i965/drm: Switch from uthash to Mesa's hash table.
authorKenneth Graunke <kenneth@whitecape.org>
Thu, 23 Mar 2017 00:29:39 +0000 (17:29 -0700)
committerKenneth Graunke <kenneth@whitecape.org>
Mon, 10 Apr 2017 21:31:45 +0000 (14:31 -0700)
No performance data has been gathered about this choice.  I just don't
want that many hash tables.  Chris points out that this is not
performance critical - we should not be recreating that many handles
from scratch.  In the past we used a linear list, which became
unreasonable in stress tests that used hundreds of thousands of BOs.
In real usage, it shouldn't matter that much.

Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
Acked-by: Jason Ekstrand <jason@jlekstrand.net>
src/mesa/drivers/dri/i965/Makefile.sources
src/mesa/drivers/dri/i965/intel_bufmgr_gem.c
src/mesa/drivers/dri/i965/uthash.h [deleted file]

index 9b550dc17cc448e521076fa2e67263782228d0e0..6cab4eb5f43b50455eb1ad7e106d57de6df59bb5 100644 (file)
@@ -160,8 +160,7 @@ i965_FILES = \
        intel_tiled_memcpy.c \
        intel_tiled_memcpy.h \
        intel_upload.c \
-       libdrm_macros.h \
-       uthash.h
+       libdrm_macros.h
 
 i965_gen6_FILES = \
        genX_blorp_exec.c
index 499cd19808f9a34d9761ad60d359973d82ace5ef..98f2e13137793a6990efb31fadb633c437641311 100644 (file)
 #include "libdrm_macros.h"
 #include "main/macros.h"
 #include "util/macros.h"
+#include "util/hash_table.h"
 #include "util/list.h"
 #include "brw_bufmgr.h"
 #include "string.h"
 
 #include "i915_drm.h"
-#include "uthash.h"
 
 #ifdef HAVE_VALGRIND
 #include <valgrind.h>
@@ -139,8 +139,8 @@ typedef struct _drm_bacon_bufmgr {
 
        struct list_head managers;
 
-       drm_bacon_bo_gem *name_table;
-       drm_bacon_bo_gem *handle_table;
+       struct hash_table *name_table;
+       struct hash_table *handle_table;
 
        struct list_head vma_cache;
        int vma_count, vma_open, vma_max;
@@ -169,9 +169,6 @@ struct _drm_bacon_bo_gem {
         */
        unsigned int global_name;
 
-       UT_hash_handle handle_hh;
-       UT_hash_handle name_hh;
-
        /**
         * Index of the buffer within the validation list while preparing a
         * batchbuffer execution.
@@ -272,6 +269,25 @@ static inline drm_bacon_bo_gem *to_bo_gem(drm_bacon_bo *bo)
         return (drm_bacon_bo_gem *)bo;
 }
 
+static uint32_t
+key_hash_uint(const void *key)
+{
+       return _mesa_hash_data(key, 4);
+}
+
+static bool
+key_uint_equal(const void *a, const void *b)
+{
+       return *((unsigned *) a) == *((unsigned *) b);
+}
+
+static drm_bacon_bo_gem *
+hash_find_bo(struct hash_table *ht, unsigned int key)
+{
+       struct hash_entry *entry = _mesa_hash_table_search(ht, &key);
+       return entry ? (drm_bacon_bo_gem *) entry->data : NULL;
+}
+
 static unsigned long
 drm_bacon_gem_bo_tile_size(drm_bacon_bufmgr *bufmgr, unsigned long size,
                           uint32_t *tiling_mode)
@@ -635,9 +651,8 @@ retry:
                }
 
                bo_gem->gem_handle = create.handle;
-               HASH_ADD(handle_hh, bufmgr->handle_table,
-                        gem_handle, sizeof(bo_gem->gem_handle),
-                        bo_gem);
+               _mesa_hash_table_insert(bufmgr->handle_table,
+                                       &bo_gem->gem_handle, bo_gem);
 
                bo_gem->bo.handle = bo_gem->gem_handle;
                bo_gem->bo.bufmgr = bufmgr;
@@ -767,8 +782,7 @@ drm_bacon_bo_gem_create_from_name(drm_bacon_bufmgr *bufmgr,
         * provides a sufficiently fast match.
         */
        pthread_mutex_lock(&bufmgr->lock);
-       HASH_FIND(name_hh, bufmgr->name_table,
-                 &handle, sizeof(handle), bo_gem);
+       bo_gem = hash_find_bo(bufmgr->name_table, handle);
        if (bo_gem) {
                drm_bacon_bo_reference(&bo_gem->bo);
                goto out;
@@ -789,8 +803,7 @@ drm_bacon_bo_gem_create_from_name(drm_bacon_bufmgr *bufmgr,
          * object from the kernel before by looking through the list
          * again for a matching gem_handle
          */
-       HASH_FIND(handle_hh, bufmgr->handle_table,
-                 &open_arg.handle, sizeof(open_arg.handle), bo_gem);
+       bo_gem = hash_find_bo(bufmgr->handle_table, open_arg.handle);
        if (bo_gem) {
                drm_bacon_bo_reference(&bo_gem->bo);
                goto out;
@@ -814,10 +827,10 @@ drm_bacon_bo_gem_create_from_name(drm_bacon_bufmgr *bufmgr,
        bo_gem->global_name = handle;
        bo_gem->reusable = false;
 
-       HASH_ADD(handle_hh, bufmgr->handle_table,
-                gem_handle, sizeof(bo_gem->gem_handle), bo_gem);
-       HASH_ADD(name_hh, bufmgr->name_table,
-                global_name, sizeof(bo_gem->global_name), bo_gem);
+       _mesa_hash_table_insert(bufmgr->handle_table,
+                               &bo_gem->gem_handle, bo_gem);
+       _mesa_hash_table_insert(bufmgr->name_table,
+                               &bo_gem->global_name, bo_gem);
 
        memclear(get_tiling);
        get_tiling.handle = bo_gem->gem_handle;
@@ -849,6 +862,7 @@ drm_bacon_gem_bo_free(drm_bacon_bo *bo)
        drm_bacon_bufmgr *bufmgr = bo->bufmgr;
        drm_bacon_bo_gem *bo_gem = (drm_bacon_bo_gem *) bo;
        struct drm_gem_close close;
+       struct hash_entry *entry;
        int ret;
 
        list_del(&bo_gem->vma_list);
@@ -867,9 +881,14 @@ drm_bacon_gem_bo_free(drm_bacon_bo *bo)
                bufmgr->vma_count--;
        }
 
-       if (bo_gem->global_name)
-               HASH_DELETE(name_hh, bufmgr->name_table, bo_gem);
-       HASH_DELETE(handle_hh, bufmgr->handle_table, bo_gem);
+       if (bo_gem->global_name) {
+               entry = _mesa_hash_table_search(bufmgr->name_table,
+                                               &bo_gem->global_name);
+               _mesa_hash_table_remove(bufmgr->name_table, entry);
+       }
+       entry = _mesa_hash_table_search(bufmgr->handle_table,
+                                       &bo_gem->gem_handle);
+       _mesa_hash_table_remove(bufmgr->handle_table, entry);
 
        /* Close this object */
        memclear(close);
@@ -1523,6 +1542,9 @@ drm_bacon_bufmgr_gem_destroy(drm_bacon_bufmgr *bufmgr)
                }
        }
 
+       _mesa_hash_table_destroy(bufmgr->name_table, NULL);
+       _mesa_hash_table_destroy(bufmgr->handle_table, NULL);
+
        free(bufmgr);
 }
 
@@ -1894,8 +1916,7 @@ drm_bacon_bo_gem_create_from_prime(drm_bacon_bufmgr *bufmgr, int prime_fd, int s
         * for named buffers, we must not create two bo's pointing at the same
         * kernel object
         */
-       HASH_FIND(handle_hh, bufmgr->handle_table,
-                 &handle, sizeof(handle), bo_gem);
+       bo_gem = hash_find_bo(bufmgr->handle_table, handle);
        if (bo_gem) {
                drm_bacon_bo_reference(&bo_gem->bo);
                goto out;
@@ -1923,8 +1944,8 @@ drm_bacon_bo_gem_create_from_prime(drm_bacon_bufmgr *bufmgr, int prime_fd, int s
        bo_gem->bo.bufmgr = bufmgr;
 
        bo_gem->gem_handle = handle;
-       HASH_ADD(handle_hh, bufmgr->handle_table,
-                gem_handle, sizeof(bo_gem->gem_handle), bo_gem);
+       _mesa_hash_table_insert(bufmgr->handle_table,
+                               &bo_gem->gem_handle, bo_gem);
 
        bo_gem->name = "prime";
        bo_gem->validate_index = -1;
@@ -1988,9 +2009,8 @@ drm_bacon_bo_flink(drm_bacon_bo *bo, uint32_t *name)
                        bo_gem->global_name = flink.name;
                        bo_gem->reusable = false;
 
