warp-lang 1.8.1__py3-none-manylinux_2_34_aarch64.whl → 1.9.0__py3-none-manylinux_2_34_aarch64.whl

warp/examples/core/example_marching_cubes.py

@@ -22,6 +22,7 @@
 # Note: requires a CUDA-capable device
 ###########################################################################
 
+
 import warp as wp
 import warp.render
 

warp/examples/core/example_render_opengl.py

@@ -18,6 +18,7 @@
 #
 # Demonstrates how to set up tiled rendering and retrieves the pixels from
 # OpenGLRenderer as a Warp array while keeping all memory on the GPU.
+# It also shows how to add an ImGui UI to the renderer.
 #
 ###########################################################################
 
@@ -25,14 +26,95 @@ import numpy as np
 
 import warp as wp
 import warp.render
+from warp.render.imgui_manager import ImGuiManager
+
+
+class ExampleImGuiManager(ImGuiManager):
+    """An example ImGui manager that displays a few float values."""
+
+    def __init__(self, renderer, window_pos=(10, 10), window_size=(300, 400)):
+        super().__init__(renderer)
+        if not self.is_available:
+            return
+
+        # UI properties
+        self.window_pos = window_pos
+        self.window_size = window_size
+
+        # Values to display in the UI
+        self.some_float = 123.456
+        self.editable_float1 = 10.0
+        self.editable_float2 = 20.0
+        self.editable_float3 = 30.0
+        self.editable_vec2 = wp.vec2(0.5, 1.2)
+        self.editable_vec3 = wp.vec3(2.1, 3.4, 4.7)
+        self.editable_vec4 = wp.vec4(1.5, 3.2, 4.8, 6.1)
+        self.warp_array_float = wp.array([0.7, 1.4, 2.8], dtype=float)
+        self.warp_array_vec2 = wp.array([wp.vec2(1.1, 2.3), wp.vec2(3.4, 4.2), wp.vec2(5.6, 6.9)], dtype=wp.vec2)
+        self.warp_array_vec3 = wp.array(
+            [wp.vec3(0.5, 1.7, 2.9), wp.vec3(3.2, 4.8, 5.1), wp.vec3(6.4, 7.6, 8.3)], dtype=wp.vec3
+        )
+        self.warp_array_vec4 = wp.array([wp.vec4(1.2, 2.4, 3.6, 4.8), wp.vec4(5.1, 6.3, 7.5, 8.7)], dtype=wp.vec4)
+
+    def draw_ui(self):
+        # set window position and size once
+        self.imgui.set_next_window_size(self.window_size[0], self.window_size[1], self.imgui.ONCE)
+        self.imgui.set_next_window_position(self.window_pos[0], self.window_pos[1], self.imgui.ONCE)
+
+        self.imgui.begin("Warp Float Values")
+
+        self.imgui.text(f"A read-only float: {self.some_float}")
+        self.imgui.separator()
+
+        self.imgui.text("Editable floats:")
+        changed1, self.editable_float1 = self.imgui.slider_float("Slider", self.editable_float1, 0.0, 100.0)
+        changed2, self.editable_float2 = self.imgui.drag_float("Drag", self.editable_float2, 0.1, 0.0, 100.0)
+        changed3, self.editable_float3 = self.imgui.input_float("Input", self.editable_float3)
+
+        changed, self.editable_vec2 = self.drag_vec2("Vec2", self.editable_vec2)
+        changed, self.editable_vec3 = self.drag_vec3("Vec3", self.editable_vec3)
+        changed, self.editable_vec4 = self.drag_vec4("Vec4", self.editable_vec4)
+
+        changed, self.warp_array_float = self.drag_float_list("Float", self.warp_array_float)
+        changed, self.warp_array_vec2 = self.drag_vec2_list("Vec2", self.warp_array_vec2)
+        changed, self.warp_array_vec3 = self.drag_vec3_list("Vec3", self.warp_array_vec3)
+        changed, self.warp_array_vec4 = self.drag_vec4_list("Vec4", self.warp_array_vec4)
+
+        self.imgui.separator()
+        self.imgui.text("File Dialog Examples:")
+
+        if self.imgui.button("Open File"):
+            file_path = self.open_load_file_dialog(
+                title="Select a File", filetypes=[("Text Files", "*.txt"), ("All Files", "*.*")]
+            )
+            if file_path:
+                print(f"Selected file to open: {file_path}")
+
+        if self.imgui.button("Save File"):
+            file_path = self.open_save_file_dialog(
+                title="Save As", defaultextension=".txt", filetypes=[("Text Files", "*.txt"), ("All Files", "*.*")]
+            )
+            if file_path:
+                print(f"Selected file to save: {file_path}")
+
+        self.imgui.end()
 
 
 class Example:
-    def __init__(self, num_tiles=4, custom_tile_arrangement=False):
+    def __init__(self, num_tiles=4, custom_tile_arrangement=False, use_imgui=True):
         if num_tiles < 1:
             raise ValueError("num_tiles must be greater than or equal to 1.")
 
