1. Reconfigurable Computing - Assignment Feedback John Morris Chung-Ang University The University of Auckland ‘Iolanthe’ at 13 knots on Cockburn Sound, Western Australia

2. Assignment Feedback  Grade  Your group has been assigned an indicative grade  A final grade for this assignment will be set when Stage B is completed  Thus you have a chance to upgrade (or downgrade!) your final result!  There is no correlation between the amount of written comments and the quality  When a report was already good, comments on many small details may have been added to show how to make it ‘perfect’ !

3. Assignment Feedback  Adders  Should have TWO architectures!  One for a ‘standard’ implementation using full adder components  One using Altera’s (or Xilinx’ or …) fast carry logic

4. Ripple Carry Adder  ENTITY  Formally describes the interface for this model  Common  Used by both implementations of the adder ENTITY adder IS GENERIC( n_bits : positive := 8 ); PORT( a, b : IN std_logic_vector; c : IN std_logic; sum: OUT std_logic_vector; c_out : OUT std_logic ); END ENTITY adder;

5. Ripple Carry Adder  First ARCHITECTURE ENTITY adder IS GENERIC( n_bits : positive := 8 ); PORT( a, b : IN std_logic_vector; c : IN std_logic; sum: OUT std_logic_vector; c_out : OUT std_logic ); END ENTITY adder; ARCHITECTURE simple OF adder IS COMPONENT full_adder … SIGNAL c_int: std_logic_vector( n_bits DOWNTO 1 ); BEGIN FOR J IN n_bits DOWNTO 1 GENERATE fa: full_adder PORT MAP( … ); END GENERATE; END ARCHITECTURE simple;

6. Ripple Carry Adder  Second ARCHITECTURE ENTITY adder IS GENERIC( n_bits : positive := 8 ); PORT( a, b : IN std_logic_vector; c : IN std_logic; sum: OUT std_logic_vector; c_out : OUT std_logic ); END ENTITY adder; ARCHITECTURE fast_carry OF adder IS COMPONENT lpm_add_sub … BEGIN rc: lpm_add_sub PORT MAP( … ); END ARCHITECTURE fast_carry ;

7. Configurations  VHDL’93 allows you to write a configuration  This capability is very important!  It allows you to  Check that the substitution of an improved architecture does not alter the function (correctness) of a system  Substitute different versions of basic modules in different parts of your design, eg  fast adder when speed is important  small adder when space is important CONFIGURATION std OF test_bench IS FOR test_bench_architecture FOR ALL: adder USE ENTITY work.adder( simple ); END FOR; END FOR; END std; Specifies the architecture to be used

8. Configurations  VHDL’93 allows you to write a configuration  The full capabilities of a configuration allow  Specification of an architecture for all instances of a model  Specification of architectures for individual (labelled) instances  An instantiation has a label adder_a : adder PORT MAP( a=>.., b=>…, … ); Used to identify a component in simulation  Substitution of equivalent entities  ‘Equivalent’ = having compatible interfaces CONFIGURATION std OF test_bench IS FOR test_bench_architecture FOR ALL: adder USE ENTITY work.adder( simple ); END FOR; END FOR; END std; Specifies the architecture to be used

9. Configurations (and Altera Limitations)  VHDL’93 allows you to write a configuration  Unfortunately, Altera’s software is, in some respects, primitive!!  It does not support configuration ie it does not allow you to specify which architecture to use when instantiating a component  It also insists that the ENTITY and ARCHITECTURE parts are in the same file  Preventing you from separating them!  Good design separates interfaces and implementations! Separating the abstract from the concrete  It probably has some good points though! eg the synthesizer and simulator appear to be integrated well CONFIGURATION std OF test_bench IS FOR test_bench_architecture FOR ALL: adder USE ENTITY work.adder( simple ); END FOR; END FOR; END std;

10. Altera - Getting around the problem 1.Copy (link) everything into a different directory 2.Alter one of the architectures 3.Fine until you change something else!  This is the reason that we want to have configuration capabilities So that we could work with two (or more) architectures  Do NOT try renaming one of the entities!  Unless you need to use BOTH architectures in the same system!  Altera only!

