The Verilog Hardware Description Language Professor Don Thomas Carnegie Mellon University (CMU) thomas@ece.cmu.edu http://www.ece.cmu.edu/~thomas This is not one cohesive presentation on Verilog. The slides contained here are collected from different CMU classes at various academic levels. These slides are provided as an alternate aid to learning the language. You may find them helpful. Send bug reports to the above address — there are some! The Verilog Hardware Description Language, Fifth Edition is available from Kluwer Academic Publishers, http://www.wkap.com. Phone: 781-871-6600. Use or reproduction of the information provided in this file for commercial gain is strictly prohibited. Explicit permission is given for the reproduction and use of this information in an instructional setting. Representation: Structural Models Structural models Are built from gate primitives and/or other modules They describe the circuit using logic gates — much as you would see in an implementation of a circuit. Identify: Gate instances, wire names, delay from a or b to f. This is a multiplexor — it selects one of n inputs (2 here) and passes it on to the output module mux (output f, input a, b, sel); a f b sel f= a • sel’ + b • sel and #5 g1 (f1, a, nsel), g2 (f2, b, sel); or #5 g3 (f, f1, f2); not g4 (nsel, sel); endmodule Representation: Gate-Level Models Need to model the gate’s: Function Delay Function Generally, HDLs have built-in gate-level primitives Verilog has NAND, NOR, AND, OR, XOR, XNOR, BUF, NOT, and some others The gates operate on input values producing an output value typical Verilog gate instantiation is: “many” optional and #delay instance-name (out, in1, in2, in3, …); and #5 g1 (f1, a, nsel); a comma here let’s you list other instance names and their port lists. Four-Valued Logic Verilog Logic Values The underlying data representation allows for any bit to have one of four values 1, 0, x (unknown), z (high impedance) x — one of: 1, 0, z, or in the state of change z — the high impedance output of a tri-state gate. What basis do these have in reality? 0, 1 … no question z … A tri-state gate drives either a zero or one on its output…and if it’s not doing that, its output is high impedance (z). Tri-state gates are real devices and z is a real electrical affect. x … not a real value. There is no real gate that drives an x on to a wire. x is used as a debugging aid. x means the simulator can’t determine the answer and so maybe you should worry! All values in a simulation start as x. BTW … Verilog keeps track of more values than these in some situations. Four-Valued Logic Logic with multi-level logic values Logic with these four values make sense Nand anything with a 0, and you get a 1. This includes having an x or z on the other input. That’s the nature of the nand gate Nand two x’s and you get an x — makes sense! Note: z treated as an x on input. Their rows and columns are the same If you forget to connect an input … it will be seen as an z. At the start of simulation, everything is an x. A B Input A Input B Nand 0 1 x z 0 1 1 1 1 1 1 0 x x x 1 x x x z 1 x x x A 4-valued truth table for a Nand gate with two inputs Concurrent activity Eval g2, g3 Do these two evaluations happen at the same time? Yes and No! Yes and No! Yes … They happen at the same simulated (or virtual) time After all, the time when they occur is 27 No … We all know the processor is only doing one thing at any given time So, which is done first? That’s undefined. We can’t assume anything except that the order is arbitrary. Concurrent Activity The point is In the real implementation, all activity will be concurrent Thus the simulator models the elements of the system as being concurrent in simulated time The simulator stands on its head trying to do this! Thus, Even though the simulator executes each element of the design one at a time … … we’ll call it concurrent Delay Transport delay — input to output delay “nand #35 (f1, a, b, c);” #35 is the transport delay What if the input changes during that time? i.e. how wide must an input spike be to affect the output. Think of the gate as having inertia. — The input change must be present long enough to get the output to change. (That “long enough” time is called inertial delay) in Verilog, this time is equal to the transport delay a b c a b c pulse too small, no output change — transport delay a b c Let’s build a wider 2-bit mux Build a 2-bit 2:1 mux OK, let’s put two 1-bit 2:1 muxes in the same module with a common select line What would it look like? a0 f0 b0 a f a1 f1 b1 b sel sel Reuse! Reuse of smaller objects Can we use the mux module that we already designed? A big idea — instantiation Modules and primitive gates can be instantiated — copied — to many sites in a design Previously, two ANDs, one OR, and a NOT gate were instantiated into module mux Now we instantiate two copies of module mux into module wideMux module wideMux (f1, f0, a1, a0, b1, b0, sel); input a1, a0, b1, b0, sel; output f1, f0; mux endmodule bit1 (f1, a1, b1, sel), bit0 (f0, a0, b0, sel); Instantiate two mux modules, name them, and specify connections (the order is important). Instantiation — Copies Modules and gate primitives are instantiated == copied Note the word “copies” The copies (also called instances) share the module (or primitive) definition If we ever change a module definition, the copies will all change too However, the internal entities (gate names, internal port names, and other things to come) are all private, separate copies Don’t think of module instantiations as subroutines that are called They are copies — there are 2 mux modules in wideMux with a total of: 4 AND gates, ______ 2 OR gates, ______ 2 Not gates ______ Why is this cool? In Verilog “Primitive” gates are predefined (NAND, NOR, …) Other modules are built by instantiating these gates Other modules are built by instantiating other modules, … The design hierarchy of modules is built using instantiation Bigger modules of useful functionality are defined You can then design with these bigger modules You can reuse modules that you’ve already built and tested You can hide the detail — why show a bunch of gates and their interconnection when you know it’s a mux! Instantiation & hierarchy control complexity. No one designs 1M+ random gates — they use hierarchy. What are the software analogies? How to wire modules together Real designs have many modules and gates module putTogether (); wire w1, w2, w3, w4; bbb aaa lucy ricky (w1, w2, w3, w4); (w3, w2, w1); … module bbb (i1, i2, o1, clk); input i1, i2, clk; output o1; xor (o1, i2, …); … Each module has it’s own namespace. Wires connect elements of namespaces. what happens when out1 is set to 1? module aaa (in1, out1, out2); input in1; output out1, out2; … nand nand … #2 (out1, in1, b); #6 (out2, in1, b); Implicit wires How come there were no wires declared in some of these modules? Gate instantiations implicitly declare wires for their outputs. All other connections must be explicitly declared as wires — for instance, connections between module ports Output and input declarations are wires module putTogether (); wire w1, w2, w3, w4; mux aaa inst1 (w1, w2, w3, w4); duh (w3, w2, w1); … wires explicitly declared