Skip to content

Latest commit

 

History

History
136 lines (85 loc) · 5.17 KB

2021-10-18-minutes.md

File metadata and controls

136 lines (85 loc) · 5.17 KB
Raw

Embench Meeting Minutes

Monday 18 October 2021

  • 08:00-09:00 Pacific Time
  • 11:00-12:00 East Coast Time
  • 15:00-16:00 UTC
  • 16:00-17:00 UK Time
  • 17:00-18:00 Central European Time
  • 23:00-00:00 Beijing Time

Summary of actions

  • Ray Simar: update on the status of the RISC-V Summit lightning talk

  • Ofer Shinaar: would like to discuss embench-rt running on Cortex M4

  • Nidal Faour: update on the new build system

  • Paolo Savini to prepare a short presentation to address the doubts about the execution speed measurements #137 Understanding Benchmark Speed score. Investigate the possibility and benefits of adding an instruction count alongside the existing cycle count measurement.

Agenda topics

Welcome

Welcome to H. Hayashi and thanks for the recent contributions to Embench.

Actions from previous meeting

  • Ray Simar: we missed the deadline for the RISC-V Summit lightning talk. More work to present for next time. Other venues could be used.

  • Ofer Shinaar: #4 Promote Latency result. #2 new context switch benchmark

    Call for comments or any concern. Any volunteer to port the latency support to the Cortex M3 is welcome. Ready to merge and publish results once these are solved.

  • Nidal Faour: demonstrated the proposed build system. Observations and key points:

    • portability. The new system has self-maintainence.
    • security when automatic download of packages is involved. Dockers can prevent internet accesses from the packages.
    • run in an RTL environment.
    • benchmark build systems need simplicity of use and structure.

    Given the key points above the build systems appears to be more mature and easy to maintain than the current python scripts. A switch to the new build system will be considered for a few months from now. Volunteers are welcome to try it. Current volunteers:

    • Bill McSpadden
    • Jeremy Bennett and Paolo Savini
    • H. Hayashi

    Some work is still needed to ratify and open source the build system.

  • Paolo Savini: presentation deferred to next meeting.

GitHub pull requests

From embench-iot:

The following new pull requests have been reviewed and merged.

  • #150 Reworded decription of the capabilities of x86 and M1 macs to run x86 and M1 binaries
  • #146 Tarfind: fix heap size to be multiple of the pointer size

The following pull requests are still open and under review:

  • #148 A Board Configuration Example for The Freedom E310 Arty FPGA Dev Kit
  • #147 md5sum: make calloc() use calloc_beebs() and use XOR for result value
  • #145 add --benchmark and --exclude option to the benchmark scripts for debugging
  • #144 Replace pyelftools with lief in prerequisites
  • #133 run_all.py: Allow run benchmark runset from .json config.
  • #130 Improve JSON output of benchmarking.

From embench-rt:

The following pull requests is still open and waiting for last comments:

  • #2 new context switch benchmark

New GitHub open issues

From embench-iot:

  • #149 MD5 speed test not working on Mac

A total of 39 issues remain open.

From embench-rt:

  • #4 Promote Latency result

    • raised by Ofer Shinaar
    • assigned to Jeremy Bennett and Nati Rapaport

Further work on embench-rt

At a meeting earlier this year, there was a student volunteer to work on the Embench Real-Time project. Unfortunately, the minutes did not record the name of that volunteer. The Embench-RT team would be very pleased to make contact with the volunteer. Any other volunteer welcome.

New benchmarks for Embench IoT 2.0

The document is now open for submissions, for both new benchmarks and benchmarks to cull.

No new benchmarks were proposed this month.

Publicity

Noted that getting the word out about Embench. Target retrospective paper at Embedded World in February 2022.

Missed the deadline for a RISC-V Summit lightning talk. Will look into next year's event and other venues.

Suggest a presentation focusing on role of Embench within the academic community and working with industry to solve issue of pre-silicon (simulation) benchmarking.

Next meeting

We currently meet regularly on the third Monday of each month 8-9am California time.

The following dates are proposed for 2021

  • Mon 15 Nov 2021
  • Mon 13 Dec 2021 (week 2, to keep clear of Christmas)

Jeremy created a calendar entry for our regular attendees, to provide automated reminders. Please ask if you wish to be added to this.

AOB

Paolo Savini\