fullwave25 1.2.6.dev3__tar.gz → 1.2.6.dev5__tar.gz

{fullwave25-1.2.6.dev3 → fullwave25-1.2.6.dev5}/.bumpversion.toml

@@ -1,5 +1,5 @@
 [tool.bumpversion]
-current_version = "1.2.6-dev3"
+current_version = "1.2.6-dev5"
 parse = """(?x)
     (?P<major>0|[1-9]\\d*)\\.
     (?P<minor>0|[1-9]\\d*)\\.

{fullwave25-1.2.6.dev3 → fullwave25-1.2.6.dev5}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: fullwave25
-Version: 1.2.6.dev3
+Version: 1.2.6.dev5
 Summary: Fullwave 2.5: Ultrasound wave propagation simulation with heterogeneous power law attenuation modelling capabilities
 Author: Masashi Sode, Gianmarco Pinton
 Maintainer: Masashi Sode
@@ -13,7 +13,6 @@ Classifier: Programming Language :: Python :: 3.11
 Classifier: Programming Language :: Python :: 3.12
 Classifier: Topic :: Scientific/Engineering :: Physics
 Requires-Python: >=3.10
-Requires-Dist: joblib>=1.5.3
 Requires-Dist: matplotlib>=3.10.7
 Requires-Dist: numba>=0.63.1
 Requires-Dist: numexpr>=2.14.1

{fullwave25-1.2.6.dev3 → fullwave25-1.2.6.dev5}/fullwave/__init__.py

@@ -60,7 +60,7 @@ try:
     __version__ = version("fullwave")
 except PackageNotFoundError:
     # Update via bump-my-version, not manually
-    __version__ = "1.2.6-dev3"
+    __version__ = "1.2.6-dev5"
 
 VERSION = __version__  # for convenience
 logger.info("Fullwave version: %s", __version__)

{fullwave25-1.2.6.dev3 → fullwave25-1.2.6.dev5}/fullwave/solver/launcher.py

@@ -31,6 +31,7 @@ class Launcher:
         use_gpu: bool = True,
         cuda_device_id: str | int | list | None = None,
         save_gpu_memory: bool = False,
+        verify_gpu: bool = True,
     ) -> None:
         """Initialize a FullwaveLauncher instance.
 
@@ -62,6 +63,11 @@ class Launcher:
             depending on the hardware and the simulation settings.
             useful in 3D simulations with large grid sizes
             where GPU memory is a limiting factor.
+        verify_gpu : bool, optional
+            Whether to verify that the specified CUDA devices exist on the system.
+            Defaults to True. Set to False when generating input files only
+            (``generate_input_only=True``) on a machine that may not have
+            the target GPUs available.
 
         """
         self._path_fullwave_simulation_bin = path_fullwave_simulation_bin
@@ -69,7 +75,10 @@ class Launcher:
         assert self._path_fullwave_simulation_bin.exists(), error_msg
         self.is_3d = is_3d
         self.use_gpu = use_gpu
-        self.cuda_device_id = self._configure_cuda_device_id(cuda_device_id)
+        self.cuda_device_id = self._configure_cuda_device_id(
+            cuda_device_id,
+            verify_gpu=verify_gpu,
+        )
         self.save_gpu_memory = save_gpu_memory
         logger.debug("Launcher instance created.")
 
@@ -152,13 +161,20 @@ class Launcher:
                 raise ValueError(message)
 
     @staticmethod
-    def _configure_cuda_device_id(cuda_device_id: str | int | list | None) -> str:
+    def _configure_cuda_device_id(
+        cuda_device_id: str | int | list | None,
+        *,
+        verify_gpu: bool = True,
+    ) -> str:
         """Verify and assign the CUDA device ID.
 
         Parameters
         ----------
         cuda_device_id : str | int | None
             The CUDA device ID to verify and assign.
+        verify_gpu : bool, optional
+            Whether to verify that the specified CUDA devices exist.
+            Defaults to True.
 
