mirror of
https://github.com/paboyle/Grid.git
synced 2025-06-19 00:07:05 +01:00
Merge branch 'feature/gpu-port' of https://github.com/paboyle/Grid into feature/gpu-port
Conflicts: Grid/stencil/Stencil.h
This commit is contained in:
Peter Boyle
@ -62,7 +62,7 @@ public:
|
||||
LatticeCoordinate(coor, nu + shift);
|
||||
ph = ph + twist[nu]*coor*((1./(in.Grid()->FullDimensions()[nu+shift])));
|
||||
}
|
||||
in_buf = exp((Real)(2.0*M_PI)*ci*ph*(-1.0))*in;
|
||||
in_buf = exp(Scalar(2.0*M_PI)*ci*ph*(-1.0))*in;
|
||||
|
||||
if(fiveD){//FFT only on temporal and spatial dimensions
|
||||
std::vector<int> mask(Nd+1,1); mask[0] = 0;
|
||||
@ -77,7 +77,7 @@ public:
|
||||
}
|
||||
|
||||
//phase for boundary condition
|
||||
out = out * exp((Real)(2.0*M_PI)*ci*ph);
|
||||
out = out * exp(Scalar(2.0*M_PI)*ci*ph);
|
||||
};
|
||||
|
||||
virtual void FreePropagator(const FermionField &in,FermionField &out,RealD mass,std::vector<double> twist) {
|
||||
|
||||
@ -323,10 +323,8 @@ public:
|
||||
this->HaloExchangeOptGather(source,compress);
|
||||
double t1=usecond();
|
||||
// Asynchronous MPI calls multidirectional, Isend etc...
|
||||
// this->CommunicateBegin(reqs);
|
||||
// this->CommunicateComplete(reqs);
|
||||
// Non-overlapped directions within a thread. Asynchronous calls except MPI3, threaded up to comm threads ways.
|
||||
this->Communicate();
|
||||
// this->Communicate();
|
||||
double t2=usecond(); timer1 += t2-t1;
|
||||
this->CommsMerge(compress);
|
||||
double t3=usecond(); timer2 += t3-t2;
|
||||
|
||||
@ -1 +0,0 @@
|
||||
../CayleyFermion5DInstantiation.cc.master
|
||||
@ -1 +0,0 @@
|
||||
../ContinuedFractionFermion5DInstantiation.cc.master
|
||||
@ -1 +0,0 @@
|
||||
../DomainWallEOFAFermionInstantiation.cc.master
|
||||
@ -1 +0,0 @@
|
||||
../MobiusEOFAFermionInstantiation.cc.master
|
||||
@ -1 +0,0 @@
|
||||
../PartialFractionFermion5DInstantiation.cc.master
|
||||
@ -1 +0,0 @@
|
||||
../WilsonCloverFermionInstantiation.cc.master
|
||||
@ -1 +0,0 @@
|
||||
../WilsonFermion5DInstantiation.cc.master
|
||||
@ -1 +0,0 @@
|
||||
../WilsonFermionInstantiation.cc.master
|
||||
@ -1 +0,0 @@
|
||||
../WilsonKernelsInstantiationGparity.cc.master
|
||||
@ -1 +0,0 @@
|
||||
../WilsonTMFermionInstantiation.cc.master
|
||||
@ -1 +0,0 @@
|
||||
../CayleyFermion5DInstantiation.cc.master
|
||||
@ -1 +0,0 @@
|
||||
../ContinuedFractionFermion5DInstantiation.cc.master
|
||||
@ -1 +0,0 @@
|
||||
../DomainWallEOFAFermionInstantiation.cc.master
|
||||
@ -1 +0,0 @@
|
||||
../MobiusEOFAFermionInstantiation.cc.master
|
||||
@ -1 +0,0 @@
|
||||
../PartialFractionFermion5DInstantiation.cc.master
|
||||
@ -1 +0,0 @@
|
||||
../WilsonCloverFermionInstantiation.cc.master
|
||||
@ -1 +0,0 @@
|
||||
../WilsonFermion5DInstantiation.cc.master
|
||||
@ -1 +0,0 @@
|
||||
../WilsonFermionInstantiation.cc.master
|
||||
