fix shared memory test

The integration test for shared memory was allocating shared memory for
the full grid instead of the block only.
The amount on exeets the typical available 64kiB, it was only passing
because on CPU we allocate dynamic shared memory via `new` and it looks
like on GPU there is no test.

Author: psychocoderHPC

test/integ/sharedMem/src/sharedMem.cpp

@@ -22,6 +22,7 @@ class SharedMemKernel
 {
 public:
     ALPAKA_NO_HOST_ACC_WARNING
+
     template<typename TAcc>
     ALPAKA_FN_ACC auto operator()(TAcc const& acc, Val* const puiBlockRetVals) const -> void
     {
@@ -94,11 +95,11 @@ namespace alpaka::trait
         template<typename TVec, typename... TArgs>
         ALPAKA_FN_HOST_ACC static auto getBlockSharedMemDynSizeBytes(
             SharedMemKernel<TnumUselessWork, Val> const& /* sharedMemKernel */,
-            TVec const& blockThreadExtent,
+            TVec const& /* blockThreadExtent */,
             TVec const& threadElemExtent,
             TArgs&&...) -> std::size_t
         {
-            return static_cast<std::size_t>(blockThreadExtent.prod() * threadElemExtent.prod()) * sizeof(Val);
+            return static_cast<std::size_t>(threadElemExtent.prod()) * sizeof(Val);
         }
     };
 } // namespace alpaka::trait