         Returns
         -------
@@ -167,7 +183,8 @@ class Launcher:
 
         """
         output = Launcher._parse_cuda_device_id(cuda_device_id)
-        Launcher._verify_cuda_devices_exist(output)
+        if verify_gpu:
+            Launcher._verify_cuda_devices_exist(output)
         return output
 
     def run(

{fullwave25-1.2.6.dev3 → fullwave25-1.2.6.dev5}/fullwave/solver/pml_builder.py

@@ -5,13 +5,11 @@ import logging
 from collections import OrderedDict
 from dataclasses import dataclass, field
 from functools import cached_property
-from itertools import starmap
 from pathlib import Path
 
 import matplotlib.pyplot as plt
 import numexpr as ne
 import numpy as np
-from joblib import Parallel, delayed
 from numpy.typing import NDArray
 
 import fullwave
@@ -834,21 +832,24 @@ class PMLBuilder:
 
         items = list(rename_dict.items())
 
-        results = Parallel(n_jobs=self.medium_org.n_jobs, backend="threading")(
-            delayed(_compute_one)(
-                key_fw2,
-                key_py,
-                relaxation_param_dict,
-                alpha_target_higher_nu,
-                d_target_higher_nu,
-                alpha_target_pml,
-                d_target_pml,
-                n_polynomial,
-                self.is_3d,
-                self._apply_transition_and_pml,
-            )
-            for key_fw2, key_py in items
-        )
+        with concurrent.futures.ThreadPoolExecutor() as executor:
+            futures = [
+                executor.submit(
+                    _compute_one,
+                    key_fw2,
+                    key_py,
+                    relaxation_param_dict,
+                    alpha_target_higher_nu,
+                    d_target_higher_nu,
+                    alpha_target_pml,
+                    d_target_pml,
+                    n_polynomial,
+                    self.is_3d,
+                    self._apply_transition_and_pml,
+                )
+                for key_fw2, key_py in items
+            ]
+            results = [f.result() for f in futures]
         out_dict = dict(results)
 
         logger.debug("Calculating PML a and b coefficients...")
@@ -876,13 +877,9 @@ class PMLBuilder:
             # Return keys + values so parent can update dict safely
             return (f"a_pml_{axis}{nu}", a, f"b_pml_{axis}{nu}", b)
 
-        results = Parallel(
-            n_jobs=self.medium_org.n_jobs,  # use all cores
-            backend="loky",  # process-based; safe default for Python code
-            prefer="processes",
-        )(
-            starmap(delayed(_worker), tasks),
-        )
+        with concurrent.futures.ThreadPoolExecutor() as executor:
+            futures = [executor.submit(_worker, nu, axis) for nu, axis in tasks]
+            results = [f.result() for f in futures]
 
         for a_key, a_val, b_key, b_val in results:
             out_dict[a_key] = a_val
@@ -1119,21 +1116,24 @@ class PMLBuilder:
             raise ValueError(error_msg)
 
         items = list(rename_dict.items())
-        results = Parallel(n_jobs=self.medium_org.n_jobs, backend="threading")(
-            delayed(_compute_one)(
-                key_fw2,
-                key_py,
-                relaxation_param_dict,
-                alpha_target_higher_nu,
-                d_target_higher_nu,
-                alpha_target_pml,
-                d_target_pml,
-                n_polynomial,
-                self.is_3d,
-                self._apply_transition_and_pml,
-            )
-            for key_fw2, key_py in items
-        )
+        with concurrent.futures.ThreadPoolExecutor() as executor:
+            futures = [
+                executor.submit(
+                    _compute_one,
+                    key_fw2,
+                    key_py,
+                    relaxation_param_dict,
+                    alpha_target_higher_nu,
+                    d_target_higher_nu,
+                    alpha_target_pml,
+                    d_target_pml,
+                    n_polynomial,
+                    self.is_3d,
+                    self._apply_transition_and_pml,
+                )
+                for key_fw2, key_py in items
+            ]
+            results = [f.result() for f in futures]
         out_dict = dict(results)
 
         logger.debug("Calculating PML a and b coefficients...")
@@ -1162,13 +1162,9 @@ class PMLBuilder:
             # Return keys + values so parent can update dict safely
             return (f"a_pml_{axis}{nu}", a, f"b_pml_{axis}{nu}", b)
 
