1. UC Berkeley BRASS Group Post Placement C-Slow Retiming for Xilinx Virtex FPGAs Nicholas Weaver Yury Markovskiy Yatish Patel John Wawrzynek UC Berkeley Reconfigurable Architectures, Systems, and Software (BRASS) Group ACM Symposium on Field Programmable Gate Arrays (FPGA) February 2x, 2003 http://www.cs.berkeley.edu/~nweaver/cslow.html

2. UC Berkeley BRASS Group Automatic C-Slow Retiming for Virtex FPGAs 2 Outline “Automatically Double Your Throughput” –“You paid for those registers, here’s how to use them” Retiming and C-slow Retiming –The transformation C-slow Retiming and the Virtex FPGA –The target Retiming 3 Benchmarks –The tests

3. UC Berkeley BRASS Group Automatic C-Slow Retiming for Virtex FPGAs 3 Retiming and Repipelining Retiming –Automatically moving registers to minimize the clock period –Benefits limited by the number of registers –Algorithm developed by Leiserson et al Repipelining –Adding registers to the front or back –Let retiming then move them around But What About Feedback Loops? –Retiming and repipelining are of limited benefit when you have feedback loops

4. UC Berkeley BRASS Group Automatic C-Slow Retiming for Virtex FPGAs 4 C-Slow Retiming Replace every register with a sequence of C registers. –With more registers retiming can break the design into finer pieces –Again proposed by Leiserson et al, to meet systolic slowdown Semantic altering transformation –But resulting semantics are predictable and useful Ideal: C-slow in synthesis, retime after placement Our prototype: C-slow and retime after placement

5. UC Berkeley BRASS Group Automatic C-Slow Retiming for Virtex FPGAs 5 Design Semantics After C-Slowing Design operates on C independent data streams –Data streams are externally interleaved on round robin basis Semantics apply to designs with Task Level Parallelism –Encryption Counter (CTR) mode works on independent blocks –Sequence matching Compare sequence vs database C-slowing improves throughput but adds latency and registers

6. UC Berkeley BRASS Group Automatic C-Slow Retiming for Virtex FPGAs 6 C-slowing, Retiming, and the Virtex FPGA Every 4-LUT has associated register –Register can, almost always, be used independently of the LUT LUTs can act as clocked shift registers (SRL16s) –Used in our AES hand-benchmark –Not used in our tool Many designs have low register utilization –Excess of registers available in unoptimized designs Retiming best performed with/after placement –Xilinx placement operates on mapped slices –Need net delay information for better results F1 F2 F3 F4 BX X XB 4-LUT

7. UC Berkeley BRASS Group Automatic C-Slow Retiming for Virtex FPGAs 7 Sketch of Tool’s Operation 1.Convert.ncd to.xdl after placement 2.Load design into graph representation 3.Replace registers with edge annotations to represent registers 4.Replace every single register with C registers 5.Compute costs based on delay model 6.Retime 7.Convert edge annotations back to instance registers 8.Write out.xdl, convert to.ncd 9.Route PlacerRouter.xdl 1 1 1 2 2 2 2 2 2 1.1 1.3 0.9 2.2 1.4 1.6 1 1 2 1.1 1.3 0.9 2.2 1.4 1.6 1 1 1.1 1.3 0.9 2.2 1.4 1.6.xdl

8. UC Berkeley BRASS Group Automatic C-Slow Retiming for Virtex FPGAs 8 Experiment 1: How Good is the Tool? Tool is a simple prototype –Manhattan distance delay estimate –No attempt to minimize flip-flops –Basic flip-flop allocation Two benchmarks: AES and Smith/Waterman –Hand mapped –(optionally) hand placed –(optionally) hand C-slowed and retimed Our Best hand AES implementation –1.3 Gb/s –<800 Slices, 10 BlockRAMs –$10 part, Spartan II-100

9. UC Berkeley BRASS Group Automatic C-Slow Retiming for Virtex FPGAs 9 Experiment 1: AES, Automatically Placed VersionClock Rate (Throughput) Stream Clock Rate (1 / Latency) Initial Design48 MHz 5-Slow by hand105 MHz21 MHz Retimed Automatically47 MHz 2-Slow Automatically64 MHz32 MHz 3-Slow Automatically75 MHz25 MHz 4-Slow Automatically87 MHz21 MHz 5-Slow Automatically88 MHz18 MHz Just retiming is of no benefit Automatic C-slowing very effective –But could do even better

10. UC Berkeley BRASS Group Automatic C-Slow Retiming for Virtex FPGAs 10 Experiment 1: Smith/Waterman, Automatically Placed VersionClock Rate (Throughput) Stream Clock Rate (1 / Latency) Initial Design43 MHz 4-Slow by hand90 MHz22 MHz Retimed Automatically40 MHz 2-Slow Automatically69 MHz34 MHz 3-Slow Automatically84 MHz28 MHz 4-Slow Automatically76 MHz25 MHz Again, just retiming is of no benefit C-slowing highly effective –Within 7% of hand-built implementation

11. UC Berkeley BRASS Group Automatic C-Slow Retiming for Virtex FPGAs 11 Experiment 1: Comments Just retiming is of no benefit –Both designs limited by single cycle feedback loops C-Slowing very effective –Able to automatically nearly double throughput Hand implementations more than doubled throughput –Reasonable numbers of additional registers Limitations of prototype tool: –Flip-flop allocation routines could be better –Some AES hand benchmarks used SRL16 delay chains Simple is pretty good –Relatively simplistic implementation gets reasonably close to hand-mapped performance

