diff --git a/include/rp_kernel_on_cpu.h b/include/rp_kernel_on_cpu.h
index 5b8361a4a..90f29b660 100644
--- a/include/rp_kernel_on_cpu.h
+++ b/include/rp_kernel_on_cpu.h
@@ -6,7 +6,7 @@
 #ifndef _RP_KERNEL_ON_CPU_H
 #define _RP_KERNEL_ON_CPU_H
 
-#define swap_workaround(n) __builtin_bswap32(n)
+u32 swap_workaround (const u32 n);
 
 u32 apply_rule (const u32 name, const u32 p0, const u32 p1, u32 buf0[4], u32 buf1[4], const u32 in_len);
 u32 apply_rules (u32 *cmds, u32 buf0[4], u32 buf1[4], const u32 len);
diff --git a/src/rp_kernel_on_cpu.c b/src/rp_kernel_on_cpu.c
index 30e46afe3..8a9d37eb1 100644
--- a/src/rp_kernel_on_cpu.c
+++ b/src/rp_kernel_on_cpu.c
@@ -8,8 +8,14 @@
 
 #include "common.h"
 #include "types.h"
+#include "bitops.h"
 #include "rp_kernel_on_cpu.h"
 
+u32 swap_workaround (const u32 n)
+{
+  return byte_swap_32 (n);
+}
+
 static u32 generate_cmask (u32 buf)
 {
   const u32 rmask =  ((buf & 0x40404040) >> 1)