portelli/Grid
1
0
Fork 0
mirror of https://github.com/paboyle/Grid.git synced 2025-06-18 07:47:06 +01:00
Code Releases Activity

Global edit with change to View usage. autoView() creates a wrapper object that closes the view when scope closes.

Browse Source
This commit is contained in:
Peter Boyle
2020-06-05 18:52:35 -04:00
parent f39c2a240b
commit 1a4c8c3387
78 changed files with 773 additions and 778 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
Grid/lattice/Lattice_ET.h
View File

@ -87,12 +87,14 @@ sobj eval(const uint64_t ss, const sobj &arg)
}
template <class lobj> accelerator_inline
const lobj & eval(const uint64_t ss, const LatticeExprView<lobj> &arg)
const lobj & eval(const uint64_t ss, const LatticeView<lobj> &arg)
{
return arg[ss];
}
// What needs this?
// Cannot be legal on accelerator
// Comparison must convert
#if 1
template <class lobj> accelerator_inline
const lobj & eval(const uint64_t ss, const Lattice<lobj> &arg)

68
Grid/lattice/Lattice_arith.h
View File

@ -36,9 +36,9 @@ NAMESPACE_BEGIN(Grid);
template<class obj1,class obj2,class obj3> inline
void mult(Lattice<obj1> &ret,const Lattice<obj2> &lhs,const Lattice<obj3> &rhs){
ret.Checkerboard() = lhs.Checkerboard();
auto ret_v = ret.View(AcceleratorWrite);
auto lhs_v = lhs.View(AcceleratorRead);
auto rhs_v = rhs.View(AcceleratorRead);
autoView( ret_v , ret, AcceleratorWrite);
autoView( lhs_v , lhs, AcceleratorRead);
autoView( rhs_v , rhs, AcceleratorRead);
conformable(ret,rhs);
conformable(lhs,rhs);
accelerator_for(ss,lhs_v.size(),obj1::Nsimd(),{
@ -55,9 +55,9 @@ void mac(Lattice<obj1> &ret,const Lattice<obj2> &lhs,const Lattice<obj3> &rhs){
ret.Checkerboard() = lhs.Checkerboard();
conformable(ret,rhs);
conformable(lhs,rhs);
auto ret_v = ret.View(AcceleratorWrite);
auto lhs_v = lhs.View(AcceleratorRead);
auto rhs_v = rhs.View(AcceleratorRead);
autoView( ret_v , ret, AcceleratorWrite);
autoView( lhs_v , lhs, AcceleratorRead);
autoView( rhs_v , rhs, AcceleratorRead);
accelerator_for(ss,lhs_v.size(),obj1::Nsimd(),{
decltype(coalescedRead(obj1())) tmp;
auto lhs_t=lhs_v(ss);
@ -72,9 +72,9 @@ void sub(Lattice<obj1> &ret,const Lattice<obj2> &lhs,const Lattice<obj3> &rhs){
ret.Checkerboard() = lhs.Checkerboard();
conformable(ret,rhs);
conformable(lhs,rhs);
auto ret_v = ret.View(AcceleratorWrite);
auto lhs_v = lhs.View(AcceleratorRead);
auto rhs_v = rhs.View(AcceleratorRead);
autoView( ret_v , ret, AcceleratorWrite);
autoView( lhs_v , lhs, AcceleratorRead);
autoView( rhs_v , rhs, AcceleratorRead);
accelerator_for(ss,lhs_v.size(),obj1::Nsimd(),{
decltype(coalescedRead(obj1())) tmp;
auto lhs_t=lhs_v(ss);
@ -88,9 +88,9 @@ void add(Lattice<obj1> &ret,const Lattice<obj2> &lhs,const Lattice<obj3> &rhs){
ret.Checkerboard() = lhs.Checkerboard();
conformable(ret,rhs);
conformable(lhs,rhs);
auto ret_v = ret.View(AcceleratorWrite);
auto lhs_v = lhs.View(AcceleratorRead);
auto rhs_v = rhs.View(AcceleratorRead);
autoView( ret_v , ret, AcceleratorWrite);
autoView( lhs_v , lhs, AcceleratorRead);
autoView( rhs_v , rhs, AcceleratorRead);
accelerator_for(ss,lhs_v.size(),obj1::Nsimd(),{
decltype(coalescedRead(obj1())) tmp;
auto lhs_t=lhs_v(ss);
@ -107,8 +107,8 @@ template<class obj1,class obj2,class obj3> inline
void mult(Lattice<obj1> &ret,const Lattice<obj2> &lhs,const obj3 &rhs){
ret.Checkerboard() = lhs.Checkerboard();
conformable(lhs,ret);
auto ret_v = ret.View(AcceleratorWrite);
auto lhs_v = lhs.View(AcceleratorRead);
autoView( ret_v , ret, AcceleratorWrite);
autoView( lhs_v , lhs, AcceleratorRead);
accelerator_for(ss,lhs_v.size(),obj1::Nsimd(),{
decltype(coalescedRead(obj1())) tmp;
mult(&tmp,&lhs_v(ss),&rhs);
@ -120,8 +120,8 @@ template<class obj1,class obj2,class obj3> inline
void mac(Lattice<obj1> &ret,const Lattice<obj2> &lhs,const obj3 &rhs){
ret.Checkerboard() = lhs.Checkerboard();
conformable(ret,lhs);
auto ret_v = ret.View(AcceleratorWrite);
auto lhs_v = lhs.View(AcceleratorRead);
autoView( ret_v , ret, AcceleratorWrite);
autoView( lhs_v , lhs, AcceleratorRead);
accelerator_for(ss,lhs_v.size(),obj1::Nsimd(),{
decltype(coalescedRead(obj1())) tmp;
auto lhs_t=lhs_v(ss);
@ -134,8 +134,8 @@ template<class obj1,class obj2,class obj3> inline
void sub(Lattice<obj1> &ret,const Lattice<obj2> &lhs,const obj3 &rhs){
ret.Checkerboard() = lhs.Checkerboard();
conformable(ret,lhs);
auto ret_v = ret.View(AcceleratorWrite);
auto lhs_v = lhs.View(AcceleratorRead);
autoView( ret_v , ret, AcceleratorWrite);
autoView( lhs_v , lhs, AcceleratorRead);
accelerator_for(ss,lhs_v.size(),obj1::Nsimd(),{
decltype(coalescedRead(obj1())) tmp;
auto lhs_t=lhs_v(ss);
@ -147,8 +147,8 @@ template<class obj1,class obj2,class obj3> inline
void add(Lattice<obj1> &ret,const Lattice<obj2> &lhs,const obj3 &rhs){
ret.Checkerboard() = lhs.Checkerboard();
conformable(lhs,ret);
auto ret_v = ret.View(AcceleratorWrite);
auto lhs_v = lhs.View(AcceleratorRead);
autoView( ret_v , ret, AcceleratorWrite);
autoView( lhs_v , lhs, AcceleratorRead);
accelerator_for(ss,lhs_v.size(),obj1::Nsimd(),{
decltype(coalescedRead(obj1())) tmp;
auto lhs_t=lhs_v(ss);
@ -164,8 +164,8 @@ template<class obj1,class obj2,class obj3> inline
void mult(Lattice<obj1> &ret,const obj2 &lhs,const Lattice<obj3> &rhs){
ret.Checkerboard() = rhs.Checkerboard();
conformable(ret,rhs);
auto ret_v = ret.View(AcceleratorWrite);
auto rhs_v = lhs.View(AcceleratorRead);
autoView( ret_v , ret, AcceleratorWrite);
autoView( rhs_v , lhs, AcceleratorRead);
accelerator_for(ss,rhs_v.size(),obj1::Nsimd(),{
decltype(coalescedRead(obj1())) tmp;
auto rhs_t=rhs_v(ss);
@ -178,8 +178,8 @@ template<class obj1,class obj2,class obj3> inline
void mac(Lattice<obj1> &ret,const obj2 &lhs,const Lattice<obj3> &rhs){
ret.Checkerboard() = rhs.Checkerboard();
conformable(ret,rhs);
auto ret_v = ret.View(AcceleratorWrite);
auto rhs_v = lhs.View(AcceleratorRead);
autoView( ret_v , ret, AcceleratorWrite);
autoView( rhs_v , lhs, AcceleratorRead);
accelerator_for(ss,rhs_v.size(),obj1::Nsimd(),{
decltype(coalescedRead(obj1())) tmp;
auto rhs_t=rhs_v(ss);
@ -192,8 +192,8 @@ template<class obj1,class obj2,class obj3> inline
void sub(Lattice<obj1> &ret,const obj2 &lhs,const Lattice<obj3> &rhs){
ret.Checkerboard() = rhs.Checkerboard();
conformable(ret,rhs);
auto ret_v = ret.View(AcceleratorWrite);
auto rhs_v = lhs.View(AcceleratorRead);
autoView( ret_v , ret, AcceleratorWrite);
autoView( rhs_v , lhs, AcceleratorRead);
accelerator_for(ss,rhs_v.size(),obj1::Nsimd(),{
decltype(coalescedRead(obj1())) tmp;
auto rhs_t=rhs_v(ss);
@ -205,8 +205,8 @@ template<class obj1,class obj2,class obj3> inline
void add(Lattice<obj1> &ret,const obj2 &lhs,const Lattice<obj3> &rhs){
ret.Checkerboard() = rhs.Checkerboard();
conformable(ret,rhs);
auto ret_v = ret.View(AcceleratorWrite);
auto rhs_v = lhs.View(AcceleratorRead);
autoView( ret_v , ret, AcceleratorWrite);
autoView( rhs_v , lhs, AcceleratorRead);
accelerator_for(ss,rhs_v.size(),obj1::Nsimd(),{
decltype(coalescedRead(obj1())) tmp;
auto rhs_t=rhs_v(ss);
@ -220,9 +220,9 @@ void axpy(Lattice<vobj> &ret,sobj a,const Lattice<vobj> &x,const Lattice<vobj> &
ret.Checkerboard() = x.Checkerboard();
conformable(ret,x);
conformable(x,y);
auto ret_v = ret.View(AcceleratorWrite);
auto x_v = x.View(AcceleratorRead);
auto y_v = y.View(AcceleratorRead);
autoView( ret_v , ret, AcceleratorWrite);
autoView( x_v , x, AcceleratorRead);
autoView( y_v , y, AcceleratorRead);
accelerator_for(ss,x_v.size(),vobj::Nsimd(),{
auto tmp = a*x_v(ss)+y_v(ss);
coalescedWrite(ret_v[ss],tmp);
@ -233,9 +233,9 @@ void axpby(Lattice<vobj> &ret,sobj a,sobj b,const Lattice<vobj> &x,const Lattice
ret.Checkerboard() = x.Checkerboard();
conformable(ret,x);
conformable(x,y);
auto ret_v = ret.View(AcceleratorWrite);
auto x_v = x.View(AcceleratorRead);
auto y_v = y.View(AcceleratorRead);
autoView( ret_v , ret, AcceleratorWrite);
autoView( x_v , x, AcceleratorRead);
autoView( y_v , y, AcceleratorRead);
accelerator_for(ss,x_v.size(),vobj::Nsimd(),{
auto tmp = a*x_v(ss)+b*y_v(ss);
coalescedWrite(ret_v[ss],tmp);

