Index: vendor-sys/ck/dist/include/gcc/ppc/ck_pr.h =================================================================== --- vendor-sys/ck/dist/include/gcc/ppc/ck_pr.h (revision 337530) +++ vendor-sys/ck/dist/include/gcc/ppc/ck_pr.h (revision 337531) @@ -1,327 +1,335 @@ /* * Copyright 2009-2015 Samy Al Bahra. * Copyright 2012 João Fernandes. * All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. */ #ifndef CK_PR_PPC_H #define CK_PR_PPC_H #ifndef CK_PR_H #error Do not include this file directly, use ck_pr.h #endif #include #include /* * The following represent supported atomic operations. * These operations may be emulated. */ #include "ck_f_pr.h" /* * Minimum interface requirement met. */ #define CK_F_PR /* * This bounces the hardware thread from low to medium * priority. I am unsure of the benefits of this approach * but it is used by the Linux kernel. */ CK_CC_INLINE static void ck_pr_stall(void) { __asm__ __volatile__("or 1, 1, 1;" "or 2, 2, 2;" ::: "memory"); return; } #define CK_PR_FENCE(T, I) \ CK_CC_INLINE static void \ ck_pr_fence_strict_##T(void) \ { \ __asm__ __volatile__(I ::: "memory"); \ } -CK_PR_FENCE(atomic, "lwsync") -CK_PR_FENCE(atomic_store, "lwsync") +#ifdef CK_MD_PPC32_LWSYNC +#define CK_PR_LWSYNCOP "lwsync" +#else /* CK_MD_PPC32_LWSYNC_DISABLE */ +#define CK_PR_LWSYNCOP "sync" +#endif + +CK_PR_FENCE(atomic, CK_PR_LWSYNCOP) +CK_PR_FENCE(atomic_store, CK_PR_LWSYNCOP) CK_PR_FENCE(atomic_load, "sync") -CK_PR_FENCE(store_atomic, "lwsync") -CK_PR_FENCE(load_atomic, "lwsync") -CK_PR_FENCE(store, "lwsync") +CK_PR_FENCE(store_atomic, CK_PR_LWSYNCOP) +CK_PR_FENCE(load_atomic, CK_PR_LWSYNCOP) +CK_PR_FENCE(store, CK_PR_LWSYNCOP) CK_PR_FENCE(store_load, "sync") -CK_PR_FENCE(load, "lwsync") -CK_PR_FENCE(load_store, "lwsync") +CK_PR_FENCE(load, CK_PR_LWSYNCOP) +CK_PR_FENCE(load_store, CK_PR_LWSYNCOP) CK_PR_FENCE(memory, "sync") -CK_PR_FENCE(acquire, "lwsync") -CK_PR_FENCE(release, "lwsync") -CK_PR_FENCE(acqrel, "lwsync") -CK_PR_FENCE(lock, "lwsync") -CK_PR_FENCE(unlock, "lwsync") +CK_PR_FENCE(acquire, CK_PR_LWSYNCOP) +CK_PR_FENCE(release, CK_PR_LWSYNCOP) +CK_PR_FENCE(acqrel, CK_PR_LWSYNCOP) +CK_PR_FENCE(lock, CK_PR_LWSYNCOP) +CK_PR_FENCE(unlock, CK_PR_LWSYNCOP) + +#undef CK_PR_LWSYNCOP #undef CK_PR_FENCE #define CK_PR_LOAD(S, M, T, C, I) \ CK_CC_INLINE static T \ ck_pr_md_load_##S(const M *target) \ { \ T r; \ __asm__ __volatile__(I "%U1%X1 %0, %1" \ : "=r" (r) \ : "m" (*(const C *)target) \ : "memory"); \ return (r); \ } CK_PR_LOAD(ptr, void, void *, uint32_t, "lwz") #define CK_PR_LOAD_S(S, T, I) CK_PR_LOAD(S, T, T, T, I) CK_PR_LOAD_S(32, uint32_t, "lwz") CK_PR_LOAD_S(16, uint16_t, "lhz") CK_PR_LOAD_S(8, uint8_t, "lbz") CK_PR_LOAD_S(uint, unsigned int, "lwz") CK_PR_LOAD_S(int, int, "lwz") CK_PR_LOAD_S(short, short, "lhz") CK_PR_LOAD_S(char, char, "lbz") #undef CK_PR_LOAD_S #undef CK_PR_LOAD #define CK_PR_STORE(S, M, T, C, I) \ CK_CC_INLINE static void \ ck_pr_md_store_##S(M *target, T v) \ { \ __asm__ __volatile__(I "%U0%X0 %1, %0" \ : "=m" (*(C *)target) \ : "r" (v) \ : "memory"); \ return; \ } CK_PR_STORE(ptr, void, const void *, uint32_t, "stw") #define CK_PR_STORE_S(S, T, I) CK_PR_STORE(S, T, T, T, I) CK_PR_STORE_S(32, uint32_t, "stw") CK_PR_STORE_S(16, uint16_t, "sth") CK_PR_STORE_S(8, uint8_t, "stb") CK_PR_STORE_S(uint, unsigned