[Darshan-commits] [Darshan] branch, dev-modular, updated. b1ae796cb558e93b5de96bd3d46077319a4acac9
Service Account
git at mcs.anl.gov
Thu Oct 30 14:24:59 CDT 2014
This is an automated email from the git hooks/post-receive script. It was
generated because a ref change was pushed to the repository containing
the project "".
The branch, dev-modular has been updated
via b1ae796cb558e93b5de96bd3d46077319a4acac9 (commit)
via 265328856281b37ae5c368db8be346239bb0fe55 (commit)
from 29cc3cd6b4a15555a3b5ad7c5d229594d420e64d (commit)
Those revisions listed above that are new to this repository have
not appeared on any other notification email; so we list those
revisions in full, below.
- Log -----------------------------------------------------------------
commit b1ae796cb558e93b5de96bd3d46077319a4acac9
Author: Shane Snyder <ssnyder at mcs.anl.gov>
Date: Thu Oct 30 14:24:33 2014 -0500
Convert posix module to use uthash
commit 265328856281b37ae5c368db8be346239bb0fe55
Author: Shane Snyder <ssnyder at mcs.anl.gov>
Date: Thu Oct 30 13:30:29 2014 -0500
simplified posix module and added core lookup
-----------------------------------------------------------------------
Summary of changes:
darshan-runtime/darshan-core.h | 2 -
darshan-runtime/darshan.h | 1 +
darshan-runtime/lib/darshan-core.c | 200 +-------
darshan-runtime/lib/darshan-posix.c | 166 ++++---
uthash.h | 966 +++++++++++++++++++++++++++++++++++
5 files changed, 1064 insertions(+), 271 deletions(-)
create mode 100644 uthash.h
Diff of changes:
diff --git a/darshan-runtime/darshan-core.h b/darshan-runtime/darshan-core.h
index 7462c41..c0fbc06 100644
--- a/darshan-runtime/darshan-core.h
+++ b/darshan-runtime/darshan-core.h
@@ -40,9 +40,7 @@ struct darshan_core_job_runtime
struct darshan_core_module *mod_list_head;
char comp_buf[CP_COMP_BUF_SIZE];
int flags;
- int file_count;
double wtime_offset;
- char* trailing_data;
};
#endif /* __DARSHAN_CORE_H */
diff --git a/darshan-runtime/darshan.h b/darshan-runtime/darshan.h
index f405968..f7e6347 100644
--- a/darshan-runtime/darshan.h
+++ b/darshan-runtime/darshan.h
@@ -12,6 +12,7 @@
#include <mpi.h>
#include "darshan-log-format.h"
+/* TODO these go where ? */
#define DARSHAN_MPI_CALL(func) func
typedef uint64_t darshan_file_id;
diff --git a/darshan-runtime/lib/darshan-core.c b/darshan-runtime/lib/darshan-core.c
index 69eca3c..1aa0918 100644
--- a/darshan-runtime/lib/darshan-core.c
+++ b/darshan-runtime/lib/darshan-core.c
@@ -25,9 +25,6 @@ extern char* __progname_full;
static void darshan_core_initialize(int *argc, char ***argv);
static void darshan_core_shutdown(void);
-static char *darshan_get_exe_and_mounts(int rank);
-static void darshan_get_exe_and_mounts_root(char* trailing_data, int space_left);
-
/* internal variables */
static struct darshan_core_job_runtime *darshan_global_job = NULL;
static pthread_mutex_t darshan_mutex = PTHREAD_MUTEX_INITIALIZER;
@@ -35,19 +32,6 @@ static pthread_mutex_t darshan_mutex = PTHREAD_MUTEX_INITIALIZER;
#define DARSHAN_LOCK() pthread_mutex_lock(&darshan_mutex)
#define DARSHAN_UNLOCK() pthread_mutex_unlock(&darshan_mutex)
-#define DARSHAN_MAX_MNTS 64
-#define DARSHAN_MAX_MNT_PATH 256
-#define DARSHAN_MAX_MNT_TYPE 32
-struct mnt_data
-{
- int64_t hash;
- int64_t block_size;
- char path[DARSHAN_MAX_MNT_PATH];
- char type[DARSHAN_MAX_MNT_TYPE];
-};
-static struct mnt_data mnt_data_array[DARSHAN_MAX_MNTS];
-static int mnt_data_count = 0;
-
/* intercept MPI initialize and finalize to initialize darshan */
int MPI_Init(int *argc, char ***argv)
{
@@ -161,11 +145,6 @@ static void darshan_core_initialize(int *argc, char ***argv)
sprintf(&darshan_global_job->exe[truncate_offset], "%s",
truncate_string);
}
-
- /* collect information about command line and
- * mounted file systems
- */
- darshan_global_job->trailing_data = darshan_get_exe_and_mounts(rank);
}
}
@@ -196,184 +175,7 @@ static void darshan_core_shutdown()
return;
}
-static int mnt_data_cmp(const void* a, const void* b)
-{
- const struct mnt_data *d_a = (const struct mnt_data*)a;
- const struct mnt_data *d_b = (const struct mnt_data*)b;
-
- if(strlen(d_a->path) > strlen(d_b->path))
- return(-1);
- else if(strlen(d_a->path) < strlen(d_b->path))
- return(1);
- else
- return(0);
-}
-
-/* adds an entry to table of mounted file systems */
-static void add_entry(char* trailing_data, int* space_left, struct mntent *entry)
-{
- int ret;
- char tmp_mnt[256];
- struct statfs statfsbuf;
-
- strncpy(mnt_data_array[mnt_data_count].path, entry->mnt_dir,
- DARSHAN_MAX_MNT_PATH-1);
- strncpy(mnt_data_array[mnt_data_count].type, entry->mnt_type,
- DARSHAN_MAX_MNT_TYPE-1);
- mnt_data_array[mnt_data_count].hash =
- darshan_hash((void*)mnt_data_array[mnt_data_count].path,
- strlen(mnt_data_array[mnt_data_count].path), 0);
- /* NOTE: we now try to detect the preferred block size for each file
- * system using fstatfs(). On Lustre we assume a size of 1 MiB
- * because fstatfs() reports 4 KiB.
