c# - Performance between Iterating through IEnumerable<T> and List<T>

Question

Today, I faced a problem with performance while iterating through a list of items. After done some diagnostic, I finally figured out the reason which slowed down performance. It turned out that iterating through an IEnumerable<T> took much more time than iterating through a List<T>. Please help me understand why IEnumerable<T> is slower than List<T>.

UPDATE benchmark context:

I'm using NHibernate to fetch a collection of items from a database into an IEnumerable<T> and sum its property's value. This is just a simple entity without any reference type:

public SimpleEntity
{
    public int Id {get;set}
    public string Name {get;set}
    public decimal Price {get;set}
}

Public Test
{
    void Main()
    {
        //this query get a list of about 200 items
        IEnumerable<SimpleEntity> entities = from entity in Session.Query<SimpleEntity>
                                             select entity;

        decimal value = 0.0;
        foreach(SimpleEntity item in entities)
        {
             //this for loop took 1.5 seconds 
             value += item.Price;
        }

        List<SimpleEntity> lstEntities = entities.ToList();

        foreach(SimpleEntity item in lstEntities)
        {
             //this for loop took less than a milisecond
             value += item.Price;
        }
    }
}

Answer 1

Enumerating an IEnumerable<T> is 2 to 3 times slower than enumerating the same List<T> directly. This is due to a subtlety on how C# selects its enumerator for a given type.

List<T> exposes 3 enumerators:

1. List<T>.Enumerator List<T>.GetEnumerator()
2. IEnumerator<T> IEnumerable<T>.GetEnumerator()
3. IEnumerator IEnumerable.GetEnumerator()

When C# compiles a foreach loop, it will select the enumerator in the above order. Note that a type doesn't need to implement IEnumerable or IEnumerable<T> to be enumerable, it just needs a method named GetEnumerator() that returns an enumerator.

Now, List<T>.GetEnumerator() has the advantage of being statically typed which makes all calls to List<T>.Enumerator.get_Current and List<T>.Enumerator.MoveNext() static-bound instead of virtual.

10M iterations (coreclr):

for(int i ...)               73 ms
foreach(... List<T>)        215 ms
foreach(... IEnumerable<T>) 698 ms
foreach(... IEnumerable)   1028 ms
for(int *p ...)              50 ms

10M iterations (Framework):

for(int i ...)              210 ms
foreach(... List<T>)        252 ms
foreach(... IEnumerable<T>) 537 ms
foreach(... IEnumerable)    844 ms
for(int *p ...)             202 ms

Disclaimer

I should point out the actual iteration in a list is rarely the bottleneck. Keep in mind those are hundreds of milliseconds over millions of iterations. Any work in the loop more complicated than a few arithmetic operations will be overwhelmingly costlier than the iteration itself.