int, "stw") CK_PR_STORE_S(int, int, "stw") CK_PR_STORE_S(short, short, "sth") CK_PR_STORE_S(char, char, "stb") #undef CK_PR_STORE_S #undef CK_PR_STORE #define CK_PR_CAS(N, T, M) \ CK_CC_INLINE static bool \ ck_pr_cas_##N##_value(M *target, T compare, T set, M *value) \ { \ T previous; \ __asm__ __volatile__("1:" \ "lwarx %0, 0, %1;" \ "cmpw 0, %0, %3;" \ "bne- 2f;" \ "stwcx. %2, 0, %1;" \ "bne- 1b;" \ "2:" \ : "=&r" (previous) \ : "r" (target), \ "r" (set), \ "r" (compare) \ : "memory", "cc"); \ *(T *)value = previous; \ return (previous == compare); \ } \ CK_CC_INLINE static bool \ ck_pr_cas_##N(M *target, T compare, T set) \ { \ T previous; \ __asm__ __volatile__("1:" \ "lwarx %0, 0, %1;" \ "cmpw 0, %0, %3;" \ "bne- 2f;" \ "stwcx. %2, 0, %1;" \ "bne- 1b;" \ "2:" \ : "=&r" (previous) \ : "r" (target), \ "r" (set), \ "r" (compare) \ : "memory", "cc"); \ return (previous == compare); \ } CK_PR_CAS(ptr, void *, void) #define CK_PR_CAS_S(a, b) CK_PR_CAS(a, b, b) CK_PR_CAS_S(32, uint32_t) CK_PR_CAS_S(uint, unsigned int) CK_PR_CAS_S(int, int) #undef CK_PR_CAS_S #undef CK_PR_CAS #define CK_PR_FAS(N, M, T, W) \ CK_CC_INLINE static T \ ck_pr_fas_##N(M *target, T v) \ { \ T previous; \ __asm__ __volatile__("1:" \ "l" W "arx %0, 0, %1;" \ "st" W "cx. %2, 0, %1;" \ "bne- 1b;" \ : "=&r" (previous) \ : "r" (target), \ "r" (v) \ : "memory", "cc"); \ return (previous); \ } CK_PR_FAS(32, uint32_t, uint32_t, "w") CK_PR_FAS(ptr, void, void *, "w") CK_PR_FAS(int, int, int, "w") CK_PR_FAS(uint, unsigned int, unsigned int, "w") #undef CK_PR_FAS #define CK_PR_UNARY(O, N, M, T, I, W) \ CK_CC_INLINE static void \ ck_pr_##O##_##N(M *target) \ { \ T previous; \ __asm__ __volatile__("1:" \ "l" W "arx %0, 0, %1;" \ I ";" \ "st" W "cx. %0, 0, %1;" \ "bne- 1b;" \ : "=&r" (previous) \ : "r" (target) \ : "memory", "cc"); \ return; \ } CK_PR_UNARY(inc, ptr, void, void *, "addic %0, %0, 1", "w") CK_PR_UNARY(dec, ptr, void, void *, "addic %0, %0, -1", "w") CK_PR_UNARY(not, ptr, void, void *, "not %0, %0", "w") CK_PR_UNARY(neg, ptr, void, void *, "neg %0, %0", "w") #define CK_PR_UNARY_S(S, T, W) \ CK_PR_UNARY(inc, S, T, T, "addic %0, %0, 1", W) \ CK_PR_UNARY(dec, S, T, T, "addic %0, %0, -1", W) \ CK_PR_UNARY(not, S, T, T, "not %0, %0", W) \ CK_PR_UNARY(neg, S, T, T, "neg %0, %0", W) CK_PR_UNARY_S(32, uint32_t, "w") CK_PR_UNARY_S(uint, unsigned int, "w") CK_PR_UNARY_S(int, int, "w") #undef CK_PR_UNARY_S #undef CK_PR_UNARY #define CK_PR_BINARY(O, N, M, T, I, W) \ CK_CC_INLINE static void \ ck_pr_##O##_##N(M *target, T delta) \ { \ T previous; \ __asm__ __volatile__("1:" \ "l" W "arx %0, 0, %1;" \ I " %0, %2, %0;" \ "st" W "cx. %0, 0, %1;" \ "bne- 1b;" \ : "=&r" (previous) \ : "r" (target), \ "r" (delta) \ : "memory", "cc"); \ return; \ } CK_PR_BINARY(and, ptr, void, uintptr_t, "and", "w") CK_PR_BINARY(add, ptr, void, uintptr_t, "add", "w") CK_PR_BINARY(or, ptr, void, uintptr_t, "or", "w") CK_PR_BINARY(sub, ptr, void, uintptr_t, "sub", "w") CK_PR_BINARY(xor, ptr, void, uintptr_t, "xor", "w") #define CK_PR_BINARY_S(S, T, W) \ CK_PR_BINARY(and, S, T, T, "and", W) \ CK_PR_BINARY(add, S, T, T, "add", W) \ CK_PR_BINARY(or, S, T, T, "or", W) \ CK_PR_BINARY(sub, S, T, T, "subf", W) \ CK_PR_BINARY(xor, S, T, T, "xor", W) CK_PR_BINARY_S(32, uint32_t, "w") CK_PR_BINARY_S(uint, unsigned int, "w") CK_PR_BINARY_S(int, int, "w") #undef CK_PR_BINARY_S #undef CK_PR_BINARY CK_CC_INLINE static void * ck_pr_faa_ptr(void *target, uintptr_t delta) { uintptr_t previous, r; __asm__ __volatile__("1:" "lwarx %0, 0, %2;" "add %1, %3, %0;" "stwcx. %1, 0, %2;" "bne- 1b;" : "=&r" (previous), "=&r" (r) : "r" (target), "r" (delta) : "memory", "cc"); return (void *)(previous); } #define CK_PR_FAA(S, T, W) \ CK_CC_INLINE static T \ ck_pr_faa_##S(T *target, T delta) \ { \ T previous, r; \ __asm__ __volatile__("1:" \ "l" W "arx %0, 0, %2;" \ "add %1, %3, %0;" \ "st" W "cx. %1, 0, %2;" \ "bne- 1b;" \ : "=&r" (previous), \ "=&r" (r) \ : "r" (target), \ "r" (delta) \ : "memory", "cc"); \ return (previous); \ } CK_PR_FAA(32, uint32_t, "w") CK_PR_FAA(uint, unsigned int, "w") CK_PR_FAA(int, int, "w") #undef CK_PR_FAA #endif /* CK_PR_PPC_H */ Index: vendor-sys/ck/dist/include/spinlock/hclh.h =================================================================== --- vendor-sys/ck/dist/include/spinlock/hclh.h (revision 337530) +++ vendor-sys/ck/dist/include/spinlock/hclh.h (revision 337531) @@ -1,145 +1,149 @@ /* * Copyright 2013-2015 Olivier Houchard * Copyright 2010-2015 Samy Al Bahra. * All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. */ #ifndef CK_SPINLOCK_HCLH_H #define CK_SPINLOCK_HCLH_H #include #include #include #include #ifndef CK_F_SPINLOCK_HCLH #define CK_F_SPINLOCK_HCLH struct ck_spinlock_hclh { unsigned int wait; unsigned int splice; int cluster_id; struct ck_spinlock_hclh *previous; }; typedef struct ck_spinlock_hclh ck_spinlock_hclh_t; CK_CC_INLINE static void ck_spinlock_hclh_init(struct ck_spinlock_hclh **lock, struct ck_spinlock_hclh *unowned, int cluster_id) { unowned->previous = NULL; unowned->wait = false; unowned->splice = false; unowned->cluster_id = cluster_id; *lock = unowned; ck_pr_barrier(); return; } CK_CC_INLINE static bool ck_spinlock_hclh_locked(struct ck_spinlock_hclh **queue) { struct ck_spinlock_hclh *head; bool r; head = ck_pr_load_ptr(queue); r = ck_pr_load_uint(&head->wait); ck_pr_fence_acquire(); return r; } CK_CC_INLINE static void ck_spinlock_hclh_lock(struct ck_spinlock_hclh **glob_queue, struct ck_spinlock_hclh **local_queue, struct ck_spinlock_hclh *thread) { struct ck_spinlock_hclh *previous, *local_tail; /* Indicate to the next thread on queue that they will have to block. */ thread->wait = true; thread->splice = false; thread->cluster_id = (*local_queue)->cluster_id; + /* Make sure previous->previous doesn't appear to be NULL */ + thread->previous = *local_queue; /* Serialize with respect to update of local queue. */ ck_pr_fence_store_atomic(); /* Mark current request as last request. Save reference to previous request. */ previous = ck_pr_fas_ptr(local_queue, thread); thread->previous = previous; /* Wait until previous thread from the local queue is done with lock. */ ck_pr_fence_load(); - if (previous->previous != NULL && - previous->cluster_id == thread->cluster_id) { - while (ck_pr_load_uint(&previous->wait) == true) + if (previous->previous != NULL) { + while (ck_pr_load_uint(&previous->wait) == true && + ck_pr_load_int(&previous->cluster_id) == thread->cluster_id && + ck_pr_load_uint(&previous->splice) == false) ck_pr_stall(); /* We're head of the global queue, we're done */ - if (ck_pr_load_uint(&previous->splice) == false) + if (ck_pr_load_int(&previous->cluster_id) == thread->cluster_id && + ck_pr_load_uint(&previous->splice) == false) return; } /* Now we need to splice the local queue into the global queue. */ local_tail = ck_pr_load_ptr(local_queue); previous = ck_pr_fas_ptr(glob_queue, local_tail); ck_pr_store_uint(&local_tail->splice, true); /* Wait until previous thread from the global queue is done with lock. */ while (ck_pr_load_uint(&previous->wait) == true) ck_pr_stall(); ck_pr_fence_lock(); return; } CK_CC_INLINE static void ck_spinlock_hclh_unlock(struct ck_spinlock_hclh **thread) { struct ck_spinlock_hclh *previous; /* * If there are waiters, they are spinning on the current node wait * flag. The flag is cleared so that the successor may complete an * acquisition. If the caller is pre-empted then the predecessor field * may be updated by a successor's lock operation. In order to avoid * this, save a copy of the predecessor before setting the flag. */ previous = thread[0]->previous; /* We have to pay this cost anyways, use it as a compiler barrier too. */ ck_pr_fence_unlock(); ck_pr_store_uint(&(*thread)->wait, false); /* * Predecessor is guaranteed not to be spinning on previous request, * so update caller to use previous structure. This allows successor * all the time in the world to successfully read updated wait flag. */ *thread = previous; return; } #endif /* CK_F_SPINLOCK_HCLH */ #endif /* CK_SPINLOCK_HCLH_H */ Index: vendor-sys/ck/dist/src/ck_barrier_combining.c =================================================================== --- vendor-sys/ck/dist/src/ck_barrier_combining.c (revision 337530) +++ vendor-sys/ck/dist/src/ck_barrier_combining.c (revision 337531) @@ -1,207 +1,207 @@ /* * Copyright 2011-2015 Samy Al Bahra. * Copyright 2011 David Joseph. * All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. */ #include #include #include #include struct ck_barrier_combining_queue { struct ck_barrier_combining_group *head; struct ck_barrier_combining_group *tail; }; -CK_CC_INLINE static struct ck_barrier_combining_group * +static struct ck_barrier_combining_group * ck_barrier_combining_queue_dequeue(struct ck_barrier_combining_queue *queue) { struct ck_barrier_combining_group *front = NULL; if (queue->head != NULL) { front = queue->head; queue->head = queue->head->next; } return front; } -CK_CC_INLINE static void +static void ck_barrier_combining_insert(struct ck_barrier_combining_group *parent, struct ck_barrier_combining_group *tnode, struct ck_barrier_combining_group **child) { *child = tnode; tnode->parent = parent; /* * After inserting, we must increment the parent group's count for * number of threads expected to reach it; otherwise, the * barrier may end prematurely. */ parent->k++; return; } /* * This implementation of software combining tree barriers * uses level order traversal to insert new thread groups * into the barrier's tree. We use a queue to implement this * traversal. */ -CK_CC_INLINE static void +static void ck_barrier_combining_queue_enqueue(struct ck_barrier_combining_queue *queue, struct ck_barrier_combining_group *node_value) { node_value->next = NULL; if (queue->head == NULL) { queue->head = queue->tail = node_value; return; } queue->tail->next = node_value; queue->tail = node_value; return; } void ck_barrier_combining_group_init(struct ck_barrier_combining *root, struct ck_barrier_combining_group *tnode, unsigned int nthr) { struct ck_barrier_combining_group *node; struct ck_barrier_combining_queue queue; queue.head = queue.tail = NULL; tnode->k = nthr; tnode->count = 0; tnode->sense = 0; tnode->left = tnode->right = NULL; /* * Finds the first available node for linkage into the combining * tree. The use of a spinlock is excusable as this is a one-time * initialization cost. */ ck_spinlock_fas_lock(&root->mutex); ck_barrier_combining_queue_enqueue(&queue, root->root); while (queue.head != NULL) { node = ck_barrier_combining_queue_dequeue(&queue); /* If the left child is free, link the group there. */ if (node->left == NULL) { ck_barrier_combining_insert(node, tnode, &node->left); goto leave; } /* If the right child is free, link the group there. */ if (node->right == NULL) { ck_barrier_combining_insert(node, tnode, &node->right); goto leave; } /* * If unsuccessful, try inserting as a child of the children of the * current node. */ ck_barrier_combining_queue_enqueue(&queue, node->left); ck_barrier_combining_queue_enqueue(&queue, node->right); } leave: ck_spinlock_fas_unlock(&root->mutex); return; } void ck_barrier_combining_init(struct ck_barrier_combining *root, struct ck_barrier_combining_group *init_root) { init_root->k = 0; init_root->count = 0; init_root->sense = 0; init_root->parent = init_root->left = init_root->right = NULL; ck_spinlock_fas_init(&root->mutex); root->root = init_root; return; } static void ck_barrier_combining_aux(struct ck_barrier_combining *barrier, struct ck_barrier_combining_group *tnode, unsigned int sense) { /* * If this is the last thread in the group, it moves on to the parent group. * Otherwise, it spins on this group's sense. */ if (ck_pr_faa_uint(&tnode->count, 1) == tnode->k - 1) { /* * If we are and will be the last thread entering the barrier for the * current group then signal the parent group if one exists. */ if (tnode->parent != NULL) ck_barrier_combining_aux(barrier, tnode->parent, sense); /* * Once the thread returns from its parent(s), it reinitializes the group's * arrival count and signals other threads to continue by flipping the group * sense. Order of these operations is not important since we assume a static * number of threads are members of a barrier for the lifetime of the barrier. * Since count is explicitly reinitialized, it is guaranteed that at any point * tnode->count is equivalent to tnode->k if and only if that many threads * are at the barrier. */ ck_pr_store_uint(&tnode->count, 0); ck_pr_fence_store(); ck_pr_store_uint(&tnode->sense, ~tnode->sense); } else { - ck_pr_fence_memory(); while (sense != ck_pr_load_uint(&tnode->sense)) ck_pr_stall(); } + ck_pr_fence_memory(); return; } void ck_barrier_combining(struct ck_barrier_combining *barrier, struct ck_barrier_combining_group *tnode, struct ck_barrier_combining_state *state) { ck_barrier_combining_aux(barrier, tnode, state->sense); /* Reverse the execution context's sense for the next barrier. */ state->sense = ~state->sense; return; }