]> git.dujemihanovic.xyz Git - u-boot.git/commitdiff
lib: Add xxhash support
authorMarek Behún <marek.behun@nic.cz>
Mon, 29 Apr 2019 20:40:43 +0000 (22:40 +0200)
committerTom Rini <trini@konsulko.com>
Sun, 5 May 2019 12:48:50 +0000 (08:48 -0400)
This adds the xxhash support from Linux. Files are almost identical to
those added to Linux in commit 5d240522 ("lib: Add xxhash module") (they
haven't been touched since in Linux). The only difference is to add some
includes to be compatible with U-Boot. Also SPDX lincese tags were
added.

Signed-off-by: Marek Behún <marek.behun@nic.cz>
include/linux/xxhash.h [new file with mode: 0644]
lib/Kconfig
lib/Makefile
lib/xxhash.c [new file with mode: 0644]

diff --git a/include/linux/xxhash.h b/include/linux/xxhash.h
new file mode 100644 (file)
index 0000000..85feb67
--- /dev/null
@@ -0,0 +1,229 @@
+/* SPDX-License-Identifier: (GPL-2.0 or BSD-2-Clause) */
+/*
+ * xxHash - Extremely Fast Hash algorithm
+ * Copyright (C) 2012-2016, Yann Collet.
+ *
+ * You can contact the author at:
+ * - xxHash homepage: http://cyan4973.github.io/xxHash/
+ * - xxHash source repository: https://github.com/Cyan4973/xxHash
+ */
+
+/*
+ * Notice extracted from xxHash homepage:
+ *
+ * xxHash is an extremely fast Hash algorithm, running at RAM speed limits.
+ * It also successfully passes all tests from the SMHasher suite.
+ *
+ * Comparison (single thread, Windows Seven 32 bits, using SMHasher on a Core 2
+ * Duo @3GHz)
+ *
+ * Name            Speed       Q.Score   Author
+ * xxHash          5.4 GB/s     10
+ * CrapWow         3.2 GB/s      2       Andrew
+ * MumurHash 3a    2.7 GB/s     10       Austin Appleby
+ * SpookyHash      2.0 GB/s     10       Bob Jenkins
+ * SBox            1.4 GB/s      9       Bret Mulvey
+ * Lookup3         1.2 GB/s      9       Bob Jenkins
+ * SuperFastHash   1.2 GB/s      1       Paul Hsieh
+ * CityHash64      1.05 GB/s    10       Pike & Alakuijala
+ * FNV             0.55 GB/s     5       Fowler, Noll, Vo
+ * CRC32           0.43 GB/s     9
+ * MD5-32          0.33 GB/s    10       Ronald L. Rivest
+ * SHA1-32         0.28 GB/s    10
+ *
+ * Q.Score is a measure of quality of the hash function.
+ * It depends on successfully passing SMHasher test set.
+ * 10 is a perfect score.
+ *
+ * A 64-bits version, named xxh64 offers much better speed,
+ * but for 64-bits applications only.
+ * Name     Speed on 64 bits    Speed on 32 bits
+ * xxh64       13.8 GB/s            1.9 GB/s
+ * xxh32        6.8 GB/s            6.0 GB/s
+ */
+
+#ifndef XXHASH_H
+#define XXHASH_H
+
+#include <linux/types.h>
+
+/*-****************************
+ * Simple Hash Functions
+ *****************************/
+
+/**
+ * xxh32() - calculate the 32-bit hash of the input with a given seed.
+ *
+ * @input:  The data to hash.
+ * @length: The length of the data to hash.
+ * @seed:   The seed can be used to alter the result predictably.
+ *
+ * Speed on Core 2 Duo @ 3 GHz (single thread, SMHasher benchmark) : 5.4 GB/s
+ *
+ * Return:  The 32-bit hash of the data.
+ */
+uint32_t xxh32(const void *input, size_t length, uint32_t seed);
+
+/**
+ * xxh64() - calculate the 64-bit hash of the input with a given seed.