         self.renderer = wp.render.OpenGLRenderer(vsync=False)
+        self.use_imgui = use_imgui
+
+        if self.use_imgui:
+            self.imgui_manager = ExampleImGuiManager(self.renderer)
+            if self.imgui_manager.is_available:
+                self.renderer.render_2d_callbacks.append(self.imgui_manager.render_frame)
+            else:
+                self.use_imgui = False
+
         instance_ids = []
 
         if custom_tile_arrangement:
@@ -81,6 +163,11 @@ class Example:
         )
         self.renderer.end_frame()
 
+    def clear(self):
+        if self.use_imgui:
+            self.imgui_manager.shutdown()
+        self.renderer.clear()
+
 
 if __name__ == "__main__":
     import argparse
@@ -103,11 +190,21 @@ if __name__ == "__main__":
         help="Whether to split tiles into subplots when --show_plot is True.",
     )
     parser.add_argument("--custom_tile_arrangement", action="store_true", help="Apply custom tile arrangement.")
+    parser.add_argument(
+        "--use_imgui",
+        type=lambda x: bool(distutils.util.strtobool(x.strip())),
+        default=True,
+        help="Enable or disable the ImGui window.",
+    )
 
     args = parser.parse_known_args()[0]
 
     with wp.ScopedDevice(args.device):
-        example = Example(num_tiles=args.num_tiles, custom_tile_arrangement=args.custom_tile_arrangement)
+        example = Example(
+            num_tiles=args.num_tiles,
+            custom_tile_arrangement=args.custom_tile_arrangement,
+            use_imgui=args.use_imgui,
+        )
 
         channels = 1 if args.render_mode == "depth" else 3
 
@@ -190,4 +287,4 @@ if __name__ == "__main__":
                 fig.canvas.draw()
                 fig.canvas.flush_events()
 
-        example.renderer.clear()
+        example.clear()

warp/examples/fem/example_apic_fluid.py

@@ -20,6 +20,7 @@
 # grid and the PicQuadrature class.
 ###########################################################################
 
+from dataclasses import dataclass
 from typing import Any
 
 import numpy as np
@@ -27,9 +28,8 @@ import numpy as np
 import warp as wp
 import warp.examples.fem.utils as fem_example_utils
 import warp.fem as fem
-import warp.sim.render
+import warp.render
 from warp.fem import Domain, Field, Sample, at_node, div, grad, integrand
-from warp.sim import Model, State
 from warp.sparse import BsrMatrix, bsr_mm, bsr_mv, bsr_transposed
 
 
@@ -186,76 +186,83 @@ def solve_incompressibility(
 
 
 class Example:
-    def __init__(self, quiet=False, stage_path="example_apic_fluid.usd", voxel_size=1.0):
+    @dataclass
+    class State:
+        particle_q: wp.array(dtype=wp.vec3)
+        particle_qd: wp.array(dtype=wp.vec3)
+        particle_qd_grad: wp.array(dtype=wp.mat33)
+
+    def __init__(self, quiet=False, stage_path="example_apic_fluid.usd", voxel_size=1.0, opengl=False):
+        self.gravity = wp.vec3(0.0, -10.0, 0.0)
+
         fps = 60
+        self.sim_substeps = 1
         self.frame_dt = 1.0 / fps
         self.current_frame = 0
-
-        self.sim_substeps = 1
         self.sim_dt = self.frame_dt / self.sim_substeps
         self.voxel_size = voxel_size
 
         self._quiet = quiet
 
         # particle emission
-        particle_grid_lo = wp.vec3(-5)
-        particle_grid_hi = wp.vec3(5)
-
-        grid_cell_size = voxel_size
-        grid_cell_volume = np.prod(grid_cell_size)
-
         PARTICLES_PER_CELL_DIM = 2
-        self.radius = float(np.max(grid_cell_size) / (2 * PARTICLES_PER_CELL_DIM))
+        self.radius = float(np.max(voxel_size) / (2 * PARTICLES_PER_CELL_DIM))
 
