“Whenever trying to make complicated systems and understand them, it’s crucial to divide the things up into as many pieces as I can, each of which I understand separately. I would like to understand the way of adding things up independently of what it is I’m adding up.” - Gerald Sussman
Implementing a 4-bit Ripple-Carry Adder
Let’s work through an example to understand how functions that solve simple problems can be combined to solve a complex problem. The problem we want to solve is simulating the behavior of a 4-bit ripple-carry adder using an Elm program. “4-bit ripple-carry adder” is a bit of a mouthful, so let’s break down what that means word by word.
4-bit: “bit” is short for binary digit. So 4-bit means a four digit binary number, like
Ripple-Carry: Remember how in the early days of learning addition, you were taught that when you add
5you carry the
1left one column, resulting in
10? Then once you were comfortable with that concept, you moved on to math problems with two carries — like adding
5. First, you carry the one that is the result of
5left one column, but then when you add
9in that column you need to carry another
1to the another left column, resulting in
100. That’s basically what a “ripple-carry” is — each number that is carried out of one column is then carried in to the next column.
Adder: A digital circuit that adds two binary numbers.
First, let’s understand how the binary system works before diving deeper into the ripple-carry adder.
The decimal numeral system is the mathematical system most people are familiar with. In the decimal system, the integers
9 are combined in different ways to represent all numbers. It uses base-10 notation, which means each digit in a number is ten times larger than the digit to its right. The figure below breaks down the decimal number
4608 so you can see for yourself – the digits start from 100 on the right, then the next digit to the left is ten times larger at 101, then the digit to the left of that is ten times larger at 102, and so on.
The binary system, however, has only two integers —
1. It uses base-2 notation, which means each digit in a binary number is two times larger than the digit to its right. Here’s an example of the binary number
If you are wondering what decimal number the binary
1101 is equivalent to, it’s easy to find out. Just evaluate the bottom row of the above figure as one mathematical expression, and you will see that you get a final result of
13 is the decimal equivalent of
1101 in binary.
Adding binary numbers is much like adding everyday decimal numbers, except that it carries on a value of
2 instead of
10. For example, in decimal, if we add
3 we get
9. But if we add
4, we’ve run out of single integers to represent what comes next! That’s when the digit on the far right resets to zero, and we now have a one to carry into the next digit spot on the left.
In binary, if we add
1 we are in the same pickle as the decimal
4 example above: we have exhausted the integers available to us. When you run out of integers, it’s time to reset the digit on the right to zero and carry the one into the next digit spot on the left.
By this logic, it’s easy to keep adding binary numbers together. Let’s find out what the decimal number
4 is equivalent to in binary. Remember, every time we run out of numbers, that means it’s time to reset to zero and carry the one:
Now we can see,
4 in decimal system is
100 in binary.
Adding Numbers with a 4-bit Ripple-Carry Adder
A 4-bit ripple-carry adder can add binary numbers up to four digits. The figure below shows an example, adding the binary numbers
1011, with the digits to carry shown in purple.
A ripple-carry adder adds two digits at a time starting from the right. If there’s a carry, it propagates that to the next addition. Once it’s done adding all four digits, the output is presented to us like this:
A0, A1, A2, and A3 represent the individual digits of the first number. B0, B1, B2, and B3 represent the individual digits of the second number. S0, S1, S2, and S3 represent the individual digits of the sum. Cout represents the most significant digit of the sum. The most significant digit is the digit that has the greatest value. In the case of a binary number, that’s the digit all the way to the left. Here is a circuit diagram for a 4-bit ripple-carry adder.
Cin is the carry-in bit from the previous addition. Cout is the carry-out bit that will be passed to the next addition.
The circuit above is what we want to simulate in Elm. Where should we start? Well, the first thing we need to do is look for ways to break the problem at hand into smaller problems. If we look closer, we can see that a ripple-carry adder uses the same adder repeatedly but with different inputs and outputs. This adder acts as a building block for a ripple-carry adder and is called 1-bit full adder. Let’s figure out how we can implement a 1-bit full adder before we implement a ripple-carry adder.
