warp-lang 1.7.0__py3-none-manylinux_2_34_aarch64.whl

warp/fem/adaptivity.py

@@ -0,0 +1,508 @@
+# SPDX-FileCopyrightText: Copyright (c) 2024 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
+# SPDX-License-Identifier: Apache-2.0
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+# http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from typing import List, Optional, Tuple
+
+import numpy as np
+
+import warp as wp
+from warp.fem import cache
+from warp.fem.domain import Cells
+from warp.fem.field import GeometryField
+from warp.fem.geometry import AdaptiveNanogrid
+from warp.fem.integrate import interpolate
+from warp.fem.operator import integrand, lookup
+from warp.fem.types import NULL_ELEMENT_INDEX, Domain, Field, Sample
+
+
+def adaptive_nanogrid_from_hierarchy(
+    grids: List[wp.Volume], grading: Optional[str] = None, temporary_store: Optional[cache.TemporaryStore] = None
+) -> AdaptiveNanogrid:
+    """
+    Constructs a :class:`warp.fem.AdaptiveNanogrid` from a non-overlapping grid hierarchy.
+
+    Warning: The result is undefined if there are partial overlaps between levels, that is if a cell at level `l` is only partially covered by cells at levels `l-1` or lower.
+
+    Args:
+        grids: List of sparse Volumes, from finest to coarsest
+        grading: Supplementary grading condition, may be ``None``, "face" or "vertex"; see :func:`enforce_nanogrid_grading`
+        temporary_store: Storage for temporary allocations
+    """
+    if not grids:
+        raise ValueError("No grids to build from!")
+
+    level_count = len(grids)
+
+    device = grids[0].device
+
+    # Concatenate voxels for each grid
+    voxel_counts = [grid.get_voxel_count() for grid in grids]
+
+    voxel_offsets = np.cumsum(np.array([0] + voxel_counts))
+    merged_ijks = cache.borrow_temporary(temporary_store, dtype=wp.vec3i, shape=int(voxel_offsets[-1]), device=device)
+    for l in range(level_count):
+        voxel_count = voxel_counts[l]
+        grid_voxels = cache.borrow_temporary(temporary_store, shape=(voxel_count,), dtype=wp.vec3i, device=device)
+        grids[l].get_voxels(out=grid_voxels.array)
+
+        wp.launch(
+            _fill_hierarchy_merged_ijk,
+            dim=voxel_count,
+            device=device,
+            inputs=[l, voxel_offsets[l], grid_voxels.array, merged_ijks.array],
+        )
+
+    # Allocate merged grid
+    grid_info = grids[0].get_grid_info()
+    cell_grid = wp.Volume.allocate_by_voxels(
+        merged_ijks.array,
+        transform=grid_info.transform_matrix,
+        translation=grid_info.translation,
+        device=device,
+    )
+
+    # Get unique voxel and corresponding level
+    cell_count = cell_grid.get_voxel_count()
+    cell_ijk = cache.borrow_temporary(temporary_store, shape=(cell_count,), dtype=wp.vec3i, device=device)
+    cell_level = wp.array(shape=(cell_count,), dtype=wp.uint8, device=device)
+
+    cell_grid.get_voxels(out=cell_ijk.array)
+
+    cell_grid_ids = wp.array([grid.id for grid in grids], dtype=wp.uint64, device=device)
+    wp.launch(
+        _fill_hierarchy_cell_level,
+        device=device,
+        dim=cell_count,
+        inputs=[level_count, cell_grid_ids, cell_ijk.array, cell_level],
+    )
+
+    cell_grid, cell_level = enforce_nanogrid_grading(
+        cell_grid, cell_level, level_count=level_count, grading=grading, temporary_store=temporary_store
+    )
+
+    return AdaptiveNanogrid(cell_grid, cell_level=cell_level, level_count=level_count, temporary_store=temporary_store)
+
+
+def adaptive_nanogrid_from_field(
+    coarse_grid: wp.Volume,
+    level_count: int,
+    refinement_field: GeometryField,
+    samples_per_voxel: int = 64,
+    grading: Optional[str] = None,
+    temporary_store: Optional[cache.TemporaryStore] = None,
+) -> AdaptiveNanogrid:
+    """
+    Constructs a :class:`warp.fem.AdaptiveNanogrid` from a coarse grid and a refinement field.