7
Grid/lattice/Lattice_base.h
View File

@ -84,6 +84,7 @@ public:
/////////////////////////////////////////////////////////////////////////////////
void SetViewMode(ViewMode mode) {
LatticeView<vobj> accessor(*( (LatticeAccelerator<vobj> *) this),mode);
accessor.ViewClose();
}
/////////////////////////////////////////////////////////////////////////////////
// Return a view object that may be dereferenced in site loops.
@ -123,6 +124,7 @@ public:
auto tmp = eval(ss,exprCopy);
vstream(me[ss],tmp);
});
me.ViewClose();
ExpressionViewClose(exprCopy);
return *this;
}
@ -145,6 +147,7 @@ public:
auto tmp = eval(ss,exprCopy);
vstream(me[ss],tmp);
});
me.ViewClose();
ExpressionViewClose(exprCopy);
return *this;
}
@ -166,6 +169,7 @@ public:
auto tmp = eval(ss,exprCopy);
vstream(me[ss],tmp);
});
me.ViewClose();
ExpressionViewClose(exprCopy);
return *this;
}
@ -221,6 +225,7 @@ public:
thread_for(ss,me.size(),{
me[ss]= r;
});
me.ViewClose();
return *this;
}
@ -278,6 +283,7 @@ public:
accelerator_for(ss,me.size(),vobj::Nsimd(),{
coalescedWrite(me[ss],him(ss));
});
me.ViewClose(); him.ViewClose();
return *this;
}
@ -292,6 +298,7 @@ public:
accelerator_for(ss,me.size(),vobj::Nsimd(),{
coalescedWrite(me[ss],him(ss));
});
me.ViewClose(); him.ViewClose();
return *this;
}
///////////////////////////////////////////