- */
-#ifndef LL_SUPER_MAGIC
-#define LL_SUPER_MAGIC 0x0BD00BD0
-#endif
- ret = statfs(entry->mnt_dir, &statfsbuf);
- if(ret == 0 && statfsbuf.f_type != LL_SUPER_MAGIC)
- mnt_data_array[mnt_data_count].block_size = statfsbuf.f_bsize;
- else if(ret == 0 && statfsbuf.f_type == LL_SUPER_MAGIC)
- mnt_data_array[mnt_data_count].block_size = 1024*1024;
- else
- mnt_data_array[mnt_data_count].block_size = 4096;
-
- /* store mount information for use in header of darshan log */
- ret = snprintf(tmp_mnt, 256, "\n%" PRId64 "\t%s\t%s",
- mnt_data_array[mnt_data_count].hash,
- entry->mnt_type, entry->mnt_dir);
- if(ret < 256 && strlen(tmp_mnt) <= (*space_left))
- {
- strcat(trailing_data, tmp_mnt);
- (*space_left) -= strlen(tmp_mnt);
- }
-
- mnt_data_count++;
- return;
-}
-
-/* darshan_get_exe_and_mounts_root()
- *
- * collects command line and list of mounted file systems into a string that
- * will be stored with the job header
- */
-static void darshan_get_exe_and_mounts_root(char* trailing_data, int space_left)
-{
- FILE* tab;
- struct mntent *entry;
- char* exclude;
- int tmp_index = 0;
- int skip = 0;
-
- /* skip these fs types */
- static char* fs_exclusions[] = {
- "tmpfs",
- "proc",
- "sysfs",
- "devpts",
- "binfmt_misc",
- "fusectl",
- "debugfs",
- "securityfs",
- "nfsd",
- "none",
- "rpc_pipefs",
- "hugetlbfs",
- "cgroup",
- NULL
- };
-
- /* length of exe has already been safety checked in darshan-posix.c */
- strcat(trailing_data, darshan_global_job->exe);
- space_left = CP_EXE_LEN - strlen(trailing_data);
-
- /* we make two passes through mounted file systems; in the first pass we
- * grab any non-nfs mount points, then on the second pass we grab nfs
- * mount points
- */
-
- tab = setmntent("/etc/mtab", "r");
- if(!tab)
- return;
- /* loop through list of mounted file systems */
- while(mnt_data_count<DARSHAN_MAX_MNTS && (entry = getmntent(tab)) != NULL)
- {
- /* filter out excluded fs types */
- tmp_index = 0;
- skip = 0;
- while((exclude = fs_exclusions[tmp_index]))
- {
- if(!(strcmp(exclude, entry->mnt_type)))
- {
- skip =1;
- break;
- }
- tmp_index++;
- }
-
- if(skip || (strcmp(entry->mnt_type, "nfs") == 0))
- continue;
-
- add_entry(trailing_data, &space_left, entry);
- }
- endmntent(tab);
-
- tab = setmntent("/etc/mtab", "r");
- if(!tab)
- return;
- /* loop through list of mounted file systems */
- while(mnt_data_count<DARSHAN_MAX_MNTS && (entry = getmntent(tab)) != NULL)
- {
- if(strcmp(entry->mnt_type, "nfs") != 0)
- continue;
-
- add_entry(trailing_data, &space_left, entry);
- }
- endmntent(tab);
-
- /* Sort mount points in order of longest path to shortest path. This is
- * necessary so that if we try to match file paths to mount points later
- * we don't match on "/" every time.
