[RFC PATCH 13/22] staging: crypto: skein: fix leading whitespace

Jason Cooper jason at lakedaemon.net
Tue Mar 11 21:32:45 UTC 2014


Signed-off-by: Jason Cooper <jason at lakedaemon.net>
---
 drivers/staging/skein/include/skein.h        |  136 +-
 drivers/staging/skein/include/skeinApi.h     |  284 +--
 drivers/staging/skein/include/skein_iv.h     |  276 +--
 drivers/staging/skein/include/threefishApi.h |  230 +--
 drivers/staging/skein/skein.c                | 1126 +++++------
 drivers/staging/skein/skeinApi.c             |  320 +--
 drivers/staging/skein/skeinBlockNo3F.c       |  286 +--
 drivers/staging/skein/skein_block.c          | 1012 +++++-----
 drivers/staging/skein/threefish1024Block.c   | 2740 +++++++++++++-------------
 drivers/staging/skein/threefish256Block.c    |  639 +++---
 drivers/staging/skein/threefish512Block.c    | 1254 ++++++------
 drivers/staging/skein/threefishApi.c         |  102 +-
 12 files changed, 4200 insertions(+), 4205 deletions(-)

diff --git a/drivers/staging/skein/include/skein.h b/drivers/staging/skein/include/skein.h
index 18bb15824e41..906bcee41c39 100644
--- a/drivers/staging/skein/include/skein.h
+++ b/drivers/staging/skein/include/skein.h
@@ -38,11 +38,11 @@
 #define Skein_Swap64(w64)  (w64)
 
 enum
-    {
-    SKEIN_SUCCESS         =      0,          /* return codes from Skein calls */
-    SKEIN_FAIL            =      1,
-    SKEIN_BAD_HASHLEN     =      2
-    };
+	{
+	SKEIN_SUCCESS         =      0,          /* return codes from Skein calls */
+	SKEIN_FAIL            =      1,
+	SKEIN_BAD_HASHLEN     =      2
+	};
 
 #define  SKEIN_MODIFIER_WORDS   (2)          /* number of modifier (tweak) words */
 
@@ -64,32 +64,32 @@ enum
 #define  SKEIN1024_BLOCK_BYTES  (8*SKEIN1024_STATE_WORDS)
 
 struct skein_ctx_hdr
-    {
-    size_t  hashBitLen;                      /* size of hash result, in bits */
-    size_t  bCnt;                            /* current byte count in buffer b[] */
-    u64  T[SKEIN_MODIFIER_WORDS];         /* tweak words: T[0]=byte cnt, T[1]=flags */
-    };
+	{
+	size_t  hashBitLen;                      /* size of hash result, in bits */
+	size_t  bCnt;                            /* current byte count in buffer b[] */
+	u64  T[SKEIN_MODIFIER_WORDS];         /* tweak words: T[0]=byte cnt, T[1]=flags */
+	};
 
 struct skein_256_ctx                               /*  256-bit Skein hash context structure */
-    {
-    struct skein_ctx_hdr h;                      /* common header context variables */
-    u64  X[SKEIN_256_STATE_WORDS];        /* chaining variables */
-    u8  b[SKEIN_256_BLOCK_BYTES];        /* partial block buffer (8-byte aligned) */
-    };
+	{
+	struct skein_ctx_hdr h;                      /* common header context variables */
+	u64  X[SKEIN_256_STATE_WORDS];        /* chaining variables */
+	u8  b[SKEIN_256_BLOCK_BYTES];        /* partial block buffer (8-byte aligned) */
+	};
 
 struct skein_512_ctx                             /*  512-bit Skein hash context structure */
-    {
-    struct skein_ctx_hdr h;                      /* common header context variables */
-    u64  X[SKEIN_512_STATE_WORDS];        /* chaining variables */
-    u8  b[SKEIN_512_BLOCK_BYTES];        /* partial block buffer (8-byte aligned) */
-    };
+	{
+	struct skein_ctx_hdr h;                      /* common header context variables */
+	u64  X[SKEIN_512_STATE_WORDS];        /* chaining variables */
+	u8  b[SKEIN_512_BLOCK_BYTES];        /* partial block buffer (8-byte aligned) */
+	};
 
 struct skein1024_ctx                              /* 1024-bit Skein hash context structure */
-    {
-    struct skein_ctx_hdr h;                      /* common header context variables */
-    u64  X[SKEIN1024_STATE_WORDS];        /* chaining variables */
-    u8  b[SKEIN1024_BLOCK_BYTES];        /* partial block buffer (8-byte aligned) */
-    };
+	{
+	struct skein_ctx_hdr h;                      /* common header context variables */
+	u64  X[SKEIN1024_STATE_WORDS];        /* chaining variables */
+	u8  b[SKEIN1024_BLOCK_BYTES];        /* partial block buffer (8-byte aligned) */
+	};
 
 /*   Skein APIs for (incremental) "straight hashing" */
 int  Skein_256_Init(struct skein_256_ctx *ctx, size_t hashBitLen);
@@ -150,18 +150,18 @@ int  Skein1024_Output(struct skein1024_ctx *ctx, u8 *hashVal);
 
 /* tweak word T[1]: bit field starting positions */
 #define SKEIN_T1_BIT(BIT)       ((BIT) - 64)            /* offset 64 because it's the second word  */
-                                
+
 #define SKEIN_T1_POS_TREE_LVL   SKEIN_T1_BIT(112)       /* bits 112..118: level in hash tree       */
 #define SKEIN_T1_POS_BIT_PAD    SKEIN_T1_BIT(119)       /* bit  119     : partial final input byte */
 #define SKEIN_T1_POS_BLK_TYPE   SKEIN_T1_BIT(120)       /* bits 120..125: type field               */
 #define SKEIN_T1_POS_FIRST      SKEIN_T1_BIT(126)       /* bits 126     : first block flag         */
 #define SKEIN_T1_POS_FINAL      SKEIN_T1_BIT(127)       /* bit  127     : final block flag         */
-                                
+
 /* tweak word T[1]: flag bit definition(s) */
 #define SKEIN_T1_FLAG_FIRST     (((u64)  1) << SKEIN_T1_POS_FIRST)
 #define SKEIN_T1_FLAG_FINAL     (((u64)  1) << SKEIN_T1_POS_FINAL)
 #define SKEIN_T1_FLAG_BIT_PAD   (((u64)  1) << SKEIN_T1_POS_BIT_PAD)
-                                
+
 /* tweak word T[1]: tree level bit field mask */
 #define SKEIN_T1_TREE_LVL_MASK  (((u64)0x7F) << SKEIN_T1_POS_TREE_LVL)
 #define SKEIN_T1_TREE_LEVEL(n)  (((u64) (n)) << SKEIN_T1_POS_TREE_LVL)
@@ -213,9 +213,9 @@ int  Skein1024_Output(struct skein1024_ctx *ctx, u8 *hashVal);
 #define SKEIN_CFG_TREE_MAX_LEVEL_MSK  (((u64) 0xFF) << SKEIN_CFG_TREE_MAX_LEVEL_POS)
 
 #define SKEIN_CFG_TREE_INFO(leaf, node, maxLvl)                   \
-    ((((u64)(leaf))   << SKEIN_CFG_TREE_LEAF_SIZE_POS) |    \
-     (((u64)(node))   << SKEIN_CFG_TREE_NODE_SIZE_POS) |    \
-     (((u64)(maxLvl)) << SKEIN_CFG_TREE_MAX_LEVEL_POS))
+	((((u64)(leaf))   << SKEIN_CFG_TREE_LEAF_SIZE_POS) |    \
+	 (((u64)(node))   << SKEIN_CFG_TREE_NODE_SIZE_POS) |    \
+	 (((u64)(maxLvl)) << SKEIN_CFG_TREE_MAX_LEVEL_POS))
 
 #define SKEIN_CFG_TREE_INFO_SEQUENTIAL SKEIN_CFG_TREE_INFO(0, 0, 0) /* use as treeInfo in InitExt() call for sequential processing */
 
@@ -233,17 +233,17 @@ int  Skein1024_Output(struct skein1024_ctx *ctx, u8 *hashVal);
 
 /* set both tweak words at once */
 #define Skein_Set_T0_T1(ctxPtr, T0, T1)           \
-    {                                           \
-    Skein_Set_T0(ctxPtr, (T0));                  \
-    Skein_Set_T1(ctxPtr, (T1));                  \
-    }
+	{                                           \
+	Skein_Set_T0(ctxPtr, (T0));                  \
+	Skein_Set_T1(ctxPtr, (T1));                  \
+	}
 
 #define Skein_Set_Type(ctxPtr, BLK_TYPE)         \
-    Skein_Set_T1(ctxPtr, SKEIN_T1_BLK_TYPE_##BLK_TYPE)
+	Skein_Set_T1(ctxPtr, SKEIN_T1_BLK_TYPE_##BLK_TYPE)
 
 /* set up for starting with a new type: h.T[0]=0; h.T[1] = NEW_TYPE; h.bCnt=0; */
 #define Skein_Start_New_Type(ctxPtr, BLK_TYPE)   \
-    { Skein_Set_T0_T1(ctxPtr, 0, SKEIN_T1_FLAG_FIRST | SKEIN_T1_BLK_TYPE_##BLK_TYPE); (ctxPtr)->h.bCnt = 0; }
+	{ Skein_Set_T0_T1(ctxPtr, 0, SKEIN_T1_FLAG_FIRST | SKEIN_T1_BLK_TYPE_##BLK_TYPE); (ctxPtr)->h.bCnt = 0; }
 
 #define Skein_Clear_First_Flag(hdr)      { (hdr).T[1] &= ~SKEIN_T1_FLAG_FIRST;       }
 #define Skein_Set_Bit_Pad_Flag(hdr)      { (hdr).T[1] |=  SKEIN_T1_FLAG_BIT_PAD;     }
@@ -270,37 +270,37 @@ int  Skein1024_Output(struct skein1024_ctx *ctx, u8 *hashVal);
 ** Skein block function constants (shared across Ref and Opt code)
 ******************************************************************/
 enum    
-    {   
-        /* Skein_256 round rotation constants */
-    R_256_0_0 = 14, R_256_0_1 = 16,
-    R_256_1_0 = 52, R_256_1_1 = 57,
-    R_256_2_0 = 23, R_256_2_1 = 40,
-    R_256_3_0 =  5, R_256_3_1 = 37,
-    R_256_4_0 = 25, R_256_4_1 = 33,
-    R_256_5_0 = 46, R_256_5_1 = 12,
-    R_256_6_0 = 58, R_256_6_1 = 22,
-    R_256_7_0 = 32, R_256_7_1 = 32,
-
-        /* Skein_512 round rotation constants */
-    R_512_0_0 = 46, R_512_0_1 = 36, R_512_0_2 = 19, R_512_0_3 = 37,
-    R_512_1_0 = 33, R_512_1_1 = 27, R_512_1_2 = 14, R_512_1_3 = 42,
-    R_512_2_0 = 17, R_512_2_1 = 49, R_512_2_2 = 36, R_512_2_3 = 39,
-    R_512_3_0 = 44, R_512_3_1 =  9, R_512_3_2 = 54, R_512_3_3 = 56,
-    R_512_4_0 = 39, R_512_4_1 = 30, R_512_4_2 = 34, R_512_4_3 = 24,
-    R_512_5_0 = 13, R_512_5_1 = 50, R_512_5_2 = 10, R_512_5_3 = 17,
-    R_512_6_0 = 25, R_512_6_1 = 29, R_512_6_2 = 39, R_512_6_3 = 43,
-    R_512_7_0 =  8, R_512_7_1 = 35, R_512_7_2 = 56, R_512_7_3 = 22,
-
-        /* Skein1024 round rotation constants */
-    R1024_0_0 = 24, R1024_0_1 = 13, R1024_0_2 =  8, R1024_0_3 = 47, R1024_0_4 =  8, R1024_0_5 = 17, R1024_0_6 = 22, R1024_0_7 = 37,
-    R1024_1_0 = 38, R1024_1_1 = 19, R1024_1_2 = 10, R1024_1_3 = 55, R1024_1_4 = 49, R1024_1_5 = 18, R1024_1_6 = 23, R1024_1_7 = 52,
-    R1024_2_0 = 33, R1024_2_1 =  4, R1024_2_2 = 51, R1024_2_3 = 13, R1024_2_4 = 34, R1024_2_5 = 41, R1024_2_6 = 59, R1024_2_7 = 17,
-    R1024_3_0 =  5, R1024_3_1 = 20, R1024_3_2 = 48, R1024_3_3 = 41, R1024_3_4 = 47, R1024_3_5 = 28, R1024_3_6 = 16, R1024_3_7 = 25,
-    R1024_4_0 = 41, R1024_4_1 =  9, R1024_4_2 = 37, R1024_4_3 = 31, R1024_4_4 = 12, R1024_4_5 = 47, R1024_4_6 = 44, R1024_4_7 = 30,
-    R1024_5_0 = 16, R1024_5_1 = 34, R1024_5_2 = 56, R1024_5_3 = 51, R1024_5_4 =  4, R1024_5_5 = 53, R1024_5_6 = 42, R1024_5_7 = 41,
-    R1024_6_0 = 31, R1024_6_1 = 44, R1024_6_2 = 47, R1024_6_3 = 46, R1024_6_4 = 19, R1024_6_5 = 42, R1024_6_6 = 44, R1024_6_7 = 25,
-    R1024_7_0 =  9, R1024_7_1 = 48, R1024_7_2 = 35, R1024_7_3 = 52, R1024_7_4 = 23, R1024_7_5 = 31, R1024_7_6 = 37, R1024_7_7 = 20
-    };
+	{   
+	    /* Skein_256 round rotation constants */
+	R_256_0_0 = 14, R_256_0_1 = 16,
+	R_256_1_0 = 52, R_256_1_1 = 57,
+	R_256_2_0 = 23, R_256_2_1 = 40,
+	R_256_3_0 =  5, R_256_3_1 = 37,
+	R_256_4_0 = 25, R_256_4_1 = 33,
+	R_256_5_0 = 46, R_256_5_1 = 12,
+	R_256_6_0 = 58, R_256_6_1 = 22,
+	R_256_7_0 = 32, R_256_7_1 = 32,
+
+	    /* Skein_512 round rotation constants */
+	R_512_0_0 = 46, R_512_0_1 = 36, R_512_0_2 = 19, R_512_0_3 = 37,
+	R_512_1_0 = 33, R_512_1_1 = 27, R_512_1_2 = 14, R_512_1_3 = 42,
+	R_512_2_0 = 17, R_512_2_1 = 49, R_512_2_2 = 36, R_512_2_3 = 39,
+	R_512_3_0 = 44, R_512_3_1 =  9, R_512_3_2 = 54, R_512_3_3 = 56,
+	R_512_4_0 = 39, R_512_4_1 = 30, R_512_4_2 = 34, R_512_4_3 = 24,
+	R_512_5_0 = 13, R_512_5_1 = 50, R_512_5_2 = 10, R_512_5_3 = 17,
+	R_512_6_0 = 25, R_512_6_1 = 29, R_512_6_2 = 39, R_512_6_3 = 43,
+	R_512_7_0 =  8, R_512_7_1 = 35, R_512_7_2 = 56, R_512_7_3 = 22,
+
+	    /* Skein1024 round rotation constants */
+	R1024_0_0 = 24, R1024_0_1 = 13, R1024_0_2 =  8, R1024_0_3 = 47, R1024_0_4 =  8, R1024_0_5 = 17, R1024_0_6 = 22, R1024_0_7 = 37,
+	R1024_1_0 = 38, R1024_1_1 = 19, R1024_1_2 = 10, R1024_1_3 = 55, R1024_1_4 = 49, R1024_1_5 = 18, R1024_1_6 = 23, R1024_1_7 = 52,
+	R1024_2_0 = 33, R1024_2_1 =  4, R1024_2_2 = 51, R1024_2_3 = 13, R1024_2_4 = 34, R1024_2_5 = 41, R1024_2_6 = 59, R1024_2_7 = 17,
+	R1024_3_0 =  5, R1024_3_1 = 20, R1024_3_2 = 48, R1024_3_3 = 41, R1024_3_4 = 47, R1024_3_5 = 28, R1024_3_6 = 16, R1024_3_7 = 25,
+	R1024_4_0 = 41, R1024_4_1 =  9, R1024_4_2 = 37, R1024_4_3 = 31, R1024_4_4 = 12, R1024_4_5 = 47, R1024_4_6 = 44, R1024_4_7 = 30,
+	R1024_5_0 = 16, R1024_5_1 = 34, R1024_5_2 = 56, R1024_5_3 = 51, R1024_5_4 =  4, R1024_5_5 = 53, R1024_5_6 = 42, R1024_5_7 = 41,
+	R1024_6_0 = 31, R1024_6_1 = 44, R1024_6_2 = 47, R1024_6_3 = 46, R1024_6_4 = 19, R1024_6_5 = 42, R1024_6_6 = 44, R1024_6_7 = 25,
+	R1024_7_0 =  9, R1024_7_1 = 48, R1024_7_2 = 35, R1024_7_3 = 52, R1024_7_4 = 23, R1024_7_5 = 31, R1024_7_6 = 37, R1024_7_7 = 20
+	};
 
 #ifndef SKEIN_ROUNDS
 #define SKEIN_256_ROUNDS_TOTAL (72)          /* number of rounds for the different block sizes */
diff --git a/drivers/staging/skein/include/skeinApi.h b/drivers/staging/skein/include/skeinApi.h
index 2c52797918cf..0d7d59eff460 100644
--- a/drivers/staging/skein/include/skeinApi.h
+++ b/drivers/staging/skein/include/skeinApi.h
@@ -81,148 +81,148 @@ OTHER DEALINGS IN THE SOFTWARE.
 #include <linux/types.h>
 #include <skein.h>
 
-    /**
-     * Which Skein size to use
-     */
-    enum skein_size {
-        Skein256 = 256,     /*!< Skein with 256 bit state */
-        Skein512 = 512,     /*!< Skein with 512 bit state */
-        Skein1024 = 1024    /*!< Skein with 1024 bit state */
-    };
-
-    /**
-     * Context for Skein.
-     *
-     * This structure was setup with some know-how of the internal
-     * Skein structures, in particular ordering of header and size dependent
-     * variables. If Skein implementation changes this, then adapt these
-     * structures as well.
-     */
-    struct skein_ctx {
-        u64 skeinSize;
-        u64  XSave[SKEIN_MAX_STATE_WORDS];   /* save area for state variables */
-        union {
-            struct skein_ctx_hdr h;
-            struct skein_256_ctx s256;
-            struct skein_512_ctx s512;
-            struct skein1024_ctx s1024;
-        } m;
-    };
-
-    /**
-     * Prepare a Skein context.
-     * 
-     * An application must call this function before it can use the Skein
-     * context. The functions clears memory and initializes size dependent
-     * variables.
-     *
-     * @param ctx
-     *     Pointer to a Skein context.
-     * @param size
-     *     Which Skein size to use.
-     * @return
-     *     SKEIN_SUCESS of SKEIN_FAIL
-     */
-    int skeinCtxPrepare(struct skein_ctx *ctx, enum skein_size size);
-
-    /**
-     * Initialize a Skein context.
-     *
-     * Initializes the context with this data and saves the resulting Skein 
-     * state variables for further use.
-     *
-     * @param ctx
-     *     Pointer to a Skein context.
-     * @param hashBitLen
-     *     Number of MAC hash bits to compute
-     * @return
-     *     SKEIN_SUCESS of SKEIN_FAIL
-     * @see skeinReset
-     */
-    int skeinInit(struct skein_ctx *ctx, size_t hashBitLen);
-
-    /**
-     * Resets a Skein context for further use.
-     * 
-     * Restores the saved chaining variables to reset the Skein context. 
-     * Thus applications can reuse the same setup to  process several 
-     * messages. This saves a complete Skein initialization cycle.
-     * 
-     * @param ctx
-     *     Pointer to a pre-initialized Skein MAC context
-     */
-    void skeinReset(struct skein_ctx *ctx);
-    
-    /**
-     * Initializes a Skein context for MAC usage.
-     * 
-     * Initializes the context with this data and saves the resulting Skein 
-     * state variables for further use.
-     *
-     * Applications call the normal Skein functions to update the MAC and
-     * get the final result.
-     *
-     * @param ctx
-     *     Pointer to an empty or preinitialized Skein MAC context
-     * @param key
-     *     Pointer to key bytes or NULL
-     * @param keyLen
-     *     Length of the key in bytes or zero
-     * @param hashBitLen
-     *     Number of MAC hash bits to compute
-     * @return
-     *     SKEIN_SUCESS of SKEIN_FAIL
-     */
-    int skeinMacInit(struct skein_ctx *ctx, const u8 *key, size_t keyLen,
-                     size_t hashBitLen);
-
-    /**
-     * Update Skein with the next part of the message.
-     *
-     * @param ctx
-     *     Pointer to initialized Skein context
-     * @param msg
-     *     Pointer to the message.
-     * @param msgByteCnt
-     *     Length of the message in @b bytes
-     * @return
-     *     Success or error code.
-     */
-    int skeinUpdate(struct skein_ctx *ctx, const u8 *msg,
-                    size_t msgByteCnt);
-
-    /**
-     * Update the hash with a message bit string.
-     *
-     * Skein can handle data not only as bytes but also as bit strings of
-     * arbitrary length (up to its maximum design size).
-     *
-     * @param ctx
-     *     Pointer to initialized Skein context
-     * @param msg
-     *     Pointer to the message.
-     * @param msgBitCnt
-     *     Length of the message in @b bits.
-     */
-    int skeinUpdateBits(struct skein_ctx *ctx, const u8 *msg,
-                        size_t msgBitCnt);
-
-    /**
-     * Finalize Skein and return the hash.
-     * 
-     * Before an application can reuse a Skein setup the application must
-     * reset the Skein context.
-     *
-     * @param ctx
-     *     Pointer to initialized Skein context
-     * @param hash
-     *     Pointer to buffer that receives the hash. The buffer must be large
-     *     enough to store @c hashBitLen bits.
-     * @return
-     *     Success or error code.
-     * @see skeinReset
-     */
-    int skeinFinal(struct skein_ctx *ctx, u8 *hash);
+/**
+ * Which Skein size to use
+ */
+enum skein_size {
+	Skein256 = 256,     /*!< Skein with 256 bit state */
+	Skein512 = 512,     /*!< Skein with 512 bit state */
+	Skein1024 = 1024    /*!< Skein with 1024 bit state */
+};
+
+/**
+ * Context for Skein.
+ *
+ * This structure was setup with some know-how of the internal
+ * Skein structures, in particular ordering of header and size dependent
+ * variables. If Skein implementation changes this, then adapt these
+ * structures as well.
+ */
+struct skein_ctx {
+	u64 skeinSize;
+	u64  XSave[SKEIN_MAX_STATE_WORDS];   /* save area for state variables */
+	union {
+		struct skein_ctx_hdr h;
+		struct skein_256_ctx s256;
+		struct skein_512_ctx s512;
+		struct skein1024_ctx s1024;
+	} m;
+};
+
+/**
+ * Prepare a Skein context.
+ * 
+ * An application must call this function before it can use the Skein
+ * context. The functions clears memory and initializes size dependent
+ * variables.
+ *
+ * @param ctx
+ *     Pointer to a Skein context.
+ * @param size
+ *     Which Skein size to use.
+ * @return
+ *     SKEIN_SUCESS of SKEIN_FAIL
+ */
+int skeinCtxPrepare(struct skein_ctx *ctx, enum skein_size size);
+
+/**
+ * Initialize a Skein context.
+ *
+ * Initializes the context with this data and saves the resulting Skein 
+ * state variables for further use.
+ *
+ * @param ctx
+ *     Pointer to a Skein context.
+ * @param hashBitLen
+ *     Number of MAC hash bits to compute
+ * @return
+ *     SKEIN_SUCESS of SKEIN_FAIL
+ * @see skeinReset
+ */
+int skeinInit(struct skein_ctx *ctx, size_t hashBitLen);
+
+/**
+ * Resets a Skein context for further use.
+ * 
+ * Restores the saved chaining variables to reset the Skein context. 
+ * Thus applications can reuse the same setup to  process several 
+ * messages. This saves a complete Skein initialization cycle.
+ * 
+ * @param ctx
+ *     Pointer to a pre-initialized Skein MAC context
+ */
+void skeinReset(struct skein_ctx *ctx);
+
+/**
+ * Initializes a Skein context for MAC usage.
+ * 
+ * Initializes the context with this data and saves the resulting Skein 
+ * state variables for further use.
+ *
+ * Applications call the normal Skein functions to update the MAC and
+ * get the final result.
+ *
+ * @param ctx
+ *     Pointer to an empty or preinitialized Skein MAC context
+ * @param key
+ *     Pointer to key bytes or NULL
+ * @param keyLen
+ *     Length of the key in bytes or zero
+ * @param hashBitLen
+ *     Number of MAC hash bits to compute
+ * @return
+ *     SKEIN_SUCESS of SKEIN_FAIL
+ */
+int skeinMacInit(struct skein_ctx *ctx, const u8 *key, size_t keyLen,
+		 size_t hashBitLen);
+
+/**
+ * Update Skein with the next part of the message.
+ *
+ * @param ctx
+ *     Pointer to initialized Skein context
+ * @param msg
+ *     Pointer to the message.
+ * @param msgByteCnt
+ *     Length of the message in @b bytes
+ * @return
+ *     Success or error code.
+ */
+int skeinUpdate(struct skein_ctx *ctx, const u8 *msg,
+		size_t msgByteCnt);
+
+/**
+ * Update the hash with a message bit string.
+ *
+ * Skein can handle data not only as bytes but also as bit strings of
+ * arbitrary length (up to its maximum design size).
+ *
+ * @param ctx
+ *     Pointer to initialized Skein context
+ * @param msg
+ *     Pointer to the message.
+ * @param msgBitCnt
+ *     Length of the message in @b bits.
+ */
+int skeinUpdateBits(struct skein_ctx *ctx, const u8 *msg,
+		    size_t msgBitCnt);
+
+/**
+ * Finalize Skein and return the hash.
+ * 
+ * Before an application can reuse a Skein setup the application must
+ * reset the Skein context.
+ *
+ * @param ctx
+ *     Pointer to initialized Skein context
+ * @param hash
+ *     Pointer to buffer that receives the hash. The buffer must be large
+ *     enough to store @c hashBitLen bits.
+ * @return
+ *     Success or error code.
+ * @see skeinReset
+ */
+int skeinFinal(struct skein_ctx *ctx, u8 *hash);
 
 /**
  * @}
diff --git a/drivers/staging/skein/include/skein_iv.h b/drivers/staging/skein/include/skein_iv.h
index 813bad528e3c..bbbba77c44d3 100644
--- a/drivers/staging/skein/include/skein_iv.h
+++ b/drivers/staging/skein/include/skein_iv.h
@@ -21,179 +21,179 @@
 
 /* blkSize =  256 bits. hashSize =  128 bits */
 const u64 SKEIN_256_IV_128[] =
-    {
-    MK_64(0xE1111906, 0x964D7260),
-    MK_64(0x883DAAA7, 0x7C8D811C),
-    MK_64(0x10080DF4, 0x91960F7A),
-    MK_64(0xCCF7DDE5, 0xB45BC1C2)
-    };
+	{
+	MK_64(0xE1111906, 0x964D7260),
+	MK_64(0x883DAAA7, 0x7C8D811C),
+	MK_64(0x10080DF4, 0x91960F7A),
+	MK_64(0xCCF7DDE5, 0xB45BC1C2)
+	};
 
 /* blkSize =  256 bits. hashSize =  160 bits */
 const u64 SKEIN_256_IV_160[] =
-    {
-    MK_64(0x14202314, 0x72825E98),
-    MK_64(0x2AC4E9A2, 0x5A77E590),
-    MK_64(0xD47A5856, 0x8838D63E),
-    MK_64(0x2DD2E496, 0x8586AB7D)
-    };
+	{
+	MK_64(0x14202314, 0x72825E98),
+	MK_64(0x2AC4E9A2, 0x5A77E590),
+	MK_64(0xD47A5856, 0x8838D63E),
+	MK_64(0x2DD2E496, 0x8586AB7D)
+	};
 
 /* blkSize =  256 bits. hashSize =  224 bits */
 const u64 SKEIN_256_IV_224[] =
-    {
-    MK_64(0xC6098A8C, 0x9AE5EA0B),
-    MK_64(0x876D5686, 0x08C5191C),
-    MK_64(0x99CB88D7, 0xD7F53884),
-    MK_64(0x384BDDB1, 0xAEDDB5DE)
-    };
+	{
+	MK_64(0xC6098A8C, 0x9AE5EA0B),
+	MK_64(0x876D5686, 0x08C5191C),
+	MK_64(0x99CB88D7, 0xD7F53884),
+	MK_64(0x384BDDB1, 0xAEDDB5DE)
+	};
 
 /* blkSize =  256 bits. hashSize =  256 bits */
 const u64 SKEIN_256_IV_256[] =
-    {
-    MK_64(0xFC9DA860, 0xD048B449),
-    MK_64(0x2FCA6647, 0x9FA7D833),
-    MK_64(0xB33BC389, 0x6656840F),
-    MK_64(0x6A54E920, 0xFDE8DA69)
-    };
+	{
+	MK_64(0xFC9DA860, 0xD048B449),
+	MK_64(0x2FCA6647, 0x9FA7D833),
+	MK_64(0xB33BC389, 0x6656840F),
+	MK_64(0x6A54E920, 0xFDE8DA69)
+	};
 
 /* blkSize =  512 bits. hashSize =  128 bits */
 const u64 SKEIN_512_IV_128[] =
-    {
-    MK_64(0xA8BC7BF3, 0x6FBF9F52),
-    MK_64(0x1E9872CE, 0xBD1AF0AA),
-    MK_64(0x309B1790, 0xB32190D3),
-    MK_64(0xBCFBB854, 0x3F94805C),
-    MK_64(0x0DA61BCD, 0x6E31B11B),
-    MK_64(0x1A18EBEA, 0xD46A32E3),
-    MK_64(0xA2CC5B18, 0xCE84AA82),
-    MK_64(0x6982AB28, 0x9D46982D)
-    };
+	{
+	MK_64(0xA8BC7BF3, 0x6FBF9F52),
+	MK_64(0x1E9872CE, 0xBD1AF0AA),
+	MK_64(0x309B1790, 0xB32190D3),
+	MK_64(0xBCFBB854, 0x3F94805C),
+	MK_64(0x0DA61BCD, 0x6E31B11B),
+	MK_64(0x1A18EBEA, 0xD46A32E3),
+	MK_64(0xA2CC5B18, 0xCE84AA82),
+	MK_64(0x6982AB28, 0x9D46982D)
+	};
 
 /* blkSize =  512 bits. hashSize =  160 bits */
 const u64 SKEIN_512_IV_160[] =
-    {
-    MK_64(0x28B81A2A, 0xE013BD91),
-    MK_64(0xC2F11668, 0xB5BDF78F),
-    MK_64(0x1760D8F3, 0xF6A56F12),
-    MK_64(0x4FB74758, 0x8239904F),
-    MK_64(0x21EDE07F, 0x7EAF5056),
-    MK_64(0xD908922E, 0x63ED70B8),
-    MK_64(0xB8EC76FF, 0xECCB52FA),
-    MK_64(0x01A47BB8, 0xA3F27A6E)
-    };
+	{
+	MK_64(0x28B81A2A, 0xE013BD91),
+	MK_64(0xC2F11668, 0xB5BDF78F),
+	MK_64(0x1760D8F3, 0xF6A56F12),
+	MK_64(0x4FB74758, 0x8239904F),
+	MK_64(0x21EDE07F, 0x7EAF5056),
+	MK_64(0xD908922E, 0x63ED70B8),
+	MK_64(0xB8EC76FF, 0xECCB52FA),
+	MK_64(0x01A47BB8, 0xA3F27A6E)
+	};
 
 /* blkSize =  512 bits. hashSize =  224 bits */
 const u64 SKEIN_512_IV_224[] =
-    {
-    MK_64(0xCCD06162, 0x48677224),
-    MK_64(0xCBA65CF3, 0xA92339EF),
-    MK_64(0x8CCD69D6, 0x52FF4B64),
-    MK_64(0x398AED7B, 0x3AB890B4),
-    MK_64(0x0F59D1B1, 0x457D2BD0),
-    MK_64(0x6776FE65, 0x75D4EB3D),
-    MK_64(0x99FBC70E, 0x997413E9),
-    MK_64(0x9E2CFCCF, 0xE1C41EF7)
-    };
+	{
+	MK_64(0xCCD06162, 0x48677224),
+	MK_64(0xCBA65CF3, 0xA92339EF),
+	MK_64(0x8CCD69D6, 0x52FF4B64),
+	MK_64(0x398AED7B, 0x3AB890B4),
+	MK_64(0x0F59D1B1, 0x457D2BD0),
+	MK_64(0x6776FE65, 0x75D4EB3D),
+	MK_64(0x99FBC70E, 0x997413E9),
+	MK_64(0x9E2CFCCF, 0xE1C41EF7)
+	};
 
 /* blkSize =  512 bits. hashSize =  256 bits */
 const u64 SKEIN_512_IV_256[] =
-    {
-    MK_64(0xCCD044A1, 0x2FDB3E13),
-    MK_64(0xE8359030, 0x1A79A9EB),
-    MK_64(0x55AEA061, 0x4F816E6F),
-    MK_64(0x2A2767A4, 0xAE9B94DB),
-    MK_64(0xEC06025E, 0x74DD7683),
-    MK_64(0xE7A436CD, 0xC4746251),
-    MK_64(0xC36FBAF9, 0x393AD185),
-    MK_64(0x3EEDBA18, 0x33EDFC13)
-    };
+	{
+	MK_64(0xCCD044A1, 0x2FDB3E13),
+	MK_64(0xE8359030, 0x1A79A9EB),
+	MK_64(0x55AEA061, 0x4F816E6F),
+	MK_64(0x2A2767A4, 0xAE9B94DB),
+	MK_64(0xEC06025E, 0x74DD7683),
+	MK_64(0xE7A436CD, 0xC4746251),
+	MK_64(0xC36FBAF9, 0x393AD185),
+	MK_64(0x3EEDBA18, 0x33EDFC13)
+	};
 
 /* blkSize =  512 bits. hashSize =  384 bits */
 const u64 SKEIN_512_IV_384[] =
-    {
-    MK_64(0xA3F6C6BF, 0x3A75EF5F),
-    MK_64(0xB0FEF9CC, 0xFD84FAA4),
-    MK_64(0x9D77DD66, 0x3D770CFE),
-    MK_64(0xD798CBF3, 0xB468FDDA),
-    MK_64(0x1BC4A666, 0x8A0E4465),
-    MK_64(0x7ED7D434, 0xE5807407),
-    MK_64(0x548FC1AC, 0xD4EC44D6),
-    MK_64(0x266E1754, 0x6AA18FF8)
-    };
+	{
+	MK_64(0xA3F6C6BF, 0x3A75EF5F),
+	MK_64(0xB0FEF9CC, 0xFD84FAA4),
+	MK_64(0x9D77DD66, 0x3D770CFE),
+	MK_64(0xD798CBF3, 0xB468FDDA),
+	MK_64(0x1BC4A666, 0x8A0E4465),
+	MK_64(0x7ED7D434, 0xE5807407),
+	MK_64(0x548FC1AC, 0xD4EC44D6),
+	MK_64(0x266E1754, 0x6AA18FF8)
+	};
 
 /* blkSize =  512 bits. hashSize =  512 bits */
 const u64 SKEIN_512_IV_512[] =
-    {
-    MK_64(0x4903ADFF, 0x749C51CE),
-    MK_64(0x0D95DE39, 0x9746DF03),
-    MK_64(0x8FD19341, 0x27C79BCE),
-    MK_64(0x9A255629, 0xFF352CB1),
-    MK_64(0x5DB62599, 0xDF6CA7B0),
-    MK_64(0xEABE394C, 0xA9D5C3F4),
-    MK_64(0x991112C7, 0x1A75B523),
-    MK_64(0xAE18A40B, 0x660FCC33)
-    };
+	{
+	MK_64(0x4903ADFF, 0x749C51CE),
+	MK_64(0x0D95DE39, 0x9746DF03),
+	MK_64(0x8FD19341, 0x27C79BCE),
+	MK_64(0x9A255629, 0xFF352CB1),
+	MK_64(0x5DB62599, 0xDF6CA7B0),
+	MK_64(0xEABE394C, 0xA9D5C3F4),
+	MK_64(0x991112C7, 0x1A75B523),
+	MK_64(0xAE18A40B, 0x660FCC33)
+	};
 
 /* blkSize = 1024 bits. hashSize =  384 bits */
 const u64 SKEIN1024_IV_384[] =
-    {
-    MK_64(0x5102B6B8, 0xC1894A35),
-    MK_64(0xFEEBC9E3, 0xFE8AF11A),
-    MK_64(0x0C807F06, 0xE32BED71),
-    MK_64(0x60C13A52, 0xB41A91F6),
-    MK_64(0x9716D35D, 0xD4917C38),
-    MK_64(0xE780DF12, 0x6FD31D3A),
-    MK_64(0x797846B6, 0xC898303A),
-    MK_64(0xB172C2A8, 0xB3572A3B),
-    MK_64(0xC9BC8203, 0xA6104A6C),
-    MK_64(0x65909338, 0xD75624F4),
-    MK_64(0x94BCC568, 0x4B3F81A0),
-    MK_64(0x3EBBF51E, 0x10ECFD46),
-    MK_64(0x2DF50F0B, 0xEEB08542),
-    MK_64(0x3B5A6530, 0x0DBC6516),
-    MK_64(0x484B9CD2, 0x167BBCE1),
-    MK_64(0x2D136947, 0xD4CBAFEA)
-    };
+	{
+	MK_64(0x5102B6B8, 0xC1894A35),
+	MK_64(0xFEEBC9E3, 0xFE8AF11A),
+	MK_64(0x0C807F06, 0xE32BED71),
+	MK_64(0x60C13A52, 0xB41A91F6),
+	MK_64(0x9716D35D, 0xD4917C38),
+	MK_64(0xE780DF12, 0x6FD31D3A),
+	MK_64(0x797846B6, 0xC898303A),
+	MK_64(0xB172C2A8, 0xB3572A3B),
+	MK_64(0xC9BC8203, 0xA6104A6C),
+	MK_64(0x65909338, 0xD75624F4),
+	MK_64(0x94BCC568, 0x4B3F81A0),
+	MK_64(0x3EBBF51E, 0x10ECFD46),
+	MK_64(0x2DF50F0B, 0xEEB08542),
+	MK_64(0x3B5A6530, 0x0DBC6516),
+	MK_64(0x484B9CD2, 0x167BBCE1),
+	MK_64(0x2D136947, 0xD4CBAFEA)
+	};
 
 /* blkSize = 1024 bits. hashSize =  512 bits */
 const u64 SKEIN1024_IV_512[] =
-    {
-    MK_64(0xCAEC0E5D, 0x7C1B1B18),
-    MK_64(0xA01B0E04, 0x5F03E802),
-    MK_64(0x33840451, 0xED912885),
-    MK_64(0x374AFB04, 0xEAEC2E1C),
-    MK_64(0xDF25A0E2, 0x813581F7),
-    MK_64(0xE4004093, 0x8B12F9D2),
-    MK_64(0xA662D539, 0xC2ED39B6),
-    MK_64(0xFA8B85CF, 0x45D8C75A),
-    MK_64(0x8316ED8E, 0x29EDE796),
-    MK_64(0x053289C0, 0x2E9F91B8),
-    MK_64(0xC3F8EF1D, 0x6D518B73),
-    MK_64(0xBDCEC3C4, 0xD5EF332E),
-    MK_64(0x549A7E52, 0x22974487),
-    MK_64(0x67070872, 0x5B749816),
-    MK_64(0xB9CD28FB, 0xF0581BD1),
-    MK_64(0x0E2940B8, 0x15804974)
-    };
+	{
+	MK_64(0xCAEC0E5D, 0x7C1B1B18),
+	MK_64(0xA01B0E04, 0x5F03E802),
+	MK_64(0x33840451, 0xED912885),
+	MK_64(0x374AFB04, 0xEAEC2E1C),
+	MK_64(0xDF25A0E2, 0x813581F7),
+	MK_64(0xE4004093, 0x8B12F9D2),
+	MK_64(0xA662D539, 0xC2ED39B6),
+	MK_64(0xFA8B85CF, 0x45D8C75A),
+	MK_64(0x8316ED8E, 0x29EDE796),
+	MK_64(0x053289C0, 0x2E9F91B8),
+	MK_64(0xC3F8EF1D, 0x6D518B73),
+	MK_64(0xBDCEC3C4, 0xD5EF332E),
+	MK_64(0x549A7E52, 0x22974487),
+	MK_64(0x67070872, 0x5B749816),
+	MK_64(0xB9CD28FB, 0xF0581BD1),
+	MK_64(0x0E2940B8, 0x15804974)
+	};
 
 /* blkSize = 1024 bits. hashSize = 1024 bits */
 const u64 SKEIN1024_IV_1024[] =
-    {
-    MK_64(0xD593DA07, 0x41E72355),
-    MK_64(0x15B5E511, 0xAC73E00C),
-    MK_64(0x5180E5AE, 0xBAF2C4F0),
-    MK_64(0x03BD41D3, 0xFCBCAFAF),
-    MK_64(0x1CAEC6FD, 0x1983A898),
-    MK_64(0x6E510B8B, 0xCDD0589F),
-    MK_64(0x77E2BDFD, 0xC6394ADA),
-    MK_64(0xC11E1DB5, 0x24DCB0A3),
-    MK_64(0xD6D14AF9, 0xC6329AB5),
-    MK_64(0x6A9B0BFC, 0x6EB67E0D),
-    MK_64(0x9243C60D, 0xCCFF1332),
-    MK_64(0x1A1F1DDE, 0x743F02D4),
-    MK_64(0x0996753C, 0x10ED0BB8),
-    MK_64(0x6572DD22, 0xF2B4969A),
-    MK_64(0x61FD3062, 0xD00A579A),
-    MK_64(0x1DE0536E, 0x8682E539)
-    };
+	{
+	MK_64(0xD593DA07, 0x41E72355),
+	MK_64(0x15B5E511, 0xAC73E00C),
+	MK_64(0x5180E5AE, 0xBAF2C4F0),
+	MK_64(0x03BD41D3, 0xFCBCAFAF),
+	MK_64(0x1CAEC6FD, 0x1983A898),
+	MK_64(0x6E510B8B, 0xCDD0589F),
+	MK_64(0x77E2BDFD, 0xC6394ADA),
+	MK_64(0xC11E1DB5, 0x24DCB0A3),
+	MK_64(0xD6D14AF9, 0xC6329AB5),
+	MK_64(0x6A9B0BFC, 0x6EB67E0D),
+	MK_64(0x9243C60D, 0xCCFF1332),
+	MK_64(0x1A1F1DDE, 0x743F02D4),
+	MK_64(0x0996753C, 0x10ED0BB8),
+	MK_64(0x6572DD22, 0xF2B4969A),
+	MK_64(0x61FD3062, 0xD00A579A),
+	MK_64(0x1DE0536E, 0x8682E539)
+	};
 
 #endif /* _SKEIN_IV_H_ */
diff --git a/drivers/staging/skein/include/threefishApi.h b/drivers/staging/skein/include/threefishApi.h
index 1f9e6e14f50b..199257e37813 100644
--- a/drivers/staging/skein/include/threefishApi.h
+++ b/drivers/staging/skein/include/threefishApi.h
@@ -33,125 +33,125 @@
 
 #define KeyScheduleConst 0x1BD11BDAA9FC1A22L
 
-    /**
-     * Which Threefish size to use
-     */
-    enum threefish_size {
-        Threefish256 = 256,     /*!< Skein with 256 bit state */
-        Threefish512 = 512,     /*!< Skein with 512 bit state */
-        Threefish1024 = 1024    /*!< Skein with 1024 bit state */
-    };
-    
-    /**
-     * Context for Threefish key and tweak words.
-     * 
-     * This structure was setup with some know-how of the internal
-     * Skein structures, in particular ordering of header and size dependent
-     * variables. If Skein implementation changes this, the adapt these
-     * structures as well.
-     */
-    struct threefish_key {
-        u64 stateSize;
-        u64 key[SKEIN_MAX_STATE_WORDS+1];   /* max number of key words*/
-        u64 tweak[3];
-    };
+/**
+ * Which Threefish size to use
+ */
+enum threefish_size {
+	Threefish256 = 256,     /*!< Skein with 256 bit state */
+	Threefish512 = 512,     /*!< Skein with 512 bit state */
+	Threefish1024 = 1024    /*!< Skein with 1024 bit state */
+};
+
+/**
+ * Context for Threefish key and tweak words.
+ * 
+ * This structure was setup with some know-how of the internal
+ * Skein structures, in particular ordering of header and size dependent
+ * variables. If Skein implementation changes this, the adapt these
+ * structures as well.
+ */
+struct threefish_key {
+	u64 stateSize;
+	u64 key[SKEIN_MAX_STATE_WORDS+1];   /* max number of key words*/
+	u64 tweak[3];
+};
+
+/**
+ * Set Threefish key and tweak data.
+ * 
+ * This function sets the key and tweak data for the Threefish cipher of
+ * the given size. The key data must have the same length (number of bits)
+ * as the state size 
+ *
+ * @param keyCtx
+ *     Pointer to a Threefish key structure.
+ * @param size
+ *     Which Skein size to use.
+ * @param keyData
+ *     Pointer to the key words (word has 64 bits).
+ * @param tweak
+ *     Pointer to the two tweak words (word has 64 bits).
+ */
+void threefishSetKey(struct threefish_key *keyCtx, enum threefish_size stateSize, u64 *keyData, u64 *tweak);
+
+/**
+ * Encrypt Threefisch block (bytes).
+ * 
+ * The buffer must have at least the same length (number of bits) aas the 
+ * state size for this key. The function uses the first @c stateSize bits
+ * of the input buffer, encrypts them and stores the result in the output
+ * buffer.
+ * 
+ * @param keyCtx
+ *     Pointer to a Threefish key structure.
+ * @param in
+ *     Poionter to plaintext data buffer.
+ * @param out
+ *     Pointer to cipher buffer.
+ */
+void threefishEncryptBlockBytes(struct threefish_key *keyCtx, u8 *in, u8 *out);
 
-    /**
-     * Set Threefish key and tweak data.
-     * 
-     * This function sets the key and tweak data for the Threefish cipher of
-     * the given size. The key data must have the same length (number of bits)
-     * as the state size 
-     *
-     * @param keyCtx
-     *     Pointer to a Threefish key structure.
-     * @param size
-     *     Which Skein size to use.
-     * @param keyData
-     *     Pointer to the key words (word has 64 bits).
-     * @param tweak
-     *     Pointer to the two tweak words (word has 64 bits).
-     */
-    void threefishSetKey(struct threefish_key *keyCtx, enum threefish_size stateSize, u64 *keyData, u64 *tweak);
-    
-    /**
-     * Encrypt Threefisch block (bytes).
-     * 
-     * The buffer must have at least the same length (number of bits) aas the 
-     * state size for this key. The function uses the first @c stateSize bits
-     * of the input buffer, encrypts them and stores the result in the output
-     * buffer.
-     * 
-     * @param keyCtx
-     *     Pointer to a Threefish key structure.
-     * @param in
-     *     Poionter to plaintext data buffer.
-     * @param out
-     *     Pointer to cipher buffer.
-     */
-    void threefishEncryptBlockBytes(struct threefish_key *keyCtx, u8 *in, u8 *out);
-    
-    /**
-     * Encrypt Threefisch block (words).
-     * 
-     * The buffer must have at least the same length (number of bits) aas the 
-     * state size for this key. The function uses the first @c stateSize bits
-     * of the input buffer, encrypts them and stores the result in the output
-     * buffer.
-     * 
-     * The wordsize ist set to 64 bits.
-     * 
-     * @param keyCtx
-     *     Pointer to a Threefish key structure.
-     * @param in
-     *     Poionter to plaintext data buffer.
-     * @param out
-     *     Pointer to cipher buffer.
-     */
-    void threefishEncryptBlockWords(struct threefish_key *keyCtx, u64 *in, u64 *out);
+/**
+ * Encrypt Threefisch block (words).
+ * 
+ * The buffer must have at least the same length (number of bits) aas the 
+ * state size for this key. The function uses the first @c stateSize bits
+ * of the input buffer, encrypts them and stores the result in the output
+ * buffer.
+ * 
+ * The wordsize ist set to 64 bits.
+ * 
+ * @param keyCtx
+ *     Pointer to a Threefish key structure.
+ * @param in
+ *     Poionter to plaintext data buffer.
+ * @param out
+ *     Pointer to cipher buffer.
+ */
+void threefishEncryptBlockWords(struct threefish_key *keyCtx, u64 *in, u64 *out);
 
-    /**
-     * Decrypt Threefisch block (bytes).
-     * 
-     * The buffer must have at least the same length (number of bits) aas the 
-     * state size for this key. The function uses the first @c stateSize bits
-     * of the input buffer, decrypts them and stores the result in the output
-     * buffer
-     * 
-     * @param keyCtx
-     *     Pointer to a Threefish key structure.
-     * @param in
-     *     Poionter to cipher data buffer.
-     * @param out
-     *     Pointer to plaintext buffer.
-     */
-    void threefishDecryptBlockBytes(struct threefish_key *keyCtx, u8 *in, u8 *out);
+/**
+ * Decrypt Threefisch block (bytes).
+ * 
+ * The buffer must have at least the same length (number of bits) aas the 
+ * state size for this key. The function uses the first @c stateSize bits
+ * of the input buffer, decrypts them and stores the result in the output
+ * buffer
+ * 
+ * @param keyCtx
+ *     Pointer to a Threefish key structure.
+ * @param in
+ *     Poionter to cipher data buffer.
+ * @param out
+ *     Pointer to plaintext buffer.
+ */
+void threefishDecryptBlockBytes(struct threefish_key *keyCtx, u8 *in, u8 *out);
 
-    /**
-     * Decrypt Threefisch block (words).
-     * 
-     * The buffer must have at least the same length (number of bits) aas the 
-     * state size for this key. The function uses the first @c stateSize bits
-     * of the input buffer, encrypts them and stores the result in the output
-     * buffer.
-     * 
-     * The wordsize ist set to 64 bits.
-     * 
-     * @param keyCtx
-     *     Pointer to a Threefish key structure.
-     * @param in
-     *     Poionter to cipher data buffer.
-     * @param out
-     *     Pointer to plaintext buffer.
-     */
-    void threefishDecryptBlockWords(struct threefish_key *keyCtx, u64 *in, u64 *out);
+/**
+ * Decrypt Threefisch block (words).
+ * 
+ * The buffer must have at least the same length (number of bits) aas the 
+ * state size for this key. The function uses the first @c stateSize bits
+ * of the input buffer, encrypts them and stores the result in the output
+ * buffer.
+ * 
+ * The wordsize ist set to 64 bits.
+ * 
+ * @param keyCtx
+ *     Pointer to a Threefish key structure.
+ * @param in
+ *     Poionter to cipher data buffer.
+ * @param out
+ *     Pointer to plaintext buffer.
+ */
+void threefishDecryptBlockWords(struct threefish_key *keyCtx, u64 *in, u64 *out);
 
-    void threefishEncrypt256(struct threefish_key *keyCtx, u64 *input, u64 *output);
-    void threefishEncrypt512(struct threefish_key *keyCtx, u64 *input, u64 *output);
-    void threefishEncrypt1024(struct threefish_key *keyCtx, u64 *input, u64 *output);
-    void threefishDecrypt256(struct threefish_key *keyCtx, u64 *input, u64 *output);
-    void threefishDecrypt512(struct threefish_key *keyCtx, u64 *input, u64 *output);
-    void threefishDecrypt1024(struct threefish_key *keyCtx, u64 *input, u64 *output);
+void threefishEncrypt256(struct threefish_key *keyCtx, u64 *input, u64 *output);
+void threefishEncrypt512(struct threefish_key *keyCtx, u64 *input, u64 *output);
+void threefishEncrypt1024(struct threefish_key *keyCtx, u64 *input, u64 *output);
+void threefishDecrypt256(struct threefish_key *keyCtx, u64 *input, u64 *output);
+void threefishDecrypt512(struct threefish_key *keyCtx, u64 *input, u64 *output);
+void threefishDecrypt1024(struct threefish_key *keyCtx, u64 *input, u64 *output);
 /**
  * @}
  */
diff --git a/drivers/staging/skein/skein.c b/drivers/staging/skein/skein.c
index e2e5685157a0..3f0f32806181 100644
--- a/drivers/staging/skein/skein.c
+++ b/drivers/staging/skein/skein.c
@@ -28,49 +28,49 @@ void    Skein1024_Process_Block(struct skein1024_ctx *ctx, const u8 *blkPtr, siz
 /* init the context for a straight hashing operation  */
 int Skein_256_Init(struct skein_256_ctx *ctx, size_t hashBitLen)
 {
-    union
-    {
-        u8  b[SKEIN_256_STATE_BYTES];
-        u64  w[SKEIN_256_STATE_WORDS];
-    } cfg;                              /* config block */
-
-    Skein_Assert(hashBitLen > 0, SKEIN_BAD_HASHLEN);
-    ctx->h.hashBitLen = hashBitLen;         /* output hash bit count */
-
-    switch (hashBitLen)
-    {             /* use pre-computed values, where available */
-    case  256:
-        memcpy(ctx->X, SKEIN_256_IV_256, sizeof(ctx->X));
-        break;
-    case  224:
-        memcpy(ctx->X, SKEIN_256_IV_224, sizeof(ctx->X));
-        break;
-    case  160:
-        memcpy(ctx->X, SKEIN_256_IV_160, sizeof(ctx->X));
-        break;
-    case  128:
-        memcpy(ctx->X, SKEIN_256_IV_128, sizeof(ctx->X));
-        break;
-    default:
-        /* here if there is no precomputed IV value available */
-        /* build/process the config block, type == CONFIG (could be precomputed) */
-        Skein_Start_New_Type(ctx, CFG_FINAL);        /* set tweaks: T0=0; T1=CFG | FINAL */
-
-        cfg.w[0] = Skein_Swap64(SKEIN_SCHEMA_VER);  /* set the schema, version */
-        cfg.w[1] = Skein_Swap64(hashBitLen);        /* hash result length in bits */
-        cfg.w[2] = Skein_Swap64(SKEIN_CFG_TREE_INFO_SEQUENTIAL);
-        memset(&cfg.w[3], 0, sizeof(cfg) - 3*sizeof(cfg.w[0])); /* zero pad config block */
-
-        /* compute the initial chaining values from config block */
-        memset(ctx->X, 0, sizeof(ctx->X));            /* zero the chaining variables */
-        Skein_256_Process_Block(ctx, cfg.b, 1, SKEIN_CFG_STR_LEN);
-        break;
-    }
-    /* The chaining vars ctx->X are now initialized for the given hashBitLen. */
-    /* Set up to process the data message portion of the hash (default) */
-    Skein_Start_New_Type(ctx, MSG);              /* T0=0, T1= MSG type */
-
-    return SKEIN_SUCCESS;
+	union
+	{
+		u8  b[SKEIN_256_STATE_BYTES];
+		u64  w[SKEIN_256_STATE_WORDS];
+	} cfg;                              /* config block */
+
+	Skein_Assert(hashBitLen > 0, SKEIN_BAD_HASHLEN);
+	ctx->h.hashBitLen = hashBitLen;         /* output hash bit count */
+
+	switch (hashBitLen)
+	{             /* use pre-computed values, where available */
+	case  256:
+		memcpy(ctx->X, SKEIN_256_IV_256, sizeof(ctx->X));
+		break;
+	case  224:
+		memcpy(ctx->X, SKEIN_256_IV_224, sizeof(ctx->X));
+		break;
+	case  160:
+		memcpy(ctx->X, SKEIN_256_IV_160, sizeof(ctx->X));
+		break;
+	case  128:
+		memcpy(ctx->X, SKEIN_256_IV_128, sizeof(ctx->X));
+		break;
+	default:
+		/* here if there is no precomputed IV value available */
+		/* build/process the config block, type == CONFIG (could be precomputed) */
+		Skein_Start_New_Type(ctx, CFG_FINAL);        /* set tweaks: T0=0; T1=CFG | FINAL */
+
+		cfg.w[0] = Skein_Swap64(SKEIN_SCHEMA_VER);  /* set the schema, version */
+		cfg.w[1] = Skein_Swap64(hashBitLen);        /* hash result length in bits */
+		cfg.w[2] = Skein_Swap64(SKEIN_CFG_TREE_INFO_SEQUENTIAL);
+		memset(&cfg.w[3], 0, sizeof(cfg) - 3*sizeof(cfg.w[0])); /* zero pad config block */
+
+		/* compute the initial chaining values from config block */
+		memset(ctx->X, 0, sizeof(ctx->X));            /* zero the chaining variables */
+		Skein_256_Process_Block(ctx, cfg.b, 1, SKEIN_CFG_STR_LEN);
+		break;
+	}
+	/* The chaining vars ctx->X are now initialized for the given hashBitLen. */
+	/* Set up to process the data message portion of the hash (default) */
+	Skein_Start_New_Type(ctx, MSG);              /* T0=0, T1= MSG type */
+
+	return SKEIN_SUCCESS;
 }
 
 /*++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++*/
@@ -78,133 +78,133 @@ int Skein_256_Init(struct skein_256_ctx *ctx, size_t hashBitLen)
 /* [identical to Skein_256_Init() when keyBytes == 0 && treeInfo == SKEIN_CFG_TREE_INFO_SEQUENTIAL] */
 int Skein_256_InitExt(struct skein_256_ctx *ctx, size_t hashBitLen, u64 treeInfo, const u8 *key, size_t keyBytes)
 {
-    union
-    {
-        u8  b[SKEIN_256_STATE_BYTES];
-        u64  w[SKEIN_256_STATE_WORDS];
-    } cfg;                              /* config block */
-
-    Skein_Assert(hashBitLen > 0, SKEIN_BAD_HASHLEN);
-    Skein_Assert(keyBytes == 0 || key != NULL, SKEIN_FAIL);
-
-    /* compute the initial chaining values ctx->X[], based on key */
-    if (keyBytes == 0)                          /* is there a key? */
-    {
-        memset(ctx->X, 0, sizeof(ctx->X));        /* no key: use all zeroes as key for config block */
-    }
-    else                                        /* here to pre-process a key */
-    {
-        Skein_assert(sizeof(cfg.b) >= sizeof(ctx->X));
-        /* do a mini-Init right here */
-        ctx->h.hashBitLen = 8*sizeof(ctx->X);     /* set output hash bit count = state size */
-        Skein_Start_New_Type(ctx, KEY);          /* set tweaks: T0 = 0; T1 = KEY type */
-        memset(ctx->X, 0, sizeof(ctx->X));        /* zero the initial chaining variables */
-        Skein_256_Update(ctx, key, keyBytes);     /* hash the key */
-        Skein_256_Final_Pad(ctx, cfg.b);         /* put result into cfg.b[] */
-        memcpy(ctx->X, cfg.b, sizeof(cfg.b));     /* copy over into ctx->X[] */
-    }
-    /* build/process the config block, type == CONFIG (could be precomputed for each key) */
-    ctx->h.hashBitLen = hashBitLen;             /* output hash bit count */
-    Skein_Start_New_Type(ctx, CFG_FINAL);
-
-    memset(&cfg.w, 0, sizeof(cfg.w));             /* pre-pad cfg.w[] with zeroes */
-    cfg.w[0] = Skein_Swap64(SKEIN_SCHEMA_VER);
-    cfg.w[1] = Skein_Swap64(hashBitLen);        /* hash result length in bits */
-    cfg.w[2] = Skein_Swap64(treeInfo);          /* tree hash config info (or SKEIN_CFG_TREE_INFO_SEQUENTIAL) */
-
-    Skein_Show_Key(256, &ctx->h, key, keyBytes);
-
-    /* compute the initial chaining values from config block */
-    Skein_256_Process_Block(ctx, cfg.b, 1, SKEIN_CFG_STR_LEN);
-
-    /* The chaining vars ctx->X are now initialized */
-    /* Set up to process the data message portion of the hash (default) */
-    Skein_Start_New_Type(ctx, MSG);
-
-    return SKEIN_SUCCESS;
+	union
+	{
+		u8  b[SKEIN_256_STATE_BYTES];
+		u64  w[SKEIN_256_STATE_WORDS];
+	} cfg;                              /* config block */
+
+	Skein_Assert(hashBitLen > 0, SKEIN_BAD_HASHLEN);
+	Skein_Assert(keyBytes == 0 || key != NULL, SKEIN_FAIL);
+
+	/* compute the initial chaining values ctx->X[], based on key */
+	if (keyBytes == 0)                          /* is there a key? */
+	{
+		memset(ctx->X, 0, sizeof(ctx->X));        /* no key: use all zeroes as key for config block */
+	}
+	else                                        /* here to pre-process a key */
+	{
+		Skein_assert(sizeof(cfg.b) >= sizeof(ctx->X));
+		/* do a mini-Init right here */
+		ctx->h.hashBitLen = 8*sizeof(ctx->X);     /* set output hash bit count = state size */
+		Skein_Start_New_Type(ctx, KEY);          /* set tweaks: T0 = 0; T1 = KEY type */
+		memset(ctx->X, 0, sizeof(ctx->X));        /* zero the initial chaining variables */
+		Skein_256_Update(ctx, key, keyBytes);     /* hash the key */
+		Skein_256_Final_Pad(ctx, cfg.b);         /* put result into cfg.b[] */
+		memcpy(ctx->X, cfg.b, sizeof(cfg.b));     /* copy over into ctx->X[] */
+	}
+	/* build/process the config block, type == CONFIG (could be precomputed for each key) */
+	ctx->h.hashBitLen = hashBitLen;             /* output hash bit count */
+	Skein_Start_New_Type(ctx, CFG_FINAL);
+
+	memset(&cfg.w, 0, sizeof(cfg.w));             /* pre-pad cfg.w[] with zeroes */
+	cfg.w[0] = Skein_Swap64(SKEIN_SCHEMA_VER);
+	cfg.w[1] = Skein_Swap64(hashBitLen);        /* hash result length in bits */
+	cfg.w[2] = Skein_Swap64(treeInfo);          /* tree hash config info (or SKEIN_CFG_TREE_INFO_SEQUENTIAL) */
+
+	Skein_Show_Key(256, &ctx->h, key, keyBytes);
+
+	/* compute the initial chaining values from config block */
+	Skein_256_Process_Block(ctx, cfg.b, 1, SKEIN_CFG_STR_LEN);
+
+	/* The chaining vars ctx->X are now initialized */
+	/* Set up to process the data message portion of the hash (default) */
+	Skein_Start_New_Type(ctx, MSG);
+
+	return SKEIN_SUCCESS;
 }
 
 /*++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++*/
 /* process the input bytes */
 int Skein_256_Update(struct skein_256_ctx *ctx, const u8 *msg, size_t msgByteCnt)
 {
-    size_t n;
-
-    Skein_Assert(ctx->h.bCnt <= SKEIN_256_BLOCK_BYTES, SKEIN_FAIL);    /* catch uninitialized context */
-
-    /* process full blocks, if any */
-    if (msgByteCnt + ctx->h.bCnt > SKEIN_256_BLOCK_BYTES)
-    {
-        if (ctx->h.bCnt)                              /* finish up any buffered message data */
-        {
-            n = SKEIN_256_BLOCK_BYTES - ctx->h.bCnt;  /* # bytes free in buffer b[] */
-            if (n)
-            {
-                Skein_assert(n < msgByteCnt);         /* check on our logic here */
-                memcpy(&ctx->b[ctx->h.bCnt], msg, n);
-                msgByteCnt  -= n;
-                msg         += n;
-                ctx->h.bCnt += n;
-            }
-            Skein_assert(ctx->h.bCnt == SKEIN_256_BLOCK_BYTES);
-            Skein_256_Process_Block(ctx, ctx->b, 1, SKEIN_256_BLOCK_BYTES);
-            ctx->h.bCnt = 0;
-        }
-        /* now process any remaining full blocks, directly from input message data */
-        if (msgByteCnt > SKEIN_256_BLOCK_BYTES)
-        {
-            n = (msgByteCnt-1) / SKEIN_256_BLOCK_BYTES;   /* number of full blocks to process */
-            Skein_256_Process_Block(ctx, msg, n, SKEIN_256_BLOCK_BYTES);
-            msgByteCnt -= n * SKEIN_256_BLOCK_BYTES;
-            msg        += n * SKEIN_256_BLOCK_BYTES;
-        }
-        Skein_assert(ctx->h.bCnt == 0);
-    }
-
-    /* copy any remaining source message data bytes into b[] */
-    if (msgByteCnt)
-    {
-        Skein_assert(msgByteCnt + ctx->h.bCnt <= SKEIN_256_BLOCK_BYTES);
-        memcpy(&ctx->b[ctx->h.bCnt], msg, msgByteCnt);
-        ctx->h.bCnt += msgByteCnt;
-    }
-
-    return SKEIN_SUCCESS;
+	size_t n;
+
+	Skein_Assert(ctx->h.bCnt <= SKEIN_256_BLOCK_BYTES, SKEIN_FAIL);    /* catch uninitialized context */
+
+	/* process full blocks, if any */
+	if (msgByteCnt + ctx->h.bCnt > SKEIN_256_BLOCK_BYTES)
+	{
+		if (ctx->h.bCnt)                              /* finish up any buffered message data */
+		{
+			n = SKEIN_256_BLOCK_BYTES - ctx->h.bCnt;  /* # bytes free in buffer b[] */
+			if (n)
+			{
+				Skein_assert(n < msgByteCnt);         /* check on our logic here */
+				memcpy(&ctx->b[ctx->h.bCnt], msg, n);
+				msgByteCnt  -= n;
+				msg         += n;
+				ctx->h.bCnt += n;
+			}
+			Skein_assert(ctx->h.bCnt == SKEIN_256_BLOCK_BYTES);
+			Skein_256_Process_Block(ctx, ctx->b, 1, SKEIN_256_BLOCK_BYTES);
+			ctx->h.bCnt = 0;
+		}
+		/* now process any remaining full blocks, directly from input message data */
+		if (msgByteCnt > SKEIN_256_BLOCK_BYTES)
+		{
+			n = (msgByteCnt-1) / SKEIN_256_BLOCK_BYTES;   /* number of full blocks to process */
+			Skein_256_Process_Block(ctx, msg, n, SKEIN_256_BLOCK_BYTES);
+			msgByteCnt -= n * SKEIN_256_BLOCK_BYTES;
+			msg        += n * SKEIN_256_BLOCK_BYTES;
+		}
+		Skein_assert(ctx->h.bCnt == 0);
+	}
+
+	/* copy any remaining source message data bytes into b[] */
+	if (msgByteCnt)
+	{
+		Skein_assert(msgByteCnt + ctx->h.bCnt <= SKEIN_256_BLOCK_BYTES);
+		memcpy(&ctx->b[ctx->h.bCnt], msg, msgByteCnt);
+		ctx->h.bCnt += msgByteCnt;
+	}
+
+	return SKEIN_SUCCESS;
 }
 
 /*++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++*/
 /* finalize the hash computation and output the result */
 int Skein_256_Final(struct skein_256_ctx *ctx, u8 *hashVal)
 {
-    size_t i, n, byteCnt;
-    u64 X[SKEIN_256_STATE_WORDS];
-    Skein_Assert(ctx->h.bCnt <= SKEIN_256_BLOCK_BYTES, SKEIN_FAIL);    /* catch uninitialized context */
-
-    ctx->h.T[1] |= SKEIN_T1_FLAG_FINAL;                 /* tag as the final block */
-    if (ctx->h.bCnt < SKEIN_256_BLOCK_BYTES)            /* zero pad b[] if necessary */
-        memset(&ctx->b[ctx->h.bCnt], 0, SKEIN_256_BLOCK_BYTES - ctx->h.bCnt);
-
-    Skein_256_Process_Block(ctx, ctx->b, 1, ctx->h.bCnt);  /* process the final block */
-
-    /* now output the result */
-    byteCnt = (ctx->h.hashBitLen + 7) >> 3;             /* total number of output bytes */
-
-    /* run Threefish in "counter mode" to generate output */
-    memset(ctx->b, 0, sizeof(ctx->b));  /* zero out b[], so it can hold the counter */
-    memcpy(X, ctx->X, sizeof(X));       /* keep a local copy of counter mode "key" */
-    for (i = 0; i*SKEIN_256_BLOCK_BYTES < byteCnt; i++)
-    {
-        ((u64 *)ctx->b)[0] = Skein_Swap64((u64) i); /* build the counter block */
-        Skein_Start_New_Type(ctx, OUT_FINAL);
-        Skein_256_Process_Block(ctx, ctx->b, 1, sizeof(u64)); /* run "counter mode" */
-        n = byteCnt - i*SKEIN_256_BLOCK_BYTES;   /* number of output bytes left to go */
-        if (n >= SKEIN_256_BLOCK_BYTES)
-            n  = SKEIN_256_BLOCK_BYTES;
-        Skein_Put64_LSB_First(hashVal+i*SKEIN_256_BLOCK_BYTES, ctx->X, n);   /* "output" the ctr mode bytes */
-        Skein_Show_Final(256, &ctx->h, n, hashVal+i*SKEIN_256_BLOCK_BYTES);
-        memcpy(ctx->X, X, sizeof(X));   /* restore the counter mode key for next time */
-    }
-    return SKEIN_SUCCESS;
+	size_t i, n, byteCnt;
+	u64 X[SKEIN_256_STATE_WORDS];
+	Skein_Assert(ctx->h.bCnt <= SKEIN_256_BLOCK_BYTES, SKEIN_FAIL);    /* catch uninitialized context */
+
+	ctx->h.T[1] |= SKEIN_T1_FLAG_FINAL;                 /* tag as the final block */
+	if (ctx->h.bCnt < SKEIN_256_BLOCK_BYTES)            /* zero pad b[] if necessary */
+		memset(&ctx->b[ctx->h.bCnt], 0, SKEIN_256_BLOCK_BYTES - ctx->h.bCnt);
+
+	Skein_256_Process_Block(ctx, ctx->b, 1, ctx->h.bCnt);  /* process the final block */
+
+	/* now output the result */
+	byteCnt = (ctx->h.hashBitLen + 7) >> 3;             /* total number of output bytes */
+
+	/* run Threefish in "counter mode" to generate output */
+	memset(ctx->b, 0, sizeof(ctx->b));  /* zero out b[], so it can hold the counter */
+	memcpy(X, ctx->X, sizeof(X));       /* keep a local copy of counter mode "key" */
+	for (i = 0; i*SKEIN_256_BLOCK_BYTES < byteCnt; i++)
+	{
+		((u64 *)ctx->b)[0] = Skein_Swap64((u64) i); /* build the counter block */
+		Skein_Start_New_Type(ctx, OUT_FINAL);
+		Skein_256_Process_Block(ctx, ctx->b, 1, sizeof(u64)); /* run "counter mode" */
+		n = byteCnt - i*SKEIN_256_BLOCK_BYTES;   /* number of output bytes left to go */
+		if (n >= SKEIN_256_BLOCK_BYTES)
+			n  = SKEIN_256_BLOCK_BYTES;
+		Skein_Put64_LSB_First(hashVal+i*SKEIN_256_BLOCK_BYTES, ctx->X, n);   /* "output" the ctr mode bytes */
+		Skein_Show_Final(256, &ctx->h, n, hashVal+i*SKEIN_256_BLOCK_BYTES);
+		memcpy(ctx->X, X, sizeof(X));   /* restore the counter mode key for next time */
+	}
+	return SKEIN_SUCCESS;
 }
 
 /*****************************************************************/
@@ -215,50 +215,50 @@ int Skein_256_Final(struct skein_256_ctx *ctx, u8 *hashVal)
 /* init the context for a straight hashing operation  */
 int Skein_512_Init(struct skein_512_ctx *ctx, size_t hashBitLen)
 {
-    union
-    {
-        u8  b[SKEIN_512_STATE_BYTES];
-        u64  w[SKEIN_512_STATE_WORDS];
-    } cfg;                              /* config block */
-
-    Skein_Assert(hashBitLen > 0, SKEIN_BAD_HASHLEN);
-    ctx->h.hashBitLen = hashBitLen;         /* output hash bit count */
-
-    switch (hashBitLen)
-    {             /* use pre-computed values, where available */
-    case  512:
-        memcpy(ctx->X, SKEIN_512_IV_512, sizeof(ctx->X));
-        break;
-    case  384:
-        memcpy(ctx->X, SKEIN_512_IV_384, sizeof(ctx->X));
-        break;
-    case  256:
-        memcpy(ctx->X, SKEIN_512_IV_256, sizeof(ctx->X));
-        break;
-    case  224:
-        memcpy(ctx->X, SKEIN_512_IV_224, sizeof(ctx->X));
-        break;
-    default:
-        /* here if there is no precomputed IV value available */
-        /* build/process the config block, type == CONFIG (could be precomputed) */
-        Skein_Start_New_Type(ctx, CFG_FINAL);        /* set tweaks: T0=0; T1=CFG | FINAL */
-
-        cfg.w[0] = Skein_Swap64(SKEIN_SCHEMA_VER);  /* set the schema, version */
-        cfg.w[1] = Skein_Swap64(hashBitLen);        /* hash result length in bits */
-        cfg.w[2] = Skein_Swap64(SKEIN_CFG_TREE_INFO_SEQUENTIAL);
-        memset(&cfg.w[3], 0, sizeof(cfg) - 3*sizeof(cfg.w[0])); /* zero pad config block */
-
-        /* compute the initial chaining values from config block */
-        memset(ctx->X, 0, sizeof(ctx->X));            /* zero the chaining variables */
-        Skein_512_Process_Block(ctx, cfg.b, 1, SKEIN_CFG_STR_LEN);
-        break;
-    }
-
-    /* The chaining vars ctx->X are now initialized for the given hashBitLen. */
-    /* Set up to process the data message portion of the hash (default) */
-    Skein_Start_New_Type(ctx, MSG);              /* T0=0, T1= MSG type */
-
-    return SKEIN_SUCCESS;
+	union
+	{
+		u8  b[SKEIN_512_STATE_BYTES];
+		u64  w[SKEIN_512_STATE_WORDS];
+	} cfg;                              /* config block */
+
+	Skein_Assert(hashBitLen > 0, SKEIN_BAD_HASHLEN);
+	ctx->h.hashBitLen = hashBitLen;         /* output hash bit count */
+
+	switch (hashBitLen)
+	{             /* use pre-computed values, where available */
+	case  512:
+		memcpy(ctx->X, SKEIN_512_IV_512, sizeof(ctx->X));
+		break;
+	case  384:
+		memcpy(ctx->X, SKEIN_512_IV_384, sizeof(ctx->X));
+		break;
+	case  256:
+		memcpy(ctx->X, SKEIN_512_IV_256, sizeof(ctx->X));
+		break;
+	case  224:
+		memcpy(ctx->X, SKEIN_512_IV_224, sizeof(ctx->X));
+		break;
+	default:
+		/* here if there is no precomputed IV value available */
+		/* build/process the config block, type == CONFIG (could be precomputed) */
+		Skein_Start_New_Type(ctx, CFG_FINAL);        /* set tweaks: T0=0; T1=CFG | FINAL */
+
+		cfg.w[0] = Skein_Swap64(SKEIN_SCHEMA_VER);  /* set the schema, version */
+		cfg.w[1] = Skein_Swap64(hashBitLen);        /* hash result length in bits */
+		cfg.w[2] = Skein_Swap64(SKEIN_CFG_TREE_INFO_SEQUENTIAL);
+		memset(&cfg.w[3], 0, sizeof(cfg) - 3*sizeof(cfg.w[0])); /* zero pad config block */
+
+		/* compute the initial chaining values from config block */
+		memset(ctx->X, 0, sizeof(ctx->X));            /* zero the chaining variables */
+		Skein_512_Process_Block(ctx, cfg.b, 1, SKEIN_CFG_STR_LEN);
+		break;
+	}
+
+	/* The chaining vars ctx->X are now initialized for the given hashBitLen. */
+	/* Set up to process the data message portion of the hash (default) */
+	Skein_Start_New_Type(ctx, MSG);              /* T0=0, T1= MSG type */
+
+	return SKEIN_SUCCESS;
 }
 
 /*++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++*/
@@ -266,133 +266,133 @@ int Skein_512_Init(struct skein_512_ctx *ctx, size_t hashBitLen)
 /* [identical to Skein_512_Init() when keyBytes == 0 && treeInfo == SKEIN_CFG_TREE_INFO_SEQUENTIAL] */
 int Skein_512_InitExt(struct skein_512_ctx *ctx, size_t hashBitLen, u64 treeInfo, const u8 *key, size_t keyBytes)
 {
-    union
-    {
-        u8  b[SKEIN_512_STATE_BYTES];
-        u64  w[SKEIN_512_STATE_WORDS];
-    } cfg;                              /* config block */
-
-    Skein_Assert(hashBitLen > 0, SKEIN_BAD_HASHLEN);
-    Skein_Assert(keyBytes == 0 || key != NULL, SKEIN_FAIL);
-
-    /* compute the initial chaining values ctx->X[], based on key */
-    if (keyBytes == 0)                          /* is there a key? */
-    {
-        memset(ctx->X, 0, sizeof(ctx->X));        /* no key: use all zeroes as key for config block */
-    }
-    else                                        /* here to pre-process a key */
-    {
-        Skein_assert(sizeof(cfg.b) >= sizeof(ctx->X));
-        /* do a mini-Init right here */
-        ctx->h.hashBitLen = 8*sizeof(ctx->X);     /* set output hash bit count = state size */
-        Skein_Start_New_Type(ctx, KEY);          /* set tweaks: T0 = 0; T1 = KEY type */
-        memset(ctx->X, 0, sizeof(ctx->X));        /* zero the initial chaining variables */
-        Skein_512_Update(ctx, key, keyBytes);     /* hash the key */
-        Skein_512_Final_Pad(ctx, cfg.b);         /* put result into cfg.b[] */
-        memcpy(ctx->X, cfg.b, sizeof(cfg.b));     /* copy over into ctx->X[] */
-    }
-    /* build/process the config block, type == CONFIG (could be precomputed for each key) */
-    ctx->h.hashBitLen = hashBitLen;             /* output hash bit count */
-    Skein_Start_New_Type(ctx, CFG_FINAL);
-
-    memset(&cfg.w, 0, sizeof(cfg.w));             /* pre-pad cfg.w[] with zeroes */
-    cfg.w[0] = Skein_Swap64(SKEIN_SCHEMA_VER);
-    cfg.w[1] = Skein_Swap64(hashBitLen);        /* hash result length in bits */
-    cfg.w[2] = Skein_Swap64(treeInfo);          /* tree hash config info (or SKEIN_CFG_TREE_INFO_SEQUENTIAL) */
-
-    Skein_Show_Key(512, &ctx->h, key, keyBytes);
-
-    /* compute the initial chaining values from config block */
-    Skein_512_Process_Block(ctx, cfg.b, 1, SKEIN_CFG_STR_LEN);
-
-    /* The chaining vars ctx->X are now initialized */
-    /* Set up to process the data message portion of the hash (default) */
-    Skein_Start_New_Type(ctx, MSG);
-
-    return SKEIN_SUCCESS;
+	union
+	{
+		u8  b[SKEIN_512_STATE_BYTES];
+		u64  w[SKEIN_512_STATE_WORDS];
+	} cfg;                              /* config block */
+
+	Skein_Assert(hashBitLen > 0, SKEIN_BAD_HASHLEN);
+	Skein_Assert(keyBytes == 0 || key != NULL, SKEIN_FAIL);
+
+	/* compute the initial chaining values ctx->X[], based on key */
+	if (keyBytes == 0)                          /* is there a key? */
+	{
+		memset(ctx->X, 0, sizeof(ctx->X));        /* no key: use all zeroes as key for config block */
+	}
+	else                                        /* here to pre-process a key */
+	{
+		Skein_assert(sizeof(cfg.b) >= sizeof(ctx->X));
+		/* do a mini-Init right here */
+		ctx->h.hashBitLen = 8*sizeof(ctx->X);     /* set output hash bit count = state size */
+		Skein_Start_New_Type(ctx, KEY);          /* set tweaks: T0 = 0; T1 = KEY type */
+		memset(ctx->X, 0, sizeof(ctx->X));        /* zero the initial chaining variables */
+		Skein_512_Update(ctx, key, keyBytes);     /* hash the key */
+		Skein_512_Final_Pad(ctx, cfg.b);         /* put result into cfg.b[] */
+		memcpy(ctx->X, cfg.b, sizeof(cfg.b));     /* copy over into ctx->X[] */
+	}
+	/* build/process the config block, type == CONFIG (could be precomputed for each key) */
+	ctx->h.hashBitLen = hashBitLen;             /* output hash bit count */
+	Skein_Start_New_Type(ctx, CFG_FINAL);
+
+	memset(&cfg.w, 0, sizeof(cfg.w));             /* pre-pad cfg.w[] with zeroes */
+	cfg.w[0] = Skein_Swap64(SKEIN_SCHEMA_VER);
+	cfg.w[1] = Skein_Swap64(hashBitLen);        /* hash result length in bits */
+	cfg.w[2] = Skein_Swap64(treeInfo);          /* tree hash config info (or SKEIN_CFG_TREE_INFO_SEQUENTIAL) */
+
+	Skein_Show_Key(512, &ctx->h, key, keyBytes);
+
+	/* compute the initial chaining values from config block */
+	Skein_512_Process_Block(ctx, cfg.b, 1, SKEIN_CFG_STR_LEN);
+
+	/* The chaining vars ctx->X are now initialized */
+	/* Set up to process the data message portion of the hash (default) */
+	Skein_Start_New_Type(ctx, MSG);
+
+	return SKEIN_SUCCESS;
 }
 
 /*++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++*/
 /* process the input bytes */
 int Skein_512_Update(struct skein_512_ctx *ctx, const u8 *msg, size_t msgByteCnt)
 {
-    size_t n;
-
-    Skein_Assert(ctx->h.bCnt <= SKEIN_512_BLOCK_BYTES, SKEIN_FAIL);    /* catch uninitialized context */
-
-    /* process full blocks, if any */
-    if (msgByteCnt + ctx->h.bCnt > SKEIN_512_BLOCK_BYTES)
-    {
-        if (ctx->h.bCnt)                              /* finish up any buffered message data */
-        {
-            n = SKEIN_512_BLOCK_BYTES - ctx->h.bCnt;  /* # bytes free in buffer b[] */
-            if (n)
-            {
-                Skein_assert(n < msgByteCnt);         /* check on our logic here */
-                memcpy(&ctx->b[ctx->h.bCnt], msg, n);
-                msgByteCnt  -= n;
-                msg         += n;
-                ctx->h.bCnt += n;
-            }
-            Skein_assert(ctx->h.bCnt == SKEIN_512_BLOCK_BYTES);
-            Skein_512_Process_Block(ctx, ctx->b, 1, SKEIN_512_BLOCK_BYTES);
-            ctx->h.bCnt = 0;
-        }
-        /* now process any remaining full blocks, directly from input message data */
-        if (msgByteCnt > SKEIN_512_BLOCK_BYTES)
-        {
-            n = (msgByteCnt-1) / SKEIN_512_BLOCK_BYTES;   /* number of full blocks to process */
-            Skein_512_Process_Block(ctx, msg, n, SKEIN_512_BLOCK_BYTES);
-            msgByteCnt -= n * SKEIN_512_BLOCK_BYTES;
-            msg        += n * SKEIN_512_BLOCK_BYTES;
-        }
-        Skein_assert(ctx->h.bCnt == 0);
-    }
-
-    /* copy any remaining source message data bytes into b[] */
-    if (msgByteCnt)
-    {
-        Skein_assert(msgByteCnt + ctx->h.bCnt <= SKEIN_512_BLOCK_BYTES);
-        memcpy(&ctx->b[ctx->h.bCnt], msg, msgByteCnt);
-        ctx->h.bCnt += msgByteCnt;
-    }
-
-    return SKEIN_SUCCESS;
+	size_t n;
+
+	Skein_Assert(ctx->h.bCnt <= SKEIN_512_BLOCK_BYTES, SKEIN_FAIL);    /* catch uninitialized context */
+
+	/* process full blocks, if any */
+	if (msgByteCnt + ctx->h.bCnt > SKEIN_512_BLOCK_BYTES)
+	{
+		if (ctx->h.bCnt)                              /* finish up any buffered message data */
+		{
+			n = SKEIN_512_BLOCK_BYTES - ctx->h.bCnt;  /* # bytes free in buffer b[] */
+			if (n)
+			{
+				Skein_assert(n < msgByteCnt);         /* check on our logic here */
+				memcpy(&ctx->b[ctx->h.bCnt], msg, n);
+				msgByteCnt  -= n;
+				msg         += n;
+				ctx->h.bCnt += n;
+			}
+			Skein_assert(ctx->h.bCnt == SKEIN_512_BLOCK_BYTES);
+			Skein_512_Process_Block(ctx, ctx->b, 1, SKEIN_512_BLOCK_BYTES);
+			ctx->h.bCnt = 0;
+		}
+		/* now process any remaining full blocks, directly from input message data */
+		if (msgByteCnt > SKEIN_512_BLOCK_BYTES)
+		{
+			n = (msgByteCnt-1) / SKEIN_512_BLOCK_BYTES;   /* number of full blocks to process */
+			Skein_512_Process_Block(ctx, msg, n, SKEIN_512_BLOCK_BYTES);
+			msgByteCnt -= n * SKEIN_512_BLOCK_BYTES;
+			msg        += n * SKEIN_512_BLOCK_BYTES;
+		}
+		Skein_assert(ctx->h.bCnt == 0);
+	}
+
+	/* copy any remaining source message data bytes into b[] */
+	if (msgByteCnt)
+	{
+		Skein_assert(msgByteCnt + ctx->h.bCnt <= SKEIN_512_BLOCK_BYTES);
+		memcpy(&ctx->b[ctx->h.bCnt], msg, msgByteCnt);
+		ctx->h.bCnt += msgByteCnt;
+	}
+
+	return SKEIN_SUCCESS;
 }
 
 /*++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++*/
 /* finalize the hash computation and output the result */
 int Skein_512_Final(struct skein_512_ctx *ctx, u8 *hashVal)
 {
-    size_t i, n, byteCnt;
-    u64 X[SKEIN_512_STATE_WORDS];
-    Skein_Assert(ctx->h.bCnt <= SKEIN_512_BLOCK_BYTES, SKEIN_FAIL);    /* catch uninitialized context */
-
-    ctx->h.T[1] |= SKEIN_T1_FLAG_FINAL;                 /* tag as the final block */
-    if (ctx->h.bCnt < SKEIN_512_BLOCK_BYTES)            /* zero pad b[] if necessary */
-        memset(&ctx->b[ctx->h.bCnt], 0, SKEIN_512_BLOCK_BYTES - ctx->h.bCnt);
-
-    Skein_512_Process_Block(ctx, ctx->b, 1, ctx->h.bCnt);  /* process the final block */
-
-    /* now output the result */
-    byteCnt = (ctx->h.hashBitLen + 7) >> 3;             /* total number of output bytes */
-
-    /* run Threefish in "counter mode" to generate output */
-    memset(ctx->b, 0, sizeof(ctx->b));  /* zero out b[], so it can hold the counter */
-    memcpy(X, ctx->X, sizeof(X));       /* keep a local copy of counter mode "key" */
-    for (i = 0; i*SKEIN_512_BLOCK_BYTES < byteCnt; i++)
-    {
-        ((u64 *)ctx->b)[0] = Skein_Swap64((u64) i); /* build the counter block */
-        Skein_Start_New_Type(ctx, OUT_FINAL);
-        Skein_512_Process_Block(ctx, ctx->b, 1, sizeof(u64)); /* run "counter mode" */
-        n = byteCnt - i*SKEIN_512_BLOCK_BYTES;   /* number of output bytes left to go */
-        if (n >= SKEIN_512_BLOCK_BYTES)
-            n  = SKEIN_512_BLOCK_BYTES;
-        Skein_Put64_LSB_First(hashVal+i*SKEIN_512_BLOCK_BYTES, ctx->X, n);   /* "output" the ctr mode bytes */
-        Skein_Show_Final(512, &ctx->h, n, hashVal+i*SKEIN_512_BLOCK_BYTES);
-        memcpy(ctx->X, X, sizeof(X));   /* restore the counter mode key for next time */
-    }
-    return SKEIN_SUCCESS;
+	size_t i, n, byteCnt;
+	u64 X[SKEIN_512_STATE_WORDS];
+	Skein_Assert(ctx->h.bCnt <= SKEIN_512_BLOCK_BYTES, SKEIN_FAIL);    /* catch uninitialized context */
+
+	ctx->h.T[1] |= SKEIN_T1_FLAG_FINAL;                 /* tag as the final block */
+	if (ctx->h.bCnt < SKEIN_512_BLOCK_BYTES)            /* zero pad b[] if necessary */
+		memset(&ctx->b[ctx->h.bCnt], 0, SKEIN_512_BLOCK_BYTES - ctx->h.bCnt);
+
+	Skein_512_Process_Block(ctx, ctx->b, 1, ctx->h.bCnt);  /* process the final block */
+
+	/* now output the result */
+	byteCnt = (ctx->h.hashBitLen + 7) >> 3;             /* total number of output bytes */
+
+	/* run Threefish in "counter mode" to generate output */
+	memset(ctx->b, 0, sizeof(ctx->b));  /* zero out b[], so it can hold the counter */
+	memcpy(X, ctx->X, sizeof(X));       /* keep a local copy of counter mode "key" */
+	for (i = 0; i*SKEIN_512_BLOCK_BYTES < byteCnt; i++)
+	{
+		((u64 *)ctx->b)[0] = Skein_Swap64((u64) i); /* build the counter block */
+		Skein_Start_New_Type(ctx, OUT_FINAL);
+		Skein_512_Process_Block(ctx, ctx->b, 1, sizeof(u64)); /* run "counter mode" */
+		n = byteCnt - i*SKEIN_512_BLOCK_BYTES;   /* number of output bytes left to go */
+		if (n >= SKEIN_512_BLOCK_BYTES)
+			n  = SKEIN_512_BLOCK_BYTES;
+		Skein_Put64_LSB_First(hashVal+i*SKEIN_512_BLOCK_BYTES, ctx->X, n);   /* "output" the ctr mode bytes */
+		Skein_Show_Final(512, &ctx->h, n, hashVal+i*SKEIN_512_BLOCK_BYTES);
+		memcpy(ctx->X, X, sizeof(X));   /* restore the counter mode key for next time */
+	}
+	return SKEIN_SUCCESS;
 }
 
 /*****************************************************************/
@@ -403,47 +403,47 @@ int Skein_512_Final(struct skein_512_ctx *ctx, u8 *hashVal)
 /* init the context for a straight hashing operation  */
 int Skein1024_Init(struct skein1024_ctx *ctx, size_t hashBitLen)
 {
-    union
-    {
-        u8  b[SKEIN1024_STATE_BYTES];
-        u64  w[SKEIN1024_STATE_WORDS];
-    } cfg;                              /* config block */
-
-    Skein_Assert(hashBitLen > 0, SKEIN_BAD_HASHLEN);
-    ctx->h.hashBitLen = hashBitLen;         /* output hash bit count */
-
-    switch (hashBitLen)
-    {              /* use pre-computed values, where available */
-    case  512:
-        memcpy(ctx->X, SKEIN1024_IV_512, sizeof(ctx->X));
-        break;
-    case  384:
-        memcpy(ctx->X, SKEIN1024_IV_384, sizeof(ctx->X));
-        break;
-    case 1024:
-        memcpy(ctx->X, SKEIN1024_IV_1024, sizeof(ctx->X));
-        break;
-    default:
-        /* here if there is no precomputed IV value available */
-        /* build/process the config block, type == CONFIG (could be precomputed) */
-        Skein_Start_New_Type(ctx, CFG_FINAL);        /* set tweaks: T0=0; T1=CFG | FINAL */
-
-        cfg.w[0] = Skein_Swap64(SKEIN_SCHEMA_VER);  /* set the schema, version */
-        cfg.w[1] = Skein_Swap64(hashBitLen);        /* hash result length in bits */
-        cfg.w[2] = Skein_Swap64(SKEIN_CFG_TREE_INFO_SEQUENTIAL);
-        memset(&cfg.w[3], 0, sizeof(cfg) - 3*sizeof(cfg.w[0])); /* zero pad config block */
-
-        /* compute the initial chaining values from config block */
-        memset(ctx->X, 0, sizeof(ctx->X));            /* zero the chaining variables */
-        Skein1024_Process_Block(ctx, cfg.b, 1, SKEIN_CFG_STR_LEN);
-        break;
-    }
-
-    /* The chaining vars ctx->X are now initialized for the given hashBitLen. */
-    /* Set up to process the data message portion of the hash (default) */
-    Skein_Start_New_Type(ctx, MSG);              /* T0=0, T1= MSG type */
-
-    return SKEIN_SUCCESS;
+	union
+	{
+		u8  b[SKEIN1024_STATE_BYTES];
+		u64  w[SKEIN1024_STATE_WORDS];
+	} cfg;                              /* config block */
+
+	Skein_Assert(hashBitLen > 0, SKEIN_BAD_HASHLEN);
+	ctx->h.hashBitLen = hashBitLen;         /* output hash bit count */
+
+	switch (hashBitLen)
+	{              /* use pre-computed values, where available */
+	case  512:
+		memcpy(ctx->X, SKEIN1024_IV_512, sizeof(ctx->X));
+		break;
+	case  384:
+		memcpy(ctx->X, SKEIN1024_IV_384, sizeof(ctx->X));
+		break;
+	case 1024:
+		memcpy(ctx->X, SKEIN1024_IV_1024, sizeof(ctx->X));
+		break;
+	default:
+		/* here if there is no precomputed IV value available */
+		/* build/process the config block, type == CONFIG (could be precomputed) */
+		Skein_Start_New_Type(ctx, CFG_FINAL);        /* set tweaks: T0=0; T1=CFG | FINAL */
+
+		cfg.w[0] = Skein_Swap64(SKEIN_SCHEMA_VER);  /* set the schema, version */
+		cfg.w[1] = Skein_Swap64(hashBitLen);        /* hash result length in bits */
+		cfg.w[2] = Skein_Swap64(SKEIN_CFG_TREE_INFO_SEQUENTIAL);
+		memset(&cfg.w[3], 0, sizeof(cfg) - 3*sizeof(cfg.w[0])); /* zero pad config block */
+
+		/* compute the initial chaining values from config block */
+		memset(ctx->X, 0, sizeof(ctx->X));            /* zero the chaining variables */
+		Skein1024_Process_Block(ctx, cfg.b, 1, SKEIN_CFG_STR_LEN);
+		break;
+	}
+
+	/* The chaining vars ctx->X are now initialized for the given hashBitLen. */
+	/* Set up to process the data message portion of the hash (default) */
+	Skein_Start_New_Type(ctx, MSG);              /* T0=0, T1= MSG type */
+
+	return SKEIN_SUCCESS;
 }
 
