Fix 32u_reverse_32u for ARM.

 * Order of operands in inline asm reversed.
 * Worked for some tune setting since gcc generated rbit lr, lr so
   order didn't matter. For other tune settings it generated rbit r3, r2
   which would fail QA test.

Signed-off-by: Philip Balister <philip@balister.org>
Gbp-Pq: Name 0009-Fix-32u_reverse_32u-for-ARM.patch

diff --git a/kernels/volk/volk_32u_reverse_32u.h b/kernels/volk/volk_32u_reverse_32u.h
index acdbacde96834ac4ed480d083a4312ccbb37f45e..417a5d153541a45826de13ac4275a6c8655c20e5 100644 (file)
--- a/kernels/volk/volk_32u_reverse_32u.h
+++ b/kernels/volk/volk_32u_reverse_32u.h
@@ -337,7 +337,7 @@ static inline void volk_32u_reverse_32u_bintree_permute_bottom_up(uint32_t* out,
 #include <arm_neon.h>
 
 #define DO_RBIT                                          \
-    asm("rbit %1,%0" : "=r" (*out_ptr) : "r" (*in_ptr)); \
+    asm("rbit %0,%1" : "=r" (*out_ptr) : "r" (*in_ptr)); \
     in_ptr++;                                            \
     out_ptr++;