-                       HASH_ADD(name_hh, bufmgr->name_table,
-                                global_name, sizeof(bo_gem->global_name),
-                                bo_gem);
+                       _mesa_hash_table_insert(bufmgr->name_table,
+                                               &bo_gem->global_name, bo_gem);
                }
                pthread_mutex_unlock(&bufmgr->lock);
        }
@@ -2565,6 +2585,11 @@ drm_bacon_bufmgr_gem_init(struct gen_device_info *devinfo,
 
        list_add(&bufmgr->managers, &bufmgr_list);
 
+       bufmgr->name_table =
+               _mesa_hash_table_create(NULL, key_hash_uint, key_uint_equal);
+       bufmgr->handle_table =
+               _mesa_hash_table_create(NULL, key_hash_uint, key_uint_equal);
+
 exit:
        pthread_mutex_unlock(&bufmgr_list_mutex);
 
diff --git a/src/mesa/drivers/dri/i965/uthash.h b/src/mesa/drivers/dri/i965/uthash.h
deleted file mode 100644 (file)
index 45d1f9f..0000000
+++ /dev/null
@@ -1,1074 +0,0 @@
-/*
-Copyright (c) 2003-2016, Troy D. Hanson     http://troydhanson.github.com/uthash/
-All rights reserved.
-
-Redistribution and use in source and binary forms, with or without
-modification, are permitted provided that the following conditions are met:
-
-    * Redistributions of source code must retain the above copyright
-      notice, this list of conditions and the following disclaimer.
-
-THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS
-IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
-TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A
-PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER
-OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL,
-EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
-PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
-PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
-LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
-NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
-SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
-*/
-
-#ifndef UTHASH_H
-#define UTHASH_H
-
-#define UTHASH_VERSION 2.0.1
-
-#include <string.h>   /* memcmp,strlen */
-#include <stddef.h>   /* ptrdiff_t */
-#include <stdlib.h>   /* exit() */
-
-/* These macros use decltype or the earlier __typeof GNU extension.
-   As decltype is only available in newer compilers (VS2010 or gcc 4.3+
-   when compiling c++ source) this code uses whatever method is needed
-   or, for VS2008 where neither is available, uses casting workarounds. */
-#if defined(_MSC_VER)   /* MS compiler */
-#if _MSC_VER >= 1600 && defined(__cplusplus)  /* VS2010 or newer in C++ mode */
-#define DECLTYPE(x) (decltype(x))
-#else                   /* VS2008 or older (or VS2010 in C mode) */
-#define NO_DECLTYPE
-#define DECLTYPE(x)
-#endif
-#elif defined(__BORLANDC__) || defined(__LCC__) || defined(__WATCOMC__)
-#define NO_DECLTYPE
-#define DECLTYPE(x)
-#else                   /* GNU, Sun and other compilers */
-#define DECLTYPE(x) (__typeof(x))
-#endif
-
-#ifdef NO_DECLTYPE
-#define DECLTYPE_ASSIGN(dst,src)                                                 \
-do {                                                                             \
-  char **_da_dst = (char**)(&(dst));                                             \
-  *_da_dst = (char*)(src);                                                       \
-} while (0)
-#else
-#define DECLTYPE_ASSIGN(dst,src)                                                 \
-do {                                                                             \
-  (dst) = DECLTYPE(dst)(src);                                                    \
-} while (0)
-#endif
-
-/* a number of the hash function use uint32_t which isn't defined on Pre VS2010 */
-#if defined(_WIN32)
-#if defined(_MSC_VER) && _MSC_VER >= 1600
-#include <stdint.h>
-#elif defined(__WATCOMC__) || defined(__MINGW32__) || defined(__CYGWIN__)
-#include <stdint.h>
-#else
-typedef unsigned int uint32_t;
-typedef unsigned char uint8_t;
-#endif
-#elif defined(__GNUC__) && !defined(__VXWORKS__)
-#include <stdint.h>
-#else
-typedef unsigned int uint32_t;
-typedef unsigned char uint8_t;
-#endif
-
-#ifndef uthash_fatal
-#define uthash_fatal(msg) exit(-1)        /* fatal error (out of memory,etc) */
-#endif
-#ifndef uthash_malloc
-#define uthash_malloc(sz) malloc(sz)      /* malloc fcn                      */
-#endif
-#ifndef uthash_free
-#define uthash_free(ptr,sz) free(ptr)     /* free fcn                        */
-#endif
-#ifndef uthash_strlen
-#define uthash_strlen(s) strlen(s)
-#endif
-#ifndef uthash_memcmp
-#define uthash_memcmp(a,b,n) memcmp(a,b,n)
-#endif
-
-#ifndef uthash_noexpand_fyi
-#define uthash_noexpand_fyi(tbl)          /* can be defined to log noexpand  */
-#endif
-#ifndef uthash_expand_fyi
-#define uthash_expand_fyi(tbl)            /* can be defined to log expands   */
-#endif
-
-/* initial number of buckets */
-#define HASH_INITIAL_NUM_BUCKETS 32U     /* initial number of buckets        */
-#define HASH_INITIAL_NUM_BUCKETS_LOG2 5U /* lg2 of initial number of buckets */
-#define HASH_BKT_CAPACITY_THRESH 10U     /* expand when bucket count reaches */
-
-/* calculate the element whose hash handle address is hhp */
-#define ELMT_FROM_HH(tbl,hhp) ((void*)(((char*)(hhp)) - ((tbl)->hho)))
-/* calculate the hash handle from element address elp */
-#define HH_FROM_ELMT(tbl,elp) ((UT_hash_handle *)(((char*)(elp)) + ((tbl)->hho)))
-
-#define HASH_VALUE(keyptr,keylen,hashv)                                          \
-do {                                                                             \
-  HASH_FCN(keyptr, keylen, hashv);                                               \
-} while (0)
-
-#define HASH_FIND_BYHASHVALUE(hh,head,keyptr,keylen,hashval,out)                 \
-do {                                                                             \
-  (out) = NULL;                                                                  \
-  if (head) {                                                                    \
-    unsigned _hf_bkt;                                                            \
-    HASH_TO_BKT(hashval, (head)->hh.tbl->num_buckets, _hf_bkt);                  \
-    if (HASH_BLOOM_TEST((head)->hh.tbl, hashval) != 0) {                         \
-      HASH_FIND_IN_BKT((head)->hh.tbl, hh, (head)->hh.tbl->buckets[ _hf_bkt ], keyptr, keylen, hashval, out); \
-    }                                                                            \
-  }                                                                              \
-} while (0)
-
-#define HASH_FIND(hh,head,keyptr,keylen,out)                                     \
-do {                                                                             \
-  unsigned _hf_hashv;                                                            \
-  HASH_VALUE(keyptr, keylen, _hf_hashv);                                         \
-  HASH_FIND_BYHASHVALUE(hh, head, keyptr, keylen, _hf_hashv, out);               \
-} while (0)
-
-#ifdef HASH_BLOOM
-#define HASH_BLOOM_BITLEN (1UL << HASH_BLOOM)
-#define HASH_BLOOM_BYTELEN (HASH_BLOOM_BITLEN/8UL) + (((HASH_BLOOM_BITLEN%8UL)!=0UL) ? 1UL : 0UL)
-#define HASH_BLOOM_MAKE(tbl)                                                     \
-do {                                                                             \
-  (tbl)->bloom_nbits = HASH_BLOOM;                                               \
-  (tbl)->bloom_bv = (uint8_t*)uthash_malloc(HASH_BLOOM_BYTELEN);                 \
-  if (!((tbl)->bloom_bv))  { uthash_fatal( "out of memory"); }                   \
-  memset((tbl)->bloom_bv, 0, HASH_BLOOM_BYTELEN);                                \
-  (tbl)->bloom_sig = HASH_BLOOM_SIGNATURE;                                       \
-} while (0)
-
-#define HASH_BLOOM_FREE(tbl)                                                     \
-do {                                                                             \
-  uthash_free((tbl)->bloom_bv, HASH_BLOOM_BYTELEN);                              \
-} while (0)
-
-#define HASH_BLOOM_BITSET(bv,idx) (bv[(idx)/8U] |= (1U << ((idx)%8U)))
-#define HASH_BLOOM_BITTEST(bv,idx) (bv[(idx)/8U] & (1U << ((idx)%8U)))
-
-#define HASH_BLOOM_ADD(tbl,hashv)                                                \
-  HASH_BLOOM_BITSET((tbl)->bloom_bv, (hashv & (uint32_t)((1ULL << (tbl)->bloom_nbits) - 1U)))
-
-#define HASH_BLOOM_TEST(tbl,hashv)                                               \
-  HASH_BLOOM_BITTEST((tbl)->bloom_bv, (hashv & (uint32_t)((1ULL << (tbl)->bloom_nbits) - 1U)))
-
-#else
-#define HASH_BLOOM_MAKE(tbl)
-#define HASH_BLOOM_FREE(tbl)
-#define HASH_BLOOM_ADD(tbl,hashv)
-#define HASH_BLOOM_TEST(tbl,hashv) (1)
-#define HASH_BLOOM_BYTELEN 0U
-#endif
-
-#define HASH_MAKE_TABLE(hh,head)                                                 \
-do {                                                                             \
-  (head)->hh.tbl = (UT_hash_table*)uthash_malloc(                                \
-                  sizeof(UT_hash_table));                                        \
-  if (!((head)->hh.tbl))  { uthash_fatal( "out of memory"); }                    \
-  memset((head)->hh.tbl, 0, sizeof(UT_hash_table));                              \
-  (head)->hh.tbl->tail = &((head)->hh);                                          \
-  (head)->hh.tbl->num_buckets = HASH_INITIAL_NUM_BUCKETS;                        \
-  (head)->hh.tbl->log2_num_buckets = HASH_INITIAL_NUM_BUCKETS_LOG2;              \
-  (head)->hh.tbl->hho = (char*)(&(head)->hh) - (char*)(head);                    \
-  (head)->hh.tbl->buckets = (UT_hash_bucket*)uthash_malloc(                      \
-          HASH_INITIAL_NUM_BUCKETS*sizeof(struct UT_hash_bucket));               \
-  if (! (head)->hh.tbl->buckets) { uthash_fatal( "out of memory"); }             \
-  memset((head)->hh.tbl->buckets, 0,                                             \
-          HASH_INITIAL_NUM_BUCKETS*sizeof(struct UT_hash_bucket));               \
-  HASH_BLOOM_MAKE((head)->hh.tbl);                                               \
-  (head)->hh.tbl->signature = HASH_SIGNATURE;                                    \
-} while (0)
-