module mux (output f, input a, b, sel); and #5 or #5 not endmodule g1 (f1, a, nsel), g2 (f2, b, sel); g3 (f, f1, f2); g4 (nsel, sel); wires implicitly declared (f1, f2, nsel) How to build and test a module Construct a “test bench” for your design Develop your hierarchical system within a module that has input and output ports (called “design” here) Develop a separate module to generate tests for the module (“test”) Connect these together within another module (“testbench”) module testbench (); wire l, m, n; design d (l, m, n); test t (l, m); initial begin //monitor and display … module design (a, b, c); input a, b; output c; … module test (q, r); output q, r; initial begin //drive the outputs with signals … Another view of this 3 chunks of Verilog, one for each of: TESTBENCH is the final piece of hardware which connects DESIGN with TEST so the inputs generated go to the thing you want to test... Another module, called TEST, to generate interesting inputs Your hardware called DESIGN An Example Module testAdd generated inputs for module halfAdd and displayed changes. Module halfAdd was the design module tBench; wire su, co, a, b; halfAdd testAdd endmodule ad(su, co, a, b); tb(a, b, su, co); module halfAdd (sum, cOut, a, b); output sum, cOut; input a, b; xor #2 and #2 endmodule (sum, a, b); (cOut, a, b); module testAdd(a, b, sum, cOut); input sum, cOut; output a, b; reg a, b; initial begin $monitor ($time,, “a=%b, b=%b, sum=%b, cOut=%b”, a, b, sum, cOut); a = 0; b = 0; #10 b = 1; #10 a = 1; #10 b = 0; #10 $finish; end endmodule The test module It’s the test generator $monitor prints its string when executed. after that, the string is printed when one of the listed values changes. only one monitor can be active at any time prints at end of current simulation time Function of this tester at time zero, print values and set a=b=0 after 10 time units, set b=1 after another 10, set a=1 after another 10 set b=0 then another 10 and finish module testAdd(a, b, sum, cOut); input sum, cOut; output a, b; reg a, b; initial begin $monitor ($time,, “a=%b, b=%b, sum=%b, cOut=%b”, a, b, sum, cOut); a = 0; b = 0; #10 b = 1; #10 a = 1; #10 b = 0; #10 $finish; end endmodule Another version of a test module Multi-bit “thingies” test is a two-bit register and output It acts as a two-bit number (counts 0001-10-11-00…) Module tBench needs to connect it correctly — mod halfAdd has 1bit ports. module tBench; wire su, co; wire [1:0] t; halfAdd testAdd endmodule ad(su, co, t[1], t[0]); tb(t, su, co); module testAdd(test, sum, cOut); input sum, cOut; output [1:0] test; reg [1:0] test; initial begin $monitor ($time,, "test=%b, sum=%b, cOut=%b", test, sum, cOut); test = 0; #10 test = test + 1; #10 test = test + 1; #10 test = test + 1; #10 $finish; end endmodule Connects bit 0 or wire t to this port (b of the module halfAdder) Another version of testAdd Other procedural statements You can use “for”, “while”, “if-then-else” and others here. This makes it easier to write if you have lots of input bits. module tBench; wire su, co; wire [1:0] t; halfAdd testAdd endmodule ad(su, co, t[1], t[0]); tb(t, su, co); module testAdd(test, sum, cOut); input sum, cOut; output [1:0] test; reg [1:0] test; initial begin $monitor ($time,, "test=%b, sum=%b, cOut=%b", test, sum, cOut); for (test = 0; test < 3; test = test + 1) #10; #10 $finish; end endmodule hmm… “<3” … ? Structural vs Behavioral Models Structural model Just specifies primitive gates and wires i.e., the structure of a logical netlist You basically know how to do this now. Behavioral model More like a procedure in a programming language Still specify a module in Verilog with inputs and outputs... ...but inside the module you write code to tell what you want to have happen, NOT what gates to connect to make it happen i.e., you specify the behavior you want, not the structure to do it Why use behavioral models For testbench modules to test structural designs For high-level specs to drive logic synthesis tools How do behavioral models fit in? How do they work with the event list and scheduler? Initial (and always) begin executing at time 0 in arbitrary order They execute until they come to a “#delay” operator They then suspend, putting themselves in the event list 10 time units in the future (for the case at the right) At 10 time units in the future, they resume executing where they left off. Some details omitted ...more to come module testAdd(a, b, sum, cOut); input sum, cOut; output a, b; reg a, b; initial begin $monitor ($time,, “a=%b, b=%b, sum=%b, cOut=%b”, a, b, sum, cOut); a = 0; b = 0; #10 b = 1; #10 a = 1; #10 b = 0; #10 $finish; end endmodule Two initial statements? … initial begin a = 0; b = 0; #5 b = 1; #13 a = 1; end … initial begin out = 1; #10 out = 0; #8 out = 1; end … 1 a 0 1 b 0 1 out 0 0 10 Things to note Which initial statement starts first? What are the values of a, b, and out when the simulation starts? These appear to be executing concurrently (at the same time). Are they? 18 arbitrary x Yes, in simulated time Other things you can do More than modeling hardware $monitor — give it a list of variables. When one of them changes, it prints the information. Can only have one of these active at a time. e.g. … $monitor ($time,,, “a=%b, b=%b, sum=%b, cOut=%b”,a, b, sum, cOut); extra commas print spaces %b is binary (also, %h, %d and others) The above will print: 2 a=0, b=0, sum=0, cOut=0<return> $display() — sort of like printf() in C $display (“Hello, world — %h”, hexvalue) display contents of data item called “hexvalue” using hex digits (0-9,A-F) What if what you print has the value x or z? newline automatically included One Modern Design Methodology always mumble mumble blah blah gates, gates, gates, … Synthesis Synthesizable Verilog Place and Route LE 1 LE 2 Logic Elements in FPGA Chip What’s a Logic Element (LE) A mux selects which element of memory to send to output Arbitrary programmable Boolean function of K inputs K=4 in our particular example. Usually see K=3, 4, 5 in real FPGAs 0000: 0 0001: 0 0 1 0 0 0010: It has a memory — you download to the memory to program the device 0011: You also program connections between these Logical Elements 0111: Synthesis tool partitions logic into groups of 5-input functions 1011: 0100: 0101: 0110: 1000: 1001: 1010: 1100: 1101: 1110: 1111: 0 1 0 0 0 16:1 mux 1 1 1 1 1 Really just a 1-bit memory BD AC F What do we mean by “Synthesis”? Logic synthesis A program that “designs” logic from abstract descriptions of the logic takes constraints (e.g. size, speed) uses a library (e.g. 3-input gates) How? You write an “abstract” Verilog description of the logic The synthesis tool provides alternative implementations constraints Verilog blah blah blah synthesis library or … An example What’s cool? You type the left, synthesis gives you the gates It used a different library than you did. (2-input gates only) One description suffices for a variety of alternate implementations! ... but this assumes you know a gate level implementation — that’s not an “abstract” Verilog description. module gate (output f input a, b, c); and A (a1, a, b, c), B (a2, a, ~b, ~c), C (a3, ~a, o1); or D (o1, b, c), E (f, a1, a2, a3); endmodule a b c f What Do We Want Here…? Goal To specify