-        results = Parallel(
-            n_jobs=self.medium_org.n_jobs,  # use all cores
-            backend="loky",  # process-based; safe default for Python code
-            prefer="processes",
-        )(
-            starmap(delayed(_worker), tasks),
-        )
+        with concurrent.futures.ThreadPoolExecutor() as executor:
+            futures = [executor.submit(_worker, nu, axis) for nu, axis in tasks]
+            results = [f.result() for f in futures]
 
         for a_key, a_val, b_key, b_val in results:
             out_dict[a_key] = a_val

{fullwave25-1.2.6.dev3 → fullwave25-1.2.6.dev5}/fullwave/solver/solver.py

@@ -329,6 +329,7 @@ class Solver:
         use_isotropic_relaxation: bool = True,
         cuda_device_id: str | int | list | None = None,
         save_gpu_memory: bool = False,
+        verify_gpu: bool = True,
     ) -> None:
         """Initialize a Solver instance for the fullwave simulation.
 
@@ -415,6 +416,11 @@ class Solver:
             depending on the hardware and the simulation settings.
             useful in 3D simulations with large grid sizes
             where GPU memory is a limiting factor.
+        verify_gpu : bool, optional
+            Whether to verify that the specified CUDA devices exist on the system.
+            Defaults to True. Set to False when generating input files only
+            (``generate_input_only=True``) on a machine that may not have
+            the target GPUs available.
 
         Raises
         ------
@@ -555,6 +561,7 @@ class Solver:
             use_gpu=self.use_gpu,
             cuda_device_id=self.cuda_device_id,
             save_gpu_memory=self.save_gpu_memory,
+            verify_gpu=verify_gpu,
         )
 
         if use_exponential_attenuation:
@@ -758,6 +765,7 @@ class Solver:
         release_after_write: bool = False,
         highpass_cutoff_mhz: float | None = None,
         bandpass_cutoff_mhz: tuple[float, float] | None = None,
+        gpu_memory_estimate: bool = True,
     ) -> NDArray[np.float64] | Path:
         r"""Run the fullwave simulation and return the result as a NumPy array.
 
@@ -829,6 +837,9 @@ class Solver:
             Uses cosine (Hann) tapers on both edges.
             Cannot be combined with ``highpass_cutoff_mhz``.
             Requires ``load_results=True``.  Default is ``None`` (no filtering).
+        gpu_memory_estimate : bool
+            Whether to estimate GPU memory usage before running the simulation.
+            Default is True. If True, it estimates the GPU memory usage.
 
         Returns
         -------
@@ -937,6 +948,9 @@ class Solver:
         )
         logger.debug(message)
 
+        if gpu_memory_estimate:
+            self._estimate_gpu_memory(sensor)
+
         if generate_input_only:
             logger.info(
                 "Input data generation completed in %s. Skipping simulation execution.",
@@ -975,6 +989,250 @@ class Solver:
         # which is a list of file names
         return sim_result
 