-#define HASH_REPLACE_BYHASHVALUE_INORDER(hh,head,fieldname,keylen_in,hashval,add,replaced,cmpfcn) \
-do {                                                                             \
-  (replaced) = NULL;                                                             \
-  HASH_FIND_BYHASHVALUE(hh, head, &((add)->fieldname), keylen_in, hashval, replaced); \
-  if (replaced) {                                                                \
-     HASH_DELETE(hh, head, replaced);                                            \
-  }                                                                              \
-  HASH_ADD_KEYPTR_BYHASHVALUE_INORDER(hh, head, &((add)->fieldname), keylen_in, hashval, add, cmpfcn); \
-} while (0)
-
-#define HASH_REPLACE_BYHASHVALUE(hh,head,fieldname,keylen_in,hashval,add,replaced) \
-do {                                                                             \
-  (replaced) = NULL;                                                             \
-  HASH_FIND_BYHASHVALUE(hh, head, &((add)->fieldname), keylen_in, hashval, replaced); \
-  if (replaced) {                                                                \
-     HASH_DELETE(hh, head, replaced);                                            \
-  }                                                                              \
-  HASH_ADD_KEYPTR_BYHASHVALUE(hh, head, &((add)->fieldname), keylen_in, hashval, add); \
-} while (0)
-
-#define HASH_REPLACE(hh,head,fieldname,keylen_in,add,replaced)                   \
-do {                                                                             \
-  unsigned _hr_hashv;                                                            \
-  HASH_VALUE(&((add)->fieldname), keylen_in, _hr_hashv);                         \
-  HASH_REPLACE_BYHASHVALUE(hh, head, fieldname, keylen_in, _hr_hashv, add, replaced); \
-} while (0)
-
-#define HASH_REPLACE_INORDER(hh,head,fieldname,keylen_in,add,replaced,cmpfcn)    \
-do {                                                                             \
-  unsigned _hr_hashv;                                                            \
-  HASH_VALUE(&((add)->fieldname), keylen_in, _hr_hashv);                         \
-  HASH_REPLACE_BYHASHVALUE_INORDER(hh, head, fieldname, keylen_in, _hr_hashv, add, replaced, cmpfcn); \
-} while (0)
-
-#define HASH_APPEND_LIST(hh, head, add)                                          \
-do {                                                                             \
-  (add)->hh.next = NULL;                                                         \
-  (add)->hh.prev = ELMT_FROM_HH((head)->hh.tbl, (head)->hh.tbl->tail);           \
-  (head)->hh.tbl->tail->next = (add);                                            \
-  (head)->hh.tbl->tail = &((add)->hh);                                           \
-} while (0)
-
-#define HASH_ADD_KEYPTR_BYHASHVALUE_INORDER(hh,head,keyptr,keylen_in,hashval,add,cmpfcn) \
-do {                                                                             \
-  unsigned _ha_bkt;                                                              \
-  (add)->hh.hashv = (hashval);                                                   \
-  (add)->hh.key = (char*) (keyptr);                                              \
-  (add)->hh.keylen = (unsigned) (keylen_in);                                     \
-  if (!(head)) {                                                                 \
-    (add)->hh.next = NULL;                                                       \
-    (add)->hh.prev = NULL;                                                       \
-    (head) = (add);                                                              \
-    HASH_MAKE_TABLE(hh, head);                                                   \
-  } else {                                                                       \
-    struct UT_hash_handle *_hs_iter = &(head)->hh;                               \
-    (add)->hh.tbl = (head)->hh.tbl;                                              \
-    do {                                                                         \
-      if (cmpfcn(DECLTYPE(head) ELMT_FROM_HH((head)->hh.tbl, _hs_iter), add) > 0) \
-        break;                                                                   \
-    } while ((_hs_iter = _hs_iter->next));                                       \
-    if (_hs_iter) {                                                              \
-      (add)->hh.next = _hs_iter;                                                 \
-      if (((add)->hh.prev = _hs_iter->prev)) {                                   \
-        HH_FROM_ELMT((head)->hh.tbl, _hs_iter->prev)->next = (add);              \
-      } else {                                                                   \
-        (head) = (add);                                                          \
-      }                                                                          \
-      _hs_iter->prev = (add);                                                    \
-    } else {                                                                     \
-      HASH_APPEND_LIST(hh, head, add);                                           \
-    }                                                                            \
-  }                                                                              \
-  (head)->hh.tbl->num_items++;                                                   \
-  HASH_TO_BKT(hashval, (head)->hh.tbl->num_buckets, _ha_bkt);                    \
-  HASH_ADD_TO_BKT((head)->hh.tbl->buckets[_ha_bkt], &(add)->hh);                 \
-  HASH_BLOOM_ADD((head)->hh.tbl, hashval);                                       \
-  HASH_EMIT_KEY(hh, head, keyptr, keylen_in);                                    \
-  HASH_FSCK(hh, head);                                                           \
-} while (0)
-
-#define HASH_ADD_KEYPTR_INORDER(hh,head,keyptr,keylen_in,add,cmpfcn)             \
-do {                                                                             \
-  unsigned _hs_hashv;                                                            \
-  HASH_VALUE(keyptr, keylen_in, _hs_hashv);                                      \
-  HASH_ADD_KEYPTR_BYHASHVALUE_INORDER(hh, head, keyptr, keylen_in, _hs_hashv, add, cmpfcn); \
-} while (0)
-
-#define HASH_ADD_BYHASHVALUE_INORDER(hh,head,fieldname,keylen_in,hashval,add,cmpfcn) \
-  HASH_ADD_KEYPTR_BYHASHVALUE_INORDER(hh, head, &((add)->fieldname), keylen_in, hashval, add, cmpfcn)
-
-#define HASH_ADD_INORDER(hh,head,fieldname,keylen_in,add,cmpfcn)                 \
-  HASH_ADD_KEYPTR_INORDER(hh, head, &((add)->fieldname), keylen_in, add, cmpfcn)
-
-#define HASH_ADD_KEYPTR_BYHASHVALUE(hh,head,keyptr,keylen_in,hashval,add)        \
-do {                                                                             \
-  unsigned _ha_bkt;                                                              \
-  (add)->hh.hashv = (hashval);                                                   \
-  (add)->hh.key = (char*) (keyptr);                                              \
-  (add)->hh.keylen = (unsigned) (keylen_in);                                     \
-  if (!(head)) {                                                                 \
-    (add)->hh.next = NULL;                                                       \
-    (add)->hh.prev = NULL;                                                       \
-    (head) = (add);                                                              \
-    HASH_MAKE_TABLE(hh, head);                                                   \
-  } else {                                                                       \
-    (add)->hh.tbl = (head)->hh.tbl;                                              \
-    HASH_APPEND_LIST(hh, head, add);                                             \
-  }                                                                              \
-  (head)->hh.tbl->num_items++;                                                   \
-  HASH_TO_BKT(hashval, (head)->hh.tbl->num_buckets, _ha_bkt);                    \
-  HASH_ADD_TO_BKT((head)->hh.tbl->buckets[_ha_bkt], &(add)->hh);                 \
-  HASH_BLOOM_ADD((head)->hh.tbl, hashval);                                       \
-  HASH_EMIT_KEY(hh, head, keyptr, keylen_in);                                    \
-  HASH_FSCK(hh, head);                                                           \
-} while (0)
-
-#define HASH_ADD_KEYPTR(hh,head,keyptr,keylen_in,add)                            \
-do {                                                                             \
-  unsigned _ha_hashv;                                                            \
-  HASH_VALUE(keyptr, keylen_in, _ha_hashv);                                      \
-  HASH_ADD_KEYPTR_BYHASHVALUE(hh, head, keyptr, keylen_in, _ha_hashv, add);      \
-} while (0)
-
-#define HASH_ADD_BYHASHVALUE(hh,head,fieldname,keylen_in,hashval,add)            \
-  HASH_ADD_KEYPTR_BYHASHVALUE(hh, head, &((add)->fieldname), keylen_in, hashval, add)
-
-#define HASH_ADD(hh,head,fieldname,keylen_in,add)                                \
-  HASH_ADD_KEYPTR(hh, head, &((add)->fieldname), keylen_in, add)
-
-#define HASH_TO_BKT(hashv,num_bkts,bkt)                                          \
-do {                                                                             \
-  bkt = ((hashv) & ((num_bkts) - 1U));                                           \
-} while (0)
-
-/* delete "delptr" from the hash table.
- * "the usual" patch-up process for the app-order doubly-linked-list.
- * The use of _hd_hh_del below deserves special explanation.
- * These used to be expressed using (delptr) but that led to a bug
- * if someone used the same symbol for the head and deletee, like
- *  HASH_DELETE(hh,users,users);
- * We want that to work, but by changing the head (users) below
- * we were forfeiting our ability to further refer to the deletee (users)
- * in the patch-up process. Solution: use scratch space to
- * copy the deletee pointer, then the latter references are via that
- * scratch pointer rather than through the repointed (users) symbol.
- */