+        particle_grid_lo = np.full(3, -5)
+        particle_grid_hi = np.full(3, 5)
         particle_grid_res = (
             np.array((particle_grid_hi - particle_grid_lo) / voxel_size, dtype=int) * PARTICLES_PER_CELL_DIM
         )
-        particle_grid_offset = wp.vec3(self.radius, self.radius, self.radius)
-
-        # Initialize warp.sim model, spawn particles
-        builder = wp.sim.ModelBuilder()
-        builder.add_particle_grid(
-            dim_x=particle_grid_res[0],
-            dim_y=particle_grid_res[1],
-            dim_z=particle_grid_res[2],
-            cell_x=self.radius * 2.0,
-            cell_y=self.radius * 2.0,
-            cell_z=self.radius * 2.0,
-            pos=particle_grid_lo + particle_grid_offset,
-            rot=wp.quat_identity(),
-            vel=wp.vec3(0.0, 0.0, 0.0),
-            mass=grid_cell_volume / PARTICLES_PER_CELL_DIM**3,
-            jitter=self.radius * 1.0,
+
+        self.particle_volumes, particle_q = self._spawn_particles(
+            particle_grid_res, particle_grid_lo, particle_grid_hi, packing_fraction=1.0
+        )
+        particle_qd = wp.zeros_like(particle_q)
+
+        particle_count = particle_q.shape[0]
+        if not self._quiet:
+            print("Particle count:", particle_count)
+
+        # Allocate states
+        self.state_0 = self.State(
+            wp.clone(particle_q),
+            wp.clone(particle_qd),
+            particle_qd_grad=wp.zeros(shape=(particle_count), dtype=wp.mat33),
+        )
+        self.state_1 = self.State(
+            wp.clone(particle_q),
+            wp.clone(particle_qd),
+            particle_qd_grad=wp.zeros(shape=(particle_count), dtype=wp.mat33),
         )
-        self.model: Model = builder.finalize()
-        self.model.ground = False
 
         # Storage for temporary variables
         self.temporary_store = fem.TemporaryStore()
 
-        if not self._quiet:
-            print("Particle count:", self.model.particle_count)
+        # initialize renderers
+        self.opengl_renderer = None
+        self.usd_renderer = None
 
-        self.state_0: State = self.model.state()
-        self.state_0.particle_qd_grad = wp.zeros(shape=(self.model.particle_count), dtype=wp.mat33)
-
-        self.state_1: State = self.model.state()
-        self.state_1.particle_qd_grad = wp.zeros(shape=(self.model.particle_count), dtype=wp.mat33)
+        try:
+            if opengl:
+                self.opengl_renderer = warp.render.OpenGLRenderer(
+                    screen_width=1024,
+                    screen_height=1024,
+                )
+        except Exception as err:
+            wp.utils.warn(f"Could not initialize OpenGL renderer: {err}.")
 
         try:
             if stage_path:
-                self.renderer = warp.sim.render.SimRenderer(self.model, stage_path, scaling=20.0)
-            else:
-                self.renderer = None
+                self.usd_renderer = warp.render.UsdRenderer(stage_path)
         except Exception as err:
-            print(f"Could not initialize SimRenderer for stage '{stage_path}': {err}.")
+            print(f"Could not initialize Usd renderer '{stage_path}': {err}.")
 
     def step(self):
         fem.set_default_temporary_store(self.temporary_store)
 
         self.current_frame = self.current_frame + 1
 
-        with wp.ScopedTimer(f"simulate frame {self.current_frame}", active=True):
+        with wp.ScopedTimer(f"simulate frame {self.current_frame}", synchronize=True):
             for _s in range(self.sim_substeps):
                 # Allocate the voxels and create the warp.fem geometry
                 volume = wp.Volume.allocate_by_voxels(
@@ -297,7 +304,7 @@ class Example:
 
                 # Bin particles to grid cells
                 pic = fem.PicQuadrature(
-                    domain=domain, positions=self.state_0.particle_q, measures=self.model.particle_mass
+                    domain=domain, positions=self.state_0.particle_q, measures=self.particle_volumes
                 )
 
