02/10/06eecs150 lab lecture #41 debugging eecs150 spring 2006 – lab lecture #4 philip godoy greg...
Post on 22-Dec-2015
216 views
TRANSCRIPT
02/10/06 EECS150 Lab Lecture #4 1
Debugging
EECS150 Spring 2006 – Lab Lecture #4
Philip GodoyGreg Gibeling
02/10/06 EECS150 Lab Lecture #4 2
Today (1)
Lab #2 Solution Simulation vs Hardware Debugging
Goals Tips Algorithm
Administrative Info
02/10/06 EECS150 Lab Lecture #4 3
Today (2)
Lab #4 Bottom Up Testing (Peak Detector) Designing Test Hardware (Broken
Adder) Exhaustive FSM Testing (Broken FSM)
02/10/06 EECS150 Lab Lecture #4 4
module Accumulator(In, Out, Enable, Clock, Reset);input [7:0] In;output [7:0] Out;
input Enable;
input Clock, Reset;
reg [7:0] Out;
always @ (posedge Clock) beginif (Reset) Out <= 8'h00; else if (Enable) Out <= Out + In;
endendmodule
Lab #2 Solution (1)
02/10/06 EECS150 Lab Lecture #4 5
Lab #2 Solution (2)
and1
not1
or1
In0
In1
EqualIn
GreaterIn
GreaterOut
and2
xor1 not3
not2
EqualOut
02/10/06 EECS150 Lab Lecture #4 6
Lab #2 Solution (3) Accumulator
Simple, easy to build What is the actual circuit?
Get used to answering this in your head RTL View from Lab #1 (Section 4.3
Synthesis) Peak Detector
Hard to build Minute control of hardware Tools couldn’t optimize though!!
02/10/06 EECS150 Lab Lecture #4 7
Simulation vs. Hardware (1) Debugging in Simulation
Slow Running Time Fast Debugging
Waveforms Text messages
Full Visibility Can examine any signal
Easy to Fix A few minutes to compile and resimulate
02/10/06 EECS150 Lab Lecture #4 8
Simulation vs. Hardware (2)
Debugging in Hardware Fast Running Time
Full speed in fact Slow Debugging
Synthesis can take hours Little or No Visibility
Very hard to probe signals Maybe Impossible to Fix (ASICs)
02/10/06 EECS150 Lab Lecture #4 9
Simulation vs. Hardware (3) Simulation
Functional Testing & Verification Test everything at least minimally Fully Verify what you can
This will save you many sleepless nights
Hardware Debugging
Treat this as a last resort It is painful
02/10/06 EECS150 Lab Lecture #4 10
Debugging (1)
Debugging Algorithm Hypothesis: What’s broken? Control: Give it controlled test inputs Expected Output: What SHOULD it
do? Observe: Did it work right? If it broke: THAT’S GREAT!
If we can’t break anything like this then the project must be working…
02/10/06 EECS150 Lab Lecture #4 11
Debugging (2)
First, check for Verilog syntax errors Run Synthesis on the module View Synthesis report to see errors
Don’t debug randomly Just changing things at random often
makes things look fixed It won’t really help Debug systematically Your first design may be the best
02/10/06 EECS150 Lab Lecture #4 12
Debugging (3)
High Level Debugging Localize the problem
N64? SDRAM? Video? Test Patterns
Lets you easily isolate the broken component
If you know exactly what’s going in you can check what’s coming out
02/10/06 EECS150 Lab Lecture #4 13
Debugging (4)
Simulate the broken component(s) Writing test benches takes less time
than sitting around wondering why its broken
Everyone hates writing testbenches (Even the TA’s) You will hate hardware debugging more Get used to it
02/10/06 EECS150 Lab Lecture #4 14
Debugging (5) Your best debugging tool is logic
If 3 out of 4 components work, what’s broken?
Question all your assumptions! Just because you think its true doesn’t
mean it is 90% of debugging time is wasted
debugging the wrong problem otherwise
Given solutions and modules may not work the way you expect!
