diff --git a/.gitignore b/.gitignore index 59b0003..a37e698 100644 --- a/.gitignore +++ b/.gitignore @@ -15,6 +15,10 @@ build/Makefile *.so *.dSYM .*.sw[op] +GPATH +GRTAGS +GTAGS +ID regressions/ck_array/validate/serial regressions/ck_backoff/validate/validate regressions/ck_bag/validate/order @@ -52,12 +56,16 @@ regressions/ck_hp/validate/ck_hp_stack regressions/ck_hp/validate/nbds_haz_test regressions/ck_hp/validate/serial regressions/ck_hs/benchmark/parallel_bytestring +regressions/ck_hs/benchmark/parallel_bytestring.delete regressions/ck_hs/benchmark/serial regressions/ck_hs/validate/serial regressions/ck_ht/benchmark/parallel_bytestring +regressions/ck_ht/benchmark/parallel_bytestring.delete regressions/ck_ht/benchmark/parallel_direct regressions/ck_ht/benchmark/serial +regressions/ck_ht/benchmark/serial.delete regressions/ck_ht/validate/serial +regressions/ck_ht/validate/serial.delete regressions/ck_pflock/benchmark/latency regressions/ck_pflock/benchmark/throughput regressions/ck_pflock/validate/validate @@ -87,6 +95,9 @@ regressions/ck_pr/validate/ck_pr_xor regressions/ck_queue/validate/ck_list regressions/ck_queue/validate/ck_slist regressions/ck_queue/validate/ck_stailq +regressions/ck_rhs/benchmark/parallel_bytestring +regressions/ck_rhs/benchmark/serial +regressions/ck_rhs/validate/serial regressions/ck_ring/benchmark/latency regressions/ck_ring/validate/ck_ring_spmc regressions/ck_ring/validate/ck_ring_spmc_template @@ -116,6 +127,8 @@ regressions/ck_spinlock/benchmark/ck_dec.LATENCY regressions/ck_spinlock/benchmark/ck_dec.THROUGHPUT regressions/ck_spinlock/benchmark/ck_fas.LATENCY regressions/ck_spinlock/benchmark/ck_fas.THROUGHPUT +regressions/ck_spinlock/benchmark/ck_hclh.LATENCY +regressions/ck_spinlock/benchmark/ck_hclh.THROUGHPUT regressions/ck_spinlock/benchmark/ck_mcs.LATENCY regressions/ck_spinlock/benchmark/ck_mcs.THROUGHPUT regressions/ck_spinlock/benchmark/ck_spinlock.LATENCY @@ -157,3 +170,6 @@ regressions/ck_stack/validate/upmc_pop regressions/ck_stack/validate/upmc_push regressions/ck_stack/validate/upmc_trypop regressions/ck_stack/validate/upmc_trypush +regressions/ck_swlock/benchmark/latency +regressions/ck_swlock/benchmark/throughput +regressions/ck_swlock/validate/validate diff --git a/include/ck_swlock.h b/include/ck_swlock.h new file mode 100644 index 0000000..134df6c --- /dev/null +++ b/include/ck_swlock.h @@ -0,0 +1,218 @@ +/* + * Copyright 2014 Jaidev Sridhar. + * Copyright 2014 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_SWLOCK_H +#define _CK_SWLOCK_H + +#include +#include +#include +#include +#include + +struct ck_swlock { + uint32_t value; +}; +typedef struct ck_swlock ck_swlock_t; + +#define CK_SWLOCK_INITIALIZER {0} +#define CK_SWLOCK_WRITER_BIT (1UL << 31) +#define CK_SWLOCK_LATCH_BIT (1UL << 30) +#define CK_SWLOCK_WRITER_MASK (CK_SWLOCK_LATCH_BIT | CK_SWLOCK_WRITER_BIT) +#define CK_SWLOCK_READER_MASK (UINT32_MAX ^ CK_SWLOCK_WRITER_MASK) + +CK_CC_INLINE static void +ck_swlock_init(struct ck_swlock *rw) +{ + + rw->value = 0; + ck_pr_barrier(); + return; +} + +CK_CC_INLINE static void +ck_swlock_write_unlock(ck_swlock_t *rw) +{ + + ck_pr_fence_release(); + ck_pr_and_32(&rw->value, CK_SWLOCK_READER_MASK); + return; +} + +CK_CC_INLINE static bool +ck_swlock_locked_writer(ck_swlock_t *rw) +{ + + ck_pr_fence_load(); + return ck_pr_load_32(&rw->value) & CK_SWLOCK_WRITER_BIT; +} + +CK_CC_INLINE static void +ck_swlock_write_downgrade(ck_swlock_t *rw) +{ + + ck_pr_inc_32(&rw->value); + ck_swlock_write_unlock(rw); + return; +} + +CK_CC_INLINE static bool +ck_swlock_locked(ck_swlock_t *rw) +{ + + ck_pr_fence_load(); + return ck_pr_load_32(&rw->value); +} + +CK_CC_INLINE static bool +ck_swlock_write_trylock(ck_swlock_t *rw) +{ + + ck_pr_fence_acquire(); + return ck_pr_cas_32(&rw->value, 0, CK_SWLOCK_WRITER_BIT); +} + +CK_ELIDE_TRYLOCK_PROTOTYPE(ck_swlock_write, ck_swlock_t, + ck_swlock_locked, ck_swlock_write_trylock) + +CK_CC_INLINE static void +ck_swlock_write_lock(ck_swlock_t *rw) +{ + + ck_pr_or_32(&rw->value, CK_SWLOCK_WRITER_BIT); + while (ck_pr_load_32(&rw->value) & CK_SWLOCK_READER_MASK) + ck_pr_stall(); + + ck_pr_fence_acquire(); + return; +} + +CK_CC_INLINE static void +ck_swlock_write_latch(ck_swlock_t *rw) +{ + + /* Publish intent to acquire lock. */ + ck_pr_or_32(&rw->value, CK_SWLOCK_WRITER_BIT); + + /* Stall until readers have seen the seen writer and cleared. */ + while (ck_pr_cas_32(&rw->value, CK_SWLOCK_WRITER_BIT, + CK_SWLOCK_WRITER_MASK) == false) { + do { + ck_pr_stall(); + } while (ck_pr_load_32(&rw->value) != CK_SWLOCK_WRITER_BIT); + } + + ck_pr_fence_acquire(); + return; +} + +CK_CC_INLINE static void +ck_swlock_write_unlatch(ck_swlock_t *rw) +{ + + ck_pr_fence_release(); + ck_pr_store_32(&rw->value, 0); + return; +} + +CK_ELIDE_PROTOTYPE(ck_swlock_write, ck_swlock_t, + ck_swlock_locked, ck_swlock_write_lock, + ck_swlock_locked_writer, ck_swlock_write_unlock) + +CK_ELIDE_TRYLOCK_PROTOTYPE(ck_swlock_read, ck_swlock_t, + ck_swlock_locked_writer, ck_swlock_read_trylock) + +CK_CC_INLINE static bool +ck_swlock_read_trylock(ck_swlock_t *rw) +{ + uint32_t l = ck_pr_load_32(&rw->value); + + if (l & CK_SWLOCK_WRITER_BIT) + return false; + + l = ck_pr_faa_32(&rw->value, 1) & CK_SWLOCK_WRITER_MASK; + if (l == 0) { + ck_pr_fence_acquire(); + return true; + } + + if (l == CK_SWLOCK_WRITER_BIT) + ck_pr_dec_32(&rw->value); + + return false; +} + +CK_CC_INLINE static void +ck_swlock_read_lock(ck_swlock_t *rw) +{ + uint32_t l; + + for (;;) { + while (ck_pr_load_32(&rw->value) & CK_SWLOCK_WRITER_BIT) + ck_pr_stall(); + + l = ck_pr_faa_32(&rw->value, 1) & CK_SWLOCK_WRITER_MASK; + if (l == 0) + break; + + /* + * If the latch bit has not been set, then the writer would + * have observed the reader and will wait to completion of + * read-side critical section. + */ + if (l == CK_SWLOCK_WRITER_BIT) + ck_pr_dec_32(&rw->value); + } + + ck_pr_fence_acquire(); + return; +} + + +CK_CC_INLINE static bool +ck_swlock_locked_reader(ck_swlock_t *rw) +{ + + ck_pr_fence_load(); + return ck_pr_load_32(&rw->value) & CK_SWLOCK_READER_MASK; +} + +CK_CC_INLINE static void +ck_swlock_read_unlock(ck_swlock_t *rw) +{ + + ck_pr_fence_release(); + ck_pr_dec_32(&rw->value); + return; +} + +CK_ELIDE_PROTOTYPE(ck_swlock_read, ck_swlock_t, + ck_swlock_locked_writer, ck_swlock_read_lock, + ck_swlock_locked_reader, ck_swlock_read_unlock) + +#endif /* _CK_SWLOCK_H */ + diff --git a/regressions/Makefile b/regressions/Makefile index 44078b7..3bc9298 100644 --- a/regressions/Makefile +++ b/regressions/Makefile @@ -16,6 +16,7 @@ DIR=array \ queue \ ring \ rwlock \ + swlock \ sequence \ spinlock \ stack \ @@ -61,6 +62,8 @@ all: $(MAKE) -C ./ck_rwlock/benchmark all $(MAKE) -C ./ck_tflock/validate all $(MAKE) -C ./ck_tflock/benchmark all + $(MAKE) -C ./ck_swlock/validate all + $(MAKE) -C ./ck_swlock/benchmark all $(MAKE) -C ./ck_pflock/validate all $(MAKE) -C ./ck_pflock/benchmark all $(MAKE) -C ./ck_hp/validate all @@ -106,6 +109,8 @@ clean: $(MAKE) -C ./ck_ring/benchmark clean $(MAKE) -C ./ck_rwlock/validate clean $(MAKE) -C ./ck_rwlock/benchmark clean + $(MAKE) -C ./ck_swlock/validate clean + $(MAKE) -C ./ck_swlock/benchmark clean $(MAKE) -C ./ck_pflock/validate clean $(MAKE) -C ./ck_pflock/benchmark clean $(MAKE) -C ./ck_hp/validate clean diff --git a/regressions/ck_swlock/benchmark/Makefile b/regressions/ck_swlock/benchmark/Makefile new file mode 100644 index 0000000..4ec728c --- /dev/null +++ b/regressions/ck_swlock/benchmark/Makefile @@ -0,0 +1,17 @@ +.PHONY: clean distribution + +OBJECTS=latency throughput + +all: $(OBJECTS) + +latency: latency.c ../../../include/ck_swlock.h + $(CC) $(CFLAGS) -o latency latency.c + +throughput: throughput.c ../../../include/ck_swlock.h + $(CC) $(CFLAGS) -o throughput throughput.c + +clean: + rm -rf *.dSYM *.exe *~ *.o $(OBJECTS) + +include ../../../build/regressions.build +CFLAGS+=$(PTHREAD_CFLAGS) -D_GNU_SOURCE diff --git a/regressions/ck_swlock/benchmark/latency.c b/regressions/ck_swlock/benchmark/latency.c new file mode 100644 index 0000000..7261546 --- /dev/null +++ b/regressions/ck_swlock/benchmark/latency.c @@ -0,0 +1,86 @@ +/* + * Copyright 2011-2014 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. + */ + +#include +#include +#include + +#include "../../common.h" + +#define CK_F_PR_RTM + +#ifndef STEPS +#define STEPS 2000000 +#endif + +int +main(void) +{ + uint64_t s_b, e_b, i; + ck_swlock_t swlock = CK_SWLOCK_INITIALIZER; + + for (i = 0; i < STEPS; i++) { + ck_swlock_write_lock(&swlock); + ck_swlock_write_unlock(&swlock); + } + + s_b = rdtsc(); + for (i = 0; i < STEPS; i++) { + ck_swlock_write_lock(&swlock); + ck_swlock_write_unlock(&swlock); + } + e_b = rdtsc(); + printf(" WRITE: swlock %15" PRIu64 "\n", (e_b - s_b) / STEPS); + + for (i = 0; i < STEPS; i++) { + ck_swlock_read_lock(&swlock); + ck_swlock_read_unlock(&swlock); + } + + s_b = rdtsc(); + for (i = 0; i < STEPS; i++) { + ck_swlock_read_lock(&swlock); + ck_swlock_read_unlock(&swlock); + } + e_b = rdtsc(); + printf(" READ: swlock %15" PRIu64 "\n", (e_b - s_b) / STEPS); + + for (i = 0; i < STEPS; i++) { + ck_swlock_write_latch(&swlock); + ck_swlock_write_unlatch(&swlock); + } + + s_b = rdtsc(); + for (i = 0; i < STEPS; i++) { + ck_swlock_write_latch(&swlock); + ck_swlock_write_unlatch(&swlock); + } + e_b = rdtsc(); + printf(" LATCH: swlock %15" PRIu64 "\n", (e_b - s_b) / STEPS); + + return 0; +} + diff --git a/regressions/ck_swlock/benchmark/throughput.c b/regressions/ck_swlock/benchmark/throughput.c