                 # Compute inverse particle volume for each grid node
@@ -318,7 +325,7 @@ class Example:
                         "velocities": self.state_0.particle_qd,
                         "velocity_gradients": self.state_0.particle_qd_grad,
                         "dt": self.sim_dt,
-                        "gravity": self.model.gravity,
+                        "gravity": self.gravity,
                     },
                     output_dtype=wp.vec3,
                 )
@@ -377,16 +384,54 @@ class Example:
 
         fem.set_default_temporary_store(None)
 
+    @staticmethod
+    def _spawn_particles(res, bounds_lo, bounds_hi, packing_fraction):
+        Nx = res[0]
+        Ny = res[1]
+        Nz = res[2]
+
+        px = np.linspace(bounds_lo[0], bounds_hi[0], Nx + 1)
+        py = np.linspace(bounds_lo[1], bounds_hi[1], Ny + 1)
+        pz = np.linspace(bounds_lo[2], bounds_hi[2], Nz + 1)
+
+        points = np.stack(np.meshgrid(px, py, pz)).reshape(3, -1).T
+
+        cell_size = (bounds_hi - bounds_lo) / res
+        cell_volume = np.prod(cell_size)
+
+        radius = np.max(cell_size) * 0.5
+        volume = np.prod(cell_volume) * packing_fraction
+
+        rng = np.random.default_rng(42)
+        points += 2.0 * radius * (rng.random(points.shape) - 0.5)
+
+        volumes = wp.full(points.shape[0], volume, dtype=float)
+        points = wp.array(np.ascontiguousarray(points), dtype=wp.vec3)
+        return volumes, points
+
     def render(self, is_live=False):
-        if self.renderer is None:
+        if self.usd_renderer is None and self.opengl_renderer is None:
             return
 
-        with wp.ScopedTimer("render", active=True):
+        with wp.ScopedTimer("render", synchronize=True):
             time = self.current_frame * self.frame_dt
 
-            self.renderer.begin_frame(time)
-            self.renderer.render(self.state_0)
-            self.renderer.end_frame()
+            if self.usd_renderer is not None:
+                self.usd_renderer.begin_frame(time)
+                self.usd_renderer.render_points(
+                    "particles",
+                    self.state_0.particle_q.numpy(),
+                    radius=self.radius,
+                )
+                self.usd_renderer.end_frame()
+            if self.opengl_renderer is not None:
+                self.opengl_renderer.begin_frame(time)
+                self.opengl_renderer.render_points(
+                    "particles",
+                    self.state_0.particle_q,
+                    radius=self.radius,
+                )
+                self.opengl_renderer.end_frame()
 
 
 if __name__ == "__main__":
@@ -404,6 +449,7 @@ if __name__ == "__main__":
     )
     parser.add_argument("--num_frames", type=int, default=250, help="Total number of frames.")
     parser.add_argument("--quiet", action="store_true")
+    parser.add_argument("--opengl", action="store_true")
     parser.add_argument(
         "--voxel_size",
         type=float,
@@ -413,11 +459,11 @@ if __name__ == "__main__":
     args = parser.parse_known_args()[0]
 
     with wp.ScopedDevice(args.device):
-        example = Example(quiet=args.quiet, stage_path=args.stage_path, voxel_size=args.voxel_size)
+        example = Example(quiet=args.quiet, stage_path=args.stage_path, voxel_size=args.voxel_size, opengl=args.opengl)
 
         for _ in range(args.num_frames):
             example.step()
             example.render()
 
-        if example.renderer:
-            example.renderer.save()
+        if example.usd_renderer is not None:
+            example.usd_renderer.save()

warp/examples/fem/example_convection_diffusion_dg.py

@@ -116,11 +116,11 @@ class Example:
             values={"ang_vel": ang_vel},
         )
 
-        side_test = fem.make_test(space=scalar_space, domain=sides)
+        self._side_test = fem.make_test(space=scalar_space, domain=sides)
         side_trial = fem.make_trial(space=scalar_space, domain=sides)
         fem.integrate(
             upwind_transport_form,
-            fields={"phi": side_trial, "psi": side_test},
+            fields={"phi": side_trial, "psi": self._side_test},
             values={"ang_vel": ang_vel},
             output=matrix_transport,
             add=True,
@@ -132,7 +132,7 @@ class Example:
         )
         matrix_diffusion += fem.integrate(
             sip_diffusion_form,
-            fields={"phi": side_trial, "psi": side_test},
+            fields={"phi": side_trial, "psi": self._side_test},
         )
         self._matrix = matrix_inertia + matrix_transport + viscosity * matrix_diffusion
 
