diff -r 675d3f76444d Modules/_sha3/kcp/KeccakP-1600-inplace32BI.c
--- a/Modules/_sha3/kcp/KeccakP-1600-inplace32BI.c	Tue Sep 27 22:05:04 2016 +0300
+++ b/Modules/_sha3/kcp/KeccakP-1600-inplace32BI.c	Tue Sep 27 23:33:56 2016 +0300
@@ -83,28 +83,34 @@ typedef unsigned int UINT32;
         lowOut = lowIn ^ temp0; \
         highOut = highIn ^ temp1;
 
+typedef union {
+    UINT8 asBytes[8];
+    UINT32 asInts[2];
+} laneUnion;
+
+#if (PLATFORM_BYTE_ORDER != IS_LITTLE_ENDIAN)
+#define SWAB4(value) do { \
+        (value) = (((value) & 0x0000FFFFu) << 16) | ((value) >> 16); \
+        (value) = (((value) & 0x00FF00FFu) << 8) | \
+                  (((value) >> 8) & 0x00FF00FFu); \
+    } while (0);
+#endif
+
 void KeccakP1600_SetBytesInLaneToZero(void *state, unsigned int lanePosition, unsigned int offset, unsigned int length)
 {
-    UINT8 laneAsBytes[8];
+    laneUnion lane;
     UINT32 low, high;
     UINT32 temp, temp0, temp1;
     UINT32 *stateAsHalfLanes = (UINT32*)state;
 
-    memset(laneAsBytes, 0xFF, offset);
-    memset(laneAsBytes+offset, 0x00, length);
-    memset(laneAsBytes+offset+length, 0xFF, 8-offset-length);
-#if (PLATFORM_BYTE_ORDER == IS_LITTLE_ENDIAN)
-    low = *((UINT32*)(laneAsBytes+0));
-    high = *((UINT32*)(laneAsBytes+4));
-#else
-    low = laneAsBytes[0]
-        | ((UINT32)(laneAsBytes[1]) << 8)
-        | ((UINT32)(laneAsBytes[2]) << 16)
-        | ((UINT32)(laneAsBytes[3]) << 24);
-    high = laneAsBytes[4]
-        | ((UINT32)(laneAsBytes[5]) << 8)
-        | ((UINT32)(laneAsBytes[6]) << 16)
-        | ((UINT32)(laneAsBytes[7]) << 24);
+    memset(lane.asBytes, 0xFF, offset);
+    memset(lane.asBytes+offset, 0x00, length);
+    memset(lane.asBytes+offset+length, 0xFF, 8-offset-length);
+    low = lane.asInts[0];
+    high = lane.asInts[1];
+#if (PLATFORM_BYTE_ORDER != IS_LITTLE_ENDIAN)
+    SWAB4(low);
+    SWAB4(high);
 #endif
     toBitInterleavingAndAND(low, high, stateAsHalfLanes[lanePosition*2+0], stateAsHalfLanes[lanePosition*2+1], temp, temp0, temp1);
 }
@@ -141,25 +147,18 @@ void KeccakP1600_AddByte(void *state, un
 
 void KeccakP1600_AddBytesInLane(void *state, unsigned int lanePosition, const unsigned char *data, unsigned int offset, unsigned int length)
 {
-    UINT8 laneAsBytes[8];
+    laneUnion lane;
     UINT32 low, high;
     UINT32 temp, temp0, temp1;
     UINT32 *stateAsHalfLanes = (UINT32*)state;
 