new file mode 100644 index 0000000..fa3cf1c --- /dev/null +++ b/regressions/ck_swlock/benchmark/throughput.c @@ -0,0 +1,183 @@ +/* + * Copyright 2011-2014 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. + */ + +#include +#include +#include +#include +#include +#include +#include + +#include "../../common.h" + +#ifndef STEPS +#define STEPS 1000000 +#endif + +static int barrier; +static int threads; +static unsigned int flag CK_CC_CACHELINE; +static struct { + ck_swlock_t lock; +} rw CK_CC_CACHELINE = { + .lock = CK_SWLOCK_INITIALIZER +}; + +static struct affinity affinity; + +static void * +thread_lock(void *pun) +{ + uint64_t s_b, e_b, a, i; + uint64_t *value = pun; + + if (aff_iterate(&affinity) != 0) { + perror("ERROR: Could not affine thread"); + exit(EXIT_FAILURE); + } + + ck_pr_inc_int(&barrier); + while (ck_pr_load_int(&barrier) != threads) + ck_pr_stall(); + + for (i = 1, a = 0;; i++) { + s_b = rdtsc(); + ck_swlock_read_lock(&rw.lock); + ck_swlock_read_unlock(&rw.lock); + ck_swlock_read_lock(&rw.lock); + ck_swlock_read_unlock(&rw.lock); + ck_swlock_read_lock(&rw.lock); + ck_swlock_read_unlock(&rw.lock); + ck_swlock_read_lock(&rw.lock); + ck_swlock_read_unlock(&rw.lock); + ck_swlock_read_lock(&rw.lock); + ck_swlock_read_unlock(&rw.lock); + ck_swlock_read_lock(&rw.lock); + ck_swlock_read_unlock(&rw.lock); + ck_swlock_read_lock(&rw.lock); + ck_swlock_read_unlock(&rw.lock); + ck_swlock_read_lock(&rw.lock); + ck_swlock_read_unlock(&rw.lock); + ck_swlock_read_lock(&rw.lock); + ck_swlock_read_unlock(&rw.lock); + ck_swlock_read_lock(&rw.lock); + ck_swlock_read_unlock(&rw.lock); + ck_swlock_read_lock(&rw.lock); + ck_swlock_read_unlock(&rw.lock); + ck_swlock_read_lock(&rw.lock); + ck_swlock_read_unlock(&rw.lock); + ck_swlock_read_lock(&rw.lock); + ck_swlock_read_unlock(&rw.lock); + ck_swlock_read_lock(&rw.lock); + ck_swlock_read_unlock(&rw.lock); + ck_swlock_read_lock(&rw.lock); + ck_swlock_read_unlock(&rw.lock); + ck_swlock_read_lock(&rw.lock); + ck_swlock_read_unlock(&rw.lock); + e_b = rdtsc(); + + a += (e_b - s_b) >> 4; + + if (ck_pr_load_uint(&flag) == 1) + break; + } + + ck_pr_inc_int(&barrier); + while (ck_pr_load_int(&barrier) != threads * 2) + ck_pr_stall(); + + *value = (a / i); + return NULL; +} + +static void +swlock_test(pthread_t *p, int d, uint64_t *latency, void *(*f)(void *), const char *label) +{ + int t; + + ck_pr_store_int(&barrier, 0); + ck_pr_store_uint(&flag, 0); + + affinity.delta = d; + affinity.request = 0; + + fprintf(stderr, "Creating threads (%s)...", label); + for (t = 0; t < threads; t++) { + if (pthread_create(&p[t], NULL, f, latency + t) != 0) { + ck_error("ERROR: Could not create thread %d\n", t); + } + } + fprintf(stderr, "done\n"); + + common_sleep(10); + ck_pr_store_uint(&flag, 1); + + fprintf(stderr, "Waiting for threads to finish acquisition regression..."); + for (t = 0; t < threads; t++) + pthread_join(p[t], NULL); + fprintf(stderr, "done\n\n"); + + for (t = 1; t <= threads; t++) + printf("%10u %20" PRIu64 "\n", t, latency[t - 1]); + + fprintf(stderr, "\n"); + return; +} + + +int +main(int argc, char *argv[]) +{ + int d; + pthread_t *p; + uint64_t *latency; + + if (argc != 3) { + ck_error("Usage: throughput \n"); + } + + threads = atoi(argv[2]); + if (threads <= 0) { + ck_error("ERROR: Threads must be a value > 0.