- */
- qsort(mnt_data_array, mnt_data_count, sizeof(mnt_data_array[0]), mnt_data_cmp);
- return;
-}
-
-/* darshan_get_exe_and_mounts()
- *
- * collects command line and list of mounted file systems into a string that
- * will be stored with the job header
- */
-static char *darshan_get_exe_and_mounts(int rank)
-{
- char* trailing_data;
- int space_left;
-
- space_left = CP_EXE_LEN + 1;
- trailing_data = malloc(space_left);
- if(!trailing_data)
- {
- return(NULL);
- }
- memset(trailing_data, 0, space_left);
-
- if(rank == 0)
- {
- darshan_get_exe_and_mounts_root(trailing_data, space_left);
- }
-
- /* broadcast trailing data to all nodes */
- DARSHAN_MPI_CALL(PMPI_Bcast)(trailing_data, space_left, MPI_CHAR, 0,
- MPI_COMM_WORLD);
- /* broadcast mount count to all nodes */
- DARSHAN_MPI_CALL(PMPI_Bcast)(&mnt_data_count, 1, MPI_INT, 0,
- MPI_COMM_WORLD);
- /* broadcast mount data to all nodes */
- DARSHAN_MPI_CALL(PMPI_Bcast)(mnt_data_array,
- mnt_data_count*sizeof(mnt_data_array[0]), MPI_BYTE, 0, MPI_COMM_WORLD);
-
- return(trailing_data);
-}
+/* ********************************************************* */
void darshan_core_register_module(
char *name,
diff --git a/darshan-runtime/lib/darshan-posix.c b/darshan-runtime/lib/darshan-posix.c
index f86628b..a3b1998 100644
--- a/darshan-runtime/lib/darshan-posix.c
+++ b/darshan-runtime/lib/darshan-posix.c
@@ -25,6 +25,7 @@
#include <pthread.h>
#include "darshan.h"
+#include "uthash.h"
#ifndef HAVE_OFF64_T
typedef int64_t off64_t;
@@ -33,6 +34,8 @@ typedef int64_t off64_t;
#define aiocb64 aiocb
#endif
+/* TODO these go where ? */
+
#define DARSHAN_FORWARD_DECL(name,ret,args) \
extern ret __real_ ## name args;
@@ -155,6 +158,7 @@ enum f_darshan_posix_indices
struct darshan_posix_file
{
+ darshan_file_id f_id;
int64_t counters[CP_NUM_INDICES];
double fcounters[CP_F_NUM_INDICES];
};
@@ -162,16 +166,21 @@ struct darshan_posix_file
struct darshan_posix_runtime_file
{
struct darshan_posix_file file_record;
+ UT_hash_handle hlink;
};
struct darshan_posix_runtime
{
- struct darshan_posix_file_runtime *file_array;
- int file_array_sz;
+ struct darshan_posix_runtime_file* file_array;
+ int file_array_size;
+ int file_count;
+ struct darshan_posix_runtime_file* file_hash;
};
static pthread_mutex_t posix_runtime_mutex = PTHREAD_RECURSIVE_MUTEX_INITIALIZER_NP;
static struct darshan_posix_runtime *posix_runtime = NULL;
+static int my_rank = -1;
+static int darshan_mem_alignment = 1;
/* these are paths that we will not trace */
static char* exclusions[] = {
@@ -188,8 +197,6 @@ static char* exclusions[] = {
NULL
};
-static int darshan_mem_alignment = 1;
-
DARSHAN_FORWARD_DECL(open, int, (const char *path, int flags, ...));
//DARSHAN_FORWARD_DECL(close, int, (int fd));
@@ -202,19 +209,6 @@ static void posix_get_output_data(void **buffer, int size);
#define POSIX_LOCK() pthread_mutex_lock(&posix_runtime_mutex)
#define POSIX_UNLOCK() pthread_mutex_unlock(&posix_runtime_mutex)
-#ifdef __DARSHAN_STAT_AT_OPEN
-#define DARSHAN_STAT_FILE(_f, _p, _r) do { \
- if(!DARSHAN_VALUE((_f), DARSHAN_POSIX_STATS) && !DARSHAN_VALUE((_f), DARSHAN_POSIX_OPENS)){ \
- if(fstat64(_r, &cp_stat_buf) == 0) { \
- DARSHAN_SET(_f, DARSHAN_FILE_ALIGNMENT, cp_stat_buf.st_blksize); \
- DARSHAN_SET(_f, DARSHAN_SIZE_AT_OPEN, cp_stat_buf.st_size); \
- }\
- }\
-}while(0)
-#else
-#define DARSHAN_STAT_FILE(_f, _p, _r) do { }while(0)
-#endif
-
#define POSIX_RECORD_OPEN(__ret, __path, __mode, __stream_flag, __tm1, __tm2) do { \
struct darshan_posix_runtime_file* file; \
char* exclude; \
@@ -226,9 +220,8 @@ static void posix_get_output_data(void **buffer, int size);
tmp_index++; \
} \
if(exclude) break; \
- file = *************darshan_file_by_name_setfd(__path, __ret); \
- if(!file_rec) break; \
- DARSHAN_STAT_FILE(file, __path, __ret); \
+ file = darshan_file_by_name(__path, __ret); \
+ if(!file) break; \
file->log_file->rank = my_rank; \
if(__mode) \
DARSHAN_SET(file, DARSHAN_MODE, __mode); \
@@ -244,39 +237,6 @@ static void posix_get_output_data(void **buffer, int size);
DARSHAN_F_INC_NO_OVERLAP(file, __tm1, __tm2, file->last_posix_meta_end, DARSHAN_F_POSIX_META_TIME); \
} while (0)
-#if 0
-int DARSHAN_DECL(close)(int fd)
-{
- struct darshan_file_runtime* file;
- int tmp_fd = fd;
- double tm1, tm2;
- int ret;
-
- MAP_OR_FAIL(close);
-
- tm1 = darshan_core_wtime();
- ret = __real_close(fd);
- tm2 = darshan_core_wtime();
-
- POSIX_LOCK();
- posix_runtime_initialize();
-
- file = darshan_file_by_fd(tmp_fd);
- if(file)
- {
- file->last_byte_written = 0;
- file->last_byte_read = 0;
- DARSHAN_F_SET(file, DARSHAN_F_CLOSE_TIMESTAMP, posix_wtime());
- DARSHAN_F_INC_NO_OVERLAP(file, tm1, tm2, file->last_posix_meta_end, DARSHAN_F_POSIX_META_TIME);
- darshan_file_close_fd(tmp_fd);
- }
-
- POSIX_UNLOCK();
-
- return(ret);
-}
-#endif
-
int DARSHAN_DECL(open)(const char *path, int flags, ...)
{
int mode = 0;
@@ -285,7 +245,7 @@ int DARSHAN_DECL(open)(const char *path, int flags, ...)
MAP_OR_FAIL(open);
- if (flags & O_CREAT)
+ if(flags & O_CREAT)
{
va_list arg;
va_start(arg, flags);
@@ -306,14 +266,47 @@ int DARSHAN_DECL(open)(const char *path, int flags, ...)
