The bounds check here is already performed by the at() we're calling, and removing it appears to have non-trivial performance impact.
@@ -559,7 +559,6 @@ public:
ALWAYS_INLINE ByteCodeValueType argument(size_t offset) const
{
- VERIFY(state().instruction_position + offset <= m_bytecode->size());
return m_bytecode->at(state().instruction_position + 1 + offset);
}