Make tests running past nvcc. Different NVCC versions proving tricky to keep happy. This is 9.2

Grid/simd/Grid_avx512.h

@@ -481,7 +481,7 @@ struct Rotate{
 // Some Template specialization
 
 // Hack for CLANG until mm512_reduce_add_ps etc... are implemented in GCC and Clang releases
-#ifndef __INTEL_COMPILER
+#if 0
 #warning "Slow reduction due to incomplete reduce intrinsics"
 //Complex float Reduce
 template<>

Grid/simd/Grid_vector_types.h

@@ -746,11 +746,12 @@ accelerator_inline Grid_simd<S, V> operator/(Grid_simd<S, V> a, Grid_simd<S, V>
 
   ret = a * conjugate(b) ;
   den = b * conjugate(b) ;
-  
-  auto real_den = toReal(den);
-
-  ret.v=binary<V>(ret.v, real_den.v, DivSIMD());
 
+  // duplicates real part
+  auto real_den  = toReal(den);
+  simd zden;
+  memcpy((void *)&zden.v,(void *)&real_den.v,sizeof(zden));
+  ret.v=binary<V>(ret.v, zden.v, DivSIMD());
   return ret;
 };
 
@@ -839,26 +840,28 @@ accelerator_inline Grid_simd<S, V> trace(const Grid_simd<S, V> &arg) {
 // insert real into complex and zero imag;
 ////////////////////////////////////////////////////////////
 
+
+template <class T> struct toRealMapper {};
+template<> struct toRealMapper<vComplexF> {  typedef vRealF Realified; };
+template<> struct toRealMapper<vComplexD> {  typedef vRealD Realified; };
 // real = toReal( complex )
-template <class S, class V, IfReal<S> = 0>
-accelerator_inline Grid_simd<S, V> toReal(const Grid_simd<complex<S>, V> &in) {
-  typedef Grid_simd<S, V> simd;
-  simd ret;
-  typename simd::conv_t conv;
-  conv.v = in.v;  // copy the vector content (bytewise)
-  for (int i = 0; i < simd::Nsimd(); i += 2) {
+template <class Csimd>  // must be a real arg
+accelerator_inline typename toRealMapper<Csimd>::Realified toReal(const Csimd &in) {
+  typedef typename toRealMapper<Csimd>::Realified Rsimd;
+  Rsimd ret;
+  typename Rsimd::conv_t conv;
+  memcpy((void *)&conv.v,(void *)&in.v,sizeof(conv.v));
+  for (int i = 0; i < Rsimd::Nsimd(); i += 2) {
     conv.s[i + 1] = conv.s[i];  // duplicate (r,r);(r,r);(r,r); etc...
   }
-  ret.v = conv.v;
+  memcpy((void *)&ret.v,(void *)&conv.v,sizeof(ret.v));
   return ret;
 }
 
-
 template <class T> struct toComplexMapper {};
 template<> struct toComplexMapper<vRealF> {  typedef vComplexF Complexified; };
 template<> struct toComplexMapper<vRealD> {  typedef vComplexD Complexified; };
 
-
 // complex = toComplex( real )
 template <class Rsimd>  // must be a real arg
 accelerator_inline typename toComplexMapper<Rsimd>::Complexified toComplex(const Rsimd &in) {

Grid/tensors/Tensor_exp.h

@@ -55,6 +55,7 @@ template<class vtype, int N> accelerator_inline iVector<vtype, N> Exponentiate(c
 
 
 // Specialisation: Cayley-Hamilton exponential for SU(3)
+#ifndef GRID_NVCC
 template<class vtype, typename std::enable_if< GridTypeMapper<vtype>::TensorLevel == 0>::type * =nullptr> 
 accelerator_inline iMatrix<vtype,3> Exponentiate(const iMatrix<vtype,3> &arg, RealD alpha  , Integer Nexp = DEFAULT_MAT_EXP )
 {
@@ -114,7 +115,7 @@ accelerator_inline iMatrix<vtype,3> Exponentiate(const iMatrix<vtype,3> &arg, Re
 
   return (f0 * unit + timesMinusI(f1) * arg*alpha - f2 * iQ2);
 }
-
+#endif
 
 
 // General exponential

tests/core/Test_main.cc

@@ -179,7 +179,7 @@ int main(int argc, char **argv) {
       std::cout << "Norm2 LatticeReal : "<< norm2(BarReal) << std::endl;
       std::cout << "Norm2 LatticeComplex : "<< norm2(BarComplex) << std::endl;
 
-      exit(0);
+      //      exit(0);
 
       TComplex tr = trace(cmat);
 

tests/forces/Test_gp_plaq_force.cc

@@ -87,9 +87,9 @@ int main (int argc, char ** argv)
     auto Uprime_v = Uprime.View();
     auto U_v = U.View();
     auto mom_v = mom.View();
-    parallel_for(auto i=mom_v.begin();i<mom_v.end();i++){ // exp(pmu dt) * Umu
+    thread_loop( (auto i=mom_v.begin();i<mom_v.end();i++),{ // exp(pmu dt) * Umu
       Uprime_v[i](mu) = U_v[i](mu) + mom_v[i](mu)*U_v[i](mu)*dt ;
-    }
+    });
   }
 
   ComplexD Sprime    = Action.S(Uprime);

tests/forces/Test_wilsonclover_force.cc

@@ -109,10 +109,10 @@ int main(int argc, char **argv)
     auto Uprime_v = Uprime.View();
     auto U_v      = U.View();
     auto mom_v    = mom.View();
-    parallel_for(int ss = 0; ss < mom.Grid()->oSites(); ss++)
+    thread_loop( (int ss = 0; ss < mom.Grid()->oSites(); ss++),
     {
       Uprime_v[ss]._internal[mu] = ProjectOnGroup(Exponentiate(mom_v[ss]._internal[mu], dt, 12) * U_v[ss]._internal[mu]);
-    }
+    });
   }
 
   std::cout << GridLogMessage << "Initial mom hamiltonian is " << Hmom << std::endl;