+    def _estimate_gpu_memory(
+        self,
+        sensor: fullwave.Sensor,
+    ) -> None:
+        """Estimate and log GPU memory usage per device.
+
+        Provides a pre-launch estimate so users can verify that the simulation
+        fits in GPU memory before the binary starts allocating.
+
+        Parameters
+        ----------
+        sensor : fullwave.Sensor
+            The sensor that will actually be written to the input files (may
+            differ from ``self.sensor`` when ``record_whole_domain=True``).
+
+        """
+        # show that this is an experimental feature
+        logger.info("Estimating GPU memory usage... (experimental feature, may be inaccurate)")
+        device_ids = self.fullwave_launcher.cuda_device_id.split(",")
+        n_gpus = len(device_ids)
+
+        grid = self.pml_builder.extended_grid
+        source = self.pml_builder.extended_source
+        medium = self.pml_builder.extended_medium
+
+        depth = grid.nx
+        lateral = grid.ny * grid.nz if self.is_3d else grid.ny
+        halo_depth = 8
+
+        float_bytes = 4
+        int_bytes = 4
+
+        c_map = medium.sound_speed
+        c_range = int(np.rint(c_map.max()) - np.rint(c_map.min()))
+        n_deriv_levels = 1 if c_range == 0 else c_range + 1
+
+        n_source_timesteps = source.icmat.shape[1]
+
+        gb = 1024.0**3
+
+        base_depth = depth // n_gpus
+        remainder = depth % n_gpus
+
+        for rank, dev_id in enumerate(device_ids):
+            depth_this = base_depth + (1 if rank < remainder else 0)
+
+            if n_gpus == 1:
+                n_halo_sides = 0
+            elif rank == 0 or rank == n_gpus - 1:
+                n_halo_sides = 1
+            else:
+                n_halo_sides = 2
+            local_depth = depth_this + n_halo_sides * halo_depth
+            slab = local_depth * lateral
+
+            n_sources = max(source.n_sources // n_gpus, 0)
+            n_sensors = max(sensor.n_sensors // n_gpus, 0)
+            n_air_local = max(medium.n_air // n_gpus, 0)
+
+            if self.use_exponential_attenuation:
+                total = self._mem_exponential(
+                    slab,
+                    n_deriv_levels,
+                    n_sources,
+                    n_source_timesteps,
+                    save_gpu_memory=self.save_gpu_memory,
+                    n_sensors=n_sensors,
+                    float_bytes=float_bytes,
+                    int_bytes=int_bytes,
+                    is_3d=self.is_3d,
+                )
+            else:
+                total = self._mem_relaxation(
+                    slab,
+                    n_deriv_levels,
+                    n_sources,
+                    n_source_timesteps,
+                    save_gpu_memory=self.save_gpu_memory,
+                    n_air=n_air_local,
+                    n_sensors=n_sensors,
+                    n_relax=self.n_relax_mechanisms,
+                    float_bytes=float_bytes,
+                    int_bytes=int_bytes,
+                    is_3d=self.is_3d,
+                )
+
+            mode = "exponential" if self.use_exponential_attenuation else "relaxation"
+            saving = ", save_gpu_memory=True" if self.save_gpu_memory else ""
+            logger.info(
+                "GPU memory estimate [GPU %s] (%s mode%s): "
+                "%.2f GB  (depth=%d +%d halo, lateral=%d)",
+                dev_id.strip(),
+                mode,
+                saving,
+                total / gb,
+                depth_this,
+                n_halo_sides * halo_depth,
+                lateral,
+            )
+
+    @staticmethod
+    def _mem_exponential(
+        slab: int,
+        n_deriv_levels: int,
+        n_sources: int,
+        n_source_timesteps: int,
+        *,
+        save_gpu_memory: bool,
+        n_sensors: int,
+        float_bytes: int,
+        int_bytes: int,
+        is_3d: bool,
+    ) -> int:
+        """Return estimated GPU bytes for exponential-attenuation solver.
+
+        Parameters
+        ----------
+        slab : int
+            Grid points per GPU slab (local_depth * lateral).
