Add new mmap config option.

Changes related to Riak 2.0 and an issue with how statistics were gathered from the backend

.gitignore

@@ -7,6 +7,8 @@ c_src/*.o
 c_src/bzip2-1.0.6
 c_src/snappy-1.0.4
 deps/
-priv/
+priv/wt
+priv/*.so*
+priv/*.dylib*
 log/
 *~

Makefile

@@ -52,19 +52,17 @@ endif
 .PHONY: all compile doc clean test dialyzer typer shell distclean pdf \
   update-deps clean-common-test-data rebuild
 
-all: deps compile test
+all: deps compile
 
 # =============================================================================
 # Rules to build the system
 # =============================================================================
 
 deps:
-	c_src/build_deps.sh get-deps
 	$(REBAR) get-deps
 	$(REBAR) compile
 
 update-deps:
-	c_src/build_deps.sh update-deps
 	$(REBAR) update-deps
 	$(REBAR) compile
 

c_src/async_nif.h

@@ -34,9 +34,18 @@ extern "C" {
 
 #define ASYNC_NIF_MAX_WORKERS 1024
 #define ASYNC_NIF_MIN_WORKERS 2
-#define ASYNC_NIF_WORKER_QUEUE_SIZE 100
+#define ASYNC_NIF_WORKER_QUEUE_SIZE 8192
 #define ASYNC_NIF_MAX_QUEUED_REQS ASYNC_NIF_WORKER_QUEUE_SIZE * ASYNC_NIF_MAX_WORKERS
 
+/* Atoms (initialized in on_load) */
+static ERL_NIF_TERM ATOM_EAGAIN;
+static ERL_NIF_TERM ATOM_ENOMEM;
+static ERL_NIF_TERM ATOM_ENQUEUED;
+static ERL_NIF_TERM ATOM_ERROR;
+static ERL_NIF_TERM ATOM_OK;
+static ERL_NIF_TERM ATOM_SHUTDOWN;
+
+
 struct async_nif_req_entry {
   ERL_NIF_TERM ref;
   ErlNifEnv *env;
@@ -53,6 +62,7 @@ struct async_nif_work_queue {
   unsigned int depth;
   ErlNifMutex *reqs_mutex;
   ErlNifCond *reqs_cnd;
+  struct async_nif_work_queue *next;
   STAILQ_HEAD(reqs, async_nif_req_entry) reqs;
 };
 
@@ -103,25 +113,20 @@ struct async_nif_state {
     argc -= 1;                                                          \
     /* Note: !!! this assumes that the first element of priv_data is ours */ \
     struct async_nif_state *async_nif = *(struct async_nif_state**)enif_priv_data(env); \
-    if (async_nif->shutdown) {                                          \
-      return enif_make_tuple2(env, enif_make_atom(env, "error"),        \
-                              enif_make_atom(env, "shutdown"));         \
-    }                                                                   \
+    if (async_nif->shutdown)						\
+	return enif_make_tuple2(env, ATOM_ERROR, ATOM_SHUTDOWN);	\
     req = async_nif_reuse_req(async_nif);                               \
-    if (!req) {                                                         \
-        return enif_make_tuple2(env, enif_make_atom(env, "error"),      \
-                                enif_make_atom(env, "eagain"));         \
-    }                                                                   \
+    if (!req)								\
+        return enif_make_tuple2(env, ATOM_ERROR, ATOM_ENOMEM);		\
     new_env = req->env;                                                 \
     DPRINTF("async_nif: calling \"%s\"", __func__);                     \
     do pre_block while(0);                                              \
     DPRINTF("async_nif: returned from \"%s\"", __func__);               \
-    copy_of_args = (struct decl ## _args *)enif_alloc(sizeof(struct decl ## _args)); \
+    copy_of_args = (struct decl ## _args *)malloc(sizeof(struct decl ## _args)); \
     if (!copy_of_args) {                                                \
       fn_post_ ## decl (args);                                          \
       async_nif_recycle_req(req, async_nif);                            \
-      return enif_make_tuple2(env, enif_make_atom(env, "error"),        \
-                              enif_make_atom(env, "enomem"));           \
+      return enif_make_tuple2(env, ATOM_ERROR, ATOM_ENOMEM);		\
     }                                                                   \
     memcpy(copy_of_args, args, sizeof(struct decl ## _args));           \
     req->ref = enif_make_copy(new_env, argv_in[0]);                     \
@@ -136,9 +141,8 @@ struct async_nif_state {
     if (!reply) {                                                      \
       fn_post_ ## decl (args);                                         \
       async_nif_recycle_req(req, async_nif);                           \
-      enif_free(copy_of_args);                                         \
-      return enif_make_tuple2(env, enif_make_atom(env, "error"),       \
-                              enif_make_atom(env, "eagain"));          \
+      free(copy_of_args);					       \
+      return enif_make_tuple2(env, ATOM_ERROR, ATOM_EAGAIN);	       \
     }                                                                  \
     return reply;                                                      \
   }
@@ -146,16 +150,16 @@ struct async_nif_state {
 #define ASYNC_NIF_INIT(name)                                            \
         static ErlNifMutex *name##_async_nif_coord = NULL;
 
-#define ASYNC_NIF_LOAD(name, priv) do {                                 \
+#define ASYNC_NIF_LOAD(name, env, priv) do {				\
         if (!name##_async_nif_coord)                                    \
-            name##_async_nif_coord = enif_mutex_create(NULL);           \
+            name##_async_nif_coord = enif_mutex_create("nif_coord load"); \
         enif_mutex_lock(name##_async_nif_coord);                        \
-        priv = async_nif_load();                                        \
+        priv = async_nif_load(env);					\
         enif_mutex_unlock(name##_async_nif_coord);                      \
     } while(0);
 #define ASYNC_NIF_UNLOAD(name, env, priv) do {                          \
         if (!name##_async_nif_coord)                                    \
-            name##_async_nif_coord = enif_mutex_create(NULL);           \
+            name##_async_nif_coord = enif_mutex_create("nif_coord unload"); \
         enif_mutex_lock(name##_async_nif_coord);                        \
         async_nif_unload(env, priv);                                    \
         enif_mutex_unlock(name##_async_nif_coord);                      \
@@ -164,7 +168,7 @@ struct async_nif_state {
     } while(0);
 #define ASYNC_NIF_UPGRADE(name, env) do {                               \
         if (!name##_async_nif_coord)                                    \
-            name##_async_nif_coord = enif_mutex_create(NULL);           \
+            name##_async_nif_coord = enif_mutex_create("nif_coord upgrade"); \
         enif_mutex_lock(name##_async_nif_coord);                        \
         async_nif_upgrade(env);                                         \
         enif_mutex_unlock(name##_async_nif_coord);                      \
@@ -191,15 +195,15 @@ async_nif_reuse_req(struct async_nif_state *async_nif)
     enif_mutex_lock(async_nif->recycled_req_mutex);
     if (STAILQ_EMPTY(&async_nif->recycled_reqs)) {
         if (async_nif->num_reqs < ASYNC_NIF_MAX_QUEUED_REQS) {
-            req = enif_alloc(sizeof(struct async_nif_req_entry));
+            req = malloc(sizeof(struct async_nif_req_entry));
             if (req) {
                 memset(req, 0, sizeof(struct async_nif_req_entry));
                 env = enif_alloc_env();
                 if (env) {
                     req->env = env;
-                    async_nif->num_reqs++;
+                    __sync_fetch_and_add(&async_nif->num_reqs, 1);
                 } else {
-                    enif_free(req);
+                    free(req);
                     req = NULL;
                 }
             }
@@ -253,7 +257,7 @@ async_nif_start_worker(struct async_nif_state *async_nif, struct async_nif_work_
     SLIST_REMOVE(&async_nif->we_joining, we, async_nif_worker_entry, entries);
     void *exit_value = 0; /* We ignore the thread_join's exit value. */
     enif_thread_join(we->tid, &exit_value);
-    enif_free(we);
+    free(we);
     async_nif->we_active--;
     we = n;
   }
@@ -263,7 +267,7 @@ async_nif_start_worker(struct async_nif_state *async_nif, struct async_nif_work_
       return EAGAIN;
   }
 
-  we = enif_alloc(sizeof(struct async_nif_worker_entry));
+  we = malloc(sizeof(struct async_nif_worker_entry));
   if (!we) {
       enif_mutex_unlock(async_nif->we_mutex);
       return ENOMEM;
@@ -287,9 +291,9 @@ static ERL_NIF_TERM
 async_nif_enqueue_req(struct async_nif_state* async_nif, struct async_nif_req_entry *req, int hint)
 {
   /* Identify the most appropriate worker for this request. */
-  unsigned int i, qid = 0;
+  unsigned int i, last_qid, qid = 0;
   struct async_nif_work_queue *q = NULL;
-  double avg_depth;
+  double avg_depth = 0.0;
 
   /* Either we're choosing a queue based on some affinity/hinted value or we
      need to select the next queue in the rotation and atomically update that
@@ -297,9 +301,10 @@ async_nif_enqueue_req(struct async_nif_state* async_nif, struct async_nif_req_en
   if (hint >= 0) {
       qid = (unsigned int)hint;
   } else {
-      qid = async_nif->next_q;
-      qid = (qid + 1) % async_nif->num_queues;
-      async_nif->next_q = qid;
+      do {
+          last_qid = __sync_fetch_and_add(&async_nif->next_q, 0);
+          qid = (last_qid + 1) % async_nif->num_queues;
+      } while (!__sync_bool_compare_and_swap(&async_nif->next_q, last_qid, qid));
   }
 
   /* Now we inspect and interate across the set of queues trying to select one
@@ -314,18 +319,13 @@ async_nif_enqueue_req(struct async_nif_state* async_nif, struct async_nif_req_en
               avg_depth += async_nif->queues[j].depth;
           }
       }
-      if (avg_depth != 0)
-          avg_depth /= n;
+      if (avg_depth) avg_depth /= n;
 
       /* Lock this queue under consideration, then check for shutdown.  While
          we hold this lock either a) we're shutting down so exit now or b) this
          queue will be valid until we release the lock. */
       q = &async_nif->queues[qid];
       enif_mutex_lock(q->reqs_mutex);
-      if (async_nif->shutdown) {
-          enif_mutex_unlock(q->reqs_mutex);
-          return 0;
-      }
 
       /* Try not to enqueue a request into a queue that isn't keeping up with
          the request volume. */
@@ -343,18 +343,25 @@ async_nif_enqueue_req(struct async_nif_state* async_nif, struct async_nif_req_en
 
   /* Add the request to the queue. */
   STAILQ_INSERT_TAIL(&q->reqs, req, entries);
-  q->depth++;
+  __sync_fetch_and_add(&q->depth, 1);
 
   /* We've selected a queue for this new request now check to make sure there are
      enough workers actively processing requests on this queue. */
-  if (q->depth > q->num_workers || q->num_workers == 0)
-      if (async_nif_start_worker(async_nif, q) == 0) q->num_workers++;
+  while (q->depth > q->num_workers) {
+      switch(async_nif_start_worker(async_nif, q)) {
+      case EINVAL: case ENOMEM: default: return 0;
+      case EAGAIN: continue;
+      case 0:      __sync_fetch_and_add(&q->num_workers, 1); goto done;
+      }
+  }done:;
 
   /* Build the term before releasing the lock so as not to race on the use of
      the req pointer (which will soon become invalid in another thread
      performing the request). */
-  ERL_NIF_TERM reply = enif_make_tuple2(req->env, enif_make_atom(req->env, "ok"),
-                                        enif_make_atom(req->env, "enqueued"));
+  double pct_full = (double)avg_depth / (double)ASYNC_NIF_WORKER_QUEUE_SIZE;
+  ERL_NIF_TERM reply = enif_make_tuple2(req->env, ATOM_OK,
+					enif_make_tuple2(req->env, ATOM_ENQUEUED,
+							 enif_make_double(req->env, pct_full)));
   enif_cond_signal(q->reqs_cnd);
   enif_mutex_unlock(q->reqs_mutex);
   return reply;
@@ -373,6 +380,7 @@ async_nif_worker_fn(void *arg)
   struct async_nif_state *async_nif = we->async_nif;
   struct async_nif_work_queue *q = we->q;
   struct async_nif_req_entry *req = NULL;
+  unsigned int tries = async_nif->num_queues;
 
   for(;;) {
     /* Examine the request queue, are there things to be done? */
@@ -384,22 +392,33 @@ async_nif_worker_fn(void *arg)
     }
     if (STAILQ_EMPTY(&q->reqs)) {
       /* Queue is empty so we wait for more work to arrive. */
-      if (q->num_workers > ASYNC_NIF_MIN_WORKERS) {
-          enif_mutex_unlock(q->reqs_mutex);
-          break;
-      } else {
-          enif_cond_wait(q->reqs_cnd, q->reqs_mutex);
-          goto check_again_for_work;
-      }
+	enif_mutex_unlock(q->reqs_mutex);
+	if (tries == 0 && q == we->q) {
+	    if (q->num_workers > ASYNC_NIF_MIN_WORKERS) {
+		/* At this point we've tried to find/execute work on all queues
+		 * and there are at least MIN_WORKERS on this queue so we
+		 * leaving this loop (break) which leads to a thread exit/join. */
+		break;
+	    } else {
+		enif_mutex_lock(q->reqs_mutex);
+		enif_cond_wait(q->reqs_cnd, q->reqs_mutex);
+		goto check_again_for_work;
+	    }
+	} else {
+	    tries--;
+	    __sync_fetch_and_add(&q->num_workers, -1);
+	    q = q->next;
+	    __sync_fetch_and_add(&q->num_workers, 1);
+	    continue; // try next queue
+	}
     } else {
       /* At this point the next req is ours to process and we hold the
          reqs_mutex lock.  Take the request off the queue. */
       req = STAILQ_FIRST(&q->reqs);
       STAILQ_REMOVE(&q->reqs, req, async_nif_req_entry, entries);
-      q->depth--;
+      __sync_fetch_and_add(&q->depth, -1);
 
-      /* Ensure that there is at least one other worker thread watching this
-         queue. */
+      /* Wake up other worker thread watching this queue to help process work. */
       enif_cond_signal(q->reqs_cnd);
       enif_mutex_unlock(q->reqs_mutex);
 
@@ -413,7 +432,7 @@ async_nif_worker_fn(void *arg)
       req->ref = 0;
       req->fn_work = 0;
       req->fn_post = 0;
-      enif_free(req->args);
+      free(req->args);
       req->args = NULL;
       async_nif_recycle_req(req, async_nif);
       req = NULL;
@@ -422,7 +441,7 @@ async_nif_worker_fn(void *arg)
   enif_mutex_lock(async_nif->we_mutex);
   SLIST_INSERT_HEAD(&async_nif->we_joining, we, entries);
   enif_mutex_unlock(async_nif->we_mutex);
-  q->num_workers--;
+  __sync_fetch_and_add(&q->num_workers, -1);
   enif_thread_exit(0);
   return 0;
 }
@@ -465,7 +484,7 @@ async_nif_unload(ErlNifEnv *env, struct async_nif_state *async_nif)
           SLIST_REMOVE(&async_nif->we_joining, we, async_nif_worker_entry, entries);
           void *exit_value = 0; /* We ignore the thread_join's exit value. */
           enif_thread_join(we->tid, &exit_value);
-          enif_free(we);
+          free(we);
           async_nif->we_active--;
           we = n;
       }
@@ -484,12 +503,11 @@ async_nif_unload(ErlNifEnv *env, struct async_nif_state *async_nif)
           struct async_nif_req_entry *n = STAILQ_NEXT(req, entries);
           enif_clear_env(req->env);
           enif_send(NULL, &req->pid, req->env,
-                    enif_make_tuple2(req->env, enif_make_atom(req->env, "error"),
-                                     enif_make_atom(req->env, "shutdown")));
+		    enif_make_tuple2(req->env, ATOM_ERROR, ATOM_SHUTDOWN));
           req->fn_post(req->args);
           enif_free_env(req->env);
-          enif_free(req->args);
-          enif_free(req);
+          free(req->args);
+          free(req);
           req = n;
       }
       enif_mutex_destroy(q->reqs_mutex);
@@ -503,18 +521,18 @@ async_nif_unload(ErlNifEnv *env, struct async_nif_state *async_nif)
   while(req != NULL) {
       struct async_nif_req_entry *n = STAILQ_NEXT(req, entries);
       enif_free_env(req->env);
-      enif_free(req);
+      free(req);
       req = n;
   }
 
   enif_mutex_unlock(async_nif->recycled_req_mutex);
   enif_mutex_destroy(async_nif->recycled_req_mutex);
   memset(async_nif, 0, sizeof(struct async_nif_state) + (sizeof(struct async_nif_work_queue) * async_nif->num_queues));
-  enif_free(async_nif);
+  free(async_nif);
 }
 
 static void *
-async_nif_load()
+async_nif_load(ErlNifEnv *env)
 {
   static int has_init = 0;
   unsigned int i, num_queues;
@@ -525,6 +543,14 @@ async_nif_load()
   if (has_init) return 0;
   else has_init = 1;
 
+  /* Init some static references to commonly used atoms. */
+  ATOM_EAGAIN = enif_make_atom(env, "eagain");
+  ATOM_ENOMEM = enif_make_atom(env, "enomem");
+  ATOM_ENQUEUED = enif_make_atom(env, "enqueued");
+  ATOM_ERROR = enif_make_atom(env, "error");
+  ATOM_OK = enif_make_atom(env, "ok");
+  ATOM_SHUTDOWN = enif_make_atom(env, "shutdown");
+
   /* Find out how many schedulers there are. */
   enif_system_info(&info, sizeof(ErlNifSysInfo));
 
@@ -542,8 +568,8 @@ async_nif_load()
   }
 
   /* Init our portion of priv_data's module-specific state. */
-  async_nif = enif_alloc(sizeof(struct async_nif_state) +
-                         sizeof(struct async_nif_work_queue) * num_queues);
+  async_nif = malloc(sizeof(struct async_nif_state) +
+		     sizeof(struct async_nif_work_queue) * num_queues);
   if (!async_nif)
       return NULL;
   memset(async_nif, 0, sizeof(struct async_nif_state) +
@@ -554,15 +580,16 @@ async_nif_load()
   async_nif->next_q = 0;
   async_nif->shutdown = 0;
   STAILQ_INIT(&async_nif->recycled_reqs);
-  async_nif->recycled_req_mutex = enif_mutex_create(NULL);
-  async_nif->we_mutex = enif_mutex_create(NULL);
+  async_nif->recycled_req_mutex = enif_mutex_create("recycled_req");
+  async_nif->we_mutex = enif_mutex_create("we");
   SLIST_INIT(&async_nif->we_joining);
 
   for (i = 0; i < async_nif->num_queues; i++) {
       struct async_nif_work_queue *q = &async_nif->queues[i];
       STAILQ_INIT(&q->reqs);
-      q->reqs_mutex = enif_mutex_create(NULL);
-      q->reqs_cnd = enif_cond_create(NULL);
+      q->reqs_mutex = enif_mutex_create("reqs");
+      q->reqs_cnd = enif_cond_create("reqs");
+      q->next = &async_nif->queues[(i + 1) % num_queues];
   }
   return async_nif;
 }

