Smalltalk Debug Lives in the Matrix

1,382 views

Published on

Loïc Lagadec, Damien Picard: Smalltalk Debug Lives in the Matrix. ESUG 2010, Barcelona, Spain

Published in: Technology
0 Comments
1 Like
Statistics
Notes
  • Be the first to comment

No Downloads
Views
Total views
1,382
On SlideShare
0
From Embeds
0
Number of Embeds
3
Actions
Shares
0
Downloads
16
Comments
0
Likes
1
Embeds 0
No embeds

No notes for slide

Smalltalk Debug Lives in the Matrix

  1. 1. Smalltalk Debug Lives in the Matrix Loïc Lagadec, Damien Picard Loic.lagadec@univ-brest.fr
  2. 2. What Matrix? “Flexible” hardware Hard to program Time to market Hard to debug Compute node i1 LUT LUT LUT I1-i2 i2 E/S µP i1 LUT LUT I1+i2 i2 LUT Programmable LUT LUT LUT interconnection
  3. 3. What Matrix? “Flexible” hardware Hard to program Time to market Hard to debug Specific languages Specific tools Performances still requires manual tuning EE skills required
  4. 4. State of the art debugging Huge challenge
  5. 5. Touching the void State of the art debugging Huge challenge Meet in the middle
  6. 6. Debug silver bullet Observability Controlability Abstract analysis Fast changes Time to market Multiple runs Long cycles Simulation is not enough
  7. 7. Debug silver bullet Observability Controlability Abstract analysis Fast changes Time to market Multiple runs Long cycles Simulation is not enough
  8. 8. Operate at-speed Keep your speed-up alive Multiple runs prohibit any over time penalty
  9. 9. Operate in-situ Because only HW brings speed-up Debug requires extra circuitry
  10. 10. Debugging facilities are circuits
  11. 11. Debugging facilities are circuits
  12. 12. Debugging facilities are circuits
  13. 13. Observe and monitor Take decision You cannot watch everything
  14. 14. Abstract analysis: semantic needed From information to knowledge Signals vs Variables, etc. Composite pattern, polymorphism, etc.
  15. 15. From a technical point of view ?  Synthesis  Software abstraction Implementation in hardware From D. Picard’s ESUG 2009 talk
  16. 16. Full observabiliy isn’t scalable Focus on POI
  17. 17. Full observabiliy isn’t scalable Focus on POI
  18. 18. Take control Select your Become a time traveller
  19. 19. Smalltalk debugger Just fit approach Run code and catch exception Code hot replacement, variable update, etc. Step on or resume execution Possible rollback Multiple runs Breakpoints update, earlier exception Same conceptual behavior
  20. 20. Dodge bullets Once the time has stopped … just operate !
  21. 21. Bullet time explained Bullet time 21
  22. 22. Conditional probes Conditional probes offer the controlabily that lacks in commercial tools Observability can be gained at the cost of adding some variable look-up wires … But also using vendor’s tool such as Chipscope
  23. 23. Decomposition Red Pill Probes Application Application Inserted probe (Generated) Control panel Active probes Probes inspector
  24. 24. RedPill Flow extends Biniou (HLS) Application Probe High-level Insertion CDFG Low-level CDFG Multi-Level Simulator Synthesis Tools Probed Netlist Reconfigurable Architecture Biniou
  25. 25. Abstract analysis Encapsulate circuit modules as smalltalk blocks Enables soft and hard objects to communicate Delivers the power of Sunit to hardware
  26. 26. Abstract analysis (example)
  27. 27. Characterization tests & SUnit
  28. 28. How Many matrix?
  29. 29. Reboot the Matrix
  30. 30. Will you take the red pill? HLS (Biniou) offers a path from HL languages to circuits Vendors tools offer observability Red Pill offers controlability Object encapsulation offers abstract analysis and polymorphism. Smalltalk debug definitively lives in the Matrix
  31. 31. Thank you for your attention http://stiff.univ-brest.fr/BINIOU

×