BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Optimization Methods for Accelerator Mapping and Hardware Design S
 pace Exploration - Grace Dinh\, Unversity of Berkley
DTSTART:20240308T130000Z
DTEND:20240308T140000Z
UID:TALK213157@talks.cam.ac.uk
CONTACT:Tobias Grosser
DESCRIPTION:Obtaining good performance for tensor problems on accelerators
  (both mapping workloads to existing accelerators\, and performing design-
 space exploration to select hardware parameters) requires optimizing an ob
 jective function over a large\, nonconvex space. This objective function r
 epresents a performance metric which may be modeled\, simulated\, or (if p
 ossible) measured. Each such objective function incurs different tradeoffs
  in terms of speed\, accuracy\, and the strength of results that can be fo
 rmally proven\, and as a result requires its own optimization methods.\n\n
 In this talk\, I will describe optimization approaches for several such mo
 dels.\nIn a simple communication model\, we derive an unconditional commun
 ication lower bound for convolutions and "projective" tensor operations\, 
 and show that this can always be attained (up to a constant factor) by sol
 ving a mathematical optimization problem.\nWe then add additional hardware
  constraints to this optimization program\, resulting in a fast one-shot m
 apper that encompasses loop tiling\, permutation\, and spatio-temporal ord
 ering.\nFurthermore\, we describe ways to incorporate measured or simulate
 d performance results into the objective for this mapping\, and discuss re
 cent progress on solving this problem when these results may be expensive 
 to collect.\n\nI’ll also describe ongoing work on extending this approac
 h to model performance\, compute lower bounds\, and determine optimization
 s for sparse tensor operations.
LOCATION:SS03\, Computer Laboratory\, William Gates Building
END:VEVENT
END:VCALENDAR