+ ldmfd sp!, {r4, r5, pc}
+
+#ifdef __ARMEB__
+#define _BE_ONLY_(code...) code
+#define _LE_ONLY_(code...)
+#define push_hbyte0 lsr #8
+#define pull_hbyte1 lsl #24
+#else
+#define _BE_ONLY_(code...)
+#define _LE_ONLY_(code...) code
+#define push_hbyte0 lsl #24
+#define pull_hbyte1 lsr #8
+#endif
+
+.insw_noalign: stmfd sp!, {r4, lr}
+ ldrccb ip, [r1, #-1]!
+ bcc 1f
+
+ ldrh ip, [r0]
+ sub r2, r2, #1
+ _BE_ONLY_( mov ip, ip, ror #8 )
+ strb ip, [r1], #1
+ _LE_ONLY_( mov ip, ip, lsr #8 )
+ _BE_ONLY_( mov ip, ip, lsr #24 )
+
+1: subs r2, r2, #2
+ bmi 3f
+ _BE_ONLY_( mov ip, ip, lsl #24 )
+
+2: ldrh r3, [r0]
+ ldrh r4, [r0]
+ subs r2, r2, #2
+ orr ip, ip, r3, lsl #8
+ orr ip, ip, r4, push_hbyte0
+ str ip, [r1], #4
+ mov ip, r4, pull_hbyte1
+ bpl 2b
+
+ _BE_ONLY_( mov ip, ip, lsr #24 )
+
+3: tst r2, #1
+ strb ip, [r1], #1
+ ldrneh ip, [r0]
+ _BE_ONLY_( movne ip, ip, ror #8 )
+ strneb ip, [r1], #1
+ _LE_ONLY_( movne ip, ip, lsr #8 )
+ _BE_ONLY_( movne ip, ip, lsr #24 )
+ strneb ip, [r1]
+ ldmfd sp!, {r4, pc}