https://bugs.winehq.org/show_bug.cgi?id=48174
--- Comment #45 from Rafał Mużyło galtgendo@o2.pl --- ...to continue the line of thought from yesterday: while I'm not sure about the part that modifies the bytecode, the other two blocks, given that the test app seems to be 32bit, are *statistically* non-ops a half of the time, as with the above, sizeof(SIZEOF_VOID_P) will be 4, so on average it will be swallowed with that rate by the 8 bytes alignment.