 /*++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++*/
@@ -451,133 +451,133 @@ int Skein1024_Init(struct skein1024_ctx *ctx, size_t hashBitLen)
 /* [identical to Skein1024_Init() when keyBytes == 0 && treeInfo == SKEIN_CFG_TREE_INFO_SEQUENTIAL] */
 int Skein1024_InitExt(struct skein1024_ctx *ctx, size_t hashBitLen, u64 treeInfo, const u8 *key, size_t keyBytes)
 {
-    union
-    {
-        u8  b[SKEIN1024_STATE_BYTES];
-        u64  w[SKEIN1024_STATE_WORDS];
-    } cfg;                              /* config block */
-
-    Skein_Assert(hashBitLen > 0, SKEIN_BAD_HASHLEN);
-    Skein_Assert(keyBytes == 0 || key != NULL, SKEIN_FAIL);
-
-    /* compute the initial chaining values ctx->X[], based on key */
-    if (keyBytes == 0)                          /* is there a key? */
-    {
-        memset(ctx->X, 0, sizeof(ctx->X));        /* no key: use all zeroes as key for config block */
-    }
-    else                                        /* here to pre-process a key */
-    {
-        Skein_assert(sizeof(cfg.b) >= sizeof(ctx->X));
-        /* do a mini-Init right here */
-        ctx->h.hashBitLen = 8*sizeof(ctx->X);     /* set output hash bit count = state size */
-        Skein_Start_New_Type(ctx, KEY);          /* set tweaks: T0 = 0; T1 = KEY type */
-        memset(ctx->X, 0, sizeof(ctx->X));        /* zero the initial chaining variables */
-        Skein1024_Update(ctx, key, keyBytes);     /* hash the key */
-        Skein1024_Final_Pad(ctx, cfg.b);         /* put result into cfg.b[] */
-        memcpy(ctx->X, cfg.b, sizeof(cfg.b));     /* copy over into ctx->X[] */
-    }
-    /* build/process the config block, type == CONFIG (could be precomputed for each key) */
-    ctx->h.hashBitLen = hashBitLen;             /* output hash bit count */
-    Skein_Start_New_Type(ctx, CFG_FINAL);
-
-    memset(&cfg.w, 0, sizeof(cfg.w));             /* pre-pad cfg.w[] with zeroes */
-    cfg.w[0] = Skein_Swap64(SKEIN_SCHEMA_VER);
-    cfg.w[1] = Skein_Swap64(hashBitLen);        /* hash result length in bits */
-    cfg.w[2] = Skein_Swap64(treeInfo);          /* tree hash config info (or SKEIN_CFG_TREE_INFO_SEQUENTIAL) */
-
-    Skein_Show_Key(1024, &ctx->h, key, keyBytes);
-
-    /* compute the initial chaining values from config block */
-    Skein1024_Process_Block(ctx, cfg.b, 1, SKEIN_CFG_STR_LEN);
-
-    /* The chaining vars ctx->X are now initialized */
-    /* Set up to process the data message portion of the hash (default) */
-    Skein_Start_New_Type(ctx, MSG);
-
-    return SKEIN_SUCCESS;
+	union
+	{
+		u8  b[SKEIN1024_STATE_BYTES];
+		u64  w[SKEIN1024_STATE_WORDS];
+	} cfg;                              /* config block */
+
+	Skein_Assert(hashBitLen > 0, SKEIN_BAD_HASHLEN);
+	Skein_Assert(keyBytes == 0 || key != NULL, SKEIN_FAIL);
+
+	/* compute the initial chaining values ctx->X[], based on key */
+	if (keyBytes == 0)                          /* is there a key? */
+	{
+		memset(ctx->X, 0, sizeof(ctx->X));        /* no key: use all zeroes as key for config block */
+	}
+	else                                        /* here to pre-process a key */
+	{
+		Skein_assert(sizeof(cfg.b) >= sizeof(ctx->X));
+		/* do a mini-Init right here */
+		ctx->h.hashBitLen = 8*sizeof(ctx->X);     /* set output hash bit count = state size */
+		Skein_Start_New_Type(ctx, KEY);          /* set tweaks: T0 = 0; T1 = KEY type */
+		memset(ctx->X, 0, sizeof(ctx->X));        /* zero the initial chaining variables */
+		Skein1024_Update(ctx, key, keyBytes);     /* hash the key */
+		Skein1024_Final_Pad(ctx, cfg.b);         /* put result into cfg.b[] */
+		memcpy(ctx->X, cfg.b, sizeof(cfg.b));     /* copy over into ctx->X[] */
+	}
+	/* build/process the config block, type == CONFIG (could be precomputed for each key) */
+	ctx->h.hashBitLen = hashBitLen;             /* output hash bit count */
+	Skein_Start_New_Type(ctx, CFG_FINAL);
+
+	memset(&cfg.w, 0, sizeof(cfg.w));             /* pre-pad cfg.w[] with zeroes */
+	cfg.w[0] = Skein_Swap64(SKEIN_SCHEMA_VER);
+	cfg.w[1] = Skein_Swap64(hashBitLen);        /* hash result length in bits */
+	cfg.w[2] = Skein_Swap64(treeInfo);          /* tree hash config info (or SKEIN_CFG_TREE_INFO_SEQUENTIAL) */
+
+	Skein_Show_Key(1024, &ctx->h, key, keyBytes);
+
+	/* compute the initial chaining values from config block */
+	Skein1024_Process_Block(ctx, cfg.b, 1, SKEIN_CFG_STR_LEN);
+
+	/* The chaining vars ctx->X are now initialized */
+	/* Set up to process the data message portion of the hash (default) */
+	Skein_Start_New_Type(ctx, MSG);
+
+	return SKEIN_SUCCESS;
 }
 
 /*++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++*/
 /* process the input bytes */
 int Skein1024_Update(struct skein1024_ctx *ctx, const u8 *msg, size_t msgByteCnt)
 {
-    size_t n;
-
-    Skein_Assert(ctx->h.bCnt <= SKEIN1024_BLOCK_BYTES, SKEIN_FAIL);    /* catch uninitialized context */
-
-    /* process full blocks, if any */
-    if (msgByteCnt + ctx->h.bCnt > SKEIN1024_BLOCK_BYTES)
-    {
-        if (ctx->h.bCnt)                              /* finish up any buffered message data */
-        {
-            n = SKEIN1024_BLOCK_BYTES - ctx->h.bCnt;  /* # bytes free in buffer b[] */
-            if (n)
-            {
-                Skein_assert(n < msgByteCnt);         /* check on our logic here */
-                memcpy(&ctx->b[ctx->h.bCnt], msg, n);
-                msgByteCnt  -= n;
-                msg         += n;
-                ctx->h.bCnt += n;
-            }
-            Skein_assert(ctx->h.bCnt == SKEIN1024_BLOCK_BYTES);
-            Skein1024_Process_Block(ctx, ctx->b, 1, SKEIN1024_BLOCK_BYTES);
-            ctx->h.bCnt = 0;
-        }
-        /* now process any remaining full blocks, directly from input message data */
-        if (msgByteCnt > SKEIN1024_BLOCK_BYTES)
-        {
-            n = (msgByteCnt-1) / SKEIN1024_BLOCK_BYTES;   /* number of full blocks to process */
-            Skein1024_Process_Block(ctx, msg, n, SKEIN1024_BLOCK_BYTES);
-            msgByteCnt -= n * SKEIN1024_BLOCK_BYTES;
-            msg        += n * SKEIN1024_BLOCK_BYTES;
-        }
-        Skein_assert(ctx->h.bCnt == 0);
-    }
-
-    /* copy any remaining source message data bytes into b[] */
-    if (msgByteCnt)
-    {
-        Skein_assert(msgByteCnt + ctx->h.bCnt <= SKEIN1024_BLOCK_BYTES);
-        memcpy(&ctx->b[ctx->h.bCnt], msg, msgByteCnt);
-        ctx->h.bCnt += msgByteCnt;
-    }
-
-    return SKEIN_SUCCESS;
+	size_t n;
+
+	Skein_Assert(ctx->h.bCnt <= SKEIN1024_BLOCK_BYTES, SKEIN_FAIL);    /* catch uninitialized context */
+
+	/* process full blocks, if any */
+	if (msgByteCnt + ctx->h.bCnt > SKEIN1024_BLOCK_BYTES)
+	{
+		if (ctx->h.bCnt)                              /* finish up any buffered message data */
+		{
+			n = SKEIN1024_BLOCK_BYTES - ctx->h.bCnt;  /* # bytes free in buffer b[] */
+			if (n)
+			{
+				Skein_assert(n < msgByteCnt);         /* check on our logic here */
+				memcpy(&ctx->b[ctx->h.bCnt], msg, n);
+				msgByteCnt  -= n;
+				msg         += n;
+				ctx->h.bCnt += n;
+			}
+			Skein_assert(ctx->h.bCnt == SKEIN1024_BLOCK_BYTES);
+			Skein1024_Process_Block(ctx, ctx->b, 1, SKEIN1024_BLOCK_BYTES);
+			ctx->h.bCnt = 0;
+		}
+		/* now process any remaining full blocks, directly from input message data */
+		if (msgByteCnt > SKEIN1024_BLOCK_BYTES)
+		{
+			n = (msgByteCnt-1) / SKEIN1024_BLOCK_BYTES;   /* number of full blocks to process */
+			Skein1024_Process_Block(ctx, msg, n, SKEIN1024_BLOCK_BYTES);
+			msgByteCnt -= n * SKEIN1024_BLOCK_BYTES;
+			msg        += n * SKEIN1024_BLOCK_BYTES;
+		}
+		Skein_assert(ctx->h.bCnt == 0);
+	}
+
+	/* copy any remaining source message data bytes into b[] */
+	if (msgByteCnt)
+	{
+		Skein_assert(msgByteCnt + ctx->h.bCnt <= SKEIN1024_BLOCK_BYTES);
+		memcpy(&ctx->b[ctx->h.bCnt], msg, msgByteCnt);
+		ctx->h.bCnt += msgByteCnt;
+	}
+
+	return SKEIN_SUCCESS;
 }
 
 /*++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++*/
 /* finalize the hash computation and output the result */
 int Skein1024_Final(struct skein1024_ctx *ctx, u8 *hashVal)
 {
-    size_t i, n, byteCnt;
-    u64 X[SKEIN1024_STATE_WORDS];
-    Skein_Assert(ctx->h.bCnt <= SKEIN1024_BLOCK_BYTES, SKEIN_FAIL);    /* catch uninitialized context */
-
-    ctx->h.T[1] |= SKEIN_T1_FLAG_FINAL;                 /* tag as the final block */
-    if (ctx->h.bCnt < SKEIN1024_BLOCK_BYTES)            /* zero pad b[] if necessary */
-        memset(&ctx->b[ctx->h.bCnt], 0, SKEIN1024_BLOCK_BYTES - ctx->h.bCnt);
-
-    Skein1024_Process_Block(ctx, ctx->b, 1, ctx->h.bCnt);  /* process the final block */
-
-    /* now output the result */
-    byteCnt = (ctx->h.hashBitLen + 7) >> 3;             /* total number of output bytes */
-
-    /* run Threefish in "counter mode" to generate output */
-    memset(ctx->b, 0, sizeof(ctx->b));  /* zero out b[], so it can hold the counter */
-    memcpy(X, ctx->X, sizeof(X));       /* keep a local copy of counter mode "key" */
-    for (i = 0; i*SKEIN1024_BLOCK_BYTES < byteCnt; i++)
-    {
-        ((u64 *)ctx->b)[0] = Skein_Swap64((u64) i); /* build the counter block */
-        Skein_Start_New_Type(ctx, OUT_FINAL);
-        Skein1024_Process_Block(ctx, ctx->b, 1, sizeof(u64)); /* run "counter mode" */
-        n = byteCnt - i*SKEIN1024_BLOCK_BYTES;   /* number of output bytes left to go */
-        if (n >= SKEIN1024_BLOCK_BYTES)
-            n  = SKEIN1024_BLOCK_BYTES;
-        Skein_Put64_LSB_First(hashVal+i*SKEIN1024_BLOCK_BYTES, ctx->X, n);   /* "output" the ctr mode bytes */
-        Skein_Show_Final(1024, &ctx->h, n, hashVal+i*SKEIN1024_BLOCK_BYTES);
-        memcpy(ctx->X, X, sizeof(X));   /* restore the counter mode key for next time */
-    }
-    return SKEIN_SUCCESS;
+	size_t i, n, byteCnt;
+	u64 X[SKEIN1024_STATE_WORDS];
+	Skein_Assert(ctx->h.bCnt <= SKEIN1024_BLOCK_BYTES, SKEIN_FAIL);    /* catch uninitialized context */
+
+	ctx->h.T[1] |= SKEIN_T1_FLAG_FINAL;                 /* tag as the final block */
+	if (ctx->h.bCnt < SKEIN1024_BLOCK_BYTES)            /* zero pad b[] if necessary */
+		memset(&ctx->b[ctx->h.bCnt], 0, SKEIN1024_BLOCK_BYTES - ctx->h.bCnt);
+
+	Skein1024_Process_Block(ctx, ctx->b, 1, ctx->h.bCnt);  /* process the final block */
+
+	/* now output the result */
+	byteCnt = (ctx->h.hashBitLen + 7) >> 3;             /* total number of output bytes */
+
+	/* run Threefish in "counter mode" to generate output */
+	memset(ctx->b, 0, sizeof(ctx->b));  /* zero out b[], so it can hold the counter */
+	memcpy(X, ctx->X, sizeof(X));       /* keep a local copy of counter mode "key" */
+	for (i = 0; i*SKEIN1024_BLOCK_BYTES < byteCnt; i++)
+	{
+		((u64 *)ctx->b)[0] = Skein_Swap64((u64) i); /* build the counter block */
+		Skein_Start_New_Type(ctx, OUT_FINAL);
+		Skein1024_Process_Block(ctx, ctx->b, 1, sizeof(u64)); /* run "counter mode" */
+		n = byteCnt - i*SKEIN1024_BLOCK_BYTES;   /* number of output bytes left to go */
+		if (n >= SKEIN1024_BLOCK_BYTES)
+			n  = SKEIN1024_BLOCK_BYTES;
+		Skein_Put64_LSB_First(hashVal+i*SKEIN1024_BLOCK_BYTES, ctx->X, n);   /* "output" the ctr mode bytes */
+		Skein_Show_Final(1024, &ctx->h, n, hashVal+i*SKEIN1024_BLOCK_BYTES);
+		memcpy(ctx->X, X, sizeof(X));   /* restore the counter mode key for next time */
+	}
+	return SKEIN_SUCCESS;
 }
 
 /**************** Functions to support MAC/tree hashing ***************/
@@ -587,48 +587,48 @@ int Skein1024_Final(struct skein1024_ctx *ctx, u8 *hashVal)
 /* finalize the hash computation and output the block, no OUTPUT stage */
 int Skein_256_Final_Pad(struct skein_256_ctx *ctx, u8 *hashVal)
 {
-    Skein_Assert(ctx->h.bCnt <= SKEIN_256_BLOCK_BYTES, SKEIN_FAIL);    /* catch uninitialized context */
+	Skein_Assert(ctx->h.bCnt <= SKEIN_256_BLOCK_BYTES, SKEIN_FAIL);    /* catch uninitialized context */
 
-    ctx->h.T[1] |= SKEIN_T1_FLAG_FINAL;        /* tag as the final block */
-    if (ctx->h.bCnt < SKEIN_256_BLOCK_BYTES)   /* zero pad b[] if necessary */
-        memset(&ctx->b[ctx->h.bCnt], 0, SKEIN_256_BLOCK_BYTES - ctx->h.bCnt);
-    Skein_256_Process_Block(ctx, ctx->b, 1, ctx->h.bCnt);    /* process the final block */
+	ctx->h.T[1] |= SKEIN_T1_FLAG_FINAL;        /* tag as the final block */
+	if (ctx->h.bCnt < SKEIN_256_BLOCK_BYTES)   /* zero pad b[] if necessary */
+		memset(&ctx->b[ctx->h.bCnt], 0, SKEIN_256_BLOCK_BYTES - ctx->h.bCnt);
+	Skein_256_Process_Block(ctx, ctx->b, 1, ctx->h.bCnt);    /* process the final block */
 
-    Skein_Put64_LSB_First(hashVal, ctx->X, SKEIN_256_BLOCK_BYTES);   /* "output" the state bytes */
+	Skein_Put64_LSB_First(hashVal, ctx->X, SKEIN_256_BLOCK_BYTES);   /* "output" the state bytes */
 
-    return SKEIN_SUCCESS;
+	return SKEIN_SUCCESS;
 }
 
 /*++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++*/
 /* finalize the hash computation and output the block, no OUTPUT stage */
 int Skein_512_Final_Pad(struct skein_512_ctx *ctx, u8 *hashVal)
 {
-    Skein_Assert(ctx->h.bCnt <= SKEIN_512_BLOCK_BYTES, SKEIN_FAIL);    /* catch uninitialized context */
+	Skein_Assert(ctx->h.bCnt <= SKEIN_512_BLOCK_BYTES, SKEIN_FAIL);    /* catch uninitialized context */
 
-    ctx->h.T[1] |= SKEIN_T1_FLAG_FINAL;        /* tag as the final block */
-    if (ctx->h.bCnt < SKEIN_512_BLOCK_BYTES)   /* zero pad b[] if necessary */
-        memset(&ctx->b[ctx->h.bCnt], 0, SKEIN_512_BLOCK_BYTES - ctx->h.bCnt);
-    Skein_512_Process_Block(ctx, ctx->b, 1, ctx->h.bCnt);    /* process the final block */
+	ctx->h.T[1] |= SKEIN_T1_FLAG_FINAL;        /* tag as the final block */
+	if (ctx->h.bCnt < SKEIN_512_BLOCK_BYTES)   /* zero pad b[] if necessary */
+		memset(&ctx->b[ctx->h.bCnt], 0, SKEIN_512_BLOCK_BYTES - ctx->h.bCnt);
+	Skein_512_Process_Block(ctx, ctx->b, 1, ctx->h.bCnt);    /* process the final block */
 
-    Skein_Put64_LSB_First(hashVal, ctx->X, SKEIN_512_BLOCK_BYTES);   /* "output" the state bytes */
+	Skein_Put64_LSB_First(hashVal, ctx->X, SKEIN_512_BLOCK_BYTES);   /* "output" the state bytes */
 
-    return SKEIN_SUCCESS;
+	return SKEIN_SUCCESS;
 }
 
 /*++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++*/
 /* finalize the hash computation and output the block, no OUTPUT stage */
 int Skein1024_Final_Pad(struct skein1024_ctx *ctx, u8 *hashVal)
 {
-    Skein_Assert(ctx->h.bCnt <= SKEIN1024_BLOCK_BYTES, SKEIN_FAIL);    /* catch uninitialized context */
+	Skein_Assert(ctx->h.bCnt <= SKEIN1024_BLOCK_BYTES, SKEIN_FAIL);    /* catch uninitialized context */
 
-    ctx->h.T[1] |= SKEIN_T1_FLAG_FINAL;        /* tag as the final block */
-    if (ctx->h.bCnt < SKEIN1024_BLOCK_BYTES)   /* zero pad b[] if necessary */
-        memset(&ctx->b[ctx->h.bCnt], 0, SKEIN1024_BLOCK_BYTES - ctx->h.bCnt);
-    Skein1024_Process_Block(ctx, ctx->b, 1, ctx->h.bCnt);    /* process the final block */
+	ctx->h.T[1] |= SKEIN_T1_FLAG_FINAL;        /* tag as the final block */
+	if (ctx->h.bCnt < SKEIN1024_BLOCK_BYTES)   /* zero pad b[] if necessary */
+		memset(&ctx->b[ctx->h.bCnt], 0, SKEIN1024_BLOCK_BYTES - ctx->h.bCnt);
+	Skein1024_Process_Block(ctx, ctx->b, 1, ctx->h.bCnt);    /* process the final block */
 
-    Skein_Put64_LSB_First(hashVal, ctx->X, SKEIN1024_BLOCK_BYTES);   /* "output" the state bytes */
+	Skein_Put64_LSB_First(hashVal, ctx->X, SKEIN1024_BLOCK_BYTES);   /* "output" the state bytes */
 
-    return SKEIN_SUCCESS;
+	return SKEIN_SUCCESS;
 }
 
 #if SKEIN_TREE_HASH
@@ -636,86 +636,86 @@ int Skein1024_Final_Pad(struct skein1024_ctx *ctx, u8 *hashVal)
 /* just do the OUTPUT stage                                       */
 int Skein_256_Output(struct skein_256_ctx *ctx, u8 *hashVal)
 {
-    size_t i, n, byteCnt;
-    u64 X[SKEIN_256_STATE_WORDS];
-    Skein_Assert(ctx->h.bCnt <= SKEIN_256_BLOCK_BYTES, SKEIN_FAIL);    /* catch uninitialized context */
-
-    /* now output the result */
-    byteCnt = (ctx->h.hashBitLen + 7) >> 3;    /* total number of output bytes */
-
-    /* run Threefish in "counter mode" to generate output */
-    memset(ctx->b, 0, sizeof(ctx->b));  /* zero out b[], so it can hold the counter */
-    memcpy(X, ctx->X, sizeof(X));       /* keep a local copy of counter mode "key" */
-    for (i = 0; i*SKEIN_256_BLOCK_BYTES < byteCnt; i++)
-    {
-        ((u64 *)ctx->b)[0] = Skein_Swap64((u64) i); /* build the counter block */
-        Skein_Start_New_Type(ctx, OUT_FINAL);
-        Skein_256_Process_Block(ctx, ctx->b, 1, sizeof(u64)); /* run "counter mode" */
-        n = byteCnt - i*SKEIN_256_BLOCK_BYTES;   /* number of output bytes left to go */
-        if (n >= SKEIN_256_BLOCK_BYTES)
-            n  = SKEIN_256_BLOCK_BYTES;
-        Skein_Put64_LSB_First(hashVal+i*SKEIN_256_BLOCK_BYTES, ctx->X, n);   /* "output" the ctr mode bytes */
-        Skein_Show_Final(256, &ctx->h, n, hashVal+i*SKEIN_256_BLOCK_BYTES);
-        memcpy(ctx->X, X, sizeof(X));   /* restore the counter mode key for next time */
-    }
-    return SKEIN_SUCCESS;
+	size_t i, n, byteCnt;
+	u64 X[SKEIN_256_STATE_WORDS];
+	Skein_Assert(ctx->h.bCnt <= SKEIN_256_BLOCK_BYTES, SKEIN_FAIL);    /* catch uninitialized context */
+
+	/* now output the result */
+	byteCnt = (ctx->h.hashBitLen + 7) >> 3;    /* total number of output bytes */
+
+	/* run Threefish in "counter mode" to generate output */
+	memset(ctx->b, 0, sizeof(ctx->b));  /* zero out b[], so it can hold the counter */
+	memcpy(X, ctx->X, sizeof(X));       /* keep a local copy of counter mode "key" */
+	for (i = 0; i*SKEIN_256_BLOCK_BYTES < byteCnt; i++)
+	{
+		((u64 *)ctx->b)[0] = Skein_Swap64((u64) i); /* build the counter block */
+		Skein_Start_New_Type(ctx, OUT_FINAL);
+		Skein_256_Process_Block(ctx, ctx->b, 1, sizeof(u64)); /* run "counter mode" */
+		n = byteCnt - i*SKEIN_256_BLOCK_BYTES;   /* number of output bytes left to go */
+		if (n >= SKEIN_256_BLOCK_BYTES)
+			n  = SKEIN_256_BLOCK_BYTES;
+		Skein_Put64_LSB_First(hashVal+i*SKEIN_256_BLOCK_BYTES, ctx->X, n);   /* "output" the ctr mode bytes */
+		Skein_Show_Final(256, &ctx->h, n, hashVal+i*SKEIN_256_BLOCK_BYTES);
+		memcpy(ctx->X, X, sizeof(X));   /* restore the counter mode key for next time */
+	}
+	return SKEIN_SUCCESS;
 }
 
 /*++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++*/
 /* just do the OUTPUT stage                                       */
 int Skein_512_Output(struct skein_512_ctx *ctx, u8 *hashVal)
 {
-    size_t i, n, byteCnt;
-    u64 X[SKEIN_512_STATE_WORDS];
-    Skein_Assert(ctx->h.bCnt <= SKEIN_512_BLOCK_BYTES, SKEIN_FAIL);    /* catch uninitialized context */
-
-    /* now output the result */
-    byteCnt = (ctx->h.hashBitLen + 7) >> 3;    /* total number of output bytes */
-
-    /* run Threefish in "counter mode" to generate output */
-    memset(ctx->b, 0, sizeof(ctx->b));  /* zero out b[], so it can hold the counter */
-    memcpy(X, ctx->X, sizeof(X));       /* keep a local copy of counter mode "key" */
-    for (i = 0; i*SKEIN_512_BLOCK_BYTES < byteCnt; i++)
-    {
-        ((u64 *)ctx->b)[0] = Skein_Swap64((u64) i); /* build the counter block */
-        Skein_Start_New_Type(ctx, OUT_FINAL);
-        Skein_512_Process_Block(ctx, ctx->b, 1, sizeof(u64)); /* run "counter mode" */
-        n = byteCnt - i*SKEIN_512_BLOCK_BYTES;   /* number of output bytes left to go */
-        if (n >= SKEIN_512_BLOCK_BYTES)
-            n  = SKEIN_512_BLOCK_BYTES;
-        Skein_Put64_LSB_First(hashVal+i*SKEIN_512_BLOCK_BYTES, ctx->X, n);   /* "output" the ctr mode bytes */
-        Skein_Show_Final(256, &ctx->h, n, hashVal+i*SKEIN_512_BLOCK_BYTES);
-        memcpy(ctx->X, X, sizeof(X));   /* restore the counter mode key for next time */
-    }
-    return SKEIN_SUCCESS;
+	size_t i, n, byteCnt;
+	u64 X[SKEIN_512_STATE_WORDS];
+	Skein_Assert(ctx->h.bCnt <= SKEIN_512_BLOCK_BYTES, SKEIN_FAIL);    /* catch uninitialized context */
+
+	/* now output the result */
+	byteCnt = (ctx->h.hashBitLen + 7) >> 3;    /* total number of output bytes */
+
+	/* run Threefish in "counter mode" to generate output */
+	memset(ctx->b, 0, sizeof(ctx->b));  /* zero out b[], so it can hold the counter */
+	memcpy(X, ctx->X, sizeof(X));       /* keep a local copy of counter mode "key" */
+	for (i = 0; i*SKEIN_512_BLOCK_BYTES < byteCnt; i++)
+	{
+		((u64 *)ctx->b)[0] = Skein_Swap64((u64) i); /* build the counter block */
+		Skein_Start_New_Type(ctx, OUT_FINAL);
+		Skein_512_Process_Block(ctx, ctx->b, 1, sizeof(u64)); /* run "counter mode" */
+		n = byteCnt - i*SKEIN_512_BLOCK_BYTES;   /* number of output bytes left to go */
+		if (n >= SKEIN_512_BLOCK_BYTES)
+			n  = SKEIN_512_BLOCK_BYTES;
+		Skein_Put64_LSB_First(hashVal+i*SKEIN_512_BLOCK_BYTES, ctx->X, n);   /* "output" the ctr mode bytes */
+		Skein_Show_Final(256, &ctx->h, n, hashVal+i*SKEIN_512_BLOCK_BYTES);
+		memcpy(ctx->X, X, sizeof(X));   /* restore the counter mode key for next time */
+	}
+	return SKEIN_SUCCESS;
 }
 
 /*++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++*/
 /* just do the OUTPUT stage                                       */
 int Skein1024_Output(struct skein1024_ctx *ctx, u8 *hashVal)
 {
-    size_t i, n, byteCnt;
-    u64 X[SKEIN1024_STATE_WORDS];
-    Skein_Assert(ctx->h.bCnt <= SKEIN1024_BLOCK_BYTES, SKEIN_FAIL);    /* catch uninitialized context */
-
-    /* now output the result */
-    byteCnt = (ctx->h.hashBitLen + 7) >> 3;    /* total number of output bytes */
-
-    /* run Threefish in "counter mode" to generate output */
-    memset(ctx->b, 0, sizeof(ctx->b));  /* zero out b[], so it can hold the counter */
-    memcpy(X, ctx->X, sizeof(X));       /* keep a local copy of counter mode "key" */
-    for (i = 0; i*SKEIN1024_BLOCK_BYTES < byteCnt; i++)
-    {
-        ((u64 *)ctx->b)[0] = Skein_Swap64((u64) i); /* build the counter block */
-        Skein_Start_New_Type(ctx, OUT_FINAL);
-        Skein1024_Process_Block(ctx, ctx->b, 1, sizeof(u64)); /* run "counter mode" */
-        n = byteCnt - i*SKEIN1024_BLOCK_BYTES;   /* number of output bytes left to go */
-        if (n >= SKEIN1024_BLOCK_BYTES)
-            n  = SKEIN1024_BLOCK_BYTES;
-        Skein_Put64_LSB_First(hashVal+i*SKEIN1024_BLOCK_BYTES, ctx->X, n);   /* "output" the ctr mode bytes */
-        Skein_Show_Final(256, &ctx->h, n, hashVal+i*SKEIN1024_BLOCK_BYTES);
-        memcpy(ctx->X, X, sizeof(X));   /* restore the counter mode key for next time */
-    }
-    return SKEIN_SUCCESS;
+	size_t i, n, byteCnt;
+	u64 X[SKEIN1024_STATE_WORDS];
+	Skein_Assert(ctx->h.bCnt <= SKEIN1024_BLOCK_BYTES, SKEIN_FAIL);    /* catch uninitialized context */
+
+	/* now output the result */
+	byteCnt = (ctx->h.hashBitLen + 7) >> 3;    /* total number of output bytes */
+
+	/* run Threefish in "counter mode" to generate output */
+	memset(ctx->b, 0, sizeof(ctx->b));  /* zero out b[], so it can hold the counter */
+	memcpy(X, ctx->X, sizeof(X));       /* keep a local copy of counter mode "key" */
+	for (i = 0; i*SKEIN1024_BLOCK_BYTES < byteCnt; i++)
+	{
+		((u64 *)ctx->b)[0] = Skein_Swap64((u64) i); /* build the counter block */
+		Skein_Start_New_Type(ctx, OUT_FINAL);
+		Skein1024_Process_Block(ctx, ctx->b, 1, sizeof(u64)); /* run "counter mode" */
+		n = byteCnt - i*SKEIN1024_BLOCK_BYTES;   /* number of output bytes left to go */
+		if (n >= SKEIN1024_BLOCK_BYTES)
+			n  = SKEIN1024_BLOCK_BYTES;
+		Skein_Put64_LSB_First(hashVal+i*SKEIN1024_BLOCK_BYTES, ctx->X, n);   /* "output" the ctr mode bytes */
+		Skein_Show_Final(256, &ctx->h, n, hashVal+i*SKEIN1024_BLOCK_BYTES);
+		memcpy(ctx->X, X, sizeof(X));   /* restore the counter mode key for next time */
+	}
+	return SKEIN_SUCCESS;
 }
 #endif
diff --git a/drivers/staging/skein/skeinApi.c b/drivers/staging/skein/skeinApi.c
index a3f471be8db3..3ebb1d60ef93 100644
--- a/drivers/staging/skein/skeinApi.c
+++ b/drivers/staging/skein/skeinApi.c
@@ -29,191 +29,191 @@ OTHER DEALINGS IN THE SOFTWARE.
 
 int skeinCtxPrepare(struct skein_ctx *ctx, enum skein_size size)
 {
-    Skein_Assert(ctx && size, SKEIN_FAIL);
+	Skein_Assert(ctx && size, SKEIN_FAIL);
 
-    memset(ctx , 0, sizeof(struct skein_ctx));
-    ctx->skeinSize = size;
+	memset(ctx , 0, sizeof(struct skein_ctx));
+	ctx->skeinSize = size;
 
-    return SKEIN_SUCCESS;
+	return SKEIN_SUCCESS;
 }
 
 int skeinInit(struct skein_ctx *ctx, size_t hashBitLen)
 {
-    int ret = SKEIN_FAIL;
-    size_t Xlen = 0;
-    u64 *X = NULL;
-    u64 treeInfo = SKEIN_CFG_TREE_INFO_SEQUENTIAL;
-
-    Skein_Assert(ctx, SKEIN_FAIL);
-    /*
-     * The following two lines rely of the fact that the real Skein contexts are
-     * a union in out context and thus have tha maximum memory available.
-     * The beauty of C :-) .
-     */
-    X = ctx->m.s256.X;
-    Xlen = ctx->skeinSize/8;
-    /*
-     * If size is the same and hash bit length is zero then reuse
-     * the save chaining variables.
-     */
-    switch (ctx->skeinSize) {
-    case Skein256:
-        ret = Skein_256_InitExt(&ctx->m.s256, hashBitLen,
-                                treeInfo, NULL, 0);
-        break;
-    case Skein512:
-        ret = Skein_512_InitExt(&ctx->m.s512, hashBitLen,
-                                treeInfo, NULL, 0);
-        break;
-    case Skein1024:
-        ret = Skein1024_InitExt(&ctx->m.s1024, hashBitLen,
-                                treeInfo, NULL, 0);
-        break;
-    }
-
-    if (ret == SKEIN_SUCCESS) {
-        /* Save chaining variables for this combination of size and hashBitLen */
-        memcpy(ctx->XSave, X, Xlen);
-    }
-    return ret;
+	int ret = SKEIN_FAIL;
+	size_t Xlen = 0;
+	u64 *X = NULL;
+	u64 treeInfo = SKEIN_CFG_TREE_INFO_SEQUENTIAL;
+
+	Skein_Assert(ctx, SKEIN_FAIL);
+	/*
+	 * The following two lines rely of the fact that the real Skein contexts are
+	 * a union in out context and thus have tha maximum memory available.
+	 * The beauty of C :-) .
+	 */
+	X = ctx->m.s256.X;
+	Xlen = ctx->skeinSize/8;
+	/*
+	 * If size is the same and hash bit length is zero then reuse
+	 * the save chaining variables.
+	 */
+	switch (ctx->skeinSize) {
+	case Skein256:
+		ret = Skein_256_InitExt(&ctx->m.s256, hashBitLen,
+					treeInfo, NULL, 0);
+		break;
+	case Skein512:
+		ret = Skein_512_InitExt(&ctx->m.s512, hashBitLen,
+					treeInfo, NULL, 0);
+		break;
+	case Skein1024:
+		ret = Skein1024_InitExt(&ctx->m.s1024, hashBitLen,
+					treeInfo, NULL, 0);
+		break;
+	}
+
+	if (ret == SKEIN_SUCCESS) {
+		/* Save chaining variables for this combination of size and hashBitLen */
+		memcpy(ctx->XSave, X, Xlen);
+	}
+	return ret;
 }
 
 int skeinMacInit(struct skein_ctx *ctx, const u8 *key, size_t keyLen,
-                 size_t hashBitLen)
+		size_t hashBitLen)
 {
-    int ret = SKEIN_FAIL;
-    u64 *X = NULL;
-    size_t Xlen = 0;
-    u64 treeInfo = SKEIN_CFG_TREE_INFO_SEQUENTIAL;
-
-    Skein_Assert(ctx, SKEIN_FAIL);
-
-    X = ctx->m.s256.X;
-    Xlen = ctx->skeinSize/8;
-
-    Skein_Assert(hashBitLen, SKEIN_BAD_HASHLEN);
-
-    switch (ctx->skeinSize) {
-    case Skein256:
-        ret = Skein_256_InitExt(&ctx->m.s256, hashBitLen,
-                                treeInfo,
-                                (const u8 *)key, keyLen);
-
-        break;
-    case Skein512:
-        ret = Skein_512_InitExt(&ctx->m.s512, hashBitLen,
-                                treeInfo,
-                                (const u8 *)key, keyLen);
-        break;
-    case Skein1024:
-        ret = Skein1024_InitExt(&ctx->m.s1024, hashBitLen,
-                                treeInfo,
-                                (const u8 *)key, keyLen);
-
-        break;
-    }
-    if (ret == SKEIN_SUCCESS) {
-        /* Save chaining variables for this combination of key, keyLen, hashBitLen */
-        memcpy(ctx->XSave, X, Xlen);
-    }
-    return ret;
+	int ret = SKEIN_FAIL;
+	u64 *X = NULL;
+	size_t Xlen = 0;
+	u64 treeInfo = SKEIN_CFG_TREE_INFO_SEQUENTIAL;
+
+	Skein_Assert(ctx, SKEIN_FAIL);
+
+	X = ctx->m.s256.X;
+	Xlen = ctx->skeinSize/8;
+
+	Skein_Assert(hashBitLen, SKEIN_BAD_HASHLEN);
+
+	switch (ctx->skeinSize) {
+	case Skein256:
+		ret = Skein_256_InitExt(&ctx->m.s256, hashBitLen,
+					treeInfo,
+					(const u8 *)key, keyLen);
+
+		break;
+	case Skein512:
+		ret = Skein_512_InitExt(&ctx->m.s512, hashBitLen,
+					treeInfo,
+					(const u8 *)key, keyLen);
+		break;
+	case Skein1024:
+		ret = Skein1024_InitExt(&ctx->m.s1024, hashBitLen,
+					treeInfo,
+					(const u8 *)key, keyLen);
+
+		break;
+	}
+	if (ret == SKEIN_SUCCESS) {
+		/* Save chaining variables for this combination of key, keyLen, hashBitLen */
+		memcpy(ctx->XSave, X, Xlen);
+	}
+	return ret;
 }
 
 void skeinReset(struct skein_ctx *ctx)
 {
-    size_t Xlen = 0;
-    u64 *X = NULL;
-
-    /*
-     * The following two lines rely of the fact that the real Skein contexts are
-     * a union in out context and thus have tha maximum memory available.
-     * The beautiy of C :-) .
-     */
-    X = ctx->m.s256.X;
-    Xlen = ctx->skeinSize/8;
-    /* Restore the chaing variable, reset byte counter */
-    memcpy(X, ctx->XSave, Xlen);
-
-    /* Setup context to process the message */
-    Skein_Start_New_Type(&ctx->m, MSG);
+	size_t Xlen = 0;
+	u64 *X = NULL;
+
+	/*
+	 * The following two lines rely of the fact that the real Skein contexts are
+	 * a union in out context and thus have tha maximum memory available.
+	 * The beautiy of C :-) .
+	 */
+	X = ctx->m.s256.X;
+	Xlen = ctx->skeinSize/8;
+	/* Restore the chaing variable, reset byte counter */
+	memcpy(X, ctx->XSave, Xlen);
+
+	/* Setup context to process the message */
+	Skein_Start_New_Type(&ctx->m, MSG);
 }
 
 int skeinUpdate(struct skein_ctx *ctx, const u8 *msg,
-                size_t msgByteCnt)
+		size_t msgByteCnt)
 {
-    int ret = SKEIN_FAIL;
-    Skein_Assert(ctx, SKEIN_FAIL);
-
-    switch (ctx->skeinSize) {
-    case Skein256:
-        ret = Skein_256_Update(&ctx->m.s256, (const u8 *)msg, msgByteCnt);
-        break;
-    case Skein512:
-        ret = Skein_512_Update(&ctx->m.s512, (const u8 *)msg, msgByteCnt);
-        break;
-    case Skein1024:
-        ret = Skein1024_Update(&ctx->m.s1024, (const u8 *)msg, msgByteCnt);
-        break;
-    }
-    return ret;
+	int ret = SKEIN_FAIL;
+	Skein_Assert(ctx, SKEIN_FAIL);
+
+	switch (ctx->skeinSize) {
+	case Skein256:
+		ret = Skein_256_Update(&ctx->m.s256, (const u8 *)msg, msgByteCnt);
+		break;
+	case Skein512:
+		ret = Skein_512_Update(&ctx->m.s512, (const u8 *)msg, msgByteCnt);
+		break;
+	case Skein1024:
+		ret = Skein1024_Update(&ctx->m.s1024, (const u8 *)msg, msgByteCnt);
+		break;
+	}
+	return ret;
 
 }
 
 int skeinUpdateBits(struct skein_ctx *ctx, const u8 *msg,
-                    size_t msgBitCnt)
+			size_t msgBitCnt)
 {
-    /*
-     * I've used the bit pad implementation from skein_test.c (see NIST CD)
-     * and modified it to use the convenience functions and added some pointer
-     * arithmetic.
-     */
-    size_t length;
-    u8 mask;
-    u8 *up;
-
-    /* only the final Update() call is allowed do partial bytes, else assert an error */
-    Skein_Assert((ctx->m.h.T[1] & SKEIN_T1_FLAG_BIT_PAD) == 0 || msgBitCnt == 0, SKEIN_FAIL);
-
-    /* if number of bits is a multiple of bytes - that's easy */
-    if ((msgBitCnt & 0x7) == 0) {
-        return skeinUpdate(ctx, msg, msgBitCnt >> 3);
-    }
-    skeinUpdate(ctx, msg, (msgBitCnt >> 3) + 1);
-
-    /*
-     * The next line rely on the fact that the real Skein contexts
-     * are a union in our context. After the addition the pointer points to
-     * Skein's real partial block buffer.
-     * If this layout ever changes we have to adapt this as well.
-     */
-    up = (u8 *)ctx->m.s256.X + ctx->skeinSize / 8;
-
-    Skein_Set_Bit_Pad_Flag(ctx->m.h);                       /* set tweak flag for the skeinFinal call */
-
-    /* now "pad" the final partial byte the way NIST likes */
-    length = ctx->m.h.bCnt;                                 /* get the bCnt value (same location for all block sizes) */
-    Skein_assert(length != 0);                              /* internal sanity check: there IS a partial byte in the buffer! */
-    mask = (u8) (1u << (7 - (msgBitCnt & 7)));         /* partial byte bit mask */
-    up[length-1]  = (u8)((up[length-1] & (0-mask))|mask);   /* apply bit padding on final byte (in the buffer) */
-
-    return SKEIN_SUCCESS;
+	/*
+	 * I've used the bit pad implementation from skein_test.c (see NIST CD)
+	 * and modified it to use the convenience functions and added some pointer
+	 * arithmetic.
+	 */
+	size_t length;
+	u8 mask;
+	u8 *up;
+
+	/* only the final Update() call is allowed do partial bytes, else assert an error */
+	Skein_Assert((ctx->m.h.T[1] & SKEIN_T1_FLAG_BIT_PAD) == 0 || msgBitCnt == 0, SKEIN_FAIL);
+
+	/* if number of bits is a multiple of bytes - that's easy */
+	if ((msgBitCnt & 0x7) == 0) {
+		return skeinUpdate(ctx, msg, msgBitCnt >> 3);
+	}
+	skeinUpdate(ctx, msg, (msgBitCnt >> 3) + 1);
+
+	/*
+	 * The next line rely on the fact that the real Skein contexts
+	 * are a union in our context. After the addition the pointer points to
+	 * Skein's real partial block buffer.
+	 * If this layout ever changes we have to adapt this as well.
+	 */
+	up = (u8 *)ctx->m.s256.X + ctx->skeinSize / 8;
+
+	Skein_Set_Bit_Pad_Flag(ctx->m.h);                       /* set tweak flag for the skeinFinal call */
+
+	/* now "pad" the final partial byte the way NIST likes */
+	length = ctx->m.h.bCnt;                                 /* get the bCnt value (same location for all block sizes) */
+	Skein_assert(length != 0);                              /* internal sanity check: there IS a partial byte in the buffer! */
+	mask = (u8) (1u << (7 - (msgBitCnt & 7)));         /* partial byte bit mask */
+	up[length-1]  = (u8)((up[length-1] & (0-mask))|mask);   /* apply bit padding on final byte (in the buffer) */
+
+	return SKEIN_SUCCESS;
 }
 
 int skeinFinal(struct skein_ctx *ctx, u8 *hash)
 {
-    int ret = SKEIN_FAIL;
-    Skein_Assert(ctx, SKEIN_FAIL);
-
-    switch (ctx->skeinSize) {
-    case Skein256:
-        ret = Skein_256_Final(&ctx->m.s256, (u8 *)hash);
-        break;
-    case Skein512:
-        ret = Skein_512_Final(&ctx->m.s512, (u8 *)hash);
-        break;
-    case Skein1024:
-        ret = Skein1024_Final(&ctx->m.s1024, (u8 *)hash);
-        break;
-    }
-    return ret;
+	int ret = SKEIN_FAIL;
+	Skein_Assert(ctx, SKEIN_FAIL);
+
+	switch (ctx->skeinSize) {
+	case Skein256:
+		ret = Skein_256_Final(&ctx->m.s256, (u8 *)hash);
+		break;
+	case Skein512:
+		ret = Skein_512_Final(&ctx->m.s512, (u8 *)hash);
+		break;
+	case Skein1024:
+		ret = Skein1024_Final(&ctx->m.s1024, (u8 *)hash);
+		break;
+	}
+	return ret;
 }
diff --git a/drivers/staging/skein/skeinBlockNo3F.c b/drivers/staging/skein/skeinBlockNo3F.c
index d98933eeb0bf..3c2878c966e1 100644
--- a/drivers/staging/skein/skeinBlockNo3F.c
+++ b/drivers/staging/skein/skeinBlockNo3F.c
@@ -6,167 +6,167 @@
 
 /*****************************  Skein_256 ******************************/
 void Skein_256_Process_Block(struct skein_256_ctx *ctx, const u8 *blkPtr,
-                             size_t blkCnt, size_t byteCntAdd)
+				size_t blkCnt, size_t byteCntAdd)
 {
-    struct threefish_key key;
-    u64 tweak[2];
-    int i;
-    u64  w[SKEIN_256_STATE_WORDS];           /* local copy of input block */
-    u64 words[3];
+	struct threefish_key key;
+	u64 tweak[2];
+	int i;
+	u64  w[SKEIN_256_STATE_WORDS];           /* local copy of input block */
+	u64 words[3];
 
-    Skein_assert(blkCnt != 0);                  /* never call with blkCnt == 0! */
-    tweak[0] = ctx->h.T[0];
-    tweak[1] = ctx->h.T[1];
+	Skein_assert(blkCnt != 0);                  /* never call with blkCnt == 0! */
+	tweak[0] = ctx->h.T[0];
+	tweak[1] = ctx->h.T[1];
 
-    do  {
-        u64 carry = byteCntAdd;
+	do  {
+		u64 carry = byteCntAdd;
 
-        words[0] = tweak[0] & 0xffffffffL;
-        words[1] = ((tweak[0] >> 32) & 0xffffffffL);
-        words[2] = (tweak[1] & 0xffffffffL);
+		words[0] = tweak[0] & 0xffffffffL;
+		words[1] = ((tweak[0] >> 32) & 0xffffffffL);
+		words[2] = (tweak[1] & 0xffffffffL);
 
-        for (i = 0; i < 3; i++) {
-            carry += words[i];
-            words[i] = carry;
-            carry >>= 32;
-        }        
-        tweak[0] = words[0] & 0xffffffffL;
-        tweak[0] |= (words[1] & 0xffffffffL) << 32;
-        tweak[1] |= words[2] & 0xffffffffL;
+		for (i = 0; i < 3; i++) {
+			carry += words[i];
+			words[i] = carry;
+			carry >>= 32;
+		}        
+		tweak[0] = words[0] & 0xffffffffL;
+		tweak[0] |= (words[1] & 0xffffffffL) << 32;
+		tweak[1] |= words[2] & 0xffffffffL;
 
-        threefishSetKey(&key, Threefish256, ctx->X, tweak);
+		threefishSetKey(&key, Threefish256, ctx->X, tweak);
 
-        Skein_Get64_LSB_First(w, blkPtr, SKEIN_256_STATE_WORDS);   /* get input block in little-endian format */
+		Skein_Get64_LSB_First(w, blkPtr, SKEIN_256_STATE_WORDS);   /* get input block in little-endian format */
 
-        threefishEncryptBlockWords(&key, w, ctx->X);
+		threefishEncryptBlockWords(&key, w, ctx->X);
 
-        blkPtr += SKEIN_256_BLOCK_BYTES;
+		blkPtr += SKEIN_256_BLOCK_BYTES;
 
-        /* do the final "feedforward" xor, update context chaining vars */
-        ctx->X[0] = ctx->X[0] ^ w[0];
-        ctx->X[1] = ctx->X[1] ^ w[1];
-        ctx->X[2] = ctx->X[2] ^ w[2];
-        ctx->X[3] = ctx->X[3] ^ w[3];
+		/* do the final "feedforward" xor, update context chaining vars */
+		ctx->X[0] = ctx->X[0] ^ w[0];
+		ctx->X[1] = ctx->X[1] ^ w[1];
+		ctx->X[2] = ctx->X[2] ^ w[2];
+		ctx->X[3] = ctx->X[3] ^ w[3];
 
-        tweak[1] &= ~SKEIN_T1_FLAG_FIRST;
-    } while (--blkCnt);
+		tweak[1] &= ~SKEIN_T1_FLAG_FIRST;
+	} while (--blkCnt);
 
-    ctx->h.T[0] = tweak[0];
-    ctx->h.T[1] = tweak[1];
+	ctx->h.T[0] = tweak[0];
+	ctx->h.T[1] = tweak[1];
 }
 
 void Skein_512_Process_Block(struct skein_512_ctx *ctx, const u8 *blkPtr,
-                             size_t blkCnt, size_t byteCntAdd)
+				size_t blkCnt, size_t byteCntAdd)
 {
-    struct threefish_key key;
-    u64 tweak[2];
-    int i;
-    u64 words[3];
-    u64  w[SKEIN_512_STATE_WORDS];           /* local copy of input block */
-
-    Skein_assert(blkCnt != 0);                  /* never call with blkCnt == 0! */
-    tweak[0] = ctx->h.T[0];
-    tweak[1] = ctx->h.T[1];
-
-    do  {
-        u64 carry = byteCntAdd;
-
-        words[0] = tweak[0] & 0xffffffffL;
-        words[1] = ((tweak[0] >> 32) & 0xffffffffL);
-        words[2] = (tweak[1] & 0xffffffffL);
-
-        for (i = 0; i < 3; i++) {
-            carry += words[i];
-            words[i] = carry;
-            carry >>= 32;
-        }        
-        tweak[0] = words[0] & 0xffffffffL;
-        tweak[0] |= (words[1] & 0xffffffffL) << 32;
-        tweak[1] |= words[2] & 0xffffffffL;
-
-        threefishSetKey(&key, Threefish512, ctx->X, tweak);
-
-        Skein_Get64_LSB_First(w, blkPtr, SKEIN_512_STATE_WORDS);   /* get input block in little-endian format */
-
-        threefishEncryptBlockWords(&key, w, ctx->X);
-
-        blkPtr += SKEIN_512_BLOCK_BYTES;
-
-        /* do the final "feedforward" xor, update context chaining vars */
-        ctx->X[0] = ctx->X[0] ^ w[0];
-        ctx->X[1] = ctx->X[1] ^ w[1];
-        ctx->X[2] = ctx->X[2] ^ w[2];
-        ctx->X[3] = ctx->X[3] ^ w[3];
-        ctx->X[4] = ctx->X[4] ^ w[4];
-        ctx->X[5] = ctx->X[5] ^ w[5];
-        ctx->X[6] = ctx->X[6] ^ w[6];
-        ctx->X[7] = ctx->X[7] ^ w[7];
-
-        tweak[1] &= ~SKEIN_T1_FLAG_FIRST;
-    } while (--blkCnt);
-
-    ctx->h.T[0] = tweak[0];
-    ctx->h.T[1] = tweak[1];
+	struct threefish_key key;
+	u64 tweak[2];
+	int i;
+	u64 words[3];
+	u64  w[SKEIN_512_STATE_WORDS];           /* local copy of input block */
+
+	Skein_assert(blkCnt != 0);                  /* never call with blkCnt == 0! */
+	tweak[0] = ctx->h.T[0];
+	tweak[1] = ctx->h.T[1];
+
+	do  {
+		u64 carry = byteCntAdd;
+
+		words[0] = tweak[0] & 0xffffffffL;
+		words[1] = ((tweak[0] >> 32) & 0xffffffffL);
+		words[2] = (tweak[1] & 0xffffffffL);
+
+		for (i = 0; i < 3; i++) {
+			carry += words[i];
+			words[i] = carry;
+			carry >>= 32;
+		}        
+		tweak[0] = words[0] & 0xffffffffL;
+		tweak[0] |= (words[1] & 0xffffffffL) << 32;
+		tweak[1] |= words[2] & 0xffffffffL;
+
+		threefishSetKey(&key, Threefish512, ctx->X, tweak);
+
+		Skein_Get64_LSB_First(w, blkPtr, SKEIN_512_STATE_WORDS);   /* get input block in little-endian format */
+
+		threefishEncryptBlockWords(&key, w, ctx->X);
+
+		blkPtr += SKEIN_512_BLOCK_BYTES;
+
+		/* do the final "feedforward" xor, update context chaining vars */
+		ctx->X[0] = ctx->X[0] ^ w[0];
+		ctx->X[1] = ctx->X[1] ^ w[1];
+		ctx->X[2] = ctx->X[2] ^ w[2];
+		ctx->X[3] = ctx->X[3] ^ w[3];
+		ctx->X[4] = ctx->X[4] ^ w[4];
+		ctx->X[5] = ctx->X[5] ^ w[5];
+		ctx->X[6] = ctx->X[6] ^ w[6];
+		ctx->X[7] = ctx->X[7] ^ w[7];
+
+		tweak[1] &= ~SKEIN_T1_FLAG_FIRST;
+	} while (--blkCnt);
+
+	ctx->h.T[0] = tweak[0];
+	ctx->h.T[1] = tweak[1];
 }
 
 void Skein1024_Process_Block(struct skein1024_ctx *ctx, const u8 *blkPtr,
-                              size_t blkCnt, size_t byteCntAdd)
+				size_t blkCnt, size_t byteCntAdd)
 {
-    struct threefish_key key;
-    u64 tweak[2];
-    int i;
-    u64 words[3];
-    u64  w[SKEIN1024_STATE_WORDS];           /* local copy of input block */
-
-    Skein_assert(blkCnt != 0);                  /* never call with blkCnt == 0! */
-    tweak[0] = ctx->h.T[0];
-    tweak[1] = ctx->h.T[1];
-
-    do  {
-        u64 carry = byteCntAdd;
-
-        words[0] = tweak[0] & 0xffffffffL;
-        words[1] = ((tweak[0] >> 32) & 0xffffffffL);
-        words[2] = (tweak[1] & 0xffffffffL);
-
-        for (i = 0; i < 3; i++) {
-            carry += words[i];
-            words[i] = carry;
-            carry >>= 32;
-        }        
-        tweak[0] = words[0] & 0xffffffffL;
-        tweak[0] |= (words[1] & 0xffffffffL) << 32;
-        tweak[1] |= words[2] & 0xffffffffL;
-
-        threefishSetKey(&key, Threefish1024, ctx->X, tweak);
-
-        Skein_Get64_LSB_First(w, blkPtr, SKEIN1024_STATE_WORDS);   /* get input block in little-endian format */
-
-        threefishEncryptBlockWords(&key, w, ctx->X);
-
-        blkPtr += SKEIN1024_BLOCK_BYTES;
-
-        /* do the final "feedforward" xor, update context chaining vars */
-        ctx->X[0]  = ctx->X[0]  ^ w[0];
-        ctx->X[1]  = ctx->X[1]  ^ w[1];
-        ctx->X[2]  = ctx->X[2]  ^ w[2];
-        ctx->X[3]  = ctx->X[3]  ^ w[3];
-        ctx->X[4]  = ctx->X[4]  ^ w[4];
-        ctx->X[5]  = ctx->X[5]  ^ w[5];
-        ctx->X[6]  = ctx->X[6]  ^ w[6];
-        ctx->X[7]  = ctx->X[7]  ^ w[7];
-        ctx->X[8]  = ctx->X[8]  ^ w[8];
-        ctx->X[9]  = ctx->X[9]  ^ w[9];
-        ctx->X[10] = ctx->X[10] ^ w[10];
-        ctx->X[11] = ctx->X[11] ^ w[11];
-        ctx->X[12] = ctx->X[12] ^ w[12];
-        ctx->X[13] = ctx->X[13] ^ w[13];
-        ctx->X[14] = ctx->X[14] ^ w[14];
-        ctx->X[15] = ctx->X[15] ^ w[15];
-
-        tweak[1] &= ~SKEIN_T1_FLAG_FIRST;
-    } while (--blkCnt);
-
-    ctx->h.T[0] = tweak[0];
-    ctx->h.T[1] = tweak[1];
+	struct threefish_key key;
+	u64 tweak[2];
+	int i;
+	u64 words[3];
+	u64  w[SKEIN1024_STATE_WORDS];           /* local copy of input block */
+
+	Skein_assert(blkCnt != 0);                  /* never call with blkCnt == 0! */
+	tweak[0] = ctx->h.T[0];
+	tweak[1] = ctx->h.T[1];
+
+	do  {
+		u64 carry = byteCntAdd;
+
+		words[0] = tweak[0] & 0xffffffffL;
+		words[1] = ((tweak[0] >> 32) & 0xffffffffL);
+		words[2] = (tweak[1] & 0xffffffffL);
+
+		for (i = 0; i < 3; i++) {
+			carry += words[i];
+			words[i] = carry;
+			carry >>= 32;
+		}        
+		tweak[0] = words[0] & 0xffffffffL;
+		tweak[0] |= (words[1] & 0xffffffffL) << 32;
+		tweak[1] |= words[2] & 0xffffffffL;
+
+		threefishSetKey(&key, Threefish1024, ctx->X, tweak);
+
+		Skein_Get64_LSB_First(w, blkPtr, SKEIN1024_STATE_WORDS);   /* get input block in little-endian format */
+
+		threefishEncryptBlockWords(&key, w, ctx->X);
+
+		blkPtr += SKEIN1024_BLOCK_BYTES;
+
+		/* do the final "feedforward" xor, update context chaining vars */
+		ctx->X[0]  = ctx->X[0]  ^ w[0];
+		ctx->X[1]  = ctx->X[1]  ^ w[1];
+		ctx->X[2]  = ctx->X[2]  ^ w[2];
+		ctx->X[3]  = ctx->X[3]  ^ w[3];
+		ctx->X[4]  = ctx->X[4]  ^ w[4];
+		ctx->X[5]  = ctx->X[5]  ^ w[5];
+		ctx->X[6]  = ctx->X[6]  ^ w[6];
+		ctx->X[7]  = ctx->X[7]  ^ w[7];
+		ctx->X[8]  = ctx->X[8]  ^ w[8];
+		ctx->X[9]  = ctx->X[9]  ^ w[9];
+		ctx->X[10] = ctx->X[10] ^ w[10];
+		ctx->X[11] = ctx->X[11] ^ w[11];
+		ctx->X[12] = ctx->X[12] ^ w[12];
+		ctx->X[13] = ctx->X[13] ^ w[13];
+		ctx->X[14] = ctx->X[14] ^ w[14];
+		ctx->X[15] = ctx->X[15] ^ w[15];
+
+		tweak[1] &= ~SKEIN_T1_FLAG_FIRST;
+	} while (--blkCnt);
+
+	ctx->h.T[0] = tweak[0];
+	ctx->h.T[1] = tweak[1];
 }
diff --git a/drivers/staging/skein/skein_block.c b/drivers/staging/skein/skein_block.c
index e62b6442783e..bb36860fafdf 100644
--- a/drivers/staging/skein/skein_block.c
+++ b/drivers/staging/skein/skein_block.c
@@ -40,10 +40,10 @@
 /*****************************  Skein_256 ******************************/
 #if !(SKEIN_USE_ASM & 256)
 void Skein_256_Process_Block(struct skein_256_ctx *ctx, const u8 *blkPtr, size_t blkCnt, size_t byteCntAdd)
-    { /* do it in C */
-    enum {
-        WCNT = SKEIN_256_STATE_WORDS
-        };
+	{ /* do it in C */
+	enum {
+		WCNT = SKEIN_256_STATE_WORDS
+	};
 #undef  RCNT
 #define RCNT  (SKEIN_256_ROUNDS_TOTAL/8)
 
@@ -57,177 +57,177 @@ void Skein_256_Process_Block(struct skein_256_ctx *ctx, const u8 *blkPtr, size_t
 #if (RCNT % SKEIN_UNROLL_256)
 #error "Invalid SKEIN_UNROLL_256"               /* sanity check on unroll count */
 #endif
-    size_t  r;
-    u64  kw[WCNT+4+RCNT*2];                  /* key schedule words : chaining vars + tweak + "rotation"*/
+	size_t  r;
+	u64  kw[WCNT+4+RCNT*2];                  /* key schedule words : chaining vars + tweak + "rotation"*/
 #else
-    u64  kw[WCNT+4];                         /* key schedule words : chaining vars + tweak */
+	u64  kw[WCNT+4];                         /* key schedule words : chaining vars + tweak */
 #endif
-    u64  X0, X1, X2, X3;                        /* local copy of context vars, for speed */
-    u64  w[WCNT];                           /* local copy of input block */
+	u64  X0, X1, X2, X3;                        /* local copy of context vars, for speed */
+	u64  w[WCNT];                           /* local copy of input block */
 #ifdef SKEIN_DEBUG
-    const u64 *Xptr[4];                      /* use for debugging (help compiler put Xn in registers) */
-    Xptr[0] = &X0;  Xptr[1] = &X1;  Xptr[2] = &X2;  Xptr[3] = &X3;
+	const u64 *Xptr[4];                      /* use for debugging (help compiler put Xn in registers) */
+	Xptr[0] = &X0;  Xptr[1] = &X1;  Xptr[2] = &X2;  Xptr[3] = &X3;
 #endif
-    Skein_assert(blkCnt != 0);                  /* never call with blkCnt == 0! */
-    ts[0] = ctx->h.T[0];
-    ts[1] = ctx->h.T[1];
-    do  {
-        /* this implementation only supports 2**64 input bytes (no carry out here) */
-        ts[0] += byteCntAdd;                    /* update processed length */
+	Skein_assert(blkCnt != 0);                  /* never call with blkCnt == 0! */
+	ts[0] = ctx->h.T[0];
+	ts[1] = ctx->h.T[1];
+	do  {
+		/* this implementation only supports 2**64 input bytes (no carry out here) */
+		ts[0] += byteCntAdd;                    /* update processed length */
 
-        /* precompute the key schedule for this block */
-        ks[0] = ctx->X[0];     
-        ks[1] = ctx->X[1];
-        ks[2] = ctx->X[2];
-        ks[3] = ctx->X[3];
-        ks[4] = ks[0] ^ ks[1] ^ ks[2] ^ ks[3] ^ SKEIN_KS_PARITY;
+		/* precompute the key schedule for this block */
+		ks[0] = ctx->X[0];     
+		ks[1] = ctx->X[1];
+		ks[2] = ctx->X[2];
+		ks[3] = ctx->X[3];
+		ks[4] = ks[0] ^ ks[1] ^ ks[2] ^ ks[3] ^ SKEIN_KS_PARITY;
 
-        ts[2] = ts[0] ^ ts[1];
+		ts[2] = ts[0] ^ ts[1];
 
-        Skein_Get64_LSB_First(w, blkPtr, WCNT);   /* get input block in little-endian format */
-        DebugSaveTweak(ctx);
-        Skein_Show_Block(BLK_BITS, &ctx->h, ctx->X, blkPtr, w, ks, ts);
+		Skein_Get64_LSB_First(w, blkPtr, WCNT);   /* get input block in little-endian format */
+		DebugSaveTweak(ctx);
+		Skein_Show_Block(BLK_BITS, &ctx->h, ctx->X, blkPtr, w, ks, ts);
 
-        X0 = w[0] + ks[0];                      /* do the first full key injection */
-        X1 = w[1] + ks[1] + ts[0];
-        X2 = w[2] + ks[2] + ts[1];
-        X3 = w[3] + ks[3];
+		X0 = w[0] + ks[0];                      /* do the first full key injection */
+		X1 = w[1] + ks[1] + ts[0];
+		X2 = w[2] + ks[2] + ts[1];
+		X3 = w[3] + ks[3];
 
-        Skein_Show_R_Ptr(BLK_BITS, &ctx->h, SKEIN_RND_KEY_INITIAL, Xptr);    /* show starting state values */
+		Skein_Show_R_Ptr(BLK_BITS, &ctx->h, SKEIN_RND_KEY_INITIAL, Xptr);    /* show starting state values */
 
-        blkPtr += SKEIN_256_BLOCK_BYTES;
+		blkPtr += SKEIN_256_BLOCK_BYTES;
 
-        /* run the rounds */
+		/* run the rounds */
 
 #define Round256(p0, p1, p2, p3, ROT, rNum)                              \
-    X##p0 += X##p1; X##p1 = RotL_64(X##p1, ROT##_0); X##p1 ^= X##p0; \
-    X##p2 += X##p3; X##p3 = RotL_64(X##p3, ROT##_1); X##p3 ^= X##p2; \
+	X##p0 += X##p1; X##p1 = RotL_64(X##p1, ROT##_0); X##p1 ^= X##p0; \
+	X##p2 += X##p3; X##p3 = RotL_64(X##p3, ROT##_1); X##p3 ^= X##p2; \
 
 #if SKEIN_UNROLL_256 == 0                       
 #define R256(p0, p1, p2, p3, ROT, rNum)           /* fully unrolled */   \
-    Round256(p0, p1, p2, p3, ROT, rNum)                                  \
-    Skein_Show_R_Ptr(BLK_BITS, &ctx->h, rNum, Xptr);
+	Round256(p0, p1, p2, p3, ROT, rNum)                                  \
+	Skein_Show_R_Ptr(BLK_BITS, &ctx->h, rNum, Xptr);
 
 #define I256(R)                                                     \
-    X0   += ks[((R)+1) % 5];    /* inject the key schedule value */ \
-    X1   += ks[((R)+2) % 5] + ts[((R)+1) % 3];                      \
-    X2   += ks[((R)+3) % 5] + ts[((R)+2) % 3];                      \
-    X3   += ks[((R)+4) % 5] +     (R)+1;                            \
-    Skein_Show_R_Ptr(BLK_BITS, &ctx->h, SKEIN_RND_KEY_INJECT, Xptr);
+	X0   += ks[((R)+1) % 5];    /* inject the key schedule value */ \
+	X1   += ks[((R)+2) % 5] + ts[((R)+1) % 3];                      \
+	X2   += ks[((R)+3) % 5] + ts[((R)+2) % 3];                      \
+	X3   += ks[((R)+4) % 5] +     (R)+1;                            \
+	Skein_Show_R_Ptr(BLK_BITS, &ctx->h, SKEIN_RND_KEY_INJECT, Xptr);
 #else                                       /* looping version */
 #define R256(p0, p1, p2, p3, ROT, rNum)                                  \
-    Round256(p0, p1, p2, p3, ROT, rNum)                                  \
-    Skein_Show_R_Ptr(BLK_BITS, &ctx->h, 4 * (r - 1) + rNum, Xptr);
+	Round256(p0, p1, p2, p3, ROT, rNum)                                  \
+	Skein_Show_R_Ptr(BLK_BITS, &ctx->h, 4 * (r - 1) + rNum, Xptr);
 
 #define I256(R)                                                     \
-    X0   += ks[r+(R)+0];        /* inject the key schedule value */ \
-    X1   += ks[r+(R)+1] + ts[r+(R)+0];                              \
-    X2   += ks[r+(R)+2] + ts[r+(R)+1];                              \
-    X3   += ks[r+(R)+3] +    r+(R);                              \
-    ks[r + (R) + 4]   = ks[r + (R) - 1];     /* rotate key schedule */\
-    ts[r + (R) + 2]   = ts[r + (R) - 1];                              \
-    Skein_Show_R_Ptr(BLK_BITS, &ctx->h, SKEIN_RND_KEY_INJECT, Xptr);
-
-    for (r = 1; r < 2 * RCNT; r += 2 * SKEIN_UNROLL_256)  /* loop thru it */
+	X0   += ks[r+(R)+0];        /* inject the key schedule value */ \
+	X1   += ks[r+(R)+1] + ts[r+(R)+0];                              \
+	X2   += ks[r+(R)+2] + ts[r+(R)+1];                              \
+	X3   += ks[r+(R)+3] +    r+(R);                              \
+	ks[r + (R) + 4]   = ks[r + (R) - 1];     /* rotate key schedule */\
+	ts[r + (R) + 2]   = ts[r + (R) - 1];                              \
+	Skein_Show_R_Ptr(BLK_BITS, &ctx->h, SKEIN_RND_KEY_INJECT, Xptr);
+
+	for (r = 1; r < 2 * RCNT; r += 2 * SKEIN_UNROLL_256)  /* loop thru it */
 #endif  
-        {    
+		{    
 #define R256_8_rounds(R)                  \
-        R256(0, 1, 2, 3, R_256_0, 8 * (R) + 1);  \
-        R256(0, 3, 2, 1, R_256_1, 8 * (R) + 2);  \
-        R256(0, 1, 2, 3, R_256_2, 8 * (R) + 3);  \
-        R256(0, 3, 2, 1, R_256_3, 8 * (R) + 4);  \
-        I256(2 * (R));                      \
-        R256(0, 1, 2, 3, R_256_4, 8 * (R) + 5);  \
-        R256(0, 3, 2, 1, R_256_5, 8 * (R) + 6);  \
-        R256(0, 1, 2, 3, R_256_6, 8 * (R) + 7);  \
-        R256(0, 3, 2, 1, R_256_7, 8 * (R) + 8);  \
-        I256(2 * (R) + 1);
-
-        R256_8_rounds(0);
+		R256(0, 1, 2, 3, R_256_0, 8 * (R) + 1);  \
+		R256(0, 3, 2, 1, R_256_1, 8 * (R) + 2);  \
+		R256(0, 1, 2, 3, R_256_2, 8 * (R) + 3);  \
+		R256(0, 3, 2, 1, R_256_3, 8 * (R) + 4);  \
+		I256(2 * (R));                      \
+		R256(0, 1, 2, 3, R_256_4, 8 * (R) + 5);  \
+		R256(0, 3, 2, 1, R_256_5, 8 * (R) + 6);  \
+		R256(0, 1, 2, 3, R_256_6, 8 * (R) + 7);  \
+		R256(0, 3, 2, 1, R_256_7, 8 * (R) + 8);  \
+		I256(2 * (R) + 1);
+
+		R256_8_rounds(0);
 
 #define R256_Unroll_R(NN) ((SKEIN_UNROLL_256 == 0 && SKEIN_256_ROUNDS_TOTAL/8 > (NN)) || (SKEIN_UNROLL_256 > (NN)))
 
-  #if   R256_Unroll_R(1)
-        R256_8_rounds(1);
-  #endif
-  #if   R256_Unroll_R(2)
-        R256_8_rounds(2);
-  #endif
-  #if   R256_Unroll_R(3)
-        R256_8_rounds(3);
-  #endif
-  #if   R256_Unroll_R(4)
-        R256_8_rounds(4);
-  #endif
-  #if   R256_Unroll_R(5)
-        R256_8_rounds(5);
-  #endif
-  #if   R256_Unroll_R(6)
-        R256_8_rounds(6);
-  #endif
-  #if   R256_Unroll_R(7)
-        R256_8_rounds(7);
-  #endif
-  #if   R256_Unroll_R(8)
-        R256_8_rounds(8);
-  #endif
-  #if   R256_Unroll_R(9)
-        R256_8_rounds(9);
-  #endif
-  #if   R256_Unroll_R(10)
-        R256_8_rounds(10);
-  #endif
-  #if   R256_Unroll_R(11)
-        R256_8_rounds(11);
-  #endif
-  #if   R256_Unroll_R(12)
-        R256_8_rounds(12);
-  #endif
-  #if   R256_Unroll_R(13)
-        R256_8_rounds(13);
-  #endif
-  #if   R256_Unroll_R(14)
-        R256_8_rounds(14);
-  #endif
-  #if  (SKEIN_UNROLL_256 > 14)
+	#if   R256_Unroll_R(1)
+		R256_8_rounds(1);
+	#endif
+	#if   R256_Unroll_R(2)
+		R256_8_rounds(2);
+	#endif
+	#if   R256_Unroll_R(3)
+		R256_8_rounds(3);
+	#endif
+	#if   R256_Unroll_R(4)
+		R256_8_rounds(4);
+	#endif
+	#if   R256_Unroll_R(5)
+		R256_8_rounds(5);
+	#endif
+	#if   R256_Unroll_R(6)
+		R256_8_rounds(6);
+	#endif
+	#if   R256_Unroll_R(7)
+		R256_8_rounds(7);
+	#endif
+	#if   R256_Unroll_R(8)
+		R256_8_rounds(8);
+	#endif
+	#if   R256_Unroll_R(9)
+		R256_8_rounds(9);
+	#endif
+	#if   R256_Unroll_R(10)
+		R256_8_rounds(10);
+	#endif
+	#if   R256_Unroll_R(11)
+		R256_8_rounds(11);
+	#endif
+	#if   R256_Unroll_R(12)
+		R256_8_rounds(12);
+	#endif
+	#if   R256_Unroll_R(13)
+		R256_8_rounds(13);
+	#endif
+	#if   R256_Unroll_R(14)
+		R256_8_rounds(14);
+	#endif
+	#if  (SKEIN_UNROLL_256 > 14)
 #error  "need more unrolling in Skein_256_Process_Block"
-  #endif
-        }
-        /* do the final "feedforward" xor, update context chaining vars */
-        ctx->X[0] = X0 ^ w[0];
-        ctx->X[1] = X1 ^ w[1];
-        ctx->X[2] = X2 ^ w[2];
-        ctx->X[3] = X3 ^ w[3];
-
-        Skein_Show_Round(BLK_BITS, &ctx->h, SKEIN_RND_FEED_FWD, ctx->X);
-
-        ts[1] &= ~SKEIN_T1_FLAG_FIRST;
-        }
-    while (--blkCnt);
-    ctx->h.T[0] = ts[0];
-    ctx->h.T[1] = ts[1];
-    }
+	#endif
+		}
+		/* do the final "feedforward" xor, update context chaining vars */
+		ctx->X[0] = X0 ^ w[0];
+		ctx->X[1] = X1 ^ w[1];
+		ctx->X[2] = X2 ^ w[2];
+		ctx->X[3] = X3 ^ w[3];
+
+		Skein_Show_Round(BLK_BITS, &ctx->h, SKEIN_RND_FEED_FWD, ctx->X);
+
+		ts[1] &= ~SKEIN_T1_FLAG_FIRST;
+	}
+	while (--blkCnt);
+	ctx->h.T[0] = ts[0];
+	ctx->h.T[1] = ts[1];
+}
 
 #if defined(SKEIN_CODE_SIZE) || defined(SKEIN_PERF)
 size_t Skein_256_Process_Block_CodeSize(void)
-    {
-    return ((u8 *) Skein_256_Process_Block_CodeSize) -
-           ((u8 *) Skein_256_Process_Block);
-    }
+{
+	return ((u8 *) Skein_256_Process_Block_CodeSize) -
+		((u8 *) Skein_256_Process_Block);
+}
 unsigned int Skein_256_Unroll_Cnt(void)
-    {
-    return SKEIN_UNROLL_256;
-    }
+{
+	return SKEIN_UNROLL_256;
+}
 #endif
 #endif
 
 /*****************************  Skein_512 ******************************/
 #if !(SKEIN_USE_ASM & 512)
 void Skein_512_Process_Block(struct skein_512_ctx *ctx, const u8 *blkPtr, size_t blkCnt, size_t byteCntAdd)
-    { /* do it in C */
-    enum {
-        WCNT = SKEIN_512_STATE_WORDS
-        };
+{ /* do it in C */
+	enum {
+		WCNT = SKEIN_512_STATE_WORDS
+	};
 #undef  RCNT
 #define RCNT  (SKEIN_512_ROUNDS_TOTAL/8)
 
@@ -241,200 +241,200 @@ void Skein_512_Process_Block(struct skein_512_ctx *ctx, const u8 *blkPtr, size_t
 #if (RCNT % SKEIN_UNROLL_512)
 #error "Invalid SKEIN_UNROLL_512"               /* sanity check on unroll count */
 #endif
-    size_t  r;
-    u64  kw[WCNT+4+RCNT*2];                  /* key schedule words : chaining vars + tweak + "rotation"*/
+	size_t  r;
+	u64  kw[WCNT+4+RCNT*2];                  /* key schedule words : chaining vars + tweak + "rotation"*/
 #else
-    u64  kw[WCNT+4];                         /* key schedule words : chaining vars + tweak */
+	u64  kw[WCNT+4];                         /* key schedule words : chaining vars + tweak */
 #endif
-    u64  X0, X1, X2, X3, X4, X5, X6, X7;            /* local copy of vars,  for speed */
-    u64  w[WCNT];                           /* local copy of input block */
+	u64  X0, X1, X2, X3, X4, X5, X6, X7;            /* local copy of vars,  for speed */
+	u64  w[WCNT];                           /* local copy of input block */
 #ifdef SKEIN_DEBUG
-    const u64 *Xptr[8];                      /* use for debugging (help compiler put Xn in registers) */
-    Xptr[0] = &X0;  Xptr[1] = &X1;  Xptr[2] = &X2;  Xptr[3] = &X3;
-    Xptr[4] = &X4;  Xptr[5] = &X5;  Xptr[6] = &X6;  Xptr[7] = &X7;
+	const u64 *Xptr[8];                      /* use for debugging (help compiler put Xn in registers) */
+	Xptr[0] = &X0;  Xptr[1] = &X1;  Xptr[2] = &X2;  Xptr[3] = &X3;
+	Xptr[4] = &X4;  Xptr[5] = &X5;  Xptr[6] = &X6;  Xptr[7] = &X7;
 #endif
 
-    Skein_assert(blkCnt != 0);                  /* never call with blkCnt == 0! */
-    ts[0] = ctx->h.T[0];
-    ts[1] = ctx->h.T[1];
-    do  {
-        /* this implementation only supports 2**64 input bytes (no carry out here) */
-        ts[0] += byteCntAdd;                    /* update processed length */
-
-        /* precompute the key schedule for this block */
-        ks[0] = ctx->X[0];
-        ks[1] = ctx->X[1];
-        ks[2] = ctx->X[2];
-        ks[3] = ctx->X[3];
-        ks[4] = ctx->X[4];
-        ks[5] = ctx->X[5];
-        ks[6] = ctx->X[6];
-        ks[7] = ctx->X[7];
-        ks[8] = ks[0] ^ ks[1] ^ ks[2] ^ ks[3] ^ 
-                ks[4] ^ ks[5] ^ ks[6] ^ ks[7] ^ SKEIN_KS_PARITY;
-
-        ts[2] = ts[0] ^ ts[1];
-
-        Skein_Get64_LSB_First(w, blkPtr, WCNT); /* get input block in little-endian format */
-        DebugSaveTweak(ctx);
-        Skein_Show_Block(BLK_BITS, &ctx->h, ctx->X, blkPtr, w, ks, ts);
-
-        X0   = w[0] + ks[0];                    /* do the first full key injection */
-        X1   = w[1] + ks[1];
-        X2   = w[2] + ks[2];
-        X3   = w[3] + ks[3];
-        X4   = w[4] + ks[4];
-        X5   = w[5] + ks[5] + ts[0];
-        X6   = w[6] + ks[6] + ts[1];
-        X7   = w[7] + ks[7];
-
-        blkPtr += SKEIN_512_BLOCK_BYTES;
-
-        Skein_Show_R_Ptr(BLK_BITS, &ctx->h, SKEIN_RND_KEY_INITIAL, Xptr);
-        /* run the rounds */
+	Skein_assert(blkCnt != 0);                  /* never call with blkCnt == 0! */
+	ts[0] = ctx->h.T[0];
+	ts[1] = ctx->h.T[1];
+	do  {
+		/* this implementation only supports 2**64 input bytes (no carry out here) */
+		ts[0] += byteCntAdd;                    /* update processed length */
+
+		/* precompute the key schedule for this block */
+		ks[0] = ctx->X[0];
+		ks[1] = ctx->X[1];
+		ks[2] = ctx->X[2];
+		ks[3] = ctx->X[3];
+		ks[4] = ctx->X[4];
+		ks[5] = ctx->X[5];
+		ks[6] = ctx->X[6];
+		ks[7] = ctx->X[7];
+		ks[8] = ks[0] ^ ks[1] ^ ks[2] ^ ks[3] ^ 
+			ks[4] ^ ks[5] ^ ks[6] ^ ks[7] ^ SKEIN_KS_PARITY;
+
+		ts[2] = ts[0] ^ ts[1];
+
+		Skein_Get64_LSB_First(w, blkPtr, WCNT); /* get input block in little-endian format */
+		DebugSaveTweak(ctx);
+		Skein_Show_Block(BLK_BITS, &ctx->h, ctx->X, blkPtr, w, ks, ts);
+
+		X0   = w[0] + ks[0];                    /* do the first full key injection */
+		X1   = w[1] + ks[1];
+		X2   = w[2] + ks[2];
+		X3   = w[3] + ks[3];
+		X4   = w[4] + ks[4];
+		X5   = w[5] + ks[5] + ts[0];
+		X6   = w[6] + ks[6] + ts[1];
+		X7   = w[7] + ks[7];
+
+		blkPtr += SKEIN_512_BLOCK_BYTES;
+
+		Skein_Show_R_Ptr(BLK_BITS, &ctx->h, SKEIN_RND_KEY_INITIAL, Xptr);
+		/* run the rounds */
 #define Round512(p0, p1, p2, p3, p4, p5, p6, p7, ROT, rNum)                  \
-    X##p0 += X##p1; X##p1 = RotL_64(X##p1, ROT##_0); X##p1 ^= X##p0; \
-    X##p2 += X##p3; X##p3 = RotL_64(X##p3, ROT##_1); X##p3 ^= X##p2; \
-    X##p4 += X##p5; X##p5 = RotL_64(X##p5, ROT##_2); X##p5 ^= X##p4; \
-    X##p6 += X##p7; X##p7 = RotL_64(X##p7, ROT##_3); X##p7 ^= X##p6; \
+		X##p0 += X##p1; X##p1 = RotL_64(X##p1, ROT##_0); X##p1 ^= X##p0; \
+		X##p2 += X##p3; X##p3 = RotL_64(X##p3, ROT##_1); X##p3 ^= X##p2; \
+		X##p4 += X##p5; X##p5 = RotL_64(X##p5, ROT##_2); X##p5 ^= X##p4; \
+		X##p6 += X##p7; X##p7 = RotL_64(X##p7, ROT##_3); X##p7 ^= X##p6; \
 
 #if SKEIN_UNROLL_512 == 0                       
 #define R512(p0, p1, p2, p3, p4, p5, p6, p7, ROT, rNum)      /* unrolled */  \
-    Round512(p0, p1, p2, p3, p4, p5, p6, p7, ROT, rNum)                      \
-    Skein_Show_R_Ptr(BLK_BITS, &ctx->h, rNum, Xptr);
+		Round512(p0, p1, p2, p3, p4, p5, p6, p7, ROT, rNum)                      \
+		Skein_Show_R_Ptr(BLK_BITS, &ctx->h, rNum, Xptr);
 
 #define I512(R)                                                     \
-    X0   += ks[((R) + 1) % 9];   /* inject the key schedule value */  \
-    X1   += ks[((R) + 2) % 9];                                        \
-    X2   += ks[((R) + 3) % 9];                                        \
-    X3   += ks[((R) + 4) % 9];                                        \
-    X4   += ks[((R) + 5) % 9];                                        \
-    X5   += ks[((R) + 6) % 9] + ts[((R) + 1) % 3];                      \
-    X6   += ks[((R) + 7) % 9] + ts[((R) + 2) % 3];                      \
-    X7   += ks[((R) + 8) % 9] +     (R) + 1;                            \
-    Skein_Show_R_Ptr(BLK_BITS, &ctx->h, SKEIN_RND_KEY_INJECT, Xptr);
+		X0   += ks[((R) + 1) % 9];   /* inject the key schedule value */  \
+		X1   += ks[((R) + 2) % 9];                                        \
+		X2   += ks[((R) + 3) % 9];                                        \
+		X3   += ks[((R) + 4) % 9];                                        \
+		X4   += ks[((R) + 5) % 9];                                        \
+		X5   += ks[((R) + 6) % 9] + ts[((R) + 1) % 3];                      \
+		X6   += ks[((R) + 7) % 9] + ts[((R) + 2) % 3];                      \
+		X7   += ks[((R) + 8) % 9] +     (R) + 1;                            \
+		Skein_Show_R_Ptr(BLK_BITS, &ctx->h, SKEIN_RND_KEY_INJECT, Xptr);
 #else                                       /* looping version */
 #define R512(p0, p1, p2, p3, p4, p5, p6, p7, ROT, rNum)                      \
-    Round512(p0, p1, p2, p3, p4, p5, p6, p7, ROT, rNum)                      \
-    Skein_Show_R_Ptr(BLK_BITS, &ctx->h, 4 * (r - 1) + rNum, Xptr);
+		Round512(p0, p1, p2, p3, p4, p5, p6, p7, ROT, rNum)                      \
+		Skein_Show_R_Ptr(BLK_BITS, &ctx->h, 4 * (r - 1) + rNum, Xptr);
 
 #define I512(R)                                                     \
-    X0   += ks[r + (R) + 0];        /* inject the key schedule value */ \
-    X1   += ks[r + (R) + 1];                                            \
-    X2   += ks[r + (R) + 2];                                            \
-    X3   += ks[r + (R) + 3];                                            \
-    X4   += ks[r + (R) + 4];                                            \
-    X5   += ks[r + (R) + 5] + ts[r + (R) + 0];                              \
-    X6   += ks[r + (R) + 6] + ts[r + (R) + 1];                              \
-    X7   += ks[r + (R) + 7] +         r + (R);                              \
-    ks[r +         (R) + 8] = ks[r + (R) - 1];  /* rotate key schedule */   \
-    ts[r +         (R) + 2] = ts[r + (R) - 1];                              \
-    Skein_Show_R_Ptr(BLK_BITS, &ctx->h, SKEIN_RND_KEY_INJECT, Xptr);
-
-    for (r = 1; r < 2 * RCNT; r += 2 * SKEIN_UNROLL_512)   /* loop thru it */
+		X0   += ks[r + (R) + 0];        /* inject the key schedule value */ \
+		X1   += ks[r + (R) + 1];                                            \
+		X2   += ks[r + (R) + 2];                                            \
+		X3   += ks[r + (R) + 3];                                            \
+		X4   += ks[r + (R) + 4];                                            \
+		X5   += ks[r + (R) + 5] + ts[r + (R) + 0];                              \
+		X6   += ks[r + (R) + 6] + ts[r + (R) + 1];                              \
+		X7   += ks[r + (R) + 7] +         r + (R);                              \
+		ks[r +         (R) + 8] = ks[r + (R) - 1];  /* rotate key schedule */   \
+		ts[r +         (R) + 2] = ts[r + (R) - 1];                              \
+		Skein_Show_R_Ptr(BLK_BITS, &ctx->h, SKEIN_RND_KEY_INJECT, Xptr);
+
+		for (r = 1; r < 2 * RCNT; r += 2 * SKEIN_UNROLL_512)   /* loop thru it */
 #endif                         /* end of looped code definitions */
-        {
+		{
 #define R512_8_rounds(R)  /* do 8 full rounds */  \
-        R512(0, 1, 2, 3, 4, 5, 6, 7, R_512_0, 8 * (R) + 1);   \
-        R512(2, 1, 4, 7, 6, 5, 0, 3, R_512_1, 8 * (R) + 2);   \
-        R512(4, 1, 6, 3, 0, 5, 2, 7, R_512_2, 8 * (R) + 3);   \
-        R512(6, 1, 0, 7, 2, 5, 4, 3, R_512_3, 8 * (R) + 4);   \
-        I512(2 * (R));                              \
-        R512(0, 1, 2, 3, 4, 5, 6, 7, R_512_4, 8 * (R) + 5);   \
-        R512(2, 1, 4, 7, 6, 5, 0, 3, R_512_5, 8 * (R) + 6);   \
-        R512(4, 1, 6, 3, 0, 5, 2, 7, R_512_6, 8 * (R) + 7);   \
-        R512(6, 1, 0, 7, 2, 5, 4, 3, R_512_7, 8 * (R) + 8);   \
-        I512(2 * (R) + 1);        /* and key injection */
-
-        R512_8_rounds(0);
+			R512(0, 1, 2, 3, 4, 5, 6, 7, R_512_0, 8 * (R) + 1);   \
+			R512(2, 1, 4, 7, 6, 5, 0, 3, R_512_1, 8 * (R) + 2);   \
+			R512(4, 1, 6, 3, 0, 5, 2, 7, R_512_2, 8 * (R) + 3);   \
+			R512(6, 1, 0, 7, 2, 5, 4, 3, R_512_3, 8 * (R) + 4);   \
+			I512(2 * (R));                              \
+			R512(0, 1, 2, 3, 4, 5, 6, 7, R_512_4, 8 * (R) + 5);   \
+			R512(2, 1, 4, 7, 6, 5, 0, 3, R_512_5, 8 * (R) + 6);   \
+			R512(4, 1, 6, 3, 0, 5, 2, 7, R_512_6, 8 * (R) + 7);   \
+			R512(6, 1, 0, 7, 2, 5, 4, 3, R_512_7, 8 * (R) + 8);   \
+			I512(2 * (R) + 1);        /* and key injection */
+
+			R512_8_rounds(0);
 
 #define R512_Unroll_R(NN) ((SKEIN_UNROLL_512 == 0 && SKEIN_512_ROUNDS_TOTAL/8 > (NN)) || (SKEIN_UNROLL_512 > (NN)))
 
-  #if   R512_Unroll_R(1)
-        R512_8_rounds(1);
-  #endif
-  #if   R512_Unroll_R(2)
-        R512_8_rounds(2);
-  #endif
-  #if   R512_Unroll_R(3)
-        R512_8_rounds(3);
-  #endif
-  #if   R512_Unroll_R(4)
-        R512_8_rounds(4);
-  #endif
-  #if   R512_Unroll_R(5)
-        R512_8_rounds(5);
-  #endif
-  #if   R512_Unroll_R(6)
-        R512_8_rounds(6);
-  #endif
-  #if   R512_Unroll_R(7)
-        R512_8_rounds(7);
-  #endif
-  #if   R512_Unroll_R(8)
-        R512_8_rounds(8);
-  #endif
-  #if   R512_Unroll_R(9)
-        R512_8_rounds(9);
-  #endif
-  #if   R512_Unroll_R(10)
-        R512_8_rounds(10);
-  #endif
-  #if   R512_Unroll_R(11)
-        R512_8_rounds(11);
-  #endif
-  #if   R512_Unroll_R(12)
-        R512_8_rounds(12);
-  #endif
-  #if   R512_Unroll_R(13)
-        R512_8_rounds(13);
-  #endif
-  #if   R512_Unroll_R(14)
-        R512_8_rounds(14);
-  #endif
-  #if  (SKEIN_UNROLL_512 > 14)
+	#if   R512_Unroll_R(1)
+			R512_8_rounds(1);
+	#endif
+	#if   R512_Unroll_R(2)
+			R512_8_rounds(2);
+	#endif
+	#if   R512_Unroll_R(3)
+			R512_8_rounds(3);
+	#endif
+	#if   R512_Unroll_R(4)
+			R512_8_rounds(4);
+	#endif
+	#if   R512_Unroll_R(5)
+			R512_8_rounds(5);
+	#endif
+	#if   R512_Unroll_R(6)
+			R512_8_rounds(6);
+	#endif
+	#if   R512_Unroll_R(7)
+			R512_8_rounds(7);
+	#endif
+	#if   R512_Unroll_R(8)
+			R512_8_rounds(8);
+	#endif
+	#if   R512_Unroll_R(9)
+			R512_8_rounds(9);
+	#endif
+	#if   R512_Unroll_R(10)
+			R512_8_rounds(10);
+	#endif
+	#if   R512_Unroll_R(11)
+			R512_8_rounds(11);
+	#endif
+	#if   R512_Unroll_R(12)
+			R512_8_rounds(12);
+	#endif
+	#if   R512_Unroll_R(13)
+			R512_8_rounds(13);
+	#endif
+	#if   R512_Unroll_R(14)
+			R512_8_rounds(14);
+	#endif
+	#if  (SKEIN_UNROLL_512 > 14)
 #error  "need more unrolling in Skein_512_Process_Block"
-  #endif
-        }
-
-        /* do the final "feedforward" xor, update context chaining vars */
-        ctx->X[0] = X0 ^ w[0];
-        ctx->X[1] = X1 ^ w[1];
-        ctx->X[2] = X2 ^ w[2];
-        ctx->X[3] = X3 ^ w[3];
-        ctx->X[4] = X4 ^ w[4];
-        ctx->X[5] = X5 ^ w[5];
-        ctx->X[6] = X6 ^ w[6];
-        ctx->X[7] = X7 ^ w[7];
-        Skein_Show_Round(BLK_BITS, &ctx->h, SKEIN_RND_FEED_FWD, ctx->X);
-
-        ts[1] &= ~SKEIN_T1_FLAG_FIRST;
-        }
-    while (--blkCnt);
-    ctx->h.T[0] = ts[0];
-    ctx->h.T[1] = ts[1];
-    }
+	#endif
+		}
+
+		/* do the final "feedforward" xor, update context chaining vars */
+		ctx->X[0] = X0 ^ w[0];
+		ctx->X[1] = X1 ^ w[1];
+		ctx->X[2] = X2 ^ w[2];
+		ctx->X[3] = X3 ^ w[3];
+		ctx->X[4] = X4 ^ w[4];
+		ctx->X[5] = X5 ^ w[5];
+		ctx->X[6] = X6 ^ w[6];
+		ctx->X[7] = X7 ^ w[7];
+		Skein_Show_Round(BLK_BITS, &ctx->h, SKEIN_RND_FEED_FWD, ctx->X);
+
+		ts[1] &= ~SKEIN_T1_FLAG_FIRST;
+	}
+	while (--blkCnt);
+	ctx->h.T[0] = ts[0];
+	ctx->h.T[1] = ts[1];
+}
 
 #if defined(SKEIN_CODE_SIZE) || defined(SKEIN_PERF)
 size_t Skein_512_Process_Block_CodeSize(void)
-    {
-    return ((u8 *) Skein_512_Process_Block_CodeSize) -
-           ((u8 *) Skein_512_Process_Block);
-    }
+{
+	return ((u8 *) Skein_512_Process_Block_CodeSize) -
+		((u8 *) Skein_512_Process_Block);
+}
 unsigned int Skein_512_Unroll_Cnt(void)
-    {
-    return SKEIN_UNROLL_512;
-    }
+{
+	return SKEIN_UNROLL_512;
+}
 #endif
 #endif
 
 /*****************************  Skein1024 ******************************/
 #if !(SKEIN_USE_ASM & 1024)
 void Skein1024_Process_Block(struct skein1024_ctx *ctx, const u8 *blkPtr, size_t blkCnt, size_t byteCntAdd)
-    { /* do it in C, always looping (unrolled is bigger AND slower!) */
-    enum {
-        WCNT = SKEIN1024_STATE_WORDS
-        };
+{ /* do it in C, always looping (unrolled is bigger AND slower!) */
+	enum {
+		WCNT = SKEIN1024_STATE_WORDS
+	};
 #undef  RCNT
 #define RCNT  (SKEIN1024_ROUNDS_TOTAL/8)
 
@@ -448,239 +448,239 @@ void Skein1024_Process_Block(struct skein1024_ctx *ctx, const u8 *blkPtr, size_t
 #if (RCNT % SKEIN_UNROLL_1024)
 #error "Invalid SKEIN_UNROLL_1024"              /* sanity check on unroll count */
 #endif
-    size_t  r;
-    u64  kw[WCNT+4+RCNT*2];                  /* key schedule words : chaining vars + tweak + "rotation"*/
+	size_t  r;
+	u64  kw[WCNT+4+RCNT*2];                  /* key schedule words : chaining vars + tweak + "rotation"*/
 #else
-    u64  kw[WCNT+4];                         /* key schedule words : chaining vars + tweak */
+	u64  kw[WCNT+4];                         /* key schedule words : chaining vars + tweak */
 #endif
 
