Prototype port to 3D

What is it with OpenCL @ GPU and arrays of 3-dimensional vectors?!
Luckily I vaguely remembered encountering this problem before when
implementing my OpenCL LBM code but why on earth is the compiler not
warning about this? I guess the underlying reason has to do with the
data alignment requirements on the GPU but still...

1 files changed, 9 insertions, 4 deletions

diff --git a/boltzgas/initial_condition.py b/boltzgas/initial_condition.py
index 4d66263..2526566 100644
--- a/boltzgas/initial_condition.py
+++ b/boltzgas/initial_condition.py
@@ -1,15 +1,20 @@
 import numpy as np
 
 def grid_of_random_velocity_particles(width, radius, u_scale):
-    np_position = np.ndarray((width**2, 2))
-    np_velocity = np.ndarray((width**2, 2))
+    np_position = np.ndarray((width**3, 4))
+    np_velocity = np.ndarray((width**3, 4))
 
     grid = np.meshgrid(np.linspace(2*radius, 1-2*radius, width),
+                       np.linspace(2*radius, 1-2*radius, width),
                        np.linspace(2*radius, 1-2*radius, width))
     np_position[:,0] = grid[0].flatten()
     np_position[:,1] = grid[1].flatten()
+    np_position[:,2] = grid[2].flatten()
+    np_position[:,3] = 0
 
-    np_velocity[:,0] = u_scale*(-0.5 + np.random.random_sample((width**2,)))
-    np_velocity[:,1] = u_scale*(-0.5 + np.random.random_sample((width**2,)))
+    np_velocity[:,0] = u_scale*(-0.5 + np.random.random_sample((width**3,)))
+    np_velocity[:,1] = u_scale*(-0.5 + np.random.random_sample((width**3,)))
+    np_velocity[:,2] = u_scale*(-0.5 + np.random.random_sample((width**3,)))
+    np_velocity[:,3] = 0
 
     return (np_position, np_velocity)