Revert back to Grid develop version since new LLVM compilers now do not require static loop count variables.

Grid/qcd/action/fermion/CloverHelpers.h

@@ -120,8 +120,8 @@ public:
     int DimRep = Impl::Dimension;
 
     autoView(in_v, in, AcceleratorWrite);
-    int osites=in.Grid()->oSites();
+
-    accelerator_for(ss, osites, 1, {
+    accelerator_for(ss, in.Grid()->oSites(), 1, {
       for (int sa=0; sa<Ns; sa++)
         for (int ca=0; ca<DimRep; ca++)
           in_v[ss]()(sa,sa)(ca,ca) = c;
@@ -246,8 +246,8 @@ public:
     int DimRep = Impl::Dimension;
 
     autoView(in_v, in, AcceleratorWrite);
-    int osites=in.Grid()->oSites();
+
-    accelerator_for(ss, osites, 1, {
+    accelerator_for(ss, in.Grid()->oSites(), 1, {
       for (int sa=0; sa<Ns; sa++)
         for (int ca=0; ca<DimRep; ca++)
           in_v[ss]()(sa,sa)(ca,ca) = c;

Grid/qcd/action/fermion/WilsonCloverHelpers.h

@@ -85,8 +85,7 @@ public:
     T = Zero();
     autoView(T_v,T,AcceleratorWrite);
     autoView(F_v,F,AcceleratorRead);
-    int osites=T.Grid()->oSites();
+    accelerator_for(i, T.Grid()->oSites(),CloverField::vector_type::Nsimd(),
-    accelerator_for(i, osites,CloverField::vector_type::Nsimd(),
     {
       coalescedWrite(T_v[i]()(0, 1), coalescedRead(timesMinusI(F_v[i]()())));
       coalescedWrite(T_v[i]()(1, 0), coalescedRead(timesMinusI(F_v[i]()())));
@@ -104,8 +103,7 @@ public:
     
     autoView(T_v, T,AcceleratorWrite);
     autoView(F_v, F,AcceleratorRead);
-    int osites=T.Grid()->oSites();
+    accelerator_for(i, T.Grid()->oSites(),CloverField::vector_type::Nsimd(),
-    accelerator_for(i, osites,CloverField::vector_type::Nsimd(),
     {
       coalescedWrite(T_v[i]()(0, 1), coalescedRead(-F_v[i]()()));
       coalescedWrite(T_v[i]()(1, 0), coalescedRead(F_v[i]()()));
@@ -123,8 +121,7 @@ public:
 
     autoView(T_v,T,AcceleratorWrite);
     autoView(F_v,F,AcceleratorRead);
-    int osites=T.Grid()->oSites();
+    accelerator_for(i, T.Grid()->oSites(),CloverField::vector_type::Nsimd(),
-    accelerator_for(i, osites,CloverField::vector_type::Nsimd(),
     {
       coalescedWrite(T_v[i]()(0, 0), coalescedRead(timesMinusI(F_v[i]()())));
       coalescedWrite(T_v[i]()(1, 1), coalescedRead(timesI(F_v[i]()())));
@@ -142,8 +139,7 @@ public:
 
     autoView( T_v , T, AcceleratorWrite);
     autoView( F_v , F, AcceleratorRead);
-    int osites=T.Grid()->oSites();
+    accelerator_for(i, T.Grid()->oSites(),CloverField::vector_type::Nsimd(),
-    accelerator_for(i, osites,CloverField::vector_type::Nsimd(),
     {
       coalescedWrite(T_v[i]()(0, 1), coalescedRead(timesI(F_v[i]()())));
       coalescedWrite(T_v[i]()(1, 0), coalescedRead(timesI(F_v[i]()())));
@@ -161,8 +157,7 @@ public:
     
     autoView( T_v ,T,AcceleratorWrite);
     autoView( F_v ,F,AcceleratorRead);
-    int osites=T.Grid()->oSites();
+    accelerator_for(i, T.Grid()->oSites(),CloverField::vector_type::Nsimd(),
-    accelerator_for(i, osites,CloverField::vector_type::Nsimd(),
     {
       coalescedWrite(T_v[i]()(0, 1), coalescedRead(-(F_v[i]()())));
       coalescedWrite(T_v[i]()(1, 0), coalescedRead((F_v[i]()())));
@@ -181,8 +176,7 @@ public:
 
