[gdal-dev] Experience with slowness of free() on Windows with lots of allocations?

[Uhrig, Stefan]

After further investigation I found that the order in which the CMyClass instances are destroyed matter. delete[] destroys them in reverse order while the vector destroys them in a forward order.

I could emulate both ways by using placement new:

#include <cassert>
#include <vector>

int SINGLE_ALLOC_SIZE = 21200;
int NUMBER_OF_ALLOCS = 21200 * 4;

class CMyClass
{
public:
    CMyClass()
    {
        lpData = new char[SINGLE_ALLOC_SIZE];
        assert(lpData);
    };

    ~CMyClass()
    {
        delete[] lpData;
    };

public:
    char* lpData;
};

void strat1()
{
    printf("strat1 start\n");
    CMyClass* lpList = new CMyClass[NUMBER_OF_ALLOCS];
    printf("after alloc. starting freeing\n");
    delete[] lpList;
    printf("end\n");
}

void strat2()
{
    printf("strat2 start\n");
    {
        std::vector<CMyClass> lpList(NUMBER_OF_ALLOCS);
        printf("after alloc. starting freeing\n");
    }
    printf("end\n");
}

void strat3()
{
    printf("strat3 start\n");
    void* block = malloc(NUMBER_OF_ALLOCS * sizeof(CMyClass));
    for (size_t i = 0; i < NUMBER_OF_ALLOCS; ++i)
    {
        void* address = (char*)block + i * sizeof(CMyClass);
        new (address) CMyClass();
    }
    printf("after alloc. starting freeing\n");
    for (size_t i = 0; i < NUMBER_OF_ALLOCS; ++i)
    {
        CMyClass* instance = (CMyClass*)((char*)block + (NUMBER_OF_ALLOCS - i - 1) * sizeof(CMyClass));
        instance->~CMyClass();
    }
    free(block);
    printf("end\n");
}

void strat4()
{
    printf("strat4 start\n");
    void* block = malloc(NUMBER_OF_ALLOCS * sizeof(CMyClass));
    for (size_t i = 0; i < NUMBER_OF_ALLOCS; ++i)
    {
        void* address = (char*)block + i * sizeof(CMyClass);
        new (address) CMyClass();
    }
    printf("after alloc. starting freeing\n");
    for (size_t i = 0; i < NUMBER_OF_ALLOCS; ++i)
    {
        CMyClass* instance = (CMyClass*)((char*)block + i * sizeof(CMyClass));
        instance->~CMyClass();
    }
    free(block);
    printf("end\n");
}

int main()
{
    do
    {
        strat1();
        strat2();
        strat3();
        strat4();
    } while (1);

}

From: gdal-dev <gdal-dev-bounces at lists.osgeo.org> On Behalf Of Uhrig, Stefan via gdal-dev
Sent: Thursday, March 21, 2024 2:17 PM
To: gdal-dev at lists.osgeo.org
Subject: Re: [gdal-dev] Experience with slowness of free() on Windows with lots of allocations?

I was curious and gave it a try. I also saw the bad performance on deallocations, but surprisingly the usage of a std::vector in the outer loop speeds things up considerably.

I could still see a peak memory usage of 1.8GiB, so it does not seem as if the compiler did optimize something out.
<snip>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.osgeo.org/pipermail/gdal-dev/attachments/20240321/2f6f6559/attachment-0001.htm>