Padding needs to be done in a way that avoids leaving gaps in the object layout. Current strategy lays out 256 byte fields, which then becomes a significant part of generated code. To reduce code we could consider padding using a mix of byte and long fields.

A PoC which assumes 16 bytes + 240 bytes in longs would be sufficient reduces generated JMH code size (both compiled and source) by ~20%.

This might be of interest in light of JDK-8345302