1-bit Full Adder
A 1-bit full adder adds only two binary digits, but it’s capable of accepting a carry-in value as well. It takes three inputs and produces two outputs as shown in the figure below:
An and Bn represent the binary digits that need adding. Cn represents the carry-in digit, and Cn+1 represents the carry-out digit. Sn represents the least significant digit of the sum. The least significant digit is the digit that has the lowest value. In a binary number, that’s the digit all the way to the right. Here’s a truth table that shows all the possible inputs and outputs produced by a full adder:
Can we break a 1-bit full adder further into something even smaller? As it turns out, we can as shown in the figure below.
A full adder can be implemented using two half adders and an OR gate. We’ll find out what an OR gate is soon, but first let’s understand how a half adder works.
A half adder also adds two binary digits, but unlike a full adder it doesn’t accept a carry-in value.
Here’s a truth table that shows all the possible inputs and outputs for a half adder:
We can break its implementation into even smaller components: AND gate, OR gate, and inverter.
An AND gate takes two input signals and returns an output that is the logical and of the inputs. It’s equivalent to the
&& operator in Elm.
An OR gate takes two input signals and returns an output that is the logical or of the inputs. It’s equivalent to the
|| operator in Elm.
An inverter takes one input signal and inverts it. If the input is
0 it returns
1 as the output. If the input is
1, it returns
0 as the output. It’s equivalent to the
not function in Elm.
Now that we have broken down our original problem (4-bit ripple-carry) into its essential building blocks (AND gate, OR gate, and inverter), we can start building a solution by first implementing the building blocks.
Implementing an AND Gate
Let’s create a separate file that will contain all the code we will be writing as we implement different parts of a 4-bit ripple-carry adder. In the
elm-examples directory, create a new file called
Add the following code to the
We declared a new module called
RippleCarryAdder and imported another module called Bitwise, which includes functions that manipulate individual bits. As it so happens, the
Bitwise.and function does exactly what our AND gate is supposed to do which is to compute a logical and of two input signals.
You may wonder why we didn’t just use the
&& operator to implement the
andGate function. That’s because it only works for boolean values. Since we are trying to simulate a digital circuit, we want to work with
1s instead of
False although conceptually they are equivalent.
Generally, it doesn’t make sense to create a new function that just wraps an existing function, but the name
andGate is bit more revealing in the context of a digital circuit than just
and. Let’s load it up in the repl and verify that it works as expected. Run the
elm-repl command from the
beginning-elm directory to launch the repl, and expose the
andGate function when importing the
RippleCarryAdder module like this:
Playground module, we didn’t write a
main function in the
RippleCarryAdder module. That’s because we will be executing all code in that module from the repl. We need a
main function only if we want to run our Elm code in the browser.
Although there’s only one function in the
RippleCarryAdder module right now, we used
(..) to expose everything in it. It’ll make it easier for us to test things from that module as we add more functions to it. Since our function names will continue to be very descriptive, there’s a very little chance of them colliding with other names that might have already been imported.
Implementing an OR Gate
Bitwise module also includes a function for computing a logical or of two signals. Let’s use that to implement an OR gate. Add the following function definition to the bottom of
Let’s load it up in the repl and verify that it works as expected.
We don’t need to reimport the
RippleCarryAdder module whenever we make any changes to it. The repl automatically reloads a module when it detects a change. Isn’t that nice?
Implementing an Inverter
There is no suitable function in the
Bitwise module that does what an inverter is supposed to do. So let’s roll our own. Add the following function definition to the bottom of
inverter function flips the input signal using a
case expression. If the input signal is
0 it returns
1 and vice versa. If the input signal is anything other than
1, it returns
-1. We used negative
1 to indicate that the input signal doesn’t represent a valid binary number, although any number other than
1 would work.
If we don’t add the
_ -> -1 catch-all pattern, Elm will complain because without it we wouldn’t be accounting for all possible integer values the parameter
a can hold. This is not an elegant way to implement an inverter, but it gets the job done. Once we get introduced to types later in this chapter, we can refine this implementation.