+ *
+ * @input:  The data to hash.
+ * @length: The length of the data to hash.
+ * @seed:   The seed can be used to alter the result predictably.
+ *
+ * This function runs 2x faster on 64-bit systems, but slower on 32-bit systems.
+ *
+ * Return:  The 64-bit hash of the data.
+ */
+uint64_t xxh64(const void *input, size_t length, uint64_t seed);
+
+/**
+ * xxhash() - calculate wordsize hash of the input with a given seed
+ * @input:  The data to hash.
+ * @length: The length of the data to hash.
+ * @seed:   The seed can be used to alter the result predictably.
+ *
+ * If the hash does not need to be comparable between machines with
+ * different word sizes, this function will call whichever of xxh32()
+ * or xxh64() is faster.
+ *
+ * Return:  wordsize hash of the data.
+ */
+
+static inline unsigned long xxhash(const void *input, size_t length,
+                                  uint64_t seed)
+{
+#if BITS_PER_LONG == 64
+       return xxh64(input, length, seed);
+#else
+       return xxh32(input, length, seed);
+#endif
+}
+
+/*-****************************
+ * Streaming Hash Functions
+ *****************************/
+
+/*
+ * These definitions are only meant to allow allocation of XXH state
+ * statically, on stack, or in a struct for example.
+ * Do not use members directly.
+ */
+
+/**
+ * struct xxh32_state - private xxh32 state, do not use members directly
+ */
+struct xxh32_state {
+       uint32_t total_len_32;
+       uint32_t large_len;
+       uint32_t v1;
+       uint32_t v2;
+       uint32_t v3;
+       uint32_t v4;
+       uint32_t mem32[4];
+       uint32_t memsize;
+};
+
+/**
+ * struct xxh32_state - private xxh64 state, do not use members directly
+ */
+struct xxh64_state {
+       uint64_t total_len;
+       uint64_t v1;
+       uint64_t v2;
+       uint64_t v3;
+       uint64_t v4;
+       uint64_t mem64[4];
+       uint32_t memsize;
+};
+
+/**
+ * xxh32_reset() - reset the xxh32 state to start a new hashing operation
+ *
+ * @state: The xxh32 state to reset.
+ * @seed:  Initialize the hash state with this seed.
+ *
+ * Call this function on any xxh32_state to prepare for a new hashing operation.
+ */
+void xxh32_reset(struct xxh32_state *state, uint32_t seed);
+
+/**
+ * xxh32_update() - hash the data given and update the xxh32 state
+ *
+ * @state:  The xxh32 state to update.
+ * @input:  The data to hash.
+ * @length: The length of the data to hash.
+ *
+ * After calling xxh32_reset() call xxh32_update() as many times as necessary.
+ *
+ * Return:  Zero on success, otherwise an error code.
+ */
+int xxh32_update(struct xxh32_state *state, const void *input, size_t length);
+
+/**
+ * xxh32_digest() - produce the current xxh32 hash
+ *
+ * @state: Produce the current xxh32 hash of this state.
+ *
+ * A hash value can be produced at any time. It is still possible to continue
+ * inserting input into the hash state after a call to xxh32_digest(), and
+ * generate new hashes later on, by calling xxh32_digest() again.
+ *
+ * Return: The xxh32 hash stored in the state.
+ */
+uint32_t xxh32_digest(const struct xxh32_state *state);
+
+/**
+ * xxh64_reset() - reset the xxh64 state to start a new hashing operation
+ *
+ * @state: The xxh64 state to reset.
+ * @seed:  Initialize the hash state with this seed.
+ */
+void xxh64_reset(struct xxh64_state *state, uint64_t seed);
+
+/**
+ * xxh64_update() - hash the data given and update the xxh64 state
+ * @state:  The xxh64 state to update.
+ * @input:  The data to hash.
+ * @length: The length of the data to hash.
+ *
+ * After calling xxh64_reset() call xxh64_update() as many times as necessary.