+        n_deriv_levels : int
+            Number of derivative-map levels.
+        n_sources : int
+            Approximate source count on this GPU.
+        n_source_timesteps : int
+            Number of source time steps.
+        save_gpu_memory : bool
+            Whether memory-saving mode is active.
+        n_sensors : int
+            Approximate sensor count on this GPU.
+        float_bytes : int
+            Bytes per float (4).
+        int_bytes : int
+            Bytes per int (4).
+        is_3d: bool,
+            Whether the simulation is 3D (affects sensor memory).
+
+        Returns
+        -------
+        int
+            Total estimated bytes.
+
+        """
+        fb = float_bytes
+        ib = int_bytes
+        ndim = 3 if is_3d else 2
+        n_fields = 4 if is_3d else 3  # p, u, [v], w
+
+        # wave fields: n_fields pairs x 2 time levels
+        mem = n_fields * 2 * slab * fb
+        # material: rho + K + beta + a_exp
+        mem += 4 * slab * fb
+        # derivative maps (dmap + dcmap)
+        mem += 9 * 2 * n_deriv_levels * fb + slab * ib
+        # source (icmat + coords)
+        if n_sources > 0:
+            mem += n_sources * fb if save_gpu_memory else n_sources * n_source_timesteps * fb
+            mem += ndim * n_sources * ib
+        # sensor (genoutframe + coordsout_local + p_idx_array)
+        if n_sensors > 0:
+            mem += n_sensors * fb
+            mem += (ndim + 1) * n_sensors * ib
+            mem += n_sensors * ib
+        return mem
+
+    @staticmethod
+    def _mem_relaxation(
+        slab: int,
+        n_deriv_levels: int,
+        n_sources: int,
+        n_source_timesteps: int,
+        *,
+        save_gpu_memory: bool,
+        n_air: int,
+        n_sensors: int,
+        n_relax: int,
+        float_bytes: int,
+        int_bytes: int,
+        is_3d: bool,
+    ) -> int:
+        """Return estimated GPU bytes for relaxation (power-law) solver.
+
+        Parameters
+        ----------
+        slab : int
+            Grid points per GPU slab (local_depth * lateral).
+        n_deriv_levels : int
+            Number of derivative-map levels.
+        n_sources : int
+            Approximate source count on this GPU.
+        n_source_timesteps : int
+            Number of source time steps.
+        save_gpu_memory : bool
+            Whether memory-saving mode is active.
+        n_air : int
+            Number of zero-pressure (air) coordinates on this GPU.
+        n_sensors : int
+            Approximate sensor count on this GPU.
+        n_relax : int
+            Number of relaxation mechanisms.
+        float_bytes : int
+            Bytes per float (4).
+        int_bytes : int
+            Bytes per int (4).
+        is_3d: bool,
+            Whether the simulation is 3D (affects sensor memory).
+
+        Returns
+        -------
+        int
+            Total estimated bytes.
+
+        """
+        fb = float_bytes
+        ib = int_bytes
+        ndim = 3 if is_3d else 2
+        n_fields = 4 if is_3d else 3  # p, u, [v], w
+
+        # wave fields: n_fields pairs x 2 time levels
+        mem = n_fields * 2 * slab * fb
+        # relaxation psi:
+        mem += 2 * (ndim * n_relax * 2 * slab * fb)
+        # material: rho + K + beta
+        mem += 3 * slab * fb
+        # kappa: 2 arrays (kappa_x1, kappa_x2)
+        mem += 2 * slab * fb
+        # PML: pml_x1 + pml_x2, each has 2 * n_relax arrays
+        mem += 2 * (2 * n_relax) * slab * fb
+        # (dmap + dcmap)
+        mem += 9 * 2 * n_deriv_levels * fb + slab * ib
+        # source (icmat + coords)
+        if n_sources > 0:
+            mem += n_sources * fb if save_gpu_memory else n_sources * n_source_timesteps * fb
+            mem += ndim * n_sources * ib
+
+        # air
+        if n_air > 0:
+            mem += ndim * n_air * ib
+        # sensor
+        if n_sensors > 0:
+            mem += n_sensors * fb
+            mem += (ndim + 1) * n_sensors * ib
+            mem += n_sensors * ib
+        return mem
+
     def print_info(self) -> None:
         """Print the Solver instance information."""
         print(str(self))