-    u64  X00, X01, X02, X03, X04, X05, X06, X07,     /* local copy of vars, for speed */
-            X08, X09, X10, X11, X12, X13, X14, X15;
-    u64  w[WCNT];                            /* local copy of input block */
+	u64  X00, X01, X02, X03, X04, X05, X06, X07,     /* local copy of vars, for speed */
+		X08, X09, X10, X11, X12, X13, X14, X15;
+	u64  w[WCNT];                            /* local copy of input block */
 #ifdef SKEIN_DEBUG
-    const u64 *Xptr[16];                     /* use for debugging (help compiler put Xn in registers) */
-    Xptr[0]  = &X00;  Xptr[1]  = &X01;  Xptr[2]  = &X02;  Xptr[3]  = &X03;
-    Xptr[4]  = &X04;  Xptr[5]  = &X05;  Xptr[6]  = &X06;  Xptr[7]  = &X07;
-    Xptr[8]  = &X08;  Xptr[9]  = &X09;  Xptr[10] = &X10;  Xptr[11] = &X11;
-    Xptr[12] = &X12;  Xptr[13] = &X13;  Xptr[14] = &X14;  Xptr[15] = &X15;
+	const u64 *Xptr[16];                     /* use for debugging (help compiler put Xn in registers) */
+	Xptr[0]  = &X00;  Xptr[1]  = &X01;  Xptr[2]  = &X02;  Xptr[3]  = &X03;
+	Xptr[4]  = &X04;  Xptr[5]  = &X05;  Xptr[6]  = &X06;  Xptr[7]  = &X07;
+	Xptr[8]  = &X08;  Xptr[9]  = &X09;  Xptr[10] = &X10;  Xptr[11] = &X11;
+	Xptr[12] = &X12;  Xptr[13] = &X13;  Xptr[14] = &X14;  Xptr[15] = &X15;
 #endif
 
-    Skein_assert(blkCnt != 0);                  /* never call with blkCnt == 0! */
-    ts[0] = ctx->h.T[0];
-    ts[1] = ctx->h.T[1];
-    do  {
-        /* this implementation only supports 2**64 input bytes (no carry out here) */
-        ts[0] += byteCntAdd;                    /* update processed length */
-
-        /* precompute the key schedule for this block */
-        ks[0]  = ctx->X[0];
-        ks[1]  = ctx->X[1];
-        ks[2]  = ctx->X[2];
-        ks[3]  = ctx->X[3];
-        ks[4]  = ctx->X[4];
-        ks[5]  = ctx->X[5];
-        ks[6]  = ctx->X[6];
-        ks[7]  = ctx->X[7];
-        ks[8]  = ctx->X[8];
-        ks[9]  = ctx->X[9];
-        ks[10] = ctx->X[10];
-        ks[11] = ctx->X[11];
-        ks[12] = ctx->X[12];
-        ks[13] = ctx->X[13];
-        ks[14] = ctx->X[14];
-        ks[15] = ctx->X[15];
-        ks[16] =  ks[0] ^  ks[1] ^  ks[2] ^  ks[3] ^
-                  ks[4] ^  ks[5] ^  ks[6] ^  ks[7] ^
-                  ks[8] ^  ks[9] ^ ks[10] ^ ks[11] ^
-                 ks[12] ^ ks[13] ^ ks[14] ^ ks[15] ^ SKEIN_KS_PARITY;
-
-        ts[2]  = ts[0] ^ ts[1];
-
-        Skein_Get64_LSB_First(w, blkPtr, WCNT); /* get input block in little-endian format */
-        DebugSaveTweak(ctx);
-        Skein_Show_Block(BLK_BITS, &ctx->h, ctx->X, blkPtr, w, ks, ts);
-
-        X00    =  w[0] +  ks[0];                 /* do the first full key injection */
-        X01    =  w[1] +  ks[1];
-        X02    =  w[2] +  ks[2];
-        X03    =  w[3] +  ks[3];
-        X04    =  w[4] +  ks[4];
-        X05    =  w[5] +  ks[5];
-        X06    =  w[6] +  ks[6];
-        X07    =  w[7] +  ks[7];
-        X08    =  w[8] +  ks[8];
-        X09    =  w[9] +  ks[9];
-        X10    = w[10] + ks[10];
-        X11    = w[11] + ks[11];
-        X12    = w[12] + ks[12];
-        X13    = w[13] + ks[13] + ts[0];
-        X14    = w[14] + ks[14] + ts[1];
-        X15    = w[15] + ks[15];
-
-        Skein_Show_R_Ptr(BLK_BITS, &ctx->h, SKEIN_RND_KEY_INITIAL, Xptr);
+	Skein_assert(blkCnt != 0);                  /* never call with blkCnt == 0! */
+	ts[0] = ctx->h.T[0];
+	ts[1] = ctx->h.T[1];
+	do  {
+		/* this implementation only supports 2**64 input bytes (no carry out here) */
+		ts[0] += byteCntAdd;                    /* update processed length */
+
+		/* precompute the key schedule for this block */
+		ks[0]  = ctx->X[0];
+		ks[1]  = ctx->X[1];
+		ks[2]  = ctx->X[2];
+		ks[3]  = ctx->X[3];
+		ks[4]  = ctx->X[4];
+		ks[5]  = ctx->X[5];
+		ks[6]  = ctx->X[6];
+		ks[7]  = ctx->X[7];
+		ks[8]  = ctx->X[8];
+		ks[9]  = ctx->X[9];
+		ks[10] = ctx->X[10];
+		ks[11] = ctx->X[11];
+		ks[12] = ctx->X[12];
+		ks[13] = ctx->X[13];
+		ks[14] = ctx->X[14];
+		ks[15] = ctx->X[15];
+		ks[16] =  ks[0] ^  ks[1] ^  ks[2] ^  ks[3] ^
+			  ks[4] ^  ks[5] ^  ks[6] ^  ks[7] ^
+			  ks[8] ^  ks[9] ^ ks[10] ^ ks[11] ^
+			  ks[12] ^ ks[13] ^ ks[14] ^ ks[15] ^ SKEIN_KS_PARITY;
+
+		ts[2]  = ts[0] ^ ts[1];
+
+		Skein_Get64_LSB_First(w, blkPtr, WCNT); /* get input block in little-endian format */
+		DebugSaveTweak(ctx);
+		Skein_Show_Block(BLK_BITS, &ctx->h, ctx->X, blkPtr, w, ks, ts);
+
+		X00    =  w[0] +  ks[0];                 /* do the first full key injection */
+		X01    =  w[1] +  ks[1];
+		X02    =  w[2] +  ks[2];
+		X03    =  w[3] +  ks[3];
+		X04    =  w[4] +  ks[4];
+		X05    =  w[5] +  ks[5];
+		X06    =  w[6] +  ks[6];
+		X07    =  w[7] +  ks[7];
+		X08    =  w[8] +  ks[8];
+		X09    =  w[9] +  ks[9];
+		X10    = w[10] + ks[10];
+		X11    = w[11] + ks[11];
+		X12    = w[12] + ks[12];
+		X13    = w[13] + ks[13] + ts[0];
+		X14    = w[14] + ks[14] + ts[1];
+		X15    = w[15] + ks[15];
+
+		Skein_Show_R_Ptr(BLK_BITS, &ctx->h, SKEIN_RND_KEY_INITIAL, Xptr);
 
 #define Round1024(p0, p1, p2, p3, p4, p5, p6, p7, p8, p9, pA, pB, pC, pD, pE, pF, ROT, rNum) \
-    X##p0 += X##p1; X##p1 = RotL_64(X##p1, ROT##_0); X##p1 ^= X##p0;   \
-    X##p2 += X##p3; X##p3 = RotL_64(X##p3, ROT##_1); X##p3 ^= X##p2;   \
-    X##p4 += X##p5; X##p5 = RotL_64(X##p5, ROT##_2); X##p5 ^= X##p4;   \
-    X##p6 += X##p7; X##p7 = RotL_64(X##p7, ROT##_3); X##p7 ^= X##p6;   \
-    X##p8 += X##p9; X##p9 = RotL_64(X##p9, ROT##_4); X##p9 ^= X##p8;   \
-    X##pA += X##pB; X##pB = RotL_64(X##pB, ROT##_5); X##pB ^= X##pA;   \
-    X##pC += X##pD; X##pD = RotL_64(X##pD, ROT##_6); X##pD ^= X##pC;   \
-    X##pE += X##pF; X##pF = RotL_64(X##pF, ROT##_7); X##pF ^= X##pE;   \
+		X##p0 += X##p1; X##p1 = RotL_64(X##p1, ROT##_0); X##p1 ^= X##p0;   \
+		X##p2 += X##p3; X##p3 = RotL_64(X##p3, ROT##_1); X##p3 ^= X##p2;   \
+		X##p4 += X##p5; X##p5 = RotL_64(X##p5, ROT##_2); X##p5 ^= X##p4;   \
+		X##p6 += X##p7; X##p7 = RotL_64(X##p7, ROT##_3); X##p7 ^= X##p6;   \
+		X##p8 += X##p9; X##p9 = RotL_64(X##p9, ROT##_4); X##p9 ^= X##p8;   \
+		X##pA += X##pB; X##pB = RotL_64(X##pB, ROT##_5); X##pB ^= X##pA;   \
+		X##pC += X##pD; X##pD = RotL_64(X##pD, ROT##_6); X##pD ^= X##pC;   \
+		X##pE += X##pF; X##pF = RotL_64(X##pF, ROT##_7); X##pF ^= X##pE;   \
 
 #if SKEIN_UNROLL_1024 == 0                      
 #define R1024(p0, p1, p2, p3, p4, p5, p6, p7, p8, p9, pA, pB, pC, pD, pE, pF, ROT, rn) \
-    Round1024(p0, p1, p2, p3, p4, p5, p6, p7, p8, p9, pA, pB, pC, pD, pE, pF, ROT, rn) \
-    Skein_Show_R_Ptr(BLK_BITS, &ctx->h, rn, Xptr);
+		Round1024(p0, p1, p2, p3, p4, p5, p6, p7, p8, p9, pA, pB, pC, pD, pE, pF, ROT, rn) \
+		Skein_Show_R_Ptr(BLK_BITS, &ctx->h, rn, Xptr);
 
 #define I1024(R)                                                        \
-    X00   += ks[((R) +  1) % 17]; /* inject the key schedule value */   \
-    X01   += ks[((R) +  2) % 17];                                       \
-    X02   += ks[((R) +  3) % 17];                                       \
-    X03   += ks[((R) +  4) % 17];                                       \
-    X04   += ks[((R) +  5) % 17];                                       \
-    X05   += ks[((R) +  6) % 17];                                       \
-    X06   += ks[((R) +  7) % 17];                                       \
-    X07   += ks[((R) +  8) % 17];                                       \
-    X08   += ks[((R) +  9) % 17];                                       \
-    X09   += ks[((R) + 10) % 17];                                       \
-    X10   += ks[((R) + 11) % 17];                                       \
-    X11   += ks[((R) + 12) % 17];                                       \
-    X12   += ks[((R) + 13) % 17];                                       \
-    X13   += ks[((R) + 14) % 17] + ts[((R) + 1) % 3];                   \
-    X14   += ks[((R) + 15) % 17] + ts[((R) + 2) % 3];                   \
-    X15   += ks[((R) + 16) % 17] +     (R) + 1;                         \
-    Skein_Show_R_Ptr(BLK_BITS, &ctx->h, SKEIN_RND_KEY_INJECT, Xptr); 
+		X00   += ks[((R) +  1) % 17]; /* inject the key schedule value */   \
+		X01   += ks[((R) +  2) % 17];                                       \
+		X02   += ks[((R) +  3) % 17];                                       \
+		X03   += ks[((R) +  4) % 17];                                       \
+		X04   += ks[((R) +  5) % 17];                                       \
+		X05   += ks[((R) +  6) % 17];                                       \
+		X06   += ks[((R) +  7) % 17];                                       \
+		X07   += ks[((R) +  8) % 17];                                       \
+		X08   += ks[((R) +  9) % 17];                                       \
+		X09   += ks[((R) + 10) % 17];                                       \
+		X10   += ks[((R) + 11) % 17];                                       \
+		X11   += ks[((R) + 12) % 17];                                       \
+		X12   += ks[((R) + 13) % 17];                                       \
+		X13   += ks[((R) + 14) % 17] + ts[((R) + 1) % 3];                   \
+		X14   += ks[((R) + 15) % 17] + ts[((R) + 2) % 3];                   \
+		X15   += ks[((R) + 16) % 17] +     (R) + 1;                         \
+		Skein_Show_R_Ptr(BLK_BITS, &ctx->h, SKEIN_RND_KEY_INJECT, Xptr); 
 #else                                       /* looping version */
 #define R1024(p0, p1, p2, p3, p4, p5, p6, p7, p8, p9, pA, pB, pC, pD, pE, pF, ROT, rn) \
-    Round1024(p0, p1, p2, p3, p4, p5, p6, p7, p8, p9, pA, pB, pC, pD, pE, pF, ROT, rn) \
-    Skein_Show_R_Ptr(BLK_BITS, &ctx->h, 4 * (r - 1) + rn, Xptr);
+		Round1024(p0, p1, p2, p3, p4, p5, p6, p7, p8, p9, pA, pB, pC, pD, pE, pF, ROT, rn) \
+		Skein_Show_R_Ptr(BLK_BITS, &ctx->h, 4 * (r - 1) + rn, Xptr);
 
 #define I1024(R)                                                      \
-    X00   += ks[r + (R) +  0];    /* inject the key schedule value */     \
-    X01   += ks[r + (R) +  1];                                            \
-    X02   += ks[r + (R) +  2];                                            \
-    X03   += ks[r + (R) +  3];                                            \
-    X04   += ks[r + (R) +  4];                                            \
-    X05   += ks[r + (R) +  5];                                            \
-    X06   += ks[r + (R) +  6];                                            \
-    X07   += ks[r + (R) +  7];                                            \
-    X08   += ks[r + (R) +  8];                                            \
-    X09   += ks[r + (R) +  9];                                            \
-    X10   += ks[r + (R) + 10];                                            \
-    X11   += ks[r + (R) + 11];                                            \
-    X12   += ks[r + (R) + 12];                                            \
-    X13   += ks[r + (R) + 13] + ts[r + (R) + 0];                          \
-    X14   += ks[r + (R) + 14] + ts[r + (R) + 1];                          \
-    X15   += ks[r + (R) + 15] +         r + (R);                          \
-    ks[r  +         (R) + 16] = ks[r + (R) - 1]; /* rotate key schedule */\
-    ts[r  +         (R) +  2] = ts[r + (R) - 1];                          \
-    Skein_Show_R_Ptr(BLK_BITSi, &ctx->h, SKEIN_RND_KEY_INJECT, Xptr);
-
-    for (r = 1; r <= 2 * RCNT; r += 2 * SKEIN_UNROLL_1024)    /* loop thru it */
+		X00   += ks[r + (R) +  0];    /* inject the key schedule value */     \
+		X01   += ks[r + (R) +  1];                                            \
+		X02   += ks[r + (R) +  2];                                            \
+		X03   += ks[r + (R) +  3];                                            \
+		X04   += ks[r + (R) +  4];                                            \
+		X05   += ks[r + (R) +  5];                                            \
+		X06   += ks[r + (R) +  6];                                            \
+		X07   += ks[r + (R) +  7];                                            \
+		X08   += ks[r + (R) +  8];                                            \
+		X09   += ks[r + (R) +  9];                                            \
+		X10   += ks[r + (R) + 10];                                            \
+		X11   += ks[r + (R) + 11];                                            \
+		X12   += ks[r + (R) + 12];                                            \
+		X13   += ks[r + (R) + 13] + ts[r + (R) + 0];                          \
+		X14   += ks[r + (R) + 14] + ts[r + (R) + 1];                          \
+		X15   += ks[r + (R) + 15] +         r + (R);                          \
+		ks[r  +         (R) + 16] = ks[r + (R) - 1]; /* rotate key schedule */\
+		ts[r  +         (R) +  2] = ts[r + (R) - 1];                          \
+		Skein_Show_R_Ptr(BLK_BITSi, &ctx->h, SKEIN_RND_KEY_INJECT, Xptr);
+
+		for (r = 1; r <= 2 * RCNT; r += 2 * SKEIN_UNROLL_1024)    /* loop thru it */
 #endif  
-        {
+		{
 #define R1024_8_rounds(R)    /* do 8 full rounds */                               \
-        R1024(00, 01, 02, 03, 04, 05, 06, 07, 08, 09, 10, 11, 12, 13, 14, 15, R1024_0, 8*(R) + 1); \
-        R1024(00, 09, 02, 13, 06, 11, 04, 15, 10, 07, 12, 03, 14, 05, 08, 01, R1024_1, 8*(R) + 2); \
-        R1024(00, 07, 02, 05, 04, 03, 06, 01, 12, 15, 14, 13, 08, 11, 10, 09, R1024_2, 8*(R) + 3); \
-        R1024(00, 15, 02, 11, 06, 13, 04, 09, 14, 01, 08, 05, 10, 03, 12, 07, R1024_3, 8*(R) + 4); \
-        I1024(2*(R));                                                             \
-        R1024(00, 01, 02, 03, 04, 05, 06, 07, 08, 09, 10, 11, 12, 13, 14, 15, R1024_4, 8*(R) + 5); \
-        R1024(00, 09, 02, 13, 06, 11, 04, 15, 10, 07, 12, 03, 14, 05, 08, 01, R1024_5, 8*(R) + 6); \
-        R1024(00, 07, 02, 05, 04, 03, 06, 01, 12, 15, 14, 13, 08, 11, 10, 09, R1024_6, 8*(R) + 7); \
-        R1024(00, 15, 02, 11, 06, 13, 04, 09, 14, 01, 08, 05, 10, 03, 12, 07, R1024_7, 8*(R) + 8); \
-        I1024(2*(R)+1);
-
-        R1024_8_rounds(0);
+			R1024(00, 01, 02, 03, 04, 05, 06, 07, 08, 09, 10, 11, 12, 13, 14, 15, R1024_0, 8*(R) + 1); \
+			R1024(00, 09, 02, 13, 06, 11, 04, 15, 10, 07, 12, 03, 14, 05, 08, 01, R1024_1, 8*(R) + 2); \
+			R1024(00, 07, 02, 05, 04, 03, 06, 01, 12, 15, 14, 13, 08, 11, 10, 09, R1024_2, 8*(R) + 3); \
+			R1024(00, 15, 02, 11, 06, 13, 04, 09, 14, 01, 08, 05, 10, 03, 12, 07, R1024_3, 8*(R) + 4); \
+			I1024(2*(R));                                                             \
+			R1024(00, 01, 02, 03, 04, 05, 06, 07, 08, 09, 10, 11, 12, 13, 14, 15, R1024_4, 8*(R) + 5); \
+			R1024(00, 09, 02, 13, 06, 11, 04, 15, 10, 07, 12, 03, 14, 05, 08, 01, R1024_5, 8*(R) + 6); \
+			R1024(00, 07, 02, 05, 04, 03, 06, 01, 12, 15, 14, 13, 08, 11, 10, 09, R1024_6, 8*(R) + 7); \
+			R1024(00, 15, 02, 11, 06, 13, 04, 09, 14, 01, 08, 05, 10, 03, 12, 07, R1024_7, 8*(R) + 8); \
+			I1024(2*(R)+1);
+
+			R1024_8_rounds(0);
 
 #define R1024_Unroll_R(NN) ((SKEIN_UNROLL_1024 == 0 && SKEIN1024_ROUNDS_TOTAL/8 > (NN)) || (SKEIN_UNROLL_1024 > (NN)))
 
-  #if   R1024_Unroll_R(1)
-        R1024_8_rounds(1);
-  #endif
-  #if   R1024_Unroll_R(2)
-        R1024_8_rounds(2);
-  #endif
-  #if   R1024_Unroll_R(3)
-        R1024_8_rounds(3);
-  #endif
-  #if   R1024_Unroll_R(4)
-        R1024_8_rounds(4);
-  #endif
-  #if   R1024_Unroll_R(5)
-        R1024_8_rounds(5);
-  #endif
-  #if   R1024_Unroll_R(6)
-        R1024_8_rounds(6);
-  #endif
-  #if   R1024_Unroll_R(7)
-        R1024_8_rounds(7);
-  #endif
-  #if   R1024_Unroll_R(8)
-        R1024_8_rounds(8);
-  #endif
-  #if   R1024_Unroll_R(9)
-        R1024_8_rounds(9);
-  #endif
-  #if   R1024_Unroll_R(10)
-        R1024_8_rounds(10);
-  #endif
-  #if   R1024_Unroll_R(11)
-        R1024_8_rounds(11);
-  #endif
-  #if   R1024_Unroll_R(12)
-        R1024_8_rounds(12);
-  #endif
-  #if   R1024_Unroll_R(13)
-        R1024_8_rounds(13);
-  #endif
-  #if   R1024_Unroll_R(14)
-        R1024_8_rounds(14);
-  #endif
-  #if  (SKEIN_UNROLL_1024 > 14)
+	#if   R1024_Unroll_R(1)
+			R1024_8_rounds(1);
+	#endif
+	#if   R1024_Unroll_R(2)
+			R1024_8_rounds(2);
+	#endif
+	#if   R1024_Unroll_R(3)
+			R1024_8_rounds(3);
+	#endif
+	#if   R1024_Unroll_R(4)
+			R1024_8_rounds(4);
+	#endif
+	#if   R1024_Unroll_R(5)
+			R1024_8_rounds(5);
+	#endif
+	#if   R1024_Unroll_R(6)
+			R1024_8_rounds(6);
+	#endif
+	#if   R1024_Unroll_R(7)
+			R1024_8_rounds(7);
+	#endif
+	#if   R1024_Unroll_R(8)
+			R1024_8_rounds(8);
+	#endif
+	#if   R1024_Unroll_R(9)
+			R1024_8_rounds(9);
+	#endif
+	#if   R1024_Unroll_R(10)
+			R1024_8_rounds(10);
+	#endif
+	#if   R1024_Unroll_R(11)
+			R1024_8_rounds(11);
+	#endif
+	#if   R1024_Unroll_R(12)
+			R1024_8_rounds(12);
+	#endif
+	#if   R1024_Unroll_R(13)
+			R1024_8_rounds(13);
+	#endif
+	#if   R1024_Unroll_R(14)
+			R1024_8_rounds(14);
+	#endif
+#if  (SKEIN_UNROLL_1024 > 14)
 #error  "need more unrolling in Skein_1024_Process_Block"
   #endif
-        }
-        /* do the final "feedforward" xor, update context chaining vars */
-
-        ctx->X[0] = X00 ^ w[0];
-        ctx->X[1] = X01 ^ w[1];
-        ctx->X[2] = X02 ^ w[2];
-        ctx->X[3] = X03 ^ w[3];
-        ctx->X[4] = X04 ^ w[4];
-        ctx->X[5] = X05 ^ w[5];
-        ctx->X[6] = X06 ^ w[6];
-        ctx->X[7] = X07 ^ w[7];
-        ctx->X[8] = X08 ^ w[8];
-        ctx->X[9] = X09 ^ w[9];
-        ctx->X[10] = X10 ^ w[10];
-        ctx->X[11] = X11 ^ w[11];
-        ctx->X[12] = X12 ^ w[12];
-        ctx->X[13] = X13 ^ w[13];
-        ctx->X[14] = X14 ^ w[14];
-        ctx->X[15] = X15 ^ w[15];
-
-        Skein_Show_Round(BLK_BITS, &ctx->h, SKEIN_RND_FEED_FWD, ctx->X);
-        
-        ts[1] &= ~SKEIN_T1_FLAG_FIRST;
-        blkPtr += SKEIN1024_BLOCK_BYTES;
-        }
-    while (--blkCnt);
-    ctx->h.T[0] = ts[0];
-    ctx->h.T[1] = ts[1];
-    }
+		}
+		/* do the final "feedforward" xor, update context chaining vars */
+
+		ctx->X[0] = X00 ^ w[0];
+		ctx->X[1] = X01 ^ w[1];
+		ctx->X[2] = X02 ^ w[2];
+		ctx->X[3] = X03 ^ w[3];
+		ctx->X[4] = X04 ^ w[4];
+		ctx->X[5] = X05 ^ w[5];
+		ctx->X[6] = X06 ^ w[6];
+		ctx->X[7] = X07 ^ w[7];
+		ctx->X[8] = X08 ^ w[8];
+		ctx->X[9] = X09 ^ w[9];
+		ctx->X[10] = X10 ^ w[10];
+		ctx->X[11] = X11 ^ w[11];
+		ctx->X[12] = X12 ^ w[12];
+		ctx->X[13] = X13 ^ w[13];
+		ctx->X[14] = X14 ^ w[14];
+		ctx->X[15] = X15 ^ w[15];
+
+		Skein_Show_Round(BLK_BITS, &ctx->h, SKEIN_RND_FEED_FWD, ctx->X);
+
+		ts[1] &= ~SKEIN_T1_FLAG_FIRST;
+		blkPtr += SKEIN1024_BLOCK_BYTES;
+	}
+	while (--blkCnt);
+	ctx->h.T[0] = ts[0];
+	ctx->h.T[1] = ts[1];
+}
 
 #if defined(SKEIN_CODE_SIZE) || defined(SKEIN_PERF)
 size_t Skein1024_Process_Block_CodeSize(void)
-    {
-    return ((u8 *) Skein1024_Process_Block_CodeSize) -
-           ((u8 *) Skein1024_Process_Block);
-    }
+{
+	return ((u8 *) Skein1024_Process_Block_CodeSize) -
+		((u8 *) Skein1024_Process_Block);
+}
 unsigned int Skein1024_Unroll_Cnt(void)
-    {
-    return SKEIN_UNROLL_1024;
-    }
+{
+	return SKEIN_UNROLL_1024;
+}
 #endif
 #endif
diff --git a/drivers/staging/skein/threefish1024Block.c b/drivers/staging/skein/threefish1024Block.c
index e3be37ea8024..1730a3120a0f 100644
--- a/drivers/staging/skein/threefish1024Block.c
+++ b/drivers/staging/skein/threefish1024Block.c
@@ -3,1382 +3,1380 @@
 
 
 void threefishEncrypt1024(struct threefish_key *keyCtx, u64 *input, u64 *output)
-        {
-
-    u64 b0 = input[0], b1 = input[1],
-      b2 = input[2], b3 = input[3],
-      b4 = input[4], b5 = input[5],
-      b6 = input[6], b7 = input[7],
-      b8 = input[8], b9 = input[9],
-      b10 = input[10], b11 = input[11],
-      b12 = input[12], b13 = input[13],
-      b14 = input[14], b15 = input[15];
-    u64 k0 = keyCtx->key[0], k1 = keyCtx->key[1],
-      k2 = keyCtx->key[2], k3 = keyCtx->key[3],
-      k4 = keyCtx->key[4], k5 = keyCtx->key[5],
-      k6 = keyCtx->key[6], k7 = keyCtx->key[7],
-      k8 = keyCtx->key[8], k9 = keyCtx->key[9],
-      k10 = keyCtx->key[10], k11 = keyCtx->key[11],
-      k12 = keyCtx->key[12], k13 = keyCtx->key[13],
-      k14 = keyCtx->key[14], k15 = keyCtx->key[15],
-      k16 = keyCtx->key[16];
-    u64 t0 = keyCtx->tweak[0], t1 = keyCtx->tweak[1],
-      t2 = keyCtx->tweak[2];
+{
+	u64 b0 = input[0], b1 = input[1],
+	  b2 = input[2], b3 = input[3],
+	  b4 = input[4], b5 = input[5],
+	  b6 = input[6], b7 = input[7],
+	  b8 = input[8], b9 = input[9],
+	  b10 = input[10], b11 = input[11],
+	  b12 = input[12], b13 = input[13],
+	  b14 = input[14], b15 = input[15];
+	u64 k0 = keyCtx->key[0], k1 = keyCtx->key[1],
+	  k2 = keyCtx->key[2], k3 = keyCtx->key[3],
+	  k4 = keyCtx->key[4], k5 = keyCtx->key[5],
+	  k6 = keyCtx->key[6], k7 = keyCtx->key[7],
+	  k8 = keyCtx->key[8], k9 = keyCtx->key[9],
+	  k10 = keyCtx->key[10], k11 = keyCtx->key[11],
+	  k12 = keyCtx->key[12], k13 = keyCtx->key[13],
+	  k14 = keyCtx->key[14], k15 = keyCtx->key[15],
+	  k16 = keyCtx->key[16];
+	u64 t0 = keyCtx->tweak[0], t1 = keyCtx->tweak[1],
+	  t2 = keyCtx->tweak[2];
 
