A problem was found with chroma handling of axis aligned photons (coming from MockNuWa). Checking with g4daeview.py –nopropagate see the problem is more specifically with vertical photons.
They result in kernels never completing, when running with GUI these cause terminations. When running in >console mode my patience never lasted long enough.
From CUDA GDB-ing they enter fill_state and never emerge getting stuck in intersect_mesh/intersect_box. The CUDA GDB session that revealed this is documented at
As the intersection code deals in reciprocal directions, clearly axis alignment results in infinities.
(chroma_env)delta:~ blyth$ RANGE=0:32 MockNuWa 1
Checking typical behaviour of chroma/chroma/cuda/mesh.h:
2014-11-25 16:37:12,973 INFO chroma.gpu.photon_hit:204 nwork 32 step 0 max_steps 30 nsteps 30
[Launch of CUDA Kernel 18 (propagate_hit<<<(1,1,1),(64,1,1)>>>) on Device 0]
node gets: 906 triangle count: 190 maxcurr: 23 hitsame: 0
node gets: 788 triangle count: 100 maxcurr: 15 hitsame: 0
node gets: 365 triangle count: 49 maxcurr: 15 hitsame: 0
node gets: 764 triangle count: 160 maxcurr: 18 hitsame: 0
node gets: 1224 triangle count: 248 maxcurr: 20 hitsame: 0
node gets: 757 triangle count: 115 maxcurr: 16 hitsame: 0
node gets: 372 triangle count: 42 maxcurr: 14 hitsame: 0
node gets: 991 triangle count: 203 maxcurr: 19 hitsame: 0
node gets: 316 triangle count: 26 maxcurr: 14 hitsame: 0
node gets: 342 triangle count: 28 maxcurr: 15 hitsame: 0
node gets: 765 triangle count: 82 maxcurr: 14 hitsame: 0
node gets: 209 triangle count: 4 maxcurr: 12 hitsame: 0
node gets: 666 triangle count: 120 maxcurr: 14 hitsame: 0
node gets: 183 triangle count: 6 maxcurr: 11 hitsame: 0
node gets: 198 triangle count: 8 maxcurr: 9 hitsame: 0
node gets: 205 triangle count: 8 maxcurr: 8 hitsame: 0
node gets: 964 triangle count: 266 maxcurr: 25 hitsame: 0
node gets: 188 triangle count: 4 maxcurr: 9 hitsame: 0
node gets: 188 triangle count: 4 maxcurr: 9 hitsame: 0
node gets: 186 triangle count: 8 maxcurr: 9 hitsame: 0
node gets: 217 triangle count: 8 maxcurr: 8 hitsame: 0
node gets: 184 triangle count: 4 maxcurr: 9 hitsame: 0
node gets: 390 triangle count: 54 maxcurr: 10 hitsame: 0
node gets: 1082 triangle count: 240 maxcurr: 18 hitsame: 0
node gets: 662 triangle count: 70 maxcurr: 16 hitsame: 0
node gets: 241 triangle count: 4 maxcurr: 9 hitsame: 0
node gets: 390 triangle count: 36 maxcurr: 13 hitsame: 0
node gets: 503 triangle count: 49 maxcurr: 13 hitsame: 0
node gets: 213 triangle count: 8 maxcurr: 9 hitsame: 0
node gets: 157 triangle count: 4 maxcurr: 8 hitsame: 0
node gets: 191 triangle count: 6 maxcurr: 11 hitsame: 0
node gets: 335 triangle count: 20 maxcurr: 12 hitsame: 0
node gets: 507 triangle count: 89 maxcurr: 11 hitsame: 1
node gets: 291 triangle count: 27 maxcurr: 8 hitsame: 1
node gets: 243 triangle count: 11 maxcurr: 7 hitsame: 1
node gets: 159 triangle count: 3 maxcurr: 9 hitsame: 1
node gets: 460 triangle count: 86 maxcurr: 11 hitsame: 1
After a few seconds of kernel run on single MOCK vertical photon:
(chroma_env)delta:~ blyth$ RANGE=0:1 MockNuWa MOCK
cuda-gdb) p count
$1 = 110695
(cuda-gdb) p tri_count
$2 = 86518
(cuda-gdb) p maxcurr
$3 = <value optimized out>
(cuda-gdb) p hitsame
$4 = 0
(cuda-gdb)
Its as if the x and y position of the photon is not coming into play, so are intersecting with most everythig in the tree ?:
(cuda-gdb) p node
$5 = {lower = {x = -22439.8477, y = -797443.5, z = -3032.02246}, upper = {x = -22425.7305, y = -797436.625, z = -3004.49121}, child = 1852655, nchild = 0}
(cuda-gdb) c
Continuing.
Breakpoint 1, intersect_mesh(const float3 * @generic, const float3 * @generic, Geometry * @generic, float * @generic, int) (origin=0x3fffbb0, direction=0x3fffbbc, g=0x1000000, min_distance=0x3fffc1c, last_hit_triangle=-1) at mesh.h:83
83 if (node.nchild == 0) { /* leaf node */
(cuda-gdb) p node
$6 = {lower = {x = -22426.6699, y = -797442.5, z = -3032.02246}, upper = {x = -22425.4941, y = -797438.5, z = -3004.25586}, child = 1852751, nchild = 0}
(cuda-gdb)