{fullwave25-1.2.6.dev3 → fullwave25-1.2.6.dev5}/pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "fullwave25"
-version = "1.2.6-dev3" # Update via bump-my-version, not manually
+version = "1.2.6-dev5" # Update via bump-my-version, not manually
 description = "Fullwave 2.5: Ultrasound wave propagation simulation with heterogeneous power law attenuation modelling capabilities"
 readme = "README.md"
 requires-python = ">=3.10"
@@ -12,7 +12,6 @@ dependencies = [
     "opencv-python-headless>=4.12.0.88",
     "tomli>=2.3.0",
     "numba>=0.63.1",
-    "joblib>=1.5.3",
     "numexpr>=2.14.1",
 ]
 authors = [{ name = "Masashi Sode" }, { name = "Gianmarco Pinton" }]
@@ -57,4 +56,13 @@ build-backend = "hatchling.build"
 packages = ["fullwave", "fullwave.utils"]
 
 [tool.hatch.build]
-exclude = ["tests", "figs", "examples", ".vscode", ".github", "fullwave/solver/bins/gpu", "fullwave/solver/bins/exponential_attenuation", "fullwave/solver/bins/_exponential_attenuation"]
+exclude = [
+    "tests",
+    "figs",
+    "examples",
+    ".vscode",
+    ".github",
+    "fullwave/solver/bins/gpu",
+    "fullwave/solver/bins/exponential_attenuation",
+    "fullwave/solver/bins/_exponential_attenuation",
+]

{fullwave25-1.2.6.dev3 → fullwave25-1.2.6.dev5}/ruff.toml

@@ -77,6 +77,7 @@ dummy-variable-rgx = "^(_+|(_+[a-zA-Z0-9_]*[a-zA-Z0-9]+?))$"
 "tests/**" = [
     "D100",
     "D101",
+    "D102",
     "D103",
     "D104",
     "D107",

{fullwave25-1.2.6.dev3 → fullwave25-1.2.6.dev5}/uv.lock

@@ -735,10 +735,9 @@ wheels = [
 
 [[package]]
 name = "fullwave25"
-version = "1.2.6.dev3"
+version = "1.2.6.dev5"
 source = { editable = "." }
 dependencies = [
-    { name = "joblib" },
     { name = "matplotlib" },
     { name = "numba" },
     { name = "numexpr" },
@@ -780,7 +779,6 @@ dev = [
 requires-dist = [
     { name = "cupy-cuda12x", marker = "extra == 'examples'", specifier = ">=13.6.0" },
     { name = "ipykernel", marker = "extra == 'examples'", specifier = ">=6.28.0" },
-    { name = "joblib", specifier = ">=1.5.3" },
     { name = "jupyter", marker = "extra == 'examples'", specifier = ">=1.0.0" },
     { name = "line-profiler", marker = "extra == 'dev'", specifier = ">=5.0.1" },
     { name = "mach-beamform", marker = "extra == 'examples'", specifier = ">=0.0.4" },
@@ -1034,15 +1032,6 @@ wheels = [
     { url = "https://files.pythonhosted.org/packages/62/a1/3d680cbfd5f4b8f15abc1d571870c5fc3e594bb582bc3b64ea099db13e56/jinja2-3.1.6-py3-none-any.whl", hash = "sha256:85ece4451f492d0c13c5dd7c13a64681a86afae63a5f347908daf103ce6d2f67", size = 134899, upload-time = "2025-03-05T20:05:00.369Z" },
 ]
 
-[[package]]
-name = "joblib"
-version = "1.5.3"
-source = { registry = "https://pypi.org/simple" }
-sdist = { url = "https://files.pythonhosted.org/packages/41/f2/d34e8b3a08a9cc79a50b2208a93dce981fe615b64d5a4d4abee421d898df/joblib-1.5.3.tar.gz", hash = "sha256:8561a3269e6801106863fd0d6d84bb737be9e7631e33aaed3fb9ce5953688da3", size = 331603, upload-time = "2025-12-15T08:41:46.427Z" }
-wheels = [
-    { url = "https://files.pythonhosted.org/packages/7b/91/984aca2ec129e2757d1e4e3c81c3fcda9d0f85b74670a094cc443d9ee949/joblib-1.5.3-py3-none-any.whl", hash = "sha256:5fc3c5039fc5ca8c0276333a188bbd59d6b7ab37fe6632daa76bc7f9ec18e713", size = 309071, upload-time = "2025-12-15T08:41:44.973Z" },
-]
-
 [[package]]
 name = "json5"
 version = "0.12.1"