-            b1 += k1; b0 += b1 + k0; b1 = ((b1 << 24) | (b1 >> (64 - 24))) ^ b0;
-            b3 += k3; b2 += b3 + k2; b3 = ((b3 << 13) | (b3 >> (64 - 13))) ^ b2;
-            b5 += k5; b4 += b5 + k4; b5 = ((b5 << 8) | (b5 >> (64 - 8))) ^ b4;
-            b7 += k7; b6 += b7 + k6; b7 = ((b7 << 47) | (b7 >> (64 - 47))) ^ b6;
-            b9 += k9; b8 += b9 + k8; b9 = ((b9 << 8) | (b9 >> (64 - 8))) ^ b8;
-            b11 += k11; b10 += b11 + k10; b11 = ((b11 << 17) | (b11 >> (64 - 17))) ^ b10;
-            b13 += k13 + t0; b12 += b13 + k12; b13 = ((b13 << 22) | (b13 >> (64 - 22))) ^ b12;
-            b15 += k15; b14 += b15 + k14 + t1; b15 = ((b15 << 37) | (b15 >> (64 - 37))) ^ b14;
-            b0 += b9; b9 = ((b9 << 38) | (b9 >> (64 - 38))) ^ b0;
-            b2 += b13; b13 = ((b13 << 19) | (b13 >> (64 - 19))) ^ b2;
-            b6 += b11; b11 = ((b11 << 10) | (b11 >> (64 - 10))) ^ b6;
-            b4 += b15; b15 = ((b15 << 55) | (b15 >> (64 - 55))) ^ b4;
-            b10 += b7; b7 = ((b7 << 49) | (b7 >> (64 - 49))) ^ b10;
-            b12 += b3; b3 = ((b3 << 18) | (b3 >> (64 - 18))) ^ b12;
-            b14 += b5; b5 = ((b5 << 23) | (b5 >> (64 - 23))) ^ b14;
-            b8 += b1; b1 = ((b1 << 52) | (b1 >> (64 - 52))) ^ b8;
-            b0 += b7; b7 = ((b7 << 33) | (b7 >> (64 - 33))) ^ b0;
-            b2 += b5; b5 = ((b5 << 4) | (b5 >> (64 - 4))) ^ b2;
-            b4 += b3; b3 = ((b3 << 51) | (b3 >> (64 - 51))) ^ b4;
-            b6 += b1; b1 = ((b1 << 13) | (b1 >> (64 - 13))) ^ b6;
-            b12 += b15; b15 = ((b15 << 34) | (b15 >> (64 - 34))) ^ b12;
-            b14 += b13; b13 = ((b13 << 41) | (b13 >> (64 - 41))) ^ b14;
-            b8 += b11; b11 = ((b11 << 59) | (b11 >> (64 - 59))) ^ b8;
-            b10 += b9; b9 = ((b9 << 17) | (b9 >> (64 - 17))) ^ b10;
-            b0 += b15; b15 = ((b15 << 5) | (b15 >> (64 - 5))) ^ b0;
-            b2 += b11; b11 = ((b11 << 20) | (b11 >> (64 - 20))) ^ b2;
-            b6 += b13; b13 = ((b13 << 48) | (b13 >> (64 - 48))) ^ b6;
-            b4 += b9; b9 = ((b9 << 41) | (b9 >> (64 - 41))) ^ b4;
-            b14 += b1; b1 = ((b1 << 47) | (b1 >> (64 - 47))) ^ b14;
-            b8 += b5; b5 = ((b5 << 28) | (b5 >> (64 - 28))) ^ b8;
-            b10 += b3; b3 = ((b3 << 16) | (b3 >> (64 - 16))) ^ b10;
-            b12 += b7; b7 = ((b7 << 25) | (b7 >> (64 - 25))) ^ b12;
-            b1 += k2; b0 += b1 + k1; b1 = ((b1 << 41) | (b1 >> (64 - 41))) ^ b0;
-            b3 += k4; b2 += b3 + k3; b3 = ((b3 << 9) | (b3 >> (64 - 9))) ^ b2;
-            b5 += k6; b4 += b5 + k5; b5 = ((b5 << 37) | (b5 >> (64 - 37))) ^ b4;
-            b7 += k8; b6 += b7 + k7; b7 = ((b7 << 31) | (b7 >> (64 - 31))) ^ b6;
-            b9 += k10; b8 += b9 + k9; b9 = ((b9 << 12) | (b9 >> (64 - 12))) ^ b8;
-            b11 += k12; b10 += b11 + k11; b11 = ((b11 << 47) | (b11 >> (64 - 47))) ^ b10;
-            b13 += k14 + t1; b12 += b13 + k13; b13 = ((b13 << 44) | (b13 >> (64 - 44))) ^ b12;
-            b15 += k16 + 1; b14 += b15 + k15 + t2; b15 = ((b15 << 30) | (b15 >> (64 - 30))) ^ b14;
-            b0 += b9; b9 = ((b9 << 16) | (b9 >> (64 - 16))) ^ b0;
-            b2 += b13; b13 = ((b13 << 34) | (b13 >> (64 - 34))) ^ b2;
-            b6 += b11; b11 = ((b11 << 56) | (b11 >> (64 - 56))) ^ b6;
-            b4 += b15; b15 = ((b15 << 51) | (b15 >> (64 - 51))) ^ b4;
-            b10 += b7; b7 = ((b7 << 4) | (b7 >> (64 - 4))) ^ b10;
-            b12 += b3; b3 = ((b3 << 53) | (b3 >> (64 - 53))) ^ b12;
-            b14 += b5; b5 = ((b5 << 42) | (b5 >> (64 - 42))) ^ b14;
-            b8 += b1; b1 = ((b1 << 41) | (b1 >> (64 - 41))) ^ b8;
-            b0 += b7; b7 = ((b7 << 31) | (b7 >> (64 - 31))) ^ b0;
-            b2 += b5; b5 = ((b5 << 44) | (b5 >> (64 - 44))) ^ b2;
-            b4 += b3; b3 = ((b3 << 47) | (b3 >> (64 - 47))) ^ b4;
-            b6 += b1; b1 = ((b1 << 46) | (b1 >> (64 - 46))) ^ b6;
-            b12 += b15; b15 = ((b15 << 19) | (b15 >> (64 - 19))) ^ b12;
-            b14 += b13; b13 = ((b13 << 42) | (b13 >> (64 - 42))) ^ b14;
-            b8 += b11; b11 = ((b11 << 44) | (b11 >> (64 - 44))) ^ b8;
-            b10 += b9; b9 = ((b9 << 25) | (b9 >> (64 - 25))) ^ b10;
-            b0 += b15; b15 = ((b15 << 9) | (b15 >> (64 - 9))) ^ b0;
-            b2 += b11; b11 = ((b11 << 48) | (b11 >> (64 - 48))) ^ b2;
-            b6 += b13; b13 = ((b13 << 35) | (b13 >> (64 - 35))) ^ b6;
-            b4 += b9; b9 = ((b9 << 52) | (b9 >> (64 - 52))) ^ b4;
-            b14 += b1; b1 = ((b1 << 23) | (b1 >> (64 - 23))) ^ b14;
-            b8 += b5; b5 = ((b5 << 31) | (b5 >> (64 - 31))) ^ b8;
-            b10 += b3; b3 = ((b3 << 37) | (b3 >> (64 - 37))) ^ b10;
-            b12 += b7; b7 = ((b7 << 20) | (b7 >> (64 - 20))) ^ b12;
-            b1 += k3; b0 += b1 + k2; b1 = ((b1 << 24) | (b1 >> (64 - 24))) ^ b0;
-            b3 += k5; b2 += b3 + k4; b3 = ((b3 << 13) | (b3 >> (64 - 13))) ^ b2;
-            b5 += k7; b4 += b5 + k6; b5 = ((b5 << 8) | (b5 >> (64 - 8))) ^ b4;
-            b7 += k9; b6 += b7 + k8; b7 = ((b7 << 47) | (b7 >> (64 - 47))) ^ b6;
-            b9 += k11; b8 += b9 + k10; b9 = ((b9 << 8) | (b9 >> (64 - 8))) ^ b8;
-            b11 += k13; b10 += b11 + k12; b11 = ((b11 << 17) | (b11 >> (64 - 17))) ^ b10;
-            b13 += k15 + t2; b12 += b13 + k14; b13 = ((b13 << 22) | (b13 >> (64 - 22))) ^ b12;
-            b15 += k0 + 2; b14 += b15 + k16 + t0; b15 = ((b15 << 37) | (b15 >> (64 - 37))) ^ b14;
-            b0 += b9; b9 = ((b9 << 38) | (b9 >> (64 - 38))) ^ b0;
-            b2 += b13; b13 = ((b13 << 19) | (b13 >> (64 - 19))) ^ b2;
-            b6 += b11; b11 = ((b11 << 10) | (b11 >> (64 - 10))) ^ b6;
-            b4 += b15; b15 = ((b15 << 55) | (b15 >> (64 - 55))) ^ b4;
-            b10 += b7; b7 = ((b7 << 49) | (b7 >> (64 - 49))) ^ b10;
-            b12 += b3; b3 = ((b3 << 18) | (b3 >> (64 - 18))) ^ b12;
-            b14 += b5; b5 = ((b5 << 23) | (b5 >> (64 - 23))) ^ b14;
-            b8 += b1; b1 = ((b1 << 52) | (b1 >> (64 - 52))) ^ b8;
-            b0 += b7; b7 = ((b7 << 33) | (b7 >> (64 - 33))) ^ b0;
-            b2 += b5; b5 = ((b5 << 4) | (b5 >> (64 - 4))) ^ b2;
-            b4 += b3; b3 = ((b3 << 51) | (b3 >> (64 - 51))) ^ b4;
-            b6 += b1; b1 = ((b1 << 13) | (b1 >> (64 - 13))) ^ b6;
-            b12 += b15; b15 = ((b15 << 34) | (b15 >> (64 - 34))) ^ b12;
-            b14 += b13; b13 = ((b13 << 41) | (b13 >> (64 - 41))) ^ b14;
-            b8 += b11; b11 = ((b11 << 59) | (b11 >> (64 - 59))) ^ b8;
-            b10 += b9; b9 = ((b9 << 17) | (b9 >> (64 - 17))) ^ b10;
-            b0 += b15; b15 = ((b15 << 5) | (b15 >> (64 - 5))) ^ b0;
-            b2 += b11; b11 = ((b11 << 20) | (b11 >> (64 - 20))) ^ b2;
-            b6 += b13; b13 = ((b13 << 48) | (b13 >> (64 - 48))) ^ b6;
-            b4 += b9; b9 = ((b9 << 41) | (b9 >> (64 - 41))) ^ b4;
-            b14 += b1; b1 = ((b1 << 47) | (b1 >> (64 - 47))) ^ b14;
-            b8 += b5; b5 = ((b5 << 28) | (b5 >> (64 - 28))) ^ b8;
-            b10 += b3; b3 = ((b3 << 16) | (b3 >> (64 - 16))) ^ b10;
-            b12 += b7; b7 = ((b7 << 25) | (b7 >> (64 - 25))) ^ b12;
-            b1 += k4; b0 += b1 + k3; b1 = ((b1 << 41) | (b1 >> (64 - 41))) ^ b0;
-            b3 += k6; b2 += b3 + k5; b3 = ((b3 << 9) | (b3 >> (64 - 9))) ^ b2;
-            b5 += k8; b4 += b5 + k7; b5 = ((b5 << 37) | (b5 >> (64 - 37))) ^ b4;
-            b7 += k10; b6 += b7 + k9; b7 = ((b7 << 31) | (b7 >> (64 - 31))) ^ b6;
-            b9 += k12; b8 += b9 + k11; b9 = ((b9 << 12) | (b9 >> (64 - 12))) ^ b8;
-            b11 += k14; b10 += b11 + k13; b11 = ((b11 << 47) | (b11 >> (64 - 47))) ^ b10;
-            b13 += k16 + t0; b12 += b13 + k15; b13 = ((b13 << 44) | (b13 >> (64 - 44))) ^ b12;
-            b15 += k1 + 3; b14 += b15 + k0 + t1; b15 = ((b15 << 30) | (b15 >> (64 - 30))) ^ b14;
-            b0 += b9; b9 = ((b9 << 16) | (b9 >> (64 - 16))) ^ b0;
-            b2 += b13; b13 = ((b13 << 34) | (b13 >> (64 - 34))) ^ b2;
-            b6 += b11; b11 = ((b11 << 56) | (b11 >> (64 - 56))) ^ b6;
-            b4 += b15; b15 = ((b15 << 51) | (b15 >> (64 - 51))) ^ b4;
-            b10 += b7; b7 = ((b7 << 4) | (b7 >> (64 - 4))) ^ b10;
-            b12 += b3; b3 = ((b3 << 53) | (b3 >> (64 - 53))) ^ b12;
-            b14 += b5; b5 = ((b5 << 42) | (b5 >> (64 - 42))) ^ b14;
-            b8 += b1; b1 = ((b1 << 41) | (b1 >> (64 - 41))) ^ b8;
-            b0 += b7; b7 = ((b7 << 31) | (b7 >> (64 - 31))) ^ b0;
-            b2 += b5; b5 = ((b5 << 44) | (b5 >> (64 - 44))) ^ b2;
-            b4 += b3; b3 = ((b3 << 47) | (b3 >> (64 - 47))) ^ b4;
-            b6 += b1; b1 = ((b1 << 46) | (b1 >> (64 - 46))) ^ b6;
-            b12 += b15; b15 = ((b15 << 19) | (b15 >> (64 - 19))) ^ b12;
-            b14 += b13; b13 = ((b13 << 42) | (b13 >> (64 - 42))) ^ b14;
-            b8 += b11; b11 = ((b11 << 44) | (b11 >> (64 - 44))) ^ b8;
-            b10 += b9; b9 = ((b9 << 25) | (b9 >> (64 - 25))) ^ b10;
-            b0 += b15; b15 = ((b15 << 9) | (b15 >> (64 - 9))) ^ b0;
-            b2 += b11; b11 = ((b11 << 48) | (b11 >> (64 - 48))) ^ b2;
-            b6 += b13; b13 = ((b13 << 35) | (b13 >> (64 - 35))) ^ b6;
-            b4 += b9; b9 = ((b9 << 52) | (b9 >> (64 - 52))) ^ b4;
-            b14 += b1; b1 = ((b1 << 23) | (b1 >> (64 - 23))) ^ b14;
-            b8 += b5; b5 = ((b5 << 31) | (b5 >> (64 - 31))) ^ b8;
-            b10 += b3; b3 = ((b3 << 37) | (b3 >> (64 - 37))) ^ b10;
-            b12 += b7; b7 = ((b7 << 20) | (b7 >> (64 - 20))) ^ b12;
-            b1 += k5; b0 += b1 + k4; b1 = ((b1 << 24) | (b1 >> (64 - 24))) ^ b0;
-            b3 += k7; b2 += b3 + k6; b3 = ((b3 << 13) | (b3 >> (64 - 13))) ^ b2;
-            b5 += k9; b4 += b5 + k8; b5 = ((b5 << 8) | (b5 >> (64 - 8))) ^ b4;
-            b7 += k11; b6 += b7 + k10; b7 = ((b7 << 47) | (b7 >> (64 - 47))) ^ b6;
-            b9 += k13; b8 += b9 + k12; b9 = ((b9 << 8) | (b9 >> (64 - 8))) ^ b8;
-            b11 += k15; b10 += b11 + k14; b11 = ((b11 << 17) | (b11 >> (64 - 17))) ^ b10;
-            b13 += k0 + t1; b12 += b13 + k16; b13 = ((b13 << 22) | (b13 >> (64 - 22))) ^ b12;
-            b15 += k2 + 4; b14 += b15 + k1 + t2; b15 = ((b15 << 37) | (b15 >> (64 - 37))) ^ b14;
-            b0 += b9; b9 = ((b9 << 38) | (b9 >> (64 - 38))) ^ b0;
-            b2 += b13; b13 = ((b13 << 19) | (b13 >> (64 - 19))) ^ b2;
-            b6 += b11; b11 = ((b11 << 10) | (b11 >> (64 - 10))) ^ b6;
-            b4 += b15; b15 = ((b15 << 55) | (b15 >> (64 - 55))) ^ b4;
-            b10 += b7; b7 = ((b7 << 49) | (b7 >> (64 - 49))) ^ b10;
-            b12 += b3; b3 = ((b3 << 18) | (b3 >> (64 - 18))) ^ b12;
-            b14 += b5; b5 = ((b5 << 23) | (b5 >> (64 - 23))) ^ b14;
-            b8 += b1; b1 = ((b1 << 52) | (b1 >> (64 - 52))) ^ b8;
-            b0 += b7; b7 = ((b7 << 33) | (b7 >> (64 - 33))) ^ b0;
-            b2 += b5; b5 = ((b5 << 4) | (b5 >> (64 - 4))) ^ b2;
-            b4 += b3; b3 = ((b3 << 51) | (b3 >> (64 - 51))) ^ b4;
-            b6 += b1; b1 = ((b1 << 13) | (b1 >> (64 - 13))) ^ b6;
-            b12 += b15; b15 = ((b15 << 34) | (b15 >> (64 - 34))) ^ b12;
-            b14 += b13; b13 = ((b13 << 41) | (b13 >> (64 - 41))) ^ b14;
-            b8 += b11; b11 = ((b11 << 59) | (b11 >> (64 - 59))) ^ b8;
-            b10 += b9; b9 = ((b9 << 17) | (b9 >> (64 - 17))) ^ b10;
-            b0 += b15; b15 = ((b15 << 5) | (b15 >> (64 - 5))) ^ b0;
-            b2 += b11; b11 = ((b11 << 20) | (b11 >> (64 - 20))) ^ b2;
-            b6 += b13; b13 = ((b13 << 48) | (b13 >> (64 - 48))) ^ b6;
-            b4 += b9; b9 = ((b9 << 41) | (b9 >> (64 - 41))) ^ b4;
-            b14 += b1; b1 = ((b1 << 47) | (b1 >> (64 - 47))) ^ b14;
-            b8 += b5; b5 = ((b5 << 28) | (b5 >> (64 - 28))) ^ b8;
-            b10 += b3; b3 = ((b3 << 16) | (b3 >> (64 - 16))) ^ b10;
-            b12 += b7; b7 = ((b7 << 25) | (b7 >> (64 - 25))) ^ b12;
-            b1 += k6; b0 += b1 + k5; b1 = ((b1 << 41) | (b1 >> (64 - 41))) ^ b0;
-            b3 += k8; b2 += b3 + k7; b3 = ((b3 << 9) | (b3 >> (64 - 9))) ^ b2;
-            b5 += k10; b4 += b5 + k9; b5 = ((b5 << 37) | (b5 >> (64 - 37))) ^ b4;
-            b7 += k12; b6 += b7 + k11; b7 = ((b7 << 31) | (b7 >> (64 - 31))) ^ b6;
-            b9 += k14; b8 += b9 + k13; b9 = ((b9 << 12) | (b9 >> (64 - 12))) ^ b8;
-            b11 += k16; b10 += b11 + k15; b11 = ((b11 << 47) | (b11 >> (64 - 47))) ^ b10;
-            b13 += k1 + t2; b12 += b13 + k0; b13 = ((b13 << 44) | (b13 >> (64 - 44))) ^ b12;
-            b15 += k3 + 5; b14 += b15 + k2 + t0; b15 = ((b15 << 30) | (b15 >> (64 - 30))) ^ b14;
-            b0 += b9; b9 = ((b9 << 16) | (b9 >> (64 - 16))) ^ b0;
-            b2 += b13; b13 = ((b13 << 34) | (b13 >> (64 - 34))) ^ b2;
-            b6 += b11; b11 = ((b11 << 56) | (b11 >> (64 - 56))) ^ b6;
-            b4 += b15; b15 = ((b15 << 51) | (b15 >> (64 - 51))) ^ b4;
-            b10 += b7; b7 = ((b7 << 4) | (b7 >> (64 - 4))) ^ b10;
-            b12 += b3; b3 = ((b3 << 53) | (b3 >> (64 - 53))) ^ b12;
-            b14 += b5; b5 = ((b5 << 42) | (b5 >> (64 - 42))) ^ b14;
-            b8 += b1; b1 = ((b1 << 41) | (b1 >> (64 - 41))) ^ b8;
-            b0 += b7; b7 = ((b7 << 31) | (b7 >> (64 - 31))) ^ b0;
-            b2 += b5; b5 = ((b5 << 44) | (b5 >> (64 - 44))) ^ b2;
-            b4 += b3; b3 = ((b3 << 47) | (b3 >> (64 - 47))) ^ b4;
-            b6 += b1; b1 = ((b1 << 46) | (b1 >> (64 - 46))) ^ b6;
-            b12 += b15; b15 = ((b15 << 19) | (b15 >> (64 - 19))) ^ b12;
-            b14 += b13; b13 = ((b13 << 42) | (b13 >> (64 - 42))) ^ b14;
-            b8 += b11; b11 = ((b11 << 44) | (b11 >> (64 - 44))) ^ b8;
-            b10 += b9; b9 = ((b9 << 25) | (b9 >> (64 - 25))) ^ b10;
-            b0 += b15; b15 = ((b15 << 9) | (b15 >> (64 - 9))) ^ b0;
-            b2 += b11; b11 = ((b11 << 48) | (b11 >> (64 - 48))) ^ b2;
-            b6 += b13; b13 = ((b13 << 35) | (b13 >> (64 - 35))) ^ b6;
-            b4 += b9; b9 = ((b9 << 52) | (b9 >> (64 - 52))) ^ b4;
-            b14 += b1; b1 = ((b1 << 23) | (b1 >> (64 - 23))) ^ b14;
-            b8 += b5; b5 = ((b5 << 31) | (b5 >> (64 - 31))) ^ b8;
-            b10 += b3; b3 = ((b3 << 37) | (b3 >> (64 - 37))) ^ b10;
-            b12 += b7; b7 = ((b7 << 20) | (b7 >> (64 - 20))) ^ b12;
-            b1 += k7; b0 += b1 + k6; b1 = ((b1 << 24) | (b1 >> (64 - 24))) ^ b0;
-            b3 += k9; b2 += b3 + k8; b3 = ((b3 << 13) | (b3 >> (64 - 13))) ^ b2;
-            b5 += k11; b4 += b5 + k10; b5 = ((b5 << 8) | (b5 >> (64 - 8))) ^ b4;
-            b7 += k13; b6 += b7 + k12; b7 = ((b7 << 47) | (b7 >> (64 - 47))) ^ b6;
-            b9 += k15; b8 += b9 + k14; b9 = ((b9 << 8) | (b9 >> (64 - 8))) ^ b8;
-            b11 += k0; b10 += b11 + k16; b11 = ((b11 << 17) | (b11 >> (64 - 17))) ^ b10;
-            b13 += k2 + t0; b12 += b13 + k1; b13 = ((b13 << 22) | (b13 >> (64 - 22))) ^ b12;
-            b15 += k4 + 6; b14 += b15 + k3 + t1; b15 = ((b15 << 37) | (b15 >> (64 - 37))) ^ b14;
-            b0 += b9; b9 = ((b9 << 38) | (b9 >> (64 - 38))) ^ b0;
-            b2 += b13; b13 = ((b13 << 19) | (b13 >> (64 - 19))) ^ b2;
-            b6 += b11; b11 = ((b11 << 10) | (b11 >> (64 - 10))) ^ b6;
-            b4 += b15; b15 = ((b15 << 55) | (b15 >> (64 - 55))) ^ b4;
-            b10 += b7; b7 = ((b7 << 49) | (b7 >> (64 - 49))) ^ b10;
-            b12 += b3; b3 = ((b3 << 18) | (b3 >> (64 - 18))) ^ b12;
-            b14 += b5; b5 = ((b5 << 23) | (b5 >> (64 - 23))) ^ b14;
-            b8 += b1; b1 = ((b1 << 52) | (b1 >> (64 - 52))) ^ b8;
-            b0 += b7; b7 = ((b7 << 33) | (b7 >> (64 - 33))) ^ b0;
-            b2 += b5; b5 = ((b5 << 4) | (b5 >> (64 - 4))) ^ b2;
-            b4 += b3; b3 = ((b3 << 51) | (b3 >> (64 - 51))) ^ b4;
-            b6 += b1; b1 = ((b1 << 13) | (b1 >> (64 - 13))) ^ b6;
-            b12 += b15; b15 = ((b15 << 34) | (b15 >> (64 - 34))) ^ b12;
-            b14 += b13; b13 = ((b13 << 41) | (b13 >> (64 - 41))) ^ b14;
-            b8 += b11; b11 = ((b11 << 59) | (b11 >> (64 - 59))) ^ b8;
-            b10 += b9; b9 = ((b9 << 17) | (b9 >> (64 - 17))) ^ b10;
-            b0 += b15; b15 = ((b15 << 5) | (b15 >> (64 - 5))) ^ b0;
-            b2 += b11; b11 = ((b11 << 20) | (b11 >> (64 - 20))) ^ b2;
-            b6 += b13; b13 = ((b13 << 48) | (b13 >> (64 - 48))) ^ b6;
-            b4 += b9; b9 = ((b9 << 41) | (b9 >> (64 - 41))) ^ b4;
-            b14 += b1; b1 = ((b1 << 47) | (b1 >> (64 - 47))) ^ b14;
-            b8 += b5; b5 = ((b5 << 28) | (b5 >> (64 - 28))) ^ b8;
-            b10 += b3; b3 = ((b3 << 16) | (b3 >> (64 - 16))) ^ b10;
-            b12 += b7; b7 = ((b7 << 25) | (b7 >> (64 - 25))) ^ b12;
-            b1 += k8; b0 += b1 + k7; b1 = ((b1 << 41) | (b1 >> (64 - 41))) ^ b0;
-            b3 += k10; b2 += b3 + k9; b3 = ((b3 << 9) | (b3 >> (64 - 9))) ^ b2;
-            b5 += k12; b4 += b5 + k11; b5 = ((b5 << 37) | (b5 >> (64 - 37))) ^ b4;
-            b7 += k14; b6 += b7 + k13; b7 = ((b7 << 31) | (b7 >> (64 - 31))) ^ b6;
-            b9 += k16; b8 += b9 + k15; b9 = ((b9 << 12) | (b9 >> (64 - 12))) ^ b8;
-            b11 += k1; b10 += b11 + k0; b11 = ((b11 << 47) | (b11 >> (64 - 47))) ^ b10;
-            b13 += k3 + t1; b12 += b13 + k2; b13 = ((b13 << 44) | (b13 >> (64 - 44))) ^ b12;
-            b15 += k5 + 7; b14 += b15 + k4 + t2; b15 = ((b15 << 30) | (b15 >> (64 - 30))) ^ b14;
-            b0 += b9; b9 = ((b9 << 16) | (b9 >> (64 - 16))) ^ b0;
-            b2 += b13; b13 = ((b13 << 34) | (b13 >> (64 - 34))) ^ b2;
-            b6 += b11; b11 = ((b11 << 56) | (b11 >> (64 - 56))) ^ b6;
-            b4 += b15; b15 = ((b15 << 51) | (b15 >> (64 - 51))) ^ b4;
-            b10 += b7; b7 = ((b7 << 4) | (b7 >> (64 - 4))) ^ b10;
-            b12 += b3; b3 = ((b3 << 53) | (b3 >> (64 - 53))) ^ b12;
-            b14 += b5; b5 = ((b5 << 42) | (b5 >> (64 - 42))) ^ b14;
-            b8 += b1; b1 = ((b1 << 41) | (b1 >> (64 - 41))) ^ b8;
-            b0 += b7; b7 = ((b7 << 31) | (b7 >> (64 - 31))) ^ b0;
-            b2 += b5; b5 = ((b5 << 44) | (b5 >> (64 - 44))) ^ b2;
-            b4 += b3; b3 = ((b3 << 47) | (b3 >> (64 - 47))) ^ b4;
-            b6 += b1; b1 = ((b1 << 46) | (b1 >> (64 - 46))) ^ b6;
-            b12 += b15; b15 = ((b15 << 19) | (b15 >> (64 - 19))) ^ b12;
-            b14 += b13; b13 = ((b13 << 42) | (b13 >> (64 - 42))) ^ b14;
-            b8 += b11; b11 = ((b11 << 44) | (b11 >> (64 - 44))) ^ b8;
-            b10 += b9; b9 = ((b9 << 25) | (b9 >> (64 - 25))) ^ b10;
-            b0 += b15; b15 = ((b15 << 9) | (b15 >> (64 - 9))) ^ b0;
-            b2 += b11; b11 = ((b11 << 48) | (b11 >> (64 - 48))) ^ b2;
-            b6 += b13; b13 = ((b13 << 35) | (b13 >> (64 - 35))) ^ b6;
-            b4 += b9; b9 = ((b9 << 52) | (b9 >> (64 - 52))) ^ b4;
-            b14 += b1; b1 = ((b1 << 23) | (b1 >> (64 - 23))) ^ b14;
-            b8 += b5; b5 = ((b5 << 31) | (b5 >> (64 - 31))) ^ b8;
-            b10 += b3; b3 = ((b3 << 37) | (b3 >> (64 - 37))) ^ b10;
-            b12 += b7; b7 = ((b7 << 20) | (b7 >> (64 - 20))) ^ b12;
-            b1 += k9; b0 += b1 + k8; b1 = ((b1 << 24) | (b1 >> (64 - 24))) ^ b0;
-            b3 += k11; b2 += b3 + k10; b3 = ((b3 << 13) | (b3 >> (64 - 13))) ^ b2;
-            b5 += k13; b4 += b5 + k12; b5 = ((b5 << 8) | (b5 >> (64 - 8))) ^ b4;
-            b7 += k15; b6 += b7 + k14; b7 = ((b7 << 47) | (b7 >> (64 - 47))) ^ b6;
-            b9 += k0; b8 += b9 + k16; b9 = ((b9 << 8) | (b9 >> (64 - 8))) ^ b8;
-            b11 += k2; b10 += b11 + k1; b11 = ((b11 << 17) | (b11 >> (64 - 17))) ^ b10;
-            b13 += k4 + t2; b12 += b13 + k3; b13 = ((b13 << 22) | (b13 >> (64 - 22))) ^ b12;
-            b15 += k6 + 8; b14 += b15 + k5 + t0; b15 = ((b15 << 37) | (b15 >> (64 - 37))) ^ b14;
-            b0 += b9; b9 = ((b9 << 38) | (b9 >> (64 - 38))) ^ b0;
-            b2 += b13; b13 = ((b13 << 19) | (b13 >> (64 - 19))) ^ b2;
-            b6 += b11; b11 = ((b11 << 10) | (b11 >> (64 - 10))) ^ b6;
-            b4 += b15; b15 = ((b15 << 55) | (b15 >> (64 - 55))) ^ b4;
-            b10 += b7; b7 = ((b7 << 49) | (b7 >> (64 - 49))) ^ b10;
-            b12 += b3; b3 = ((b3 << 18) | (b3 >> (64 - 18))) ^ b12;
-            b14 += b5; b5 = ((b5 << 23) | (b5 >> (64 - 23))) ^ b14;
-            b8 += b1; b1 = ((b1 << 52) | (b1 >> (64 - 52))) ^ b8;
-            b0 += b7; b7 = ((b7 << 33) | (b7 >> (64 - 33))) ^ b0;
-            b2 += b5; b5 = ((b5 << 4) | (b5 >> (64 - 4))) ^ b2;
-            b4 += b3; b3 = ((b3 << 51) | (b3 >> (64 - 51))) ^ b4;
-            b6 += b1; b1 = ((b1 << 13) | (b1 >> (64 - 13))) ^ b6;
-            b12 += b15; b15 = ((b15 << 34) | (b15 >> (64 - 34))) ^ b12;
-            b14 += b13; b13 = ((b13 << 41) | (b13 >> (64 - 41))) ^ b14;
-            b8 += b11; b11 = ((b11 << 59) | (b11 >> (64 - 59))) ^ b8;
-            b10 += b9; b9 = ((b9 << 17) | (b9 >> (64 - 17))) ^ b10;
-            b0 += b15; b15 = ((b15 << 5) | (b15 >> (64 - 5))) ^ b0;
-            b2 += b11; b11 = ((b11 << 20) | (b11 >> (64 - 20))) ^ b2;
-            b6 += b13; b13 = ((b13 << 48) | (b13 >> (64 - 48))) ^ b6;
-            b4 += b9; b9 = ((b9 << 41) | (b9 >> (64 - 41))) ^ b4;
-            b14 += b1; b1 = ((b1 << 47) | (b1 >> (64 - 47))) ^ b14;
-            b8 += b5; b5 = ((b5 << 28) | (b5 >> (64 - 28))) ^ b8;
-            b10 += b3; b3 = ((b3 << 16) | (b3 >> (64 - 16))) ^ b10;
-            b12 += b7; b7 = ((b7 << 25) | (b7 >> (64 - 25))) ^ b12;
-            b1 += k10; b0 += b1 + k9; b1 = ((b1 << 41) | (b1 >> (64 - 41))) ^ b0;
-            b3 += k12; b2 += b3 + k11; b3 = ((b3 << 9) | (b3 >> (64 - 9))) ^ b2;
-            b5 += k14; b4 += b5 + k13; b5 = ((b5 << 37) | (b5 >> (64 - 37))) ^ b4;
-            b7 += k16; b6 += b7 + k15; b7 = ((b7 << 31) | (b7 >> (64 - 31))) ^ b6;
-            b9 += k1; b8 += b9 + k0; b9 = ((b9 << 12) | (b9 >> (64 - 12))) ^ b8;
-            b11 += k3; b10 += b11 + k2; b11 = ((b11 << 47) | (b11 >> (64 - 47))) ^ b10;
-            b13 += k5 + t0; b12 += b13 + k4; b13 = ((b13 << 44) | (b13 >> (64 - 44))) ^ b12;
-            b15 += k7 + 9; b14 += b15 + k6 + t1; b15 = ((b15 << 30) | (b15 >> (64 - 30))) ^ b14;
-            b0 += b9; b9 = ((b9 << 16) | (b9 >> (64 - 16))) ^ b0;
-            b2 += b13; b13 = ((b13 << 34) | (b13 >> (64 - 34))) ^ b2;
-            b6 += b11; b11 = ((b11 << 56) | (b11 >> (64 - 56))) ^ b6;
-            b4 += b15; b15 = ((b15 << 51) | (b15 >> (64 - 51))) ^ b4;
-            b10 += b7; b7 = ((b7 << 4) | (b7 >> (64 - 4))) ^ b10;
-            b12 += b3; b3 = ((b3 << 53) | (b3 >> (64 - 53))) ^ b12;
-            b14 += b5; b5 = ((b5 << 42) | (b5 >> (64 - 42))) ^ b14;
-            b8 += b1; b1 = ((b1 << 41) | (b1 >> (64 - 41))) ^ b8;
-            b0 += b7; b7 = ((b7 << 31) | (b7 >> (64 - 31))) ^ b0;
-            b2 += b5; b5 = ((b5 << 44) | (b5 >> (64 - 44))) ^ b2;
-            b4 += b3; b3 = ((b3 << 47) | (b3 >> (64 - 47))) ^ b4;
-            b6 += b1; b1 = ((b1 << 46) | (b1 >> (64 - 46))) ^ b6;
-            b12 += b15; b15 = ((b15 << 19) | (b15 >> (64 - 19))) ^ b12;
-            b14 += b13; b13 = ((b13 << 42) | (b13 >> (64 - 42))) ^ b14;
-            b8 += b11; b11 = ((b11 << 44) | (b11 >> (64 - 44))) ^ b8;
-            b10 += b9; b9 = ((b9 << 25) | (b9 >> (64 - 25))) ^ b10;
-            b0 += b15; b15 = ((b15 << 9) | (b15 >> (64 - 9))) ^ b0;
-            b2 += b11; b11 = ((b11 << 48) | (b11 >> (64 - 48))) ^ b2;
-            b6 += b13; b13 = ((b13 << 35) | (b13 >> (64 - 35))) ^ b6;
-            b4 += b9; b9 = ((b9 << 52) | (b9 >> (64 - 52))) ^ b4;
-            b14 += b1; b1 = ((b1 << 23) | (b1 >> (64 - 23))) ^ b14;
-            b8 += b5; b5 = ((b5 << 31) | (b5 >> (64 - 31))) ^ b8;
-            b10 += b3; b3 = ((b3 << 37) | (b3 >> (64 - 37))) ^ b10;
-            b12 += b7; b7 = ((b7 << 20) | (b7 >> (64 - 20))) ^ b12;
-            b1 += k11; b0 += b1 + k10; b1 = ((b1 << 24) | (b1 >> (64 - 24))) ^ b0;
-            b3 += k13; b2 += b3 + k12; b3 = ((b3 << 13) | (b3 >> (64 - 13))) ^ b2;
-            b5 += k15; b4 += b5 + k14; b5 = ((b5 << 8) | (b5 >> (64 - 8))) ^ b4;
-            b7 += k0; b6 += b7 + k16; b7 = ((b7 << 47) | (b7 >> (64 - 47))) ^ b6;
-            b9 += k2; b8 += b9 + k1; b9 = ((b9 << 8) | (b9 >> (64 - 8))) ^ b8;
-            b11 += k4; b10 += b11 + k3; b11 = ((b11 << 17) | (b11 >> (64 - 17))) ^ b10;
-            b13 += k6 + t1; b12 += b13 + k5; b13 = ((b13 << 22) | (b13 >> (64 - 22))) ^ b12;
-            b15 += k8 + 10; b14 += b15 + k7 + t2; b15 = ((b15 << 37) | (b15 >> (64 - 37))) ^ b14;
-            b0 += b9; b9 = ((b9 << 38) | (b9 >> (64 - 38))) ^ b0;
-            b2 += b13; b13 = ((b13 << 19) | (b13 >> (64 - 19))) ^ b2;
-            b6 += b11; b11 = ((b11 << 10) | (b11 >> (64 - 10))) ^ b6;
-            b4 += b15; b15 = ((b15 << 55) | (b15 >> (64 - 55))) ^ b4;
-            b10 += b7; b7 = ((b7 << 49) | (b7 >> (64 - 49))) ^ b10;
-            b12 += b3; b3 = ((b3 << 18) | (b3 >> (64 - 18))) ^ b12;
-            b14 += b5; b5 = ((b5 << 23) | (b5 >> (64 - 23))) ^ b14;
-            b8 += b1; b1 = ((b1 << 52) | (b1 >> (64 - 52))) ^ b8;
-            b0 += b7; b7 = ((b7 << 33) | (b7 >> (64 - 33))) ^ b0;
-            b2 += b5; b5 = ((b5 << 4) | (b5 >> (64 - 4))) ^ b2;
-            b4 += b3; b3 = ((b3 << 51) | (b3 >> (64 - 51))) ^ b4;
-            b6 += b1; b1 = ((b1 << 13) | (b1 >> (64 - 13))) ^ b6;
-            b12 += b15; b15 = ((b15 << 34) | (b15 >> (64 - 34))) ^ b12;
-            b14 += b13; b13 = ((b13 << 41) | (b13 >> (64 - 41))) ^ b14;
-            b8 += b11; b11 = ((b11 << 59) | (b11 >> (64 - 59))) ^ b8;
-            b10 += b9; b9 = ((b9 << 17) | (b9 >> (64 - 17))) ^ b10;
-            b0 += b15; b15 = ((b15 << 5) | (b15 >> (64 - 5))) ^ b0;
-            b2 += b11; b11 = ((b11 << 20) | (b11 >> (64 - 20))) ^ b2;
-            b6 += b13; b13 = ((b13 << 48) | (b13 >> (64 - 48))) ^ b6;
-            b4 += b9; b9 = ((b9 << 41) | (b9 >> (64 - 41))) ^ b4;
-            b14 += b1; b1 = ((b1 << 47) | (b1 >> (64 - 47))) ^ b14;
-            b8 += b5; b5 = ((b5 << 28) | (b5 >> (64 - 28))) ^ b8;
-            b10 += b3; b3 = ((b3 << 16) | (b3 >> (64 - 16))) ^ b10;
-            b12 += b7; b7 = ((b7 << 25) | (b7 >> (64 - 25))) ^ b12;
-            b1 += k12; b0 += b1 + k11; b1 = ((b1 << 41) | (b1 >> (64 - 41))) ^ b0;
-            b3 += k14; b2 += b3 + k13; b3 = ((b3 << 9) | (b3 >> (64 - 9))) ^ b2;
-            b5 += k16; b4 += b5 + k15; b5 = ((b5 << 37) | (b5 >> (64 - 37))) ^ b4;
-            b7 += k1; b6 += b7 + k0; b7 = ((b7 << 31) | (b7 >> (64 - 31))) ^ b6;
-            b9 += k3; b8 += b9 + k2; b9 = ((b9 << 12) | (b9 >> (64 - 12))) ^ b8;
-            b11 += k5; b10 += b11 + k4; b11 = ((b11 << 47) | (b11 >> (64 - 47))) ^ b10;
-            b13 += k7 + t2; b12 += b13 + k6; b13 = ((b13 << 44) | (b13 >> (64 - 44))) ^ b12;
-            b15 += k9 + 11; b14 += b15 + k8 + t0; b15 = ((b15 << 30) | (b15 >> (64 - 30))) ^ b14;
-            b0 += b9; b9 = ((b9 << 16) | (b9 >> (64 - 16))) ^ b0;
-            b2 += b13; b13 = ((b13 << 34) | (b13 >> (64 - 34))) ^ b2;
-            b6 += b11; b11 = ((b11 << 56) | (b11 >> (64 - 56))) ^ b6;
-            b4 += b15; b15 = ((b15 << 51) | (b15 >> (64 - 51))) ^ b4;
-            b10 += b7; b7 = ((b7 << 4) | (b7 >> (64 - 4))) ^ b10;
-            b12 += b3; b3 = ((b3 << 53) | (b3 >> (64 - 53))) ^ b12;
-            b14 += b5; b5 = ((b5 << 42) | (b5 >> (64 - 42))) ^ b14;
-            b8 += b1; b1 = ((b1 << 41) | (b1 >> (64 - 41))) ^ b8;
-            b0 += b7; b7 = ((b7 << 31) | (b7 >> (64 - 31))) ^ b0;
-            b2 += b5; b5 = ((b5 << 44) | (b5 >> (64 - 44))) ^ b2;
-            b4 += b3; b3 = ((b3 << 47) | (b3 >> (64 - 47))) ^ b4;
-            b6 += b1; b1 = ((b1 << 46) | (b1 >> (64 - 46))) ^ b6;
-            b12 += b15; b15 = ((b15 << 19) | (b15 >> (64 - 19))) ^ b12;
-            b14 += b13; b13 = ((b13 << 42) | (b13 >> (64 - 42))) ^ b14;
-            b8 += b11; b11 = ((b11 << 44) | (b11 >> (64 - 44))) ^ b8;
-            b10 += b9; b9 = ((b9 << 25) | (b9 >> (64 - 25))) ^ b10;
-            b0 += b15; b15 = ((b15 << 9) | (b15 >> (64 - 9))) ^ b0;
-            b2 += b11; b11 = ((b11 << 48) | (b11 >> (64 - 48))) ^ b2;
-            b6 += b13; b13 = ((b13 << 35) | (b13 >> (64 - 35))) ^ b6;
-            b4 += b9; b9 = ((b9 << 52) | (b9 >> (64 - 52))) ^ b4;
-            b14 += b1; b1 = ((b1 << 23) | (b1 >> (64 - 23))) ^ b14;
-            b8 += b5; b5 = ((b5 << 31) | (b5 >> (64 - 31))) ^ b8;
-            b10 += b3; b3 = ((b3 << 37) | (b3 >> (64 - 37))) ^ b10;
-            b12 += b7; b7 = ((b7 << 20) | (b7 >> (64 - 20))) ^ b12;
-            b1 += k13; b0 += b1 + k12; b1 = ((b1 << 24) | (b1 >> (64 - 24))) ^ b0;
-            b3 += k15; b2 += b3 + k14; b3 = ((b3 << 13) | (b3 >> (64 - 13))) ^ b2;
-            b5 += k0; b4 += b5 + k16; b5 = ((b5 << 8) | (b5 >> (64 - 8))) ^ b4;
-            b7 += k2; b6 += b7 + k1; b7 = ((b7 << 47) | (b7 >> (64 - 47))) ^ b6;
-            b9 += k4; b8 += b9 + k3; b9 = ((b9 << 8) | (b9 >> (64 - 8))) ^ b8;
-            b11 += k6; b10 += b11 + k5; b11 = ((b11 << 17) | (b11 >> (64 - 17))) ^ b10;
-            b13 += k8 + t0; b12 += b13 + k7; b13 = ((b13 << 22) | (b13 >> (64 - 22))) ^ b12;
-            b15 += k10 + 12; b14 += b15 + k9 + t1; b15 = ((b15 << 37) | (b15 >> (64 - 37))) ^ b14;
-            b0 += b9; b9 = ((b9 << 38) | (b9 >> (64 - 38))) ^ b0;
-            b2 += b13; b13 = ((b13 << 19) | (b13 >> (64 - 19))) ^ b2;
-            b6 += b11; b11 = ((b11 << 10) | (b11 >> (64 - 10))) ^ b6;
-            b4 += b15; b15 = ((b15 << 55) | (b15 >> (64 - 55))) ^ b4;
-            b10 += b7; b7 = ((b7 << 49) | (b7 >> (64 - 49))) ^ b10;
-            b12 += b3; b3 = ((b3 << 18) | (b3 >> (64 - 18))) ^ b12;
-            b14 += b5; b5 = ((b5 << 23) | (b5 >> (64 - 23))) ^ b14;
-            b8 += b1; b1 = ((b1 << 52) | (b1 >> (64 - 52))) ^ b8;
-            b0 += b7; b7 = ((b7 << 33) | (b7 >> (64 - 33))) ^ b0;
-            b2 += b5; b5 = ((b5 << 4) | (b5 >> (64 - 4))) ^ b2;
-            b4 += b3; b3 = ((b3 << 51) | (b3 >> (64 - 51))) ^ b4;
-            b6 += b1; b1 = ((b1 << 13) | (b1 >> (64 - 13))) ^ b6;
-            b12 += b15; b15 = ((b15 << 34) | (b15 >> (64 - 34))) ^ b12;
-            b14 += b13; b13 = ((b13 << 41) | (b13 >> (64 - 41))) ^ b14;
-            b8 += b11; b11 = ((b11 << 59) | (b11 >> (64 - 59))) ^ b8;
-            b10 += b9; b9 = ((b9 << 17) | (b9 >> (64 - 17))) ^ b10;
-            b0 += b15; b15 = ((b15 << 5) | (b15 >> (64 - 5))) ^ b0;
-            b2 += b11; b11 = ((b11 << 20) | (b11 >> (64 - 20))) ^ b2;
-            b6 += b13; b13 = ((b13 << 48) | (b13 >> (64 - 48))) ^ b6;
-            b4 += b9; b9 = ((b9 << 41) | (b9 >> (64 - 41))) ^ b4;
-            b14 += b1; b1 = ((b1 << 47) | (b1 >> (64 - 47))) ^ b14;
-            b8 += b5; b5 = ((b5 << 28) | (b5 >> (64 - 28))) ^ b8;
-            b10 += b3; b3 = ((b3 << 16) | (b3 >> (64 - 16))) ^ b10;
-            b12 += b7; b7 = ((b7 << 25) | (b7 >> (64 - 25))) ^ b12;
-            b1 += k14; b0 += b1 + k13; b1 = ((b1 << 41) | (b1 >> (64 - 41))) ^ b0;
-            b3 += k16; b2 += b3 + k15; b3 = ((b3 << 9) | (b3 >> (64 - 9))) ^ b2;
-            b5 += k1; b4 += b5 + k0; b5 = ((b5 << 37) | (b5 >> (64 - 37))) ^ b4;
-            b7 += k3; b6 += b7 + k2; b7 = ((b7 << 31) | (b7 >> (64 - 31))) ^ b6;
-            b9 += k5; b8 += b9 + k4; b9 = ((b9 << 12) | (b9 >> (64 - 12))) ^ b8;
-            b11 += k7; b10 += b11 + k6; b11 = ((b11 << 47) | (b11 >> (64 - 47))) ^ b10;
-            b13 += k9 + t1; b12 += b13 + k8; b13 = ((b13 << 44) | (b13 >> (64 - 44))) ^ b12;
-            b15 += k11 + 13; b14 += b15 + k10 + t2; b15 = ((b15 << 30) | (b15 >> (64 - 30))) ^ b14;
-            b0 += b9; b9 = ((b9 << 16) | (b9 >> (64 - 16))) ^ b0;
-            b2 += b13; b13 = ((b13 << 34) | (b13 >> (64 - 34))) ^ b2;
-            b6 += b11; b11 = ((b11 << 56) | (b11 >> (64 - 56))) ^ b6;
-            b4 += b15; b15 = ((b15 << 51) | (b15 >> (64 - 51))) ^ b4;
-            b10 += b7; b7 = ((b7 << 4) | (b7 >> (64 - 4))) ^ b10;
-            b12 += b3; b3 = ((b3 << 53) | (b3 >> (64 - 53))) ^ b12;
-            b14 += b5; b5 = ((b5 << 42) | (b5 >> (64 - 42))) ^ b14;
-            b8 += b1; b1 = ((b1 << 41) | (b1 >> (64 - 41))) ^ b8;
-            b0 += b7; b7 = ((b7 << 31) | (b7 >> (64 - 31))) ^ b0;
-            b2 += b5; b5 = ((b5 << 44) | (b5 >> (64 - 44))) ^ b2;
-            b4 += b3; b3 = ((b3 << 47) | (b3 >> (64 - 47))) ^ b4;
-            b6 += b1; b1 = ((b1 << 46) | (b1 >> (64 - 46))) ^ b6;
-            b12 += b15; b15 = ((b15 << 19) | (b15 >> (64 - 19))) ^ b12;
-            b14 += b13; b13 = ((b13 << 42) | (b13 >> (64 - 42))) ^ b14;
-            b8 += b11; b11 = ((b11 << 44) | (b11 >> (64 - 44))) ^ b8;
-            b10 += b9; b9 = ((b9 << 25) | (b9 >> (64 - 25))) ^ b10;
-            b0 += b15; b15 = ((b15 << 9) | (b15 >> (64 - 9))) ^ b0;
-            b2 += b11; b11 = ((b11 << 48) | (b11 >> (64 - 48))) ^ b2;
-            b6 += b13; b13 = ((b13 << 35) | (b13 >> (64 - 35))) ^ b6;
-            b4 += b9; b9 = ((b9 << 52) | (b9 >> (64 - 52))) ^ b4;
-            b14 += b1; b1 = ((b1 << 23) | (b1 >> (64 - 23))) ^ b14;
-            b8 += b5; b5 = ((b5 << 31) | (b5 >> (64 - 31))) ^ b8;
-            b10 += b3; b3 = ((b3 << 37) | (b3 >> (64 - 37))) ^ b10;
-            b12 += b7; b7 = ((b7 << 20) | (b7 >> (64 - 20))) ^ b12;
-            b1 += k15; b0 += b1 + k14; b1 = ((b1 << 24) | (b1 >> (64 - 24))) ^ b0;
-            b3 += k0; b2 += b3 + k16; b3 = ((b3 << 13) | (b3 >> (64 - 13))) ^ b2;
-            b5 += k2; b4 += b5 + k1; b5 = ((b5 << 8) | (b5 >> (64 - 8))) ^ b4;
-            b7 += k4; b6 += b7 + k3; b7 = ((b7 << 47) | (b7 >> (64 - 47))) ^ b6;
-            b9 += k6; b8 += b9 + k5; b9 = ((b9 << 8) | (b9 >> (64 - 8))) ^ b8;
-            b11 += k8; b10 += b11 + k7; b11 = ((b11 << 17) | (b11 >> (64 - 17))) ^ b10;
-            b13 += k10 + t2; b12 += b13 + k9; b13 = ((b13 << 22) | (b13 >> (64 - 22))) ^ b12;
-            b15 += k12 + 14; b14 += b15 + k11 + t0; b15 = ((b15 << 37) | (b15 >> (64 - 37))) ^ b14;
-            b0 += b9; b9 = ((b9 << 38) | (b9 >> (64 - 38))) ^ b0;
-            b2 += b13; b13 = ((b13 << 19) | (b13 >> (64 - 19))) ^ b2;
-            b6 += b11; b11 = ((b11 << 10) | (b11 >> (64 - 10))) ^ b6;
-            b4 += b15; b15 = ((b15 << 55) | (b15 >> (64 - 55))) ^ b4;
-            b10 += b7; b7 = ((b7 << 49) | (b7 >> (64 - 49))) ^ b10;
-            b12 += b3; b3 = ((b3 << 18) | (b3 >> (64 - 18))) ^ b12;
-            b14 += b5; b5 = ((b5 << 23) | (b5 >> (64 - 23))) ^ b14;
-            b8 += b1; b1 = ((b1 << 52) | (b1 >> (64 - 52))) ^ b8;
-            b0 += b7; b7 = ((b7 << 33) | (b7 >> (64 - 33))) ^ b0;
-            b2 += b5; b5 = ((b5 << 4) | (b5 >> (64 - 4))) ^ b2;
-            b4 += b3; b3 = ((b3 << 51) | (b3 >> (64 - 51))) ^ b4;
-            b6 += b1; b1 = ((b1 << 13) | (b1 >> (64 - 13))) ^ b6;
-            b12 += b15; b15 = ((b15 << 34) | (b15 >> (64 - 34))) ^ b12;
-            b14 += b13; b13 = ((b13 << 41) | (b13 >> (64 - 41))) ^ b14;
-            b8 += b11; b11 = ((b11 << 59) | (b11 >> (64 - 59))) ^ b8;
-            b10 += b9; b9 = ((b9 << 17) | (b9 >> (64 - 17))) ^ b10;
-            b0 += b15; b15 = ((b15 << 5) | (b15 >> (64 - 5))) ^ b0;
-            b2 += b11; b11 = ((b11 << 20) | (b11 >> (64 - 20))) ^ b2;
-            b6 += b13; b13 = ((b13 << 48) | (b13 >> (64 - 48))) ^ b6;
-            b4 += b9; b9 = ((b9 << 41) | (b9 >> (64 - 41))) ^ b4;
-            b14 += b1; b1 = ((b1 << 47) | (b1 >> (64 - 47))) ^ b14;
-            b8 += b5; b5 = ((b5 << 28) | (b5 >> (64 - 28))) ^ b8;
-            b10 += b3; b3 = ((b3 << 16) | (b3 >> (64 - 16))) ^ b10;
-            b12 += b7; b7 = ((b7 << 25) | (b7 >> (64 - 25))) ^ b12;
-            b1 += k16; b0 += b1 + k15; b1 = ((b1 << 41) | (b1 >> (64 - 41))) ^ b0;
-            b3 += k1; b2 += b3 + k0; b3 = ((b3 << 9) | (b3 >> (64 - 9))) ^ b2;
-            b5 += k3; b4 += b5 + k2; b5 = ((b5 << 37) | (b5 >> (64 - 37))) ^ b4;
-            b7 += k5; b6 += b7 + k4; b7 = ((b7 << 31) | (b7 >> (64 - 31))) ^ b6;
-            b9 += k7; b8 += b9 + k6; b9 = ((b9 << 12) | (b9 >> (64 - 12))) ^ b8;
-            b11 += k9; b10 += b11 + k8; b11 = ((b11 << 47) | (b11 >> (64 - 47))) ^ b10;
-            b13 += k11 + t0; b12 += b13 + k10; b13 = ((b13 << 44) | (b13 >> (64 - 44))) ^ b12;
-            b15 += k13 + 15; b14 += b15 + k12 + t1; b15 = ((b15 << 30) | (b15 >> (64 - 30))) ^ b14;
-            b0 += b9; b9 = ((b9 << 16) | (b9 >> (64 - 16))) ^ b0;
-            b2 += b13; b13 = ((b13 << 34) | (b13 >> (64 - 34))) ^ b2;
-            b6 += b11; b11 = ((b11 << 56) | (b11 >> (64 - 56))) ^ b6;
-            b4 += b15; b15 = ((b15 << 51) | (b15 >> (64 - 51))) ^ b4;
-            b10 += b7; b7 = ((b7 << 4) | (b7 >> (64 - 4))) ^ b10;
-            b12 += b3; b3 = ((b3 << 53) | (b3 >> (64 - 53))) ^ b12;
-            b14 += b5; b5 = ((b5 << 42) | (b5 >> (64 - 42))) ^ b14;
-            b8 += b1; b1 = ((b1 << 41) | (b1 >> (64 - 41))) ^ b8;
-            b0 += b7; b7 = ((b7 << 31) | (b7 >> (64 - 31))) ^ b0;
-            b2 += b5; b5 = ((b5 << 44) | (b5 >> (64 - 44))) ^ b2;
-            b4 += b3; b3 = ((b3 << 47) | (b3 >> (64 - 47))) ^ b4;
-            b6 += b1; b1 = ((b1 << 46) | (b1 >> (64 - 46))) ^ b6;
-            b12 += b15; b15 = ((b15 << 19) | (b15 >> (64 - 19))) ^ b12;
-            b14 += b13; b13 = ((b13 << 42) | (b13 >> (64 - 42))) ^ b14;
-            b8 += b11; b11 = ((b11 << 44) | (b11 >> (64 - 44))) ^ b8;
-            b10 += b9; b9 = ((b9 << 25) | (b9 >> (64 - 25))) ^ b10;
-            b0 += b15; b15 = ((b15 << 9) | (b15 >> (64 - 9))) ^ b0;
-            b2 += b11; b11 = ((b11 << 48) | (b11 >> (64 - 48))) ^ b2;
-            b6 += b13; b13 = ((b13 << 35) | (b13 >> (64 - 35))) ^ b6;
-            b4 += b9; b9 = ((b9 << 52) | (b9 >> (64 - 52))) ^ b4;
-            b14 += b1; b1 = ((b1 << 23) | (b1 >> (64 - 23))) ^ b14;
-            b8 += b5; b5 = ((b5 << 31) | (b5 >> (64 - 31))) ^ b8;
-            b10 += b3; b3 = ((b3 << 37) | (b3 >> (64 - 37))) ^ b10;
-            b12 += b7; b7 = ((b7 << 20) | (b7 >> (64 - 20))) ^ b12;
-            b1 += k0; b0 += b1 + k16; b1 = ((b1 << 24) | (b1 >> (64 - 24))) ^ b0;
-            b3 += k2; b2 += b3 + k1; b3 = ((b3 << 13) | (b3 >> (64 - 13))) ^ b2;
-            b5 += k4; b4 += b5 + k3; b5 = ((b5 << 8) | (b5 >> (64 - 8))) ^ b4;
-            b7 += k6; b6 += b7 + k5; b7 = ((b7 << 47) | (b7 >> (64 - 47))) ^ b6;
-            b9 += k8; b8 += b9 + k7; b9 = ((b9 << 8) | (b9 >> (64 - 8))) ^ b8;
-            b11 += k10; b10 += b11 + k9; b11 = ((b11 << 17) | (b11 >> (64 - 17))) ^ b10;
-            b13 += k12 + t1; b12 += b13 + k11; b13 = ((b13 << 22) | (b13 >> (64 - 22))) ^ b12;
-            b15 += k14 + 16; b14 += b15 + k13 + t2; b15 = ((b15 << 37) | (b15 >> (64 - 37))) ^ b14;
-            b0 += b9; b9 = ((b9 << 38) | (b9 >> (64 - 38))) ^ b0;
-            b2 += b13; b13 = ((b13 << 19) | (b13 >> (64 - 19))) ^ b2;
-            b6 += b11; b11 = ((b11 << 10) | (b11 >> (64 - 10))) ^ b6;
-            b4 += b15; b15 = ((b15 << 55) | (b15 >> (64 - 55))) ^ b4;
-            b10 += b7; b7 = ((b7 << 49) | (b7 >> (64 - 49))) ^ b10;
-            b12 += b3; b3 = ((b3 << 18) | (b3 >> (64 - 18))) ^ b12;
-            b14 += b5; b5 = ((b5 << 23) | (b5 >> (64 - 23))) ^ b14;
-            b8 += b1; b1 = ((b1 << 52) | (b1 >> (64 - 52))) ^ b8;
-            b0 += b7; b7 = ((b7 << 33) | (b7 >> (64 - 33))) ^ b0;
-            b2 += b5; b5 = ((b5 << 4) | (b5 >> (64 - 4))) ^ b2;
-            b4 += b3; b3 = ((b3 << 51) | (b3 >> (64 - 51))) ^ b4;
-            b6 += b1; b1 = ((b1 << 13) | (b1 >> (64 - 13))) ^ b6;
-            b12 += b15; b15 = ((b15 << 34) | (b15 >> (64 - 34))) ^ b12;
-            b14 += b13; b13 = ((b13 << 41) | (b13 >> (64 - 41))) ^ b14;
-            b8 += b11; b11 = ((b11 << 59) | (b11 >> (64 - 59))) ^ b8;
-            b10 += b9; b9 = ((b9 << 17) | (b9 >> (64 - 17))) ^ b10;
-            b0 += b15; b15 = ((b15 << 5) | (b15 >> (64 - 5))) ^ b0;
-            b2 += b11; b11 = ((b11 << 20) | (b11 >> (64 - 20))) ^ b2;
-            b6 += b13; b13 = ((b13 << 48) | (b13 >> (64 - 48))) ^ b6;
-            b4 += b9; b9 = ((b9 << 41) | (b9 >> (64 - 41))) ^ b4;
-            b14 += b1; b1 = ((b1 << 47) | (b1 >> (64 - 47))) ^ b14;
-            b8 += b5; b5 = ((b5 << 28) | (b5 >> (64 - 28))) ^ b8;
-            b10 += b3; b3 = ((b3 << 16) | (b3 >> (64 - 16))) ^ b10;
-            b12 += b7; b7 = ((b7 << 25) | (b7 >> (64 - 25))) ^ b12;
-            b1 += k1; b0 += b1 + k0; b1 = ((b1 << 41) | (b1 >> (64 - 41))) ^ b0;
-            b3 += k3; b2 += b3 + k2; b3 = ((b3 << 9) | (b3 >> (64 - 9))) ^ b2;
-            b5 += k5; b4 += b5 + k4; b5 = ((b5 << 37) | (b5 >> (64 - 37))) ^ b4;
-            b7 += k7; b6 += b7 + k6; b7 = ((b7 << 31) | (b7 >> (64 - 31))) ^ b6;
-            b9 += k9; b8 += b9 + k8; b9 = ((b9 << 12) | (b9 >> (64 - 12))) ^ b8;
-            b11 += k11; b10 += b11 + k10; b11 = ((b11 << 47) | (b11 >> (64 - 47))) ^ b10;
-            b13 += k13 + t2; b12 += b13 + k12; b13 = ((b13 << 44) | (b13 >> (64 - 44))) ^ b12;
-            b15 += k15 + 17; b14 += b15 + k14 + t0; b15 = ((b15 << 30) | (b15 >> (64 - 30))) ^ b14;
-            b0 += b9; b9 = ((b9 << 16) | (b9 >> (64 - 16))) ^ b0;
-            b2 += b13; b13 = ((b13 << 34) | (b13 >> (64 - 34))) ^ b2;
-            b6 += b11; b11 = ((b11 << 56) | (b11 >> (64 - 56))) ^ b6;
-            b4 += b15; b15 = ((b15 << 51) | (b15 >> (64 - 51))) ^ b4;
-            b10 += b7; b7 = ((b7 << 4) | (b7 >> (64 - 4))) ^ b10;
-            b12 += b3; b3 = ((b3 << 53) | (b3 >> (64 - 53))) ^ b12;
-            b14 += b5; b5 = ((b5 << 42) | (b5 >> (64 - 42))) ^ b14;
-            b8 += b1; b1 = ((b1 << 41) | (b1 >> (64 - 41))) ^ b8;
-            b0 += b7; b7 = ((b7 << 31) | (b7 >> (64 - 31))) ^ b0;
-            b2 += b5; b5 = ((b5 << 44) | (b5 >> (64 - 44))) ^ b2;
-            b4 += b3; b3 = ((b3 << 47) | (b3 >> (64 - 47))) ^ b4;
-            b6 += b1; b1 = ((b1 << 46) | (b1 >> (64 - 46))) ^ b6;
-            b12 += b15; b15 = ((b15 << 19) | (b15 >> (64 - 19))) ^ b12;
-            b14 += b13; b13 = ((b13 << 42) | (b13 >> (64 - 42))) ^ b14;
-            b8 += b11; b11 = ((b11 << 44) | (b11 >> (64 - 44))) ^ b8;
-            b10 += b9; b9 = ((b9 << 25) | (b9 >> (64 - 25))) ^ b10;
-            b0 += b15; b15 = ((b15 << 9) | (b15 >> (64 - 9))) ^ b0;
-            b2 += b11; b11 = ((b11 << 48) | (b11 >> (64 - 48))) ^ b2;
-            b6 += b13; b13 = ((b13 << 35) | (b13 >> (64 - 35))) ^ b6;
-            b4 += b9; b9 = ((b9 << 52) | (b9 >> (64 - 52))) ^ b4;
-            b14 += b1; b1 = ((b1 << 23) | (b1 >> (64 - 23))) ^ b14;
-            b8 += b5; b5 = ((b5 << 31) | (b5 >> (64 - 31))) ^ b8;
-            b10 += b3; b3 = ((b3 << 37) | (b3 >> (64 - 37))) ^ b10;
-            b12 += b7; b7 = ((b7 << 20) | (b7 >> (64 - 20))) ^ b12;
-            b1 += k2; b0 += b1 + k1; b1 = ((b1 << 24) | (b1 >> (64 - 24))) ^ b0;
-            b3 += k4; b2 += b3 + k3; b3 = ((b3 << 13) | (b3 >> (64 - 13))) ^ b2;
-            b5 += k6; b4 += b5 + k5; b5 = ((b5 << 8) | (b5 >> (64 - 8))) ^ b4;
-            b7 += k8; b6 += b7 + k7; b7 = ((b7 << 47) | (b7 >> (64 - 47))) ^ b6;
-            b9 += k10; b8 += b9 + k9; b9 = ((b9 << 8) | (b9 >> (64 - 8))) ^ b8;
-            b11 += k12; b10 += b11 + k11; b11 = ((b11 << 17) | (b11 >> (64 - 17))) ^ b10;
-            b13 += k14 + t0; b12 += b13 + k13; b13 = ((b13 << 22) | (b13 >> (64 - 22))) ^ b12;
-            b15 += k16 + 18; b14 += b15 + k15 + t1; b15 = ((b15 << 37) | (b15 >> (64 - 37))) ^ b14;
-            b0 += b9; b9 = ((b9 << 38) | (b9 >> (64 - 38))) ^ b0;
-            b2 += b13; b13 = ((b13 << 19) | (b13 >> (64 - 19))) ^ b2;
-            b6 += b11; b11 = ((b11 << 10) | (b11 >> (64 - 10))) ^ b6;
-            b4 += b15; b15 = ((b15 << 55) | (b15 >> (64 - 55))) ^ b4;
-            b10 += b7; b7 = ((b7 << 49) | (b7 >> (64 - 49))) ^ b10;
-            b12 += b3; b3 = ((b3 << 18) | (b3 >> (64 - 18))) ^ b12;
-            b14 += b5; b5 = ((b5 << 23) | (b5 >> (64 - 23))) ^ b14;
-            b8 += b1; b1 = ((b1 << 52) | (b1 >> (64 - 52))) ^ b8;
-            b0 += b7; b7 = ((b7 << 33) | (b7 >> (64 - 33))) ^ b0;
-            b2 += b5; b5 = ((b5 << 4) | (b5 >> (64 - 4))) ^ b2;
-            b4 += b3; b3 = ((b3 << 51) | (b3 >> (64 - 51))) ^ b4;
-            b6 += b1; b1 = ((b1 << 13) | (b1 >> (64 - 13))) ^ b6;
-            b12 += b15; b15 = ((b15 << 34) | (b15 >> (64 - 34))) ^ b12;
-            b14 += b13; b13 = ((b13 << 41) | (b13 >> (64 - 41))) ^ b14;
-            b8 += b11; b11 = ((b11 << 59) | (b11 >> (64 - 59))) ^ b8;
-            b10 += b9; b9 = ((b9 << 17) | (b9 >> (64 - 17))) ^ b10;
-            b0 += b15; b15 = ((b15 << 5) | (b15 >> (64 - 5))) ^ b0;
-            b2 += b11; b11 = ((b11 << 20) | (b11 >> (64 - 20))) ^ b2;
-            b6 += b13; b13 = ((b13 << 48) | (b13 >> (64 - 48))) ^ b6;
-            b4 += b9; b9 = ((b9 << 41) | (b9 >> (64 - 41))) ^ b4;
-            b14 += b1; b1 = ((b1 << 47) | (b1 >> (64 - 47))) ^ b14;
-            b8 += b5; b5 = ((b5 << 28) | (b5 >> (64 - 28))) ^ b8;
-            b10 += b3; b3 = ((b3 << 16) | (b3 >> (64 - 16))) ^ b10;
-            b12 += b7; b7 = ((b7 << 25) | (b7 >> (64 - 25))) ^ b12;
-            b1 += k3; b0 += b1 + k2; b1 = ((b1 << 41) | (b1 >> (64 - 41))) ^ b0;
-            b3 += k5; b2 += b3 + k4; b3 = ((b3 << 9) | (b3 >> (64 - 9))) ^ b2;
-            b5 += k7; b4 += b5 + k6; b5 = ((b5 << 37) | (b5 >> (64 - 37))) ^ b4;
-            b7 += k9; b6 += b7 + k8; b7 = ((b7 << 31) | (b7 >> (64 - 31))) ^ b6;
-            b9 += k11; b8 += b9 + k10; b9 = ((b9 << 12) | (b9 >> (64 - 12))) ^ b8;
-            b11 += k13; b10 += b11 + k12; b11 = ((b11 << 47) | (b11 >> (64 - 47))) ^ b10;
-            b13 += k15 + t1; b12 += b13 + k14; b13 = ((b13 << 44) | (b13 >> (64 - 44))) ^ b12;
-            b15 += k0 + 19; b14 += b15 + k16 + t2; b15 = ((b15 << 30) | (b15 >> (64 - 30))) ^ b14;
-            b0 += b9; b9 = ((b9 << 16) | (b9 >> (64 - 16))) ^ b0;
-            b2 += b13; b13 = ((b13 << 34) | (b13 >> (64 - 34))) ^ b2;
-            b6 += b11; b11 = ((b11 << 56) | (b11 >> (64 - 56))) ^ b6;
-            b4 += b15; b15 = ((b15 << 51) | (b15 >> (64 - 51))) ^ b4;
-            b10 += b7; b7 = ((b7 << 4) | (b7 >> (64 - 4))) ^ b10;
-            b12 += b3; b3 = ((b3 << 53) | (b3 >> (64 - 53))) ^ b12;
-            b14 += b5; b5 = ((b5 << 42) | (b5 >> (64 - 42))) ^ b14;
-            b8 += b1; b1 = ((b1 << 41) | (b1 >> (64 - 41))) ^ b8;
-            b0 += b7; b7 = ((b7 << 31) | (b7 >> (64 - 31))) ^ b0;
-            b2 += b5; b5 = ((b5 << 44) | (b5 >> (64 - 44))) ^ b2;
-            b4 += b3; b3 = ((b3 << 47) | (b3 >> (64 - 47))) ^ b4;
-            b6 += b1; b1 = ((b1 << 46) | (b1 >> (64 - 46))) ^ b6;
-            b12 += b15; b15 = ((b15 << 19) | (b15 >> (64 - 19))) ^ b12;
-            b14 += b13; b13 = ((b13 << 42) | (b13 >> (64 - 42))) ^ b14;
-            b8 += b11; b11 = ((b11 << 44) | (b11 >> (64 - 44))) ^ b8;
-            b10 += b9; b9 = ((b9 << 25) | (b9 >> (64 - 25))) ^ b10;
-            b0 += b15; b15 = ((b15 << 9) | (b15 >> (64 - 9))) ^ b0;
-            b2 += b11; b11 = ((b11 << 48) | (b11 >> (64 - 48))) ^ b2;
-            b6 += b13; b13 = ((b13 << 35) | (b13 >> (64 - 35))) ^ b6;
-            b4 += b9; b9 = ((b9 << 52) | (b9 >> (64 - 52))) ^ b4;
-            b14 += b1; b1 = ((b1 << 23) | (b1 >> (64 - 23))) ^ b14;
-            b8 += b5; b5 = ((b5 << 31) | (b5 >> (64 - 31))) ^ b8;
-            b10 += b3; b3 = ((b3 << 37) | (b3 >> (64 - 37))) ^ b10;
-            b12 += b7; b7 = ((b7 << 20) | (b7 >> (64 - 20))) ^ b12;
+	b1 += k1; b0 += b1 + k0; b1 = ((b1 << 24) | (b1 >> (64 - 24))) ^ b0;
+	b3 += k3; b2 += b3 + k2; b3 = ((b3 << 13) | (b3 >> (64 - 13))) ^ b2;
+	b5 += k5; b4 += b5 + k4; b5 = ((b5 << 8) | (b5 >> (64 - 8))) ^ b4;
+	b7 += k7; b6 += b7 + k6; b7 = ((b7 << 47) | (b7 >> (64 - 47))) ^ b6;
+	b9 += k9; b8 += b9 + k8; b9 = ((b9 << 8) | (b9 >> (64 - 8))) ^ b8;
+	b11 += k11; b10 += b11 + k10; b11 = ((b11 << 17) | (b11 >> (64 - 17))) ^ b10;
+	b13 += k13 + t0; b12 += b13 + k12; b13 = ((b13 << 22) | (b13 >> (64 - 22))) ^ b12;
+	b15 += k15; b14 += b15 + k14 + t1; b15 = ((b15 << 37) | (b15 >> (64 - 37))) ^ b14;
+	b0 += b9; b9 = ((b9 << 38) | (b9 >> (64 - 38))) ^ b0;
+	b2 += b13; b13 = ((b13 << 19) | (b13 >> (64 - 19))) ^ b2;
+	b6 += b11; b11 = ((b11 << 10) | (b11 >> (64 - 10))) ^ b6;
+	b4 += b15; b15 = ((b15 << 55) | (b15 >> (64 - 55))) ^ b4;
+	b10 += b7; b7 = ((b7 << 49) | (b7 >> (64 - 49))) ^ b10;
+	b12 += b3; b3 = ((b3 << 18) | (b3 >> (64 - 18))) ^ b12;
+	b14 += b5; b5 = ((b5 << 23) | (b5 >> (64 - 23))) ^ b14;
+	b8 += b1; b1 = ((b1 << 52) | (b1 >> (64 - 52))) ^ b8;
+	b0 += b7; b7 = ((b7 << 33) | (b7 >> (64 - 33))) ^ b0;
+	b2 += b5; b5 = ((b5 << 4) | (b5 >> (64 - 4))) ^ b2;
+	b4 += b3; b3 = ((b3 << 51) | (b3 >> (64 - 51))) ^ b4;
+	b6 += b1; b1 = ((b1 << 13) | (b1 >> (64 - 13))) ^ b6;
+	b12 += b15; b15 = ((b15 << 34) | (b15 >> (64 - 34))) ^ b12;
+	b14 += b13; b13 = ((b13 << 41) | (b13 >> (64 - 41))) ^ b14;
+	b8 += b11; b11 = ((b11 << 59) | (b11 >> (64 - 59))) ^ b8;
+	b10 += b9; b9 = ((b9 << 17) | (b9 >> (64 - 17))) ^ b10;
+	b0 += b15; b15 = ((b15 << 5) | (b15 >> (64 - 5))) ^ b0;
+	b2 += b11; b11 = ((b11 << 20) | (b11 >> (64 - 20))) ^ b2;
+	b6 += b13; b13 = ((b13 << 48) | (b13 >> (64 - 48))) ^ b6;
+	b4 += b9; b9 = ((b9 << 41) | (b9 >> (64 - 41))) ^ b4;
+	b14 += b1; b1 = ((b1 << 47) | (b1 >> (64 - 47))) ^ b14;
+	b8 += b5; b5 = ((b5 << 28) | (b5 >> (64 - 28))) ^ b8;
+	b10 += b3; b3 = ((b3 << 16) | (b3 >> (64 - 16))) ^ b10;
+	b12 += b7; b7 = ((b7 << 25) | (b7 >> (64 - 25))) ^ b12;
+	b1 += k2; b0 += b1 + k1; b1 = ((b1 << 41) | (b1 >> (64 - 41))) ^ b0;
+	b3 += k4; b2 += b3 + k3; b3 = ((b3 << 9) | (b3 >> (64 - 9))) ^ b2;
+	b5 += k6; b4 += b5 + k5; b5 = ((b5 << 37) | (b5 >> (64 - 37))) ^ b4;
+	b7 += k8; b6 += b7 + k7; b7 = ((b7 << 31) | (b7 >> (64 - 31))) ^ b6;
+	b9 += k10; b8 += b9 + k9; b9 = ((b9 << 12) | (b9 >> (64 - 12))) ^ b8;
+	b11 += k12; b10 += b11 + k11; b11 = ((b11 << 47) | (b11 >> (64 - 47))) ^ b10;
+	b13 += k14 + t1; b12 += b13 + k13; b13 = ((b13 << 44) | (b13 >> (64 - 44))) ^ b12;
+	b15 += k16 + 1; b14 += b15 + k15 + t2; b15 = ((b15 << 30) | (b15 >> (64 - 30))) ^ b14;
+	b0 += b9; b9 = ((b9 << 16) | (b9 >> (64 - 16))) ^ b0;
+	b2 += b13; b13 = ((b13 << 34) | (b13 >> (64 - 34))) ^ b2;
+	b6 += b11; b11 = ((b11 << 56) | (b11 >> (64 - 56))) ^ b6;
+	b4 += b15; b15 = ((b15 << 51) | (b15 >> (64 - 51))) ^ b4;
+	b10 += b7; b7 = ((b7 << 4) | (b7 >> (64 - 4))) ^ b10;
+	b12 += b3; b3 = ((b3 << 53) | (b3 >> (64 - 53))) ^ b12;
+	b14 += b5; b5 = ((b5 << 42) | (b5 >> (64 - 42))) ^ b14;
+	b8 += b1; b1 = ((b1 << 41) | (b1 >> (64 - 41))) ^ b8;
+	b0 += b7; b7 = ((b7 << 31) | (b7 >> (64 - 31))) ^ b0;
+	b2 += b5; b5 = ((b5 << 44) | (b5 >> (64 - 44))) ^ b2;
+	b4 += b3; b3 = ((b3 << 47) | (b3 >> (64 - 47))) ^ b4;
+	b6 += b1; b1 = ((b1 << 46) | (b1 >> (64 - 46))) ^ b6;
+	b12 += b15; b15 = ((b15 << 19) | (b15 >> (64 - 19))) ^ b12;
+	b14 += b13; b13 = ((b13 << 42) | (b13 >> (64 - 42))) ^ b14;
+	b8 += b11; b11 = ((b11 << 44) | (b11 >> (64 - 44))) ^ b8;
+	b10 += b9; b9 = ((b9 << 25) | (b9 >> (64 - 25))) ^ b10;
+	b0 += b15; b15 = ((b15 << 9) | (b15 >> (64 - 9))) ^ b0;
+	b2 += b11; b11 = ((b11 << 48) | (b11 >> (64 - 48))) ^ b2;
+	b6 += b13; b13 = ((b13 << 35) | (b13 >> (64 - 35))) ^ b6;
+	b4 += b9; b9 = ((b9 << 52) | (b9 >> (64 - 52))) ^ b4;
+	b14 += b1; b1 = ((b1 << 23) | (b1 >> (64 - 23))) ^ b14;
+	b8 += b5; b5 = ((b5 << 31) | (b5 >> (64 - 31))) ^ b8;
+	b10 += b3; b3 = ((b3 << 37) | (b3 >> (64 - 37))) ^ b10;
+	b12 += b7; b7 = ((b7 << 20) | (b7 >> (64 - 20))) ^ b12;
+	b1 += k3; b0 += b1 + k2; b1 = ((b1 << 24) | (b1 >> (64 - 24))) ^ b0;
+	b3 += k5; b2 += b3 + k4; b3 = ((b3 << 13) | (b3 >> (64 - 13))) ^ b2;
+	b5 += k7; b4 += b5 + k6; b5 = ((b5 << 8) | (b5 >> (64 - 8))) ^ b4;
+	b7 += k9; b6 += b7 + k8; b7 = ((b7 << 47) | (b7 >> (64 - 47))) ^ b6;
+	b9 += k11; b8 += b9 + k10; b9 = ((b9 << 8) | (b9 >> (64 - 8))) ^ b8;
+	b11 += k13; b10 += b11 + k12; b11 = ((b11 << 17) | (b11 >> (64 - 17))) ^ b10;
+	b13 += k15 + t2; b12 += b13 + k14; b13 = ((b13 << 22) | (b13 >> (64 - 22))) ^ b12;
+	b15 += k0 + 2; b14 += b15 + k16 + t0; b15 = ((b15 << 37) | (b15 >> (64 - 37))) ^ b14;
+	b0 += b9; b9 = ((b9 << 38) | (b9 >> (64 - 38))) ^ b0;
+	b2 += b13; b13 = ((b13 << 19) | (b13 >> (64 - 19))) ^ b2;
+	b6 += b11; b11 = ((b11 << 10) | (b11 >> (64 - 10))) ^ b6;
+	b4 += b15; b15 = ((b15 << 55) | (b15 >> (64 - 55))) ^ b4;
+	b10 += b7; b7 = ((b7 << 49) | (b7 >> (64 - 49))) ^ b10;
+	b12 += b3; b3 = ((b3 << 18) | (b3 >> (64 - 18))) ^ b12;
+	b14 += b5; b5 = ((b5 << 23) | (b5 >> (64 - 23))) ^ b14;
+	b8 += b1; b1 = ((b1 << 52) | (b1 >> (64 - 52))) ^ b8;
+	b0 += b7; b7 = ((b7 << 33) | (b7 >> (64 - 33))) ^ b0;
+	b2 += b5; b5 = ((b5 << 4) | (b5 >> (64 - 4))) ^ b2;
+	b4 += b3; b3 = ((b3 << 51) | (b3 >> (64 - 51))) ^ b4;
+	b6 += b1; b1 = ((b1 << 13) | (b1 >> (64 - 13))) ^ b6;
+	b12 += b15; b15 = ((b15 << 34) | (b15 >> (64 - 34))) ^ b12;
+	b14 += b13; b13 = ((b13 << 41) | (b13 >> (64 - 41))) ^ b14;
+	b8 += b11; b11 = ((b11 << 59) | (b11 >> (64 - 59))) ^ b8;
+	b10 += b9; b9 = ((b9 << 17) | (b9 >> (64 - 17))) ^ b10;
+	b0 += b15; b15 = ((b15 << 5) | (b15 >> (64 - 5))) ^ b0;
+	b2 += b11; b11 = ((b11 << 20) | (b11 >> (64 - 20))) ^ b2;
+	b6 += b13; b13 = ((b13 << 48) | (b13 >> (64 - 48))) ^ b6;
+	b4 += b9; b9 = ((b9 << 41) | (b9 >> (64 - 41))) ^ b4;
+	b14 += b1; b1 = ((b1 << 47) | (b1 >> (64 - 47))) ^ b14;
+	b8 += b5; b5 = ((b5 << 28) | (b5 >> (64 - 28))) ^ b8;
+	b10 += b3; b3 = ((b3 << 16) | (b3 >> (64 - 16))) ^ b10;
+	b12 += b7; b7 = ((b7 << 25) | (b7 >> (64 - 25))) ^ b12;
+	b1 += k4; b0 += b1 + k3; b1 = ((b1 << 41) | (b1 >> (64 - 41))) ^ b0;
+	b3 += k6; b2 += b3 + k5; b3 = ((b3 << 9) | (b3 >> (64 - 9))) ^ b2;
+	b5 += k8; b4 += b5 + k7; b5 = ((b5 << 37) | (b5 >> (64 - 37))) ^ b4;
+	b7 += k10; b6 += b7 + k9; b7 = ((b7 << 31) | (b7 >> (64 - 31))) ^ b6;
+	b9 += k12; b8 += b9 + k11; b9 = ((b9 << 12) | (b9 >> (64 - 12))) ^ b8;
+	b11 += k14; b10 += b11 + k13; b11 = ((b11 << 47) | (b11 >> (64 - 47))) ^ b10;
+	b13 += k16 + t0; b12 += b13 + k15; b13 = ((b13 << 44) | (b13 >> (64 - 44))) ^ b12;
+	b15 += k1 + 3; b14 += b15 + k0 + t1; b15 = ((b15 << 30) | (b15 >> (64 - 30))) ^ b14;
+	b0 += b9; b9 = ((b9 << 16) | (b9 >> (64 - 16))) ^ b0;
+	b2 += b13; b13 = ((b13 << 34) | (b13 >> (64 - 34))) ^ b2;
+	b6 += b11; b11 = ((b11 << 56) | (b11 >> (64 - 56))) ^ b6;
+	b4 += b15; b15 = ((b15 << 51) | (b15 >> (64 - 51))) ^ b4;
+	b10 += b7; b7 = ((b7 << 4) | (b7 >> (64 - 4))) ^ b10;
+	b12 += b3; b3 = ((b3 << 53) | (b3 >> (64 - 53))) ^ b12;
+	b14 += b5; b5 = ((b5 << 42) | (b5 >> (64 - 42))) ^ b14;
+	b8 += b1; b1 = ((b1 << 41) | (b1 >> (64 - 41))) ^ b8;
+	b0 += b7; b7 = ((b7 << 31) | (b7 >> (64 - 31))) ^ b0;
+	b2 += b5; b5 = ((b5 << 44) | (b5 >> (64 - 44))) ^ b2;
+	b4 += b3; b3 = ((b3 << 47) | (b3 >> (64 - 47))) ^ b4;
+	b6 += b1; b1 = ((b1 << 46) | (b1 >> (64 - 46))) ^ b6;
+	b12 += b15; b15 = ((b15 << 19) | (b15 >> (64 - 19))) ^ b12;
+	b14 += b13; b13 = ((b13 << 42) | (b13 >> (64 - 42))) ^ b14;
+	b8 += b11; b11 = ((b11 << 44) | (b11 >> (64 - 44))) ^ b8;
+	b10 += b9; b9 = ((b9 << 25) | (b9 >> (64 - 25))) ^ b10;
+	b0 += b15; b15 = ((b15 << 9) | (b15 >> (64 - 9))) ^ b0;
+	b2 += b11; b11 = ((b11 << 48) | (b11 >> (64 - 48))) ^ b2;
+	b6 += b13; b13 = ((b13 << 35) | (b13 >> (64 - 35))) ^ b6;
+	b4 += b9; b9 = ((b9 << 52) | (b9 >> (64 - 52))) ^ b4;
+	b14 += b1; b1 = ((b1 << 23) | (b1 >> (64 - 23))) ^ b14;
+	b8 += b5; b5 = ((b5 << 31) | (b5 >> (64 - 31))) ^ b8;
+	b10 += b3; b3 = ((b3 << 37) | (b3 >> (64 - 37))) ^ b10;
+	b12 += b7; b7 = ((b7 << 20) | (b7 >> (64 - 20))) ^ b12;
+	b1 += k5; b0 += b1 + k4; b1 = ((b1 << 24) | (b1 >> (64 - 24))) ^ b0;
+	b3 += k7; b2 += b3 + k6; b3 = ((b3 << 13) | (b3 >> (64 - 13))) ^ b2;
+	b5 += k9; b4 += b5 + k8; b5 = ((b5 << 8) | (b5 >> (64 - 8))) ^ b4;
+	b7 += k11; b6 += b7 + k10; b7 = ((b7 << 47) | (b7 >> (64 - 47))) ^ b6;
+	b9 += k13; b8 += b9 + k12; b9 = ((b9 << 8) | (b9 >> (64 - 8))) ^ b8;
+	b11 += k15; b10 += b11 + k14; b11 = ((b11 << 17) | (b11 >> (64 - 17))) ^ b10;
+	b13 += k0 + t1; b12 += b13 + k16; b13 = ((b13 << 22) | (b13 >> (64 - 22))) ^ b12;
+	b15 += k2 + 4; b14 += b15 + k1 + t2; b15 = ((b15 << 37) | (b15 >> (64 - 37))) ^ b14;
+	b0 += b9; b9 = ((b9 << 38) | (b9 >> (64 - 38))) ^ b0;
+	b2 += b13; b13 = ((b13 << 19) | (b13 >> (64 - 19))) ^ b2;
+	b6 += b11; b11 = ((b11 << 10) | (b11 >> (64 - 10))) ^ b6;
+	b4 += b15; b15 = ((b15 << 55) | (b15 >> (64 - 55))) ^ b4;
+	b10 += b7; b7 = ((b7 << 49) | (b7 >> (64 - 49))) ^ b10;
+	b12 += b3; b3 = ((b3 << 18) | (b3 >> (64 - 18))) ^ b12;
+	b14 += b5; b5 = ((b5 << 23) | (b5 >> (64 - 23))) ^ b14;
+	b8 += b1; b1 = ((b1 << 52) | (b1 >> (64 - 52))) ^ b8;
+	b0 += b7; b7 = ((b7 << 33) | (b7 >> (64 - 33))) ^ b0;
+	b2 += b5; b5 = ((b5 << 4) | (b5 >> (64 - 4))) ^ b2;
+	b4 += b3; b3 = ((b3 << 51) | (b3 >> (64 - 51))) ^ b4;
+	b6 += b1; b1 = ((b1 << 13) | (b1 >> (64 - 13))) ^ b6;
+	b12 += b15; b15 = ((b15 << 34) | (b15 >> (64 - 34))) ^ b12;
+	b14 += b13; b13 = ((b13 << 41) | (b13 >> (64 - 41))) ^ b14;
+	b8 += b11; b11 = ((b11 << 59) | (b11 >> (64 - 59))) ^ b8;
+	b10 += b9; b9 = ((b9 << 17) | (b9 >> (64 - 17))) ^ b10;
+	b0 += b15; b15 = ((b15 << 5) | (b15 >> (64 - 5))) ^ b0;
+	b2 += b11; b11 = ((b11 << 20) | (b11 >> (64 - 20))) ^ b2;
+	b6 += b13; b13 = ((b13 << 48) | (b13 >> (64 - 48))) ^ b6;
+	b4 += b9; b9 = ((b9 << 41) | (b9 >> (64 - 41))) ^ b4;
+	b14 += b1; b1 = ((b1 << 47) | (b1 >> (64 - 47))) ^ b14;
+	b8 += b5; b5 = ((b5 << 28) | (b5 >> (64 - 28))) ^ b8;
+	b10 += b3; b3 = ((b3 << 16) | (b3 >> (64 - 16))) ^ b10;
+	b12 += b7; b7 = ((b7 << 25) | (b7 >> (64 - 25))) ^ b12;
+	b1 += k6; b0 += b1 + k5; b1 = ((b1 << 41) | (b1 >> (64 - 41))) ^ b0;
+	b3 += k8; b2 += b3 + k7; b3 = ((b3 << 9) | (b3 >> (64 - 9))) ^ b2;
+	b5 += k10; b4 += b5 + k9; b5 = ((b5 << 37) | (b5 >> (64 - 37))) ^ b4;
+	b7 += k12; b6 += b7 + k11; b7 = ((b7 << 31) | (b7 >> (64 - 31))) ^ b6;
+	b9 += k14; b8 += b9 + k13; b9 = ((b9 << 12) | (b9 >> (64 - 12))) ^ b8;
+	b11 += k16; b10 += b11 + k15; b11 = ((b11 << 47) | (b11 >> (64 - 47))) ^ b10;
+	b13 += k1 + t2; b12 += b13 + k0; b13 = ((b13 << 44) | (b13 >> (64 - 44))) ^ b12;
+	b15 += k3 + 5; b14 += b15 + k2 + t0; b15 = ((b15 << 30) | (b15 >> (64 - 30))) ^ b14;
+	b0 += b9; b9 = ((b9 << 16) | (b9 >> (64 - 16))) ^ b0;
+	b2 += b13; b13 = ((b13 << 34) | (b13 >> (64 - 34))) ^ b2;
+	b6 += b11; b11 = ((b11 << 56) | (b11 >> (64 - 56))) ^ b6;
+	b4 += b15; b15 = ((b15 << 51) | (b15 >> (64 - 51))) ^ b4;
+	b10 += b7; b7 = ((b7 << 4) | (b7 >> (64 - 4))) ^ b10;
+	b12 += b3; b3 = ((b3 << 53) | (b3 >> (64 - 53))) ^ b12;
+	b14 += b5; b5 = ((b5 << 42) | (b5 >> (64 - 42))) ^ b14;
+	b8 += b1; b1 = ((b1 << 41) | (b1 >> (64 - 41))) ^ b8;
+	b0 += b7; b7 = ((b7 << 31) | (b7 >> (64 - 31))) ^ b0;
+	b2 += b5; b5 = ((b5 << 44) | (b5 >> (64 - 44))) ^ b2;
+	b4 += b3; b3 = ((b3 << 47) | (b3 >> (64 - 47))) ^ b4;
+	b6 += b1; b1 = ((b1 << 46) | (b1 >> (64 - 46))) ^ b6;
+	b12 += b15; b15 = ((b15 << 19) | (b15 >> (64 - 19))) ^ b12;
+	b14 += b13; b13 = ((b13 << 42) | (b13 >> (64 - 42))) ^ b14;
+	b8 += b11; b11 = ((b11 << 44) | (b11 >> (64 - 44))) ^ b8;
+	b10 += b9; b9 = ((b9 << 25) | (b9 >> (64 - 25))) ^ b10;
+	b0 += b15; b15 = ((b15 << 9) | (b15 >> (64 - 9))) ^ b0;
+	b2 += b11; b11 = ((b11 << 48) | (b11 >> (64 - 48))) ^ b2;
+	b6 += b13; b13 = ((b13 << 35) | (b13 >> (64 - 35))) ^ b6;
+	b4 += b9; b9 = ((b9 << 52) | (b9 >> (64 - 52))) ^ b4;
+	b14 += b1; b1 = ((b1 << 23) | (b1 >> (64 - 23))) ^ b14;
+	b8 += b5; b5 = ((b5 << 31) | (b5 >> (64 - 31))) ^ b8;
+	b10 += b3; b3 = ((b3 << 37) | (b3 >> (64 - 37))) ^ b10;
+	b12 += b7; b7 = ((b7 << 20) | (b7 >> (64 - 20))) ^ b12;
+	b1 += k7; b0 += b1 + k6; b1 = ((b1 << 24) | (b1 >> (64 - 24))) ^ b0;
+	b3 += k9; b2 += b3 + k8; b3 = ((b3 << 13) | (b3 >> (64 - 13))) ^ b2;
+	b5 += k11; b4 += b5 + k10; b5 = ((b5 << 8) | (b5 >> (64 - 8))) ^ b4;
+	b7 += k13; b6 += b7 + k12; b7 = ((b7 << 47) | (b7 >> (64 - 47))) ^ b6;
+	b9 += k15; b8 += b9 + k14; b9 = ((b9 << 8) | (b9 >> (64 - 8))) ^ b8;
+	b11 += k0; b10 += b11 + k16; b11 = ((b11 << 17) | (b11 >> (64 - 17))) ^ b10;
+	b13 += k2 + t0; b12 += b13 + k1; b13 = ((b13 << 22) | (b13 >> (64 - 22))) ^ b12;
+	b15 += k4 + 6; b14 += b15 + k3 + t1; b15 = ((b15 << 37) | (b15 >> (64 - 37))) ^ b14;
+	b0 += b9; b9 = ((b9 << 38) | (b9 >> (64 - 38))) ^ b0;
+	b2 += b13; b13 = ((b13 << 19) | (b13 >> (64 - 19))) ^ b2;
+	b6 += b11; b11 = ((b11 << 10) | (b11 >> (64 - 10))) ^ b6;
+	b4 += b15; b15 = ((b15 << 55) | (b15 >> (64 - 55))) ^ b4;
+	b10 += b7; b7 = ((b7 << 49) | (b7 >> (64 - 49))) ^ b10;
+	b12 += b3; b3 = ((b3 << 18) | (b3 >> (64 - 18))) ^ b12;
+	b14 += b5; b5 = ((b5 << 23) | (b5 >> (64 - 23))) ^ b14;
+	b8 += b1; b1 = ((b1 << 52) | (b1 >> (64 - 52))) ^ b8;
+	b0 += b7; b7 = ((b7 << 33) | (b7 >> (64 - 33))) ^ b0;
+	b2 += b5; b5 = ((b5 << 4) | (b5 >> (64 - 4))) ^ b2;
+	b4 += b3; b3 = ((b3 << 51) | (b3 >> (64 - 51))) ^ b4;
+	b6 += b1; b1 = ((b1 << 13) | (b1 >> (64 - 13))) ^ b6;
+	b12 += b15; b15 = ((b15 << 34) | (b15 >> (64 - 34))) ^ b12;
+	b14 += b13; b13 = ((b13 << 41) | (b13 >> (64 - 41))) ^ b14;
+	b8 += b11; b11 = ((b11 << 59) | (b11 >> (64 - 59))) ^ b8;
+	b10 += b9; b9 = ((b9 << 17) | (b9 >> (64 - 17))) ^ b10;
+	b0 += b15; b15 = ((b15 << 5) | (b15 >> (64 - 5))) ^ b0;
+	b2 += b11; b11 = ((b11 << 20) | (b11 >> (64 - 20))) ^ b2;
+	b6 += b13; b13 = ((b13 << 48) | (b13 >> (64 - 48))) ^ b6;
+	b4 += b9; b9 = ((b9 << 41) | (b9 >> (64 - 41))) ^ b4;
+	b14 += b1; b1 = ((b1 << 47) | (b1 >> (64 - 47))) ^ b14;
+	b8 += b5; b5 = ((b5 << 28) | (b5 >> (64 - 28))) ^ b8;
+	b10 += b3; b3 = ((b3 << 16) | (b3 >> (64 - 16))) ^ b10;
+	b12 += b7; b7 = ((b7 << 25) | (b7 >> (64 - 25))) ^ b12;
+	b1 += k8; b0 += b1 + k7; b1 = ((b1 << 41) | (b1 >> (64 - 41))) ^ b0;
+	b3 += k10; b2 += b3 + k9; b3 = ((b3 << 9) | (b3 >> (64 - 9))) ^ b2;
+	b5 += k12; b4 += b5 + k11; b5 = ((b5 << 37) | (b5 >> (64 - 37))) ^ b4;
+	b7 += k14; b6 += b7 + k13; b7 = ((b7 << 31) | (b7 >> (64 - 31))) ^ b6;
+	b9 += k16; b8 += b9 + k15; b9 = ((b9 << 12) | (b9 >> (64 - 12))) ^ b8;
+	b11 += k1; b10 += b11 + k0; b11 = ((b11 << 47) | (b11 >> (64 - 47))) ^ b10;
+	b13 += k3 + t1; b12 += b13 + k2; b13 = ((b13 << 44) | (b13 >> (64 - 44))) ^ b12;
+	b15 += k5 + 7; b14 += b15 + k4 + t2; b15 = ((b15 << 30) | (b15 >> (64 - 30))) ^ b14;
+	b0 += b9; b9 = ((b9 << 16) | (b9 >> (64 - 16))) ^ b0;
+	b2 += b13; b13 = ((b13 << 34) | (b13 >> (64 - 34))) ^ b2;
+	b6 += b11; b11 = ((b11 << 56) | (b11 >> (64 - 56))) ^ b6;
+	b4 += b15; b15 = ((b15 << 51) | (b15 >> (64 - 51))) ^ b4;
+	b10 += b7; b7 = ((b7 << 4) | (b7 >> (64 - 4))) ^ b10;
+	b12 += b3; b3 = ((b3 << 53) | (b3 >> (64 - 53))) ^ b12;
+	b14 += b5; b5 = ((b5 << 42) | (b5 >> (64 - 42))) ^ b14;
+	b8 += b1; b1 = ((b1 << 41) | (b1 >> (64 - 41))) ^ b8;
+	b0 += b7; b7 = ((b7 << 31) | (b7 >> (64 - 31))) ^ b0;
+	b2 += b5; b5 = ((b5 << 44) | (b5 >> (64 - 44))) ^ b2;
+	b4 += b3; b3 = ((b3 << 47) | (b3 >> (64 - 47))) ^ b4;
+	b6 += b1; b1 = ((b1 << 46) | (b1 >> (64 - 46))) ^ b6;
+	b12 += b15; b15 = ((b15 << 19) | (b15 >> (64 - 19))) ^ b12;
+	b14 += b13; b13 = ((b13 << 42) | (b13 >> (64 - 42))) ^ b14;
+	b8 += b11; b11 = ((b11 << 44) | (b11 >> (64 - 44))) ^ b8;
+	b10 += b9; b9 = ((b9 << 25) | (b9 >> (64 - 25))) ^ b10;
+	b0 += b15; b15 = ((b15 << 9) | (b15 >> (64 - 9))) ^ b0;
+	b2 += b11; b11 = ((b11 << 48) | (b11 >> (64 - 48))) ^ b2;
+	b6 += b13; b13 = ((b13 << 35) | (b13 >> (64 - 35))) ^ b6;
+	b4 += b9; b9 = ((b9 << 52) | (b9 >> (64 - 52))) ^ b4;
+	b14 += b1; b1 = ((b1 << 23) | (b1 >> (64 - 23))) ^ b14;
+	b8 += b5; b5 = ((b5 << 31) | (b5 >> (64 - 31))) ^ b8;
+	b10 += b3; b3 = ((b3 << 37) | (b3 >> (64 - 37))) ^ b10;
+	b12 += b7; b7 = ((b7 << 20) | (b7 >> (64 - 20))) ^ b12;
+	b1 += k9; b0 += b1 + k8; b1 = ((b1 << 24) | (b1 >> (64 - 24))) ^ b0;
+	b3 += k11; b2 += b3 + k10; b3 = ((b3 << 13) | (b3 >> (64 - 13))) ^ b2;
+	b5 += k13; b4 += b5 + k12; b5 = ((b5 << 8) | (b5 >> (64 - 8))) ^ b4;
+	b7 += k15; b6 += b7 + k14; b7 = ((b7 << 47) | (b7 >> (64 - 47))) ^ b6;
+	b9 += k0; b8 += b9 + k16; b9 = ((b9 << 8) | (b9 >> (64 - 8))) ^ b8;
+	b11 += k2; b10 += b11 + k1; b11 = ((b11 << 17) | (b11 >> (64 - 17))) ^ b10;
+	b13 += k4 + t2; b12 += b13 + k3; b13 = ((b13 << 22) | (b13 >> (64 - 22))) ^ b12;
+	b15 += k6 + 8; b14 += b15 + k5 + t0; b15 = ((b15 << 37) | (b15 >> (64 - 37))) ^ b14;
+	b0 += b9; b9 = ((b9 << 38) | (b9 >> (64 - 38))) ^ b0;
+	b2 += b13; b13 = ((b13 << 19) | (b13 >> (64 - 19))) ^ b2;
+	b6 += b11; b11 = ((b11 << 10) | (b11 >> (64 - 10))) ^ b6;
+	b4 += b15; b15 = ((b15 << 55) | (b15 >> (64 - 55))) ^ b4;
+	b10 += b7; b7 = ((b7 << 49) | (b7 >> (64 - 49))) ^ b10;
+	b12 += b3; b3 = ((b3 << 18) | (b3 >> (64 - 18))) ^ b12;
+	b14 += b5; b5 = ((b5 << 23) | (b5 >> (64 - 23))) ^ b14;
+	b8 += b1; b1 = ((b1 << 52) | (b1 >> (64 - 52))) ^ b8;
+	b0 += b7; b7 = ((b7 << 33) | (b7 >> (64 - 33))) ^ b0;
+	b2 += b5; b5 = ((b5 << 4) | (b5 >> (64 - 4))) ^ b2;
+	b4 += b3; b3 = ((b3 << 51) | (b3 >> (64 - 51))) ^ b4;
+	b6 += b1; b1 = ((b1 << 13) | (b1 >> (64 - 13))) ^ b6;
+	b12 += b15; b15 = ((b15 << 34) | (b15 >> (64 - 34))) ^ b12;
+	b14 += b13; b13 = ((b13 << 41) | (b13 >> (64 - 41))) ^ b14;
+	b8 += b11; b11 = ((b11 << 59) | (b11 >> (64 - 59))) ^ b8;
+	b10 += b9; b9 = ((b9 << 17) | (b9 >> (64 - 17))) ^ b10;
+	b0 += b15; b15 = ((b15 << 5) | (b15 >> (64 - 5))) ^ b0;
+	b2 += b11; b11 = ((b11 << 20) | (b11 >> (64 - 20))) ^ b2;
+	b6 += b13; b13 = ((b13 << 48) | (b13 >> (64 - 48))) ^ b6;
+	b4 += b9; b9 = ((b9 << 41) | (b9 >> (64 - 41))) ^ b4;
+	b14 += b1; b1 = ((b1 << 47) | (b1 >> (64 - 47))) ^ b14;
+	b8 += b5; b5 = ((b5 << 28) | (b5 >> (64 - 28))) ^ b8;
+	b10 += b3; b3 = ((b3 << 16) | (b3 >> (64 - 16))) ^ b10;
+	b12 += b7; b7 = ((b7 << 25) | (b7 >> (64 - 25))) ^ b12;
+	b1 += k10; b0 += b1 + k9; b1 = ((b1 << 41) | (b1 >> (64 - 41))) ^ b0;
+	b3 += k12; b2 += b3 + k11; b3 = ((b3 << 9) | (b3 >> (64 - 9))) ^ b2;
+	b5 += k14; b4 += b5 + k13; b5 = ((b5 << 37) | (b5 >> (64 - 37))) ^ b4;
+	b7 += k16; b6 += b7 + k15; b7 = ((b7 << 31) | (b7 >> (64 - 31))) ^ b6;
+	b9 += k1; b8 += b9 + k0; b9 = ((b9 << 12) | (b9 >> (64 - 12))) ^ b8;
+	b11 += k3; b10 += b11 + k2; b11 = ((b11 << 47) | (b11 >> (64 - 47))) ^ b10;
+	b13 += k5 + t0; b12 += b13 + k4; b13 = ((b13 << 44) | (b13 >> (64 - 44))) ^ b12;
+	b15 += k7 + 9; b14 += b15 + k6 + t1; b15 = ((b15 << 30) | (b15 >> (64 - 30))) ^ b14;
+	b0 += b9; b9 = ((b9 << 16) | (b9 >> (64 - 16))) ^ b0;
+	b2 += b13; b13 = ((b13 << 34) | (b13 >> (64 - 34))) ^ b2;
+	b6 += b11; b11 = ((b11 << 56) | (b11 >> (64 - 56))) ^ b6;
+	b4 += b15; b15 = ((b15 << 51) | (b15 >> (64 - 51))) ^ b4;
+	b10 += b7; b7 = ((b7 << 4) | (b7 >> (64 - 4))) ^ b10;
+	b12 += b3; b3 = ((b3 << 53) | (b3 >> (64 - 53))) ^ b12;
+	b14 += b5; b5 = ((b5 << 42) | (b5 >> (64 - 42))) ^ b14;
+	b8 += b1; b1 = ((b1 << 41) | (b1 >> (64 - 41))) ^ b8;
+	b0 += b7; b7 = ((b7 << 31) | (b7 >> (64 - 31))) ^ b0;
+	b2 += b5; b5 = ((b5 << 44) | (b5 >> (64 - 44))) ^ b2;
+	b4 += b3; b3 = ((b3 << 47) | (b3 >> (64 - 47))) ^ b4;
+	b6 += b1; b1 = ((b1 << 46) | (b1 >> (64 - 46))) ^ b6;
+	b12 += b15; b15 = ((b15 << 19) | (b15 >> (64 - 19))) ^ b12;
+	b14 += b13; b13 = ((b13 << 42) | (b13 >> (64 - 42))) ^ b14;
+	b8 += b11; b11 = ((b11 << 44) | (b11 >> (64 - 44))) ^ b8;
+	b10 += b9; b9 = ((b9 << 25) | (b9 >> (64 - 25))) ^ b10;
+	b0 += b15; b15 = ((b15 << 9) | (b15 >> (64 - 9))) ^ b0;
+	b2 += b11; b11 = ((b11 << 48) | (b11 >> (64 - 48))) ^ b2;
+	b6 += b13; b13 = ((b13 << 35) | (b13 >> (64 - 35))) ^ b6;
+	b4 += b9; b9 = ((b9 << 52) | (b9 >> (64 - 52))) ^ b4;
+	b14 += b1; b1 = ((b1 << 23) | (b1 >> (64 - 23))) ^ b14;
+	b8 += b5; b5 = ((b5 << 31) | (b5 >> (64 - 31))) ^ b8;
+	b10 += b3; b3 = ((b3 << 37) | (b3 >> (64 - 37))) ^ b10;
+	b12 += b7; b7 = ((b7 << 20) | (b7 >> (64 - 20))) ^ b12;
+	b1 += k11; b0 += b1 + k10; b1 = ((b1 << 24) | (b1 >> (64 - 24))) ^ b0;
+	b3 += k13; b2 += b3 + k12; b3 = ((b3 << 13) | (b3 >> (64 - 13))) ^ b2;
+	b5 += k15; b4 += b5 + k14; b5 = ((b5 << 8) | (b5 >> (64 - 8))) ^ b4;
+	b7 += k0; b6 += b7 + k16; b7 = ((b7 << 47) | (b7 >> (64 - 47))) ^ b6;
+	b9 += k2; b8 += b9 + k1; b9 = ((b9 << 8) | (b9 >> (64 - 8))) ^ b8;
+	b11 += k4; b10 += b11 + k3; b11 = ((b11 << 17) | (b11 >> (64 - 17))) ^ b10;
+	b13 += k6 + t1; b12 += b13 + k5; b13 = ((b13 << 22) | (b13 >> (64 - 22))) ^ b12;
+	b15 += k8 + 10; b14 += b15 + k7 + t2; b15 = ((b15 << 37) | (b15 >> (64 - 37))) ^ b14;
+	b0 += b9; b9 = ((b9 << 38) | (b9 >> (64 - 38))) ^ b0;
+	b2 += b13; b13 = ((b13 << 19) | (b13 >> (64 - 19))) ^ b2;
+	b6 += b11; b11 = ((b11 << 10) | (b11 >> (64 - 10))) ^ b6;
+	b4 += b15; b15 = ((b15 << 55) | (b15 >> (64 - 55))) ^ b4;
+	b10 += b7; b7 = ((b7 << 49) | (b7 >> (64 - 49))) ^ b10;
+	b12 += b3; b3 = ((b3 << 18) | (b3 >> (64 - 18))) ^ b12;
+	b14 += b5; b5 = ((b5 << 23) | (b5 >> (64 - 23))) ^ b14;
+	b8 += b1; b1 = ((b1 << 52) | (b1 >> (64 - 52))) ^ b8;
+	b0 += b7; b7 = ((b7 << 33) | (b7 >> (64 - 33))) ^ b0;
+	b2 += b5; b5 = ((b5 << 4) | (b5 >> (64 - 4))) ^ b2;
+	b4 += b3; b3 = ((b3 << 51) | (b3 >> (64 - 51))) ^ b4;
+	b6 += b1; b1 = ((b1 << 13) | (b1 >> (64 - 13))) ^ b6;
+	b12 += b15; b15 = ((b15 << 34) | (b15 >> (64 - 34))) ^ b12;
+	b14 += b13; b13 = ((b13 << 41) | (b13 >> (64 - 41))) ^ b14;
+	b8 += b11; b11 = ((b11 << 59) | (b11 >> (64 - 59))) ^ b8;
+	b10 += b9; b9 = ((b9 << 17) | (b9 >> (64 - 17))) ^ b10;
+	b0 += b15; b15 = ((b15 << 5) | (b15 >> (64 - 5))) ^ b0;
+	b2 += b11; b11 = ((b11 << 20) | (b11 >> (64 - 20))) ^ b2;
+	b6 += b13; b13 = ((b13 << 48) | (b13 >> (64 - 48))) ^ b6;
+	b4 += b9; b9 = ((b9 << 41) | (b9 >> (64 - 41))) ^ b4;
+	b14 += b1; b1 = ((b1 << 47) | (b1 >> (64 - 47))) ^ b14;
+	b8 += b5; b5 = ((b5 << 28) | (b5 >> (64 - 28))) ^ b8;
+	b10 += b3; b3 = ((b3 << 16) | (b3 >> (64 - 16))) ^ b10;
+	b12 += b7; b7 = ((b7 << 25) | (b7 >> (64 - 25))) ^ b12;
+	b1 += k12; b0 += b1 + k11; b1 = ((b1 << 41) | (b1 >> (64 - 41))) ^ b0;
+	b3 += k14; b2 += b3 + k13; b3 = ((b3 << 9) | (b3 >> (64 - 9))) ^ b2;
+	b5 += k16; b4 += b5 + k15; b5 = ((b5 << 37) | (b5 >> (64 - 37))) ^ b4;
+	b7 += k1; b6 += b7 + k0; b7 = ((b7 << 31) | (b7 >> (64 - 31))) ^ b6;
+	b9 += k3; b8 += b9 + k2; b9 = ((b9 << 12) | (b9 >> (64 - 12))) ^ b8;
+	b11 += k5; b10 += b11 + k4; b11 = ((b11 << 47) | (b11 >> (64 - 47))) ^ b10;
+	b13 += k7 + t2; b12 += b13 + k6; b13 = ((b13 << 44) | (b13 >> (64 - 44))) ^ b12;
+	b15 += k9 + 11; b14 += b15 + k8 + t0; b15 = ((b15 << 30) | (b15 >> (64 - 30))) ^ b14;
+	b0 += b9; b9 = ((b9 << 16) | (b9 >> (64 - 16))) ^ b0;
+	b2 += b13; b13 = ((b13 << 34) | (b13 >> (64 - 34))) ^ b2;
+	b6 += b11; b11 = ((b11 << 56) | (b11 >> (64 - 56))) ^ b6;
+	b4 += b15; b15 = ((b15 << 51) | (b15 >> (64 - 51))) ^ b4;
+	b10 += b7; b7 = ((b7 << 4) | (b7 >> (64 - 4))) ^ b10;
+	b12 += b3; b3 = ((b3 << 53) | (b3 >> (64 - 53))) ^ b12;
+	b14 += b5; b5 = ((b5 << 42) | (b5 >> (64 - 42))) ^ b14;
+	b8 += b1; b1 = ((b1 << 41) | (b1 >> (64 - 41))) ^ b8;
+	b0 += b7; b7 = ((b7 << 31) | (b7 >> (64 - 31))) ^ b0;
+	b2 += b5; b5 = ((b5 << 44) | (b5 >> (64 - 44))) ^ b2;
+	b4 += b3; b3 = ((b3 << 47) | (b3 >> (64 - 47))) ^ b4;
+	b6 += b1; b1 = ((b1 << 46) | (b1 >> (64 - 46))) ^ b6;
+	b12 += b15; b15 = ((b15 << 19) | (b15 >> (64 - 19))) ^ b12;
+	b14 += b13; b13 = ((b13 << 42) | (b13 >> (64 - 42))) ^ b14;
+	b8 += b11; b11 = ((b11 << 44) | (b11 >> (64 - 44))) ^ b8;
+	b10 += b9; b9 = ((b9 << 25) | (b9 >> (64 - 25))) ^ b10;
+	b0 += b15; b15 = ((b15 << 9) | (b15 >> (64 - 9))) ^ b0;
+	b2 += b11; b11 = ((b11 << 48) | (b11 >> (64 - 48))) ^ b2;
+	b6 += b13; b13 = ((b13 << 35) | (b13 >> (64 - 35))) ^ b6;
+	b4 += b9; b9 = ((b9 << 52) | (b9 >> (64 - 52))) ^ b4;
+	b14 += b1; b1 = ((b1 << 23) | (b1 >> (64 - 23))) ^ b14;
+	b8 += b5; b5 = ((b5 << 31) | (b5 >> (64 - 31))) ^ b8;
+	b10 += b3; b3 = ((b3 << 37) | (b3 >> (64 - 37))) ^ b10;
+	b12 += b7; b7 = ((b7 << 20) | (b7 >> (64 - 20))) ^ b12;
+	b1 += k13; b0 += b1 + k12; b1 = ((b1 << 24) | (b1 >> (64 - 24))) ^ b0;
+	b3 += k15; b2 += b3 + k14; b3 = ((b3 << 13) | (b3 >> (64 - 13))) ^ b2;
+	b5 += k0; b4 += b5 + k16; b5 = ((b5 << 8) | (b5 >> (64 - 8))) ^ b4;
+	b7 += k2; b6 += b7 + k1; b7 = ((b7 << 47) | (b7 >> (64 - 47))) ^ b6;
+	b9 += k4; b8 += b9 + k3; b9 = ((b9 << 8) | (b9 >> (64 - 8))) ^ b8;
+	b11 += k6; b10 += b11 + k5; b11 = ((b11 << 17) | (b11 >> (64 - 17))) ^ b10;
+	b13 += k8 + t0; b12 += b13 + k7; b13 = ((b13 << 22) | (b13 >> (64 - 22))) ^ b12;
+	b15 += k10 + 12; b14 += b15 + k9 + t1; b15 = ((b15 << 37) | (b15 >> (64 - 37))) ^ b14;
+	b0 += b9; b9 = ((b9 << 38) | (b9 >> (64 - 38))) ^ b0;
+	b2 += b13; b13 = ((b13 << 19) | (b13 >> (64 - 19))) ^ b2;
+	b6 += b11; b11 = ((b11 << 10) | (b11 >> (64 - 10))) ^ b6;
+	b4 += b15; b15 = ((b15 << 55) | (b15 >> (64 - 55))) ^ b4;
+	b10 += b7; b7 = ((b7 << 49) | (b7 >> (64 - 49))) ^ b10;
+	b12 += b3; b3 = ((b3 << 18) | (b3 >> (64 - 18))) ^ b12;
+	b14 += b5; b5 = ((b5 << 23) | (b5 >> (64 - 23))) ^ b14;
+	b8 += b1; b1 = ((b1 << 52) | (b1 >> (64 - 52))) ^ b8;
+	b0 += b7; b7 = ((b7 << 33) | (b7 >> (64 - 33))) ^ b0;
+	b2 += b5; b5 = ((b5 << 4) | (b5 >> (64 - 4))) ^ b2;
+	b4 += b3; b3 = ((b3 << 51) | (b3 >> (64 - 51))) ^ b4;
+	b6 += b1; b1 = ((b1 << 13) | (b1 >> (64 - 13))) ^ b6;
+	b12 += b15; b15 = ((b15 << 34) | (b15 >> (64 - 34))) ^ b12;
+	b14 += b13; b13 = ((b13 << 41) | (b13 >> (64 - 41))) ^ b14;
+	b8 += b11; b11 = ((b11 << 59) | (b11 >> (64 - 59))) ^ b8;
+	b10 += b9; b9 = ((b9 << 17) | (b9 >> (64 - 17))) ^ b10;
+	b0 += b15; b15 = ((b15 << 5) | (b15 >> (64 - 5))) ^ b0;
+	b2 += b11; b11 = ((b11 << 20) | (b11 >> (64 - 20))) ^ b2;
+	b6 += b13; b13 = ((b13 << 48) | (b13 >> (64 - 48))) ^ b6;
+	b4 += b9; b9 = ((b9 << 41) | (b9 >> (64 - 41))) ^ b4;
+	b14 += b1; b1 = ((b1 << 47) | (b1 >> (64 - 47))) ^ b14;
+	b8 += b5; b5 = ((b5 << 28) | (b5 >> (64 - 28))) ^ b8;
+	b10 += b3; b3 = ((b3 << 16) | (b3 >> (64 - 16))) ^ b10;
+	b12 += b7; b7 = ((b7 << 25) | (b7 >> (64 - 25))) ^ b12;
+	b1 += k14; b0 += b1 + k13; b1 = ((b1 << 41) | (b1 >> (64 - 41))) ^ b0;
+	b3 += k16; b2 += b3 + k15; b3 = ((b3 << 9) | (b3 >> (64 - 9))) ^ b2;
+	b5 += k1; b4 += b5 + k0; b5 = ((b5 << 37) | (b5 >> (64 - 37))) ^ b4;
+	b7 += k3; b6 += b7 + k2; b7 = ((b7 << 31) | (b7 >> (64 - 31))) ^ b6;
+	b9 += k5; b8 += b9 + k4; b9 = ((b9 << 12) | (b9 >> (64 - 12))) ^ b8;
+	b11 += k7; b10 += b11 + k6; b11 = ((b11 << 47) | (b11 >> (64 - 47))) ^ b10;
+	b13 += k9 + t1; b12 += b13 + k8; b13 = ((b13 << 44) | (b13 >> (64 - 44))) ^ b12;
+	b15 += k11 + 13; b14 += b15 + k10 + t2; b15 = ((b15 << 30) | (b15 >> (64 - 30))) ^ b14;
+	b0 += b9; b9 = ((b9 << 16) | (b9 >> (64 - 16))) ^ b0;
+	b2 += b13; b13 = ((b13 << 34) | (b13 >> (64 - 34))) ^ b2;
+	b6 += b11; b11 = ((b11 << 56) | (b11 >> (64 - 56))) ^ b6;
+	b4 += b15; b15 = ((b15 << 51) | (b15 >> (64 - 51))) ^ b4;
+	b10 += b7; b7 = ((b7 << 4) | (b7 >> (64 - 4))) ^ b10;
+	b12 += b3; b3 = ((b3 << 53) | (b3 >> (64 - 53))) ^ b12;
+	b14 += b5; b5 = ((b5 << 42) | (b5 >> (64 - 42))) ^ b14;
+	b8 += b1; b1 = ((b1 << 41) | (b1 >> (64 - 41))) ^ b8;
+	b0 += b7; b7 = ((b7 << 31) | (b7 >> (64 - 31))) ^ b0;
+	b2 += b5; b5 = ((b5 << 44) | (b5 >> (64 - 44))) ^ b2;
+	b4 += b3; b3 = ((b3 << 47) | (b3 >> (64 - 47))) ^ b4;
+	b6 += b1; b1 = ((b1 << 46) | (b1 >> (64 - 46))) ^ b6;
+	b12 += b15; b15 = ((b15 << 19) | (b15 >> (64 - 19))) ^ b12;
+	b14 += b13; b13 = ((b13 << 42) | (b13 >> (64 - 42))) ^ b14;
+	b8 += b11; b11 = ((b11 << 44) | (b11 >> (64 - 44))) ^ b8;
+	b10 += b9; b9 = ((b9 << 25) | (b9 >> (64 - 25))) ^ b10;
+	b0 += b15; b15 = ((b15 << 9) | (b15 >> (64 - 9))) ^ b0;
+	b2 += b11; b11 = ((b11 << 48) | (b11 >> (64 - 48))) ^ b2;
+	b6 += b13; b13 = ((b13 << 35) | (b13 >> (64 - 35))) ^ b6;
+	b4 += b9; b9 = ((b9 << 52) | (b9 >> (64 - 52))) ^ b4;
+	b14 += b1; b1 = ((b1 << 23) | (b1 >> (64 - 23))) ^ b14;
+	b8 += b5; b5 = ((b5 << 31) | (b5 >> (64 - 31))) ^ b8;
+	b10 += b3; b3 = ((b3 << 37) | (b3 >> (64 - 37))) ^ b10;
+	b12 += b7; b7 = ((b7 << 20) | (b7 >> (64 - 20))) ^ b12;
+	b1 += k15; b0 += b1 + k14; b1 = ((b1 << 24) | (b1 >> (64 - 24))) ^ b0;
+	b3 += k0; b2 += b3 + k16; b3 = ((b3 << 13) | (b3 >> (64 - 13))) ^ b2;
+	b5 += k2; b4 += b5 + k1; b5 = ((b5 << 8) | (b5 >> (64 - 8))) ^ b4;
+	b7 += k4; b6 += b7 + k3; b7 = ((b7 << 47) | (b7 >> (64 - 47))) ^ b6;
+	b9 += k6; b8 += b9 + k5; b9 = ((b9 << 8) | (b9 >> (64 - 8))) ^ b8;
+	b11 += k8; b10 += b11 + k7; b11 = ((b11 << 17) | (b11 >> (64 - 17))) ^ b10;
+	b13 += k10 + t2; b12 += b13 + k9; b13 = ((b13 << 22) | (b13 >> (64 - 22))) ^ b12;
+	b15 += k12 + 14; b14 += b15 + k11 + t0; b15 = ((b15 << 37) | (b15 >> (64 - 37))) ^ b14;
+	b0 += b9; b9 = ((b9 << 38) | (b9 >> (64 - 38))) ^ b0;
+	b2 += b13; b13 = ((b13 << 19) | (b13 >> (64 - 19))) ^ b2;
+	b6 += b11; b11 = ((b11 << 10) | (b11 >> (64 - 10))) ^ b6;
+	b4 += b15; b15 = ((b15 << 55) | (b15 >> (64 - 55))) ^ b4;
+	b10 += b7; b7 = ((b7 << 49) | (b7 >> (64 - 49))) ^ b10;
+	b12 += b3; b3 = ((b3 << 18) | (b3 >> (64 - 18))) ^ b12;
+	b14 += b5; b5 = ((b5 << 23) | (b5 >> (64 - 23))) ^ b14;
+	b8 += b1; b1 = ((b1 << 52) | (b1 >> (64 - 52))) ^ b8;
+	b0 += b7; b7 = ((b7 << 33) | (b7 >> (64 - 33))) ^ b0;
+	b2 += b5; b5 = ((b5 << 4) | (b5 >> (64 - 4))) ^ b2;
+	b4 += b3; b3 = ((b3 << 51) | (b3 >> (64 - 51))) ^ b4;
+	b6 += b1; b1 = ((b1 << 13) | (b1 >> (64 - 13))) ^ b6;
+	b12 += b15; b15 = ((b15 << 34) | (b15 >> (64 - 34))) ^ b12;
+	b14 += b13; b13 = ((b13 << 41) | (b13 >> (64 - 41))) ^ b14;
+	b8 += b11; b11 = ((b11 << 59) | (b11 >> (64 - 59))) ^ b8;
+	b10 += b9; b9 = ((b9 << 17) | (b9 >> (64 - 17))) ^ b10;
+	b0 += b15; b15 = ((b15 << 5) | (b15 >> (64 - 5))) ^ b0;
+	b2 += b11; b11 = ((b11 << 20) | (b11 >> (64 - 20))) ^ b2;
+	b6 += b13; b13 = ((b13 << 48) | (b13 >> (64 - 48))) ^ b6;
+	b4 += b9; b9 = ((b9 << 41) | (b9 >> (64 - 41))) ^ b4;
+	b14 += b1; b1 = ((b1 << 47) | (b1 >> (64 - 47))) ^ b14;
+	b8 += b5; b5 = ((b5 << 28) | (b5 >> (64 - 28))) ^ b8;
+	b10 += b3; b3 = ((b3 << 16) | (b3 >> (64 - 16))) ^ b10;
+	b12 += b7; b7 = ((b7 << 25) | (b7 >> (64 - 25))) ^ b12;
+	b1 += k16; b0 += b1 + k15; b1 = ((b1 << 41) | (b1 >> (64 - 41))) ^ b0;
+	b3 += k1; b2 += b3 + k0; b3 = ((b3 << 9) | (b3 >> (64 - 9))) ^ b2;
+	b5 += k3; b4 += b5 + k2; b5 = ((b5 << 37) | (b5 >> (64 - 37))) ^ b4;
+	b7 += k5; b6 += b7 + k4; b7 = ((b7 << 31) | (b7 >> (64 - 31))) ^ b6;
+	b9 += k7; b8 += b9 + k6; b9 = ((b9 << 12) | (b9 >> (64 - 12))) ^ b8;
+	b11 += k9; b10 += b11 + k8; b11 = ((b11 << 47) | (b11 >> (64 - 47))) ^ b10;
+	b13 += k11 + t0; b12 += b13 + k10; b13 = ((b13 << 44) | (b13 >> (64 - 44))) ^ b12;
+	b15 += k13 + 15; b14 += b15 + k12 + t1; b15 = ((b15 << 30) | (b15 >> (64 - 30))) ^ b14;
+	b0 += b9; b9 = ((b9 << 16) | (b9 >> (64 - 16))) ^ b0;
+	b2 += b13; b13 = ((b13 << 34) | (b13 >> (64 - 34))) ^ b2;
+	b6 += b11; b11 = ((b11 << 56) | (b11 >> (64 - 56))) ^ b6;
+	b4 += b15; b15 = ((b15 << 51) | (b15 >> (64 - 51))) ^ b4;
+	b10 += b7; b7 = ((b7 << 4) | (b7 >> (64 - 4))) ^ b10;
+	b12 += b3; b3 = ((b3 << 53) | (b3 >> (64 - 53))) ^ b12;
+	b14 += b5; b5 = ((b5 << 42) | (b5 >> (64 - 42))) ^ b14;
+	b8 += b1; b1 = ((b1 << 41) | (b1 >> (64 - 41))) ^ b8;
+	b0 += b7; b7 = ((b7 << 31) | (b7 >> (64 - 31))) ^ b0;
+	b2 += b5; b5 = ((b5 << 44) | (b5 >> (64 - 44))) ^ b2;
+	b4 += b3; b3 = ((b3 << 47) | (b3 >> (64 - 47))) ^ b4;
+	b6 += b1; b1 = ((b1 << 46) | (b1 >> (64 - 46))) ^ b6;
+	b12 += b15; b15 = ((b15 << 19) | (b15 >> (64 - 19))) ^ b12;
+	b14 += b13; b13 = ((b13 << 42) | (b13 >> (64 - 42))) ^ b14;
+	b8 += b11; b11 = ((b11 << 44) | (b11 >> (64 - 44))) ^ b8;
+	b10 += b9; b9 = ((b9 << 25) | (b9 >> (64 - 25))) ^ b10;
+	b0 += b15; b15 = ((b15 << 9) | (b15 >> (64 - 9))) ^ b0;
+	b2 += b11; b11 = ((b11 << 48) | (b11 >> (64 - 48))) ^ b2;
+	b6 += b13; b13 = ((b13 << 35) | (b13 >> (64 - 35))) ^ b6;
+	b4 += b9; b9 = ((b9 << 52) | (b9 >> (64 - 52))) ^ b4;
+	b14 += b1; b1 = ((b1 << 23) | (b1 >> (64 - 23))) ^ b14;
+	b8 += b5; b5 = ((b5 << 31) | (b5 >> (64 - 31))) ^ b8;
+	b10 += b3; b3 = ((b3 << 37) | (b3 >> (64 - 37))) ^ b10;
+	b12 += b7; b7 = ((b7 << 20) | (b7 >> (64 - 20))) ^ b12;
+	b1 += k0; b0 += b1 + k16; b1 = ((b1 << 24) | (b1 >> (64 - 24))) ^ b0;
+	b3 += k2; b2 += b3 + k1; b3 = ((b3 << 13) | (b3 >> (64 - 13))) ^ b2;
+	b5 += k4; b4 += b5 + k3; b5 = ((b5 << 8) | (b5 >> (64 - 8))) ^ b4;
+	b7 += k6; b6 += b7 + k5; b7 = ((b7 << 47) | (b7 >> (64 - 47))) ^ b6;
+	b9 += k8; b8 += b9 + k7; b9 = ((b9 << 8) | (b9 >> (64 - 8))) ^ b8;
+	b11 += k10; b10 += b11 + k9; b11 = ((b11 << 17) | (b11 >> (64 - 17))) ^ b10;
+	b13 += k12 + t1; b12 += b13 + k11; b13 = ((b13 << 22) | (b13 >> (64 - 22))) ^ b12;
+	b15 += k14 + 16; b14 += b15 + k13 + t2; b15 = ((b15 << 37) | (b15 >> (64 - 37))) ^ b14;
+	b0 += b9; b9 = ((b9 << 38) | (b9 >> (64 - 38))) ^ b0;
+	b2 += b13; b13 = ((b13 << 19) | (b13 >> (64 - 19))) ^ b2;
+	b6 += b11; b11 = ((b11 << 10) | (b11 >> (64 - 10))) ^ b6;
+	b4 += b15; b15 = ((b15 << 55) | (b15 >> (64 - 55))) ^ b4;
+	b10 += b7; b7 = ((b7 << 49) | (b7 >> (64 - 49))) ^ b10;
+	b12 += b3; b3 = ((b3 << 18) | (b3 >> (64 - 18))) ^ b12;
+	b14 += b5; b5 = ((b5 << 23) | (b5 >> (64 - 23))) ^ b14;
+	b8 += b1; b1 = ((b1 << 52) | (b1 >> (64 - 52))) ^ b8;
+	b0 += b7; b7 = ((b7 << 33) | (b7 >> (64 - 33))) ^ b0;
+	b2 += b5; b5 = ((b5 << 4) | (b5 >> (64 - 4))) ^ b2;
+	b4 += b3; b3 = ((b3 << 51) | (b3 >> (64 - 51))) ^ b4;
+	b6 += b1; b1 = ((b1 << 13) | (b1 >> (64 - 13))) ^ b6;
+	b12 += b15; b15 = ((b15 << 34) | (b15 >> (64 - 34))) ^ b12;
+	b14 += b13; b13 = ((b13 << 41) | (b13 >> (64 - 41))) ^ b14;
+	b8 += b11; b11 = ((b11 << 59) | (b11 >> (64 - 59))) ^ b8;
+	b10 += b9; b9 = ((b9 << 17) | (b9 >> (64 - 17))) ^ b10;
+	b0 += b15; b15 = ((b15 << 5) | (b15 >> (64 - 5))) ^ b0;
+	b2 += b11; b11 = ((b11 << 20) | (b11 >> (64 - 20))) ^ b2;
+	b6 += b13; b13 = ((b13 << 48) | (b13 >> (64 - 48))) ^ b6;
+	b4 += b9; b9 = ((b9 << 41) | (b9 >> (64 - 41))) ^ b4;
+	b14 += b1; b1 = ((b1 << 47) | (b1 >> (64 - 47))) ^ b14;
+	b8 += b5; b5 = ((b5 << 28) | (b5 >> (64 - 28))) ^ b8;
+	b10 += b3; b3 = ((b3 << 16) | (b3 >> (64 - 16))) ^ b10;
+	b12 += b7; b7 = ((b7 << 25) | (b7 >> (64 - 25))) ^ b12;
+	b1 += k1; b0 += b1 + k0; b1 = ((b1 << 41) | (b1 >> (64 - 41))) ^ b0;
+	b3 += k3; b2 += b3 + k2; b3 = ((b3 << 9) | (b3 >> (64 - 9))) ^ b2;
+	b5 += k5; b4 += b5 + k4; b5 = ((b5 << 37) | (b5 >> (64 - 37))) ^ b4;
+	b7 += k7; b6 += b7 + k6; b7 = ((b7 << 31) | (b7 >> (64 - 31))) ^ b6;
+	b9 += k9; b8 += b9 + k8; b9 = ((b9 << 12) | (b9 >> (64 - 12))) ^ b8;
+	b11 += k11; b10 += b11 + k10; b11 = ((b11 << 47) | (b11 >> (64 - 47))) ^ b10;
+	b13 += k13 + t2; b12 += b13 + k12; b13 = ((b13 << 44) | (b13 >> (64 - 44))) ^ b12;
+	b15 += k15 + 17; b14 += b15 + k14 + t0; b15 = ((b15 << 30) | (b15 >> (64 - 30))) ^ b14;
+	b0 += b9; b9 = ((b9 << 16) | (b9 >> (64 - 16))) ^ b0;
+	b2 += b13; b13 = ((b13 << 34) | (b13 >> (64 - 34))) ^ b2;
+	b6 += b11; b11 = ((b11 << 56) | (b11 >> (64 - 56))) ^ b6;
+	b4 += b15; b15 = ((b15 << 51) | (b15 >> (64 - 51))) ^ b4;
+	b10 += b7; b7 = ((b7 << 4) | (b7 >> (64 - 4))) ^ b10;
+	b12 += b3; b3 = ((b3 << 53) | (b3 >> (64 - 53))) ^ b12;
+	b14 += b5; b5 = ((b5 << 42) | (b5 >> (64 - 42))) ^ b14;
+	b8 += b1; b1 = ((b1 << 41) | (b1 >> (64 - 41))) ^ b8;
+	b0 += b7; b7 = ((b7 << 31) | (b7 >> (64 - 31))) ^ b0;
+	b2 += b5; b5 = ((b5 << 44) | (b5 >> (64 - 44))) ^ b2;
+	b4 += b3; b3 = ((b3 << 47) | (b3 >> (64 - 47))) ^ b4;
+	b6 += b1; b1 = ((b1 << 46) | (b1 >> (64 - 46))) ^ b6;
+	b12 += b15; b15 = ((b15 << 19) | (b15 >> (64 - 19))) ^ b12;
+	b14 += b13; b13 = ((b13 << 42) | (b13 >> (64 - 42))) ^ b14;
+	b8 += b11; b11 = ((b11 << 44) | (b11 >> (64 - 44))) ^ b8;
+	b10 += b9; b9 = ((b9 << 25) | (b9 >> (64 - 25))) ^ b10;
+	b0 += b15; b15 = ((b15 << 9) | (b15 >> (64 - 9))) ^ b0;
+	b2 += b11; b11 = ((b11 << 48) | (b11 >> (64 - 48))) ^ b2;
+	b6 += b13; b13 = ((b13 << 35) | (b13 >> (64 - 35))) ^ b6;
+	b4 += b9; b9 = ((b9 << 52) | (b9 >> (64 - 52))) ^ b4;
+	b14 += b1; b1 = ((b1 << 23) | (b1 >> (64 - 23))) ^ b14;
+	b8 += b5; b5 = ((b5 << 31) | (b5 >> (64 - 31))) ^ b8;
+	b10 += b3; b3 = ((b3 << 37) | (b3 >> (64 - 37))) ^ b10;
+	b12 += b7; b7 = ((b7 << 20) | (b7 >> (64 - 20))) ^ b12;
+	b1 += k2; b0 += b1 + k1; b1 = ((b1 << 24) | (b1 >> (64 - 24))) ^ b0;
+	b3 += k4; b2 += b3 + k3; b3 = ((b3 << 13) | (b3 >> (64 - 13))) ^ b2;
+	b5 += k6; b4 += b5 + k5; b5 = ((b5 << 8) | (b5 >> (64 - 8))) ^ b4;
+	b7 += k8; b6 += b7 + k7; b7 = ((b7 << 47) | (b7 >> (64 - 47))) ^ b6;
+	b9 += k10; b8 += b9 + k9; b9 = ((b9 << 8) | (b9 >> (64 - 8))) ^ b8;
+	b11 += k12; b10 += b11 + k11; b11 = ((b11 << 17) | (b11 >> (64 - 17))) ^ b10;
+	b13 += k14 + t0; b12 += b13 + k13; b13 = ((b13 << 22) | (b13 >> (64 - 22))) ^ b12;
+	b15 += k16 + 18; b14 += b15 + k15 + t1; b15 = ((b15 << 37) | (b15 >> (64 - 37))) ^ b14;
+	b0 += b9; b9 = ((b9 << 38) | (b9 >> (64 - 38))) ^ b0;
+	b2 += b13; b13 = ((b13 << 19) | (b13 >> (64 - 19))) ^ b2;
+	b6 += b11; b11 = ((b11 << 10) | (b11 >> (64 - 10))) ^ b6;
+	b4 += b15; b15 = ((b15 << 55) | (b15 >> (64 - 55))) ^ b4;
+	b10 += b7; b7 = ((b7 << 49) | (b7 >> (64 - 49))) ^ b10;
+	b12 += b3; b3 = ((b3 << 18) | (b3 >> (64 - 18))) ^ b12;
+	b14 += b5; b5 = ((b5 << 23) | (b5 >> (64 - 23))) ^ b14;
+	b8 += b1; b1 = ((b1 << 52) | (b1 >> (64 - 52))) ^ b8;
+	b0 += b7; b7 = ((b7 << 33) | (b7 >> (64 - 33))) ^ b0;
+	b2 += b5; b5 = ((b5 << 4) | (b5 >> (64 - 4))) ^ b2;
+	b4 += b3; b3 = ((b3 << 51) | (b3 >> (64 - 51))) ^ b4;
+	b6 += b1; b1 = ((b1 << 13) | (b1 >> (64 - 13))) ^ b6;
+	b12 += b15; b15 = ((b15 << 34) | (b15 >> (64 - 34))) ^ b12;
+	b14 += b13; b13 = ((b13 << 41) | (b13 >> (64 - 41))) ^ b14;
+	b8 += b11; b11 = ((b11 << 59) | (b11 >> (64 - 59))) ^ b8;
+	b10 += b9; b9 = ((b9 << 17) | (b9 >> (64 - 17))) ^ b10;
+	b0 += b15; b15 = ((b15 << 5) | (b15 >> (64 - 5))) ^ b0;
+	b2 += b11; b11 = ((b11 << 20) | (b11 >> (64 - 20))) ^ b2;
+	b6 += b13; b13 = ((b13 << 48) | (b13 >> (64 - 48))) ^ b6;
+	b4 += b9; b9 = ((b9 << 41) | (b9 >> (64 - 41))) ^ b4;
+	b14 += b1; b1 = ((b1 << 47) | (b1 >> (64 - 47))) ^ b14;
+	b8 += b5; b5 = ((b5 << 28) | (b5 >> (64 - 28))) ^ b8;
+	b10 += b3; b3 = ((b3 << 16) | (b3 >> (64 - 16))) ^ b10;
+	b12 += b7; b7 = ((b7 << 25) | (b7 >> (64 - 25))) ^ b12;
+	b1 += k3; b0 += b1 + k2; b1 = ((b1 << 41) | (b1 >> (64 - 41))) ^ b0;
+	b3 += k5; b2 += b3 + k4; b3 = ((b3 << 9) | (b3 >> (64 - 9))) ^ b2;
+	b5 += k7; b4 += b5 + k6; b5 = ((b5 << 37) | (b5 >> (64 - 37))) ^ b4;
+	b7 += k9; b6 += b7 + k8; b7 = ((b7 << 31) | (b7 >> (64 - 31))) ^ b6;
+	b9 += k11; b8 += b9 + k10; b9 = ((b9 << 12) | (b9 >> (64 - 12))) ^ b8;
+	b11 += k13; b10 += b11 + k12; b11 = ((b11 << 47) | (b11 >> (64 - 47))) ^ b10;
+	b13 += k15 + t1; b12 += b13 + k14; b13 = ((b13 << 44) | (b13 >> (64 - 44))) ^ b12;
+	b15 += k0 + 19; b14 += b15 + k16 + t2; b15 = ((b15 << 30) | (b15 >> (64 - 30))) ^ b14;
+	b0 += b9; b9 = ((b9 << 16) | (b9 >> (64 - 16))) ^ b0;
+	b2 += b13; b13 = ((b13 << 34) | (b13 >> (64 - 34))) ^ b2;
+	b6 += b11; b11 = ((b11 << 56) | (b11 >> (64 - 56))) ^ b6;
+	b4 += b15; b15 = ((b15 << 51) | (b15 >> (64 - 51))) ^ b4;
+	b10 += b7; b7 = ((b7 << 4) | (b7 >> (64 - 4))) ^ b10;
+	b12 += b3; b3 = ((b3 << 53) | (b3 >> (64 - 53))) ^ b12;
+	b14 += b5; b5 = ((b5 << 42) | (b5 >> (64 - 42))) ^ b14;
+	b8 += b1; b1 = ((b1 << 41) | (b1 >> (64 - 41))) ^ b8;
+	b0 += b7; b7 = ((b7 << 31) | (b7 >> (64 - 31))) ^ b0;
+	b2 += b5; b5 = ((b5 << 44) | (b5 >> (64 - 44))) ^ b2;
+	b4 += b3; b3 = ((b3 << 47) | (b3 >> (64 - 47))) ^ b4;
+	b6 += b1; b1 = ((b1 << 46) | (b1 >> (64 - 46))) ^ b6;
+	b12 += b15; b15 = ((b15 << 19) | (b15 >> (64 - 19))) ^ b12;
+	b14 += b13; b13 = ((b13 << 42) | (b13 >> (64 - 42))) ^ b14;
+	b8 += b11; b11 = ((b11 << 44) | (b11 >> (64 - 44))) ^ b8;
+	b10 += b9; b9 = ((b9 << 25) | (b9 >> (64 - 25))) ^ b10;
+	b0 += b15; b15 = ((b15 << 9) | (b15 >> (64 - 9))) ^ b0;
+	b2 += b11; b11 = ((b11 << 48) | (b11 >> (64 - 48))) ^ b2;
+	b6 += b13; b13 = ((b13 << 35) | (b13 >> (64 - 35))) ^ b6;
+	b4 += b9; b9 = ((b9 << 52) | (b9 >> (64 - 52))) ^ b4;
+	b14 += b1; b1 = ((b1 << 23) | (b1 >> (64 - 23))) ^ b14;
+	b8 += b5; b5 = ((b5 << 31) | (b5 >> (64 - 31))) ^ b8;
+	b10 += b3; b3 = ((b3 << 37) | (b3 >> (64 - 37))) ^ b10;
+	b12 += b7; b7 = ((b7 << 20) | (b7 >> (64 - 20))) ^ b12;
 
-            output[0] = b0 + k3;
-            output[1] = b1 + k4;
-            output[2] = b2 + k5;
-            output[3] = b3 + k6;
-            output[4] = b4 + k7;
-            output[5] = b5 + k8;
-            output[6] = b6 + k9;
-            output[7] = b7 + k10;
-            output[8] = b8 + k11;
-            output[9] = b9 + k12;
-            output[10] = b10 + k13;
-            output[11] = b11 + k14;
-            output[12] = b12 + k15;
-            output[13] = b13 + k16 + t2;
-            output[14] = b14 + k0 + t0;
-            output[15] = b15 + k1 + 20;
-        }
+	output[0] = b0 + k3;
+	output[1] = b1 + k4;
+	output[2] = b2 + k5;
+	output[3] = b3 + k6;
+	output[4] = b4 + k7;
+	output[5] = b5 + k8;
+	output[6] = b6 + k9;
+	output[7] = b7 + k10;
+	output[8] = b8 + k11;
+	output[9] = b9 + k12;
+	output[10] = b10 + k13;
+	output[11] = b11 + k14;
+	output[12] = b12 + k15;
+	output[13] = b13 + k16 + t2;
+	output[14] = b14 + k0 + t0;
+	output[15] = b15 + k1 + 20;
+}
 
 void threefishDecrypt1024(struct threefish_key *keyCtx, u64 *input, u64 *output)
 {
+	u64 b0 = input[0], b1 = input[1],
+	  b2 = input[2], b3 = input[3],
+	  b4 = input[4], b5 = input[5],
+	  b6 = input[6], b7 = input[7],
+	  b8 = input[8], b9 = input[9],
+	  b10 = input[10], b11 = input[11],
+	  b12 = input[12], b13 = input[13],
+	  b14 = input[14], b15 = input[15];
+	u64 k0 = keyCtx->key[0], k1 = keyCtx->key[1],
+	  k2 = keyCtx->key[2], k3 = keyCtx->key[3],
+	  k4 = keyCtx->key[4], k5 = keyCtx->key[5],
+	  k6 = keyCtx->key[6], k7 = keyCtx->key[7],
+	  k8 = keyCtx->key[8], k9 = keyCtx->key[9],
+	  k10 = keyCtx->key[10], k11 = keyCtx->key[11],
+	  k12 = keyCtx->key[12], k13 = keyCtx->key[13],
+	  k14 = keyCtx->key[14], k15 = keyCtx->key[15],
+	  k16 = keyCtx->key[16];
+	u64 t0 = keyCtx->tweak[0], t1 = keyCtx->tweak[1],
+	  t2 = keyCtx->tweak[2];
+	u64 tmp;
 