+
+    Args:
+        coarse_grid: Base grid from which to start refining. No voxels will be added outside of the base grid.
+        level_count: Maximum number of refinement levels
+        refinement_field: Scalar field used as a refinement oracle. If the returned value is negative, the corresponding voxel will be carved out.
+            Positive values indicate the desired refinement with 0.0 corresponding to the finest level and 1.0 to the coarsest level.
+        samples_per_voxel: How many samples to use for evaluating the refinement field within each voxel
+        grading: Supplementary grading condition, may be ``None``, "face" or "vertex"; see :func:`enforce_nanogrid_grading`
+        temporary_store: Storage for temporary allocations
+    """
+
+    device = coarse_grid.device
+    cell_count = coarse_grid.get_voxel_count()
+
+    cell_ijk = cache.borrow_temporary(temporary_store, shape=(cell_count,), dtype=wp.vec3i, device=device)
+    cell_level = cache.borrow_temporary(temporary_store, shape=(cell_count,), dtype=wp.uint8, device=device)
+
+    cell_level.array.fill_(level_count - 1)
+    coarse_grid.get_voxels(out=cell_ijk.array)
+
+    domain = Cells(refinement_field.geometry)
+
+    fine_count = cache.borrow_temporary(temporary_store, dtype=int, shape=1, device=device)
+    fine_count.array.zero_()
+
+    for _ in range(level_count):
+        cell_count = cell_ijk.array.shape[0]
+        cell_refinement = cache.borrow_temporary(temporary_store, shape=(cell_count,), dtype=wp.int8, device=device)
+
+        with wp.ScopedDevice(device):
+            interpolate(
+                _count_refined_voxels,
+                domain=domain,
+                dim=cell_count,
+                fields={"field": refinement_field},
+                values={
+                    "sample_count": samples_per_voxel,
+                    "level_count": level_count,
+                    "coarse_grid": coarse_grid.id,
+                    "coarse_ijk": cell_ijk.array,
+                    "coarse_level": cell_level.array,
+                    "coarse_refinement": cell_refinement.array,
+                    "fine_count": fine_count.array,
+                },
+            )
+
+        fine_shape = int(fine_count.array.numpy()[0])
+        fine_ijk = cache.borrow_temporary(temporary_store, shape=fine_shape, dtype=wp.vec3i, device=device)
+        fine_level = cache.borrow_temporary(temporary_store, shape=fine_shape, dtype=wp.uint8, device=device)
+
+        wp.launch(
+            _fill_refined_voxels,
+            dim=cell_count,
+            device=device,
+            inputs=[
+                cell_ijk.array,
+                cell_level.array,
+                cell_refinement.array,
+                fine_count.array,
+                fine_ijk.array,
+                fine_level.array,
+            ],
+        )
+
+        # Fine is the new coarse
+        cell_ijk = fine_ijk
+        cell_level = fine_level
+
+    wp.launch(_adjust_refined_ijk, dim=fine_shape, device=device, inputs=[cell_ijk.array, cell_level.array])
+
+    # We now have our refined voxels, allocate the grid
+    coarse_info = coarse_grid.get_grid_info()
+    fine_scale = 1.0 / (1 << (level_count - 1))
+    fine_transform = coarse_info.transform_matrix * fine_scale
+    fine_translation = coarse_info.translation + (fine_scale - 1.0) * 0.5 * wp.vec3(coarse_grid.get_voxel_size())
+    fine_grid = wp.Volume.allocate_by_voxels(
+        cell_ijk.array, translation=fine_translation, transform=fine_transform, device=device
+    )
+
+    # Reorder cell_levels (voxels will have moved)
+    fine_count = fine_grid.get_voxel_count()
+    fine_level = wp.array(dtype=wp.uint8, shape=fine_count, device=device)
+    wp.launch(
+        _fill_refined_level,
+        dim=fine_count,
+        device=device,
+        inputs=[fine_grid.id, cell_ijk.array, cell_level.array, fine_level],
+    )
+
+    fine_grid, fine_level = enforce_nanogrid_grading(
+        fine_grid, fine_level, level_count=level_count, grading=grading, temporary_store=temporary_store
+    )
+
+    return AdaptiveNanogrid(fine_grid, cell_level=fine_level, level_count=level_count, temporary_store=temporary_store)
+
+
+def enforce_nanogrid_grading(
+    cell_grid: wp.Volume,
+    cell_level: wp.array,
+    level_count: int,
+    grading: Optional[str] = None,
+    temporary_store: Optional[cache.TemporaryStore] = None,
+) -> Tuple[wp.Volume, wp.array]:
+    """
+    Refines an adaptive grid such that if satisfies a grading condition.