+ *
+ * Return:  Zero on success, otherwise an error code.
+ */
+int xxh64_update(struct xxh64_state *state, const void *input, size_t length);
+
+/**
+ * xxh64_digest() - produce the current xxh64 hash
+ *
+ * @state: Produce the current xxh64 hash of this state.
+ *
+ * A hash value can be produced at any time. It is still possible to continue
+ * inserting input into the hash state after a call to xxh64_digest(), and
+ * generate new hashes later on, by calling xxh64_digest() again.
+ *
+ * Return: The xxh64 hash stored in the state.
+ */
+uint64_t xxh64_digest(const struct xxh64_state *state);
+
+/*-**************************
+ * Utils
+ ***************************/
+
+/**
+ * xxh32_copy_state() - copy the source state into the destination state
+ *
+ * @src: The source xxh32 state.
+ * @dst: The destination xxh32 state.
+ */
+void xxh32_copy_state(struct xxh32_state *dst, const struct xxh32_state *src);
+
+/**
+ * xxh64_copy_state() - copy the source state into the destination state
+ *
+ * @src: The source xxh64 state.
+ * @dst: The destination xxh64 state.
+ */
+void xxh64_copy_state(struct xxh64_state *dst, const struct xxh64_state *src);
+
+#endif /* XXHASH_H */
index 38012506d294ccf1884aa9f633ae5915e0e261a0..1ed69fafc73622effbab07059b177cdff427990f 100644 (file)
@@ -327,6 +327,9 @@ config MD5
 config CRC32C
        bool
 
+config XXHASH
+       bool
+
 endmenu
 
 menu "Compression Support"
index 47829bfed52d43132682802933b970ee146af87d..6a52543530d810260dc8a5382b71ba371cb727bf 100644 (file)
@@ -37,6 +37,7 @@ obj-$(CONFIG_GENERATE_SMBIOS_TABLE) += smbios.o
 obj-$(CONFIG_IMAGE_SPARSE) += image-sparse.o
 obj-y += ldiv.o
 obj-$(CONFIG_MD5) += md5.o
+obj-$(CONFIG_XXHASH) += xxhash.o
 obj-y += net_utils.o
 obj-$(CONFIG_PHYSMEM) += physmem.o
 obj-y += rc4.o
diff --git a/lib/xxhash.c b/lib/xxhash.c
new file mode 100644 (file)
index 0000000..2fb4dc6
--- /dev/null
@@ -0,0 +1,467 @@
+// SPDX-License-Identifier: (GPL-2.0 or BSD-2-Clause)
+/*
+ * xxHash - Extremely Fast Hash algorithm
+ * Copyright (C) 2012-2016, Yann Collet.
+ *
+ * You can contact the author at:
+ * - xxHash homepage: http://cyan4973.github.io/xxHash/
+ * - xxHash source repository: https://github.com/Cyan4973/xxHash
+ */
+
+#include <asm/unaligned.h>
+#include <linux/errno.h>
+#include <linux/compiler.h>
+#include <linux/kernel.h>
+#include <linux/compat.h>
+#include <linux/string.h>
+#include <linux/xxhash.h>
+
+/*-*************************************
+ * Macros
+ **************************************/
+#define xxh_rotl32(x, r) ((x << r) | (x >> (32 - r)))
+#define xxh_rotl64(x, r) ((x << r) | (x >> (64 - r)))
+
+#ifdef __LITTLE_ENDIAN
+# define XXH_CPU_LITTLE_ENDIAN 1
+#else
+# define XXH_CPU_LITTLE_ENDIAN 0
+#endif
+
+/*-*************************************
+ * Constants
+ **************************************/
+static const uint32_t PRIME32_1 = 2654435761U;
+static const uint32_t PRIME32_2 = 2246822519U;
+static const uint32_t PRIME32_3 = 3266489917U;
+static const uint32_t PRIME32_4 =  668265263U;
+static const uint32_t PRIME32_5 =  374761393U;
+
+static const uint64_t PRIME64_1 = 11400714785074694791ULL;
+static const uint64_t PRIME64_2 = 14029467366897019727ULL;
+static const uint64_t PRIME64_3 =  1609587929392839161ULL;
+static const uint64_t PRIME64_4 =  9650029242287828579ULL;
+static const uint64_t PRIME64_5 =  2870177450012600261ULL;