c_src/atomic.h

@@ -1,100 +0,0 @@
-/*
- * File:
- *   atomic.h
- * Author(s):
- *   Pascal Felber <pascal.felber@unine.ch>
- *   Patrick Marlier <patrick.marlier@unine.ch>
- * Description:
- *   Atomic operations.
- *
- * Copyright (c) 2007-2012.
- *
- * This program is free software; you can redistribute it and/or
- * modify it under the terms of the GNU General Public License
- * as published by the Free Software Foundation, version 2
- * of the License.
- *
- * This program is distributed in the hope that it will be useful,
- * but WITHOUT ANY WARRANTY; without even the implied warranty of
- * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
- * GNU General Public License for more details.
- *
- * This program has a dual license and can also be distributed
- * under the terms of the MIT license.
- */
-
-#ifndef _ATOMIC_H_
-# define _ATOMIC_H_
-
-# ifdef ATOMIC_BUILTIN
-typedef volatile size_t atomic_t;
-#  ifdef __INTEL_COMPILER
-#   define ATOMIC_CB                     __memory_barrier()
-#  else /* ! __INTEL_COMPILER, assuming __GNUC__ */
-#   define ATOMIC_CB                     __asm__ __volatile__("": : :"memory")
-#  endif /* ! __INTEL_COMPILER */
-#  ifndef UNSAFE
-#   warning "This is experimental and shouldn't be used"
-/*
-   Note: __sync_ is available for GCC 4.2+ and ICC 11.1+
-   But these definitions are not 100% safe:
-    * need 'a' to be volatile
-    * no fence for read/store proposed (only full fence)
-   C11 and C++11 also propose atomic operations.
-*/
-#   define ATOMIC_CAS_FULL(a, e, v)      (__sync_bool_compare_and_swap(a, e, v))
-#   define ATOMIC_FETCH_INC_FULL(a)      (__sync_fetch_and_add(a, 1))
-#   define ATOMIC_FETCH_DEC_FULL(a)      (__sync_fetch_and_add(a, -1))
-#   define ATOMIC_FETCH_ADD_FULL(a, v)   (__sync_fetch_and_add(a, v))
-#   define ATOMIC_LOAD_ACQ(a)            (*(a))
-#   define ATOMIC_LOAD(a)                (*(a))
-#   define ATOMIC_STORE_REL(a, v)        (*(a) = (v))
-#   define ATOMIC_STORE(a, v)            (*(a) = (v))
-#   define ATOMIC_MB_READ                /* Nothing */
-#   define ATOMIC_MB_WRITE               /* Nothing */
-#   define ATOMIC_MB_FULL                __sync_synchronize()
-#  else
-/* Use only for testing purposes (single thread benchmarks) */
-#   define ATOMIC_CAS_FULL(a, e, v)      (*(a) = (v), 1)
-#   define ATOMIC_FETCH_INC_FULL(a)      ((*(a))++)
-#   define ATOMIC_FETCH_DEC_FULL(a)      ((*(a))--)
-#   define ATOMIC_FETCH_ADD_FULL(a, v)   ((*(a)) += (v))
-#   define ATOMIC_LOAD_ACQ(a)            (*(a))
-#   define ATOMIC_LOAD(a)                (*(a))
-#   define ATOMIC_STORE_REL(a, v)        (*(a) = (v))
-#   define ATOMIC_STORE(a, v)            (*(a) = (v))
-#   define ATOMIC_MB_READ                /* Nothing */
-#   define ATOMIC_MB_WRITE               /* Nothing */
-#   define ATOMIC_MB_FULL                /* Nothing */
-#  endif /* UNSAFE */
-
-# else /* ! ATOMIC_BUILTIN */
-/* NOTE: enable fence instructions for i386 and amd64 but the mfence instructions seems costly. */
-/* # define AO_USE_PENTIUM4_INSTRS */
-#  include "atomic_ops/atomic_ops.h"
-typedef AO_t atomic_t;
-#  define ATOMIC_CB                     AO_compiler_barrier()
-#  define ATOMIC_CAS_FULL(a, e, v)      (AO_compare_and_swap_full((volatile AO_t *)(a), (AO_t)(e), (AO_t)(v)))
-#  define ATOMIC_FETCH_INC_FULL(a)      (AO_fetch_and_add1_full((volatile AO_t *)(a)))
-#  define ATOMIC_FETCH_DEC_FULL(a)      (AO_fetch_and_sub1_full((volatile AO_t *)(a)))
-#  define ATOMIC_FETCH_ADD_FULL(a, v)   (AO_fetch_and_add_full((volatile AO_t *)(a), (AO_t)(v)))
-#  ifdef SAFE
-#   define ATOMIC_LOAD_ACQ(a)           (AO_load_full((volatile AO_t *)(a)))
-#   define ATOMIC_LOAD(a)               (AO_load_full((volatile AO_t *)(a)))
-#   define ATOMIC_STORE_REL(a, v)       (AO_store_full((volatile AO_t *)(a), (AO_t)(v)))
-#   define ATOMIC_STORE(a, v)           (AO_store_full((volatile AO_t *)(a), (AO_t)(v)))
-#   define ATOMIC_MB_READ               AO_nop_full()
-#   define ATOMIC_MB_WRITE              AO_nop_full()
-#   define ATOMIC_MB_FULL               AO_nop_full()
-#  else /* ! SAFE */
-#   define ATOMIC_LOAD_ACQ(a)           (AO_load_acquire_read((volatile AO_t *)(a)))
-#   define ATOMIC_LOAD(a)               (*((volatile AO_t *)(a)))
-#   define ATOMIC_STORE_REL(a, v)       (AO_store_release((volatile AO_t *)(a), (AO_t)(v)))
-#   define ATOMIC_STORE(a, v)           (*((volatile AO_t *)(a)) = (AO_t)(v))
-#   define ATOMIC_MB_READ               AO_nop_read()
-#   define ATOMIC_MB_WRITE              AO_nop_write()
-#   define ATOMIC_MB_FULL               AO_nop_full()
-#  endif /* ! SAFE */
-# endif /* ! NO_AO */
-
-#endif /* _ATOMIC_H_ */

c_src/atomic_ops/AUTHORS

@@ -1,4 +0,0 @@
-Originally written by Hans Boehm, with some platform-dependent code
-imported from the Boehm-Demers-Weiser GC, where it was contributed
-by many others.
-

c_src/atomic_ops/COPYING

@@ -1,340 +0,0 @@
-		    GNU GENERAL PUBLIC LICENSE
-		       Version 2, June 1991
-
- Copyright (C) 1989, 1991 Free Software Foundation, Inc.
-     59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
- Everyone is permitted to copy and distribute verbatim copies
- of this license document, but changing it is not allowed.
-
-			    Preamble
-
-  The licenses for most software are designed to take away your
-freedom to share and change it.  By contrast, the GNU General Public
-License is intended to guarantee your freedom to share and change free
-software--to make sure the software is free for all its users.  This
-General Public License applies to most of the Free Software
-Foundation's software and to any other program whose authors commit to
-using it.  (Some other Free Software Foundation software is covered by
-the GNU Library General Public License instead.)  You can apply it to
-your programs, too.
-
-  When we speak of free software, we are referring to freedom, not
-price.  Our General Public Licenses are designed to make sure that you
-have the freedom to distribute copies of free software (and charge for
-this service if you wish), that you receive source code or can get it
-if you want it, that you can change the software or use pieces of it
-in new free programs; and that you know you can do these things.
-
-  To protect your rights, we need to make restrictions that forbid
-anyone to deny you these rights or to ask you to surrender the rights.
-These restrictions translate to certain responsibilities for you if you
-distribute copies of the software, or if you modify it.
-
-  For example, if you distribute copies of such a program, whether
-gratis or for a fee, you must give the recipients all the rights that
-you have.  You must make sure that they, too, receive or can get the
-source code.  And you must show them these terms so they know their
-rights.
-
-  We protect your rights with two steps: (1) copyright the software, and
-(2) offer you this license which gives you legal permission to copy,
-distribute and/or modify the software.
-
-  Also, for each author's protection and ours, we want to make certain
-that everyone understands that there is no warranty for this free
-software.  If the software is modified by someone else and passed on, we
-want its recipients to know that what they have is not the original, so
-that any problems introduced by others will not reflect on the original
-authors' reputations.
-
-  Finally, any free program is threatened constantly by software
-patents.  We wish to avoid the danger that redistributors of a free
-program will individually obtain patent licenses, in effect making the
-program proprietary.  To prevent this, we have made it clear that any
-patent must be licensed for everyone's free use or not licensed at all.
-
-  The precise terms and conditions for copying, distribution and
-modification follow.
-
-		    GNU GENERAL PUBLIC LICENSE
-   TERMS AND CONDITIONS FOR COPYING, DISTRIBUTION AND MODIFICATION
-
-  0. This License applies to any program or other work which contains
-a notice placed by the copyright holder saying it may be distributed
-under the terms of this General Public License.  The "Program", below,
-refers to any such program or work, and a "work based on the Program"
-means either the Program or any derivative work under copyright law:
-that is to say, a work containing the Program or a portion of it,
-either verbatim or with modifications and/or translated into another
-language.  (Hereinafter, translation is included without limitation in
-the term "modification".)  Each licensee is addressed as "you".
-
-Activities other than copying, distribution and modification are not
-covered by this License; they are outside its scope.  The act of
-running the Program is not restricted, and the output from the Program
-is covered only if its contents constitute a work based on the
-Program (independent of having been made by running the Program).
-Whether that is true depends on what the Program does.
-
-  1. You may copy and distribute verbatim copies of the Program's
-source code as you receive it, in any medium, provided that you
-conspicuously and appropriately publish on each copy an appropriate
-copyright notice and disclaimer of warranty; keep intact all the
-notices that refer to this License and to the absence of any warranty;
-and give any other recipients of the Program a copy of this License
-along with the Program.
-
-You may charge a fee for the physical act of transferring a copy, and
-you may at your option offer warranty protection in exchange for a fee.
-
-  2. You may modify your copy or copies of the Program or any portion
-of it, thus forming a work based on the Program, and copy and
-distribute such modifications or work under the terms of Section 1
-above, provided that you also meet all of these conditions:
-
-    a) You must cause the modified files to carry prominent notices
-    stating that you changed the files and the date of any change.
-
-    b) You must cause any work that you distribute or publish, that in
-    whole or in part contains or is derived from the Program or any
-    part thereof, to be licensed as a whole at no charge to all third
-    parties under the terms of this License.
-
-    c) If the modified program normally reads commands interactively
-    when run, you must cause it, when started running for such
-    interactive use in the most ordinary way, to print or display an
-    announcement including an appropriate copyright notice and a
-    notice that there is no warranty (or else, saying that you provide
-    a warranty) and that users may redistribute the program under
-    these conditions, and telling the user how to view a copy of this
-    License.  (Exception: if the Program itself is interactive but
-    does not normally print such an announcement, your work based on
-    the Program is not required to print an announcement.)
-
-These requirements apply to the modified work as a whole.  If
-identifiable sections of that work are not derived from the Program,
-and can be reasonably considered independent and separate works in
-themselves, then this License, and its terms, do not apply to those
-sections when you distribute them as separate works.  But when you
-distribute the same sections as part of a whole which is a work based
-on the Program, the distribution of the whole must be on the terms of
-this License, whose permissions for other licensees extend to the
-entire whole, and thus to each and every part regardless of who wrote it.
-
-Thus, it is not the intent of this section to claim rights or contest
-your rights to work written entirely by you; rather, the intent is to
-exercise the right to control the distribution of derivative or
-collective works based on the Program.
-
-In addition, mere aggregation of another work not based on the Program
-with the Program (or with a work based on the Program) on a volume of
-a storage or distribution medium does not bring the other work under
-the scope of this License.
-
-  3. You may copy and distribute the Program (or a work based on it,
-under Section 2) in object code or executable form under the terms of
-Sections 1 and 2 above provided that you also do one of the following:
-
-    a) Accompany it with the complete corresponding machine-readable
-    source code, which must be distributed under the terms of Sections
-    1 and 2 above on a medium customarily used for software interchange; or,
-
-    b) Accompany it with a written offer, valid for at least three
-    years, to give any third party, for a charge no more than your
-    cost of physically performing source distribution, a complete
-    machine-readable copy of the corresponding source code, to be
-    distributed under the terms of Sections 1 and 2 above on a medium
-    customarily used for software interchange; or,
-
-    c) Accompany it with the information you received as to the offer
-    to distribute corresponding source code.  (This alternative is
-    allowed only for noncommercial distribution and only if you
-    received the program in object code or executable form with such
-    an offer, in accord with Subsection b above.)
-
-The source code for a work means the preferred form of the work for
-making modifications to it.  For an executable work, complete source
-code means all the source code for all modules it contains, plus any
-associated interface definition files, plus the scripts used to
-control compilation and installation of the executable.  However, as a
-special exception, the source code distributed need not include
-anything that is normally distributed (in either source or binary
-form) with the major components (compiler, kernel, and so on) of the
-operating system on which the executable runs, unless that component
-itself accompanies the executable.
-
-If distribution of executable or object code is made by offering
-access to copy from a designated place, then offering equivalent
-access to copy the source code from the same place counts as
-distribution of the source code, even though third parties are not
-compelled to copy the source along with the object code.
-
-  4. You may not copy, modify, sublicense, or distribute the Program
-except as expressly provided under this License.  Any attempt
-otherwise to copy, modify, sublicense or distribute the Program is
-void, and will automatically terminate your rights under this License.
-However, parties who have received copies, or rights, from you under
-this License will not have their licenses terminated so long as such
-parties remain in full compliance.
-
-  5. You are not required to accept this License, since you have not
-signed it.  However, nothing else grants you permission to modify or
-distribute the Program or its derivative works.  These actions are
-prohibited by law if you do not accept this License.  Therefore, by
-modifying or distributing the Program (or any work based on the
-Program), you indicate your acceptance of this License to do so, and
-all its terms and conditions for copying, distributing or modifying
-the Program or works based on it.
-
-  6. Each time you redistribute the Program (or any work based on the
-Program), the recipient automatically receives a license from the
-original licensor to copy, distribute or modify the Program subject to
-these terms and conditions.  You may not impose any further
-restrictions on the recipients' exercise of the rights granted herein.
-You are not responsible for enforcing compliance by third parties to
-this License.
-
-  7. If, as a consequence of a court judgment or allegation of patent
-infringement or for any other reason (not limited to patent issues),
-conditions are imposed on you (whether by court order, agreement or
-otherwise) that contradict the conditions of this License, they do not
-excuse you from the conditions of this License.  If you cannot
-distribute so as to satisfy simultaneously your obligations under this
-License and any other pertinent obligations, then as a consequence you
-may not distribute the Program at all.  For example, if a patent
-license would not permit royalty-free redistribution of the Program by
-all those who receive copies directly or indirectly through you, then
-the only way you could satisfy both it and this License would be to
-refrain entirely from distribution of the Program.
-
-If any portion of this section is held invalid or unenforceable under
-any particular circumstance, the balance of the section is intended to
-apply and the section as a whole is intended to apply in other
-circumstances.
-
-It is not the purpose of this section to induce you to infringe any
-patents or other property right claims or to contest validity of any
-such claims; this section has the sole purpose of protecting the
-integrity of the free software distribution system, which is
-implemented by public license practices.  Many people have made
-generous contributions to the wide range of software distributed
-through that system in reliance on consistent application of that
-system; it is up to the author/donor to decide if he or she is willing
-to distribute software through any other system and a licensee cannot
-impose that choice.
-
-This section is intended to make thoroughly clear what is believed to
-be a consequence of the rest of this License.
-
-  8. If the distribution and/or use of the Program is restricted in
-certain countries either by patents or by copyrighted interfaces, the
-original copyright holder who places the Program under this License
-may add an explicit geographical distribution limitation excluding
-those countries, so that distribution is permitted only in or among
-countries not thus excluded.  In such case, this License incorporates
-the limitation as if written in the body of this License.
-
-  9. The Free Software Foundation may publish revised and/or new versions
-of the General Public License from time to time.  Such new versions will
-be similar in spirit to the present version, but may differ in detail to
-address new problems or concerns.
-
-Each version is given a distinguishing version number.  If the Program
-specifies a version number of this License which applies to it and "any
-later version", you have the option of following the terms and conditions
-either of that version or of any later version published by the Free
-Software Foundation.  If the Program does not specify a version number of
-this License, you may choose any version ever published by the Free Software
-Foundation.
-
-  10. If you wish to incorporate parts of the Program into other free
-programs whose distribution conditions are different, write to the author
-to ask for permission.  For software which is copyrighted by the Free
-Software Foundation, write to the Free Software Foundation; we sometimes
-make exceptions for this.  Our decision will be guided by the two goals
-of preserving the free status of all derivatives of our free software and
-of promoting the sharing and reuse of software generally.
-
-			    NO WARRANTY
-
-  11. BECAUSE THE PROGRAM IS LICENSED FREE OF CHARGE, THERE IS NO WARRANTY
-FOR THE PROGRAM, TO THE EXTENT PERMITTED BY APPLICABLE LAW.  EXCEPT WHEN
-OTHERWISE STATED IN WRITING THE COPYRIGHT HOLDERS AND/OR OTHER PARTIES
-PROVIDE THE PROGRAM "AS IS" WITHOUT WARRANTY OF ANY KIND, EITHER EXPRESSED
-OR IMPLIED, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF
-MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE.  THE ENTIRE RISK AS
-TO THE QUALITY AND PERFORMANCE OF THE PROGRAM IS WITH YOU.  SHOULD THE
-PROGRAM PROVE DEFECTIVE, YOU ASSUME THE COST OF ALL NECESSARY SERVICING,
-REPAIR OR CORRECTION.
-
-  12. IN NO EVENT UNLESS REQUIRED BY APPLICABLE LAW OR AGREED TO IN WRITING
-WILL ANY COPYRIGHT HOLDER, OR ANY OTHER PARTY WHO MAY MODIFY AND/OR
-REDISTRIBUTE THE PROGRAM AS PERMITTED ABOVE, BE LIABLE TO YOU FOR DAMAGES,
-INCLUDING ANY GENERAL, SPECIAL, INCIDENTAL OR CONSEQUENTIAL DAMAGES ARISING
-OUT OF THE USE OR INABILITY TO USE THE PROGRAM (INCLUDING BUT NOT LIMITED
-TO LOSS OF DATA OR DATA BEING RENDERED INACCURATE OR LOSSES SUSTAINED BY
-YOU OR THIRD PARTIES OR A FAILURE OF THE PROGRAM TO OPERATE WITH ANY OTHER
-PROGRAMS), EVEN IF SUCH HOLDER OR OTHER PARTY HAS BEEN ADVISED OF THE
-POSSIBILITY OF SUCH DAMAGES.
-
-		     END OF TERMS AND CONDITIONS
-
-	    How to Apply These Terms to Your New Programs
-
-  If you develop a new program, and you want it to be of the greatest
-possible use to the public, the best way to achieve this is to make it
-free software which everyone can redistribute and change under these terms.
-
-  To do so, attach the following notices to the program.  It is safest
-to attach them to the start of each source file to most effectively
-convey the exclusion of warranty; and each file should have at least
-the "copyright" line and a pointer to where the full notice is found.
-
-    <one line to give the program's name and a brief idea of what it does.>
-    Copyright (C) <year>  <name of author>
-
-    This program is free software; you can redistribute it and/or modify
-    it under the terms of the GNU General Public License as published by
-    the Free Software Foundation; either version 2 of the License, or
-    (at your option) any later version.
-
-    This program is distributed in the hope that it will be useful,
-    but WITHOUT ANY WARRANTY; without even the implied warranty of
-    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
-    GNU General Public License for more details.
-
-    You should have received a copy of the GNU General Public License
-    along with this program; if not, write to the Free Software
-    Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
-
-
-Also add information on how to contact you by electronic and paper mail.
-
-If the program is interactive, make it output a short notice like this
-when it starts in an interactive mode:
-
-    Gnomovision version 69, Copyright (C) year  name of author
-    Gnomovision comes with ABSOLUTELY NO WARRANTY; for details type `show w'.
-    This is free software, and you are welcome to redistribute it
-    under certain conditions; type `show c' for details.
-
-The hypothetical commands `show w' and `show c' should show the appropriate
-parts of the General Public License.  Of course, the commands you use may
-be called something other than `show w' and `show c'; they could even be
-mouse-clicks or menu items--whatever suits your program.
-
-You should also get your employer (if you work as a programmer) or your
-school, if any, to sign a "copyright disclaimer" for the program, if
-necessary.  Here is a sample; alter the names:
-
-  Yoyodyne, Inc., hereby disclaims all copyright interest in the program
-  `Gnomovision' (which makes passes at compilers) written by James Hacker.
-
-  <signature of Ty Coon>, 1 April 1989
-  Ty Coon, President of Vice
-
-This General Public License does not permit incorporating your program into
-proprietary programs.  If your program is a subroutine library, you may
-consider it more useful to permit linking proprietary applications with the
-library.  If this is what you want to do, use the GNU Library General
-Public License instead of this License.

c_src/atomic_ops/README

@@ -1,2 +0,0 @@
-This directory contains a stripped-down (support only gcc) version of libatomic_ops by Hans Boehm.
-The official release is available from http://www.hpl.hp.com/research/linux/atomic_ops/.