-    u64 b0 = input[0], b1 = input[1],
-      b2 = input[2], b3 = input[3],
-      b4 = input[4], b5 = input[5],
-      b6 = input[6], b7 = input[7],
-      b8 = input[8], b9 = input[9],
-      b10 = input[10], b11 = input[11],
-      b12 = input[12], b13 = input[13],
-      b14 = input[14], b15 = input[15];
-    u64 k0 = keyCtx->key[0], k1 = keyCtx->key[1],
-      k2 = keyCtx->key[2], k3 = keyCtx->key[3],
-      k4 = keyCtx->key[4], k5 = keyCtx->key[5],
-      k6 = keyCtx->key[6], k7 = keyCtx->key[7],
-      k8 = keyCtx->key[8], k9 = keyCtx->key[9],
-      k10 = keyCtx->key[10], k11 = keyCtx->key[11],
-      k12 = keyCtx->key[12], k13 = keyCtx->key[13],
-      k14 = keyCtx->key[14], k15 = keyCtx->key[15],
-      k16 = keyCtx->key[16];
-    u64 t0 = keyCtx->tweak[0], t1 = keyCtx->tweak[1],
-      t2 = keyCtx->tweak[2];
-    u64 tmp;
-
-            b0 -= k3;
-            b1 -= k4;
-            b2 -= k5;
-            b3 -= k6;
-            b4 -= k7;
-            b5 -= k8;
-            b6 -= k9;
-            b7 -= k10;
-            b8 -= k11;
-            b9 -= k12;
-            b10 -= k13;
-            b11 -= k14;
-            b12 -= k15;
-            b13 -= k16 + t2;
-            b14 -= k0 + t0;
-            b15 -= k1 + 20;
-            tmp = b7 ^ b12; b7 = (tmp >> 20) | (tmp << (64 - 20)); b12 -= b7;
-            tmp = b3 ^ b10; b3 = (tmp >> 37) | (tmp << (64 - 37)); b10 -= b3;
-            tmp = b5 ^ b8; b5 = (tmp >> 31) | (tmp << (64 - 31)); b8 -= b5;
-            tmp = b1 ^ b14; b1 = (tmp >> 23) | (tmp << (64 - 23)); b14 -= b1;
-            tmp = b9 ^ b4; b9 = (tmp >> 52) | (tmp << (64 - 52)); b4 -= b9;
-            tmp = b13 ^ b6; b13 = (tmp >> 35) | (tmp << (64 - 35)); b6 -= b13;
-            tmp = b11 ^ b2; b11 = (tmp >> 48) | (tmp << (64 - 48)); b2 -= b11;
-            tmp = b15 ^ b0; b15 = (tmp >> 9) | (tmp << (64 - 9)); b0 -= b15;
-            tmp = b9 ^ b10; b9 = (tmp >> 25) | (tmp << (64 - 25)); b10 -= b9;
-            tmp = b11 ^ b8; b11 = (tmp >> 44) | (tmp << (64 - 44)); b8 -= b11;
-            tmp = b13 ^ b14; b13 = (tmp >> 42) | (tmp << (64 - 42)); b14 -= b13;
-            tmp = b15 ^ b12; b15 = (tmp >> 19) | (tmp << (64 - 19)); b12 -= b15;
-            tmp = b1 ^ b6; b1 = (tmp >> 46) | (tmp << (64 - 46)); b6 -= b1;
-            tmp = b3 ^ b4; b3 = (tmp >> 47) | (tmp << (64 - 47)); b4 -= b3;
-            tmp = b5 ^ b2; b5 = (tmp >> 44) | (tmp << (64 - 44)); b2 -= b5;
-            tmp = b7 ^ b0; b7 = (tmp >> 31) | (tmp << (64 - 31)); b0 -= b7;
-            tmp = b1 ^ b8; b1 = (tmp >> 41) | (tmp << (64 - 41)); b8 -= b1;
-            tmp = b5 ^ b14; b5 = (tmp >> 42) | (tmp << (64 - 42)); b14 -= b5;
-            tmp = b3 ^ b12; b3 = (tmp >> 53) | (tmp << (64 - 53)); b12 -= b3;
-            tmp = b7 ^ b10; b7 = (tmp >> 4) | (tmp << (64 - 4)); b10 -= b7;
-            tmp = b15 ^ b4; b15 = (tmp >> 51) | (tmp << (64 - 51)); b4 -= b15;
-            tmp = b11 ^ b6; b11 = (tmp >> 56) | (tmp << (64 - 56)); b6 -= b11;
-            tmp = b13 ^ b2; b13 = (tmp >> 34) | (tmp << (64 - 34)); b2 -= b13;
-            tmp = b9 ^ b0; b9 = (tmp >> 16) | (tmp << (64 - 16)); b0 -= b9;
-            tmp = b15 ^ b14; b15 = (tmp >> 30) | (tmp << (64 - 30)); b14 -= b15 + k16 + t2; b15 -= k0 + 19;
-            tmp = b13 ^ b12; b13 = (tmp >> 44) | (tmp << (64 - 44)); b12 -= b13 + k14; b13 -= k15 + t1;
-            tmp = b11 ^ b10; b11 = (tmp >> 47) | (tmp << (64 - 47)); b10 -= b11 + k12; b11 -= k13;
-            tmp = b9 ^ b8; b9 = (tmp >> 12) | (tmp << (64 - 12)); b8 -= b9 + k10; b9 -= k11;
-            tmp = b7 ^ b6; b7 = (tmp >> 31) | (tmp << (64 - 31)); b6 -= b7 + k8; b7 -= k9;
-            tmp = b5 ^ b4; b5 = (tmp >> 37) | (tmp << (64 - 37)); b4 -= b5 + k6; b5 -= k7;
-            tmp = b3 ^ b2; b3 = (tmp >> 9) | (tmp << (64 - 9)); b2 -= b3 + k4; b3 -= k5;
-            tmp = b1 ^ b0; b1 = (tmp >> 41) | (tmp << (64 - 41)); b0 -= b1 + k2; b1 -= k3;
-            tmp = b7 ^ b12; b7 = (tmp >> 25) | (tmp << (64 - 25)); b12 -= b7;
-            tmp = b3 ^ b10; b3 = (tmp >> 16) | (tmp << (64 - 16)); b10 -= b3;
-            tmp = b5 ^ b8; b5 = (tmp >> 28) | (tmp << (64 - 28)); b8 -= b5;
-            tmp = b1 ^ b14; b1 = (tmp >> 47) | (tmp << (64 - 47)); b14 -= b1;
-            tmp = b9 ^ b4; b9 = (tmp >> 41) | (tmp << (64 - 41)); b4 -= b9;
-            tmp = b13 ^ b6; b13 = (tmp >> 48) | (tmp << (64 - 48)); b6 -= b13;
-            tmp = b11 ^ b2; b11 = (tmp >> 20) | (tmp << (64 - 20)); b2 -= b11;
-            tmp = b15 ^ b0; b15 = (tmp >> 5) | (tmp << (64 - 5)); b0 -= b15;
-            tmp = b9 ^ b10; b9 = (tmp >> 17) | (tmp << (64 - 17)); b10 -= b9;
-            tmp = b11 ^ b8; b11 = (tmp >> 59) | (tmp << (64 - 59)); b8 -= b11;
-            tmp = b13 ^ b14; b13 = (tmp >> 41) | (tmp << (64 - 41)); b14 -= b13;
-            tmp = b15 ^ b12; b15 = (tmp >> 34) | (tmp << (64 - 34)); b12 -= b15;
-            tmp = b1 ^ b6; b1 = (tmp >> 13) | (tmp << (64 - 13)); b6 -= b1;
-            tmp = b3 ^ b4; b3 = (tmp >> 51) | (tmp << (64 - 51)); b4 -= b3;
-            tmp = b5 ^ b2; b5 = (tmp >> 4) | (tmp << (64 - 4)); b2 -= b5;
-            tmp = b7 ^ b0; b7 = (tmp >> 33) | (tmp << (64 - 33)); b0 -= b7;
-            tmp = b1 ^ b8; b1 = (tmp >> 52) | (tmp << (64 - 52)); b8 -= b1;
-            tmp = b5 ^ b14; b5 = (tmp >> 23) | (tmp << (64 - 23)); b14 -= b5;
-            tmp = b3 ^ b12; b3 = (tmp >> 18) | (tmp << (64 - 18)); b12 -= b3;
-            tmp = b7 ^ b10; b7 = (tmp >> 49) | (tmp << (64 - 49)); b10 -= b7;
-            tmp = b15 ^ b4; b15 = (tmp >> 55) | (tmp << (64 - 55)); b4 -= b15;
-            tmp = b11 ^ b6; b11 = (tmp >> 10) | (tmp << (64 - 10)); b6 -= b11;
-            tmp = b13 ^ b2; b13 = (tmp >> 19) | (tmp << (64 - 19)); b2 -= b13;
-            tmp = b9 ^ b0; b9 = (tmp >> 38) | (tmp << (64 - 38)); b0 -= b9;
-            tmp = b15 ^ b14; b15 = (tmp >> 37) | (tmp << (64 - 37)); b14 -= b15 + k15 + t1; b15 -= k16 + 18;
-            tmp = b13 ^ b12; b13 = (tmp >> 22) | (tmp << (64 - 22)); b12 -= b13 + k13; b13 -= k14 + t0;
-            tmp = b11 ^ b10; b11 = (tmp >> 17) | (tmp << (64 - 17)); b10 -= b11 + k11; b11 -= k12;
-            tmp = b9 ^ b8; b9 = (tmp >> 8) | (tmp << (64 - 8)); b8 -= b9 + k9; b9 -= k10;
-            tmp = b7 ^ b6; b7 = (tmp >> 47) | (tmp << (64 - 47)); b6 -= b7 + k7; b7 -= k8;
-            tmp = b5 ^ b4; b5 = (tmp >> 8) | (tmp << (64 - 8)); b4 -= b5 + k5; b5 -= k6;
-            tmp = b3 ^ b2; b3 = (tmp >> 13) | (tmp << (64 - 13)); b2 -= b3 + k3; b3 -= k4;
-            tmp = b1 ^ b0; b1 = (tmp >> 24) | (tmp << (64 - 24)); b0 -= b1 + k1; b1 -= k2;
-            tmp = b7 ^ b12; b7 = (tmp >> 20) | (tmp << (64 - 20)); b12 -= b7;
-            tmp = b3 ^ b10; b3 = (tmp >> 37) | (tmp << (64 - 37)); b10 -= b3;
-            tmp = b5 ^ b8; b5 = (tmp >> 31) | (tmp << (64 - 31)); b8 -= b5;
-            tmp = b1 ^ b14; b1 = (tmp >> 23) | (tmp << (64 - 23)); b14 -= b1;
-            tmp = b9 ^ b4; b9 = (tmp >> 52) | (tmp << (64 - 52)); b4 -= b9;
-            tmp = b13 ^ b6; b13 = (tmp >> 35) | (tmp << (64 - 35)); b6 -= b13;
-            tmp = b11 ^ b2; b11 = (tmp >> 48) | (tmp << (64 - 48)); b2 -= b11;
-            tmp = b15 ^ b0; b15 = (tmp >> 9) | (tmp << (64 - 9)); b0 -= b15;
-            tmp = b9 ^ b10; b9 = (tmp >> 25) | (tmp << (64 - 25)); b10 -= b9;
-            tmp = b11 ^ b8; b11 = (tmp >> 44) | (tmp << (64 - 44)); b8 -= b11;
-            tmp = b13 ^ b14; b13 = (tmp >> 42) | (tmp << (64 - 42)); b14 -= b13;
-            tmp = b15 ^ b12; b15 = (tmp >> 19) | (tmp << (64 - 19)); b12 -= b15;
-            tmp = b1 ^ b6; b1 = (tmp >> 46) | (tmp << (64 - 46)); b6 -= b1;
-            tmp = b3 ^ b4; b3 = (tmp >> 47) | (tmp << (64 - 47)); b4 -= b3;
-            tmp = b5 ^ b2; b5 = (tmp >> 44) | (tmp << (64 - 44)); b2 -= b5;
-            tmp = b7 ^ b0; b7 = (tmp >> 31) | (tmp << (64 - 31)); b0 -= b7;
-            tmp = b1 ^ b8; b1 = (tmp >> 41) | (tmp << (64 - 41)); b8 -= b1;
-            tmp = b5 ^ b14; b5 = (tmp >> 42) | (tmp << (64 - 42)); b14 -= b5;
-            tmp = b3 ^ b12; b3 = (tmp >> 53) | (tmp << (64 - 53)); b12 -= b3;
-            tmp = b7 ^ b10; b7 = (tmp >> 4) | (tmp << (64 - 4)); b10 -= b7;
-            tmp = b15 ^ b4; b15 = (tmp >> 51) | (tmp << (64 - 51)); b4 -= b15;
-            tmp = b11 ^ b6; b11 = (tmp >> 56) | (tmp << (64 - 56)); b6 -= b11;
-            tmp = b13 ^ b2; b13 = (tmp >> 34) | (tmp << (64 - 34)); b2 -= b13;
-            tmp = b9 ^ b0; b9 = (tmp >> 16) | (tmp << (64 - 16)); b0 -= b9;
-            tmp = b15 ^ b14; b15 = (tmp >> 30) | (tmp << (64 - 30)); b14 -= b15 + k14 + t0; b15 -= k15 + 17;
-            tmp = b13 ^ b12; b13 = (tmp >> 44) | (tmp << (64 - 44)); b12 -= b13 + k12; b13 -= k13 + t2;
-            tmp = b11 ^ b10; b11 = (tmp >> 47) | (tmp << (64 - 47)); b10 -= b11 + k10; b11 -= k11;
-            tmp = b9 ^ b8; b9 = (tmp >> 12) | (tmp << (64 - 12)); b8 -= b9 + k8; b9 -= k9;
-            tmp = b7 ^ b6; b7 = (tmp >> 31) | (tmp << (64 - 31)); b6 -= b7 + k6; b7 -= k7;
-            tmp = b5 ^ b4; b5 = (tmp >> 37) | (tmp << (64 - 37)); b4 -= b5 + k4; b5 -= k5;
-            tmp = b3 ^ b2; b3 = (tmp >> 9) | (tmp << (64 - 9)); b2 -= b3 + k2; b3 -= k3;
-            tmp = b1 ^ b0; b1 = (tmp >> 41) | (tmp << (64 - 41)); b0 -= b1 + k0; b1 -= k1;
-            tmp = b7 ^ b12; b7 = (tmp >> 25) | (tmp << (64 - 25)); b12 -= b7;
-            tmp = b3 ^ b10; b3 = (tmp >> 16) | (tmp << (64 - 16)); b10 -= b3;
-            tmp = b5 ^ b8; b5 = (tmp >> 28) | (tmp << (64 - 28)); b8 -= b5;
-            tmp = b1 ^ b14; b1 = (tmp >> 47) | (tmp << (64 - 47)); b14 -= b1;
-            tmp = b9 ^ b4; b9 = (tmp >> 41) | (tmp << (64 - 41)); b4 -= b9;
-            tmp = b13 ^ b6; b13 = (tmp >> 48) | (tmp << (64 - 48)); b6 -= b13;
-            tmp = b11 ^ b2; b11 = (tmp >> 20) | (tmp << (64 - 20)); b2 -= b11;
-            tmp = b15 ^ b0; b15 = (tmp >> 5) | (tmp << (64 - 5)); b0 -= b15;
-            tmp = b9 ^ b10; b9 = (tmp >> 17) | (tmp << (64 - 17)); b10 -= b9;
-            tmp = b11 ^ b8; b11 = (tmp >> 59) | (tmp << (64 - 59)); b8 -= b11;
-            tmp = b13 ^ b14; b13 = (tmp >> 41) | (tmp << (64 - 41)); b14 -= b13;
-            tmp = b15 ^ b12; b15 = (tmp >> 34) | (tmp << (64 - 34)); b12 -= b15;
-            tmp = b1 ^ b6; b1 = (tmp >> 13) | (tmp << (64 - 13)); b6 -= b1;
-            tmp = b3 ^ b4; b3 = (tmp >> 51) | (tmp << (64 - 51)); b4 -= b3;
-            tmp = b5 ^ b2; b5 = (tmp >> 4) | (tmp << (64 - 4)); b2 -= b5;
-            tmp = b7 ^ b0; b7 = (tmp >> 33) | (tmp << (64 - 33)); b0 -= b7;
-            tmp = b1 ^ b8; b1 = (tmp >> 52) | (tmp << (64 - 52)); b8 -= b1;
-            tmp = b5 ^ b14; b5 = (tmp >> 23) | (tmp << (64 - 23)); b14 -= b5;
-            tmp = b3 ^ b12; b3 = (tmp >> 18) | (tmp << (64 - 18)); b12 -= b3;
-            tmp = b7 ^ b10; b7 = (tmp >> 49) | (tmp << (64 - 49)); b10 -= b7;
-            tmp = b15 ^ b4; b15 = (tmp >> 55) | (tmp << (64 - 55)); b4 -= b15;
-            tmp = b11 ^ b6; b11 = (tmp >> 10) | (tmp << (64 - 10)); b6 -= b11;
-            tmp = b13 ^ b2; b13 = (tmp >> 19) | (tmp << (64 - 19)); b2 -= b13;
-            tmp = b9 ^ b0; b9 = (tmp >> 38) | (tmp << (64 - 38)); b0 -= b9;
-            tmp = b15 ^ b14; b15 = (tmp >> 37) | (tmp << (64 - 37)); b14 -= b15 + k13 + t2; b15 -= k14 + 16;
-            tmp = b13 ^ b12; b13 = (tmp >> 22) | (tmp << (64 - 22)); b12 -= b13 + k11; b13 -= k12 + t1;
-            tmp = b11 ^ b10; b11 = (tmp >> 17) | (tmp << (64 - 17)); b10 -= b11 + k9; b11 -= k10;
-            tmp = b9 ^ b8; b9 = (tmp >> 8) | (tmp << (64 - 8)); b8 -= b9 + k7; b9 -= k8;
-            tmp = b7 ^ b6; b7 = (tmp >> 47) | (tmp << (64 - 47)); b6 -= b7 + k5; b7 -= k6;
-            tmp = b5 ^ b4; b5 = (tmp >> 8) | (tmp << (64 - 8)); b4 -= b5 + k3; b5 -= k4;
-            tmp = b3 ^ b2; b3 = (tmp >> 13) | (tmp << (64 - 13)); b2 -= b3 + k1; b3 -= k2;
-            tmp = b1 ^ b0; b1 = (tmp >> 24) | (tmp << (64 - 24)); b0 -= b1 + k16; b1 -= k0;
-            tmp = b7 ^ b12; b7 = (tmp >> 20) | (tmp << (64 - 20)); b12 -= b7;
-            tmp = b3 ^ b10; b3 = (tmp >> 37) | (tmp << (64 - 37)); b10 -= b3;
-            tmp = b5 ^ b8; b5 = (tmp >> 31) | (tmp << (64 - 31)); b8 -= b5;
-            tmp = b1 ^ b14; b1 = (tmp >> 23) | (tmp << (64 - 23)); b14 -= b1;
-            tmp = b9 ^ b4; b9 = (tmp >> 52) | (tmp << (64 - 52)); b4 -= b9;
-            tmp = b13 ^ b6; b13 = (tmp >> 35) | (tmp << (64 - 35)); b6 -= b13;
-            tmp = b11 ^ b2; b11 = (tmp >> 48) | (tmp << (64 - 48)); b2 -= b11;
-            tmp = b15 ^ b0; b15 = (tmp >> 9) | (tmp << (64 - 9)); b0 -= b15;
-            tmp = b9 ^ b10; b9 = (tmp >> 25) | (tmp << (64 - 25)); b10 -= b9;
-            tmp = b11 ^ b8; b11 = (tmp >> 44) | (tmp << (64 - 44)); b8 -= b11;
-            tmp = b13 ^ b14; b13 = (tmp >> 42) | (tmp << (64 - 42)); b14 -= b13;
-            tmp = b15 ^ b12; b15 = (tmp >> 19) | (tmp << (64 - 19)); b12 -= b15;
-            tmp = b1 ^ b6; b1 = (tmp >> 46) | (tmp << (64 - 46)); b6 -= b1;
-            tmp = b3 ^ b4; b3 = (tmp >> 47) | (tmp << (64 - 47)); b4 -= b3;
-            tmp = b5 ^ b2; b5 = (tmp >> 44) | (tmp << (64 - 44)); b2 -= b5;
-            tmp = b7 ^ b0; b7 = (tmp >> 31) | (tmp << (64 - 31)); b0 -= b7;
-            tmp = b1 ^ b8; b1 = (tmp >> 41) | (tmp << (64 - 41)); b8 -= b1;
-            tmp = b5 ^ b14; b5 = (tmp >> 42) | (tmp << (64 - 42)); b14 -= b5;
-            tmp = b3 ^ b12; b3 = (tmp >> 53) | (tmp << (64 - 53)); b12 -= b3;
-            tmp = b7 ^ b10; b7 = (tmp >> 4) | (tmp << (64 - 4)); b10 -= b7;
-            tmp = b15 ^ b4; b15 = (tmp >> 51) | (tmp << (64 - 51)); b4 -= b15;
-            tmp = b11 ^ b6; b11 = (tmp >> 56) | (tmp << (64 - 56)); b6 -= b11;
-            tmp = b13 ^ b2; b13 = (tmp >> 34) | (tmp << (64 - 34)); b2 -= b13;
-            tmp = b9 ^ b0; b9 = (tmp >> 16) | (tmp << (64 - 16)); b0 -= b9;
-            tmp = b15 ^ b14; b15 = (tmp >> 30) | (tmp << (64 - 30)); b14 -= b15 + k12 + t1; b15 -= k13 + 15;
-            tmp = b13 ^ b12; b13 = (tmp >> 44) | (tmp << (64 - 44)); b12 -= b13 + k10; b13 -= k11 + t0;
-            tmp = b11 ^ b10; b11 = (tmp >> 47) | (tmp << (64 - 47)); b10 -= b11 + k8; b11 -= k9;
-            tmp = b9 ^ b8; b9 = (tmp >> 12) | (tmp << (64 - 12)); b8 -= b9 + k6; b9 -= k7;
-            tmp = b7 ^ b6; b7 = (tmp >> 31) | (tmp << (64 - 31)); b6 -= b7 + k4; b7 -= k5;
-            tmp = b5 ^ b4; b5 = (tmp >> 37) | (tmp << (64 - 37)); b4 -= b5 + k2; b5 -= k3;
-            tmp = b3 ^ b2; b3 = (tmp >> 9) | (tmp << (64 - 9)); b2 -= b3 + k0; b3 -= k1;
-            tmp = b1 ^ b0; b1 = (tmp >> 41) | (tmp << (64 - 41)); b0 -= b1 + k15; b1 -= k16;
-            tmp = b7 ^ b12; b7 = (tmp >> 25) | (tmp << (64 - 25)); b12 -= b7;
-            tmp = b3 ^ b10; b3 = (tmp >> 16) | (tmp << (64 - 16)); b10 -= b3;
-            tmp = b5 ^ b8; b5 = (tmp >> 28) | (tmp << (64 - 28)); b8 -= b5;
-            tmp = b1 ^ b14; b1 = (tmp >> 47) | (tmp << (64 - 47)); b14 -= b1;
-            tmp = b9 ^ b4; b9 = (tmp >> 41) | (tmp << (64 - 41)); b4 -= b9;
-            tmp = b13 ^ b6; b13 = (tmp >> 48) | (tmp << (64 - 48)); b6 -= b13;
-            tmp = b11 ^ b2; b11 = (tmp >> 20) | (tmp << (64 - 20)); b2 -= b11;
-            tmp = b15 ^ b0; b15 = (tmp >> 5) | (tmp << (64 - 5)); b0 -= b15;
-            tmp = b9 ^ b10; b9 = (tmp >> 17) | (tmp << (64 - 17)); b10 -= b9;
-            tmp = b11 ^ b8; b11 = (tmp >> 59) | (tmp << (64 - 59)); b8 -= b11;
-            tmp = b13 ^ b14; b13 = (tmp >> 41) | (tmp << (64 - 41)); b14 -= b13;
-            tmp = b15 ^ b12; b15 = (tmp >> 34) | (tmp << (64 - 34)); b12 -= b15;
-            tmp = b1 ^ b6; b1 = (tmp >> 13) | (tmp << (64 - 13)); b6 -= b1;
-            tmp = b3 ^ b4; b3 = (tmp >> 51) | (tmp << (64 - 51)); b4 -= b3;
-            tmp = b5 ^ b2; b5 = (tmp >> 4) | (tmp << (64 - 4)); b2 -= b5;
-            tmp = b7 ^ b0; b7 = (tmp >> 33) | (tmp << (64 - 33)); b0 -= b7;
-            tmp = b1 ^ b8; b1 = (tmp >> 52) | (tmp << (64 - 52)); b8 -= b1;
-            tmp = b5 ^ b14; b5 = (tmp >> 23) | (tmp << (64 - 23)); b14 -= b5;
-            tmp = b3 ^ b12; b3 = (tmp >> 18) | (tmp << (64 - 18)); b12 -= b3;
-            tmp = b7 ^ b10; b7 = (tmp >> 49) | (tmp << (64 - 49)); b10 -= b7;
-            tmp = b15 ^ b4; b15 = (tmp >> 55) | (tmp << (64 - 55)); b4 -= b15;
-            tmp = b11 ^ b6; b11 = (tmp >> 10) | (tmp << (64 - 10)); b6 -= b11;
-            tmp = b13 ^ b2; b13 = (tmp >> 19) | (tmp << (64 - 19)); b2 -= b13;
-            tmp = b9 ^ b0; b9 = (tmp >> 38) | (tmp << (64 - 38)); b0 -= b9;
-            tmp = b15 ^ b14; b15 = (tmp >> 37) | (tmp << (64 - 37)); b14 -= b15 + k11 + t0; b15 -= k12 + 14;
-            tmp = b13 ^ b12; b13 = (tmp >> 22) | (tmp << (64 - 22)); b12 -= b13 + k9; b13 -= k10 + t2;
-            tmp = b11 ^ b10; b11 = (tmp >> 17) | (tmp << (64 - 17)); b10 -= b11 + k7; b11 -= k8;
-            tmp = b9 ^ b8; b9 = (tmp >> 8) | (tmp << (64 - 8)); b8 -= b9 + k5; b9 -= k6;
-            tmp = b7 ^ b6; b7 = (tmp >> 47) | (tmp << (64 - 47)); b6 -= b7 + k3; b7 -= k4;
-            tmp = b5 ^ b4; b5 = (tmp >> 8) | (tmp << (64 - 8)); b4 -= b5 + k1; b5 -= k2;
-            tmp = b3 ^ b2; b3 = (tmp >> 13) | (tmp << (64 - 13)); b2 -= b3 + k16; b3 -= k0;
-            tmp = b1 ^ b0; b1 = (tmp >> 24) | (tmp << (64 - 24)); b0 -= b1 + k14; b1 -= k15;
-            tmp = b7 ^ b12; b7 = (tmp >> 20) | (tmp << (64 - 20)); b12 -= b7;
-            tmp = b3 ^ b10; b3 = (tmp >> 37) | (tmp << (64 - 37)); b10 -= b3;
-            tmp = b5 ^ b8; b5 = (tmp >> 31) | (tmp << (64 - 31)); b8 -= b5;
-            tmp = b1 ^ b14; b1 = (tmp >> 23) | (tmp << (64 - 23)); b14 -= b1;
-            tmp = b9 ^ b4; b9 = (tmp >> 52) | (tmp << (64 - 52)); b4 -= b9;
-            tmp = b13 ^ b6; b13 = (tmp >> 35) | (tmp << (64 - 35)); b6 -= b13;
-            tmp = b11 ^ b2; b11 = (tmp >> 48) | (tmp << (64 - 48)); b2 -= b11;
-            tmp = b15 ^ b0; b15 = (tmp >> 9) | (tmp << (64 - 9)); b0 -= b15;
-            tmp = b9 ^ b10; b9 = (tmp >> 25) | (tmp << (64 - 25)); b10 -= b9;
-            tmp = b11 ^ b8; b11 = (tmp >> 44) | (tmp << (64 - 44)); b8 -= b11;
-            tmp = b13 ^ b14; b13 = (tmp >> 42) | (tmp << (64 - 42)); b14 -= b13;
-            tmp = b15 ^ b12; b15 = (tmp >> 19) | (tmp << (64 - 19)); b12 -= b15;
-            tmp = b1 ^ b6; b1 = (tmp >> 46) | (tmp << (64 - 46)); b6 -= b1;
-            tmp = b3 ^ b4; b3 = (tmp >> 47) | (tmp << (64 - 47)); b4 -= b3;
-            tmp = b5 ^ b2; b5 = (tmp >> 44) | (tmp << (64 - 44)); b2 -= b5;
-            tmp = b7 ^ b0; b7 = (tmp >> 31) | (tmp << (64 - 31)); b0 -= b7;
-            tmp = b1 ^ b8; b1 = (tmp >> 41) | (tmp << (64 - 41)); b8 -= b1;
-            tmp = b5 ^ b14; b5 = (tmp >> 42) | (tmp << (64 - 42)); b14 -= b5;
-            tmp = b3 ^ b12; b3 = (tmp >> 53) | (tmp << (64 - 53)); b12 -= b3;
-            tmp = b7 ^ b10; b7 = (tmp >> 4) | (tmp << (64 - 4)); b10 -= b7;
-            tmp = b15 ^ b4; b15 = (tmp >> 51) | (tmp << (64 - 51)); b4 -= b15;
-            tmp = b11 ^ b6; b11 = (tmp >> 56) | (tmp << (64 - 56)); b6 -= b11;
-            tmp = b13 ^ b2; b13 = (tmp >> 34) | (tmp << (64 - 34)); b2 -= b13;
-            tmp = b9 ^ b0; b9 = (tmp >> 16) | (tmp << (64 - 16)); b0 -= b9;
-            tmp = b15 ^ b14; b15 = (tmp >> 30) | (tmp << (64 - 30)); b14 -= b15 + k10 + t2; b15 -= k11 + 13;
-            tmp = b13 ^ b12; b13 = (tmp >> 44) | (tmp << (64 - 44)); b12 -= b13 + k8; b13 -= k9 + t1;
-            tmp = b11 ^ b10; b11 = (tmp >> 47) | (tmp << (64 - 47)); b10 -= b11 + k6; b11 -= k7;
-            tmp = b9 ^ b8; b9 = (tmp >> 12) | (tmp << (64 - 12)); b8 -= b9 + k4; b9 -= k5;
-            tmp = b7 ^ b6; b7 = (tmp >> 31) | (tmp << (64 - 31)); b6 -= b7 + k2; b7 -= k3;
-            tmp = b5 ^ b4; b5 = (tmp >> 37) | (tmp << (64 - 37)); b4 -= b5 + k0; b5 -= k1;
-            tmp = b3 ^ b2; b3 = (tmp >> 9) | (tmp << (64 - 9)); b2 -= b3 + k15; b3 -= k16;
-            tmp = b1 ^ b0; b1 = (tmp >> 41) | (tmp << (64 - 41)); b0 -= b1 + k13; b1 -= k14;
-            tmp = b7 ^ b12; b7 = (tmp >> 25) | (tmp << (64 - 25)); b12 -= b7;
-            tmp = b3 ^ b10; b3 = (tmp >> 16) | (tmp << (64 - 16)); b10 -= b3;
-            tmp = b5 ^ b8; b5 = (tmp >> 28) | (tmp << (64 - 28)); b8 -= b5;
-            tmp = b1 ^ b14; b1 = (tmp >> 47) | (tmp << (64 - 47)); b14 -= b1;
-            tmp = b9 ^ b4; b9 = (tmp >> 41) | (tmp << (64 - 41)); b4 -= b9;
-            tmp = b13 ^ b6; b13 = (tmp >> 48) | (tmp << (64 - 48)); b6 -= b13;
-            tmp = b11 ^ b2; b11 = (tmp >> 20) | (tmp << (64 - 20)); b2 -= b11;
-            tmp = b15 ^ b0; b15 = (tmp >> 5) | (tmp << (64 - 5)); b0 -= b15;
-            tmp = b9 ^ b10; b9 = (tmp >> 17) | (tmp << (64 - 17)); b10 -= b9;
-            tmp = b11 ^ b8; b11 = (tmp >> 59) | (tmp << (64 - 59)); b8 -= b11;
-            tmp = b13 ^ b14; b13 = (tmp >> 41) | (tmp << (64 - 41)); b14 -= b13;
-            tmp = b15 ^ b12; b15 = (tmp >> 34) | (tmp << (64 - 34)); b12 -= b15;
-            tmp = b1 ^ b6; b1 = (tmp >> 13) | (tmp << (64 - 13)); b6 -= b1;
-            tmp = b3 ^ b4; b3 = (tmp >> 51) | (tmp << (64 - 51)); b4 -= b3;
-            tmp = b5 ^ b2; b5 = (tmp >> 4) | (tmp << (64 - 4)); b2 -= b5;
-            tmp = b7 ^ b0; b7 = (tmp >> 33) | (tmp << (64 - 33)); b0 -= b7;
-            tmp = b1 ^ b8; b1 = (tmp >> 52) | (tmp << (64 - 52)); b8 -= b1;
-            tmp = b5 ^ b14; b5 = (tmp >> 23) | (tmp << (64 - 23)); b14 -= b5;
-            tmp = b3 ^ b12; b3 = (tmp >> 18) | (tmp << (64 - 18)); b12 -= b3;
-            tmp = b7 ^ b10; b7 = (tmp >> 49) | (tmp << (64 - 49)); b10 -= b7;
-            tmp = b15 ^ b4; b15 = (tmp >> 55) | (tmp << (64 - 55)); b4 -= b15;
-            tmp = b11 ^ b6; b11 = (tmp >> 10) | (tmp << (64 - 10)); b6 -= b11;
-            tmp = b13 ^ b2; b13 = (tmp >> 19) | (tmp << (64 - 19)); b2 -= b13;
-            tmp = b9 ^ b0; b9 = (tmp >> 38) | (tmp << (64 - 38)); b0 -= b9;
-            tmp = b15 ^ b14; b15 = (tmp >> 37) | (tmp << (64 - 37)); b14 -= b15 + k9 + t1; b15 -= k10 + 12;
-            tmp = b13 ^ b12; b13 = (tmp >> 22) | (tmp << (64 - 22)); b12 -= b13 + k7; b13 -= k8 + t0;
-            tmp = b11 ^ b10; b11 = (tmp >> 17) | (tmp << (64 - 17)); b10 -= b11 + k5; b11 -= k6;
-            tmp = b9 ^ b8; b9 = (tmp >> 8) | (tmp << (64 - 8)); b8 -= b9 + k3; b9 -= k4;
-            tmp = b7 ^ b6; b7 = (tmp >> 47) | (tmp << (64 - 47)); b6 -= b7 + k1; b7 -= k2;
-            tmp = b5 ^ b4; b5 = (tmp >> 8) | (tmp << (64 - 8)); b4 -= b5 + k16; b5 -= k0;
-            tmp = b3 ^ b2; b3 = (tmp >> 13) | (tmp << (64 - 13)); b2 -= b3 + k14; b3 -= k15;
-            tmp = b1 ^ b0; b1 = (tmp >> 24) | (tmp << (64 - 24)); b0 -= b1 + k12; b1 -= k13;
-            tmp = b7 ^ b12; b7 = (tmp >> 20) | (tmp << (64 - 20)); b12 -= b7;
-            tmp = b3 ^ b10; b3 = (tmp >> 37) | (tmp << (64 - 37)); b10 -= b3;
-            tmp = b5 ^ b8; b5 = (tmp >> 31) | (tmp << (64 - 31)); b8 -= b5;
-            tmp = b1 ^ b14; b1 = (tmp >> 23) | (tmp << (64 - 23)); b14 -= b1;
-            tmp = b9 ^ b4; b9 = (tmp >> 52) | (tmp << (64 - 52)); b4 -= b9;
-            tmp = b13 ^ b6; b13 = (tmp >> 35) | (tmp << (64 - 35)); b6 -= b13;
-            tmp = b11 ^ b2; b11 = (tmp >> 48) | (tmp << (64 - 48)); b2 -= b11;
-            tmp = b15 ^ b0; b15 = (tmp >> 9) | (tmp << (64 - 9)); b0 -= b15;
-            tmp = b9 ^ b10; b9 = (tmp >> 25) | (tmp << (64 - 25)); b10 -= b9;
-            tmp = b11 ^ b8; b11 = (tmp >> 44) | (tmp << (64 - 44)); b8 -= b11;
-            tmp = b13 ^ b14; b13 = (tmp >> 42) | (tmp << (64 - 42)); b14 -= b13;
-            tmp = b15 ^ b12; b15 = (tmp >> 19) | (tmp << (64 - 19)); b12 -= b15;
-            tmp = b1 ^ b6; b1 = (tmp >> 46) | (tmp << (64 - 46)); b6 -= b1;
-            tmp = b3 ^ b4; b3 = (tmp >> 47) | (tmp << (64 - 47)); b4 -= b3;
-            tmp = b5 ^ b2; b5 = (tmp >> 44) | (tmp << (64 - 44)); b2 -= b5;
-            tmp = b7 ^ b0; b7 = (tmp >> 31) | (tmp << (64 - 31)); b0 -= b7;
-            tmp = b1 ^ b8; b1 = (tmp >> 41) | (tmp << (64 - 41)); b8 -= b1;
-            tmp = b5 ^ b14; b5 = (tmp >> 42) | (tmp << (64 - 42)); b14 -= b5;
-            tmp = b3 ^ b12; b3 = (tmp >> 53) | (tmp << (64 - 53)); b12 -= b3;
-            tmp = b7 ^ b10; b7 = (tmp >> 4) | (tmp << (64 - 4)); b10 -= b7;
-            tmp = b15 ^ b4; b15 = (tmp >> 51) | (tmp << (64 - 51)); b4 -= b15;
-            tmp = b11 ^ b6; b11 = (tmp >> 56) | (tmp << (64 - 56)); b6 -= b11;
-            tmp = b13 ^ b2; b13 = (tmp >> 34) | (tmp << (64 - 34)); b2 -= b13;
-            tmp = b9 ^ b0; b9 = (tmp >> 16) | (tmp << (64 - 16)); b0 -= b9;
-            tmp = b15 ^ b14; b15 = (tmp >> 30) | (tmp << (64 - 30)); b14 -= b15 + k8 + t0; b15 -= k9 + 11;
-            tmp = b13 ^ b12; b13 = (tmp >> 44) | (tmp << (64 - 44)); b12 -= b13 + k6; b13 -= k7 + t2;
-            tmp = b11 ^ b10; b11 = (tmp >> 47) | (tmp << (64 - 47)); b10 -= b11 + k4; b11 -= k5;
-            tmp = b9 ^ b8; b9 = (tmp >> 12) | (tmp << (64 - 12)); b8 -= b9 + k2; b9 -= k3;
-            tmp = b7 ^ b6; b7 = (tmp >> 31) | (tmp << (64 - 31)); b6 -= b7 + k0; b7 -= k1;
-            tmp = b5 ^ b4; b5 = (tmp >> 37) | (tmp << (64 - 37)); b4 -= b5 + k15; b5 -= k16;
-            tmp = b3 ^ b2; b3 = (tmp >> 9) | (tmp << (64 - 9)); b2 -= b3 + k13; b3 -= k14;
-            tmp = b1 ^ b0; b1 = (tmp >> 41) | (tmp << (64 - 41)); b0 -= b1 + k11; b1 -= k12;
-            tmp = b7 ^ b12; b7 = (tmp >> 25) | (tmp << (64 - 25)); b12 -= b7;
-            tmp = b3 ^ b10; b3 = (tmp >> 16) | (tmp << (64 - 16)); b10 -= b3;
-            tmp = b5 ^ b8; b5 = (tmp >> 28) | (tmp << (64 - 28)); b8 -= b5;
-            tmp = b1 ^ b14; b1 = (tmp >> 47) | (tmp << (64 - 47)); b14 -= b1;
-            tmp = b9 ^ b4; b9 = (tmp >> 41) | (tmp << (64 - 41)); b4 -= b9;
-            tmp = b13 ^ b6; b13 = (tmp >> 48) | (tmp << (64 - 48)); b6 -= b13;
-            tmp = b11 ^ b2; b11 = (tmp >> 20) | (tmp << (64 - 20)); b2 -= b11;
-            tmp = b15 ^ b0; b15 = (tmp >> 5) | (tmp << (64 - 5)); b0 -= b15;
-            tmp = b9 ^ b10; b9 = (tmp >> 17) | (tmp << (64 - 17)); b10 -= b9;
-            tmp = b11 ^ b8; b11 = (tmp >> 59) | (tmp << (64 - 59)); b8 -= b11;
-            tmp = b13 ^ b14; b13 = (tmp >> 41) | (tmp << (64 - 41)); b14 -= b13;
-            tmp = b15 ^ b12; b15 = (tmp >> 34) | (tmp << (64 - 34)); b12 -= b15;
-            tmp = b1 ^ b6; b1 = (tmp >> 13) | (tmp << (64 - 13)); b6 -= b1;
-            tmp = b3 ^ b4; b3 = (tmp >> 51) | (tmp << (64 - 51)); b4 -= b3;
-            tmp = b5 ^ b2; b5 = (tmp >> 4) | (tmp << (64 - 4)); b2 -= b5;
-            tmp = b7 ^ b0; b7 = (tmp >> 33) | (tmp << (64 - 33)); b0 -= b7;
-            tmp = b1 ^ b8; b1 = (tmp >> 52) | (tmp << (64 - 52)); b8 -= b1;
-            tmp = b5 ^ b14; b5 = (tmp >> 23) | (tmp << (64 - 23)); b14 -= b5;
-            tmp = b3 ^ b12; b3 = (tmp >> 18) | (tmp << (64 - 18)); b12 -= b3;
-            tmp = b7 ^ b10; b7 = (tmp >> 49) | (tmp << (64 - 49)); b10 -= b7;
-            tmp = b15 ^ b4; b15 = (tmp >> 55) | (tmp << (64 - 55)); b4 -= b15;
-            tmp = b11 ^ b6; b11 = (tmp >> 10) | (tmp << (64 - 10)); b6 -= b11;
-            tmp = b13 ^ b2; b13 = (tmp >> 19) | (tmp << (64 - 19)); b2 -= b13;
-            tmp = b9 ^ b0; b9 = (tmp >> 38) | (tmp << (64 - 38)); b0 -= b9;
-            tmp = b15 ^ b14; b15 = (tmp >> 37) | (tmp << (64 - 37)); b14 -= b15 + k7 + t2; b15 -= k8 + 10;
-            tmp = b13 ^ b12; b13 = (tmp >> 22) | (tmp << (64 - 22)); b12 -= b13 + k5; b13 -= k6 + t1;
-            tmp = b11 ^ b10; b11 = (tmp >> 17) | (tmp << (64 - 17)); b10 -= b11 + k3; b11 -= k4;
-            tmp = b9 ^ b8; b9 = (tmp >> 8) | (tmp << (64 - 8)); b8 -= b9 + k1; b9 -= k2;
-            tmp = b7 ^ b6; b7 = (tmp >> 47) | (tmp << (64 - 47)); b6 -= b7 + k16; b7 -= k0;
-            tmp = b5 ^ b4; b5 = (tmp >> 8) | (tmp << (64 - 8)); b4 -= b5 + k14; b5 -= k15;
-            tmp = b3 ^ b2; b3 = (tmp >> 13) | (tmp << (64 - 13)); b2 -= b3 + k12; b3 -= k13;
-            tmp = b1 ^ b0; b1 = (tmp >> 24) | (tmp << (64 - 24)); b0 -= b1 + k10; b1 -= k11;
-            tmp = b7 ^ b12; b7 = (tmp >> 20) | (tmp << (64 - 20)); b12 -= b7;
-            tmp = b3 ^ b10; b3 = (tmp >> 37) | (tmp << (64 - 37)); b10 -= b3;
-            tmp = b5 ^ b8; b5 = (tmp >> 31) | (tmp << (64 - 31)); b8 -= b5;
-            tmp = b1 ^ b14; b1 = (tmp >> 23) | (tmp << (64 - 23)); b14 -= b1;
-            tmp = b9 ^ b4; b9 = (tmp >> 52) | (tmp << (64 - 52)); b4 -= b9;
-            tmp = b13 ^ b6; b13 = (tmp >> 35) | (tmp << (64 - 35)); b6 -= b13;
-            tmp = b11 ^ b2; b11 = (tmp >> 48) | (tmp << (64 - 48)); b2 -= b11;
-            tmp = b15 ^ b0; b15 = (tmp >> 9) | (tmp << (64 - 9)); b0 -= b15;
-            tmp = b9 ^ b10; b9 = (tmp >> 25) | (tmp << (64 - 25)); b10 -= b9;
-            tmp = b11 ^ b8; b11 = (tmp >> 44) | (tmp << (64 - 44)); b8 -= b11;
-            tmp = b13 ^ b14; b13 = (tmp >> 42) | (tmp << (64 - 42)); b14 -= b13;
-            tmp = b15 ^ b12; b15 = (tmp >> 19) | (tmp << (64 - 19)); b12 -= b15;
-            tmp = b1 ^ b6; b1 = (tmp >> 46) | (tmp << (64 - 46)); b6 -= b1;
-            tmp = b3 ^ b4; b3 = (tmp >> 47) | (tmp << (64 - 47)); b4 -= b3;
-            tmp = b5 ^ b2; b5 = (tmp >> 44) | (tmp << (64 - 44)); b2 -= b5;
-            tmp = b7 ^ b0; b7 = (tmp >> 31) | (tmp << (64 - 31)); b0 -= b7;
-            tmp = b1 ^ b8; b1 = (tmp >> 41) | (tmp << (64 - 41)); b8 -= b1;
-            tmp = b5 ^ b14; b5 = (tmp >> 42) | (tmp << (64 - 42)); b14 -= b5;
-            tmp = b3 ^ b12; b3 = (tmp >> 53) | (tmp << (64 - 53)); b12 -= b3;
-            tmp = b7 ^ b10; b7 = (tmp >> 4) | (tmp << (64 - 4)); b10 -= b7;
-            tmp = b15 ^ b4; b15 = (tmp >> 51) | (tmp << (64 - 51)); b4 -= b15;
-            tmp = b11 ^ b6; b11 = (tmp >> 56) | (tmp << (64 - 56)); b6 -= b11;
-            tmp = b13 ^ b2; b13 = (tmp >> 34) | (tmp << (64 - 34)); b2 -= b13;
-            tmp = b9 ^ b0; b9 = (tmp >> 16) | (tmp << (64 - 16)); b0 -= b9;
-            tmp = b15 ^ b14; b15 = (tmp >> 30) | (tmp << (64 - 30)); b14 -= b15 + k6 + t1; b15 -= k7 + 9;
-            tmp = b13 ^ b12; b13 = (tmp >> 44) | (tmp << (64 - 44)); b12 -= b13 + k4; b13 -= k5 + t0;
-            tmp = b11 ^ b10; b11 = (tmp >> 47) | (tmp << (64 - 47)); b10 -= b11 + k2; b11 -= k3;
-            tmp = b9 ^ b8; b9 = (tmp >> 12) | (tmp << (64 - 12)); b8 -= b9 + k0; b9 -= k1;
-            tmp = b7 ^ b6; b7 = (tmp >> 31) | (tmp << (64 - 31)); b6 -= b7 + k15; b7 -= k16;
-            tmp = b5 ^ b4; b5 = (tmp >> 37) | (tmp << (64 - 37)); b4 -= b5 + k13; b5 -= k14;
-            tmp = b3 ^ b2; b3 = (tmp >> 9) | (tmp << (64 - 9)); b2 -= b3 + k11; b3 -= k12;
-            tmp = b1 ^ b0; b1 = (tmp >> 41) | (tmp << (64 - 41)); b0 -= b1 + k9; b1 -= k10;
-            tmp = b7 ^ b12; b7 = (tmp >> 25) | (tmp << (64 - 25)); b12 -= b7;
-            tmp = b3 ^ b10; b3 = (tmp >> 16) | (tmp << (64 - 16)); b10 -= b3;
-            tmp = b5 ^ b8; b5 = (tmp >> 28) | (tmp << (64 - 28)); b8 -= b5;
-            tmp = b1 ^ b14; b1 = (tmp >> 47) | (tmp << (64 - 47)); b14 -= b1;
-            tmp = b9 ^ b4; b9 = (tmp >> 41) | (tmp << (64 - 41)); b4 -= b9;
-            tmp = b13 ^ b6; b13 = (tmp >> 48) | (tmp << (64 - 48)); b6 -= b13;
-            tmp = b11 ^ b2; b11 = (tmp >> 20) | (tmp << (64 - 20)); b2 -= b11;
-            tmp = b15 ^ b0; b15 = (tmp >> 5) | (tmp << (64 - 5)); b0 -= b15;
-            tmp = b9 ^ b10; b9 = (tmp >> 17) | (tmp << (64 - 17)); b10 -= b9;
-            tmp = b11 ^ b8; b11 = (tmp >> 59) | (tmp << (64 - 59)); b8 -= b11;
-            tmp = b13 ^ b14; b13 = (tmp >> 41) | (tmp << (64 - 41)); b14 -= b13;
-            tmp = b15 ^ b12; b15 = (tmp >> 34) | (tmp << (64 - 34)); b12 -= b15;
-            tmp = b1 ^ b6; b1 = (tmp >> 13) | (tmp << (64 - 13)); b6 -= b1;
-            tmp = b3 ^ b4; b3 = (tmp >> 51) | (tmp << (64 - 51)); b4 -= b3;
-            tmp = b5 ^ b2; b5 = (tmp >> 4) | (tmp << (64 - 4)); b2 -= b5;
-            tmp = b7 ^ b0; b7 = (tmp >> 33) | (tmp << (64 - 33)); b0 -= b7;
-            tmp = b1 ^ b8; b1 = (tmp >> 52) | (tmp << (64 - 52)); b8 -= b1;
-            tmp = b5 ^ b14; b5 = (tmp >> 23) | (tmp << (64 - 23)); b14 -= b5;
-            tmp = b3 ^ b12; b3 = (tmp >> 18) | (tmp << (64 - 18)); b12 -= b3;
-            tmp = b7 ^ b10; b7 = (tmp >> 49) | (tmp << (64 - 49)); b10 -= b7;
-            tmp = b15 ^ b4; b15 = (tmp >> 55) | (tmp << (64 - 55)); b4 -= b15;
-            tmp = b11 ^ b6; b11 = (tmp >> 10) | (tmp << (64 - 10)); b6 -= b11;
-            tmp = b13 ^ b2; b13 = (tmp >> 19) | (tmp << (64 - 19)); b2 -= b13;
-            tmp = b9 ^ b0; b9 = (tmp >> 38) | (tmp << (64 - 38)); b0 -= b9;
-            tmp = b15 ^ b14; b15 = (tmp >> 37) | (tmp << (64 - 37)); b14 -= b15 + k5 + t0; b15 -= k6 + 8;
-            tmp = b13 ^ b12; b13 = (tmp >> 22) | (tmp << (64 - 22)); b12 -= b13 + k3; b13 -= k4 + t2;
-            tmp = b11 ^ b10; b11 = (tmp >> 17) | (tmp << (64 - 17)); b10 -= b11 + k1; b11 -= k2;
-            tmp = b9 ^ b8; b9 = (tmp >> 8) | (tmp << (64 - 8)); b8 -= b9 + k16; b9 -= k0;
-            tmp = b7 ^ b6; b7 = (tmp >> 47) | (tmp << (64 - 47)); b6 -= b7 + k14; b7 -= k15;
-            tmp = b5 ^ b4; b5 = (tmp >> 8) | (tmp << (64 - 8)); b4 -= b5 + k12; b5 -= k13;
-            tmp = b3 ^ b2; b3 = (tmp >> 13) | (tmp << (64 - 13)); b2 -= b3 + k10; b3 -= k11;
-            tmp = b1 ^ b0; b1 = (tmp >> 24) | (tmp << (64 - 24)); b0 -= b1 + k8; b1 -= k9;
-            tmp = b7 ^ b12; b7 = (tmp >> 20) | (tmp << (64 - 20)); b12 -= b7;
-            tmp = b3 ^ b10; b3 = (tmp >> 37) | (tmp << (64 - 37)); b10 -= b3;
-            tmp = b5 ^ b8; b5 = (tmp >> 31) | (tmp << (64 - 31)); b8 -= b5;
-            tmp = b1 ^ b14; b1 = (tmp >> 23) | (tmp << (64 - 23)); b14 -= b1;
-            tmp = b9 ^ b4; b9 = (tmp >> 52) | (tmp << (64 - 52)); b4 -= b9;
-            tmp = b13 ^ b6; b13 = (tmp >> 35) | (tmp << (64 - 35)); b6 -= b13;
-            tmp = b11 ^ b2; b11 = (tmp >> 48) | (tmp << (64 - 48)); b2 -= b11;
-            tmp = b15 ^ b0; b15 = (tmp >> 9) | (tmp << (64 - 9)); b0 -= b15;
-            tmp = b9 ^ b10; b9 = (tmp >> 25) | (tmp << (64 - 25)); b10 -= b9;
-            tmp = b11 ^ b8; b11 = (tmp >> 44) | (tmp << (64 - 44)); b8 -= b11;
-            tmp = b13 ^ b14; b13 = (tmp >> 42) | (tmp << (64 - 42)); b14 -= b13;
-            tmp = b15 ^ b12; b15 = (tmp >> 19) | (tmp << (64 - 19)); b12 -= b15;
-            tmp = b1 ^ b6; b1 = (tmp >> 46) | (tmp << (64 - 46)); b6 -= b1;
-            tmp = b3 ^ b4; b3 = (tmp >> 47) | (tmp << (64 - 47)); b4 -= b3;
-            tmp = b5 ^ b2; b5 = (tmp >> 44) | (tmp << (64 - 44)); b2 -= b5;
-            tmp = b7 ^ b0; b7 = (tmp >> 31) | (tmp << (64 - 31)); b0 -= b7;
-            tmp = b1 ^ b8; b1 = (tmp >> 41) | (tmp << (64 - 41)); b8 -= b1;
-            tmp = b5 ^ b14; b5 = (tmp >> 42) | (tmp << (64 - 42)); b14 -= b5;
-            tmp = b3 ^ b12; b3 = (tmp >> 53) | (tmp << (64 - 53)); b12 -= b3;
-            tmp = b7 ^ b10; b7 = (tmp >> 4) | (tmp << (64 - 4)); b10 -= b7;
-            tmp = b15 ^ b4; b15 = (tmp >> 51) | (tmp << (64 - 51)); b4 -= b15;
-            tmp = b11 ^ b6; b11 = (tmp >> 56) | (tmp << (64 - 56)); b6 -= b11;
-            tmp = b13 ^ b2; b13 = (tmp >> 34) | (tmp << (64 - 34)); b2 -= b13;
-            tmp = b9 ^ b0; b9 = (tmp >> 16) | (tmp << (64 - 16)); b0 -= b9;
-            tmp = b15 ^ b14; b15 = (tmp >> 30) | (tmp << (64 - 30)); b14 -= b15 + k4 + t2; b15 -= k5 + 7;
-            tmp = b13 ^ b12; b13 = (tmp >> 44) | (tmp << (64 - 44)); b12 -= b13 + k2; b13 -= k3 + t1;
-            tmp = b11 ^ b10; b11 = (tmp >> 47) | (tmp << (64 - 47)); b10 -= b11 + k0; b11 -= k1;
-            tmp = b9 ^ b8; b9 = (tmp >> 12) | (tmp << (64 - 12)); b8 -= b9 + k15; b9 -= k16;
-            tmp = b7 ^ b6; b7 = (tmp >> 31) | (tmp << (64 - 31)); b6 -= b7 + k13; b7 -= k14;
-            tmp = b5 ^ b4; b5 = (tmp >> 37) | (tmp << (64 - 37)); b4 -= b5 + k11; b5 -= k12;
-            tmp = b3 ^ b2; b3 = (tmp >> 9) | (tmp << (64 - 9)); b2 -= b3 + k9; b3 -= k10;
-            tmp = b1 ^ b0; b1 = (tmp >> 41) | (tmp << (64 - 41)); b0 -= b1 + k7; b1 -= k8;
-            tmp = b7 ^ b12; b7 = (tmp >> 25) | (tmp << (64 - 25)); b12 -= b7;
-            tmp = b3 ^ b10; b3 = (tmp >> 16) | (tmp << (64 - 16)); b10 -= b3;
-            tmp = b5 ^ b8; b5 = (tmp >> 28) | (tmp << (64 - 28)); b8 -= b5;
-            tmp = b1 ^ b14; b1 = (tmp >> 47) | (tmp << (64 - 47)); b14 -= b1;
-            tmp = b9 ^ b4; b9 = (tmp >> 41) | (tmp << (64 - 41)); b4 -= b9;
-            tmp = b13 ^ b6; b13 = (tmp >> 48) | (tmp << (64 - 48)); b6 -= b13;
-            tmp = b11 ^ b2; b11 = (tmp >> 20) | (tmp << (64 - 20)); b2 -= b11;
-            tmp = b15 ^ b0; b15 = (tmp >> 5) | (tmp << (64 - 5)); b0 -= b15;
-            tmp = b9 ^ b10; b9 = (tmp >> 17) | (tmp << (64 - 17)); b10 -= b9;
-            tmp = b11 ^ b8; b11 = (tmp >> 59) | (tmp << (64 - 59)); b8 -= b11;
-            tmp = b13 ^ b14; b13 = (tmp >> 41) | (tmp << (64 - 41)); b14 -= b13;
-            tmp = b15 ^ b12; b15 = (tmp >> 34) | (tmp << (64 - 34)); b12 -= b15;
-            tmp = b1 ^ b6; b1 = (tmp >> 13) | (tmp << (64 - 13)); b6 -= b1;
-            tmp = b3 ^ b4; b3 = (tmp >> 51) | (tmp << (64 - 51)); b4 -= b3;
-            tmp = b5 ^ b2; b5 = (tmp >> 4) | (tmp << (64 - 4)); b2 -= b5;
-            tmp = b7 ^ b0; b7 = (tmp >> 33) | (tmp << (64 - 33)); b0 -= b7;
-            tmp = b1 ^ b8; b1 = (tmp >> 52) | (tmp << (64 - 52)); b8 -= b1;
-            tmp = b5 ^ b14; b5 = (tmp >> 23) | (tmp << (64 - 23)); b14 -= b5;
-            tmp = b3 ^ b12; b3 = (tmp >> 18) | (tmp << (64 - 18)); b12 -= b3;
-            tmp = b7 ^ b10; b7 = (tmp >> 49) | (tmp << (64 - 49)); b10 -= b7;
-            tmp = b15 ^ b4; b15 = (tmp >> 55) | (tmp << (64 - 55)); b4 -= b15;
-            tmp = b11 ^ b6; b11 = (tmp >> 10) | (tmp << (64 - 10)); b6 -= b11;
-            tmp = b13 ^ b2; b13 = (tmp >> 19) | (tmp << (64 - 19)); b2 -= b13;
-            tmp = b9 ^ b0; b9 = (tmp >> 38) | (tmp << (64 - 38)); b0 -= b9;
-            tmp = b15 ^ b14; b15 = (tmp >> 37) | (tmp << (64 - 37)); b14 -= b15 + k3 + t1; b15 -= k4 + 6;
-            tmp = b13 ^ b12; b13 = (tmp >> 22) | (tmp << (64 - 22)); b12 -= b13 + k1; b13 -= k2 + t0;
-            tmp = b11 ^ b10; b11 = (tmp >> 17) | (tmp << (64 - 17)); b10 -= b11 + k16; b11 -= k0;
-            tmp = b9 ^ b8; b9 = (tmp >> 8) | (tmp << (64 - 8)); b8 -= b9 + k14; b9 -= k15;
-            tmp = b7 ^ b6; b7 = (tmp >> 47) | (tmp << (64 - 47)); b6 -= b7 + k12; b7 -= k13;
-            tmp = b5 ^ b4; b5 = (tmp >> 8) | (tmp << (64 - 8)); b4 -= b5 + k10; b5 -= k11;
-            tmp = b3 ^ b2; b3 = (tmp >> 13) | (tmp << (64 - 13)); b2 -= b3 + k8; b3 -= k9;
-            tmp = b1 ^ b0; b1 = (tmp >> 24) | (tmp << (64 - 24)); b0 -= b1 + k6; b1 -= k7;
-            tmp = b7 ^ b12; b7 = (tmp >> 20) | (tmp << (64 - 20)); b12 -= b7;
-            tmp = b3 ^ b10; b3 = (tmp >> 37) | (tmp << (64 - 37)); b10 -= b3;
-            tmp = b5 ^ b8; b5 = (tmp >> 31) | (tmp << (64 - 31)); b8 -= b5;
-            tmp = b1 ^ b14; b1 = (tmp >> 23) | (tmp << (64 - 23)); b14 -= b1;
-            tmp = b9 ^ b4; b9 = (tmp >> 52) | (tmp << (64 - 52)); b4 -= b9;
-            tmp = b13 ^ b6; b13 = (tmp >> 35) | (tmp << (64 - 35)); b6 -= b13;
-            tmp = b11 ^ b2; b11 = (tmp >> 48) | (tmp << (64 - 48)); b2 -= b11;
-            tmp = b15 ^ b0; b15 = (tmp >> 9) | (tmp << (64 - 9)); b0 -= b15;
-            tmp = b9 ^ b10; b9 = (tmp >> 25) | (tmp << (64 - 25)); b10 -= b9;
-            tmp = b11 ^ b8; b11 = (tmp >> 44) | (tmp << (64 - 44)); b8 -= b11;
-            tmp = b13 ^ b14; b13 = (tmp >> 42) | (tmp << (64 - 42)); b14 -= b13;
-            tmp = b15 ^ b12; b15 = (tmp >> 19) | (tmp << (64 - 19)); b12 -= b15;
-            tmp = b1 ^ b6; b1 = (tmp >> 46) | (tmp << (64 - 46)); b6 -= b1;
-            tmp = b3 ^ b4; b3 = (tmp >> 47) | (tmp << (64 - 47)); b4 -= b3;
-            tmp = b5 ^ b2; b5 = (tmp >> 44) | (tmp << (64 - 44)); b2 -= b5;
-            tmp = b7 ^ b0; b7 = (tmp >> 31) | (tmp << (64 - 31)); b0 -= b7;
-            tmp = b1 ^ b8; b1 = (tmp >> 41) | (tmp << (64 - 41)); b8 -= b1;
-            tmp = b5 ^ b14; b5 = (tmp >> 42) | (tmp << (64 - 42)); b14 -= b5;
-            tmp = b3 ^ b12; b3 = (tmp >> 53) | (tmp << (64 - 53)); b12 -= b3;
-            tmp = b7 ^ b10; b7 = (tmp >> 4) | (tmp << (64 - 4)); b10 -= b7;
-            tmp = b15 ^ b4; b15 = (tmp >> 51) | (tmp << (64 - 51)); b4 -= b15;
-            tmp = b11 ^ b6; b11 = (tmp >> 56) | (tmp << (64 - 56)); b6 -= b11;
-            tmp = b13 ^ b2; b13 = (tmp >> 34) | (tmp << (64 - 34)); b2 -= b13;
-            tmp = b9 ^ b0; b9 = (tmp >> 16) | (tmp << (64 - 16)); b0 -= b9;
-            tmp = b15 ^ b14; b15 = (tmp >> 30) | (tmp << (64 - 30)); b14 -= b15 + k2 + t0; b15 -= k3 + 5;
-            tmp = b13 ^ b12; b13 = (tmp >> 44) | (tmp << (64 - 44)); b12 -= b13 + k0; b13 -= k1 + t2;
-            tmp = b11 ^ b10; b11 = (tmp >> 47) | (tmp << (64 - 47)); b10 -= b11 + k15; b11 -= k16;
-            tmp = b9 ^ b8; b9 = (tmp >> 12) | (tmp << (64 - 12)); b8 -= b9 + k13; b9 -= k14;
-            tmp = b7 ^ b6; b7 = (tmp >> 31) | (tmp << (64 - 31)); b6 -= b7 + k11; b7 -= k12;
-            tmp = b5 ^ b4; b5 = (tmp >> 37) | (tmp << (64 - 37)); b4 -= b5 + k9; b5 -= k10;
-            tmp = b3 ^ b2; b3 = (tmp >> 9) | (tmp << (64 - 9)); b2 -= b3 + k7; b3 -= k8;
-            tmp = b1 ^ b0; b1 = (tmp >> 41) | (tmp << (64 - 41)); b0 -= b1 + k5; b1 -= k6;
-            tmp = b7 ^ b12; b7 = (tmp >> 25) | (tmp << (64 - 25)); b12 -= b7;
-            tmp = b3 ^ b10; b3 = (tmp >> 16) | (tmp << (64 - 16)); b10 -= b3;
-            tmp = b5 ^ b8; b5 = (tmp >> 28) | (tmp << (64 - 28)); b8 -= b5;
-            tmp = b1 ^ b14; b1 = (tmp >> 47) | (tmp << (64 - 47)); b14 -= b1;
-            tmp = b9 ^ b4; b9 = (tmp >> 41) | (tmp << (64 - 41)); b4 -= b9;
-            tmp = b13 ^ b6; b13 = (tmp >> 48) | (tmp << (64 - 48)); b6 -= b13;
-            tmp = b11 ^ b2; b11 = (tmp >> 20) | (tmp << (64 - 20)); b2 -= b11;
-            tmp = b15 ^ b0; b15 = (tmp >> 5) | (tmp << (64 - 5)); b0 -= b15;
-            tmp = b9 ^ b10; b9 = (tmp >> 17) | (tmp << (64 - 17)); b10 -= b9;
-            tmp = b11 ^ b8; b11 = (tmp >> 59) | (tmp << (64 - 59)); b8 -= b11;
-            tmp = b13 ^ b14; b13 = (tmp >> 41) | (tmp << (64 - 41)); b14 -= b13;
-            tmp = b15 ^ b12; b15 = (tmp >> 34) | (tmp << (64 - 34)); b12 -= b15;
-            tmp = b1 ^ b6; b1 = (tmp >> 13) | (tmp << (64 - 13)); b6 -= b1;
-            tmp = b3 ^ b4; b3 = (tmp >> 51) | (tmp << (64 - 51)); b4 -= b3;
-            tmp = b5 ^ b2; b5 = (tmp >> 4) | (tmp << (64 - 4)); b2 -= b5;
-            tmp = b7 ^ b0; b7 = (tmp >> 33) | (tmp << (64 - 33)); b0 -= b7;
-            tmp = b1 ^ b8; b1 = (tmp >> 52) | (tmp << (64 - 52)); b8 -= b1;
-            tmp = b5 ^ b14; b5 = (tmp >> 23) | (tmp << (64 - 23)); b14 -= b5;
-            tmp = b3 ^ b12; b3 = (tmp >> 18) | (tmp << (64 - 18)); b12 -= b3;
-            tmp = b7 ^ b10; b7 = (tmp >> 49) | (tmp << (64 - 49)); b10 -= b7;
-            tmp = b15 ^ b4; b15 = (tmp >> 55) | (tmp << (64 - 55)); b4 -= b15;
-            tmp = b11 ^ b6; b11 = (tmp >> 10) | (tmp << (64 - 10)); b6 -= b11;
-            tmp = b13 ^ b2; b13 = (tmp >> 19) | (tmp << (64 - 19)); b2 -= b13;
-            tmp = b9 ^ b0; b9 = (tmp >> 38) | (tmp << (64 - 38)); b0 -= b9;
-            tmp = b15 ^ b14; b15 = (tmp >> 37) | (tmp << (64 - 37)); b14 -= b15 + k1 + t2; b15 -= k2 + 4;
-            tmp = b13 ^ b12; b13 = (tmp >> 22) | (tmp << (64 - 22)); b12 -= b13 + k16; b13 -= k0 + t1;
-            tmp = b11 ^ b10; b11 = (tmp >> 17) | (tmp << (64 - 17)); b10 -= b11 + k14; b11 -= k15;
-            tmp = b9 ^ b8; b9 = (tmp >> 8) | (tmp << (64 - 8)); b8 -= b9 + k12; b9 -= k13;
-            tmp = b7 ^ b6; b7 = (tmp >> 47) | (tmp << (64 - 47)); b6 -= b7 + k10; b7 -= k11;
-            tmp = b5 ^ b4; b5 = (tmp >> 8) | (tmp << (64 - 8)); b4 -= b5 + k8; b5 -= k9;
-            tmp = b3 ^ b2; b3 = (tmp >> 13) | (tmp << (64 - 13)); b2 -= b3 + k6; b3 -= k7;
-            tmp = b1 ^ b0; b1 = (tmp >> 24) | (tmp << (64 - 24)); b0 -= b1 + k4; b1 -= k5;
-            tmp = b7 ^ b12; b7 = (tmp >> 20) | (tmp << (64 - 20)); b12 -= b7;
-            tmp = b3 ^ b10; b3 = (tmp >> 37) | (tmp << (64 - 37)); b10 -= b3;
-            tmp = b5 ^ b8; b5 = (tmp >> 31) | (tmp << (64 - 31)); b8 -= b5;
-            tmp = b1 ^ b14; b1 = (tmp >> 23) | (tmp << (64 - 23)); b14 -= b1;
-            tmp = b9 ^ b4; b9 = (tmp >> 52) | (tmp << (64 - 52)); b4 -= b9;
-            tmp = b13 ^ b6; b13 = (tmp >> 35) | (tmp << (64 - 35)); b6 -= b13;
-            tmp = b11 ^ b2; b11 = (tmp >> 48) | (tmp << (64 - 48)); b2 -= b11;
-            tmp = b15 ^ b0; b15 = (tmp >> 9) | (tmp << (64 - 9)); b0 -= b15;
-            tmp = b9 ^ b10; b9 = (tmp >> 25) | (tmp << (64 - 25)); b10 -= b9;
-            tmp = b11 ^ b8; b11 = (tmp >> 44) | (tmp << (64 - 44)); b8 -= b11;
-            tmp = b13 ^ b14; b13 = (tmp >> 42) | (tmp << (64 - 42)); b14 -= b13;
-            tmp = b15 ^ b12; b15 = (tmp >> 19) | (tmp << (64 - 19)); b12 -= b15;
-            tmp = b1 ^ b6; b1 = (tmp >> 46) | (tmp << (64 - 46)); b6 -= b1;
-            tmp = b3 ^ b4; b3 = (tmp >> 47) | (tmp << (64 - 47)); b4 -= b3;
-            tmp = b5 ^ b2; b5 = (tmp >> 44) | (tmp << (64 - 44)); b2 -= b5;
-            tmp = b7 ^ b0; b7 = (tmp >> 31) | (tmp << (64 - 31)); b0 -= b7;
-            tmp = b1 ^ b8; b1 = (tmp >> 41) | (tmp << (64 - 41)); b8 -= b1;
-            tmp = b5 ^ b14; b5 = (tmp >> 42) | (tmp << (64 - 42)); b14 -= b5;
-            tmp = b3 ^ b12; b3 = (tmp >> 53) | (tmp << (64 - 53)); b12 -= b3;
-            tmp = b7 ^ b10; b7 = (tmp >> 4) | (tmp << (64 - 4)); b10 -= b7;
-            tmp = b15 ^ b4; b15 = (tmp >> 51) | (tmp << (64 - 51)); b4 -= b15;
-            tmp = b11 ^ b6; b11 = (tmp >> 56) | (tmp << (64 - 56)); b6 -= b11;
-            tmp = b13 ^ b2; b13 = (tmp >> 34) | (tmp << (64 - 34)); b2 -= b13;
-            tmp = b9 ^ b0; b9 = (tmp >> 16) | (tmp << (64 - 16)); b0 -= b9;
-            tmp = b15 ^ b14; b15 = (tmp >> 30) | (tmp << (64 - 30)); b14 -= b15 + k0 + t1; b15 -= k1 + 3;
-            tmp = b13 ^ b12; b13 = (tmp >> 44) | (tmp << (64 - 44)); b12 -= b13 + k15; b13 -= k16 + t0;
-            tmp = b11 ^ b10; b11 = (tmp >> 47) | (tmp << (64 - 47)); b10 -= b11 + k13; b11 -= k14;
-            tmp = b9 ^ b8; b9 = (tmp >> 12) | (tmp << (64 - 12)); b8 -= b9 + k11; b9 -= k12;
-            tmp = b7 ^ b6; b7 = (tmp >> 31) | (tmp << (64 - 31)); b6 -= b7 + k9; b7 -= k10;
-            tmp = b5 ^ b4; b5 = (tmp >> 37) | (tmp << (64 - 37)); b4 -= b5 + k7; b5 -= k8;
-            tmp = b3 ^ b2; b3 = (tmp >> 9) | (tmp << (64 - 9)); b2 -= b3 + k5; b3 -= k6;
-            tmp = b1 ^ b0; b1 = (tmp >> 41) | (tmp << (64 - 41)); b0 -= b1 + k3; b1 -= k4;
-            tmp = b7 ^ b12; b7 = (tmp >> 25) | (tmp << (64 - 25)); b12 -= b7;
-            tmp = b3 ^ b10; b3 = (tmp >> 16) | (tmp << (64 - 16)); b10 -= b3;
-            tmp = b5 ^ b8; b5 = (tmp >> 28) | (tmp << (64 - 28)); b8 -= b5;
-            tmp = b1 ^ b14; b1 = (tmp >> 47) | (tmp << (64 - 47)); b14 -= b1;
-            tmp = b9 ^ b4; b9 = (tmp >> 41) | (tmp << (64 - 41)); b4 -= b9;
-            tmp = b13 ^ b6; b13 = (tmp >> 48) | (tmp << (64 - 48)); b6 -= b13;
-            tmp = b11 ^ b2; b11 = (tmp >> 20) | (tmp << (64 - 20)); b2 -= b11;
-            tmp = b15 ^ b0; b15 = (tmp >> 5) | (tmp << (64 - 5)); b0 -= b15;
-            tmp = b9 ^ b10; b9 = (tmp >> 17) | (tmp << (64 - 17)); b10 -= b9;
-            tmp = b11 ^ b8; b11 = (tmp >> 59) | (tmp << (64 - 59)); b8 -= b11;
-            tmp = b13 ^ b14; b13 = (tmp >> 41) | (tmp << (64 - 41)); b14 -= b13;
-            tmp = b15 ^ b12; b15 = (tmp >> 34) | (tmp << (64 - 34)); b12 -= b15;
-            tmp = b1 ^ b6; b1 = (tmp >> 13) | (tmp << (64 - 13)); b6 -= b1;
-            tmp = b3 ^ b4; b3 = (tmp >> 51) | (tmp << (64 - 51)); b4 -= b3;
-            tmp = b5 ^ b2; b5 = (tmp >> 4) | (tmp << (64 - 4)); b2 -= b5;
-            tmp = b7 ^ b0; b7 = (tmp >> 33) | (tmp << (64 - 33)); b0 -= b7;
-            tmp = b1 ^ b8; b1 = (tmp >> 52) | (tmp << (64 - 52)); b8 -= b1;
-            tmp = b5 ^ b14; b5 = (tmp >> 23) | (tmp << (64 - 23)); b14 -= b5;
-            tmp = b3 ^ b12; b3 = (tmp >> 18) | (tmp << (64 - 18)); b12 -= b3;
-            tmp = b7 ^ b10; b7 = (tmp >> 49) | (tmp << (64 - 49)); b10 -= b7;
-            tmp = b15 ^ b4; b15 = (tmp >> 55) | (tmp << (64 - 55)); b4 -= b15;
-            tmp = b11 ^ b6; b11 = (tmp >> 10) | (tmp << (64 - 10)); b6 -= b11;
-            tmp = b13 ^ b2; b13 = (tmp >> 19) | (tmp << (64 - 19)); b2 -= b13;
-            tmp = b9 ^ b0; b9 = (tmp >> 38) | (tmp << (64 - 38)); b0 -= b9;
-            tmp = b15 ^ b14; b15 = (tmp >> 37) | (tmp << (64 - 37)); b14 -= b15 + k16 + t0; b15 -= k0 + 2;
-            tmp = b13 ^ b12; b13 = (tmp >> 22) | (tmp << (64 - 22)); b12 -= b13 + k14; b13 -= k15 + t2;
-            tmp = b11 ^ b10; b11 = (tmp >> 17) | (tmp << (64 - 17)); b10 -= b11 + k12; b11 -= k13;
-            tmp = b9 ^ b8; b9 = (tmp >> 8) | (tmp << (64 - 8)); b8 -= b9 + k10; b9 -= k11;
-            tmp = b7 ^ b6; b7 = (tmp >> 47) | (tmp << (64 - 47)); b6 -= b7 + k8; b7 -= k9;
-            tmp = b5 ^ b4; b5 = (tmp >> 8) | (tmp << (64 - 8)); b4 -= b5 + k6; b5 -= k7;
-            tmp = b3 ^ b2; b3 = (tmp >> 13) | (tmp << (64 - 13)); b2 -= b3 + k4; b3 -= k5;
-            tmp = b1 ^ b0; b1 = (tmp >> 24) | (tmp << (64 - 24)); b0 -= b1 + k2; b1 -= k3;
-            tmp = b7 ^ b12; b7 = (tmp >> 20) | (tmp << (64 - 20)); b12 -= b7;
-            tmp = b3 ^ b10; b3 = (tmp >> 37) | (tmp << (64 - 37)); b10 -= b3;
-            tmp = b5 ^ b8; b5 = (tmp >> 31) | (tmp << (64 - 31)); b8 -= b5;
-            tmp = b1 ^ b14; b1 = (tmp >> 23) | (tmp << (64 - 23)); b14 -= b1;
-            tmp = b9 ^ b4; b9 = (tmp >> 52) | (tmp << (64 - 52)); b4 -= b9;
-            tmp = b13 ^ b6; b13 = (tmp >> 35) | (tmp << (64 - 35)); b6 -= b13;
-            tmp = b11 ^ b2; b11 = (tmp >> 48) | (tmp << (64 - 48)); b2 -= b11;
-            tmp = b15 ^ b0; b15 = (tmp >> 9) | (tmp << (64 - 9)); b0 -= b15;
-            tmp = b9 ^ b10; b9 = (tmp >> 25) | (tmp << (64 - 25)); b10 -= b9;
-            tmp = b11 ^ b8; b11 = (tmp >> 44) | (tmp << (64 - 44)); b8 -= b11;
-            tmp = b13 ^ b14; b13 = (tmp >> 42) | (tmp << (64 - 42)); b14 -= b13;
-            tmp = b15 ^ b12; b15 = (tmp >> 19) | (tmp << (64 - 19)); b12 -= b15;
-            tmp = b1 ^ b6; b1 = (tmp >> 46) | (tmp << (64 - 46)); b6 -= b1;
-            tmp = b3 ^ b4; b3 = (tmp >> 47) | (tmp << (64 - 47)); b4 -= b3;
-            tmp = b5 ^ b2; b5 = (tmp >> 44) | (tmp << (64 - 44)); b2 -= b5;
-            tmp = b7 ^ b0; b7 = (tmp >> 31) | (tmp << (64 - 31)); b0 -= b7;
-            tmp = b1 ^ b8; b1 = (tmp >> 41) | (tmp << (64 - 41)); b8 -= b1;
-            tmp = b5 ^ b14; b5 = (tmp >> 42) | (tmp << (64 - 42)); b14 -= b5;
-            tmp = b3 ^ b12; b3 = (tmp >> 53) | (tmp << (64 - 53)); b12 -= b3;
-            tmp = b7 ^ b10; b7 = (tmp >> 4) | (tmp << (64 - 4)); b10 -= b7;
-            tmp = b15 ^ b4; b15 = (tmp >> 51) | (tmp << (64 - 51)); b4 -= b15;
-            tmp = b11 ^ b6; b11 = (tmp >> 56) | (tmp << (64 - 56)); b6 -= b11;
-            tmp = b13 ^ b2; b13 = (tmp >> 34) | (tmp << (64 - 34)); b2 -= b13;
-            tmp = b9 ^ b0; b9 = (tmp >> 16) | (tmp << (64 - 16)); b0 -= b9;
-            tmp = b15 ^ b14; b15 = (tmp >> 30) | (tmp << (64 - 30)); b14 -= b15 + k15 + t2; b15 -= k16 + 1;
-            tmp = b13 ^ b12; b13 = (tmp >> 44) | (tmp << (64 - 44)); b12 -= b13 + k13; b13 -= k14 + t1;
-            tmp = b11 ^ b10; b11 = (tmp >> 47) | (tmp << (64 - 47)); b10 -= b11 + k11; b11 -= k12;
-            tmp = b9 ^ b8; b9 = (tmp >> 12) | (tmp << (64 - 12)); b8 -= b9 + k9; b9 -= k10;
-            tmp = b7 ^ b6; b7 = (tmp >> 31) | (tmp << (64 - 31)); b6 -= b7 + k7; b7 -= k8;
-            tmp = b5 ^ b4; b5 = (tmp >> 37) | (tmp << (64 - 37)); b4 -= b5 + k5; b5 -= k6;
-            tmp = b3 ^ b2; b3 = (tmp >> 9) | (tmp << (64 - 9)); b2 -= b3 + k3; b3 -= k4;
-            tmp = b1 ^ b0; b1 = (tmp >> 41) | (tmp << (64 - 41)); b0 -= b1 + k1; b1 -= k2;
-            tmp = b7 ^ b12; b7 = (tmp >> 25) | (tmp << (64 - 25)); b12 -= b7;
-            tmp = b3 ^ b10; b3 = (tmp >> 16) | (tmp << (64 - 16)); b10 -= b3;
-            tmp = b5 ^ b8; b5 = (tmp >> 28) | (tmp << (64 - 28)); b8 -= b5;
-            tmp = b1 ^ b14; b1 = (tmp >> 47) | (tmp << (64 - 47)); b14 -= b1;
-            tmp = b9 ^ b4; b9 = (tmp >> 41) | (tmp << (64 - 41)); b4 -= b9;
-            tmp = b13 ^ b6; b13 = (tmp >> 48) | (tmp << (64 - 48)); b6 -= b13;
-            tmp = b11 ^ b2; b11 = (tmp >> 20) | (tmp << (64 - 20)); b2 -= b11;
-            tmp = b15 ^ b0; b15 = (tmp >> 5) | (tmp << (64 - 5)); b0 -= b15;
-            tmp = b9 ^ b10; b9 = (tmp >> 17) | (tmp << (64 - 17)); b10 -= b9;
-            tmp = b11 ^ b8; b11 = (tmp >> 59) | (tmp << (64 - 59)); b8 -= b11;
-            tmp = b13 ^ b14; b13 = (tmp >> 41) | (tmp << (64 - 41)); b14 -= b13;
-            tmp = b15 ^ b12; b15 = (tmp >> 34) | (tmp << (64 - 34)); b12 -= b15;
-            tmp = b1 ^ b6; b1 = (tmp >> 13) | (tmp << (64 - 13)); b6 -= b1;
-            tmp = b3 ^ b4; b3 = (tmp >> 51) | (tmp << (64 - 51)); b4 -= b3;
-            tmp = b5 ^ b2; b5 = (tmp >> 4) | (tmp << (64 - 4)); b2 -= b5;
-            tmp = b7 ^ b0; b7 = (tmp >> 33) | (tmp << (64 - 33)); b0 -= b7;
-            tmp = b1 ^ b8; b1 = (tmp >> 52) | (tmp << (64 - 52)); b8 -= b1;
-            tmp = b5 ^ b14; b5 = (tmp >> 23) | (tmp << (64 - 23)); b14 -= b5;
-            tmp = b3 ^ b12; b3 = (tmp >> 18) | (tmp << (64 - 18)); b12 -= b3;
-            tmp = b7 ^ b10; b7 = (tmp >> 49) | (tmp << (64 - 49)); b10 -= b7;
-            tmp = b15 ^ b4; b15 = (tmp >> 55) | (tmp << (64 - 55)); b4 -= b15;
-            tmp = b11 ^ b6; b11 = (tmp >> 10) | (tmp << (64 - 10)); b6 -= b11;
-            tmp = b13 ^ b2; b13 = (tmp >> 19) | (tmp << (64 - 19)); b2 -= b13;
-            tmp = b9 ^ b0; b9 = (tmp >> 38) | (tmp << (64 - 38)); b0 -= b9;
-            tmp = b15 ^ b14; b15 = (tmp >> 37) | (tmp << (64 - 37)); b14 -= b15 + k14 + t1; b15 -= k15;
-            tmp = b13 ^ b12; b13 = (tmp >> 22) | (tmp << (64 - 22)); b12 -= b13 + k12; b13 -= k13 + t0;
-            tmp = b11 ^ b10; b11 = (tmp >> 17) | (tmp << (64 - 17)); b10 -= b11 + k10; b11 -= k11;
-            tmp = b9 ^ b8; b9 = (tmp >> 8) | (tmp << (64 - 8)); b8 -= b9 + k8; b9 -= k9;
-            tmp = b7 ^ b6; b7 = (tmp >> 47) | (tmp << (64 - 47)); b6 -= b7 + k6; b7 -= k7;
-            tmp = b5 ^ b4; b5 = (tmp >> 8) | (tmp << (64 - 8)); b4 -= b5 + k4; b5 -= k5;
-            tmp = b3 ^ b2; b3 = (tmp >> 13) | (tmp << (64 - 13)); b2 -= b3 + k2; b3 -= k3;
-            tmp = b1 ^ b0; b1 = (tmp >> 24) | (tmp << (64 - 24)); b0 -= b1 + k0; b1 -= k1;
+	b0 -= k3;
+	b1 -= k4;
+	b2 -= k5;
+	b3 -= k6;
+	b4 -= k7;
+	b5 -= k8;
+	b6 -= k9;
+	b7 -= k10;
+	b8 -= k11;
+	b9 -= k12;
+	b10 -= k13;
+	b11 -= k14;
+	b12 -= k15;
+	b13 -= k16 + t2;
+	b14 -= k0 + t0;
+	b15 -= k1 + 20;
+	tmp = b7 ^ b12; b7 = (tmp >> 20) | (tmp << (64 - 20)); b12 -= b7;
+	tmp = b3 ^ b10; b3 = (tmp >> 37) | (tmp << (64 - 37)); b10 -= b3;
+	tmp = b5 ^ b8; b5 = (tmp >> 31) | (tmp << (64 - 31)); b8 -= b5;
+	tmp = b1 ^ b14; b1 = (tmp >> 23) | (tmp << (64 - 23)); b14 -= b1;
+	tmp = b9 ^ b4; b9 = (tmp >> 52) | (tmp << (64 - 52)); b4 -= b9;
+	tmp = b13 ^ b6; b13 = (tmp >> 35) | (tmp << (64 - 35)); b6 -= b13;
+	tmp = b11 ^ b2; b11 = (tmp >> 48) | (tmp << (64 - 48)); b2 -= b11;
+	tmp = b15 ^ b0; b15 = (tmp >> 9) | (tmp << (64 - 9)); b0 -= b15;
+	tmp = b9 ^ b10; b9 = (tmp >> 25) | (tmp << (64 - 25)); b10 -= b9;
+	tmp = b11 ^ b8; b11 = (tmp >> 44) | (tmp << (64 - 44)); b8 -= b11;
+	tmp = b13 ^ b14; b13 = (tmp >> 42) | (tmp << (64 - 42)); b14 -= b13;
+	tmp = b15 ^ b12; b15 = (tmp >> 19) | (tmp << (64 - 19)); b12 -= b15;
+	tmp = b1 ^ b6; b1 = (tmp >> 46) | (tmp << (64 - 46)); b6 -= b1;
+	tmp = b3 ^ b4; b3 = (tmp >> 47) | (tmp << (64 - 47)); b4 -= b3;
+	tmp = b5 ^ b2; b5 = (tmp >> 44) | (tmp << (64 - 44)); b2 -= b5;
+	tmp = b7 ^ b0; b7 = (tmp >> 31) | (tmp << (64 - 31)); b0 -= b7;
+	tmp = b1 ^ b8; b1 = (tmp >> 41) | (tmp << (64 - 41)); b8 -= b1;
+	tmp = b5 ^ b14; b5 = (tmp >> 42) | (tmp << (64 - 42)); b14 -= b5;
+	tmp = b3 ^ b12; b3 = (tmp >> 53) | (tmp << (64 - 53)); b12 -= b3;
+	tmp = b7 ^ b10; b7 = (tmp >> 4) | (tmp << (64 - 4)); b10 -= b7;
+	tmp = b15 ^ b4; b15 = (tmp >> 51) | (tmp << (64 - 51)); b4 -= b15;
+	tmp = b11 ^ b6; b11 = (tmp >> 56) | (tmp << (64 - 56)); b6 -= b11;
+	tmp = b13 ^ b2; b13 = (tmp >> 34) | (tmp << (64 - 34)); b2 -= b13;
+	tmp = b9 ^ b0; b9 = (tmp >> 16) | (tmp << (64 - 16)); b0 -= b9;
+	tmp = b15 ^ b14; b15 = (tmp >> 30) | (tmp << (64 - 30)); b14 -= b15 + k16 + t2; b15 -= k0 + 19;
+	tmp = b13 ^ b12; b13 = (tmp >> 44) | (tmp << (64 - 44)); b12 -= b13 + k14; b13 -= k15 + t1;
+	tmp = b11 ^ b10; b11 = (tmp >> 47) | (tmp << (64 - 47)); b10 -= b11 + k12; b11 -= k13;
+	tmp = b9 ^ b8; b9 = (tmp >> 12) | (tmp << (64 - 12)); b8 -= b9 + k10; b9 -= k11;
+	tmp = b7 ^ b6; b7 = (tmp >> 31) | (tmp << (64 - 31)); b6 -= b7 + k8; b7 -= k9;
+	tmp = b5 ^ b4; b5 = (tmp >> 37) | (tmp << (64 - 37)); b4 -= b5 + k6; b5 -= k7;
+	tmp = b3 ^ b2; b3 = (tmp >> 9) | (tmp << (64 - 9)); b2 -= b3 + k4; b3 -= k5;
+	tmp = b1 ^ b0; b1 = (tmp >> 41) | (tmp << (64 - 41)); b0 -= b1 + k2; b1 -= k3;
+	tmp = b7 ^ b12; b7 = (tmp >> 25) | (tmp << (64 - 25)); b12 -= b7;
+	tmp = b3 ^ b10; b3 = (tmp >> 16) | (tmp << (64 - 16)); b10 -= b3;
+	tmp = b5 ^ b8; b5 = (tmp >> 28) | (tmp << (64 - 28)); b8 -= b5;
+	tmp = b1 ^ b14; b1 = (tmp >> 47) | (tmp << (64 - 47)); b14 -= b1;
+	tmp = b9 ^ b4; b9 = (tmp >> 41) | (tmp << (64 - 41)); b4 -= b9;
+	tmp = b13 ^ b6; b13 = (tmp >> 48) | (tmp << (64 - 48)); b6 -= b13;
+	tmp = b11 ^ b2; b11 = (tmp >> 20) | (tmp << (64 - 20)); b2 -= b11;
+	tmp = b15 ^ b0; b15 = (tmp >> 5) | (tmp << (64 - 5)); b0 -= b15;
+	tmp = b9 ^ b10; b9 = (tmp >> 17) | (tmp << (64 - 17)); b10 -= b9;
+	tmp = b11 ^ b8; b11 = (tmp >> 59) | (tmp << (64 - 59)); b8 -= b11;
+	tmp = b13 ^ b14; b13 = (tmp >> 41) | (tmp << (64 - 41)); b14 -= b13;
+	tmp = b15 ^ b12; b15 = (tmp >> 34) | (tmp << (64 - 34)); b12 -= b15;
+	tmp = b1 ^ b6; b1 = (tmp >> 13) | (tmp << (64 - 13)); b6 -= b1;
+	tmp = b3 ^ b4; b3 = (tmp >> 51) | (tmp << (64 - 51)); b4 -= b3;
+	tmp = b5 ^ b2; b5 = (tmp >> 4) | (tmp << (64 - 4)); b2 -= b5;
+	tmp = b7 ^ b0; b7 = (tmp >> 33) | (tmp << (64 - 33)); b0 -= b7;
+	tmp = b1 ^ b8; b1 = (tmp >> 52) | (tmp << (64 - 52)); b8 -= b1;
+	tmp = b5 ^ b14; b5 = (tmp >> 23) | (tmp << (64 - 23)); b14 -= b5;
+	tmp = b3 ^ b12; b3 = (tmp >> 18) | (tmp << (64 - 18)); b12 -= b3;
+	tmp = b7 ^ b10; b7 = (tmp >> 49) | (tmp << (64 - 49)); b10 -= b7;
+	tmp = b15 ^ b4; b15 = (tmp >> 55) | (tmp << (64 - 55)); b4 -= b15;
+	tmp = b11 ^ b6; b11 = (tmp >> 10) | (tmp << (64 - 10)); b6 -= b11;
+	tmp = b13 ^ b2; b13 = (tmp >> 19) | (tmp << (64 - 19)); b2 -= b13;
+	tmp = b9 ^ b0; b9 = (tmp >> 38) | (tmp << (64 - 38)); b0 -= b9;
+	tmp = b15 ^ b14; b15 = (tmp >> 37) | (tmp << (64 - 37)); b14 -= b15 + k15 + t1; b15 -= k16 + 18;
+	tmp = b13 ^ b12; b13 = (tmp >> 22) | (tmp << (64 - 22)); b12 -= b13 + k13; b13 -= k14 + t0;
+	tmp = b11 ^ b10; b11 = (tmp >> 17) | (tmp << (64 - 17)); b10 -= b11 + k11; b11 -= k12;
+	tmp = b9 ^ b8; b9 = (tmp >> 8) | (tmp << (64 - 8)); b8 -= b9 + k9; b9 -= k10;
+	tmp = b7 ^ b6; b7 = (tmp >> 47) | (tmp << (64 - 47)); b6 -= b7 + k7; b7 -= k8;
+	tmp = b5 ^ b4; b5 = (tmp >> 8) | (tmp << (64 - 8)); b4 -= b5 + k5; b5 -= k6;
+	tmp = b3 ^ b2; b3 = (tmp >> 13) | (tmp << (64 - 13)); b2 -= b3 + k3; b3 -= k4;
+	tmp = b1 ^ b0; b1 = (tmp >> 24) | (tmp << (64 - 24)); b0 -= b1 + k1; b1 -= k2;
+	tmp = b7 ^ b12; b7 = (tmp >> 20) | (tmp << (64 - 20)); b12 -= b7;
+	tmp = b3 ^ b10; b3 = (tmp >> 37) | (tmp << (64 - 37)); b10 -= b3;
+	tmp = b5 ^ b8; b5 = (tmp >> 31) | (tmp << (64 - 31)); b8 -= b5;
+	tmp = b1 ^ b14; b1 = (tmp >> 23) | (tmp << (64 - 23)); b14 -= b1;
+	tmp = b9 ^ b4; b9 = (tmp >> 52) | (tmp << (64 - 52)); b4 -= b9;
+	tmp = b13 ^ b6; b13 = (tmp >> 35) | (tmp << (64 - 35)); b6 -= b13;
+	tmp = b11 ^ b2; b11 = (tmp >> 48) | (tmp << (64 - 48)); b2 -= b11;
+	tmp = b15 ^ b0; b15 = (tmp >> 9) | (tmp << (64 - 9)); b0 -= b15;
+	tmp = b9 ^ b10; b9 = (tmp >> 25) | (tmp << (64 - 25)); b10 -= b9;
+	tmp = b11 ^ b8; b11 = (tmp >> 44) | (tmp << (64 - 44)); b8 -= b11;
+	tmp = b13 ^ b14; b13 = (tmp >> 42) | (tmp << (64 - 42)); b14 -= b13;
+	tmp = b15 ^ b12; b15 = (tmp >> 19) | (tmp << (64 - 19)); b12 -= b15;
+	tmp = b1 ^ b6; b1 = (tmp >> 46) | (tmp << (64 - 46)); b6 -= b1;
+	tmp = b3 ^ b4; b3 = (tmp >> 47) | (tmp << (64 - 47)); b4 -= b3;
+	tmp = b5 ^ b2; b5 = (tmp >> 44) | (tmp << (64 - 44)); b2 -= b5;
+	tmp = b7 ^ b0; b7 = (tmp >> 31) | (tmp << (64 - 31)); b0 -= b7;
+	tmp = b1 ^ b8; b1 = (tmp >> 41) | (tmp << (64 - 41)); b8 -= b1;
+	tmp = b5 ^ b14; b5 = (tmp >> 42) | (tmp << (64 - 42)); b14 -= b5;
+	tmp = b3 ^ b12; b3 = (tmp >> 53) | (tmp << (64 - 53)); b12 -= b3;
+	tmp = b7 ^ b10; b7 = (tmp >> 4) | (tmp << (64 - 4)); b10 -= b7;
+	tmp = b15 ^ b4; b15 = (tmp >> 51) | (tmp << (64 - 51)); b4 -= b15;
+	tmp = b11 ^ b6; b11 = (tmp >> 56) | (tmp << (64 - 56)); b6 -= b11;
+	tmp = b13 ^ b2; b13 = (tmp >> 34) | (tmp << (64 - 34)); b2 -= b13;
+	tmp = b9 ^ b0; b9 = (tmp >> 16) | (tmp << (64 - 16)); b0 -= b9;
+	tmp = b15 ^ b14; b15 = (tmp >> 30) | (tmp << (64 - 30)); b14 -= b15 + k14 + t0; b15 -= k15 + 17;
+	tmp = b13 ^ b12; b13 = (tmp >> 44) | (tmp << (64 - 44)); b12 -= b13 + k12; b13 -= k13 + t2;
+	tmp = b11 ^ b10; b11 = (tmp >> 47) | (tmp << (64 - 47)); b10 -= b11 + k10; b11 -= k11;
+	tmp = b9 ^ b8; b9 = (tmp >> 12) | (tmp << (64 - 12)); b8 -= b9 + k8; b9 -= k9;
+	tmp = b7 ^ b6; b7 = (tmp >> 31) | (tmp << (64 - 31)); b6 -= b7 + k6; b7 -= k7;
+	tmp = b5 ^ b4; b5 = (tmp >> 37) | (tmp << (64 - 37)); b4 -= b5 + k4; b5 -= k5;
+	tmp = b3 ^ b2; b3 = (tmp >> 9) | (tmp << (64 - 9)); b2 -= b3 + k2; b3 -= k3;
+	tmp = b1 ^ b0; b1 = (tmp >> 41) | (tmp << (64 - 41)); b0 -= b1 + k0; b1 -= k1;
+	tmp = b7 ^ b12; b7 = (tmp >> 25) | (tmp << (64 - 25)); b12 -= b7;
+	tmp = b3 ^ b10; b3 = (tmp >> 16) | (tmp << (64 - 16)); b10 -= b3;
+	tmp = b5 ^ b8; b5 = (tmp >> 28) | (tmp << (64 - 28)); b8 -= b5;
+	tmp = b1 ^ b14; b1 = (tmp >> 47) | (tmp << (64 - 47)); b14 -= b1;
+	tmp = b9 ^ b4; b9 = (tmp >> 41) | (tmp << (64 - 41)); b4 -= b9;
+	tmp = b13 ^ b6; b13 = (tmp >> 48) | (tmp << (64 - 48)); b6 -= b13;
+	tmp = b11 ^ b2; b11 = (tmp >> 20) | (tmp << (64 - 20)); b2 -= b11;
+	tmp = b15 ^ b0; b15 = (tmp >> 5) | (tmp << (64 - 5)); b0 -= b15;
+	tmp = b9 ^ b10; b9 = (tmp >> 17) | (tmp << (64 - 17)); b10 -= b9;
+	tmp = b11 ^ b8; b11 = (tmp >> 59) | (tmp << (64 - 59)); b8 -= b11;
+	tmp = b13 ^ b14; b13 = (tmp >> 41) | (tmp << (64 - 41)); b14 -= b13;
+	tmp = b15 ^ b12; b15 = (tmp >> 34) | (tmp << (64 - 34)); b12 -= b15;
+	tmp = b1 ^ b6; b1 = (tmp >> 13) | (tmp << (64 - 13)); b6 -= b1;
+	tmp = b3 ^ b4; b3 = (tmp >> 51) | (tmp << (64 - 51)); b4 -= b3;
+	tmp = b5 ^ b2; b5 = (tmp >> 4) | (tmp << (64 - 4)); b2 -= b5;
+	tmp = b7 ^ b0; b7 = (tmp >> 33) | (tmp << (64 - 33)); b0 -= b7;
+	tmp = b1 ^ b8; b1 = (tmp >> 52) | (tmp << (64 - 52)); b8 -= b1;
+	tmp = b5 ^ b14; b5 = (tmp >> 23) | (tmp << (64 - 23)); b14 -= b5;
+	tmp = b3 ^ b12; b3 = (tmp >> 18) | (tmp << (64 - 18)); b12 -= b3;
+	tmp = b7 ^ b10; b7 = (tmp >> 49) | (tmp << (64 - 49)); b10 -= b7;
+	tmp = b15 ^ b4; b15 = (tmp >> 55) | (tmp << (64 - 55)); b4 -= b15;
+	tmp = b11 ^ b6; b11 = (tmp >> 10) | (tmp << (64 - 10)); b6 -= b11;
+	tmp = b13 ^ b2; b13 = (tmp >> 19) | (tmp << (64 - 19)); b2 -= b13;
+	tmp = b9 ^ b0; b9 = (tmp >> 38) | (tmp << (64 - 38)); b0 -= b9;
+	tmp = b15 ^ b14; b15 = (tmp >> 37) | (tmp << (64 - 37)); b14 -= b15 + k13 + t2; b15 -= k14 + 16;
+	tmp = b13 ^ b12; b13 = (tmp >> 22) | (tmp << (64 - 22)); b12 -= b13 + k11; b13 -= k12 + t1;
+	tmp = b11 ^ b10; b11 = (tmp >> 17) | (tmp << (64 - 17)); b10 -= b11 + k9; b11 -= k10;
+	tmp = b9 ^ b8; b9 = (tmp >> 8) | (tmp << (64 - 8)); b8 -= b9 + k7; b9 -= k8;
+	tmp = b7 ^ b6; b7 = (tmp >> 47) | (tmp << (64 - 47)); b6 -= b7 + k5; b7 -= k6;
+	tmp = b5 ^ b4; b5 = (tmp >> 8) | (tmp << (64 - 8)); b4 -= b5 + k3; b5 -= k4;
+	tmp = b3 ^ b2; b3 = (tmp >> 13) | (tmp << (64 - 13)); b2 -= b3 + k1; b3 -= k2;
+	tmp = b1 ^ b0; b1 = (tmp >> 24) | (tmp << (64 - 24)); b0 -= b1 + k16; b1 -= k0;
+	tmp = b7 ^ b12; b7 = (tmp >> 20) | (tmp << (64 - 20)); b12 -= b7;
+	tmp = b3 ^ b10; b3 = (tmp >> 37) | (tmp << (64 - 37)); b10 -= b3;
+	tmp = b5 ^ b8; b5 = (tmp >> 31) | (tmp << (64 - 31)); b8 -= b5;
+	tmp = b1 ^ b14; b1 = (tmp >> 23) | (tmp << (64 - 23)); b14 -= b1;
+	tmp = b9 ^ b4; b9 = (tmp >> 52) | (tmp << (64 - 52)); b4 -= b9;
+	tmp = b13 ^ b6; b13 = (tmp >> 35) | (tmp << (64 - 35)); b6 -= b13;
+	tmp = b11 ^ b2; b11 = (tmp >> 48) | (tmp << (64 - 48)); b2 -= b11;
+	tmp = b15 ^ b0; b15 = (tmp >> 9) | (tmp << (64 - 9)); b0 -= b15;
+	tmp = b9 ^ b10; b9 = (tmp >> 25) | (tmp << (64 - 25)); b10 -= b9;
+	tmp = b11 ^ b8; b11 = (tmp >> 44) | (tmp << (64 - 44)); b8 -= b11;
+	tmp = b13 ^ b14; b13 = (tmp >> 42) | (tmp << (64 - 42)); b14 -= b13;
+	tmp = b15 ^ b12; b15 = (tmp >> 19) | (tmp << (64 - 19)); b12 -= b15;
+	tmp = b1 ^ b6; b1 = (tmp >> 46) | (tmp << (64 - 46)); b6 -= b1;
+	tmp = b3 ^ b4; b3 = (tmp >> 47) | (tmp << (64 - 47)); b4 -= b3;
+	tmp = b5 ^ b2; b5 = (tmp >> 44) | (tmp << (64 - 44)); b2 -= b5;
+	tmp = b7 ^ b0; b7 = (tmp >> 31) | (tmp << (64 - 31)); b0 -= b7;
+	tmp = b1 ^ b8; b1 = (tmp >> 41) | (tmp << (64 - 41)); b8 -= b1;
+	tmp = b5 ^ b14; b5 = (tmp >> 42) | (tmp << (64 - 42)); b14 -= b5;
+	tmp = b3 ^ b12; b3 = (tmp >> 53) | (tmp << (64 - 53)); b12 -= b3;
+	tmp = b7 ^ b10; b7 = (tmp >> 4) | (tmp << (64 - 4)); b10 -= b7;
+	tmp = b15 ^ b4; b15 = (tmp >> 51) | (tmp << (64 - 51)); b4 -= b15;
+	tmp = b11 ^ b6; b11 = (tmp >> 56) | (tmp << (64 - 56)); b6 -= b11;
+	tmp = b13 ^ b2; b13 = (tmp >> 34) | (tmp << (64 - 34)); b2 -= b13;
+	tmp = b9 ^ b0; b9 = (tmp >> 16) | (tmp << (64 - 16)); b0 -= b9;
+	tmp = b15 ^ b14; b15 = (tmp >> 30) | (tmp << (64 - 30)); b14 -= b15 + k12 + t1; b15 -= k13 + 15;
+	tmp = b13 ^ b12; b13 = (tmp >> 44) | (tmp << (64 - 44)); b12 -= b13 + k10; b13 -= k11 + t0;
+	tmp = b11 ^ b10; b11 = (tmp >> 47) | (tmp << (64 - 47)); b10 -= b11 + k8; b11 -= k9;
+	tmp = b9 ^ b8; b9 = (tmp >> 12) | (tmp << (64 - 12)); b8 -= b9 + k6; b9 -= k7;
+	tmp = b7 ^ b6; b7 = (tmp >> 31) | (tmp << (64 - 31)); b6 -= b7 + k4; b7 -= k5;
+	tmp = b5 ^ b4; b5 = (tmp >> 37) | (tmp << (64 - 37)); b4 -= b5 + k2; b5 -= k3;
+	tmp = b3 ^ b2; b3 = (tmp >> 9) | (tmp << (64 - 9)); b2 -= b3 + k0; b3 -= k1;
+	tmp = b1 ^ b0; b1 = (tmp >> 41) | (tmp << (64 - 41)); b0 -= b1 + k15; b1 -= k16;
+	tmp = b7 ^ b12; b7 = (tmp >> 25) | (tmp << (64 - 25)); b12 -= b7;
+	tmp = b3 ^ b10; b3 = (tmp >> 16) | (tmp << (64 - 16)); b10 -= b3;
+	tmp = b5 ^ b8; b5 = (tmp >> 28) | (tmp << (64 - 28)); b8 -= b5;
+	tmp = b1 ^ b14; b1 = (tmp >> 47) | (tmp << (64 - 47)); b14 -= b1;
+	tmp = b9 ^ b4; b9 = (tmp >> 41) | (tmp << (64 - 41)); b4 -= b9;
+	tmp = b13 ^ b6; b13 = (tmp >> 48) | (tmp << (64 - 48)); b6 -= b13;
+	tmp = b11 ^ b2; b11 = (tmp >> 20) | (tmp << (64 - 20)); b2 -= b11;
+	tmp = b15 ^ b0; b15 = (tmp >> 5) | (tmp << (64 - 5)); b0 -= b15;
+	tmp = b9 ^ b10; b9 = (tmp >> 17) | (tmp << (64 - 17)); b10 -= b9;
+	tmp = b11 ^ b8; b11 = (tmp >> 59) | (tmp << (64 - 59)); b8 -= b11;
+	tmp = b13 ^ b14; b13 = (tmp >> 41) | (tmp << (64 - 41)); b14 -= b13;
+	tmp = b15 ^ b12; b15 = (tmp >> 34) | (tmp << (64 - 34)); b12 -= b15;
+	tmp = b1 ^ b6; b1 = (tmp >> 13) | (tmp << (64 - 13)); b6 -= b1;
+	tmp = b3 ^ b4; b3 = (tmp >> 51) | (tmp << (64 - 51)); b4 -= b3;
+	tmp = b5 ^ b2; b5 = (tmp >> 4) | (tmp << (64 - 4)); b2 -= b5;
+	tmp = b7 ^ b0; b7 = (tmp >> 33) | (tmp << (64 - 33)); b0 -= b7;
+	tmp = b1 ^ b8; b1 = (tmp >> 52) | (tmp << (64 - 52)); b8 -= b1;
+	tmp = b5 ^ b14; b5 = (tmp >> 23) | (tmp << (64 - 23)); b14 -= b5;
+	tmp = b3 ^ b12; b3 = (tmp >> 18) | (tmp << (64 - 18)); b12 -= b3;
+	tmp = b7 ^ b10; b7 = (tmp >> 49) | (tmp << (64 - 49)); b10 -= b7;
+	tmp = b15 ^ b4; b15 = (tmp >> 55) | (tmp << (64 - 55)); b4 -= b15;
+	tmp = b11 ^ b6; b11 = (tmp >> 10) | (tmp << (64 - 10)); b6 -= b11;
+	tmp = b13 ^ b2; b13 = (tmp >> 19) | (tmp << (64 - 19)); b2 -= b13;
+	tmp = b9 ^ b0; b9 = (tmp >> 38) | (tmp << (64 - 38)); b0 -= b9;
+	tmp = b15 ^ b14; b15 = (tmp >> 37) | (tmp << (64 - 37)); b14 -= b15 + k11 + t0; b15 -= k12 + 14;
+	tmp = b13 ^ b12; b13 = (tmp >> 22) | (tmp << (64 - 22)); b12 -= b13 + k9; b13 -= k10 + t2;
+	tmp = b11 ^ b10; b11 = (tmp >> 17) | (tmp << (64 - 17)); b10 -= b11 + k7; b11 -= k8;
+	tmp = b9 ^ b8; b9 = (tmp >> 8) | (tmp << (64 - 8)); b8 -= b9 + k5; b9 -= k6;
+	tmp = b7 ^ b6; b7 = (tmp >> 47) | (tmp << (64 - 47)); b6 -= b7 + k3; b7 -= k4;
+	tmp = b5 ^ b4; b5 = (tmp >> 8) | (tmp << (64 - 8)); b4 -= b5 + k1; b5 -= k2;
+	tmp = b3 ^ b2; b3 = (tmp >> 13) | (tmp << (64 - 13)); b2 -= b3 + k16; b3 -= k0;
+	tmp = b1 ^ b0; b1 = (tmp >> 24) | (tmp << (64 - 24)); b0 -= b1 + k14; b1 -= k15;
+	tmp = b7 ^ b12; b7 = (tmp >> 20) | (tmp << (64 - 20)); b12 -= b7;
+	tmp = b3 ^ b10; b3 = (tmp >> 37) | (tmp << (64 - 37)); b10 -= b3;
+	tmp = b5 ^ b8; b5 = (tmp >> 31) | (tmp << (64 - 31)); b8 -= b5;
+	tmp = b1 ^ b14; b1 = (tmp >> 23) | (tmp << (64 - 23)); b14 -= b1;
+	tmp = b9 ^ b4; b9 = (tmp >> 52) | (tmp << (64 - 52)); b4 -= b9;
+	tmp = b13 ^ b6; b13 = (tmp >> 35) | (tmp << (64 - 35)); b6 -= b13;
+	tmp = b11 ^ b2; b11 = (tmp >> 48) | (tmp << (64 - 48)); b2 -= b11;
+	tmp = b15 ^ b0; b15 = (tmp >> 9) | (tmp << (64 - 9)); b0 -= b15;
+	tmp = b9 ^ b10; b9 = (tmp >> 25) | (tmp << (64 - 25)); b10 -= b9;
+	tmp = b11 ^ b8; b11 = (tmp >> 44) | (tmp << (64 - 44)); b8 -= b11;
+	tmp = b13 ^ b14; b13 = (tmp >> 42) | (tmp << (64 - 42)); b14 -= b13;
+	tmp = b15 ^ b12; b15 = (tmp >> 19) | (tmp << (64 - 19)); b12 -= b15;
+	tmp = b1 ^ b6; b1 = (tmp >> 46) | (tmp << (64 - 46)); b6 -= b1;
+	tmp = b3 ^ b4; b3 = (tmp >> 47) | (tmp << (64 - 47)); b4 -= b3;
+	tmp = b5 ^ b2; b5 = (tmp >> 44) | (tmp << (64 - 44)); b2 -= b5;
+	tmp = b7 ^ b0; b7 = (tmp >> 31) | (tmp << (64 - 31)); b0 -= b7;
+	tmp = b1 ^ b8; b1 = (tmp >> 41) | (tmp << (64 - 41)); b8 -= b1;
+	tmp = b5 ^ b14; b5 = (tmp >> 42) | (tmp << (64 - 42)); b14 -= b5;
+	tmp = b3 ^ b12; b3 = (tmp >> 53) | (tmp << (64 - 53)); b12 -= b3;
+	tmp = b7 ^ b10; b7 = (tmp >> 4) | (tmp << (64 - 4)); b10 -= b7;
+	tmp = b15 ^ b4; b15 = (tmp >> 51) | (tmp << (64 - 51)); b4 -= b15;
+	tmp = b11 ^ b6; b11 = (tmp >> 56) | (tmp << (64 - 56)); b6 -= b11;
+	tmp = b13 ^ b2; b13 = (tmp >> 34) | (tmp << (64 - 34)); b2 -= b13;
+	tmp = b9 ^ b0; b9 = (tmp >> 16) | (tmp << (64 - 16)); b0 -= b9;
+	tmp = b15 ^ b14; b15 = (tmp >> 30) | (tmp << (64 - 30)); b14 -= b15 + k10 + t2; b15 -= k11 + 13;
+	tmp = b13 ^ b12; b13 = (tmp >> 44) | (tmp << (64 - 44)); b12 -= b13 + k8; b13 -= k9 + t1;
+	tmp = b11 ^ b10; b11 = (tmp >> 47) | (tmp << (64 - 47)); b10 -= b11 + k6; b11 -= k7;
+	tmp = b9 ^ b8; b9 = (tmp >> 12) | (tmp << (64 - 12)); b8 -= b9 + k4; b9 -= k5;
+	tmp = b7 ^ b6; b7 = (tmp >> 31) | (tmp << (64 - 31)); b6 -= b7 + k2; b7 -= k3;
+	tmp = b5 ^ b4; b5 = (tmp >> 37) | (tmp << (64 - 37)); b4 -= b5 + k0; b5 -= k1;
+	tmp = b3 ^ b2; b3 = (tmp >> 9) | (tmp << (64 - 9)); b2 -= b3 + k15; b3 -= k16;
+	tmp = b1 ^ b0; b1 = (tmp >> 41) | (tmp << (64 - 41)); b0 -= b1 + k13; b1 -= k14;
+	tmp = b7 ^ b12; b7 = (tmp >> 25) | (tmp << (64 - 25)); b12 -= b7;
+	tmp = b3 ^ b10; b3 = (tmp >> 16) | (tmp << (64 - 16)); b10 -= b3;
+	tmp = b5 ^ b8; b5 = (tmp >> 28) | (tmp << (64 - 28)); b8 -= b5;
+	tmp = b1 ^ b14; b1 = (tmp >> 47) | (tmp << (64 - 47)); b14 -= b1;
+	tmp = b9 ^ b4; b9 = (tmp >> 41) | (tmp << (64 - 41)); b4 -= b9;
+	tmp = b13 ^ b6; b13 = (tmp >> 48) | (tmp << (64 - 48)); b6 -= b13;
+	tmp = b11 ^ b2; b11 = (tmp >> 20) | (tmp << (64 - 20)); b2 -= b11;
+	tmp = b15 ^ b0; b15 = (tmp >> 5) | (tmp << (64 - 5)); b0 -= b15;
+	tmp = b9 ^ b10; b9 = (tmp >> 17) | (tmp << (64 - 17)); b10 -= b9;
+	tmp = b11 ^ b8; b11 = (tmp >> 59) | (tmp << (64 - 59)); b8 -= b11;
+	tmp = b13 ^ b14; b13 = (tmp >> 41) | (tmp << (64 - 41)); b14 -= b13;
+	tmp = b15 ^ b12; b15 = (tmp >> 34) | (tmp << (64 - 34)); b12 -= b15;
+	tmp = b1 ^ b6; b1 = (tmp >> 13) | (tmp << (64 - 13)); b6 -= b1;
+	tmp = b3 ^ b4; b3 = (tmp >> 51) | (tmp << (64 - 51)); b4 -= b3;
+	tmp = b5 ^ b2; b5 = (tmp >> 4) | (tmp << (64 - 4)); b2 -= b5;
+	tmp = b7 ^ b0; b7 = (tmp >> 33) | (tmp << (64 - 33)); b0 -= b7;
+	tmp = b1 ^ b8; b1 = (tmp >> 52) | (tmp << (64 - 52)); b8 -= b1;
+	tmp = b5 ^ b14; b5 = (tmp >> 23) | (tmp << (64 - 23)); b14 -= b5;
+	tmp = b3 ^ b12; b3 = (tmp >> 18) | (tmp << (64 - 18)); b12 -= b3;
+	tmp = b7 ^ b10; b7 = (tmp >> 49) | (tmp << (64 - 49)); b10 -= b7;
+	tmp = b15 ^ b4; b15 = (tmp >> 55) | (tmp << (64 - 55)); b4 -= b15;
+	tmp = b11 ^ b6; b11 = (tmp >> 10) | (tmp << (64 - 10)); b6 -= b11;
+	tmp = b13 ^ b2; b13 = (tmp >> 19) | (tmp << (64 - 19)); b2 -= b13;
+	tmp = b9 ^ b0; b9 = (tmp >> 38) | (tmp << (64 - 38)); b0 -= b9;
+	tmp = b15 ^ b14; b15 = (tmp >> 37) | (tmp << (64 - 37)); b14 -= b15 + k9 + t1; b15 -= k10 + 12;
+	tmp = b13 ^ b12; b13 = (tmp >> 22) | (tmp << (64 - 22)); b12 -= b13 + k7; b13 -= k8 + t0;
+	tmp = b11 ^ b10; b11 = (tmp >> 17) | (tmp << (64 - 17)); b10 -= b11 + k5; b11 -= k6;
+	tmp = b9 ^ b8; b9 = (tmp >> 8) | (tmp << (64 - 8)); b8 -= b9 + k3; b9 -= k4;
+	tmp = b7 ^ b6; b7 = (tmp >> 47) | (tmp << (64 - 47)); b6 -= b7 + k1; b7 -= k2;
+	tmp = b5 ^ b4; b5 = (tmp >> 8) | (tmp << (64 - 8)); b4 -= b5 + k16; b5 -= k0;
+	tmp = b3 ^ b2; b3 = (tmp >> 13) | (tmp << (64 - 13)); b2 -= b3 + k14; b3 -= k15;
+	tmp = b1 ^ b0; b1 = (tmp >> 24) | (tmp << (64 - 24)); b0 -= b1 + k12; b1 -= k13;
+	tmp = b7 ^ b12; b7 = (tmp >> 20) | (tmp << (64 - 20)); b12 -= b7;
+	tmp = b3 ^ b10; b3 = (tmp >> 37) | (tmp << (64 - 37)); b10 -= b3;
+	tmp = b5 ^ b8; b5 = (tmp >> 31) | (tmp << (64 - 31)); b8 -= b5;
+	tmp = b1 ^ b14; b1 = (tmp >> 23) | (tmp << (64 - 23)); b14 -= b1;
+	tmp = b9 ^ b4; b9 = (tmp >> 52) | (tmp << (64 - 52)); b4 -= b9;
+	tmp = b13 ^ b6; b13 = (tmp >> 35) | (tmp << (64 - 35)); b6 -= b13;
+	tmp = b11 ^ b2; b11 = (tmp >> 48) | (tmp << (64 - 48)); b2 -= b11;
+	tmp = b15 ^ b0; b15 = (tmp >> 9) | (tmp << (64 - 9)); b0 -= b15;
+	tmp = b9 ^ b10; b9 = (tmp >> 25) | (tmp << (64 - 25)); b10 -= b9;
+	tmp = b11 ^ b8; b11 = (tmp >> 44) | (tmp << (64 - 44)); b8 -= b11;
+	tmp = b13 ^ b14; b13 = (tmp >> 42) | (tmp << (64 - 42)); b14 -= b13;
+	tmp = b15 ^ b12; b15 = (tmp >> 19) | (tmp << (64 - 19)); b12 -= b15;
+	tmp = b1 ^ b6; b1 = (tmp >> 46) | (tmp << (64 - 46)); b6 -= b1;
+	tmp = b3 ^ b4; b3 = (tmp >> 47) | (tmp << (64 - 47)); b4 -= b3;
+	tmp = b5 ^ b2; b5 = (tmp >> 44) | (tmp << (64 - 44)); b2 -= b5;
+	tmp = b7 ^ b0; b7 = (tmp >> 31) | (tmp << (64 - 31)); b0 -= b7;
+	tmp = b1 ^ b8; b1 = (tmp >> 41) | (tmp << (64 - 41)); b8 -= b1;
+	tmp = b5 ^ b14; b5 = (tmp >> 42) | (tmp << (64 - 42)); b14 -= b5;
+	tmp = b3 ^ b12; b3 = (tmp >> 53) | (tmp << (64 - 53)); b12 -= b3;
+	tmp = b7 ^ b10; b7 = (tmp >> 4) | (tmp << (64 - 4)); b10 -= b7;
+	tmp = b15 ^ b4; b15 = (tmp >> 51) | (tmp << (64 - 51)); b4 -= b15;
+	tmp = b11 ^ b6; b11 = (tmp >> 56) | (tmp << (64 - 56)); b6 -= b11;
+	tmp = b13 ^ b2; b13 = (tmp >> 34) | (tmp << (64 - 34)); b2 -= b13;
+	tmp = b9 ^ b0; b9 = (tmp >> 16) | (tmp << (64 - 16)); b0 -= b9;
+	tmp = b15 ^ b14; b15 = (tmp >> 30) | (tmp << (64 - 30)); b14 -= b15 + k8 + t0; b15 -= k9 + 11;
+	tmp = b13 ^ b12; b13 = (tmp >> 44) | (tmp << (64 - 44)); b12 -= b13 + k6; b13 -= k7 + t2;
+	tmp = b11 ^ b10; b11 = (tmp >> 47) | (tmp << (64 - 47)); b10 -= b11 + k4; b11 -= k5;
+	tmp = b9 ^ b8; b9 = (tmp >> 12) | (tmp << (64 - 12)); b8 -= b9 + k2; b9 -= k3;
+	tmp = b7 ^ b6; b7 = (tmp >> 31) | (tmp << (64 - 31)); b6 -= b7 + k0; b7 -= k1;
+	tmp = b5 ^ b4; b5 = (tmp >> 37) | (tmp << (64 - 37)); b4 -= b5 + k15; b5 -= k16;
+	tmp = b3 ^ b2; b3 = (tmp >> 9) | (tmp << (64 - 9)); b2 -= b3 + k13; b3 -= k14;
+	tmp = b1 ^ b0; b1 = (tmp >> 41) | (tmp << (64 - 41)); b0 -= b1 + k11; b1 -= k12;
+	tmp = b7 ^ b12; b7 = (tmp >> 25) | (tmp << (64 - 25)); b12 -= b7;
+	tmp = b3 ^ b10; b3 = (tmp >> 16) | (tmp << (64 - 16)); b10 -= b3;
+	tmp = b5 ^ b8; b5 = (tmp >> 28) | (tmp << (64 - 28)); b8 -= b5;
+	tmp = b1 ^ b14; b1 = (tmp >> 47) | (tmp << (64 - 47)); b14 -= b1;
+	tmp = b9 ^ b4; b9 = (tmp >> 41) | (tmp << (64 - 41)); b4 -= b9;
+	tmp = b13 ^ b6; b13 = (tmp >> 48) | (tmp << (64 - 48)); b6 -= b13;
+	tmp = b11 ^ b2; b11 = (tmp >> 20) | (tmp << (64 - 20)); b2 -= b11;
+	tmp = b15 ^ b0; b15 = (tmp >> 5) | (tmp << (64 - 5)); b0 -= b15;
+	tmp = b9 ^ b10; b9 = (tmp >> 17) | (tmp << (64 - 17)); b10 -= b9;
+	tmp = b11 ^ b8; b11 = (tmp >> 59) | (tmp << (64 - 59)); b8 -= b11;
+	tmp = b13 ^ b14; b13 = (tmp >> 41) | (tmp << (64 - 41)); b14 -= b13;
+	tmp = b15 ^ b12; b15 = (tmp >> 34) | (tmp << (64 - 34)); b12 -= b15;
+	tmp = b1 ^ b6; b1 = (tmp >> 13) | (tmp << (64 - 13)); b6 -= b1;
+	tmp = b3 ^ b4; b3 = (tmp >> 51) | (tmp << (64 - 51)); b4 -= b3;
+	tmp = b5 ^ b2; b5 = (tmp >> 4) | (tmp << (64 - 4)); b2 -= b5;
+	tmp = b7 ^ b0; b7 = (tmp >> 33) | (tmp << (64 - 33)); b0 -= b7;
+	tmp = b1 ^ b8; b1 = (tmp >> 52) | (tmp << (64 - 52)); b8 -= b1;
+	tmp = b5 ^ b14; b5 = (tmp >> 23) | (tmp << (64 - 23)); b14 -= b5;
+	tmp = b3 ^ b12; b3 = (tmp >> 18) | (tmp << (64 - 18)); b12 -= b3;
+	tmp = b7 ^ b10; b7 = (tmp >> 49) | (tmp << (64 - 49)); b10 -= b7;
+	tmp = b15 ^ b4; b15 = (tmp >> 55) | (tmp << (64 - 55)); b4 -= b15;
+	tmp = b11 ^ b6; b11 = (tmp >> 10) | (tmp << (64 - 10)); b6 -= b11;
+	tmp = b13 ^ b2; b13 = (tmp >> 19) | (tmp << (64 - 19)); b2 -= b13;
+	tmp = b9 ^ b0; b9 = (tmp >> 38) | (tmp << (64 - 38)); b0 -= b9;
+	tmp = b15 ^ b14; b15 = (tmp >> 37) | (tmp << (64 - 37)); b14 -= b15 + k7 + t2; b15 -= k8 + 10;
+	tmp = b13 ^ b12; b13 = (tmp >> 22) | (tmp << (64 - 22)); b12 -= b13 + k5; b13 -= k6 + t1;
+	tmp = b11 ^ b10; b11 = (tmp >> 17) | (tmp << (64 - 17)); b10 -= b11 + k3; b11 -= k4;
+	tmp = b9 ^ b8; b9 = (tmp >> 8) | (tmp << (64 - 8)); b8 -= b9 + k1; b9 -= k2;
+	tmp = b7 ^ b6; b7 = (tmp >> 47) | (tmp << (64 - 47)); b6 -= b7 + k16; b7 -= k0;
+	tmp = b5 ^ b4; b5 = (tmp >> 8) | (tmp << (64 - 8)); b4 -= b5 + k14; b5 -= k15;
+	tmp = b3 ^ b2; b3 = (tmp >> 13) | (tmp << (64 - 13)); b2 -= b3 + k12; b3 -= k13;
+	tmp = b1 ^ b0; b1 = (tmp >> 24) | (tmp << (64 - 24)); b0 -= b1 + k10; b1 -= k11;
+	tmp = b7 ^ b12; b7 = (tmp >> 20) | (tmp << (64 - 20)); b12 -= b7;
+	tmp = b3 ^ b10; b3 = (tmp >> 37) | (tmp << (64 - 37)); b10 -= b3;
+	tmp = b5 ^ b8; b5 = (tmp >> 31) | (tmp << (64 - 31)); b8 -= b5;
+	tmp = b1 ^ b14; b1 = (tmp >> 23) | (tmp << (64 - 23)); b14 -= b1;
+	tmp = b9 ^ b4; b9 = (tmp >> 52) | (tmp << (64 - 52)); b4 -= b9;
+	tmp = b13 ^ b6; b13 = (tmp >> 35) | (tmp << (64 - 35)); b6 -= b13;
+	tmp = b11 ^ b2; b11 = (tmp >> 48) | (tmp << (64 - 48)); b2 -= b11;
+	tmp = b15 ^ b0; b15 = (tmp >> 9) | (tmp << (64 - 9)); b0 -= b15;
+	tmp = b9 ^ b10; b9 = (tmp >> 25) | (tmp << (64 - 25)); b10 -= b9;
+	tmp = b11 ^ b8; b11 = (tmp >> 44) | (tmp << (64 - 44)); b8 -= b11;
+	tmp = b13 ^ b14; b13 = (tmp >> 42) | (tmp << (64 - 42)); b14 -= b13;
+	tmp = b15 ^ b12; b15 = (tmp >> 19) | (tmp << (64 - 19)); b12 -= b15;
+	tmp = b1 ^ b6; b1 = (tmp >> 46) | (tmp << (64 - 46)); b6 -= b1;
+	tmp = b3 ^ b4; b3 = (tmp >> 47) | (tmp << (64 - 47)); b4 -= b3;
+	tmp = b5 ^ b2; b5 = (tmp >> 44) | (tmp << (64 - 44)); b2 -= b5;
+	tmp = b7 ^ b0; b7 = (tmp >> 31) | (tmp << (64 - 31)); b0 -= b7;
+	tmp = b1 ^ b8; b1 = (tmp >> 41) | (tmp << (64 - 41)); b8 -= b1;
+	tmp = b5 ^ b14; b5 = (tmp >> 42) | (tmp << (64 - 42)); b14 -= b5;
+	tmp = b3 ^ b12; b3 = (tmp >> 53) | (tmp << (64 - 53)); b12 -= b3;
+	tmp = b7 ^ b10; b7 = (tmp >> 4) | (tmp << (64 - 4)); b10 -= b7;
+	tmp = b15 ^ b4; b15 = (tmp >> 51) | (tmp << (64 - 51)); b4 -= b15;
+	tmp = b11 ^ b6; b11 = (tmp >> 56) | (tmp << (64 - 56)); b6 -= b11;
+	tmp = b13 ^ b2; b13 = (tmp >> 34) | (tmp << (64 - 34)); b2 -= b13;
+	tmp = b9 ^ b0; b9 = (tmp >> 16) | (tmp << (64 - 16)); b0 -= b9;
+	tmp = b15 ^ b14; b15 = (tmp >> 30) | (tmp << (64 - 30)); b14 -= b15 + k6 + t1; b15 -= k7 + 9;
+	tmp = b13 ^ b12; b13 = (tmp >> 44) | (tmp << (64 - 44)); b12 -= b13 + k4; b13 -= k5 + t0;
+	tmp = b11 ^ b10; b11 = (tmp >> 47) | (tmp << (64 - 47)); b10 -= b11 + k2; b11 -= k3;
+	tmp = b9 ^ b8; b9 = (tmp >> 12) | (tmp << (64 - 12)); b8 -= b9 + k0; b9 -= k1;
+	tmp = b7 ^ b6; b7 = (tmp >> 31) | (tmp << (64 - 31)); b6 -= b7 + k15; b7 -= k16;
+	tmp = b5 ^ b4; b5 = (tmp >> 37) | (tmp << (64 - 37)); b4 -= b5 + k13; b5 -= k14;
+	tmp = b3 ^ b2; b3 = (tmp >> 9) | (tmp << (64 - 9)); b2 -= b3 + k11; b3 -= k12;
+	tmp = b1 ^ b0; b1 = (tmp >> 41) | (tmp << (64 - 41)); b0 -= b1 + k9; b1 -= k10;
+	tmp = b7 ^ b12; b7 = (tmp >> 25) | (tmp << (64 - 25)); b12 -= b7;
+	tmp = b3 ^ b10; b3 = (tmp >> 16) | (tmp << (64 - 16)); b10 -= b3;
+	tmp = b5 ^ b8; b5 = (tmp >> 28) | (tmp << (64 - 28)); b8 -= b5;
+	tmp = b1 ^ b14; b1 = (tmp >> 47) | (tmp << (64 - 47)); b14 -= b1;
+	tmp = b9 ^ b4; b9 = (tmp >> 41) | (tmp << (64 - 41)); b4 -= b9;
+	tmp = b13 ^ b6; b13 = (tmp >> 48) | (tmp << (64 - 48)); b6 -= b13;
+	tmp = b11 ^ b2; b11 = (tmp >> 20) | (tmp << (64 - 20)); b2 -= b11;
+	tmp = b15 ^ b0; b15 = (tmp >> 5) | (tmp << (64 - 5)); b0 -= b15;
+	tmp = b9 ^ b10; b9 = (tmp >> 17) | (tmp << (64 - 17)); b10 -= b9;
+	tmp = b11 ^ b8; b11 = (tmp >> 59) | (tmp << (64 - 59)); b8 -= b11;
+	tmp = b13 ^ b14; b13 = (tmp >> 41) | (tmp << (64 - 41)); b14 -= b13;
+	tmp = b15 ^ b12; b15 = (tmp >> 34) | (tmp << (64 - 34)); b12 -= b15;
+	tmp = b1 ^ b6; b1 = (tmp >> 13) | (tmp << (64 - 13)); b6 -= b1;
+	tmp = b3 ^ b4; b3 = (tmp >> 51) | (tmp << (64 - 51)); b4 -= b3;
+	tmp = b5 ^ b2; b5 = (tmp >> 4) | (tmp << (64 - 4)); b2 -= b5;
+	tmp = b7 ^ b0; b7 = (tmp >> 33) | (tmp << (64 - 33)); b0 -= b7;
+	tmp = b1 ^ b8; b1 = (tmp >> 52) | (tmp << (64 - 52)); b8 -= b1;
+	tmp = b5 ^ b14; b5 = (tmp >> 23) | (tmp << (64 - 23)); b14 -= b5;
+	tmp = b3 ^ b12; b3 = (tmp >> 18) | (tmp << (64 - 18)); b12 -= b3;
+	tmp = b7 ^ b10; b7 = (tmp >> 49) | (tmp << (64 - 49)); b10 -= b7;
+	tmp = b15 ^ b4; b15 = (tmp >> 55) | (tmp << (64 - 55)); b4 -= b15;
+	tmp = b11 ^ b6; b11 = (tmp >> 10) | (tmp << (64 - 10)); b6 -= b11;
+	tmp = b13 ^ b2; b13 = (tmp >> 19) | (tmp << (64 - 19)); b2 -= b13;
+	tmp = b9 ^ b0; b9 = (tmp >> 38) | (tmp << (64 - 38)); b0 -= b9;
+	tmp = b15 ^ b14; b15 = (tmp >> 37) | (tmp << (64 - 37)); b14 -= b15 + k5 + t0; b15 -= k6 + 8;
+	tmp = b13 ^ b12; b13 = (tmp >> 22) | (tmp << (64 - 22)); b12 -= b13 + k3; b13 -= k4 + t2;
+	tmp = b11 ^ b10; b11 = (tmp >> 17) | (tmp << (64 - 17)); b10 -= b11 + k1; b11 -= k2;
+	tmp = b9 ^ b8; b9 = (tmp >> 8) | (tmp << (64 - 8)); b8 -= b9 + k16; b9 -= k0;
+	tmp = b7 ^ b6; b7 = (tmp >> 47) | (tmp << (64 - 47)); b6 -= b7 + k14; b7 -= k15;
+	tmp = b5 ^ b4; b5 = (tmp >> 8) | (tmp << (64 - 8)); b4 -= b5 + k12; b5 -= k13;
+	tmp = b3 ^ b2; b3 = (tmp >> 13) | (tmp << (64 - 13)); b2 -= b3 + k10; b3 -= k11;
+	tmp = b1 ^ b0; b1 = (tmp >> 24) | (tmp << (64 - 24)); b0 -= b1 + k8; b1 -= k9;
+	tmp = b7 ^ b12; b7 = (tmp >> 20) | (tmp << (64 - 20)); b12 -= b7;
+	tmp = b3 ^ b10; b3 = (tmp >> 37) | (tmp << (64 - 37)); b10 -= b3;
+	tmp = b5 ^ b8; b5 = (tmp >> 31) | (tmp << (64 - 31)); b8 -= b5;
+	tmp = b1 ^ b14; b1 = (tmp >> 23) | (tmp << (64 - 23)); b14 -= b1;
+	tmp = b9 ^ b4; b9 = (tmp >> 52) | (tmp << (64 - 52)); b4 -= b9;
+	tmp = b13 ^ b6; b13 = (tmp >> 35) | (tmp << (64 - 35)); b6 -= b13;
+	tmp = b11 ^ b2; b11 = (tmp >> 48) | (tmp << (64 - 48)); b2 -= b11;
+	tmp = b15 ^ b0; b15 = (tmp >> 9) | (tmp << (64 - 9)); b0 -= b15;
+	tmp = b9 ^ b10; b9 = (tmp >> 25) | (tmp << (64 - 25)); b10 -= b9;
+	tmp = b11 ^ b8; b11 = (tmp >> 44) | (tmp << (64 - 44)); b8 -= b11;
+	tmp = b13 ^ b14; b13 = (tmp >> 42) | (tmp << (64 - 42)); b14 -= b13;
+	tmp = b15 ^ b12; b15 = (tmp >> 19) | (tmp << (64 - 19)); b12 -= b15;
+	tmp = b1 ^ b6; b1 = (tmp >> 46) | (tmp << (64 - 46)); b6 -= b1;
+	tmp = b3 ^ b4; b3 = (tmp >> 47) | (tmp << (64 - 47)); b4 -= b3;
+	tmp = b5 ^ b2; b5 = (tmp >> 44) | (tmp << (64 - 44)); b2 -= b5;
+	tmp = b7 ^ b0; b7 = (tmp >> 31) | (tmp << (64 - 31)); b0 -= b7;
+	tmp = b1 ^ b8; b1 = (tmp >> 41) | (tmp << (64 - 41)); b8 -= b1;
+	tmp = b5 ^ b14; b5 = (tmp >> 42) | (tmp << (64 - 42)); b14 -= b5;
+	tmp = b3 ^ b12; b3 = (tmp >> 53) | (tmp << (64 - 53)); b12 -= b3;
+	tmp = b7 ^ b10; b7 = (tmp >> 4) | (tmp << (64 - 4)); b10 -= b7;
+	tmp = b15 ^ b4; b15 = (tmp >> 51) | (tmp << (64 - 51)); b4 -= b15;
+	tmp = b11 ^ b6; b11 = (tmp >> 56) | (tmp << (64 - 56)); b6 -= b11;
+	tmp = b13 ^ b2; b13 = (tmp >> 34) | (tmp << (64 - 34)); b2 -= b13;
+	tmp = b9 ^ b0; b9 = (tmp >> 16) | (tmp << (64 - 16)); b0 -= b9;
+	tmp = b15 ^ b14; b15 = (tmp >> 30) | (tmp << (64 - 30)); b14 -= b15 + k4 + t2; b15 -= k5 + 7;
+	tmp = b13 ^ b12; b13 = (tmp >> 44) | (tmp << (64 - 44)); b12 -= b13 + k2; b13 -= k3 + t1;
+	tmp = b11 ^ b10; b11 = (tmp >> 47) | (tmp << (64 - 47)); b10 -= b11 + k0; b11 -= k1;
+	tmp = b9 ^ b8; b9 = (tmp >> 12) | (tmp << (64 - 12)); b8 -= b9 + k15; b9 -= k16;
+	tmp = b7 ^ b6; b7 = (tmp >> 31) | (tmp << (64 - 31)); b6 -= b7 + k13; b7 -= k14;
+	tmp = b5 ^ b4; b5 = (tmp >> 37) | (tmp << (64 - 37)); b4 -= b5 + k11; b5 -= k12;
+	tmp = b3 ^ b2; b3 = (tmp >> 9) | (tmp << (64 - 9)); b2 -= b3 + k9; b3 -= k10;
+	tmp = b1 ^ b0; b1 = (tmp >> 41) | (tmp << (64 - 41)); b0 -= b1 + k7; b1 -= k8;
+	tmp = b7 ^ b12; b7 = (tmp >> 25) | (tmp << (64 - 25)); b12 -= b7;
+	tmp = b3 ^ b10; b3 = (tmp >> 16) | (tmp << (64 - 16)); b10 -= b3;
+	tmp = b5 ^ b8; b5 = (tmp >> 28) | (tmp << (64 - 28)); b8 -= b5;
+	tmp = b1 ^ b14; b1 = (tmp >> 47) | (tmp << (64 - 47)); b14 -= b1;
+	tmp = b9 ^ b4; b9 = (tmp >> 41) | (tmp << (64 - 41)); b4 -= b9;
+	tmp = b13 ^ b6; b13 = (tmp >> 48) | (tmp << (64 - 48)); b6 -= b13;
+	tmp = b11 ^ b2; b11 = (tmp >> 20) | (tmp << (64 - 20)); b2 -= b11;
+	tmp = b15 ^ b0; b15 = (tmp >> 5) | (tmp << (64 - 5)); b0 -= b15;
+	tmp = b9 ^ b10; b9 = (tmp >> 17) | (tmp << (64 - 17)); b10 -= b9;
+	tmp = b11 ^ b8; b11 = (tmp >> 59) | (tmp << (64 - 59)); b8 -= b11;
+	tmp = b13 ^ b14; b13 = (tmp >> 41) | (tmp << (64 - 41)); b14 -= b13;
+	tmp = b15 ^ b12; b15 = (tmp >> 34) | (tmp << (64 - 34)); b12 -= b15;
+	tmp = b1 ^ b6; b1 = (tmp >> 13) | (tmp << (64 - 13)); b6 -= b1;
+	tmp = b3 ^ b4; b3 = (tmp >> 51) | (tmp << (64 - 51)); b4 -= b3;
+	tmp = b5 ^ b2; b5 = (tmp >> 4) | (tmp << (64 - 4)); b2 -= b5;
+	tmp = b7 ^ b0; b7 = (tmp >> 33) | (tmp << (64 - 33)); b0 -= b7;
+	tmp = b1 ^ b8; b1 = (tmp >> 52) | (tmp << (64 - 52)); b8 -= b1;
+	tmp = b5 ^ b14; b5 = (tmp >> 23) | (tmp << (64 - 23)); b14 -= b5;
+	tmp = b3 ^ b12; b3 = (tmp >> 18) | (tmp << (64 - 18)); b12 -= b3;
+	tmp = b7 ^ b10; b7 = (tmp >> 49) | (tmp << (64 - 49)); b10 -= b7;
+	tmp = b15 ^ b4; b15 = (tmp >> 55) | (tmp << (64 - 55)); b4 -= b15;
+	tmp = b11 ^ b6; b11 = (tmp >> 10) | (tmp << (64 - 10)); b6 -= b11;
+	tmp = b13 ^ b2; b13 = (tmp >> 19) | (tmp << (64 - 19)); b2 -= b13;
+	tmp = b9 ^ b0; b9 = (tmp >> 38) | (tmp << (64 - 38)); b0 -= b9;
+	tmp = b15 ^ b14; b15 = (tmp >> 37) | (tmp << (64 - 37)); b14 -= b15 + k3 + t1; b15 -= k4 + 6;
+	tmp = b13 ^ b12; b13 = (tmp >> 22) | (tmp << (64 - 22)); b12 -= b13 + k1; b13 -= k2 + t0;
+	tmp = b11 ^ b10; b11 = (tmp >> 17) | (tmp << (64 - 17)); b10 -= b11 + k16; b11 -= k0;
+	tmp = b9 ^ b8; b9 = (tmp >> 8) | (tmp << (64 - 8)); b8 -= b9 + k14; b9 -= k15;
+	tmp = b7 ^ b6; b7 = (tmp >> 47) | (tmp << (64 - 47)); b6 -= b7 + k12; b7 -= k13;
+	tmp = b5 ^ b4; b5 = (tmp >> 8) | (tmp << (64 - 8)); b4 -= b5 + k10; b5 -= k11;
+	tmp = b3 ^ b2; b3 = (tmp >> 13) | (tmp << (64 - 13)); b2 -= b3 + k8; b3 -= k9;
+	tmp = b1 ^ b0; b1 = (tmp >> 24) | (tmp << (64 - 24)); b0 -= b1 + k6; b1 -= k7;
+	tmp = b7 ^ b12; b7 = (tmp >> 20) | (tmp << (64 - 20)); b12 -= b7;
+	tmp = b3 ^ b10; b3 = (tmp >> 37) | (tmp << (64 - 37)); b10 -= b3;
+	tmp = b5 ^ b8; b5 = (tmp >> 31) | (tmp << (64 - 31)); b8 -= b5;
+	tmp = b1 ^ b14; b1 = (tmp >> 23) | (tmp << (64 - 23)); b14 -= b1;
+	tmp = b9 ^ b4; b9 = (tmp >> 52) | (tmp << (64 - 52)); b4 -= b9;
+	tmp = b13 ^ b6; b13 = (tmp >> 35) | (tmp << (64 - 35)); b6 -= b13;
+	tmp = b11 ^ b2; b11 = (tmp >> 48) | (tmp << (64 - 48)); b2 -= b11;
+	tmp = b15 ^ b0; b15 = (tmp >> 9) | (tmp << (64 - 9)); b0 -= b15;
+	tmp = b9 ^ b10; b9 = (tmp >> 25) | (tmp << (64 - 25)); b10 -= b9;
+	tmp = b11 ^ b8; b11 = (tmp >> 44) | (tmp << (64 - 44)); b8 -= b11;
+	tmp = b13 ^ b14; b13 = (tmp >> 42) | (tmp << (64 - 42)); b14 -= b13;
+	tmp = b15 ^ b12; b15 = (tmp >> 19) | (tmp << (64 - 19)); b12 -= b15;
+	tmp = b1 ^ b6; b1 = (tmp >> 46) | (tmp << (64 - 46)); b6 -= b1;
+	tmp = b3 ^ b4; b3 = (tmp >> 47) | (tmp << (64 - 47)); b4 -= b3;
+	tmp = b5 ^ b2; b5 = (tmp >> 44) | (tmp << (64 - 44)); b2 -= b5;
+	tmp = b7 ^ b0; b7 = (tmp >> 31) | (tmp << (64 - 31)); b0 -= b7;
+	tmp = b1 ^ b8; b1 = (tmp >> 41) | (tmp << (64 - 41)); b8 -= b1;
+	tmp = b5 ^ b14; b5 = (tmp >> 42) | (tmp << (64 - 42)); b14 -= b5;
+	tmp = b3 ^ b12; b3 = (tmp >> 53) | (tmp << (64 - 53)); b12 -= b3;
+	tmp = b7 ^ b10; b7 = (tmp >> 4) | (tmp << (64 - 4)); b10 -= b7;
+	tmp = b15 ^ b4; b15 = (tmp >> 51) | (tmp << (64 - 51)); b4 -= b15;
+	tmp = b11 ^ b6; b11 = (tmp >> 56) | (tmp << (64 - 56)); b6 -= b11;
+	tmp = b13 ^ b2; b13 = (tmp >> 34) | (tmp << (64 - 34)); b2 -= b13;
+	tmp = b9 ^ b0; b9 = (tmp >> 16) | (tmp << (64 - 16)); b0 -= b9;
+	tmp = b15 ^ b14; b15 = (tmp >> 30) | (tmp << (64 - 30)); b14 -= b15 + k2 + t0; b15 -= k3 + 5;
+	tmp = b13 ^ b12; b13 = (tmp >> 44) | (tmp << (64 - 44)); b12 -= b13 + k0; b13 -= k1 + t2;
+	tmp = b11 ^ b10; b11 = (tmp >> 47) | (tmp << (64 - 47)); b10 -= b11 + k15; b11 -= k16;
+	tmp = b9 ^ b8; b9 = (tmp >> 12) | (tmp << (64 - 12)); b8 -= b9 + k13; b9 -= k14;
+	tmp = b7 ^ b6; b7 = (tmp >> 31) | (tmp << (64 - 31)); b6 -= b7 + k11; b7 -= k12;
+	tmp = b5 ^ b4; b5 = (tmp >> 37) | (tmp << (64 - 37)); b4 -= b5 + k9; b5 -= k10;
+	tmp = b3 ^ b2; b3 = (tmp >> 9) | (tmp << (64 - 9)); b2 -= b3 + k7; b3 -= k8;
+	tmp = b1 ^ b0; b1 = (tmp >> 41) | (tmp << (64 - 41)); b0 -= b1 + k5; b1 -= k6;
+	tmp = b7 ^ b12; b7 = (tmp >> 25) | (tmp << (64 - 25)); b12 -= b7;
+	tmp = b3 ^ b10; b3 = (tmp >> 16) | (tmp << (64 - 16)); b10 -= b3;
+	tmp = b5 ^ b8; b5 = (tmp >> 28) | (tmp << (64 - 28)); b8 -= b5;
+	tmp = b1 ^ b14; b1 = (tmp >> 47) | (tmp << (64 - 47)); b14 -= b1;
+	tmp = b9 ^ b4; b9 = (tmp >> 41) | (tmp << (64 - 41)); b4 -= b9;
+	tmp = b13 ^ b6; b13 = (tmp >> 48) | (tmp << (64 - 48)); b6 -= b13;
+	tmp = b11 ^ b2; b11 = (tmp >> 20) | (tmp << (64 - 20)); b2 -= b11;
+	tmp = b15 ^ b0; b15 = (tmp >> 5) | (tmp << (64 - 5)); b0 -= b15;
+	tmp = b9 ^ b10; b9 = (tmp >> 17) | (tmp << (64 - 17)); b10 -= b9;
+	tmp = b11 ^ b8; b11 = (tmp >> 59) | (tmp << (64 - 59)); b8 -= b11;
+	tmp = b13 ^ b14; b13 = (tmp >> 41) | (tmp << (64 - 41)); b14 -= b13;
+	tmp = b15 ^ b12; b15 = (tmp >> 34) | (tmp << (64 - 34)); b12 -= b15;
+	tmp = b1 ^ b6; b1 = (tmp >> 13) | (tmp << (64 - 13)); b6 -= b1;
+	tmp = b3 ^ b4; b3 = (tmp >> 51) | (tmp << (64 - 51)); b4 -= b3;
+	tmp = b5 ^ b2; b5 = (tmp >> 4) | (tmp << (64 - 4)); b2 -= b5;
+	tmp = b7 ^ b0; b7 = (tmp >> 33) | (tmp << (64 - 33)); b0 -= b7;
+	tmp = b1 ^ b8; b1 = (tmp >> 52) | (tmp << (64 - 52)); b8 -= b1;
+	tmp = b5 ^ b14; b5 = (tmp >> 23) | (tmp << (64 - 23)); b14 -= b5;
+	tmp = b3 ^ b12; b3 = (tmp >> 18) | (tmp << (64 - 18)); b12 -= b3;
+	tmp = b7 ^ b10; b7 = (tmp >> 49) | (tmp << (64 - 49)); b10 -= b7;
+	tmp = b15 ^ b4; b15 = (tmp >> 55) | (tmp << (64 - 55)); b4 -= b15;
+	tmp = b11 ^ b6; b11 = (tmp >> 10) | (tmp << (64 - 10)); b6 -= b11;
+	tmp = b13 ^ b2; b13 = (tmp >> 19) | (tmp << (64 - 19)); b2 -= b13;
+	tmp = b9 ^ b0; b9 = (tmp >> 38) | (tmp << (64 - 38)); b0 -= b9;
+	tmp = b15 ^ b14; b15 = (tmp >> 37) | (tmp << (64 - 37)); b14 -= b15 + k1 + t2; b15 -= k2 + 4;
+	tmp = b13 ^ b12; b13 = (tmp >> 22) | (tmp << (64 - 22)); b12 -= b13 + k16; b13 -= k0 + t1;
+	tmp = b11 ^ b10; b11 = (tmp >> 17) | (tmp << (64 - 17)); b10 -= b11 + k14; b11 -= k15;
+	tmp = b9 ^ b8; b9 = (tmp >> 8) | (tmp << (64 - 8)); b8 -= b9 + k12; b9 -= k13;
+	tmp = b7 ^ b6; b7 = (tmp >> 47) | (tmp << (64 - 47)); b6 -= b7 + k10; b7 -= k11;
+	tmp = b5 ^ b4; b5 = (tmp >> 8) | (tmp << (64 - 8)); b4 -= b5 + k8; b5 -= k9;
+	tmp = b3 ^ b2; b3 = (tmp >> 13) | (tmp << (64 - 13)); b2 -= b3 + k6; b3 -= k7;
+	tmp = b1 ^ b0; b1 = (tmp >> 24) | (tmp << (64 - 24)); b0 -= b1 + k4; b1 -= k5;
+	tmp = b7 ^ b12; b7 = (tmp >> 20) | (tmp << (64 - 20)); b12 -= b7;
+	tmp = b3 ^ b10; b3 = (tmp >> 37) | (tmp << (64 - 37)); b10 -= b3;
+	tmp = b5 ^ b8; b5 = (tmp >> 31) | (tmp << (64 - 31)); b8 -= b5;
+	tmp = b1 ^ b14; b1 = (tmp >> 23) | (tmp << (64 - 23)); b14 -= b1;
+	tmp = b9 ^ b4; b9 = (tmp >> 52) | (tmp << (64 - 52)); b4 -= b9;
+	tmp = b13 ^ b6; b13 = (tmp >> 35) | (tmp << (64 - 35)); b6 -= b13;
+	tmp = b11 ^ b2; b11 = (tmp >> 48) | (tmp << (64 - 48)); b2 -= b11;
+	tmp = b15 ^ b0; b15 = (tmp >> 9) | (tmp << (64 - 9)); b0 -= b15;
+	tmp = b9 ^ b10; b9 = (tmp >> 25) | (tmp << (64 - 25)); b10 -= b9;
+	tmp = b11 ^ b8; b11 = (tmp >> 44) | (tmp << (64 - 44)); b8 -= b11;
+	tmp = b13 ^ b14; b13 = (tmp >> 42) | (tmp << (64 - 42)); b14 -= b13;
+	tmp = b15 ^ b12; b15 = (tmp >> 19) | (tmp << (64 - 19)); b12 -= b15;
+	tmp = b1 ^ b6; b1 = (tmp >> 46) | (tmp << (64 - 46)); b6 -= b1;
+	tmp = b3 ^ b4; b3 = (tmp >> 47) | (tmp << (64 - 47)); b4 -= b3;
+	tmp = b5 ^ b2; b5 = (tmp >> 44) | (tmp << (64 - 44)); b2 -= b5;
+	tmp = b7 ^ b0; b7 = (tmp >> 31) | (tmp << (64 - 31)); b0 -= b7;
+	tmp = b1 ^ b8; b1 = (tmp >> 41) | (tmp << (64 - 41)); b8 -= b1;
+	tmp = b5 ^ b14; b5 = (tmp >> 42) | (tmp << (64 - 42)); b14 -= b5;
+	tmp = b3 ^ b12; b3 = (tmp >> 53) | (tmp << (64 - 53)); b12 -= b3;
+	tmp = b7 ^ b10; b7 = (tmp >> 4) | (tmp << (64 - 4)); b10 -= b7;
+	tmp = b15 ^ b4; b15 = (tmp >> 51) | (tmp << (64 - 51)); b4 -= b15;
+	tmp = b11 ^ b6; b11 = (tmp >> 56) | (tmp << (64 - 56)); b6 -= b11;
+	tmp = b13 ^ b2; b13 = (tmp >> 34) | (tmp << (64 - 34)); b2 -= b13;
+	tmp = b9 ^ b0; b9 = (tmp >> 16) | (tmp << (64 - 16)); b0 -= b9;
+	tmp = b15 ^ b14; b15 = (tmp >> 30) | (tmp << (64 - 30)); b14 -= b15 + k0 + t1; b15 -= k1 + 3;
+	tmp = b13 ^ b12; b13 = (tmp >> 44) | (tmp << (64 - 44)); b12 -= b13 + k15; b13 -= k16 + t0;
+	tmp = b11 ^ b10; b11 = (tmp >> 47) | (tmp << (64 - 47)); b10 -= b11 + k13; b11 -= k14;
+	tmp = b9 ^ b8; b9 = (tmp >> 12) | (tmp << (64 - 12)); b8 -= b9 + k11; b9 -= k12;
+	tmp = b7 ^ b6; b7 = (tmp >> 31) | (tmp << (64 - 31)); b6 -= b7 + k9; b7 -= k10;
+	tmp = b5 ^ b4; b5 = (tmp >> 37) | (tmp << (64 - 37)); b4 -= b5 + k7; b5 -= k8;
+	tmp = b3 ^ b2; b3 = (tmp >> 9) | (tmp << (64 - 9)); b2 -= b3 + k5; b3 -= k6;
+	tmp = b1 ^ b0; b1 = (tmp >> 41) | (tmp << (64 - 41)); b0 -= b1 + k3; b1 -= k4;
+	tmp = b7 ^ b12; b7 = (tmp >> 25) | (tmp << (64 - 25)); b12 -= b7;
+	tmp = b3 ^ b10; b3 = (tmp >> 16) | (tmp << (64 - 16)); b10 -= b3;
+	tmp = b5 ^ b8; b5 = (tmp >> 28) | (tmp << (64 - 28)); b8 -= b5;
+	tmp = b1 ^ b14; b1 = (tmp >> 47) | (tmp << (64 - 47)); b14 -= b1;
+	tmp = b9 ^ b4; b9 = (tmp >> 41) | (tmp << (64 - 41)); b4 -= b9;
+	tmp = b13 ^ b6; b13 = (tmp >> 48) | (tmp << (64 - 48)); b6 -= b13;
+	tmp = b11 ^ b2; b11 = (tmp >> 20) | (tmp << (64 - 20)); b2 -= b11;
+	tmp = b15 ^ b0; b15 = (tmp >> 5) | (tmp << (64 - 5)); b0 -= b15;
+	tmp = b9 ^ b10; b9 = (tmp >> 17) | (tmp << (64 - 17)); b10 -= b9;
+	tmp = b11 ^ b8; b11 = (tmp >> 59) | (tmp << (64 - 59)); b8 -= b11;
+	tmp = b13 ^ b14; b13 = (tmp >> 41) | (tmp << (64 - 41)); b14 -= b13;
+	tmp = b15 ^ b12; b15 = (tmp >> 34) | (tmp << (64 - 34)); b12 -= b15;
+	tmp = b1 ^ b6; b1 = (tmp >> 13) | (tmp << (64 - 13)); b6 -= b1;
+	tmp = b3 ^ b4; b3 = (tmp >> 51) | (tmp << (64 - 51)); b4 -= b3;
+	tmp = b5 ^ b2; b5 = (tmp >> 4) | (tmp << (64 - 4)); b2 -= b5;
+	tmp = b7 ^ b0; b7 = (tmp >> 33) | (tmp << (64 - 33)); b0 -= b7;
+	tmp = b1 ^ b8; b1 = (tmp >> 52) | (tmp << (64 - 52)); b8 -= b1;
+	tmp = b5 ^ b14; b5 = (tmp >> 23) | (tmp << (64 - 23)); b14 -= b5;
+	tmp = b3 ^ b12; b3 = (tmp >> 18) | (tmp << (64 - 18)); b12 -= b3;
+	tmp = b7 ^ b10; b7 = (tmp >> 49) | (tmp << (64 - 49)); b10 -= b7;
+	tmp = b15 ^ b4; b15 = (tmp >> 55) | (tmp << (64 - 55)); b4 -= b15;
+	tmp = b11 ^ b6; b11 = (tmp >> 10) | (tmp << (64 - 10)); b6 -= b11;
+	tmp = b13 ^ b2; b13 = (tmp >> 19) | (tmp << (64 - 19)); b2 -= b13;
+	tmp = b9 ^ b0; b9 = (tmp >> 38) | (tmp << (64 - 38)); b0 -= b9;
+	tmp = b15 ^ b14; b15 = (tmp >> 37) | (tmp << (64 - 37)); b14 -= b15 + k16 + t0; b15 -= k0 + 2;
+	tmp = b13 ^ b12; b13 = (tmp >> 22) | (tmp << (64 - 22)); b12 -= b13 + k14; b13 -= k15 + t2;
+	tmp = b11 ^ b10; b11 = (tmp >> 17) | (tmp << (64 - 17)); b10 -= b11 + k12; b11 -= k13;
+	tmp = b9 ^ b8; b9 = (tmp >> 8) | (tmp << (64 - 8)); b8 -= b9 + k10; b9 -= k11;
+	tmp = b7 ^ b6; b7 = (tmp >> 47) | (tmp << (64 - 47)); b6 -= b7 + k8; b7 -= k9;
+	tmp = b5 ^ b4; b5 = (tmp >> 8) | (tmp << (64 - 8)); b4 -= b5 + k6; b5 -= k7;
+	tmp = b3 ^ b2; b3 = (tmp >> 13) | (tmp << (64 - 13)); b2 -= b3 + k4; b3 -= k5;
+	tmp = b1 ^ b0; b1 = (tmp >> 24) | (tmp << (64 - 24)); b0 -= b1 + k2; b1 -= k3;
+	tmp = b7 ^ b12; b7 = (tmp >> 20) | (tmp << (64 - 20)); b12 -= b7;
+	tmp = b3 ^ b10; b3 = (tmp >> 37) | (tmp << (64 - 37)); b10 -= b3;
+	tmp = b5 ^ b8; b5 = (tmp >> 31) | (tmp << (64 - 31)); b8 -= b5;
+	tmp = b1 ^ b14; b1 = (tmp >> 23) | (tmp << (64 - 23)); b14 -= b1;
+	tmp = b9 ^ b4; b9 = (tmp >> 52) | (tmp << (64 - 52)); b4 -= b9;
+	tmp = b13 ^ b6; b13 = (tmp >> 35) | (tmp << (64 - 35)); b6 -= b13;
+	tmp = b11 ^ b2; b11 = (tmp >> 48) | (tmp << (64 - 48)); b2 -= b11;
+	tmp = b15 ^ b0; b15 = (tmp >> 9) | (tmp << (64 - 9)); b0 -= b15;
+	tmp = b9 ^ b10; b9 = (tmp >> 25) | (tmp << (64 - 25)); b10 -= b9;
+	tmp = b11 ^ b8; b11 = (tmp >> 44) | (tmp << (64 - 44)); b8 -= b11;
+	tmp = b13 ^ b14; b13 = (tmp >> 42) | (tmp << (64 - 42)); b14 -= b13;
+	tmp = b15 ^ b12; b15 = (tmp >> 19) | (tmp << (64 - 19)); b12 -= b15;
+	tmp = b1 ^ b6; b1 = (tmp >> 46) | (tmp << (64 - 46)); b6 -= b1;
+	tmp = b3 ^ b4; b3 = (tmp >> 47) | (tmp << (64 - 47)); b4 -= b3;
+	tmp = b5 ^ b2; b5 = (tmp >> 44) | (tmp << (64 - 44)); b2 -= b5;
+	tmp = b7 ^ b0; b7 = (tmp >> 31) | (tmp << (64 - 31)); b0 -= b7;
+	tmp = b1 ^ b8; b1 = (tmp >> 41) | (tmp << (64 - 41)); b8 -= b1;
+	tmp = b5 ^ b14; b5 = (tmp >> 42) | (tmp << (64 - 42)); b14 -= b5;
+	tmp = b3 ^ b12; b3 = (tmp >> 53) | (tmp << (64 - 53)); b12 -= b3;
+	tmp = b7 ^ b10; b7 = (tmp >> 4) | (tmp << (64 - 4)); b10 -= b7;
+	tmp = b15 ^ b4; b15 = (tmp >> 51) | (tmp << (64 - 51)); b4 -= b15;
+	tmp = b11 ^ b6; b11 = (tmp >> 56) | (tmp << (64 - 56)); b6 -= b11;
+	tmp = b13 ^ b2; b13 = (tmp >> 34) | (tmp << (64 - 34)); b2 -= b13;
+	tmp = b9 ^ b0; b9 = (tmp >> 16) | (tmp << (64 - 16)); b0 -= b9;
+	tmp = b15 ^ b14; b15 = (tmp >> 30) | (tmp << (64 - 30)); b14 -= b15 + k15 + t2; b15 -= k16 + 1;
+	tmp = b13 ^ b12; b13 = (tmp >> 44) | (tmp << (64 - 44)); b12 -= b13 + k13; b13 -= k14 + t1;
+	tmp = b11 ^ b10; b11 = (tmp >> 47) | (tmp << (64 - 47)); b10 -= b11 + k11; b11 -= k12;
+	tmp = b9 ^ b8; b9 = (tmp >> 12) | (tmp << (64 - 12)); b8 -= b9 + k9; b9 -= k10;
+	tmp = b7 ^ b6; b7 = (tmp >> 31) | (tmp << (64 - 31)); b6 -= b7 + k7; b7 -= k8;
+	tmp = b5 ^ b4; b5 = (tmp >> 37) | (tmp << (64 - 37)); b4 -= b5 + k5; b5 -= k6;
+	tmp = b3 ^ b2; b3 = (tmp >> 9) | (tmp << (64 - 9)); b2 -= b3 + k3; b3 -= k4;
+	tmp = b1 ^ b0; b1 = (tmp >> 41) | (tmp << (64 - 41)); b0 -= b1 + k1; b1 -= k2;
+	tmp = b7 ^ b12; b7 = (tmp >> 25) | (tmp << (64 - 25)); b12 -= b7;
+	tmp = b3 ^ b10; b3 = (tmp >> 16) | (tmp << (64 - 16)); b10 -= b3;
+	tmp = b5 ^ b8; b5 = (tmp >> 28) | (tmp << (64 - 28)); b8 -= b5;
+	tmp = b1 ^ b14; b1 = (tmp >> 47) | (tmp << (64 - 47)); b14 -= b1;
+	tmp = b9 ^ b4; b9 = (tmp >> 41) | (tmp << (64 - 41)); b4 -= b9;
+	tmp = b13 ^ b6; b13 = (tmp >> 48) | (tmp << (64 - 48)); b6 -= b13;
+	tmp = b11 ^ b2; b11 = (tmp >> 20) | (tmp << (64 - 20)); b2 -= b11;
+	tmp = b15 ^ b0; b15 = (tmp >> 5) | (tmp << (64 - 5)); b0 -= b15;
+	tmp = b9 ^ b10; b9 = (tmp >> 17) | (tmp << (64 - 17)); b10 -= b9;
+	tmp = b11 ^ b8; b11 = (tmp >> 59) | (tmp << (64 - 59)); b8 -= b11;
+	tmp = b13 ^ b14; b13 = (tmp >> 41) | (tmp << (64 - 41)); b14 -= b13;
+	tmp = b15 ^ b12; b15 = (tmp >> 34) | (tmp << (64 - 34)); b12 -= b15;
+	tmp = b1 ^ b6; b1 = (tmp >> 13) | (tmp << (64 - 13)); b6 -= b1;
+	tmp = b3 ^ b4; b3 = (tmp >> 51) | (tmp << (64 - 51)); b4 -= b3;
+	tmp = b5 ^ b2; b5 = (tmp >> 4) | (tmp << (64 - 4)); b2 -= b5;
+	tmp = b7 ^ b0; b7 = (tmp >> 33) | (tmp << (64 - 33)); b0 -= b7;
+	tmp = b1 ^ b8; b1 = (tmp >> 52) | (tmp << (64 - 52)); b8 -= b1;
+	tmp = b5 ^ b14; b5 = (tmp >> 23) | (tmp << (64 - 23)); b14 -= b5;
+	tmp = b3 ^ b12; b3 = (tmp >> 18) | (tmp << (64 - 18)); b12 -= b3;
+	tmp = b7 ^ b10; b7 = (tmp >> 49) | (tmp << (64 - 49)); b10 -= b7;
+	tmp = b15 ^ b4; b15 = (tmp >> 55) | (tmp << (64 - 55)); b4 -= b15;
+	tmp = b11 ^ b6; b11 = (tmp >> 10) | (tmp << (64 - 10)); b6 -= b11;
+	tmp = b13 ^ b2; b13 = (tmp >> 19) | (tmp << (64 - 19)); b2 -= b13;
+	tmp = b9 ^ b0; b9 = (tmp >> 38) | (tmp << (64 - 38)); b0 -= b9;
+	tmp = b15 ^ b14; b15 = (tmp >> 37) | (tmp << (64 - 37)); b14 -= b15 + k14 + t1; b15 -= k15;
+	tmp = b13 ^ b12; b13 = (tmp >> 22) | (tmp << (64 - 22)); b12 -= b13 + k12; b13 -= k13 + t0;
+	tmp = b11 ^ b10; b11 = (tmp >> 17) | (tmp << (64 - 17)); b10 -= b11 + k10; b11 -= k11;
+	tmp = b9 ^ b8; b9 = (tmp >> 8) | (tmp << (64 - 8)); b8 -= b9 + k8; b9 -= k9;
+	tmp = b7 ^ b6; b7 = (tmp >> 47) | (tmp << (64 - 47)); b6 -= b7 + k6; b7 -= k7;
+	tmp = b5 ^ b4; b5 = (tmp >> 8) | (tmp << (64 - 8)); b4 -= b5 + k4; b5 -= k5;
+	tmp = b3 ^ b2; b3 = (tmp >> 13) | (tmp << (64 - 13)); b2 -= b3 + k2; b3 -= k3;
+	tmp = b1 ^ b0; b1 = (tmp >> 24) | (tmp << (64 - 24)); b0 -= b1 + k0; b1 -= k1;
 