12. UC Berkeley BRASS Group Automatic C-Slow Retiming for Virtex FPGAs 12 Experiment 2: Retiming LEON Can we automatically C-slow a large, synthesized design? Leon 1: A synthesized, GPLed SPARC compatible microprocessor core [1] –5 stage pipeline, integer only –Modify register file to use BlockRAMs BlockRAMs are used as negative edge devices –Remove caches, I/O, etc –Synthesize, using Symplify with CEs disabled –Edit EDIF to replace Sets/Resets Retime and C-slow with prototype tool –Prototype tool converts BlockRAMs to positive edge C-slow a microprocessor core... –Get an interleaved multithreaded architecture [1] Leon 1, by Jiri Gaisler, http://www.gaisler.com/leonmain.html

13. UC Berkeley BRASS Group Automatic C-Slow Retiming for Virtex FPGAs 13 Experiment 2: Results VersionClock Rate (Throughput) Thread Clock Rate (Latency) Lut Associated Flip Flops Lut Independent Flip Flops Initial Design23 MHz 1611NA Retimed Automatically25 MHz 2398194 2-Slow Automatically46 MHz23 MHz2150388 3-Slow Automatically47 MHz16 MHz24383713 6132 Luts for all designs Retiming alone worked surprisingly well 2-slowing very effective 3-slowing hit diminishing returns

14. UC Berkeley BRASS Group Automatic C-Slow Retiming for Virtex FPGAs 14 Experiment 2: Comments Retiming alone worked surprisingly well –Tool automatically converted BlockRAMs to positive-edge clocking and rebalanced the pipeline 2-slowing very effective –Effectively doubled the initial throughput NO slowdown in latency over initial design because retiming was effective without C-slowing –Used more many registers, but fewer registers than LUTs 3-slowing hit diminishing returns –Too many registers required combined with poor register allocation  poor performance

15. UC Berkeley BRASS Group Automatic C-Slow Retiming for Virtex FPGAs 15 Conclusions: C-slow retiming is very effective –"Automatically double your throughput" Benefits: More throughput Costs: More Flip Flops, worse latency Post-placement retiming appropriate –Independent Flip Flop usage critical –Have delay model for interconnect as well as logic Some room for improvement –Faster/Better implementation Minimize Flip Flop usage as well as delay Use SRL16s Better placement of Flip Flops –Experience suggests more Flip Flops/LUT would be useful

16. UC Berkeley BRASS Group Automatic C-Slow Retiming for Virtex FPGAs 16 Backup Slide: Why Not Use (Current) Synthesis Tools? Many synthesis tools support retiming, but with caveats: –ONLY works for synthesized items AES and Smith/Waterman didn't use synthesis –Can't automatically C-slow –Can't retime through memory blocks –Can't accurately guesstimate interconnect delay before placement >½ of the delay is the interconnect –Can't effectively scavenge unused flip-flops before placement Xilinx placement operates on slices, not luts

17. UC Berkeley BRASS Group Automatic C-Slow Retiming for Virtex FPGAs 17 Backup Slide: Why the limitations on total speedup? Absolute maximum –Interconnect + LUT + Flip-Flop Practical maximums –Too many flip-flops to allocate “Only” one flip-flop per LUT available –Flip-flop allocation poor Quick and dirty greedy heuristic –Works well for mild C-slowing –Fails with highly aggressive C-slowing –Tool doesn’t minimize flip-flops –Critical path is defined by the single worst path –Tool uses “Cheap and dirty” interconnect delay model

18. UC Berkeley BRASS Group Automatic C-Slow Retiming for Virtex FPGAs 18 (Backup Slide) :Design Restrictions to Enable C-slowing Resets and Clock Enables –Convert to explicit logic Memories –Increase by a factor of C Add high bits of addr to provide round-robin access Every stream sees an independent memory Global Set/Reset –Convert to individual resets –Still highly restrictive Interleave/deinterleave IO –Requires external logic No asynchronous sets/resets Din Dout Addr WE Din Dout Addr WE Thread Counter

19. UC Berkeley BRASS Group Automatic C-Slow Retiming for Virtex FPGAs 19 Scrap Image

20. UC Berkeley BRASS Group Automatic C-Slow Retiming for Virtex FPGAs 20 Scrap Image 2-

21. UC Berkeley BRASS Group Automatic C-Slow Retiming for Virtex FPGAs 21 Scrap Image 3 Din Dout Addr WE Din Dout Addr WE Thread Counter

22. UC Berkeley BRASS Group Automatic C-Slow Retiming for Virtex FPGAs 22 Scrap Image 4

23. UC Berkeley BRASS Group Automatic C-Slow Retiming for Virtex FPGAs 23 Scrap 5 1 1 1 2 2 2 1.1 1.3 0.9 2.2 1.4 1.6 1 1 2 1.1 1.3 0.9 2.2 1.4 1.6 1 1

24. UC Berkeley BRASS Group Automatic C-Slow Retiming for Virtex FPGAs 24 Scrap 6 1.1 1.3 0.9 2.2 1.4 1.6