Make CUTLASS compileable with Clang.

Requires a recent clang build (r359248 or newer).

Enable compilation with clang with these options:
cmake -DCUDA_COMPILER=clang -DCMAKE_CXX_COMPILER=/path/to/clang++

tools/test/unit/core/layout_verification.cu

@@ -142,12 +142,17 @@ int Layout::operator()(Layout::Coordinate const &_coord) const {
 
 }
 
+// test::Layout::Coordinate is actually a std::vector<>, so for ADL lookup to
+// work, the operator<< must be in std::. GCC does look it up in global
+// namespace, but that's a bug.
+namespace std {
 std::ostream & operator<<(std::ostream &out, test::Layout::Coordinate const &coord) {
     for (int i = 0; i < coord.size(); ++i) {
         out << (i ? ", " : "") << coord.at(i);
     }
     return out;
 }
+} // namespace std
 
 ///////////////////////////////////////////////////////////////////////////////////////////////////
 

tools/test/unit/core/layout_verification.h

@@ -94,7 +94,9 @@ class Layout {
 }
 
 /// Implemented in layout_verification.cu
+namespace std {
 std::ostream& operator<<(std::ostream& out, test::Layout::Coordinate const& coord);
+}
 
 namespace test {
 

tools/test/unit/core/tile_iterator.cu

@@ -42,9 +42,7 @@ __global__ void load_store_global(
             typename cutlass::TileStoreIterator<Traits, Scalar, cutlass::IteratorAdvance::kH,
             cutlass::MemorySpace::kGlobal>::Scalar *output,
             int kW,
-            int kH,
-            typename cutlass::TileStoreIterator<Traits, Scalar, cutlass::IteratorAdvance::kH,
-            cutlass::MemorySpace::kGlobal>::Scalar identity = 0
+            int kH
         ) {
 
     /// Load iterator