*fh[idx_fh_F_ord2(iF,jF,kF-1,ex)]-fh[idx_fh_F_ord2(iF,jF,kF-2,ex)]);
-                        else if (k0>=kminF) f_rhs[p] -= sfz*d2dz*(-fh[idx_fh_F_ord2(iF,jF,kF,ex)]+fh[idx_fh_F_ord2(iF,jF,kF-1,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 (r=2, cof=16, sign=-) ---- */
-        if (eps > ZEO) {
-            const double cof = 16.0;
-            const double F4k = 4.0, F6k = 6.0;
-            const int i0_lo = (iminF+1>0)?iminF+1:0, j0_lo=(jminF+1>0)?jminF+1:0, k0_lo=(kminF+1>0)?kminF+1:0;
-            const int i0_hi=imaxF-3, j0_hi=jmaxF-3, k0_hi=kmaxF-3;
-            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=((fh[idx_fh_F_ord2(iF-2,jF,kF,ex)]+fh[idx_fh_F_ord2(iF+2,jF,kF,ex)])-F4k*(fh[idx_fh_F_ord2(iF-1,jF,kF,ex)]+fh[idx_fh_F_ord2(iF+1,jF,kF,ex)])+F6k*fh[idx_fh_F_ord2(iF,jF,kF,ex)])/dX;
-                    const double Dy=((fh[idx_fh_F_ord2(iF,jF-2,kF,ex)]+fh[idx_fh_F_ord2(iF,jF+2,kF,ex)])-F4k*(fh[idx_fh_F_ord2(iF,jF-1,kF,ex)]+fh[idx_fh_F_ord2(iF,jF+1,kF,ex)])+F6k*fh[idx_fh_F_ord2(iF,jF,kF,ex)])/dY;
-                    const double Dz=((fh[idx_fh_F_ord2(iF,jF,kF-2,ex)]+fh[idx_fh_F_ord2(iF,jF,kF+2,ex)])-F4k*(fh[idx_fh_F_ord2(iF,jF,kF-1,ex)]+fh[idx_fh_F_ord2(iF,jF,kF+1,ex)])+F6k*fh[idx_fh_F_ord2(iF,jF,kF,ex)])/dZ;
-                    f_rhs[p] -= (eps/cof)*(Dx+Dy+Dz);
-                }}}
-            }
-        }
-        free(fh);
-        return;
     }
-#elif (ghost_width == 3)
-    /* ---- 4th-order advection + r=3 KO (original code) ----------------- */
-    {
-        const int ord = 3;
-        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;
 
-        const size_t nx = (size_t)ex1 + ord;
-        const size_t ny = (size_t)ex2 + ord;
-        const size_t nz = (size_t)ex3 + ord;
-        double *fh = (double*)malloc(nx*ny*nz*sizeof(double));
-        if (!fh) return;
-        symmetry_bd(ord, ex, f, fh, SoA);
+    // 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;
 
-        const double d12dx = ONE/F12/dX, d12dy = ONE/F12/dY, d12dz = ONE/F12/dZ;
+        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);
 
-        /* ---- advection ---- */
-        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 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 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)]);
+                        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);
                     }
                 }
             }
         }
-
-        /* ---- KO dissipation (r=3, cof=64, sign=+) ---- */
-        if (eps > ZEO) {
-            const double cof = 64.0;
-            const double SIX = 6.0, FIT = 15.0, TWT = 20.0;
-            const int i0_lo=(iminF+2>0)?iminF+2:0, j0_lo=(jminF+2>0)?jminF+2:0, k0_lo=(kminF+2>0)?kminF+2:0;
-            const int i0_hi=imaxF-4, j0_hi=jmaxF-4, 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=((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=((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=((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+Dy+Dz);
-                }}}
-            }
-        }
-        free(fh);
-        return;
     }
-#elif (ghost_width == 4)
-    {
-        const int ord = 4;
-        int iminF = 1, jminF = 1, kminF = 1;
-        if (Symmetry > NO_SYMM && fabs(Z[0]) < dZ) kminF = -3;
-        if (Symmetry > EQ_SYMM && fabs(X[0]) < dX) iminF = -3;
-        if (Symmetry > EQ_SYMM && fabs(Y[0]) < dY) jminF = -3;
 
-        const size_t nx = (size_t)ex1 + ord;
-        const size_t ny = (size_t)ex2 + ord;
-        const size_t nz = (size_t)ex3 + ord;
-        double *fh = (double*)malloc(nx*ny*nz*sizeof(double));
-        if (!fh) return;
-        symmetry_bd(ord, ex, f, fh, SoA);
-
-        const double d60dx=ONE/F60/dX, d60dy=ONE/F60/dY, d60dz=ONE/F60/dZ;
-        const double d12dx=ONE/F12/dX, d12dy=ONE/F12/dY, d12dz=ONE/F12/dZ;
-        const double d2dx=ONE/TWO/dX, d2dy=ONE/TWO/dY, d2dz=ONE/TWO/dZ;
-
-        /* ---- advection (6th-order lopsided) ---- */
-        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 */
-            const double sfx=Sfx[p];
-            if (sfx>ZEO) {
-                if (i0<=ex1-5&&(i0-1)>=iminF) f_rhs[p]+=sfx*d60dx*(+F2*fh[idx_fh_F_ord4(iF-2,jF,kF,ex)]-F24*fh[idx_fh_F_ord4(iF-1,jF,kF,ex)]-F35*fh[idx_fh_F_ord4(iF,jF,kF,ex)]+F80*fh[idx_fh_F_ord4(iF+1,jF,kF,ex)]-F30*fh[idx_fh_F_ord4(iF+2,jF,kF,ex)]+EIT*fh[idx_fh_F_ord4(iF+3,jF,kF,ex)]-fh[idx_fh_F_ord4(iF+4,jF,kF,ex)]);
-                else if (i0<=ex1-6&&i0>=iminF) f_rhs[p]+=sfx*d60dx*(-F10*fh[idx_fh_F_ord4(iF-1,jF,kF,ex)]-F77*fh[idx_fh_F_ord4(iF,jF,kF,ex)]+F150*fh[idx_fh_F_ord4(iF+1,jF,kF,ex)]-F100*fh[idx_fh_F_ord4(iF+2,jF,kF,ex)]+F50*fh[idx_fh_F_ord4(iF+3,jF,kF,ex)]-F15*fh[idx_fh_F_ord4(iF+4,jF,kF,ex)]+F2*fh[idx_fh_F_ord4(iF+5,jF,kF,ex)]);
-                else if (i0<=ex1-4&&(i0-2)>=iminF) f_rhs[p]+=sfx*d60dx*(-fh[idx_fh_F_ord4(iF-3,jF,kF,ex)]+F9*fh[idx_fh_F_ord4(iF-2,jF,kF,ex)]-F45*fh[idx_fh_F_ord4(iF-1,jF,kF,ex)]+F45*fh[idx_fh_F_ord4(iF+1,jF,kF,ex)]-F9*fh[idx_fh_F_ord4(iF+2,jF,kF,ex)]+fh[idx_fh_F_ord4(iF+3,jF,kF,ex)]);
-                else if (i0<=ex1-3&&(i0-1)>=iminF) f_rhs[p]+=sfx*d12dx*(fh[idx_fh_F_ord4(iF-2,jF,kF,ex)]-EIT*fh[idx_fh_F_ord4(iF-1,jF,kF,ex)]+EIT*fh[idx_fh_F_ord4(iF+1,jF,kF,ex)]-fh[idx_fh_F_ord4(iF+2,jF,kF,ex)]);
-                else if (i0<=ex1-2&&i0>=iminF) f_rhs[p]+=sfx*d2dx*(-fh[idx_fh_F_ord4(iF-1,jF,kF,ex)]+fh[idx_fh_F_ord4(iF+1,jF,kF,ex)]);
-            } else if (sfx<ZEO) {
-                if ((i0-3)>=iminF&&i0<=ex1-3) f_rhs[p]-=sfx*d60dx*(+F2*fh[idx_fh_F_ord4(iF+2,jF,kF,ex)]-F24*fh[idx_fh_F_ord4(iF+1,jF,kF,ex)]-F35*fh[idx_fh_F_ord4(iF,jF,kF,ex)]+F80*fh[idx_fh_F_ord4(iF-1,jF,kF,ex)]-F30*fh[idx_fh_F_ord4(iF-2,jF,kF,ex)]+EIT*fh[idx_fh_F_ord4(iF-3,jF,kF,ex)]-fh[idx_fh_F_ord4(iF-4,jF,kF,ex)]);
-                else if ((i0-4)>=iminF&&i0<=ex1-2) f_rhs[p]-=sfx*d60dx*(-F10*fh[idx_fh_F_ord4(iF+1,jF,kF,ex)]-F77*fh[idx_fh_F_ord4(iF,jF,kF,ex)]+F150*fh[idx_fh_F_ord4(iF-1,jF,kF,ex)]-F100*fh[idx_fh_F_ord4(iF-2,jF,kF,ex)]+F50*fh[idx_fh_F_ord4(iF-3,jF,kF,ex)]-F15*fh[idx_fh_F_ord4(iF-4,jF,kF,ex)]+F2*fh[idx_fh_F_ord4(iF-5,jF,kF,ex)]);
-                else if ((i0-2)>=iminF&&i0<=ex1-4) f_rhs[p]+=sfx*d60dx*(-fh[idx_fh_F_ord4(iF-3,jF,kF,ex)]+F9*fh[idx_fh_F_ord4(iF-2,jF,kF,ex)]-F45*fh[idx_fh_F_ord4(iF-1,jF,kF,ex)]+F45*fh[idx_fh_F_ord4(iF+1,jF,kF,ex)]-F9*fh[idx_fh_F_ord4(iF+2,jF,kF,ex)]+fh[idx_fh_F_ord4(iF+3,jF,kF,ex)]);
-                else if ((i0-1)>=iminF&&i0<=ex1-3) f_rhs[p]+=sfx*d12dx*(fh[idx_fh_F_ord4(iF-2,jF,kF,ex)]-EIT*fh[idx_fh_F_ord4(iF-1,jF,kF,ex)]+EIT*fh[idx_fh_F_ord4(iF+1,jF,kF,ex)]-fh[idx_fh_F_ord4(iF+2,jF,kF,ex)]);
-                else if (i0>=iminF&&i0<=ex1-2) f_rhs[p]+=sfx*d2dx*(-fh[idx_fh_F_ord4(iF-1,jF,kF,ex)]+fh[idx_fh_F_ord4(iF+1,jF,kF,ex)]);
-            }
-            /* y */
-            const double sfy=Sfy[p];
-            if (sfy>ZEO) {
-                if (j0<=ex2-5&&(j0-1)>=jminF) f_rhs[p]+=sfy*d60dy*(F2*fh[idx_fh_F_ord4(iF,jF-2,kF,ex)]-F24*fh[idx_fh_F_ord4(iF,jF-1,kF,ex)]-F35*fh[idx_fh_F_ord4(iF,jF,kF,ex)]+F80*fh[idx_fh_F_ord4(iF,jF+1,kF,ex)]-F30*fh[idx_fh_F_ord4(iF,jF+2,kF,ex)]+EIT*fh[idx_fh_F_ord4(iF,jF+3,kF,ex)]-fh[idx_fh_F_ord4(iF,jF+4,kF,ex)]);
-                else if (j0<=ex2-6&&j0>=jminF) f_rhs[p]+=sfy*d60dy*(-F10*fh[idx_fh_F_ord4(iF,jF-1,kF,ex)]-F77*fh[idx_fh_F_ord4(iF,jF,kF,ex)]+F150*fh[idx_fh_F_ord4(iF,jF+1,kF,ex)]-F100*fh[idx_fh_F_ord4(iF,jF+2,kF,ex)]+F50*fh[idx_fh_F_ord4(iF,jF+3,kF,ex)]-F15*fh[idx_fh_F_ord4(iF,jF+4,kF,ex)]+F2*fh[idx_fh_F_ord4(iF,jF+5,kF,ex)]);
-                else if (j0<=ex2-4&&(j0-2)>=jminF) f_rhs[p]+=sfy*d60dy*(-fh[idx_fh_F_ord4(iF,jF-3,kF,ex)]+F9*fh[idx_fh_F_ord4(iF,jF-2,kF,ex)]-F45*fh[idx_fh_F_ord4(iF,jF-1,kF,ex)]+F45*fh[idx_fh_F_ord4(iF,jF+1,kF,ex)]-F9*fh[idx_fh_F_ord4(iF,jF+2,kF,ex)]+fh[idx_fh_F_ord4(iF,jF+3,kF,ex)]);
-                else if (j0<=ex2-3&&(j0-1)>=jminF) f_rhs[p]+=sfy*d12dy*(fh[idx_fh_F_ord4(iF,jF-2,kF,ex)]-EIT*fh[idx_fh_F_ord4(iF,jF-1,kF,ex)]+EIT*fh[idx_fh_F_ord4(iF,jF+1,kF,ex)]-fh[idx_fh_F_ord4(iF,jF+2,kF,ex)]);
-                else if (j0<=ex2-2&&j0>=jminF) f_rhs[p]+=sfy*d2dy*(-fh[idx_fh_F_ord4(iF,jF-1,kF,ex)]+fh[idx_fh_F_ord4(iF,jF+1,kF,ex)]);
-            } else if (sfy<ZEO) {
-                if ((j0-3)>=jminF&&j0<=ex2-3) f_rhs[p]-=sfy*d60dy*(F2*fh[idx_fh_F_ord4(iF,jF+2,kF,ex)]-F24*fh[idx_fh_F_ord4(iF,jF+1,kF,ex)]-F35*fh[idx_fh_F_ord4(iF,jF,kF,ex)]+F80*fh[idx_fh_F_ord4(iF,jF-1,kF,ex)]-F30*fh[idx_fh_F_ord4(iF,jF-2,kF,ex)]+EIT*fh[idx_fh_F_ord4(iF,jF-3,kF,ex)]-fh[idx_fh_F_ord4(iF,jF-4,kF,ex)]);
-                else if ((j0-4)>=jminF&&j0<=ex2-2) f_rhs[p]-=sfy*d60dy*(-F10*fh[idx_fh_F_ord4(iF,jF+1,kF,ex)]-F77*fh[idx_fh_F_ord4(iF,jF,kF,ex)]+F150*fh[idx_fh_F_ord4(iF,jF-1,kF,ex)]-F100*fh[idx_fh_F_ord4(iF,jF-2,kF,ex)]+F50*fh[idx_fh_F_ord4(iF,jF-3,kF,ex)]-F15*fh[idx_fh_F_ord4(iF,jF-4,kF,ex)]+F2*fh[idx_fh_F_ord4(iF,jF-5,kF,ex)]);
-                else if ((j0-2)>=jminF&&j0<=ex2-4) f_rhs[p]+=sfy*d60dy*(-fh[idx_fh_F_ord4(iF,jF-3,kF,ex)]+F9*fh[idx_fh_F_ord4(iF,jF-2,kF,ex)]-F45*fh[idx_fh_F_ord4(iF,jF-1,kF,ex)]+F45*fh[idx_fh_F_ord4(iF,jF+1,kF,ex)]-F9*fh[idx_fh_F_ord4(iF,jF+2,kF,ex)]+fh[idx_fh_F_ord4(iF,jF+3,kF,ex)]);
-                else if ((j0-1)>=jminF&&j0<=ex2-3) f_rhs[p]+=sfy*d12dy*(fh[idx_fh_F_ord4(iF,jF-2,kF,ex)]-EIT*fh[idx_fh_F_ord4(iF,jF-1,kF,ex)]+EIT*fh[idx_fh_F_ord4(iF,jF+1,kF,ex)]-fh[idx_fh_F_ord4(iF,jF+2,kF,ex)]);
-                else if (j0>=jminF&&j0<=ex2-2) f_rhs[p]+=sfy*d2dy*(-fh[idx_fh_F_ord4(iF,jF-1,kF,ex)]+fh[idx_fh_F_ord4(iF,jF+1,kF,ex)]);
-            }
-            /* z */
-            const double sfz=Sfz[p];
-            if (sfz>ZEO) {
-                if (k0<=ex3-5&&(k0-1)>=kminF) f_rhs[p]+=sfz*d60dz*(F2*fh[idx_fh_F_ord4(iF,jF,kF-2,ex)]-F24*fh[idx_fh_F_ord4(iF,jF,kF-1,ex)]-F35*fh[idx_fh_F_ord4(iF,jF,kF,ex)]+F80*fh[idx_fh_F_ord4(iF,jF,kF+1,ex)]-F30*fh[idx_fh_F_ord4(iF,jF,kF+2,ex)]+EIT*fh[idx_fh_F_ord4(iF,jF,kF+3,ex)]-fh[idx_fh_F_ord4(iF,jF,kF+4,ex)]);
-                else if (k0<=ex3-6&&k0>=kminF) f_rhs[p]+=sfz*d60dz*(-F10*fh[idx_fh_F_ord4(iF,jF,kF-1,ex)]-F77*fh[idx_fh_F_ord4(iF,jF,kF,ex)]+F150*fh[idx_fh_F_ord4(iF,jF,kF+1,ex)]-F100*fh[idx_fh_F_ord4(iF,jF,kF+2,ex)]+F50*fh[idx_fh_F_ord4(iF,jF,kF+3,ex)]-F15*fh[idx_fh_F_ord4(iF,jF,kF+4,ex)]+F2*fh[idx_fh_F_ord4(iF,jF,kF+5,ex)]);
-                else if (k0<=ex3-4&&(k0-2)>=kminF) f_rhs[p]+=sfz*d60dz*(-fh[idx_fh_F_ord4(iF,jF,kF-3,ex)]+F9*fh[idx_fh_F_ord4(iF,jF,kF-2,ex)]-F45*fh[idx_fh_F_ord4(iF,jF,kF-1,ex)]+F45*fh[idx_fh_F_ord4(iF,jF,kF+1,ex)]-F9*fh[idx_fh_F_ord4(iF,jF,kF+2,ex)]+fh[idx_fh_F_ord4(iF,jF,kF+3,ex)]);
-                else if (k0<=ex3-3&&(k0-1)>=kminF) f_rhs[p]+=sfz*d12dz*(fh[idx_fh_F_ord4(iF,jF,kF-2,ex)]-EIT*fh[idx_fh_F_ord4(iF,jF,kF-1,ex)]+EIT*fh[idx_fh_F_ord4(iF,jF,kF+1,ex)]-fh[idx_fh_F_ord4(iF,jF,kF+2,ex)]);
-                else if (k0<=ex3-2&&k0>=kminF) f_rhs[p]+=sfz*d2dz*(-fh[idx_fh_F_ord4(iF,jF,kF-1,ex)]+fh[idx_fh_F_ord4(iF,jF,kF+1,ex)]);
-            } else if (sfz<ZEO) {
-                if ((k0-3)>=kminF&&k0<=ex3-3) f_rhs[p]-=sfz*d60dz*(F2*fh[idx_fh_F_ord4(iF,jF,kF+2,ex)]-F24*fh[idx_fh_F_ord4(iF,jF,kF+1,ex)]-F35*fh[idx_fh_F_ord4(iF,jF,kF,ex)]+F80*fh[idx_fh_F_ord4(iF,jF,kF-1,ex)]-F30*fh[idx_fh_F_ord4(iF,jF,kF-2,ex)]+EIT*fh[idx_fh_F_ord4(iF,jF,kF-3,ex)]-fh[idx_fh_F_ord4(iF,jF,kF-4,ex)]);
-                else if ((k0-4)>=kminF&&k0<=ex3-2) f_rhs[p]-=sfz*d60dz*(-F10*fh[idx_fh_F_ord4(iF,jF,kF+1,ex)]-F77*fh[idx_fh_F_ord4(iF,jF,kF,ex)]+F150*fh[idx_fh_F_ord4(iF,jF,kF-1,ex)]-F100*fh[idx_fh_F_ord4(iF,jF,kF-2,ex)]+F50*fh[idx_fh_F_ord4(iF,jF,kF-3,ex)]-F15*fh[idx_fh_F_ord4(iF,jF,kF-4,ex)]+F2*fh[idx_fh_F_ord4(iF,jF,kF-5,ex)]);
-                else if ((k0-2)>=kminF&&k0<=ex3-4) f_rhs[p]+=sfz*d60dz*(-fh[idx_fh_F_ord4(iF,jF,kF-3,ex)]+F9*fh[idx_fh_F_ord4(iF,jF,kF-2,ex)]-F45*fh[idx_fh_F_ord4(iF,jF,kF-1,ex)]+F45*fh[idx_fh_F_ord4(iF,jF,kF+1,ex)]-F9*fh[idx_fh_F_ord4(iF,jF,kF+2,ex)]+fh[idx_fh_F_ord4(iF,jF,kF+3,ex)]);
-                else if ((k0-1)>=kminF&&k0<=ex3-3) f_rhs[p]+=sfz*d12dz*(fh[idx_fh_F_ord4(iF,jF,kF-2,ex)]-EIT*fh[idx_fh_F_ord4(iF,jF,kF-1,ex)]+EIT*fh[idx_fh_F_ord4(iF,jF,kF+1,ex)]-fh[idx_fh_F_ord4(iF,jF,kF+2,ex)]);
-                else if (k0>=kminF&&k0<=ex3-2) f_rhs[p]+=sfz*d2dz*(-fh[idx_fh_F_ord4(iF,jF,kF-1,ex)]+fh[idx_fh_F_ord4(iF,jF,kF+1,ex)]);
-            }
-        }}}
-
-        /* ---- KO dissipation (r=4, cof=256, sign=-) ---- */
-        if (eps > ZEO) {
-            const double cof = 256.0;
-            const double F8k = 8.0, F28 = 28.0, F56 = 56.0, F70 = 70.0;
-            const int i0_lo=(iminF+3>0)?iminF+3:0, j0_lo=(jminF+3>0)?jminF+3:0, k0_lo=(kminF+3>0)?kminF+3:0;
-            const int i0_hi=imaxF-5, j0_hi=jmaxF-5, k0_hi=kmaxF-5;
-            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=((fh[idx_fh_F_ord4(iF-4,jF,kF,ex)]+fh[idx_fh_F_ord4(iF+4,jF,kF,ex)])-F8k*(fh[idx_fh_F_ord4(iF-3,jF,kF,ex)]+fh[idx_fh_F_ord4(iF+3,jF,kF,ex)])+F28*(fh[idx_fh_F_ord4(iF-2,jF,kF,ex)]+fh[idx_fh_F_ord4(iF+2,jF,kF,ex)])-F56*(fh[idx_fh_F_ord4(iF-1,jF,kF,ex)]+fh[idx_fh_F_ord4(iF+1,jF,kF,ex)])+F70*fh[idx_fh_F_ord4(iF,jF,kF,ex)])/dX;
-                    const double Dy=((fh[idx_fh_F_ord4(iF,jF-4,kF,ex)]+fh[idx_fh_F_ord4(iF,jF+4,kF,ex)])-F8k*(fh[idx_fh_F_ord4(iF,jF-3,kF,ex)]+fh[idx_fh_F_ord4(iF,jF+3,kF,ex)])+F28*(fh[idx_fh_F_ord4(iF,jF-2,kF,ex)]+fh[idx_fh_F_ord4(iF,jF+2,kF,ex)])-F56*(fh[idx_fh_F_ord4(iF,jF-1,kF,ex)]+fh[idx_fh_F_ord4(iF,jF+1,kF,ex)])+F70*fh[idx_fh_F_ord4(iF,jF,kF,ex)])/dY;
-                    const double Dz=((fh[idx_fh_F_ord4(iF,jF,kF-4,ex)]+fh[idx_fh_F_ord4(iF,jF,kF+4,ex)])-F8k*(fh[idx_fh_F_ord4(iF,jF,kF-3,ex)]+fh[idx_fh_F_ord4(iF,jF,kF+3,ex)])+F28*(fh[idx_fh_F_ord4(iF,jF,kF-2,ex)]+fh[idx_fh_F_ord4(iF,jF,kF+2,ex)])-F56*(fh[idx_fh_F_ord4(iF,jF,kF-1,ex)]+fh[idx_fh_F_ord4(iF,jF,kF+1,ex)])+F70*fh[idx_fh_F_ord4(iF,jF,kF,ex)])/dZ;
-                    f_rhs[p] -= (eps/cof)*(Dx+Dy+Dz);
-                }}}
-            }
-        }
-        free(fh);
-        return;
-    }
-#elif (ghost_width == 5)
-    {
-        lopsided(ex, X, Y, Z, f, f_rhs, Sfx, Sfy, Sfz, Symmetry, SoA);
-        if (eps > ZEO) kodis(ex, X, Y, Z, f, f_rhs, SoA, Symmetry, eps);
-        return;
-    }
-#else
-#error "lopsided_kodis_c.C: unsupported ghost_width (must be 2, 3, 4, or 5)"
-#endif
+    free(fh);
 }

