Bug#944095: intel-mkl: Complex SVD gives incorrect results (LAPACK/LAPACKE zgesvd and cgesvd)

Richard Furse richard at blueripplesound.com
Mon Nov 4 09:32:52 GMT 2019


Package: intel-mkl
Version: 2019.2.187-1
Severity: normal
Tags: upstream

Dear Maintainer,

Hi there! I was testing MKL's LAPACK routines as an alternative to ATLAS and
found regression test failures. The apparently worst issue was that the
complex SVD of 2x2 matrix ((1 0) (0 -2i)) gives singular values (0 1) rather
than (2 1), which is badly wrong. I've discussed this on the Intel forum and
it turns out that, although not obviously in their bug fix list, this is
fixed in 2019u5. The CPU is a Xeon W-2156.

Many thanks



-- System Information:
Debian Release: 10.1
  APT prefers stable-updates
  APT policy: (500, 'stable-updates'), (500, 'proposed-updates'), (500,
'stable')
Architecture: amd64 (x86_64)
Foreign Architectures: i386

Kernel: Linux 4.19.0-6-amd64 (SMP w/16 CPU cores)
Kernel taint flags: TAINT_PROPRIETARY_MODULE, TAINT_OOT_MODULE,
TAINT_UNSIGNED_MODULE
Locale: LANG=en_GB.UTF-8, LC_CTYPE=en_GB.UTF-8 (charmap=UTF-8),
LANGUAGE=en_GB.UTF-8 (charmap=UTF-8)
Shell: /bin/sh linked to /usr/bin/dash
Init: systemd (via /run/systemd/system)
LSM: AppArmor: enabled

Versions of packages intel-mkl depends on:
ii  libmkl-dev  2019.2.187-1

intel-mkl recommends no packages.

intel-mkl suggests no packages.



More information about the debian-science-maintainers mailing list