[MISC] Enable taichi debug mode in tests if possible. (Genesis-Embodied-AI#1541)

* Improve huggingface download fault tolerance.
* Enable taichi debug mode if possible.
* Add opt-in flag for debug mode in unit tests.
* Fix small nits.
* Require 'taichi>=1.7.4'.

Author: duburcqa

.github/workflows/generic.yml

@@ -32,6 +32,7 @@ jobs:
       TI_ENABLE_METAL: "0"
       TI_ENABLE_OPENGL: "0"
       TI_ENABLE_VULKAN: "0"
+      TI_DEBUG: "0"
 
     runs-on: ${{ matrix.OS }}
     if: github.event_name != 'release'
@@ -112,7 +113,7 @@ jobs:
 
       - name: Run unit tests
         run: |
-          pytest -v --logical --forked -m required ./tests
+          pytest -v --logical --dev --forked -m required ./tests
 
       - name: Save Updated Taichi Kernel Cache
         if: always()

.github/workflows/production.yml

@@ -67,7 +67,7 @@ jobs:
             bash -c "
               pip install --extra-index-url https://pypi.nvidia.com/ omniverse-kit && \
               pip install -e '.[dev,render,usd]' && \
-              pytest -v --forked ./tests
+              pytest -v --dev --forked ./tests
             "
 
       - name: Run benchmarks

genesis/__init__.py

@@ -211,8 +211,8 @@ def init(
     with redirect_stdout(_ti_outputs):
         ti.init(
             arch=ti_arch,
-            # debug is causing segfault on some machines
-            debug=False,
+            # Add a (hidden) mechanism to forceable disable taichi debug mode as it is still a bit experimental
+            debug=debug and backend == gs.cpu and (os.environ.get("TI_DEBUG") != "0"),
             check_out_of_bound=debug,
             # force_scalarize_matrix=True for speeding up kernel compilation
             # Turning off 'force_scalarize_matrix' is causing numerical instabilities ('nan') on MacOS

genesis/engine/couplers/legacy_coupler.py

@@ -454,7 +454,7 @@ def fem_surface_force(self, f: ti.i32):
                             mpm_grid_I = mpm_base - self.mpm_solver.grid_offset + mpm_offset
                             mpm_grid_mass = self.mpm_solver.grid[f, mpm_grid_I, i_b].mass / self.mpm_solver.p_vol_scale
 
-                            mpm_weight = ti.cast(1.0, gs.ti_float)
+                            mpm_weight = gs.ti_float(1.0)
                             for d in ti.static(range(3)):
                                 mpm_weight *= mpm_w[mpm_offset[d]][d]
 

genesis/engine/entities/pbd_entity.py

@@ -114,7 +114,7 @@ def _kernel_add_particles_edges_to_solver(
             self.solver.particles[i_p, i_b].dpos = ti.Vector.zero(gs.ti_float, 3)
             self.solver.particles[i_p, i_b].free = True
 
-            self.solver.particles_ng[i_p, i_b].active = active
+            self.solver.particles_ng[i_p, i_b].active = ti.cast(active, gs.ti_bool)
 
         for i_e_ in range(self.n_edges):
             i_e = i_e_ + self._edge_start

genesis/engine/entities/tool_entity/mesh.py

@@ -86,7 +86,7 @@ def sdf(self, f, pos_world, i_b):
     def sdf_(self, pos_voxels):
         # sdf value from voxels coordinate
         base = ti.floor(pos_voxels, gs.ti_int)
-        signed_dist = ti.cast(0.0, gs.ti_float)
+        signed_dist = gs.ti_float(0.0)
         if (base >= self.sdf_res - 1).any() or (base < 0).any():
             signed_dist = 1.0
         else:
@@ -117,7 +117,7 @@ def normal(self, f, pos_world, i_b):
     @ti.func
     def normal_(self, pos_voxels):
         # since we are in voxels frame, delta can be a relatively big value
-        delta = ti.cast(1e-2, gs.ti_float)
+        delta = gs.ti_float(1e-2)
         normal_vec = ti.Vector([0, 0, 0], dt=gs.ti_float)
 
         for i in ti.static(range(3)):
@@ -146,15 +146,15 @@ def collide(self, f, pos_world, vel_mat, i_b):
             signed_dist = self.sdf(f, pos_world, i_b)
             # bigger coup_softness implies that the coupling influence extends further away from the object.
             influence = ti.min(ti.exp(-signed_dist / max(gs.EPS, self.material.coup_softness)), 1)
-            if signed_dist <= 0 or influence > 0.1:
+            if signed_dist <= 0.0 or influence > 0.1:
                 vel_collider = self.vel_collider(f, pos_world, i_b)
 
                 # v w.r.t collider
                 rel_v = vel_mat - vel_collider
                 normal_vec = self.normal(f, pos_world, i_b)
                 normal_component = rel_v.dot(normal_vec)
 
-                if normal_component < 0:
+                if normal_component < 0.0:
                     # remove inward velocity
                     rel_v_t = rel_v - normal_component * normal_vec
                     rel_v_t_norm = rel_v_t.norm(gs.EPS)
@@ -165,9 +165,10 @@ def collide(self, f, pos_world, vel_mat, i_b):
                     )
 