AMSS_NCKU_source/makefile

@@ -1,293 +1,270 @@
+
+
+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 ($(TOOLCHAIN),intel)
+OMP_FLAG = -qopenmp
+
+ifeq ($(PGO_MODE),instrument)
+## Intel Phase 1: instrumentation — omit -ipo/-fp-model fast=2 for faster build and numerical stability
+CXXAPPFLAGS = -O3 -march=znver5  -fma -fprofile-instr-generate -ipo \
+              -Dfortran3 -Dnewc $(MKL_INC) $(INTERP_LB_FLAGS)
+f90appflags = -O3 -march=znver5  -fma -fprofile-instr-generate -ipo \
+              -align array64byte -fpp $(MKL_INC) $(POLINT6_FLAG)
+else
+## opt (default): maximum performance with PGO profile data -fprofile-instr-use=$(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 -march=znver5  -fp-model fast=2 -fma -ipo \
+              -Dfortran3 -Dnewc $(MKL_INC) $(INTERP_LB_FLAGS)
+f90appflags = -O3 -march=znver5  -fp-model fast=2 -fma -ipo \
+              -align array64byte -fpp $(MKL_INC) $(POLINT6_FLAG)
+endif
+
+TP_OPTFLAGS = -O3 -march=znver5  -fp-model fast=2 -fma -ipo \
+              -Dfortran3 -Dnewc $(MKL_INC)
+else
+## NVHPC defaults: mpicc/mpicxx/mpifort wrappers
+## PGO_MODE is ignored in this branch.
+OMP_FLAG = -mp
+CXXAPPFLAGS = -O3 -march=znver5 -tp=host -Mcache_align -Mfma \
+              -Dfortran3 -Dnewc $(MKL_INC) $(INTERP_LB_FLAGS)
+f90appflags = -O3 -march=znver5 -tp=host -Mcache_align -Mfma -Mpreprocess \
+              $(MKL_INC) $(POLINT6_FLAG)
+TP_OPTFLAGS = -O3 -march=znver5 -tp=host -Mcache_align -Mfma \
+              -Dfortran3 -Dnewc $(MKL_INC)
+endif
+
+.SUFFIXES: .o .f90 .C .for .cu
+
+.f90.o:
+	$(f90) $(f90appflags) -c $< -o $@
+
+.C.o:
+	${CXX} $(CXXAPPFLAGS) -c $< $(filein) -o $@
+
+# ShellPatch.C uses OpenMP for setupintintstuff search loops
+ShellPatch.o: ShellPatch.C
+	${CXX} $(CXXAPPFLAGS) $(OMP_FLAG) -c $< $(filein) -o $@
+
+.for.o:
+	$(f77) -c $< -o $@
+
+.cu.o:
+	$(Cu) $(CUDA_APP_FLAGS) -c $< -o $@ $(CUDA_LIB_PATH)
+
+# CUDA rewrite of BSSN RHS (drop-in replacement for bssn_rhs_c + stencil helpers)
+bssn_rhs_cuda.o: bssn_rhs_cuda.cu bssn_rhs.h macrodef.h fd_cuda_helpers.cuh
+	$(Cu) $(CUDA_APP_FLAGS) -c $< -o $@ $(CUDA_LIB_PATH)
+
+# CUDA rewrite of BSSN Shell-Patch RHS (drop-in replacement for bssn_rhs_ss)
+bssn_gpu_rhs_ss.o: bssn_gpu_rhs_ss.cu bssn_gpu.h gpu_rhsSS_mem.h bssn_macro.h macrodef.fh
+	$(Cu) $(CUDA_APP_FLAGS) -c $< -o $@ $(CUDA_LIB_PATH)
+
+# CUDA rewrite of Z4C Cartesian RHS
+z4c_rhs_cuda.o: z4c_rhs_cuda.cu z4c_rhs_cuda.h bssn_rhs.h macrodef.h ricci_gamma.h fd_cuda_helpers.cuh
+	$(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 $@
+
+z4c_rhs_c.o: z4c_rhs_c.C
+	${CXX} $(CXXAPPFLAGS) -c $< $(filein) -o $@
+
+#interp_lb_profile.o: interp_lb_profile.C interp_lb_profile.h
+#	${CXX} $(CXXAPPFLAGS) -c $< $(filein) -o $@
+
+TwoPunctures.o: TwoPunctures.C
+	${CXX} $(TP_OPTFLAGS) $(OMP_FLAG) -c $< -o $@
+
+TwoPunctureABE.o: TwoPunctureABE.C
+	${CXX} $(TP_OPTFLAGS) $(OMP_FLAG) -c $< -o $@
+
+# Input files
+
+## CUDA BSSN RHS switch
+##   1 : use the rewritten CUDA bssn_rhs backend
+##   0 : keep the normal CPU/Fortran selection below
+USE_CUDA_BSSN ?= 0
+USE_CUDA_Z4C ?= 0
+AMSS_Z4C_MRBD ?= 0
 
