+ * optimization: MAP and MAP-INTO are more efficient for non-simple vectors,
+ when (> SPEED SPACE).
+ * optimization: local call trampolines (x86 and x86-64) are emitted
+ inline.
+ * optimization: implicit value cells for dynamic-extent closed-over bindings
+ on x86 and x86-64 can hold unboxed values as well.
+ * meta-optimization: improved compilation speed, especially for large
+ functions. (lp#792363 and lp#394206)