orr lr, lr, lsl #16
blt 1f
+ @ src can be unaligned, but that's very rare, so just force it.
+ @ The manual says unaligned ldm should fault, and it does on
+ @ cortex-a78's 32bit mode, but curiously on cortex-a8 it just
+ @ works and loads the data correctly.
+ bic r1, r1, #3
+
0:
ldmia r1!, {r3-r10}
subs r2, #4*8