ck_swlock: A single writer rwlock.

This lock is copy-safe when the latch operations are used.

Simplified write side operations lead to lower latencies than ck_rwlock
for single writer workloads.

.gitignore

@@ -15,6 +15,10 @@ build/Makefile
 *.so
 *.dSYM
 .*.sw[op]
+GPATH
+GRTAGS
+GTAGS
+ID
 regressions/ck_array/validate/serial
 regressions/ck_cohort/benchmark/ck_cohort.LATENCY
 regressions/ck_cohort/benchmark/ck_cohort.THROUGHPUT
@@ -156,3 +160,16 @@ regressions/ck_rwcohort/benchmark/ck_rp.LATENCY
 regressions/ck_rwcohort/benchmark/ck_rp.THROUGHPUT
 regressions/ck_rwcohort/benchmark/ck_wp.LATENCY
 regressions/ck_rwcohort/benchmark/ck_wp.THROUGHPUT
+regressions/ck_hs/benchmark/parallel_bytestring.delete
+regressions/ck_ht/benchmark/parallel_bytestring.delete
+regressions/ck_ht/benchmark/serial.delete
+regressions/ck_ht/validate/serial.delete
+regressions/ck_rhs/benchmark/parallel_bytestring
+regressions/ck_rhs/benchmark/serial
+regressions/ck_rhs/validate/serial
+regressions/ck_spinlock/benchmark/ck_hclh.LATENCY
+regressions/ck_spinlock/benchmark/ck_hclh.THROUGHPUT
+regressions/ck_spinlock/validate/ck_hclh
+regressions/ck_swlock/benchmark/latency
+regressions/ck_swlock/benchmark/throughput
+regressions/ck_swlock/validate/validate