11. Abstract designs  Copying and renaming defeats the idea that you can have  the abstract part - a common interface  represented by an ENTITY  and  concrete parts or different implementations  represented by different ARCHITECTURES  Abstraction allows you to design at a high level  Concentrate on the major functions of a system … and leave low level details until later  Example  A washing machine clearly needs a timer  Specify a generic timer By writing out its entity  Decide how to implement the timer later

12. Abstract designs  Concept: designs have two parts  the abstract part - a common interface and  (several) concrete parts or implementations  Designing at a high level: Example B  At this point in the system, we need an adder but we know several types of adder that we could use  Defer the decision as to what type of adder until more is known about the full system  We may need a fast one, a compact one, a really small one (bit serial?)  At an early stage in design, the only critical thing is that it adds correctly!  You might even use the multiple architectures capability to test various implementations to find the best one

13. Assignment Summary  The assignment specification called for a short report ‘summarizing what you have done so far’  Some of you interpreted ‘short’ in an extreme way!!  Ideally, your summary should have consisted of 1.a table with resource usage (# of logic blocks) and times for various adder configurations 2.some text commenting on this table Which configuration is best for each adder width? Configuration with minimum delay should have been highlighted, marked or mentioned How much does the resource usage differ? If the time advantage is small, you might prefer the smaller structure!  If there were anomalies or unexpected results  They should have been explained or at least (if you were unable to explain them)  mentioned as unexpected

14. Assignment Summary  Most of you omitted the ‘base’ configuration for an n -bit adder – a simple n -bit ripple carry adder  The 1  n configuration  Requires you to synthesize the ripple carry adder only!  Actually important because  Fast carry logic  almost as fast as the CS adder  Only ~ half the resources (or space) required!  One important design question  Use a simple ripple carry adder? Not too slow if the fast carry logic is used and Uses considerably less space Simple to implement ( 1 line of VHDL) Regular or  Try to implement a faster adder? Is it really needed?

15. Assignment - Details  The following details are important and were commonly omitted  Which software package was used?  Altera’s MaxPlus, Quartus, Xilinx’s Webpack, etc  Precise explanation of measurement unit for resources  What does a slice mean?  Even the term ‘logic block’ has a different sense for different FPGAs Xilinx CLB has 2 FF’s, 9 inputs, 2 outputs Altera Logic Element has 1 FF, 4 inputs, 1 output Quicklogic has …  So resource columns should be headed Altera Flex10K logic cells or Xilinx 4000 series CLBs or …. ie be precise!

16. Delay Estimates and Accuracy  Do not take the figures for delays reported by the tools as perfectly accurate!!  Delays are computed by summing the propagation delays through every element on a signal’s path These elements include  Transmission gates (probably several types)  Multiplexers  Look up tables (LUTs)  Flip flops and sometimes  Buffers  Other logic gates  Allowance must also be made for heavily loaded lines  Thus each reported path delay might be a sum of a 100 or more elementary delays

17. Delay Estimates and Accuracy  Delays  For a given technology, the manufacturers estimate (by measurement or calculation) the propagation delays of each of these types of elements  t pd for a real circuit is a function of  Temperature  Supply Voltage  Circuit dimensions Width of gates Length of channels, etc  Even if the voltage and circuit dimensions are precisely known, t pd values are only accurate for one temperature  Circuits are usually rated for  Commercial - 0 o C – 70 o C  Industrial - 0 o C - ~100 o C  Military, space – even wider range

18. Delay Estimates and Accuracy  Delays  Combined with manufacturing tolerances for circuit dimensions  Track and gate widths  Size of etched, implanted, … regions  It is unlikely that a propagation delay for an individual circuit element is accurate to < 100ps or 0.1ns over even a small T range  If several hundred figures with errors of  0.1ns are added, the error in the sum can be very large!  A predicted delay is probably only good to  1ns  … and could be much less accurate!  Values in your report should reflect this!!  Thus you should NOT copy the 55.828ns figure from the synthesis report  A realistic estimate is probably 56ns  This is the value which should appear in your report  55.8ns is tolerable, but probably completely unrealistic!

19. Delay Estimates and Accuracy  Delays  Selecting the best configuration  If the synthesizer reports t pd = 23ns for configuration A and t pd = 25ns for configuration B  Is it safe to assume that A at 23ns is really faster?  Almost certainly not!  It may be … but on a chip with slightly different track widths or running at a slightly different T or V dd  Then the relative speeds may be different and B may be faster  Again your report should reflect this  It should not claim that a configuration differing from another by a small margin is faster  It should note the similarity  and  look at other factors – such as size, regularity, …

20. Finally, English - The hardest bit? 1.Technical Reports should be mainly written in the simple past tense  Use the simple past tense in the active voice  Errors in measurement caused …  I/we implemented …  This result confirmed … and in the passive voice  this effect was caused by …  the designs were implemented in …  our hypotheses were supported by these observations 2.Acronyms  Explain them before you use them!  Several used ‘RCA’ for ripple-carry adder  Not generally accepted (even in this context) so must be spelt out in full before use  VHDL and FPGA are OK (in this context) though!

21. Finally, English …  The hardest bit? 1.Articles  Even native speakers have difficulty explaining the rules!  I’ve tried to correct most errors  See if you can follow the patterns in my use