Optimizing C and C++ Code

Embedded software often runs on processors with limited computation power, thus optimizing the code becomes a necessity. In this article we will explore the following optimization techniques for C and C++ code developed for Real-time and Embedded Systems. 

1. Adjust structure sizes to power of two
2. Place case labels in narrow range
3. Place frequent case labels first
4. Break big switch statements into nested switches
5. Minimize local variables
6. Declare local variables in the inner most scope
7. Reduce the number of parameters
8. Use references for parameter passing and return value for types bigger than 4 bytes
9. Don't define a return value if not used
10. Consider locality of reference for code and data
11. Prefer int over char and short
12. Define lightweight constructors
13. Prefer initialization over assignment
14. Use constructor initialization lists
15. Do not declare "just in case" virtual functions
16. In-line 1 to 3 line functions

Many techniques discussed here have roots in the material we covered in the articles dealing with C to Assembly translation. A good understanding of the following articles will help:

• C To Assembly Translation
• C To Assembly Translation II
• C To Assembly Translation III

Adjust structure sizes to power of two

When arrays of structures are involved, the compiler performs a multiply by the structure size to perform the array indexing. If the structure size is a power of 2, an expensive multiply operation will be replaced by an inexpensive shift operation. Thus keeping structure sizes aligned to a power of 2 will improve performance in array indexing. 

Place case labels in narrow range

If the case labels are in a narrow range, the compiler does not generate a if-else-if cascade for the switch statement. Instead, it generates a jump table of case labels along with manipulating the value of the switch to index the table. This code generated is faster than if-else-if cascade code that is generated in cases where the case labels are far apart. Also, performance of a jump table based switch statement is independent of the number of case entries in switch statement.

Place frequent case labels first

If the case labels are placed far apart, the compiler will generate if-else-if cascaded code with comparing for each case label and jumping to the action for leg on hitting a label match. By placing the frequent case labels first, you can reduce the number of comparisons that will be performed for frequently occurring scenarios. Typically this means that cases corresponding to the success of an operation should be placed before cases of failure handling. 

Break big switch statements into nested switches

The previous technique does not work for some compilers as they do not generate the cascade of if-else-if in the order specified in the switch statement. In such cases nested switch statements can be used to get the same effect.

To reduce the number of comparisons being performed, judiciously break big switch statements into nested switches. Put frequently occurring case labels into one switch and keep the rest of case labels into another switch which is the default leg of the first switch.

 // This switch statement performs a switch on frequent messages and handles the // infrequent messages with another switch statement in the default leg of the outer // switch statement pMsg = ReceiveMessage(); switch (pMsg->type) { case FREQUENT_MSG1: handleFrequentMsg1(); break; case FREQUENT_MSG2: handleFrequentMsg2(); break; . . . case FREQUENT_MSGn: handleFrequentMsgn(); break; default: // Nested switch statement for handling infrequent messages. switch (pMsg->type) { case INFREQUENT_MSG1: handleInfrequentMsg1(); break; case INFREQUENT_MSG2: handleInfrequentMsg2(); break; . . . case INFREQUENT_MSGm: handleInfrequentMsgm(); break; } } 

Minimize local variables

If the number of local variables in a function is less, the compiler will be able to fit them into registers. Hence, it will be avoiding frame pointer operations on local variables that are kept on stack. This can result in considerable improvement due to two reasons:

• All local variables are in registers so this improves performance over accessing them from memory.
• If no local variables need to be saved on the stack, the compiler will not incur the overhead of setting up and restoring the frame pointer.

Declare local variables in the inner most scope

Do not declare all the local variables in the outermost function scope. You will get better performance if local variables are declared in the inner most scope. Consider the example below; here object a is needed only in the error case, so it should be invoked only inside the error check. If this parameter was declared in the outermost scope, all function calls would have incurred the overhead of object a's creation (i.e. invoking the default constructor for a).

int foo(char *pName) { if (pName == NULL) { A a; ... return ERROR; } ... return SUCCESS; }

Reduce the number of parameters

Function calls with large number of parameters may be expensive due to large number of parameter pushes on stack on each call. For the same reason, avoid passing complete structures as parameters. Use pointers and references in such cases.

Use references for parameter passing and return value for types bigger than 4 bytes

Passing parameters by value results in the complete parameter being copied on to the stack. This is fine for regular types like integer, pointer etc. These types are generally restricted to four bytes. When passing bigger types, the cost of copying the object on the stack can be prohibitive. In case of classes there will be an additional overhead of invoking the constructor for the temporary copy that is created on the stack. When the function exits the destructor will also be invoked.

