Some instructions must be distinct in the bytecode for the purpose of subsequent compilation decisions (the loop instructions come to mind) but are identical in the LLInt. Currently we use this pattern: _llint_op_loop: nop _llint_op_jmp: # stuff This makes _llint_op_loop distinct from _llint_op_jmp, except that it also assumes that the linker will lay out jmp right after loop. It will do this on many (all?) platforms, but we should not rely on this, particularly since we're using inline assembly, and so we may be subject to whatever crazy assembly directives the C++ compiler feels like emitting.
Created attachment 161021 [details] the patch
Landed in http://trac.webkit.org/changeset/126924