portelli/Grid
1
0
Fork 0
mirror of https://github.com/paboyle/Grid.git synced 2025-08-22 22:17:10 +01:00
Code Releases Activity

First compile against SYCL

Browse Source
This commit is contained in:
Peter Boyle
2020-05-05 11:13:27 -07:00
parent 04927d2e40
commit 28a1fcaaff
24 changed files with 205 additions and 87 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
Grid/lattice/Lattice_base.h
View File

@@ -89,7 +89,7 @@ public:
// Rvalue
#ifdef __CUDA_ARCH__
#ifdef GRID_SIMT
accelerator_inline const typename vobj::scalar_object operator()(size_t i) const { return coalescedRead(this->_odata[i]); }
#else
accelerator_inline const vobj & operator()(size_t i) const { return this->_odata[i]; }
@@ -211,7 +211,7 @@ public:
LatticeView<vobj> accessor(*( (LatticeAccelerator<vobj> *) this));
return accessor;
}
~Lattice() {
if ( this->_odata_size ) {
dealloc();

8
Grid/lattice/Lattice_reduction.h
View File

@@ -24,7 +24,7 @@ Author: paboyle <paboyle@ph.ed.ac.uk>
#include <Grid/Grid_Eigen_Dense.h>
#ifdef GRID_NVCC
#ifdef GRID_CUDA
#include <Grid/lattice/Lattice_reduction_gpu.h>
#endif
@@ -67,7 +67,7 @@ inline typename vobj::scalar_object sum_cpu(const vobj *arg, Integer osites)
template<class vobj>
inline typename vobj::scalar_object sum(const vobj *arg, Integer osites)
{
#ifdef GRID_NVCC
#ifdef GRID_CUDA
return sum_gpu(arg,osites);
#else
return sum_cpu(arg,osites);
@@ -108,7 +108,7 @@ inline ComplexD innerProduct(const Lattice<vobj> &left,const Lattice<vobj> &righ
const uint64_t nsimd = grid->Nsimd();
const uint64_t sites = grid->oSites();
#ifdef GRID_NVCC
#ifdef GRID_CUDA
// GPU - SIMT lane compliance...
typedef decltype(innerProduct(left_v[0],right_v[0])) inner_t;
Vector<inner_t> inner_tmp(sites);
@@ -174,7 +174,7 @@ axpby_norm_fast(Lattice<vobj> &z,sobj a,sobj b,const Lattice<vobj> &x,const Latt
const uint64_t nsimd = grid->Nsimd();
const uint64_t sites = grid->oSites();
#ifdef GRID_NVCC
#ifdef GRID_CUDA
// GPU
typedef decltype(innerProduct(x_v[0],y_v[0])) inner_t;
Vector<inner_t> inner_tmp(sites);