...

sycl::property_list properties{sycl::property::queue::enable_profiling()};
auto qu = sycl::queue{sycl::gpu_selector_v, properties};

auto sBuf  = sycl::malloc_device<quint16>(pixAll, qu); //
auto sFrc  = sycl::malloc_device<quint16>(pixImg, qu); //fractal

benchmark([&]() {
    e1 = qu.memcpy(sBuf, buf16, pixAll * sizeof(quint16));
    e2 = tMandelbrot(qu, -0.9, -0.9, 0.9, 0.9, width, height, 4095, sFrc, {});
    e1.wait();
    e2.wait();
}, "Queue_prl");
    
int64_t e1_end    = e1.get_profiling_info<sycl::info::event_profiling::command_end>();
int64_t e2_start  = e2.get_profiling_info<sycl::info::event_profiling::command_start>();
    
std::cout << "E2_start - E1_end: " << e2_start-e1_end << std::endl;

memcpy       : 13526 mcs
tMandelbrot  : 30530 mcs
Queue_prl    : 42079 mcs
E2_start - E1_end: 109997

    e1 = tMandelbrot(qu, -0.9, -0.9, 0.9, 0.9, width, height, 4095, sFrc, {});
    e2 = qu.memcpy(sBuf, buf16, pixAll * sizeof(quint16));
    e1.wait();
    e2.wait();

memcpy       : 14009 mcs
tMandelbrot  : 31086 mcs
Queue_prl    : 30888 mcs
E2_start - E1_end: -30846426

template <typename Func>
auto benchmark(Func &&func, string str, int n=12, int d=0)
{
    auto tm_beg = chrono::steady_clock::now();
    func();
    auto tm_end = chrono::steady_clock::now();
    auto tm_time = chrono::duration_cast<chrono::microseconds>(tm_end - tm_beg);
    cout << addVsps(str, n, d) << " : " << tm_time.count() << " mcs" << endl;
}

sycl::event tMandelbrot(sycl::queue &qu, float x1, float y1, float x2, float y2,
                            int width, int height, int maxIters, unsigned short * image, const std::vector<sycl::event> &evt)
{
    float dx = (x2-x1)/width, dy = (y2-y1)/height;
    int pxc = width*height;
    sycl::range nRng = sycl::range<1>(pxc);
    return qu.parallel_for(nRng, evt, [=](sycl::id<1> idx) {
        auto gid = idx[0];
        int j = gid/width;    
        int i = gid - j*width; 
        //
        float sx = x1+dx*i;
        float sy = y1+dy*j;
        float zx=0, zy=0;
        int cnt = 0;
        while (zx*zx + zy*zy <= 4.0f && cnt <= maxIters) {
            float nx = zx*zx - zy*zy + sx;
            zy = 2.0f * zx * zy + sy;
            zx = nx;
            cnt += 1;
        }
        image[gid] = cnt - tMCLR;
    });
}