+ * The procesor natively handles unaligned loads, so just use memcpy()
+ * and memcmp(), to enable those optimizations.
+ *
+ * XXX - are those all the x86 tests we need?
+ * XXX - do we need to worry about ARMv1 through ARMv5, which didn't
+ * support unaligned loads, and, if so, do we need to worry about all
+ * of them, or just some of them, e.g. ARMv5?
+ * XXX - are those the only 68k tests we need not to generated
+ * unaligned accesses if the target is the 68000 or 68010?
+ * XXX - are there any tests we don't need, because some definitions are for
+ * compilers that also predefine the GCC symbols?
+ * XXX - do we need to test for both 32-bit and 64-bit versions of those
+ * architectures in all cases?
+ */
+#else
+/*
+ * The processor doesn't natively handle unaligned loads,
+ * and the compiler might "helpfully" optimize memcpy()
+ * and memcmp(), when handed pointers that would normally
+ * be properly aligned, into sequences that assume proper
+ * alignment.
+ *
+ * Do copies and compares of possibly-unaligned data by
+ * calling routines that wrap memcpy() and memcmp(), to
+ * prevent that optimization.