Let’s verify that
inverter works as expected in the repl.
Now that we have implemented the basic building blocks, we can start implementing the more complex parts.
Implementing a Half Adder
Add the following function definition to the bottom of
halfAdder function mimics the half adder circuit we saw earlier. It uses all three logic gates we implemented before to compute a sum and a carry which are then returned as part of a record. The constants
e hold onto the intermediate results that get passed to an AND gate to compute the final sum.
In the Using Tuples section, we learned that when we want to return multiple values from a function, we should consider using a tuple. But records are also useful for returning multiple values. If we use a record in this instance, when we look at the output of
halfAdder, it’ll be easier for us to recognize which digit is a sum and which digit is a carry.
In the record above we added the comma in front of
sum = sumDigit instead of adding it at the end of the previous line. In most languages, it’s best practice to add commas at the end when we create a collection, but the Elm style guide recommends adding them in the front. That’s because it’s easier to spot a missing comma when they’re all in the front. For example, here’s a bigger record with commas in the end:
It’s hard to notice that the third line is missing a comma. But if we move them all to the front it’s much easier to spot a missing comma.
The Elm compiler will easily catch errors like this, but by putting the commas in the front we save ourselves an extra trip to the compiler land. Let’s verify that the
halfAdder function works as expected.
Implementing a 1-bit Full Adder
Let’s implement a full adder next by using two half adders and an OR gate. Add the following function definition to the bottom of
fullAdder function also mimics the full adder circuit we saw earlier. A full adder is capable of taking a carry-in value as one of the inputs. So we give it a third parameter that represents a carry-in value in addition to the two input signals:
b. Let’s verify that the logic in
fullAdder works as expected.
Implementing a 4-bit Ripple-Carry Adder
Finally, we’re ready to implement a 4-bit ripple-carry adder. Add the following function definition to the bottom of
rippleCarryAdderrepeatedly uses the
fullAdder function to compute the final sum and a carry-out. Let’s verify that it works as expected.
To summarize, we took a complex problem of building a 4-bit ripple-carry adder and broke it down to its essential building blocks (AND gate, OR gate, and inverter). We used those building blocks to implement a half adder. We then used a half adder and an OR gate to implement a full adder. Finally, we combined four full adders to implement a 4-bit ripple-carry adder.
We used small reliable functions to build bigger and equally reliable functions. This pattern of solving a complex problem by combining small functions is quite common in functional programming languages such as Elm. The main reason this pattern works is because all functions in Elm are pure.
Improving Inputs and Outputs
rippleCarryAdder function above accepts the input signals as tuples and returns the output as a record. This makes reading the inputs and outputs a bit difficult. It would be much more readable if they looked something like this:
0b as the prefix:
Therefore, we will have to do some extra work to be able to pass binary numbers as inputs to the
rippleCarryAdder function. The first thing we need to do is extract digits from a binary number so that we can pass them individually to different full adders. Modify the
rippleCarryAdder function as shown below so that it now accepts numbers instead of tuples:
We used a non-existent function called
extractDigits to extract digits from the input numbers. Let’s implement that function next. Add the following function definition to the bottom of
extractDigits accepts a number representing binary digits and takes it through several transformations until all digits have been extracted. Let’s understand how each transformation works.
You don’t have to try the code listed in each step below. It’s there to show how
extractDigits works. Some of it won’t even compile because we haven’t implemented all the necessary functions used inside
Step 1: Convert the input number into a string.
Step 2: Split the string into a list.
Step 3: Convert each string element inside the list back to a number.
It’s a roundabout way of converting a number into a list of digits because Elm doesn’t provide a function that does that.
Step 4: Convert the list into an array so that we can access each digit by specifying an index.
Step 5: Convert the array into a tuple.
