BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:PSLP: Padded SLP Automatic Vectorization - Vasileios Porpodas\, Un
 iversity of Cambridge
DTSTART:20150204T110000Z
DTEND:20150204T120000Z
UID:TALK57832@talks.cam.ac.uk
CONTACT:Timothy Jones
DESCRIPTION:The need to increase performance and power efficiency in moder
 n processors has led to a wide adoption of SIMD vector units.  However\, w
 riting code that makes efficient use of these units is hard and leads to p
 latform specific implementations.  Compiler-based automatic vectorization 
 is one solution for this problem.  In particular the Superword-Level Paral
 lelism (SLP) vectorization algorithm is the primary way to automatically g
 enerate vector code starting form straight-line scalar code\, relying on f
 inding sequences of isomorphic instructions to pack together into vectors.
   However\, this hinders the applicability of the algorithm as isomorphic 
 code sequences are not common in practice.\n\nIn this talk we propose a so
 lution to overcome this limitation.  We introduce Padded SLP (PSLP)\, a no
 vel vectorization algorithm that injects a near-minimal number of redundan
 t instructions into the code to transform non-isomorphic sequences into is
 omorphic versions\, enabling their successful vectorization.  Our experime
 nts show that PSLP improves vectorization coverage and improves performanc
 e across several benchmarks and kernels.\n
LOCATION:FW11\, Computer Laboratory
END:VEVENT
END:VCALENDAR