the `merge` program, when compiled with the intel compiler (icx) sometimes crashes or hangs
The following command line gives inconsistent result.
$build_tree/filter/merge -mat /tmp/c30.purged.gz -out /tmp/h -skip 32 -t 64
with the attached data file.
I observed the following possible outcomes:
- crashes
N=745 W=42609 (0MB) W/N=57.19 fill-in=230.00 cpu=11.8s wct=0.3s mem=4751M [pass=106,cwmax=32]
Starting collection with 1% of waste...Examined -nan% of full pages, recycled 0% of waste. 0% of examined data was garbage
GC took 0.0s (cpu), 0.0s (wct) [cpu/wct=43.6]
compute_R took 0.0s (cpu), 0.0s (wct) [cpu/wct=44.1]
compute_merges took 0.0s (cpu), 0.0s (wct) [cpu/wct=44.2]
apply_merges took 0.0s (cpu), 0.0s (wct) [cpu/wct=45.5]
buffer_flush took 0.0s (cpu), 0.0s (wct) [cpu/wct=43.0]
Error, no merge done while n_possible_merges > 0
Please check the entries in your purged file are sorted
- segfaults. Since I'm lucky enough that the segfaults still exists in debug mode, I can say that the crash occurs in lines 1174 (rarely) or 1175 (more often) there
N=737 W=43514 (0MB) W/N=59.04 fill-in=303.00 cpu=12.7s wct=0.3s mem=4807M [pass=122,cwmax=32]
Starting collection with 1% of waste...Examined -nan% of full pages, recycled 0% of waste. 0% of examined data was garbage
GC took 0.0s (cpu), 0.0s (wct) [cpu/wct=48.7]
compute_R took 0.0s (cpu), 0.0s (wct) [cpu/wct=48.9]
compute_merges took 0.1s (cpu), 0.0s (wct) [cpu/wct=47.6]
Segmentation fault (core dumped)
- and sometimes the computation works correctly.
Edited by Emmanuel Thomé