-#define HASH_DELETE(hh,head,delptr)                                              \
-do {                                                                             \
-    struct UT_hash_handle *_hd_hh_del;                                           \
-    if ( ((delptr)->hh.prev == NULL) && ((delptr)->hh.next == NULL) )  {         \
-        uthash_free((head)->hh.tbl->buckets,                                     \
-                    (head)->hh.tbl->num_buckets*sizeof(struct UT_hash_bucket) ); \
-        HASH_BLOOM_FREE((head)->hh.tbl);                                         \
-        uthash_free((head)->hh.tbl, sizeof(UT_hash_table));                      \
-        head = NULL;                                                             \
-    } else {                                                                     \
-        unsigned _hd_bkt;                                                        \
-        _hd_hh_del = &((delptr)->hh);                                            \
-        if ((delptr) == ELMT_FROM_HH((head)->hh.tbl,(head)->hh.tbl->tail)) {     \
-            (head)->hh.tbl->tail =                                               \
-                (UT_hash_handle*)((ptrdiff_t)((delptr)->hh.prev) +               \
-                (head)->hh.tbl->hho);                                            \
-        }                                                                        \
-        if ((delptr)->hh.prev != NULL) {                                         \
-            ((UT_hash_handle*)((ptrdiff_t)((delptr)->hh.prev) +                  \
-                    (head)->hh.tbl->hho))->next = (delptr)->hh.next;             \
-        } else {                                                                 \
-            DECLTYPE_ASSIGN(head,(delptr)->hh.next);                             \
-        }                                                                        \
-        if (_hd_hh_del->next != NULL) {                                          \
-            ((UT_hash_handle*)((ptrdiff_t)_hd_hh_del->next +                     \
-                    (head)->hh.tbl->hho))->prev =                                \
-                    _hd_hh_del->prev;                                            \
-        }                                                                        \
-        HASH_TO_BKT( _hd_hh_del->hashv, (head)->hh.tbl->num_buckets, _hd_bkt);   \
-        HASH_DEL_IN_BKT(hh,(head)->hh.tbl->buckets[_hd_bkt], _hd_hh_del);        \
-        (head)->hh.tbl->num_items--;                                             \
-    }                                                                            \
-    HASH_FSCK(hh,head);                                                          \
-} while (0)
-
-
-/* convenience forms of HASH_FIND/HASH_ADD/HASH_DEL */
-#define HASH_FIND_STR(head,findstr,out)                                          \
-    HASH_FIND(hh,head,findstr,(unsigned)uthash_strlen(findstr),out)
-#define HASH_ADD_STR(head,strfield,add)                                          \
-    HASH_ADD(hh,head,strfield[0],(unsigned)uthash_strlen(add->strfield),add)
-#define HASH_REPLACE_STR(head,strfield,add,replaced)                             \
-    HASH_REPLACE(hh,head,strfield[0],(unsigned)uthash_strlen(add->strfield),add,replaced)
-#define HASH_FIND_INT(head,findint,out)                                          \
-    HASH_FIND(hh,head,findint,sizeof(int),out)
-#define HASH_ADD_INT(head,intfield,add)                                          \
-    HASH_ADD(hh,head,intfield,sizeof(int),add)
-#define HASH_REPLACE_INT(head,intfield,add,replaced)                             \
-    HASH_REPLACE(hh,head,intfield,sizeof(int),add,replaced)
-#define HASH_FIND_PTR(head,findptr,out)                                          \
-    HASH_FIND(hh,head,findptr,sizeof(void *),out)
-#define HASH_ADD_PTR(head,ptrfield,add)                                          \
-    HASH_ADD(hh,head,ptrfield,sizeof(void *),add)
-#define HASH_REPLACE_PTR(head,ptrfield,add,replaced)                             \
-    HASH_REPLACE(hh,head,ptrfield,sizeof(void *),add,replaced)
-#define HASH_DEL(head,delptr)                                                    \
-    HASH_DELETE(hh,head,delptr)
-
-/* HASH_FSCK checks hash integrity on every add/delete when HASH_DEBUG is defined.
- * This is for uthash developer only; it compiles away if HASH_DEBUG isn't defined.
- */
-#ifdef HASH_DEBUG
-#define HASH_OOPS(...) do { fprintf(stderr,__VA_ARGS__); exit(-1); } while (0)
-#define HASH_FSCK(hh,head)                                                       \
-do {                                                                             \
-    struct UT_hash_handle *_thh;                                                 \
-    if (head) {                                                                  \
-        unsigned _bkt_i;                                                         \
-        unsigned _count;                                                         \
-        char *_prev;                                                             \
-        _count = 0;                                                              \
-        for( _bkt_i = 0; _bkt_i < (head)->hh.tbl->num_buckets; _bkt_i++) {       \
-            unsigned _bkt_count = 0;                                             \
-            _thh = (head)->hh.tbl->buckets[_bkt_i].hh_head;                      \
-            _prev = NULL;                                                        \
-            while (_thh) {                                                       \
-               if (_prev != (char*)(_thh->hh_prev)) {                            \
-                   HASH_OOPS("invalid hh_prev %p, actual %p\n",                  \
-                    _thh->hh_prev, _prev );                                      \
-               }                                                                 \
-               _bkt_count++;                                                     \
-               _prev = (char*)(_thh);                                            \
-               _thh = _thh->hh_next;                                             \
-            }                                                                    \
-            _count += _bkt_count;                                                \
-            if ((head)->hh.tbl->buckets[_bkt_i].count !=  _bkt_count) {          \
-               HASH_OOPS("invalid bucket count %u, actual %u\n",                 \
-                (head)->hh.tbl->buckets[_bkt_i].count, _bkt_count);              \
-            }                                                                    \
-        }                                                                        \
-        if (_count != (head)->hh.tbl->num_items) {                               \
-            HASH_OOPS("invalid hh item count %u, actual %u\n",                   \
-                (head)->hh.tbl->num_items, _count );                             \
-        }                                                                        \
-        /* traverse hh in app order; check next/prev integrity, count */         \
-        _count = 0;                                                              \
-        _prev = NULL;                                                            \
-        _thh =  &(head)->hh;                                                     \
-        while (_thh) {                                                           \
-           _count++;                                                             \
-           if (_prev !=(char*)(_thh->prev)) {                                    \
-              HASH_OOPS("invalid prev %p, actual %p\n",                          \
-                    _thh->prev, _prev );                                         \
-           }                                                                     \
-           _prev = (char*)ELMT_FROM_HH((head)->hh.tbl, _thh);                    \
-           _thh = ( _thh->next ?  (UT_hash_handle*)((char*)(_thh->next) +        \
-                                  (head)->hh.tbl->hho) : NULL );                 \
-        }                                                                        \
-        if (_count != (head)->hh.tbl->num_items) {                               \
-            HASH_OOPS("invalid app item count %u, actual %u\n",                  \
-                (head)->hh.tbl->num_items, _count );                             \
-        }                                                                        \
-    }                                                                            \
-} while (0)
-#else
-#define HASH_FSCK(hh,head)
-#endif
-
-/* When compiled with -DHASH_EMIT_KEYS, length-prefixed keys are emitted to
- * the descriptor to which this macro is defined for tuning the hash function.
- * The app can #include <unistd.h> to get the prototype for write(2). */
-#ifdef HASH_EMIT_KEYS
-#define HASH_EMIT_KEY(hh,head,keyptr,fieldlen)                                   \
-do {                                                                             \
-    unsigned _klen = fieldlen;                                                   \
-    write(HASH_EMIT_KEYS, &_klen, sizeof(_klen));                                \
-    write(HASH_EMIT_KEYS, keyptr, (unsigned long)fieldlen);                      \
-} while (0)
-#else
-#define HASH_EMIT_KEY(hh,head,keyptr,fieldlen)
-#endif
-
-/* default to Jenkin's hash unless overridden e.g. DHASH_FUNCTION=HASH_SAX */
-#ifdef HASH_FUNCTION
-#define HASH_FCN HASH_FUNCTION
-#else
-#define HASH_FCN HASH_JEN
-#endif
-
-/* The Bernstein hash function, used in Perl prior to v5.6. Note (x<<5+x)=x*33. */
-#define HASH_BER(key,keylen,hashv)                                               \
-do {                                                                             \
-  unsigned _hb_keylen=(unsigned)keylen;                                          \
-  const unsigned char *_hb_key=(const unsigned char*)(key);                      \
-  (hashv) = 0;                                                                   \
