From f6a53fd9de3d01826f232278857f625a0231e299 Mon Sep 17 00:00:00 2001 From: Samy Al Bahra Date: Sun, 24 Feb 2013 17:43:00 -0500 Subject: [PATCH 1/3] ck_cohort: First-cut implementation of generalized cohort interface. This work was contributed by Brendon Scheinman. This work is based off "Lock Cohorting: A General Technique for Designing NUMA Locks". --- .gitignore | 5 + include/ck_cohort.h | 126 ++++++++++ regressions/Makefile | 5 + regressions/ck_cohort/benchmark/Makefile | 17 ++ regressions/ck_cohort/benchmark/ck_cohort.c | 8 + regressions/ck_cohort/benchmark/throughput.c | 231 +++++++++++++++++++ regressions/ck_cohort/ck_cohort.h | 25 ++ regressions/ck_cohort/validate/Makefile | 17 ++ regressions/ck_cohort/validate/validate.c | 180 +++++++++++++++ regressions/common.h | 32 +++ 10 files changed, 646 insertions(+) create mode 100644 include/ck_cohort.h create mode 100644 regressions/ck_cohort/benchmark/Makefile create mode 100644 regressions/ck_cohort/benchmark/ck_cohort.c create mode 100644 regressions/ck_cohort/benchmark/throughput.c create mode 100644 regressions/ck_cohort/ck_cohort.h create mode 100644 regressions/ck_cohort/validate/Makefile create mode 100644 regressions/ck_cohort/validate/validate.c diff --git a/.gitignore b/.gitignore index 7c54903..1f8fc99 100644 --- a/.gitignore +++ b/.gitignore @@ -13,6 +13,11 @@ build/Makefile *.a *.so *.dSYM +regressions/ck_cohort/benchmark/ck_cohort.LATENCY +regressions/ck_cohort/benchmark/ck_cohort.THROUGHPUT +regressions/ck_pflock/benchmark/latency +regressions/ck_pflock/benchmark/throughput +regressions/ck_pflock/validate/validate regressions/ck_barrier/benchmark/throughput regressions/ck_barrier/validate/barrier_centralized regressions/ck_barrier/validate/barrier_combining diff --git a/include/ck_cohort.h b/include/ck_cohort.h new file mode 100644 index 0000000..7ad8aea --- /dev/null +++ b/include/ck_cohort.h @@ -0,0 +1,126 @@ +/* + * Copyright 2013 Samy Al Bahra. + * Copyright 2013 Brendon Scheinman. + * 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_COHORT_H +#define _CK_COHORT_H + +/* + * This is an implementation of lock cohorts as described in: + * Dice, D.; Marathe, V.; and Shavit, N. 2012. + * Lock Cohorting: A General Technique for Designing NUMA Locks + */ + +#include +#include +#include + +enum ck_cohort_state { + CK_COHORT_STATE_GLOBAL = 0, + CK_COHORT_STATE_LOCAL = 1 +}; + +#define CK_COHORT_DEFAULT_LOCAL_PASS_LIMIT 10 + +#define CK_COHORT_NAME(N) ck_cohort_##N +#define CK_COHORT_INSTANCE(N) struct CK_COHORT_NAME(N) +#define CK_COHORT_INIT(N, C, GL, LL, P) ck_cohort_##N##_init(C, GL, LL, P) +#define CK_COHORT_LOCK(N, C, GC, LC) ck_cohort_##N##_lock(C, GC, LC) +#define CK_COHORT_UNLOCK(N, C, GC, LC) ck_cohort_##N##_unlock(C, GC, LC) + +#define CK_COHORT_PROTOTYPE(N, GT, GL, GU, LT, LL, LU) \ + CK_COHORT_INSTANCE(N) { \ + GT *global_lock; \ + LT *local_lock; \ + enum ck_cohort_state release_state; \ + unsigned int waiting_threads; \ + unsigned int acquire_count; \ + unsigned int local_pass_limit; \ + }; \ + \ + CK_CC_INLINE static void \ + ck_cohort_##N##_init(struct ck_cohort_##N *cohort, \ + GT *global_lock, LT *local_lock, unsigned int pass_limit) \ + { \ + cohort->global_lock = global_lock; \ + cohort->local_lock = local_lock; \ + cohort->release_state = CK_COHORT_STATE_GLOBAL; \ + cohort->waiting_threads = 0; \ + cohort->acquire_count = 0; \ + cohort->local_pass_limit = pass_limit; \ + ck_pr_barrier(); \ + return; \ + } \ + \ + CK_CC_INLINE static void \ + ck_cohort_##N##_lock(CK_COHORT_INSTANCE(N) *cohort, \ + void *global_context, void *local_context) \ + { \ + \ + ck_pr_inc_uint(&cohort->waiting_threads); \ + LL(cohort->local_lock, local_context); \ + ck_pr_dec_uint(&cohort->waiting_threads); \ + \ + if (cohort->release_state == CK_COHORT_STATE_GLOBAL) { \ + GL(cohort->global_lock, global_context); \ + } \ + \ + ++cohort->acquire_count; \ + return; \ + } \ + \ + CK_CC_INLINE static void \ + ck_cohort_##N##_unlock(CK_COHORT_INSTANCE(N) *cohort, \ + void *global_context, void *local_context) \ + { \ + \ + if (ck_pr_load_uint(&cohort->waiting_threads) > 0 \ + && cohort->acquire_count < cohort->local_pass_limit) { \ + cohort->release_state = CK_COHORT_STATE_LOCAL; \ + } else { \ + GU(cohort->global_lock, global_context); \ + cohort->release_state = CK_COHORT_STATE_GLOBAL; \ + cohort->acquire_count = 0; \ + } \ + \ + ck_pr_fence_memory(); \ + LU(cohort->local_lock, local_context); \ + \ + return; \ + } + +#define CK_COHORT_INITIALIZER { \ + .global_lock = NULL, \ + .local_lock = NULL, \ + .release_state = CK_COHORT_STATE_GLOBAL, \ + .waiting_threads = 0, \ + .acquire_count = 0, \ + .local_pass_limit = CK_COHORT_DEFAULT_LOCAL_PASS_LIMIT \ +} + + +#endif /* _CK_COHORT_H */ + diff --git a/regressions/Makefile b/regressions/Makefile index 005314c..e1c5574 100644 --- a/regressions/Makefile +++ b/regressions/Makefile @@ -4,6 +4,7 @@ DIR=backoff \ bitmap \ brlock \ bytelock \ + cohort \ epoch \ fifo \ hp \ @@ -21,6 +22,8 @@ DIR=backoff \ .PHONY: all clean check all: + $(MAKE) -C ./ck_cohort/validate all + $(MAKE) -C ./ck_cohort/benchmark all $(MAKE) -C ./ck_bitmap/validate all $(MAKE) -C ./ck_backoff/validate all $(MAKE) -C ./ck_queue/validate all @@ -58,6 +61,8 @@ clean: $(MAKE) -C ./ck_backoff/validate clean $(MAKE) -C ./ck_bitmap/validate clean $(MAKE) -C ./ck_queue/validate clean + $(MAKE) -C ./ck_cohort/validate clean + $(MAKE) -C ./ck_cohort/benchmark clean $(MAKE) -C ./ck_brlock/validate clean $(MAKE) -C ./ck_ht/validate clean $(MAKE) -C ./ck_ht/benchmark clean diff --git a/regressions/ck_cohort/benchmark/Makefile b/regressions/ck_cohort/benchmark/Makefile new file mode 100644 index 0000000..7cf62da --- /dev/null +++ b/regressions/ck_cohort/benchmark/Makefile @@ -0,0 +1,17 @@ +.PHONY: all clean + +OBJECTS=ck_cohort.THROUGHPUT ck_cohort.LATENCY + +all: $(OBJECTS) + +ck_cohort.THROUGHPUT: ck_cohort.c + $(CC) $(CFLAGS) -o ck_cohort.THROUGHPUT throughput.c -lm + +ck_cohort.LATENCY: ck_cohort.c + $(CC) -DLATENCY $(CFLAGS) -o ck_cohort.LATENCY ck_cohort.c + +clean: + rm -rf *.dSYM $(OBJECTS) + +include ../../../build/regressions.build +CFLAGS+=$(PTHREAD_CFLAGS) -D_GNU_SOURCE -lm diff --git a/regressions/ck_cohort/benchmark/ck_cohort.c b/regressions/ck_cohort/benchmark/ck_cohort.c new file mode 100644 index 0000000..954b616 --- /dev/null +++ b/regressions/ck_cohort/benchmark/ck_cohort.c @@ -0,0 +1,8 @@ +#include "../ck_cohort.h" + +#include +#ifdef THROUGHPUT +#include "../../ck_spinlock/benchmark/throughput.h" +#elif defined(LATENCY) +#include "../../ck_spinlock/benchmark/latency.h" +#endif diff --git a/regressions/ck_cohort/benchmark/throughput.c