-
-include makefile.inc
-
--include AMSS_NCKU_build.mk
-
-ABE_TYPE ?= $(shell awk '/^[[:space:]]*\#define[[:space:]]+ABEtype/ {print $$3; exit}' macrodef.h 2>/dev/null)
-
-ifeq ($(USE_TRANSFER_CACHE),auto)
-ifeq ($(ABE_TYPE),0)
-EFFECTIVE_USE_TRANSFER_CACHE = 1
-else
-EFFECTIVE_USE_TRANSFER_CACHE = 0
-endif
-else
-EFFECTIVE_USE_TRANSFER_CACHE = $(USE_TRANSFER_CACHE)
-endif
-
-ifeq ($(USE_CXX_ESCALAR_KERNEL),1)
-ifeq ($(ABE_TYPE),1)
-EFFECTIVE_USE_CXX_ESCALAR_KERNEL = 1
-else
-EFFECTIVE_USE_CXX_ESCALAR_KERNEL = 0
-endif
-else
-EFFECTIVE_USE_CXX_ESCALAR_KERNEL = 0
-endif
-
-ifeq ($(EFFECTIVE_USE_CXX_ESCALAR_KERNEL),1)
-ifeq ($(USE_CXX_KERNELS),0)
-$(error USE_CXX_ESCALAR_KERNEL=1 requires USE_CXX_KERNELS=1 because bssn_escalar_rhs_c.C reuses the C BSSN kernel)
-endif
-endif
-
-ifeq ($(USE_CXX_EM_KERNEL),1)
-ifeq ($(ABE_TYPE),3)
-EFFECTIVE_USE_CXX_EM_KERNEL = 1
-else
-EFFECTIVE_USE_CXX_EM_KERNEL = 0
-endif
-else
-EFFECTIVE_USE_CXX_EM_KERNEL = 0
-endif
-
-ifeq ($(EFFECTIVE_USE_CXX_EM_KERNEL),1)
-ifeq ($(USE_CXX_KERNELS),0)
-$(error USE_CXX_EM_KERNEL=1 requires USE_CXX_KERNELS=1 because bssn_em_rhs_c.C reuses the C BSSN kernel)
-endif
-endif
-
-EM_KERNEL_FLAG = -DBSSN_USE_EM_C_KERNEL=$(EFFECTIVE_USE_CXX_EM_KERNEL)
-
-## polint(ordn=6) kernel selector:
-##   1 (default): barycentric fast path
-##   0          : fallback to Neville path
-POLINT6_USE_BARY ?= 1
-POLINT6_FLAG = -DPOLINT6_USE_BARYCENTRIC=$(POLINT6_USE_BARY)
-TRANSFER_CACHE_FLAG = -DBSSN_USE_TRANSFER_CACHE=$(EFFECTIVE_USE_TRANSFER_CACHE)
-ESCALAR_KERNEL_FLAG = -DBSSN_USE_ESCALAR_C_KERNEL=$(EFFECTIVE_USE_CXX_ESCALAR_KERNEL)
-
-## ABE build flags selected by PGO_MODE (set in makefile.inc, default: opt)
-##   make                        -> opt  (PGO-guided, maximum performance)
-##   make PGO_MODE=instrument    -> instrument (Phase 1: collect fresh profile data)
-PROFDATA = /home/$(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) \
-              $(TRANSFER_CACHE_FLAG) $(ESCALAR_KERNEL_FLAG) $(EM_KERNEL_FLAG)
-f90appflags = -O3 -xHost -fma -fprofile-instr-generate -ipo \
-              -align array64byte -fpp -I${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) \
-              $(TRANSFER_CACHE_FLAG) $(ESCALAR_KERNEL_FLAG) $(EM_KERNEL_FLAG)
-f90appflags = -O3 -xHost -fp-model fast=2 -fma -ipo \
-              -align array64byte -fpp -I${MKLROOT}/include $(POLINT6_FLAG)
-endif
-
-.SUFFIXES: .o .f90 .C .for .cu
-
-.f90.o:
-	$(f90) $(f90appflags) -c $< -o $@
-
-.C.o:
-	${CXX} $(CXXAPPFLAGS) -c $< $(filein) -o $@
-
-# ShellPatch.C uses OpenMP for setupintintstuff search loops
-ShellPatch.o: ShellPatch.C
-	${CXX} $(CXXAPPFLAGS) $(OMP_FLAG) -c $< $(filein) -o $@
-
-.for.o:
-	$(f77) -c $< -o $@
-
-.cu.o:
-	$(Cu) $(CUDA_APP_FLAGS) -c $< -o $@ $(CUDA_LIB_PATH)
-
-# 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 $@
-
-# C rewrite of shell-patch derivative kernels
-fderivs_sh_c.o: fderivs_sh_c.C
-	${CXX} $(CXXAPPFLAGS) -c $< $(filein) -o $@
-
-fdderivs_sh_c.o: fdderivs_sh_c.C
-	${CXX} $(CXXAPPFLAGS) -c $< $(filein) -o $@
-
-fderivs_shc_c.o: fderivs_shc_c.C
-	${CXX} $(CXXAPPFLAGS) -c $< $(filein) -o $@
-
-fdderivs_shc_c.o: fdderivs_shc_c.C
-	${CXX} $(CXXAPPFLAGS) -c $< $(filein) -o $@
-
-kodiss_sh_c.o: kodiss_sh_c.C
-	${CXX} $(CXXAPPFLAGS) -c $< $(filein) -o $@
-
-
-bssn_em_rhs_c.o: bssn_em_rhs_c.C
-	${CXX} $(CXXAPPFLAGS) -c $< $(filein) -o $@
-z4c_rhs_c.o: z4c_rhs_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 $@
-
-TwoPunctureABE.o: TwoPunctureABE.C
-	${CXX} $(TP_OPTFLAGS) -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/bssn-escalar 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
-ifeq ($(EFFECTIVE_USE_CXX_ESCALAR_KERNEL),1)
-CFILES += bssn_escalar_rhs_c.o
-endif
-ifeq ($(EFFECTIVE_USE_CXX_EM_KERNEL),1)
-CFILES += bssn_em_rhs_c.o
-endif
-endif
-
-ifeq ($(USE_CXX_Z4C_KERNELS),1)
-CFILES += z4c_rhs_c.o
-Z4C_F90_OBJ =
-else
-Z4C_F90_OBJ = Z4c_rhs.o
-endif
-
-## RK4 kernel switch (independent from USE_CXX_KERNELS)
-ifeq ($(USE_CXX_RK4),1)
-CFILES += rungekutta4_rout_c.o
-RK4_F90_OBJ =
-else
-RK4_F90_OBJ = rungekutta4_rout.o
-endif
-
-## Shell-patch derivative kernel switch (independent from USE_CXX_KERNELS)
-##   1 : use C++ rewrite of shell derivative functions (experimental)
-##   0 : use original Fortran diff_new_sh.o and kodiss_sh.o (default)
-USE_CXX_SHELL_KERNELS ?= 0
-ifeq ($(USE_CXX_SHELL_KERNELS),1)
-CFILES += fderivs_sh_c.o fdderivs_sh_c.o fderivs_shc_c.o fdderivs_shc_c.o kodiss_sh_c.o
-SH_F90_OBJ =
-else
-SH_F90_OBJ = diff_new_sh.o kodiss_sh.o point_diff_new_sh.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
-	   
-C++FILES_GPU = ABE.o Ansorg.o Block.o misc.o monitor.o Parallel.o MPatch.o var.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_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\
-	   prolongrestrict_cell.o prolongrestrict_vertex.o\
-	   $(RK4_F90_OBJ) diff_new.o kodiss.o\
-	   lopsidediff.o sommerfeld_rout.o getnp4.o $(SH_F90_OBJ)\
-	   shellfunctions.o bssn_rhs_ss.o Set_Rho_ADM.o\
-           getnp4EScalar.o bssnEScalar_rhs.o bssn_constraint.o ricci_gamma.o\
-           fadmquantites_bssn.o $(Z4C_F90_OBJ) Z4c_rhs_ss.o\
-	   cpbc.o getnp4old.o NullEvol.o initial_null.o initial_maxwell.o\
-	   getnpem2.o empart.o NullNews.o fourdcurvature.o\
-	   bssn2adm.o adm_constraint.o adm_ricci_gamma.o\
-	   scalar_rhs.o initial_scalar.o NullEvol2.o initial_null2.o\
-	   NullNews2.o tool_f.o
-
-ifeq ($(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 = $(F90FILES_BASE)
-endif
-
-F77FILES = zbesh.o
-
-AHFDOBJS = expansion.o expansion_Jacobian.o patch.o coords.o patch_info.o patch_interp.o patch_system.o \
-tgrid.o fd_grid.o ghost_zone.o array.o round.o norm.o fuzzy.o error_exit.o miscfp.o \
-linear_map.o cpm_map.o BH_diagnostics.o setup.o horizon_sequence.o find_horizons.o \
-initial_guess.o Newton.o Jacobian.o ilucg.o IntPnts0.o IntPnts.o
-
-TwoPunctureFILES = TwoPunctureABE.o TwoPunctures.o 
-
-CUDAFILES = bssn_gpu.o bssn_gpu_rhs_ss.o
-
-# file dependences
-$(C++FILES) $(C++FILES_GPU) $(F90FILES) $(CFILES) $(AHFDOBJS) $(CUDAFILES): macrodef.fh
-
-$(C++FILES): Block.h enforce_algebra.h fmisc.h initial_puncture.h macrodef.h\
-	     misc.h monitor.h MyList.h Parallel.h MPatch.h prolongrestrict.h\
-	     rungekutta4_rout.h var.h bssn_class.h bssn_rhs.h sommerfeld_rout.h\
-	     cgh.h surface_integral.h ShellPatch.h shellfunctions.h perf.h\
-             fadmquantites_bssn.h cpbc.h getnp4.h initial_null.h NullEvol.h\
-	     NullShellPatch.h initial_maxwell.h bssnEM_class.h getnpem2.h\
-	     empart.h NullNews.h kodiss.h Parallel_bam.h ricci_gamma.h\
-             initial_null2.h NullShellPatch2.h 
-             
-$(C++FILES_GPU): Block.h enforce_algebra.h fmisc.h initial_puncture.h macrodef.h\
-	     misc.h monitor.h MyList.h Parallel.h MPatch.h prolongrestrict.h\
-	     rungekutta4_rout.h var.h bssn_rhs.h sommerfeld_rout.h\
-	     cgh.h surface_integral.h ShellPatch.h shellfunctions.h perf.h\
-             fadmquantites_bssn.h cpbc.h getnp4.h initial_null.h NullEvol.h\
-	     NullShellPatch.h initial_maxwell.h bssnEM_class.h getnpem2.h\
-	     empart.h NullNews.h kodiss.h Parallel_bam.h ricci_gamma.h\
-             initial_null2.h NullShellPatch2.h \
-             bssn_gpu_class.h bssn_macro.h
-             
-$(AHFDOBJS): cctk.h cctk_Config.h cctk_Types.h cctk_Constants.h myglobal.h
-
-$(C++FILES) $(C++FILES_GPU) $(CFILES) $(AHFDOBJS) $(CUDAFILES): macrodef.h
-
-TwoPunctureFILES: TwoPunctures.h
-
-$(CUDAFILES): bssn_gpu.h gpu_mem.h gpu_rhsSS_mem.h
-
-misc.o : zbesh.o
-
-# projects
-ABE: $(C++FILES) $(CFILES) $(F90FILES) $(F77FILES) $(AHFDOBJS)
-	$(CLINKER) $(CXXAPPFLAGS) -o $@ $(C++FILES) $(CFILES) $(F90FILES) $(F77FILES) $(AHFDOBJS) $(LDLIBS)
-	
-ABEGPU: $(C++FILES_GPU) $(CFILES) $(F90FILES) $(F77FILES) $(AHFDOBJS) $(CUDAFILES)
-	$(CLINKER) $(CXXAPPFLAGS) -o $@ $(C++FILES_GPU) $(CFILES) $(F90FILES) $(F77FILES) $(AHFDOBJS) $(CUDAFILES) $(LDLIBS)
-
-TwoPunctureABE: $(TwoPunctureFILES)
-	$(CLINKER) $(TP_OPTFLAGS) -qopenmp -o $@ $(TwoPunctureFILES) $(LDLIBS)
-
-clean:
-	rm *.o ABE ABEGPU TwoPunctureABE make.log -f
+CXXAPPFLAGS += -DUSE_CUDA_BSSN=$(USE_CUDA_BSSN)
+CUDA_APP_FLAGS += -DUSE_CUDA_BSSN=$(USE_CUDA_BSSN)
+CXXAPPFLAGS += -DUSE_CUDA_Z4C=$(USE_CUDA_Z4C)
+CUDA_APP_FLAGS += -DUSE_CUDA_Z4C=$(USE_CUDA_Z4C)
+CXXAPPFLAGS += -DAMSS_Z4C_MRBD=$(AMSS_Z4C_MRBD)
+CUDA_APP_FLAGS += -DAMSS_Z4C_MRBD=$(AMSS_Z4C_MRBD)
+
+## Kernel implementation switch (set USE_CXX_KERNELS=0 to fall back to Fortran)
+ifeq ($(USE_CXX_KERNELS),0)
+# Fortran mode: no C rewrite files; bssn_rhs.o is included via F90FILES below
+CFILES_CPU =
+else
+# C++ mode (default): C rewrite of bssn_rhs and helper kernels
+CFILES_CPU = bssn_rhs_c.o fderivs_c.o fdderivs_c.o kodiss_c.o lopsided_c.o lopsided_kodis_c.o
+endif
+
+CFILES_CUDA_BSSN = bssn_rhs_cuda.o bssn_gpu_rhs_ss.o
+
+ifeq ($(USE_CUDA_BSSN),1)
+CFILES = $(CFILES_CUDA_BSSN)
+else
+CFILES = $(CFILES_CPU)
+endif
+
+ifeq ($(USE_CUDA_Z4C),1)
+CFILES += z4c_rhs_cuda.o
+Z4C_F90_OBJ =
+else ifeq ($(USE_CXX_Z4C_KERNELS),1)
+CFILES += z4c_rhs_c.o
+Z4C_F90_OBJ =
+else
+Z4C_F90_OBJ = Z4c_rhs.o
+endif
+
+## RK4 kernel switch (independent from USE_CXX_KERNELS)
+ifeq ($(USE_CXX_RK4),1)
+RK4_C_OBJ = rungekutta4_rout_c.o
+RK4_F90_OBJ =
+else
+RK4_C_OBJ =
+RK4_F90_OBJ = rungekutta4_rout.o
+endif
+
+CFILES += $(RK4_C_OBJ)
+ABE_CUDA_CFILES = $(CFILES_CUDA_BSSN) z4c_rhs_cuda.o $(RK4_C_OBJ)
+
+ABE_LDLIBS = $(LDLIBS)
+ifeq ($(USE_CUDA_BSSN),1)
+ABE_LDLIBS += -lcudart $(CUDA_LIB_PATH)
+endif
+ifeq ($(USE_CUDA_Z4C),1)
+ABE_LDLIBS += -lcudart $(CUDA_LIB_PATH)
+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
+	   
+#C++FILES_GPU = ABE.o Ansorg.o Block.o misc.o monitor.o Parallel.o MPatch.o var.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_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\
+	   prolongrestrict_cell.o prolongrestrict_vertex.o\
+	   $(RK4_F90_OBJ) 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\
+           fadmquantites_bssn.o $(Z4C_F90_OBJ) Z4c_rhs_ss.o point_diff_new_sh.o\
+	   cpbc.o getnp4old.o NullEvol.o initial_null.o initial_maxwell.o\
+	   getnpem2.o empart.o NullNews.o fourdcurvature.o\
+	   bssn2adm.o adm_constraint.o adm_ricci_gamma.o\
+	   scalar_rhs.o initial_scalar.o NullEvol2.o initial_null2.o\
+	   NullNews2.o tool_f.o
+
+ifeq ($(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 = $(F90FILES_BASE)
+endif
+
+F77FILES = zbesh.o
+
+AHFDOBJS = expansion.o expansion_Jacobian.o patch.o coords.o patch_info.o patch_interp.o patch_system.o \
+tgrid.o fd_grid.o ghost_zone.o array.o round.o norm.o fuzzy.o error_exit.o miscfp.o \
+linear_map.o cpm_map.o BH_diagnostics.o setup.o horizon_sequence.o find_horizons.o \
+initial_guess.o Newton.o Jacobian.o ilucg.o IntPnts0.o IntPnts.o
+
+TwoPunctureFILES = TwoPunctureABE.o TwoPunctures.o 
+
+#CUDAFILES = bssn_gpu.o bssn_gpu_rhs_ss.o
+
+# file dependences
+$(C++FILES) $(C++FILES_GPU) $(F90FILES) $(CFILES) $(ABE_CUDA_CFILES) $(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\
+	     rungekutta4_rout.h var.h bssn_class.h bssn_rhs.h sommerfeld_rout.h\
+	     cgh.h surface_integral.h ShellPatch.h shellfunctions.h perf.h\
+             fadmquantites_bssn.h cpbc.h getnp4.h initial_null.h NullEvol.h\
+	     NullShellPatch.h initial_maxwell.h bssnEM_class.h getnpem2.h\
+	     empart.h NullNews.h kodiss.h Parallel_bam.h ricci_gamma.h\
+             initial_null2.h NullShellPatch2.h 
+             
+#$(C++FILES_GPU): Block.h enforce_algebra.h fmisc.h initial_puncture.h macrodef.h\
+	     misc.h monitor.h MyList.h Parallel.h MPatch.h prolongrestrict.h\
+	     rungekutta4_rout.h var.h bssn_rhs.h sommerfeld_rout.h\
+	     cgh.h surface_integral.h ShellPatch.h shellfunctions.h perf.h\
+             fadmquantites_bssn.h cpbc.h getnp4.h initial_null.h NullEvol.h\
+	     NullShellPatch.h initial_maxwell.h bssnEM_class.h getnpem2.h\
+	     empart.h NullNews.h kodiss.h Parallel_bam.h ricci_gamma.h\
+             initial_null2.h NullShellPatch2.h \
+             bssn_gpu_class.h bssn_macro.h
+             
+$(AHFDOBJS): cctk.h cctk_Config.h cctk_Types.h cctk_Constants.h myglobal.h
+
+$(C++FILES) $(C++FILES_GPU) $(CFILES) $(ABE_CUDA_CFILES) $(AHFDOBJS) $(CUDAFILES): macrodef.h
+
+TwoPunctureFILES: TwoPunctures.h
+
+$(CUDAFILES): bssn_gpu.h gpu_mem.h gpu_rhsSS_mem.h
+
+misc.o : zbesh.o
+
+# projects
+ABE: $(C++FILES) $(CFILES) $(F90FILES) $(F77FILES) $(AHFDOBJS)
+	$(CLINKER) $(CXXAPPFLAGS) -o $@ $(C++FILES) $(CFILES) $(F90FILES) $(F77FILES) $(AHFDOBJS) $(ABE_LDLIBS)
+
+ABE_CUDA: USE_CUDA_BSSN=1
+ABE_CUDA: USE_CUDA_Z4C=1
+ABE_CUDA: $(C++FILES) $(ABE_CUDA_CFILES) $(F90FILES) $(F77FILES) $(AHFDOBJS)
+	$(CLINKER) $(CXXAPPFLAGS) -o $@ $(C++FILES) $(ABE_CUDA_CFILES) $(F90FILES) $(F77FILES) $(AHFDOBJS) $(LDLIBS) -lcudart $(CUDA_LIB_PATH)
+	
+#ABEGPU: $(C++FILES_GPU) $(CFILES) $(F90FILES) $(F77FILES) $(AHFDOBJS) $(CUDAFILES)
+#	$(CLINKER) $(CXXAPPFLAGS) -o $@ $(C++FILES_GPU) $(CFILES) $(F90FILES) $(F77FILES) $(AHFDOBJS) $(CUDAFILES) $(LDLIBS)
+
+TwoPunctureABE: $(TwoPunctureFILES)
+	$(CLINKER) $(TP_OPTFLAGS) $(OMP_FLAG) -o $@ $(TwoPunctureFILES) $(LDLIBS)
+
+clean:
+	rm *.o ABE ABE_CUDA ABEGPU TwoPunctureABE make.log -f

