cory-master/BabelStream
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

[SYCL 2020] use new reduction for dot kernel

Browse Source
This commit is contained in:
Tom Deakin 2021-01-21 10:37:56 +00:00
parent aa0ab6a8e3
commit 42c8954789
1 changed files with 12 additions and 31 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

45
SYCLStream.cpp
View File

@ -69,7 +69,7 @@ SYCLStream<T>::SYCLStream(const int ARRAY_SIZE, const int device_index)
d_a = new sycl::buffer<T>(array_size); d_a = new sycl::buffer<T>(array_size);
d_b = new sycl::buffer<T>(array_size); d_b = new sycl::buffer<T>(array_size);
d_c = new sycl::buffer<T>(array_size); d_c = new sycl::buffer<T>(array_size);
d_sum = new sycl::buffer<T>(dot_num_groups); d_sum = new sycl::buffer<T>(1);
} }
template <class T> template <class T>
@ -150,47 +150,28 @@ void SYCLStream<T>::triad()
template <class T> template <class T>
T SYCLStream<T>::dot() T SYCLStream<T>::dot()
{ {
queue->submit([&](sycl::handler &cgh) queue->submit([&](sycl::handler &cgh)
{ {
sycl::accessor ka {*d_a, cgh, sycl::read_only}; sycl::accessor ka {*d_a, cgh, sycl::read_only};
sycl::accessor kb {*d_b, cgh, sycl::read_only}; sycl::accessor kb {*d_b, cgh, sycl::read_only};
sycl::accessor ksum {*d_sum, cgh, sycl::write_only};
//sycl::local_accessor wg_sum {sycl::range<1>(dot_wgsize), cgh}; // Reduction object, to perform summation
auto wg_sum = sycl::accessor<T, 1, sycl::access::mode::read_write, sycl::access::target::local>(sycl::range<1>(dot_wgsize), cgh); // Initialises the result to zero
auto sumReducer = sycl::reduction(*d_sum, cgh, std::plus<T>(), sycl::property::reduction::initialize_to_identity);
size_t N = array_size; cgh.parallel_for(sycl::range<1>{array_size}, sumReducer, [=](sycl::id<1> idx, auto& sum)
cgh.parallel_for(sycl::nd_range<1>(dot_num_groups*dot_wgsize, dot_wgsize), [=](sycl::nd_item<1> item)
{ {
size_t i = item.get_global_id(0); sum += ka[idx] * kb[idx];
size_t li = item.get_local_id(0);
size_t global_size = item.get_global_range()[0];
wg_sum[li] = 0.0;
for (; i < N; i += global_size)
wg_sum[li] += ka[i] * kb[i];
size_t local_size = item.get_local_range()[0];
for (int offset = local_size / 2; offset > 0; offset /= 2)
{
item.barrier(sycl::access::fence_space::local_space);
if (li < offset)
wg_sum[li] += wg_sum[li + offset];
}
if (li == 0)
ksum[item.get_group(0)] = wg_sum[0];
});
}); });
T sum = 0.0; });
auto h_sum = d_sum->template get_access<sycl::access::mode::read>();
for (int i = 0; i < dot_num_groups; i++) // Get access on the host, and return a copy of the data (single number)
{ // This will block until the result is available, so no need to wait on the queue.
sum += h_sum[i]; sycl::host_accessor result {*d_sum, sycl::read_only};
} return result[0];
return sum;
} }
template <class T> template <class T>