点解GPU计算会比CPU更慢..?

复制内容到剪贴板代码:template <typename type, size_t Rank, typename Allocator>
void print(const multiarraycase<type, Rank, Allocator> &a, ostream &output)
{
typedef multiarraycase<type, Rank, Allocator>::base_array base_array;
__if_exists(base_array::type)
{
for (size_t i = 0; i < a.size(0); ++i)
print(a[i], output);
output << "\n";
}
__if_not_exists(base_array::type)
{
std::transform(a.begin(), a.end(), std::ostream_iterator<type>(output, "\t"),
[](type value)
{
return (type)round(value * 100) / 100;
});
output << "\n";
}
}
template <typename type, size_t Rank, typename Allocator>
void show(const multiarraycase<type, Rank, Allocator> &a)
{
print(a, std::cout);
}

void main()
{
default_random_engine generator;
auto GetRandom = bind(uniform_int_distribution<int>(1, 9), std::ref(generator));

high_resolution_clock::time_point record;
duration<double> time_diff;

size_t N = 10000000;
multiarray<float> a(N), b(N), c(N);

array_view<const float, 1> _a(a.size(), a.data());
array_view<const float, 1> _b(b.size(), b.data());
array_view<float, 1> _c(c.size(), c.data());

while (true)
{

system("cls");

for (size_t i = 0; i < N; ++i)
{
a[i] = GetRandom();
b[i] = GetRandom();
}

record = high_resolution_clock::now();
_a.refresh();
_b.refresh();
parallel_for_each(
// Define the compute domain, which is the set of threads that are created.
_c.extent,
// Define the code to run on each thread on the accelerator.
[=](index<1> idx) restrict(amp)
{
_c[idx] = _a[idx] + _b[idx];
}
);
_c.synchronize();
time_diff = duration_cast<duration<double>>(high_resolution_clock::now() - record);
cout << "GPU: " << time_diff.count() << endl;

record = high_resolution_clock::now();
for (size_t i = 0; i < N; ++i)
{
c[i] = a[i] + b[i];
}
time_diff = duration_cast<duration<double>>(high_resolution_clock::now() - record);
cout << "CPU: " << time_diff.count() << endl;

cout << endl;

system("pause");

}
}

引用:原帖由 烟民比食屎9更贱 於 2014-4-16 12:05 AM 发表
呢个似乎唔系 openCL, 亦唔系 Cuda, 何解你会认为系用到 GPU?

引用:原帖由 Susan﹏汪汪 於 2014-4-16 12:06 AM 发表

Visual studio 的C++ amp功能

引用:原帖由 烟民比食屎9更贱 於 2014-4-16 03:05 AM 发表
呢个 amp lib 几令我怀疑其效能, 好似无咩人提过。同埋你 hardware 有否合乎哂佢既要求?有无用 GPU-Z 睇下个 GPU loading 有无增加?

C++ Accelerated Massive Parallelism (C++ AMP) is a library implemented on ...

引用:原帖由 a8d7e8 於 2014-4-16 01:33 PM 发表
Data dependency?

引用:原帖由 Susan﹏汪汪 於 2014-4-16 13:43 发表

唔明想问咩

引用:原帖由 Susan﹏汪汪 於 2014-4-16 04:06 AM 发表

VS可以直接睇到GPU堆叠

引用:原帖由 烟民比食屎9更贱 於 2014-4-16 10:53 PM 发表
我谂佢模拟出黎, 有无睇过实际 GPU realtime loading 先?

复制内容到剪贴板代码:c[i] = a[i] + b[i]

复制内容到剪贴板代码:_a.refresh();
_b.refresh();
和
_c.synchronize();