+
+    Arguments are similar to the :class:`warp.fem.AdaptiveNanogrid` constructor, with the
+    addition of the `grading` condition which can be:
+         - "face": two cells sharing a common face must have a level difference of at most 1
+         - "vertex": two cells sharing a common vertex must have a level difference of at most 1
+         - "none" or ``None``: no grading condition
+
+    Returns the refined grid and levels
+    """
+
+    if not grading or grading == "none" or level_count <= 2:
+        # skip
+        return cell_grid, cell_level
+
+    device = cell_grid.device
+
+    grid_info = cell_grid.get_grid_info()
+
+    fine_count = cache.borrow_temporary(temporary_store, shape=(1,), dtype=int, device=device)
+    grading_kernel = _count_ungraded_faces if grading == "face" else _count_ungraded_vertices
+
+    for _ in range(level_count - 2):
+        cell_count = cell_grid.get_voxel_count()
+        cell_ijk = cache.borrow_temporary(temporary_store, shape=(cell_count,), dtype=wp.vec3i, device=device)
+        cell_grid.get_voxels(out=cell_ijk.array)
+
+        refinement = cache.borrow_temporary(temporary_store, shape=(cell_count,), dtype=int, device=device)
+        refinement.array.zero_()
+
+        wp.launch(
+            grading_kernel,
+            dim=cell_count,
+            device=device,
+            inputs=[cell_grid.id, cell_ijk.array, cell_level, level_count, refinement.array],
+        )
+
+        fine_count.array.fill_(cell_count)
+        wp.launch(
+            _count_graded_cells,
+            dim=cell_count,
+            device=device,
+            inputs=[
+                refinement.array,
+                fine_count.array,
+            ],
+        )
+
+        # Add new coordinates
+        fine_shape = int(fine_count.array.numpy()[0])
+        if fine_shape == cell_count:
+            break
+
+        fine_ijk = cache.borrow_temporary(temporary_store, shape=fine_shape, dtype=wp.vec3i, device=device)
+        fine_level = cache.borrow_temporary(temporary_store, shape=fine_shape, dtype=wp.uint8, device=device)
+
+        wp.launch(
+            _fill_graded_cells,
+            dim=cell_count,
+            device=device,
+            inputs=[
+                cell_ijk.array,
+                cell_level,
+                refinement.array,
+                fine_count.array,
+                fine_ijk.array,
+                fine_level.array,
+            ],
+        )
+
+        # Rebuild grid and levels
+        cell_grid = wp.Volume.allocate_by_voxels(
+            fine_ijk.array, translation=grid_info.translation, transform=grid_info.transform_matrix, device=device
+        )
+        cell_level = wp.empty(fine_shape, dtype=wp.uint8, device=device)
+        wp.launch(
+            _fill_refined_level,
+            dim=fine_shape,
+            device=device,
+            inputs=[cell_grid.id, fine_ijk.array, fine_level.array, cell_level],
+        )
+
+    return cell_grid, cell_level
+
+
+@wp.kernel
+def _count_ungraded_faces(
+    cell_grid: wp.uint64,
+    cell_ijk: wp.array(dtype=wp.vec3i),
+    cell_level: wp.array(dtype=wp.uint8),
+    level_count: int,
