c++ ~ 1M在unordered_map中查找字符串键要比.net代码慢得多

我有一个perf test函数的。net和c++实现，该函数使用6838个键池中的字符串键在字典中进行854,750次查找。我编写这些函数是为了调查真实应用程序中的性能瓶颈。

。.NET实现是用f#编写的，使用字典，并为。NET 4.0编译

c++实现使用std::unordered_map，并在VS2010发布模式下构建。

如果我在c++实现中缩短键长度并使用"key_"前缀而不是"key_prefix_"它将在140毫秒内运行

我尝试的另一个技巧是用自定义的不可变字符串实现替换std::string，该实现具有指向源的const char*指针和一次性计算散列。使用这个字符串可以将c++实现的性能降低到190毫秒。

c++代码:

struct SomeData
{
public:
    float Value;
};
typedef std::string KeyString;
typedef std::unordered_map<KeyString, SomeData> DictionaryT;
const int MaxNumberOfRuns = 125;
const int MaxNumberOfKeys = 6838;
DictionaryT dictionary;
dictionary.rehash(MaxNumberOfKeys);
auto timer = Stopwatch::StartNew();
int lookupCount = 0;
char keyBuffer[100] = "key_prefix_";
size_t keyPrefixLen = std::strlen(keyBuffer);
/// run MaxNumberOfRuns * MaxNumberOfKeys iterations
for(int runId = 0; runId < MaxNumberOfRuns; runId++)
{
    for(int keyId = 0; keyId < MaxNumberOfKeys; keyId++)
    {
        /// get a new key from the pool of MaxNumberOfKeys keys           
        int randomKeySuffix = (std::rand() % MaxNumberOfKeys);
        ::itoa(randomKeySuffix, keyBuffer + keyPrefixLen, 10);
        KeyString key = keyBuffer;
        /// lookup key in the dictionary         
        auto dataIter = dictionary.find(key);
        SomeData* data;
        if(dataIter != dictionary.end())
        {
            /// get existing value           
            data = &dataIter->second;
        }
        else
        {
            /// add a new value
            data = &dictionary.insert(dataIter, DictionaryT::value_type(key, SomeData()))->second;
        }
        /// update corresponding value in the dictionary
        data->Value += keyId * runId;
        lookupCount++;
    }
}
timer.Stop();
std::cout << "Time: " << timer.GetElapsedMilleseconds() << " ms" << std::endl;
std::cout << "Lookup count: " << lookupCount << std::endl;

打印:

时间:636 ms
查找计数:854750

open System
open System.Diagnostics
open System.Collections.Generic
type SomeData =
    struct
        val mutable Value : float
    end
let dictionary = new Dictionary<string, SomeData>()
let randomGen = new Random()
let MaxNumberOfRuns = 125
let MaxNumberOfKeys = 6838
let timer = Stopwatch.StartNew()
let mutable lookupCount = 0
/// run MaxNumberOfRuns * MaxNumberOfKeys iterations
for runId in 1 .. MaxNumberOfRuns do
    for keyId in 1 .. MaxNumberOfKeys do
        /// get a new key from the pool of MaxNumberOfKeys keys
        let randomKeySuffix = randomGen.Next(0, MaxNumberOfKeys).ToString()        
        let key = "key_prefix_" + randomKeySuffix
        /// lookup key in the dictionary
        let mutable found, someData = dictionary.TryGetValue (key)
        if not(found) then
            /// add a new value
            someData <- new SomeData()
            dictionary.[key] <- someData
        /// update corresponding value in the dictionary
        someData.Value <- someData.Value + float(keyId) * float(runId)
        lookupCount <- lookupCount + 1
timer.Stop()
printfn "Time: %d ms" timer.ElapsedMilliseconds
printfn "Lookup count: %d" lookupCount