AMSS_NCKU_source/makefile.inc

@@ -1,28 +1,7 @@
-## GCC version (commented out)
-## filein  = -I/usr/include -I/usr/lib/x86_64-linux-gnu/mpich/include -I/usr/lib/x86_64-linux-gnu/openmpi/lib/ -I/usr/lib/gcc/x86_64-linux-gnu/11/ -I/usr/include/c++/11/
-## filein  = -I/usr/include/ -I/usr/include/openmpi-x86_64/ -I/usr/lib/x86_64-linux-gnu/openmpi/include/ -I/usr/lib/x86_64-linux-gnu/openmpi/lib/ -I/usr/lib/gcc/x86_64-linux-gnu/11/ -I/usr/include/c++/11/
-## LDLIBS  = -L/usr/lib/x86_64-linux-gnu -L/usr/lib64 -L/usr/lib/gcc/x86_64-linux-gnu/11 -lgfortran -lmpi -lgfortran
-
-## Intel oneAPI version with oneMKL (Optimized for performance)
-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
-
-## 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
+## Toolchain selection
+##   nvhpc : NVIDIA HPC SDK + CUDA-aware MPI (default)
+##   intel : Intel oneAPI toolchain (legacy path)
+TOOLCHAIN ?= intel
 
 ## PGO build mode switch (ABE only; TwoPunctureABE always uses opt flags)
 ##   opt        : (default) maximum performance with PGO profile-guided optimization
@@ -43,6 +22,14 @@ else
 INTERP_LB_FLAGS =
 endif
 
+MKLROOT ?= /home/intel/oneapi/mkl/latest
+MKL_LIBDIR ?= $(MKLROOT)/lib/intel64
+MKL_INC ?= -I$(MKLROOT)/include
+
+NVHPC_ROOT ?= /home/nvidia/hpc_sdk/Linux_x86_64/25.11
+CUDA_HOME  ?= $(NVHPC_ROOT)/cuda
+CUDA_ARCH  ?= sm_80
+
 ## Kernel implementation switch
 ##   1 (default) : use C++ rewrite of bssn_rhs and helper kernels (faster)
 ##   0           : fall back to original Fortran kernels
@@ -53,36 +40,52 @@ USE_CXX_KERNELS ?= 1
 ##   0           : use original Fortran Z4c_rhs.o
 USE_CXX_Z4C_KERNELS ?= 1
 
-## BSSN-EScalar RHS switch
-##   1 (default) : use BSSN-EScalar C wrapper on the normal patch path
-##   0           : keep the original Fortran BSSN-EScalar RHS for precision-safe runs
-## Note: this requires USE_CXX_KERNELS=1 because the wrapper reuses the C BSSN kernel.
-USE_CXX_ESCALAR_KERNEL ?= 1
-
-## BSSN-EM RHS switch
-##   1 : use BSSN-EM C kernel (bssn_em_rhs_c.C) on the normal patch path
-##   0 : keep the original Fortran empart.f90 RHS for the EM fields (default)
-## Note: experimental, requires USE_CXX_KERNELS=1
-USE_CXX_EM_KERNEL ?= 0
-
-## Cached transfer switch
-##   auto (default): enable for BSSN vacuum, keep other paths on the safe uncached path
-##   1             : force cached Sync/Restrict/OutBd transfer on evolution hot paths
-##   0             : force the original uncached transfer path
-USE_TRANSFER_CACHE ?= auto
-
 ## 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
 
+## 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 $(TBBMALLOC_SO) -Wl,--as-needed
+else
+TBBMALLOC_LIBS = -Wl,--no-as-needed -ltbbmalloc -Wl,--as-needed
+endif
+
+ifeq ($(TOOLCHAIN),intel)
 f90          = ifx
 f77          = ifx
 CXX          = icpx
 CC           = icx
 CLINKER      = mpiicpx
+filein  = -I/usr/include/ $(MKL_INC) -I$(CUDA_HOME)/include
+LDLIBS       = -L$(MKL_LIBDIR) -Wl,-rpath,$(MKL_LIBDIR) \
+               -lmkl_intel_lp64 -lmkl_sequential -lmkl_core \
+               -lifcore -limf -liomp5 -lpthread -lm -ldl \
+               -L$(CUDA_HOME)/lib64 -Wl,-rpath,$(CUDA_HOME)/lib64 -lcuda -lcudart
+else ifeq ($(TOOLCHAIN),nvhpc)
+f90          = mpifort
+f77          = mpifort
+CXX          = mpicxx
+CC           = mpicc
+CLINKER      = mpicxx
 
-Cu = nvcc
-CUDA_LIB_PATH = -L/usr/lib/cuda/lib64 -I/usr/include -I/usr/lib/cuda/include
-#CUDA_APP_FLAGS = -c -g -O3 --ptxas-options=-v -arch compute_13 -code compute_13,sm_13 -Dfortran3 -Dnewc
-CUDA_APP_FLAGS = -c -g -O3 --ptxas-options=-v -Dfortran3 -Dnewc
+filein       = -I/usr/include/ $(MKL_INC) -I$(CUDA_HOME)/include
+LDLIBS       = -L$(MKL_LIBDIR) -Wl,-rpath,$(MKL_LIBDIR) \
+               -lmkl_intel_lp64 -lmkl_sequential -lmkl_core \
+               -lpthread -lm -ldl \
+               -L$(CUDA_HOME)/lib64 -Wl,-rpath,$(CUDA_HOME)/lib64 -lcuda -lcudart \
+               -fortranlibs
+endif
+
+ifeq ($(USE_TBBMALLOC),1)
+LDLIBS := $(TBBMALLOC_LIBS) $(LDLIBS)
+endif
+
+Cu = $(NVHPC_ROOT)/compilers/bin/nvcc
+CUDA_LIB_PATH = -L$(CUDA_HOME)/lib64 -I$(CUDA_HOME)/include
+CUDA_APP_FLAGS = -c -g -O3 --ptxas-options=-v -Dfortran3 -Dnewc -arch=$(CUDA_ARCH)

AMSS_NCKU_source/monitor.C

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

AMSS_NCKU_source/share_func.h

@@ -46,45 +46,6 @@ static inline size_t idx_fh_F(int iF, int jF, int kF, const int ex[3]) {
     return (size_t)ii + (size_t)jj * (size_t)nx + (size_t)kk * (size_t)nx * (size_t)ny;
 }
 
-/*
- * fh 对应 Fortran: fh(0:ex1, 0:ex2, 0:ex3)
- * ord=1 => shift=0
- * iF/jF/kF 为 Fortran 索引 (0..ex)
- */
-static inline size_t idx_fh_F_ord1(int iF, int jF, int kF, const int ex[3]) {
-    const int nx = ex[0] + 1;            // ex1 + ord
-    const int ny = ex[1] + 1;
-    return (size_t)iF + (size_t)jF * (size_t)nx + (size_t)kF * (size_t)nx * (size_t)ny;
-}
-
-/*
- * fh 对应 Fortran: fh(-3:ex1, -3:ex2, -3:ex3)
- * ord=4 => shift=3
- */
-static inline size_t idx_fh_F_ord4(int iF, int jF, int kF, const int ex[3]) {
-    const int shift = 3;
-    const int nx = ex[0] + 4;            // ex1 + ord
-    const int ny = ex[1] + 4;
-    const int ii = iF + shift;           // 0..ex1+3
-    const int jj = jF + shift;           // 0..ex2+3
-    const int kk = kF + shift;           // 0..ex3+3
-    return (size_t)ii + (size_t)jj * (size_t)nx + (size_t)kk * (size_t)nx * (size_t)ny;
-}
-
-/*
- * fh 对应 Fortran: fh(-4:ex1, -4:ex2, -4:ex3)
- * ord=5 => shift=4
- */
-static inline size_t idx_fh_F_ord5(int iF, int jF, int kF, const int ex[3]) {
-    const int shift = 4;
-    const int nx = ex[0] + 5;            // ex1 + ord
-    const int ny = ex[1] + 5;
-    const int ii = iF + shift;           // 0..ex1+4
-    const int jj = jF + shift;           // 0..ex2+4
-    const int kk = kF + shift;           // 0..ex3+4
-    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
@@ -270,10 +231,7 @@ static inline void symmetry_bd(int ord,
 {
     if (ord <= 0) return;
 
