具有和不具有参数C / C ++的函数之间的执行时间

Question

I'd just like to ask what's the difference in execution time when:

a) We have variables declared in main() and send them in functions using arguments

b) We have global variables and access them directly from functions with no arguments

The thing is I started writing few aps and just saw a few discussions about this, but I haven't yet written any application that makes the difference bigger then few ms.

Answer 1

Passing parameters to function involves usually one of two assembler instructions: push param (and later pop param ) or mov ax, param . Since processor is able to do (a lot) more in a second, such "optimization" mostly probably will go unnoticed (a few ms on the whole program is below the error margin)

Using global variables in the place of function parameters will cause a huge mess in the code with almost unnoticeable or possibly no performance gain.

Answer 2

All of this depends on what CPU and compiler that are used.

When you pass a parameter to a function, one of the following could happen:

• The parameter is stored in a CPU register. This is very efficient.
• The parameter is stored on the stack. This is the most common. It involves some minor overhead of pushing/popping the parameter to/from the stack when the function is started/finished.
• The parameter isn't a new variable at all. Instead, the compiler inlines the function and uses the original variable for modifications. This is about as efficient as you can get.

Using a global variable is will be ever so slightly faster than using the stack. It will unlikley be faster than using a CPU register: inside the function the value might need to be loaded into such a register before calculations anyhow.

It should be noted we are talking about a few CPU ticks here and there.

My advise:

• You should never attempt any kind of manual optimizations unless you have in-depth hardware knowledge of the specific CPU used. If you have no such knowledge, the compiler will optimize the code better than you do in 99% of the cases. Because the compiler port was most likely written by an expert of the given system. The compiler also knows the overall performance picture, which the programmer does not, so the compiler is more suitable for doing the optimization.
• You should never attempt any kind of manual optimization unless you have actually performed formal benchmarking and found a bottleneck in the program.
• Global variables are incredibly bad and dangerous to use. They lead to spaghetti code and they are not thread-safe.
• If you are writing some kind of high-end desktop application, such as PC programs or phone apps, then using global variables to increase performance is complete nonsense . You are on a system which has non-existent real-time performance to begin with! At any given time, your OS might decide to chew up many billions of CPU ticks while giving your program the finger. So don't go chasing 1 or 2 CPU ticks.
• These kind of manual optimizations only make sense if you are developing embedded systems applications that are very close to the hardware, and at the same time have hard real-time requirements.

Answer 3

Stack operations are very efficient, and they are made even more efficient because the stack is likely to be in cache memory. That could make using the stack even faster than not using it.

Answer 4

Expect parameter passed variables to be a lot faster than global variables. In Modern ABIs, function parameters are passed mostly using CPU registers which are immediately available to the CPU.

Global variables must be read from (static) memory. Worse than that, static memory is allocated on its own memory page, which is generally far from your stack (or heap) memory. This means that cache misses are more likely which in turn means that access can consume a large number of CPU cycles.

Obviously this depends a lot on your usage patterns.