+
+/*-**************************
+ *  Utils
+ ***************************/
+void xxh32_copy_state(struct xxh32_state *dst, const struct xxh32_state *src)
+{
+       memcpy(dst, src, sizeof(*dst));
+}
+EXPORT_SYMBOL(xxh32_copy_state);
+
+void xxh64_copy_state(struct xxh64_state *dst, const struct xxh64_state *src)
+{
+       memcpy(dst, src, sizeof(*dst));
+}
+EXPORT_SYMBOL(xxh64_copy_state);
+
+/*-***************************
+ * Simple Hash Functions
+ ****************************/
+static uint32_t xxh32_round(uint32_t seed, const uint32_t input)
+{
+       seed += input * PRIME32_2;
+       seed = xxh_rotl32(seed, 13);
+       seed *= PRIME32_1;
+       return seed;
+}
+
+uint32_t xxh32(const void *input, const size_t len, const uint32_t seed)
+{
+       const uint8_t *p = (const uint8_t *)input;
+       const uint8_t *b_end = p + len;
+       uint32_t h32;
+
+       if (len >= 16) {
+               const uint8_t *const limit = b_end - 16;
+               uint32_t v1 = seed + PRIME32_1 + PRIME32_2;
+               uint32_t v2 = seed + PRIME32_2;
+               uint32_t v3 = seed + 0;
+               uint32_t v4 = seed - PRIME32_1;
+
+               do {
+                       v1 = xxh32_round(v1, get_unaligned_le32(p));
+                       p += 4;
+                       v2 = xxh32_round(v2, get_unaligned_le32(p));
+                       p += 4;
+                       v3 = xxh32_round(v3, get_unaligned_le32(p));
+                       p += 4;
+                       v4 = xxh32_round(v4, get_unaligned_le32(p));
+                       p += 4;
+               } while (p <= limit);
+
+               h32 = xxh_rotl32(v1, 1) + xxh_rotl32(v2, 7) +
+                       xxh_rotl32(v3, 12) + xxh_rotl32(v4, 18);
+       } else {
+               h32 = seed + PRIME32_5;
+       }
+
+       h32 += (uint32_t)len;
+
+       while (p + 4 <= b_end) {
+               h32 += get_unaligned_le32(p) * PRIME32_3;
+               h32 = xxh_rotl32(h32, 17) * PRIME32_4;
+               p += 4;
+       }
+
+       while (p < b_end) {
+               h32 += (*p) * PRIME32_5;
+               h32 = xxh_rotl32(h32, 11) * PRIME32_1;
+               p++;
+       }
+
+       h32 ^= h32 >> 15;
+       h32 *= PRIME32_2;
+       h32 ^= h32 >> 13;
+       h32 *= PRIME32_3;
+       h32 ^= h32 >> 16;
+
+       return h32;
+}
+EXPORT_SYMBOL(xxh32);
+
+static uint64_t xxh64_round(uint64_t acc, const uint64_t input)
+{
+       acc += input * PRIME64_2;
+       acc = xxh_rotl64(acc, 31);
+       acc *= PRIME64_1;
+       return acc;
+}
+
+static uint64_t xxh64_merge_round(uint64_t acc, uint64_t val)
+{
+       val = xxh64_round(0, val);
+       acc ^= val;
+       acc = acc * PRIME64_1 + PRIME64_4;
+       return acc;
+}
+
+uint64_t xxh64(const void *input, const size_t len, const uint64_t seed)
+{
+       const uint8_t *p = (const uint8_t *)input;
+       const uint8_t *const b_end = p + len;
+       uint64_t h64;
+
+       if (len >= 32) {
+               const uint8_t *const limit = b_end - 32;
+               uint64_t v1 = seed + PRIME64_1 + PRIME64_2;
+               uint64_t v2 = seed + PRIME64_2;
+               uint64_t v3 = seed + 0;
+               uint64_t v4 = seed - PRIME64_1;
+
+               do {
+                       v1 = xxh64_round(v1, get_unaligned_le64(p));
+                       p += 8;