-    if (ord == 1) {
-        symmetry_bd_impl(1, 0, extc, func, funcc, SoA);
-        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;
@@ -282,91 +240,7 @@ static inline void symmetry_bd(int ord,
         symmetry_bd_impl(3, 2, extc, func, funcc, SoA);
         return;
     }
-    if (ord == 4) {
-        symmetry_bd_impl(4, 3, extc, func, funcc, SoA);
-        return;
-    }
 
     symmetry_bd_impl(ord, ord - 1, extc, func, funcc, SoA);
 }
-
-/*
- * symmetry_stbd — shell-patch (staggered boundary) ghost fill.
- *
- * Fortran: funcc(-ord+1:extc1+ord, -ord+1:extc2+ord, extc3)
- * Only 2 SoA values (x/y). No z symmetry fill.
- * Ghost on BOTH positive and negative sides of x and y.
- * Reflection uses i+2 (skips boundary) instead of i+1.
- * nx = extc1 + 2*ord, ny = extc2 + 2*ord
- */
-static inline void symmetry_stbd(int ord,
-                                 const int extc[3],
-                                 const double *func,
-                                 double *funcc,
-                                 const double SoA[2])
-{
-    const int extc1 = extc[0], extc2 = extc[1], extc3 = extc[2];
-    const int nx = extc1 + 2 * ord;
-    const int ny = extc2 + 2 * ord;
-    const int sh = ord - 1;
-    const size_t snx = (size_t)nx;
-    const size_t splane = snx * (size_t)ny;
-
-    /* 1) Copy interior: funcc(1:extc1, 1:extc2, 1:extc3) = func */
-    for (int k0 = 0; k0 < extc3; ++k0) {
-        const double *src = func + (size_t)k0 * (size_t)extc2 * (size_t)extc1;
-        const size_t kbase = (size_t)k0 * splane;
-        for (int j0 = 0; j0 < extc2; ++j0) {
-            double *dst = funcc + kbase + (size_t)(sh + j0 + 1) * snx + (size_t)(sh + 1);
-            const double *s = src + (size_t)j0 * (size_t)extc1;
-            for (int i0 = 0; i0 < extc1; ++i0) dst[i0] = s[i0];
-        }
-    }
-
-    /* 2) x-direction ghost fill */
-    const double s1 = SoA[0];
-    for (int k0 = 0; k0 < extc3; ++k0) {
-        const size_t kbase = (size_t)k0 * splane;
-        for (int j0 = 0; j0 < extc2; ++j0) {
-            const size_t off = kbase + (size_t)(sh + j0 + 1) * snx;
-            /* left side: funcc(-i) = funcc(i+2) * s1 */
-            for (int i = 0; i < ord; ++i) {
-                funcc[off + (size_t)(sh - i)]       = funcc[off + (size_t)(sh + i + 2)] * s1;
-                /* right side: funcc(extc1+1+i) = funcc(extc1-1-i) * s1 */
-                funcc[off + (size_t)(sh + extc1 + 1 + i)] = funcc[off + (size_t)(sh + extc1 - 1 - i)] * s1;
-            }
-        }
-    }
-
-    /* 3) y-direction ghost fill */
-    const double s2 = SoA[1];
-    for (int i = 0; i < nx; ++i) {
-        for (int k0 = 0; k0 < extc3; ++k0) {
-            const size_t kbase = (size_t)k0 * splane;
-            /* bottom: funcc(:,-i,:) = funcc(:,i+2,:) * s2 */
-            for (int jj = 0; jj < ord; ++jj) {
-                funcc[kbase + (size_t)(sh - jj) * snx + (size_t)i] =
-                    funcc[kbase + (size_t)(sh + jj + 2) * snx + (size_t)i] * s2;
-                /* top: funcc(:,extc2+1+jj,:) = funcc(:,extc2-1-jj,:) * s2 */
-                funcc[kbase + (size_t)(sh + extc2 + 1 + jj) * snx + (size_t)i] =
-                    funcc[kbase + (size_t)(sh + extc2 - 1 - jj) * snx + (size_t)i] * s2;
-            }
-        }
-    }
-}
-
-/*
- * Indexing for shell fh buffer: Fortran fh(-ord+1:extc1+ord, -ord+1:extc2+ord, extc3)
- * C 0-based: ii = iF + ord - 1
- * nx = extc1 + 2*ord, ny = extc2 + 2*ord
- */
-static inline size_t idx_fh_stbd(int iF, int jF, int kF, int ord, const int extc[3]) {
-    const int sh = ord - 1;
-    const int nx = extc[0] + 2 * ord;
-    const int ny = extc[1] + 2 * ord;
-    const int ii = iF + sh;
-    const int jj = jF + sh;
-    const int kk = kF - 1;  // Fortran 1-based kF → C 0-based
-    return (size_t)ii + (size_t)jj * (size_t)nx + (size_t)kk * (size_t)nx * (size_t)ny;
-}
 #endif

AMSS_NCKU_source/surface_integral.C

@@ -8,10 +8,11 @@
 #include <iostream>
 #include <iomanip>
 #include <fstream>
-#include <strstream>
-#include <cmath>
-#include <map>
-using namespace std;
+#include <strstream>
+#include <cmath>
+#include <map>
+#include <cstdlib>
+using namespace std;
 #else
 #include <iostream.h>
 #include <iomanip.h>
@@ -29,12 +30,26 @@ using namespace std;
 #include "fadmquantites_bssn.h"
 #include "getnpem2.h"
 #include "getnp4.h"
-#include "parameters.h"
-
-#define PI M_PI
-//|============================================================================
-//| Constructor
-//|============================================================================
+#include "parameters.h"
+
+#define PI M_PI
+
+namespace
+{
+bool amss_surface_timing_enabled()
+{
+  static int enabled = -1;
+  if (enabled < 0)
+  {
+    const char *env = getenv("AMSS_SURFACE_TIMING");
+    enabled = (env && atoi(env) != 0) ? 1 : 0;
+  }
+  return enabled != 0;
+}
+}
+//|============================================================================
+//| Constructor
+//|============================================================================
 
 surface_integral::surface_integral(int iSymmetry) : Symmetry(iSymmetry),
                                                     wave_cache_spinw(-1),
@@ -484,9 +499,9 @@ void surface_integral::surf_Wave(double rex, int lev, cgh *GH, var *Rpsi4, var *
   delete[] IP_out;
   DG_List->clearList();
 }
-void surface_integral::surf_Wave(double rex, int lev, cgh *GH, var *Rpsi4, var *Ipsi4,
-                                 int spinw, int maxl, int NN, double *RP, double *IP,
-                                 monitor *Monitor, MPI_Comm Comm_here) // NN is the length of RP and IP
+void surface_integral::surf_Wave(double rex, int lev, cgh *GH, var *Rpsi4, var *Ipsi4,
+                                 int spinw, int maxl, int NN, double *RP, double *IP,
+                                 monitor *Monitor, MPI_Comm Comm_here) // NN is the length of RP and IP
 {
   //   misc::tillherecheck(GH->Commlev[lev],GH->start_rank[lev],"start surface_integral::surf_Wave");
 
@@ -720,10 +735,10 @@ void surface_integral::surf_Wave(double rex, int lev, cgh *GH, var *Rpsi4, var *
   delete[] IP_out;
   DG_List->clearList();
 }
-//|----------------------------------------------------------------
-//  for shell patch
-//|----------------------------------------------------------------
-void surface_integral::surf_Wave(double rex, int lev, ShellPatch *GH, var *Rpsi4, var *Ipsi4,
+//|----------------------------------------------------------------
+//  for shell patch
+//|----------------------------------------------------------------
+void surface_integral::surf_Wave(double rex, int lev, ShellPatch *GH, var *Rpsi4, var *Ipsi4,
                                  int spinw, int maxl, int NN, double *RP, double *IP,
                                  monitor *Monitor) // NN is the length of RP and IP
 {
@@ -3281,6 +3296,8 @@ void surface_integral::surf_WaveMassPAng(double rex, int lev, cgh *GH,
                                          var *Sfx_rhs, var *Sfy_rhs, var *Sfz_rhs,
                                          double *Rout, monitor *Monitor, bool refresh_mass_fields)
 {
+  const bool timing = amss_surface_timing_enabled();
+  const double t_start = timing ? MPI_Wtime() : 0.0;
   if (Symmetry != 0 && Symmetry != 1)
   {
     surf_Wave(rex, lev, GH, Rpsi4, Ipsi4, spinw, maxl, NN, RP, IP, Monitor);
@@ -3325,6 +3342,7 @@ void surface_integral::surf_WaveMassPAng(double rex, int lev, cgh *GH,
       Pp = Pp->next;
     }
   }
+  const double t_refresh_done = timing ? MPI_Wtime() : 0.0;
 
   const int InList = 19;
   const int idx_rpsi4 = 0, idx_ipsi4 = 1;
@@ -3380,6 +3398,7 @@ void surface_integral::surf_WaveMassPAng(double rex, int lev, cgh *GH,
 
   double *shellf = new double[n_tot * InList];
   GH->PatL[lev]->data->Interp_Points(DG_List, n_tot, pox, shellf, Symmetry, Nmin, Nmax);
+  const double t_interp_done = timing ? MPI_Wtime() : 0.0;
 
   double *RP_out = new double[NN];
   double *IP_out = new double[NN];
@@ -3496,6 +3515,7 @@ void surface_integral::surf_WaveMassPAng(double rex, int lev, cgh *GH,
     if (Symmetry == 0)
       p_outz += f1o8 * Psi * (nx_g[n] * axz + ny_g[n] * ayz + nz_g[n] * azz) * theta_weight;
   }
+  const double t_integral_done = timing ? MPI_Wtime() : 0.0;
 
   for (int ii = 0; ii < NN; ii++)
   {
@@ -3534,6 +3554,7 @@ void surface_integral::surf_WaveMassPAng(double rex, int lev, cgh *GH,
     delete[] reduce_out;
     delete[] reduce_in;
   }
+  const double t_reduce_done = timing ? MPI_Wtime() : 0.0;
 
 #ifdef GaussInt
   mass = mass * rex * rex * dphi * factor;
@@ -3565,6 +3586,19 @@ void surface_integral::surf_WaveMassPAng(double rex, int lev, cgh *GH,
   Rout[5] = sy;
   Rout[6] = sz;
 
+  if (timing)
+  {
+    fprintf(stderr,
+            "[AMSS-SURFACE][rank %d] rex=%.6g lev=%d refresh=%.6f interp=%.6f integral=%.6f reduce=%.6f total=%.6f nlocal=%d ntotal=%d modes=%d\n",
+            myrank, rex, lev,
+            t_refresh_done - t_start,
+            t_interp_done - t_refresh_done,
+            t_integral_done - t_interp_done,
+            t_reduce_done - t_integral_done,
+            t_reduce_done - t_start,
+            Nmax - Nmin + 1, n_tot, NN);
+  }
+
   delete[] pox[0];
   delete[] pox[1];
   delete[] pox[2];

AMSS_NCKU_source/surface_integral.h

@@ -46,10 +46,10 @@ public:
 	surface_integral(int iSymmetry);
 	~surface_integral();
 
-	void surf_Wave(double rex, int lev, cgh *GH, var *Rpsi4, var *Ipsi4,
-				   int spinw, int maxl, int NN, double *RP, double *IP,
-				   monitor *Monitor); // NN is the length of RP and IP
-									  // this routine can only deal with the symmetry of Psi4
+	void surf_Wave(double rex, int lev, cgh *GH, var *Rpsi4, var *Ipsi4,
+				   int spinw, int maxl, int NN, double *RP, double *IP,
+				   monitor *Monitor); // NN is the length of RP and IP
+									  // this routine can only deal with the symmetry of Psi4
 	void surf_Wave(double rex, int lev, ShellPatch *GH, var *Rpsi4, var *Ipsi4,
 				   int spinw, int maxl, int NN, double *RP, double *IP,
 				   monitor *Monitor);

AMSS_NCKU_source/z4c_rhs_c.C

@@ -327,9 +327,6 @@ static int compute_rhs_z4c_cartesian(
     double Axxx[all], Axxy[all], Axxz[all], Axyx[all], Axyy[all], Axyz[all];
     double Axzx[all], Axzy[all], Axzz[all], Ayyx[all], Ayyy[all], Ayyz[all];
     double Ayzx[all], Ayzy[all], Ayzz[all], Azzx[all], Azzy[all], Azzz[all];
-#if (GAUGE == 2 || GAUGE == 3 || GAUGE == 4 || GAUGE == 5)
-    double reta[all];
-#endif
 