-            output[15] = b15;
-            output[14] = b14;
-            output[13] = b13;
-            output[12] = b12;
-            output[11] = b11;
-            output[10] = b10;
-            output[9] = b9;
-            output[8] = b8;
-            output[7] = b7;
-            output[6] = b6;
-            output[5] = b5;
-            output[4] = b4;
-            output[3] = b3;
-            output[2] = b2;
-            output[1] = b1;
-            output[0] = b0;
+	output[15] = b15;
+	output[14] = b14;
+	output[13] = b13;
+	output[12] = b12;
+	output[11] = b11;
+	output[10] = b10;
+	output[9] = b9;
+	output[8] = b8;
+	output[7] = b7;
+	output[6] = b6;
+	output[5] = b5;
+	output[4] = b4;
+	output[3] = b3;
+	output[2] = b2;
+	output[1] = b1;
+	output[0] = b0;
 }
diff --git a/drivers/staging/skein/threefish256Block.c b/drivers/staging/skein/threefish256Block.c
index 09ea5099bc76..da3b8357e47f 100644
--- a/drivers/staging/skein/threefish256Block.c
+++ b/drivers/staging/skein/threefish256Block.c
@@ -3,346 +3,345 @@
 
 
 void threefishEncrypt256(struct threefish_key *keyCtx, u64 *input, u64 *output)
-  {
+{
+	u64 b0 = input[0], b1 = input[1],
+	  b2 = input[2], b3 = input[3];
+	u64 k0 = keyCtx->key[0], k1 = keyCtx->key[1],
+	  k2 = keyCtx->key[2], k3 = keyCtx->key[3],
+	  k4 = keyCtx->key[4];
+	u64 t0 = keyCtx->tweak[0], t1 = keyCtx->tweak[1],
+	  t2 = keyCtx->tweak[2];
 
-    u64 b0 = input[0], b1 = input[1],
-      b2 = input[2], b3 = input[3];
-    u64 k0 = keyCtx->key[0], k1 = keyCtx->key[1],
-      k2 = keyCtx->key[2], k3 = keyCtx->key[3],
-      k4 = keyCtx->key[4];
-    u64 t0 = keyCtx->tweak[0], t1 = keyCtx->tweak[1],
-      t2 = keyCtx->tweak[2];
+	b1 += k1 + t0; b0 += b1 + k0; b1 = ((b1 << 14) | (b1 >> (64 - 14))) ^ b0;
+	b3 += k3; b2 += b3 + k2 + t1; b3 = ((b3 << 16) | (b3 >> (64 - 16))) ^ b2;
+	b0 += b3; b3 = ((b3 << 52) | (b3 >> (64 - 52))) ^ b0;
+	b2 += b1; b1 = ((b1 << 57) | (b1 >> (64 - 57))) ^ b2;
+	b0 += b1; b1 = ((b1 << 23) | (b1 >> (64 - 23))) ^ b0;
+	b2 += b3; b3 = ((b3 << 40) | (b3 >> (64 - 40))) ^ b2;
+	b0 += b3; b3 = ((b3 << 5) | (b3 >> (64 - 5))) ^ b0;
+	b2 += b1; b1 = ((b1 << 37) | (b1 >> (64 - 37))) ^ b2;
+	b1 += k2 + t1; b0 += b1 + k1; b1 = ((b1 << 25) | (b1 >> (64 - 25))) ^ b0;
+	b3 += k4 + 1; b2 += b3 + k3 + t2; b3 = ((b3 << 33) | (b3 >> (64 - 33))) ^ b2;
+	b0 += b3; b3 = ((b3 << 46) | (b3 >> (64 - 46))) ^ b0;
+	b2 += b1; b1 = ((b1 << 12) | (b1 >> (64 - 12))) ^ b2;
+	b0 += b1; b1 = ((b1 << 58) | (b1 >> (64 - 58))) ^ b0;
+	b2 += b3; b3 = ((b3 << 22) | (b3 >> (64 - 22))) ^ b2;
+	b0 += b3; b3 = ((b3 << 32) | (b3 >> (64 - 32))) ^ b0;
+	b2 += b1; b1 = ((b1 << 32) | (b1 >> (64 - 32))) ^ b2;
 
-    b1 += k1 + t0; b0 += b1 + k0; b1 = ((b1 << 14) | (b1 >> (64 - 14))) ^ b0;
-    b3 += k3; b2 += b3 + k2 + t1; b3 = ((b3 << 16) | (b3 >> (64 - 16))) ^ b2;
-    b0 += b3; b3 = ((b3 << 52) | (b3 >> (64 - 52))) ^ b0;
-    b2 += b1; b1 = ((b1 << 57) | (b1 >> (64 - 57))) ^ b2;
-    b0 += b1; b1 = ((b1 << 23) | (b1 >> (64 - 23))) ^ b0;
-    b2 += b3; b3 = ((b3 << 40) | (b3 >> (64 - 40))) ^ b2;
-    b0 += b3; b3 = ((b3 << 5) | (b3 >> (64 - 5))) ^ b0;
-    b2 += b1; b1 = ((b1 << 37) | (b1 >> (64 - 37))) ^ b2;
-    b1 += k2 + t1; b0 += b1 + k1; b1 = ((b1 << 25) | (b1 >> (64 - 25))) ^ b0;
-    b3 += k4 + 1; b2 += b3 + k3 + t2; b3 = ((b3 << 33) | (b3 >> (64 - 33))) ^ b2;
-    b0 += b3; b3 = ((b3 << 46) | (b3 >> (64 - 46))) ^ b0;
-    b2 += b1; b1 = ((b1 << 12) | (b1 >> (64 - 12))) ^ b2;
-    b0 += b1; b1 = ((b1 << 58) | (b1 >> (64 - 58))) ^ b0;
-    b2 += b3; b3 = ((b3 << 22) | (b3 >> (64 - 22))) ^ b2;
-    b0 += b3; b3 = ((b3 << 32) | (b3 >> (64 - 32))) ^ b0;
-    b2 += b1; b1 = ((b1 << 32) | (b1 >> (64 - 32))) ^ b2;
+	b1 += k3 + t2; b0 += b1 + k2; b1 = ((b1 << 14) | (b1 >> (64 - 14))) ^ b0;
+	b3 += k0 + 2; b2 += b3 + k4 + t0; b3 = ((b3 << 16) | (b3 >> (64 - 16))) ^ b2;
+	b0 += b3; b3 = ((b3 << 52) | (b3 >> (64 - 52))) ^ b0;
+	b2 += b1; b1 = ((b1 << 57) | (b1 >> (64 - 57))) ^ b2;
+	b0 += b1; b1 = ((b1 << 23) | (b1 >> (64 - 23))) ^ b0;
+	b2 += b3; b3 = ((b3 << 40) | (b3 >> (64 - 40))) ^ b2;
+	b0 += b3; b3 = ((b3 << 5) | (b3 >> (64 - 5))) ^ b0;
+	b2 += b1; b1 = ((b1 << 37) | (b1 >> (64 - 37))) ^ b2;
+	b1 += k4 + t0; b0 += b1 + k3; b1 = ((b1 << 25) | (b1 >> (64 - 25))) ^ b0;
+	b3 += k1 + 3; b2 += b3 + k0 + t1; b3 = ((b3 << 33) | (b3 >> (64 - 33))) ^ b2;
+	b0 += b3; b3 = ((b3 << 46) | (b3 >> (64 - 46))) ^ b0;
+	b2 += b1; b1 = ((b1 << 12) | (b1 >> (64 - 12))) ^ b2;
+	b0 += b1; b1 = ((b1 << 58) | (b1 >> (64 - 58))) ^ b0;
+	b2 += b3; b3 = ((b3 << 22) | (b3 >> (64 - 22))) ^ b2;
+	b0 += b3; b3 = ((b3 << 32) | (b3 >> (64 - 32))) ^ b0;
+	b2 += b1; b1 = ((b1 << 32) | (b1 >> (64 - 32))) ^ b2;
 
-    b1 += k3 + t2; b0 += b1 + k2; b1 = ((b1 << 14) | (b1 >> (64 - 14))) ^ b0;
-    b3 += k0 + 2; b2 += b3 + k4 + t0; b3 = ((b3 << 16) | (b3 >> (64 - 16))) ^ b2;
-    b0 += b3; b3 = ((b3 << 52) | (b3 >> (64 - 52))) ^ b0;
-    b2 += b1; b1 = ((b1 << 57) | (b1 >> (64 - 57))) ^ b2;
-    b0 += b1; b1 = ((b1 << 23) | (b1 >> (64 - 23))) ^ b0;
-    b2 += b3; b3 = ((b3 << 40) | (b3 >> (64 - 40))) ^ b2;
-    b0 += b3; b3 = ((b3 << 5) | (b3 >> (64 - 5))) ^ b0;
-    b2 += b1; b1 = ((b1 << 37) | (b1 >> (64 - 37))) ^ b2;
-    b1 += k4 + t0; b0 += b1 + k3; b1 = ((b1 << 25) | (b1 >> (64 - 25))) ^ b0;
-    b3 += k1 + 3; b2 += b3 + k0 + t1; b3 = ((b3 << 33) | (b3 >> (64 - 33))) ^ b2;
-    b0 += b3; b3 = ((b3 << 46) | (b3 >> (64 - 46))) ^ b0;
-    b2 += b1; b1 = ((b1 << 12) | (b1 >> (64 - 12))) ^ b2;
-    b0 += b1; b1 = ((b1 << 58) | (b1 >> (64 - 58))) ^ b0;
-    b2 += b3; b3 = ((b3 << 22) | (b3 >> (64 - 22))) ^ b2;
-    b0 += b3; b3 = ((b3 << 32) | (b3 >> (64 - 32))) ^ b0;
-    b2 += b1; b1 = ((b1 << 32) | (b1 >> (64 - 32))) ^ b2;
+	b1 += k0 + t1; b0 += b1 + k4; b1 = ((b1 << 14) | (b1 >> (64 - 14))) ^ b0;
+	b3 += k2 + 4; b2 += b3 + k1 + t2; b3 = ((b3 << 16) | (b3 >> (64 - 16))) ^ b2;
+	b0 += b3; b3 = ((b3 << 52) | (b3 >> (64 - 52))) ^ b0;
+	b2 += b1; b1 = ((b1 << 57) | (b1 >> (64 - 57))) ^ b2;
+	b0 += b1; b1 = ((b1 << 23) | (b1 >> (64 - 23))) ^ b0;
+	b2 += b3; b3 = ((b3 << 40) | (b3 >> (64 - 40))) ^ b2;
+	b0 += b3; b3 = ((b3 << 5) | (b3 >> (64 - 5))) ^ b0;
+	b2 += b1; b1 = ((b1 << 37) | (b1 >> (64 - 37))) ^ b2;
+	b1 += k1 + t2; b0 += b1 + k0; b1 = ((b1 << 25) | (b1 >> (64 - 25))) ^ b0;
+	b3 += k3 + 5; b2 += b3 + k2 + t0; b3 = ((b3 << 33) | (b3 >> (64 - 33))) ^ b2;
+	b0 += b3; b3 = ((b3 << 46) | (b3 >> (64 - 46))) ^ b0;
+	b2 += b1; b1 = ((b1 << 12) | (b1 >> (64 - 12))) ^ b2;
+	b0 += b1; b1 = ((b1 << 58) | (b1 >> (64 - 58))) ^ b0;
+	b2 += b3; b3 = ((b3 << 22) | (b3 >> (64 - 22))) ^ b2;
+	b0 += b3; b3 = ((b3 << 32) | (b3 >> (64 - 32))) ^ b0;
+	b2 += b1; b1 = ((b1 << 32) | (b1 >> (64 - 32))) ^ b2;
 
-    b1 += k0 + t1; b0 += b1 + k4; b1 = ((b1 << 14) | (b1 >> (64 - 14))) ^ b0;
-    b3 += k2 + 4; b2 += b3 + k1 + t2; b3 = ((b3 << 16) | (b3 >> (64 - 16))) ^ b2;
-    b0 += b3; b3 = ((b3 << 52) | (b3 >> (64 - 52))) ^ b0;
-    b2 += b1; b1 = ((b1 << 57) | (b1 >> (64 - 57))) ^ b2;
-    b0 += b1; b1 = ((b1 << 23) | (b1 >> (64 - 23))) ^ b0;
-    b2 += b3; b3 = ((b3 << 40) | (b3 >> (64 - 40))) ^ b2;
-    b0 += b3; b3 = ((b3 << 5) | (b3 >> (64 - 5))) ^ b0;
-    b2 += b1; b1 = ((b1 << 37) | (b1 >> (64 - 37))) ^ b2;
-    b1 += k1 + t2; b0 += b1 + k0; b1 = ((b1 << 25) | (b1 >> (64 - 25))) ^ b0;
-    b3 += k3 + 5; b2 += b3 + k2 + t0; b3 = ((b3 << 33) | (b3 >> (64 - 33))) ^ b2;
-    b0 += b3; b3 = ((b3 << 46) | (b3 >> (64 - 46))) ^ b0;
-    b2 += b1; b1 = ((b1 << 12) | (b1 >> (64 - 12))) ^ b2;
-    b0 += b1; b1 = ((b1 << 58) | (b1 >> (64 - 58))) ^ b0;
-    b2 += b3; b3 = ((b3 << 22) | (b3 >> (64 - 22))) ^ b2;
-    b0 += b3; b3 = ((b3 << 32) | (b3 >> (64 - 32))) ^ b0;
-    b2 += b1; b1 = ((b1 << 32) | (b1 >> (64 - 32))) ^ b2;
+	b1 += k2 + t0; b0 += b1 + k1; b1 = ((b1 << 14) | (b1 >> (64 - 14))) ^ b0;
+	b3 += k4 + 6; b2 += b3 + k3 + t1; b3 = ((b3 << 16) | (b3 >> (64 - 16))) ^ b2;
+	b0 += b3; b3 = ((b3 << 52) | (b3 >> (64 - 52))) ^ b0;
+	b2 += b1; b1 = ((b1 << 57) | (b1 >> (64 - 57))) ^ b2;
+	b0 += b1; b1 = ((b1 << 23) | (b1 >> (64 - 23))) ^ b0;
+	b2 += b3; b3 = ((b3 << 40) | (b3 >> (64 - 40))) ^ b2;
+	b0 += b3; b3 = ((b3 << 5) | (b3 >> (64 - 5))) ^ b0;
+	b2 += b1; b1 = ((b1 << 37) | (b1 >> (64 - 37))) ^ b2;
+	b1 += k3 + t1; b0 += b1 + k2; b1 = ((b1 << 25) | (b1 >> (64 - 25))) ^ b0;
+	b3 += k0 + 7; b2 += b3 + k4 + t2; b3 = ((b3 << 33) | (b3 >> (64 - 33))) ^ b2;
+	b0 += b3; b3 = ((b3 << 46) | (b3 >> (64 - 46))) ^ b0;
+	b2 += b1; b1 = ((b1 << 12) | (b1 >> (64 - 12))) ^ b2;
+	b0 += b1; b1 = ((b1 << 58) | (b1 >> (64 - 58))) ^ b0;
+	b2 += b3; b3 = ((b3 << 22) | (b3 >> (64 - 22))) ^ b2;
+	b0 += b3; b3 = ((b3 << 32) | (b3 >> (64 - 32))) ^ b0;
+	b2 += b1; b1 = ((b1 << 32) | (b1 >> (64 - 32))) ^ b2;
 
-    b1 += k2 + t0; b0 += b1 + k1; b1 = ((b1 << 14) | (b1 >> (64 - 14))) ^ b0;
-    b3 += k4 + 6; b2 += b3 + k3 + t1; b3 = ((b3 << 16) | (b3 >> (64 - 16))) ^ b2;
-    b0 += b3; b3 = ((b3 << 52) | (b3 >> (64 - 52))) ^ b0;
-    b2 += b1; b1 = ((b1 << 57) | (b1 >> (64 - 57))) ^ b2;
-    b0 += b1; b1 = ((b1 << 23) | (b1 >> (64 - 23))) ^ b0;
-    b2 += b3; b3 = ((b3 << 40) | (b3 >> (64 - 40))) ^ b2;
-    b0 += b3; b3 = ((b3 << 5) | (b3 >> (64 - 5))) ^ b0;
-    b2 += b1; b1 = ((b1 << 37) | (b1 >> (64 - 37))) ^ b2;
-    b1 += k3 + t1; b0 += b1 + k2; b1 = ((b1 << 25) | (b1 >> (64 - 25))) ^ b0;
-    b3 += k0 + 7; b2 += b3 + k4 + t2; b3 = ((b3 << 33) | (b3 >> (64 - 33))) ^ b2;
-    b0 += b3; b3 = ((b3 << 46) | (b3 >> (64 - 46))) ^ b0;
-    b2 += b1; b1 = ((b1 << 12) | (b1 >> (64 - 12))) ^ b2;
-    b0 += b1; b1 = ((b1 << 58) | (b1 >> (64 - 58))) ^ b0;
-    b2 += b3; b3 = ((b3 << 22) | (b3 >> (64 - 22))) ^ b2;
-    b0 += b3; b3 = ((b3 << 32) | (b3 >> (64 - 32))) ^ b0;
-    b2 += b1; b1 = ((b1 << 32) | (b1 >> (64 - 32))) ^ b2;
+	b1 += k4 + t2; b0 += b1 + k3; b1 = ((b1 << 14) | (b1 >> (64 - 14))) ^ b0;
+	b3 += k1 + 8; b2 += b3 + k0 + t0; b3 = ((b3 << 16) | (b3 >> (64 - 16))) ^ b2;
+	b0 += b3; b3 = ((b3 << 52) | (b3 >> (64 - 52))) ^ b0;
+	b2 += b1; b1 = ((b1 << 57) | (b1 >> (64 - 57))) ^ b2;
+	b0 += b1; b1 = ((b1 << 23) | (b1 >> (64 - 23))) ^ b0;
+	b2 += b3; b3 = ((b3 << 40) | (b3 >> (64 - 40))) ^ b2;
+	b0 += b3; b3 = ((b3 << 5) | (b3 >> (64 - 5))) ^ b0;
+	b2 += b1; b1 = ((b1 << 37) | (b1 >> (64 - 37))) ^ b2;
+	b1 += k0 + t0; b0 += b1 + k4; b1 = ((b1 << 25) | (b1 >> (64 - 25))) ^ b0;
+	b3 += k2 + 9; b2 += b3 + k1 + t1; b3 = ((b3 << 33) | (b3 >> (64 - 33))) ^ b2;
+	b0 += b3; b3 = ((b3 << 46) | (b3 >> (64 - 46))) ^ b0;
+	b2 += b1; b1 = ((b1 << 12) | (b1 >> (64 - 12))) ^ b2;
+	b0 += b1; b1 = ((b1 << 58) | (b1 >> (64 - 58))) ^ b0;
+	b2 += b3; b3 = ((b3 << 22) | (b3 >> (64 - 22))) ^ b2;
+	b0 += b3; b3 = ((b3 << 32) | (b3 >> (64 - 32))) ^ b0;
+	b2 += b1; b1 = ((b1 << 32) | (b1 >> (64 - 32))) ^ b2;
 
-    b1 += k4 + t2; b0 += b1 + k3; b1 = ((b1 << 14) | (b1 >> (64 - 14))) ^ b0;
-    b3 += k1 + 8; b2 += b3 + k0 + t0; b3 = ((b3 << 16) | (b3 >> (64 - 16))) ^ b2;
-    b0 += b3; b3 = ((b3 << 52) | (b3 >> (64 - 52))) ^ b0;
-    b2 += b1; b1 = ((b1 << 57) | (b1 >> (64 - 57))) ^ b2;
-    b0 += b1; b1 = ((b1 << 23) | (b1 >> (64 - 23))) ^ b0;
-    b2 += b3; b3 = ((b3 << 40) | (b3 >> (64 - 40))) ^ b2;
-    b0 += b3; b3 = ((b3 << 5) | (b3 >> (64 - 5))) ^ b0;
-    b2 += b1; b1 = ((b1 << 37) | (b1 >> (64 - 37))) ^ b2;
-    b1 += k0 + t0; b0 += b1 + k4; b1 = ((b1 << 25) | (b1 >> (64 - 25))) ^ b0;
-    b3 += k2 + 9; b2 += b3 + k1 + t1; b3 = ((b3 << 33) | (b3 >> (64 - 33))) ^ b2;
-    b0 += b3; b3 = ((b3 << 46) | (b3 >> (64 - 46))) ^ b0;
-    b2 += b1; b1 = ((b1 << 12) | (b1 >> (64 - 12))) ^ b2;
-    b0 += b1; b1 = ((b1 << 58) | (b1 >> (64 - 58))) ^ b0;
-    b2 += b3; b3 = ((b3 << 22) | (b3 >> (64 - 22))) ^ b2;
-    b0 += b3; b3 = ((b3 << 32) | (b3 >> (64 - 32))) ^ b0;
-    b2 += b1; b1 = ((b1 << 32) | (b1 >> (64 - 32))) ^ b2;
+	b1 += k1 + t1; b0 += b1 + k0; b1 = ((b1 << 14) | (b1 >> (64 - 14))) ^ b0;
+	b3 += k3 + 10; b2 += b3 + k2 + t2; b3 = ((b3 << 16) | (b3 >> (64 - 16))) ^ b2;
+	b0 += b3; b3 = ((b3 << 52) | (b3 >> (64 - 52))) ^ b0;
+	b2 += b1; b1 = ((b1 << 57) | (b1 >> (64 - 57))) ^ b2;
+	b0 += b1; b1 = ((b1 << 23) | (b1 >> (64 - 23))) ^ b0;
+	b2 += b3; b3 = ((b3 << 40) | (b3 >> (64 - 40))) ^ b2;
+	b0 += b3; b3 = ((b3 << 5) | (b3 >> (64 - 5))) ^ b0;
+	b2 += b1; b1 = ((b1 << 37) | (b1 >> (64 - 37))) ^ b2;
+	b1 += k2 + t2; b0 += b1 + k1; b1 = ((b1 << 25) | (b1 >> (64 - 25))) ^ b0;
+	b3 += k4 + 11; b2 += b3 + k3 + t0; b3 = ((b3 << 33) | (b3 >> (64 - 33))) ^ b2;
+	b0 += b3; b3 = ((b3 << 46) | (b3 >> (64 - 46))) ^ b0;
+	b2 += b1; b1 = ((b1 << 12) | (b1 >> (64 - 12))) ^ b2;
+	b0 += b1; b1 = ((b1 << 58) | (b1 >> (64 - 58))) ^ b0;
+	b2 += b3; b3 = ((b3 << 22) | (b3 >> (64 - 22))) ^ b2;
+	b0 += b3; b3 = ((b3 << 32) | (b3 >> (64 - 32))) ^ b0;
+	b2 += b1; b1 = ((b1 << 32) | (b1 >> (64 - 32))) ^ b2;
 
-    b1 += k1 + t1; b0 += b1 + k0; b1 = ((b1 << 14) | (b1 >> (64 - 14))) ^ b0;
-    b3 += k3 + 10; b2 += b3 + k2 + t2; b3 = ((b3 << 16) | (b3 >> (64 - 16))) ^ b2;
-    b0 += b3; b3 = ((b3 << 52) | (b3 >> (64 - 52))) ^ b0;
-    b2 += b1; b1 = ((b1 << 57) | (b1 >> (64 - 57))) ^ b2;
-    b0 += b1; b1 = ((b1 << 23) | (b1 >> (64 - 23))) ^ b0;
-    b2 += b3; b3 = ((b3 << 40) | (b3 >> (64 - 40))) ^ b2;
-    b0 += b3; b3 = ((b3 << 5) | (b3 >> (64 - 5))) ^ b0;
-    b2 += b1; b1 = ((b1 << 37) | (b1 >> (64 - 37))) ^ b2;
-    b1 += k2 + t2; b0 += b1 + k1; b1 = ((b1 << 25) | (b1 >> (64 - 25))) ^ b0;
-    b3 += k4 + 11; b2 += b3 + k3 + t0; b3 = ((b3 << 33) | (b3 >> (64 - 33))) ^ b2;
-    b0 += b3; b3 = ((b3 << 46) | (b3 >> (64 - 46))) ^ b0;
-    b2 += b1; b1 = ((b1 << 12) | (b1 >> (64 - 12))) ^ b2;
-    b0 += b1; b1 = ((b1 << 58) | (b1 >> (64 - 58))) ^ b0;
-    b2 += b3; b3 = ((b3 << 22) | (b3 >> (64 - 22))) ^ b2;
-    b0 += b3; b3 = ((b3 << 32) | (b3 >> (64 - 32))) ^ b0;
-    b2 += b1; b1 = ((b1 << 32) | (b1 >> (64 - 32))) ^ b2;
+	b1 += k3 + t0; b0 += b1 + k2; b1 = ((b1 << 14) | (b1 >> (64 - 14))) ^ b0;
+	b3 += k0 + 12; b2 += b3 + k4 + t1; b3 = ((b3 << 16) | (b3 >> (64 - 16))) ^ b2;
+	b0 += b3; b3 = ((b3 << 52) | (b3 >> (64 - 52))) ^ b0;
+	b2 += b1; b1 = ((b1 << 57) | (b1 >> (64 - 57))) ^ b2;
+	b0 += b1; b1 = ((b1 << 23) | (b1 >> (64 - 23))) ^ b0;
+	b2 += b3; b3 = ((b3 << 40) | (b3 >> (64 - 40))) ^ b2;
+	b0 += b3; b3 = ((b3 << 5) | (b3 >> (64 - 5))) ^ b0;
+	b2 += b1; b1 = ((b1 << 37) | (b1 >> (64 - 37))) ^ b2;
+	b1 += k4 + t1; b0 += b1 + k3; b1 = ((b1 << 25) | (b1 >> (64 - 25))) ^ b0;
+	b3 += k1 + 13; b2 += b3 + k0 + t2; b3 = ((b3 << 33) | (b3 >> (64 - 33))) ^ b2;
+	b0 += b3; b3 = ((b3 << 46) | (b3 >> (64 - 46))) ^ b0;
+	b2 += b1; b1 = ((b1 << 12) | (b1 >> (64 - 12))) ^ b2;
+	b0 += b1; b1 = ((b1 << 58) | (b1 >> (64 - 58))) ^ b0;
+	b2 += b3; b3 = ((b3 << 22) | (b3 >> (64 - 22))) ^ b2;
+	b0 += b3; b3 = ((b3 << 32) | (b3 >> (64 - 32))) ^ b0;
+	b2 += b1; b1 = ((b1 << 32) | (b1 >> (64 - 32))) ^ b2;
 
-    b1 += k3 + t0; b0 += b1 + k2; b1 = ((b1 << 14) | (b1 >> (64 - 14))) ^ b0;
-    b3 += k0 + 12; b2 += b3 + k4 + t1; b3 = ((b3 << 16) | (b3 >> (64 - 16))) ^ b2;
-    b0 += b3; b3 = ((b3 << 52) | (b3 >> (64 - 52))) ^ b0;
-    b2 += b1; b1 = ((b1 << 57) | (b1 >> (64 - 57))) ^ b2;
-    b0 += b1; b1 = ((b1 << 23) | (b1 >> (64 - 23))) ^ b0;
-    b2 += b3; b3 = ((b3 << 40) | (b3 >> (64 - 40))) ^ b2;
-    b0 += b3; b3 = ((b3 << 5) | (b3 >> (64 - 5))) ^ b0;
-    b2 += b1; b1 = ((b1 << 37) | (b1 >> (64 - 37))) ^ b2;
-    b1 += k4 + t1; b0 += b1 + k3; b1 = ((b1 << 25) | (b1 >> (64 - 25))) ^ b0;
-    b3 += k1 + 13; b2 += b3 + k0 + t2; b3 = ((b3 << 33) | (b3 >> (64 - 33))) ^ b2;
-    b0 += b3; b3 = ((b3 << 46) | (b3 >> (64 - 46))) ^ b0;
-    b2 += b1; b1 = ((b1 << 12) | (b1 >> (64 - 12))) ^ b2;
-    b0 += b1; b1 = ((b1 << 58) | (b1 >> (64 - 58))) ^ b0;
-    b2 += b3; b3 = ((b3 << 22) | (b3 >> (64 - 22))) ^ b2;
-    b0 += b3; b3 = ((b3 << 32) | (b3 >> (64 - 32))) ^ b0;
-    b2 += b1; b1 = ((b1 << 32) | (b1 >> (64 - 32))) ^ b2;
+	b1 += k0 + t2; b0 += b1 + k4; b1 = ((b1 << 14) | (b1 >> (64 - 14))) ^ b0;
+	b3 += k2 + 14; b2 += b3 + k1 + t0; b3 = ((b3 << 16) | (b3 >> (64 - 16))) ^ b2;
+	b0 += b3; b3 = ((b3 << 52) | (b3 >> (64 - 52))) ^ b0;
+	b2 += b1; b1 = ((b1 << 57) | (b1 >> (64 - 57))) ^ b2;
+	b0 += b1; b1 = ((b1 << 23) | (b1 >> (64 - 23))) ^ b0;
+	b2 += b3; b3 = ((b3 << 40) | (b3 >> (64 - 40))) ^ b2;
+	b0 += b3; b3 = ((b3 << 5) | (b3 >> (64 - 5))) ^ b0;
+	b2 += b1; b1 = ((b1 << 37) | (b1 >> (64 - 37))) ^ b2;
+	b1 += k1 + t0; b0 += b1 + k0; b1 = ((b1 << 25) | (b1 >> (64 - 25))) ^ b0;
+	b3 += k3 + 15; b2 += b3 + k2 + t1; b3 = ((b3 << 33) | (b3 >> (64 - 33))) ^ b2;
+	b0 += b3; b3 = ((b3 << 46) | (b3 >> (64 - 46))) ^ b0;
+	b2 += b1; b1 = ((b1 << 12) | (b1 >> (64 - 12))) ^ b2;
+	b0 += b1; b1 = ((b1 << 58) | (b1 >> (64 - 58))) ^ b0;
+	b2 += b3; b3 = ((b3 << 22) | (b3 >> (64 - 22))) ^ b2;
+	b0 += b3; b3 = ((b3 << 32) | (b3 >> (64 - 32))) ^ b0;
+	b2 += b1; b1 = ((b1 << 32) | (b1 >> (64 - 32))) ^ b2;
 
-    b1 += k0 + t2; b0 += b1 + k4; b1 = ((b1 << 14) | (b1 >> (64 - 14))) ^ b0;
-    b3 += k2 + 14; b2 += b3 + k1 + t0; b3 = ((b3 << 16) | (b3 >> (64 - 16))) ^ b2;
-    b0 += b3; b3 = ((b3 << 52) | (b3 >> (64 - 52))) ^ b0;
-    b2 += b1; b1 = ((b1 << 57) | (b1 >> (64 - 57))) ^ b2;
-    b0 += b1; b1 = ((b1 << 23) | (b1 >> (64 - 23))) ^ b0;
-    b2 += b3; b3 = ((b3 << 40) | (b3 >> (64 - 40))) ^ b2;
-    b0 += b3; b3 = ((b3 << 5) | (b3 >> (64 - 5))) ^ b0;
-    b2 += b1; b1 = ((b1 << 37) | (b1 >> (64 - 37))) ^ b2;
-    b1 += k1 + t0; b0 += b1 + k0; b1 = ((b1 << 25) | (b1 >> (64 - 25))) ^ b0;
-    b3 += k3 + 15; b2 += b3 + k2 + t1; b3 = ((b3 << 33) | (b3 >> (64 - 33))) ^ b2;
-    b0 += b3; b3 = ((b3 << 46) | (b3 >> (64 - 46))) ^ b0;
-    b2 += b1; b1 = ((b1 << 12) | (b1 >> (64 - 12))) ^ b2;
-    b0 += b1; b1 = ((b1 << 58) | (b1 >> (64 - 58))) ^ b0;
-    b2 += b3; b3 = ((b3 << 22) | (b3 >> (64 - 22))) ^ b2;
-    b0 += b3; b3 = ((b3 << 32) | (b3 >> (64 - 32))) ^ b0;
-    b2 += b1; b1 = ((b1 << 32) | (b1 >> (64 - 32))) ^ b2;
+	b1 += k2 + t1; b0 += b1 + k1; b1 = ((b1 << 14) | (b1 >> (64 - 14))) ^ b0;
+	b3 += k4 + 16; b2 += b3 + k3 + t2; b3 = ((b3 << 16) | (b3 >> (64 - 16))) ^ b2;
+	b0 += b3; b3 = ((b3 << 52) | (b3 >> (64 - 52))) ^ b0;
+	b2 += b1; b1 = ((b1 << 57) | (b1 >> (64 - 57))) ^ b2;
+	b0 += b1; b1 = ((b1 << 23) | (b1 >> (64 - 23))) ^ b0;
+	b2 += b3; b3 = ((b3 << 40) | (b3 >> (64 - 40))) ^ b2;
+	b0 += b3; b3 = ((b3 << 5) | (b3 >> (64 - 5))) ^ b0;
+	b2 += b1; b1 = ((b1 << 37) | (b1 >> (64 - 37))) ^ b2;
+	b1 += k3 + t2; b0 += b1 + k2; b1 = ((b1 << 25) | (b1 >> (64 - 25))) ^ b0;
+	b3 += k0 + 17; b2 += b3 + k4 + t0; b3 = ((b3 << 33) | (b3 >> (64 - 33))) ^ b2;
+	b0 += b3; b3 = ((b3 << 46) | (b3 >> (64 - 46))) ^ b0;
+	b2 += b1; b1 = ((b1 << 12) | (b1 >> (64 - 12))) ^ b2;
+	b0 += b1; b1 = ((b1 << 58) | (b1 >> (64 - 58))) ^ b0;
+	b2 += b3; b3 = ((b3 << 22) | (b3 >> (64 - 22))) ^ b2;
+	b0 += b3; b3 = ((b3 << 32) | (b3 >> (64 - 32))) ^ b0;
+	b2 += b1; b1 = ((b1 << 32) | (b1 >> (64 - 32))) ^ b2;
 
-    b1 += k2 + t1; b0 += b1 + k1; b1 = ((b1 << 14) | (b1 >> (64 - 14))) ^ b0;
-    b3 += k4 + 16; b2 += b3 + k3 + t2; b3 = ((b3 << 16) | (b3 >> (64 - 16))) ^ b2;
-    b0 += b3; b3 = ((b3 << 52) | (b3 >> (64 - 52))) ^ b0;
-    b2 += b1; b1 = ((b1 << 57) | (b1 >> (64 - 57))) ^ b2;
-    b0 += b1; b1 = ((b1 << 23) | (b1 >> (64 - 23))) ^ b0;
-    b2 += b3; b3 = ((b3 << 40) | (b3 >> (64 - 40))) ^ b2;
-    b0 += b3; b3 = ((b3 << 5) | (b3 >> (64 - 5))) ^ b0;
-    b2 += b1; b1 = ((b1 << 37) | (b1 >> (64 - 37))) ^ b2;
-    b1 += k3 + t2; b0 += b1 + k2; b1 = ((b1 << 25) | (b1 >> (64 - 25))) ^ b0;
-    b3 += k0 + 17; b2 += b3 + k4 + t0; b3 = ((b3 << 33) | (b3 >> (64 - 33))) ^ b2;
-    b0 += b3; b3 = ((b3 << 46) | (b3 >> (64 - 46))) ^ b0;
-    b2 += b1; b1 = ((b1 << 12) | (b1 >> (64 - 12))) ^ b2;
-    b0 += b1; b1 = ((b1 << 58) | (b1 >> (64 - 58))) ^ b0;
-    b2 += b3; b3 = ((b3 << 22) | (b3 >> (64 - 22))) ^ b2;
-    b0 += b3; b3 = ((b3 << 32) | (b3 >> (64 - 32))) ^ b0;
-    b2 += b1; b1 = ((b1 << 32) | (b1 >> (64 - 32))) ^ b2;
-
-    output[0] = b0 + k3;
-    output[1] = b1 + k4 + t0;
-    output[2] = b2 + k0 + t1;
-    output[3] = b3 + k1 + 18;
-  }
+	output[0] = b0 + k3;
+	output[1] = b1 + k4 + t0;
+	output[2] = b2 + k0 + t1;
+	output[3] = b3 + k1 + 18;
+}
 
 void threefishDecrypt256(struct threefish_key *keyCtx, u64 *input, u64 *output)
-  {
-    u64 b0 = input[0], b1 = input[1],
-      b2 = input[2], b3 = input[3];
-    u64 k0 = keyCtx->key[0], k1 = keyCtx->key[1],
-      k2 = keyCtx->key[2], k3 = keyCtx->key[3],
-      k4 = keyCtx->key[4];
-    u64 t0 = keyCtx->tweak[0], t1 = keyCtx->tweak[1],
-      t2 = keyCtx->tweak[2];
+{
+	u64 b0 = input[0], b1 = input[1],
+	  b2 = input[2], b3 = input[3];
+	u64 k0 = keyCtx->key[0], k1 = keyCtx->key[1],
+	  k2 = keyCtx->key[2], k3 = keyCtx->key[3],
+	  k4 = keyCtx->key[4];
+	u64 t0 = keyCtx->tweak[0], t1 = keyCtx->tweak[1],
+	  t2 = keyCtx->tweak[2];
 