                     # tangential component after friction
-                    flag = ti.cast(normal_component < 0, gs.ti_float)
-                    rel_v_t = rel_v_t_friction * flag + rel_v_t * (1 - flag)
-                    vel_mat = vel_collider + rel_v_t * influence + rel_v * (1 - influence)
+                    # FIXME: This formula could be simplified since flag = 1.0 systematically.
+                    flag = ti.cast(normal_component < 0.0, gs.ti_float)
+                    rel_v_t = rel_v_t_friction * flag + rel_v_t * (1.0 - flag)
+                    vel_mat = vel_collider + rel_v_t * influence + rel_v * (1.0 - influence)
 
         return vel_mat
 

genesis/engine/solvers/fem_solver.py

@@ -314,12 +314,12 @@ def _init_surface_info(self):
     @ti.kernel
     def compute_surface_vertices(self):
         for i_v in range(self.n_vertices):
-            self.vertices_on_surface[i_v] = 0
+            self.vertices_on_surface[i_v] = False
 
         for i_s in range(self.n_surfaces):
             tri2v = self.surface[i_s].tri2v
             for i in ti.static(range(3)):
-                self.vertices_on_surface[tri2v[i]] = 1
+                self.vertices_on_surface[tri2v[i]] = True
 
     @ti.kernel
     def compute_surface_elements(self):

genesis/engine/solvers/mpm_solver.py

@@ -383,7 +383,7 @@ def p2g(self, f: ti.i32):
                 w = [0.5 * (1.5 - fx) ** 2, 0.75 - (fx - 1) ** 2, 0.5 * (fx - 0.5) ** 2]
                 for offset in ti.static(ti.grouped(self.stencil_range())):
                     dpos = (offset.cast(gs.ti_float) - fx) * self._dx
-                    weight = ti.cast(1.0, gs.ti_float)
+                    weight = gs.ti_float(1.0)
                     for d in ti.static(range(3)):
                         weight *= w[offset[d]][d]
 
@@ -439,7 +439,7 @@ def g2p(self, f: ti.i32):
                 for offset in ti.static(ti.grouped(self.stencil_range())):
                     dpos = offset.cast(gs.ti_float) - fx
                     grid_vel = self.grid[f, base - self._grid_offset + offset, i_b].vel_out
-                    weight = ti.cast(1.0, gs.ti_float)
+                    weight = gs.ti_float(1.0)
                     for d in ti.static(range(3)):
                         weight *= w[offset[d]][d]
 
@@ -962,11 +962,11 @@ def set_state(self, f, state, envs_idx=None):
     def _kernel_update_render_fields(self, f: ti.i32):
         for i_p, i_b in ti.ndrange(self._n_particles, self._B):
             if self.particles_ng[f, i_p, i_b].active:
-                self.particles_render[i_b, i_p].pos = self.particles[f, i_p, i_b].pos
-                self.particles_render[i_b, i_p].vel = self.particles[f, i_p, i_b].vel
+                self.particles_render[i_p, i_b].pos = self.particles[f, i_p, i_b].pos
+                self.particles_render[i_p, i_b].vel = self.particles[f, i_p, i_b].vel
             else:
-                self.particles_render[i_b, i_p].pos = gu.ti_nowhere()
-            self.particles_render[i_b, i_p].active = self.particles_ng[f, i_p, i_b].active
+                self.particles_render[i_p, i_b].pos = gu.ti_nowhere()
+            self.particles_render[i_p, i_b].active = self.particles_ng[f, i_p, i_b].active
 
         for i_v, i_b in ti.ndrange(self._n_vverts, self._B):
             vvert_pos = ti.Vector.zero(gs.ti_float, 3)

genesis/engine/solvers/rigid/rigid_solver_decomp.py

@@ -5349,11 +5349,11 @@ def func_torque_and_passive_force(
                 joint_type = joints_info.type[I_j]
 
                 if joint_type != gs.JOINT_TYPE.FREE and joint_type != gs.JOINT_TYPE.FIXED:
-                    dof_start = links_info.dof_start[I_l]
                     q_start = links_info.q_start[I_l]
-                    q_end = links_info.q_end[I_l]
+                    dof_start = links_info.dof_start[I_l]
+                    dof_end = links_info.dof_end[I_l]
 
-                    for j_d in range(q_end - q_start):
+                    for j_d in range(dof_end - dof_start):
                         I_d = (
                             [dof_start + j_d, i_b]
                             if ti.static(static_rigid_sim_config.batch_dofs_info)
@@ -5379,11 +5379,11 @@ def func_torque_and_passive_force(
             joint_type = joints_info.type[I_j]
 
             if joint_type != gs.JOINT_TYPE.FREE and joint_type != gs.JOINT_TYPE.FIXED:
-                dof_start = links_info.dof_start[I_l]
                 q_start = links_info.q_start[I_l]
-                q_end = links_info.q_end[I_l]
+                dof_start = links_info.dof_start[I_l]
+                dof_end = links_info.dof_end[I_l]
 
-                for j_d in range(q_end - q_start):
+                for j_d in range(dof_end - dof_start):
                     I_d = (
                         [dof_start + j_d, i_b]
                         if ti.static(static_rigid_sim_config.batch_dofs_info)

genesis/engine/solvers/sph_solver.py

@@ -628,7 +628,7 @@ def _density_solve(self, f: ti.i32):
 
     @ti.func
     def cubic_kernel(self, r_norm):
-        res = ti.cast(0.0, gs.ti_float)
+        res = gs.ti_float(0.0)
         h = self._support_radius
         # value of cubic spline smoothing kernel
         k = 1.0