It’s quite convenient to use a tuple to define multiple constants in one go like this:
Hopefully you now understand how the
extractDigits function works. Next, let’s implement the two non-existent functions (
arrayToTuple) we used inside
extractDigits. Add the following function definitions to the bottom of
Array module isn’t imported automatically by Elm. So we need to do that ourselves. Import it right below the line that imports the
Bitwise module in
As its name suggests the
stringToInt function takes a string and converts it into an integer. Because the
String.toInt function returns a value of type
Result instead of an integer, we need to do a little bit of extra work to convert a string to an integer.
We will cover the
Result type in detail later in this chapter, but for now think of it as a container that either contains the successful outcome of an operation or an error. It’s defined in Elm like this:
Don’t enter the
Result type definition shown below into the repl.
So if the operation succeeded as it did in
String.toInt "1", it contains
Ok followed by the output value:
Ok 1. Otherwise, it contains
Err followed by the error message. Here’s an example:
As you can see,
String.toInt cannot guarantee that it will always be able to convert a string into a number. That’s why it returns a
Result instead of an integer. But what about this code in the
Result.withDefault is a convenient function that returns the value if the result is
OK, but if the result is an
Err it returns a default value,
-1 in our case to indicate an invalid binary number.
Next, we will go through the code in
arrayToTuple function. As its name suggests, the
arrayToTuple function takes an array, reads values at each index and puts them in a tuple. As we learned in
the Array section, we can use the
Array.get function to read a value from an array at a specific index.
Array.get also doesn’t return the value we’re looking for directly. It returns a
Maybe works very similarly to
Result. It’s defined in Elm like this:
So if the value we’re looking for exists, it contains
Just followed by the value. Otherwise, it contains
Array.get cannot guarantee that there will be a value at a given index.
Maybe.withDefault is also a convenient function that returns the value if it exists, but if the value isn’t there it returns a default value instead.
Now if we reload the
rippleCarryAdder and run it again, we should be able to pass input signals as numbers instead of tuples.
Ah, much better! Not sure if you noticed, but
rippleCarryAdder has a bug: if an input binary number has zeros in the front, the output is incorrect.
When we add
0011, the output should be
1100 but we’re getting
11000. The problem is most likely in the
extractDigits function. Let’s find out which step is causing this issue.
The root cause of the issue is
toString. It gets rid of leading zeros when converting a number to a string. One way to fix this issue is by padding the output of
toString with enough zeros before passing it to the
String.split function like this:
But this whole approach of converting a number to a string which gets padded with zeros and then gets converted back to numbers feels a bit clumsy. Why don’t we write a function that directly converts a number into a list of digits without involving any string operation? Add the following function definition to the bottom of
digits function takes a number and returns a list of digits.
When we set out to implement a 4-bit ripple-carry adder, we broke the problem at hand into smaller sub-problems, solved them individually and combined the results. The
digits function follows a similar approach, but instead of using other functions to solve the sub-problems it solves them by itself. To understand how it does that, let’s break the problem of extracting digits from
1100 into sub-problems.
We start with the number
1100 which is our original problem. We take the last digit from it and reduce the problem to
110. We then take the last digit from
110 and further reduce the problem to
11. We continue to do this until there are no digits left in which case we simply return an empty list. Then we combine the solution to each sub-problem one at a time moving from left to right.
This process of solving a problem by first solving the smaller versions of the same problem is called recursion. The problem eventually gets reduced to something so small that we can just solve it directly. This is called a base case. In the example above, when there’s no digit left in the number we know that we’ve arrived at the base case. We solve it by simply returning an empty list.
Like most programming languages, Elm supports recursion by allowing a function to apply itself as we have done in the definition of
digits. When using recursion, we need to keep the following three things in mind.
1. How do we reduce the problem - If a problem can’t be reduced to smaller versions of itself then recursion probably isn’t the best tool to solve that problem. Therefore, we need to know how exactly are we going to reduce the problem. The
digits function does this by dividing the
// operator performs integer division, which truncates everything after the decimal point. We didn’t use recursion to build a 4-bit ripple-carry adder because we couldn’t reduce the problem into smaller versions of itself. Therefore, we had to assemble a bunch of other functions to solve the original problem instead of applying itself repeatedly.