b/regressions/ck_cohort/benchmark/throughput.c new file mode 100644 index 0000000..e27fbb2 --- /dev/null +++ b/regressions/ck_cohort/benchmark/throughput.c @@ -0,0 +1,231 @@ +/* + * Copyright 2013 Samy Al Bahra. + * Copyright 2013 Brendon Scheinman. + * 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 +#include +#include +#include +#include +#include +#include +#include + +#include +#include +#include +#include + +#include "../../common.h" + +#define max(x, y) (((x) > (y)) ? (x) : (y)) + +static struct affinity a; +static unsigned int ready; + +struct counters { + uint64_t value; +} CK_CC_CACHELINE; + +static struct counters *count; +static uint64_t nthr; +static unsigned int n_cohorts; +static unsigned int barrier; +static int critical CK_CC_CACHELINE; + +static void +ck_spinlock_lock_with_context(ck_spinlock_t *lock, void *context) +{ + + (void)context; + ck_spinlock_lock(lock); + return; +} + +static void +ck_spinlock_unlock_with_context(ck_spinlock_t *lock, void *context) +{ + + (void)context; + ck_spinlock_unlock(lock); + return; +} + +CK_COHORT_PROTOTYPE(basic, + ck_spinlock_t, ck_spinlock_lock_with_context, ck_spinlock_unlock_with_context, + ck_spinlock_t, ck_spinlock_lock_with_context, ck_spinlock_unlock_with_context) + +struct cohort_record { + CK_COHORT_INSTANCE(basic) cohort; +} CK_CC_CACHELINE; +static struct cohort_record *cohorts; + +static ck_spinlock_t global_lock = CK_SPINLOCK_INITIALIZER; + +struct block { + unsigned int tid; +}; + +static void * +fairness(void *null) +{ + struct block *context = null; + unsigned int i = context->tid; + volatile int j; + long int base; + unsigned int core; + CK_COHORT_INSTANCE(basic) *cohort; + + if (aff_iterate_core(&a, &core)) { + perror("ERROR: Could not affine thread"); + exit(EXIT_FAILURE); + } + + cohort = &((cohorts + (core / (int)(a.delta)) % n_cohorts)->cohort); + + while (ck_pr_load_uint(&ready) == 0); + + ck_pr_inc_uint(&barrier); + while (ck_pr_load_uint(&barrier) != nthr); + + while (ready) { + CK_COHORT_LOCK(basic, cohort, NULL, NULL); + + count[i].value++; + if (critical) { + base = common_lrand48() % critical; + for (j = 0; j < base; j++); + } + + CK_COHORT_UNLOCK(basic, cohort, NULL, NULL); + } + + return NULL; +} + +int +main(int argc, char *argv[]) +{ + uint64_t v, d; + unsigned int i; + pthread_t *threads; + struct block *context; + ck_spinlock_t *local_lock; + + if (argc != 5) { + ck_error("Usage: ck_cohort " + " \n"); + } + + n_cohorts = atoi(argv[1]); + if (n_cohorts <= 0) { + ck_error("ERROR: Number of cohorts must be greater than 0\n"); + } + + nthr = n_cohorts * atoi(argv[2]); + if (nthr <= 0) { + ck_error("ERROR: Number of threads must be greater than 0\n"); + } + + critical = atoi(argv[4]); + if (critical < 0) { + ck_error("ERROR: critical section cannot be negative\n"); + } + + threads = malloc(sizeof(pthread_t) * nthr); + if (threads == NULL) { + ck_error("ERROR: Could not allocate thread structures\n"); + } + + cohorts = malloc(sizeof(struct cohort_record) * n_cohorts); + if (cohorts == NULL) { + ck_error("ERROR: Could not allocate cohort structures\n"); + } + + context = malloc(sizeof(struct block) * nthr); + if (context == NULL) { + ck_error("ERROR: Could not allocate thread contexts\n"); + } + + a.delta = atoi(argv[2]); + a.request = 0; + + count = malloc(sizeof(*count) * nthr); + if (count == NULL) { + ck_error("ERROR: Could not create acquisition buffer\n"); + } + memset(count, 0, sizeof(*count) * nthr); + + fprintf(stderr, "Creating cohorts..."); + for (i = 0 ; i < n_cohorts ; i++) { + local_lock = malloc(max(CK_MD_CACHELINE, sizeof(ck_spinlock_t))); + if (local_lock == NULL) { + ck_error("ERROR: Could not allocate local lock\n"); + } + CK_COHORT_INIT(basic, &((cohorts + i)->cohort), &global_lock, local_lock, + CK_COHORT_DEFAULT_LOCAL_PASS_LIMIT); + local_lock = NULL; + } + fprintf(stderr, "done\n"); + + fprintf(stderr, "Creating threads (fairness)..."); + for (i = 0; i < nthr; i++) { + context[i].tid = i; + if (pthread_create(&threads[i], NULL, fairness, context + i)) { + ck_error("ERROR: Could not create thread %d\n", i); + } + } + fprintf(stderr, "done\n"); + + ck_pr_store_uint(&ready, 1); + common_sleep(10); + ck_pr_store_uint(&ready, 0); + + fprintf(stderr, "Waiting for threads to finish acquisition regression..."); + for (i = 0; i < nthr; i++) + pthread_join(threads[i], NULL); + fprintf(stderr, "done\n\n"); + + for (i = 0, v = 0; i < nthr; i++) { + printf("%d %15" PRIu64 "\n", i, count[i].value); + v += count[i].value; + } + + printf("\n# total : %15" PRIu64 "\n", v); + printf("# throughput : %15" PRIu64 " a/s\n", (v /= nthr) / 10); + + for (i = 0, d = 0; i < nthr; i++) + d += (count[i].value - v) * (count[i].value - v); + + printf("# average : %15" PRIu64 "\n", v); + printf("# deviation : %.2f (%.2f%%)\n\n", sqrt(d / nthr), (sqrt(d / nthr) / v) * 100.00); + + return 0; +} + diff --git a/regressions/ck_cohort/ck_cohort.h b/regressions/ck_cohort/ck_cohort.h new file mode 100644 index 0000000..3194311 --- /dev/null +++ b/regressions/ck_cohort/ck_cohort.h @@ -0,0 +1,25 @@ +#define LOCK_NAME "ck_cohort" +#define LOCK_DEFINE\ + static ck_spinlock_fas_t global_fas_lock = CK_SPINLOCK_FAS_INITIALIZER;\ + static ck_spinlock_fas_t local_fas_lock = CK_SPINLOCK_FAS_INITIALIZER;\ + static void\ + ck_spinlock_fas_lock_with_context(ck_spinlock_fas_t *lock, void *context)\ + {\ + (void)context;\ + ck_spinlock_fas_lock(lock);\ + }\ +\ + static void\ + ck_spinlock_fas_unlock_with_context(ck_spinlock_fas_t *lock, void *context)\ + {\ + (void)context;\ + ck_spinlock_fas_unlock(lock);\ + }\ + CK_COHORT_PROTOTYPE(fas_fas,\ + ck_spinlock_fas_t, ck_spinlock_fas_lock_with_context, ck_spinlock_fas_unlock_with_context,\ + ck_spinlock_fas_t, ck_spinlock_fas_lock_with_context, ck_spinlock_fas_unlock_with_context)\ + static CK_COHORT_INSTANCE(fas_fas) CK_CC_CACHELINE cohort = CK_COHORT_INITIALIZER +#define LOCK_INIT CK_COHORT_INIT(fas_fas, &cohort, &global_fas_lock, &local_fas_lock,\ + CK_COHORT_DEFAULT_LOCAL_PASS_LIMIT) +#define LOCK CK_COHORT_LOCK(fas_fas, &cohort, NULL, NULL) +#define UNLOCK CK_COHORT_UNLOCK(fas_fas, &cohort, NULL, NULL) diff --git a/regressions/ck_cohort/validate/Makefile b/regressions/ck_cohort/validate/Makefile new file mode 100644 index 0000000..794e3e6 --- /dev/null +++ b/regressions/ck_cohort/validate/Makefile @@ -0,0 +1,17 @@ +.PHONY: check clean distribution + +OBJECTS=validate + +all: $(OBJECTS) + +validate: validate.c ../../../include/ck_cohort.h + $(CC) $(CFLAGS) -o validate validate.c + +check: all + ./validate `expr $(CORES) / 2` 2 1 + +clean: + rm -rf *.dSYM *~ *.o $(OBJECTS) + +include ../../../build/regressions.build +CFLAGS+=$(PTHREAD_CFLAGS) -D_GNU_SOURCE diff --git a/regressions/ck_cohort/validate/validate.c b/regressions/ck_cohort/validate/validate.c new file mode 100644 index 0000000..8b31e90 --- /dev/null +++ b/regressions/ck_cohort/validate/validate.c @@ -0,0 +1,180 @@ +/* + * Copyright 2013 Samy Al Bahra. + * Copyright 2013 Brendon Scheinman. + * 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 + +#include +#include +#include + +#include "../../common.h" + +#ifndef ITERATE +#define ITERATE 1000000 +#endif + +static struct affinity a; +static unsigned int locked; +static int nthr; +static ck_spinlock_fas_t global_fas_lock = CK_SPINLOCK_FAS_INITIALIZER; + +static void +ck_spinlock_fas_lock_with_context(ck_spinlock_fas_t *lock, void *context) +{ + (void)context; + ck_spinlock_fas_lock(lock); +} + +static void +ck_spinlock_fas_unlock_with_context(ck_spinlock_fas_t *lock, void *context) +{ + (void)context; + ck_spinlock_fas_unlock(lock); +} + +CK_COHORT_PROTOTYPE(fas_fas, + ck_spinlock_fas_t, ck_spinlock_fas_lock_with_context, ck_spinlock_fas_unlock_with_context, + ck_spinlock_fas_t, ck_spinlock_fas_lock_with_context, ck_spinlock_fas_unlock_with_context) +static CK_COHORT_INSTANCE(fas_fas) *cohorts; +static int n_cohorts; + +static void * +thread(void *null CK_CC_UNUSED) +{ + int i = ITERATE; + unsigned int l; + unsigned int core; + CK_COHORT_INSTANCE(fas_fas) *cohort; + + if (aff_iterate_core(&a, &core)) { + perror("ERROR: Could not affine thread"); + exit(EXIT_FAILURE); + } + + cohort = cohorts + (core / (int)(a.delta)) % n_cohorts; + + while (i--) { + CK_COHORT_LOCK(fas_fas, cohort, NULL, NULL); + { + l = ck_pr_load_uint(&locked); + if (l != 0) { + ck_error("ERROR [WR:%d]: %u != 0\n", __LINE__, l); + } + + ck_pr_inc_uint(&locked); + ck_pr_inc_uint(&locked); + ck_pr_inc_uint(&locked); + ck_pr_inc_uint(&locked); + ck_pr_inc_uint(&locked); + ck_pr_inc_uint(&locked); + ck_pr_inc_uint(&locked); + ck_pr_inc_uint(&locked); + + l = ck_pr_load_uint(&locked); + if (l != 8) { + ck_error("ERROR [WR:%d]: %u != 2\n", __LINE__, l); + } + + ck_pr_dec_uint(&locked); + ck_pr_dec_uint(&locked); + ck_pr_dec_uint(&locked); + ck_pr_dec_uint(&locked); + ck_pr_dec_uint(&locked); + ck_pr_dec_uint(&locked); + ck_pr_dec_uint(&locked); + ck_pr_dec_uint(&locked); + + l = ck_pr_load_uint(&locked); + if (l != 0) { + ck_error("ERROR [WR:%d]: %u != 0\n", __LINE__, l); + } + } + CK_COHORT_UNLOCK(fas_fas, cohort, NULL, NULL); + } + + return (NULL); +} + +int +main(int argc, char *argv[]) +{ + pthread_t *threads; + int threads_per_cohort; + ck_spinlock_fas_t *local_lock; + int i; + + if (argc != 4) { + ck_error("Usage: validate \n"); + } + + n_cohorts = atoi(argv[1]); + if (n_cohorts <= 0) { + ck_error("ERROR: Number of cohorts must be greater than 0\n"); + } + + threads_per_cohort = atoi(argv[2]); + if (threads_per_cohort <= 0) { + ck_error("ERROR: Threads per cohort must be greater than 0\n"); + } + + nthr = n_cohorts * threads_per_cohort; + + threads = malloc(sizeof(pthread_t) * nthr); + if (threads == NULL) { + ck_error("ERROR: Could not allocate thread structures\n"); + } + + a.delta = atoi(argv[3]); + + fprintf(stderr, "Creating cohorts..."); + cohorts = malloc(sizeof(CK_COHORT_INSTANCE(fas_fas)) * n_cohorts); + for (i = 0 ; i < n_cohorts ; i++) { + local_lock = malloc(sizeof(ck_spinlock_fas_t)); + CK_COHORT_INIT(fas_fas, cohorts + i, &global_fas_lock, local_lock, + CK_COHORT_DEFAULT_LOCAL_PASS_LIMIT); + } + fprintf(stderr, "done\n"); + + fprintf(stderr, "Creating threads..."); + for (i = 0; i < nthr; i++) { + if (pthread_create(&threads[i], NULL, thread, NULL)) { + ck_error("ERROR: Could not create thread %d\n", i); + } + } + fprintf(stderr, "done\n"); + + fprintf(stderr, "Waiting for threads to finish correctness regression..."); + for (i = 0; i < nthr; i++) + pthread_join(threads[i], NULL); + fprintf(stderr, "done (passed)\n"); + + return (0); +} + diff --git a/regressions/common.h b/regressions/common.h index 9994563..c06a145 100644 --- a/regressions/common.h +++ b/regressions/common.h @@ -229,6 +229,18 @@ aff_iterate(struct affinity *acb) return sched_setaffinity(gettid(), sizeof(s), &s); } + +CK_CC_UNUSED static int +aff_iterate_core(struct affinity *acb, unsigned int *core) +{ + cpu_set_t s; + + *core = ck_pr_faa_uint(&acb->request, acb->delta); + CPU_ZERO(&s); + CPU_SET((*core) % CORES, &s); + + return sched_setaffinity(gettid(), sizeof(s), &s); +} #elif defined(__MACH__) CK_CC_UNUSED static int aff_iterate(struct affinity *acb) @@ -243,6 +255,19 @@ aff_iterate(struct affinity *acb) (thread_policy_t)&policy, THREAD_AFFINITY_POLICY_COUNT); } + +CK_CC_UNUSED static int +aff_iterate_core(struct affinity *acb, unsigned int *core) +{ + thread_affinity_policy_data_t policy; + + *core = ck_pr_faa_uint(&acb->request, acb->delta) % CORES; + policy.affinity_tag = *core; + return thread_policy_set(mach_thread_self(), + THREAD_AFFINITY_POLICY, + (thread_policy_t)&policy, + THREAD_AFFINITY_POLICY_COUNT); +} #else CK_CC_UNUSED static int aff_iterate(struct affinity *acb CK_CC_UNUSED) @@ -250,6 +275,13 @@ aff_iterate(struct affinity *acb CK_CC_UNUSED) return (0); } + +CK_CC_UNUSED static int +aff_iterate_core(struct affinity *acb CK_CC_UNUSED, unsigned int *core) +{ + *core = 0; + return (0); +} #endif CK_CC_INLINE static uint64_t From 7e9a020abde2f7b5adc2f0ea75e352fa1c794cb8 Mon Sep 17 00:00:00 2001 From: Samy Al Bahra Date: Sun, 24 Feb 2013 17:47:33 -0500 Subject: [PATCH 2/3] regressions: Add ck_pflock clean target. --- regressions/Makefile | 2 ++ 1 file changed, 2 insertions(+) diff --git a/regressions/Makefile b/regressions/Makefile index e1c5574..30639a9 100644 --- a/regressions/Makefile +++ b/regressions/Makefile @@ -58,6 +58,8 @@ all: $(MAKE) -C ./ck_bag/validate all clean: + $(MAKE) -C ./ck_pflock/validate clean + $(MAKE) -C ./ck_pflock/benchmark clean $(MAKE) -C ./ck_backoff/validate clean $(MAKE) -C ./ck_bitmap/validate clean $(MAKE) -C ./ck_queue/validate clean From c393a97cb751192d0d292afd3dbc3083869c5ae7 Mon Sep 17 00:00:00 2001 From: Samy Al Bahra Date: Sun, 24 Feb 2013 18:09:36 -0500 Subject: [PATCH 3/3] ck_pflock: Fix Copyright typo from original patch. --- include/ck_pflock.h | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/include/ck_pflock.h b/include/ck_pflock.h index 91983f0..b352339 100644 --- a/include/ck_pflock.h +++ b/include/ck_pflock.h @@ -13,7 +13,7 @@ * documentation and/or other materials provided with the distribution. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND - * ANY EXCK_PFLOCK_PRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE + * 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