pulled a lot of the photon propagation code out of src/kernel.cu into src/photon.h so that photon propagation by propagate() in kernel.cu and the hybrid monte carlo ray tracing use the same code. instead of a single state, photons now carry the history of the processes they've undergone. this history is stored as a bitmask; see src/photon.h. start_node and first_node of the mesh are now stored as global variables in mesh.h instead of being passed to kernel functions.

1 files changed, 3 insertions, 3 deletions

diff --git a/render.py b/render.py
index 3ba0276..fe70492 100755
--- a/render.py
+++ b/render.py
@@ -50,10 +50,10 @@ def render(viewable, size=(800,600), name='', bits=8, make_movie=False):
     rgb_lookup_gpu = gpuarray.zeros(source_positions_gpu.size, dtype=gpuarray.vec.float3)
 
     max_steps = 10
-    rgb_runs = 100
+    rgb_runs = 1000
 
     print 'building rgb lookup.'
-    cuda_build_rgb_lookup(np.int32(source_positions_gpu.size), rng_states_gpu, source_positions_gpu, source_directions_gpu, np.int32(geometry.node_map.size-1), np.int32(geometry.first_node), rgb_lookup_gpu, np.int32(rgb_runs), np.int32(max_steps), block=(64,1,1), grid=(source_positions_gpu.size//64+1,1))
+    cuda_build_rgb_lookup(np.int32(source_positions_gpu.size), rng_states_gpu, source_positions_gpu, source_directions_gpu, rgb_lookup_gpu, np.int32(rgb_runs), np.int32(max_steps), block=(64,1,1), grid=(source_positions_gpu.size//64+1,1))
     cuda.Context.synchronize()
     print 'done.'
 
@@ -97,7 +97,7 @@ def render(viewable, size=(800,600), name='', bits=8, make_movie=False):
         global image_index
 
         t0 = time.time()
-        cuda_render(np.int32(pixels_gpu.size), rng_states_gpu, origins_gpu, directions_gpu, np.int32(geometry.node_map.size-1), np.int32(geometry.first_node), rgb_lookup_gpu, np.int32(nruns), pixels_gpu, np.int32(max_steps), block=(nblocks,1,1), grid=(pixels_gpu.size//nblocks+1,1))
+        cuda_render(np.int32(pixels_gpu.size), rng_states_gpu, origins_gpu, directions_gpu, rgb_lookup_gpu, np.int32(nruns), pixels_gpu, np.int32(max_steps), block=(nblocks,1,1), grid=(pixels_gpu.size//nblocks+1,1))
         cuda.Context.synchronize()
         elapsed = time.time() - t0