Some code using undocumented NT profiling functions to count LLC misses. I don't know how accurate it is, haven't used it for anything yet.

nt_profile_cache_miss.cpp

typedef enum _KPROFILE_SOURCE
{
    ProfileTime                     = 0x00,
    ProfileTotalIssues              = 0x02,
    ProfileBranchInstructions       = 0x06,
    ProfileCacheMisses              = 0x0A,
    ProfileBranchMispredictions     = 0x0B,
    ProfileTotalCycles              = 0x13,
    ProfileUnhaltedCoreCycles       = 0x19,
    ProfileInstructionRetired       = 0x1A,
    ProfileUnhaltedReferenceCycles  = 0x1B,
    ProfileLLCReference             = 0x1C,
    ProfileLLCMisses                = 0x1D,
    ProfileBranchInstructionRetired = 0x1E,
    ProfileBranchMispredictsRetired = 0x1F,
} KPROFILE_SOURCE, *PKPROFILE_SOURCE;


typedef NTSYSAPI NTSTATUS NTAPI F_NtCreateProfile(
    OUT PHANDLE        ProfileHandle,
    IN HANDLE          Process OPTIONAL,
    IN PVOID           ImageBase,
    IN ULONG           ImageSize,
    IN ULONG           BucketSize,
    IN PVOID           Buffer,
    IN ULONG           BufferSize,
    IN KPROFILE_SOURCE ProfileSource,
    IN KAFFINITY       Affinity
);

typedef NTSYSAPI NTSTATUS NTAPI F_NtStartProfile(
    IN HANDLE           ProfileHandle
);

typedef NTSYSAPI NTSTATUS NTAPI F_NtStopProfile(
    IN HANDLE           ProfileHandle
);


HMODULE GetCurrentModule()
{
    HMODULE module = NULL;
    GetModuleHandleExW(GET_MODULE_HANDLE_EX_FLAG_FROM_ADDRESS, (LPCWSTR) GetCurrentModule, &module);
    return module;
}


void test()
{
    HMODULE nt = LoadLibraryA("ntdll.dll");

    F_NtCreateProfile* NtCreateProfile = (F_NtCreateProfile*) GetProcAddress(nt, "NtCreateProfile");
    F_NtStartProfile* NtStartProfile = (F_NtStartProfile*) GetProcAddress(nt, "NtStartProfile");
    F_NtStopProfile* NtStopProfile = (F_NtStopProfile*) GetProcAddress(nt, "NtStopProfile");

    HANDLE process = GetCurrentProcess();
    HMODULE module = GetCurrentModule();
    printf("%p %p\n", process, module);

    MODULEINFO module_info;
    BOOL module_info_success = GetModuleInformation(process, module, &module_info, sizeof(module_info));
    if (!module_info_success)
    {
        printf("GetModuleInformation() failed!\n");
        return;
    }

    void* image_base = module_info.lpBaseOfDll;
    umm image_size = module_info.SizeOfImage;

    printf("%p\n", image_base);
    printf("%d\n", (int) image_size);

    HANDLE profile;

    ULONG buffer[1] = {};
    NTSTATUS create_status = NtCreateProfile(&profile, process, image_base, image_size, 30, buffer, 4, ProfileLLCMisses, -1);
    printf("NtCreateProfile: %x\n", create_status);

    NtStartProfile(profile);
    {
        int* matrix = (int*) malloc(sizeof(int) * 16 * 1024 * 16 * 1024);
        for (int i = 0; i < 10000; i++)
            for (int j = 0; j < 10000; j++)
                matrix[i * 16 * 1024 + j] *= i * j;
        free(matrix);
    }
    NtStopProfile(profile);
    printf("Cache misses 1: %d\n", buffer[0]);
    buffer[0] = 0;


    NtStartProfile(profile);
    {
        int* matrix = (int*) malloc(sizeof(int) * 16 * 1024 * 16 * 1024);
        for (int i = 0; i < 10000; i++)
            for (int j = 0; j < 10000; j++)
                matrix[j * 16 * 1024 + i] *= i * j;
        free(matrix);
    }
    NtStopProfile(profile);
    printf("Cache misses 2: %d\n", buffer[0]);
    buffer[0] = 0;

    exit(0);
}