Christoph Conrads

Results 71 comments of Christoph Conrads

The work space size was changed twice, more recently in commit f46ce64a4f7e1187047f1a527ffbca919e97fe40 (2016) by Mark Gates. The less recent change was made in 2009 in commit 1d4ed33a5cfc65afcd46a0babb74979dfbed73f3. The message of...

> So there is an issue here. We had 3mnmn in 3.1.1 and we now have 3*mn in 3.10.1. Oops. This change originates in commit 1d4ed33a5cfc65afcd46a0babb74979dfbed73f3 (2009). The commit message...

Thank you for providing this reference. Apparently the changes to xLARFGP are on target whereas the changes to xORBDB5 and xORBDB6 from #930 are just working around the problem. >...

Computations with Householder reflectors by xLARFGP are only conditionally backward stable. The problem is not the computation of the reflector itself but the entries large in modulus it can contain....

(1) Frankly I do not want to look at alternative representations now and finish #406 before involving myself in something else. Besides the current representation has very nice properties (the...

> * If the xLARF* implementations were not updated to handle τ with a sign bit set, then they will compute garbage. [...] Two scenarios come to mind how this...

# A 3x3 Example with Rapid Divergence The stability proofs for the computations of QR decomposition by means of Householder reflections do not apply when the sign choice by slarfgP...

> comments on the specific way I introduced C++ and my implementation of the algorithm should be posted here. C++11 introduced the `auto` keyword. Throughout the code you use the...

> comments on the specific way I introduced C++ and my implementation of the algorithm should be posted here. I can'tquite wrap my head around the file names, specifically the...

The major problems with Fortran are the *massively error prone* code duplication (real/complex, single/double), error prone variable initialization, and the lack of conditional compilation including assertions and initializing memory in...