c_src/atomic_ops/aligned_atomic_load_store.h

@@ -1,46 +0,0 @@
-/*
- * Copyright (c) 2003 Hewlett-Packard Development Company, L.P.
- *
- * Permission is hereby granted, free of charge, to any person obtaining a copy
- * of this software and associated documentation files (the "Software"), to deal
- * in the Software without restriction, including without limitation the rights
- * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
- * copies of the Software, and to permit persons to whom the Software is
- * furnished to do so, subject to the following conditions:
- *
- * The above copyright notice and this permission notice shall be included in
- * all copies or substantial portions of the Software.
- *
- * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
- * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
- * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
- * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
- * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
- * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
- * SOFTWARE.
- */
-
-/*
- * Definitions for architectures on which loads and stores of AO_t are
- * atomic for all legal alignments.
- */
-
-AO_INLINE AO_t
-AO_load(const volatile AO_t *addr)
-{
-  assert(((size_t)addr & (sizeof(AO_t) - 1)) == 0);
-  /* Cast away the volatile for architectures where volatile adds barrier
-     semantics. */
-  return *(AO_t *)addr;
-}
-
-#define AO_HAVE_load
-
-AO_INLINE void
-AO_store(volatile AO_t *addr, AO_t new_val)
-{
-  assert(((size_t)addr & (sizeof(AO_t) - 1)) == 0);
-  (*(AO_t *)addr) = new_val;
-}
-
-#define AO_HAVE_store

c_src/atomic_ops/all_acquire_release_volatile.h

@@ -1,168 +0,0 @@
-/*
- * Copyright (c) 2004 Hewlett-Packard Development Company, L.P.
- * 
- * Permission is hereby granted, free of charge, to any person obtaining a copy
- * of this software and associated documentation files (the "Software"), to deal
- * in the Software without restriction, including without limitation the rights
- * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
- * copies of the Software, and to permit persons to whom the Software is
- * furnished to do so, subject to the following conditions:
- * 
- * The above copyright notice and this permission notice shall be included in
- * all copies or substantial portions of the Software.
- * 
- * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
- * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
- * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
- * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
- * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
- * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
- * SOFTWARE. 
- */
-
-/*
- * Describes architectures on which volatile AO_t, unsigned char, unsigned
- * short, and unsigned int loads and stores have acquire/release semantics for
- * all normally legal alignments.
- */
-//#include "acquire_release_volatile.h"
-//#include "char_acquire_release_volatile.h"
-//#include "short_acquire_release_volatile.h"
-//#include "int_acquire_release_volatile.h"
-
-/*
- * This file adds definitions appropriate for environments in which an AO_t
- * volatile load has acquire semantics, and an AO_t volatile store has release
- * semantics.  This is arguably supposed to be true with the standard Itanium
- * software conventions.
- */
-
-/*
- * Empirically gcc/ia64 does some reordering of ordinary operations around volatiles
- * even when we think it shouldn't.  Gcc 3.3 and earlier could reorder a volatile store
- * with another store.  As of March 2005, gcc pre-4 reused previously computed
- * common subexpressions across a volatile load.
- * Hence we now add compiler barriers for gcc.
- */
-#if !defined(AO_GCC_BARRIER)
-#  if defined(__GNUC__)
-#    define AO_GCC_BARRIER() AO_compiler_barrier()
-#  else
-#    define AO_GCC_BARRIER()
-#  endif
-#endif
-
-AO_INLINE AO_t
-AO_load_acquire(const volatile AO_t *p)
-{
-  AO_t result = *p;
-  /* A normal volatile load generates an ld.acq         */
-  AO_GCC_BARRIER();
-  return result;
-}
-#define AO_HAVE_load_acquire
-
-AO_INLINE void
-AO_store_release(volatile AO_t *p, AO_t val)
-{
-  AO_GCC_BARRIER();
-  /* A normal volatile store generates an st.rel        */
-  *p = val;
-}
-#define AO_HAVE_store_release
-
-/*
- * This file adds definitions appropriate for environments in which an unsigned char
- * volatile load has acquire semantics, and an unsigned char volatile store has release
- * semantics.  This is true with the standard Itanium ABI.
- */
-#if !defined(AO_GCC_BARRIER)
-#  if defined(__GNUC__)
-#    define AO_GCC_BARRIER() AO_compiler_barrier()
-#  else
-#    define AO_GCC_BARRIER()
-#  endif
-#endif
-
-AO_INLINE unsigned char
-AO_char_load_acquire(const volatile unsigned char *p)
-{
-  unsigned char result = *p;
-  /* A normal volatile load generates an ld.acq         */
-  AO_GCC_BARRIER();
-  return result;
-}
-#define AO_HAVE_char_load_acquire
-
-AO_INLINE void
-AO_char_store_release(volatile unsigned char *p, unsigned char val)
-{
-  AO_GCC_BARRIER();
-  /* A normal volatile store generates an st.rel        */
-  *p = val;
-}
-#define AO_HAVE_char_store_release
-
-/*
- * This file adds definitions appropriate for environments in which an unsigned short
- * volatile load has acquire semantics, and an unsigned short volatile store has release
- * semantics.  This is true with the standard Itanium ABI.
- */
-#if !defined(AO_GCC_BARRIER)
-#  if defined(__GNUC__)
-#    define AO_GCC_BARRIER() AO_compiler_barrier()
-#  else
-#    define AO_GCC_BARRIER()
-#  endif
-#endif
-
-AO_INLINE unsigned short
-AO_short_load_acquire(const volatile unsigned short *p)
-{
-  unsigned short result = *p;
-  /* A normal volatile load generates an ld.acq         */
-  AO_GCC_BARRIER();
-  return result;
-}
-#define AO_HAVE_short_load_acquire
-
-AO_INLINE void
-AO_short_store_release(volatile unsigned short *p, unsigned short val)
-{
-  AO_GCC_BARRIER();
-  /* A normal volatile store generates an st.rel        */
-  *p = val;
-}
-#define AO_HAVE_short_store_release
-
-/*
- * This file adds definitions appropriate for environments in which an unsigned
- * int volatile load has acquire semantics, and an unsigned short volatile
- * store has release semantics.  This is true with the standard Itanium ABI.
- */
-#if !defined(AO_GCC_BARRIER)
-#  if defined(__GNUC__)
-#    define AO_GCC_BARRIER() AO_compiler_barrier()
-#  else
-#    define AO_GCC_BARRIER()
-#  endif
-#endif
-
-AO_INLINE unsigned int
-AO_int_load_acquire(const volatile unsigned int *p)
-{
-  unsigned int result = *p;
-  /* A normal volatile load generates an ld.acq         */
-  AO_GCC_BARRIER();
-  return result;
-}
-#define AO_HAVE_int_load_acquire
-
-AO_INLINE void
-AO_int_store_release(volatile unsigned int *p, unsigned int val)
-{
-  AO_GCC_BARRIER();
-  /* A normal volatile store generates an st.rel        */
-  *p = val;
-}
-#define AO_HAVE_int_store_release

c_src/atomic_ops/ao_t_is_int.h

@@ -1,126 +0,0 @@
-/*
- * Copyright (c) 2003-2004 Hewlett-Packard Development Company, L.P.
- *
- * Permission is hereby granted, free of charge, to any person obtaining a copy
- * of this software and associated documentation files (the "Software"), to deal
- * in the Software without restriction, including without limitation the rights
- * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
- * copies of the Software, and to permit persons to whom the Software is
- * furnished to do so, subject to the following conditions:
- *
- * The above copyright notice and this permission notice shall be included in
- * all copies or substantial portions of the Software.
- *
- * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
- * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
- * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
- * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
- * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
- * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
- * SOFTWARE.
- */
-
-/*
- * Inclusion of this file signifies that AO_t is in fact int.  Hence
- * any AO_... operations can also server as AO_int_... operations.
- * We currently define only the more important ones here, and allow for
- * the normal generalization process to define the others.
- * We should probably add others in the future.
- */
-
-#if defined(AO_HAVE_compare_and_swap_full) && \
-    !defined(AO_HAVE_int_compare_and_swap_full)
-#  define AO_int_compare_and_swap_full(addr, old, new_val) \
-                AO_compare_and_swap_full((volatile AO_t *)(addr), \
-                                        (AO_t)(old), (AO_t)(new_val))
-#  define AO_HAVE_int_compare_and_swap_full
-# endif
-
-#if defined(AO_HAVE_compare_and_swap_acquire) && \
-    !defined(AO_HAVE_int_compare_and_swap_acquire)
-#  define AO_int_compare_and_swap_acquire(addr, old, new_val) \
-                AO_compare_and_swap_acquire((volatile AO_t *)(addr), \
-                                            (AO_t)(old), (AO_t)(new_val))
-#  define AO_HAVE_int_compare_and_swap_acquire
-# endif
-
-#if defined(AO_HAVE_compare_and_swap_release) && \
-    !defined(AO_HAVE_int_compare_and_swap_release)
-#  define AO_int_compare_and_swap_release(addr, old, new_val) \
-                AO_compare_and_swap_release((volatile AO_t *)(addr), \
-                                         (AO_t)(old), (AO_t)(new_val))
-#  define AO_HAVE_int_compare_and_swap_release
-# endif
-
-#if defined(AO_HAVE_compare_and_swap_write) && \
-    !defined(AO_HAVE_int_compare_and_swap_write)
-#  define AO_int_compare_and_swap_write(addr, old, new_val) \
-                AO_compare_and_swap_write((volatile AO_t *)(addr), \
-                                          (AO_t)(old), (AO_t)(new_val))
-#  define AO_HAVE_int_compare_and_swap_write
-# endif
-
-#if defined(AO_HAVE_compare_and_swap_read) && \
-    !defined(AO_HAVE_int_compare_and_swap_read)
-#  define AO_int_compare_and_swap_read(addr, old, new_val) \
-                AO_compare_and_swap_read((volatile AO_t *)(addr), \
-                                         (AO_t)(old), (AO_t)(new_val))
-#  define AO_HAVE_int_compare_and_swap_read
-# endif
-
-#if defined(AO_HAVE_compare_and_swap) && \
-    !defined(AO_HAVE_int_compare_and_swap)
-#  define AO_int_compare_and_swap(addr, old, new_val) \
-                AO_compare_and_swap((volatile AO_t *)(addr), \
-                                    (AO_t)(old), (AO_t)(new_val))
-#  define AO_HAVE_int_compare_and_swap
-# endif
-
-#if defined(AO_HAVE_load_acquire) && \
-    !defined(AO_HAVE_int_load_acquire)
-#  define AO_int_load_acquire(addr) \
-        (int)AO_load_acquire((const volatile AO_t *)(addr))
-#  define AO_HAVE_int_load_acquire
-# endif
-
-#if defined(AO_HAVE_store_release) && \
-    !defined(AO_HAVE_int_store_release)
-#  define AO_int_store_release(addr, val) \
-        AO_store_release((volatile AO_t *)(addr), (AO_t)(val))
-#  define AO_HAVE_int_store_release
-# endif
-
-#if defined(AO_HAVE_fetch_and_add_full) && \
-    !defined(AO_HAVE_int_fetch_and_add_full)
-#  define AO_int_fetch_and_add_full(addr, incr) \
-        (int)AO_fetch_and_add_full((volatile AO_t *)(addr), (AO_t)(incr))
-#  define AO_HAVE_int_fetch_and_add_full
-# endif
-
-#if defined(AO_HAVE_fetch_and_add1_acquire) && \
-    !defined(AO_HAVE_int_fetch_and_add1_acquire)
-#  define AO_int_fetch_and_add1_acquire(addr) \
-        (int)AO_fetch_and_add1_acquire((volatile AO_t *)(addr))
-#  define AO_HAVE_int_fetch_and_add1_acquire
-# endif
-
-#if defined(AO_HAVE_fetch_and_add1_release) && \
-    !defined(AO_HAVE_int_fetch_and_add1_release)
-#  define AO_int_fetch_and_add1_release(addr) \
-        (int)AO_fetch_and_add1_release((volatile AO_t *)(addr))
-#  define AO_HAVE_int_fetch_and_add1_release
-# endif
-
-#if defined(AO_HAVE_fetch_and_sub1_acquire) && \
-    !defined(AO_HAVE_int_fetch_and_sub1_acquire)
-#  define AO_int_fetch_and_sub1_acquire(addr) \
-        (int)AO_fetch_and_sub1_acquire((volatile AO_t *)(addr))
-#  define AO_HAVE_int_fetch_and_sub1_acquire
-# endif
-
-#if defined(AO_HAVE_fetch_and_sub1_release) && \
-    !defined(AO_HAVE_int_fetch_and_sub1_release)
-#  define AO_int_fetch_and_sub1_release(addr) \
-        (int)AO_fetch_and_sub1_release((volatile AO_t *)(addr))
-#  define AO_HAVE_int_fetch_and_sub1_release
-# endif

c_src/atomic_ops/atomic_ops.h

@@ -1,348 +0,0 @@
-/*
- * Copyright (c) 2003 Hewlett-Packard Development Company, L.P.
- *
- * Permission is hereby granted, free of charge, to any person obtaining a copy
- * of this software and associated documentation files (the "Software"), to deal
- * in the Software without restriction, including without limitation the rights
- * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
- * copies of the Software, and to permit persons to whom the Software is
- * furnished to do so, subject to the following conditions:
- *
- * The above copyright notice and this permission notice shall be included in
- * all copies or substantial portions of the Software.
- *
- * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
- * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
- * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
- * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
- * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
- * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
- * SOFTWARE.
- */
-
-#ifndef ATOMIC_OPS_H
-
-#define ATOMIC_OPS_H
-
-#include <assert.h>
-#include <stddef.h>
-
-/* We define various atomic operations on memory in a           */
-/* machine-specific way.  Unfortunately, this is complicated    */
-/* by the fact that these may or may not be combined with       */
-/* various memory barriers.  Thus the actual operations we      */
-/* define have the form AO_<atomic-op>_<barrier>, for all       */
-/* plausible combinations of <atomic-op> and <barrier>.         */
-/* This of course results in a mild combinatorial explosion.    */
-/* To deal with it, we try to generate derived                  */
-/* definitions for as many of the combinations as we can, as    */
-/* automatically as possible.                                   */
-/*                                                              */
-/* Our assumption throughout is that the programmer will        */
-/* specify the least demanding operation and memory barrier     */
-/* that will guarantee correctness for the implementation.      */
-/* Our job is to find the least expensive way to implement it   */
-/* on the applicable hardware.  In many cases that will         */
-/* involve, for example, a stronger memory barrier, or a        */
-/* combination of hardware primitives.                          */
-/*                                                              */
-/* Conventions:                                                 */
-/* "plain" atomic operations are not guaranteed to include      */
-/* a barrier.  The suffix in the name specifies the barrier     */
-/* type.  Suffixes are:                                         */
-/* _release: Earlier operations may not be delayed past it.     */
-/* _acquire: Later operations may not move ahead of it.         */
-/* _read: Subsequent reads must follow this operation and       */
-/*        preceding reads.                                      */
-/* _write: Earlier writes precede both this operation and       */
-/*        later writes.                                         */
-/* _full: Ordered with respect to both earlier and later memops.*/
-/* _release_write: Ordered with respect to earlier writes.      */
-/* _acquire_read: Ordered with respect to later reads.          */
-/*                                                              */
-/* Currently we try to define the following atomic memory       */
-/* operations, in combination with the above barriers:          */
-/* AO_nop                                                       */
-/* AO_load                                                      */
-/* AO_store                                                     */
-/* AO_test_and_set (binary)                                     */
-/* AO_fetch_and_add                                             */
-/* AO_fetch_and_add1                                            */
-/* AO_fetch_and_sub1                                            */
-/* AO_or                                                        */
-/* AO_compare_and_swap                                          */
-/*                                                              */
-/* Note that atomicity guarantees are valid only if both        */
-/* readers and writers use AO_ operations to access the         */
-/* shared value, while ordering constraints are intended to     */
-/* apply all memory operations.  If a location can potentially  */
-/* be accessed simultaneously from multiple threads, and one of */
-/* those accesses may be a write access, then all such          */
-/* accesses to that location should be through AO_ primitives.  */
-/* However if AO_ operations enforce sufficient ordering to     */
-/* ensure that a location x cannot be accessed concurrently,    */
-/* or can only be read concurrently, then x can be accessed     */
-/* via ordinary references and assignments.                     */
-/*                                                              */
-/* Compare_and_exchange takes an address and an expected old    */
-/* value and a new value, and returns an int.  Nonzero          */
-/* indicates that it succeeded.                                 */
-/* Test_and_set takes an address, atomically replaces it by     */
-/* AO_TS_SET, and returns the prior value.                      */
-/* An AO_TS_t location can be reset with the                    */
-/* AO_CLEAR macro, which normally uses AO_store_release.        */
-/* AO_fetch_and_add takes an address and an AO_t increment      */
-/* value.  The AO_fetch_and_add1 and AO_fetch_and_sub1 variants */
-/* are provided, since they allow faster implementations on     */
-/* some hardware. AO_or atomically ors an AO_t value into a     */
-/* memory location, but does not provide access to the original.*/
-/*                                                              */
-/* We expect this list to grow slowly over time.                */
-/*                                                              */
-/* Note that AO_nop_full is a full memory barrier.              */
-/*                                                              */
-/* Note that if some data is initialized with                   */
-/*      data.x = ...; data.y = ...; ...                         */
-/*      AO_store_release_write(&data_is_initialized, 1)         */
-/* then data is guaranteed to be initialized after the test     */
-/*      if (AO_load_release_read(&data_is_initialized)) ...     */
-/* succeeds.  Furthermore, this should generate near-optimal    */
-/* code on all common platforms.                                */
-/*                                                              */
-/* All operations operate on unsigned AO_t, which               */
-/* is the natural word size, and usually unsigned long.         */
-/* It is possible to check whether a particular operation op    */
-/* is available on a particular platform by checking whether    */
-/* AO_HAVE_op is defined.  We make heavy use of these macros    */
-/* internally.                                                  */
-
-/* The rest of this file basically has three sections:          */
-/*                                                              */
-/* Some utility and default definitions.                        */
-/*                                                              */
-/* The architecture dependent section:                          */
-/* This defines atomic operations that have direct hardware     */
-/* support on a particular platform, mostly by including the    */
-/* appropriate compiler- and hardware-dependent file.           */
-/*                                                              */
-/* The synthesis section:                                       */
-/* This tries to define other atomic operations in terms of     */
-/* those that are explicitly available on the platform.         */
-/* This section is hardware independent.                        */
-/* We make no attempt to synthesize operations in ways that     */
-/* effectively introduce locks, except for the debugging/demo   */
-/* pthread-based implementation at the beginning.  A more       */
-/* realistic implementation that falls back to locks could be   */
-/* added as a higher layer.  But that would sacrifice           */
-/* usability from signal handlers.                              */
-/* The synthesis section is implemented almost entirely in      */
-/* atomic_ops_generalize.h.                                     */
-
-/* Some common defaults.  Overridden for some architectures.    */
-#define AO_t size_t
-
-/* The test_and_set primitive returns an AO_TS_VAL_t value.     */
-/* AO_TS_t is the type of an in-memory test-and-set location.   */
-
-#define AO_TS_INITIALIZER (AO_t)AO_TS_CLEAR
-
-/* Platform-dependent stuff:                                    */
-#if defined(__GNUC__) || defined(_MSC_VER) || defined(__INTEL_COMPILER) \
-        || defined(__DMC__) || defined(__WATCOMC__)
-# define AO_INLINE static __inline
-#elif defined(__sun)
-# define AO_INLINE static inline
-#else
-# define AO_INLINE static
-#endif
-
-#if defined(__GNUC__) && !defined(__INTEL_COMPILER)
-# define AO_compiler_barrier() __asm__ __volatile__("" : : : "memory")
-#elif defined(_MSC_VER) || defined(__DMC__) || defined(__BORLANDC__) \
-        || defined(__WATCOMC__)
-# if defined(_AMD64_) || defined(_M_X64) || _MSC_VER >= 1400
-#   if defined(_WIN32_WCE)
-/* #     include <cmnintrin.h> */
-#   elif defined(_MSC_VER)
-#     include <intrin.h>
-#   endif
-#   pragma intrinsic(_ReadWriteBarrier)
-#   define AO_compiler_barrier() _ReadWriteBarrier()
-        /* We assume this does not generate a fence instruction.        */
-        /* The documentation is a bit unclear.                          */
-# else
-#   define AO_compiler_barrier() __asm { }
-        /* The preceding implementation may be preferable here too.     */
-        /* But the documentation warns about VC++ 2003 and earlier.     */
-# endif
-#elif defined(__INTEL_COMPILER)
-# define AO_compiler_barrier() __memory_barrier() /* Too strong? IA64-only? */
-#elif defined(_HPUX_SOURCE)
-# if defined(__ia64)
-#   include <machine/sys/inline.h>
-#   define AO_compiler_barrier() _Asm_sched_fence()
-# else
-    /* FIXME - We dont know how to do this.  This is a guess.   */
-    /* And probably a bad one.                                  */
-    static volatile int AO_barrier_dummy;
-#   define AO_compiler_barrier() AO_barrier_dummy = AO_barrier_dummy
-# endif
-#else
-  /* We conjecture that the following usually gives us the right        */
-  /* semantics or an error.                                             */
-# define AO_compiler_barrier() asm("")
-#endif
-
-#if defined(AO_USE_PTHREAD_DEFS)
-# include "atomic_ops/sysdeps/generic_pthread.h"
-#endif /* AO_USE_PTHREAD_DEFS */
-
-#if defined(__GNUC__) && !defined(AO_USE_PTHREAD_DEFS) \
-    && !defined(__INTEL_COMPILER)
-# if defined(__i386__)
-    /* We don't define AO_USE_SYNC_CAS_BUILTIN for x86 here because     */
-    /* it might require specifying additional options (like -march)     */
-    /* or additional link libraries (if -march is not specified).       */
-#   include "./x86.h"
-# endif /* __i386__ */
-# if defined(__x86_64__)
-#   if __GNUC__ > 4 || (__GNUC__ == 4 && __GNUC_MINOR__ >= 2)
-      /* It is safe to use __sync CAS built-in on this architecture.    */
-#     define AO_USE_SYNC_CAS_BUILTIN
-#   endif
-#   include "./x86_64.h"
-# endif /* __x86_64__ */
-# if defined(__ia64__)
-#   include "./ia64.h"
-#   define AO_GENERALIZE_TWICE
-# endif /* __ia64__ */
-# if defined(__hppa__)
-#   include "atomic_ops/sysdeps/gcc/hppa.h"
-#   define AO_CAN_EMUL_CAS
-# endif /* __hppa__ */
-# if defined(__alpha__)
-#   include "atomic_ops/sysdeps/gcc/alpha.h"
-#   define AO_GENERALIZE_TWICE
-# endif /* __alpha__ */
-# if defined(__s390__)
-#   include "atomic_ops/sysdeps/gcc/s390.h"
-# endif /* __s390__ */
-# if defined(__sparc__)
-#   include "./sparc.h"
-#   define AO_CAN_EMUL_CAS
-# endif /* __sparc__ */
-# if defined(__m68k__)
-#   include "atomic_ops/sysdeps/gcc/m68k.h"
-# endif /* __m68k__ */
-# if defined(__powerpc__) || defined(__ppc__) || defined(__PPC__) \
-     || defined(__powerpc64__) || defined(__ppc64__)
-#   include "./powerpc.h"
-# endif /* __powerpc__ */
-# if defined(__arm__) && !defined(AO_USE_PTHREAD_DEFS)
-#   include "atomic_ops/sysdeps/gcc/arm.h"
-#   define AO_CAN_EMUL_CAS
-# endif /* __arm__ */
-# if defined(__cris__) || defined(CRIS)
-#   include "atomic_ops/sysdeps/gcc/cris.h"
-# endif
-# if defined(__mips__)
-#   include "atomic_ops/sysdeps/gcc/mips.h"
-# endif /* __mips__ */
-# if defined(__sh__) || defined(SH4)
-#   include "atomic_ops/sysdeps/gcc/sh.h"
-#   define AO_CAN_EMUL_CAS
-# endif /* __sh__ */
-#endif /* __GNUC__ && !AO_USE_PTHREAD_DEFS */
-
-#if defined(__INTEL_COMPILER) && !defined(AO_USE_PTHREAD_DEFS)
-# if defined(__ia64__)
-#   include "./ia64.h"
-#   define AO_GENERALIZE_TWICE
-# endif
-# if defined(__GNUC__)
-    /* Intel Compiler in GCC compatible mode */
-#   if defined(__i386__)
-#     include "./x86.h"
-#   endif /* __i386__ */
-#   if defined(__x86_64__)
-#     if __INTEL_COMPILER > 1110
-#       define AO_USE_SYNC_CAS_BUILTIN
-#     endif
-#     include "./x86_64.h"
-#   endif /* __x86_64__ */
-# endif
-#endif
-
-#if defined(_HPUX_SOURCE) && !defined(__GNUC__) && !defined(AO_USE_PTHREAD_DEFS)
-# if defined(__ia64)
-#   include "atomic_ops/sysdeps/hpc/ia64.h"
-#   define AO_GENERALIZE_TWICE
-# else
-#   include "atomic_ops/sysdeps/hpc/hppa.h"
-#   define AO_CAN_EMUL_CAS
-# endif
-#endif
-
-#if defined(__sun) && !defined(__GNUC__) && !defined(AO_USE_PTHREAD_DEFS)
-  /* Note: use -DAO_USE_PTHREAD_DEFS if Sun CC does not handle inline asm. */
-# if defined(__i386)
-#   include "atomic_ops/sysdeps/sunc/x86.h"
-# endif /* __i386 */
-# if defined(__x86_64) || defined(__amd64)
-#   include "atomic_ops/sysdeps/sunc/x86_64.h"
-# endif /* __x86_64 */
-#endif
-
-#if !defined(__GNUC__) && (defined(sparc) || defined(__sparc)) \
-    && !defined(AO_USE_PTHREAD_DEFS)
-#   include "atomic_ops/sysdeps/sunc/sparc.h"
-#   define AO_CAN_EMUL_CAS
-#endif
-
-#if defined(_MSC_VER) || defined(__DMC__) || defined(__BORLANDC__) \
-        || (defined(__WATCOMC__) && defined(__NT__))
-# if defined(_AMD64_) || defined(_M_X64)
-#   include "atomic_ops/sysdeps/msftc/x86_64.h"
-# elif defined(_M_IX86) || defined(x86)
-#   include "atomic_ops/sysdeps/msftc/x86.h"
-# elif defined(_M_ARM) || defined(ARM) || defined(_ARM_)
-#   include "atomic_ops/sysdeps/msftc/arm.h"
-# endif
-#endif
-
-#if defined(AO_REQUIRE_CAS) && !defined(AO_HAVE_compare_and_swap) \
-    && !defined(AO_HAVE_compare_and_swap_full) \
-    && !defined(AO_HAVE_compare_and_swap_acquire)
-# if defined(AO_CAN_EMUL_CAS)
-#   include "atomic_ops/sysdeps/emul_cas.h"
-# else
-#  error Cannot implement AO_compare_and_swap_full on this architecture.
-# endif
-#endif  /* AO_REQUIRE_CAS && !AO_HAVE_compare_and_swap ... */
-
-/* The most common way to clear a test-and-set location         */
-/* at the end of a critical section.                            */
-#if AO_AO_TS_T && !defined(AO_CLEAR)
-# define AO_CLEAR(addr) AO_store_release((AO_TS_t *)(addr), AO_TS_CLEAR)
-#endif
-#if AO_CHAR_TS_T && !defined(AO_CLEAR)
-# define AO_CLEAR(addr) AO_char_store_release((AO_TS_t *)(addr), AO_TS_CLEAR)
-#endif
-
-/*
- * The generalization section.
- * Theoretically this should repeatedly include atomic_ops_generalize.h.
- * In fact, we observe that this converges after a small fixed number
- * of iterations, usually one.
- */
-#include "./generalize.h"
-#ifdef AO_GENERALIZE_TWICE
-# include "./generalize.h"
-#endif
-
-/* For compatibility with version 0.4 and earlier       */
-#define AO_TS_T AO_TS_t
-#define AO_T AO_t
-#define AO_TS_VAL AO_TS_VAL_t
-
-#endif /* ATOMIC_OPS_H */

