Source: adolc Version: 2.5.2 Severity: important Tags: patch Dear Maintainer,
ADOL-C computes wrong second derivatives for the log function in vector mode.
I consider this bug 'important', because you may not even notice right away
that your results are wrong.
The bug has been fixed upstream in commit
1fd1caefc8d6874162463a9f7123f54ed5f05964
after the 2.5.2 release (which is the one in Debian). The trivial patch is
appended below. Please consider applying it to the version that will be in
Jessie.
Thanks.
commit 1fd1caefc8d6874162463a9f7123f54ed5f05964
Author: kulshres <kulshres@94ac48a7-3327-4b6a-8511-9a4036a20e83>
Date: Fri Aug 29 06:47:22 2014 +0000
increment the correct pointer in log_op reverse mode
this op has no use for Targ2 but Targ is used in the statements above
so incrementing Targ2 is the silliest mistake I've seen so far.
Signed-off-by: Kshitij Kulshreshtha <kshi...@math.upb.de>
git-svn-id: https://projects.coin-or.org/svn/ADOL-C/trunk@568
94ac48a7-3327-4b6a-8511-9a4036a20e83
diff --git a/ADOL-C/src/ho_rev.c b/ADOL-C/src/ho_rev.c
index 28591e5..a5d5c03 100644
--- a/ADOL-C/src/ho_rev.c
+++ b/ADOL-C/src/ho_rev.c
@@ -1688,7 +1688,7 @@ int hov_ti_reverse(
HOV_INC(Ares, k)
HOV_INC(Aarg, k)
- HOS_OV_INC(Targ2, k)
+ HOS_OV_INC(Targ, k)
}
}
break;
-- System Information:
Debian Release: jessie/sid
APT prefers testing
APT policy: (500, 'testing'), (300, 'unstable')
Architecture: amd64 (x86_64)
Foreign Architectures: i386
Kernel: Linux 3.16-3-amd64 (SMP w/4 CPU cores)
Locale: LANG=de_DE.UTF-8, LC_CTYPE=de_DE.UTF-8 (charmap=UTF-8)
Shell: /bin/sh linked to /bin/dash
signature.asc
Description: OpenPGP digital signature