-    u64 tmp;
+	u64 tmp;
 
-    b0 -= k3;
-    b1 -= k4 + t0;
-    b2 -= k0 + t1;
-    b3 -= k1 + 18;
-    tmp = b3 ^ b0; b3 = (tmp >> 32) | (tmp << (64 - 32)); b0 -= b3;
-    tmp = b1 ^ b2; b1 = (tmp >> 32) | (tmp << (64 - 32)); b2 -= b1;
-    tmp = b1 ^ b0; b1 = (tmp >> 58) | (tmp << (64 - 58)); b0 -= b1;
-    tmp = b3 ^ b2; b3 = (tmp >> 22) | (tmp << (64 - 22)); b2 -= b3;
-    tmp = b3 ^ b0; b3 = (tmp >> 46) | (tmp << (64 - 46)); b0 -= b3;
-    tmp = b1 ^ b2; b1 = (tmp >> 12) | (tmp << (64 - 12)); b2 -= b1;
-    tmp = b1 ^ b0; b1 = (tmp >> 25) | (tmp << (64 - 25)); b0 -= b1 + k2; b1 -= k3 + t2;
-    tmp = b3 ^ b2; b3 = (tmp >> 33) | (tmp << (64 - 33)); b2 -= b3 + k4 + t0; b3 -= k0 + 17;
-    tmp = b3 ^ b0; b3 = (tmp >> 5) | (tmp << (64 - 5)); b0 -= b3;
-    tmp = b1 ^ b2; b1 = (tmp >> 37) | (tmp << (64 - 37)); b2 -= b1;
-    tmp = b1 ^ b0; b1 = (tmp >> 23) | (tmp << (64 - 23)); b0 -= b1;
-    tmp = b3 ^ b2; b3 = (tmp >> 40) | (tmp << (64 - 40)); b2 -= b3;
-    tmp = b3 ^ b0; b3 = (tmp >> 52) | (tmp << (64 - 52)); b0 -= b3;
-    tmp = b1 ^ b2; b1 = (tmp >> 57) | (tmp << (64 - 57)); b2 -= b1;
-    tmp = b1 ^ b0; b1 = (tmp >> 14) | (tmp << (64 - 14)); b0 -= b1 + k1; b1 -= k2 + t1;
-    tmp = b3 ^ b2; b3 = (tmp >> 16) | (tmp << (64 - 16)); b2 -= b3 + k3 + t2; b3 -= k4 + 16;
+	b0 -= k3;
+	b1 -= k4 + t0;
+	b2 -= k0 + t1;
+	b3 -= k1 + 18;
+	tmp = b3 ^ b0; b3 = (tmp >> 32) | (tmp << (64 - 32)); b0 -= b3;
+	tmp = b1 ^ b2; b1 = (tmp >> 32) | (tmp << (64 - 32)); b2 -= b1;
+	tmp = b1 ^ b0; b1 = (tmp >> 58) | (tmp << (64 - 58)); b0 -= b1;
+	tmp = b3 ^ b2; b3 = (tmp >> 22) | (tmp << (64 - 22)); b2 -= b3;
+	tmp = b3 ^ b0; b3 = (tmp >> 46) | (tmp << (64 - 46)); b0 -= b3;
+	tmp = b1 ^ b2; b1 = (tmp >> 12) | (tmp << (64 - 12)); b2 -= b1;
+	tmp = b1 ^ b0; b1 = (tmp >> 25) | (tmp << (64 - 25)); b0 -= b1 + k2; b1 -= k3 + t2;
+	tmp = b3 ^ b2; b3 = (tmp >> 33) | (tmp << (64 - 33)); b2 -= b3 + k4 + t0; b3 -= k0 + 17;
+	tmp = b3 ^ b0; b3 = (tmp >> 5) | (tmp << (64 - 5)); b0 -= b3;
+	tmp = b1 ^ b2; b1 = (tmp >> 37) | (tmp << (64 - 37)); b2 -= b1;
+	tmp = b1 ^ b0; b1 = (tmp >> 23) | (tmp << (64 - 23)); b0 -= b1;
+	tmp = b3 ^ b2; b3 = (tmp >> 40) | (tmp << (64 - 40)); b2 -= b3;
+	tmp = b3 ^ b0; b3 = (tmp >> 52) | (tmp << (64 - 52)); b0 -= b3;
+	tmp = b1 ^ b2; b1 = (tmp >> 57) | (tmp << (64 - 57)); b2 -= b1;
+	tmp = b1 ^ b0; b1 = (tmp >> 14) | (tmp << (64 - 14)); b0 -= b1 + k1; b1 -= k2 + t1;
+	tmp = b3 ^ b2; b3 = (tmp >> 16) | (tmp << (64 - 16)); b2 -= b3 + k3 + t2; b3 -= k4 + 16;
 
-    tmp = b3 ^ b0; b3 = (tmp >> 32) | (tmp << (64 - 32)); b0 -= b3;
-    tmp = b1 ^ b2; b1 = (tmp >> 32) | (tmp << (64 - 32)); b2 -= b1;
-    tmp = b1 ^ b0; b1 = (tmp >> 58) | (tmp << (64 - 58)); b0 -= b1;
-    tmp = b3 ^ b2; b3 = (tmp >> 22) | (tmp << (64 - 22)); b2 -= b3;
-    tmp = b3 ^ b0; b3 = (tmp >> 46) | (tmp << (64 - 46)); b0 -= b3;
-    tmp = b1 ^ b2; b1 = (tmp >> 12) | (tmp << (64 - 12)); b2 -= b1;
-    tmp = b1 ^ b0; b1 = (tmp >> 25) | (tmp << (64 - 25)); b0 -= b1 + k0; b1 -= k1 + t0;
-    tmp = b3 ^ b2; b3 = (tmp >> 33) | (tmp << (64 - 33)); b2 -= b3 + k2 + t1; b3 -= k3 + 15;
-    tmp = b3 ^ b0; b3 = (tmp >> 5) | (tmp << (64 - 5)); b0 -= b3;
-    tmp = b1 ^ b2; b1 = (tmp >> 37) | (tmp << (64 - 37)); b2 -= b1;
-    tmp = b1 ^ b0; b1 = (tmp >> 23) | (tmp << (64 - 23)); b0 -= b1;
-    tmp = b3 ^ b2; b3 = (tmp >> 40) | (tmp << (64 - 40)); b2 -= b3;
-    tmp = b3 ^ b0; b3 = (tmp >> 52) | (tmp << (64 - 52)); b0 -= b3;
-    tmp = b1 ^ b2; b1 = (tmp >> 57) | (tmp << (64 - 57)); b2 -= b1;
-    tmp = b1 ^ b0; b1 = (tmp >> 14) | (tmp << (64 - 14)); b0 -= b1 + k4; b1 -= k0 + t2;
-    tmp = b3 ^ b2; b3 = (tmp >> 16) | (tmp << (64 - 16)); b2 -= b3 + k1 + t0; b3 -= k2 + 14;
+	tmp = b3 ^ b0; b3 = (tmp >> 32) | (tmp << (64 - 32)); b0 -= b3;
+	tmp = b1 ^ b2; b1 = (tmp >> 32) | (tmp << (64 - 32)); b2 -= b1;
+	tmp = b1 ^ b0; b1 = (tmp >> 58) | (tmp << (64 - 58)); b0 -= b1;
+	tmp = b3 ^ b2; b3 = (tmp >> 22) | (tmp << (64 - 22)); b2 -= b3;
+	tmp = b3 ^ b0; b3 = (tmp >> 46) | (tmp << (64 - 46)); b0 -= b3;
+	tmp = b1 ^ b2; b1 = (tmp >> 12) | (tmp << (64 - 12)); b2 -= b1;
+	tmp = b1 ^ b0; b1 = (tmp >> 25) | (tmp << (64 - 25)); b0 -= b1 + k0; b1 -= k1 + t0;
+	tmp = b3 ^ b2; b3 = (tmp >> 33) | (tmp << (64 - 33)); b2 -= b3 + k2 + t1; b3 -= k3 + 15;
+	tmp = b3 ^ b0; b3 = (tmp >> 5) | (tmp << (64 - 5)); b0 -= b3;
+	tmp = b1 ^ b2; b1 = (tmp >> 37) | (tmp << (64 - 37)); b2 -= b1;
+	tmp = b1 ^ b0; b1 = (tmp >> 23) | (tmp << (64 - 23)); b0 -= b1;
+	tmp = b3 ^ b2; b3 = (tmp >> 40) | (tmp << (64 - 40)); b2 -= b3;
+	tmp = b3 ^ b0; b3 = (tmp >> 52) | (tmp << (64 - 52)); b0 -= b3;
+	tmp = b1 ^ b2; b1 = (tmp >> 57) | (tmp << (64 - 57)); b2 -= b1;
+	tmp = b1 ^ b0; b1 = (tmp >> 14) | (tmp << (64 - 14)); b0 -= b1 + k4; b1 -= k0 + t2;
+	tmp = b3 ^ b2; b3 = (tmp >> 16) | (tmp << (64 - 16)); b2 -= b3 + k1 + t0; b3 -= k2 + 14;
 
-    tmp = b3 ^ b0; b3 = (tmp >> 32) | (tmp << (64 - 32)); b0 -= b3;
-    tmp = b1 ^ b2; b1 = (tmp >> 32) | (tmp << (64 - 32)); b2 -= b1;
-    tmp = b1 ^ b0; b1 = (tmp >> 58) | (tmp << (64 - 58)); b0 -= b1;
-    tmp = b3 ^ b2; b3 = (tmp >> 22) | (tmp << (64 - 22)); b2 -= b3;
-    tmp = b3 ^ b0; b3 = (tmp >> 46) | (tmp << (64 - 46)); b0 -= b3;
-    tmp = b1 ^ b2; b1 = (tmp >> 12) | (tmp << (64 - 12)); b2 -= b1;
-    tmp = b1 ^ b0; b1 = (tmp >> 25) | (tmp << (64 - 25)); b0 -= b1 + k3; b1 -= k4 + t1;
-    tmp = b3 ^ b2; b3 = (tmp >> 33) | (tmp << (64 - 33)); b2 -= b3 + k0 + t2; b3 -= k1 + 13;
-    tmp = b3 ^ b0; b3 = (tmp >> 5) | (tmp << (64 - 5)); b0 -= b3;
-    tmp = b1 ^ b2; b1 = (tmp >> 37) | (tmp << (64 - 37)); b2 -= b1;
-    tmp = b1 ^ b0; b1 = (tmp >> 23) | (tmp << (64 - 23)); b0 -= b1;
-    tmp = b3 ^ b2; b3 = (tmp >> 40) | (tmp << (64 - 40)); b2 -= b3;
-    tmp = b3 ^ b0; b3 = (tmp >> 52) | (tmp << (64 - 52)); b0 -= b3;
-    tmp = b1 ^ b2; b1 = (tmp >> 57) | (tmp << (64 - 57)); b2 -= b1;
-    tmp = b1 ^ b0; b1 = (tmp >> 14) | (tmp << (64 - 14)); b0 -= b1 + k2; b1 -= k3 + t0;
-    tmp = b3 ^ b2; b3 = (tmp >> 16) | (tmp << (64 - 16)); b2 -= b3 + k4 + t1; b3 -= k0 + 12;
+	tmp = b3 ^ b0; b3 = (tmp >> 32) | (tmp << (64 - 32)); b0 -= b3;
+	tmp = b1 ^ b2; b1 = (tmp >> 32) | (tmp << (64 - 32)); b2 -= b1;
+	tmp = b1 ^ b0; b1 = (tmp >> 58) | (tmp << (64 - 58)); b0 -= b1;
+	tmp = b3 ^ b2; b3 = (tmp >> 22) | (tmp << (64 - 22)); b2 -= b3;
+	tmp = b3 ^ b0; b3 = (tmp >> 46) | (tmp << (64 - 46)); b0 -= b3;
+	tmp = b1 ^ b2; b1 = (tmp >> 12) | (tmp << (64 - 12)); b2 -= b1;
+	tmp = b1 ^ b0; b1 = (tmp >> 25) | (tmp << (64 - 25)); b0 -= b1 + k3; b1 -= k4 + t1;
+	tmp = b3 ^ b2; b3 = (tmp >> 33) | (tmp << (64 - 33)); b2 -= b3 + k0 + t2; b3 -= k1 + 13;
+	tmp = b3 ^ b0; b3 = (tmp >> 5) | (tmp << (64 - 5)); b0 -= b3;
+	tmp = b1 ^ b2; b1 = (tmp >> 37) | (tmp << (64 - 37)); b2 -= b1;
+	tmp = b1 ^ b0; b1 = (tmp >> 23) | (tmp << (64 - 23)); b0 -= b1;
+	tmp = b3 ^ b2; b3 = (tmp >> 40) | (tmp << (64 - 40)); b2 -= b3;
+	tmp = b3 ^ b0; b3 = (tmp >> 52) | (tmp << (64 - 52)); b0 -= b3;
+	tmp = b1 ^ b2; b1 = (tmp >> 57) | (tmp << (64 - 57)); b2 -= b1;
+	tmp = b1 ^ b0; b1 = (tmp >> 14) | (tmp << (64 - 14)); b0 -= b1 + k2; b1 -= k3 + t0;
+	tmp = b3 ^ b2; b3 = (tmp >> 16) | (tmp << (64 - 16)); b2 -= b3 + k4 + t1; b3 -= k0 + 12;
 
-    tmp = b3 ^ b0; b3 = (tmp >> 32) | (tmp << (64 - 32)); b0 -= b3;
-    tmp = b1 ^ b2; b1 = (tmp >> 32) | (tmp << (64 - 32)); b2 -= b1;
-    tmp = b1 ^ b0; b1 = (tmp >> 58) | (tmp << (64 - 58)); b0 -= b1;
-    tmp = b3 ^ b2; b3 = (tmp >> 22) | (tmp << (64 - 22)); b2 -= b3;
-    tmp = b3 ^ b0; b3 = (tmp >> 46) | (tmp << (64 - 46)); b0 -= b3;
-    tmp = b1 ^ b2; b1 = (tmp >> 12) | (tmp << (64 - 12)); b2 -= b1;
-    tmp = b1 ^ b0; b1 = (tmp >> 25) | (tmp << (64 - 25)); b0 -= b1 + k1; b1 -= k2 + t2;
-    tmp = b3 ^ b2; b3 = (tmp >> 33) | (tmp << (64 - 33)); b2 -= b3 + k3 + t0; b3 -= k4 + 11;
-    tmp = b3 ^ b0; b3 = (tmp >> 5) | (tmp << (64 - 5)); b0 -= b3;
-    tmp = b1 ^ b2; b1 = (tmp >> 37) | (tmp << (64 - 37)); b2 -= b1;
-    tmp = b1 ^ b0; b1 = (tmp >> 23) | (tmp << (64 - 23)); b0 -= b1;
-    tmp = b3 ^ b2; b3 = (tmp >> 40) | (tmp << (64 - 40)); b2 -= b3;
-    tmp = b3 ^ b0; b3 = (tmp >> 52) | (tmp << (64 - 52)); b0 -= b3;
-    tmp = b1 ^ b2; b1 = (tmp >> 57) | (tmp << (64 - 57)); b2 -= b1;
-    tmp = b1 ^ b0; b1 = (tmp >> 14) | (tmp << (64 - 14)); b0 -= b1 + k0; b1 -= k1 + t1;
-    tmp = b3 ^ b2; b3 = (tmp >> 16) | (tmp << (64 - 16)); b2 -= b3 + k2 + t2; b3 -= k3 + 10;
+	tmp = b3 ^ b0; b3 = (tmp >> 32) | (tmp << (64 - 32)); b0 -= b3;
+	tmp = b1 ^ b2; b1 = (tmp >> 32) | (tmp << (64 - 32)); b2 -= b1;
+	tmp = b1 ^ b0; b1 = (tmp >> 58) | (tmp << (64 - 58)); b0 -= b1;
+	tmp = b3 ^ b2; b3 = (tmp >> 22) | (tmp << (64 - 22)); b2 -= b3;
+	tmp = b3 ^ b0; b3 = (tmp >> 46) | (tmp << (64 - 46)); b0 -= b3;
+	tmp = b1 ^ b2; b1 = (tmp >> 12) | (tmp << (64 - 12)); b2 -= b1;
+	tmp = b1 ^ b0; b1 = (tmp >> 25) | (tmp << (64 - 25)); b0 -= b1 + k1; b1 -= k2 + t2;
+	tmp = b3 ^ b2; b3 = (tmp >> 33) | (tmp << (64 - 33)); b2 -= b3 + k3 + t0; b3 -= k4 + 11;
+	tmp = b3 ^ b0; b3 = (tmp >> 5) | (tmp << (64 - 5)); b0 -= b3;
+	tmp = b1 ^ b2; b1 = (tmp >> 37) | (tmp << (64 - 37)); b2 -= b1;
+	tmp = b1 ^ b0; b1 = (tmp >> 23) | (tmp << (64 - 23)); b0 -= b1;
+	tmp = b3 ^ b2; b3 = (tmp >> 40) | (tmp << (64 - 40)); b2 -= b3;
+	tmp = b3 ^ b0; b3 = (tmp >> 52) | (tmp << (64 - 52)); b0 -= b3;
+	tmp = b1 ^ b2; b1 = (tmp >> 57) | (tmp << (64 - 57)); b2 -= b1;
+	tmp = b1 ^ b0; b1 = (tmp >> 14) | (tmp << (64 - 14)); b0 -= b1 + k0; b1 -= k1 + t1;
+	tmp = b3 ^ b2; b3 = (tmp >> 16) | (tmp << (64 - 16)); b2 -= b3 + k2 + t2; b3 -= k3 + 10;
 
-    tmp = b3 ^ b0; b3 = (tmp >> 32) | (tmp << (64 - 32)); b0 -= b3;
-    tmp = b1 ^ b2; b1 = (tmp >> 32) | (tmp << (64 - 32)); b2 -= b1;
-    tmp = b1 ^ b0; b1 = (tmp >> 58) | (tmp << (64 - 58)); b0 -= b1;
-    tmp = b3 ^ b2; b3 = (tmp >> 22) | (tmp << (64 - 22)); b2 -= b3;
-    tmp = b3 ^ b0; b3 = (tmp >> 46) | (tmp << (64 - 46)); b0 -= b3;
-    tmp = b1 ^ b2; b1 = (tmp >> 12) | (tmp << (64 - 12)); b2 -= b1;
-    tmp = b1 ^ b0; b1 = (tmp >> 25) | (tmp << (64 - 25)); b0 -= b1 + k4; b1 -= k0 + t0;
-    tmp = b3 ^ b2; b3 = (tmp >> 33) | (tmp << (64 - 33)); b2 -= b3 + k1 + t1; b3 -= k2 + 9;
-    tmp = b3 ^ b0; b3 = (tmp >> 5) | (tmp << (64 - 5)); b0 -= b3;
-    tmp = b1 ^ b2; b1 = (tmp >> 37) | (tmp << (64 - 37)); b2 -= b1;
-    tmp = b1 ^ b0; b1 = (tmp >> 23) | (tmp << (64 - 23)); b0 -= b1;
-    tmp = b3 ^ b2; b3 = (tmp >> 40) | (tmp << (64 - 40)); b2 -= b3;
-    tmp = b3 ^ b0; b3 = (tmp >> 52) | (tmp << (64 - 52)); b0 -= b3;
-    tmp = b1 ^ b2; b1 = (tmp >> 57) | (tmp << (64 - 57)); b2 -= b1;
-    tmp = b1 ^ b0; b1 = (tmp >> 14) | (tmp << (64 - 14)); b0 -= b1 + k3; b1 -= k4 + t2;
-    tmp = b3 ^ b2; b3 = (tmp >> 16) | (tmp << (64 - 16)); b2 -= b3 + k0 + t0; b3 -= k1 + 8;
+	tmp = b3 ^ b0; b3 = (tmp >> 32) | (tmp << (64 - 32)); b0 -= b3;
+	tmp = b1 ^ b2; b1 = (tmp >> 32) | (tmp << (64 - 32)); b2 -= b1;
+	tmp = b1 ^ b0; b1 = (tmp >> 58) | (tmp << (64 - 58)); b0 -= b1;
+	tmp = b3 ^ b2; b3 = (tmp >> 22) | (tmp << (64 - 22)); b2 -= b3;
+	tmp = b3 ^ b0; b3 = (tmp >> 46) | (tmp << (64 - 46)); b0 -= b3;
+	tmp = b1 ^ b2; b1 = (tmp >> 12) | (tmp << (64 - 12)); b2 -= b1;
+	tmp = b1 ^ b0; b1 = (tmp >> 25) | (tmp << (64 - 25)); b0 -= b1 + k4; b1 -= k0 + t0;
+	tmp = b3 ^ b2; b3 = (tmp >> 33) | (tmp << (64 - 33)); b2 -= b3 + k1 + t1; b3 -= k2 + 9;
+	tmp = b3 ^ b0; b3 = (tmp >> 5) | (tmp << (64 - 5)); b0 -= b3;
+	tmp = b1 ^ b2; b1 = (tmp >> 37) | (tmp << (64 - 37)); b2 -= b1;
+	tmp = b1 ^ b0; b1 = (tmp >> 23) | (tmp << (64 - 23)); b0 -= b1;
+	tmp = b3 ^ b2; b3 = (tmp >> 40) | (tmp << (64 - 40)); b2 -= b3;
+	tmp = b3 ^ b0; b3 = (tmp >> 52) | (tmp << (64 - 52)); b0 -= b3;
+	tmp = b1 ^ b2; b1 = (tmp >> 57) | (tmp << (64 - 57)); b2 -= b1;
+	tmp = b1 ^ b0; b1 = (tmp >> 14) | (tmp << (64 - 14)); b0 -= b1 + k3; b1 -= k4 + t2;
+	tmp = b3 ^ b2; b3 = (tmp >> 16) | (tmp << (64 - 16)); b2 -= b3 + k0 + t0; b3 -= k1 + 8;
 
-    tmp = b3 ^ b0; b3 = (tmp >> 32) | (tmp << (64 - 32)); b0 -= b3;
-    tmp = b1 ^ b2; b1 = (tmp >> 32) | (tmp << (64 - 32)); b2 -= b1;
-    tmp = b1 ^ b0; b1 = (tmp >> 58) | (tmp << (64 - 58)); b0 -= b1;
-    tmp = b3 ^ b2; b3 = (tmp >> 22) | (tmp << (64 - 22)); b2 -= b3;
-    tmp = b3 ^ b0; b3 = (tmp >> 46) | (tmp << (64 - 46)); b0 -= b3;
-    tmp = b1 ^ b2; b1 = (tmp >> 12) | (tmp << (64 - 12)); b2 -= b1;
-    tmp = b1 ^ b0; b1 = (tmp >> 25) | (tmp << (64 - 25)); b0 -= b1 + k2; b1 -= k3 + t1;
-    tmp = b3 ^ b2; b3 = (tmp >> 33) | (tmp << (64 - 33)); b2 -= b3 + k4 + t2; b3 -= k0 + 7;
-    tmp = b3 ^ b0; b3 = (tmp >> 5) | (tmp << (64 - 5)); b0 -= b3;
-    tmp = b1 ^ b2; b1 = (tmp >> 37) | (tmp << (64 - 37)); b2 -= b1;
-    tmp = b1 ^ b0; b1 = (tmp >> 23) | (tmp << (64 - 23)); b0 -= b1;
-    tmp = b3 ^ b2; b3 = (tmp >> 40) | (tmp << (64 - 40)); b2 -= b3;
-    tmp = b3 ^ b0; b3 = (tmp >> 52) | (tmp << (64 - 52)); b0 -= b3;
-    tmp = b1 ^ b2; b1 = (tmp >> 57) | (tmp << (64 - 57)); b2 -= b1;
-    tmp = b1 ^ b0; b1 = (tmp >> 14) | (tmp << (64 - 14)); b0 -= b1 + k1; b1 -= k2 + t0;
-    tmp = b3 ^ b2; b3 = (tmp >> 16) | (tmp << (64 - 16)); b2 -= b3 + k3 + t1; b3 -= k4 + 6;
+	tmp = b3 ^ b0; b3 = (tmp >> 32) | (tmp << (64 - 32)); b0 -= b3;
+	tmp = b1 ^ b2; b1 = (tmp >> 32) | (tmp << (64 - 32)); b2 -= b1;
+	tmp = b1 ^ b0; b1 = (tmp >> 58) | (tmp << (64 - 58)); b0 -= b1;
+	tmp = b3 ^ b2; b3 = (tmp >> 22) | (tmp << (64 - 22)); b2 -= b3;
+	tmp = b3 ^ b0; b3 = (tmp >> 46) | (tmp << (64 - 46)); b0 -= b3;
+	tmp = b1 ^ b2; b1 = (tmp >> 12) | (tmp << (64 - 12)); b2 -= b1;
+	tmp = b1 ^ b0; b1 = (tmp >> 25) | (tmp << (64 - 25)); b0 -= b1 + k2; b1 -= k3 + t1;
+	tmp = b3 ^ b2; b3 = (tmp >> 33) | (tmp << (64 - 33)); b2 -= b3 + k4 + t2; b3 -= k0 + 7;
+	tmp = b3 ^ b0; b3 = (tmp >> 5) | (tmp << (64 - 5)); b0 -= b3;
+	tmp = b1 ^ b2; b1 = (tmp >> 37) | (tmp << (64 - 37)); b2 -= b1;
+	tmp = b1 ^ b0; b1 = (tmp >> 23) | (tmp << (64 - 23)); b0 -= b1;
+	tmp = b3 ^ b2; b3 = (tmp >> 40) | (tmp << (64 - 40)); b2 -= b3;
+	tmp = b3 ^ b0; b3 = (tmp >> 52) | (tmp << (64 - 52)); b0 -= b3;
+	tmp = b1 ^ b2; b1 = (tmp >> 57) | (tmp << (64 - 57)); b2 -= b1;
+	tmp = b1 ^ b0; b1 = (tmp >> 14) | (tmp << (64 - 14)); b0 -= b1 + k1; b1 -= k2 + t0;
+	tmp = b3 ^ b2; b3 = (tmp >> 16) | (tmp << (64 - 16)); b2 -= b3 + k3 + t1; b3 -= k4 + 6;
 
-    tmp = b3 ^ b0; b3 = (tmp >> 32) | (tmp << (64 - 32)); b0 -= b3;
-    tmp = b1 ^ b2; b1 = (tmp >> 32) | (tmp << (64 - 32)); b2 -= b1;
-    tmp = b1 ^ b0; b1 = (tmp >> 58) | (tmp << (64 - 58)); b0 -= b1;
-    tmp = b3 ^ b2; b3 = (tmp >> 22) | (tmp << (64 - 22)); b2 -= b3;
-    tmp = b3 ^ b0; b3 = (tmp >> 46) | (tmp << (64 - 46)); b0 -= b3;
-    tmp = b1 ^ b2; b1 = (tmp >> 12) | (tmp << (64 - 12)); b2 -= b1;
-    tmp = b1 ^ b0; b1 = (tmp >> 25) | (tmp << (64 - 25)); b0 -= b1 + k0; b1 -= k1 + t2;
-    tmp = b3 ^ b2; b3 = (tmp >> 33) | (tmp << (64 - 33)); b2 -= b3 + k2 + t0; b3 -= k3 + 5;
-    tmp = b3 ^ b0; b3 = (tmp >> 5) | (tmp << (64 - 5)); b0 -= b3;
-    tmp = b1 ^ b2; b1 = (tmp >> 37) | (tmp << (64 - 37)); b2 -= b1;
-    tmp = b1 ^ b0; b1 = (tmp >> 23) | (tmp << (64 - 23)); b0 -= b1;
-    tmp = b3 ^ b2; b3 = (tmp >> 40) | (tmp << (64 - 40)); b2 -= b3;
-    tmp = b3 ^ b0; b3 = (tmp >> 52) | (tmp << (64 - 52)); b0 -= b3;
-    tmp = b1 ^ b2; b1 = (tmp >> 57) | (tmp << (64 - 57)); b2 -= b1;
-    tmp = b1 ^ b0; b1 = (tmp >> 14) | (tmp << (64 - 14)); b0 -= b1 + k4; b1 -= k0 + t1;
-    tmp = b3 ^ b2; b3 = (tmp >> 16) | (tmp << (64 - 16)); b2 -= b3 + k1 + t2; b3 -= k2 + 4;
+	tmp = b3 ^ b0; b3 = (tmp >> 32) | (tmp << (64 - 32)); b0 -= b3;
+	tmp = b1 ^ b2; b1 = (tmp >> 32) | (tmp << (64 - 32)); b2 -= b1;
+	tmp = b1 ^ b0; b1 = (tmp >> 58) | (tmp << (64 - 58)); b0 -= b1;
+	tmp = b3 ^ b2; b3 = (tmp >> 22) | (tmp << (64 - 22)); b2 -= b3;
+	tmp = b3 ^ b0; b3 = (tmp >> 46) | (tmp << (64 - 46)); b0 -= b3;
+	tmp = b1 ^ b2; b1 = (tmp >> 12) | (tmp << (64 - 12)); b2 -= b1;
+	tmp = b1 ^ b0; b1 = (tmp >> 25) | (tmp << (64 - 25)); b0 -= b1 + k0; b1 -= k1 + t2;
+	tmp = b3 ^ b2; b3 = (tmp >> 33) | (tmp << (64 - 33)); b2 -= b3 + k2 + t0; b3 -= k3 + 5;
+	tmp = b3 ^ b0; b3 = (tmp >> 5) | (tmp << (64 - 5)); b0 -= b3;
+	tmp = b1 ^ b2; b1 = (tmp >> 37) | (tmp << (64 - 37)); b2 -= b1;
+	tmp = b1 ^ b0; b1 = (tmp >> 23) | (tmp << (64 - 23)); b0 -= b1;
+	tmp = b3 ^ b2; b3 = (tmp >> 40) | (tmp << (64 - 40)); b2 -= b3;
+	tmp = b3 ^ b0; b3 = (tmp >> 52) | (tmp << (64 - 52)); b0 -= b3;
+	tmp = b1 ^ b2; b1 = (tmp >> 57) | (tmp << (64 - 57)); b2 -= b1;
+	tmp = b1 ^ b0; b1 = (tmp >> 14) | (tmp << (64 - 14)); b0 -= b1 + k4; b1 -= k0 + t1;
+	tmp = b3 ^ b2; b3 = (tmp >> 16) | (tmp << (64 - 16)); b2 -= b3 + k1 + t2; b3 -= k2 + 4;
 
-    tmp = b3 ^ b0; b3 = (tmp >> 32) | (tmp << (64 - 32)); b0 -= b3;
-    tmp = b1 ^ b2; b1 = (tmp >> 32) | (tmp << (64 - 32)); b2 -= b1;
-    tmp = b1 ^ b0; b1 = (tmp >> 58) | (tmp << (64 - 58)); b0 -= b1;
-    tmp = b3 ^ b2; b3 = (tmp >> 22) | (tmp << (64 - 22)); b2 -= b3;
-    tmp = b3 ^ b0; b3 = (tmp >> 46) | (tmp << (64 - 46)); b0 -= b3;
-    tmp = b1 ^ b2; b1 = (tmp >> 12) | (tmp << (64 - 12)); b2 -= b1;
-    tmp = b1 ^ b0; b1 = (tmp >> 25) | (tmp << (64 - 25)); b0 -= b1 + k3; b1 -= k4 + t0;
-    tmp = b3 ^ b2; b3 = (tmp >> 33) | (tmp << (64 - 33)); b2 -= b3 + k0 + t1; b3 -= k1 + 3;
-    tmp = b3 ^ b0; b3 = (tmp >> 5) | (tmp << (64 - 5)); b0 -= b3;
-    tmp = b1 ^ b2; b1 = (tmp >> 37) | (tmp << (64 - 37)); b2 -= b1;
-    tmp = b1 ^ b0; b1 = (tmp >> 23) | (tmp << (64 - 23)); b0 -= b1;
-    tmp = b3 ^ b2; b3 = (tmp >> 40) | (tmp << (64 - 40)); b2 -= b3;
-    tmp = b3 ^ b0; b3 = (tmp >> 52) | (tmp << (64 - 52)); b0 -= b3;
-    tmp = b1 ^ b2; b1 = (tmp >> 57) | (tmp << (64 - 57)); b2 -= b1;
-    tmp = b1 ^ b0; b1 = (tmp >> 14) | (tmp << (64 - 14)); b0 -= b1 + k2; b1 -= k3 + t2;
-    tmp = b3 ^ b2; b3 = (tmp >> 16) | (tmp << (64 - 16)); b2 -= b3 + k4 + t0; b3 -= k0 + 2;
+	tmp = b3 ^ b0; b3 = (tmp >> 32) | (tmp << (64 - 32)); b0 -= b3;
+	tmp = b1 ^ b2; b1 = (tmp >> 32) | (tmp << (64 - 32)); b2 -= b1;
+	tmp = b1 ^ b0; b1 = (tmp >> 58) | (tmp << (64 - 58)); b0 -= b1;
+	tmp = b3 ^ b2; b3 = (tmp >> 22) | (tmp << (64 - 22)); b2 -= b3;
+	tmp = b3 ^ b0; b3 = (tmp >> 46) | (tmp << (64 - 46)); b0 -= b3;
+	tmp = b1 ^ b2; b1 = (tmp >> 12) | (tmp << (64 - 12)); b2 -= b1;
+	tmp = b1 ^ b0; b1 = (tmp >> 25) | (tmp << (64 - 25)); b0 -= b1 + k3; b1 -= k4 + t0;
+	tmp = b3 ^ b2; b3 = (tmp >> 33) | (tmp << (64 - 33)); b2 -= b3 + k0 + t1; b3 -= k1 + 3;
+	tmp = b3 ^ b0; b3 = (tmp >> 5) | (tmp << (64 - 5)); b0 -= b3;
+	tmp = b1 ^ b2; b1 = (tmp >> 37) | (tmp << (64 - 37)); b2 -= b1;
+	tmp = b1 ^ b0; b1 = (tmp >> 23) | (tmp << (64 - 23)); b0 -= b1;
+	tmp = b3 ^ b2; b3 = (tmp >> 40) | (tmp << (64 - 40)); b2 -= b3;
+	tmp = b3 ^ b0; b3 = (tmp >> 52) | (tmp << (64 - 52)); b0 -= b3;
+	tmp = b1 ^ b2; b1 = (tmp >> 57) | (tmp << (64 - 57)); b2 -= b1;
+	tmp = b1 ^ b0; b1 = (tmp >> 14) | (tmp << (64 - 14)); b0 -= b1 + k2; b1 -= k3 + t2;
+	tmp = b3 ^ b2; b3 = (tmp >> 16) | (tmp << (64 - 16)); b2 -= b3 + k4 + t0; b3 -= k0 + 2;
 
-    tmp = b3 ^ b0; b3 = (tmp >> 32) | (tmp << (64 - 32)); b0 -= b3;
-    tmp = b1 ^ b2; b1 = (tmp >> 32) | (tmp << (64 - 32)); b2 -= b1;
-    tmp = b1 ^ b0; b1 = (tmp >> 58) | (tmp << (64 - 58)); b0 -= b1;
-    tmp = b3 ^ b2; b3 = (tmp >> 22) | (tmp << (64 - 22)); b2 -= b3;
-    tmp = b3 ^ b0; b3 = (tmp >> 46) | (tmp << (64 - 46)); b0 -= b3;
-    tmp = b1 ^ b2; b1 = (tmp >> 12) | (tmp << (64 - 12)); b2 -= b1;
-    tmp = b1 ^ b0; b1 = (tmp >> 25) | (tmp << (64 - 25)); b0 -= b1 + k1; b1 -= k2 + t1;
-    tmp = b3 ^ b2; b3 = (tmp >> 33) | (tmp << (64 - 33)); b2 -= b3 + k3 + t2; b3 -= k4 + 1;
-    tmp = b3 ^ b0; b3 = (tmp >> 5) | (tmp << (64 - 5)); b0 -= b3;
-    tmp = b1 ^ b2; b1 = (tmp >> 37) | (tmp << (64 - 37)); b2 -= b1;
-    tmp = b1 ^ b0; b1 = (tmp >> 23) | (tmp << (64 - 23)); b0 -= b1;
-    tmp = b3 ^ b2; b3 = (tmp >> 40) | (tmp << (64 - 40)); b2 -= b3;
-    tmp = b3 ^ b0; b3 = (tmp >> 52) | (tmp << (64 - 52)); b0 -= b3;
-    tmp = b1 ^ b2; b1 = (tmp >> 57) | (tmp << (64 - 57)); b2 -= b1;
-    tmp = b1 ^ b0; b1 = (tmp >> 14) | (tmp << (64 - 14)); b0 -= b1 + k0; b1 -= k1 + t0;
-    tmp = b3 ^ b2; b3 = (tmp >> 16) | (tmp << (64 - 16)); b2 -= b3 + k2 + t1; b3 -= k3;
+	tmp = b3 ^ b0; b3 = (tmp >> 32) | (tmp << (64 - 32)); b0 -= b3;
+	tmp = b1 ^ b2; b1 = (tmp >> 32) | (tmp << (64 - 32)); b2 -= b1;
+	tmp = b1 ^ b0; b1 = (tmp >> 58) | (tmp << (64 - 58)); b0 -= b1;
+	tmp = b3 ^ b2; b3 = (tmp >> 22) | (tmp << (64 - 22)); b2 -= b3;
+	tmp = b3 ^ b0; b3 = (tmp >> 46) | (tmp << (64 - 46)); b0 -= b3;
+	tmp = b1 ^ b2; b1 = (tmp >> 12) | (tmp << (64 - 12)); b2 -= b1;
+	tmp = b1 ^ b0; b1 = (tmp >> 25) | (tmp << (64 - 25)); b0 -= b1 + k1; b1 -= k2 + t1;
+	tmp = b3 ^ b2; b3 = (tmp >> 33) | (tmp << (64 - 33)); b2 -= b3 + k3 + t2; b3 -= k4 + 1;
+	tmp = b3 ^ b0; b3 = (tmp >> 5) | (tmp << (64 - 5)); b0 -= b3;
+	tmp = b1 ^ b2; b1 = (tmp >> 37) | (tmp << (64 - 37)); b2 -= b1;
+	tmp = b1 ^ b0; b1 = (tmp >> 23) | (tmp << (64 - 23)); b0 -= b1;
+	tmp = b3 ^ b2; b3 = (tmp >> 40) | (tmp << (64 - 40)); b2 -= b3;
+	tmp = b3 ^ b0; b3 = (tmp >> 52) | (tmp << (64 - 52)); b0 -= b3;
+	tmp = b1 ^ b2; b1 = (tmp >> 57) | (tmp << (64 - 57)); b2 -= b1;
+	tmp = b1 ^ b0; b1 = (tmp >> 14) | (tmp << (64 - 14)); b0 -= b1 + k0; b1 -= k1 + t0;
+	tmp = b3 ^ b2; b3 = (tmp >> 16) | (tmp << (64 - 16)); b2 -= b3 + k2 + t1; b3 -= k3;
 
-    output[0] = b0;
-    output[1] = b1;
-    output[2] = b2;
-    output[3] = b3;
-  }
+	output[0] = b0;
+	output[1] = b1;
+	output[2] = b2;
+	output[3] = b3;
+}
diff --git a/drivers/staging/skein/threefish512Block.c b/drivers/staging/skein/threefish512Block.c
index 5262f5a8f21b..dc96ba279720 100644
--- a/drivers/staging/skein/threefish512Block.c
+++ b/drivers/staging/skein/threefish512Block.c
@@ -3,640 +3,638 @@
 
 
 void threefishEncrypt512(struct threefish_key *keyCtx, u64 *input, u64 *output)
-    {
+{
+	u64 b0 = input[0], b1 = input[1],
+	  b2 = input[2], b3 = input[3],
+	  b4 = input[4], b5 = input[5],
+	  b6 = input[6], b7 = input[7];
+	u64 k0 = keyCtx->key[0], k1 = keyCtx->key[1],
+	  k2 = keyCtx->key[2], k3 = keyCtx->key[3],
+	  k4 = keyCtx->key[4], k5 = keyCtx->key[5],
+	  k6 = keyCtx->key[6], k7 = keyCtx->key[7],
+	  k8 = keyCtx->key[8];
+	u64 t0 = keyCtx->tweak[0], t1 = keyCtx->tweak[1],
+	  t2 = keyCtx->tweak[2];
 
-    u64 b0 = input[0], b1 = input[1],
-      b2 = input[2], b3 = input[3],
-      b4 = input[4], b5 = input[5],
-      b6 = input[6], b7 = input[7];
-    u64 k0 = keyCtx->key[0], k1 = keyCtx->key[1],
-      k2 = keyCtx->key[2], k3 = keyCtx->key[3],
-      k4 = keyCtx->key[4], k5 = keyCtx->key[5],
-      k6 = keyCtx->key[6], k7 = keyCtx->key[7],
-      k8 = keyCtx->key[8];
-    u64 t0 = keyCtx->tweak[0], t1 = keyCtx->tweak[1],
-      t2 = keyCtx->tweak[2];
+	b1 += k1; b0 += b1 + k0; b1 = ((b1 << 46) | (b1 >> (64 - 46))) ^ b0;
+	b3 += k3; b2 += b3 + k2; b3 = ((b3 << 36) | (b3 >> (64 - 36))) ^ b2;
+	b5 += k5 + t0; b4 += b5 + k4; b5 = ((b5 << 19) | (b5 >> (64 - 19))) ^ b4;
+	b7 += k7; b6 += b7 + k6 + t1; b7 = ((b7 << 37) | (b7 >> (64 - 37))) ^ b6;
+	b2 += b1; b1 = ((b1 << 33) | (b1 >> (64 - 33))) ^ b2;
+	b4 += b7; b7 = ((b7 << 27) | (b7 >> (64 - 27))) ^ b4;
+	b6 += b5; b5 = ((b5 << 14) | (b5 >> (64 - 14))) ^ b6;
+	b0 += b3; b3 = ((b3 << 42) | (b3 >> (64 - 42))) ^ b0;
+	b4 += b1; b1 = ((b1 << 17) | (b1 >> (64 - 17))) ^ b4;
+	b6 += b3; b3 = ((b3 << 49) | (b3 >> (64 - 49))) ^ b6;
+	b0 += b5; b5 = ((b5 << 36) | (b5 >> (64 - 36))) ^ b0;
+	b2 += b7; b7 = ((b7 << 39) | (b7 >> (64 - 39))) ^ b2;
+	b6 += b1; b1 = ((b1 << 44) | (b1 >> (64 - 44))) ^ b6;
+	b0 += b7; b7 = ((b7 << 9) | (b7 >> (64 - 9))) ^ b0;
+	b2 += b5; b5 = ((b5 << 54) | (b5 >> (64 - 54))) ^ b2;
+	b4 += b3; b3 = ((b3 << 56) | (b3 >> (64 - 56))) ^ b4;
+	b1 += k2; b0 += b1 + k1; b1 = ((b1 << 39) | (b1 >> (64 - 39))) ^ b0;
+	b3 += k4; b2 += b3 + k3; b3 = ((b3 << 30) | (b3 >> (64 - 30))) ^ b2;
+	b5 += k6 + t1; b4 += b5 + k5; b5 = ((b5 << 34) | (b5 >> (64 - 34))) ^ b4;
+	b7 += k8 + 1; b6 += b7 + k7 + t2; b7 = ((b7 << 24) | (b7 >> (64 - 24))) ^ b6;
+	b2 += b1; b1 = ((b1 << 13) | (b1 >> (64 - 13))) ^ b2;
+	b4 += b7; b7 = ((b7 << 50) | (b7 >> (64 - 50))) ^ b4;
+	b6 += b5; b5 = ((b5 << 10) | (b5 >> (64 - 10))) ^ b6;
+	b0 += b3; b3 = ((b3 << 17) | (b3 >> (64 - 17))) ^ b0;
+	b4 += b1; b1 = ((b1 << 25) | (b1 >> (64 - 25))) ^ b4;
+	b6 += b3; b3 = ((b3 << 29) | (b3 >> (64 - 29))) ^ b6;
+	b0 += b5; b5 = ((b5 << 39) | (b5 >> (64 - 39))) ^ b0;
+	b2 += b7; b7 = ((b7 << 43) | (b7 >> (64 - 43))) ^ b2;
+	b6 += b1; b1 = ((b1 << 8) | (b1 >> (64 - 8))) ^ b6;
+	b0 += b7; b7 = ((b7 << 35) | (b7 >> (64 - 35))) ^ b0;
+	b2 += b5; b5 = ((b5 << 56) | (b5 >> (64 - 56))) ^ b2;
+	b4 += b3; b3 = ((b3 << 22) | (b3 >> (64 - 22))) ^ b4;
+	b1 += k3; b0 += b1 + k2; b1 = ((b1 << 46) | (b1 >> (64 - 46))) ^ b0;
+	b3 += k5; b2 += b3 + k4; b3 = ((b3 << 36) | (b3 >> (64 - 36))) ^ b2;
+	b5 += k7 + t2; b4 += b5 + k6; b5 = ((b5 << 19) | (b5 >> (64 - 19))) ^ b4;
+	b7 += k0 + 2; b6 += b7 + k8 + t0; b7 = ((b7 << 37) | (b7 >> (64 - 37))) ^ b6;
+	b2 += b1; b1 = ((b1 << 33) | (b1 >> (64 - 33))) ^ b2;
+	b4 += b7; b7 = ((b7 << 27) | (b7 >> (64 - 27))) ^ b4;
+	b6 += b5; b5 = ((b5 << 14) | (b5 >> (64 - 14))) ^ b6;
+	b0 += b3; b3 = ((b3 << 42) | (b3 >> (64 - 42))) ^ b0;
+	b4 += b1; b1 = ((b1 << 17) | (b1 >> (64 - 17))) ^ b4;
+	b6 += b3; b3 = ((b3 << 49) | (b3 >> (64 - 49))) ^ b6;
+	b0 += b5; b5 = ((b5 << 36) | (b5 >> (64 - 36))) ^ b0;
+	b2 += b7; b7 = ((b7 << 39) | (b7 >> (64 - 39))) ^ b2;
+	b6 += b1; b1 = ((b1 << 44) | (b1 >> (64 - 44))) ^ b6;
+	b0 += b7; b7 = ((b7 << 9) | (b7 >> (64 - 9))) ^ b0;
+	b2 += b5; b5 = ((b5 << 54) | (b5 >> (64 - 54))) ^ b2;
+	b4 += b3; b3 = ((b3 << 56) | (b3 >> (64 - 56))) ^ b4;
+	b1 += k4; b0 += b1 + k3; b1 = ((b1 << 39) | (b1 >> (64 - 39))) ^ b0;
+	b3 += k6; b2 += b3 + k5; b3 = ((b3 << 30) | (b3 >> (64 - 30))) ^ b2;
+	b5 += k8 + t0; b4 += b5 + k7; b5 = ((b5 << 34) | (b5 >> (64 - 34))) ^ b4;
+	b7 += k1 + 3; b6 += b7 + k0 + t1; b7 = ((b7 << 24) | (b7 >> (64 - 24))) ^ b6;
+	b2 += b1; b1 = ((b1 << 13) | (b1 >> (64 - 13))) ^ b2;
+	b4 += b7; b7 = ((b7 << 50) | (b7 >> (64 - 50))) ^ b4;
+	b6 += b5; b5 = ((b5 << 10) | (b5 >> (64 - 10))) ^ b6;
+	b0 += b3; b3 = ((b3 << 17) | (b3 >> (64 - 17))) ^ b0;
+	b4 += b1; b1 = ((b1 << 25) | (b1 >> (64 - 25))) ^ b4;
+	b6 += b3; b3 = ((b3 << 29) | (b3 >> (64 - 29))) ^ b6;
+	b0 += b5; b5 = ((b5 << 39) | (b5 >> (64 - 39))) ^ b0;
+	b2 += b7; b7 = ((b7 << 43) | (b7 >> (64 - 43))) ^ b2;
+	b6 += b1; b1 = ((b1 << 8) | (b1 >> (64 - 8))) ^ b6;
+	b0 += b7; b7 = ((b7 << 35) | (b7 >> (64 - 35))) ^ b0;
+	b2 += b5; b5 = ((b5 << 56) | (b5 >> (64 - 56))) ^ b2;
+	b4 += b3; b3 = ((b3 << 22) | (b3 >> (64 - 22))) ^ b4;
+	b1 += k5; b0 += b1 + k4; b1 = ((b1 << 46) | (b1 >> (64 - 46))) ^ b0;
+	b3 += k7; b2 += b3 + k6; b3 = ((b3 << 36) | (b3 >> (64 - 36))) ^ b2;
+	b5 += k0 + t1; b4 += b5 + k8; b5 = ((b5 << 19) | (b5 >> (64 - 19))) ^ b4;
+	b7 += k2 + 4; b6 += b7 + k1 + t2; b7 = ((b7 << 37) | (b7 >> (64 - 37))) ^ b6;
+	b2 += b1; b1 = ((b1 << 33) | (b1 >> (64 - 33))) ^ b2;
+	b4 += b7; b7 = ((b7 << 27) | (b7 >> (64 - 27))) ^ b4;
+	b6 += b5; b5 = ((b5 << 14) | (b5 >> (64 - 14))) ^ b6;
+	b0 += b3; b3 = ((b3 << 42) | (b3 >> (64 - 42))) ^ b0;
+	b4 += b1; b1 = ((b1 << 17) | (b1 >> (64 - 17))) ^ b4;
+	b6 += b3; b3 = ((b3 << 49) | (b3 >> (64 - 49))) ^ b6;
+	b0 += b5; b5 = ((b5 << 36) | (b5 >> (64 - 36))) ^ b0;
+	b2 += b7; b7 = ((b7 << 39) | (b7 >> (64 - 39))) ^ b2;
+	b6 += b1; b1 = ((b1 << 44) | (b1 >> (64 - 44))) ^ b6;
+	b0 += b7; b7 = ((b7 << 9) | (b7 >> (64 - 9))) ^ b0;
+	b2 += b5; b5 = ((b5 << 54) | (b5 >> (64 - 54))) ^ b2;
+	b4 += b3; b3 = ((b3 << 56) | (b3 >> (64 - 56))) ^ b4;
+	b1 += k6; b0 += b1 + k5; b1 = ((b1 << 39) | (b1 >> (64 - 39))) ^ b0;
+	b3 += k8; b2 += b3 + k7; b3 = ((b3 << 30) | (b3 >> (64 - 30))) ^ b2;
+	b5 += k1 + t2; b4 += b5 + k0; b5 = ((b5 << 34) | (b5 >> (64 - 34))) ^ b4;
+	b7 += k3 + 5; b6 += b7 + k2 + t0; b7 = ((b7 << 24) | (b7 >> (64 - 24))) ^ b6;
+	b2 += b1; b1 = ((b1 << 13) | (b1 >> (64 - 13))) ^ b2;
+	b4 += b7; b7 = ((b7 << 50) | (b7 >> (64 - 50))) ^ b4;
+	b6 += b5; b5 = ((b5 << 10) | (b5 >> (64 - 10))) ^ b6;
+	b0 += b3; b3 = ((b3 << 17) | (b3 >> (64 - 17))) ^ b0;
+	b4 += b1; b1 = ((b1 << 25) | (b1 >> (64 - 25))) ^ b4;
+	b6 += b3; b3 = ((b3 << 29) | (b3 >> (64 - 29))) ^ b6;
+	b0 += b5; b5 = ((b5 << 39) | (b5 >> (64 - 39))) ^ b0;
+	b2 += b7; b7 = ((b7 << 43) | (b7 >> (64 - 43))) ^ b2;
+	b6 += b1; b1 = ((b1 << 8) | (b1 >> (64 - 8))) ^ b6;
+	b0 += b7; b7 = ((b7 << 35) | (b7 >> (64 - 35))) ^ b0;
+	b2 += b5; b5 = ((b5 << 56) | (b5 >> (64 - 56))) ^ b2;
+	b4 += b3; b3 = ((b3 << 22) | (b3 >> (64 - 22))) ^ b4;
+	b1 += k7; b0 += b1 + k6; b1 = ((b1 << 46) | (b1 >> (64 - 46))) ^ b0;
+	b3 += k0; b2 += b3 + k8; b3 = ((b3 << 36) | (b3 >> (64 - 36))) ^ b2;
+	b5 += k2 + t0; b4 += b5 + k1; b5 = ((b5 << 19) | (b5 >> (64 - 19))) ^ b4;
+	b7 += k4 + 6; b6 += b7 + k3 + t1; b7 = ((b7 << 37) | (b7 >> (64 - 37))) ^ b6;
+	b2 += b1; b1 = ((b1 << 33) | (b1 >> (64 - 33))) ^ b2;
+	b4 += b7; b7 = ((b7 << 27) | (b7 >> (64 - 27))) ^ b4;
+	b6 += b5; b5 = ((b5 << 14) | (b5 >> (64 - 14))) ^ b6;
+	b0 += b3; b3 = ((b3 << 42) | (b3 >> (64 - 42))) ^ b0;
+	b4 += b1; b1 = ((b1 << 17) | (b1 >> (64 - 17))) ^ b4;
+	b6 += b3; b3 = ((b3 << 49) | (b3 >> (64 - 49))) ^ b6;
+	b0 += b5; b5 = ((b5 << 36) | (b5 >> (64 - 36))) ^ b0;
+	b2 += b7; b7 = ((b7 << 39) | (b7 >> (64 - 39))) ^ b2;
+	b6 += b1; b1 = ((b1 << 44) | (b1 >> (64 - 44))) ^ b6;
+	b0 += b7; b7 = ((b7 << 9) | (b7 >> (64 - 9))) ^ b0;
+	b2 += b5; b5 = ((b5 << 54) | (b5 >> (64 - 54))) ^ b2;
+	b4 += b3; b3 = ((b3 << 56) | (b3 >> (64 - 56))) ^ b4;
+	b1 += k8; b0 += b1 + k7; b1 = ((b1 << 39) | (b1 >> (64 - 39))) ^ b0;
+	b3 += k1; b2 += b3 + k0; b3 = ((b3 << 30) | (b3 >> (64 - 30))) ^ b2;
+	b5 += k3 + t1; b4 += b5 + k2; b5 = ((b5 << 34) | (b5 >> (64 - 34))) ^ b4;
+	b7 += k5 + 7; b6 += b7 + k4 + t2; b7 = ((b7 << 24) | (b7 >> (64 - 24))) ^ b6;
+	b2 += b1; b1 = ((b1 << 13) | (b1 >> (64 - 13))) ^ b2;
+	b4 += b7; b7 = ((b7 << 50) | (b7 >> (64 - 50))) ^ b4;
+	b6 += b5; b5 = ((b5 << 10) | (b5 >> (64 - 10))) ^ b6;
+	b0 += b3; b3 = ((b3 << 17) | (b3 >> (64 - 17))) ^ b0;
+	b4 += b1; b1 = ((b1 << 25) | (b1 >> (64 - 25))) ^ b4;
+	b6 += b3; b3 = ((b3 << 29) | (b3 >> (64 - 29))) ^ b6;
+	b0 += b5; b5 = ((b5 << 39) | (b5 >> (64 - 39))) ^ b0;
+	b2 += b7; b7 = ((b7 << 43) | (b7 >> (64 - 43))) ^ b2;
+	b6 += b1; b1 = ((b1 << 8) | (b1 >> (64 - 8))) ^ b6;
+	b0 += b7; b7 = ((b7 << 35) | (b7 >> (64 - 35))) ^ b0;
+	b2 += b5; b5 = ((b5 << 56) | (b5 >> (64 - 56))) ^ b2;
+	b4 += b3; b3 = ((b3 << 22) | (b3 >> (64 - 22))) ^ b4;
+	b1 += k0; b0 += b1 + k8; b1 = ((b1 << 46) | (b1 >> (64 - 46))) ^ b0;
+	b3 += k2; b2 += b3 + k1; b3 = ((b3 << 36) | (b3 >> (64 - 36))) ^ b2;
+	b5 += k4 + t2; b4 += b5 + k3; b5 = ((b5 << 19) | (b5 >> (64 - 19))) ^ b4;
+	b7 += k6 + 8; b6 += b7 + k5 + t0; b7 = ((b7 << 37) | (b7 >> (64 - 37))) ^ b6;
+	b2 += b1; b1 = ((b1 << 33) | (b1 >> (64 - 33))) ^ b2;
+	b4 += b7; b7 = ((b7 << 27) | (b7 >> (64 - 27))) ^ b4;
+	b6 += b5; b5 = ((b5 << 14) | (b5 >> (64 - 14))) ^ b6;
+	b0 += b3; b3 = ((b3 << 42) | (b3 >> (64 - 42))) ^ b0;
+	b4 += b1; b1 = ((b1 << 17) | (b1 >> (64 - 17))) ^ b4;
+	b6 += b3; b3 = ((b3 << 49) | (b3 >> (64 - 49))) ^ b6;
+	b0 += b5; b5 = ((b5 << 36) | (b5 >> (64 - 36))) ^ b0;
+	b2 += b7; b7 = ((b7 << 39) | (b7 >> (64 - 39))) ^ b2;
+	b6 += b1; b1 = ((b1 << 44) | (b1 >> (64 - 44))) ^ b6;
+	b0 += b7; b7 = ((b7 << 9) | (b7 >> (64 - 9))) ^ b0;
+	b2 += b5; b5 = ((b5 << 54) | (b5 >> (64 - 54))) ^ b2;
+	b4 += b3; b3 = ((b3 << 56) | (b3 >> (64 - 56))) ^ b4;
+	b1 += k1; b0 += b1 + k0; b1 = ((b1 << 39) | (b1 >> (64 - 39))) ^ b0;
+	b3 += k3; b2 += b3 + k2; b3 = ((b3 << 30) | (b3 >> (64 - 30))) ^ b2;
+	b5 += k5 + t0; b4 += b5 + k4; b5 = ((b5 << 34) | (b5 >> (64 - 34))) ^ b4;
+	b7 += k7 + 9; b6 += b7 + k6 + t1; b7 = ((b7 << 24) | (b7 >> (64 - 24))) ^ b6;
+	b2 += b1; b1 = ((b1 << 13) | (b1 >> (64 - 13))) ^ b2;
+	b4 += b7; b7 = ((b7 << 50) | (b7 >> (64 - 50))) ^ b4;
+	b6 += b5; b5 = ((b5 << 10) | (b5 >> (64 - 10))) ^ b6;
+	b0 += b3; b3 = ((b3 << 17) | (b3 >> (64 - 17))) ^ b0;
+	b4 += b1; b1 = ((b1 << 25) | (b1 >> (64 - 25))) ^ b4;
+	b6 += b3; b3 = ((b3 << 29) | (b3 >> (64 - 29))) ^ b6;
+	b0 += b5; b5 = ((b5 << 39) | (b5 >> (64 - 39))) ^ b0;
+	b2 += b7; b7 = ((b7 << 43) | (b7 >> (64 - 43))) ^ b2;
+	b6 += b1; b1 = ((b1 << 8) | (b1 >> (64 - 8))) ^ b6;
+	b0 += b7; b7 = ((b7 << 35) | (b7 >> (64 - 35))) ^ b0;
+	b2 += b5; b5 = ((b5 << 56) | (b5 >> (64 - 56))) ^ b2;
+	b4 += b3; b3 = ((b3 << 22) | (b3 >> (64 - 22))) ^ b4;
+	b1 += k2; b0 += b1 + k1; b1 = ((b1 << 46) | (b1 >> (64 - 46))) ^ b0;
+	b3 += k4; b2 += b3 + k3; b3 = ((b3 << 36) | (b3 >> (64 - 36))) ^ b2;
+	b5 += k6 + t1; b4 += b5 + k5; b5 = ((b5 << 19) | (b5 >> (64 - 19))) ^ b4;
+	b7 += k8 + 10; b6 += b7 + k7 + t2; b7 = ((b7 << 37) | (b7 >> (64 - 37))) ^ b6;
+	b2 += b1; b1 = ((b1 << 33) | (b1 >> (64 - 33))) ^ b2;
+	b4 += b7; b7 = ((b7 << 27) | (b7 >> (64 - 27))) ^ b4;
+	b6 += b5; b5 = ((b5 << 14) | (b5 >> (64 - 14))) ^ b6;
+	b0 += b3; b3 = ((b3 << 42) | (b3 >> (64 - 42))) ^ b0;
+	b4 += b1; b1 = ((b1 << 17) | (b1 >> (64 - 17))) ^ b4;
+	b6 += b3; b3 = ((b3 << 49) | (b3 >> (64 - 49))) ^ b6;
+	b0 += b5; b5 = ((b5 << 36) | (b5 >> (64 - 36))) ^ b0;
+	b2 += b7; b7 = ((b7 << 39) | (b7 >> (64 - 39))) ^ b2;
+	b6 += b1; b1 = ((b1 << 44) | (b1 >> (64 - 44))) ^ b6;
+	b0 += b7; b7 = ((b7 << 9) | (b7 >> (64 - 9))) ^ b0;
+	b2 += b5; b5 = ((b5 << 54) | (b5 >> (64 - 54))) ^ b2;
+	b4 += b3; b3 = ((b3 << 56) | (b3 >> (64 - 56))) ^ b4;
+	b1 += k3; b0 += b1 + k2; b1 = ((b1 << 39) | (b1 >> (64 - 39))) ^ b0;
+	b3 += k5; b2 += b3 + k4; b3 = ((b3 << 30) | (b3 >> (64 - 30))) ^ b2;
+	b5 += k7 + t2; b4 += b5 + k6; b5 = ((b5 << 34) | (b5 >> (64 - 34))) ^ b4;
+	b7 += k0 + 11; b6 += b7 + k8 + t0; b7 = ((b7 << 24) | (b7 >> (64 - 24))) ^ b6;
+	b2 += b1; b1 = ((b1 << 13) | (b1 >> (64 - 13))) ^ b2;
+	b4 += b7; b7 = ((b7 << 50) | (b7 >> (64 - 50))) ^ b4;
+	b6 += b5; b5 = ((b5 << 10) | (b5 >> (64 - 10))) ^ b6;
+	b0 += b3; b3 = ((b3 << 17) | (b3 >> (64 - 17))) ^ b0;
+	b4 += b1; b1 = ((b1 << 25) | (b1 >> (64 - 25))) ^ b4;
+	b6 += b3; b3 = ((b3 << 29) | (b3 >> (64 - 29))) ^ b6;
+	b0 += b5; b5 = ((b5 << 39) | (b5 >> (64 - 39))) ^ b0;
+	b2 += b7; b7 = ((b7 << 43) | (b7 >> (64 - 43))) ^ b2;
+	b6 += b1; b1 = ((b1 << 8) | (b1 >> (64 - 8))) ^ b6;
+	b0 += b7; b7 = ((b7 << 35) | (b7 >> (64 - 35))) ^ b0;
+	b2 += b5; b5 = ((b5 << 56) | (b5 >> (64 - 56))) ^ b2;
+	b4 += b3; b3 = ((b3 << 22) | (b3 >> (64 - 22))) ^ b4;
+	b1 += k4; b0 += b1 + k3; b1 = ((b1 << 46) | (b1 >> (64 - 46))) ^ b0;
+	b3 += k6; b2 += b3 + k5; b3 = ((b3 << 36) | (b3 >> (64 - 36))) ^ b2;
+	b5 += k8 + t0; b4 += b5 + k7; b5 = ((b5 << 19) | (b5 >> (64 - 19))) ^ b4;
+	b7 += k1 + 12; b6 += b7 + k0 + t1; b7 = ((b7 << 37) | (b7 >> (64 - 37))) ^ b6;
+	b2 += b1; b1 = ((b1 << 33) | (b1 >> (64 - 33))) ^ b2;
+	b4 += b7; b7 = ((b7 << 27) | (b7 >> (64 - 27))) ^ b4;
+	b6 += b5; b5 = ((b5 << 14) | (b5 >> (64 - 14))) ^ b6;
+	b0 += b3; b3 = ((b3 << 42) | (b3 >> (64 - 42))) ^ b0;
+	b4 += b1; b1 = ((b1 << 17) | (b1 >> (64 - 17))) ^ b4;
+	b6 += b3; b3 = ((b3 << 49) | (b3 >> (64 - 49))) ^ b6;
+	b0 += b5; b5 = ((b5 << 36) | (b5 >> (64 - 36))) ^ b0;
+	b2 += b7; b7 = ((b7 << 39) | (b7 >> (64 - 39))) ^ b2;
+	b6 += b1; b1 = ((b1 << 44) | (b1 >> (64 - 44))) ^ b6;
+	b0 += b7; b7 = ((b7 << 9) | (b7 >> (64 - 9))) ^ b0;
+	b2 += b5; b5 = ((b5 << 54) | (b5 >> (64 - 54))) ^ b2;
+	b4 += b3; b3 = ((b3 << 56) | (b3 >> (64 - 56))) ^ b4;
+	b1 += k5; b0 += b1 + k4; b1 = ((b1 << 39) | (b1 >> (64 - 39))) ^ b0;
+	b3 += k7; b2 += b3 + k6; b3 = ((b3 << 30) | (b3 >> (64 - 30))) ^ b2;
+	b5 += k0 + t1; b4 += b5 + k8; b5 = ((b5 << 34) | (b5 >> (64 - 34))) ^ b4;
+	b7 += k2 + 13; b6 += b7 + k1 + t2; b7 = ((b7 << 24) | (b7 >> (64 - 24))) ^ b6;
+	b2 += b1; b1 = ((b1 << 13) | (b1 >> (64 - 13))) ^ b2;
+	b4 += b7; b7 = ((b7 << 50) | (b7 >> (64 - 50))) ^ b4;
+	b6 += b5; b5 = ((b5 << 10) | (b5 >> (64 - 10))) ^ b6;
+	b0 += b3; b3 = ((b3 << 17) | (b3 >> (64 - 17))) ^ b0;
+	b4 += b1; b1 = ((b1 << 25) | (b1 >> (64 - 25))) ^ b4;
+	b6 += b3; b3 = ((b3 << 29) | (b3 >> (64 - 29))) ^ b6;
+	b0 += b5; b5 = ((b5 << 39) | (b5 >> (64 - 39))) ^ b0;
+	b2 += b7; b7 = ((b7 << 43) | (b7 >> (64 - 43))) ^ b2;
+	b6 += b1; b1 = ((b1 << 8) | (b1 >> (64 - 8))) ^ b6;
+	b0 += b7; b7 = ((b7 << 35) | (b7 >> (64 - 35))) ^ b0;
+	b2 += b5; b5 = ((b5 << 56) | (b5 >> (64 - 56))) ^ b2;
+	b4 += b3; b3 = ((b3 << 22) | (b3 >> (64 - 22))) ^ b4;
+	b1 += k6; b0 += b1 + k5; b1 = ((b1 << 46) | (b1 >> (64 - 46))) ^ b0;
+	b3 += k8; b2 += b3 + k7; b3 = ((b3 << 36) | (b3 >> (64 - 36))) ^ b2;
+	b5 += k1 + t2; b4 += b5 + k0; b5 = ((b5 << 19) | (b5 >> (64 - 19))) ^ b4;
+	b7 += k3 + 14; b6 += b7 + k2 + t0; b7 = ((b7 << 37) | (b7 >> (64 - 37))) ^ b6;
+	b2 += b1; b1 = ((b1 << 33) | (b1 >> (64 - 33))) ^ b2;
+	b4 += b7; b7 = ((b7 << 27) | (b7 >> (64 - 27))) ^ b4;
+	b6 += b5; b5 = ((b5 << 14) | (b5 >> (64 - 14))) ^ b6;
+	b0 += b3; b3 = ((b3 << 42) | (b3 >> (64 - 42))) ^ b0;
+	b4 += b1; b1 = ((b1 << 17) | (b1 >> (64 - 17))) ^ b4;
+	b6 += b3; b3 = ((b3 << 49) | (b3 >> (64 - 49))) ^ b6;
+	b0 += b5; b5 = ((b5 << 36) | (b5 >> (64 - 36))) ^ b0;
+	b2 += b7; b7 = ((b7 << 39) | (b7 >> (64 - 39))) ^ b2;
+	b6 += b1; b1 = ((b1 << 44) | (b1 >> (64 - 44))) ^ b6;
+	b0 += b7; b7 = ((b7 << 9) | (b7 >> (64 - 9))) ^ b0;
+	b2 += b5; b5 = ((b5 << 54) | (b5 >> (64 - 54))) ^ b2;
+	b4 += b3; b3 = ((b3 << 56) | (b3 >> (64 - 56))) ^ b4;
+	b1 += k7; b0 += b1 + k6; b1 = ((b1 << 39) | (b1 >> (64 - 39))) ^ b0;
+	b3 += k0; b2 += b3 + k8; b3 = ((b3 << 30) | (b3 >> (64 - 30))) ^ b2;
+	b5 += k2 + t0; b4 += b5 + k1; b5 = ((b5 << 34) | (b5 >> (64 - 34))) ^ b4;
+	b7 += k4 + 15; b6 += b7 + k3 + t1; b7 = ((b7 << 24) | (b7 >> (64 - 24))) ^ b6;
+	b2 += b1; b1 = ((b1 << 13) | (b1 >> (64 - 13))) ^ b2;
+	b4 += b7; b7 = ((b7 << 50) | (b7 >> (64 - 50))) ^ b4;
+	b6 += b5; b5 = ((b5 << 10) | (b5 >> (64 - 10))) ^ b6;
+	b0 += b3; b3 = ((b3 << 17) | (b3 >> (64 - 17))) ^ b0;
+	b4 += b1; b1 = ((b1 << 25) | (b1 >> (64 - 25))) ^ b4;
+	b6 += b3; b3 = ((b3 << 29) | (b3 >> (64 - 29))) ^ b6;
+	b0 += b5; b5 = ((b5 << 39) | (b5 >> (64 - 39))) ^ b0;
+	b2 += b7; b7 = ((b7 << 43) | (b7 >> (64 - 43))) ^ b2;
+	b6 += b1; b1 = ((b1 << 8) | (b1 >> (64 - 8))) ^ b6;
+	b0 += b7; b7 = ((b7 << 35) | (b7 >> (64 - 35))) ^ b0;
+	b2 += b5; b5 = ((b5 << 56) | (b5 >> (64 - 56))) ^ b2;
+	b4 += b3; b3 = ((b3 << 22) | (b3 >> (64 - 22))) ^ b4;
+	b1 += k8; b0 += b1 + k7; b1 = ((b1 << 46) | (b1 >> (64 - 46))) ^ b0;
+	b3 += k1; b2 += b3 + k0; b3 = ((b3 << 36) | (b3 >> (64 - 36))) ^ b2;
+	b5 += k3 + t1; b4 += b5 + k2; b5 = ((b5 << 19) | (b5 >> (64 - 19))) ^ b4;
+	b7 += k5 + 16; b6 += b7 + k4 + t2; b7 = ((b7 << 37) | (b7 >> (64 - 37))) ^ b6;
+	b2 += b1; b1 = ((b1 << 33) | (b1 >> (64 - 33))) ^ b2;
+	b4 += b7; b7 = ((b7 << 27) | (b7 >> (64 - 27))) ^ b4;
+	b6 += b5; b5 = ((b5 << 14) | (b5 >> (64 - 14))) ^ b6;
+	b0 += b3; b3 = ((b3 << 42) | (b3 >> (64 - 42))) ^ b0;
+	b4 += b1; b1 = ((b1 << 17) | (b1 >> (64 - 17))) ^ b4;
+	b6 += b3; b3 = ((b3 << 49) | (b3 >> (64 - 49))) ^ b6;
+	b0 += b5; b5 = ((b5 << 36) | (b5 >> (64 - 36))) ^ b0;
+	b2 += b7; b7 = ((b7 << 39) | (b7 >> (64 - 39))) ^ b2;
+	b6 += b1; b1 = ((b1 << 44) | (b1 >> (64 - 44))) ^ b6;
+	b0 += b7; b7 = ((b7 << 9) | (b7 >> (64 - 9))) ^ b0;
+	b2 += b5; b5 = ((b5 << 54) | (b5 >> (64 - 54))) ^ b2;
+	b4 += b3; b3 = ((b3 << 56) | (b3 >> (64 - 56))) ^ b4;
+	b1 += k0; b0 += b1 + k8; b1 = ((b1 << 39) | (b1 >> (64 - 39))) ^ b0;
+	b3 += k2; b2 += b3 + k1; b3 = ((b3 << 30) | (b3 >> (64 - 30))) ^ b2;
+	b5 += k4 + t2; b4 += b5 + k3; b5 = ((b5 << 34) | (b5 >> (64 - 34))) ^ b4;
+	b7 += k6 + 17; b6 += b7 + k5 + t0; b7 = ((b7 << 24) | (b7 >> (64 - 24))) ^ b6;
+	b2 += b1; b1 = ((b1 << 13) | (b1 >> (64 - 13))) ^ b2;
+	b4 += b7; b7 = ((b7 << 50) | (b7 >> (64 - 50))) ^ b4;
+	b6 += b5; b5 = ((b5 << 10) | (b5 >> (64 - 10))) ^ b6;
+	b0 += b3; b3 = ((b3 << 17) | (b3 >> (64 - 17))) ^ b0;
+	b4 += b1; b1 = ((b1 << 25) | (b1 >> (64 - 25))) ^ b4;
+	b6 += b3; b3 = ((b3 << 29) | (b3 >> (64 - 29))) ^ b6;
+	b0 += b5; b5 = ((b5 << 39) | (b5 >> (64 - 39))) ^ b0;
+	b2 += b7; b7 = ((b7 << 43) | (b7 >> (64 - 43))) ^ b2;
+	b6 += b1; b1 = ((b1 << 8) | (b1 >> (64 - 8))) ^ b6;
+	b0 += b7; b7 = ((b7 << 35) | (b7 >> (64 - 35))) ^ b0;
+	b2 += b5; b5 = ((b5 << 56) | (b5 >> (64 - 56))) ^ b2;
+	b4 += b3; b3 = ((b3 << 22) | (b3 >> (64 - 22))) ^ b4;
 
-        b1 += k1; b0 += b1 + k0; b1 = ((b1 << 46) | (b1 >> (64 - 46))) ^ b0;
-        b3 += k3; b2 += b3 + k2; b3 = ((b3 << 36) | (b3 >> (64 - 36))) ^ b2;
-        b5 += k5 + t0; b4 += b5 + k4; b5 = ((b5 << 19) | (b5 >> (64 - 19))) ^ b4;
-        b7 += k7; b6 += b7 + k6 + t1; b7 = ((b7 << 37) | (b7 >> (64 - 37))) ^ b6;
-        b2 += b1; b1 = ((b1 << 33) | (b1 >> (64 - 33))) ^ b2;
-        b4 += b7; b7 = ((b7 << 27) | (b7 >> (64 - 27))) ^ b4;
-        b6 += b5; b5 = ((b5 << 14) | (b5 >> (64 - 14))) ^ b6;
-        b0 += b3; b3 = ((b3 << 42) | (b3 >> (64 - 42))) ^ b0;
-        b4 += b1; b1 = ((b1 << 17) | (b1 >> (64 - 17))) ^ b4;
-        b6 += b3; b3 = ((b3 << 49) | (b3 >> (64 - 49))) ^ b6;
-        b0 += b5; b5 = ((b5 << 36) | (b5 >> (64 - 36))) ^ b0;
-        b2 += b7; b7 = ((b7 << 39) | (b7 >> (64 - 39))) ^ b2;
-        b6 += b1; b1 = ((b1 << 44) | (b1 >> (64 - 44))) ^ b6;
-        b0 += b7; b7 = ((b7 << 9) | (b7 >> (64 - 9))) ^ b0;
-        b2 += b5; b5 = ((b5 << 54) | (b5 >> (64 - 54))) ^ b2;
-        b4 += b3; b3 = ((b3 << 56) | (b3 >> (64 - 56))) ^ b4;
-        b1 += k2; b0 += b1 + k1; b1 = ((b1 << 39) | (b1 >> (64 - 39))) ^ b0;
-        b3 += k4; b2 += b3 + k3; b3 = ((b3 << 30) | (b3 >> (64 - 30))) ^ b2;
-        b5 += k6 + t1; b4 += b5 + k5; b5 = ((b5 << 34) | (b5 >> (64 - 34))) ^ b4;
-        b7 += k8 + 1; b6 += b7 + k7 + t2; b7 = ((b7 << 24) | (b7 >> (64 - 24))) ^ b6;
-        b2 += b1; b1 = ((b1 << 13) | (b1 >> (64 - 13))) ^ b2;
-        b4 += b7; b7 = ((b7 << 50) | (b7 >> (64 - 50))) ^ b4;
-        b6 += b5; b5 = ((b5 << 10) | (b5 >> (64 - 10))) ^ b6;
-        b0 += b3; b3 = ((b3 << 17) | (b3 >> (64 - 17))) ^ b0;
-        b4 += b1; b1 = ((b1 << 25) | (b1 >> (64 - 25))) ^ b4;
-        b6 += b3; b3 = ((b3 << 29) | (b3 >> (64 - 29))) ^ b6;
-        b0 += b5; b5 = ((b5 << 39) | (b5 >> (64 - 39))) ^ b0;
-        b2 += b7; b7 = ((b7 << 43) | (b7 >> (64 - 43))) ^ b2;
-        b6 += b1; b1 = ((b1 << 8) | (b1 >> (64 - 8))) ^ b6;
-        b0 += b7; b7 = ((b7 << 35) | (b7 >> (64 - 35))) ^ b0;
-        b2 += b5; b5 = ((b5 << 56) | (b5 >> (64 - 56))) ^ b2;
-        b4 += b3; b3 = ((b3 << 22) | (b3 >> (64 - 22))) ^ b4;
-        b1 += k3; b0 += b1 + k2; b1 = ((b1 << 46) | (b1 >> (64 - 46))) ^ b0;
-        b3 += k5; b2 += b3 + k4; b3 = ((b3 << 36) | (b3 >> (64 - 36))) ^ b2;
-        b5 += k7 + t2; b4 += b5 + k6; b5 = ((b5 << 19) | (b5 >> (64 - 19))) ^ b4;
-        b7 += k0 + 2; b6 += b7 + k8 + t0; b7 = ((b7 << 37) | (b7 >> (64 - 37))) ^ b6;
-        b2 += b1; b1 = ((b1 << 33) | (b1 >> (64 - 33))) ^ b2;
-        b4 += b7; b7 = ((b7 << 27) | (b7 >> (64 - 27))) ^ b4;
-        b6 += b5; b5 = ((b5 << 14) | (b5 >> (64 - 14))) ^ b6;
-        b0 += b3; b3 = ((b3 << 42) | (b3 >> (64 - 42))) ^ b0;
-        b4 += b1; b1 = ((b1 << 17) | (b1 >> (64 - 17))) ^ b4;
-        b6 += b3; b3 = ((b3 << 49) | (b3 >> (64 - 49))) ^ b6;
-        b0 += b5; b5 = ((b5 << 36) | (b5 >> (64 - 36))) ^ b0;
-        b2 += b7; b7 = ((b7 << 39) | (b7 >> (64 - 39))) ^ b2;
-        b6 += b1; b1 = ((b1 << 44) | (b1 >> (64 - 44))) ^ b6;
-        b0 += b7; b7 = ((b7 << 9) | (b7 >> (64 - 9))) ^ b0;
-        b2 += b5; b5 = ((b5 << 54) | (b5 >> (64 - 54))) ^ b2;
-        b4 += b3; b3 = ((b3 << 56) | (b3 >> (64 - 56))) ^ b4;
-        b1 += k4; b0 += b1 + k3; b1 = ((b1 << 39) | (b1 >> (64 - 39))) ^ b0;
-        b3 += k6; b2 += b3 + k5; b3 = ((b3 << 30) | (b3 >> (64 - 30))) ^ b2;
-        b5 += k8 + t0; b4 += b5 + k7; b5 = ((b5 << 34) | (b5 >> (64 - 34))) ^ b4;
-        b7 += k1 + 3; b6 += b7 + k0 + t1; b7 = ((b7 << 24) | (b7 >> (64 - 24))) ^ b6;
-        b2 += b1; b1 = ((b1 << 13) | (b1 >> (64 - 13))) ^ b2;
-        b4 += b7; b7 = ((b7 << 50) | (b7 >> (64 - 50))) ^ b4;
-        b6 += b5; b5 = ((b5 << 10) | (b5 >> (64 - 10))) ^ b6;
-        b0 += b3; b3 = ((b3 << 17) | (b3 >> (64 - 17))) ^ b0;
-        b4 += b1; b1 = ((b1 << 25) | (b1 >> (64 - 25))) ^ b4;
-        b6 += b3; b3 = ((b3 << 29) | (b3 >> (64 - 29))) ^ b6;
-        b0 += b5; b5 = ((b5 << 39) | (b5 >> (64 - 39))) ^ b0;
-        b2 += b7; b7 = ((b7 << 43) | (b7 >> (64 - 43))) ^ b2;
-        b6 += b1; b1 = ((b1 << 8) | (b1 >> (64 - 8))) ^ b6;
-        b0 += b7; b7 = ((b7 << 35) | (b7 >> (64 - 35))) ^ b0;
-        b2 += b5; b5 = ((b5 << 56) | (b5 >> (64 - 56))) ^ b2;
-        b4 += b3; b3 = ((b3 << 22) | (b3 >> (64 - 22))) ^ b4;
-        b1 += k5; b0 += b1 + k4; b1 = ((b1 << 46) | (b1 >> (64 - 46))) ^ b0;
-        b3 += k7; b2 += b3 + k6; b3 = ((b3 << 36) | (b3 >> (64 - 36))) ^ b2;
-        b5 += k0 + t1; b4 += b5 + k8; b5 = ((b5 << 19) | (b5 >> (64 - 19))) ^ b4;
-        b7 += k2 + 4; b6 += b7 + k1 + t2; b7 = ((b7 << 37) | (b7 >> (64 - 37))) ^ b6;
-        b2 += b1; b1 = ((b1 << 33) | (b1 >> (64 - 33))) ^ b2;
-        b4 += b7; b7 = ((b7 << 27) | (b7 >> (64 - 27))) ^ b4;
-        b6 += b5; b5 = ((b5 << 14) | (b5 >> (64 - 14))) ^ b6;
-        b0 += b3; b3 = ((b3 << 42) | (b3 >> (64 - 42))) ^ b0;
-        b4 += b1; b1 = ((b1 << 17) | (b1 >> (64 - 17))) ^ b4;
-        b6 += b3; b3 = ((b3 << 49) | (b3 >> (64 - 49))) ^ b6;
-        b0 += b5; b5 = ((b5 << 36) | (b5 >> (64 - 36))) ^ b0;
-        b2 += b7; b7 = ((b7 << 39) | (b7 >> (64 - 39))) ^ b2;
-        b6 += b1; b1 = ((b1 << 44) | (b1 >> (64 - 44))) ^ b6;
-        b0 += b7; b7 = ((b7 << 9) | (b7 >> (64 - 9))) ^ b0;
-        b2 += b5; b5 = ((b5 << 54) | (b5 >> (64 - 54))) ^ b2;
-        b4 += b3; b3 = ((b3 << 56) | (b3 >> (64 - 56))) ^ b4;
-        b1 += k6; b0 += b1 + k5; b1 = ((b1 << 39) | (b1 >> (64 - 39))) ^ b0;
-        b3 += k8; b2 += b3 + k7; b3 = ((b3 << 30) | (b3 >> (64 - 30))) ^ b2;
-        b5 += k1 + t2; b4 += b5 + k0; b5 = ((b5 << 34) | (b5 >> (64 - 34))) ^ b4;
-        b7 += k3 + 5; b6 += b7 + k2 + t0; b7 = ((b7 << 24) | (b7 >> (64 - 24))) ^ b6;
-        b2 += b1; b1 = ((b1 << 13) | (b1 >> (64 - 13))) ^ b2;
-        b4 += b7; b7 = ((b7 << 50) | (b7 >> (64 - 50))) ^ b4;
-        b6 += b5; b5 = ((b5 << 10) | (b5 >> (64 - 10))) ^ b6;
-        b0 += b3; b3 = ((b3 << 17) | (b3 >> (64 - 17))) ^ b0;
-        b4 += b1; b1 = ((b1 << 25) | (b1 >> (64 - 25))) ^ b4;
-        b6 += b3; b3 = ((b3 << 29) | (b3 >> (64 - 29))) ^ b6;
-        b0 += b5; b5 = ((b5 << 39) | (b5 >> (64 - 39))) ^ b0;
-        b2 += b7; b7 = ((b7 << 43) | (b7 >> (64 - 43))) ^ b2;
-        b6 += b1; b1 = ((b1 << 8) | (b1 >> (64 - 8))) ^ b6;
-        b0 += b7; b7 = ((b7 << 35) | (b7 >> (64 - 35))) ^ b0;
-        b2 += b5; b5 = ((b5 << 56) | (b5 >> (64 - 56))) ^ b2;
-        b4 += b3; b3 = ((b3 << 22) | (b3 >> (64 - 22))) ^ b4;
-        b1 += k7; b0 += b1 + k6; b1 = ((b1 << 46) | (b1 >> (64 - 46))) ^ b0;
-        b3 += k0; b2 += b3 + k8; b3 = ((b3 << 36) | (b3 >> (64 - 36))) ^ b2;
-        b5 += k2 + t0; b4 += b5 + k1; b5 = ((b5 << 19) | (b5 >> (64 - 19))) ^ b4;
-        b7 += k4 + 6; b6 += b7 + k3 + t1; b7 = ((b7 << 37) | (b7 >> (64 - 37))) ^ b6;
-        b2 += b1; b1 = ((b1 << 33) | (b1 >> (64 - 33))) ^ b2;
-        b4 += b7; b7 = ((b7 << 27) | (b7 >> (64 - 27))) ^ b4;
-        b6 += b5; b5 = ((b5 << 14) | (b5 >> (64 - 14))) ^ b6;
-        b0 += b3; b3 = ((b3 << 42) | (b3 >> (64 - 42))) ^ b0;
-        b4 += b1; b1 = ((b1 << 17) | (b1 >> (64 - 17))) ^ b4;
-        b6 += b3; b3 = ((b3 << 49) | (b3 >> (64 - 49))) ^ b6;
-        b0 += b5; b5 = ((b5 << 36) | (b5 >> (64 - 36))) ^ b0;
-        b2 += b7; b7 = ((b7 << 39) | (b7 >> (64 - 39))) ^ b2;
-        b6 += b1; b1 = ((b1 << 44) | (b1 >> (64 - 44))) ^ b6;
-        b0 += b7; b7 = ((b7 << 9) | (b7 >> (64 - 9))) ^ b0;
-        b2 += b5; b5 = ((b5 << 54) | (b5 >> (64 - 54))) ^ b2;
-        b4 += b3; b3 = ((b3 << 56) | (b3 >> (64 - 56))) ^ b4;
-        b1 += k8; b0 += b1 + k7; b1 = ((b1 << 39) | (b1 >> (64 - 39))) ^ b0;
-        b3 += k1; b2 += b3 + k0; b3 = ((b3 << 30) | (b3 >> (64 - 30))) ^ b2;
-        b5 += k3 + t1; b4 += b5 + k2; b5 = ((b5 << 34) | (b5 >> (64 - 34))) ^ b4;
-        b7 += k5 + 7; b6 += b7 + k4 + t2; b7 = ((b7 << 24) | (b7 >> (64 - 24))) ^ b6;
-        b2 += b1; b1 = ((b1 << 13) | (b1 >> (64 - 13))) ^ b2;
-        b4 += b7; b7 = ((b7 << 50) | (b7 >> (64 - 50))) ^ b4;
-        b6 += b5; b5 = ((b5 << 10) | (b5 >> (64 - 10))) ^ b6;
-        b0 += b3; b3 = ((b3 << 17) | (b3 >> (64 - 17))) ^ b0;
-        b4 += b1; b1 = ((b1 << 25) | (b1 >> (64 - 25))) ^ b4;
-        b6 += b3; b3 = ((b3 << 29) | (b3 >> (64 - 29))) ^ b6;
-        b0 += b5; b5 = ((b5 << 39) | (b5 >> (64 - 39))) ^ b0;
-        b2 += b7; b7 = ((b7 << 43) | (b7 >> (64 - 43))) ^ b2;
-        b6 += b1; b1 = ((b1 << 8) | (b1 >> (64 - 8))) ^ b6;
-        b0 += b7; b7 = ((b7 << 35) | (b7 >> (64 - 35))) ^ b0;
-        b2 += b5; b5 = ((b5 << 56) | (b5 >> (64 - 56))) ^ b2;
-        b4 += b3; b3 = ((b3 << 22) | (b3 >> (64 - 22))) ^ b4;
-        b1 += k0; b0 += b1 + k8; b1 = ((b1 << 46) | (b1 >> (64 - 46))) ^ b0;
-        b3 += k2; b2 += b3 + k1; b3 = ((b3 << 36) | (b3 >> (64 - 36))) ^ b2;
-        b5 += k4 + t2; b4 += b5 + k3; b5 = ((b5 << 19) | (b5 >> (64 - 19))) ^ b4;
-        b7 += k6 + 8; b6 += b7 + k5 + t0; b7 = ((b7 << 37) | (b7 >> (64 - 37))) ^ b6;
-        b2 += b1; b1 = ((b1 << 33) | (b1 >> (64 - 33))) ^ b2;
-        b4 += b7; b7 = ((b7 << 27) | (b7 >> (64 - 27))) ^ b4;
-        b6 += b5; b5 = ((b5 << 14) | (b5 >> (64 - 14))) ^ b6;
-        b0 += b3; b3 = ((b3 << 42) | (b3 >> (64 - 42))) ^ b0;
-        b4 += b1; b1 = ((b1 << 17) | (b1 >> (64 - 17))) ^ b4;
-        b6 += b3; b3 = ((b3 << 49) | (b3 >> (64 - 49))) ^ b6;
-        b0 += b5; b5 = ((b5 << 36) | (b5 >> (64 - 36))) ^ b0;
-        b2 += b7; b7 = ((b7 << 39) | (b7 >> (64 - 39))) ^ b2;
-        b6 += b1; b1 = ((b1 << 44) | (b1 >> (64 - 44))) ^ b6;
-        b0 += b7; b7 = ((b7 << 9) | (b7 >> (64 - 9))) ^ b0;
-        b2 += b5; b5 = ((b5 << 54) | (b5 >> (64 - 54))) ^ b2;
-        b4 += b3; b3 = ((b3 << 56) | (b3 >> (64 - 56))) ^ b4;
-        b1 += k1; b0 += b1 + k0; b1 = ((b1 << 39) | (b1 >> (64 - 39))) ^ b0;
-        b3 += k3; b2 += b3 + k2; b3 = ((b3 << 30) | (b3 >> (64 - 30))) ^ b2;
-        b5 += k5 + t0; b4 += b5 + k4; b5 = ((b5 << 34) | (b5 >> (64 - 34))) ^ b4;
-        b7 += k7 + 9; b6 += b7 + k6 + t1; b7 = ((b7 << 24) | (b7 >> (64 - 24))) ^ b6;
-        b2 += b1; b1 = ((b1 << 13) | (b1 >> (64 - 13))) ^ b2;
-        b4 += b7; b7 = ((b7 << 50) | (b7 >> (64 - 50))) ^ b4;
-        b6 += b5; b5 = ((b5 << 10) | (b5 >> (64 - 10))) ^ b6;
-        b0 += b3; b3 = ((b3 << 17) | (b3 >> (64 - 17))) ^ b0;
-        b4 += b1; b1 = ((b1 << 25) | (b1 >> (64 - 25))) ^ b4;
-        b6 += b3; b3 = ((b3 << 29) | (b3 >> (64 - 29))) ^ b6;
-        b0 += b5; b5 = ((b5 << 39) | (b5 >> (64 - 39))) ^ b0;
-        b2 += b7; b7 = ((b7 << 43) | (b7 >> (64 - 43))) ^ b2;
-        b6 += b1; b1 = ((b1 << 8) | (b1 >> (64 - 8))) ^ b6;
-        b0 += b7; b7 = ((b7 << 35) | (b7 >> (64 - 35))) ^ b0;
-        b2 += b5; b5 = ((b5 << 56) | (b5 >> (64 - 56))) ^ b2;
-        b4 += b3; b3 = ((b3 << 22) | (b3 >> (64 - 22))) ^ b4;
-        b1 += k2; b0 += b1 + k1; b1 = ((b1 << 46) | (b1 >> (64 - 46))) ^ b0;
-        b3 += k4; b2 += b3 + k3; b3 = ((b3 << 36) | (b3 >> (64 - 36))) ^ b2;
-        b5 += k6 + t1; b4 += b5 + k5; b5 = ((b5 << 19) | (b5 >> (64 - 19))) ^ b4;
-        b7 += k8 + 10; b6 += b7 + k7 + t2; b7 = ((b7 << 37) | (b7 >> (64 - 37))) ^ b6;
-        b2 += b1; b1 = ((b1 << 33) | (b1 >> (64 - 33))) ^ b2;
-        b4 += b7; b7 = ((b7 << 27) | (b7 >> (64 - 27))) ^ b4;
-        b6 += b5; b5 = ((b5 << 14) | (b5 >> (64 - 14))) ^ b6;
-        b0 += b3; b3 = ((b3 << 42) | (b3 >> (64 - 42))) ^ b0;
-        b4 += b1; b1 = ((b1 << 17) | (b1 >> (64 - 17))) ^ b4;
-        b6 += b3; b3 = ((b3 << 49) | (b3 >> (64 - 49))) ^ b6;
-        b0 += b5; b5 = ((b5 << 36) | (b5 >> (64 - 36))) ^ b0;
-        b2 += b7; b7 = ((b7 << 39) | (b7 >> (64 - 39))) ^ b2;
-        b6 += b1; b1 = ((b1 << 44) | (b1 >> (64 - 44))) ^ b6;
-        b0 += b7; b7 = ((b7 << 9) | (b7 >> (64 - 9))) ^ b0;
-        b2 += b5; b5 = ((b5 << 54) | (b5 >> (64 - 54))) ^ b2;
-        b4 += b3; b3 = ((b3 << 56) | (b3 >> (64 - 56))) ^ b4;
-        b1 += k3; b0 += b1 + k2; b1 = ((b1 << 39) | (b1 >> (64 - 39))) ^ b0;
-        b3 += k5; b2 += b3 + k4; b3 = ((b3 << 30) | (b3 >> (64 - 30))) ^ b2;
-        b5 += k7 + t2; b4 += b5 + k6; b5 = ((b5 << 34) | (b5 >> (64 - 34))) ^ b4;
-        b7 += k0 + 11; b6 += b7 + k8 + t0; b7 = ((b7 << 24) | (b7 >> (64 - 24))) ^ b6;
-        b2 += b1; b1 = ((b1 << 13) | (b1 >> (64 - 13))) ^ b2;
-        b4 += b7; b7 = ((b7 << 50) | (b7 >> (64 - 50))) ^ b4;
-        b6 += b5; b5 = ((b5 << 10) | (b5 >> (64 - 10))) ^ b6;
-        b0 += b3; b3 = ((b3 << 17) | (b3 >> (64 - 17))) ^ b0;
-        b4 += b1; b1 = ((b1 << 25) | (b1 >> (64 - 25))) ^ b4;
-        b6 += b3; b3 = ((b3 << 29) | (b3 >> (64 - 29))) ^ b6;
-        b0 += b5; b5 = ((b5 << 39) | (b5 >> (64 - 39))) ^ b0;
-        b2 += b7; b7 = ((b7 << 43) | (b7 >> (64 - 43))) ^ b2;
-        b6 += b1; b1 = ((b1 << 8) | (b1 >> (64 - 8))) ^ b6;
-        b0 += b7; b7 = ((b7 << 35) | (b7 >> (64 - 35))) ^ b0;
-        b2 += b5; b5 = ((b5 << 56) | (b5 >> (64 - 56))) ^ b2;
-        b4 += b3; b3 = ((b3 << 22) | (b3 >> (64 - 22))) ^ b4;
-        b1 += k4; b0 += b1 + k3; b1 = ((b1 << 46) | (b1 >> (64 - 46))) ^ b0;
-        b3 += k6; b2 += b3 + k5; b3 = ((b3 << 36) | (b3 >> (64 - 36))) ^ b2;
-        b5 += k8 + t0; b4 += b5 + k7; b5 = ((b5 << 19) | (b5 >> (64 - 19))) ^ b4;
-        b7 += k1 + 12; b6 += b7 + k0 + t1; b7 = ((b7 << 37) | (b7 >> (64 - 37))) ^ b6;
-        b2 += b1; b1 = ((b1 << 33) | (b1 >> (64 - 33))) ^ b2;
-        b4 += b7; b7 = ((b7 << 27) | (b7 >> (64 - 27))) ^ b4;
-        b6 += b5; b5 = ((b5 << 14) | (b5 >> (64 - 14))) ^ b6;
-        b0 += b3; b3 = ((b3 << 42) | (b3 >> (64 - 42))) ^ b0;
-        b4 += b1; b1 = ((b1 << 17) | (b1 >> (64 - 17))) ^ b4;
-        b6 += b3; b3 = ((b3 << 49) | (b3 >> (64 - 49))) ^ b6;
-        b0 += b5; b5 = ((b5 << 36) | (b5 >> (64 - 36))) ^ b0;
-        b2 += b7; b7 = ((b7 << 39) | (b7 >> (64 - 39))) ^ b2;
-        b6 += b1; b1 = ((b1 << 44) | (b1 >> (64 - 44))) ^ b6;
-        b0 += b7; b7 = ((b7 << 9) | (b7 >> (64 - 9))) ^ b0;
-        b2 += b5; b5 = ((b5 << 54) | (b5 >> (64 - 54))) ^ b2;
-        b4 += b3; b3 = ((b3 << 56) | (b3 >> (64 - 56))) ^ b4;
-        b1 += k5; b0 += b1 + k4; b1 = ((b1 << 39) | (b1 >> (64 - 39))) ^ b0;
-        b3 += k7; b2 += b3 + k6; b3 = ((b3 << 30) | (b3 >> (64 - 30))) ^ b2;
-        b5 += k0 + t1; b4 += b5 + k8; b5 = ((b5 << 34) | (b5 >> (64 - 34))) ^ b4;
-        b7 += k2 + 13; b6 += b7 + k1 + t2; b7 = ((b7 << 24) | (b7 >> (64 - 24))) ^ b6;
-        b2 += b1; b1 = ((b1 << 13) | (b1 >> (64 - 13))) ^ b2;
-        b4 += b7; b7 = ((b7 << 50) | (b7 >> (64 - 50))) ^ b4;
-        b6 += b5; b5 = ((b5 << 10) | (b5 >> (64 - 10))) ^ b6;
-        b0 += b3; b3 = ((b3 << 17) | (b3 >> (64 - 17))) ^ b0;
-        b4 += b1; b1 = ((b1 << 25) | (b1 >> (64 - 25))) ^ b4;
-        b6 += b3; b3 = ((b3 << 29) | (b3 >> (64 - 29))) ^ b6;
-        b0 += b5; b5 = ((b5 << 39) | (b5 >> (64 - 39))) ^ b0;
-        b2 += b7; b7 = ((b7 << 43) | (b7 >> (64 - 43))) ^ b2;
-        b6 += b1; b1 = ((b1 << 8) | (b1 >> (64 - 8))) ^ b6;
-        b0 += b7; b7 = ((b7 << 35) | (b7 >> (64 - 35))) ^ b0;
-        b2 += b5; b5 = ((b5 << 56) | (b5 >> (64 - 56))) ^ b2;
-        b4 += b3; b3 = ((b3 << 22) | (b3 >> (64 - 22))) ^ b4;
-        b1 += k6; b0 += b1 + k5; b1 = ((b1 << 46) | (b1 >> (64 - 46))) ^ b0;
-        b3 += k8; b2 += b3 + k7; b3 = ((b3 << 36) | (b3 >> (64 - 36))) ^ b2;
-        b5 += k1 + t2; b4 += b5 + k0; b5 = ((b5 << 19) | (b5 >> (64 - 19))) ^ b4;
-        b7 += k3 + 14; b6 += b7 + k2 + t0; b7 = ((b7 << 37) | (b7 >> (64 - 37))) ^ b6;
-        b2 += b1; b1 = ((b1 << 33) | (b1 >> (64 - 33))) ^ b2;
-        b4 += b7; b7 = ((b7 << 27) | (b7 >> (64 - 27))) ^ b4;
-        b6 += b5; b5 = ((b5 << 14) | (b5 >> (64 - 14))) ^ b6;
-        b0 += b3; b3 = ((b3 << 42) | (b3 >> (64 - 42))) ^ b0;
-        b4 += b1; b1 = ((b1 << 17) | (b1 >> (64 - 17))) ^ b4;
-        b6 += b3; b3 = ((b3 << 49) | (b3 >> (64 - 49))) ^ b6;
-        b0 += b5; b5 = ((b5 << 36) | (b5 >> (64 - 36))) ^ b0;
-        b2 += b7; b7 = ((b7 << 39) | (b7 >> (64 - 39))) ^ b2;
-        b6 += b1; b1 = ((b1 << 44) | (b1 >> (64 - 44))) ^ b6;
-        b0 += b7; b7 = ((b7 << 9) | (b7 >> (64 - 9))) ^ b0;
-        b2 += b5; b5 = ((b5 << 54) | (b5 >> (64 - 54))) ^ b2;
-        b4 += b3; b3 = ((b3 << 56) | (b3 >> (64 - 56))) ^ b4;
-        b1 += k7; b0 += b1 + k6; b1 = ((b1 << 39) | (b1 >> (64 - 39))) ^ b0;
-        b3 += k0; b2 += b3 + k8; b3 = ((b3 << 30) | (b3 >> (64 - 30))) ^ b2;
-        b5 += k2 + t0; b4 += b5 + k1; b5 = ((b5 << 34) | (b5 >> (64 - 34))) ^ b4;
-        b7 += k4 + 15; b6 += b7 + k3 + t1; b7 = ((b7 << 24) | (b7 >> (64 - 24))) ^ b6;
-        b2 += b1; b1 = ((b1 << 13) | (b1 >> (64 - 13))) ^ b2;
-        b4 += b7; b7 = ((b7 << 50) | (b7 >> (64 - 50))) ^ b4;
-        b6 += b5; b5 = ((b5 << 10) | (b5 >> (64 - 10))) ^ b6;
-        b0 += b3; b3 = ((b3 << 17) | (b3 >> (64 - 17))) ^ b0;
-        b4 += b1; b1 = ((b1 << 25) | (b1 >> (64 - 25))) ^ b4;
-        b6 += b3; b3 = ((b3 << 29) | (b3 >> (64 - 29))) ^ b6;
-        b0 += b5; b5 = ((b5 << 39) | (b5 >> (64 - 39))) ^ b0;
-        b2 += b7; b7 = ((b7 << 43) | (b7 >> (64 - 43))) ^ b2;
-        b6 += b1; b1 = ((b1 << 8) | (b1 >> (64 - 8))) ^ b6;
-        b0 += b7; b7 = ((b7 << 35) | (b7 >> (64 - 35))) ^ b0;
-        b2 += b5; b5 = ((b5 << 56) | (b5 >> (64 - 56))) ^ b2;
-        b4 += b3; b3 = ((b3 << 22) | (b3 >> (64 - 22))) ^ b4;
-        b1 += k8; b0 += b1 + k7; b1 = ((b1 << 46) | (b1 >> (64 - 46))) ^ b0;
-        b3 += k1; b2 += b3 + k0; b3 = ((b3 << 36) | (b3 >> (64 - 36))) ^ b2;
-        b5 += k3 + t1; b4 += b5 + k2; b5 = ((b5 << 19) | (b5 >> (64 - 19))) ^ b4;
-        b7 += k5 + 16; b6 += b7 + k4 + t2; b7 = ((b7 << 37) | (b7 >> (64 - 37))) ^ b6;
-        b2 += b1; b1 = ((b1 << 33) | (b1 >> (64 - 33))) ^ b2;
-        b4 += b7; b7 = ((b7 << 27) | (b7 >> (64 - 27))) ^ b4;
-        b6 += b5; b5 = ((b5 << 14) | (b5 >> (64 - 14))) ^ b6;
-        b0 += b3; b3 = ((b3 << 42) | (b3 >> (64 - 42))) ^ b0;
-        b4 += b1; b1 = ((b1 << 17) | (b1 >> (64 - 17))) ^ b4;
-        b6 += b3; b3 = ((b3 << 49) | (b3 >> (64 - 49))) ^ b6;
-        b0 += b5; b5 = ((b5 << 36) | (b5 >> (64 - 36))) ^ b0;
-        b2 += b7; b7 = ((b7 << 39) | (b7 >> (64 - 39))) ^ b2;
-        b6 += b1; b1 = ((b1 << 44) | (b1 >> (64 - 44))) ^ b6;
-        b0 += b7; b7 = ((b7 << 9) | (b7 >> (64 - 9))) ^ b0;
-        b2 += b5; b5 = ((b5 << 54) | (b5 >> (64 - 54))) ^ b2;
-        b4 += b3; b3 = ((b3 << 56) | (b3 >> (64 - 56))) ^ b4;
-        b1 += k0; b0 += b1 + k8; b1 = ((b1 << 39) | (b1 >> (64 - 39))) ^ b0;
-        b3 += k2; b2 += b3 + k1; b3 = ((b3 << 30) | (b3 >> (64 - 30))) ^ b2;
-        b5 += k4 + t2; b4 += b5 + k3; b5 = ((b5 << 34) | (b5 >> (64 - 34))) ^ b4;
-        b7 += k6 + 17; b6 += b7 + k5 + t0; b7 = ((b7 << 24) | (b7 >> (64 - 24))) ^ b6;
-        b2 += b1; b1 = ((b1 << 13) | (b1 >> (64 - 13))) ^ b2;
-        b4 += b7; b7 = ((b7 << 50) | (b7 >> (64 - 50))) ^ b4;
-        b6 += b5; b5 = ((b5 << 10) | (b5 >> (64 - 10))) ^ b6;
-        b0 += b3; b3 = ((b3 << 17) | (b3 >> (64 - 17))) ^ b0;
-        b4 += b1; b1 = ((b1 << 25) | (b1 >> (64 - 25))) ^ b4;
-        b6 += b3; b3 = ((b3 << 29) | (b3 >> (64 - 29))) ^ b6;
-        b0 += b5; b5 = ((b5 << 39) | (b5 >> (64 - 39))) ^ b0;
-        b2 += b7; b7 = ((b7 << 43) | (b7 >> (64 - 43))) ^ b2;
-        b6 += b1; b1 = ((b1 << 8) | (b1 >> (64 - 8))) ^ b6;
-        b0 += b7; b7 = ((b7 << 35) | (b7 >> (64 - 35))) ^ b0;
-        b2 += b5; b5 = ((b5 << 56) | (b5 >> (64 - 56))) ^ b2;
-        b4 += b3; b3 = ((b3 << 22) | (b3 >> (64 - 22))) ^ b4;
-
-        output[0] = b0 + k0;
-        output[1] = b1 + k1;
-        output[2] = b2 + k2;
-        output[3] = b3 + k3;
-        output[4] = b4 + k4;
-        output[5] = b5 + k5 + t0;
-        output[6] = b6 + k6 + t1;
-        output[7] = b7 + k7 + 18;
-    }
+	output[0] = b0 + k0;
+	output[1] = b1 + k1;
+	output[2] = b2 + k2;
+	output[3] = b3 + k3;
+	output[4] = b4 + k4;
+	output[5] = b5 + k5 + t0;
+	output[6] = b6 + k6 + t1;
+	output[7] = b7 + k7 + 18;
+}
 
 void threefishDecrypt512(struct threefish_key *keyCtx, u64 *input, u64 *output)
-    {
-
-    u64 b0 = input[0], b1 = input[1],
-      b2 = input[2], b3 = input[3],
-      b4 = input[4], b5 = input[5],
-      b6 = input[6], b7 = input[7];
-    u64 k0 = keyCtx->key[0], k1 = keyCtx->key[1],
-      k2 = keyCtx->key[2], k3 = keyCtx->key[3],
-      k4 = keyCtx->key[4], k5 = keyCtx->key[5],
-      k6 = keyCtx->key[6], k7 = keyCtx->key[7],
-      k8 = keyCtx->key[8];
-    u64 t0 = keyCtx->tweak[0], t1 = keyCtx->tweak[1],
-      t2 = keyCtx->tweak[2];
+{
+	u64 b0 = input[0], b1 = input[1],
+	  b2 = input[2], b3 = input[3],
+	  b4 = input[4], b5 = input[5],
+	  b6 = input[6], b7 = input[7];
+	u64 k0 = keyCtx->key[0], k1 = keyCtx->key[1],
+	  k2 = keyCtx->key[2], k3 = keyCtx->key[3],
+	  k4 = keyCtx->key[4], k5 = keyCtx->key[5],
+	  k6 = keyCtx->key[6], k7 = keyCtx->key[7],
+	  k8 = keyCtx->key[8];
+	u64 t0 = keyCtx->tweak[0], t1 = keyCtx->tweak[1],
+	  t2 = keyCtx->tweak[2];
 