c_src/atomic_ops/ia64.h

@@ -1,297 +0,0 @@
-/*
- * Copyright (c) 2003 Hewlett-Packard Development Company, L.P.
- *
- * Permission is hereby granted, free of charge, to any person obtaining a copy
- * of this software and associated documentation files (the "Software"), to deal
- * in the Software without restriction, including without limitation the rights
- * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
- * copies of the Software, and to permit persons to whom the Software is
- * furnished to do so, subject to the following conditions:
- *
- * The above copyright notice and this permission notice shall be included in
- * all copies or substantial portions of the Software.
- *
- * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
- * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
- * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
- * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
- * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
- * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
- * SOFTWARE.
- */
-
-#include "./aligned_atomic_load_store.h"
-
-#include "./all_acquire_release_volatile.h"
-
-#include "./test_and_set_t_is_char.h"
-
-#ifdef _ILP32
-  /* 32-bit HP/UX code. */
-  /* This requires pointer "swizzling".  Pointers need to be expanded   */
-  /* to 64 bits using the addp4 instruction before use.  This makes it  */
-  /* hard to share code, but we try anyway.                             */
-# define AO_LEN "4"
-  /* We assume that addr always appears in argument position 1 in asm   */
-  /* code.  If it is clobbered due to swizzling, we also need it in     */
-  /* second position.  Any later arguments are referenced symbolically, */
-  /* so that we don't have to worry about their position.  This requires*/
-  /* gcc 3.1, but you shouldn't be using anything older than that on    */
-  /* IA64 anyway.                                                       */
-  /* The AO_MASK macro is a workaround for the fact that HP/UX gcc      */
-  /* appears to otherwise store 64-bit pointers in ar.ccv, i.e. it      */
-  /* doesn't appear to clear high bits in a pointer value we pass into  */
-  /* assembly code, even if it is supposedly of type AO_t.              */
-# define AO_IN_ADDR "1"(addr)
-# define AO_OUT_ADDR , "=r"(addr)
-# define AO_SWIZZLE "addp4 %1=0,%1;;\n"
-# define AO_MASK(ptr) __asm__("zxt4 %1=%1": "=r"(ptr) : "0"(ptr));
-#else
-# define AO_LEN "8"
-# define AO_IN_ADDR "r"(addr)
-# define AO_OUT_ADDR
-# define AO_SWIZZLE
-# define AO_MASK(ptr)
-#endif
-
-AO_INLINE void
-AO_nop_full(void)
-{
-  __asm__ __volatile__("mf" : : : "memory");
-}
-#define AO_HAVE_nop_full
-
-AO_INLINE AO_t
-AO_fetch_and_add1_acquire (volatile AO_t *addr)
-{
-  AO_t result;
-
-  __asm__ __volatile__ (AO_SWIZZLE
-                        "fetchadd" AO_LEN ".acq %0=[%1],1":
-                        "=r" (result) AO_OUT_ADDR: AO_IN_ADDR :"memory");
-  return result;
-}
-#define AO_HAVE_fetch_and_add1_acquire
-
-AO_INLINE AO_t
-AO_fetch_and_add1_release (volatile AO_t *addr)
-{
-  AO_t result;
-
-  __asm__ __volatile__ (AO_SWIZZLE
-                        "fetchadd" AO_LEN ".rel %0=[%1],1":
-                        "=r" (result) AO_OUT_ADDR: AO_IN_ADDR :"memory");
-  return result;
-}
-
-#define AO_HAVE_fetch_and_add1_release
-
-AO_INLINE AO_t
-AO_fetch_and_sub1_acquire (volatile AO_t *addr)
-{
-  AO_t result;
-
-  __asm__ __volatile__ (AO_SWIZZLE
-                        "fetchadd" AO_LEN ".acq %0=[%1],-1":
-                        "=r" (result) AO_OUT_ADDR: AO_IN_ADDR :"memory");
-  return result;
-}
-
-#define AO_HAVE_fetch_and_sub1_acquire
-
-AO_INLINE AO_t
-AO_fetch_and_sub1_release (volatile AO_t *addr)
-{
-  AO_t result;
-
-  __asm__ __volatile__ (AO_SWIZZLE
-                        "fetchadd" AO_LEN ".rel %0=[%1],-1":
-                        "=r" (result) AO_OUT_ADDR: AO_IN_ADDR :"memory");
-  return result;
-}
-
-#define AO_HAVE_fetch_and_sub1_release
-
-#ifndef _ILP32
-
-AO_INLINE unsigned int
-AO_int_fetch_and_add1_acquire (volatile unsigned int *addr)
-{
-  unsigned int result;
-
-  __asm__ __volatile__ ("fetchadd4.acq %0=[%1],1":
-                        "=r" (result): AO_IN_ADDR :"memory");
-  return result;
-}
-#define AO_HAVE_int_fetch_and_add1_acquire
-
-AO_INLINE unsigned int
-AO_int_fetch_and_add1_release (volatile unsigned int *addr)
-{
-  unsigned int result;
-
-  __asm__ __volatile__ ("fetchadd4.rel %0=[%1],1":
-                        "=r" (result): AO_IN_ADDR :"memory");
-  return result;
-}
-
-#define AO_HAVE_int_fetch_and_add1_release
-
-AO_INLINE unsigned int
-AO_int_fetch_and_sub1_acquire (volatile unsigned int *addr)
-{
-  unsigned int result;
-
-  __asm__ __volatile__ ("fetchadd4.acq %0=[%1],-1":
-                        "=r" (result): AO_IN_ADDR :"memory");
-  return result;
-}
-
-#define AO_HAVE_int_fetch_and_sub1_acquire
-
-AO_INLINE unsigned int
-AO_int_fetch_and_sub1_release (volatile unsigned int *addr)
-{
-  unsigned int result;
-
-  __asm__ __volatile__ ("fetchadd4.rel %0=[%1],-1":
-                        "=r" (result): AO_IN_ADDR :"memory");
-  return result;
-}
-
-#define AO_HAVE_int_fetch_and_sub1_release
-
-#endif /* !_ILP32 */
-
-AO_INLINE int
-AO_compare_and_swap_acquire(volatile AO_t *addr,
-                             AO_t old, AO_t new_val)
-{
-  AO_t oldval;
-  AO_MASK(old);
-  __asm__ __volatile__(AO_SWIZZLE
-                       "mov ar.ccv=%[old] ;; cmpxchg" AO_LEN
-                       ".acq %0=[%1],%[new_val],ar.ccv"
-                       : "=r"(oldval) AO_OUT_ADDR
-                       : AO_IN_ADDR, [new_val]"r"(new_val), [old]"r"(old)
-                       : "memory");
-  return (oldval == old);
-}
-
-#define AO_HAVE_compare_and_swap_acquire
-
-AO_INLINE int
-AO_compare_and_swap_release(volatile AO_t *addr,
-                             AO_t old, AO_t new_val)
-{
-  AO_t oldval;
-  AO_MASK(old);
-  __asm__ __volatile__(AO_SWIZZLE
-                       "mov ar.ccv=%[old] ;; cmpxchg" AO_LEN
-                       ".rel %0=[%1],%[new_val],ar.ccv"
-                       : "=r"(oldval) AO_OUT_ADDR
-                       : AO_IN_ADDR, [new_val]"r"(new_val), [old]"r"(old)
-                       : "memory");
-  return (oldval == old);
-}
-
-#define AO_HAVE_compare_and_swap_release
-
-AO_INLINE int
-AO_char_compare_and_swap_acquire(volatile unsigned char *addr,
-                                 unsigned char old, unsigned char new_val)
-{
-  unsigned char oldval;
-  __asm__ __volatile__(AO_SWIZZLE
-               "mov ar.ccv=%[old] ;; cmpxchg1.acq %0=[%1],%[new_val],ar.ccv"
-               : "=r"(oldval) AO_OUT_ADDR
-               : AO_IN_ADDR, [new_val]"r"(new_val), [old]"r"((AO_t)old)
-               : "memory");
-  return (oldval == old);
-}
-
-#define AO_HAVE_char_compare_and_swap_acquire
-
-AO_INLINE int
-AO_char_compare_and_swap_release(volatile unsigned char *addr,
-                                 unsigned char old, unsigned char new_val)
-{
-  unsigned char oldval;
-  __asm__ __volatile__(AO_SWIZZLE
-                "mov ar.ccv=%[old] ;; cmpxchg1.rel %0=[%1],%[new_val],ar.ccv"
-                : "=r"(oldval) AO_OUT_ADDR
-                : AO_IN_ADDR, [new_val]"r"(new_val), [old]"r"((AO_t)old)
-                : "memory");
-  return (oldval == old);
-}
-
-#define AO_HAVE_char_compare_and_swap_release
-
-AO_INLINE int
-AO_short_compare_and_swap_acquire(volatile unsigned short *addr,
-                                  unsigned short old, unsigned short new_val)
-{
-  unsigned short oldval;
-  __asm__ __volatile__(AO_SWIZZLE
-                "mov ar.ccv=%[old] ;; cmpxchg2.acq %0=[%1],%[new_val],ar.ccv"
-                : "=r"(oldval) AO_OUT_ADDR
-                : AO_IN_ADDR, [new_val]"r"(new_val), [old]"r"((AO_t)old)
-                : "memory");
-  return (oldval == old);
-}
-
-#define AO_HAVE_short_compare_and_swap_acquire
-
-AO_INLINE int
-AO_short_compare_and_swap_release(volatile unsigned short *addr,
-                                  unsigned short old, unsigned short new_val)
-{
-  unsigned short oldval;
-  __asm__ __volatile__(AO_SWIZZLE
-                "mov ar.ccv=%[old] ;; cmpxchg2.rel %0=[%1],%[new_val],ar.ccv"
-                : "=r"(oldval) AO_OUT_ADDR
-                : AO_IN_ADDR, [new_val]"r"(new_val), [old]"r"((AO_t)old)
-                : "memory");
-  return (oldval == old);
-}
-
-#define AO_HAVE_short_compare_and_swap_release
-
-#ifndef _ILP32
-
-AO_INLINE int
-AO_int_compare_and_swap_acquire(volatile unsigned int *addr,
-                                unsigned int old, unsigned int new_val)
-{
-  unsigned int oldval;
-  __asm__ __volatile__("mov ar.ccv=%3 ;; cmpxchg4.acq %0=[%1],%2,ar.ccv"
-                       : "=r"(oldval)
-                       : AO_IN_ADDR, "r"(new_val), "r"((AO_t)old) : "memory");
-  return (oldval == old);
-}
-
-#define AO_HAVE_int_compare_and_swap_acquire
-
-AO_INLINE int
-AO_int_compare_and_swap_release(volatile unsigned int *addr,
-                                unsigned int old, unsigned int new_val)
-{
-  unsigned int oldval;
-  __asm__ __volatile__("mov ar.ccv=%3 ;; cmpxchg4.rel %0=[%1],%2,ar.ccv"
-                       : "=r"(oldval)
-                       : AO_IN_ADDR, "r"(new_val), "r"((AO_t)old) : "memory");
-  return (oldval == old);
-}
-
-#define AO_HAVE_int_compare_and_swap_release
-
-#endif /* !_ILP32 */
-
-/* FIXME: Add compare_and_swap_double as soon as there is widely        */
-/* available hardware that implements it.                               */
-
-/* FIXME: Add compare_double_and_swap_double for the _ILP32 case.       */
-
-#ifdef _ILP32
-# include "./ao_t_is_int.h"
-#endif

c_src/atomic_ops/ordered_except_wr.h

@@ -1,100 +0,0 @@
-/*
- * Copyright (c) 2003 by Hewlett-Packard Company.  All rights reserved.
- *
- * Permission is hereby granted, free of charge, to any person obtaining a copy
- * of this software and associated documentation files (the "Software"), to deal
- * in the Software without restriction, including without limitation the rights
- * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
- * copies of the Software, and to permit persons to whom the Software is
- * furnished to do so, subject to the following conditions:
- *
- * The above copyright notice and this permission notice shall be included in
- * all copies or substantial portions of the Software.
- *
- * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
- * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
- * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
- * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
- * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
- * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
- * SOFTWARE.
- */
-
-/*
- * These are common definitions for architectures that provide processor
- * ordered memory operations except that a later read may pass an
- * earlier write.  Real x86 implementations seem to be in this category,
- * except apparently for some IDT WinChips, which we ignore.
- */
-
-#include "read_ordered.h"
-
-AO_INLINE void
-AO_nop_write(void)
-{
-  AO_compiler_barrier();
-  /* sfence according to Intel docs.  Pentium 3 and up. */
-  /* Unnecessary for cached accesses?                   */
-}
-
-#define AO_HAVE_NOP_WRITE
-
-#if defined(AO_HAVE_store)
-
-AO_INLINE void
-AO_store_write(volatile AO_t *addr, AO_t val)
-{
-  AO_compiler_barrier();
-  AO_store(addr, val);
-}
-# define AO_HAVE_store_write
-
-# define AO_store_release(addr, val) AO_store_write(addr, val)
-# define AO_HAVE_store_release
-
-#endif /* AO_HAVE_store */
-
-#if defined(AO_HAVE_char_store)
-
-AO_INLINE void
-AO_char_store_write(volatile unsigned char *addr, unsigned char val)
-{
-  AO_compiler_barrier();
-  AO_char_store(addr, val);
-}
-# define AO_HAVE_char_store_write
-
-# define AO_char_store_release(addr, val) AO_char_store_write(addr, val)
-# define AO_HAVE_char_store_release
-
-#endif /* AO_HAVE_char_store */
-
-#if defined(AO_HAVE_short_store)
-
-AO_INLINE void
-AO_short_store_write(volatile unsigned short *addr, unsigned short val)
-{
-  AO_compiler_barrier();
-  AO_short_store(addr, val);
-}
-# define AO_HAVE_short_store_write
-
-# define AO_short_store_release(addr, val) AO_short_store_write(addr, val)
-# define AO_HAVE_short_store_release
-
-#endif /* AO_HAVE_short_store */
-
-#if defined(AO_HAVE_int_store)
-
-AO_INLINE void
-AO_int_store_write(volatile unsigned int *addr, unsigned int val)
-{
-  AO_compiler_barrier();
-  AO_int_store(addr, val);
-}
-# define AO_HAVE_int_store_write
-
-# define AO_int_store_release(addr, val) AO_int_store_write(addr, val)
-# define AO_HAVE_int_store_release
-
-#endif /* AO_HAVE_int_store */