include/ck_swlock.h

@@ -0,0 +1,374 @@
+/*
+ * 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.
+ */
+
+#ifndef _CK_SWLOCK_H
+#define _CK_SWLOCK_H
+
+#include <ck_elide.h>
+#include <ck_pr.h>
+#include <stdbool.h>
+#include <stddef.h>
+
+struct ck_swlock {
+	uint32_t writer;
+	uint32_t n_readers;
+};
+typedef struct ck_swlock ck_swlock_t;
+
+#define CK_SWLOCK_INITIALIZER {0, 0}
+#define CK_SWLOCK_LATCH_BIT (1 << 31)
+#define CK_SWLOCK_READER_BITS (UINT32_MAX ^ CK_SWLOCK_LATCH_BIT)
+
+CK_CC_INLINE static void
+ck_swlock_init(struct ck_swlock *rw)
+{
+
+	rw->writer = 0;
+	rw->n_readers = 0;
+	ck_pr_fence_store();
+	return;
+}
+
+CK_CC_INLINE static void
+ck_swlock_write_unlock(ck_swlock_t *rw)
+{
+
+	ck_pr_store_32(&rw->writer, 0);
+	return;
+}
+
+CK_CC_INLINE static bool
+ck_swlock_locked_writer(ck_swlock_t *rw)
+{
+
+	return ck_pr_load_32(&rw->writer);
+}
+
+CK_CC_INLINE static void
+ck_swlock_write_downgrade(ck_swlock_t *rw)
+{
+
+	ck_pr_inc_32(&rw->n_readers);
+	ck_swlock_write_unlock(rw);
+	return;
+}
+
+CK_CC_INLINE static bool
+ck_swlock_locked(ck_swlock_t *rw)
+{
+	uint32_t r;
+
+	r = ck_pr_load_32(&rw->writer);
+
+	return ck_pr_load_32(&rw->n_readers) | r;
+}
+
+CK_CC_INLINE static bool
+ck_swlock_write_trylock(ck_swlock_t *rw)
+{
+
+	ck_pr_store_32(&rw->writer, 1);
+
+	ck_pr_fence_atomic_load();
+
+	if (ck_pr_load_32(&rw->n_readers) != 0) {
+		ck_swlock_write_unlock(rw);
+		return false;
+	}
+
+	return true;
+}
+
+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_store_32(&rw->writer, 1);
+	
+	ck_pr_fence_atomic_load();
+
+	while (ck_pr_load_32(&rw->n_readers) != 0)
+		ck_pr_stall();
+
+	return;
+}
+
+CK_CC_INLINE static void
+ck_swlock_write_latch(ck_swlock_t *rw)
+{
+
+	ck_pr_store_32(&rw->writer, 1);
+	
+	ck_pr_fence_atomic_load();
+	
+	while (ck_pr_cas_32(&rw->n_readers, 0, CK_SWLOCK_LATCH_BIT) == false) {
+		/* Stall until readers have seen the latch and cleared */
+		ck_pr_stall();
+	}
+
+	return;
+}
+
+CK_CC_INLINE static void
+ck_swlock_write_unlatch(ck_swlock_t *rw)
+{
+
+	ck_pr_store_32(&rw->n_readers, 0);
+	
+	ck_swlock_write_unlock(rw);
+	
+	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_CC_INLINE static bool
+ck_swlock_read_trylock(ck_swlock_t *rw)
+{
+
+	if (ck_pr_load_32(&rw->writer) != 0)
+		return false;
+
+	if (ck_pr_faa_32(&rw->n_readers, 1) & CK_SWLOCK_LATCH_BIT) {
+		return false;
+	}
+
+	/*
+	 * Serialize with respect to concurrent write
+	 * lock operation.
+	 */
+	ck_pr_fence_atomic_load();
+
+	if (ck_pr_load_32(&rw->writer) == 0) {
+		ck_pr_fence_load();
+		return true;
+	}
+
+	ck_pr_dec_32(&rw->n_readers);
+	return false;
+}
+
+CK_ELIDE_TRYLOCK_PROTOTYPE(ck_swlock_read, ck_swlock_t,
+    ck_swlock_locked_writer, ck_swlock_read_trylock)
+
+CK_CC_INLINE static void
+ck_swlock_read_lock(ck_swlock_t *rw)
+{
+
+	for (;;) {
+		while (ck_pr_load_32(&rw->writer) != 0)
+			ck_pr_stall();
+
+		ck_pr_inc_32(&rw->n_readers);
+
+		/*
+		 * Serialize with respect to concurrent write
+		 * lock operation.
+		 */
+		ck_pr_fence_atomic_load();
+
+		if (ck_pr_load_32(&rw->writer) == 0)
+			break;
+
+		ck_pr_dec_32(&rw->n_readers);
+	}
+
+	/* Acquire semantics are necessary. */
+	ck_pr_fence_load();
+	return;
+}
+
+CK_CC_INLINE static void
+ck_swlock_read_latchlock(ck_swlock_t *rw)
+{
+
+	for (;;) {
+
+		while (ck_pr_load_32(&rw->writer) != 0)
+			ck_pr_stall();
+
+		if (ck_pr_faa_32(&rw->n_readers, 1) & CK_SWLOCK_LATCH_BIT) {
+			/* Writer has latched, stall the reader */
+			continue;
+		}
+
+		/*
+		 * Serialize with respect to concurrent write
+		 * lock operation.
+		 */
+		ck_pr_fence_atomic_load();
+
+		if (ck_pr_load_32(&rw->writer) == 0)
+			break;
+		
+		ck_pr_dec_32(&rw->n_readers);
+	}
+
+	/* Acquire semantics are necessary. */
+	ck_pr_fence_load();
+	return;
+}
+
+
+CK_CC_INLINE static bool
+ck_swlock_locked_reader(ck_swlock_t *rw)
+{
+
+	ck_pr_fence_load();
+	return (ck_pr_load_32(&rw->n_readers) & CK_SWLOCK_READER_BITS);
+}
+
+CK_CC_INLINE static void
+ck_swlock_read_unlock(ck_swlock_t *rw)
+{
+
+	ck_pr_fence_load_atomic();
+	ck_pr_dec_32(&rw->n_readers);
+	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)
+
+/*
+ * Recursive writer reader-writer lock implementation.
+ */
+struct ck_swlock_recursive {
+	struct ck_swlock rw;
+	uint32_t wc;
+};
+typedef struct ck_swlock_recursive ck_swlock_recursive_t;
+
+#define CK_SWLOCK_RECURSIVE_INITIALIZER {CK_SWLOCK_INITIALIZER, 0}
+
+CK_CC_INLINE static void
+ck_swlock_recursive_write_lock(ck_swlock_recursive_t *rw)
+{
+
+	ck_pr_store_32(&rw->rw.writer, 1);
+
+	ck_pr_fence_atomic_load();
+
+	while (ck_pr_load_32(&rw->rw.n_readers) != 0)
+		ck_pr_stall();
+
+	rw->wc++;
+	return;
+}
+
+CK_CC_INLINE static void
+ck_swlock_recursive_write_latch(ck_swlock_recursive_t *rw)
+{
+	ck_pr_store_32(&rw->rw.writer, 1);
+
+	ck_pr_fence_atomic_load();
+
+	while (ck_pr_cas_32(&rw->rw.n_readers, 0, CK_SWLOCK_LATCH_BIT) == false)
+		ck_pr_stall();
+
+	rw->wc++;
+	return;
+}
+
+CK_CC_INLINE static bool
+ck_swlock_recursive_write_trylock(ck_swlock_recursive_t *rw)
+{
+
+	ck_pr_store_32(&rw->rw.writer, 1);
+
+	ck_pr_fence_atomic_load();
+
+	if (ck_pr_load_32(&rw->rw.n_readers) != 0) {
+		ck_pr_store_32(&rw->rw.writer, 0);
+		return false;
+	}
+
+	rw->wc++;
+	return true;
+}
+
+CK_CC_INLINE static void
+ck_swlock_recursive_write_unlock(ck_swlock_recursive_t *rw)
+{
+
+	if (--rw->wc == 0) {
+		ck_pr_fence_release();
+		ck_pr_store_32(&rw->rw.writer, 0);
+	}
+
+	return;
+}
+
+CK_CC_INLINE static void
+ck_swlock_recursive_write_unlatch(ck_swlock_recursive_t *rw)
+{
+	ck_pr_store_32(&rw->rw.n_readers, 0);
+
+	ck_swlock_recursive_write_unlock(rw);
+
+	return;
+}
+
+
+CK_CC_INLINE static void
+ck_swlock_recursive_read_lock(ck_swlock_recursive_t *rw)
+{
+
+	ck_swlock_read_lock(&rw->rw);
+	return;
+}
+
+CK_CC_INLINE static void
+ck_swlock_recursive_read_latchlock(ck_swlock_recursive_t *rw)
+{
+
+	ck_swlock_read_latchlock(&rw->rw);
+	return;
+}
+
+CK_CC_INLINE static bool
+ck_swlock_recursive_read_trylock(ck_swlock_recursive_t *rw)
+{
+
+	return ck_swlock_read_trylock(&rw->rw);
+}
+
+CK_CC_INLINE static void
+ck_swlock_recursive_read_unlock(ck_swlock_recursive_t *rw)
+{
+
+	ck_swlock_read_unlock(&rw->rw);
+	return;
+}
+
+#endif /* _CK_SWLOCK_H */
+