-      u64 tmp;
+	u64 tmp;
 
-        b0 -= k0;
-        b1 -= k1;
-        b2 -= k2;
-        b3 -= k3;
-        b4 -= k4;
-        b5 -= k5 + t0;
-        b6 -= k6 + t1;
-        b7 -= k7 + 18;
-        tmp = b3 ^ b4; b3 = (tmp >> 22) | (tmp << (64 - 22)); b4 -= b3;
-        tmp = b5 ^ b2; b5 = (tmp >> 56) | (tmp << (64 - 56)); b2 -= b5;
-        tmp = b7 ^ b0; b7 = (tmp >> 35) | (tmp << (64 - 35)); b0 -= b7;
-        tmp = b1 ^ b6; b1 = (tmp >> 8) | (tmp << (64 - 8)); b6 -= b1;
-        tmp = b7 ^ b2; b7 = (tmp >> 43) | (tmp << (64 - 43)); b2 -= b7;
-        tmp = b5 ^ b0; b5 = (tmp >> 39) | (tmp << (64 - 39)); b0 -= b5;
-        tmp = b3 ^ b6; b3 = (tmp >> 29) | (tmp << (64 - 29)); b6 -= b3;
-        tmp = b1 ^ b4; b1 = (tmp >> 25) | (tmp << (64 - 25)); b4 -= b1;
-        tmp = b3 ^ b0; b3 = (tmp >> 17) | (tmp << (64 - 17)); b0 -= b3;
-        tmp = b5 ^ b6; b5 = (tmp >> 10) | (tmp << (64 - 10)); b6 -= b5;
-        tmp = b7 ^ b4; b7 = (tmp >> 50) | (tmp << (64 - 50)); b4 -= b7;
-        tmp = b1 ^ b2; b1 = (tmp >> 13) | (tmp << (64 - 13)); b2 -= b1;
-        tmp = b7 ^ b6; b7 = (tmp >> 24) | (tmp << (64 - 24)); b6 -= b7 + k5 + t0; b7 -= k6 + 17;
-        tmp = b5 ^ b4; b5 = (tmp >> 34) | (tmp << (64 - 34)); b4 -= b5 + k3; b5 -= k4 + t2;
-        tmp = b3 ^ b2; b3 = (tmp >> 30) | (tmp << (64 - 30)); b2 -= b3 + k1; b3 -= k2;
-        tmp = b1 ^ b0; b1 = (tmp >> 39) | (tmp << (64 - 39)); b0 -= b1 + k8; b1 -= k0;
-        tmp = b3 ^ b4; b3 = (tmp >> 56) | (tmp << (64 - 56)); b4 -= b3;
-        tmp = b5 ^ b2; b5 = (tmp >> 54) | (tmp << (64 - 54)); b2 -= b5;
-        tmp = b7 ^ b0; b7 = (tmp >> 9) | (tmp << (64 - 9)); b0 -= b7;
-        tmp = b1 ^ b6; b1 = (tmp >> 44) | (tmp << (64 - 44)); b6 -= b1;
-        tmp = b7 ^ b2; b7 = (tmp >> 39) | (tmp << (64 - 39)); b2 -= b7;
-        tmp = b5 ^ b0; b5 = (tmp >> 36) | (tmp << (64 - 36)); b0 -= b5;
-        tmp = b3 ^ b6; b3 = (tmp >> 49) | (tmp << (64 - 49)); b6 -= b3;
-        tmp = b1 ^ b4; b1 = (tmp >> 17) | (tmp << (64 - 17)); b4 -= b1;
-        tmp = b3 ^ b0; b3 = (tmp >> 42) | (tmp << (64 - 42)); b0 -= b3;
-        tmp = b5 ^ b6; b5 = (tmp >> 14) | (tmp << (64 - 14)); b6 -= b5;
-        tmp = b7 ^ b4; b7 = (tmp >> 27) | (tmp << (64 - 27)); b4 -= b7;
-        tmp = b1 ^ b2; b1 = (tmp >> 33) | (tmp << (64 - 33)); b2 -= b1;
-        tmp = b7 ^ b6; b7 = (tmp >> 37) | (tmp << (64 - 37)); b6 -= b7 + k4 + t2; b7 -= k5 + 16;
-        tmp = b5 ^ b4; b5 = (tmp >> 19) | (tmp << (64 - 19)); b4 -= b5 + k2; b5 -= k3 + t1;
-        tmp = b3 ^ b2; b3 = (tmp >> 36) | (tmp << (64 - 36)); b2 -= b3 + k0; b3 -= k1;
-        tmp = b1 ^ b0; b1 = (tmp >> 46) | (tmp << (64 - 46)); b0 -= b1 + k7; b1 -= k8;
-        tmp = b3 ^ b4; b3 = (tmp >> 22) | (tmp << (64 - 22)); b4 -= b3;
-        tmp = b5 ^ b2; b5 = (tmp >> 56) | (tmp << (64 - 56)); b2 -= b5;
-        tmp = b7 ^ b0; b7 = (tmp >> 35) | (tmp << (64 - 35)); b0 -= b7;
-        tmp = b1 ^ b6; b1 = (tmp >> 8) | (tmp << (64 - 8)); b6 -= b1;
-        tmp = b7 ^ b2; b7 = (tmp >> 43) | (tmp << (64 - 43)); b2 -= b7;
-        tmp = b5 ^ b0; b5 = (tmp >> 39) | (tmp << (64 - 39)); b0 -= b5;
-        tmp = b3 ^ b6; b3 = (tmp >> 29) | (tmp << (64 - 29)); b6 -= b3;
-        tmp = b1 ^ b4; b1 = (tmp >> 25) | (tmp << (64 - 25)); b4 -= b1;
-        tmp = b3 ^ b0; b3 = (tmp >> 17) | (tmp << (64 - 17)); b0 -= b3;
-        tmp = b5 ^ b6; b5 = (tmp >> 10) | (tmp << (64 - 10)); b6 -= b5;
-        tmp = b7 ^ b4; b7 = (tmp >> 50) | (tmp << (64 - 50)); b4 -= b7;
-        tmp = b1 ^ b2; b1 = (tmp >> 13) | (tmp << (64 - 13)); b2 -= b1;
-        tmp = b7 ^ b6; b7 = (tmp >> 24) | (tmp << (64 - 24)); b6 -= b7 + k3 + t1; b7 -= k4 + 15;
-        tmp = b5 ^ b4; b5 = (tmp >> 34) | (tmp << (64 - 34)); b4 -= b5 + k1; b5 -= k2 + t0;
-        tmp = b3 ^ b2; b3 = (tmp >> 30) | (tmp << (64 - 30)); b2 -= b3 + k8; b3 -= k0;
-        tmp = b1 ^ b0; b1 = (tmp >> 39) | (tmp << (64 - 39)); b0 -= b1 + k6; b1 -= k7;
-        tmp = b3 ^ b4; b3 = (tmp >> 56) | (tmp << (64 - 56)); b4 -= b3;
-        tmp = b5 ^ b2; b5 = (tmp >> 54) | (tmp << (64 - 54)); b2 -= b5;
-        tmp = b7 ^ b0; b7 = (tmp >> 9) | (tmp << (64 - 9)); b0 -= b7;
-        tmp = b1 ^ b6; b1 = (tmp >> 44) | (tmp << (64 - 44)); b6 -= b1;
-        tmp = b7 ^ b2; b7 = (tmp >> 39) | (tmp << (64 - 39)); b2 -= b7;
-        tmp = b5 ^ b0; b5 = (tmp >> 36) | (tmp << (64 - 36)); b0 -= b5;
-        tmp = b3 ^ b6; b3 = (tmp >> 49) | (tmp << (64 - 49)); b6 -= b3;
-        tmp = b1 ^ b4; b1 = (tmp >> 17) | (tmp << (64 - 17)); b4 -= b1;
-        tmp = b3 ^ b0; b3 = (tmp >> 42) | (tmp << (64 - 42)); b0 -= b3;
-        tmp = b5 ^ b6; b5 = (tmp >> 14) | (tmp << (64 - 14)); b6 -= b5;
-        tmp = b7 ^ b4; b7 = (tmp >> 27) | (tmp << (64 - 27)); b4 -= b7;
-        tmp = b1 ^ b2; b1 = (tmp >> 33) | (tmp << (64 - 33)); b2 -= b1;
-        tmp = b7 ^ b6; b7 = (tmp >> 37) | (tmp << (64 - 37)); b6 -= b7 + k2 + t0; b7 -= k3 + 14;
-        tmp = b5 ^ b4; b5 = (tmp >> 19) | (tmp << (64 - 19)); b4 -= b5 + k0; b5 -= k1 + t2;
-        tmp = b3 ^ b2; b3 = (tmp >> 36) | (tmp << (64 - 36)); b2 -= b3 + k7; b3 -= k8;
-        tmp = b1 ^ b0; b1 = (tmp >> 46) | (tmp << (64 - 46)); b0 -= b1 + k5; b1 -= k6;
-        tmp = b3 ^ b4; b3 = (tmp >> 22) | (tmp << (64 - 22)); b4 -= b3;
-        tmp = b5 ^ b2; b5 = (tmp >> 56) | (tmp << (64 - 56)); b2 -= b5;
-        tmp = b7 ^ b0; b7 = (tmp >> 35) | (tmp << (64 - 35)); b0 -= b7;
-        tmp = b1 ^ b6; b1 = (tmp >> 8) | (tmp << (64 - 8)); b6 -= b1;
-        tmp = b7 ^ b2; b7 = (tmp >> 43) | (tmp << (64 - 43)); b2 -= b7;
-        tmp = b5 ^ b0; b5 = (tmp >> 39) | (tmp << (64 - 39)); b0 -= b5;
-        tmp = b3 ^ b6; b3 = (tmp >> 29) | (tmp << (64 - 29)); b6 -= b3;
-        tmp = b1 ^ b4; b1 = (tmp >> 25) | (tmp << (64 - 25)); b4 -= b1;
-        tmp = b3 ^ b0; b3 = (tmp >> 17) | (tmp << (64 - 17)); b0 -= b3;
-        tmp = b5 ^ b6; b5 = (tmp >> 10) | (tmp << (64 - 10)); b6 -= b5;
-        tmp = b7 ^ b4; b7 = (tmp >> 50) | (tmp << (64 - 50)); b4 -= b7;
-        tmp = b1 ^ b2; b1 = (tmp >> 13) | (tmp << (64 - 13)); b2 -= b1;
-        tmp = b7 ^ b6; b7 = (tmp >> 24) | (tmp << (64 - 24)); b6 -= b7 + k1 + t2; b7 -= k2 + 13;
-        tmp = b5 ^ b4; b5 = (tmp >> 34) | (tmp << (64 - 34)); b4 -= b5 + k8; b5 -= k0 + t1;
-        tmp = b3 ^ b2; b3 = (tmp >> 30) | (tmp << (64 - 30)); b2 -= b3 + k6; b3 -= k7;
-        tmp = b1 ^ b0; b1 = (tmp >> 39) | (tmp << (64 - 39)); b0 -= b1 + k4; b1 -= k5;
-        tmp = b3 ^ b4; b3 = (tmp >> 56) | (tmp << (64 - 56)); b4 -= b3;
-        tmp = b5 ^ b2; b5 = (tmp >> 54) | (tmp << (64 - 54)); b2 -= b5;
-        tmp = b7 ^ b0; b7 = (tmp >> 9) | (tmp << (64 - 9)); b0 -= b7;
-        tmp = b1 ^ b6; b1 = (tmp >> 44) | (tmp << (64 - 44)); b6 -= b1;
-        tmp = b7 ^ b2; b7 = (tmp >> 39) | (tmp << (64 - 39)); b2 -= b7;
-        tmp = b5 ^ b0; b5 = (tmp >> 36) | (tmp << (64 - 36)); b0 -= b5;
-        tmp = b3 ^ b6; b3 = (tmp >> 49) | (tmp << (64 - 49)); b6 -= b3;
-        tmp = b1 ^ b4; b1 = (tmp >> 17) | (tmp << (64 - 17)); b4 -= b1;
-        tmp = b3 ^ b0; b3 = (tmp >> 42) | (tmp << (64 - 42)); b0 -= b3;
-        tmp = b5 ^ b6; b5 = (tmp >> 14) | (tmp << (64 - 14)); b6 -= b5;
-        tmp = b7 ^ b4; b7 = (tmp >> 27) | (tmp << (64 - 27)); b4 -= b7;
-        tmp = b1 ^ b2; b1 = (tmp >> 33) | (tmp << (64 - 33)); b2 -= b1;
-        tmp = b7 ^ b6; b7 = (tmp >> 37) | (tmp << (64 - 37)); b6 -= b7 + k0 + t1; b7 -= k1 + 12;
-        tmp = b5 ^ b4; b5 = (tmp >> 19) | (tmp << (64 - 19)); b4 -= b5 + k7; b5 -= k8 + t0;
-        tmp = b3 ^ b2; b3 = (tmp >> 36) | (tmp << (64 - 36)); b2 -= b3 + k5; b3 -= k6;
-        tmp = b1 ^ b0; b1 = (tmp >> 46) | (tmp << (64 - 46)); b0 -= b1 + k3; b1 -= k4;
-        tmp = b3 ^ b4; b3 = (tmp >> 22) | (tmp << (64 - 22)); b4 -= b3;
-        tmp = b5 ^ b2; b5 = (tmp >> 56) | (tmp << (64 - 56)); b2 -= b5;
-        tmp = b7 ^ b0; b7 = (tmp >> 35) | (tmp << (64 - 35)); b0 -= b7;
-        tmp = b1 ^ b6; b1 = (tmp >> 8) | (tmp << (64 - 8)); b6 -= b1;
-        tmp = b7 ^ b2; b7 = (tmp >> 43) | (tmp << (64 - 43)); b2 -= b7;
-        tmp = b5 ^ b0; b5 = (tmp >> 39) | (tmp << (64 - 39)); b0 -= b5;
-        tmp = b3 ^ b6; b3 = (tmp >> 29) | (tmp << (64 - 29)); b6 -= b3;
-        tmp = b1 ^ b4; b1 = (tmp >> 25) | (tmp << (64 - 25)); b4 -= b1;
-        tmp = b3 ^ b0; b3 = (tmp >> 17) | (tmp << (64 - 17)); b0 -= b3;
-        tmp = b5 ^ b6; b5 = (tmp >> 10) | (tmp << (64 - 10)); b6 -= b5;
-        tmp = b7 ^ b4; b7 = (tmp >> 50) | (tmp << (64 - 50)); b4 -= b7;
-        tmp = b1 ^ b2; b1 = (tmp >> 13) | (tmp << (64 - 13)); b2 -= b1;
-        tmp = b7 ^ b6; b7 = (tmp >> 24) | (tmp << (64 - 24)); b6 -= b7 + k8 + t0; b7 -= k0 + 11;
-        tmp = b5 ^ b4; b5 = (tmp >> 34) | (tmp << (64 - 34)); b4 -= b5 + k6; b5 -= k7 + t2;
-        tmp = b3 ^ b2; b3 = (tmp >> 30) | (tmp << (64 - 30)); b2 -= b3 + k4; b3 -= k5;
-        tmp = b1 ^ b0; b1 = (tmp >> 39) | (tmp << (64 - 39)); b0 -= b1 + k2; b1 -= k3;
-        tmp = b3 ^ b4; b3 = (tmp >> 56) | (tmp << (64 - 56)); b4 -= b3;
-        tmp = b5 ^ b2; b5 = (tmp >> 54) | (tmp << (64 - 54)); b2 -= b5;
-        tmp = b7 ^ b0; b7 = (tmp >> 9) | (tmp << (64 - 9)); b0 -= b7;
-        tmp = b1 ^ b6; b1 = (tmp >> 44) | (tmp << (64 - 44)); b6 -= b1;
-        tmp = b7 ^ b2; b7 = (tmp >> 39) | (tmp << (64 - 39)); b2 -= b7;
-        tmp = b5 ^ b0; b5 = (tmp >> 36) | (tmp << (64 - 36)); b0 -= b5;
-        tmp = b3 ^ b6; b3 = (tmp >> 49) | (tmp << (64 - 49)); b6 -= b3;
-        tmp = b1 ^ b4; b1 = (tmp >> 17) | (tmp << (64 - 17)); b4 -= b1;
-        tmp = b3 ^ b0; b3 = (tmp >> 42) | (tmp << (64 - 42)); b0 -= b3;
-        tmp = b5 ^ b6; b5 = (tmp >> 14) | (tmp << (64 - 14)); b6 -= b5;
-        tmp = b7 ^ b4; b7 = (tmp >> 27) | (tmp << (64 - 27)); b4 -= b7;
-        tmp = b1 ^ b2; b1 = (tmp >> 33) | (tmp << (64 - 33)); b2 -= b1;
-        tmp = b7 ^ b6; b7 = (tmp >> 37) | (tmp << (64 - 37)); b6 -= b7 + k7 + t2; b7 -= k8 + 10;
-        tmp = b5 ^ b4; b5 = (tmp >> 19) | (tmp << (64 - 19)); b4 -= b5 + k5; b5 -= k6 + t1;
-        tmp = b3 ^ b2; b3 = (tmp >> 36) | (tmp << (64 - 36)); b2 -= b3 + k3; b3 -= k4;
-        tmp = b1 ^ b0; b1 = (tmp >> 46) | (tmp << (64 - 46)); b0 -= b1 + k1; b1 -= k2;
-        tmp = b3 ^ b4; b3 = (tmp >> 22) | (tmp << (64 - 22)); b4 -= b3;
-        tmp = b5 ^ b2; b5 = (tmp >> 56) | (tmp << (64 - 56)); b2 -= b5;
-        tmp = b7 ^ b0; b7 = (tmp >> 35) | (tmp << (64 - 35)); b0 -= b7;
-        tmp = b1 ^ b6; b1 = (tmp >> 8) | (tmp << (64 - 8)); b6 -= b1;
-        tmp = b7 ^ b2; b7 = (tmp >> 43) | (tmp << (64 - 43)); b2 -= b7;
-        tmp = b5 ^ b0; b5 = (tmp >> 39) | (tmp << (64 - 39)); b0 -= b5;
-        tmp = b3 ^ b6; b3 = (tmp >> 29) | (tmp << (64 - 29)); b6 -= b3;
-        tmp = b1 ^ b4; b1 = (tmp >> 25) | (tmp << (64 - 25)); b4 -= b1;
-        tmp = b3 ^ b0; b3 = (tmp >> 17) | (tmp << (64 - 17)); b0 -= b3;
-        tmp = b5 ^ b6; b5 = (tmp >> 10) | (tmp << (64 - 10)); b6 -= b5;
-        tmp = b7 ^ b4; b7 = (tmp >> 50) | (tmp << (64 - 50)); b4 -= b7;
-        tmp = b1 ^ b2; b1 = (tmp >> 13) | (tmp << (64 - 13)); b2 -= b1;
-        tmp = b7 ^ b6; b7 = (tmp >> 24) | (tmp << (64 - 24)); b6 -= b7 + k6 + t1; b7 -= k7 + 9;
-        tmp = b5 ^ b4; b5 = (tmp >> 34) | (tmp << (64 - 34)); b4 -= b5 + k4; b5 -= k5 + t0;
-        tmp = b3 ^ b2; b3 = (tmp >> 30) | (tmp << (64 - 30)); b2 -= b3 + k2; b3 -= k3;
-        tmp = b1 ^ b0; b1 = (tmp >> 39) | (tmp << (64 - 39)); b0 -= b1 + k0; b1 -= k1;
-        tmp = b3 ^ b4; b3 = (tmp >> 56) | (tmp << (64 - 56)); b4 -= b3;
-        tmp = b5 ^ b2; b5 = (tmp >> 54) | (tmp << (64 - 54)); b2 -= b5;
-        tmp = b7 ^ b0; b7 = (tmp >> 9) | (tmp << (64 - 9)); b0 -= b7;
-        tmp = b1 ^ b6; b1 = (tmp >> 44) | (tmp << (64 - 44)); b6 -= b1;
-        tmp = b7 ^ b2; b7 = (tmp >> 39) | (tmp << (64 - 39)); b2 -= b7;
-        tmp = b5 ^ b0; b5 = (tmp >> 36) | (tmp << (64 - 36)); b0 -= b5;
-        tmp = b3 ^ b6; b3 = (tmp >> 49) | (tmp << (64 - 49)); b6 -= b3;
-        tmp = b1 ^ b4; b1 = (tmp >> 17) | (tmp << (64 - 17)); b4 -= b1;
-        tmp = b3 ^ b0; b3 = (tmp >> 42) | (tmp << (64 - 42)); b0 -= b3;
-        tmp = b5 ^ b6; b5 = (tmp >> 14) | (tmp << (64 - 14)); b6 -= b5;
-        tmp = b7 ^ b4; b7 = (tmp >> 27) | (tmp << (64 - 27)); b4 -= b7;
-        tmp = b1 ^ b2; b1 = (tmp >> 33) | (tmp << (64 - 33)); b2 -= b1;
-        tmp = b7 ^ b6; b7 = (tmp >> 37) | (tmp << (64 - 37)); b6 -= b7 + k5 + t0; b7 -= k6 + 8;
-        tmp = b5 ^ b4; b5 = (tmp >> 19) | (tmp << (64 - 19)); b4 -= b5 + k3; b5 -= k4 + t2;
-        tmp = b3 ^ b2; b3 = (tmp >> 36) | (tmp << (64 - 36)); b2 -= b3 + k1; b3 -= k2;
-        tmp = b1 ^ b0; b1 = (tmp >> 46) | (tmp << (64 - 46)); b0 -= b1 + k8; b1 -= k0;
-        tmp = b3 ^ b4; b3 = (tmp >> 22) | (tmp << (64 - 22)); b4 -= b3;
-        tmp = b5 ^ b2; b5 = (tmp >> 56) | (tmp << (64 - 56)); b2 -= b5;
-        tmp = b7 ^ b0; b7 = (tmp >> 35) | (tmp << (64 - 35)); b0 -= b7;
-        tmp = b1 ^ b6; b1 = (tmp >> 8) | (tmp << (64 - 8)); b6 -= b1;
-        tmp = b7 ^ b2; b7 = (tmp >> 43) | (tmp << (64 - 43)); b2 -= b7;
-        tmp = b5 ^ b0; b5 = (tmp >> 39) | (tmp << (64 - 39)); b0 -= b5;
-        tmp = b3 ^ b6; b3 = (tmp >> 29) | (tmp << (64 - 29)); b6 -= b3;
-        tmp = b1 ^ b4; b1 = (tmp >> 25) | (tmp << (64 - 25)); b4 -= b1;
-        tmp = b3 ^ b0; b3 = (tmp >> 17) | (tmp << (64 - 17)); b0 -= b3;
-        tmp = b5 ^ b6; b5 = (tmp >> 10) | (tmp << (64 - 10)); b6 -= b5;
-        tmp = b7 ^ b4; b7 = (tmp >> 50) | (tmp << (64 - 50)); b4 -= b7;
-        tmp = b1 ^ b2; b1 = (tmp >> 13) | (tmp << (64 - 13)); b2 -= b1;
-        tmp = b7 ^ b6; b7 = (tmp >> 24) | (tmp << (64 - 24)); b6 -= b7 + k4 + t2; b7 -= k5 + 7;
-        tmp = b5 ^ b4; b5 = (tmp >> 34) | (tmp << (64 - 34)); b4 -= b5 + k2; b5 -= k3 + t1;
-        tmp = b3 ^ b2; b3 = (tmp >> 30) | (tmp << (64 - 30)); b2 -= b3 + k0; b3 -= k1;
-        tmp = b1 ^ b0; b1 = (tmp >> 39) | (tmp << (64 - 39)); b0 -= b1 + k7; b1 -= k8;
-        tmp = b3 ^ b4; b3 = (tmp >> 56) | (tmp << (64 - 56)); b4 -= b3;
-        tmp = b5 ^ b2; b5 = (tmp >> 54) | (tmp << (64 - 54)); b2 -= b5;
-        tmp = b7 ^ b0; b7 = (tmp >> 9) | (tmp << (64 - 9)); b0 -= b7;
-        tmp = b1 ^ b6; b1 = (tmp >> 44) | (tmp << (64 - 44)); b6 -= b1;
-        tmp = b7 ^ b2; b7 = (tmp >> 39) | (tmp << (64 - 39)); b2 -= b7;
-        tmp = b5 ^ b0; b5 = (tmp >> 36) | (tmp << (64 - 36)); b0 -= b5;
-        tmp = b3 ^ b6; b3 = (tmp >> 49) | (tmp << (64 - 49)); b6 -= b3;
-        tmp = b1 ^ b4; b1 = (tmp >> 17) | (tmp << (64 - 17)); b4 -= b1;
-        tmp = b3 ^ b0; b3 = (tmp >> 42) | (tmp << (64 - 42)); b0 -= b3;
-        tmp = b5 ^ b6; b5 = (tmp >> 14) | (tmp << (64 - 14)); b6 -= b5;
-        tmp = b7 ^ b4; b7 = (tmp >> 27) | (tmp << (64 - 27)); b4 -= b7;
-        tmp = b1 ^ b2; b1 = (tmp >> 33) | (tmp << (64 - 33)); b2 -= b1;
-        tmp = b7 ^ b6; b7 = (tmp >> 37) | (tmp << (64 - 37)); b6 -= b7 + k3 + t1; b7 -= k4 + 6;
-        tmp = b5 ^ b4; b5 = (tmp >> 19) | (tmp << (64 - 19)); b4 -= b5 + k1; b5 -= k2 + t0;
-        tmp = b3 ^ b2; b3 = (tmp >> 36) | (tmp << (64 - 36)); b2 -= b3 + k8; b3 -= k0;
-        tmp = b1 ^ b0; b1 = (tmp >> 46) | (tmp << (64 - 46)); b0 -= b1 + k6; b1 -= k7;
-        tmp = b3 ^ b4; b3 = (tmp >> 22) | (tmp << (64 - 22)); b4 -= b3;
-        tmp = b5 ^ b2; b5 = (tmp >> 56) | (tmp << (64 - 56)); b2 -= b5;
-        tmp = b7 ^ b0; b7 = (tmp >> 35) | (tmp << (64 - 35)); b0 -= b7;
-        tmp = b1 ^ b6; b1 = (tmp >> 8) | (tmp << (64 - 8)); b6 -= b1;
-        tmp = b7 ^ b2; b7 = (tmp >> 43) | (tmp << (64 - 43)); b2 -= b7;
-        tmp = b5 ^ b0; b5 = (tmp >> 39) | (tmp << (64 - 39)); b0 -= b5;
-        tmp = b3 ^ b6; b3 = (tmp >> 29) | (tmp << (64 - 29)); b6 -= b3;
-        tmp = b1 ^ b4; b1 = (tmp >> 25) | (tmp << (64 - 25)); b4 -= b1;
-        tmp = b3 ^ b0; b3 = (tmp >> 17) | (tmp << (64 - 17)); b0 -= b3;
-        tmp = b5 ^ b6; b5 = (tmp >> 10) | (tmp << (64 - 10)); b6 -= b5;
-        tmp = b7 ^ b4; b7 = (tmp >> 50) | (tmp << (64 - 50)); b4 -= b7;
-        tmp = b1 ^ b2; b1 = (tmp >> 13) | (tmp << (64 - 13)); b2 -= b1;
-        tmp = b7 ^ b6; b7 = (tmp >> 24) | (tmp << (64 - 24)); b6 -= b7 + k2 + t0; b7 -= k3 + 5;
-        tmp = b5 ^ b4; b5 = (tmp >> 34) | (tmp << (64 - 34)); b4 -= b5 + k0; b5 -= k1 + t2;
-        tmp = b3 ^ b2; b3 = (tmp >> 30) | (tmp << (64 - 30)); b2 -= b3 + k7; b3 -= k8;
-        tmp = b1 ^ b0; b1 = (tmp >> 39) | (tmp << (64 - 39)); b0 -= b1 + k5; b1 -= k6;
-        tmp = b3 ^ b4; b3 = (tmp >> 56) | (tmp << (64 - 56)); b4 -= b3;
-        tmp = b5 ^ b2; b5 = (tmp >> 54) | (tmp << (64 - 54)); b2 -= b5;
-        tmp = b7 ^ b0; b7 = (tmp >> 9) | (tmp << (64 - 9)); b0 -= b7;
-        tmp = b1 ^ b6; b1 = (tmp >> 44) | (tmp << (64 - 44)); b6 -= b1;
-        tmp = b7 ^ b2; b7 = (tmp >> 39) | (tmp << (64 - 39)); b2 -= b7;
-        tmp = b5 ^ b0; b5 = (tmp >> 36) | (tmp << (64 - 36)); b0 -= b5;
-        tmp = b3 ^ b6; b3 = (tmp >> 49) | (tmp << (64 - 49)); b6 -= b3;
-        tmp = b1 ^ b4; b1 = (tmp >> 17) | (tmp << (64 - 17)); b4 -= b1;
-        tmp = b3 ^ b0; b3 = (tmp >> 42) | (tmp << (64 - 42)); b0 -= b3;
-        tmp = b5 ^ b6; b5 = (tmp >> 14) | (tmp << (64 - 14)); b6 -= b5;
-        tmp = b7 ^ b4; b7 = (tmp >> 27) | (tmp << (64 - 27)); b4 -= b7;
-        tmp = b1 ^ b2; b1 = (tmp >> 33) | (tmp << (64 - 33)); b2 -= b1;
-        tmp = b7 ^ b6; b7 = (tmp >> 37) | (tmp << (64 - 37)); b6 -= b7 + k1 + t2; b7 -= k2 + 4;
-        tmp = b5 ^ b4; b5 = (tmp >> 19) | (tmp << (64 - 19)); b4 -= b5 + k8; b5 -= k0 + t1;
-        tmp = b3 ^ b2; b3 = (tmp >> 36) | (tmp << (64 - 36)); b2 -= b3 + k6; b3 -= k7;
-        tmp = b1 ^ b0; b1 = (tmp >> 46) | (tmp << (64 - 46)); b0 -= b1 + k4; b1 -= k5;
-        tmp = b3 ^ b4; b3 = (tmp >> 22) | (tmp << (64 - 22)); b4 -= b3;
-        tmp = b5 ^ b2; b5 = (tmp >> 56) | (tmp << (64 - 56)); b2 -= b5;
-        tmp = b7 ^ b0; b7 = (tmp >> 35) | (tmp << (64 - 35)); b0 -= b7;
-        tmp = b1 ^ b6; b1 = (tmp >> 8) | (tmp << (64 - 8)); b6 -= b1;
-        tmp = b7 ^ b2; b7 = (tmp >> 43) | (tmp << (64 - 43)); b2 -= b7;
-        tmp = b5 ^ b0; b5 = (tmp >> 39) | (tmp << (64 - 39)); b0 -= b5;
-        tmp = b3 ^ b6; b3 = (tmp >> 29) | (tmp << (64 - 29)); b6 -= b3;
-        tmp = b1 ^ b4; b1 = (tmp >> 25) | (tmp << (64 - 25)); b4 -= b1;
-        tmp = b3 ^ b0; b3 = (tmp >> 17) | (tmp << (64 - 17)); b0 -= b3;
-        tmp = b5 ^ b6; b5 = (tmp >> 10) | (tmp << (64 - 10)); b6 -= b5;
-        tmp = b7 ^ b4; b7 = (tmp >> 50) | (tmp << (64 - 50)); b4 -= b7;
-        tmp = b1 ^ b2; b1 = (tmp >> 13) | (tmp << (64 - 13)); b2 -= b1;
-        tmp = b7 ^ b6; b7 = (tmp >> 24) | (tmp << (64 - 24)); b6 -= b7 + k0 + t1; b7 -= k1 + 3;
-        tmp = b5 ^ b4; b5 = (tmp >> 34) | (tmp << (64 - 34)); b4 -= b5 + k7; b5 -= k8 + t0;
-        tmp = b3 ^ b2; b3 = (tmp >> 30) | (tmp << (64 - 30)); b2 -= b3 + k5; b3 -= k6;
-        tmp = b1 ^ b0; b1 = (tmp >> 39) | (tmp << (64 - 39)); b0 -= b1 + k3; b1 -= k4;
-        tmp = b3 ^ b4; b3 = (tmp >> 56) | (tmp << (64 - 56)); b4 -= b3;
-        tmp = b5 ^ b2; b5 = (tmp >> 54) | (tmp << (64 - 54)); b2 -= b5;
-        tmp = b7 ^ b0; b7 = (tmp >> 9) | (tmp << (64 - 9)); b0 -= b7;
-        tmp = b1 ^ b6; b1 = (tmp >> 44) | (tmp << (64 - 44)); b6 -= b1;
-        tmp = b7 ^ b2; b7 = (tmp >> 39) | (tmp << (64 - 39)); b2 -= b7;
-        tmp = b5 ^ b0; b5 = (tmp >> 36) | (tmp << (64 - 36)); b0 -= b5;
-        tmp = b3 ^ b6; b3 = (tmp >> 49) | (tmp << (64 - 49)); b6 -= b3;
-        tmp = b1 ^ b4; b1 = (tmp >> 17) | (tmp << (64 - 17)); b4 -= b1;
-        tmp = b3 ^ b0; b3 = (tmp >> 42) | (tmp << (64 - 42)); b0 -= b3;
-        tmp = b5 ^ b6; b5 = (tmp >> 14) | (tmp << (64 - 14)); b6 -= b5;
-        tmp = b7 ^ b4; b7 = (tmp >> 27) | (tmp << (64 - 27)); b4 -= b7;
-        tmp = b1 ^ b2; b1 = (tmp >> 33) | (tmp << (64 - 33)); b2 -= b1;
-        tmp = b7 ^ b6; b7 = (tmp >> 37) | (tmp << (64 - 37)); b6 -= b7 + k8 + t0; b7 -= k0 + 2;
-        tmp = b5 ^ b4; b5 = (tmp >> 19) | (tmp << (64 - 19)); b4 -= b5 + k6; b5 -= k7 + t2;
-        tmp = b3 ^ b2; b3 = (tmp >> 36) | (tmp << (64 - 36)); b2 -= b3 + k4; b3 -= k5;
-        tmp = b1 ^ b0; b1 = (tmp >> 46) | (tmp << (64 - 46)); b0 -= b1 + k2; b1 -= k3;
-        tmp = b3 ^ b4; b3 = (tmp >> 22) | (tmp << (64 - 22)); b4 -= b3;
-        tmp = b5 ^ b2; b5 = (tmp >> 56) | (tmp << (64 - 56)); b2 -= b5;
-        tmp = b7 ^ b0; b7 = (tmp >> 35) | (tmp << (64 - 35)); b0 -= b7;
-        tmp = b1 ^ b6; b1 = (tmp >> 8) | (tmp << (64 - 8)); b6 -= b1;
-        tmp = b7 ^ b2; b7 = (tmp >> 43) | (tmp << (64 - 43)); b2 -= b7;
-        tmp = b5 ^ b0; b5 = (tmp >> 39) | (tmp << (64 - 39)); b0 -= b5;
-        tmp = b3 ^ b6; b3 = (tmp >> 29) | (tmp << (64 - 29)); b6 -= b3;
-        tmp = b1 ^ b4; b1 = (tmp >> 25) | (tmp << (64 - 25)); b4 -= b1;
-        tmp = b3 ^ b0; b3 = (tmp >> 17) | (tmp << (64 - 17)); b0 -= b3;
-        tmp = b5 ^ b6; b5 = (tmp >> 10) | (tmp << (64 - 10)); b6 -= b5;
-        tmp = b7 ^ b4; b7 = (tmp >> 50) | (tmp << (64 - 50)); b4 -= b7;
-        tmp = b1 ^ b2; b1 = (tmp >> 13) | (tmp << (64 - 13)); b2 -= b1;
-        tmp = b7 ^ b6; b7 = (tmp >> 24) | (tmp << (64 - 24)); b6 -= b7 + k7 + t2; b7 -= k8 + 1;
-        tmp = b5 ^ b4; b5 = (tmp >> 34) | (tmp << (64 - 34)); b4 -= b5 + k5; b5 -= k6 + t1;
-        tmp = b3 ^ b2; b3 = (tmp >> 30) | (tmp << (64 - 30)); b2 -= b3 + k3; b3 -= k4;
-        tmp = b1 ^ b0; b1 = (tmp >> 39) | (tmp << (64 - 39)); b0 -= b1 + k1; b1 -= k2;
-        tmp = b3 ^ b4; b3 = (tmp >> 56) | (tmp << (64 - 56)); b4 -= b3;
-        tmp = b5 ^ b2; b5 = (tmp >> 54) | (tmp << (64 - 54)); b2 -= b5;
-        tmp = b7 ^ b0; b7 = (tmp >> 9) | (tmp << (64 - 9)); b0 -= b7;
-        tmp = b1 ^ b6; b1 = (tmp >> 44) | (tmp << (64 - 44)); b6 -= b1;
-        tmp = b7 ^ b2; b7 = (tmp >> 39) | (tmp << (64 - 39)); b2 -= b7;
-        tmp = b5 ^ b0; b5 = (tmp >> 36) | (tmp << (64 - 36)); b0 -= b5;
-        tmp = b3 ^ b6; b3 = (tmp >> 49) | (tmp << (64 - 49)); b6 -= b3;
-        tmp = b1 ^ b4; b1 = (tmp >> 17) | (tmp << (64 - 17)); b4 -= b1;
-        tmp = b3 ^ b0; b3 = (tmp >> 42) | (tmp << (64 - 42)); b0 -= b3;
-        tmp = b5 ^ b6; b5 = (tmp >> 14) | (tmp << (64 - 14)); b6 -= b5;
-        tmp = b7 ^ b4; b7 = (tmp >> 27) | (tmp << (64 - 27)); b4 -= b7;
-        tmp = b1 ^ b2; b1 = (tmp >> 33) | (tmp << (64 - 33)); b2 -= b1;
-        tmp = b7 ^ b6; b7 = (tmp >> 37) | (tmp << (64 - 37)); b6 -= b7 + k6 + t1; b7 -= k7;
-        tmp = b5 ^ b4; b5 = (tmp >> 19) | (tmp << (64 - 19)); b4 -= b5 + k4; b5 -= k5 + t0;
-        tmp = b3 ^ b2; b3 = (tmp >> 36) | (tmp << (64 - 36)); b2 -= b3 + k2; b3 -= k3;
-        tmp = b1 ^ b0; b1 = (tmp >> 46) | (tmp << (64 - 46)); b0 -= b1 + k0; b1 -= k1;
+	b0 -= k0;
+	b1 -= k1;
+	b2 -= k2;
+	b3 -= k3;
+	b4 -= k4;
+	b5 -= k5 + t0;
+	b6 -= k6 + t1;
+	b7 -= k7 + 18;
+	tmp = b3 ^ b4; b3 = (tmp >> 22) | (tmp << (64 - 22)); b4 -= b3;
+	tmp = b5 ^ b2; b5 = (tmp >> 56) | (tmp << (64 - 56)); b2 -= b5;
+	tmp = b7 ^ b0; b7 = (tmp >> 35) | (tmp << (64 - 35)); b0 -= b7;
+	tmp = b1 ^ b6; b1 = (tmp >> 8) | (tmp << (64 - 8)); b6 -= b1;
+	tmp = b7 ^ b2; b7 = (tmp >> 43) | (tmp << (64 - 43)); b2 -= b7;
+	tmp = b5 ^ b0; b5 = (tmp >> 39) | (tmp << (64 - 39)); b0 -= b5;
+	tmp = b3 ^ b6; b3 = (tmp >> 29) | (tmp << (64 - 29)); b6 -= b3;
+	tmp = b1 ^ b4; b1 = (tmp >> 25) | (tmp << (64 - 25)); b4 -= b1;
+	tmp = b3 ^ b0; b3 = (tmp >> 17) | (tmp << (64 - 17)); b0 -= b3;
+	tmp = b5 ^ b6; b5 = (tmp >> 10) | (tmp << (64 - 10)); b6 -= b5;
+	tmp = b7 ^ b4; b7 = (tmp >> 50) | (tmp << (64 - 50)); b4 -= b7;
+	tmp = b1 ^ b2; b1 = (tmp >> 13) | (tmp << (64 - 13)); b2 -= b1;
+	tmp = b7 ^ b6; b7 = (tmp >> 24) | (tmp << (64 - 24)); b6 -= b7 + k5 + t0; b7 -= k6 + 17;
+	tmp = b5 ^ b4; b5 = (tmp >> 34) | (tmp << (64 - 34)); b4 -= b5 + k3; b5 -= k4 + t2;
+	tmp = b3 ^ b2; b3 = (tmp >> 30) | (tmp << (64 - 30)); b2 -= b3 + k1; b3 -= k2;
+	tmp = b1 ^ b0; b1 = (tmp >> 39) | (tmp << (64 - 39)); b0 -= b1 + k8; b1 -= k0;
+	tmp = b3 ^ b4; b3 = (tmp >> 56) | (tmp << (64 - 56)); b4 -= b3;
+	tmp = b5 ^ b2; b5 = (tmp >> 54) | (tmp << (64 - 54)); b2 -= b5;
+	tmp = b7 ^ b0; b7 = (tmp >> 9) | (tmp << (64 - 9)); b0 -= b7;
+	tmp = b1 ^ b6; b1 = (tmp >> 44) | (tmp << (64 - 44)); b6 -= b1;
+	tmp = b7 ^ b2; b7 = (tmp >> 39) | (tmp << (64 - 39)); b2 -= b7;
+	tmp = b5 ^ b0; b5 = (tmp >> 36) | (tmp << (64 - 36)); b0 -= b5;
+	tmp = b3 ^ b6; b3 = (tmp >> 49) | (tmp << (64 - 49)); b6 -= b3;
+	tmp = b1 ^ b4; b1 = (tmp >> 17) | (tmp << (64 - 17)); b4 -= b1;
+	tmp = b3 ^ b0; b3 = (tmp >> 42) | (tmp << (64 - 42)); b0 -= b3;
+	tmp = b5 ^ b6; b5 = (tmp >> 14) | (tmp << (64 - 14)); b6 -= b5;
+	tmp = b7 ^ b4; b7 = (tmp >> 27) | (tmp << (64 - 27)); b4 -= b7;
+	tmp = b1 ^ b2; b1 = (tmp >> 33) | (tmp << (64 - 33)); b2 -= b1;
+	tmp = b7 ^ b6; b7 = (tmp >> 37) | (tmp << (64 - 37)); b6 -= b7 + k4 + t2; b7 -= k5 + 16;
+	tmp = b5 ^ b4; b5 = (tmp >> 19) | (tmp << (64 - 19)); b4 -= b5 + k2; b5 -= k3 + t1;
+	tmp = b3 ^ b2; b3 = (tmp >> 36) | (tmp << (64 - 36)); b2 -= b3 + k0; b3 -= k1;
+	tmp = b1 ^ b0; b1 = (tmp >> 46) | (tmp << (64 - 46)); b0 -= b1 + k7; b1 -= k8;
+	tmp = b3 ^ b4; b3 = (tmp >> 22) | (tmp << (64 - 22)); b4 -= b3;
+	tmp = b5 ^ b2; b5 = (tmp >> 56) | (tmp << (64 - 56)); b2 -= b5;
+	tmp = b7 ^ b0; b7 = (tmp >> 35) | (tmp << (64 - 35)); b0 -= b7;
+	tmp = b1 ^ b6; b1 = (tmp >> 8) | (tmp << (64 - 8)); b6 -= b1;
+	tmp = b7 ^ b2; b7 = (tmp >> 43) | (tmp << (64 - 43)); b2 -= b7;
+	tmp = b5 ^ b0; b5 = (tmp >> 39) | (tmp << (64 - 39)); b0 -= b5;
+	tmp = b3 ^ b6; b3 = (tmp >> 29) | (tmp << (64 - 29)); b6 -= b3;
+	tmp = b1 ^ b4; b1 = (tmp >> 25) | (tmp << (64 - 25)); b4 -= b1;
+	tmp = b3 ^ b0; b3 = (tmp >> 17) | (tmp << (64 - 17)); b0 -= b3;
+	tmp = b5 ^ b6; b5 = (tmp >> 10) | (tmp << (64 - 10)); b6 -= b5;
+	tmp = b7 ^ b4; b7 = (tmp >> 50) | (tmp << (64 - 50)); b4 -= b7;
+	tmp = b1 ^ b2; b1 = (tmp >> 13) | (tmp << (64 - 13)); b2 -= b1;
+	tmp = b7 ^ b6; b7 = (tmp >> 24) | (tmp << (64 - 24)); b6 -= b7 + k3 + t1; b7 -= k4 + 15;
+	tmp = b5 ^ b4; b5 = (tmp >> 34) | (tmp << (64 - 34)); b4 -= b5 + k1; b5 -= k2 + t0;
+	tmp = b3 ^ b2; b3 = (tmp >> 30) | (tmp << (64 - 30)); b2 -= b3 + k8; b3 -= k0;
+	tmp = b1 ^ b0; b1 = (tmp >> 39) | (tmp << (64 - 39)); b0 -= b1 + k6; b1 -= k7;
+	tmp = b3 ^ b4; b3 = (tmp >> 56) | (tmp << (64 - 56)); b4 -= b3;
+	tmp = b5 ^ b2; b5 = (tmp >> 54) | (tmp << (64 - 54)); b2 -= b5;
+	tmp = b7 ^ b0; b7 = (tmp >> 9) | (tmp << (64 - 9)); b0 -= b7;
+	tmp = b1 ^ b6; b1 = (tmp >> 44) | (tmp << (64 - 44)); b6 -= b1;
+	tmp = b7 ^ b2; b7 = (tmp >> 39) | (tmp << (64 - 39)); b2 -= b7;
+	tmp = b5 ^ b0; b5 = (tmp >> 36) | (tmp << (64 - 36)); b0 -= b5;
+	tmp = b3 ^ b6; b3 = (tmp >> 49) | (tmp << (64 - 49)); b6 -= b3;
+	tmp = b1 ^ b4; b1 = (tmp >> 17) | (tmp << (64 - 17)); b4 -= b1;
+	tmp = b3 ^ b0; b3 = (tmp >> 42) | (tmp << (64 - 42)); b0 -= b3;
+	tmp = b5 ^ b6; b5 = (tmp >> 14) | (tmp << (64 - 14)); b6 -= b5;
+	tmp = b7 ^ b4; b7 = (tmp >> 27) | (tmp << (64 - 27)); b4 -= b7;
+	tmp = b1 ^ b2; b1 = (tmp >> 33) | (tmp << (64 - 33)); b2 -= b1;
+	tmp = b7 ^ b6; b7 = (tmp >> 37) | (tmp << (64 - 37)); b6 -= b7 + k2 + t0; b7 -= k3 + 14;
+	tmp = b5 ^ b4; b5 = (tmp >> 19) | (tmp << (64 - 19)); b4 -= b5 + k0; b5 -= k1 + t2;
+	tmp = b3 ^ b2; b3 = (tmp >> 36) | (tmp << (64 - 36)); b2 -= b3 + k7; b3 -= k8;
+	tmp = b1 ^ b0; b1 = (tmp >> 46) | (tmp << (64 - 46)); b0 -= b1 + k5; b1 -= k6;
+	tmp = b3 ^ b4; b3 = (tmp >> 22) | (tmp << (64 - 22)); b4 -= b3;
+	tmp = b5 ^ b2; b5 = (tmp >> 56) | (tmp << (64 - 56)); b2 -= b5;
+	tmp = b7 ^ b0; b7 = (tmp >> 35) | (tmp << (64 - 35)); b0 -= b7;
+	tmp = b1 ^ b6; b1 = (tmp >> 8) | (tmp << (64 - 8)); b6 -= b1;
+	tmp = b7 ^ b2; b7 = (tmp >> 43) | (tmp << (64 - 43)); b2 -= b7;
+	tmp = b5 ^ b0; b5 = (tmp >> 39) | (tmp << (64 - 39)); b0 -= b5;
+	tmp = b3 ^ b6; b3 = (tmp >> 29) | (tmp << (64 - 29)); b6 -= b3;
+	tmp = b1 ^ b4; b1 = (tmp >> 25) | (tmp << (64 - 25)); b4 -= b1;
+	tmp = b3 ^ b0; b3 = (tmp >> 17) | (tmp << (64 - 17)); b0 -= b3;
+	tmp = b5 ^ b6; b5 = (tmp >> 10) | (tmp << (64 - 10)); b6 -= b5;
+	tmp = b7 ^ b4; b7 = (tmp >> 50) | (tmp << (64 - 50)); b4 -= b7;
+	tmp = b1 ^ b2; b1 = (tmp >> 13) | (tmp << (64 - 13)); b2 -= b1;
+	tmp = b7 ^ b6; b7 = (tmp >> 24) | (tmp << (64 - 24)); b6 -= b7 + k1 + t2; b7 -= k2 + 13;
+	tmp = b5 ^ b4; b5 = (tmp >> 34) | (tmp << (64 - 34)); b4 -= b5 + k8; b5 -= k0 + t1;
+	tmp = b3 ^ b2; b3 = (tmp >> 30) | (tmp << (64 - 30)); b2 -= b3 + k6; b3 -= k7;
+	tmp = b1 ^ b0; b1 = (tmp >> 39) | (tmp << (64 - 39)); b0 -= b1 + k4; b1 -= k5;
+	tmp = b3 ^ b4; b3 = (tmp >> 56) | (tmp << (64 - 56)); b4 -= b3;
+	tmp = b5 ^ b2; b5 = (tmp >> 54) | (tmp << (64 - 54)); b2 -= b5;
+	tmp = b7 ^ b0; b7 = (tmp >> 9) | (tmp << (64 - 9)); b0 -= b7;
+	tmp = b1 ^ b6; b1 = (tmp >> 44) | (tmp << (64 - 44)); b6 -= b1;
+	tmp = b7 ^ b2; b7 = (tmp >> 39) | (tmp << (64 - 39)); b2 -= b7;
+	tmp = b5 ^ b0; b5 = (tmp >> 36) | (tmp << (64 - 36)); b0 -= b5;
+	tmp = b3 ^ b6; b3 = (tmp >> 49) | (tmp << (64 - 49)); b6 -= b3;
+	tmp = b1 ^ b4; b1 = (tmp >> 17) | (tmp << (64 - 17)); b4 -= b1;
+	tmp = b3 ^ b0; b3 = (tmp >> 42) | (tmp << (64 - 42)); b0 -= b3;
+	tmp = b5 ^ b6; b5 = (tmp >> 14) | (tmp << (64 - 14)); b6 -= b5;
+	tmp = b7 ^ b4; b7 = (tmp >> 27) | (tmp << (64 - 27)); b4 -= b7;
+	tmp = b1 ^ b2; b1 = (tmp >> 33) | (tmp << (64 - 33)); b2 -= b1;
+	tmp = b7 ^ b6; b7 = (tmp >> 37) | (tmp << (64 - 37)); b6 -= b7 + k0 + t1; b7 -= k1 + 12;
+	tmp = b5 ^ b4; b5 = (tmp >> 19) | (tmp << (64 - 19)); b4 -= b5 + k7; b5 -= k8 + t0;
+	tmp = b3 ^ b2; b3 = (tmp >> 36) | (tmp << (64 - 36)); b2 -= b3 + k5; b3 -= k6;
+	tmp = b1 ^ b0; b1 = (tmp >> 46) | (tmp << (64 - 46)); b0 -= b1 + k3; b1 -= k4;
+	tmp = b3 ^ b4; b3 = (tmp >> 22) | (tmp << (64 - 22)); b4 -= b3;
+	tmp = b5 ^ b2; b5 = (tmp >> 56) | (tmp << (64 - 56)); b2 -= b5;
+	tmp = b7 ^ b0; b7 = (tmp >> 35) | (tmp << (64 - 35)); b0 -= b7;
+	tmp = b1 ^ b6; b1 = (tmp >> 8) | (tmp << (64 - 8)); b6 -= b1;
+	tmp = b7 ^ b2; b7 = (tmp >> 43) | (tmp << (64 - 43)); b2 -= b7;
+	tmp = b5 ^ b0; b5 = (tmp >> 39) | (tmp << (64 - 39)); b0 -= b5;
+	tmp = b3 ^ b6; b3 = (tmp >> 29) | (tmp << (64 - 29)); b6 -= b3;
+	tmp = b1 ^ b4; b1 = (tmp >> 25) | (tmp << (64 - 25)); b4 -= b1;
+	tmp = b3 ^ b0; b3 = (tmp >> 17) | (tmp << (64 - 17)); b0 -= b3;
+	tmp = b5 ^ b6; b5 = (tmp >> 10) | (tmp << (64 - 10)); b6 -= b5;
+	tmp = b7 ^ b4; b7 = (tmp >> 50) | (tmp << (64 - 50)); b4 -= b7;
+	tmp = b1 ^ b2; b1 = (tmp >> 13) | (tmp << (64 - 13)); b2 -= b1;
+	tmp = b7 ^ b6; b7 = (tmp >> 24) | (tmp << (64 - 24)); b6 -= b7 + k8 + t0; b7 -= k0 + 11;
+	tmp = b5 ^ b4; b5 = (tmp >> 34) | (tmp << (64 - 34)); b4 -= b5 + k6; b5 -= k7 + t2;
+	tmp = b3 ^ b2; b3 = (tmp >> 30) | (tmp << (64 - 30)); b2 -= b3 + k4; b3 -= k5;
+	tmp = b1 ^ b0; b1 = (tmp >> 39) | (tmp << (64 - 39)); b0 -= b1 + k2; b1 -= k3;
+	tmp = b3 ^ b4; b3 = (tmp >> 56) | (tmp << (64 - 56)); b4 -= b3;
+	tmp = b5 ^ b2; b5 = (tmp >> 54) | (tmp << (64 - 54)); b2 -= b5;
+	tmp = b7 ^ b0; b7 = (tmp >> 9) | (tmp << (64 - 9)); b0 -= b7;
+	tmp = b1 ^ b6; b1 = (tmp >> 44) | (tmp << (64 - 44)); b6 -= b1;
+	tmp = b7 ^ b2; b7 = (tmp >> 39) | (tmp << (64 - 39)); b2 -= b7;
+	tmp = b5 ^ b0; b5 = (tmp >> 36) | (tmp << (64 - 36)); b0 -= b5;
+	tmp = b3 ^ b6; b3 = (tmp >> 49) | (tmp << (64 - 49)); b6 -= b3;
+	tmp = b1 ^ b4; b1 = (tmp >> 17) | (tmp << (64 - 17)); b4 -= b1;
+	tmp = b3 ^ b0; b3 = (tmp >> 42) | (tmp << (64 - 42)); b0 -= b3;
+	tmp = b5 ^ b6; b5 = (tmp >> 14) | (tmp << (64 - 14)); b6 -= b5;
+	tmp = b7 ^ b4; b7 = (tmp >> 27) | (tmp << (64 - 27)); b4 -= b7;
+	tmp = b1 ^ b2; b1 = (tmp >> 33) | (tmp << (64 - 33)); b2 -= b1;
+	tmp = b7 ^ b6; b7 = (tmp >> 37) | (tmp << (64 - 37)); b6 -= b7 + k7 + t2; b7 -= k8 + 10;
+	tmp = b5 ^ b4; b5 = (tmp >> 19) | (tmp << (64 - 19)); b4 -= b5 + k5; b5 -= k6 + t1;
+	tmp = b3 ^ b2; b3 = (tmp >> 36) | (tmp << (64 - 36)); b2 -= b3 + k3; b3 -= k4;
+	tmp = b1 ^ b0; b1 = (tmp >> 46) | (tmp << (64 - 46)); b0 -= b1 + k1; b1 -= k2;
+	tmp = b3 ^ b4; b3 = (tmp >> 22) | (tmp << (64 - 22)); b4 -= b3;
+	tmp = b5 ^ b2; b5 = (tmp >> 56) | (tmp << (64 - 56)); b2 -= b5;
+	tmp = b7 ^ b0; b7 = (tmp >> 35) | (tmp << (64 - 35)); b0 -= b7;
+	tmp = b1 ^ b6; b1 = (tmp >> 8) | (tmp << (64 - 8)); b6 -= b1;
+	tmp = b7 ^ b2; b7 = (tmp >> 43) | (tmp << (64 - 43)); b2 -= b7;
+	tmp = b5 ^ b0; b5 = (tmp >> 39) | (tmp << (64 - 39)); b0 -= b5;
+	tmp = b3 ^ b6; b3 = (tmp >> 29) | (tmp << (64 - 29)); b6 -= b3;
+	tmp = b1 ^ b4; b1 = (tmp >> 25) | (tmp << (64 - 25)); b4 -= b1;
+	tmp = b3 ^ b0; b3 = (tmp >> 17) | (tmp << (64 - 17)); b0 -= b3;
+	tmp = b5 ^ b6; b5 = (tmp >> 10) | (tmp << (64 - 10)); b6 -= b5;
+	tmp = b7 ^ b4; b7 = (tmp >> 50) | (tmp << (64 - 50)); b4 -= b7;
+	tmp = b1 ^ b2; b1 = (tmp >> 13) | (tmp << (64 - 13)); b2 -= b1;
+	tmp = b7 ^ b6; b7 = (tmp >> 24) | (tmp << (64 - 24)); b6 -= b7 + k6 + t1; b7 -= k7 + 9;
+	tmp = b5 ^ b4; b5 = (tmp >> 34) | (tmp << (64 - 34)); b4 -= b5 + k4; b5 -= k5 + t0;
+	tmp = b3 ^ b2; b3 = (tmp >> 30) | (tmp << (64 - 30)); b2 -= b3 + k2; b3 -= k3;
+	tmp = b1 ^ b0; b1 = (tmp >> 39) | (tmp << (64 - 39)); b0 -= b1 + k0; b1 -= k1;
+	tmp = b3 ^ b4; b3 = (tmp >> 56) | (tmp << (64 - 56)); b4 -= b3;
+	tmp = b5 ^ b2; b5 = (tmp >> 54) | (tmp << (64 - 54)); b2 -= b5;
+	tmp = b7 ^ b0; b7 = (tmp >> 9) | (tmp << (64 - 9)); b0 -= b7;
+	tmp = b1 ^ b6; b1 = (tmp >> 44) | (tmp << (64 - 44)); b6 -= b1;
+	tmp = b7 ^ b2; b7 = (tmp >> 39) | (tmp << (64 - 39)); b2 -= b7;
+	tmp = b5 ^ b0; b5 = (tmp >> 36) | (tmp << (64 - 36)); b0 -= b5;
+	tmp = b3 ^ b6; b3 = (tmp >> 49) | (tmp << (64 - 49)); b6 -= b3;
+	tmp = b1 ^ b4; b1 = (tmp >> 17) | (tmp << (64 - 17)); b4 -= b1;
+	tmp = b3 ^ b0; b3 = (tmp >> 42) | (tmp << (64 - 42)); b0 -= b3;
+	tmp = b5 ^ b6; b5 = (tmp >> 14) | (tmp << (64 - 14)); b6 -= b5;
+	tmp = b7 ^ b4; b7 = (tmp >> 27) | (tmp << (64 - 27)); b4 -= b7;
+	tmp = b1 ^ b2; b1 = (tmp >> 33) | (tmp << (64 - 33)); b2 -= b1;
+	tmp = b7 ^ b6; b7 = (tmp >> 37) | (tmp << (64 - 37)); b6 -= b7 + k5 + t0; b7 -= k6 + 8;
+	tmp = b5 ^ b4; b5 = (tmp >> 19) | (tmp << (64 - 19)); b4 -= b5 + k3; b5 -= k4 + t2;
+	tmp = b3 ^ b2; b3 = (tmp >> 36) | (tmp << (64 - 36)); b2 -= b3 + k1; b3 -= k2;
+	tmp = b1 ^ b0; b1 = (tmp >> 46) | (tmp << (64 - 46)); b0 -= b1 + k8; b1 -= k0;
+	tmp = b3 ^ b4; b3 = (tmp >> 22) | (tmp << (64 - 22)); b4 -= b3;
+	tmp = b5 ^ b2; b5 = (tmp >> 56) | (tmp << (64 - 56)); b2 -= b5;
+	tmp = b7 ^ b0; b7 = (tmp >> 35) | (tmp << (64 - 35)); b0 -= b7;
+	tmp = b1 ^ b6; b1 = (tmp >> 8) | (tmp << (64 - 8)); b6 -= b1;
+	tmp = b7 ^ b2; b7 = (tmp >> 43) | (tmp << (64 - 43)); b2 -= b7;
+	tmp = b5 ^ b0; b5 = (tmp >> 39) | (tmp << (64 - 39)); b0 -= b5;
+	tmp = b3 ^ b6; b3 = (tmp >> 29) | (tmp << (64 - 29)); b6 -= b3;
+	tmp = b1 ^ b4; b1 = (tmp >> 25) | (tmp << (64 - 25)); b4 -= b1;
+	tmp = b3 ^ b0; b3 = (tmp >> 17) | (tmp << (64 - 17)); b0 -= b3;
+	tmp = b5 ^ b6; b5 = (tmp >> 10) | (tmp << (64 - 10)); b6 -= b5;
+	tmp = b7 ^ b4; b7 = (tmp >> 50) | (tmp << (64 - 50)); b4 -= b7;
+	tmp = b1 ^ b2; b1 = (tmp >> 13) | (tmp << (64 - 13)); b2 -= b1;
+	tmp = b7 ^ b6; b7 = (tmp >> 24) | (tmp << (64 - 24)); b6 -= b7 + k4 + t2; b7 -= k5 + 7;
+	tmp = b5 ^ b4; b5 = (tmp >> 34) | (tmp << (64 - 34)); b4 -= b5 + k2; b5 -= k3 + t1;
+	tmp = b3 ^ b2; b3 = (tmp >> 30) | (tmp << (64 - 30)); b2 -= b3 + k0; b3 -= k1;
+	tmp = b1 ^ b0; b1 = (tmp >> 39) | (tmp << (64 - 39)); b0 -= b1 + k7; b1 -= k8;
+	tmp = b3 ^ b4; b3 = (tmp >> 56) | (tmp << (64 - 56)); b4 -= b3;
+	tmp = b5 ^ b2; b5 = (tmp >> 54) | (tmp << (64 - 54)); b2 -= b5;
+	tmp = b7 ^ b0; b7 = (tmp >> 9) | (tmp << (64 - 9)); b0 -= b7;
+	tmp = b1 ^ b6; b1 = (tmp >> 44) | (tmp << (64 - 44)); b6 -= b1;
+	tmp = b7 ^ b2; b7 = (tmp >> 39) | (tmp << (64 - 39)); b2 -= b7;
+	tmp = b5 ^ b0; b5 = (tmp >> 36) | (tmp << (64 - 36)); b0 -= b5;
+	tmp = b3 ^ b6; b3 = (tmp >> 49) | (tmp << (64 - 49)); b6 -= b3;
+	tmp = b1 ^ b4; b1 = (tmp >> 17) | (tmp << (64 - 17)); b4 -= b1;
+	tmp = b3 ^ b0; b3 = (tmp >> 42) | (tmp << (64 - 42)); b0 -= b3;
+	tmp = b5 ^ b6; b5 = (tmp >> 14) | (tmp << (64 - 14)); b6 -= b5;
+	tmp = b7 ^ b4; b7 = (tmp >> 27) | (tmp << (64 - 27)); b4 -= b7;
+	tmp = b1 ^ b2; b1 = (tmp >> 33) | (tmp << (64 - 33)); b2 -= b1;
+	tmp = b7 ^ b6; b7 = (tmp >> 37) | (tmp << (64 - 37)); b6 -= b7 + k3 + t1; b7 -= k4 + 6;
+	tmp = b5 ^ b4; b5 = (tmp >> 19) | (tmp << (64 - 19)); b4 -= b5 + k1; b5 -= k2 + t0;
+	tmp = b3 ^ b2; b3 = (tmp >> 36) | (tmp << (64 - 36)); b2 -= b3 + k8; b3 -= k0;
+	tmp = b1 ^ b0; b1 = (tmp >> 46) | (tmp << (64 - 46)); b0 -= b1 + k6; b1 -= k7;
+	tmp = b3 ^ b4; b3 = (tmp >> 22) | (tmp << (64 - 22)); b4 -= b3;
+	tmp = b5 ^ b2; b5 = (tmp >> 56) | (tmp << (64 - 56)); b2 -= b5;
+	tmp = b7 ^ b0; b7 = (tmp >> 35) | (tmp << (64 - 35)); b0 -= b7;
+	tmp = b1 ^ b6; b1 = (tmp >> 8) | (tmp << (64 - 8)); b6 -= b1;
+	tmp = b7 ^ b2; b7 = (tmp >> 43) | (tmp << (64 - 43)); b2 -= b7;
+	tmp = b5 ^ b0; b5 = (tmp >> 39) | (tmp << (64 - 39)); b0 -= b5;
+	tmp = b3 ^ b6; b3 = (tmp >> 29) | (tmp << (64 - 29)); b6 -= b3;
+	tmp = b1 ^ b4; b1 = (tmp >> 25) | (tmp << (64 - 25)); b4 -= b1;
+	tmp = b3 ^ b0; b3 = (tmp >> 17) | (tmp << (64 - 17)); b0 -= b3;
+	tmp = b5 ^ b6; b5 = (tmp >> 10) | (tmp << (64 - 10)); b6 -= b5;
+	tmp = b7 ^ b4; b7 = (tmp >> 50) | (tmp << (64 - 50)); b4 -= b7;
+	tmp = b1 ^ b2; b1 = (tmp >> 13) | (tmp << (64 - 13)); b2 -= b1;
+	tmp = b7 ^ b6; b7 = (tmp >> 24) | (tmp << (64 - 24)); b6 -= b7 + k2 + t0; b7 -= k3 + 5;
+	tmp = b5 ^ b4; b5 = (tmp >> 34) | (tmp << (64 - 34)); b4 -= b5 + k0; b5 -= k1 + t2;
+	tmp = b3 ^ b2; b3 = (tmp >> 30) | (tmp << (64 - 30)); b2 -= b3 + k7; b3 -= k8;
+	tmp = b1 ^ b0; b1 = (tmp >> 39) | (tmp << (64 - 39)); b0 -= b1 + k5; b1 -= k6;
+	tmp = b3 ^ b4; b3 = (tmp >> 56) | (tmp << (64 - 56)); b4 -= b3;
+	tmp = b5 ^ b2; b5 = (tmp >> 54) | (tmp << (64 - 54)); b2 -= b5;
+	tmp = b7 ^ b0; b7 = (tmp >> 9) | (tmp << (64 - 9)); b0 -= b7;
+	tmp = b1 ^ b6; b1 = (tmp >> 44) | (tmp << (64 - 44)); b6 -= b1;
+	tmp = b7 ^ b2; b7 = (tmp >> 39) | (tmp << (64 - 39)); b2 -= b7;
+	tmp = b5 ^ b0; b5 = (tmp >> 36) | (tmp << (64 - 36)); b0 -= b5;
+	tmp = b3 ^ b6; b3 = (tmp >> 49) | (tmp << (64 - 49)); b6 -= b3;
+	tmp = b1 ^ b4; b1 = (tmp >> 17) | (tmp << (64 - 17)); b4 -= b1;
+	tmp = b3 ^ b0; b3 = (tmp >> 42) | (tmp << (64 - 42)); b0 -= b3;
+	tmp = b5 ^ b6; b5 = (tmp >> 14) | (tmp << (64 - 14)); b6 -= b5;
+	tmp = b7 ^ b4; b7 = (tmp >> 27) | (tmp << (64 - 27)); b4 -= b7;
+	tmp = b1 ^ b2; b1 = (tmp >> 33) | (tmp << (64 - 33)); b2 -= b1;
+	tmp = b7 ^ b6; b7 = (tmp >> 37) | (tmp << (64 - 37)); b6 -= b7 + k1 + t2; b7 -= k2 + 4;
+	tmp = b5 ^ b4; b5 = (tmp >> 19) | (tmp << (64 - 19)); b4 -= b5 + k8; b5 -= k0 + t1;
+	tmp = b3 ^ b2; b3 = (tmp >> 36) | (tmp << (64 - 36)); b2 -= b3 + k6; b3 -= k7;
+	tmp = b1 ^ b0; b1 = (tmp >> 46) | (tmp << (64 - 46)); b0 -= b1 + k4; b1 -= k5;
+	tmp = b3 ^ b4; b3 = (tmp >> 22) | (tmp << (64 - 22)); b4 -= b3;
+	tmp = b5 ^ b2; b5 = (tmp >> 56) | (tmp << (64 - 56)); b2 -= b5;
+	tmp = b7 ^ b0; b7 = (tmp >> 35) | (tmp << (64 - 35)); b0 -= b7;
+	tmp = b1 ^ b6; b1 = (tmp >> 8) | (tmp << (64 - 8)); b6 -= b1;
+	tmp = b7 ^ b2; b7 = (tmp >> 43) | (tmp << (64 - 43)); b2 -= b7;
+	tmp = b5 ^ b0; b5 = (tmp >> 39) | (tmp << (64 - 39)); b0 -= b5;
+	tmp = b3 ^ b6; b3 = (tmp >> 29) | (tmp << (64 - 29)); b6 -= b3;
+	tmp = b1 ^ b4; b1 = (tmp >> 25) | (tmp << (64 - 25)); b4 -= b1;
+	tmp = b3 ^ b0; b3 = (tmp >> 17) | (tmp << (64 - 17)); b0 -= b3;
+	tmp = b5 ^ b6; b5 = (tmp >> 10) | (tmp << (64 - 10)); b6 -= b5;
+	tmp = b7 ^ b4; b7 = (tmp >> 50) | (tmp << (64 - 50)); b4 -= b7;
+	tmp = b1 ^ b2; b1 = (tmp >> 13) | (tmp << (64 - 13)); b2 -= b1;
+	tmp = b7 ^ b6; b7 = (tmp >> 24) | (tmp << (64 - 24)); b6 -= b7 + k0 + t1; b7 -= k1 + 3;
+	tmp = b5 ^ b4; b5 = (tmp >> 34) | (tmp << (64 - 34)); b4 -= b5 + k7; b5 -= k8 + t0;
+	tmp = b3 ^ b2; b3 = (tmp >> 30) | (tmp << (64 - 30)); b2 -= b3 + k5; b3 -= k6;
+	tmp = b1 ^ b0; b1 = (tmp >> 39) | (tmp << (64 - 39)); b0 -= b1 + k3; b1 -= k4;
+	tmp = b3 ^ b4; b3 = (tmp >> 56) | (tmp << (64 - 56)); b4 -= b3;
+	tmp = b5 ^ b2; b5 = (tmp >> 54) | (tmp << (64 - 54)); b2 -= b5;
+	tmp = b7 ^ b0; b7 = (tmp >> 9) | (tmp << (64 - 9)); b0 -= b7;
+	tmp = b1 ^ b6; b1 = (tmp >> 44) | (tmp << (64 - 44)); b6 -= b1;
+	tmp = b7 ^ b2; b7 = (tmp >> 39) | (tmp << (64 - 39)); b2 -= b7;
+	tmp = b5 ^ b0; b5 = (tmp >> 36) | (tmp << (64 - 36)); b0 -= b5;
+	tmp = b3 ^ b6; b3 = (tmp >> 49) | (tmp << (64 - 49)); b6 -= b3;
+	tmp = b1 ^ b4; b1 = (tmp >> 17) | (tmp << (64 - 17)); b4 -= b1;
+	tmp = b3 ^ b0; b3 = (tmp >> 42) | (tmp << (64 - 42)); b0 -= b3;
+	tmp = b5 ^ b6; b5 = (tmp >> 14) | (tmp << (64 - 14)); b6 -= b5;
+	tmp = b7 ^ b4; b7 = (tmp >> 27) | (tmp << (64 - 27)); b4 -= b7;
+	tmp = b1 ^ b2; b1 = (tmp >> 33) | (tmp << (64 - 33)); b2 -= b1;
+	tmp = b7 ^ b6; b7 = (tmp >> 37) | (tmp << (64 - 37)); b6 -= b7 + k8 + t0; b7 -= k0 + 2;
+	tmp = b5 ^ b4; b5 = (tmp >> 19) | (tmp << (64 - 19)); b4 -= b5 + k6; b5 -= k7 + t2;
+	tmp = b3 ^ b2; b3 = (tmp >> 36) | (tmp << (64 - 36)); b2 -= b3 + k4; b3 -= k5;
+	tmp = b1 ^ b0; b1 = (tmp >> 46) | (tmp << (64 - 46)); b0 -= b1 + k2; b1 -= k3;
+	tmp = b3 ^ b4; b3 = (tmp >> 22) | (tmp << (64 - 22)); b4 -= b3;
+	tmp = b5 ^ b2; b5 = (tmp >> 56) | (tmp << (64 - 56)); b2 -= b5;
+	tmp = b7 ^ b0; b7 = (tmp >> 35) | (tmp << (64 - 35)); b0 -= b7;
+	tmp = b1 ^ b6; b1 = (tmp >> 8) | (tmp << (64 - 8)); b6 -= b1;
+	tmp = b7 ^ b2; b7 = (tmp >> 43) | (tmp << (64 - 43)); b2 -= b7;
+	tmp = b5 ^ b0; b5 = (tmp >> 39) | (tmp << (64 - 39)); b0 -= b5;
+	tmp = b3 ^ b6; b3 = (tmp >> 29) | (tmp << (64 - 29)); b6 -= b3;
+	tmp = b1 ^ b4; b1 = (tmp >> 25) | (tmp << (64 - 25)); b4 -= b1;
+	tmp = b3 ^ b0; b3 = (tmp >> 17) | (tmp << (64 - 17)); b0 -= b3;
+	tmp = b5 ^ b6; b5 = (tmp >> 10) | (tmp << (64 - 10)); b6 -= b5;
+	tmp = b7 ^ b4; b7 = (tmp >> 50) | (tmp << (64 - 50)); b4 -= b7;
+	tmp = b1 ^ b2; b1 = (tmp >> 13) | (tmp << (64 - 13)); b2 -= b1;
+	tmp = b7 ^ b6; b7 = (tmp >> 24) | (tmp << (64 - 24)); b6 -= b7 + k7 + t2; b7 -= k8 + 1;
+	tmp = b5 ^ b4; b5 = (tmp >> 34) | (tmp << (64 - 34)); b4 -= b5 + k5; b5 -= k6 + t1;
+	tmp = b3 ^ b2; b3 = (tmp >> 30) | (tmp << (64 - 30)); b2 -= b3 + k3; b3 -= k4;
+	tmp = b1 ^ b0; b1 = (tmp >> 39) | (tmp << (64 - 39)); b0 -= b1 + k1; b1 -= k2;
+	tmp = b3 ^ b4; b3 = (tmp >> 56) | (tmp << (64 - 56)); b4 -= b3;
+	tmp = b5 ^ b2; b5 = (tmp >> 54) | (tmp << (64 - 54)); b2 -= b5;
+	tmp = b7 ^ b0; b7 = (tmp >> 9) | (tmp << (64 - 9)); b0 -= b7;
+	tmp = b1 ^ b6; b1 = (tmp >> 44) | (tmp << (64 - 44)); b6 -= b1;
+	tmp = b7 ^ b2; b7 = (tmp >> 39) | (tmp << (64 - 39)); b2 -= b7;
+	tmp = b5 ^ b0; b5 = (tmp >> 36) | (tmp << (64 - 36)); b0 -= b5;
+	tmp = b3 ^ b6; b3 = (tmp >> 49) | (tmp << (64 - 49)); b6 -= b3;
+	tmp = b1 ^ b4; b1 = (tmp >> 17) | (tmp << (64 - 17)); b4 -= b1;
+	tmp = b3 ^ b0; b3 = (tmp >> 42) | (tmp << (64 - 42)); b0 -= b3;
+	tmp = b5 ^ b6; b5 = (tmp >> 14) | (tmp << (64 - 14)); b6 -= b5;
+	tmp = b7 ^ b4; b7 = (tmp >> 27) | (tmp << (64 - 27)); b4 -= b7;
+	tmp = b1 ^ b2; b1 = (tmp >> 33) | (tmp << (64 - 33)); b2 -= b1;
+	tmp = b7 ^ b6; b7 = (tmp >> 37) | (tmp << (64 - 37)); b6 -= b7 + k6 + t1; b7 -= k7;
+	tmp = b5 ^ b4; b5 = (tmp >> 19) | (tmp << (64 - 19)); b4 -= b5 + k4; b5 -= k5 + t0;
+	tmp = b3 ^ b2; b3 = (tmp >> 36) | (tmp << (64 - 36)); b2 -= b3 + k2; b3 -= k3;
+	tmp = b1 ^ b0; b1 = (tmp >> 46) | (tmp << (64 - 46)); b0 -= b1 + k0; b1 -= k1;
 
-    output[0] = b0;
-    output[1] = b1;
-    output[2] = b2;
-    output[3] = b3;
+	output[0] = b0;
+	output[1] = b1;
+	output[2] = b2;
+	output[3] = b3;
 
-        output[7] = b7;
-        output[6] = b6;
-        output[5] = b5;
-        output[4] = b4;
+	output[7] = b7;
+	output[6] = b6;
+	output[5] = b5;
+	output[4] = b4;
 }
diff --git a/drivers/staging/skein/threefishApi.c b/drivers/staging/skein/threefishApi.c
index 53f46f6cb9ca..e8ce06a9122f 100644
--- a/drivers/staging/skein/threefishApi.c
+++ b/drivers/staging/skein/threefishApi.c
@@ -4,75 +4,75 @@
 #include <threefishApi.h>
 
 void threefishSetKey(struct threefish_key *keyCtx, enum threefish_size stateSize,
-                     u64 *keyData, u64 *tweak)
+		     u64 *keyData, u64 *tweak)
 {
-    int keyWords = stateSize / 64;
-    int i;
-    u64 parity = KeyScheduleConst;
+	int keyWords = stateSize / 64;
+	int i;
+	u64 parity = KeyScheduleConst;
 
-    keyCtx->tweak[0] = tweak[0];
-    keyCtx->tweak[1] = tweak[1];
-    keyCtx->tweak[2] = tweak[0] ^ tweak[1];
+	keyCtx->tweak[0] = tweak[0];
+	keyCtx->tweak[1] = tweak[1];
+	keyCtx->tweak[2] = tweak[0] ^ tweak[1];
 
-    for (i = 0; i < keyWords; i++) {
-        keyCtx->key[i] = keyData[i];
-        parity ^= keyData[i];
-    }
-    keyCtx->key[i] = parity;
-    keyCtx->stateSize = stateSize;
+	for (i = 0; i < keyWords; i++) {
+		keyCtx->key[i] = keyData[i];
+		parity ^= keyData[i];
+	}
+	keyCtx->key[i] = parity;
+	keyCtx->stateSize = stateSize;
 }
 
 void threefishEncryptBlockBytes(struct threefish_key *keyCtx, u8 *in,
-                                u8 *out)
+				u8 *out)
 {
-    u64 plain[SKEIN_MAX_STATE_WORDS];        /* max number of words*/
-    u64 cipher[SKEIN_MAX_STATE_WORDS];
-    
-    Skein_Get64_LSB_First(plain, in, keyCtx->stateSize / 64);   /* bytes to words */
-    threefishEncryptBlockWords(keyCtx, plain, cipher);
-    Skein_Put64_LSB_First(out, cipher, keyCtx->stateSize / 8);  /* words to bytes */
+	u64 plain[SKEIN_MAX_STATE_WORDS];        /* max number of words*/
+	u64 cipher[SKEIN_MAX_STATE_WORDS];
+
+	Skein_Get64_LSB_First(plain, in, keyCtx->stateSize / 64);   /* bytes to words */
+	threefishEncryptBlockWords(keyCtx, plain, cipher);
+	Skein_Put64_LSB_First(out, cipher, keyCtx->stateSize / 8);  /* words to bytes */
 }
 
 void threefishEncryptBlockWords(struct threefish_key *keyCtx, u64 *in,
-                                u64 *out)
+				u64 *out)
 {
-    switch (keyCtx->stateSize) {
-        case Threefish256:
-            threefishEncrypt256(keyCtx, in, out);
-            break;
-        case Threefish512:
-            threefishEncrypt512(keyCtx, in, out);
-            break;
-        case Threefish1024:
-            threefishEncrypt1024(keyCtx, in, out);
-            break;
-    }
+	switch (keyCtx->stateSize) {
+	case Threefish256:
+		threefishEncrypt256(keyCtx, in, out);
+		break;
+	case Threefish512:
+		threefishEncrypt512(keyCtx, in, out);
+		break;
+	case Threefish1024:
+		threefishEncrypt1024(keyCtx, in, out);
+		break;
+	}
 }
 
 void threefishDecryptBlockBytes(struct threefish_key *keyCtx, u8 *in,
-                                u8 *out)
+				u8 *out)
 {
-    u64 plain[SKEIN_MAX_STATE_WORDS];        /* max number of words*/
-    u64 cipher[SKEIN_MAX_STATE_WORDS];
-    
-    Skein_Get64_LSB_First(cipher, in, keyCtx->stateSize / 64);  /* bytes to words */
-    threefishDecryptBlockWords(keyCtx, cipher, plain);
-    Skein_Put64_LSB_First(out, plain, keyCtx->stateSize / 8);   /* words to bytes */
+	u64 plain[SKEIN_MAX_STATE_WORDS];        /* max number of words*/
+	u64 cipher[SKEIN_MAX_STATE_WORDS];
+
+	Skein_Get64_LSB_First(cipher, in, keyCtx->stateSize / 64);  /* bytes to words */
+	threefishDecryptBlockWords(keyCtx, cipher, plain);
+	Skein_Put64_LSB_First(out, plain, keyCtx->stateSize / 8);   /* words to bytes */
 }
 
 void threefishDecryptBlockWords(struct threefish_key *keyCtx, u64 *in,
-                                u64 *out)
+				u64 *out)
 {
-    switch (keyCtx->stateSize) {
-        case Threefish256:
-            threefishDecrypt256(keyCtx, in, out);
-            break;
-        case Threefish512:
-            threefishDecrypt512(keyCtx, in, out);
-            break;
-        case Threefish1024:
-            threefishDecrypt1024(keyCtx, in, out);
-            break;
-    }
+	switch (keyCtx->stateSize) {
+	case Threefish256:
+		threefishDecrypt256(keyCtx, in, out);
+		break;
+	case Threefish512:
+		threefishDecrypt512(keyCtx, in, out);
+		break;
+	case Threefish1024:
+		threefishDecrypt1024(keyCtx, in, out);
+		break;
+	}
 }
 
-- 
1.9.0



More information about the devel mailing list