     autoView( T_v , T,AcceleratorWrite);
     autoView( F_v , F,AcceleratorRead);
-    int osites=T.Grid()->oSites();
+    accelerator_for(i, T.Grid()->oSites(),CloverField::vector_type::Nsimd(),
-    accelerator_for(i, osites,CloverField::vector_type::Nsimd(),
     {
       coalescedWrite(T_v[i]()(0, 0), coalescedRead(timesI(F_v[i]()())));
       coalescedWrite(T_v[i]()(1, 1), coalescedRead(timesMinusI(F_v[i]()())));
@@ -206,8 +200,7 @@ public:
     autoView(phi_v, phi, AcceleratorRead);
     autoView(C_v,   C,   AcceleratorRead);
     typedef decltype(coalescedRead(out_v[0])) calcSpinor;
-    int osites=out.Grid()->oSites();
+    accelerator_for(sss,out.Grid()->oSites(),Nsimd,{
-    accelerator_for(sss,osites,Nsimd,{
       calcSpinor tmp;
       multClover(tmp,C_v[sss],phi_v(sss));
       coalescedWrite(out_v[sss],tmp);
@@ -617,8 +610,7 @@ public:
     autoView(triangle_v, triangle, AcceleratorWrite);
 
     // NOTE: this function cannot be 'private' since nvcc forbids this for kernels
-    int osites=full.Grid()->oSites();
+    accelerator_for(ss, full.Grid()->oSites(), 1, {
-    accelerator_for(ss, osites, 1, {
       for(int s_row = 0; s_row < Ns; s_row++) {
         for(int s_col = 0; s_col < Ns; s_col++) {
           if(abs(s_row - s_col) > 1 || s_row + s_col == 3) continue;
@@ -658,8 +650,7 @@ public:
     autoView(full_v,     full,     AcceleratorWrite);
 
     // NOTE: this function cannot be 'private' since nvcc forbids this for kernels
-    int osites=full.Grid()->oSites();
+    accelerator_for(ss, full.Grid()->oSites(), 1, {
-    accelerator_for(ss, osites, 1, {
       for(int s_row = 0; s_row < Ns; s_row++) {
         for(int s_col = 0; s_col < Ns; s_col++) {
           if(abs(s_row - s_col) > 1 || s_row + s_col == 3) continue;

Grid/qcd/action/fermion/WilsonImpl.h

@@ -110,10 +110,8 @@ public:
     autoView( out_v, out, AcceleratorWrite);
     autoView( phi_v, phi, AcceleratorRead);
     autoView( Umu_v, Umu, AcceleratorRead);
-    int size=out.Grid()->oSites();
-
     typedef decltype(coalescedRead(out_v[0]))   calcSpinor;
-    accelerator_for(sss,size,Nsimd,{
+    accelerator_for(sss,out.Grid()->oSites(),Nsimd,{
 	calcSpinor tmp;
 	multLink(tmp,Umu_v[sss],phi_v(sss),mu);
 	coalescedWrite(out_v[sss],tmp);
@@ -205,8 +203,7 @@ public:
       autoView( tmp_v , tmp, AcceleratorWrite);
       autoView( Btilde_v , Btilde, AcceleratorRead);
       autoView( Atilde_v , Atilde, AcceleratorRead);
-      int size=tmp.Grid()->oSites();
+      accelerator_for(sss,tmp.Grid()->oSites(),1,{
-      accelerator_for(sss,size,1,{
 	  int sU=sss;
 	  for(int s=0;s<Ls;s++){
 	    int sF = s+Ls*sU;
@@ -220,8 +217,7 @@ public:
       const int Nsimd = SiteSpinor::Nsimd();
       autoView( Btilde_v , Btilde, AcceleratorRead);
       autoView( Atilde_v , Atilde, AcceleratorRead);
-      int size=mat.Grid()->oSites();
+      accelerator_for(sss,mat.Grid()->oSites(),Nsimd,{
-      accelerator_for(sss,size,Nsimd,{
 	  int sU=sss;
   	  typedef decltype(coalescedRead(mat_v[sU](mu)() )) ColorMatrixType;
   	  ColorMatrixType sum;

Grid/qcd/action/gauge/GaugeImplTypes.h

@@ -88,8 +88,7 @@ public:
   static inline void AddLink(Field &U, LinkField &W, int mu) { // U[mu] += W
     autoView(U_v,U,AcceleratorWrite);
     autoView(W_v,W,AcceleratorRead);
-    int size=U.Grid()->oSites();
+    accelerator_for( ss, U.Grid()->oSites(), 1, {
-    accelerator_for( ss, size, 1, {
       U_v[ss](mu) = U_v[ss](mu) + W_v[ss]();
     });
   }
@@ -139,8 +138,7 @@ public:
     //auto start = std::chrono::high_resolution_clock::now();
     autoView(U_v,U,AcceleratorWrite);
     autoView(P_v,P,AcceleratorRead);
-    int size=P.Grid()->oSites();
+    accelerator_for(ss, P.Grid()->oSites(),1,{
-    accelerator_for(ss, size,1,{
       for (int mu = 0; mu < Nd; mu++) {
           U_v[ss](mu) = Exponentiate(P_v[ss](mu), ep, Nexp) * U_v[ss](mu);
           U_v[ss](mu) = Group::ProjectOnGeneralGroup(U_v[ss](mu));