regressions/Makefile

@@ -16,6 +16,7 @@ DIR=array	\
     queue	\
     ring	\
     rwlock	\
+    swlock	\
     sequence	\
     spinlock	\
     stack
@@ -58,6 +59,8 @@ all:
 	$(MAKE) -C ./ck_ring/benchmark all
 	$(MAKE) -C ./ck_rwlock/validate all
 	$(MAKE) -C ./ck_rwlock/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
@@ -101,6 +104,8 @@ 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

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

regressions/ck_swlock/benchmark/latency.c

@@ -0,0 +1,100 @@
+/*
+ * 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 <ck_swlock.h>
+#include <inttypes.h>
+#include <stdio.h>
+
+#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);
+
+	for (i = 0; i < STEPS; i++) {
+		ck_swlock_read_latchlock(&swlock);
+		ck_swlock_read_unlock(&swlock);
+	}
+
+	s_b = rdtsc();
+	for (i = 0; i < STEPS; i++) {
+		ck_swlock_read_latchlock(&swlock);
+		ck_swlock_read_unlock(&swlock);
+	}
+	e_b = rdtsc();
+	printf("                READ_LATCHLOCK:  swlock   %15" PRIu64 "\n", (e_b - s_b) / STEPS);
+
+
+	return 0;
+}
+