+                       v2 = xxh64_round(v2, get_unaligned_le64(p));
+                       p += 8;
+                       v3 = xxh64_round(v3, get_unaligned_le64(p));
+                       p += 8;
+                       v4 = xxh64_round(v4, get_unaligned_le64(p));
+                       p += 8;
+               } while (p <= limit);
+
+               h64 = xxh_rotl64(v1, 1) + xxh_rotl64(v2, 7) +
+                       xxh_rotl64(v3, 12) + xxh_rotl64(v4, 18);
+               h64 = xxh64_merge_round(h64, v1);
+               h64 = xxh64_merge_round(h64, v2);
+               h64 = xxh64_merge_round(h64, v3);
+               h64 = xxh64_merge_round(h64, v4);
+
+       } else {
+               h64  = seed + PRIME64_5;
+       }
+
+       h64 += (uint64_t)len;
+
+       while (p + 8 <= b_end) {
+               const uint64_t k1 = xxh64_round(0, get_unaligned_le64(p));
+
+               h64 ^= k1;
+               h64 = xxh_rotl64(h64, 27) * PRIME64_1 + PRIME64_4;
+               p += 8;
+       }
+
+       if (p + 4 <= b_end) {
+               h64 ^= (uint64_t)(get_unaligned_le32(p)) * PRIME64_1;
+               h64 = xxh_rotl64(h64, 23) * PRIME64_2 + PRIME64_3;
+               p += 4;
+       }
+
+       while (p < b_end) {
+               h64 ^= (*p) * PRIME64_5;
+               h64 = xxh_rotl64(h64, 11) * PRIME64_1;
+               p++;
+       }
+
+       h64 ^= h64 >> 33;
+       h64 *= PRIME64_2;
+       h64 ^= h64 >> 29;
+       h64 *= PRIME64_3;
+       h64 ^= h64 >> 32;
+
+       return h64;
+}
+EXPORT_SYMBOL(xxh64);
+
+/*-**************************************************
+ * Advanced Hash Functions
+ ***************************************************/
+void xxh32_reset(struct xxh32_state *statePtr, const uint32_t seed)
+{
+       /* use a local state for memcpy() to avoid strict-aliasing warnings */
+       struct xxh32_state state;
+
+       memset(&state, 0, sizeof(state));
+       state.v1 = seed + PRIME32_1 + PRIME32_2;
+       state.v2 = seed + PRIME32_2;
+       state.v3 = seed + 0;
+       state.v4 = seed - PRIME32_1;
+       memcpy(statePtr, &state, sizeof(state));
+}
+EXPORT_SYMBOL(xxh32_reset);
+
+void xxh64_reset(struct xxh64_state *statePtr, const uint64_t seed)
+{
+       /* use a local state for memcpy() to avoid strict-aliasing warnings */
+       struct xxh64_state state;
+
+       memset(&state, 0, sizeof(state));
+       state.v1 = seed + PRIME64_1 + PRIME64_2;
+       state.v2 = seed + PRIME64_2;
+       state.v3 = seed + 0;
+       state.v4 = seed - PRIME64_1;
+       memcpy(statePtr, &state, sizeof(state));
+}
+EXPORT_SYMBOL(xxh64_reset);
+
+int xxh32_update(struct xxh32_state *state, const void *input, const size_t len)
+{
+       const uint8_t *p = (const uint8_t *)input;
+       const uint8_t *const b_end = p + len;
+
+       if (input == NULL)
+               return -EINVAL;
+
+       state->total_len_32 += (uint32_t)len;
+       state->large_len |= (len >= 16) | (state->total_len_32 >= 16);
+
+       if (state->memsize + len < 16) { /* fill in tmp buffer */
+               memcpy((uint8_t *)(state->mem32) + state->memsize, input, len);
+               state->memsize += (uint32_t)len;
+               return 0;
+       }
+
+       if (state->memsize) { /* some data left from previous update */
+               const uint32_t *p32 = state->mem32;
+