POSIX_LOCK();
posix_runtime_initialize();
-// POSIX_RECORD_OPEN(ret, path, mode, 0, tm1, tm2);
+ //POSIX_RECORD_OPEN(ret, path, mode, 0, tm1, tm2);
+
POSIX_UNLOCK();
return(ret);
}
-/* ***************************************************** */
+#if 0
+int DARSHAN_DECL(close)(int fd)
+{
+ struct darshan_file_runtime* file;
+ int tmp_fd = fd;
+ double tm1, tm2;
+ int ret;
+
+ MAP_OR_FAIL(close);
+
+ tm1 = darshan_core_wtime();
+ ret = __real_close(fd);
+ tm2 = darshan_core_wtime();
+ POSIX_LOCK();
+ posix_runtime_initialize();
+
+ file = darshan_file_by_fd(tmp_fd);
+ if(file)
+ {
+ file->last_byte_written = 0;
+ file->last_byte_read = 0;
+ DARSHAN_F_SET(file, DARSHAN_F_CLOSE_TIMESTAMP, posix_wtime());
+ DARSHAN_F_INC_NO_OVERLAP(file, tm1, tm2, file->last_posix_meta_end, DARSHAN_F_POSIX_META_TIME);
+ darshan_file_close_fd(tmp_fd);
+ }
+
+ POSIX_UNLOCK();
+
+ return(ret);
+}
+#endif
+
+/* ***************************************************** */
static void posix_runtime_initialize()
{
@@ -327,15 +320,16 @@ static void posix_runtime_initialize()
.get_output_data = &posix_get_output_data,
};
- if (posix_runtime)
+ if(posix_runtime)
return;
+#if 0
/* set the memory alignment according to config or environment variables */
#if (__CP_MEM_ALIGNMENT < 1)
#error Darshan must be configured with a positive value for --with-mem-align
#endif
alignstr = getenv("DARSHAN_MEMALIGN");
- if (alignstr)
+ if(alignstr)
{
ret = sscanf(alignstr, "%d", &tmpval);
/* silently ignore if the env variable is set poorly */
@@ -350,14 +344,16 @@ static void posix_runtime_initialize()
}
/* avoid floating point errors on faulty input */
- if (darshan_mem_alignment < 1)
+ if(darshan_mem_alignment < 1)
{
darshan_mem_alignment = 1;
}
+#endif
posix_runtime = malloc(sizeof(*posix_runtime));
- if (!posix_runtime)
+ if(!posix_runtime)
return;
+ memset(posix_runtime, 0, sizeof(*posix_runtime));
/* register the posix module with darshan core */
darshan_core_register_module(
@@ -365,38 +361,68 @@ static void posix_runtime_initialize()
&posix_mod_fns,
&mem_limit);
- /* allocate array of runtime file records no larger than the returned mem_limit */
- posix_runtime->file_array_sz = mem_limit / sizeof(struct darshan_posix_runtime_file);
+ /* set maximum number of file records according to max memory limit */
+ posix_runtime->file_array_size = mem_limit / sizeof(struct darshan_posix_runtime_file);
+
+ /* allocate array of runtime file records */
posix_runtime->file_array = malloc(sizeof(struct darshan_posix_runtime_file) *
- posix_runtime->file_array_sz);
- if (!posix_runtime->file_array)
+ posix_runtime->file_array_size);
+ if(!posix_runtime->file_array)
{
- posix_runtime->file_array_sz = 0;
+ posix_runtime->file_array_size = 0;
return;
}
memset(posix_runtime->file_array, 0, sizeof(struct darshan_posix_runtime_file) *
- posix_runtime->file_array_sz);
+ posix_runtime->file_array_size);
return;
}
static struct darshan_posix_runtime_file* posix_file_by_name(const char *name)
{
- struct darshan_posix_runtime_file *tmp_file;
+ struct darshan_posix_runtime_file *tmp_file = NULL;
char *newname = NULL;
+ darshan_file_id tmp_id;
- if (!posix_runtime)
+ if(!posix_runtime)
return(NULL);
newname = darshan_clean_file_path(name);
- if (!newname)
+ if(!newname)
newname = (char*)name;
- if (newname != name)
+ /* get a unique id for this file from darshan core */
+ darshan_core_lookup_id(
+ (void*)newname,
+ strlen(newname),
+ 1,
+ &tmp_id);
+
+ /* search the hash table for this file record, and return if found */
+ HASH_FIND(hlink, posix_runtime->file_hash, &tmp_id, sizeof(darshan_file_id), tmp_file);
+ if (tmp_file)
+ {
+ if (newname != name)
+ free(newname);
+ return(tmp_file);
+ }
+
+ /* no existing record, assign a new file record from the global array */
+ tmp_file = &posix_runtime->file_array[posix_runtime->file_count];
+ tmp_file->file_record.f_id = tmp_id;
+
+ /* add new record to file hash table */
+ HASH_ADD(hlink, posix_runtime->file_hash, file_record.f_id, sizeof(darshan_file_id), tmp_file);
+
+ posix_runtime->file_count++;
+
+ if(newname != name)
free(newname);
return(tmp_file);
}
