devtools/eigen
1
0
Fork 0
mirror of https://gitlab.com/libeigen/eigen.git synced 2026-04-10 11:34:33 +08:00
Code Issues Packages Projects Releases Wiki Activity

Adding asynchronous execution as it improves the performance.

Browse Source
This commit is contained in:
Mehdi Goli
2016-12-14 17:38:53 +00:00
parent 2d4a091beb
commit 730eb9fe1c
5 changed files with 12 additions and 10 deletions
Download Patch File Download Diff File Expand all files Collapse all files

10
unsupported/Eigen/CXX11/src/Tensor/TensorDeviceSycl.h
View File

@@ -214,7 +214,7 @@ struct SyclDevice {
auto dst_acc =it2->second.template get_access<cl::sycl::access::mode::discard_write, cl::sycl::access::target::global_buffer>(cgh);
cgh.parallel_for(cl::sycl::nd_range<1>(cl::sycl::range<1>(GRange), cl::sycl::range<1>(tileSize)), MemCopyFunctor<T>(src_acc, dst_acc, rng, 0, offset));
});
synchronize();
asynchronousExec();
}
/// The memcpyHostToDevice is used to copy the device only pointer to a host pointer. Using the device
@@ -245,7 +245,7 @@ struct SyclDevice {
auto dst_acc =dest_buf.template get_access<cl::sycl::access::mode::discard_write, cl::sycl::access::target::global_buffer>(cgh);
cgh.parallel_for( cl::sycl::nd_range<1>(cl::sycl::range<1>(GRange), cl::sycl::range<1>(tileSize)), MemCopyFunctor<T>(src_acc, dst_acc, rng, 0, 0));
});
synchronize();
asynchronousExec();
}
/// returning the sycl queue
EIGEN_STRONG_INLINE cl::sycl::queue& sycl_queue() const { return m_queue_stream->m_queue;}
@@ -263,7 +263,7 @@ struct SyclDevice {
}
});
});
synchronize();
asynchronousExec();
}
EIGEN_STRONG_INLINE size_t firstLevelCacheSize() const {
@@ -282,6 +282,10 @@ struct SyclDevice {
EIGEN_STRONG_INLINE void synchronize() const {
sycl_queue().wait_and_throw(); //pass
}
EIGEN_STRONG_INLINE void asynchronousExec() const {
sycl_queue().throw_asynchronous();//pass
}
// This function checks if the runtime recorded an error for the
// underlying stream device.
EIGEN_STRONG_INLINE bool ok() const {