-  while (_hb_keylen-- != 0U) {                                                   \
-      (hashv) = (((hashv) << 5) + (hashv)) + *_hb_key++;                         \
-  }                                                                              \
-} while (0)
-
-
-/* SAX/FNV/OAT/JEN hash functions are macro variants of those listed at
- * http://eternallyconfuzzled.com/tuts/algorithms/jsw_tut_hashing.aspx */
-#define HASH_SAX(key,keylen,hashv)                                               \
-do {                                                                             \
-  unsigned _sx_i;                                                                \
-  const unsigned char *_hs_key=(const unsigned char*)(key);                      \
-  hashv = 0;                                                                     \
-  for(_sx_i=0; _sx_i < keylen; _sx_i++) {                                        \
-      hashv ^= (hashv << 5) + (hashv >> 2) + _hs_key[_sx_i];                     \
-  }                                                                              \
-} while (0)
-/* FNV-1a variation */
-#define HASH_FNV(key,keylen,hashv)                                               \
-do {                                                                             \
-  unsigned _fn_i;                                                                \
-  const unsigned char *_hf_key=(const unsigned char*)(key);                      \
-  hashv = 2166136261U;                                                           \
-  for(_fn_i=0; _fn_i < keylen; _fn_i++) {                                        \
-      hashv = hashv ^ _hf_key[_fn_i];                                            \
-      hashv = hashv * 16777619U;                                                 \
-  }                                                                              \
-} while (0)
-
-#define HASH_OAT(key,keylen,hashv)                                               \
-do {                                                                             \
-  unsigned _ho_i;                                                                \
-  const unsigned char *_ho_key=(const unsigned char*)(key);                      \
-  hashv = 0;                                                                     \
-  for(_ho_i=0; _ho_i < keylen; _ho_i++) {                                        \
-      hashv += _ho_key[_ho_i];                                                   \
-      hashv += (hashv << 10);                                                    \
-      hashv ^= (hashv >> 6);                                                     \
-  }                                                                              \
-  hashv += (hashv << 3);                                                         \
-  hashv ^= (hashv >> 11);                                                        \
-  hashv += (hashv << 15);                                                        \
-} while (0)
-
-#define HASH_JEN_MIX(a,b,c)                                                      \
-do {                                                                             \
-  a -= b; a -= c; a ^= ( c >> 13 );                                              \
-  b -= c; b -= a; b ^= ( a << 8 );                                               \
-  c -= a; c -= b; c ^= ( b >> 13 );                                              \
-  a -= b; a -= c; a ^= ( c >> 12 );                                              \
-  b -= c; b -= a; b ^= ( a << 16 );                                              \
-  c -= a; c -= b; c ^= ( b >> 5 );                                               \
-  a -= b; a -= c; a ^= ( c >> 3 );                                               \
-  b -= c; b -= a; b ^= ( a << 10 );                                              \
-  c -= a; c -= b; c ^= ( b >> 15 );                                              \
-} while (0)
-
-#define HASH_JEN(key,keylen,hashv)                                               \
-do {                                                                             \
-  unsigned _hj_i,_hj_j,_hj_k;                                                    \
-  unsigned const char *_hj_key=(unsigned const char*)(key);                      \
-  hashv = 0xfeedbeefu;                                                           \
-  _hj_i = _hj_j = 0x9e3779b9u;                                                   \
-  _hj_k = (unsigned)(keylen);                                                    \
-  while (_hj_k >= 12U) {                                                         \
-    _hj_i +=    (_hj_key[0] + ( (unsigned)_hj_key[1] << 8 )                      \
-        + ( (unsigned)_hj_key[2] << 16 )                                         \
-        + ( (unsigned)_hj_key[3] << 24 ) );                                      \
-    _hj_j +=    (_hj_key[4] + ( (unsigned)_hj_key[5] << 8 )                      \
-        + ( (unsigned)_hj_key[6] << 16 )                                         \
-        + ( (unsigned)_hj_key[7] << 24 ) );                                      \
-    hashv += (_hj_key[8] + ( (unsigned)_hj_key[9] << 8 )                         \
-        + ( (unsigned)_hj_key[10] << 16 )                                        \
-        + ( (unsigned)_hj_key[11] << 24 ) );                                     \
-                                                                                 \
-     HASH_JEN_MIX(_hj_i, _hj_j, hashv);                                          \
-                                                                                 \
-     _hj_key += 12;                                                              \
-     _hj_k -= 12U;                                                               \
-  }                                                                              \
-  hashv += (unsigned)(keylen);                                                   \
-  switch ( _hj_k ) {                                                             \
-     case 11: hashv += ( (unsigned)_hj_key[10] << 24 ); /* FALLTHROUGH */        \
-     case 10: hashv += ( (unsigned)_hj_key[9] << 16 );  /* FALLTHROUGH */        \
-     case 9:  hashv += ( (unsigned)_hj_key[8] << 8 );   /* FALLTHROUGH */        \
-     case 8:  _hj_j += ( (unsigned)_hj_key[7] << 24 );  /* FALLTHROUGH */        \
-     case 7:  _hj_j += ( (unsigned)_hj_key[6] << 16 );  /* FALLTHROUGH */        \
-     case 6:  _hj_j += ( (unsigned)_hj_key[5] << 8 );   /* FALLTHROUGH */        \
-     case 5:  _hj_j += _hj_key[4];                      /* FALLTHROUGH */        \
-     case 4:  _hj_i += ( (unsigned)_hj_key[3] << 24 );  /* FALLTHROUGH */        \
-     case 3:  _hj_i += ( (unsigned)_hj_key[2] << 16 );  /* FALLTHROUGH */        \
-     case 2:  _hj_i += ( (unsigned)_hj_key[1] << 8 );   /* FALLTHROUGH */        \
-     case 1:  _hj_i += _hj_key[0];                                               \
-  }                                                                              \
-  HASH_JEN_MIX(_hj_i, _hj_j, hashv);                                             \
-} while (0)
-
-/* The Paul Hsieh hash function */
-#undef get16bits
-#if (defined(__GNUC__) && defined(__i386__)) || defined(__WATCOMC__)             \
-  || defined(_MSC_VER) || defined (__BORLANDC__) || defined (__TURBOC__)
-#define get16bits(d) (*((const uint16_t *) (d)))
-#endif
-
-#if !defined (get16bits)
-#define get16bits(d) ((((uint32_t)(((const uint8_t *)(d))[1])) << 8)             \
-                       +(uint32_t)(((const uint8_t *)(d))[0]) )
-#endif
-#define HASH_SFH(key,keylen,hashv)                                               \
-do {                                                                             \
-  unsigned const char *_sfh_key=(unsigned const char*)(key);                     \
-  uint32_t _sfh_tmp, _sfh_len = (uint32_t)keylen;                                \
-                                                                                 \
-  unsigned _sfh_rem = _sfh_len & 3U;                                             \
-  _sfh_len >>= 2;                                                                \
-  hashv = 0xcafebabeu;                                                           \
-                                                                                 \
-  /* Main loop */                                                                \
-  for (;_sfh_len > 0U; _sfh_len--) {                                             \
-    hashv    += get16bits (_sfh_key);                                            \
-    _sfh_tmp  = ((uint32_t)(get16bits (_sfh_key+2)) << 11) ^ hashv;              \
-    hashv     = (hashv << 16) ^ _sfh_tmp;                                        \
-    _sfh_key += 2U*sizeof (uint16_t);                                            \
-    hashv    += hashv >> 11;                                                     \
-  }                                                                              \
-                                                                                 \
-  /* Handle end cases */                                                         \
-  switch (_sfh_rem) {                                                            \
-    case 3: hashv += get16bits (_sfh_key);                                       \
-            hashv ^= hashv << 16;                                                \
-            hashv ^= (uint32_t)(_sfh_key[sizeof (uint16_t)]) << 18;              \
-            hashv += hashv >> 11;                                                \
-            break;                                                               \
-    case 2: hashv += get16bits (_sfh_key);                                       \
-            hashv ^= hashv << 11;                                                \
-            hashv += hashv >> 17;                                                \
-            break;                                                               \
-    case 1: hashv += *_sfh_key;                                                  \
-            hashv ^= hashv << 10;                                                \