14
Grid/lattice/Lattice_comparison.h
View File

@ -78,9 +78,9 @@ template<class vfunctor,class lobj,class robj>
inline Lattice<vPredicate> LLComparison(vfunctor op,const Lattice<lobj> &lhs,const Lattice<robj> &rhs)
{
Lattice<vPredicate> ret(rhs.Grid());
auto lhs_v = lhs.View(CpuRead);
auto rhs_v = rhs.View(CpuRead);
auto ret_v = ret.View(CpuWrite);
autoView( lhs_v, lhs, CpuRead);
autoView( rhs_v, rhs, CpuRead);
autoView( ret_v, ret, CpuWrite);
thread_for( ss, rhs_v.size(), {
ret_v[ss]=op(lhs_v[ss],rhs_v[ss]);
});
@ -93,8 +93,8 @@ template<class vfunctor,class lobj,class robj>
inline Lattice<vPredicate> LSComparison(vfunctor op,const Lattice<lobj> &lhs,const robj &rhs)
{
Lattice<vPredicate> ret(lhs.Grid());
auto lhs_v = lhs.View(CpuRead);
auto ret_v = ret.View(CpuWrite);
autoView( lhs_v, lhs, CpuRead);
autoView( ret_v, ret, CpuWrite);
thread_for( ss, lhs_v.size(), {
ret_v[ss]=op(lhs_v[ss],rhs);
});
@ -107,8 +107,8 @@ template<class vfunctor,class lobj,class robj>
inline Lattice<vPredicate> SLComparison(vfunctor op,const lobj &lhs,const Lattice<robj> &rhs)
{
Lattice<vPredicate> ret(rhs.Grid());
auto rhs_v = rhs.View(CpuRead);
auto ret_v = ret.View(CpuWrite);
autoView( rhs_v, rhs, CpuRead);
autoView( ret_v, ret, CpuWrite);
thread_for( ss, rhs_v.size(), {
ret_v[ss]=op(lhs,rhs_v[ss]);
});

2
Grid/lattice/Lattice_coordinate.h
View File

@ -37,7 +37,7 @@ template<class iobj> inline void LatticeCoordinate(Lattice<iobj> &l,int mu)
GridBase *grid = l.Grid();
int Nsimd = grid->iSites();
auto l_v = l.View(CpuWrite);
autoView(l_v, l, CpuWrite);
thread_for( o, grid->oSites(), {
vector_type vI;
Coordinate gcoor;

16
Grid/lattice/Lattice_local.h
View File

@ -43,8 +43,8 @@ template<class vobj>
inline auto localNorm2 (const Lattice<vobj> &rhs)-> Lattice<typename vobj::tensor_reduced>
{
Lattice<typename vobj::tensor_reduced> ret(rhs.Grid());
auto rhs_v = rhs.View(AcceleratorRead);
auto ret_v = ret.View(AcceleratorWrite);
autoView( rhs_v , rhs, AcceleratorRead);
autoView( ret_v , ret, AcceleratorWrite);
accelerator_for(ss,rhs_v.size(),vobj::Nsimd(),{
coalescedWrite(ret_v[ss],innerProduct(rhs_v(ss),rhs_v(ss)));
});
@ -56,9 +56,9 @@ template<class vobj>
inline auto localInnerProduct (const Lattice<vobj> &lhs,const Lattice<vobj> &rhs) -> Lattice<typename vobj::tensor_reduced>
{
Lattice<typename vobj::tensor_reduced> ret(rhs.Grid());
auto lhs_v = lhs.View(AcceleratorRead);
auto rhs_v = rhs.View(AcceleratorRead);
auto ret_v = ret.View(AcceleratorWrite);
autoView( lhs_v , lhs, AcceleratorRead);
autoView( rhs_v , rhs, AcceleratorRead);
autoView( ret_v , ret, AcceleratorWrite);
accelerator_for(ss,rhs_v.size(),vobj::Nsimd(),{
coalescedWrite(ret_v[ss],innerProduct(lhs_v(ss),rhs_v(ss)));
});
@ -73,9 +73,9 @@ inline auto outerProduct (const Lattice<ll> &lhs,const Lattice<rr> &rhs) -> Latt
typedef decltype(coalescedRead(ll())) sll;
typedef decltype(coalescedRead(rr())) srr;
Lattice<decltype(outerProduct(ll(),rr()))> ret(rhs.Grid());
auto lhs_v = lhs.View(AcceleratorRead);
auto rhs_v = rhs.View(AcceleratorRead);
auto ret_v = ret.View(AcceleratorWrite);
autoView( lhs_v , lhs, AcceleratorRead);
autoView( rhs_v , rhs, AcceleratorRead);
autoView( ret_v , ret, AcceleratorWrite);
accelerator_for(ss,rhs_v.size(),1,{
// FIXME had issues with scalar version of outer
// Use vector [] operator and don't read coalesce this loop

14
Grid/lattice/Lattice_matrix_reduction.h
View File

@ -51,9 +51,9 @@ static void sliceMaddMatrix (Lattice<vobj> &R,Eigen::MatrixXcd &aa,const Lattice
int block =FullGrid->_slice_block [Orthog];
int nblock=FullGrid->_slice_nblock[Orthog];
int ostride=FullGrid->_ostride[Orthog];
auto X_v = X.View(CpuRead);
auto Y_v = Y.View(CpuRead);
auto R_v = R.View(CpuWrite);
autoView( X_v , X, CpuRead);
autoView( Y_v , Y, CpuRead);
autoView( R_v , R, CpuWrite);
thread_region
{
std::vector<vobj> s_x(Nblock);
@ -97,8 +97,8 @@ static void sliceMulMatrix (Lattice<vobj> &R,Eigen::MatrixXcd &aa,const Lattice<
int nblock=FullGrid->_slice_nblock[Orthog];
int ostride=FullGrid->_ostride[Orthog];
auto X_v = X.View(CpuRead);
auto R_v = R.View(CpuWrite);
autoView( X_v , X, CpuRead);
autoView( R_v , R, CpuWrite);
thread_region
{
@ -156,8 +156,8 @@ static void sliceInnerProductMatrix( Eigen::MatrixXcd &mat, const Lattice<vobj>
int ostride=FullGrid->_ostride[Orthog];
typedef typename vobj::vector_typeD vector_typeD;
auto lhs_v = lhs.View(CpuRead);
auto rhs_v = rhs.View(CpuRead);
autoView( lhs_v , lhs, CpuRead);
autoView( rhs_v , rhs, CpuRead);
thread_region {
std::vector<vobj> Left(Nblock);
std::vector<vobj> Right(Nblock);