+/* ***************************************************** */
+
static void posix_prepare_for_shutdown()
{
diff --git a/uthash.h b/uthash.h
new file mode 100644
index 0000000..94ec645
--- /dev/null
+++ b/uthash.h
@@ -0,0 +1,966 @@
+/*
+Copyright (c) 2003-2010, Troy D. Hanson http://uthash.sourceforge.net
+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
+
+#include <string.h> /* memcmp,strlen */
+#include <stddef.h> /* ptrdiff_t */
+
+/* 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. */
+#ifdef _MSC_VER /* MS compiler */
+#if _MSC_VER >= 1600 && __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
+#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 win32 */
+#ifdef _MSC_VER
+typedef unsigned int uint32_t;
+#else
+#include <inttypes.h> /* uint32_t */
+#endif
+
+#define UTHASH_VERSION 1.9.1
+
+#define uthash_fatal(msg) exit(-1) /* fatal error (out of memory,etc) */
+#define uthash_malloc(sz) malloc(sz) /* malloc fcn */
+#define uthash_free(ptr,sz) free(ptr) /* free fcn */
+
+#define uthash_noexpand_fyi(tbl) /* can be defined to log noexpand */
+#define uthash_expand_fyi(tbl) /* can be defined to log expands */
+
+/* initial number of buckets */
+#define HASH_INITIAL_NUM_BUCKETS 32 /* initial number of buckets */
+#define HASH_INITIAL_NUM_BUCKETS_LOG2 5 /* lg2 of initial number of buckets */
+#define HASH_BKT_CAPACITY_THRESH 10 /* expand when bucket count reaches */
+
+/* calculate the element whose hash handle address is hhe */
+#define ELMT_FROM_HH(tbl,hhp) ((void*)(((char*)(hhp)) - ((tbl)->hho)))
+
+#define HASH_FIND(hh,head,keyptr,keylen,out) \
+do { \
+ unsigned _hf_bkt,_hf_hashv; \
+ out=NULL; \
+ if (head) { \
+ HASH_FCN(keyptr,keylen, (head)->hh.tbl->num_buckets, _hf_hashv, _hf_bkt); \
+ if (HASH_BLOOM_TEST((head)->hh.tbl, _hf_hashv)) { \
+ HASH_FIND_IN_BKT((head)->hh.tbl, hh, (head)->hh.tbl->buckets[ _hf_bkt ], \
+ keyptr,keylen,out); \
+ } \
+ } \
+} while (0)
+
+#ifdef HASH_BLOOM
+#define HASH_BLOOM_BITLEN (1ULL << HASH_BLOOM)
+#define HASH_BLOOM_BYTELEN (HASH_BLOOM_BITLEN/8) + ((HASH_BLOOM_BITLEN%8) ? 1:0)
+#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)/8] |= (1U << ((idx)%8)))
+#define HASH_BLOOM_BITTEST(bv,idx) (bv[(idx)/8] & (1U << ((idx)%8)))
+
+#define HASH_BLOOM_ADD(tbl,hashv) \
+ HASH_BLOOM_BITSET((tbl)->bloom_bv, (hashv & (uint32_t)((1ULL << (tbl)->bloom_nbits) - 1)))
+
+#define HASH_BLOOM_TEST(tbl,hashv) \
+ HASH_BLOOM_BITTEST((tbl)->bloom_bv, (hashv & (uint32_t)((1ULL << (tbl)->bloom_nbits) - 1)))
+
+#else
+#define HASH_BLOOM_MAKE(tbl)
+#define HASH_BLOOM_FREE(tbl)
+#define HASH_BLOOM_ADD(tbl,hashv)
+#define HASH_BLOOM_TEST(tbl,hashv) (1)
+#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_ADD(hh,head,fieldname,keylen_in,add) \
+ HASH_ADD_KEYPTR(hh,head,&add->fieldname,keylen_in,add)
+
+#define HASH_ADD_KEYPTR(hh,head,keyptr,keylen_in,add) \
+do { \
+ unsigned _ha_bkt; \
+ (add)->hh.next = NULL; \
+ (add)->hh.key = (char*)keyptr; \
+ (add)->hh.keylen = keylen_in; \
+ if (!(head)) { \
+ head = (add); \
+ (head)->hh.prev = NULL; \
+ HASH_MAKE_TABLE(hh,head); \
+ } else { \
+ (head)->hh.tbl->tail->next = (add); \
+ (add)->hh.prev = ELMT_FROM_HH((head)->hh.tbl, (head)->hh.tbl->tail); \
+ (head)->hh.tbl->tail = &((add)->hh); \
+ } \
+ (head)->hh.tbl->num_items++; \
+ (add)->hh.tbl = (head)->hh.tbl; \
+ HASH_FCN(keyptr,keylen_in, (head)->hh.tbl->num_buckets, \
+ (add)->hh.hashv, _ha_bkt); \
+ HASH_ADD_TO_BKT((head)->hh.tbl->buckets[_ha_bkt],&(add)->hh); \
+ HASH_BLOOM_ADD((head)->hh.tbl,(add)->hh.hashv); \
+ HASH_EMIT_KEY(hh,head,keyptr,keylen_in); \
+ HASH_FSCK(hh,head); \
+} while(0)
+
+#define HASH_TO_BKT( hashv, num_bkts, bkt ) \
+do { \
+ bkt = ((hashv) & ((num_bkts) - 1)); \
+} 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 { \
+ unsigned _hd_bkt; \
+ 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 { \
+ _hd_hh_del = &((delptr)->hh); \
+ if ((delptr) == ELMT_FROM_HH((head)->hh.tbl,(head)->hh.tbl->tail)) { \
+ (head)->hh.tbl->tail = \
+ (UT_hash_handle*)((char*)((delptr)->hh.prev) + \
+ (head)->hh.tbl->hho); \
+ } \
+ if ((delptr)->hh.prev) { \