c_src/atomic_ops/powerpc.h

@@ -1,346 +0,0 @@
-/*
- * Copyright (c) 1991-1994 by Xerox Corporation.  All rights reserved.
- * Copyright (c) 1996-1999 by Silicon Graphics.  All rights reserved.
- * Copyright (c) 1999-2004 Hewlett-Packard Development Company, L.P.
- *
- *
- * THIS MATERIAL IS PROVIDED AS IS, WITH ABSOLUTELY NO WARRANTY EXPRESSED
- * OR IMPLIED.  ANY USE IS AT YOUR OWN RISK.
- *
- * Permission is hereby granted to use or copy this program
- * for any purpose,  provided the above notices are retained on all copies.
- * Permission to modify the code and to distribute modified code is granted,
- * provided the above notices are retained, and a notice that the code was
- * modified is included with the above copyright notice.
- *
- */
-
-/* Memory model documented at http://www-106.ibm.com/developerworks/    */
-/* eserver/articles/archguide.html and (clearer)                        */
-/* http://www-106.ibm.com/developerworks/eserver/articles/powerpc.html. */
-/* There appears to be no implicit ordering between any kind of         */
-/* independent memory references.                                       */
-/* Architecture enforces some ordering based on control dependence.     */
-/* I don't know if that could help.                                     */
-/* Data-dependent loads are always ordered.                             */
-/* Based on the above references, eieio is intended for use on          */
-/* uncached memory, which we don't support.  It does not order loads    */
-/* from cached memory.                                                  */
-/* Thanks to Maged Michael, Doug Lea, and Roger Hoover for helping to   */
-/* track some of this down and correcting my misunderstandings. -HB     */
-/* Earl Chew subsequently contributed further fixes & additions.        */
-
-#include "./aligned_atomic_load_store.h"
-
-#include "./test_and_set_t_is_ao_t.h"
-        /* There seems to be no byte equivalent of lwarx, so this       */
-        /* may really be what we want, at least in the 32-bit case.     */
-
-AO_INLINE void
-AO_nop_full(void)
-{
-  __asm__ __volatile__("sync" : : : "memory");
-}
-
-#define AO_HAVE_nop_full
-
-/* lwsync apparently works for everything but a StoreLoad barrier.      */
-AO_INLINE void
-AO_lwsync(void)
-{
-#ifdef __NO_LWSYNC__
-  __asm__ __volatile__("sync" : : : "memory");
-#else
-  __asm__ __volatile__("lwsync" : : : "memory");
-#endif
-}
-
-#define AO_nop_write() AO_lwsync()
-#define AO_HAVE_nop_write
-
-#define AO_nop_read() AO_lwsync()
-#define AO_HAVE_nop_read
-
-/* We explicitly specify load_acquire, since it is important, and can   */
-/* be implemented relatively cheaply.  It could be implemented          */
-/* with an ordinary load followed by a lwsync.  But the general wisdom  */
-/* seems to be that a data dependent branch followed by an isync is     */
-/* cheaper.  And the documentation is fairly explicit that this also    */
-/* has acquire semantics.                                               */
-/* ppc64 uses ld not lwz */
-#if defined(__powerpc64__) || defined(__ppc64__) || defined(__64BIT__)
-AO_INLINE AO_t
-AO_load_acquire(const volatile AO_t *addr)
-{
-  AO_t result;
-
-   __asm__ __volatile__ (
-    "ld%U1%X1 %0,%1\n"
-    "cmpw %0,%0\n"
-    "bne- 1f\n"
-    "1: isync\n"
-    : "=r" (result)
-    : "m"(*addr) : "memory", "cr0");
-  return result;
-}
-#else
-AO_INLINE AO_t
-AO_load_acquire(const volatile AO_t *addr)
-{
-  AO_t result;
-
-  /* FIXME: We should get gcc to allocate one of the condition  */
-  /* registers.  I always got "impossible constraint" when I    */
-  /* tried the "y" constraint.                                  */
-  __asm__ __volatile__ (
-    "lwz%U1%X1 %0,%1\n"
-    "cmpw %0,%0\n"
-    "bne- 1f\n"
-    "1: isync\n"
-    : "=r" (result)
-    : "m"(*addr) : "memory", "cc");
-  return result;
-}
-#endif
-#define AO_HAVE_load_acquire
-
-/* We explicitly specify store_release, since it relies         */
-/* on the fact that lwsync is also a LoadStore barrier.         */
-AO_INLINE void
-AO_store_release(volatile AO_t *addr, AO_t value)
-{
-  AO_lwsync();
-  *addr = value;
-}
-
-#define AO_HAVE_load_acquire
-
-/* This is similar to the code in the garbage collector.  Deleting      */
-/* this and having it synthesized from compare_and_swap would probably  */
-/* only cost us a load immediate instruction.                           */
-#if defined(__powerpc64__) || defined(__ppc64__) || defined(__64BIT__)
-/* Completely untested.  And we should be using smaller objects anyway. */
-AO_INLINE AO_TS_VAL_t
-AO_test_and_set(volatile AO_TS_t *addr) {
-  unsigned long oldval;
-  unsigned long temp = 1; /* locked value */
-
-  __asm__ __volatile__(
-               "1:ldarx %0,0,%1\n"   /* load and reserve               */
-               "cmpdi %0, 0\n"       /* if load is                     */
-               "bne 2f\n"            /*   non-zero, return already set */
-               "stdcx. %2,0,%1\n"    /* else store conditional         */
-               "bne- 1b\n"           /* retry if lost reservation      */
-               "2:\n"                /* oldval is zero if we set       */
-              : "=&r"(oldval)
-              : "r"(addr), "r"(temp)
-              : "memory", "cr0");
-
-  return (AO_TS_VAL_t)oldval;
-}
-
-#else
-
-AO_INLINE AO_TS_VAL_t
-AO_test_and_set(volatile AO_TS_t *addr) {
-  int oldval;
-  int temp = 1; /* locked value */
-
-  __asm__ __volatile__(
-               "1:lwarx %0,0,%1\n"   /* load and reserve               */
-               "cmpwi %0, 0\n"       /* if load is                     */
-               "bne 2f\n"            /*   non-zero, return already set */
-               "stwcx. %2,0,%1\n"    /* else store conditional         */
-               "bne- 1b\n"           /* retry if lost reservation      */
-               "2:\n"                /* oldval is zero if we set       */
-              : "=&r"(oldval)
-              : "r"(addr), "r"(temp)
-              : "memory", "cr0");
-
-  return (AO_TS_VAL_t)oldval;
-}
-
-#endif
-
-#define AO_HAVE_test_and_set
-
-AO_INLINE AO_TS_VAL_t
-AO_test_and_set_acquire(volatile AO_TS_t *addr) {
-  AO_TS_VAL_t result = AO_test_and_set(addr);
-  AO_lwsync();
-  return result;
-}
-
-#define AO_HAVE_test_and_set_acquire
-
-AO_INLINE AO_TS_VAL_t
-AO_test_and_set_release(volatile AO_TS_t *addr) {
-  AO_lwsync();
-  return AO_test_and_set(addr);
-}
-
-#define AO_HAVE_test_and_set_release
-
-AO_INLINE AO_TS_VAL_t
-AO_test_and_set_full(volatile AO_TS_t *addr) {
-  AO_TS_VAL_t result;
-  AO_lwsync();
-  result = AO_test_and_set(addr);
-  AO_lwsync();
-  return result;
-}
-
-#define AO_HAVE_test_and_set_full
-
-#if defined(__powerpc64__) || defined(__ppc64__) || defined(__64BIT__)
-/* FIXME: Completely untested.  */
-AO_INLINE int
-AO_compare_and_swap(volatile AO_t *addr, AO_t old, AO_t new_val) {
-  AO_t oldval;
-  int result = 0;
-
-  __asm__ __volatile__(
-               "1:ldarx %0,0,%2\n"   /* load and reserve              */
-               "cmpd %0, %4\n"      /* if load is not equal to  */
-               "bne 2f\n"            /*   old, fail                     */
-               "stdcx. %3,0,%2\n"    /* else store conditional         */
-               "bne- 1b\n"           /* retry if lost reservation      */
-               "li %1,1\n"           /* result = 1;                     */
-               "2:\n"
-              : "=&r"(oldval), "=&r"(result)
-              : "r"(addr), "r"(new_val), "r"(old), "1"(result)
-              : "memory", "cr0");
-
-  return result;
-}
-
-#else
-
-AO_INLINE int
-AO_compare_and_swap(volatile AO_t *addr, AO_t old, AO_t new_val) {
-  AO_t oldval;
-  int result = 0;
-
-  __asm__ __volatile__(
-               "1:lwarx %0,0,%2\n"   /* load and reserve              */
-               "cmpw %0, %4\n"      /* if load is not equal to  */
-               "bne 2f\n"            /*   old, fail                     */
-               "stwcx. %3,0,%2\n"    /* else store conditional         */
-               "bne- 1b\n"           /* retry if lost reservation      */
-               "li %1,1\n"           /* result = 1;                     */
-               "2:\n"
-              : "=&r"(oldval), "=&r"(result)
-              : "r"(addr), "r"(new_val), "r"(old), "1"(result)
-              : "memory", "cr0");
-
-  return result;
-}
-#endif
-
-#define AO_HAVE_compare_and_swap
-
-AO_INLINE int
-AO_compare_and_swap_acquire(volatile AO_t *addr, AO_t old, AO_t new_val) {
-  int result = AO_compare_and_swap(addr, old, new_val);
-  AO_lwsync();
-  return result;
-}
-
-#define AO_HAVE_compare_and_swap_acquire
-
-AO_INLINE int
-AO_compare_and_swap_release(volatile AO_t *addr, AO_t old, AO_t new_val) {
-  AO_lwsync();
-  return AO_compare_and_swap(addr, old, new_val);
-}
-
-#define AO_HAVE_compare_and_swap_release
-
-AO_INLINE int
-AO_compare_and_swap_full(volatile AO_t *addr, AO_t old, AO_t new_val) {
-  AO_t result;
-  AO_lwsync();
-  result = AO_compare_and_swap(addr, old, new_val);
-  AO_lwsync();
-  return result;
-}
-
-#define AO_HAVE_compare_and_swap_full
-
-#if defined(__powerpc64__) || defined(__ppc64__) || defined(__64BIT__)
-/* FIXME: Completely untested.                                          */
-
-AO_INLINE AO_t
-AO_fetch_and_add(volatile AO_t *addr, AO_t incr) {
-  AO_t oldval;
-  AO_t newval;
-
-  __asm__ __volatile__(
-               "1:ldarx %0,0,%2\n"   /* load and reserve                */
-               "add %1,%0,%3\n"      /* increment                       */
-               "stdcx. %1,0,%2\n"    /* store conditional               */
-               "bne- 1b\n"           /* retry if lost reservation       */
-              : "=&r"(oldval), "=&r"(newval)
-               : "r"(addr), "r"(incr)
-              : "memory", "cr0");
-
-  return oldval;
-}
-
-#define AO_HAVE_fetch_and_add
-
-#else
-
-AO_INLINE AO_t
-AO_fetch_and_add(volatile AO_t *addr, AO_t incr) {
-  AO_t oldval;
-  AO_t newval;
-
-  __asm__ __volatile__(
-               "1:lwarx %0,0,%2\n"   /* load and reserve                */
-               "add %1,%0,%3\n"      /* increment                       */
-               "stwcx. %1,0,%2\n"    /* store conditional               */
-               "bne- 1b\n"           /* retry if lost reservation       */
-              : "=&r"(oldval), "=&r"(newval)
-               : "r"(addr), "r"(incr)
-              : "memory", "cr0");
-
-  return oldval;
-}
-
-#define AO_HAVE_fetch_and_add
-
-#endif
-
-AO_INLINE AO_t
-AO_fetch_and_add_acquire(volatile AO_t *addr, AO_t incr) {
-  AO_t result = AO_fetch_and_add(addr, incr);
-  AO_lwsync();
-  return result;
-}
-
-#define AO_HAVE_fetch_and_add_acquire
-
-AO_INLINE AO_t
-AO_fetch_and_add_release(volatile AO_t *addr, AO_t incr) {
-  AO_lwsync();
-  return AO_fetch_and_add(addr, incr);
-}
-
-#define AO_HAVE_fetch_and_add_release
-
-AO_INLINE AO_t
-AO_fetch_and_add_full(volatile AO_t *addr, AO_t incr) {
-  AO_t result;
-  AO_lwsync();
-  result = AO_fetch_and_add(addr, incr);
-  AO_lwsync();
-  return result;
-}
-
-#define AO_HAVE_fetch_and_add_full
-
-#if defined(__powerpc64__) || defined(__ppc64__) || defined(__64BIT__)
-#else
-# include "./ao_t_is_int.h"
-#endif

c_src/atomic_ops/read_ordered.h

@@ -1,100 +0,0 @@
-/*
- * Copyright (c) 2003 by Hewlett-Packard Company.  All rights reserved.
- *
- * Permission is hereby granted, free of charge, to any person obtaining a copy
- * of this software and associated documentation files (the "Software"), to deal
- * in the Software without restriction, including without limitation the rights
- * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
- * copies of the Software, and to permit persons to whom the Software is
- * furnished to do so, subject to the following conditions:
- *
- * The above copyright notice and this permission notice shall be included in
- * all copies or substantial portions of the Software.
- *
- * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
- * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
- * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
- * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
- * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
- * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
- * SOFTWARE.
- */
-
-/*
- * These are common definitions for architectures that provide processor
- * ordered memory operations except that a later read may pass an
- * earlier write.  Real x86 implementations seem to be in this category,
- * except apparently for some IDT WinChips, which we ignore.
- */
-
-AO_INLINE void
-AO_nop_read(void)
-{
-  AO_compiler_barrier();
-}
-
-#define AO_HAVE_NOP_READ
-
-#ifdef AO_HAVE_load
-
-AO_INLINE AO_t
-AO_load_read(const volatile AO_t *addr)
-{
-  AO_t result = AO_load(addr);
-  AO_compiler_barrier();
-  return result;
-}
-#define AO_HAVE_load_read
-
-#define AO_load_acquire(addr) AO_load_read(addr)
-#define AO_HAVE_load_acquire
-
-#endif /* AO_HAVE_load */
-
-#ifdef AO_HAVE_char_load
-
-AO_INLINE AO_t
-AO_char_load_read(const volatile unsigned char *addr)
-{
-  AO_t result = AO_char_load(addr);
-  AO_compiler_barrier();
-  return result;
-}
-#define AO_HAVE_char_load_read
-
-#define AO_char_load_acquire(addr) AO_char_load_read(addr)
-#define AO_HAVE_char_load_acquire
-
-#endif /* AO_HAVE_char_load */
-
-#ifdef AO_HAVE_short_load
-
-AO_INLINE AO_t
-AO_short_load_read(const volatile unsigned short *addr)
-{
-  AO_t result = AO_short_load(addr);
-  AO_compiler_barrier();
-  return result;
-}
-#define AO_HAVE_short_load_read
-
-#define AO_short_load_acquire(addr) AO_short_load_read(addr)
-#define AO_HAVE_short_load_acquire
-
-#endif /* AO_HAVE_short_load */
-
-#ifdef AO_HAVE_int_load
-
-AO_INLINE AO_t
-AO_int_load_read(const volatile unsigned int *addr)
-{
-  AO_t result = AO_int_load(addr);
-  AO_compiler_barrier();
-  return result;
-}
-#define AO_HAVE_int_load_read
-
-#define AO_int_load_acquire(addr) AO_int_load_read(addr)
-#define AO_HAVE_int_load_acquire
-
-#endif /* AO_HAVE_int_load */

c_src/atomic_ops/sparc.h

@@ -1,72 +0,0 @@
-/*
- * Copyright (c) 1991-1994 by Xerox Corporation.  All rights reserved.
- * Copyright (c) 1996-1999 by Silicon Graphics.  All rights reserved.
- * Copyright (c) 1999-2003 by Hewlett-Packard Company. All rights reserved.
- *
- *
- * THIS MATERIAL IS PROVIDED AS IS, WITH ABSOLUTELY NO WARRANTY EXPRESSED
- * OR IMPLIED.  ANY USE IS AT YOUR OWN RISK.
- *
- * Permission is hereby granted to use or copy this program
- * for any purpose,  provided the above notices are retained on all copies.
- * Permission to modify the code and to distribute modified code is granted,
- * provided the above notices are retained, and a notice that the code was
- * modified is included with the above copyright notice.
- *
- */
-
-/* FIXME.  Very incomplete.  No support for sparc64.    */
-/* Non-ancient SPARCs provide compare-and-swap (casa).  */
-/* We should make that available.                       */
-
-#include "./aligned_atomic_load_store.h"
-
-/* Real SPARC code uses TSO:                            */
-#include "./ordered_except_wr.h"
-
-/* Test_and_set location is just a byte.                */
-#include "./test_and_set_t_is_char.h"
-
-AO_INLINE AO_TS_VAL_t
-AO_test_and_set_full(volatile AO_TS_t *addr) {
-   AO_TS_VAL_t oldval;
-
-   __asm__ __volatile__("ldstub %1,%0"
-                        : "=r"(oldval), "=m"(*addr)
-                        : "m"(*addr) : "memory");
-   return oldval;
-}
-
-#define AO_HAVE_test_and_set_full
-
-#ifndef AO_NO_SPARC_V9
-/* Returns nonzero if the comparison succeeded. */
-AO_INLINE int
-AO_compare_and_swap_full(volatile AO_t *addr, AO_t old, AO_t new_val) {
-  char ret;
-  __asm__ __volatile__ ("membar #StoreLoad | #LoadLoad\n\t"
-#                       if defined(__arch64__)
-                          "casx [%2],%0,%1\n\t"
-#                       else
-                          "cas [%2],%0,%1\n\t" /* 32-bit version */
-#                       endif
-                        "membar #StoreLoad | #StoreStore\n\t"
-                        "cmp %0,%1\n\t"
-                        "be,a 0f\n\t"
-                        "mov 1,%0\n\t"/* one insn after branch always executed */
-                        "clr %0\n\t"
-                        "0:\n\t"
-                        : "=r" (ret), "+r" (new_val)
-                        : "r" (addr), "0" (old)
-                        : "memory", "cc");
-  return (int)ret;
-}
-
-#define AO_HAVE_compare_and_swap_full
-#endif /* AO_NO_SPARC_V9 */
-
-/* FIXME: This needs to be extended for SPARC v8 and v9.        */
-/* SPARC V8 also has swap.  V9 has CAS.                         */
-/* There are barriers like membar #LoadStore.                   */
-/* CASA (32-bit) and CASXA(64-bit) instructions were            */
-/* added in V9.                                                 */

c_src/atomic_ops/standard_ao_double_t.h

@@ -1,25 +0,0 @@
-/* NEC LE-IT: For 64Bit OS we extend the double type to hold two int64's
-*
-*  x86-64: __m128 serves as placeholder which also requires the compiler
-*          to align     it on 16 byte boundary (as required by cmpxchg16.
-* Similar things could be done for PowerPC 64bit using a VMX data type...       */
-
-#if (defined(__x86_64__) && defined(__GNUC__)) || defined(_WIN64)
-# include <xmmintrin.h>
-  typedef __m128 double_ptr_storage;
-#elif defined(_WIN32) && !defined(__GNUC__)
-  typedef unsigned __int64 double_ptr_storage;
-#else
-  typedef unsigned long long double_ptr_storage;
-#endif
-
-# define AO_HAVE_DOUBLE_PTR_STORAGE
-
-typedef union {
-    double_ptr_storage AO_whole;
-    struct {AO_t AO_v1; AO_t AO_v2;} AO_parts;
-} AO_double_t;
-
-#define AO_HAVE_double_t
-#define AO_val1 AO_parts.AO_v1
-#define AO_val2 AO_parts.AO_v2

c_src/atomic_ops/test_and_set_t_is_ao_t.h

@@ -1,36 +0,0 @@
-/*
- * Copyright (c) 2004 Hewlett-Packard Development Company, L.P.
- *
- * Permission is hereby granted, free of charge, to any person obtaining a copy
- * of this software and associated documentation files (the "Software"), to deal
- * in the Software without restriction, including without limitation the rights
- * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
- * copies of the Software, and to permit persons to whom the Software is
- * furnished to do so, subject to the following conditions:
- *
- * The above copyright notice and this permission notice shall be included in
- * all copies or substantial portions of the Software.
- *
- * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
- * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
- * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
- * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
- * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
- * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
- * SOFTWARE.
- */
-
-/*
- * These are common definitions for architectures on which test_and_set
- * operates on pointer-sized quantities, the "clear" value contains
- * all zeroes, and the "set" value contains only one lowest bit set.
- * This can be used if test_and_set is synthesized from compare_and_swap.
- */
-typedef enum {AO_TS_clear = 0, AO_TS_set = 1} AO_TS_val;
-#define AO_TS_VAL_t AO_TS_val
-#define AO_TS_CLEAR AO_TS_clear
-#define AO_TS_SET AO_TS_set
-
-#define AO_TS_t AO_t
-
-#define AO_AO_TS_T 1