02/10/06 EECS150 Lab Lecture #4 15
Debugging (6)
Before you change anything Understand exactly what the problem is Find an efficient solution Evaluate alternative solutions
After the change Fixes may make things worse
sometimes May uncover a second bug May be an incorrect fix
Repeat the debugging process
02/10/06 EECS150 Lab Lecture #4 16
Debugging (7)
Ask around Someone else may have had the same
bug They’ll probably at least know about
where the problem is Different bugs may produce the same
results TAs
The TAs know common problems We’re here to help, not solve it for you
02/10/06 EECS150 Lab Lecture #4 17
Administrative Info Midterm I
Thursday 02/16, 2-3:30pm, Room 125 Cory Review Session TBA, 125 Cory
Partners You MUST have one for this week
Try someone other than your best friend Restrictions
You can change partners until the project starts You must be in the same lab
Project in 2 weeks
02/10/06 EECS150 Lab Lecture #4 18
Part1: Bottom Up Testing (1)
What if EqualOut = 1’b0 and GreaterOut = 1’b0?
Lab4Comp1
EqualIni+1
GreaterIni+1
Ai
Bi
EqualOuti
GreaterOuti
A=B
A<B
02/10/06 EECS150 Lab Lecture #4 19
Part1: Bottom Up Testing (2) Exhaustive Testing
Ideal Testing Method Circuit is 100% tested!
Requires us to test a LOT! Can we do it here? (24 possible inputs)
Method Make a truth table Have the testbench generate all inputs Make sure outputs match truth table
02/10/06 EECS150 Lab Lecture #4 20
Part1: Bottom Up Testing (3)
Equ
alO
ut[3
]G
reat
erO
ut[3
]
Equ
alO
ut[2
]G
reat
erO
ut[2
]
Equ
alO
ut[1
]G
reat
erO
ut[1
]A[0] B[0]
GreaterEqual
Lab4Comp4
Lab4Comp1Lab4Comp1Lab4Comp1Lab4Comp1
A[1] B[1]A[2] B[2]A[3] B[3]
02/10/06 EECS150 Lab Lecture #4 21
Part1: Bottom Up Testing (4)
Exhaustive Testing? 28 = 256 Possible Inputs
Method Use a for loop to generate all inputs
Loops allowed only in testbenches They will not synthesize
Compare against a “>=“ Print a message if they differ
02/10/06 EECS150 Lab Lecture #4 22
Part1: Bottom Up Testing (5)
Register
Lab4PeakDetector
In
Clock
Out
Reset
≥4
4
4
4
02/10/06 EECS150 Lab Lecture #4 23
Part1: Bottom Up Testing (6)
Exhaustive Testing? 24 = 16 Possible Inputs 24 = 16 Possible States 16*16 = 256 combinations We could do it in this case
Can’t exhaustively test FSMs Too many state/input combinations Must rely on directed testing
02/10/06 EECS150 Lab Lecture #4 24
initial begin
end
Part1: Bottom Up Testing (7)
integer i;
reg [3:0] TestValues[1:16];
$readmemh("TestValues.txt", TestValues);
for(i = 1; i <= 16; i = i + 1) begin#(`Cycle);In = TestValues[i]; $display("In = %d, Peak = %d", In, Peak);
end
02/10/06 EECS150 Lab Lecture #4 25
Part1: Bottom Up Testing (8)
Read Test Vectors from a File Designing Test Vectors
Make sure to cover most cases We want 95%+ coverage
Designing test vectors is a “black art” $ Processes
Not synthesizeable More information in IEEE Verilog
Reference
02/10/06 EECS150 Lab Lecture #4 26
Part2: Test Hardware (1)
Lab4 Part2 Adder
Test
Control
ABSum
A
B
Sum
Error
Running
Go
Reset
Lab4Part2Tester
02/10/06 EECS150 Lab Lecture #4 27
Part2: Test Hardware (2)
Test Procedure Hit Reset (SW1) Hit Go (SW2) Record an error
DD1-8 show {A, B} SW10[1] selects the sum on DD4-8
Hit Go Repeat until the tester stops
02/10/06 EECS150 Lab Lecture #4 28
Part2: Test Hardware (3)
The Broken Adder 16bit Adder
232 ≈4 Billion Test Vectors Can’t simulate this much 2:40 to test this at 27MHz
Fail Modes 0: No Errors 2: Will claim 1 + 1 = 3 1-3: Can have anywhere from 0 to 4 errors
02/10/06 EECS150 Lab Lecture #4 29
Part3: FSM Testing (1)
Exhaustive Testing Again! Check every arc Check every output
You don’t need to correct this one… We’re not giving you the source code
Boring (and Easy) You will have FSM bugs Get used to debugging them
02/10/06 EECS150 Lab Lecture #4 30
Part3: FSM Testing (2)
S0
S1 S4
S2 S5
1 1 0
S3[Output 1'b1]
S6
0 1
1
0
0
0 1
0 1
X