2. What is the base case - A properly designed recursive function must simplify the problem with each invocation so that it will eventually reach the base case. Once the execution hits the base case, then it pops back out and starts gathering the results from each invocation. If we don’t provide a base case, a recursive function will create an infinite loop. Here’s the base case from the
After the third invocation of the
digits function, the number
1100 gets reduced to
1. When integer division is performed on that number by
0. At that point the condition for the base case is satisfied and an empty list is returned. If we didn’t have this condition, the
digits function would run forever.
3. How do we combine the result of each sub problem - Once the base case is reached, we need to provide a mechanism for combining the result from each invocation. The
digits function does this by using the
To see how recursion works behind the scenes, let’s walk through each step in the
digits function’s execution.
Breaking down a recursive function’s invocations like this is fine when we are trying to understand how recursion works in the beginning, but once we have developed an intuition for recursion we should avoid thinking explicitly about the sequence of invocations. We should instead focus on the three things we discussed above.
Hopefully you now understand how recursion works. If not, don’t worry. It’s a tricky concept to master. You just need more practice writing recursive functions and reasoning through them. There are many resources out there for diving deeper into recursion. I recommend checking out Khan Academy’s mini-course that not only explains recursion in simple terms, but also provides exercises for you to practice.
digits Function’s Performance
In Elm, it’s more efficient from a performance standpoint to build a list by adding an element to the front using the cons (
::) operator than by appending values at the end using the
++ operator. Let’s say we have a list like this:
[ 2, 3, 4, 5 ]. And we want to add
1 to the front of that list.
The cons operator doesn’t traverse the entire list before adding the given number to the front. Whereas
++ does. Therefore this code:
Essentially turns into this behind the scenes:
++ to append lists with single elements in the
digits function. Now that we know the cons operator is more performant, let’s change the
digits function to use that. In the above examples, the calculations being done are so small we won’t see much of a speed difference between the two operators. It’s when we are passing a very big number through the
digits function that the speed differences really start to add up. Change the
digits function in
RippleCarryAdder.elm to use the cons operator like this:
With your new found understanding of recursion, you should be able to figure out how the new implementation works. Change the
extractDigits function in
RippleCarryAdder.elm to use the new
extractDigits function now looks much more succinct with the
digits function. Here’s how it looked before:
The new version still doesn’t solve the problem of adding a number with leading zeros, but it’s quite easy to pad zeros to the front of a list. Add the following function definition to the bottom of
padZeros function pads a list with zeros until it has a given length represented by the parameter named
total. Let’s apply
Now when we add a binary number with leading zeros, we get a correct result.
Let’s turn our attention to the output of the
rippleCarryAdder function. Currently it looks like this:
We want it to look like this instead:
Ideally, we want it to look like this:
01100, but Elm truncates the leading zeros from both the repl and code file.
It’s not that difficult to show the output as a number. All we have to do is assemble the sum and carry-out digits in a list and turn that list into a number. Modify the
rippleCarryAdder function to look like this:
The only thing that has changed is the part inside the
in area. Let’s go through the new code step by step.
You don’t have to try the code listed in each step below. It’s there to show how
rippleCarryAdder produces a binary number as an output. Some of it won’t even compile because we haven’t implemented all the necessary functions used inside
Step 1: Put the records produced by each full adder into a list.
Step 2: Map the list of records into a list of sum digits.
As mentioned in the Record section, the
map function can be used to transform a list of records into a new list that contains values from a specific property. We used a special function called
.sum to extract the sum digits from each record.
Step 3: Add the carry-out digit to the front of the list.
Step 4: Convert the list of digits into a number.
numberFromDigits is a function we haven’t implemented yet. Let’s do that next. Add the following function definition to the bottom of
numberFromDigits function uses
foldl to reduce a list of digits into a single number. If you don’t remember how
foldl works, you might want to refresh your memory. The figure below shows how
numberFromDigits combines the digits to create a number by repeatedly applying the anonymous function given to
Now both inputs and outputs generated by the
rippleCarryAdder function look much nicer.
In the next section we will learn how to verify that the
rippleCarryAdder function behaves as expected using tests.