+    refinement: wp.array(dtype=wp.int32),
+):
+    cell = wp.tid()
+
+    ijk = cell_ijk[cell]
+    level = int(cell_level[cell])
+
+    for axis in range(3):
+        for j in range(2):
+            nijk = ijk
+            nijk[axis] += (2 * j - 1) << level
+
+            nidx = AdaptiveNanogrid.find_cell(cell_grid, nijk, level_count, cell_level)
+            if nidx != -1:
+                n_level = cell_level[nidx]
+                if n_level > (level + 1):
+                    wp.atomic_add(refinement, nidx, 1)
+
+
+@wp.kernel
+def _count_ungraded_vertices(
+    cell_grid: wp.uint64,
+    cell_ijk: wp.array(dtype=wp.vec3i),
+    cell_level: wp.array(dtype=wp.uint8),
+    level_count: int,
+    refinement: wp.array(dtype=wp.int32),
+):
+    cell = wp.tid()
+
+    ijk = cell_ijk[cell]
+    level = int(cell_level[cell])
+
+    for i in range(-1, 2):
+        for j in range(-1, 2):
+            for k in range(-1, 2):
+                nijk = ijk + AdaptiveNanogrid.fine_ijk(wp.vec3i(i, j, k), level)
+
+                nidx = AdaptiveNanogrid.find_cell(cell_grid, nijk, level_count, cell_level)
+                if nidx != -1:
+                    n_level = cell_level[nidx]
+                    if n_level > (level + 1):
+                        wp.atomic_add(refinement, nidx, 1)
+
+
+@wp.kernel
+def _count_graded_cells(
+    refinement: wp.array(dtype=wp.int32),
+    fine_count: wp.array(dtype=int),
+):
+    cell = wp.tid()
+    if refinement[cell] > 0:
+        wp.atomic_add(fine_count, 0, 7)
+
+
+@wp.kernel
+def _fill_graded_cells(
+    coarse_ijk: wp.array(dtype=wp.vec3i),
+    coarse_level: wp.array(dtype=wp.uint8),
+    coarse_refinement: wp.array(dtype=wp.int32),
+    fine_count: wp.array(dtype=int),
+    fine_ijk: wp.array(dtype=wp.vec3i),
+    fine_level: wp.array(dtype=wp.uint8),
+):
+    cell = wp.tid()
+    ijk = coarse_ijk[cell]
+    level = int(coarse_level[cell])
+    refinement = wp.min(1, coarse_refinement[cell])
+
+    count = wp.where(refinement > 0, 8, 1)
+    offset = wp.atomic_sub(fine_count, 0, count) - count
+
+    f_level = level - refinement
+    for k in range(count):
+        f_ijk = ijk + AdaptiveNanogrid.fine_ijk(wp.vec3i(k >> 2, (k & 2) >> 1, k & 1), f_level)
+        fine_ijk[offset + k] = f_ijk
+        fine_level[offset + k] = wp.uint8(f_level)
+
+
+@integrand
+def _sample_refinement(
+    rng: wp.uint32,
+    sample_count: int,
+    ijk: wp.vec3i,
+    domain: Domain,
+    field: Field,
+    coarse_grid: wp.uint64,
+    cur_level: int,
+    level_count: int,
+):
+    min_level = level_count
+
+    scale = 1.0 / float(1 << (level_count - 1 - cur_level))
+    uvw = wp.vec3(ijk) * scale + wp.vec3(0.5 * (scale - 1.0))
+
+    for _ in range(sample_count):
+        trial_uvw = uvw + wp.sample_unit_cube(rng) * scale
+        pos = wp.volume_index_to_world(coarse_grid, trial_uvw)
+        field_s = lookup(domain, pos)
+        if field_s.element_index != NULL_ELEMENT_INDEX:
+            sampled_level = wp.min(level_count - 1, int(wp.floor(field(field_s) * float(level_count))))
+            if sampled_level >= 0:
+                min_level = wp.min(sampled_level, min_level)