24
Grid/lattice/Lattice_peekpoke.h
View File

@ -46,8 +46,8 @@ auto PeekIndex(const Lattice<vobj> &lhs,int i) -> Lattice<decltype(peekIndex<Ind
{
Lattice<decltype(peekIndex<Index>(vobj(),i))> ret(lhs.Grid());
ret.Checkerboard()=lhs.Checkerboard();
auto ret_v = ret.View(CpuWrite);
auto lhs_v = lhs.View(CpuRead);
autoView( ret_v, ret, CpuWrite);
autoView( lhs_v, lhs, CpuRead);
thread_for( ss, lhs_v.size(), {
ret_v[ss] = peekIndex<Index>(lhs_v[ss],i);
});
@ -58,8 +58,8 @@ auto PeekIndex(const Lattice<vobj> &lhs,int i,int j) -> Lattice<decltype(peekInd
{
Lattice<decltype(peekIndex<Index>(vobj(),i,j))> ret(lhs.Grid());
ret.Checkerboard()=lhs.Checkerboard();
auto ret_v = ret.View(CpuWrite);
auto lhs_v = lhs.View(CpuRead);
autoView( ret_v, ret, CpuWrite);
autoView( lhs_v, lhs, CpuRead);
thread_for( ss, lhs_v.size(), {
ret_v[ss] = peekIndex<Index>(lhs_v[ss],i,j);
});
@ -72,8 +72,8 @@ auto PeekIndex(const Lattice<vobj> &lhs,int i,int j) -> Lattice<decltype(peekInd
template<int Index,class vobj>
void PokeIndex(Lattice<vobj> &lhs,const Lattice<decltype(peekIndex<Index>(vobj(),0))> & rhs,int i)
{
auto rhs_v = rhs.View(CpuRead);
auto lhs_v = lhs.View(CpuWrite);
autoView( rhs_v, rhs, CpuRead);
autoView( lhs_v, lhs, CpuWrite);
thread_for( ss, lhs_v.size(), {
pokeIndex<Index>(lhs_v[ss],rhs_v[ss],i);
});
@ -81,8 +81,8 @@ void PokeIndex(Lattice<vobj> &lhs,const Lattice<decltype(peekIndex<Index>(vobj()
template<int Index,class vobj>
void PokeIndex(Lattice<vobj> &lhs,const Lattice<decltype(peekIndex<Index>(vobj(),0,0))> & rhs,int i,int j)
{
auto rhs_v = rhs.View(CpuRead);
auto lhs_v = lhs.View(CpuWrite);
autoView( rhs_v, rhs, CpuRead);
autoView( lhs_v, lhs, CpuWrite);
thread_for( ss, lhs_v.size(), {
pokeIndex<Index>(lhs_v[ss],rhs_v[ss],i,j);
});
@ -111,7 +111,7 @@ void pokeSite(const sobj &s,Lattice<vobj> &l,const Coordinate &site){
// extract-modify-merge cycle is easiest way and this is not perf critical
ExtractBuffer<sobj> buf(Nsimd);
auto l_v = l.View(CpuWrite);
autoView( l_v , l, CpuWrite);
if ( rank == grid->ThisRank() ) {
extract(l_v[odx],buf);
buf[idx] = s;
@ -141,7 +141,7 @@ void peekSite(sobj &s,const Lattice<vobj> &l,const Coordinate &site){
grid->GlobalCoorToRankIndex(rank,odx,idx,site);
ExtractBuffer<sobj> buf(Nsimd);
auto l_v = l.View(CpuWrite);
autoView( l_v , l, CpuWrite);
extract(l_v[odx],buf);
s = buf[idx];
@ -173,7 +173,7 @@ inline void peekLocalSite(sobj &s,const Lattice<vobj> &l,Coordinate &site){
idx= grid->iIndex(site);
odx= grid->oIndex(site);
auto l_v = l.View(CpuRead);
autoView( l_v , l, CpuRead);
scalar_type * vp = (scalar_type *)&l_v[odx];
scalar_type * pt = (scalar_type *)&s;
@ -202,7 +202,7 @@ inline void pokeLocalSite(const sobj &s,Lattice<vobj> &l,Coordinate &site){
idx= grid->iIndex(site);
odx= grid->oIndex(site);
auto l_v = l.View(CpuWrite);
autoView( l_v , l, CpuWrite);
scalar_type * vp = (scalar_type *)&l_v[odx];
scalar_type * pt = (scalar_type *)&s;
for(int w=0;w<words;w++){

8
Grid/lattice/Lattice_reality.h
View File

@ -40,8 +40,8 @@ NAMESPACE_BEGIN(Grid);
template<class vobj> inline Lattice<vobj> adj(const Lattice<vobj> &lhs){
Lattice<vobj> ret(lhs.Grid());
auto lhs_v = lhs.View(AcceleratorRead);
auto ret_v = ret.View(AcceleratorWrite);
autoView( lhs_v, lhs, AcceleratorRead);
autoView( ret_v, ret, AcceleratorWrite);
accelerator_for( ss, lhs_v.size(), vobj::Nsimd(), {
coalescedWrite(ret_v[ss], adj(lhs_v(ss)));
});
@ -50,8 +50,8 @@ template<class vobj> inline Lattice<vobj> adj(const Lattice<vobj> &lhs){
template<class vobj> inline Lattice<vobj> conjugate(const Lattice<vobj> &lhs){
Lattice<vobj> ret(lhs.Grid());
auto lhs_v = lhs.View(AcceleratorRead);
auto ret_v = ret.View(AcceleratorWrite);
autoView( lhs_v, lhs, AcceleratorRead);
autoView( ret_v, ret, AcceleratorWrite);
accelerator_for( ss, lhs_v.size(), vobj::Nsimd(), {
coalescedWrite( ret_v[ss] , conjugate(lhs_v(ss)));
});