-            hashv += hashv >> 1;                                                 \
-  }                                                                              \
-                                                                                 \
-    /* Force "avalanching" of final 127 bits */                                  \
-    hashv ^= hashv << 3;                                                         \
-    hashv += hashv >> 5;                                                         \
-    hashv ^= hashv << 4;                                                         \
-    hashv += hashv >> 17;                                                        \
-    hashv ^= hashv << 25;                                                        \
-    hashv += hashv >> 6;                                                         \
-} while (0)
-
-#ifdef HASH_USING_NO_STRICT_ALIASING
-/* The MurmurHash exploits some CPU's (x86,x86_64) tolerance for unaligned reads.
- * For other types of CPU's (e.g. Sparc) an unaligned read causes a bus error.
- * MurmurHash uses the faster approach only on CPU's where we know it's safe.
- *
- * Note the preprocessor built-in defines can be emitted using:
- *
- *   gcc -m64 -dM -E - < /dev/null                  (on gcc)
- *   cc -## a.c (where a.c is a simple test file)   (Sun Studio)
- */
-#if (defined(__i386__) || defined(__x86_64__)  || defined(_M_IX86))
-#define MUR_GETBLOCK(p,i) p[i]
-#else /* non intel */
-#define MUR_PLUS0_ALIGNED(p) (((unsigned long)p & 3UL) == 0UL)
-#define MUR_PLUS1_ALIGNED(p) (((unsigned long)p & 3UL) == 1UL)
-#define MUR_PLUS2_ALIGNED(p) (((unsigned long)p & 3UL) == 2UL)
-#define MUR_PLUS3_ALIGNED(p) (((unsigned long)p & 3UL) == 3UL)
-#define WP(p) ((uint32_t*)((unsigned long)(p) & ~3UL))
-#if (defined(__BIG_ENDIAN__) || defined(SPARC) || defined(__ppc__) || defined(__ppc64__))
-#define MUR_THREE_ONE(p) ((((*WP(p))&0x00ffffff) << 8) | (((*(WP(p)+1))&0xff000000) >> 24))
-#define MUR_TWO_TWO(p)   ((((*WP(p))&0x0000ffff) <<16) | (((*(WP(p)+1))&0xffff0000) >> 16))
-#define MUR_ONE_THREE(p) ((((*WP(p))&0x000000ff) <<24) | (((*(WP(p)+1))&0xffffff00) >>  8))
-#else /* assume little endian non-intel */
-#define MUR_THREE_ONE(p) ((((*WP(p))&0xffffff00) >> 8) | (((*(WP(p)+1))&0x000000ff) << 24))
-#define MUR_TWO_TWO(p)   ((((*WP(p))&0xffff0000) >>16) | (((*(WP(p)+1))&0x0000ffff) << 16))
-#define MUR_ONE_THREE(p) ((((*WP(p))&0xff000000) >>24) | (((*(WP(p)+1))&0x00ffffff) <<  8))
-#endif
-#define MUR_GETBLOCK(p,i) (MUR_PLUS0_ALIGNED(p) ? ((p)[i]) :           \
-                            (MUR_PLUS1_ALIGNED(p) ? MUR_THREE_ONE(p) : \
-                             (MUR_PLUS2_ALIGNED(p) ? MUR_TWO_TWO(p) :  \
-                                                      MUR_ONE_THREE(p))))
-#endif
-#define MUR_ROTL32(x,r) (((x) << (r)) | ((x) >> (32 - (r))))
-#define MUR_FMIX(_h) \
-do {                 \
-  _h ^= _h >> 16;    \
-  _h *= 0x85ebca6bu; \
-  _h ^= _h >> 13;    \
-  _h *= 0xc2b2ae35u; \
-  _h ^= _h >> 16;    \
-} while (0)
-
-#define HASH_MUR(key,keylen,hashv)                                     \
-do {                                                                   \
-  const uint8_t *_mur_data = (const uint8_t*)(key);                    \
-  const int _mur_nblocks = (int)(keylen) / 4;                          \
-  uint32_t _mur_h1 = 0xf88D5353u;                                      \
-  uint32_t _mur_c1 = 0xcc9e2d51u;                                      \
-  uint32_t _mur_c2 = 0x1b873593u;                                      \
-  uint32_t _mur_k1 = 0;                                                \
-  const uint8_t *_mur_tail;                                            \
-  const uint32_t *_mur_blocks = (const uint32_t*)(_mur_data+(_mur_nblocks*4)); \
-  int _mur_i;                                                          \
-  for(_mur_i = -_mur_nblocks; _mur_i!=0; _mur_i++) {                   \
-    _mur_k1 = MUR_GETBLOCK(_mur_blocks,_mur_i);                        \
-    _mur_k1 *= _mur_c1;                                                \
-    _mur_k1 = MUR_ROTL32(_mur_k1,15);                                  \
-    _mur_k1 *= _mur_c2;                                                \
-                                                                       \
-    _mur_h1 ^= _mur_k1;                                                \
-    _mur_h1 = MUR_ROTL32(_mur_h1,13);                                  \
-    _mur_h1 = (_mur_h1*5U) + 0xe6546b64u;                              \
-  }                                                                    \
-  _mur_tail = (const uint8_t*)(_mur_data + (_mur_nblocks*4));          \
-  _mur_k1=0;                                                           \
-  switch((keylen) & 3U) {                                              \
-    case 3: _mur_k1 ^= (uint32_t)_mur_tail[2] << 16; /* FALLTHROUGH */ \
-    case 2: _mur_k1 ^= (uint32_t)_mur_tail[1] << 8;  /* FALLTHROUGH */ \
-    case 1: _mur_k1 ^= (uint32_t)_mur_tail[0];                         \
-    _mur_k1 *= _mur_c1;                                                \
-    _mur_k1 = MUR_ROTL32(_mur_k1,15);                                  \
-    _mur_k1 *= _mur_c2;                                                \
-    _mur_h1 ^= _mur_k1;                                                \
-  }                                                                    \
-  _mur_h1 ^= (uint32_t)(keylen);                                       \
-  MUR_FMIX(_mur_h1);                                                   \
-  hashv = _mur_h1;                                                     \
-} while (0)
-#endif  /* HASH_USING_NO_STRICT_ALIASING */
-
-/* iterate over items in a known bucket to find desired item */
-#define HASH_FIND_IN_BKT(tbl,hh,head,keyptr,keylen_in,hashval,out)               \
-do {                                                                             \
-  if ((head).hh_head != NULL) {                                                  \
-    DECLTYPE_ASSIGN(out, ELMT_FROM_HH(tbl, (head).hh_head));                     \
-  } else {                                                                       \
-    (out) = NULL;                                                                \
-  }                                                                              \
-  while ((out) != NULL) {                                                        \
-    if ((out)->hh.hashv == (hashval) && (out)->hh.keylen == (keylen_in)) {       \
-      if (uthash_memcmp((out)->hh.key, keyptr, keylen_in) == 0) {                \
-        break;                                                                   \
-      }                                                                          \
-    }                                                                            \
-    if ((out)->hh.hh_next != NULL) {                                             \
-      DECLTYPE_ASSIGN(out, ELMT_FROM_HH(tbl, (out)->hh.hh_next));                \
-    } else {                                                                     \
-      (out) = NULL;                                                              \
-    }                                                                            \
-  }                                                                              \
-} while (0)
-
-/* add an item to a bucket  */
-#define HASH_ADD_TO_BKT(head,addhh)                                              \
-do {                                                                             \
- head.count++;                                                                   \
- (addhh)->hh_next = head.hh_head;                                                \
- (addhh)->hh_prev = NULL;                                                        \
- if (head.hh_head != NULL) { (head).hh_head->hh_prev = (addhh); }                \
- (head).hh_head=addhh;                                                           \
- if ((head.count >= ((head.expand_mult+1U) * HASH_BKT_CAPACITY_THRESH))          \
-     && ((addhh)->tbl->noexpand != 1U)) {                                        \
-       HASH_EXPAND_BUCKETS((addhh)->tbl);                                        \
- }                                                                               \
-} while (0)
-
-/* remove an item from a given bucket */
-#define HASH_DEL_IN_BKT(hh,head,hh_del)                                          \
-    (head).count--;                                                              \
-    if ((head).hh_head == hh_del) {                                              \
-      (head).hh_head = hh_del->hh_next;                                          \
-    }                                                                            \
-    if (hh_del->hh_prev) {                                                       \
-        hh_del->hh_prev->hh_next = hh_del->hh_next;                              \
-    }                                                                            \
-    if (hh_del->hh_next) {                                                       \
-        hh_del->hh_next->hh_prev = hh_del->hh_prev;                              \
-    }
-
-/* Bucket expansion has the effect of doubling the number of buckets
- * and redistributing the items into the new buckets. Ideally the
- * items will distribute more or less evenly into the new buckets
- * (the extent to which this is true is a measure of the quality of
- * the hash function as it applies to the key domain).
- *
- * With the items distributed into more buckets, the chain length
- * (item count) in each bucket is reduced. Thus by expanding buckets
- * the hash keeps a bound on the chain length. This bounded chain
- * length is the essence of how a hash provides constant time lookup.
- *
- * The calculation of tbl->ideal_chain_maxlen below deserves some
- * explanation. First, keep in mind that we're calculating the ideal
- * maximum chain length based on the *new* (doubled) bucket count.
- * In fractions this is just n/b (n=number of items,b=new num buckets).
- * Since the ideal chain length is an integer, we want to calculate