c_src/atomic_ops/test_and_set_t_is_char.h

@@ -1,38 +0,0 @@
-/*
- * Copyright (c) 2004 Hewlett-Packard Development Company, L.P.
- * 
- * Permission is hereby granted, free of charge, to any person obtaining a copy
- * of this software and associated documentation files (the "Software"), to deal
- * in the Software without restriction, including without limitation the rights
- * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
- * copies of the Software, and to permit persons to whom the Software is
- * furnished to do so, subject to the following conditions:
- * 
- * The above copyright notice and this permission notice shall be included in
- * all copies or substantial portions of the Software.
- * 
- * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
- * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
- * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
- * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
- * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
- * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
- * SOFTWARE. 
- */ 
-
-/*
- * These are common definitions for architectures on which test_and_set
- * operates on byte sized quantities, the "clear" value contains
- * all zeroes, and the "set" value contains all ones.
- */
-
-#define AO_TS_t unsigned char
-typedef enum {AO_BYTE_TS_clear = 0, AO_BYTE_TS_set = 0xff} AO_BYTE_TS_val;
-#define AO_TS_VAL_t AO_BYTE_TS_val
-#define AO_TS_CLEAR AO_BYTE_TS_clear
-#define AO_TS_SET AO_BYTE_TS_set
-
-#define AO_CHAR_TS_T 1
-
-
-

c_src/atomic_ops/x86.h

@@ -1,173 +0,0 @@
-/*
- * Copyright (c) 1991-1994 by Xerox Corporation.  All rights reserved.
- * Copyright (c) 1996-1999 by Silicon Graphics.  All rights reserved.
- * Copyright (c) 1999-2003 by Hewlett-Packard Company. All rights reserved.
- *
- *
- * THIS MATERIAL IS PROVIDED AS IS, WITH ABSOLUTELY NO WARRANTY EXPRESSED
- * OR IMPLIED.  ANY USE IS AT YOUR OWN RISK.
- *
- * Permission is hereby granted to use or copy this program
- * for any purpose,  provided the above notices are retained on all copies.
- * Permission to modify the code and to distribute modified code is granted,
- * provided the above notices are retained, and a notice that the code was
- * modified is included with the above copyright notice.
- *
- * Some of the machine specific code was borrowed from our GC distribution.
- */
-
-/* The following really assume we have a 486 or better.  Unfortunately  */
-/* gcc doesn't define a suitable feature test macro based on command    */
-/* line options.                                                        */
-/* We should perhaps test dynamically.                                  */
-
-#include "./aligned_atomic_load_store.h"
-
-/* Real X86 implementations, except for some old WinChips, appear       */
-/* to enforce ordering between memory operations, EXCEPT that a later   */
-/* read can pass earlier writes, presumably due to the visible          */
-/* presence of store buffers.                                           */
-/* We ignore both the WinChips, and the fact that the official specs    */
-/* seem to be much weaker (and arguably too weak to be usable).         */
-
-#include "./ordered_except_wr.h"
-
-#include "./test_and_set_t_is_char.h"
-
-#include "./standard_ao_double_t.h"
-
-#if defined(AO_USE_PENTIUM4_INSTRS)
-AO_INLINE void
-AO_nop_full(void)
-{
-  __asm__ __volatile__("mfence" : : : "memory");
-}
-
-#define AO_HAVE_nop_full
-
-#else
-
-/* We could use the cpuid instruction.  But that seems to be slower     */
-/* than the default implementation based on test_and_set_full.  Thus    */
-/* we omit that bit of misinformation here.                             */
-
-#endif
-
-/* As far as we can tell, the lfence and sfence instructions are not    */
-/* currently needed or useful for cached memory accesses.               */
-
-/* Really only works for 486 and later */
-AO_INLINE AO_t
-AO_fetch_and_add_full (volatile AO_t *p, AO_t incr)
-{
-  AO_t result;
-
-  __asm__ __volatile__ ("lock; xaddl %0, %1" :
-                        "=r" (result), "=m" (*p) : "0" (incr), "m" (*p)
-                        : "memory");
-  return result;
-}
-
-#define AO_HAVE_fetch_and_add_full
-
-AO_INLINE unsigned char
-AO_char_fetch_and_add_full (volatile unsigned char *p, unsigned char incr)
-{
-  unsigned char result;
-
-  __asm__ __volatile__ ("lock; xaddb %0, %1" :
-                        "=q" (result), "=m" (*p) : "0" (incr), "m" (*p)
-                        : "memory");
-  return result;
-}
-
-#define AO_HAVE_char_fetch_and_add_full
-
-AO_INLINE unsigned short
-AO_short_fetch_and_add_full (volatile unsigned short *p, unsigned short incr)
-{
-  unsigned short result;
-
-  __asm__ __volatile__ ("lock; xaddw %0, %1" :
-                        "=r" (result), "=m" (*p) : "0" (incr), "m" (*p)
-                        : "memory");
-  return result;
-}
-
-#define AO_HAVE_short_fetch_and_add_full
-
-/* Really only works for 486 and later */
-AO_INLINE void
-AO_or_full (volatile AO_t *p, AO_t incr)
-{
-  __asm__ __volatile__ ("lock; orl %1, %0" :
-                        "=m" (*p) : "r" (incr), "m" (*p) : "memory");
-}
-
-#define AO_HAVE_or_full
-
-AO_INLINE AO_TS_VAL_t
-AO_test_and_set_full(volatile AO_TS_t *addr)
-{
-  unsigned char oldval;
-  /* Note: the "xchg" instruction does not need a "lock" prefix */
-  __asm__ __volatile__("xchgb %0, %1"
-                : "=q"(oldval), "=m"(*addr)
-                : "0"((unsigned char)0xff), "m"(*addr) : "memory");
-  return (AO_TS_VAL_t)oldval;
-}
-
-#define AO_HAVE_test_and_set_full
-
-/* Returns nonzero if the comparison succeeded. */
-AO_INLINE int
-AO_compare_and_swap_full(volatile AO_t *addr, AO_t old, AO_t new_val)
-{
-# ifdef AO_USE_SYNC_CAS_BUILTIN
-    return (int)__sync_bool_compare_and_swap(addr, old, new_val);
-# else
-    char result;
-    __asm__ __volatile__("lock; cmpxchgl %3, %0; setz %1"
-                         : "=m" (*addr), "=a" (result)
-                         : "m" (*addr), "r" (new_val), "a" (old) : "memory");
-    return (int)result;
-# endif
-}
-
-#define AO_HAVE_compare_and_swap_full
-
-/* Returns nonzero if the comparison succeeded. */
-/* Really requires at least a Pentium.          */
-AO_INLINE int
-AO_compare_double_and_swap_double_full(volatile AO_double_t *addr,
-                                       AO_t old_val1, AO_t old_val2,
-                                       AO_t new_val1, AO_t new_val2)
-{
-  char result;
-#if __PIC__
-  /* If PIC is turned on, we can't use %ebx as it is reserved for the
-     GOT pointer.  We can save and restore %ebx because GCC won't be
-     using it for anything else (such as any of the m operands) */
-  __asm__ __volatile__("pushl %%ebx;"   /* save ebx used for PIC GOT ptr */
-                       "movl %6,%%ebx;" /* move new_val2 to %ebx */
-                       "lock; cmpxchg8b %0; setz %1;"
-                       "pop %%ebx;"     /* restore %ebx */
-                       : "=m"(*addr), "=a"(result)
-                       : "m"(*addr), "d" (old_val2), "a" (old_val1),
-                         "c" (new_val2), "m" (new_val1) : "memory");
-#else
-  /* We can't just do the same thing in non-PIC mode, because GCC
-   * might be using %ebx as the memory operand.  We could have ifdef'd
-   * in a clobber, but there's no point doing the push/pop if we don't
-   * have to. */
-  __asm__ __volatile__("lock; cmpxchg8b %0; setz %1;"
-                       : "=m"(*addr), "=a"(result)
-                       : "m"(*addr), "d" (old_val2), "a" (old_val1),
-                         "c" (new_val2), "b" (new_val1) : "memory");
-#endif
-  return (int) result;
-}
-
-#define AO_HAVE_compare_double_and_swap_double_full
-
-#include "./ao_t_is_int.h"

c_src/atomic_ops/x86_64.h

@@ -1,181 +0,0 @@
-/*
- * Copyright (c) 1991-1994 by Xerox Corporation.  All rights reserved.
- * Copyright (c) 1996-1999 by Silicon Graphics.  All rights reserved.
- * Copyright (c) 1999-2003 by Hewlett-Packard Company. All rights reserved.
- *
- *
- * THIS MATERIAL IS PROVIDED AS IS, WITH ABSOLUTELY NO WARRANTY EXPRESSED
- * OR IMPLIED.  ANY USE IS AT YOUR OWN RISK.
- *
- * Permission is hereby granted to use or copy this program
- * for any purpose,  provided the above notices are retained on all copies.
- * Permission to modify the code and to distribute modified code is granted,
- * provided the above notices are retained, and a notice that the code was
- * modified is included with the above copyright notice.
- *
- * Some of the machine specific code was borrowed from our GC distribution.
- */
-
-#include "./aligned_atomic_load_store.h"
-
-/* Real X86 implementations appear                                      */
-/* to enforce ordering between memory operations, EXCEPT that a later   */
-/* read can pass earlier writes, presumably due to the visible          */
-/* presence of store buffers.                                           */
-/* We ignore the fact that the official specs                           */
-/* seem to be much weaker (and arguably too weak to be usable).         */
-
-#include "./ordered_except_wr.h"
-
-#include "./test_and_set_t_is_char.h"
-
-#include "./standard_ao_double_t.h"
-
-AO_INLINE void
-AO_nop_full(void)
-{
-  /* Note: "mfence" (SSE2) is supported on all x86_64/amd64 chips.      */
-  __asm__ __volatile__("mfence" : : : "memory");
-}
-
-#define AO_HAVE_nop_full
-
-/* As far as we can tell, the lfence and sfence instructions are not    */
-/* currently needed or useful for cached memory accesses.               */
-
-AO_INLINE AO_t
-AO_fetch_and_add_full (volatile AO_t *p, AO_t incr)
-{
-  AO_t result;
-
-  __asm__ __volatile__ ("lock; xaddq %0, %1" :
-                        "=r" (result), "=m" (*p) : "0" (incr), "m" (*p)
-                        : "memory");
-  return result;
-}
-
-#define AO_HAVE_fetch_and_add_full
-
-AO_INLINE unsigned char
-AO_char_fetch_and_add_full (volatile unsigned char *p, unsigned char incr)
-{
-  unsigned char result;
-
-  __asm__ __volatile__ ("lock; xaddb %0, %1" :
-                        "=q" (result), "=m" (*p) : "0" (incr), "m" (*p)
-                        : "memory");
-  return result;
-}
-
-#define AO_HAVE_char_fetch_and_add_full
-
-AO_INLINE unsigned short
-AO_short_fetch_and_add_full (volatile unsigned short *p, unsigned short incr)
-{
-  unsigned short result;
-
-  __asm__ __volatile__ ("lock; xaddw %0, %1" :
-                        "=r" (result), "=m" (*p) : "0" (incr), "m" (*p)
-                        : "memory");
-  return result;
-}
-
-#define AO_HAVE_short_fetch_and_add_full
-
-AO_INLINE unsigned int
-AO_int_fetch_and_add_full (volatile unsigned int *p, unsigned int incr)
-{
-  unsigned int result;
-
-  __asm__ __volatile__ ("lock; xaddl %0, %1" :
-                        "=r" (result), "=m" (*p) : "0" (incr), "m" (*p)
-                        : "memory");
-  return result;
-}
-
-#define AO_HAVE_int_fetch_and_add_full
-
-AO_INLINE void
-AO_or_full (volatile AO_t *p, AO_t incr)
-{
-  __asm__ __volatile__ ("lock; orq %1, %0" :
-                        "=m" (*p) : "r" (incr), "m" (*p) : "memory");
-}
-
-#define AO_HAVE_or_full
-
-AO_INLINE AO_TS_VAL_t
-AO_test_and_set_full(volatile AO_TS_t *addr)
-{
-  unsigned char oldval;
-  /* Note: the "xchg" instruction does not need a "lock" prefix */
-  __asm__ __volatile__("xchgb %0, %1"
-                : "=q"(oldval), "=m"(*addr)
-                : "0"((unsigned char)0xff), "m"(*addr) : "memory");
-  return (AO_TS_VAL_t)oldval;
-}
-
-#define AO_HAVE_test_and_set_full
-
-/* Returns nonzero if the comparison succeeded. */
-AO_INLINE int
-AO_compare_and_swap_full(volatile AO_t *addr, AO_t old, AO_t new_val)
-{
-# ifdef AO_USE_SYNC_CAS_BUILTIN
-    return (int)__sync_bool_compare_and_swap(addr, old, new_val);
-# else
-    char result;
-    __asm__ __volatile__("lock; cmpxchgq %3, %0; setz %1"
-                         : "=m" (*addr), "=a" (result)
-                         : "m" (*addr), "r" (new_val), "a" (old) : "memory");
-    return (int) result;
-# endif
-}
-
-#define AO_HAVE_compare_and_swap_full
-
-#ifdef AO_CMPXCHG16B_AVAILABLE
-/* NEC LE-IT: older AMD Opterons are missing this instruction.
- * On these machines SIGILL will be thrown.
- * Define AO_WEAK_DOUBLE_CAS_EMULATION to have an emulated
- * (lock based) version available */
-/* HB: Changed this to not define either by default.  There are
- * enough machines and tool chains around on which cmpxchg16b
- * doesn't work.  And the emulation is unsafe by our usual rules.
- * Hoewever both are clearly useful in certain cases.
- */
-AO_INLINE int
-AO_compare_double_and_swap_double_full(volatile AO_double_t *addr,
-                                       AO_t old_val1, AO_t old_val2,
-                                       AO_t new_val1, AO_t new_val2)
-{
-  char result;
-  __asm__ __volatile__("lock; cmpxchg16b %0; setz %1"
-                       : "=m"(*addr), "=a"(result)
-                       : "m"(*addr), "d" (old_val2), "a" (old_val1),
-                         "c" (new_val2), "b" (new_val1) : "memory");
-  return (int) result;
-}
-#define AO_HAVE_compare_double_and_swap_double_full
-#else
-/* this one provides spinlock based emulation of CAS implemented in     */
-/* atomic_ops.c.  We probably do not want to do this here, since it is  */
-/* not atomic with respect to other kinds of updates of *addr.  On the  */
-/* other hand, this may be a useful facility on occasion.               */
-#ifdef AO_WEAK_DOUBLE_CAS_EMULATION
-int AO_compare_double_and_swap_double_emulation(volatile AO_double_t *addr,
-                                                AO_t old_val1, AO_t old_val2,
-                                                AO_t new_val1, AO_t new_val2);
-
-AO_INLINE int
-AO_compare_double_and_swap_double_full(volatile AO_double_t *addr,
-                                       AO_t old_val1, AO_t old_val2,
-                                       AO_t new_val1, AO_t new_val2)
-{
-        return AO_compare_double_and_swap_double_emulation(addr,
-                                                           old_val1, old_val2,
-                                                           new_val1, new_val2);
-}
-#define AO_HAVE_compare_double_and_swap_double_full
-#endif /* AO_WEAK_DOUBLE_CAS_EMULATION */
-#endif /* AO_CMPXCHG16B_AVAILABLE */

c_src/build_deps.sh

@@ -11,9 +11,10 @@ unset POSIX_SHELL # clear it so if we invoke other scripts, they run as ksh as w
 set -e
 
 WT_REPO=http://github.com/wiredtiger/wiredtiger.git
-WT_BRANCH=
-WT_REF="tags/1.6.3"
-WT_DIR=wiredtiger-`basename $WT_REF`
+WT_BRANCH=develop
+WT_DIR=wiredtiger-`basename $WT_BRANCH`
+#WT_REF="tags/1.6.6"
+#WT_DIR=wiredtiger-`basename $WT_REF`
 
 SNAPPY_VSN="1.0.4"
 SNAPPY_DIR=snappy-$SNAPPY_VSN
@@ -25,8 +26,7 @@ export BASEDIR="$PWD"
 which gmake 1>/dev/null 2>/dev/null && MAKE=gmake
 MAKE=${MAKE:-make}
 
-export CFLAGS="$CFLAGS -I $BASEDIR/system/include"
-export CXXFLAGS="$CXXFLAGS -I $BASEDIR/system/include"
+export CPPFLAGS="$CPPLAGS -I $BASEDIR/system/include -O3 -mtune=native -march=native"
 export LDFLAGS="$LDFLAGS -L$BASEDIR/system/lib"
 export LD_LIBRARY_PATH="$LD_LIBRARY_PATH:$BASEDIR/system/lib:$LD_LIBRARY_PATH"
 
@@ -57,7 +57,7 @@ get_wt ()
 wt_configure ()
 {
     (cd $BASEDIR/$WT_DIR/build_posix
-        ../configure --with-pic \
+        CFLAGS+=-g $BASEDIR/$WT_DIR/configure --with-pic \
                      --enable-snappy \
                      --prefix=${BASEDIR}/system || exit 1)
 }
@@ -75,8 +75,8 @@ get_snappy ()
 
 get_deps ()
 {
-    get_wt;
     get_snappy;
+    get_wt;
 }
 
 update_deps ()
@@ -109,7 +109,7 @@ build_snappy ()
 case "$1" in
     clean)
         [ -e $BASEDIR/$WT_DIR/build_posix/Makefile ] && \
-            (cd $BASEDIR/$WT_DIR/build_posix && $MAKE distclean)
+            (cd $BASEDIR/$WT_DIR/build_posix && $MAKE clean)
         rm -rf system $SNAPPY_DIR
         rm -f ${BASEDIR}/../priv/wt
         rm -f ${BASEDIR}/../priv/libwiredtiger-*.so
@@ -130,22 +130,23 @@ case "$1" in
         ;;
 
     *)
-        [ -d $WT_DIR ] || get_wt;
-        [ -d $SNAPPY_DIR ] || get_snappy;
+	shopt -s extglob
+	SUFFIXES='@(so|dylib)'
 
         # Build Snappy
+        [ -d $SNAPPY_DIR ] || get_snappy;
         [ -d $BASEDIR/$SNAPPY_DIR ] || (echo "Missing Snappy source directory" && exit 1)
-        test -f $BASEDIR/system/lib/libsnappy.so.[0-9].[0-9].[0-9] || build_snappy;
+        test -f $BASEDIR/system/lib/libsnappy.so.[0-9].[0-9].[0-9].* || build_snappy;
 