40
Grid/lattice/Lattice_reduction.h
View File

@ -79,11 +79,11 @@ template<class vobj>
inline typename vobj::scalar_object sum(const Lattice<vobj> &arg)
{
#if defined(GRID_CUDA)||defined(GRID_HIP)
auto arg_v = arg.View(AcceleratorRead);
autoView( arg_v, arg, AcceleratorRead);
Integer osites = arg.Grid()->oSites();
auto ssum= sum_gpu(&arg_v[0],osites);
#else
auto arg_v = arg.View(CpuRead);
autoView(arg_v, arg, CpuRead);
Integer osites = arg.Grid()->oSites();
auto ssum= sum_cpu(&arg_v[0],osites);
#endif
@ -113,8 +113,8 @@ inline ComplexD innerProduct(const Lattice<vobj> &left,const Lattice<vobj> &righ
const uint64_t sites = grid->oSites();
// Might make all code paths go this way.
auto left_v = left.View(AcceleratorRead);
auto right_v=right.View(AcceleratorRead);
autoView( left_v , left, AcceleratorRead);
autoView( right_v,right, AcceleratorRead);
// GPU - SIMT lane compliance...
typedef decltype(innerProduct(left_v[0],right_v[0])) inner_t;
@ -168,9 +168,9 @@ axpby_norm_fast(Lattice<vobj> &z,sobj a,sobj b,const Lattice<vobj> &x,const Latt
const uint64_t sites = grid->oSites();
// GPU
auto x_v=x.View(AcceleratorRead);
auto y_v=y.View(AcceleratorRead);
auto z_v=z.View(AcceleratorWrite);
autoView( x_v, x, AcceleratorRead);
autoView( y_v, y, AcceleratorRead);
autoView( z_v, z, AcceleratorWrite);
typedef decltype(innerProduct(x_v[0],y_v[0])) inner_t;
Vector<inner_t> inner_tmp(sites);
@ -257,7 +257,7 @@ template<class vobj> inline void sliceSum(const Lattice<vobj> &Data,std::vector<
// sum over reduced dimension planes, breaking out orthog dir
// Parallel over orthog direction
auto Data_v=Data.View(CpuRead);
autoView( Data_v, Data, CpuRead);
thread_for( r,rd, {
int so=r*grid->_ostride[orthogdim]; // base offset for start of plane
for(int n=0;n<e1;n++){
@ -335,8 +335,8 @@ static void sliceInnerProductVector( std::vector<ComplexD> & result, const Latti
int e2= grid->_slice_block [orthogdim];
int stride=grid->_slice_stride[orthogdim];
auto lhv=lhs.View(CpuRead);
auto rhv=rhs.View(CpuRead);
autoView( lhv, lhs, CpuRead);
autoView( rhv, rhs, CpuRead);
thread_for( r,rd,{
int so=r*grid->_ostride[orthogdim]; // base offset for start of plane
@ -443,9 +443,9 @@ static void sliceMaddVector(Lattice<vobj> &R,std::vector<RealD> &a,const Lattice
tensor_reduced at; at=av;
auto Rv=R.View(CpuWrite);
auto Xv=X.View(CpuRead);
auto Yv=Y.View(CpuRead);
autoView( Rv, R, CpuWrite);
autoView( Xv, X, CpuRead);
autoView( Yv, Y, CpuRead);
thread_for2d( n, e1, b,e2, {
int ss= so+n*stride+b;
Rv[ss] = at*Xv[ss]+Yv[ss];
@ -501,9 +501,9 @@ static void sliceMaddMatrix (Lattice<vobj> &R,Eigen::MatrixXcd &aa,const Lattice
int nblock=FullGrid->_slice_nblock[Orthog];
int ostride=FullGrid->_ostride[Orthog];
auto X_v=X.View(CpuRead);
auto Y_v=Y.View(CpuRead);
auto R_v=R.View(CpuWrite);
autoView( X_v, X, CpuRead);
autoView( Y_v, Y, CpuRead);
autoView( R_v, R, CpuWrite);
thread_region
{
Vector<vobj> s_x(Nblock);
@ -554,8 +554,8 @@ static void sliceMulMatrix (Lattice<vobj> &R,Eigen::MatrixXcd &aa,const Lattice<
int block =FullGrid->_slice_block [Orthog];
int nblock=FullGrid->_slice_nblock[Orthog];
int ostride=FullGrid->_ostride[Orthog];
auto R_v = R.View(CpuWrite);
auto X_v = X.View(CpuRead);
autoView( R_v, R, CpuWrite);
autoView( X_v, X, CpuRead);
thread_region
{
std::vector<vobj> s_x(Nblock);
@ -613,8 +613,8 @@ static void sliceInnerProductMatrix( Eigen::MatrixXcd &mat, const Lattice<vobj>
typedef typename vobj::vector_typeD vector_typeD;
auto lhs_v=lhs.View(CpuRead);
auto rhs_v=rhs.View(CpuRead);
autoView( lhs_v, lhs, CpuRead);
autoView( rhs_v, rhs, CpuRead);
thread_region
{
std::vector<vobj> Left(Nblock);

2
Grid/lattice/Lattice_rng.h
View File

@ -375,7 +375,7 @@ public:
int osites = _grid->oSites(); // guaranteed to be <= l.Grid()->oSites() by a factor multiplicity
int words = sizeof(scalar_object) / sizeof(scalar_type);
auto l_v = l.View(CpuWrite);
autoView(l_v, l, CpuWrite);
thread_for( ss, osites, {
ExtractBuffer<scalar_object> buf(Nsimd);
for (int m = 0; m < multiplicity; m++) { // Draw from same generator multiplicity times

8
Grid/lattice/Lattice_trace.h
View File

@ -41,8 +41,8 @@ template<class vobj>
inline auto trace(const Lattice<vobj> &lhs) -> Lattice<decltype(trace(vobj()))>
{
Lattice<decltype(trace(vobj()))> ret(lhs.Grid());
auto ret_v = ret.View(AcceleratorWrite);
auto lhs_v = lhs.View(AcceleratorRead);
autoView(ret_v , ret, AcceleratorWrite);
autoView(lhs_v , lhs, AcceleratorRead);
accelerator_for( ss, lhs_v.size(), vobj::Nsimd(), {
coalescedWrite(ret_v[ss], trace(lhs_v(ss)));
});
@ -56,8 +56,8 @@ template<int Index,class vobj>
inline auto TraceIndex(const Lattice<vobj> &lhs) -> Lattice<decltype(traceIndex<Index>(vobj()))>
{
Lattice<decltype(traceIndex<Index>(vobj()))> ret(lhs.Grid());
auto ret_v = ret.View(AcceleratorWrite);
auto lhs_v = lhs.View(AcceleratorRead);
autoView( ret_v , ret, AcceleratorWrite);
autoView( lhs_v , lhs, AcceleratorRead);
accelerator_for( ss, lhs_v.size(), vobj::Nsimd(), {
coalescedWrite(ret_v[ss], traceIndex<Index>(lhs_v(ss)));
});