- * ceil(n/b). We don't depend on floating point arithmetic in this
- * hash, so to calculate ceil(n/b) with integers we could write
- *
- *      ceil(n/b) = (n/b) + ((n%b)?1:0)
- *
- * and in fact a previous version of this hash did just that.
- * But now we have improved things a bit by recognizing that b is
- * always a power of two. We keep its base 2 log handy (call it lb),
- * so now we can write this with a bit shift and logical AND:
- *
- *      ceil(n/b) = (n>>lb) + ( (n & (b-1)) ? 1:0)
- *
- */
-#define HASH_EXPAND_BUCKETS(tbl)                                                 \
-do {                                                                             \
-    unsigned _he_bkt;                                                            \
-    unsigned _he_bkt_i;                                                          \
-    struct UT_hash_handle *_he_thh, *_he_hh_nxt;                                 \
-    UT_hash_bucket *_he_new_buckets, *_he_newbkt;                                \
-    _he_new_buckets = (UT_hash_bucket*)uthash_malloc(                            \
-             2UL * tbl->num_buckets * sizeof(struct UT_hash_bucket));            \
-    if (!_he_new_buckets) { uthash_fatal( "out of memory"); }                    \
-    memset(_he_new_buckets, 0,                                                   \
-            2UL * tbl->num_buckets * sizeof(struct UT_hash_bucket));             \
-    tbl->ideal_chain_maxlen =                                                    \
-       (tbl->num_items >> (tbl->log2_num_buckets+1U)) +                          \
-       (((tbl->num_items & ((tbl->num_buckets*2U)-1U)) != 0U) ? 1U : 0U);        \
-    tbl->nonideal_items = 0;                                                     \
-    for(_he_bkt_i = 0; _he_bkt_i < tbl->num_buckets; _he_bkt_i++)                \
-    {                                                                            \
-        _he_thh = tbl->buckets[ _he_bkt_i ].hh_head;                             \
-        while (_he_thh != NULL) {                                                \
-           _he_hh_nxt = _he_thh->hh_next;                                        \
-           HASH_TO_BKT( _he_thh->hashv, tbl->num_buckets*2U, _he_bkt);           \
-           _he_newbkt = &(_he_new_buckets[ _he_bkt ]);                           \
-           if (++(_he_newbkt->count) > tbl->ideal_chain_maxlen) {                \
-             tbl->nonideal_items++;                                              \
-             _he_newbkt->expand_mult = _he_newbkt->count /                       \
-                                        tbl->ideal_chain_maxlen;                 \
-           }                                                                     \
-           _he_thh->hh_prev = NULL;                                              \
-           _he_thh->hh_next = _he_newbkt->hh_head;                               \
-           if (_he_newbkt->hh_head != NULL) { _he_newbkt->hh_head->hh_prev =     \
-                _he_thh; }                                                       \
-           _he_newbkt->hh_head = _he_thh;                                        \
-           _he_thh = _he_hh_nxt;                                                 \
-        }                                                                        \
-    }                                                                            \
-    uthash_free( tbl->buckets, tbl->num_buckets*sizeof(struct UT_hash_bucket) ); \
-    tbl->num_buckets *= 2U;                                                      \
-    tbl->log2_num_buckets++;                                                     \
-    tbl->buckets = _he_new_buckets;                                              \
-    tbl->ineff_expands = (tbl->nonideal_items > (tbl->num_items >> 1)) ?         \
-        (tbl->ineff_expands+1U) : 0U;                                            \
-    if (tbl->ineff_expands > 1U) {                                               \
-        tbl->noexpand=1;                                                         \
-        uthash_noexpand_fyi(tbl);                                                \
-    }                                                                            \
-    uthash_expand_fyi(tbl);                                                      \
-} while (0)
-
-
-/* This is an adaptation of Simon Tatham's O(n log(n)) mergesort */
-/* Note that HASH_SORT assumes the hash handle name to be hh.
- * HASH_SRT was added to allow the hash handle name to be passed in. */
-#define HASH_SORT(head,cmpfcn) HASH_SRT(hh,head,cmpfcn)
-#define HASH_SRT(hh,head,cmpfcn)                                                 \
-do {                                                                             \
-  unsigned _hs_i;                                                                \
-  unsigned _hs_looping,_hs_nmerges,_hs_insize,_hs_psize,_hs_qsize;               \
-  struct UT_hash_handle *_hs_p, *_hs_q, *_hs_e, *_hs_list, *_hs_tail;            \
-  if (head != NULL) {                                                            \
-      _hs_insize = 1;                                                            \
-      _hs_looping = 1;                                                           \
-      _hs_list = &((head)->hh);                                                  \
-      while (_hs_looping != 0U) {                                                \
-          _hs_p = _hs_list;                                                      \
-          _hs_list = NULL;                                                       \
-          _hs_tail = NULL;                                                       \
-          _hs_nmerges = 0;                                                       \
-          while (_hs_p != NULL) {                                                \
-              _hs_nmerges++;                                                     \
-              _hs_q = _hs_p;                                                     \
-              _hs_psize = 0;                                                     \
-              for ( _hs_i = 0; _hs_i  < _hs_insize; _hs_i++ ) {                  \
-                  _hs_psize++;                                                   \
-                  _hs_q = (UT_hash_handle*)((_hs_q->next != NULL) ?              \
-                          ((void*)((char*)(_hs_q->next) +                        \
-                          (head)->hh.tbl->hho)) : NULL);                         \
-                  if (! (_hs_q) ) { break; }                                     \
-              }                                                                  \
-              _hs_qsize = _hs_insize;                                            \
-              while ((_hs_psize > 0U) || ((_hs_qsize > 0U) && (_hs_q != NULL))) {\
-                  if (_hs_psize == 0U) {                                         \
-                      _hs_e = _hs_q;                                             \
-                      _hs_q = (UT_hash_handle*)((_hs_q->next != NULL) ?          \
-                              ((void*)((char*)(_hs_q->next) +                    \
-                              (head)->hh.tbl->hho)) : NULL);                     \
-                      _hs_qsize--;                                               \
-                  } else if ( (_hs_qsize == 0U) || (_hs_q == NULL) ) {           \
-                      _hs_e = _hs_p;                                             \
-                      if (_hs_p != NULL){                                        \
-                        _hs_p = (UT_hash_handle*)((_hs_p->next != NULL) ?        \
-                                ((void*)((char*)(_hs_p->next) +                  \
-                                (head)->hh.tbl->hho)) : NULL);                   \
-                       }                                                         \
-                      _hs_psize--;                                               \
-                  } else if ((                                                   \
-                      cmpfcn(DECLTYPE(head)(ELMT_FROM_HH((head)->hh.tbl,_hs_p)), \
-                             DECLTYPE(head)(ELMT_FROM_HH((head)->hh.tbl,_hs_q))) \
-                             ) <= 0) {                                           \
-                      _hs_e = _hs_p;                                             \
-                      if (_hs_p != NULL){                                        \
-                        _hs_p = (UT_hash_handle*)((_hs_p->next != NULL) ?        \
-                               ((void*)((char*)(_hs_p->next) +                   \
-                               (head)->hh.tbl->hho)) : NULL);                    \
-                       }                                                         \
-                      _hs_psize--;                                               \
-                  } else {                                                       \
-                      _hs_e = _hs_q;                                             \
-                      _hs_q = (UT_hash_handle*)((_hs_q->next != NULL) ?          \
-                              ((void*)((char*)(_hs_q->next) +                    \
-                              (head)->hh.tbl->hho)) : NULL);                     \
-                      _hs_qsize--;                                               \
-                  }                                                              \
-                  if ( _hs_tail != NULL ) {                                      \
-                      _hs_tail->next = ((_hs_e != NULL) ?                        \
-                            ELMT_FROM_HH((head)->hh.tbl,_hs_e) : NULL);          \
-                  } else {                                                       \
-                      _hs_list = _hs_e;                                          \
-                  }                                                              \
-                  if (_hs_e != NULL) {                                           \