+ ((UT_hash_handle*)((char*)((delptr)->hh.prev) + \
+ (head)->hh.tbl->hho))->next = (delptr)->hh.next; \
+ } else { \
+ DECLTYPE_ASSIGN(head,(delptr)->hh.next); \
+ } \
+ if (_hd_hh_del->next) { \
+ ((UT_hash_handle*)((char*)_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,strlen(findstr),out)
+#define HASH_ADD_STR(head,strfield,add) \
+ HASH_ADD(hh,head,strfield,strlen(add->strfield),add)
+#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_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_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 { \
+ unsigned _bkt_i; \
+ unsigned _count, _bkt_count; \
+ char *_prev; \
+ struct UT_hash_handle *_thh; \
+ if (head) { \
+ _count = 0; \
+ for( _bkt_i = 0; _bkt_i < (head)->hh.tbl->num_buckets; _bkt_i++) { \
+ _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 %d, actual %d\n", \
+ (head)->hh.tbl->buckets[_bkt_i].count, _bkt_count); \
+ } \
+ } \
+ if (_count != (head)->hh.tbl->num_items) { \
+ HASH_OOPS("invalid hh item count %d, actual %d\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 %d, actual %d\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, 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 */
+#define HASH_BER(key,keylen,num_bkts,hashv,bkt) \
+do { \
+ unsigned _hb_keylen=keylen; \
+ char *_hb_key=(char*)(key); \
+ (hashv) = 0; \
+ while (_hb_keylen--) { (hashv) = ((hashv) * 33) + *_hb_key++; } \
+ bkt = (hashv) & (num_bkts-1); \
+} 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,num_bkts,hashv,bkt) \
+do { \
+ unsigned _sx_i; \
+ char *_hs_key=(char*)(key); \
+ hashv = 0; \
+ for(_sx_i=0; _sx_i < keylen; _sx_i++) \
+ hashv ^= (hashv << 5) + (hashv >> 2) + _hs_key[_sx_i]; \
+ bkt = hashv & (num_bkts-1); \
+} while (0)
+
+#define HASH_FNV(key,keylen,num_bkts,hashv,bkt) \
+do { \
+ unsigned _fn_i; \
+ char *_hf_key=(char*)(key); \
+ hashv = 2166136261UL; \
+ for(_fn_i=0; _fn_i < keylen; _fn_i++) \
+ hashv = (hashv * 16777619) ^ _hf_key[_fn_i]; \
+ bkt = hashv & (num_bkts-1); \
+} while(0);
+
+#define HASH_OAT(key,keylen,num_bkts,hashv,bkt) \
+do { \
+ unsigned _ho_i; \
+ char *_ho_key=(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); \
+ bkt = hashv & (num_bkts-1); \
+} 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,num_bkts,hashv,bkt) \
+do { \
+ unsigned _hj_i,_hj_j,_hj_k; \
+ char *_hj_key=(char*)(key); \
+ hashv = 0xfeedbeef; \
+ _hj_i = _hj_j = 0x9e3779b9; \
+ _hj_k = keylen; \
+ while (_hj_k >= 12) { \
+ _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 -= 12; \
+ } \
+ hashv += keylen; \
+ switch ( _hj_k ) { \
+ case 11: hashv += ( (unsigned)_hj_key[10] << 24 ); \
+ case 10: hashv += ( (unsigned)_hj_key[9] << 16 ); \
+ case 9: hashv += ( (unsigned)_hj_key[8] << 8 ); \
+ case 8: _hj_j += ( (unsigned)_hj_key[7] << 24 ); \
+ case 7: _hj_j += ( (unsigned)_hj_key[6] << 16 ); \
+ case 6: _hj_j += ( (unsigned)_hj_key[5] << 8 ); \
+ case 5: _hj_j += _hj_key[4]; \
+ case 4: _hj_i += ( (unsigned)_hj_key[3] << 24 ); \
+ case 3: _hj_i += ( (unsigned)_hj_key[2] << 16 ); \
+ case 2: _hj_i += ( (unsigned)_hj_key[1] << 8 ); \
+ case 1: _hj_i += _hj_key[0]; \
+ } \
+ HASH_JEN_MIX(_hj_i, _hj_j, hashv); \
+ bkt = hashv & (num_bkts-1); \
+} 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,num_bkts,hashv,bkt) \
+do { \
+ char *_sfh_key=(char*)(key); \
+ uint32_t _sfh_tmp, _sfh_len = keylen; \
+ \
+ int _sfh_rem = _sfh_len & 3; \
+ _sfh_len >>= 2; \
+ hashv = 0xcafebabe; \
+ \
+ /* Main loop */ \
+ for (;_sfh_len > 0; _sfh_len--) { \
+ hashv += get16bits (_sfh_key); \
+ _sfh_tmp = (get16bits (_sfh_key+2) << 11) ^ hashv; \
+ hashv = (hashv << 16) ^ _sfh_tmp; \
+ _sfh_key += 2*sizeof (uint16_t); \
+ hashv += hashv >> 11; \
+ } \
+ \
+ /* Handle end cases */ \
+ switch (_sfh_rem) { \
+ case 3: hashv += get16bits (_sfh_key); \
+ hashv ^= hashv << 16; \
+ hashv ^= _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; \
+ bkt = hashv & (num_bkts-1); \
+} while(0);
+
+#ifdef HASH_USING_NO_STRICT_ALIASING
+/* The MurmurHash exploits some CPU's (e.g. x86) tolerance for unaligned reads.
+ * For other types of CPU's (e.g. Sparc) an unaligned read causes a bus error.
