TODO.md

Current Roadmap

Article

Choice for Parallel Computing

• Template in output folder (Jens)
• Check if limit of 15 pages is string or not ?
• First title and structure (Tibo)
• Test case for performance model validation
  • definition of test case
  • link with multi-modes Dahlquist analysis

Library, Analysis

• Migration to plotly for every plots
  • contours for in log scale (discretization error, PinT error)
  • contours in linear scale (nIter, speedup, efficiency)
  • 2D plots
• Code cleaning / experiments to checks every edge cases
• 3-level analysis :
  • symbolic computation (Jens)
  • computation per hand for reference (Tibo)
• template for memo format
• reduction of computation time for schedule (Jens)
• addition of other scheduler (bonus, Jens)
• proper definition and settings for block costs
• validation blockops
  • pyMGRIT reference script with Dahlquist (Jens)
  • comparison 2-level F(CF) relaxation VS pyMGRIT
  • comparison 2-level PFASST VS pySDC

Website

• First application with time discretization accuracy / stability (Tibo)
• Second application with cost and schedule (Jens)

Additional Ideas

Compare PinT efficiency

• Script to plot the speedup/efficiency in function of lambda*dt on the with one given lambda value (1D)
• Choose T and nProcs, with fixed discretization error (dt fine), and compute speedup for different block size (with restard and windowing ...)