a combination ckt, inputs->outputs… … in a form of Verilog that synthesis tools will correctly read … and then use to make the right logic And... We know the function we want, and can specify in C-like form... … but we don’t now the exact gates (nor logic elements); we want the tool to figure this out. A B C Combinational Logic F Behavioral Modeling Procedural statements are used Statements using “always” Verilog construct Can specify both combinational and sequential circuits Normally don’t think of procedural stuff as “logic” They look like C: mix of ifs, case statements, assignments … … but there is a semantic interpretation to put on them to allow them to be used for simulation and synthesis (giving equivalent results) Current technology You can do combinational (and later, sequential) design Sizable designs can take hours … days … to run Companies pay $50K - 80K per copy for such software This ain’t shrink-wrap software! The software we use is more like $15-20K Behavioral Constructs Behavioral descriptions are introduced by initial and always statements Statement Looks like initial initial begin … end always always begin … end Starts How it works Execute once and stop Starts when simulation Continually loop— starts … in while (power on) arbitrary do statements; order Use in Synthesis? No, used as testbench Yes, used in synthesis Points: They all execute concurrently They contain procedural statements like if-then-else, case, loops, functions, … Statements, Registers and Wires Registers Define storage, can be 1-bit or more Can only be changed by assigning value to them on the left-hand side of a behavioral expression. Wires (actually “nets”) Electrically connect things together Can be used on the right-hand side of an expression Thus we can tie primitive gates and behavioral blocks together! Statements left-hand side = right-hand side left-hand side must be a register Four-valued logic Multi-bit registers and wires Logic with registers and wires module silly (input [3:0] q, r); reg [3:0] a, b; always begin … a = (b & r) | q; … q = b; … end endmodule Can’t do — why? Behavioral Statements if-then-else What you would expect, except that it’s doing 4valued logic. 1 is interpreted as True; 0, x, and z are interpreted as False if (select == 1) f = in1; else f = in0; case What you would expect, except that it’s doing 4valued logic If “selector” is 2 bits, there are 42 possible caseitems to select between There is no break statement — it is assumed. Sized, 4-valued constants The first number is the number of bits, the letter is the base of the following number that will be converted into the bits. 8’b00x0zx10 case (selector) 2’b00: a = b + c; 2’b01: q = r + s; 2’bx1: r = 5; default: r = 0; endcase assume f, a, q, and r are registers for this slide Behavioral Statements Loops There are some restrictions on using these for synthesis for this course — don’t. They are mentioned here for use in test modules Two main ones — for and while Just like in C There is also repeat and forever — see the book reg [4:0] testOutput, i; … for (i = 0; i < 15; i = i + 1) begin testOutput = i; #20; end reg [4:0] testOutput, i; … i = 0; while (i < 15) begin testOutput = i; #20 i = i + 1; end Important: Loops must have a delay operator (or as we’ll see later, an @ or wait(FALSE)). Otherwise, the simulator never stops executing them. Concurrent Constructs We already saw #delay Others @ … Waiting for a change in a value — used in synthesis @ (var) w = 4; This says wait for var to change from its current value. When it does, resume execution of the statement by setting w = 4. Wait … Waiting for a value to be a certain level — not used in synthesis wait (f == 0) q = 3; This says that if f is equal to zero, then continue executing and set q = 3. But if f is not equal to zero, then suspend execution until it does. When it does, this statement resumes by setting q = 3. Why are these concurrent? While one model waits for something, that something happens (concurrently) in another model FAQs: behavioral model execution How does an always or initial statement start That just happens at the start of simulation — arbitrary order Once executing, what stops it? Executing either a #delay, @event, or wait(FALSE). All always blocks need to have at least one of these. Otherwise, the simulator will never stop running the model — (it’s an infinite loop!) More FAQs How long will it stay stopped? Until the condition that stopped it has been resolved #delay … until the delay time has been reached @(var) … until var changes wait(var) … until var becomes TRUE Does time pass when a behavioral model is executing? No. The statements (if, case, etc) execute in zero time. Time passes when the model stops for #, @, or wait. Will an always stop looping? No. But an initial will only execute once. A Combinational Circuit Using behavioral constructs Logic for a simple MUX is specified procedurally here This example is synthesizable module mux (output reg f, input sel, b, c); always @ (sel or b or c) begin if (sel == 1) f = b; else f = c; end endmodule Read this as follows: Wait for any change on sel, b, or c, then execute the begin-end block containing the if. Then wait for another change. This “if” functionally describes the MUX c f b sel Example of Logic Synthesized Huh? Is it really correct? WWWWaaaaaiiiiiitttt a minute? Where’s the register? The synthesis tool figures out that this is a combinational circuit. Therefore, it doesn’t need a register. The register is there as an “artifact” of the description — things on the left-hand side have to be registers. How does it figure out that this is combinational? The output is only a function of the inputs (and not of previous values) Anytime an input changes, the output is reevaluated From outside the module, could you tell that there is a register in there? Does the loop have state? c f b ? module mux (output reg f, input sel, b, c); always @ (sel or b or c) begin if (sel == 1) f = b; else f = c; end endmodule Synthesis Template Using procedural statements in Verilog Logic is specified in “always” statements (“Initial” statements are not allowed). Each “always” statement turns into Boolean functions module blah (output reg f, input a, b, c); always @ (a or b or c) begin stuff... stuff... stuff... end endmodule You have to declare the combinational outputs like this, for synthesis. i.e., tool needs to think you are putting these computed outputs someplace. You have to list all the block’s inputs here in the “sensitivity list”. (*) also works! Actually do logic in here. There are a bunch of subtle rules to ensure that synthesis won’t mess this up... We’ll see how… How? … A Few Definitions There are some restrictions on specification Input set of an “always” statement — the set of all variables that are used on the right-hand side of procedural assignments or in conditionals. i.e. anything “sourced”. Sensitivity list of an “always” statement — the set of all names that appear in the event (“@”) list. module mux (output reg f, input sel, b, c); always @ (sel or b or c) begin if (sel == 1) f = b; else f = c; end endmodule The elements in these lists are: input: sel, b, c sensitivity: sel, b, c No coincidence here: a combinational