BEGIN:VCALENDAR
VERSION:2.0
PRODID:Linklings LLC
BEGIN:VTIMEZONE
TZID:Europe/Stockholm
X-LIC-LOCATION:Europe/Stockholm
BEGIN:DAYLIGHT
TZOFFSETFROM:+0100
TZOFFSETTO:+0200
TZNAME:CEST
DTSTART:19700308T020000
RRULE:FREQ=YEARLY;BYMONTH=3;BYDAY=-1SU
END:DAYLIGHT
BEGIN:STANDARD
TZOFFSETFROM:+0200
TZOFFSETTO:+0100
TZNAME:CET
DTSTART:19701101T020000
RRULE:FREQ=YEARLY;BYMONTH=10;BYDAY=-1SU
END:STANDARD
END:VTIMEZONE
BEGIN:VEVENT
DTSTAMP:20190719T085744Z
LOCATION:HG D 3.2
DTSTART;TZID=Europe/Stockholm:20190613T111500
DTEND;TZID=Europe/Stockholm:20190613T114500
UID:submissions.pasc-conference.org_PASC19_sess128_msa188@linklings.com
SUMMARY:Towards an Accelerated Library for Sparse Computations
DESCRIPTION:Minisymposium\nComputer Science and Applied Mathematics, Emerg
 ing Application Domains, Chemistry and Materials, Climate and Weather, Phy
 sics, Solid Earth Dynamics, Life Sciences, Engineering\n\nTowards an Accel
 erated Library for Sparse Computations\n\nMpakos, Alverti, Papadopoulou, G
 oumas\n\nSparse-Matrix-Vector-Multiplication (SpMV) is one of the most wid
 ely used and well-studied linear-algebra scientific computing kernels. The
  sparsity of the matrix and the algorithmic nature of the kernel makes it 
 a heavily bandwidth-bound problem, with a relatively low operational inten
 sity (FLOPs/byte). Porting and optimizing the SpMV kernel on emerging para
 llel architectures poses performance and scalability challenges, tightly c
 oupled to the sparsity pattern of the matrix and the employed sparse matri
 x storage scheme. We will present preliminary results from our SpMV librar
 y for the EuroEXA platform, focusing on accelerating the SpMV kernel on FP
 GAs. To fully exploit the available memory bandwidth, we use wide datatype
 s, burst read/writes to/from memory and perform dataflow optimization. To 
 achieve high area utilization, and overcome the limited FPGA memory capaci
 ty, we break down the sparse matrix into 2D-blocks and use multiple comput
 e units. We use a custom matrix representation to exploit vectorization.
END:VEVENT
END:VCALENDAR