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

Minor performance tuning for SYCL benchmark

Browse Source 
* Pre-compiling kernel binaries when setting up the benchmark,
   like OpenCL equivalent

* Using the linear access syntax for buffers
This commit is contained in:
Ruyman Reyes 2016-10-18 13:09:19 +01:00
parent 23a43bfa6d
commit d562283cde
1 changed files with 34 additions and 17 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

51
SYCLStream.cpp
View File

@ -11,12 +11,20 @@
using namespace cl::sycl;
#define WGSIZE 256
// Cache list of devices
bool cached = false;
std::vector<device> devices;
void getDeviceList(void);
program * p;
/* Forward declaration of SYCL kernels */
namespace kernels {
class copy;
class mul;
class add;
class triad;
}
template <class T>
SYCLStream<T>::SYCLStream(const unsigned int ARRAY_SIZE, const int device_index)
@ -24,14 +32,6 @@ SYCLStream<T>::SYCLStream(const unsigned int ARRAY_SIZE, const int device_index)
if (!cached)
getDeviceList();
// The array size must be divisible by WGSIZE
if (ARRAY_SIZE % WGSIZE != 0)
{
std::stringstream ss;
ss << "Array size must be a multiple of " << WGSIZE;
throw std::runtime_error(ss.str());
}
array_size = ARRAY_SIZE;
if (device_index >= devices.size())
@ -44,6 +44,14 @@ SYCLStream<T>::SYCLStream(const unsigned int ARRAY_SIZE, const int device_index)
queue = new cl::sycl::queue(dev);
/* Pre-build the kernels */
p = new program(queue->get_context());
p->build_from_kernel_name<kernels::copy>();
p->build_from_kernel_name<kernels::mul>();
p->build_from_kernel_name<kernels::add>();
p->build_from_kernel_name<kernels::triad>();
// Create buffers
d_a = new buffer<T>(array_size);
d_b = new buffer<T>(array_size);
@ -57,6 +65,7 @@ SYCLStream<T>::~SYCLStream()
delete d_b;
delete d_c;
delete p;
delete queue;
}
@ -67,9 +76,11 @@ void SYCLStream<T>::copy()
{
auto ka = d_a->template get_access<access::mode::read>(cgh);
auto kc = d_c->template get_access<access::mode::write>(cgh);
cgh.parallel_for<class copy>(nd_range<1>{array_size, WGSIZE}, [=](nd_item<1> item)
cgh.parallel_for<kernels::copy>(p->get_kernel<kernels::copy>(),
range<1>{array_size}, [=](item<1> item)
{
kc[item.get_global()] = ka[item.get_global()];
auto id = item.get();
kc[id[0]] = ka[id[0]];
});
});
queue->wait();
@ -83,9 +94,11 @@ void SYCLStream<T>::mul()
{
auto kb = d_b->template get_access<access::mode::write>(cgh);
auto kc = d_c->template get_access<access::mode::read>(cgh);
cgh.parallel_for<class mul>(nd_range<1>{array_size, WGSIZE}, [=](nd_item<1> item)
cgh.parallel_for<kernels::mul>(p->get_kernel<kernels::mul>(),
range<1>{array_size}, [=](item<1> item)
{
kb[item.get_global()] = scalar * kc[item.get_global()];
auto id = item.get();
kb[id[0]] = scalar * kc[id[0]];
});
});
queue->wait();
@ -99,9 +112,11 @@ void SYCLStream<T>::add()
auto ka = d_a->template get_access<access::mode::read>(cgh);
auto kb = d_b->template get_access<access::mode::read>(cgh);
auto kc = d_c->template get_access<access::mode::write>(cgh);
cgh.parallel_for<class add>(nd_range<1>{array_size, WGSIZE}, [=](nd_item<1> item)
cgh.parallel_for<kernels::add>(p->get_kernel<kernels::add>(),
range<1>{array_size}, [=](item<1> item)
{
kc[item.get_global()] = ka[item.get_global()] + kb[item.get_global()];
auto id = item.get();
kc[id[0]] = ka[id[0]] + kb[id[0]];
});
});
queue->wait();
@ -116,9 +131,11 @@ void SYCLStream<T>::triad()
auto ka = d_a->template get_access<access::mode::write>(cgh);
auto kb = d_b->template get_access<access::mode::read>(cgh);
auto kc = d_c->template get_access<access::mode::read>(cgh);
cgh.parallel_for<class triad>(nd_range<1>{array_size, WGSIZE}, [=](nd_item<1> item)
cgh.parallel_for<kernels::triad>(p->get_kernel<kernels::triad>(),
range<1>{array_size}, [=](item<1> item)
{
ka[item.get_global()] = kb[item.get_global()] + scalar * kc[item.get_global()];
auto id = item.get();
ka[id] = kb[id[0]] + scalar * kc[id[0]];
});
});
queue->wait();