+
+    return wp.where(min_level < level_count, cur_level - wp.clamp(min_level, 0, cur_level), -1)
+
+
+@integrand
+def _count_refined_voxels(
+    s: Sample,
+    domain: Domain,
+    field: Field,
+    sample_count: int,
+    level_count: int,
+    coarse_grid: wp.uint64,
+    coarse_ijk: wp.array(dtype=wp.vec3i),
+    coarse_level: wp.array(dtype=wp.uint8),
+    coarse_refinement: wp.array(dtype=wp.int8),
+    fine_count: wp.array(dtype=int),
+):
+    cell = s.qp_index
+
+    ijk = coarse_ijk[cell]
+    cur_level = int(coarse_level[cell])
+
+    seed = (cur_level << 24) ^ (ijk[0] << 12) ^ (ijk[1] << 6) ^ ijk[2]
+    rng = wp.rand_init(seed)
+
+    refinement = _sample_refinement(rng, sample_count, ijk, domain, field, coarse_grid, cur_level, level_count)
+
+    coarse_refinement[cell] = wp.int8(refinement)
+    if refinement >= 0:
+        wp.atomic_add(fine_count, 0, wp.where(refinement > 0, 8, 1))
+
+
+@wp.kernel
+def _fill_refined_voxels(
+    coarse_ijk: wp.array(dtype=wp.vec3i),
+    coarse_level: wp.array(dtype=wp.uint8),
+    coarse_refinement: wp.array(dtype=wp.int8),
+    fine_count: wp.array(dtype=int),
+    fine_ijk: wp.array(dtype=wp.vec3i),
+    fine_level: wp.array(dtype=wp.uint8),
+):
+    cell = wp.tid()
+    ijk = coarse_ijk[cell]
+    level = int(coarse_level[cell])
+    refinement = wp.min(1, int(coarse_refinement[cell]))
+
+    if refinement >= 0:
+        count = wp.where(refinement > 0, 8, 1)
+        offset = wp.atomic_sub(fine_count, 0, count) - count
+
+        f_level = level - refinement
+        for k in range(count):
+            f_ijk = AdaptiveNanogrid.fine_ijk(ijk, refinement) + wp.vec3i(k >> 2, (k & 2) >> 1, k & 1)
+            fine_ijk[offset + k] = f_ijk
+            fine_level[offset + k] = wp.uint8(f_level)
+
+
+@wp.kernel
+def _adjust_refined_ijk(
+    cell_ijk: wp.array(dtype=wp.vec3i),
+    cell_level: wp.array(dtype=wp.uint8),
+):
+    cell = wp.tid()
+    cell_ijk[cell] = AdaptiveNanogrid.fine_ijk(cell_ijk[cell], int(cell_level[cell]))
+
+
+@wp.kernel
+def _fill_refined_level(
+    fine_grid: wp.uint64,
+    cell_ijk: wp.array(dtype=wp.vec3i),
+    cell_level: wp.array(dtype=wp.uint8),
+    fine_level: wp.array(dtype=wp.uint8),
+):
+    cell = wp.tid()
+    ijk = cell_ijk[cell]
+    level = cell_level[cell]
+
+    fine_level[wp.volume_lookup_index(fine_grid, ijk[0], ijk[1], ijk[2])] = level
+
+
+@wp.kernel
+def _fill_hierarchy_merged_ijk(
+    level: int, level_offset: int, level_ijk: wp.array(dtype=wp.vec3i), merged_ijk: wp.array(dtype=wp.vec3i)
+):
+    cell = wp.tid()
+    merged_ijk[cell + level_offset] = AdaptiveNanogrid.fine_ijk(level_ijk[cell], level)
+
+
+@wp.kernel
+def _fill_hierarchy_cell_level(
+    level_count: int,
+    grids: wp.array(dtype=wp.uint64),
+    cells_ijk: wp.array(dtype=wp.vec3i),
+    levels: wp.array(dtype=wp.uint8),
+):
+    cell = wp.tid()
+    ijk = cells_ijk[cell]
+
+    for k in range(level_count):
+        idx = wp.volume_lookup_index(grids[k], ijk[0], ijk[1], ijk[2])
+        if idx != -1:
+            levels[cell] = wp.uint8(k)
+            break
+        ijk = AdaptiveNanogrid.coarse_ijk(ijk, 1)