On Monday, 18 November 2013 at 05:26:30 UTC, Chris Cain wrote:
On my computer (`-release -inline -noboundscheck` ... `-O` is omitted because it removes the summation all together since it isn't used anywhere):
The better way to avoid this problem is usually to compile the operation to be benchmarked separately, in such a way that the result can't be ignored (e.g. by making it read the parameters from function arguments and returning the result as the return value). Comparing non-optimized (i.e. -O) builds for execution speed is of questionable relevance for most cases.
David
