Question

Outside of the ensuring that they cannot be changed (to the tune of a compiler error), does the JIT make any optimisations for const locals?

Eg.

public static int Main(string[] args)
{
    const int timesToLoop = 50;

    for (int i=0; i<timesToLoop; i++)
    {
        // ...
    }
}

Answer 1

The generated IL is different (using Release mode):

using constant local                   using normal local
---------------------------------------------------------------------
.entrypoint                            .entrypoint
.maxstack 2                            .maxstack 2
.locals init (                         .locals init (
    [0] int32 i)                           [0] int32 timesToLoop,
L_0000: ldc.i4.0                           [1] int32 i)
L_0001: stloc.0                        L_0000: ldc.i4.s 50 
L_0002: br.s L_0008                    L_0002: stloc.0 
L_0004: ldloc.0                        L_0003: ldc.i4.0  
L_0005: ldc.i4.1                       L_0004: stloc.1 
L_0006: add                            L_0005: br.s L_000b 
L_0007: stloc.0                        L_0007: ldloc.1 
L_0008: ldloc.0                        L_0008: ldc.i4.1 
L_0009: ldc.i4.s 50                    L_0009: add
L_000b: blt.s L_0004                   L_000a: stloc.1 
L_000d: ret                            L_000b: ldloc.1 
                                       L_000c: ldloc.0 
                                       L_000d: blt.s L_0007
                                       L_000f: ret

As you can see the compiler replaces all variable usages by the value of the constant which results in a smaller stack.

Answer 2

I gave the code a quick performance test, using Snippet Compiler. The code I used is as follows:

    public static void Main()
    {
        DateTime datStart = DateTime.UtcNow;
        const int timesToLoop = 1000000;

        for (int i=0; i < timesToLoop; i++)
        {
            WL("Line Number " + i.ToString());
        }

        DateTime datEnd = DateTime.UtcNow;
        TimeSpan tsTimeTaken = datEnd - datStart;
        WL("Time Taken: " + tsTimeTaken.TotalSeconds);
        RL();
    }

Note, WL and RL are simply helper methods to read and write to the console.

To test the non-constant version, I simply removed the const keyword. The results were surprising:

                        Time Taken (average of 3 runs)

Using const keyword         26.340s
Without const keyword       28.276s

I m aware that this is very rough n ready test, but the use of the const keyword appears to count as a valid micro-optimization.

Answer 3

Your code (using const) will actually be compiled as:

public static int Main(string[] args){    
    for (int i=0; i < 50; i++)  
    {

    }
}

while a variable will compile as a variable:

public static int Main(string[] args){
    int timesToLoop = 50;    
    for (int i=0; i < timesToLoop; i++)  
    {

    }
}

Answer 4

This is not anywhere near an answer, just thought it would be nice to share this however the article does not mention runtime benefits explicitly:

Coding Standard Rule #2: Use const Wherever Possible

Excerpt:

Reasoning: The upside of using const as much as possible is compiler-enforced protection from unintended writes to data that should be read-only.

Answer 5

One difference is that if you had an assembly that referenced a const field in another assembly and you later changed that value, the referencing assembly would still use the old value until it was rebuilt.

友情链接