+               memcpy((uint8_t *)(state->mem32) + state->memsize, input,
+                       16 - state->memsize);
+
+               state->v1 = xxh32_round(state->v1, get_unaligned_le32(p32));
+               p32++;
+               state->v2 = xxh32_round(state->v2, get_unaligned_le32(p32));
+               p32++;
+               state->v3 = xxh32_round(state->v3, get_unaligned_le32(p32));
+               p32++;
+               state->v4 = xxh32_round(state->v4, get_unaligned_le32(p32));
+               p32++;
+
+               p += 16-state->memsize;
+               state->memsize = 0;
+       }
+
+       if (p <= b_end - 16) {
+               const uint8_t *const limit = b_end - 16;
+               uint32_t v1 = state->v1;
+               uint32_t v2 = state->v2;
+               uint32_t v3 = state->v3;
+               uint32_t v4 = state->v4;
+
+               do {
+                       v1 = xxh32_round(v1, get_unaligned_le32(p));
+                       p += 4;
+                       v2 = xxh32_round(v2, get_unaligned_le32(p));
+                       p += 4;
+                       v3 = xxh32_round(v3, get_unaligned_le32(p));
+                       p += 4;
+                       v4 = xxh32_round(v4, get_unaligned_le32(p));
+                       p += 4;
+               } while (p <= limit);
+
+               state->v1 = v1;
+               state->v2 = v2;
+               state->v3 = v3;
+               state->v4 = v4;
+       }
+
+       if (p < b_end) {
+               memcpy(state->mem32, p, (size_t)(b_end-p));
+               state->memsize = (uint32_t)(b_end-p);
+       }
+
+       return 0;
+}
+EXPORT_SYMBOL(xxh32_update);
+
+uint32_t xxh32_digest(const struct xxh32_state *state)
+{
+       const uint8_t *p = (const uint8_t *)state->mem32;
+       const uint8_t *const b_end = (const uint8_t *)(state->mem32) +
+               state->memsize;
+       uint32_t h32;
+
+       if (state->large_len) {
+               h32 = xxh_rotl32(state->v1, 1) + xxh_rotl32(state->v2, 7) +
+                       xxh_rotl32(state->v3, 12) + xxh_rotl32(state->v4, 18);
+       } else {
+               h32 = state->v3 /* == seed */ + PRIME32_5;
+       }
+
+       h32 += state->total_len_32;
+
+       while (p + 4 <= b_end) {
+               h32 += get_unaligned_le32(p) * PRIME32_3;
+               h32 = xxh_rotl32(h32, 17) * PRIME32_4;
+               p += 4;
+       }
+
+       while (p < b_end) {
+               h32 += (*p) * PRIME32_5;
+               h32 = xxh_rotl32(h32, 11) * PRIME32_1;
+               p++;
+       }
+
+       h32 ^= h32 >> 15;
+       h32 *= PRIME32_2;
+       h32 ^= h32 >> 13;
+       h32 *= PRIME32_3;
+       h32 ^= h32 >> 16;
+
+       return h32;
+}
+EXPORT_SYMBOL(xxh32_digest);
+
+int xxh64_update(struct xxh64_state *state, const void *input, const size_t len)
+{
+       const uint8_t *p = (const uint8_t *)input;
+       const uint8_t *const b_end = p + len;
+
+       if (input == NULL)
+               return -EINVAL;
+
+       state->total_len += len;
+
+       if (state->memsize + len < 32) { /* fill in tmp buffer */
+               memcpy(((uint8_t *)state->mem64) + state->memsize, input, len);
+               state->memsize += (uint32_t)len;
+               return 0;
+       }
+
+       if (state->memsize) { /* tmp buffer is full */
+               uint64_t *p64 = state->mem64;