Thus it is efficient to pass references as parameters. This way you save on the overhead of a temporary object creation, copying and destruction. This optimization can be performed easily without a major impact to the code by replacing pass by value parameters by const references. (It is important to pass const references so that a bug in the called function does not change the actual value of the parameter.

Passing bigger objects as return values also has the same performance issues. A temporary return object is created in this case too.

Don't define a return value if not used

The called function does not "know" if the return value is being used. So, it will always pass the return value. This return value passing may be avoided by not defining a return value which is not being used.

Consider locality of reference for code and data

The processor keeps data or code that is referenced in cache so that on its next reference if gets it from cache. These cache references are faster. Hence it is recommended that code and data that are being used together should actually be placed together physically. This is actually enforced into the language in C++. In C++, all the object's data is in one place and so is code. When coding is C, the declaration order of related code and functions can be arranged so that closely coupled code and data are declared together.

Prefer int over char and short

With C and C++ prefer use of int over char and short. The main reason behind this is that C and C++ perform arithmetic operations and parameter passing at integer level, If you have an integer value that can fit in a byte, you should still consider using an int to hold the number. If you use a char, the compiler will first convert the values into integer, perform the operations and then convert back the result to char.

Lets consider the following code which presents two functions that perform the same operation with char and int.

char sum_char(char a, char b) { char c; c = a + b; return c; } int sum_int(int a, int b) { int c; c = a + b; return c; } 

A call to sum_char involves the following operations:

1. Convert the second parameter into an int by sign extension (C and C++ push parameters in reverse)
2. Push the sign extended parameter on the stack as b.
3. Convert the first parameter into an int by sign extension.
4. Push the sign extended parameter on to the stack as a.
5. The called function adds a and b
6. The result is cast to a char.
7. The result is stored in char c.
8. c is again sign extended
9. Sign extended c is copied into the return value register and function returns to caller.
10. The caller now converts again from int to char.
11. The result is stored.

A call to sum_int involves the following operations:

1. Push int b on stack
2. Push int a on stack
3. Called function adds a and b
4. Result is stored in int c
5. c is copied into the return value register and function returns to caller.
6. The called function stores the returned value.

Thus we can conclude that int should be used for all interger variables unless storage requirements force us to use a char or short. When char and short have to be used, consider the impact of byte alignment and ordering to see if you would really save space. (Many processors align structure elements at 16 byte boundaries) 

Define lightweight constructors

As far as possible, keep the constructor light weight. The constructor will be invoked for every object creation. Keep in mind that many times the compiler might be creating temporary object over and above the explicit object creations in your program. Thus optimizing the constructor might give you a big boost in performance. If you have an array of objects, the default constructor for the object should be optimized first as the constructor gets invoked for every object in the array.

Prefer initialization over assignment

Consider the following example of a complex number::

void foo() { Complex c; c = (Complex)5; } void foo_optimized() { Complex c = 5; } 

In the function foo, the complex number c is being initialized first by the instantiation and then by the assignment. In foo_optimized, c is being initialized directly to the final value, thus saving a call to the default constructor of Complex.

Use constructor initialization lists

Use constructor initialization lists to initialize the embedded variables to the final initialization values. Assignments within the constructor body will result in lower performance as the default constructor for the embedded objects would have been invoked anyway. Using constructor initialization lists will directly result in invoking the right constructor, thus saving the overhead of default constructor invocation.  

In the example given below, the optimized version of the Employee constructor saves the default constructor calls for m_name and m_designation strings.

Employee::Employee(String name, String designation) { m_name = name; m_designation = designation; } /* === Optimized Version === */ Employee::Employee(String name, String designation): m_name(name), m_destignation (designation) { } 

Do not declare "just in case" virtual functions

Virtual function calls are more expensive than regular function calls so do not make functions virtual "just in case" somebody needs to override the default behavior. If the need arises, the developer can just as well edit the additional base class header file to change the declaration to virtual.

In-line 1 to 3 line functions

Converting small functions (1 to 3 lines) into in-line will give you big improvements in throughput. In-lining will remove the overhead of a function call and associated parameter passing. But using this technique for bigger functions can have negative impact on performance due to the associated code bloat. Also keep in mind that making a method inline should not increase the dependencies by requiring a explicit header file inclusion when you could have managed by just using a forward reference in the non-inline version. (See the article on header file include patterns for more details).