\n"); + } + + p = malloc(sizeof(pthread_t) * threads); + if (p == NULL) { + ck_error("ERROR: Failed to initialize thread.\n"); + } + + latency = malloc(sizeof(uint64_t) * threads); + if (latency == NULL) { + ck_error("ERROR: Failed to create latency buffer.\n"); + } + + d = atoi(argv[1]); + swlock_test(p, d, latency, thread_lock, "swlock"); + + return 0; +} + diff --git a/regressions/ck_swlock/validate/Makefile b/regressions/ck_swlock/validate/Makefile new file mode 100644 index 0000000..54d62f2 --- /dev/null +++ b/regressions/ck_swlock/validate/Makefile @@ -0,0 +1,17 @@ +.PHONY: check clean distribution + +OBJECTS=validate + +all: $(OBJECTS) + +validate: validate.c ../../../include/ck_swlock.h + $(CC) $(CFLAGS) -o validate validate.c + +check: all + ./validate $(CORES) 1 + +clean: + rm -rf *.dSYM *.exe *~ *.o $(OBJECTS) + +include ../../../build/regressions.build +CFLAGS+=$(PTHREAD_CFLAGS) -D_GNU_SOURCE diff --git a/regressions/ck_swlock/validate/validate.c b/regressions/ck_swlock/validate/validate.c new file mode 100644 index 0000000..bc2f496 --- /dev/null +++ b/regressions/ck_swlock/validate/validate.c @@ -0,0 +1,453 @@ +/* + * Copyright 2014 Jaidev Sridhar. + * 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 "../../common.h" + +#ifndef ITERATE +#define ITERATE 1000000 +#endif + +static struct affinity a; +static unsigned int locked; +static int nthr; +static ck_swlock_t lock = CK_SWLOCK_INITIALIZER; + +#ifdef CK_F_PR_RTM +static void * +thread_rtm_adaptive(void *arg) +{ + unsigned int i = ITERATE; + unsigned int l; + int tid = ck_pr_load_int(arg); + + struct ck_elide_config config = CK_ELIDE_CONFIG_DEFAULT_INITIALIZER; + struct ck_elide_stat st = CK_ELIDE_STAT_INITIALIZER; + + if (aff_iterate(&a)) { + perror("ERROR: Could not affine thread"); + exit(EXIT_FAILURE); + } + + while (i--) { + if (tid == 0) { + CK_ELIDE_LOCK_ADAPTIVE(ck_swlock_write, &st, &config, &lock); + { + 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_ELIDE_UNLOCK_ADAPTIVE(ck_swlock_write, &st, &lock); + } + + CK_ELIDE_LOCK(ck_swlock_read, &lock); + { + l = ck_pr_load_uint(&locked); + if (l != 0) { + ck_error("ERROR [RD:%d]: %u != 0\n", __LINE__, l); + } + } + CK_ELIDE_UNLOCK(ck_swlock_read, &lock); + } + + return NULL; +} + +static void * +thread_rtm_mix(void *arg) +{ + unsigned int i = ITERATE; + unsigned int l; + int tid = ck_pr_load_int(arg); + + if (aff_iterate(&a)) { + perror("ERROR: Could not affine thread"); + exit(EXIT_FAILURE); + } + + while (i--) { + if (tid == 0) { + if (i & 1) { + CK_ELIDE_LOCK(ck_swlock_write, &lock); + } else { + ck_swlock_write_lock(&lock); + } + + { + 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); + } + } + + if (i & 1) { + CK_ELIDE_UNLOCK(ck_swlock_write, &lock); + } else { + ck_swlock_write_unlock(&lock); + } + } + if (i & 1) { + CK_ELIDE_LOCK(ck_swlock_read, &lock); + } else { + ck_swlock_read_lock(&lock); + } + + { + l = ck_pr_load_uint(&locked); + if (l != 0) { + ck_error("ERROR [RD:%d]: %u != 