120
Grid/lattice/Lattice_transfer.h
View File

@ -46,11 +46,12 @@ inline void subdivides(GridBase *coarse,GridBase *fine)
////////////////////////////////////////////////////////////////////////////////////////////
// remove and insert a half checkerboard
////////////////////////////////////////////////////////////////////////////////////////////
template<class vobj> inline void pickCheckerboard(int cb,Lattice<vobj> &half,const Lattice<vobj> &full){
template<class vobj> inline void pickCheckerboard(int cb,Lattice<vobj> &half,const Lattice<vobj> &full)
{
half.Checkerboard() = cb;
auto half_v = half.View(CpuWrite);
auto full_v = full.View(CpuRead);
autoView( half_v, half, CpuWrite);
autoView( full_v, full, CpuRead);
thread_for(ss, full.Grid()->oSites(),{
int cbos;
Coordinate coor;
@ -63,10 +64,11 @@ template<class vobj> inline void pickCheckerboard(int cb,Lattice<vobj> &half,con
}
});
}
template<class vobj> inline void setCheckerboard(Lattice<vobj> &full,const Lattice<vobj> &half){
template<class vobj> inline void setCheckerboard(Lattice<vobj> &full,const Lattice<vobj> &half)
{
int cb = half.Checkerboard();
auto half_v = half.View(CpuRead);
auto full_v = full.View(CpuWrite);
autoView( half_v , half, CpuRead);
autoView( full_v , full, CpuWrite);
thread_for(ss,full.Grid()->oSites(),{
Coordinate coor;
@ -92,79 +94,15 @@ inline void blockProject(Lattice<iVector<CComplex,nbasis > > &coarseData,
Lattice<CComplex> ip(coarse);
auto coarseData_ = coarseData.View(AcceleratorWrite);
auto ip_ = ip.View(AcceleratorWrite);
autoView( coarseData_ , coarseData, AcceleratorWrite);
autoView( ip_ , ip, AcceleratorWrite);
for(int v=0;v<nbasis;v++) {
blockInnerProduct(ip,Basis[v],fineData);
accelerator_for( sc, coarse->oSites(), vobj::Nsimd(), {
coalescedWrite(coarseData_[sc](v),ip_(sc));
});
});
}
}
#if 0
template<class vobj,class CComplex,int nbasis>
inline void blockProject1(Lattice<iVector<CComplex,nbasis > > &coarseData,
const Lattice<vobj> &fineData,
const std::vector<Lattice<vobj> > &Basis)
{
typedef iVector<CComplex,nbasis > coarseSiteData;
coarseSiteData elide;
typedef decltype(coalescedRead(elide)) ScalarComplex;
GridBase * fine = fineData.Grid();
GridBase * coarse= coarseData.Grid();
int _ndimension = coarse->_ndimension;
// checks
assert( nbasis == Basis.size() );
subdivides(coarse,fine);
for(int i=0;i<nbasis;i++){
conformable(Basis[i],fineData);
}
Coordinate block_r (_ndimension);
for(int d=0 ; d<_ndimension;d++){
block_r[d] = fine->_rdimensions[d] / coarse->_rdimensions[d];
assert(block_r[d]*coarse->_rdimensions[d] == fine->_rdimensions[d]);
}
int blockVol = fine->oSites()/coarse->oSites();
coarseData=Zero();
auto fineData_ = fineData.View(AcceleratorRead);
auto coarseData_ = coarseData.View(AcceleratorWrite);
////////////////////////////////////////////////////////////////////////////////////////////////////////
// To make this lock free, loop over coars parallel, and then loop over fine associated with coarse.
// Otherwise do fine inner product per site, and make the update atomic
////////////////////////////////////////////////////////////////////////////////////////////////////////
accelerator_for( sci, nbasis*coarse->oSites(), vobj::Nsimd(), {
auto sc=sci/nbasis;
auto i=sci%nbasis;
auto Basis_ = Basis[i].View(AcceleratorRead);
Coordinate coor_c(_ndimension);
Lexicographic::CoorFromIndex(coor_c,sc,coarse->_rdimensions); // Block coordinate
int sf;
decltype(innerProduct(Basis_(sf),fineData_(sf))) reduce=Zero();
for(int sb=0;sb<blockVol;sb++){
Coordinate coor_b(_ndimension);
Coordinate coor_f(_ndimension);
Lexicographic::CoorFromIndex(coor_b,sb,block_r);
for(int d=0;d<_ndimension;d++) coor_f[d]=coor_c[d]*block_r[d]+coor_b[d];
Lexicographic::IndexFromCoor(coor_f,sf,fine->_rdimensions);
reduce=reduce+innerProduct(Basis_(sf),fineData_(sf));
}
coalescedWrite(coarseData_[sc](i),reduce);
});
return;
}
#endif
template<class vobj,class CComplex>
inline void blockZAXPY(Lattice<vobj> &fineZ,
@ -191,10 +129,10 @@ inline void blockZAXPY(Lattice<vobj> &fineZ,
assert(block_r[d]*coarse->_rdimensions[d]==fine->_rdimensions[d]);
}
auto fineZ_ = fineZ.View(AcceleratorWrite);
auto fineX_ = fineX.View(AcceleratorRead);
auto fineY_ = fineY.View(AcceleratorRead);
auto coarseA_= coarseA.View(AcceleratorRead);
autoView( fineZ_ , fineZ, AcceleratorWrite);
autoView( fineX_ , fineX, AcceleratorRead);
autoView( fineY_ , fineY, AcceleratorRead);
autoView( coarseA_, coarseA, AcceleratorRead);
accelerator_for(sf, fine->oSites(), CComplex::Nsimd(), {
@ -229,8 +167,8 @@ inline void blockInnerProduct(Lattice<CComplex> &CoarseInner,
// Precision promotion?
fine_inner = localInnerProduct(fineX,fineY);
blockSum(coarse_inner,fine_inner);
auto CoarseInner_ = CoarseInner.View(AcceleratorWrite);
auto coarse_inner_ = coarse_inner.View(AcceleratorRead);
autoView( CoarseInner_ , CoarseInner, AcceleratorWrite);
autoView( coarse_inner_ , coarse_inner, AcceleratorRead);
accelerator_for(ss, coarse->oSites(), 1, {
CoarseInner_[ss] = coarse_inner_[ss];
});
@ -265,8 +203,8 @@ inline void blockSum(Lattice<vobj> &coarseData,const Lattice<vobj> &fineData)
// Turn this around to loop threaded over sc and interior loop
// over sf would thread better
auto coarseData_ = coarseData.View(AcceleratorWrite);
auto fineData_ = fineData.View(AcceleratorRead);
autoView( coarseData_ , coarseData, AcceleratorWrite);
autoView( fineData_ , fineData, AcceleratorRead);
accelerator_for(sc,coarse->oSites(),1,{
@ -359,8 +297,8 @@ inline void blockPromote(const Lattice<iVector<CComplex,nbasis > > &coarseData,
for(int d=0 ; d<_ndimension;d++){
block_r[d] = fine->_rdimensions[d] / coarse->_rdimensions[d];
}
auto fineData_ = fineData.View(AcceleratorWrite);
auto coarseData_ = coarseData.View(AcceleratorRead);
autoView( fineData_ , fineData, AcceleratorWrite);
autoView( coarseData_ , coarseData, AcceleratorRead);
// Loop with a cache friendly loop ordering
accelerator_for(sf,fine->oSites(),1,{
@ -373,7 +311,7 @@ inline void blockPromote(const Lattice<iVector<CComplex,nbasis > > &coarseData,
Lexicographic::IndexFromCoor(coor_c,sc,coarse->_rdimensions);
for(int i=0;i<nbasis;i++) {
/* auto basis_ = Basis[i].View( );*/
/* auto basis_ = Basis[i], );*/
if(i==0) fineData_[sf]=coarseData_[sc](i) *basis_[sf]);
else fineData_[sf]=fineData_[sf]+coarseData_[sc](i)*basis_[sf]);
}
@ -394,8 +332,8 @@ inline void blockPromote(const Lattice<iVector<CComplex,nbasis > > &coarseData,
for(int i=0;i<nbasis;i++) {
Lattice<iScalar<CComplex> > ip = PeekIndex<0>(coarseData,i);
Lattice<CComplex> cip(coarse);
auto cip_ = cip.View(AcceleratorWrite);
auto ip_ = ip.View(AcceleratorRead);
autoView( cip_ , cip, AcceleratorWrite);
autoView( ip_ , ip, AcceleratorRead);
accelerator_forNB(sc,coarse->oSites(),CComplex::Nsimd(),{
coalescedWrite(cip_[sc], ip_(sc)());
});
@ -469,8 +407,8 @@ void localCopyRegion(const Lattice<vobj> &From,Lattice<vobj> & To,Coordinate Fro
Coordinate rdt = Tg->_rdimensions;
Coordinate ist = Tg->_istride;
Coordinate ost = Tg->_ostride;
auto t_v = To.View(AcceleratorWrite);
auto f_v = From.View(AcceleratorRead);
autoView( t_v , To, AcceleratorWrite);
autoView( f_v , From, AcceleratorRead);
accelerator_for(idx,Fg->lSites(),1,{
sobj s;
Coordinate Fcoor(nd);
@ -717,7 +655,7 @@ unvectorizeToLexOrdArray(std::vector<sobj> &out, const Lattice<vobj> &in)
}
//loop over outer index
auto in_v = in.View(CpuRead);
autoView( in_v , in, CpuRead);
thread_for(in_oidx,in_grid->oSites(),{
//Assemble vector of pointers to output elements
ExtractPointerArray<sobj> out_ptrs(in_nsimd);
@ -810,7 +748,7 @@ vectorizeFromLexOrdArray( std::vector<sobj> &in, Lattice<vobj> &out)
icoor[lane].resize(ndim);
grid->iCoorFromIindex(icoor[lane],lane);
}
auto out_v = out.View(CpuWrite);
autoView( out_v , out, CpuWrite);
thread_for(oidx, grid->oSites(),{
//Assemble vector of pointers to output elements
ExtractPointerArray<sobj> ptrs(nsimd);
@ -913,7 +851,7 @@ void precisionChange(Lattice<VobjOut> &out, const Lattice<VobjIn> &in)
std::vector<SobjOut> in_slex_conv(in_grid->lSites());
unvectorizeToLexOrdArray(in_slex_conv, in);
auto out_v = out.View(CpuWrite);
autoView( out_v , out, CpuWrite);
thread_for(out_oidx,out_grid->oSites(),{
Coordinate out_ocoor(ndim);
out_grid->oCoorFromOindex(out_ocoor, out_oidx);