     const double SSS[3] = {1.0, 1.0, 1.0};
     const double AAS[3] = {-1.0, -1.0, 1.0};
@@ -479,181 +476,8 @@ static int compute_rhs_z4c_cartesian(
         dtSfy_rhs[idx] = ZEO;
         dtSfz_rhs[idx] = ZEO;
     }
-#elif (GAUGE == 2)
-    /* Variable-eta gamma-driver, chi-sqrt denominator */
-    for (int idx = 0; idx < all; ++idx)
-    {
-        const double chin1i = chin1[idx];
-        const double det = gxx[idx] * gyy[idx] * gzz[idx]
-                         + gxy[idx] * gyz[idx] * gxz[idx] * 2.0
-                         - gxz[idx] * gyy[idx] * gxz[idx]
-                         - gxy[idx] * gxy[idx] * gzz[idx]
-                         - gxx[idx] * gyz[idx] * gyz[idx];
-        const double idet = ONE / det;
-        const double upxx = (gyy[idx] * gzz[idx] - gyz[idx] * gyz[idx]) * idet;
-        const double upxy = -(gxy[idx] * gzz[idx] - gyz[idx] * gxz[idx]) * idet;
-        const double upxz = (gxy[idx] * gyz[idx] - gyy[idx] * gxz[idx]) * idet;
-        const double upyy = (gxx[idx] * gzz[idx] - gxz[idx] * gxz[idx]) * idet;
-        const double upyz = -(gxx[idx] * gyz[idx] - gxy[idx] * gxz[idx]) * idet;
-        const double upzz = (gxx[idx] * gyy[idx] - gxy[idx] * gxy[idx]) * idet;
-        const double grdchi2 =
-            upxx * chix[idx] * chix[idx] + upyy * chiy[idx] * chiy[idx] + upzz * chiz[idx] * chiz[idx]
-          + TWO * (upxy * chix[idx] * chiy[idx] + upxz * chix[idx] * chiz[idx] + upyz * chiy[idx] * chiz[idx]);
-        const double sqchi = sqrt(chin1i);
-        reta[idx] = 1.31 / TWO * sqrt(grdchi2 / chin1i) / ((ONE - sqchi) * (ONE - sqchi));
-        betax_rhs[idx] = FF * dtSfx[idx];
-        betay_rhs[idx] = FF * dtSfy[idx];
-        betaz_rhs[idx] = FF * dtSfz[idx];
-        dtSfx_rhs[idx] = Gamx_rhs[idx] - reta[idx] * dtSfx[idx];
-        dtSfy_rhs[idx] = Gamy_rhs[idx] - reta[idx] * dtSfy[idx];
-        dtSfz_rhs[idx] = Gamz_rhs[idx] - reta[idx] * dtSfz[idx];
-    }
-#elif (GAUGE == 3)
-    /* Variable-eta gamma-driver, chi-linear denominator */
-    for (int idx = 0; idx < all; ++idx)
-    {
-        const double chin1i = chin1[idx];
-        const double det = gxx[idx] * gyy[idx] * gzz[idx]
-                         + gxy[idx] * gyz[idx] * gxz[idx] * 2.0
-                         - gxz[idx] * gyy[idx] * gxz[idx]
-                         - gxy[idx] * gxy[idx] * gzz[idx]
-                         - gxx[idx] * gyz[idx] * gyz[idx];
-        const double idet = ONE / det;
-        const double upxx = (gyy[idx] * gzz[idx] - gyz[idx] * gyz[idx]) * idet;
-        const double upxy = -(gxy[idx] * gzz[idx] - gyz[idx] * gxz[idx]) * idet;
-        const double upxz = (gxy[idx] * gyz[idx] - gyy[idx] * gxz[idx]) * idet;
-        const double upyy = (gxx[idx] * gzz[idx] - gxz[idx] * gxz[idx]) * idet;
-        const double upyz = -(gxx[idx] * gyz[idx] - gxy[idx] * gxz[idx]) * idet;
-        const double upzz = (gxx[idx] * gyy[idx] - gxy[idx] * gxy[idx]) * idet;
-        const double grdchi2 =
-            upxx * chix[idx] * chix[idx] + upyy * chiy[idx] * chiy[idx] + upzz * chiz[idx] * chiz[idx]
-          + TWO * (upxy * chix[idx] * chiy[idx] + upxz * chix[idx] * chiz[idx] + upyz * chiy[idx] * chiz[idx]);
-        reta[idx] = 1.31 / TWO * sqrt(grdchi2 / chin1i) / ((ONE - chin1i) * (ONE - chin1i));
-        betax_rhs[idx] = FF * dtSfx[idx];
-        betay_rhs[idx] = FF * dtSfy[idx];
-        betaz_rhs[idx] = FF * dtSfz[idx];
-        dtSfx_rhs[idx] = Gamx_rhs[idx] - reta[idx] * dtSfx[idx];
-        dtSfy_rhs[idx] = Gamy_rhs[idx] - reta[idx] * dtSfy[idx];
-        dtSfz_rhs[idx] = Gamz_rhs[idx] - reta[idx] * dtSfz[idx];
-    }
-#elif (GAUGE == 4)
-    /* Variable-eta gamma-driver, first-order, chi-sqrt denominator */
-    for (int idx = 0; idx < all; ++idx)
-    {
-        const double chin1i = chin1[idx];
-        const double det = gxx[idx] * gyy[idx] * gzz[idx]
-                         + gxy[idx] * gyz[idx] * gxz[idx] * 2.0
-                         - gxz[idx] * gyy[idx] * gxz[idx]
-                         - gxy[idx] * gxy[idx] * gzz[idx]
-                         - gxx[idx] * gyz[idx] * gyz[idx];
-        const double idet = ONE / det;
-        const double upxx = (gyy[idx] * gzz[idx] - gyz[idx] * gyz[idx]) * idet;
-        const double upxy = -(gxy[idx] * gzz[idx] - gyz[idx] * gxz[idx]) * idet;
-        const double upxz = (gxy[idx] * gyz[idx] - gyy[idx] * gxz[idx]) * idet;
-        const double upyy = (gxx[idx] * gzz[idx] - gxz[idx] * gxz[idx]) * idet;
-        const double upyz = -(gxx[idx] * gyz[idx] - gxy[idx] * gxz[idx]) * idet;
-        const double upzz = (gxx[idx] * gyy[idx] - gxy[idx] * gxy[idx]) * idet;
-        const double grdchi2 =
-            upxx * chix[idx] * chix[idx] + upyy * chiy[idx] * chiy[idx] + upzz * chiz[idx] * chiz[idx]
-          + TWO * (upxy * chix[idx] * chiy[idx] + upxz * chix[idx] * chiz[idx] + upyz * chiy[idx] * chiz[idx]);
-        const double sqchi = sqrt(chin1i);
-        reta[idx] = 1.31 / TWO * sqrt(grdchi2 / chin1i) / ((ONE - sqchi) * (ONE - sqchi));
-        betax_rhs[idx] = Gamx_rhs[idx] - reta[idx] * betax[idx];
-        betay_rhs[idx] = Gamy_rhs[idx] - reta[idx] * betay[idx];
-        betaz_rhs[idx] = Gamz_rhs[idx] - reta[idx] * betaz[idx];
-        dtSfx_rhs[idx] = ZEO;
-        dtSfy_rhs[idx] = ZEO;
-        dtSfz_rhs[idx] = ZEO;
-    }
-#elif (GAUGE == 5)
-    /* Variable-eta gamma-driver, first-order, chi-linear denominator */
-    for (int idx = 0; idx < all; ++idx)
-    {
-        const double chin1i = chin1[idx];
-        const double det = gxx[idx] * gyy[idx] * gzz[idx]
-                         + gxy[idx] * gyz[idx] * gxz[idx] * 2.0
-                         - gxz[idx] * gyy[idx] * gxz[idx]
-                         - gxy[idx] * gxy[idx] * gzz[idx]
-                         - gxx[idx] * gyz[idx] * gyz[idx];
-        const double idet = ONE / det;
-        const double upxx = (gyy[idx] * gzz[idx] - gyz[idx] * gyz[idx]) * idet;
-        const double upxy = -(gxy[idx] * gzz[idx] - gyz[idx] * gxz[idx]) * idet;
-        const double upxz = (gxy[idx] * gyz[idx] - gyy[idx] * gxz[idx]) * idet;
-        const double upyy = (gxx[idx] * gzz[idx] - gxz[idx] * gxz[idx]) * idet;
-        const double upyz = -(gxx[idx] * gyz[idx] - gxy[idx] * gxz[idx]) * idet;
-        const double upzz = (gxx[idx] * gyy[idx] - gxy[idx] * gxy[idx]) * idet;
-        const double grdchi2 =
-            upxx * chix[idx] * chix[idx] + upyy * chiy[idx] * chiy[idx] + upzz * chiz[idx] * chiz[idx]
-          + TWO * (upxy * chix[idx] * chiy[idx] + upxz * chix[idx] * chiz[idx] + upyz * chiy[idx] * chiz[idx]);
-        reta[idx] = 1.31 / TWO * sqrt(grdchi2 / chin1i) / ((ONE - chin1i) * (ONE - chin1i));
-        betax_rhs[idx] = Gamx_rhs[idx] - reta[idx] * betax[idx];
-        betay_rhs[idx] = Gamy_rhs[idx] - reta[idx] * betay[idx];
-        betaz_rhs[idx] = Gamz_rhs[idx] - reta[idx] * betaz[idx];
-        dtSfx_rhs[idx] = ZEO;
-        dtSfy_rhs[idx] = ZEO;
-        dtSfz_rhs[idx] = ZEO;
-    }
-#elif (GAUGE == 6 || GAUGE == 7)
-    {
-        /* Jason's position-dependent damping: rational (6) or exponential (7) */
-        int BHN = 0;
-        double Porg[9] = {0.0};
-        double Mass[3] = {0.0};
-    #ifdef fortran1
-        extern "C" { void getpbh(int &, double *, double *); }
-    #elif defined(fortran2)
-        extern "C" { void GETPBH(int &, double *, double *); }
-    #else
-        extern "C" { void getpbh_(int &, double *, double *); }
-    #endif
-        {
-    #ifdef fortran1
-            getpbh(BHN, Porg, Mass);
-    #elif defined(fortran2)
-            GETPBH(BHN, Porg, Mass);
-    #else
-            getpbh_(BHN, Porg, Mass);
-    #endif
-        }
-        if (BHN == 2)
-        {
-            const double M = Mass[0] + Mass[1];
-            const double A = 2.0 / M;
-            const double w1 = 12.0, w2 = 12.0;
-            const double C1 = 1.0 / Mass[0] - A;
-            const double C2 = 1.0 / Mass[1] - A;
-            const double BH_sep2 = (Porg[3] - Porg[0]) * (Porg[3] - Porg[0])
-                                 + (Porg[4] - Porg[1]) * (Porg[4] - Porg[1])
-                                 + (Porg[5] - Porg[2]) * (Porg[5] - Porg[2]);
-            const double inv_BH_sep2 = 1.0 / BH_sep2;
-            for (int k0 = 0; k0 < nz; ++k0) {
-            for (int j0 = 0; j0 < ny; ++j0) {
-            for (int i0 = 0; i0 < nx; ++i0) {
-                const size_t idx = idx_ex(i0, j0, k0, ex);
-                const double xp = X[i0], yp = Y[j0], zp = Z[k0];
-                const double r1 = ((Porg[0]-xp)*(Porg[0]-xp) + (Porg[1]-yp)*(Porg[1]-yp) + (Porg[2]-zp)*(Porg[2]-zp)) * inv_BH_sep2;
-                const double r2 = ((Porg[3]-xp)*(Porg[3]-xp) + (Porg[4]-yp)*(Porg[4]-yp) + (Porg[5]-zp)*(Porg[5]-zp)) * inv_BH_sep2;
-            #if (GAUGE == 6)
-                const double reta_val = A + C1 / (1.0 + w1 * r1) + C2 / (1.0 + w2 * r2);
-            #else
-                const double reta_val = A + C1 * exp(-w1 * r1) + C2 * exp(-w2 * r2);
-            #endif
-                betax_rhs[idx] = FF * dtSfx[idx];
-                betay_rhs[idx] = FF * dtSfy[idx];
-                betaz_rhs[idx] = FF * dtSfz[idx];
-                dtSfx_rhs[idx] = Gamx_rhs[idx] - reta_val * dtSfx[idx];
-                dtSfy_rhs[idx] = Gamy_rhs[idx] - reta_val * dtSfy[idx];
-                dtSfz_rhs[idx] = Gamz_rhs[idx] - reta_val * dtSfz[idx];
-            }}}
-        }
-        else
-        {
-            fprintf(stderr, "z4c_rhs_c: GAUGE %d requires BHN=2, got BHN=%d\n", (int)GAUGE, BHN);
-            return 1;
-        }
-    }
 #else
-#error "z4c_rhs_c.C: unsupported GAUGE value"
+#error "z4c_rhs_c.C currently supports GAUGE == 0 or GAUGE == 1 for Z4C"
 #endif
 
     lopsided(ex, X, Y, Z, gxx, gxx_rhs, betax, betay, betaz, Symmetry, SSS);
@@ -681,7 +505,7 @@ static int compute_rhs_z4c_cartesian(
     lopsided(ex, X, Y, Z, betax, betax_rhs, betax, betay, betaz, Symmetry, ASS);
     lopsided(ex, X, Y, Z, betay, betay_rhs, betax, betay, betaz, Symmetry, SAS);
     lopsided(ex, X, Y, Z, betaz, betaz_rhs, betax, betay, betaz, Symmetry, SSA);
-#if (GAUGE == 0 || GAUGE == 2 || GAUGE == 3 || GAUGE == 6 || GAUGE == 7)
+#if (GAUGE == 0)
     lopsided(ex, X, Y, Z, dtSfx, dtSfx_rhs, betax, betay, betaz, Symmetry, ASS);
     lopsided(ex, X, Y, Z, dtSfy, dtSfy_rhs, betax, betay, betaz, Symmetry, SAS);
     lopsided(ex, X, Y, Z, dtSfz, dtSfz_rhs, betax, betay, betaz, Symmetry, SSA);
@@ -728,7 +552,7 @@ static int compute_rhs_z4c_cartesian(
         kodis(ex, X, Y, Z, betax, betax_rhs, ASS, Symmetry, eps);
         kodis(ex, X, Y, Z, betay, betay_rhs, SAS, Symmetry, eps);
         kodis(ex, X, Y, Z, betaz, betaz_rhs, SSA, Symmetry, eps);
-#if (GAUGE == 0 || GAUGE == 2 || GAUGE == 3 || GAUGE == 6 || GAUGE == 7)
+#if (GAUGE == 0)
         kodis(ex, X, Y, Z, dtSfx, dtSfx_rhs, ASS, Symmetry, eps);
         kodis(ex, X, Y, Z, dtSfy, dtSfy_rhs, SAS, Symmetry, eps);
         kodis(ex, X, Y, Z, dtSfz, dtSfz_rhs, SSA, Symmetry, eps);

AMSS_NCKU_source/z4c_rhs_cuda.h

@@ -0,0 +1,114 @@
+#ifndef Z4C_RHS_CUDA_H
+#define Z4C_RHS_CUDA_H
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+enum {
+    Z4C_CUDA_STATE_COUNT = 25
+};
+
+int z4c_cuda_rk4_substep(void *block_tag,
+                         int *ex, double *X, double *Y, double *Z,
+                         double **state_host_in,
+                         double **state_host_out,
+                         const double *propspeed,
+                         const double *soa_flat,
+                         const double *bbox,
+                         double &dT,
+                         double &T,
+                         int &RK4,
+                         int &apply_bam_bc,
+                         int &Symmetry,
+                         int &Lev,
+                         double &eps,
+                         int &co,
+                         int &keep_resident_state,
+                         int &apply_enforce_ga,
+                         double &chitiny);
+
+int z4c_cuda_download_resident_state(void *block_tag,
+                                     int *ex,
+                                     double **state_host_out);
+
+int z4c_cuda_pack_state_region_to_host_buffer(void *block_tag,
+                                              int state_index,
+                                              double *host_buffer,
+                                              int *ex,
+                                              int i0, int j0, int k0,
+                                              int sx, int sy, int sz);
+
+int z4c_cuda_unpack_state_region_from_host_buffer(void *block_tag,
+                                                  int state_index,
+                                                  double *host_buffer,
+                                                  int *ex,
+                                                  int i0, int j0, int k0,
+                                                  int sx, int sy, int sz);
+
+int z4c_cuda_pack_state_batch_to_host_buffer(void *block_tag,
+                                             int state_count,
+                                             double *host_buffer,
+                                             int *ex,
+                                             int i0, int j0, int k0,
+                                             int sx, int sy, int sz);
+
+int z4c_cuda_unpack_state_batch_from_host_buffer(void *block_tag,
+                                                 int state_count,
+                                                 double *host_buffer,
+                                                 int *ex,
+                                                 int i0, int j0, int k0,
+                                                 int sx, int sy, int sz);
+
+int z4c_cuda_pack_state_batch_to_device_buffer(void *block_tag,
+                                               int state_count,
+                                               double *device_buffer,
+                                               int *ex,
+                                               int i0, int j0, int k0,
+                                               int sx, int sy, int sz);
+
+int z4c_cuda_unpack_state_batch_from_device_buffer(void *block_tag,
+                                                   int state_count,
+                                                   double *device_buffer,
+                                                   int *ex,
+                                                   int i0, int j0, int k0,
+                                                   int sx, int sy, int sz);
+
+int z4c_cuda_restrict_state_batch_to_device_buffer(void *block_tag,
+                                                   int state_count,
+                                                   double *device_buffer,
+                                                   int *ex,
+                                                   int sx, int sy, int sz,
+                                                   int fi0, int fj0, int fk0,
+                                                   const double *state_soa);
+
+int z4c_cuda_prolong_state_batch_to_device_buffer(void *block_tag,
+                                                  int state_count,
+                                                  double *device_buffer,
+                                                  int *ex,
+                                                  int sx, int sy, int sz,
+                                                  int ii0, int jj0, int kk0,
+                                                  int lbc_i, int lbc_j, int lbc_k,
+                                                  const double *state_soa);
+
+int z4c_cuda_download_state_subset(void *block_tag,
+                                   int *ex,
+                                   int subset_count,
+                                   const int *state_indices,
+                                   double **state_host_out);
+
+int z4c_cuda_upload_state_subset(void *block_tag,
+                                 int *ex,
+                                 int subset_count,
+                                 const int *state_indices,
+                                 double **state_host_in);
+
+int z4c_cuda_has_resident_state(void *block_tag);
+
+void z4c_cuda_release_step_ctx(void *block_tag);
+
+#ifdef __cplusplus
+}
+#endif
+
+#endif

