BEGIN:VCALENDAR
PRODID:-//Microsoft Corporation//Outlook MIMEDIR//EN
VERSION:1.0
BEGIN:VEVENT
DTSTART:20111115T190000Z
DTEND:20111115T193000Z
LOCATION:TCC 305
DESCRIPTION;ENCODING=QUOTED-PRINTABLE:ABSTRACT: This work revisits existing algorithms for the QR factorization of=0A rectangular matrices composed of $p times q$ tiles, where $p geq q$.=0A Within this framework, we study the critical paths and performance of=0A algorithms such as SK, MC, Greedy, and those found within PLASMA.=0A Although neither MC nor Greedy is optimal, both are shown to=0A be asymptotically optimal for all matrices of size $p = q^2 f(q)$, where=0A $f$ is any function such that $lim_{+infty} f= 0$. This result applies to=0A all matrices where $p$ and $q$ are proportional, $ p = lambda q$, with=0A $lambda geq 1$, thereby encompassing many important situations in=0A practice (least squares). We provide an extensive set of experiments that=0A show the superiority of the new algorithms for tall matrices.
SUMMARY:Tiled QR factorization algorithms
PRIORITY:3
END:VEVENT
END:VCALENDAR