Unstable behavior in local alignment mode on x64 linux, possibly due to uninitialized values

[ctsa]

Hi @yangao07 ,

I'm using the latest binary release of abPOA 1.4..1 on x64 linux (Centos 7). I was getting unstable, non-deterministic results, especially seeing the error [simd_abpoa_align_sequence_to_subgraph] Error in cg_backtrack quite often for my use case, and I was able to trace this to an un-initialized memory error which seems to occur in local but not global alignment mode.

I reduced this down to a simple test case, using the following input:

test.fa:

>1
A
>2
A

And running:

valgrind ./abpoa -m1 test.fa

The top of the valgrind results show:

$ valgrind ./abpoa -m1 test.fa
==121699== Memcheck, a memory error detector
==121699== Copyright (C) 2002-2017, and GNU GPL'd, by Julian Seward et al.
==121699== Using Valgrind-3.15.0 and LibVEX; rerun with -h for copyright info   
==121699== Command: ./abpoa -m1 test.fa
==121699==
[main] CMD:  ./abpoa -m1 test.fa
==121699== Conditional jump or move depends on uninitialised value(s)
==121699==    at 0x13EC10: simd_abpoa_align_sequence_to_subgraph (in /home/csaunders/devel/github/abPOA/tmp/abPOA-v1.4.1_x64-linux/abpoa)
==121699==    by 0x141AB7: simd_abpoa_align_sequence_to_graph (in /home/csaunders/devel/github/abPOA/tmp/abPOA-v1.4.1_x64-linux/abpoa)
==121699==    by 0x10CE31: abpoa_poa (in /home/csaunders/devel/github/abPOA/tmp/abPOA-v1.4.1_x64-linux/abpoa)
==121699==    by 0x10DD4A: abpoa_msa1 (in /home/csaunders/devel/github/abPOA/tmp/abPOA-v1.4.1_x64-linux/abpoa)
==121699==    by 0x10A694: abpoa_main (in /home/csaunders/devel/github/abPOA/tmp/abPOA-v1.4.1_x64-linux/abpoa)
==121699==    by 0x10A008: main (in /home/csaunders/devel/github/abPOA/tmp/abPOA-v1.4.1_x64-linux/abpoa)

...followed by many similar memory errors that I assume are related. No issues appear for the same test in global mode.

Note also that the same error occurs for code I compiled from the head of the main branch (d2e0186).

Thanks for any help or suggestions you might have to stabilize this case!

[yangao07]

Thanks for pointing this out!
It is actually a bug related to initialization.
Fixed it now, please try out the latest commit 6f6a06c.

[ctsa]

Nice thank you! I'll give it a try.

[ctsa]

Okay, all of my test cases are running cleanly through valgrind now. I'll close this as fixed.

Thanks again for the quick fix!