1. The `pragma(inline, false)` doesn't do what the author thinks it does
(Issue 23520).
2. Change that introduced `blockopt` seems to have a slight
misunderstanding about what constant propagation is (`a /= 0.0L` can
always be constant propagated regardless of what surrounds it).
3. Undocument tests that depend on the need for forcing operations to
disarm the optimizer, and replace with simple tests behind StdDdoc.