0\n", __LINE__, l); + } + } + + if (i & 1) { + CK_ELIDE_UNLOCK(ck_swlock_read, &lock); + } else { + ck_swlock_read_unlock(&lock); + } + } + + return (NULL); +} + +static void * +thread_rtm(void *arg) +{ + unsigned int i = ITERATE; + unsigned int l; + int tid = ck_pr_load_int(arg); + + if (aff_iterate(&a)) { + perror("ERROR: Could not affine thread"); + exit(EXIT_FAILURE); + } + + while (i--) { + if (tid == 0) { + CK_ELIDE_LOCK(ck_swlock_write, &lock); + { + 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_ELIDE_UNLOCK(ck_swlock_write, &lock); + } + + CK_ELIDE_LOCK(ck_swlock_read, &lock); + { + l = ck_pr_load_uint(&locked); + if (l != 0) { + ck_error("ERROR [RD:%d]: %u != 0\n", __LINE__, l); + } + } + CK_ELIDE_UNLOCK(ck_swlock_read, &lock); + } + + return (NULL); +} +#endif /* CK_F_PR_RTM */ + +static void * +thread_latch(void *arg) +{ + unsigned int i = ITERATE; + unsigned int l; + int tid = ck_pr_load_int(arg); + + if (aff_iterate(&a)) { + perror("ERROR: Could not affine thread"); + exit(EXIT_FAILURE); + } + + while (i--) { + if (tid == 0) { + /* Writer */ + ck_swlock_write_latch(&lock); + { + 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_swlock_write_unlatch(&lock); + } + + ck_swlock_read_lock(&lock); + { + l = ck_pr_load_uint(&locked); + if (l != 0) { + ck_error("ERROR [RD:%d]: %u != 0\n", __LINE__, l); + } + } + ck_swlock_read_unlock(&lock); + } + + return (NULL); +} + +static void * +thread(void *arg) +{ + unsigned int i = ITERATE; + unsigned int l; + int tid = ck_pr_load_int(arg); + + if (aff_iterate(&a)) { + perror("ERROR: Could not affine thread"); + exit(EXIT_FAILURE); + } + + while (i--) { + if (tid == 0) { + /* Writer */ + ck_swlock_write_lock(&lock); + { + 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_swlock_write_unlock(&lock); + } + + ck_swlock_read_lock(&lock); + { + l = ck_pr_load_uint(&locked); + if (l != 0) { + ck_error("ERROR [RD:%d]: %u != 0\n", __LINE__, l); + } + } + ck_swlock_read_unlock(&lock); + } + + return (NULL); +} + +static void +swlock_test(pthread_t *threads, void *(*f)(void *), const char *test) +{ + int i, tid[nthr]; + + fprintf(stderr, "Creating threads (%s)...", test); + for (i = 0; i < nthr; i++) { + ck_pr_store_int(&tid[i], i); + if (pthread_create(&threads[i], NULL, f, &tid[i])) { + ck_error("ERROR: Could not create thread %d\n", i); + } + } + fprintf(stderr, "."); + + for (i = 0; i < nthr; i++) + pthread_join(threads[i], NULL); + fprintf(stderr, "done (passed)\n"); + return; +} + +int +main(int argc, char *argv[]) +{ + pthread_t *threads; + + if (argc != 3) { + ck_error("Usage: validate \n"); + } + + nthr = atoi(argv[1]); + if (nthr <= 0) { + ck_error("ERROR: Number of threads must be greater than 0\n"); + } + + threads = malloc(sizeof(pthread_t) * nthr); + if (threads == NULL) { + ck_error("ERROR: Could not allocate thread structures\n"); + } + + a.delta = atoi(argv[2]); + + swlock_test(threads, thread, "regular"); + swlock_test(threads, thread_latch, "latch"); +#ifdef CK_F_PR_RTM + swlock_test(threads, thread_rtm, "rtm"); + swlock_test(threads, thread_rtm_mix, "rtm-mix"); + swlock_test(threads, thread_rtm_adaptive, "rtm-adaptive"); +#endif + return 0; +} +