BSSN_BUILD_CONFIG_MIGRATION.md

@@ -1,211 +0,0 @@
-# BSSN Build Config Migration
-
-This note records the build-configuration fix needed when replacing
-`AMSS_NCKU_Input.py` or `generate_macrodef.py` with a newer upstream version.
-
-## Problem
-
-`AMSS_NCKU_source/macrodef.h` is not the authoritative file used by normal
-runs. `AMSS_NCKU_Program.py` first generates macro files under
-`input_data.File_directory`, copies `AMSS_NCKU_source` to
-`<File_directory>/AMSS_NCKU_source_copy`, then copies the generated macro files
-into that copied source tree and compiles there.
-
-Therefore, makefile logic must not depend only on the stale
-`AMSS_NCKU_source/macrodef.h`. The actual equation path must be passed to the
-copied build tree from the same generation step that creates `macrodef.h`.
-
-The performance regression was caused by compiling/linking the
-`BSSN-EScalar` C wrapper into BSSN vacuum builds. For BSSN vacuum (`ABEtype=0`),
-the build must use:
-
-```make
-BSSN_USE_TRANSFER_CACHE=1
-BSSN_USE_ESCALAR_C_KERNEL=0
-```
-
-and must not link `bssn_escalar_rhs_c.o`.
-
-## Required Migration Steps
-
-### 1. Add an ABE type helper in `generate_macrodef.py`
-
-Add a helper that maps `input_data.Equation_Class` to the numeric `ABEtype`.
-Use the same mapping as `macrodef.h`:
-
-```python
-def get_abe_type():
-    if ( input_data.Equation_Class == "BSSN" ):
-        return 0
-    elif ( input_data.Equation_Class == "BSSN-EScalar" ):
-        return 1
-    elif ( input_data.Equation_Class == "BSSN-EM" ):
-        return 3
-    elif ( input_data.Equation_Class == "Z4C" ):
-        return 2
-    else:
-        raise ValueError("Equation_Class setting error!!!")
-```
-
-Update `generate_macrodef_h()` to print `#define ABEtype {get_abe_type()}`
-instead of duplicating the if/elif mapping.
-
-### 2. Generate a makefile fragment
-
-In `generate_macrodef.py`, add:
-
-```python
-def generate_build_config():
-    file1 = open(os.path.join(input_data.File_directory, "AMSS_NCKU_build.mk"), "w")
-    print("# Generated by generate_macrodef.py; do not edit manually.", file=file1)
-    print(f"ABE_TYPE := {get_abe_type()}", file=file1)
-    file1.close()
-```
-
-This file is the build-time authority for the equation path.
-
-### 3. Call and copy the generated build config
-
-In `AMSS_NCKU_Program.py`, after generating `macrodef.h` and `macrodef.fh`, call:
-
-```python
-generate_macrodef.generate_build_config()
-print(" AMSS-NCKU build config AMSS_NCKU_build.mk has been generated. ")
-```
-
-When copying generated files into `AMSS_NCKU_source_copy`, also copy:
-
-```python
-build_config_path = os.path.join(File_directory, "AMSS_NCKU_build.mk")
-shutil.copy2(build_config_path, AMSS_NCKU_source_copy)
-```
-
-### 4. Make the source makefile consume the generated config
-
-At the top of `AMSS_NCKU_source/makefile`, after `include makefile.inc`, add:
-
-```make
--include AMSS_NCKU_build.mk
-
-ABE_TYPE ?= $(shell awk '/^[[:space:]]*\#define[[:space:]]+ABEtype/ {print $$3; exit}' macrodef.h 2>/dev/null)
-```
-
-The generated `AMSS_NCKU_build.mk` is used during normal Python-driven builds.
-The fallback keeps manual source-tree builds usable.
-
-### 5. Gate path-specific build options by `ABE_TYPE`
-
-Use effective build switches:
-
-```make
-ifeq ($(USE_TRANSFER_CACHE),auto)
-ifeq ($(ABE_TYPE),0)
-EFFECTIVE_USE_TRANSFER_CACHE = 1
-else
-EFFECTIVE_USE_TRANSFER_CACHE = 0
-endif
-else
-EFFECTIVE_USE_TRANSFER_CACHE = $(USE_TRANSFER_CACHE)
-endif
-
-ifeq ($(USE_CXX_ESCALAR_KERNEL),1)
-ifeq ($(ABE_TYPE),1)
-EFFECTIVE_USE_CXX_ESCALAR_KERNEL = 1
-else
-EFFECTIVE_USE_CXX_ESCALAR_KERNEL = 0
-endif
-else
-EFFECTIVE_USE_CXX_ESCALAR_KERNEL = 0
-endif
-
-TRANSFER_CACHE_FLAG = -DBSSN_USE_TRANSFER_CACHE=$(EFFECTIVE_USE_TRANSFER_CACHE)
-ESCALAR_KERNEL_FLAG = -DBSSN_USE_ESCALAR_C_KERNEL=$(EFFECTIVE_USE_CXX_ESCALAR_KERNEL)
-```
-
-Only add `bssn_escalar_rhs_c.o` when the effective EScalar C kernel switch is
-enabled:
-
-```make
-ifeq ($(EFFECTIVE_USE_CXX_ESCALAR_KERNEL),1)
-CFILES += bssn_escalar_rhs_c.o
-endif
-```
-
-### 6. Use safe transfer-cache default
-
-In `AMSS_NCKU_source/makefile.inc`, keep:
-
-```make
-USE_TRANSFER_CACHE ?= auto
-```
-
-With the effective switch logic above, this enables cached transfer for BSSN
-vacuum while keeping non-BSSN paths on the uncached path by default.
-
-## Verification Checklist
-
-Run these checks after migrating:
-
-```bash
-python3 -c "import generate_macrodef; generate_macrodef.generate_build_config()"
-cat GW150914/AMSS_NCKU_build.mk
-```
-
-For BSSN, the generated file should contain:
-
-```make
-ABE_TYPE := 0
-```
-
-Dry-run the copied or source makefile:
-
-```bash
-make -n -B INTERP_LB_MODE=off ABE | grep -E 'BSSN_USE_TRANSFER_CACHE|BSSN_USE_ESCALAR_C_KERNEL|bssn_escalar_rhs_c'
-```
-
-Expected BSSN result:
-
-```text
--DBSSN_USE_TRANSFER_CACHE=1 -DBSSN_USE_ESCALAR_C_KERNEL=0
-```
-
-and no `bssn_escalar_rhs_c.o` in the final link command.
-
-Run the full workflow:
-
-```bash
-python3 AMSS_NCKU_Program.py
-```
-
-For the 10-step BSSN test, compare coordinate output:
-
-```bash
-python3 - <<'PY'
-from pathlib import Path
-old = Path('../GW150914-06457/AMSS_NCKU_output/bssn_BH.dat')
-new = Path('GW150914/AMSS_NCKU_output/bssn_BH.dat')
-
-def rows(path):
-    out = []
-    for line in path.read_text().splitlines():
-        if not line.strip() or line.lstrip().startswith('#'):
-            continue
-        out.append([float(x) for x in line.split()])
-    return out
-
-ro, rn = rows(old), rows(new)
-n = min(len(ro), len(rn))
-max_abs = 0.0
-for i in range(n):
-    for a, b in zip(ro[i], rn[i]):
-        max_abs = max(max_abs, abs(a - b))
-print(f"old_rows={len(ro)} new_rows={len(rn)} compared_rows={n}")
-print(f"max_abs_diff={max_abs:.17g}")
-PY
-```
-
-For the validated migration, the first 10 rows matched exactly:
-
-```text
-max_abs_diff=0
-```

generate_macrodef.py

@@ -12,37 +12,6 @@ import os
 import AMSS_NCKU_Input as input_data          ## import program input file
 
 
-##################################################################
-
-def get_abe_type():
-    if ( input_data.Equation_Class == "BSSN" ):
-        return 0
-    elif ( input_data.Equation_Class == "BSSN-EScalar" ):
-        return 1
-    elif ( input_data.Equation_Class == "BSSN-EM" ):
-        return 3
-    elif ( input_data.Equation_Class == "Z4C" ):
-        return 2
-    else:
-        raise ValueError("Equation_Class setting error!!!")
-
-
-##################################################################
-
-## Generate the makefile fragment used by the copied source tree.
-## The source-tree macrodef.h is not authoritative because macro files
-## are regenerated under File_directory for each run.
-
-def generate_build_config():
-
-    file1 = open( os.path.join(input_data.File_directory, "AMSS_NCKU_build.mk"), "w")
-
-    print( "# Generated by generate_macrodef.py; do not edit manually.", file=file1 )
-    print( f"ABE_TYPE := {get_abe_type()}",                             file=file1 )
-
-    file1.close()
-
-
 ##################################################################
 
 ## Generate the macro file macrodef.h according to user settings
@@ -89,10 +58,19 @@ def generate_macrodef_h():
     # 2: Z4c vacuum
     # 3: coupled to Maxwell field
 
-    try:
-        print( f"#define ABEtype {get_abe_type()}", file=file1 )
-        print(                                      file=file1 )
-    except ValueError:
+    if ( input_data.Equation_Class == "BSSN" ):
+        print( "#define ABEtype 0", file=file1 )
+        print(                      file=file1 )
+    elif ( input_data.Equation_Class == "BSSN-EScalar" ):
+        print( "#define ABEtype 1", file=file1 )
+        print(                      file=file1 )
+    elif ( input_data.Equation_Class == "BSSN-EM" ):
+        print( "#define ABEtype 3", file=file1 )
+        print(                      file=file1 )
+    elif ( input_data.Equation_Class == "Z4C" ):
+        print( "#define ABEtype 2", file=file1 )
+        print(                      file=file1 )
+    else:
         print( "Equation_Class setting error!!!"                )
         print()
         print( "# Equation type #define ABEtype setting error!!!", file=file1 )
@@ -226,7 +204,7 @@ def generate_macrodef_h():
     # use GPU or not
     
     if ( input_data.GPU_Calculation == "yes"):
-        print( "#define USE_GPU",   file=file1 )
+        print( "//#define USE_GPU",   file=file1 )
         print(                      file=file1 )
     elif ( input_data.GPU_Calculation == "no"):
         print( "//#define USE_GPU", file=file1 )

makefile_and_run.py

@@ -9,6 +9,8 @@
 
 
 import AMSS_NCKU_Input as input_data
+import os
+import shutil
 import subprocess
 import time
 