@@ -140,8 +140,12 @@ class Example:
         self._phi_field = scalar_space.make_field()
         fem.interpolate(initial_condition, dest=self._phi_field)
 
+        self._phi_curvature_field = scalar_space.make_field()
+        self._compute_phi_curvature()
+
         self.renderer = fem_example_utils.Plot()
         self.renderer.add_field("phi", self._phi_field)
+        self.renderer.add_field("phi_curvature", self._phi_curvature_field)
 
     def step(self):
         self.current_frame += 1
@@ -154,12 +158,29 @@ class Example:
 
         phi = wp.zeros_like(rhs)
         fem_example_utils.bsr_cg(self._matrix, b=rhs, x=phi, method="bicgstab", quiet=self._quiet)
-
         wp.utils.array_cast(in_array=phi, out_array=self._phi_field.dof_values)
 
+        # for visualization purposes only
+        self._compute_phi_curvature()
+
+    def _compute_phi_curvature(self):
+        fem.integrate(
+            diffusion_form,
+            fields={"u": self._phi_field, "v": self._test},
+            output=self._phi_curvature_field.dof_values,
+        )
+        fem.integrate(
+            sip_diffusion_form,
+            fields={"phi": self._phi_field.trace(), "psi": self._side_test},
+            output=self._phi_curvature_field.dof_values,
+            add=True,
+        )
+
     def render(self):
         self.renderer.begin_frame(time=self.current_frame * self.sim_dt)
         self.renderer.add_field("phi", self._phi_field)
+        self.renderer.add_field("phi_curvature", self._phi_curvature_field)
+
         self.renderer.end_frame()
 
 

warp/examples/fem/example_diffusion_mgpu.py

@@ -77,8 +77,7 @@ class DistributedSystem:
         stream = wp.get_stream()
 
         for mat_i, x_i, y_i, idx in zip(*self.rank_data):
-            # WAR copy with indexed array requiring matching shape
-            tmp_i = wp.array(ptr=tmp.ptr, device=tmp.device, capacity=tmp.capacity, dtype=tmp.dtype, shape=idx.shape)
+            tmp_i = tmp[: idx.size]
 
             # Compress rhs on rank 0
             x_idx = wp.indexedarray(x, idx)
@@ -161,7 +160,13 @@ class Example:
 
         with wp.ScopedDevice(main_device):
             fem_example_utils.bsr_cg(
-                A, x=global_res, b=glob_rhs, use_diag_precond=False, quiet=self._quiet, mv_routine=A.mv_routine
+                A,
+                x=global_res,
+                b=glob_rhs,
+                use_diag_precond=False,
+                quiet=self._quiet,
+                mv_routine=A.mv_routine,
+                mv_routine_uses_multiple_cuda_contexts=True,
             )
 
         array_cast(in_array=global_res, out_array=self._scalar_field.dof_values)

warp/examples/fem/utils.py

@@ -14,12 +14,14 @@
 # limitations under the License.
 
 
+import gc
 from typing import Any, Dict, Optional, Tuple
 
 import numpy as np
 
 import warp as wp
 import warp.fem as fem
+from warp.context import assert_conditional_graph_support
 from warp.optim.linear import LinearOperator, aslinearoperator, preconditioner
 from warp.sparse import BsrMatrix, bsr_get_diag, bsr_mv, bsr_transposed
 
@@ -230,6 +232,7 @@ def bsr_cg(
     quiet=False,
     method: str = "cg",
     M: BsrMatrix = None,
+    mv_routine_uses_multiple_cuda_contexts: bool = False,
 ) -> Tuple[float, int]:
     """Solves the linear system A x = b using an iterative solver, optionally with diagonal preconditioning
 
@@ -244,6 +247,8 @@ def bsr_cg(
         mv_routine: Matrix-vector multiplication routine to use for multiplications with ``A``
         quiet: if True, do not print iteration residuals
         method: Iterative solver method to use, defaults to Conjugate Gradient
+        mv_routine_uses_multiple_cuda_contexts: Whether the matrix-vector multiplication routine uses multiple CUDA contexts,
+          which prevents the use of conditional CUDA graphs.
 