+ * So MurmurHash comes in two versions, the faster unaligned one and the slower
+ * aligned one. We only use the faster one 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__))
+#define HASH_MUR HASH_MUR_UNALIGNED
+#else
+#define HASH_MUR HASH_MUR_ALIGNED
+#endif
+
+/* Appleby's MurmurHash fast version for unaligned-tolerant archs like i386 */
+#define HASH_MUR_UNALIGNED(key,keylen,num_bkts,hashv,bkt) \
+do { \
+ const unsigned int _mur_m = 0x5bd1e995; \
+ const int _mur_r = 24; \
+ hashv = 0xcafebabe ^ keylen; \
+ char *_mur_key = (char *)(key); \
+ uint32_t _mur_tmp, _mur_len = keylen; \
+ \
+ for (;_mur_len >= 4; _mur_len-=4) { \
+ _mur_tmp = *(uint32_t *)_mur_key; \
+ _mur_tmp *= _mur_m; \
+ _mur_tmp ^= _mur_tmp >> _mur_r; \
+ _mur_tmp *= _mur_m; \
+ hashv *= _mur_m; \
+ hashv ^= _mur_tmp; \
+ _mur_key += 4; \
+ } \
+ \
+ switch(_mur_len) \
+ { \
+ case 3: hashv ^= _mur_key[2] << 16; \
+ case 2: hashv ^= _mur_key[1] << 8; \
+ case 1: hashv ^= _mur_key[0]; \
+ hashv *= _mur_m; \
+ }; \
+ \
+ hashv ^= hashv >> 13; \
+ hashv *= _mur_m; \
+ hashv ^= hashv >> 15; \
+ \
+ bkt = hashv & (num_bkts-1); \
+} while(0)
+
+/* Appleby's MurmurHash version for alignment-sensitive archs like Sparc */
+#define HASH_MUR_ALIGNED(key,keylen,num_bkts,hashv,bkt) \
+do { \
+ const unsigned int _mur_m = 0x5bd1e995; \
+ const int _mur_r = 24; \
+ hashv = 0xcafebabe ^ (keylen); \
+ char *_mur_key = (char *)(key); \
+ uint32_t _mur_len = keylen; \
+ int _mur_align = (int)_mur_key & 3; \
+ \
+ if (_mur_align && (_mur_len >= 4)) { \
+ unsigned _mur_t = 0, _mur_d = 0; \
+ switch(_mur_align) { \
+ case 1: _mur_t |= _mur_key[2] << 16; \
+ case 2: _mur_t |= _mur_key[1] << 8; \
+ case 3: _mur_t |= _mur_key[0]; \
+ } \
+ _mur_t <<= (8 * _mur_align); \
+ _mur_key += 4-_mur_align; \
+ _mur_len -= 4-_mur_align; \
+ int _mur_sl = 8 * (4-_mur_align); \
+ int _mur_sr = 8 * _mur_align; \
+ \
+ for (;_mur_len >= 4; _mur_len-=4) { \
+ _mur_d = *(unsigned *)_mur_key; \
+ _mur_t = (_mur_t >> _mur_sr) | (_mur_d << _mur_sl); \
+ unsigned _mur_k = _mur_t; \
+ _mur_k *= _mur_m; \
+ _mur_k ^= _mur_k >> _mur_r; \
+ _mur_k *= _mur_m; \
+ hashv *= _mur_m; \
+ hashv ^= _mur_k; \
+ _mur_t = _mur_d; \
+ _mur_key += 4; \
+ } \
+ _mur_d = 0; \
+ if(_mur_len >= _mur_align) { \
+ switch(_mur_align) { \
+ case 3: _mur_d |= _mur_key[2] << 16; \
+ case 2: _mur_d |= _mur_key[1] << 8; \
+ case 1: _mur_d |= _mur_key[0]; \
+ } \
+ unsigned _mur_k = (_mur_t >> _mur_sr) | (_mur_d << _mur_sl); \
+ _mur_k *= _mur_m; \
+ _mur_k ^= _mur_k >> _mur_r; \
+ _mur_k *= _mur_m; \
+ hashv *= _mur_m; \
+ hashv ^= _mur_k; \
+ _mur_k += _mur_align; \
+ _mur_len -= _mur_align; \
+ \
+ switch(_mur_len) \
+ { \
+ case 3: hashv ^= _mur_key[2] << 16; \
+ case 2: hashv ^= _mur_key[1] << 8; \
+ case 1: hashv ^= _mur_key[0]; \
+ hashv *= _mur_m; \
+ } \
+ } else { \
+ switch(_mur_len) \
+ { \
+ case 3: _mur_d ^= _mur_key[2] << 16; \
+ case 2: _mur_d ^= _mur_key[1] << 8; \
+ case 1: _mur_d ^= _mur_key[0]; \
+ case 0: hashv ^= (_mur_t >> _mur_sr) | (_mur_d << _mur_sl); \
+ hashv *= _mur_m; \
+ } \
+ } \
+ \
+ hashv ^= hashv >> 13; \
+ hashv *= _mur_m; \
+ hashv ^= hashv >> 15; \
+ } else { \
+ for (;_mur_len >= 4; _mur_len-=4) { \
+ unsigned _mur_k = *(unsigned*)_mur_key; \
+ _mur_k *= _mur_m; \
+ _mur_k ^= _mur_k >> _mur_r; \
+ _mur_k *= _mur_m; \
+ hashv *= _mur_m; \
+ hashv ^= _mur_k; \
+ _mur_key += 4; \
+ } \
+ switch(_mur_len) \
+ { \
+ case 3: hashv ^= _mur_key[2] << 16; \
+ case 2: hashv ^= _mur_key[1] << 8; \
+ case 1: hashv ^= _mur_key[0]; \
+ hashv *= _mur_m; \
+ } \
+ \
+ hashv ^= hashv >> 13; \
+ hashv *= _mur_m; \
+ hashv ^= hashv >> 15; \
+ } \
+ bkt = hashv & (num_bkts-1); \
+} while(0)
+#endif /* HASH_USING_NO_STRICT_ALIASING */
+
+/* key comparison function; return 0 if keys equal */
+#define HASH_KEYCMP(a,b,len) memcmp(a,b,len)
+
+/* iterate over items in a known bucket to find desired item */
+#define HASH_FIND_IN_BKT(tbl,hh,head,keyptr,keylen_in,out) \
+do { \
