Normalise sum result to mitigate errors with large iteration counts

CHANGELOG.md

@@ -25,6 +25,7 @@ All notable changes to this project will be documented in this file.
 - Clang compiler OpenMP flags corrected for NVIDIA target.
 - Reorder OpenCL objects in class so destructors are called in safe order.
 - Ensure all OpenCL kernels are present in destructor.
+- Normalise sum result by expected value to help false negative errors.
 
 ### Removed
 - Pre-building of kernels in SYCL version to ensure compatibility with SYCL 1.2.1.

main.cpp

@@ -496,7 +496,7 @@ void check_solution(const unsigned int ntimes, std::vector<T>& a, std::vector<T>
   errB /= b.size();
   double errC = std::accumulate(c.begin(), c.end(), 0.0, [&](double sum, const T val){ return sum + fabs(val - goldC); });
   errC /= c.size();
-  double errSum = fabs(sum - goldSum);
+  double errSum = fabs((sum - goldSum)/goldSum);
 
   double epsi = std::numeric_limits<T>::epsilon() * 100.0;