         # Build WiredTiger
+        [ -d $WT_DIR ] || get_wt;
         [ -d $BASEDIR/$WT_DIR ] || (echo "Missing WiredTiger source directory" && exit 1)
-        test -f $BASEDIR/system/lib/libwiredtiger-[0-9].[0-9].[0-9].so \
-             -a -f $BASEDIR/system/lib/libwiredtiger_snappy.so || build_wt;
-
+        test -f  $BASEDIR/system/lib/libwiredtiger-[0-9].[0-9].[0-9].${SUFFIXES} -a \
+	    -f $BASEDIR/system/lib/libwiredtiger_snappy.${SUFFIXES} || build_wt;
         [ -d $BASEDIR/../priv ] || mkdir ${BASEDIR}/../priv
-        cp -p -P $BASEDIR/system/bin/wt ${BASEDIR}/../priv
-        cp -p -P $BASEDIR/system/lib/libwiredtiger-[0-9].[0-9].[0-9].so ${BASEDIR}/../priv
-        cp -p -P $BASEDIR/system/lib/libwiredtiger_snappy.so* ${BASEDIR}/../priv
-        cp -p -P $BASEDIR/system/lib/libsnappy.so* ${BASEDIR}/../priv
+	cp -p -P $BASEDIR/system/bin/wt ${BASEDIR}/../priv
+	cp -p -P ${BASEDIR}/system/lib/libwiredtiger-[0-9].[0-9].[0-9].${SUFFIXES} ${BASEDIR}/../priv
+	cp -p -P ${BASEDIR}/system/lib/libwiredtiger_snappy.${SUFFIXES} ${BASEDIR}/../priv
+	cp -p -P ${BASEDIR}/system/lib/libsnappy.${SUFFIXES}* ${BASEDIR}/../priv
         ;;
 esac

c_src/wiredtiger-build.patch

@@ -1,5 +1,5 @@
 diff --git a/ext/compressors/snappy/Makefile.am b/ext/compressors/snappy/Makefile.am
-index 6d78823..2122cf8 100644
+index 6d78823..c423590 100644
 --- a/ext/compressors/snappy/Makefile.am
 +++ b/ext/compressors/snappy/Makefile.am
 @@ -2,5 +2,6 @@ AM_CPPFLAGS = -I$(top_builddir) -I$(top_srcdir)/src/include
@@ -7,100 +7,6 @@ index 6d78823..2122cf8 100644
  lib_LTLIBRARIES = libwiredtiger_snappy.la
  libwiredtiger_snappy_la_SOURCES = snappy_compress.c
 -libwiredtiger_snappy_la_LDFLAGS = -avoid-version -module
-+libwiredtiger_snappy_la_CFLAGS = -I$(src_builddir)/../../system/include
-+libwiredtiger_snappy_la_LDFLAGS = -avoid-version -module -L$(src_builddir)/../../system/lib -Wl,-rpath,lib/wterl-0.9.0/priv:lib/wterl/priv:priv
++libwiredtiger_snappy_la_CFLAGS = -I$(abs_top_builddir)/../../system/include
++libwiredtiger_snappy_la_LDFLAGS = -avoid-version -module -L$(abs_top_builddir)/../../system/lib -Wl,-rpath,lib/wterl-0.9.0/priv:lib/wterl/priv:priv
  libwiredtiger_snappy_la_LIBADD = -lsnappy
-diff --git a/src/support/cksum.c b/src/support/cksum.c
-index 7e9befe..b924db7 100644
---- a/src/support/cksum.c
-+++ b/src/support/cksum.c
-@@ -27,6 +27,13 @@
- 
- #include "wt_internal.h"
- 
-+#if defined(__amd64) || defined(__x86_64)
-+#define USE_HARDWARE_CRC32 1
-+#else
-+#undef USE_HARDWARE_CRC32
-+#endif
-+
-+#ifdef USE_HARDWARE_CRC32
- static const uint32_t g_crc_slicing[8][256] = {
- #ifdef WORDS_BIGENDIAN
- 	/*
-@@ -1078,6 +1085,7 @@ static const uint32_t g_crc_slicing[8][256] = {
- 	}
- #endif
- };
-+#endif /* USE_HARDWARE_CRC32 */
- 
- /*
-  * __wt_cksum --
-@@ -1106,15 +1114,29 @@ __wt_cksum(const void *chunk, size_t len)
- 	/* Checksum one byte at a time to the first 4B boundary. */
- 	for (p = chunk;
- 	    ((uintptr_t)p & (sizeof(uint32_t) - 1)) != 0 &&
--	    len > 0; ++p, --len)
-+	    len > 0; ++p, --len) {
-+#ifdef USE_HARDWARE_CRC32
-+	    __asm__ __volatile__(
-+		".byte 0xF2, 0x0F, 0x38, 0xF0, 0xF1"
-+		: "=S" (crc)
-+		: "0" (crc), "c" (*p));
-+#else
- #ifdef WORDS_BIGENDIAN
- 		crc = g_crc_slicing[0][((crc >> 24) ^ *p) & 0xFF] ^ (crc << 8);
- #else
- 		crc = g_crc_slicing[0][(crc ^ *p) & 0xFF] ^ (crc >> 8);
- #endif
-+#endif
-+	}
- 
- 	/* Checksum in 8B chunks. */
- 	for (nqwords = len / sizeof(uint64_t); nqwords; nqwords--) {
-+#ifdef USE_HARDWARE_CRC32
-+	    __asm__ __volatile__ (
-+		".byte 0xf2, 0x48, 0x0f, 0x38, 0xf0, 0xf1;"
-+		: "=S"(crc)
-+		: "S"(crc), "c"(*p));
-+#else
- 		crc ^= *(uint32_t *)p;
- 		p += sizeof(uint32_t);
- 		next = *(uint32_t *)p;
-@@ -1139,22 +1161,32 @@ __wt_cksum(const void *chunk, size_t len)
- 			g_crc_slicing[1][(next >> 16) & 0xFF] ^
- 			g_crc_slicing[0][(next >> 24)];
- #endif
-+#endif
- 	}
- 
- 	/* Checksum trailing bytes one byte at a time. */
-+	for (len &= 0x7; len > 0; ++p, len--) {
-+#ifdef USE_HARDWARE_CRC32
-+	    __asm__ __volatile__(
-+		".byte 0xF2, 0x0F, 0x38, 0xF0, 0xF1"
-+		: "=S" (crc)
-+		: "0" (crc), "c" (*p));
-+#else
- #ifdef WORDS_BIGENDIAN
--	for (len &= 0x7; len > 0; ++p, len--)
- 		crc = g_crc_slicing[0][((crc >> 24) ^ *p) & 0xFF] ^ (crc << 8);
-+#else
-+		crc = g_crc_slicing[0][(crc ^ *p) & 0xFF] ^ (crc >> 8);
-+#endif
-+#endif
-+	}
- 
-+#ifdef WORDS_BIGENDIAN
- 	/* Do final byte swap to produce a result identical to little endian */
- 	crc =
- 		((crc << 24) & 0xFF000000) |
- 		((crc <<  8) & 0x00FF0000) |
- 		((crc >>  8) & 0x0000FF00) |
- 		((crc >> 24) & 0x000000FF);
--#else
--	for (len &= 0x7; len > 0; ++p, len--)
--		crc = g_crc_slicing[0][(crc ^ *p) & 0xFF] ^ (crc >> 8);
- #endif
- 	return (~crc);
- }

c_src/wterl.c

@@ -31,7 +31,6 @@
 #include "common.h"
 #include "async_nif.h"
 #include "queue.h"
-#include "atomic.h"
 
 #define MAX_CACHE_SIZE ASYNC_NIF_MAX_WORKERS
 
@@ -60,7 +59,6 @@ typedef struct wterl_conn {
     STAILQ_HEAD(ctxs, wterl_ctx) cache;
     ErlNifMutex *cache_mutex;
     uint32_t cache_size;
-    struct wterl_ctx *mru_ctx[ASYNC_NIF_MAX_WORKERS];
 } WterlConnHandle;
 
 typedef struct {
@@ -206,7 +204,7 @@ __ctx_cache_evict(WterlConnHandle *conn_handle)
             STAILQ_REMOVE(&conn_handle->cache, c, wterl_ctx, entries);
             if (c->session)
                 c->session->close(c->session, NULL);
-            enif_free(c);
+            free(c);
             num_evicted++;
         }
     }
@@ -288,6 +286,7 @@ __retain_ctx(WterlConnHandle *conn_handle, uint32_t worker_id,
              struct wterl_ctx **ctx,
              int count, const char *session_config, ...)
 {
+    UNUSED(worker_id);
     int i = 0;
     uint32_t hash = 0;
     uint32_t crc = 0;
@@ -324,64 +323,50 @@ __retain_ctx(WterlConnHandle *conn_handle, uint32_t worker_id,
     DPRINTF("sig %llu [%u:%u]", PRIuint64(sig), crc, hash);
     va_end(ap);
 
-    c = NULL;
-    do c = conn_handle->mru_ctx[worker_id];
-    while(c && !ATOMIC_CAS_FULL(&conn_handle->mru_ctx[worker_id], c, 0));
-
-    if (c && c->sig == sig) {
-	// mru hit:
-	DPRINTF("[%.4u] mru hit: %llu found", worker_id, PRIuint64(sig));
-	*ctx = c;
+    // check the cache
+    c = __ctx_cache_find(conn_handle, sig);
+    if (c == NULL) {
+	// cache miss:
+	DPRINTF("[%.4u] cache miss: %llu [cache size: %d]", worker_id, PRIuint64(sig), conn_handle->cache_size);
+	WT_CONNECTION *conn = conn_handle->conn;
+	WT_SESSION *session = NULL;
+	int rc = conn->open_session(conn, NULL, session_config, &session);
+	if (rc != 0) return rc;
+	size_t s = sizeof(struct wterl_ctx) + (count * sizeof(struct cursor_info)) + sig_len;
+	c = malloc(s); // TODO: enif_alloc_resource()
+	if (c == NULL) {
+	    session->close(session, NULL);
+	    return ENOMEM;
+	}
+	memset(c, 0, s);
+	c->sig = sig;
+	c->session = session;
+	c->sig_len = sig_len;
+	char *p = (char *)c + (s - sig_len);
+	c->session_config = __copy_str_into(&p, session_config);
+	c->num_cursors = count;
+	session_config = arg;
+	va_start(ap, session_config);
+	for (i = 0; i < count; i++) {
+	    const char *uri = va_arg(ap, const char *);
+	    const char *config = va_arg(ap, const char *);
+	    // TODO: what to do (if anything) when uri or config is NULL?
+	    c->ci[i].uri = __copy_str_into(&p, uri);
+	    c->ci[i].config = __copy_str_into(&p, config);
+	    rc = session->open_cursor(session, uri, NULL, config, &c->ci[i].cursor);
+	    if (rc != 0) {
+		free(c);
+		session->close(session, NULL); // this will free the cursors too
+		va_end(ap);
+		return rc;
+	    }
+	}
+	va_end(ap);
     } else {
-	// mru miss: check the cache
-	DPRINTF("[%.4u] mru miss or empty", worker_id);
-        c = __ctx_cache_find(conn_handle, sig);
-        if (c == NULL) {
-            // cache miss:
-            DPRINTF("[%.4u] cache miss: %llu [cache size: %d]", worker_id, PRIuint64(sig), conn_handle->cache_size);
-            WT_CONNECTION *conn = conn_handle->conn;
-            WT_SESSION *session = NULL;
-            int rc = conn->open_session(conn, NULL, session_config, &session);
-            if (rc != 0) {
-                return rc;
-            }
-            size_t s = sizeof(struct wterl_ctx) + (count * sizeof(struct cursor_info)) + sig_len;
-            c = enif_alloc(s); // TODO: enif_alloc_resource()
-            if (c == NULL) {
-                session->close(session, NULL);
-                return ENOMEM;
-            }
-            memset(c, 0, s);
-            c->sig = sig;
-            c->session = session;
-            c->sig_len = sig_len;
-            char *p = (char *)c + (s - sig_len);
-            c->session_config = __copy_str_into(&p, session_config);
-            c->num_cursors = count;
-            session_config = arg;
-            va_start(ap, session_config);
-            for (i = 0; i < count; i++) {
-                const char *uri = va_arg(ap, const char *);
-                const char *config = va_arg(ap, const char *);
-                // TODO: what to do (if anything) when uri or config is NULL?
-                c->ci[i].uri = __copy_str_into(&p, uri);
-                c->ci[i].config = __copy_str_into(&p, config);
-                rc = session->open_cursor(session, uri, NULL, config, &c->ci[i].cursor);
-                if (rc != 0) {
-                    enif_free(c);
-                    session->close(session, NULL); // this will free the cursors too
-		    va_end(ap);
-                    return rc;
-                }
-            }
-            va_end(ap);
-	    *ctx = c;
-        } else {
-            // cache hit:
-            DPRINTF("[%.4u] cache hit: %llu [cache size: %d]", worker_id, PRIuint64(sig), conn_handle->cache_size);
-	    *ctx = c;
-        }
+	// cache hit:
+	DPRINTF("[%.4u] cache hit: %llu [cache size: %d]", worker_id, PRIuint64(sig), conn_handle->cache_size);
     }
+    *ctx = c;
     return 0;
 }
 
@@ -391,6 +376,7 @@ __retain_ctx(WterlConnHandle *conn_handle, uint32_t worker_id,
 static void
 __release_ctx(WterlConnHandle *conn_handle, uint32_t worker_id, struct wterl_ctx *ctx)
 {
+    UNUSED(worker_id);
     uint32_t i;
     WT_CURSOR *cursor;
 
@@ -398,13 +384,8 @@ __release_ctx(WterlConnHandle *conn_handle, uint32_t worker_id, struct wterl_ctx
         cursor = ctx->ci[i].cursor;
         cursor->reset(cursor);
     }
-
-    struct wterl_ctx *c = conn_handle->mru_ctx[worker_id];
-    if (!ATOMIC_CAS_FULL(&conn_handle->mru_ctx[worker_id], c, ctx)) {
-	if (c) __ctx_cache_add(conn_handle, c);
-    } else __ctx_cache_add(conn_handle, ctx);
-
-    DPRINTF("[%.4u] reset %d cursors, returned ctx to cache", worker_id, ctx->num_cursors);
+    __ctx_cache_add(conn_handle, ctx);
+    DPRINTF("[%.4u] reset %d cursors, returnd ctx to cache", worker_id, ctx->num_cursors);
 }
 
 /**
@@ -416,14 +397,6 @@ void
 __close_all_sessions(WterlConnHandle *conn_handle)
 {
     struct wterl_ctx *c, *n;
-    int worker_id;
-
-    // clear out the mru
-    for (worker_id = 0; worker_id < ASYNC_NIF_MAX_WORKERS; worker_id++) {
-	do c = conn_handle->mru_ctx[worker_id];
-	while(c && !ATOMIC_CAS_FULL(&conn_handle->mru_ctx[worker_id], c, 0));
-	if (c) { c->session->close(c->session, NULL); enif_free(c); }
-    }
 
     // clear out the cache
     c = STAILQ_FIRST(&conn_handle->cache);
@@ -432,7 +405,7 @@ __close_all_sessions(WterlConnHandle *conn_handle)
         STAILQ_REMOVE(&conn_handle->cache, c, wterl_ctx, entries);
         conn_handle->cache_size -= 1;
         c->session->close(c->session, NULL);
-        enif_free(c);
+        free(c);
         c = n;
     }
 }
@@ -446,30 +419,9 @@ void
 __close_cursors_on(WterlConnHandle *conn_handle, const char *uri)
 {
     struct wterl_ctx *c, *n;
-    int worker_id, idx, cnt;
+    int idx, cnt;
 
-    // walk the mru first, look for open cursors on matching uri
-    for (worker_id = 0; worker_id < ASYNC_NIF_MAX_WORKERS; worker_id++) {
-	do c = conn_handle->mru_ctx[worker_id];
-	while(c && !ATOMIC_CAS_FULL(&conn_handle->mru_ctx[worker_id], c, 0));
-
-	if (c) {
-            cnt = c->num_cursors;
-            for(idx = 0; idx < cnt; idx++) {
-                if (!strcmp(c->ci[idx].uri, uri)) {
-                    c->session->close(c->session, NULL);
-                    enif_free(c);
-                    break;
-                } else {
-		    // not a match, be lazy and add it to the cache rather than
-		    // putting it back on the mru
-		    __ctx_cache_add(conn_handle, c);
-                }
-            }
-        }
-    }
-
-    // next we walk the cache, look for open cursors on matching uri
+    // walk the entries in the cache, look for open cursors on matching uri
     c = STAILQ_FIRST(&conn_handle->cache);
     while (c != NULL) {
         n = STAILQ_NEXT(c, entries);
@@ -479,7 +431,7 @@ __close_cursors_on(WterlConnHandle *conn_handle, const char *uri)
                 STAILQ_REMOVE(&conn_handle->cache, c, wterl_ctx, entries);
                 conn_handle->cache_size -= 1;
                 c->session->close(c->session, NULL);
-                enif_free(c);
+                free(c);
                 break;
             }
         }
@@ -488,6 +440,7 @@ __close_cursors_on(WterlConnHandle *conn_handle, const char *uri)
     return;
 }
 
+
 /**
  * Callback to handle error messages.
  *
@@ -502,13 +455,15 @@ __close_cursors_on(WterlConnHandle *conn_handle, const char *uri)
  *          operation or library failure.
  */
 int
-__wterl_error_handler(WT_EVENT_HANDLER *handler, int error, const char *message)
+__wterl_error_handler(WT_EVENT_HANDLER *handler, WT_SESSION *session,
+    int error, const char *message)
 {
     struct wterl_event_handlers *eh = (struct wterl_event_handlers *)handler;
     ErlNifEnv *msg_env;
     ErlNifPid *to_pid;
     int rc = 0;
 
+    UNUSED(session);
     enif_mutex_lock(eh->error_mutex);
     msg_env = eh->msg_env_error;
     to_pid = &eh->to_pid;
@@ -540,13 +495,14 @@ __wterl_error_handler(WT_EVENT_HANDLER *handler, int error, const char *message)
  *          operation or library failure.
  */
 int
-__wterl_message_handler(WT_EVENT_HANDLER *handler, const char *message)
+__wterl_message_handler(WT_EVENT_HANDLER *handler, WT_SESSION *session, const char *message)
 {
     struct wterl_event_handlers *eh = (struct wterl_event_handlers *)handler;
     ErlNifEnv *msg_env;
     ErlNifPid *to_pid;
     int rc = 0;
 
+    UNUSED(session);
     enif_mutex_lock(eh->message_mutex);
     msg_env = eh->msg_env_message;
     to_pid = &eh->to_pid;
@@ -577,13 +533,14 @@ __wterl_message_handler(WT_EVENT_HANDLER *handler, const char *message)
  *          operation or library failure.
  */
 int
-__wterl_progress_handler(WT_EVENT_HANDLER *handler, const char *operation, uint64_t counter)
+__wterl_progress_handler(WT_EVENT_HANDLER *handler, WT_SESSION *session, const char *operation, uint64_t counter)
 {
     struct wterl_event_handlers *eh = (struct wterl_event_handlers *)handler;
     ErlNifEnv *msg_env;
     ErlNifPid *to_pid;
     int rc = 0;
 
+    UNUSED(session);
     enif_mutex_lock(eh->progress_mutex);
     msg_env = eh->msg_env_progress;
     to_pid = &eh->to_pid;
@@ -685,7 +642,7 @@ ASYNC_NIF_DECL(
           return;
       }
       if (session_config.size > 1) {
-          char *sc = enif_alloc(session_config.size);
+          char *sc = malloc(session_config.size);
           if (!sc) {
               enif_release_resource(conn_handle);
               ASYNC_NIF_REPLY(__strerror_term(env, ENOMEM));
@@ -696,7 +653,7 @@ ASYNC_NIF_DECL(
       } else {
           conn_handle->session_config = NULL;
       }
-      conn_handle->cache_mutex = enif_mutex_create(NULL);
+      conn_handle->cache_mutex = enif_mutex_create("conn_handle");
       enif_mutex_lock(conn_handle->cache_mutex);
       conn_handle->conn = conn;
       ERL_NIF_TERM result = enif_make_resource(env, conn_handle);
@@ -744,8 +701,9 @@ ASYNC_NIF_DECL(
 
     /* Free up the shared sessions and cursors. */
     enif_mutex_lock(args->conn_handle->cache_mutex);
+    __close_all_sessions(args->conn_handle);
     if (args->conn_handle->session_config) {
-        enif_free((char *)args->conn_handle->session_config);
+        free((char *)args->conn_handle->session_config);
         args->conn_handle->session_config = NULL;
     }
     WT_CONNECTION* conn = args->conn_handle->conn;
@@ -1383,6 +1341,10 @@ ASYNC_NIF_DECL(
       ASYNC_NIF_REPLY(enif_make_badarg(env));
       return;
     }
+    if (key.size == 0) {
+      ASYNC_NIF_REPLY(enif_make_badarg(env));
+      return;
+    }
 
     struct wterl_ctx *ctx = NULL;
     WT_CURSOR *cursor = NULL;
@@ -1442,6 +1404,10 @@ ASYNC_NIF_DECL(
       ASYNC_NIF_REPLY(enif_make_badarg(env));
       return;
     }
+    if (key.size == 0) {
+      ASYNC_NIF_REPLY(enif_make_badarg(env));
+      return;
+    }
 
     struct wterl_ctx *ctx = NULL;
     WT_CURSOR *cursor = NULL;
@@ -1527,6 +1493,10 @@ ASYNC_NIF_DECL(
       ASYNC_NIF_REPLY(enif_make_badarg(env));
       return;
     }
+    if (key.size == 0 || value.size == 0) {
+      ASYNC_NIF_REPLY(enif_make_badarg(env));
+      return;
+    }
 
     struct wterl_ctx *ctx = NULL;
     WT_CURSOR *cursor = NULL;
@@ -1600,7 +1570,7 @@ ASYNC_NIF_DECL(
     }
 
     WT_CURSOR* cursor;
-    rc = session->open_cursor(session, args->uri, NULL, (config.data[0] != 0) ? (char *)config.data : "overwrite,raw", &cursor);
+    rc = session->open_cursor(session, args->uri, NULL, (config.data[0] != 0) ? (char *)config.data : "raw", &cursor);
     if (rc != 0) {
       session->close(session, NULL);
       ASYNC_NIF_REPLY(__strerror_term(env, rc));
@@ -2300,15 +2270,15 @@ on_load(ErlNifEnv *env, void **priv_data, ERL_NIF_TERM load_info)
     ATOM_WIREDTIGER_VSN = enif_make_atom(env, "wiredtiger_vsn");
     ATOM_MSG_PID = enif_make_atom(env, "message_pid");
 
-    struct wterl_priv_data *priv = enif_alloc(sizeof(struct wterl_priv_data));
+    struct wterl_priv_data *priv = malloc(sizeof(struct wterl_priv_data));
     if (!priv)
         return ENOMEM;
     memset(priv, 0, sizeof(struct wterl_priv_data));
 
     struct wterl_event_handlers *eh = &priv->eh;
-    eh->error_mutex = enif_mutex_create(NULL);
-    eh->message_mutex = enif_mutex_create(NULL);
-    eh->progress_mutex = enif_mutex_create(NULL);
+    eh->error_mutex = enif_mutex_create("error_mutex");
+    eh->message_mutex = enif_mutex_create("message_mutex");
+    eh->progress_mutex = enif_mutex_create("progress_mutex");
 
     /* Process the load_info array of tuples, we expect:
        [{wterl_vsn, "a version string"},
@@ -2328,17 +2298,17 @@ on_load(ErlNifEnv *env, void **priv_data, ERL_NIF_TERM load_info)
 
     /* Note: !!! the first element of our priv_data struct *must* be the
        pointer to the async_nif's private data which we set here. */
-    ASYNC_NIF_LOAD(wterl, priv->async_nif_priv);
+    ASYNC_NIF_LOAD(wterl, env, priv->async_nif_priv);
     if (!priv->async_nif_priv) {
         memset(priv, 0, sizeof(struct wterl_priv_data));
-        enif_free(priv);
+        free(priv);
         return ENOMEM;
     }
     *priv_data = priv;
 
     char msg[1024];
     snprintf(msg, 1024, "NIF on_load complete (wterl version: %s, wiredtiger version: %s)", priv->wterl_vsn, priv->wiredtiger_vsn);
-    __wterl_message_handler((WT_EVENT_HANDLER *)&priv->eh, msg);
+    __wterl_message_handler((WT_EVENT_HANDLER *)&priv->eh, NULL, msg);
     return 0;
 }
 
@@ -2377,7 +2347,7 @@ on_unload(ErlNifEnv *env, void *priv_data)
         enif_free_env(eh->msg_env_progress);
 
     memset(priv, 0, sizeof(struct wterl_priv_data));
-    enif_free(priv);
+    free(priv);
 
     priv_data = NULL;
 }

priv/wterl.schema

@@ -0,0 +1,6 @@
+%%%% This is the WiredTiger section
+
+%% @doc wiredtiger data_root
+{mapping, "wiredtiger.data_root", "wterl.data_root", [
+  {default, "{{platform_data_dir}}/wiredtiger"}
+]}.

rebar.config

@@ -5,13 +5,15 @@
 
 {cover_enabled, true}.
 