circuit is sensitive to its inputs More Definitions... … A control path of an “always” statement — a sequence of operations performed when executing the always statement Combinational output of an “always” statement — a variable (or variables) assigned to in every control path module mux (output reg f, input sel, b, c); always @ (sel or b or c) begin if (sel == 1) f = b; else f = c; end endmodule What are they here... Control paths: through “then” and “else” of if statement Combinational output: f The Basic Rules The rules for specifying combinational logic using procedural statements Every element of the input set must be in the sensitivity list The combinational output must be assigned in every control path module mux (output reg f, input sel, b, c); always @ (sel or b or c) begin if (sel == 1) f = b; else f = c; end endmodule So, we’re saying that if any input changes, then the output is reevaluated. — That’s the definition of combinational logic. Walking this narrow line allows you to specify and synthesize combinational logic What If You Mess Up? If you don’t follow the rules...? … you’re dead meat Verilog assumes you are trying to do something clever with the timing It’s legal, but it won’t be combinational The rules for what it does make sense — but not yet for us. module blah (output reg f, g; input a, b, c); This says: as long as a==1, then f follows b. (i.e. when b changes, so does f.) But, when a==0, f remembers the old value of b. always @ (a or b or c) begin if (a == 1) f = b; else g = c; end endmodule Combinational circuits don’t remember anything! What’s wrong? f doesn’t appear in every control path in the always block (neither does g). Typical Style Your Verilog for combinational stuff will look like this: module blah (<output names>, <input names>); output <output names>; input <input names>; reg <output names>; always @ (<names of all input vars>) begin < LHS = RHS assignments> < if ... else statements> < case statements > end endmodule Yes...it’s a pretty restricted subset of the language... A Difficulty Assigning in every control path If the function is complex, you don’t know if you assigned to the outputs in every control path. So, set all outputs to some known value (zero here) and write the code to set them to other values as needed. Synthesis tools will figure it out, but try to write clearly. always @(coke or cola) begin if (coke) blah1 = 1; else if (cola > 2’b01) blah2 = coke; else if ( … … end always @(coke or cola) begin blah1 = 0; blah2 = 0; if (coke) blah1 = 1; else if (cola > 2’b01) blah2 = coke; else if ( … … end Using a case statement Truth table method List each input combination Assign to output(s) in each case item. Concatenation {a, b, c} concatenates a, b, and c together, considering them as a single item Example a = 4’b0111 b = 6’b 1x0001 c = 2’bzx then {a, b, c} = 12’b01111x0001zx module fred (output reg f, input a, b, c); always @ (a or b or c) case ({a, b, c}) 3’b000: f = 1’b0; 3’b001: f = 1’b1; 3’b010: f = 1’b1; 3’b011: f = 1’b1; 3’b100: f = 1’b1; 3’b101: f = 1’b0; 3’b110: f = 1’b0; 3’b111: f = 1’b1; endcase endmodule Check the rules … How about a Case Statement Ex? module fred (output reg f, input a, b, c); always @ (a or b or c) case ({a, b, c}) 3’b000: f = 1’b0; 3’b001: f = 1’b1; 3’b010: f = 1’b1; 3’b011: f = 1’b1; 3’b100: f = 1’b1; 3’b101: f = 1’b0; 3’b110: f = 1’b0; 3’b111: f = 1’b1; endcase endmodule check the rules… Could put a function here too module fred (output reg f, input a, b, c); always @ (a or b or c) case ({a, b, c}) 3’b000: f = 1’b0; 3’b101: f = 1’b0; 3’b110: f = 1’b0; default: f = 1’b1; endcase endmodule Important: every control path is specified x and z not considered in case enumeration! Don’t Cares in Synthesis Rules You can’t say “if (a == 1’bx)…” — this has meaning in simulation, but not in synthesis. However, an unknown x on the right-hand side will be interpreted as a don’t care. a b f ~c ab 00 01 11 10 c 0 x 1 0 1 1 1 1 1 x module caseExample( (output reg f, input a, b, c); always @ (a or b or c) case ({a, b, c}) 3’b001: f = 1’b1; 3’b010: f = 1’b1; 3’b011: f = 1’b1; 3’b100: f = 1’b1; 3’b111: f = 1’b1; 3’b110: f = 1’b0; default: f = 1’bx; endcase endmodule The inverse function was implemented; x’s taken as ones. Alternatively… module fred1 (output reg f, input a, b, c); always @ (a or b or c) f = ~(a & b & ~c); endmodule ab These aren’t quite equivalent to the previous slide…why? module fred2 (output reg f, input a, b, c); module fred3 (output reg f, input a, b, c); always @ (a or b or c) f = ~a | c | ~b; endmodule always @ (a or b or c) begin if (c ==0) f = a~&b; else f = 1’b1; end endmodule 00 01 11 10 c 0 x 1 0 1 1 1 1 1 x Two input bits, Three output bits reg [1:0] newJ; reg out; input i, j; always @(i or j) case (j) 2’b00: begin newJ = (i == 0) ? 2’b00 : 2’b01; out = 0; end 2’b01 : begin newJ = (i == 0) ? 2’b10 : 2’b01; out = 1; end 2’b10 : begin newJ = 2’b00; out = 0; end default: begin newJ = 2’b00; out = 1'bx; end endcase Works like the C conditional operator. (expr) ? a : b; If the expr is true, then the resulting value is a, else it’s b. What about time delay Could we have described the module as shown here? Note the delays. There is a different delay from the b input than from the c input. Yes, you could write this But, Synthesis tools will ignore the time delays. Generally, they try to minimize the propagation from any combinational input to any combinational output in the system. module mux (output reg f, input sel, b, c); always @ (sel or b or c) begin if (sel == 1) #5 f = b; else #88 f = c; end endmodule Model Organization Here’s an always block for a combinational function. What Boolean functions can it model? Can I have more than one of these always blocks in a module? Can two separate always blocks Yes calculate function f? always @(b1 or b2 or b3) begin yadda yadda end Only those with inputs b1, b2, and b3 Nope! No module xyzzy (ports); … always @(b1 or b2 or b3) begin f = yadda; end always @(r1 or r2 or r3) begin f = yadda yadda; end module xyzzy (ports); … always @(b1 or b2 or b3) begin q = b1 … b2 … b3 r = b2 … b3 end always @(r1 or b2 or b3) begin s = yadda yadda yadda end FSM — Review In the abstract, an FSM can be defined by: A set of states or actions that the system will perform The inputs to the FSM that will cause a sequence to occur The outputs that the states (and possibly, the inputs) produce There are also two special inputs A clock event, sometimes called the sequencing event, that causes the FSM to go from one state to the next A reset event, that causes the FSM to go to a known state outputs inputs FSM clock reset FSM Review The traditional FSM diagram We looked at State Transition Diagrams (and Tables) last time … and ended up with a diagram that looked like this We’ll talk about how to design this inputs clock reset outputs Combinational Logic Current State Register next state We’ll talk about what these are D Flip Flops What are they? A one-bit storage devices, sometimes called a sequential element It “flip flops” between two states. What are the ports? D is the data input — the value to be remembered C is the clock input. The triangle indicates that it’s edge sensitive. Q is the output (and Q’ is its