Fix/blocking (#187)

Author: mcw92

artist/core/heliostat_ray_tracer.py

@@ -438,19 +438,18 @@ def trace_rays(
             )
 
         flux_distributions = []
+        global_active_indices = torch.nonzero(active_heliostats_mask, as_tuple=True)[0]
         for batch_index, (batch_u, batch_e) in enumerate(self.distortions_loader):
             sampler_indices = list(self.distortions_sampler)
-
+            batch_mask_indices = sampler_indices[
+                batch_index * self.batch_size : (batch_index + 1) * self.batch_size
+            ]
             active_heliostats_mask_batch = torch.zeros(
                 self.heliostat_group.number_of_active_heliostats,
                 dtype=torch.bool,
                 device=device,
             )
-            active_heliostats_mask_batch[
-                sampler_indices[
-                    batch_index * self.batch_size : (batch_index + 1) * self.batch_size
-                ]
-            ] = True
+            active_heliostats_mask_batch[batch_mask_indices] = True
 
             rays = self.scatter_rays(
                 distortion_u=batch_u,
@@ -486,9 +485,10 @@ def trace_rays(
                 points_at_ray_origins = self.heliostat_group.active_surface_points[
                     active_heliostats_mask_batch, None, :, :3
                 ].expand(-1, self.light_source.number_of_rays, -1, -1)
-                ray_to_heliostat_mapping = torch.arange(
-                    number_of_heliostats, device=device
-                ).repeat_interleave(number_of_rays * number_of_points)
+                batch_global_indices = global_active_indices[batch_mask_indices]
+                ray_to_heliostat_mapping = batch_global_indices.repeat_interleave(
+                    number_of_rays * number_of_points
+                )
 
                 # Filter out the blocking primitives that are relevant for blocking.
                 filtered_blocking_primitive_indices = (
@@ -499,7 +499,7 @@ def trace_rays(
                             ..., :3
                         ],
                         ray_to_heliostat_mapping=ray_to_heliostat_mapping,
-                        max_stack_size=128,
+                        max_stack_size=64,
                         device=device,
                     )
                 )

docs/heliostats.rst

@@ -26,7 +26,7 @@ As you can see, each facet has a ``position`` relative to the heliostat's center
 (``canting_e``, ``canting_n``) that define its orientation.
 
 Kinematics
-^^^^^^^^^
+^^^^^^^^^^
 The heliostat's kinematics model describes the motion of its mechanical system. It's used to predict the final orientation
 of the heliostat surface based on variable inputs. The kinematics model is also used to calculate the aligned surface points
 and normals for a predicted orientation. The choice of kinematics type can depend on the type and number of actuators or

docs/images/kinematic_translation.pdf docs/images/kinematics_translation.pdfdocs/images/kinematic_translation.pdf renamed to docs/images/kinematics_translation.pdf

@@ -13,6 +13,9 @@ NURBS in ``ARTIST``
 
 Here we first consider some **NURBS Theory** before investigating **NURBS in
 ARTIST**.
+All NURBS figures on this page were generated with the tool provided by Yu-Sung Chang (2007), "B-Spline Curve with Knots"
+`Wolfram Demonstrations Project`_.
+
 
 NURBS Theory
 ------------
@@ -247,3 +250,5 @@ For this calculation of the surface points and surface normals, three internal s
    ``basis_function_and_derivatives()``.
 3. Lastly, the surface points and normals are calculated from the basis functions, their derivatives, and the control
    points.
+
+.. _Wolfram Demonstrations Project: https://demonstrations.wolfram.com/BSplineCurveWithKnots/

docs/nurbs_tutorial.rst

@@ -103,7 +103,7 @@ in the generated image individually.
 
 
 Optimizer, Scheduler, Regularizer, and Constraints Configuration
-^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
+^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
 
 The surface reconstruction internally uses the ``torch.optim.Adam`` optimizer. Depending on the data you use, different
 parameters may perform better for the optimizer - including a different learning rate scheduler. Therefore, we first have

docs/tutorial_surface_reconstruction.rst

@@ -37,7 +37,7 @@ def device(request: pytest.FixtureRequest) -> torch.device:
 
         os_name = platform.system()
         if os_name in {config_dictionary.linux, config_dictionary.windows}:
-            return torch.device("cuda:1" if torch.cuda.is_available() else "cpu")
+            return torch.device("cuda" if torch.cuda.is_available() else "cpu")
         elif os_name == config_dictionary.mac:
             return torch.device("cpu")
         else: