Xilinx/BSC HLS and OmpSs tutorial

Data: 12/Feb/2019 Time: 10:00 - 13/Feb/2019 Time: 18:00

Place:

Barcelona - UPC/BSC Campus

Building A6. Classroom A6204 (second floor of Building A6, Aula A6204).

Cost: Cost: Free of charge. BSC will assume the cost of coffee breaks & lunch for each of the attendees.

Primary tabs

 Detailed agenda:

Feb. 12th, 2019 - Day 1

 10:00h Welcome (BSC/Xilinx)
 
 10:05h
  1- Introduction of Vivado HLS (Xilinx)
  2- Introduction of OmpSs@FPGA (BSC/UPC)
  3- Environment explanation of OmpSs@FPGA (BSC/UPC)

      - Generate to HLS a Matrix Multiply (no optimizations)

 11:30h Coffee Break

  4-Simulation with HLS (Hands-on Xilinx/OmpSs@FPGA guided)
    Analysis of the HLS project created by the OmpSs@FPGA toolchain

 13:30h Lunch

 14:00-18:00h

 5- Opt. RGB-YUV filter using Vivado HLS and OmpSs@FPGA (Hands-on Xilinx + OmpSs@FPGA team)
Objective: Analysis and apply optimizations to be able, later on, to execute and obtain a benefit.

        5.1- Timing of naive solution (ARM binary + bitstream) of the code 
            using OmpSs@FPGA (provided with the material)
        5.2- Make a OmpSs@FPGA porting annotating the original code with OmpSs
            directives
        5.3- Use OmpSs@FPGA toolchain to compile the code up to HLS to obtain 
            the Vivado HLS project
        5.4- Open Vivado HLS to perform the analysis using the Vivado HLS 
            features
        5.5- Add Vivado HLS and analyse their performance impact:
                Directives: tripcount, offload inlining, pipelining, dataflow,
                   fifo depth
        5.6- Use OmpSs@FPGA toolchain to compile from HLS to bitstream and 
                compare execution time to the naive version


Feb. 13th, 2019 - Day 2

 10:00h Recap first day

   10:15h
    6- Optimize DCT using OmpSs@FPGA and Vivado HLS directives

       Objective: Analysis and optimization 

        6.1- Timing of naive solution (ARM binary + bitstream) of the code 
           using OmpSs@FPGA (provided with the material)
        6.2- Make a OmpSs@FPGA porting annotating the original code with OmpSs
           directives
        6.3- Use OmpSs@FPGA toolchain to compile the code up to HLS to obtain
           the Vivado HLS project


   11:30h Coffee Break

   12:00h - 16:00h (Lunch at 13:30h)

        6.4- Open Vivado HLS to perform the analysis using the Vivado HLS 
           features
        6.5- Add Vivado HLS and analyse their performance impact:
           Directives:  pipeline (inner and outer loop), data partition/
              resource analysis, array partition, dataflow, inline, reshaping
        6.6- Use OmpSs@FPGA toolchain to compile from HLS to bitstream and 
           compare execution time to the naive version


   16:00h
     7-Hardware instrumentation (Internal to the accelerators)

     8-Work on user provided applications 
           (or propose optimizations for Matrix Multiply)

   17:30h Recap of the tutorial

  18:00h Adjourn