complement) Reset makes Q equal to zero anytime reset is asserted (some “preset” to one also) Edge triggered FF symbol data D clock Q Q and Q’ C reset asserted low Edge Triggered? How’s it work Clock transition determines when input is acquired Edge triggered FF symbol data clock The point in time where clock changes value is called the “edge”. This is a “positive edge” since clock is rising 0->1. The input value D at the time of the edge is remembered and driven on the output. D clock Q Q and Q’ C reset asserted low Triggering: How Inputs —> Outputs edge triggered FF symbol data D clock Q Q and Q’ C Timing Diagram: reset asserted low Current Next state, after state, now clock event D 0 0 1 1 Q 0 1 0 1 Q+ 0 0 1 1 D CLK Q time Some Terminology The clock is a special input It unifies all the changes in an FSM Two types of changes in logic signals Synchronous — The change is synchronized to an event. The event is typically a repetitive one — like the clock In normal operation, the output Q is synchronized to the clock event In normal operation, Q only changes when the event occurs, not when the input D changes Asynchronous — The change is not synchronized to an event The reset is an “asynchronous reset” The D flip flop we’ve shown is … synchronous with an asynchronous reset, meaning the normal operation of the data input and output are synchronized to the clock event, but the reset will cause the output to be set to zero immediately (without waiting for the clock event). Verilog for the D Flip FLop Current Next state, after state (now) clock event Q+ D Q 0 0 0 0 0 1 1 1 0 1 1 1 module DFF (output reg q, input d, clk, reset); always @(posedge clk or negedge reset) if (~reset) q <= 0; else q <= d; endmodule Note that it doesn’t matter what the current state (Q) is. The new state after the clock event will be the value on the D input. The change in q is synchronized to the clk input. The reset is an asynchronous reset (q doesn’t wait for the clk to change). Clock Events on Other Planets Trigger alternatives For flip flops, the clock event can either be a positive or negative edge Both have same next-state table D Q C clk Current Next state, after state, now clock event Positive edge triggered Flip Flop D clk Q C Negative edge triggered Flip Flop (bubble indicates negative edge) D 0 0 1 1 Q 0 1 0 1 Q+ 0 0 1 1 Where Are Flip Flops in FSMs? Current State Register Each bit of the register is implemented with a flip flop The clocks of all flip flops are tied together into one clock line The reset (or preset) of all flip flops are tied together into one reset line We’ll talk about how to design this inputs clock reset outputs Combinational Logic Current State Register next state Moore Machine — 111 Detector Q2 D1 Q1 Z reset X Q1 D2 Q2 Q2’ clock reset reset Note how the reset is connected Reset will make both of the FFs zero, thus putting them into state A. Most FFs have both reset and “preset’ inputs (preset sets the FF to one). The reset connections (to FF reset and preset) are determined by the state assignment of the reset state. Verilog Organization for FSM Q2 D1 Q1 reset X Q1 D2 Q2 Q2’ clock reset reset Two always blocks One for the combinational logic — next state and output logic One for the state register Z Verilog Behavioral Specification module FSM (x, z, clk, reset); input clk, reset, x; output z; reg [1:2] q, d; reg z; always @(posedge clk or negedge reset) if (~reset) q <= 0; else q <= d; always @(x or q) begin d[1] = q[1] & x | q[2] & x; d[2] = q[1] & x | ~q[2] & x; z = q[1] & q[2]; end endmodule Things to note reg [1:2] — matches numbering in state assignment (Q2 is least significant bit in counting) <= vs. = The sequential part (the D flip flop) The combinational logic part next state output Verilog Overview Verilog is a concurrent language Aimed at modeling hardware — optimized for it! Typical of hardware description languages (HDLs), it: controls time provides for the specification of concurrent activities stands on its head to make the activities look like they happened at the same time — Why? allows for intricate timing specifications — Why? A concurrent language allows for: Multiple concurrent “elements” An event in one element to cause activity in another. (An event is an output or state change at a given time) based on interconnection of the element’s ports Logical concurrency — software True physical concurrency — e.g., “<=” in Verilog Discrete Time Simulation Discrete Time Simulation Models evaluated and state updated only at time intervals — n Even if there is no change on an input Even if there is no state to be changed Need to execute at finest time granularity Might think of this as cycle accurate — things only happen @(posedge clock) You could do logic circuits this way, but either: Lots of gate detail lost — as with cycle accurate above (no gates!) Lots of simulation where nothing happens — every gate is executed whether an input changes or not. Discrete Event Simulation… picks up simulation efficiency due to its selective evaluation only execute models when inputs change Discrete Event (DE) Simulation Discrete Event Simulation Events — changes in state at discrete times. These cause other events to occur. Only execute something when an event has occurred at its input Events are maintained in time order Time advances in discrete steps when all events for a given time have been processed Quick example Gate A changes its output. Only then will B and C execute Observations B A C The elements in the diagram don’t need to be logic gates DE simulation works because there is a sparseness to gate execution — maybe only 12% of gates change at any one time. The overhead of the event list pays off then. Observations Hmm… Implicit model execution of fanout elements B Implicit? Concurrency — is it guaranteed? How? Time — a fundamental thingie Can’t you represent this all in C? After all, the simulator is written in it! Or assembly language? What’s the issue? Or how about Java? Ya-know, aren’t objects like things you instantiate just like in Verilog? Can’t A call the port-2 update method on object B to make a change? A C A Gate Level Model A Verilog description of an SR latch A module is defined name of the module Draw the circuit module nandLatch (output q, qBar, input set, reset); nand #2 The module has ports that are typed type and delay of primitive gates g1 (q, qBar, set), g2 (qBar, q, reset); endmodule primitive gates with names and interconnections A Gate Level Model Things to note: It doesn’t appear “executable” — no for loops, if-then-else, etc. it’s not in a programming language sense, rather it describes the interconnection of elements A new module made up of other modules (gates) has been defined software engineering aspect — we can hide detail module nandLatch (output q, qBar, input set, rese)t; nand #2 g1 (q, qBar, set), g2 (qBar, q, reset); endmodule Kinds of delays Transport delay input to output delay (sometimes “propagation”) Zero delay models (all transport delays = 0) functional testing there’s no delay, not cool for circuits with feedback! a b c — transport delay Unit delay models (all transport delays = 1) all gates have delay 1. OK for feedback Edge sensitive — delay is value sensitive nand #(3, 4, 5) (c, a, b); rising delay delay to z falling delay Some more gate level examples instance names and delays optional An adder module adder (output carryOut, sum, input aInput, bInput, carryIn); sum xor (sum, aInput, bInput, carryIn); or (carryOut, ab, bc, ac); and (ab, aInput, bInput), ab carryOut bc (bc, bInput, carryIn), ac (ac, aInput, carryIn); endmodule aInput list of gate instances of same function implicit wire declarations carryIn bInput Adder with delays An adder with delays module adder (output carryOut, sum, input aInput, bInput, carryIn); xor #(3, 5) (sum, aInput, bInput, carryIn); or #2 (carryOut, ab, bc, ac); and #(3, 2) (ab, aInput, bInput), (bc, bInput, carryIn), (ac, aInput, carryIn); endmodule each AND gate instance has the same delay Adder, continuous assign Using “continuous assignment” Continuous assignment allows you to specify combinational logic in equation form Anytime an input (value on the right-hand side) changes, the simulator re-evaluates the output No gate structure is implied — logic synthesis can design it. the description is more abstract A behavioral function may be called — details later module adder (output carryOut, sum, input aInput, bInput, carryIn); assign sum = aInput ^ bInput ^ carryIn, carryOut = (aInput & bInput) | (bInput & carryIn) | (aInput & carryIn); endmodule I’m sick of this adder Continuous assignment assigns continuously delays can be specified (same format as for gates) on whole equation no instances names — nothing is being instantiated. given the same delays in this and the gate-level model of an adder, there is no functional difference between the models FYI, the gate-level model gives names to gate instances, allowing back annotation of times. module adder (output carryOut, sum, input aInput, bInput, carryIn); assign #(3, 5) sum = aInput ^ bInput ^ carryIn; assign #(4, 8) carryOut = (aInput & bInput) | (bInput & carryIn) | (aInput & carryIn); endmodule Structure vs. Behavior Structure — Look at it from the module (adder) ports Strong physical connotations The internal structure of a system includes its state and state transition mechanism as well as the state to output mapping Behavior — again from the module ports Outer manifestation of a system The external behavior of a system is the relationship it imposes between its input time histories and output time histories module adder (output carryOut, sum, input aInput, bInput, carryIn); xor or and (sum, aInput, bInput, carryIn); (carryOut, ab, bc, ac); (ab, aInput, bInput), (bc, bInput, carryIn), (ac, aInput, carryIn); endmodule Structural model module adder (output input assign carryOut, sum, aInput, bInput, carryIn); sum = aInput ^ bInput ^ carryIn, carryOut = (aInput & bInput) | (bInput & carryIn) | (aInput & carryIn); endmodule Behavioral model Where is the state in these models? Verilog Structure vs. Behavior Structure gate level — built-in models for AND, OR, … modules and instantiations wires Behavior C-like programs or Boolean algebra (but with a few extra operators) assign statements always blocks — procedural statements (next time) Hmm… If a module has an assign statement in it, is it behavior or structure? On the outside, it appears as structure — it’s wired in, takes up space (it’s physical) — maybe it is an ALU slice On the inside, it appears as behavior — we only know the translation of inputs to outputs, but without physical connotations Mixing Levels Generally there is a mix of levels in a model e.g. part of the system is at the gate level and another part is at the behavioral level. Why? Early in design process you might not have fully-detailed models — you don’t actually know all the gate implementations of the multipliers, adders, register files You might want to think of the design at a conceptual level before doing all the work to obtain the gate implementations There might be a family of implementations planned Finer grain of distinction Levels — switch, gate, functional block (e.g. ALUs), registertransfer, behavioral for now, we’ll deal with gate and behavioral models An execution model for gates/assigns Execution model “Execution” (sometimes “timing”) model — how time is advanced, what triggers new processing and the generation of new state in the model State is held on wires, gates and continuous assigns advance state Definition — when an input changes, the simulator will evaluate the gate or continuous assign, calculating a new output if the output value is different, it is propagated to elements on the fanout module nandLatch (output q, qBar, input set, reset); nand #2 g1 (q, qBar, set), g2 (qBar, q, reset); endmodule Gate level timing model For gates and continuous assigns… What’s an input? What’s an output? What’s state? Gate inputs and RHS of assign equation Gate outputs and LHS of assign equation Wires Outputs on this “side” of the language are all … Wires … no registers are latched/loaded, no need to know about a clock event Gate level timing model Contrast At the gate level, there’s nothing special about two cross-coupled gates R Q S Q’ A register is an abstraction above this “side” of the language The left-hand sides on the behavioral “side” of the language are all registers Approach to Simulating a System Two pieces of a simulation The model — an executable specification including timing, interconnect, and input vectors Written in a language like Verilog or VHDL What’s a VHDL? The simulation scheduler — keeps track of when events occur, communicates events to appropriate parts of the model, executes the model of those parts, and as a result, possibly schedules more events for a future time. it maintains “simulated time” (sometimes “virtual time”) and the event list. Parts of the scheduler function define the language How’s the simulator work? A gate level model doesn’t look like a program No if’s or loops — what get’s executed? Here’s how gate-level Verilog is executed — You specify a bunch of primitive gates that are interconnected When an input of a gate changes, the simulator will evaluate the gate primitive and calculate a new output If the output value is different from the current, it is scheduled to propagate at some time in the future (or possibly now). After the specified time delay (possibly zero), the new value is propagated along wires to other gate-primitive inputs Simulator keeps track of time … and what has been scheduled to happen at any time Inputs and Outputs? An input to a gate primitive, the output of a gate primitive Are these two modules the same? module muxA (output f, input a, b, sel); module muxB (output f, input a, b, sel); or #5 g3 (f, f1, f2); not g4 (nsel, sel); and #5 g1 (f1, a, nsel), g2 (f2, b, sel); endmodule and #5 g1 (f1, a, nsel), g2 (f2, b, sel); or #5 g3 (f, f1, f2); not g4 (nsel, sel); endmodule a a f f b b sel Alternate drawings of a mux sel Inside the Simulator A time-ordered list of events is maintained Event — a value-change scheduled to occur at a given time All events for a given time are kept together The scheduler removes events for a given time… …propagates values, and executes gate models, creates new events… time-ordered event list remove current events ••• ti tj tk all the events for time tj tn updates Gate Outputs schedules new event Scheduler looks at Network Connections (fanouts) executes Gate Models Event-Driven Simulation while (something in time-ordered event list) { advance simulation time to soonest event’s time retrieve all events e for this time e For each event e in arbitrary order { update the value specified follow fanout evaluate the model(s) schedule resulting events } } evaluate these One traversal of the while loop is a simulation cycle. In 1 cycle, we remove all events for a time & execute them. New events may be scheduled for the current time — they are put in the event list and retrieved in the next sim. cycle. New event Event-Driven Simulation the event list initial A=1 at values as 25 shown 1 1 A=0 g1 #2 0 Eval g1 B=0 at 27 initial A=1 at values as 25 shown (at 27) g2 #3 B=1 1 g1 #2 C=0 g2 #3 B=0 A=1 1 1 g1 #2 A=1 0 D=1 g3 #5 0 (at 30) D=1 g3 #5 1 Eval g2, g3 C=1 at B=0 at A=1 at initial 25 30 values as 27 shown C=0 C=1 g2 #3 B=0 g3 #5 final D=1 How does it keep track of time? … Explicitly Events are stored in an event list (actually a 2-D list) ordered by time Events execute at a time and possibly schedule their output to change at a later time (a new event) When no more events for the current time, move to the next Events within a time are executed in arbitrary order time a event event time a + 2 event event event Let’s say A changes to 0 here. B and C have delay 2. 1 B A time a+75 time a+75492 event event Events to event update B and C are added. 1 C Two types of Events Update events — Action: update state and propagate new values along a fanout. Possibly produces new events Evaluation events — Action: evaluate, or execute, a model. Possibly produces new events Plan Will deal with update events now Evaluation events come in with behavioral models Event-Driven Simulation 1 while something in time-ordered event list { advance simulation time to top event’s time B=0 #2 A= 1 0 C=0 retrieve all events for this time update 1 #2 For each event in arbitrary order { If it’s an update event { 1 update the value specified B= 01 #2 A= 0 C= 01 follow fanout, evaluate gates there If an output changes 1 #2 schedule update event for it } update else // it’s an evaluation event evaluate the model } } time A=0 time + 2 B=1 C=1 What about zero delay events? while something in time-ordered event list { advance simulation time to top event’s time retrieve all events for this time But it’s not retrieved and executed until the next sim cycle For each event in arbitrary order { If it’s an update event { update the value specified follow fanout, evaluate gates there A gate with #0 delay gets scheduled for the current time If an output changes schedule update event for it } Ain #0 Aout else // it’s an evaluation event evaluate the model } Ain Aout == 01 time } The simulator can spend several iterations at the same simulation time Verilog Gate level timing model What if an update event is already scheduled for an primitive gate output? if the value being scheduled is different, the currently scheduled value is removed from the event list; the new event is not scheduled Called inertial delay — oddly named, how wide must an input spike be to be seen? Deviation from pure discrete a event simulation. a=1 c b nand #5 (c, a, b); b=1 update scheduled c propagation delay = 5 update removed, final value a b alternate c Instantiation — Hierarchy module above (out, …); output [2:0] out; wire [2:0] h, I, j; module r(o1,i1, i2, i3); input i1, i2, i3; output o1; assign o1 = i1 | i2 | i3; endmodule r a(out[0], h[0], I[0], j[0]), b(out[1], h[1], I[1], j[1]), c(out[2], h[2], I[2], j[2]); endmodule out[2:0] a r r r above out[2] b c o1 Not all connections shown Hierarchical name o1 is really … above_inst.c.o1 Used for debugging… why just debugging? Hierarchy Why? Hides detail Supports alternate implementations Encapsulates — side effects understood Observations Hardware resources allocated (instantiated) to perform a function exclusively No other function will use it Thus, physical concurrency and structure are established module r (output o1, input i1, i2, i3); assign o1 = i1 | i2 | i3; endmodule module r (output o1, input i1, i2, i3); or #(2, 5) (first, i1, i2), (o1, first, i3); endmodule Summary on gate evaluation Simulation languages — concurrent Maintain explicit notion of time Describe models with physically concurrent activity Interconnection of models allows for data-driven activity Timing model timing-execution model how time is advanced and new state created Any gate input or assign righthand-side change causes the model to be evaluated during the time step this is not the case for behavioral models Fanout list is static — design never changes What if you don’t like Verilog’s gates? e.g., inertial delays? use behavioral models (or user defined primitives…?) Procedural models: what’s needed? Obvious things like operator set that matches hardware functionality Bit hacking, etc. a = { b[3], b[1], c[4] }; Concurrent operators Similar to what you’d find in other “threaded” languages … plus hardware functionlity — such as: Edge triggering Concurrent/buffered state update Control of time … …minus a few — such as: Support for critical sections — P,V Procedural Models This is the “other side” of the language Always and initial statements are concurrent They start when the simulation starts, in arbitrary order Assignments are made to registers Everything on left hand side is a register Statements execute sequentially Atomicity — only one element (gate, always, initial) executing at a time. No preemption — continues executing until done. Stuff between concurrent statements executes in zero time Because statements execute in zero time and are atomic, it looks like lots of parallel stuff is happening always begin @ (posedge clock) h = f + k; g = f * g; @ (posedge clock) f = g; q = f * s; … Why is this important? At first look, it is a lot like C Most of the operators are the same as C ^ is XOR, etc. makes it easy to read But there are major differences (quick list, we’ll get to these) Concurrent statements like #delay, @event, wait(level) four-valued logic (1, 0, x, z) and the operators to go with them arbitrary bit width operations there are a couple of procedural assignments (=, <=) with subtle differences a different timing model — in fact, C doesn’t have one It has a sequencing model — sequence being a more abstract view of time. hmm, do we even know if the program sequencing holds? Review from before Behavior vs. Structure These two models are functionally interchangable — either could have been instantiated into a register ports in same order same delay from clock to q one is abstract, clear one is structurally specific there are subtle differences module d_type_FF (output reg q, input clock, data); always @(negedge clock) q = #10 data; endmodule Behavioral module d_type_FF (output q, input clock, data); nor #10 a (q, qBar, r); nor b (qBar, q, s), c (s, r, clock, s1), d (s1, s, data), e (r, r1, clock), f (r1, s1, r); endmodule Structural Procedural Timing Model How does the procedural model advance time? # — delaying a specific amount of time @ — delaying until an event occurs “posedge”, “negedge”, or any change this is edge-sensitive behavior