-                  _hs_e->prev = ((_hs_tail != NULL) ?                            \
-                     ELMT_FROM_HH((head)->hh.tbl,_hs_tail) : NULL);              \
-                  }                                                              \
-                  _hs_tail = _hs_e;                                              \
-              }                                                                  \
-              _hs_p = _hs_q;                                                     \
-          }                                                                      \
-          if (_hs_tail != NULL){                                                 \
-            _hs_tail->next = NULL;                                               \
-          }                                                                      \
-          if ( _hs_nmerges <= 1U ) {                                             \
-              _hs_looping=0;                                                     \
-              (head)->hh.tbl->tail = _hs_tail;                                   \
-              DECLTYPE_ASSIGN(head,ELMT_FROM_HH((head)->hh.tbl, _hs_list));      \
-          }                                                                      \
-          _hs_insize *= 2U;                                                      \
-      }                                                                          \
-      HASH_FSCK(hh,head);                                                        \
- }                                                                               \
-} while (0)
-
-/* This function selects items from one hash into another hash.
- * The end result is that the selected items have dual presence
- * in both hashes. There is no copy of the items made; rather
- * they are added into the new hash through a secondary hash
- * hash handle that must be present in the structure. */
-#define HASH_SELECT(hh_dst, dst, hh_src, src, cond)                              \
-do {                                                                             \
-  unsigned _src_bkt, _dst_bkt;                                                   \
-  void *_last_elt=NULL, *_elt;                                                   \
-  UT_hash_handle *_src_hh, *_dst_hh, *_last_elt_hh=NULL;                         \
-  ptrdiff_t _dst_hho = ((char*)(&(dst)->hh_dst) - (char*)(dst));                 \
-  if (src != NULL) {                                                             \
-    for(_src_bkt=0; _src_bkt < (src)->hh_src.tbl->num_buckets; _src_bkt++) {     \
-      for(_src_hh = (src)->hh_src.tbl->buckets[_src_bkt].hh_head;                \
-          _src_hh != NULL;                                                       \
-          _src_hh = _src_hh->hh_next) {                                          \
-          _elt = ELMT_FROM_HH((src)->hh_src.tbl, _src_hh);                       \
-          if (cond(_elt)) {                                                      \
-            _dst_hh = (UT_hash_handle*)(((char*)_elt) + _dst_hho);               \
-            _dst_hh->key = _src_hh->key;                                         \
-            _dst_hh->keylen = _src_hh->keylen;                                   \
-            _dst_hh->hashv = _src_hh->hashv;                                     \
-            _dst_hh->prev = _last_elt;                                           \
-            _dst_hh->next = NULL;                                                \
-            if (_last_elt_hh != NULL) { _last_elt_hh->next = _elt; }             \
-            if (dst == NULL) {                                                   \
-              DECLTYPE_ASSIGN(dst,_elt);                                         \
-              HASH_MAKE_TABLE(hh_dst,dst);                                       \
-            } else {                                                             \
-              _dst_hh->tbl = (dst)->hh_dst.tbl;                                  \
-            }                                                                    \
-            HASH_TO_BKT(_dst_hh->hashv, _dst_hh->tbl->num_buckets, _dst_bkt);    \
-            HASH_ADD_TO_BKT(_dst_hh->tbl->buckets[_dst_bkt],_dst_hh);            \
-            (dst)->hh_dst.tbl->num_items++;                                      \
-            _last_elt = _elt;                                                    \
-            _last_elt_hh = _dst_hh;                                              \
-          }                                                                      \
-      }                                                                          \
-    }                                                                            \
-  }                                                                              \
-  HASH_FSCK(hh_dst,dst);                                                         \
-} while (0)
-
-#define HASH_CLEAR(hh,head)                                                      \
-do {                                                                             \
-  if (head != NULL) {                                                            \
-    uthash_free((head)->hh.tbl->buckets,                                         \
-                (head)->hh.tbl->num_buckets*sizeof(struct UT_hash_bucket));      \
-    HASH_BLOOM_FREE((head)->hh.tbl);                                             \
-    uthash_free((head)->hh.tbl, sizeof(UT_hash_table));                          \
-    (head)=NULL;                                                                 \
-  }                                                                              \
-} while (0)
-
-#define HASH_OVERHEAD(hh,head)                                                   \
- ((head != NULL) ? (                                                             \
- (size_t)(((head)->hh.tbl->num_items   * sizeof(UT_hash_handle))   +             \
-          ((head)->hh.tbl->num_buckets * sizeof(UT_hash_bucket))   +             \
-           sizeof(UT_hash_table)                                   +             \
-           (HASH_BLOOM_BYTELEN))) : 0U)
-
-#ifdef NO_DECLTYPE
-#define HASH_ITER(hh,head,el,tmp)                                                \
-for(((el)=(head)), ((*(char**)(&(tmp)))=(char*)((head!=NULL)?(head)->hh.next:NULL)); \
-  (el) != NULL; ((el)=(tmp)), ((*(char**)(&(tmp)))=(char*)((tmp!=NULL)?(tmp)->hh.next:NULL)))
-#else
-#define HASH_ITER(hh,head,el,tmp)                                                \
-for(((el)=(head)), ((tmp)=DECLTYPE(el)((head!=NULL)?(head)->hh.next:NULL));      \
-  (el) != NULL; ((el)=(tmp)), ((tmp)=DECLTYPE(el)((tmp!=NULL)?(tmp)->hh.next:NULL)))
-#endif
-
-/* obtain a count of items in the hash */
-#define HASH_COUNT(head) HASH_CNT(hh,head)
-#define HASH_CNT(hh,head) ((head != NULL)?((head)->hh.tbl->num_items):0U)
-
-typedef struct UT_hash_bucket {
-   struct UT_hash_handle *hh_head;
-   unsigned count;
-
-   /* expand_mult is normally set to 0. In this situation, the max chain length
-    * threshold is enforced at its default value, HASH_BKT_CAPACITY_THRESH. (If
-    * the bucket's chain exceeds this length, bucket expansion is triggered).
-    * However, setting expand_mult to a non-zero value delays bucket expansion
-    * (that would be triggered by additions to this particular bucket)
-    * until its chain length reaches a *multiple* of HASH_BKT_CAPACITY_THRESH.
-    * (The multiplier is simply expand_mult+1). The whole idea of this
-    * multiplier is to reduce bucket expansions, since they are expensive, in
-    * situations where we know that a particular bucket tends to be overused.
-    * It is better to let its chain length grow to a longer yet-still-bounded
-    * value, than to do an O(n) bucket expansion too often.
-    */
-   unsigned expand_mult;
-
-} UT_hash_bucket;
-
-/* random signature used only to find hash tables in external analysis */
-#define HASH_SIGNATURE 0xa0111fe1u
-#define HASH_BLOOM_SIGNATURE 0xb12220f2u
-
-typedef struct UT_hash_table {
-   UT_hash_bucket *buckets;
-   unsigned num_buckets, log2_num_buckets;
-   unsigned num_items;
-   struct UT_hash_handle *tail; /* tail hh in app order, for fast append    */
-   ptrdiff_t hho; /* hash handle offset (byte pos of hash handle in element */
-
-   /* in an ideal situation (all buckets used equally), no bucket would have
-    * more than ceil(#items/#buckets) items. that's the ideal chain length. */
-   unsigned ideal_chain_maxlen;
-
-   /* nonideal_items is the number of items in the hash whose chain position
-    * exceeds the ideal chain maxlen. these items pay the penalty for an uneven
-    * hash distribution; reaching them in a chain traversal takes >ideal steps */
-   unsigned nonideal_items;
-
-   /* ineffective expands occur when a bucket doubling was performed, but
-    * afterward, more than half the items in the hash had nonideal chain
-    * positions. If this happens on two consecutive expansions we inhibit any
-    * further expansion, as it's not helping; this happens when the hash
-    * function isn't a good fit for the key domain. When expansion is inhibited
-    * the hash will still work, albeit no longer in constant time. */
-   unsigned ineff_expands, noexpand;
-
-   uint32_t signature; /* used only to find hash tables in external analysis */
-#ifdef HASH_BLOOM
-   uint32_t bloom_sig; /* used only to test bloom exists in external analysis */
-   uint8_t *bloom_bv;
-   uint8_t bloom_nbits;
-#endif
-
-} UT_hash_table;
-
-typedef struct UT_hash_handle {
-   struct UT_hash_table *tbl;
-   void *prev;                       /* prev element in app order      */
-   void *next;                       /* next element in app order      */
-   struct UT_hash_handle *hh_prev;   /* previous hh in bucket order    */
-   struct UT_hash_handle *hh_next;   /* next hh in bucket order        */
-   void *key;                        /* ptr to enclosing struct's key  */
-   unsigned keylen;                  /* enclosing struct's key len     */
-   unsigned hashv;                   /* result of hash-fcn(key)        */
-} UT_hash_handle;
-
-#endif /* UTHASH_H */