-    memset(laneAsBytes, 0, 8);
-    memcpy(laneAsBytes+offset, data, length);
-#if (PLATFORM_BYTE_ORDER == IS_LITTLE_ENDIAN)
-    low = *((UINT32*)(laneAsBytes+0));
-    high = *((UINT32*)(laneAsBytes+4));
-#else
-    low = laneAsBytes[0]
-        | ((UINT32)(laneAsBytes[1]) << 8)
-        | ((UINT32)(laneAsBytes[2]) << 16)
-        | ((UINT32)(laneAsBytes[3]) << 24);
-    high = laneAsBytes[4]
-        | ((UINT32)(laneAsBytes[5]) << 8)
-        | ((UINT32)(laneAsBytes[6]) << 16)
-        | ((UINT32)(laneAsBytes[7]) << 24);
+    memset(lane.asBytes, 0, 8);
+    memcpy(lane.asBytes+offset, data, length);
+    low = lane.asInts[0];
+    high = lane.asInts[1];
+#if (PLATFORM_BYTE_ORDER != IS_LITTLE_ENDIAN)
+    SWAB4(low);
+    SWAB4(high);
 #endif
     toBitInterleavingAndXOR(low, high, stateAsHalfLanes[lanePosition*2+0], stateAsHalfLanes[lanePosition*2+1], temp, temp0, temp1);
 }
@@ -289,23 +288,16 @@ void KeccakP1600_ExtractBytesInLane(cons
 {
     UINT32 *stateAsHalfLanes = (UINT32*)state;
     UINT32 low, high, temp, temp0, temp1;
-    UINT8 laneAsBytes[8];
+    laneUnion lane;
 
     fromBitInterleaving(stateAsHalfLanes[lanePosition*2], stateAsHalfLanes[lanePosition*2+1], low, high, temp, temp0, temp1);
-#if (PLATFORM_BYTE_ORDER == IS_LITTLE_ENDIAN)
-    *((UINT32*)(laneAsBytes+0)) = low;
-    *((UINT32*)(laneAsBytes+4)) = high;
-#else
-    laneAsBytes[0] = low & 0xFF;
-    laneAsBytes[1] = (low >> 8) & 0xFF;
-    laneAsBytes[2] = (low >> 16) & 0xFF;
-    laneAsBytes[3] = (low >> 24) & 0xFF;
-    laneAsBytes[4] = high & 0xFF;
-    laneAsBytes[5] = (high >> 8) & 0xFF;
-    laneAsBytes[6] = (high >> 16) & 0xFF;
-    laneAsBytes[7] = (high >> 24) & 0xFF;
+#if (PLATFORM_BYTE_ORDER != IS_LITTLE_ENDIAN)
+    SWAB4(low);
+    SWAB4(high);
 #endif
-    memcpy(data, laneAsBytes+offset, length);
+    lane.asInts[0] = low;
+    lane.asInts[1] = high;
+    memcpy(data, lane.asBytes+offset, length);
 }
 
 /* ---------------------------------------------------------------- */
@@ -361,25 +353,18 @@ void KeccakP1600_ExtractAndAddBytesInLan
 {
     UINT32 *stateAsHalfLanes = (UINT32*)state;
     UINT32 low, high, temp, temp0, temp1;
-    UINT8 laneAsBytes[8];
+    laneUnion lane;
     unsigned int i;
 
     fromBitInterleaving(stateAsHalfLanes[lanePosition*2], stateAsHalfLanes[lanePosition*2+1], low, high, temp, temp0, temp1);
-#if (PLATFORM_BYTE_ORDER == IS_LITTLE_ENDIAN)
-    *((UINT32*)(laneAsBytes+0)) = low;
-    *((UINT32*)(laneAsBytes+4)) = high;
-#else
-    laneAsBytes[0] = low & 0xFF;
-    laneAsBytes[1] = (low >> 8) & 0xFF;
-    laneAsBytes[2] = (low >> 16) & 0xFF;
-    laneAsBytes[3] = (low >> 24) & 0xFF;
-    laneAsBytes[4] = high & 0xFF;
-    laneAsBytes[5] = (high >> 8) & 0xFF;
-    laneAsBytes[6] = (high >> 16) & 0xFF;
-    laneAsBytes[7] = (high >> 24) & 0xFF;
+#if (PLATFORM_BYTE_ORDER != IS_LITTLE_ENDIAN)
+    SWAB4(low);
+    SWAB4(high);
 #endif
+    lane.asInts[0] = low;
+    lane.asInts[1] = high;
     for(i=0; i<length; i++)
-        output[i] = input[i] ^ laneAsBytes[offset+i];
+        output[i] = input[i] ^ lane.asBytes[offset+i];
 }
 
 /* ---------------------------------------------------------------- */