When the statement is encountered, the value v is sampled. When v changes in the specified way, execution continues. always begin #5 q = w; @ (negedge v) q = y; wait (c == 0) q = 3; end wait — possibly delaying until an event occurs this is level sensitive behavior While one model is waiting for one of the above reasons, other models execute — values change, time marches on Everything executes in zero time — time advances when you’re not executing! An example of wait Semantics wait (expression) statement; — e.g. wait (a == 35) q = q + 4; if the expression is FALSE, the process is stopped when a becomes 35, it resumes with q = q + 4 if the expression is TRUE, the process is not stopped it continues executing module handshake (ready, dataOut, …); (input ready, output reg [7:0] dataOut); reg [7:0] someValueWeCalculated; always begin … wait (ready); … dataOut = someValueWeCalculated; wait (~ready) … end… ready No. Not if ready is already true when the first wait is executed. You’re not guaranteed to get the value at the edge Do you always get the value at the edge when ready goes from 0 to 1? Isn’t this edge behavior? Wait vs. While Are these equivalent? No: The left example is correct, the right one isn’t — it won’t work Wait is used to wait for an expression to become TRUE the expression eventually becomes TRUE because a variable in the expression is changed by another process While is used in the normal programming sense in the case shown, if the expression is TRUE, the simulator will continuously execute the loop. Another process will never have the chance to change “in”. Infinite loop! while can’t be used to wait for a change on an input to the process. Need other variable in loop, or # or @ in loop. module yes (input in); … wait (in == 1); … endmodule module no (input in); … while (in != 1); … endmodule Blocking assignments and # We’ve seen #delay Delay for specified time … and blocking assignments — they use = Options for specifying delay Wait #10, then do the statement #10 a = b + c; a = #10 b + c; Note the action of the second one: Calculate b+c, wait 10, then do assignment an intra-assignment time delay The event list is used for temporary storage! The differences: • #10 a = b + c; Values b and c are from time (now + 10) • a = #10 b + c; Values b and c are from time (now) Blocking — what’s it mean? Blocking — the always or initial block stops (blocks) for some reason #, @, wait(FALSE) always begin q = blahblah; r = q - someInput; It blocks (stops) here, other things (always, gates, assigns) execute. Finally at t+10, this continues executing a = #10 q + r; t = a - someOtherInput; … end Intra assignment delay – delay within an assignment. Events — @something Action when first encountered, sample the expression wait for expression to change in the indicated fashion This always blocks — you never execute straight through — guaranteed edge sensitivity Examples always @(posedge ck) q <= d; always @(hello) a = b; always a = @(hello) b; always @(coke or cola) a = b; always begin yadda = yadda; @(posedge hello or negedge goodbye) a = b; … end Sensitivity Lists In the gate level timing model… model execution was sensitive to any change on any of the inputs at any time. sensitivity list — a list of inputs that a model is sensitive to a change on any of them will cause execution of the model In the gate level timing model, the lists don’t change. Ditto with continuous assign In procedural models … the sensitivity list changes as as function of time and execution module d_type_FF (output q, input clock, data); nor #10 a (q, qBar, r); nor b (qBar, q, s), c (s, r, clock, s1), d (s1, s, data), e (r, r1, clock), f (r1, s1, r); endmodule Structural Procedural Timing Model What is the behavioral model sensitive to? The behavioral statements execute in sequence Therefore, a behavioral model is sensitive to its context i.e. it is only sensitive to what it is currently waiting for time, edge, level — (#, @, wait) The following model is not sensitive to a change on y or w. always begin @ (negedge clock1) q = y; @ (negedge clock2) q = w; @ (posedge clock1) /*nothing*/ ; @ (posedge clock2) q = 3; end Here, it is only sensitive to clock1 Here, it is only sensitive to clock2. A posedge on clock1 will have no effect when waiting here. Fanout Lists Outputs of things are connected to inputs of other things No surprise The simulator maintains a fanout list of inputs driven by each “output” Why maintain a fanout list? When the output changes, it’s easy to figure out what other models need (to be) evaluated Because of procedural models … …Sensitivity lists change Fanout lists change Sensitivity lists <—> Fanout lists What’s an “output” in a behavioral model? List Changes Change in sensitivity lists in procedural models cause fanout lists to change clock1 fanout is A, B, D; clock2 fanout is C. clock1 A B clock2 C always begin:D @ (negedge clock1) q = y; @ (negedge clock2) q = w; … end clock1 fanout is A, B; clock2 fanout is C, D. Scheduling #, @, and Wait How are #, @, and wait tied into the event list? # delay schedule the resumption of the process — put it in the event queue delay units into the future. Essentially an evaluation event scheduled in the future @ change when suspended for an @v, the behavioral model is put on the fanout list of the variable v. i.e., the behavioral model is now sensitive to v. When an update event for v occurs, (e.g. posedge), then the behavioral model resumes at the current time. Wait (exp) if exp is TRUE, don’t stop if exp is FALSE, then the behavioral model is put on the fanout list(s) of the variable(s) in exp. (it’s now sensitive to the variable(s)) When there is an update event for any of the variables in exp , exp is evaluated. If exp is TRUE, resume executing in the current time , else go back to sleep Procedural Model Sensitivity? Quick example Gate A changes its output What models get executed? B Yes A C always @(A) begin R = ~A; end always @(posedge clock) Q <= A; No Maybe always begin @(A) R = ~A; @(D) R = ~B; end Order of Execution B Assume A changes. In what order do these models execute? Arbitrary, don’t count on any specific order The simulator will try to make them look like they all occur at the same time — how? By controlling virtual time. A C always @(A) begin R = ~A; end always @(posedge clock) Q <= A; Arbitrary Order? Oops! Sometimes you need to exert some control Consider the interconnections of this DFF At the positive edge of c, what models are ready to execute? Does it matter which one is done first? shiftin Oops — The order of execution can matter! clock module dff(q, d, c); … always @(posedge c) q = d; endmodule module sreg (…); … dff e (q0, shiftin, clock), f (q1, q0, clock), g (shiftout, q1, clock); endmodule D Q D Q D Q shiftout Non-blocking Concurrent Asg. module fsm (output reg Q1, Q0, input clock, in); always @(posedge clock) begin Q1 <= in & Q0; Q0 <= in | Q1; end endmodule Q0 Q Q1 D Q Q0 in Q1 clock Values after the clock edge (t+) — calculated in response to the clock edge, using values at the clock edge D Values at the clock edge. (At t -) Concurrent Assignment — primary use of <= The assignment is “guarded” by an edge All assignments guarded by the edge happen concurrently