     Returns:
         Tuple (residual norm, iteration count)
@@ -260,10 +265,53 @@ def bsr_cg(
 
     func = _get_linear_solver_func(method_name=method)
 
-    def print_callback(i, err, tol):
-        print(f"{func.__name__}: at iteration {i} error = \t {err}  \t tol: {tol}")
+    callback = None
 
-    callback = None if quiet else print_callback
+    use_cuda_graph = A.device.is_cuda and not wp.config.verify_cuda
+    capturable = use_cuda_graph and not mv_routine_uses_multiple_cuda_contexts
+
+    if capturable:
+        try:
+            assert_conditional_graph_support()
+        except RuntimeError:
+            capturable = False
+
+    if not quiet:
+        if capturable:
+
+            @wp.func_native(snippet=f'printf("%s: ", "{func.__name__}");')
+            def print_method_name():
+                pass
+
+            @fem.cache.dynamic_kernel(suffix=f"{check_every}{func.__name__}")
+            def device_cg_callback(
+                cur_iter: wp.array(dtype=int),
+                err_sq: wp.array(dtype=Any),
+                atol_sq: wp.array(dtype=Any),
+            ):
+                if cur_iter[0] % check_every == 0:
+                    print_method_name()
+                    wp.printf(
+                        "at iteration %d error = \t %f  \t tol: %f\n",
+                        cur_iter[0],
+                        wp.sqrt(err_sq[0]),
+                        wp.sqrt(atol_sq[0]),
+                    )
+
+            if check_every > 0:
+                callback = device_cg_callback
+        else:
+
+            def print_callback(i, err, tol):
+                print(f"{func.__name__}: at iteration {i} error = \t {err}  \t tol: {tol}")
+
+            callback = print_callback
+
+    if use_cuda_graph:
+        # Temporarily disable garbage collection
+        # Garbage collection of externally-allocated objects during graph capture may lead to
+        # invalid operations or memory access errors.
+        gc.disable()
 
     end_iter, err, atol = func(
         A=A,
@@ -271,12 +319,20 @@ def bsr_cg(
         x=x,
         maxiter=max_iters,
         tol=tol,
-        check_every=check_every,
+        check_every=0 if capturable else check_every,
         M=M,
         callback=callback,
-        use_cuda_graph=not wp.config.verify_cuda,
+        use_cuda_graph=use_cuda_graph,
     )
 
+    if use_cuda_graph:
+        gc.enable()
+
+    if isinstance(end_iter, wp.array):
+        end_iter = end_iter.numpy()[0]
+        err = np.sqrt(err.numpy()[0])
+        atol = np.sqrt(atol.numpy()[0])
+
     if not quiet:
         res_str = "OK" if err <= atol else "TRUNCATED"
         print(f"{func.__name__}: terminated after {end_iter} iterations with error = \t {err} ({res_str})")
@@ -437,28 +493,18 @@ def bsr_solve_saddle(
     wp.copy(src=b_u, dest=saddle_system.u_slice(b))
     wp.copy(src=b_p, dest=saddle_system.p_slice(b))
 
-    func = _get_linear_solver_func(method_name=method)
-
-    def print_callback(i, err, tol):
-        print(f"{func.__name__}: at iteration {i} error = \t {err}  \t tol: {tol}")
-
-    callback = None if quiet else print_callback
-
-    end_iter, err, atol = func(
-        A=saddle_system,
-        b=b,
-        x=x,
-        maxiter=max_iters,
+    err, end_iter = bsr_cg(
+        saddle_system,
+        x,
+        b,
+        max_iters=max_iters,
         tol=tol,
         check_every=check_every,
+        quiet=quiet,
+        method=method,
         M=saddle_system.preconditioner,
-        callback=callback,
     )
 
-    if not quiet:
-        res_str = "OK" if err <= atol else "TRUNCATED"
-        print(f"{func.__name__}: terminated after {end_iter} iterations with absolute error = \t {err} ({res_str})")
-
     wp.copy(dest=x_u, src=saddle_system.u_slice(x))
     wp.copy(dest=x_p, src=saddle_system.p_slice(x))
 

warp/fabric.py

@@ -211,7 +211,7 @@ class fabricarray(noncontiguous_array_base[T]):
                     allocator = self.device.get_allocator()
                     buckets_ptr = allocator.alloc(buckets_size)
                     cuda_stream = self.device.stream.cuda_stream
-                    runtime.core.memcpy_h2d(
+                    runtime.core.wp_memcpy_h2d(
                         self.device.context, buckets_ptr, ctypes.addressof(buckets), buckets_size, cuda_stream
                     )
                     self.deleter = allocator.deleter