+
+               memcpy(((uint8_t *)p64) + state->memsize, input,
+                       32 - state->memsize);
+
+               state->v1 = xxh64_round(state->v1, get_unaligned_le64(p64));
+               p64++;
+               state->v2 = xxh64_round(state->v2, get_unaligned_le64(p64));
+               p64++;
+               state->v3 = xxh64_round(state->v3, get_unaligned_le64(p64));
+               p64++;
+               state->v4 = xxh64_round(state->v4, get_unaligned_le64(p64));
+
+               p += 32 - state->memsize;
+               state->memsize = 0;
+       }
+
+       if (p + 32 <= b_end) {
+               const uint8_t *const limit = b_end - 32;
+               uint64_t v1 = state->v1;
+               uint64_t v2 = state->v2;
+               uint64_t v3 = state->v3;
+               uint64_t v4 = state->v4;
+
+               do {
+                       v1 = xxh64_round(v1, get_unaligned_le64(p));
+                       p += 8;
+                       v2 = xxh64_round(v2, get_unaligned_le64(p));
+                       p += 8;
+                       v3 = xxh64_round(v3, get_unaligned_le64(p));
+                       p += 8;
+                       v4 = xxh64_round(v4, get_unaligned_le64(p));
+                       p += 8;
+               } while (p <= limit);
+
+               state->v1 = v1;
+               state->v2 = v2;
+               state->v3 = v3;
+               state->v4 = v4;
+       }
+
+       if (p < b_end) {
+               memcpy(state->mem64, p, (size_t)(b_end-p));
+               state->memsize = (uint32_t)(b_end - p);
+       }
+
+       return 0;
+}
+EXPORT_SYMBOL(xxh64_update);
+
+uint64_t xxh64_digest(const struct xxh64_state *state)
+{
+       const uint8_t *p = (const uint8_t *)state->mem64;
+       const uint8_t *const b_end = (const uint8_t *)state->mem64 +
+               state->memsize;
+       uint64_t h64;
+
+       if (state->total_len >= 32) {
+               const uint64_t v1 = state->v1;
+               const uint64_t v2 = state->v2;
+               const uint64_t v3 = state->v3;
+               const uint64_t v4 = state->v4;
+
+               h64 = xxh_rotl64(v1, 1) + xxh_rotl64(v2, 7) +
+                       xxh_rotl64(v3, 12) + xxh_rotl64(v4, 18);
+               h64 = xxh64_merge_round(h64, v1);
+               h64 = xxh64_merge_round(h64, v2);
+               h64 = xxh64_merge_round(h64, v3);
+               h64 = xxh64_merge_round(h64, v4);
+       } else {
+               h64  = state->v3 + PRIME64_5;
+       }
+
+       h64 += (uint64_t)state->total_len;
+
+       while (p + 8 <= b_end) {
+               const uint64_t k1 = xxh64_round(0, get_unaligned_le64(p));
+
+               h64 ^= k1;
+               h64 = xxh_rotl64(h64, 27) * PRIME64_1 + PRIME64_4;
+               p += 8;
+       }
+
+       if (p + 4 <= b_end) {
+               h64 ^= (uint64_t)(get_unaligned_le32(p)) * PRIME64_1;
+               h64 = xxh_rotl64(h64, 23) * PRIME64_2 + PRIME64_3;
+               p += 4;
+       }
+
+       while (p < b_end) {
+               h64 ^= (*p) * PRIME64_5;
+               h64 = xxh_rotl64(h64, 11) * PRIME64_1;
+               p++;
+       }
+
+       h64 ^= h64 >> 33;
+       h64 *= PRIME64_2;
+       h64 ^= h64 >> 29;
+       h64 *= PRIME64_3;
+       h64 ^= h64 >> 32;
+
+       return h64;
+}
+EXPORT_SYMBOL(xxh64_digest);