@@ -56,6 +58,198 @@ BUILD_JOBS = 64
 
 ##################################################################
 
+def _truthy(value, default=False):
+    if value is None:
+        return default
+    if isinstance(value, bool):
+        return value
+    text = str(value).strip().lower()
+    if text == "":
+        return default
+    return text in ("1", "yes", "y", "true", "on", "enable", "enabled")
+
+
+def _input_or_env(input_name, env_name, default=None):
+    if env_name in os.environ:
+        return os.environ[env_name]
+    return getattr(input_data, input_name, default)
+
+
+def _input_env_passthrough(runtime_env, env_name):
+    if env_name in runtime_env:
+        return
+    if hasattr(input_data, env_name):
+        runtime_env[env_name] = str(getattr(input_data, env_name))
+
+
+def _start_cuda_mps_if_requested(runtime_env):
+    if input_data.GPU_Calculation != "yes":
+        return False
+
+    default_auto_mps = int(getattr(input_data, "MPI_processes", 1)) > 1
+    auto_mps = _truthy(
+        _input_or_env("CUDA_Auto_MPS", "AMSS_CUDA_AUTO_MPS", default_auto_mps),
+        default=default_auto_mps,
+    )
+    if not auto_mps:
+        return False
+
+    mps_control = shutil.which("nvidia-cuda-mps-control")
+    if not mps_control:
+        print(" CUDA MPS control command was not found; running without MPS.")
+        return False
+
+    uid = os.getuid()
+    pipe_dir = str(_input_or_env("CUDA_MPS_PIPE_DIRECTORY", "CUDA_MPS_PIPE_DIRECTORY",
+                                 f"/tmp/amss-ncku-mps-{uid}"))
+    log_dir = str(_input_or_env("CUDA_MPS_LOG_DIRECTORY", "CUDA_MPS_LOG_DIRECTORY",
+                                f"/tmp/amss-ncku-mps-log-{uid}"))
+    os.makedirs(pipe_dir, exist_ok=True)
+    os.makedirs(log_dir, exist_ok=True)
+
+    mps_env = runtime_env.copy()
+    mps_env["CUDA_MPS_PIPE_DIRECTORY"] = pipe_dir
+    mps_env["CUDA_MPS_LOG_DIRECTORY"] = log_dir
+
+    if os.path.exists(os.path.join(pipe_dir, "control")):
+        runtime_env.update({
+            "CUDA_MPS_PIPE_DIRECTORY": pipe_dir,
+            "CUDA_MPS_LOG_DIRECTORY": log_dir,
+        })
+        print(f" Reusing CUDA MPS daemon: {pipe_dir}")
+        return False
+
+    print(f" Starting CUDA MPS daemon for this run: {pipe_dir}")
+    result = subprocess.run([mps_control, "-d"], env=mps_env, text=True,
+                            stdout=subprocess.PIPE, stderr=subprocess.STDOUT)
+    if result.returncode != 0:
+        print(" CUDA MPS daemon did not start; running without MPS.")
+        if result.stdout:
+            print(result.stdout, end="")
+        return False
+
+    runtime_env.update({
+        "CUDA_MPS_PIPE_DIRECTORY": pipe_dir,
+        "CUDA_MPS_LOG_DIRECTORY": log_dir,
+    })
+    return True
+
+
+def _stop_cuda_mps(runtime_env):
+    mps_control = shutil.which("nvidia-cuda-mps-control")
+    if not mps_control:
+        return
+    subprocess.run([mps_control], input="quit\n", env=runtime_env, text=True,
+                   stdout=subprocess.PIPE, stderr=subprocess.STDOUT)
+
+
+def _gpu_runtime_env():
+    runtime_env = os.environ.copy()
+    finite_difference = str(getattr(input_data, "Finite_Diffenence_Method", "4th-order")).strip()
+
+    defaults = {
+        "AMSS_EVOLVE_TIMING": "1",
+        "AMSS_ESCALAR_STEP_TIMING": "0",
+        "AMSS_INTERP_FAST": "1",
+        "AMSS_INTERP_GPU": "1",
+        "AMSS_ANALYSIS_MAP_EVERY": "1000000",
+        "AMSS_CUDA_AWARE_MPI": "1",
+        "AMSS_CUDA_KEEP_RESIDENT_AFTER_STEP": "1",
+        "AMSS_CUDA_KEEP_ALL_LEVELS": "1",
+        "AMSS_CUDA_ESCALAR_KEEP_RESIDENT_AFTER_STEP": "1",
+        "AMSS_CUDA_ESCALAR_KEEP_ALL_LEVELS": "1",
+        "AMSS_CUDA_EM_CACHE_SOURCES": "1",
+        "AMSS_CUDA_EM_ZERO_FASTPATH": "1",
+        "AMSS_EM_ZERO_ANALYSIS_FASTPATH": "1",
+        "AMSS_EM_ZERO_RESIDENT_DOWNLOAD_FASTPATH": "1",
+        "AMSS_CUDA_AMR_HOST_STAGED": "1",
+        "AMSS_CUDA_AMR_RESTRICT_DEVICE": "0",
+        "AMSS_CUDA_AMR_RESTRICT_BATCH": "0",
+        "AMSS_CUDA_DEVICE_SEGMENT_BATCH": "0",
+        "AMSS_CUDA_UNCACHED_DEVICE_BUFFERS": "1",
+        "AMSS_SHELL_FAST_INTERP": "0",
+        "AMSS_SHELL_PARALLEL_INTERP": "0",
+        "AMSS_SHELL_CUDA_INTERP": "0",
+    }
+    if finite_difference in ("2nd-order", "8th-order"):
+        defaults.update({
+            "AMSS_INTERP_FAST": "0",
+            "AMSS_INTERP_GPU": "0",
+            "AMSS_CUDA_AWARE_MPI": "0",
+        })
+    if finite_difference == "8th-order" and getattr(input_data, "Equation_Class", "") == "BSSN-EM":
+        defaults.update({
+            "AMSS_CUDA_AMR_RESTRICT_DEVICE": "1",
+            "AMSS_CUDA_AMR_RESTRICT_BATCH": "1",
+            "AMSS_CUDA_DEVICE_SEGMENT_BATCH": "1",
+        })
+    if getattr(input_data, "basic_grid_set", "") == "Shell-Patch":
+        defaults.update({
+            "AMSS_CUDA_AWARE_MPI": "0",
+            "AMSS_SHELL_FAST_INTERP": "1",
+            "AMSS_SHELL_PARALLEL_INTERP": "1",
+            "AMSS_SHELL_INTERP_THREADS": "16",
+        })
+    if getattr(input_data, "Equation_Class", "") in ("BSSN", "BSSN-EScalar", "Z4C"):
+        defaults["AMSS_CUDA_AMR_RESTRICT_DEVICE"] = "1"
+    if getattr(input_data, "Equation_Class", "") == "Z4C":
+        defaults.update({
+            "AMSS_Z4C_CUDA_RESIDENT": "1",
+            "AMSS_CONSTRAINT_OUT_EVERY": "1000000",
+        })
+    for key, value in defaults.items():
+        runtime_env.setdefault(key, value)
+
+    passthrough_envs = [
+        "AMSS_CUDA_RESIDENT_SYNC",
+        "AMSS_CUDA_BSSN_RESIDENT_SYNC",
+        "AMSS_CUDA_EM_RESIDENT_SYNC",
+        "AMSS_CUDA_ESCALAR_RESIDENT_SYNC",
+        "AMSS_CUDA_BH_INTERP_RESIDENT",
+        "AMSS_CUDA_KEEP_RESIDENT_AFTER_STEP",
+        "AMSS_CUDA_KEEP_ALL_LEVELS",
+        "AMSS_CUDA_EM_KEEP_RESIDENT_AFTER_STEP",
+        "AMSS_CUDA_EM_KEEP_ALL_LEVELS",
+        "AMSS_CUDA_ESCALAR_KEEP_RESIDENT_AFTER_STEP",
+        "AMSS_CUDA_ESCALAR_KEEP_ALL_LEVELS",
+        "AMSS_CUDA_AMR_HOST_STAGED",
+        "AMSS_CUDA_AMR_RESTRICT_DEVICE",
+        "AMSS_CUDA_AMR_RESTRICT_BATCH",
+        "AMSS_CUDA_DEVICE_SEGMENT_BATCH",
+        "AMSS_CUDA_UNCACHED_DEVICE_BUFFERS",
+        "AMSS_CUDA_EM_CACHE_SOURCES",
+        "AMSS_CUDA_EM_ZERO_FASTPATH",
+        "AMSS_CUDA_AWARE_MPI",
+        "AMSS_CUDA_REGRID_FLUSH_ALWAYS",
+        "AMSS_Z4C_CUDA_RESIDENT",
+        "AMSS_SHELL_FAST_INTERP",
+        "AMSS_SHELL_PARALLEL_INTERP",
+        "AMSS_SHELL_CUDA_INTERP",
+        "AMSS_SHELL_INTERP_THREADS",
+        "AMSS_EM_ZERO_ANALYSIS_FASTPATH",
+        "AMSS_EM_ZERO_RESIDENT_DOWNLOAD_FASTPATH",
+        "AMSS_INTERP_FAST",
+        "AMSS_INTERP_GPU",
+    ]
+    for env_name in passthrough_envs:
+        _input_env_passthrough(runtime_env, env_name)
+
+    optional_overrides = {
+        "AMSS_INTERP_FAST_COMPARE": "AMSS_Interp_Fast_Compare",
+        "AMSS_INTERP_FAST_COMPARE_LIMIT": "AMSS_Interp_Fast_Compare_Limit",
+        "AMSS_INTERP_FAST_COMPARE_TOL": "AMSS_Interp_Fast_Compare_Tol",
+        "AMSS_GPU_STAGE_TIMING": "AMSS_GPU_Stage_Timing",
+        "AMSS_GPU_STAGE_TIMING_EVERY": "AMSS_GPU_Stage_Timing_Every",
+    }
+    for env_name, input_name in optional_overrides.items():
+        if env_name not in runtime_env and hasattr(input_data, input_name):
+            runtime_env[env_name] = str(getattr(input_data, input_name))
+
+    return runtime_env
+
+
+##################################################################
+
 
 
 ##################################################################
@@ -68,11 +262,13 @@ def makefile_ABE():
     print( " Compiling the AMSS-NCKU executable file ABE/ABEGPU " ) 
     print(                                                        )
 
+    z4c_mrbd = int(getattr(input_data, "AMSS_Z4C_MRBD", 0))
+
     ## 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} env AMSS_Z4C_MRBD={z4c_mrbd} make -j{BUILD_JOBS} INTERP_LB_MODE=off USE_CUDA_BSSN=0 USE_CUDA_Z4C=0 ABE"
     elif (input_data.GPU_Calculation == "yes"):
-        makefile_command  = f"{NUMACTL_CPU_BIND} make -j{BUILD_JOBS} ABEGPU"
+        makefile_command  = f"{NUMACTL_CPU_BIND} env AMSS_Z4C_MRBD={z4c_mrbd} make -j{BUILD_JOBS} INTERP_LB_MODE=off USE_CUDA_BSSN=1 USE_CUDA_Z4C=1 ABE_CUDA"
     else:
         print( " CPU/GPU numerical calculation setting is wrong " )
         print(                                                    )
@@ -145,29 +341,84 @@ def run_ABE():
     print(                                                      )
 
     ## Define the command to run; cast other values to strings as needed
+    mpi_env = None
+    started_mps = False
     
+    mpi_processes = int(input_data.MPI_processes)
+    if (input_data.GPU_Calculation == "yes" and
+        getattr(input_data, "Equation_Class", "") == "Z4C"):
+        z4c_env_np = os.environ.get("AMSS_Z4C_GPU_MPI_PROCESSES")
+        if z4c_env_np and int(z4c_env_np) > 0:
+            mpi_processes = int(z4c_env_np)
+        elif mpi_processes < 4:
+            mpi_processes = 4
+    if (input_data.GPU_Calculation == "yes" and
+        getattr(input_data, "basic_grid_set", "") == "Shell-Patch"):
+        shell_env_np = os.environ.get("AMSS_SHELL_GPU_MPI_PROCESSES")
+        if shell_env_np and int(shell_env_np) > 0:
+            mpi_processes = int(shell_env_np)
+        elif mpi_processes < 4:
+            mpi_processes = 4
+
     if (input_data.GPU_Calculation == "no"):
-        mpi_command         = NUMACTL_CPU_BIND + " mpirun -np " + str(input_data.MPI_processes) + " ./ABE"
+        mpi_command         = NUMACTL_CPU_BIND + " mpirun -np " + str(mpi_processes) + " ./ABE"
         #mpi_command         = " mpirun -np " + str(input_data.MPI_processes) + " ./ABE"
         mpi_command_outfile = "ABE_out.log"
     elif (input_data.GPU_Calculation == "yes"):
-        mpi_command         = NUMACTL_CPU_BIND + " mpirun -np " + str(input_data.MPI_processes) + " ./ABEGPU"
+        mpi_command         = NUMACTL_CPU_BIND + " I_MPI_OFFLOAD=1 I_MPI_OFFLOAD_IPC=0 mpirun -np " + str(mpi_processes) + " ./ABE_CUDA"
         mpi_command_outfile = "ABEGPU_out.log"
+        mpi_env = _gpu_runtime_env()
+        started_mps = _start_cuda_mps_if_requested(mpi_env)
+        print(" GPU optimized runtime switches:")
+        print(f"   MPI processes={mpi_processes}")
+        print(f"   AMSS_INTERP_FAST={mpi_env.get('AMSS_INTERP_FAST', '')}")
+        print(f"   AMSS_INTERP_GPU={mpi_env.get('AMSS_INTERP_GPU', '')}")
+        print(f"   AMSS_ANALYSIS_MAP_EVERY={mpi_env.get('AMSS_ANALYSIS_MAP_EVERY', '')}")
+        print(f"   AMSS_EVOLVE_TIMING={mpi_env.get('AMSS_EVOLVE_TIMING', '')}")
+        print(f"   AMSS_ESCALAR_STEP_TIMING={mpi_env.get('AMSS_ESCALAR_STEP_TIMING', '')}")
+        print(f"   AMSS_CUDA_AWARE_MPI={mpi_env.get('AMSS_CUDA_AWARE_MPI', '')}")
+        print(f"   AMSS_CUDA_KEEP_RESIDENT_AFTER_STEP={mpi_env.get('AMSS_CUDA_KEEP_RESIDENT_AFTER_STEP', '')}")
+        print(f"   AMSS_CUDA_KEEP_ALL_LEVELS={mpi_env.get('AMSS_CUDA_KEEP_ALL_LEVELS', '')}")
+        print(f"   AMSS_CUDA_ESCALAR_KEEP_RESIDENT_AFTER_STEP={mpi_env.get('AMSS_CUDA_ESCALAR_KEEP_RESIDENT_AFTER_STEP', '')}")
+        print(f"   AMSS_CUDA_ESCALAR_KEEP_ALL_LEVELS={mpi_env.get('AMSS_CUDA_ESCALAR_KEEP_ALL_LEVELS', '')}")
+        print(f"   AMSS_CUDA_EM_CACHE_SOURCES={mpi_env.get('AMSS_CUDA_EM_CACHE_SOURCES', '')}")
+        print(f"   AMSS_CUDA_EM_ZERO_FASTPATH={mpi_env.get('AMSS_CUDA_EM_ZERO_FASTPATH', '')}")
+        print(f"   AMSS_EM_ZERO_ANALYSIS_FASTPATH={mpi_env.get('AMSS_EM_ZERO_ANALYSIS_FASTPATH', '')}")
+        print(f"   AMSS_EM_ZERO_RESIDENT_DOWNLOAD_FASTPATH={mpi_env.get('AMSS_EM_ZERO_RESIDENT_DOWNLOAD_FASTPATH', '')}")
+        print(f"   AMSS_CUDA_AMR_HOST_STAGED={mpi_env.get('AMSS_CUDA_AMR_HOST_STAGED', '')}")
+        print(f"   AMSS_CUDA_AMR_RESTRICT_DEVICE={mpi_env.get('AMSS_CUDA_AMR_RESTRICT_DEVICE', '')}")
+        print(f"   AMSS_CUDA_AMR_RESTRICT_BATCH={mpi_env.get('AMSS_CUDA_AMR_RESTRICT_BATCH', '')}")
+        print(f"   AMSS_CUDA_DEVICE_SEGMENT_BATCH={mpi_env.get('AMSS_CUDA_DEVICE_SEGMENT_BATCH', '')}")
+        print(f"   AMSS_CUDA_UNCACHED_DEVICE_BUFFERS={mpi_env.get('AMSS_CUDA_UNCACHED_DEVICE_BUFFERS', '')}")
+        print(f"   AMSS_SHELL_FAST_INTERP={mpi_env.get('AMSS_SHELL_FAST_INTERP', '')}")
+        print(f"   AMSS_SHELL_PARALLEL_INTERP={mpi_env.get('AMSS_SHELL_PARALLEL_INTERP', '')}")
+        print(f"   AMSS_SHELL_CUDA_INTERP={mpi_env.get('AMSS_SHELL_CUDA_INTERP', '')}")
+        print(f"   AMSS_SHELL_INTERP_THREADS={mpi_env.get('AMSS_SHELL_INTERP_THREADS', '')}")
+        print(f"   AMSS_Z4C_CUDA_RESIDENT={mpi_env.get('AMSS_Z4C_CUDA_RESIDENT', '')}")
+        print(f"   AMSS_CONSTRAINT_OUT_EVERY={mpi_env.get('AMSS_CONSTRAINT_OUT_EVERY', '')}")
+        if "CUDA_MPS_PIPE_DIRECTORY" in mpi_env:
+            print(f"   CUDA_MPS_PIPE_DIRECTORY={mpi_env['CUDA_MPS_PIPE_DIRECTORY']}")
  
-    ## Execute the MPI command and stream output
-    mpi_process = subprocess.Popen(mpi_command, shell=True, stdout=subprocess.PIPE, stderr=subprocess.STDOUT, text=True)
+    try:
+        ## Execute the MPI command and stream output
+        mpi_process = subprocess.Popen(mpi_command, shell=True, stdout=subprocess.PIPE,
+                                       stderr=subprocess.STDOUT, text=True, env=mpi_env)
 
-    ## Write ABE run output to file while printing to stdout
-    with open(mpi_command_outfile, 'w') as file0:  
-        ## Read and print output lines; also write each line to file
-        for line in mpi_process.stdout:
-            print(line, end='')  # stream output in real time
-            file0.write(line)    # write the line to file
-            file0.flush()        # flush to ensure each line is written immediately (optional)            
-    file0.close()
+        ## Write ABE run output to file while printing to stdout
+        with open(mpi_command_outfile, 'w') as file0:
+            ## Read and print output lines; also write each line to file
+            for line in mpi_process.stdout:
+                print(line, end='')  # stream output in real time
+                file0.write(line)    # write the line to file
+                file0.flush()        # flush to ensure each line is written immediately (optional)
 
-    ## Wait for the process to finish
-    mpi_return_code = mpi_process.wait()
+        ## Wait for the process to finish
+        mpi_return_code = mpi_process.wait()
+        if mpi_return_code != 0:
+            raise subprocess.CalledProcessError(mpi_return_code, mpi_command)
+    finally:
+        if started_mps:
+            _stop_cuda_mps(mpi_env)
     
     print(                                           )
     print( " The ABE/ABEGPU simulation is finished " ) 

plot_xiaoqu.py

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