8
Grid/lattice/Lattice_transpose.h
View File

@ -41,8 +41,8 @@ NAMESPACE_BEGIN(Grid);
template<class vobj>
inline Lattice<vobj> transpose(const Lattice<vobj> &lhs){
Lattice<vobj> ret(lhs.Grid());
auto ret_v = ret.View(AcceleratorWrite);
auto lhs_v = lhs.View(AcceleratorRead);
autoView( ret_v, ret, AcceleratorWrite);
autoView( lhs_v, lhs, AcceleratorRead);
accelerator_for(ss,lhs_v.size(),vobj::Nsimd(),{
coalescedWrite(ret_v[ss], transpose(lhs_v(ss)));
});
@ -56,8 +56,8 @@ template<int Index,class vobj>
inline auto TransposeIndex(const Lattice<vobj> &lhs) -> Lattice<decltype(transposeIndex<Index>(vobj()))>
{
Lattice<decltype(transposeIndex<Index>(vobj()))> ret(lhs.Grid());
auto ret_v = ret.View(AcceleratorWrite);
auto lhs_v = lhs.View(AcceleratorRead);
autoView( ret_v, ret, AcceleratorWrite);
autoView( lhs_v, lhs, AcceleratorRead);
accelerator_for(ss,lhs_v.size(),vobj::Nsimd(),{
coalescedWrite(ret_v[ss] , transposeIndex<Index>(lhs_v(ss)));
});

