Since
28321c9..., the context pointer in ECX was overwritten with the
GOT pointer, used to find the end-swapping table, resulting in an abort.
Reallocate so that the round count, which is loaded /after/ the endswap
pointer, ends up in ECX, and the context pointer goes in EAX, which
doesn't get clobbered.
This doesn't affect 64-bit targets.
// Arguments come in on the stack, and need to be collected. We
// don't have a shortage of registers.
-# define K ecx
+# define K eax
# define SRC edx
# define DST edx
-# define NR eax
+# define NR ecx
mov K, [esp + 4]
mov SRC, [esp + 8]