+ if (head.hh_head) DECLTYPE_ASSIGN(out,ELMT_FROM_HH(tbl,head.hh_head)); \
+ else out=NULL; \
+ while (out) { \
+ if (out->hh.keylen == keylen_in) { \
+ if ((HASH_KEYCMP(out->hh.key,keyptr,keylen_in)) == 0) break; \
+ } \
+ if (out->hh.hh_next) 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) { (head).hh_head->hh_prev = (addhh); } \
+ (head).hh_head=addhh; \
+ if (head.count >= ((head.expand_mult+1) * HASH_BKT_CAPACITY_THRESH) \
+ && (addhh)->tbl->noexpand != 1) { \
+ 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( \
+ 2 * tbl->num_buckets * sizeof(struct UT_hash_bucket)); \
+ if (!_he_new_buckets) { uthash_fatal( "out of memory"); } \
+ memset(_he_new_buckets, 0, \
+ 2 * tbl->num_buckets * sizeof(struct UT_hash_bucket)); \
+ tbl->ideal_chain_maxlen = \
+ (tbl->num_items >> (tbl->log2_num_buckets+1)) + \
+ ((tbl->num_items & ((tbl->num_buckets*2)-1)) ? 1 : 0); \
+ 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) { \
+ _he_hh_nxt = _he_thh->hh_next; \
+ HASH_TO_BKT( _he_thh->hashv, tbl->num_buckets*2, _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) _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 *= 2; \
+ tbl->log2_num_buckets++; \
+ tbl->buckets = _he_new_buckets; \
+ tbl->ineff_expands = (tbl->nonideal_items > (tbl->num_items >> 1)) ? \
+ (tbl->ineff_expands+1) : 0; \
+ if (tbl->ineff_expands > 1) { \
+ 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) { \
+ _hs_insize = 1; \
+ _hs_looping = 1; \
+ _hs_list = &((head)->hh); \
+ while (_hs_looping) { \
+ _hs_p = _hs_list; \
+ _hs_list = NULL; \
+ _hs_tail = NULL; \
+ _hs_nmerges = 0; \
+ while (_hs_p) { \
+ _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) ? \
+ ((void*)((char*)(_hs_q->next) + \
+ (head)->hh.tbl->hho)) : NULL); \
+ if (! (_hs_q) ) break; \
+ } \
+ _hs_qsize = _hs_insize; \
+ while ((_hs_psize > 0) || ((_hs_qsize > 0) && _hs_q )) { \
+ if (_hs_psize == 0) { \
+ _hs_e = _hs_q; \
+ _hs_q = (UT_hash_handle*)((_hs_q->next) ? \
+ ((void*)((char*)(_hs_q->next) + \
+ (head)->hh.tbl->hho)) : NULL); \
+ _hs_qsize--; \
+ } else if ( (_hs_qsize == 0) || !(_hs_q) ) { \
+ _hs_e = _hs_p; \
+ _hs_p = (UT_hash_handle*)((_hs_p->next) ? \
+ ((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; \
+ _hs_p = (UT_hash_handle*)((_hs_p->next) ? \
+ ((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) ? \
+ ((void*)((char*)(_hs_q->next) + \
+ (head)->hh.tbl->hho)) : NULL); \
+ _hs_qsize--; \
+ } \
+ if ( _hs_tail ) { \
+ _hs_tail->next = ((_hs_e) ? \
+ ELMT_FROM_HH((head)->hh.tbl,_hs_e) : NULL); \
+ } else { \
+ _hs_list = _hs_e; \
+ } \
+ _hs_e->prev = ((_hs_tail) ? \
+ ELMT_FROM_HH((head)->hh.tbl,_hs_tail) : NULL); \
+ _hs_tail = _hs_e; \
+ } \
+ _hs_p = _hs_q; \
+ } \
+ _hs_tail->next = NULL; \
+ if ( _hs_nmerges <= 1 ) { \
+ _hs_looping=0; \
+ (head)->hh.tbl->tail = _hs_tail; \
+ DECLTYPE_ASSIGN(head,ELMT_FROM_HH((head)->hh.tbl, _hs_list)); \
+ } \
+ _hs_insize *= 2; \
+ } \
+ 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) { \
+ 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; \
+ _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) { _last_elt_hh->next = _elt; } \
+ if (!dst) { \
+ 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) { \
+ uthash_free((head)->hh.tbl->buckets, \
+ (head)->hh.tbl->num_buckets*sizeof(struct UT_hash_bucket)); \
+ uthash_free((head)->hh.tbl, sizeof(UT_hash_table)); \
+ (head)=NULL; \
+ } \
+} while(0)
+
+#define HASH_ITER(hh,head,el,tmp) \
+for((el)=(head),(tmp)=(head)?(head)->hh.next:NULL; \
+ el; (el)=(tmp),(tmp)=(tmp)?(tmp)->hh.next:NULL)
+
+/* obtain a count of items in the hash */
+#define HASH_COUNT(head) HASH_CNT(hh,head)
+#define HASH_CNT(hh,head) ((head)?((head)->hh.tbl->num_items):0)
+
+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 0xa0111fe1
+#define HASH_BLOOM_SIGNATURE 0xb12220f2
+
+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;
+ char 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 */
hooks/post-receive
--
More information about the Darshan-commits
mailing list