-{eunit_opts, [verbose, {report, {eunit_surefire, [{dir, "."}]}}]}.
+%{eunit_opts, [verbose, {report, {eunit_surefire, [{dir, "."}]}}]}.
 
 {erl_opts, [
+	    %native, {hipe, [o3,verbose]}, inline, {inline_size, 1024},
             {parse_transform, lager_transform},
-            debug_info, %{d,'DEBUG',true},
-            %strict_validation,
-            %fail_on_warning,
+            debug_info,
+	    {d,'DEBUG',true},
+            strict_validation,
+            fail_on_warning,
 	    %warn_missing_spec,
             warn_bif_clash,
             warn_deprecated_function,
@@ -22,7 +24,7 @@
             warn_shadow_vars,
             warn_untyped_record,
             warn_unused_function,
-            %warn_unused_import,
+            warn_unused_import,
             warn_unused_record,
             warn_unused_vars
             ]}.
@@ -36,8 +38,8 @@
 {port_specs, [{"priv/wterl.so", ["c_src/*.c"]}]}.
 
 {port_env, [
-            {"DRV_CFLAGS",  "$DRV_CFLAGS -fPIC -Wall -Wextra -Werror -I c_src/system/include"},
-            {"DRV_LDFLAGS", "$DRV_LDFLAGS -Wl,-rpath,lib/wterl/priv:priv -Lc_src/system/lib -lwiredtiger"}
+            {"DRV_CFLAGS",  "$DRV_CFLAGS  -O3 -mtune=native -march=native -fPIC -Wall -Wextra -Werror -I c_src/system/include"},
+            {"DRV_LDFLAGS", "$DRV_LDFLAGS -Wl,-rpath,lib/wterl/priv:lib/wterl-0.9.0/priv:priv -Lc_src/system/lib -lwiredtiger"}
            ]}.
 
 {pre_hooks, [{compile, "c_src/build_deps.sh compile"}]}.

src/async_nif.hrl

@@ -21,27 +21,34 @@
 %%
 %% -------------------------------------------------------------------
 
--spec async_nif_enqueue(reference(), function(), [term()]) -> term() | {error, term()}.
-async_nif_enqueue(R, F, A) ->
-    case erlang:apply(F, [R|A]) of
-        {ok, enqueued} ->
-            receive
-                {R, {error, shutdown}=Error} ->
-                    %% Work unit was queued, but not executed.
-                    Error;
-                {R, {error, _Reason}=Error} ->
-                    %% Work unit returned an error.
-                    Error;
-                {R, Reply} ->
-                    Reply
-            end;
-        {error, eagain} ->
-            %% Work unit was not queued, try again.
-            async_nif_enqueue(R, F, A);
-        %{error, enomem} ->
-        %{error, shutdown} ->
-        Other ->
-            Other
-    end.
-
--define(ASYNC_NIF_CALL(Fun, Args), async_nif_enqueue(erlang:make_ref(), Fun, Args)).
+-define(ASYNC_NIF_CALL(Fun, Args),
+	F = fun(F, T) ->
+		    R = erlang:make_ref(),
+		    case erlang:apply(Fun, [R|Args]) of
+			{ok, {enqueued, PctBusy}} ->
+			    if
+				PctBusy > 0.25 andalso PctBusy =< 1.0 ->
+				    erlang:bump_reductions(erlang:trunc(2000 * PctBusy));
+			       true ->
+				    ok
+			    end,
+			    receive
+				{R, {error, shutdown}=Error} ->
+				    %% Work unit was queued, but not executed.
+				    Error;
+				{R, {error, _Reason}=Error} ->
+				    %% Work unit returned an error.
+				    Error;
+				{R, Reply} ->
+				    Reply
+			    end;
+			{error, eagain} ->
+			    case T of
+				3 -> not_found;
+				_ -> F(F, T + 1)
+			    end;
+			Other ->
+			    Other
+		    end
+	    end,
+	F(F, 1)).

src/riak_kv_wterl_backend.erl

@@ -22,6 +22,7 @@
 
 -module(riak_kv_wterl_backend).
 -behavior(temp_riak_kv_backend).
+-compile([{parse_transform, lager_transform}]).
 
 %% KV Backend API
 -export([api_version/0,
@@ -42,7 +43,7 @@
 
 -ifdef(TEST).
 -include_lib("eunit/include/eunit.hrl").
--compiel(export_all).
+-compile(export_all).
 -endif.
 
 -define(API_VERSION, 1).
@@ -119,14 +120,14 @@ start(Partition, Config) ->
                     "lsm" ->
                         [{internal_page_max, "128K"},
                          {leaf_page_max, "16K"},
-                         {lsm_chunk_size, "100MB"},
-                         {lsm_merge_threads, 2},
-                         {prefix_compression, true},
-                         {lsm_bloom_newest, true},
-                         {lsm_bloom_oldest, true} ,
-                         {lsm_bloom_bit_count, 28},
-                         {lsm_bloom_hash_count, 19},
-                         {lsm_bloom_config, [{leaf_page_max, "8MB"}]}
+                         {lsm, [
+                             {bloom_config, [{leaf_page_max, "8MB"}]},
+                             {bloom_bit_count, 28},
+                             {bloom_hash_count, 19},
+                             {bloom_oldest, true},
+                             {chunk_size, "100MB"},
+                             {merge_threads, 2}
+                         ]}
                         ] ++ Compressor;
                     "table" ->
                         Compressor
@@ -341,22 +342,23 @@ is_empty(#state{connection=Connection, table=Table}) ->
 %% @doc Get the status information for this wterl backend
 -spec status(state()) -> [{atom(), term()}].
 status(#state{connection=Connection, table=Table}) ->
-    case wterl:cursor_open(Connection, Table) of
-        {ok, Cursor} ->
-	    TheStats =
-		case fetch_status(Cursor) of
-		    {ok, Stats} ->
-			Stats;
-		    {error, {eperm, _}} -> % TODO: review/fix this logic
-			{ok, []};
-		    _ ->
-			{ok, []}
-		end,
-	    wterl:cursor_close(Cursor),
-	    TheStats;
-        {error, Reason2} ->
-            {error, Reason2}
-    end.
+    [].
+%%     case wterl:cursor_open(Connection, "statistics:" ++ Table, [{statistics_fast, true}]) of
+%%         {ok, Cursor} ->
+%% 	    TheStats =
+%% 		case fetch_status(Cursor) of
+%% 		    {ok, Stats} ->
+%% 			Stats;
+%% 		    {error, {eperm, _}} -> % TODO: review/fix this logic
+%% 			{ok, []};
+%% 		    _ ->
+%% 			{ok, []}
+%% 		end,
+%% 	    wterl:cursor_close(Cursor),
+%% 	    TheStats;
+%%         {error, Reason2} ->
+%%             {error, Reason2}
+%%    end.
 
 %% @doc Register an asynchronous callback
 -spec callback(reference(), any(), state()) -> {ok, state()}.
@@ -399,32 +401,41 @@ establish_connection(Config, Type) ->
             ok = filelib:ensure_dir(filename:join(DataRoot, "x")),
 
             %% WT Connection Options:
-            %% NOTE: LSM auto-checkpoints, so we don't have too.
+	    LogSetting = app_helper:get_prop_or_env(log, Config, wterl, false),
             CheckpointSetting =
                 case Type =:= "lsm" of
                     true ->
-                        [];
+			case LogSetting of
+			    true ->
+				%% Turn checkpoints on if logging is on, checkpoints enable log archival.
+				app_helper:get_prop_or_env(checkpoint, Config, wterl, [{wait, 30}]);  % in seconds
+			    _ ->
+				[]
+			end;
                     false ->
-                        app_helper:get_prop_or_env(checkpoint, Config, wterl, [{wait, 10}])
+                        app_helper:get_prop_or_env(checkpoint, Config, wterl, [{wait, 30}])
                 end,
             RequestedCacheSize = app_helper:get_prop_or_env(cache_size, Config, wterl),
             ConnectionOpts =
                 orddict:from_list(
                   [ wterl:config_value(create, Config, true),
-                    wterl:config_value(sync, Config, false),
-                    wterl:config_value(logging, Config, true),
+                    wterl:config_value(checkpoint_sync, Config, false),
+                    wterl:config_value(transaction_sync, Config, "none"),
+                    wterl:config_value(log, Config, [{enabled, LogSetting}]),
+                    wterl:config_value(mmap, Config, false),
+                    wterl:config_value(checkpoint, Config, CheckpointSetting),
                     wterl:config_value(session_max, Config, max_sessions(Config)),
                     wterl:config_value(cache_size, Config, size_cache(RequestedCacheSize)),
-                    wterl:config_value(statistics_log, Config, [{wait, 300}]), % sec
+                    wterl:config_value(statistics, Config, [ "fast", "clear"]),
+                    wterl:config_value(statistics_log, Config, [{wait, 600}]), % in seconds
                     wterl:config_value(verbose, Config, [ "salvage", "verify"
                          % Note: for some unknown reason, if you add these additional
                          % verbose flags Erlang SEGV's "size_object: bad tag for 0x80"
-                         % no idea why... yet... you've been warned.
-
+                         % no idea why... you've been warned.
                          %"block", "shared_cache", "reconcile", "evict", "lsm",
                          %"fileops", "read", "write", "readserver", "evictserver",
                          %"hazard", "mutex", "ckpt"
-                         ]) ] ++ CheckpointSetting ++ proplists:get_value(wterl, Config, [])), % sec
+                         ]) ] ++ proplists:get_value(wterl, Config, [])), % sec
 
             %% WT Session Options:
             SessionOpts = [{isolation, "snapshot"}],
@@ -545,15 +556,15 @@ from_index_key(LKey) ->
 
 %% @private
 %% Return all status from wterl statistics cursor
-fetch_status(Cursor) ->
-    {ok, fetch_status(Cursor, wterl:cursor_next_value(Cursor), [])}.
-fetch_status(_Cursor, {error, _}, Acc) ->
-    lists:reverse(Acc);
-fetch_status(_Cursor, not_found, Acc) ->
-    lists:reverse(Acc);
-fetch_status(Cursor, {ok, Stat}, Acc) ->
-    [What,Val|_] = [binary_to_list(B) || B <- binary:split(Stat, [<<0>>], [global])],
-    fetch_status(Cursor, wterl:cursor_next_value(Cursor), [{What,Val}|Acc]).
+%% fetch_status(Cursor) ->
+%%    {ok, fetch_status(Cursor, wterl:cursor_next_value(Cursor), [])}.
+%% fetch_status(_Cursor, {error, _}, Acc) ->
+%%     lists:reverse(Acc);
+%% fetch_status(_Cursor, not_found, Acc) ->
+%%     lists:reverse(Acc);
+%% fetch_status(Cursor, {ok, Stat}, Acc) ->
+%%     [What,Val|_] = [binary_to_list(B) || B <- binary:split(Stat, [<<0>>], [global])],
+%%     fetch_status(Cursor, wterl:cursor_next_value(Cursor), [{What,Val}|Acc]).
 
 size_cache(RequestedSize) ->
     Size =

src/wterl.erl

@@ -96,8 +96,8 @@ nif_stub_error(Line) ->
 -spec init() -> ok | {error, any()}.
 init() ->
     erlang:load_nif(filename:join([priv_dir(), atom_to_list(?MODULE)]),
-           [{wterl_vsn, "53307e8"},
-	    {wiredtiger_vsn, "1.6.2-0-g07cb0a5"}]).
+           [{wterl_vsn, "942e51b"},
+	    {wiredtiger_vsn, "1.6.4-275-g9c44420"}]). %% TODO automate these
 
 -spec connection_open(string(), config_list()) -> {ok, connection()} | {error, term()}.
 -spec connection_open(string(), config_list(), config_list()) -> {ok, connection()} | {error, term()}.
@@ -256,6 +256,7 @@ verify_nif(_AsyncRef, _Ref, _Name, _Config) ->
 -spec cursor_open(connection(), string(), config_list()) -> {ok, cursor()} | {error, term()}.
 cursor_open(Ref, Table) ->
     cursor_open(Ref, Table, []).
+
 cursor_open(Ref, Table, Config) ->
     ?ASYNC_NIF_CALL(fun cursor_open_nif/4, [Ref, Table, config_to_bin(Config)]).
 
@@ -453,17 +454,26 @@ config_to_bin([], Acc) ->
 config_to_bin([{Key, Value} | Rest], Acc) ->
     ConfigTypes =
     [{block_compressor, {string, quoted}},
+     {bloom_bit_count, integer},
+     {bloom_config, config},
+     {bloom_hash_count, integer},
+     {bloom_newest, bool},
+     {bloom_oldest, bool},
      {cache_size, string},
      {checkpoint, config},
+     {checkpoint_sync, bool},
      {checksum, string},
+     {chunk_size, string},
      {create, bool},
      {direct_io, list},
      {drop, list},
+     {enabled, bool},
      {error_prefix, string},
      {eviction_target, integer},
      {eviction_trigger, integer},
      {extensions, {list, quoted}},
      {statistics_fast, bool},
+     {file_max, string},
      {force, bool},
      {from, string},
      {hazard_max, integer},
@@ -473,22 +483,21 @@ config_to_bin([{Key, Value} | Rest], Acc) ->
      {isolation, string},
      {key_type, string},
      {leaf_page_max, string},
-     {logging, bool},
-     {lsm_bloom_bit_count, integer},
-     {lsm_bloom_config, config},
-     {lsm_bloom_hash_count, integer},
-     {lsm_bloom_newest, bool},
-     {lsm_bloom_oldest, bool},
-     {lsm_chunk_size, string},
-     {prefix_compression, bool},
-     {lsm_merge_threads, integer},
+     {log, config},
+     {lsm, config},
+     {mmap, bool},
+     {merge_threads, integer},
      {multiprocess, bool},
      {name, string},
+     {overwrite, bool},
+     {prefix_compression, bool},
+     {raw, bool},
      {session_max, integer},
+     {statistics, list},
      {statistics_log, config},
-     {sync, bool},
      {target, {list, quoted}},
      {to, string},
+     {transaction_sync, string},
      {transactional, bool},
      {verbose, list},
      {wait, integer}],
@@ -610,7 +619,7 @@ many_open_tables_test_() ->
 	     DataDir = ?TEST_DATA_DIR,
 	     KeyGen =
 		 fun(X) ->
-			 crypto:sha(<<X>>)
+			 crypto:hash(sha, <<X>>)
 		 end,
 	     ValGen =
 		 fun() ->
@@ -872,7 +881,7 @@ various_cursor_test_() ->
                 end},
                {"update an item using a cursor",
                 fun() ->
-                        {ok, Cursor} = cursor_open(ConnRef, "table:test"),
+                        {ok, Cursor} = cursor_open(ConnRef, "table:test", [{overwrite, false}, {raw,true}]),
                         ?assertMatch(ok, cursor_update(Cursor, <<"g">>, <<"goji berries">>)),
                         ?assertMatch(not_found, cursor_update(Cursor, <<"k">>, <<"kumquat">>)),
                         ?assertMatch(ok, cursor_close(Cursor)),
@@ -880,7 +889,7 @@ various_cursor_test_() ->
                 end},
                {"remove an item using a cursor",
                 fun() ->
-                        {ok, Cursor} = cursor_open(ConnRef, "table:test"),
+                        {ok, Cursor} = cursor_open(ConnRef, "table:test", [{overwrite, false}, {raw,true}]),
                         ?assertMatch(ok, cursor_remove(Cursor, <<"g">>)),
                         ?assertMatch(not_found, cursor_remove(Cursor, <<"l">>)),
                         ?assertMatch(ok, cursor_close(Cursor)),

tools/wterl-b_b.config

@@ -25,7 +25,13 @@
 
 {mode, max}.
 {duration, 10}.
-{concurrent, 4}.
+{concurrent, 16}.
+{report_interval, 1}.
+{pb_timeout_general, 1000}. % ms
+%{pb_timeout_read, ?}.
+%{pb_timeout_write, ?}.
+%{pb_timeout_listkeys, ?}.
+%{pb_timeout_mapreduce, ?}.
 {driver, basho_bench_driver_wterl}.
 {key_generator, {int_to_bin_littleendian,{uniform_int, 5000000}}}.
 {value_generator, {fixed_bin, 10000}}.
@@ -37,9 +43,9 @@
 {wterl, [
         {connection, [
                       {create, true},
-                      {sync, false},
-                      {logging, true},
-                      {transactional, true},
+                      {session_sync, false},
+		      {transaction_sync, "none"},
+		      {log, [{enabled, false}]},
                       {session_max, 1024},
                       {cache_size, 4294967296},
                       {verbose, []},
@@ -54,11 +60,11 @@
                      ]},
         {session, [ {isolation, "snapshot"} ]},
         {table_uri, "lsm:test"},
+	{lsm_merge_threads, 2},
         {table, [
                  {internal_page_max, "128K"},
                  {leaf_page_max, "128K"},
                  {lsm_chunk_size, "25MB"},
-                 {prefix_compression, false},
                  {lsm_bloom_newest, true},
                  {lsm_bloom_oldest, true} ,
                  {lsm_bloom_bit_count, 128},
@@ -72,9 +78,9 @@
 {wterl_, [
         {connection, [
                       {create, true},
-                      {sync, false},
-                      {logging, true},
-                      {transactional, true},
+                      {session_sync, false},
+		      {transaction_sync, "none"},
+		      {log, [{enabled, false}]},
                       {session_max, 1024},
                       {cache_size, 4294967296},
                       {verbose, []},
@@ -91,7 +97,6 @@
         {session, [ {isolation, "snapshot"} ]},
         {table_uri, "table:test"},
         {table, [
-                 {prefix_compression, false},
                  {block_compressor, "snappy"} % bzip2
                 ]}
         ]}.