16
Grid/lattice/Lattice_unary.h
View File

@ -35,8 +35,8 @@ NAMESPACE_BEGIN(Grid);
template<class obj> Lattice<obj> pow(const Lattice<obj> &rhs_i,RealD y){
Lattice<obj> ret_i(rhs_i.Grid());
auto rhs = rhs_i.View(AcceleratorRead);
auto ret = ret_i.View(AcceleratorWrite);
autoView( rhs, rhs_i, AcceleratorRead);
autoView( ret, ret_i, AcceleratorWrite);
ret.Checkerboard() = rhs.Checkerboard();
accelerator_for(ss,rhs.size(),1,{
ret[ss]=pow(rhs[ss],y);
@ -45,8 +45,8 @@ template<class obj> Lattice<obj> pow(const Lattice<obj> &rhs_i,RealD y){
}
template<class obj> Lattice<obj> mod(const Lattice<obj> &rhs_i,Integer y){
Lattice<obj> ret_i(rhs_i.Grid());
auto rhs = rhs_i.View(AcceleratorRead);
auto ret = ret_i.View(AcceleratorWrite);
autoView( rhs , rhs_i, AcceleratorRead);
autoView( ret , ret_i, AcceleratorWrite);
ret.Checkerboard() = rhs.Checkerboard();
accelerator_for(ss,rhs.size(),obj::Nsimd(),{
coalescedWrite(ret[ss],mod(rhs(ss),y));
@ -56,8 +56,8 @@ template<class obj> Lattice<obj> mod(const Lattice<obj> &rhs_i,Integer y){
template<class obj> Lattice<obj> div(const Lattice<obj> &rhs_i,Integer y){
Lattice<obj> ret_i(rhs_i.Grid());
auto ret = ret_i.View(AcceleratorWrite);
auto rhs = rhs_i.View(AcceleratorRead);
autoView( ret , ret_i, AcceleratorWrite);
autoView( rhs , rhs_i, AcceleratorRead);
ret.Checkerboard() = rhs_i.Checkerboard();
accelerator_for(ss,rhs.size(),obj::Nsimd(),{
coalescedWrite(ret[ss],div(rhs(ss),y));
@ -67,8 +67,8 @@ template<class obj> Lattice<obj> div(const Lattice<obj> &rhs_i,Integer y){
template<class obj> Lattice<obj> expMat(const Lattice<obj> &rhs_i, RealD alpha, Integer Nexp = DEFAULT_MAT_EXP){
Lattice<obj> ret_i(rhs_i.Grid());
auto rhs = rhs_i.View(AcceleratorRead);
auto ret = ret_i.View(AcceleratorWrite);
autoView( rhs , rhs_i, AcceleratorRead);
autoView( ret , ret_i, AcceleratorWrite);
ret.Checkerboard() = rhs.Checkerboard();
accelerator_for(ss,rhs.size(),obj::Nsimd(),{
coalescedWrite(ret[ss],Exponentiate(rhs(ss),alpha, Nexp));

60
Grid/lattice/Lattice_view.h
View File

@ -25,6 +25,7 @@ void accelerator_inline conformable(GridBase *lhs,GridBase *rhs)
template<class vobj> class LatticeAccelerator : public LatticeBase
{
protected:
//public:
GridBase *_grid;
int checkerboard;
vobj *_odata; // A managed pointer
@ -47,7 +48,7 @@ public:
// The copy constructor for this will need to be used by device lambda functions
/////////////////////////////////////////////////////////////////////////////////////////
template<class vobj>
class LatticeExprView : public LatticeAccelerator<vobj>
class LatticeView : public LatticeAccelerator<vobj>
{
public:
// Rvalue
@ -68,7 +69,12 @@ public:
accelerator_inline uint64_t end(void) const { return this->_odata_size; };
accelerator_inline uint64_t size(void) const { return this->_odata_size; };
LatticeExprView(const LatticeAccelerator<vobj> &refer_to_me) : LatticeAccelerator<vobj> (refer_to_me){}
LatticeView(const LatticeAccelerator<vobj> &refer_to_me) : LatticeAccelerator<vobj> (refer_to_me){}
LatticeView(const LatticeView<vobj> &refer_to_me) = default; // Trivially copyable
LatticeView(const LatticeAccelerator<vobj> &refer_to_me,ViewMode mode) : LatticeAccelerator<vobj> (refer_to_me)
{
this->ViewOpen(mode);
}
// Host functions
void ViewOpen(ViewMode mode)
@ -89,46 +95,20 @@ public:
}
};
///////////////////////////////////////////////////////////////////////
// An object to be stored in a shared_ptr to clean up after last view.
// UserView constructor,destructor updates view manager
// Non-copyable object??? Second base with copy/= deleted?
///////////////////////////////////////////////////////////////////////
class MemViewDeleter {
public:
void *cpu_ptr;
ViewMode mode;
~MemViewDeleter(){
MemoryManager::ViewClose(cpu_ptr,mode);
}
};
template<class vobj>
class LatticeView : public LatticeExprView<vobj>
// Little autoscope assister
template<class View>
class ViewCloser
{
#ifndef GRID_UVM
std::shared_ptr<MemViewDeleter> Deleter;
#endif
public:
#ifdef GRID_UVM
LatticeView(const LatticeAccelerator<vobj> &refer_to_me,ViewMode mode) :
LatticeExprView<vobj> (refer_to_me)
{
}
#else
LatticeView(const LatticeView<vobj> &orig) : LatticeExprView<vobj>(orig) { }
LatticeView(const LatticeAccelerator<vobj> &refer_to_me,ViewMode mode) :
LatticeExprView<vobj> (refer_to_me), Deleter(new MemViewDeleter)
{
// std::cout << "FIXME - copy shared pointer? View Open in LatticeView"<<std::hex<<this->_odata<<std::dec<<" mode "<<mode <<std::endl;
this->ViewOpen(mode);
Deleter->cpu_ptr = this->cpu_ptr;
Deleter->mode = mode;
}
#endif
View v; // Take a copy of view and call view close when I go out of scope automatically
public:
ViewCloser(View &_v) : v(_v) {};
~ViewCloser() { v.ViewClose(); }
};
#define autoView(l_v,l,mode) \
auto l_v = l.View(mode); \
ViewCloser<decltype(l_v)> _autoView##l_v(l_v);
/////////////////////////////////////////////////////////////////////////////////////////
// Lattice expression types used by ET to assemble the AST
//
@ -142,7 +122,7 @@ template <typename T> using is_lattice = std::is_base_of<LatticeBase, T>;
template <typename T> using is_lattice_expr = std::is_base_of<LatticeExpressionBase,T >;
template<class T, bool isLattice> struct ViewMapBase { typedef T Type; };
template<class T> struct ViewMapBase<T,true> { typedef LatticeExprView<typename T::vector_object> Type; };
template<class T> struct ViewMapBase<T,true> { typedef LatticeView<typename T::vector_object> Type; };
template<class T> using ViewMap = ViewMapBase<T,std::is_base_of<LatticeBase, T>::value >;
template <typename Op, typename _T1>