@ -1 +0,0 @@
|
||||
../WilsonKernelsInstantiationGparity.cc.master
|
||||
@ -1 +0,0 @@
|
||||
../WilsonTMFermionInstantiation.cc.master
|
||||
@ -1 +0,0 @@
|
||||
../CayleyFermion5DInstantiation.cc.master
|
||||
@ -1 +0,0 @@
|
||||
../ContinuedFractionFermion5DInstantiation.cc.master
|
||||
@ -1 +0,0 @@
|
||||
../DomainWallEOFAFermionInstantiation.cc.master
|
||||
@ -1 +0,0 @@
|
||||
../MobiusEOFAFermionInstantiation.cc.master
|
||||
@ -1 +0,0 @@
|
||||
../PartialFractionFermion5DInstantiation.cc.master
|
||||
@ -1 +0,0 @@
|
||||
../WilsonCloverFermionInstantiation.cc.master
|
||||
@ -1 +0,0 @@
|
||||
../WilsonFermion5DInstantiation.cc.master
|
||||
@ -1 +0,0 @@
|
||||
../WilsonFermionInstantiation.cc.master
|
||||
@ -1 +0,0 @@
|
||||
../WilsonKernelsInstantiationGparity.cc.master
|
||||
@ -1 +0,0 @@
|
||||
../WilsonTMFermionInstantiation.cc.master
|
||||
@ -1 +0,0 @@
|
||||
../CayleyFermion5DInstantiation.cc.master
|
||||
@ -1 +0,0 @@
|
||||
../ContinuedFractionFermion5DInstantiation.cc.master
|
||||
@ -1 +0,0 @@
|
||||
../DomainWallEOFAFermionInstantiation.cc.master
|
||||
@ -1 +0,0 @@
|
||||
../MobiusEOFAFermionInstantiation.cc.master
|
||||
@ -1 +0,0 @@
|
||||
../PartialFractionFermion5DInstantiation.cc.master
|
||||
@ -1 +0,0 @@
|
||||
../WilsonCloverFermionInstantiation.cc.master
|
||||
@ -1 +0,0 @@
|
||||
../WilsonFermion5DInstantiation.cc.master
|
||||
@ -1 +0,0 @@
|
||||
../WilsonFermionInstantiation.cc.master
|
||||
@ -1 +0,0 @@
|
||||
../WilsonKernelsInstantiationGparity.cc.master
|
||||
@ -1 +0,0 @@
|
||||
../WilsonTMFermionInstantiation.cc.master
|
||||
@ -777,7 +777,7 @@ public:
|
||||
|
||||
int permute_slice=0;
|
||||
if(permute_dim){
|
||||
int wrap = sshift/rd;
|
||||
int wrap = sshift/rd; wrap=wrap % ly; // but it is local anyway
|
||||
int num = sshift%rd;
|
||||
if ( x< rd-num ) permute_slice=wrap;
|
||||
else permute_slice = (wrap+1)%ly;
|
||||
|
||||
@ -69,6 +69,11 @@ void coalescedWrite(vobj & __restrict__ vec,const vobj & __restrict__ extracted,
|
||||
// vstream(vec, extracted);
|
||||
vec = extracted;
|
||||
}
|
||||
template<class vobj> accelerator_inline
|
||||
void coalescedWriteNonTemporal(vobj & __restrict__ vec,const vobj & __restrict__ extracted,int lane=0)
|
||||
{
|
||||
vstream(vec, extracted);
|
||||
}
|
||||
#else
|
||||
accelerator_inline int SIMTlane(int Nsimd) { return threadIdx.y; } // CUDA specific
|
||||
|
||||
@ -92,6 +97,11 @@ void coalescedWrite(vobj & __restrict__ vec,const typename vobj::scalar_object &
|
||||
{
|
||||
insertLane(lane,vec,extracted);
|
||||
}
|
||||
template<class vobj> accelerator_inline
|
||||
void coalescedWriteNonTemporal(vobj & __restrict__ vec,const vobj & __restrict__ extracted,int lane=0)
|
||||
{
|
||||
insertLane(lane,vec,extracted);
|
||||
}
|
||||
#endif
|
||||
|
||||
|
||||
|
||||
Reference in New Issue
Block a user