regressions/ck_swlock/benchmark/throughput.c

@@ -0,0 +1,249 @@
+/*
+ * 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 <ck_swlock.h>
+#include <inttypes.h>
+#include <pthread.h>
+#include <stdio.h>
+#include <stdint.h>
+#include <stdlib.h>
+#include <unistd.h>
+
+#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 *
+thread_latchlock(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_latchlock(&rw.lock);
+		ck_swlock_read_unlock(&rw.lock);
+		ck_swlock_read_latchlock(&rw.lock);
+		ck_swlock_read_unlock(&rw.lock);
+		ck_swlock_read_latchlock(&rw.lock);
+		ck_swlock_read_unlock(&rw.lock);
+		ck_swlock_read_latchlock(&rw.lock);
+		ck_swlock_read_unlock(&rw.lock);
+		ck_swlock_read_latchlock(&rw.lock);
+		ck_swlock_read_unlock(&rw.lock);
+		ck_swlock_read_latchlock(&rw.lock);
+		ck_swlock_read_unlock(&rw.lock);
+		ck_swlock_read_latchlock(&rw.lock);
+		ck_swlock_read_unlock(&rw.lock);
+		ck_swlock_read_latchlock(&rw.lock);
+		ck_swlock_read_unlock(&rw.lock);
+		ck_swlock_read_latchlock(&rw.lock);
+		ck_swlock_read_unlock(&rw.lock);
+		ck_swlock_read_latchlock(&rw.lock);
+		ck_swlock_read_unlock(&rw.lock);
+		ck_swlock_read_latchlock(&rw.lock);
+		ck_swlock_read_unlock(&rw.lock);
+		ck_swlock_read_latchlock(&rw.lock);
+		ck_swlock_read_unlock(&rw.lock);
+		ck_swlock_read_latchlock(&rw.lock);
+		ck_swlock_read_unlock(&rw.lock);
+		ck_swlock_read_latchlock(&rw.lock);
+		ck_swlock_read_unlock(&rw.lock);
+		ck_swlock_read_latchlock(&rw.lock);
+		ck_swlock_read_unlock(&rw.lock);
+		ck_swlock_read_latchlock(&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 <delta> <threads>\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");
+	swlock_test(p, d, latency, thread_latchlock, "swlock");
+
+	return 0;
+}
+

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 -O0

regressions/ck_swlock/validate/validate.c

@@ -0,0 +1,239 @@
+/*
+ * 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 <errno.h>
+#include <inttypes.h>
+#include <pthread.h>
+#include <math.h>
+#include <stdint.h>
+#include <stdio.h>
+#include <stdlib.h>
+#include <strings.h>
+#include <unistd.h>
+#include <sys/time.h>
+
+#include <ck_pr.h>
+#include <ck_swlock.h>
+
+#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;
+static ck_swlock_recursive_t r_lock = CK_SWLOCK_RECURSIVE_INITIALIZER;
+
+static void *
+thread_recursive(void *arg)
+{
+	int i = ITERATE;
+	unsigned int l;
+	unsigned int tid = *(int *) arg;
+
+        if (aff_iterate(&a)) {
+                perror("ERROR: Could not affine thread");
+                exit(EXIT_FAILURE);
+        }
+
+	while (i--) {
+		if (tid == 0) {
+			/* Writer */
+			while (ck_swlock_recursive_write_trylock(&r_lock) == false)
+				ck_pr_stall();
+
+			ck_swlock_recursive_write_lock(&r_lock);
+			ck_swlock_recursive_write_latch(&r_lock);
+			ck_swlock_recursive_write_lock(&r_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_recursive_write_unlock(&r_lock);
+			ck_swlock_recursive_write_unlatch(&r_lock);
+			ck_swlock_recursive_write_unlock(&r_lock);
+			ck_swlock_recursive_write_unlock(&r_lock);
+		}
+		
+		ck_swlock_recursive_read_latchlock(&r_lock);
+		{
+			l = ck_pr_load_uint(&locked);
+			if (l != 0) {
+				ck_error("ERROR [RD:%d]: %u != 0\n", __LINE__, l);
+			}
+		}
+		ck_swlock_recursive_read_unlock(&r_lock);
+	}
+
+	return (NULL);
+}
+
+
+static void *
+thread(void *arg)
+{
+	unsigned int i = ITERATE;
+	unsigned int l;
+	int tid = *(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_latchlock(&lock);
+		{
+			l = ck_pr_load_uint(&locked);
+			if (l != 0) {
+				ck_error("ERROR [RD:%d]: %u != 0 r:%x w:%d\n", __LINE__, l, lock.n_readers, lock.writer);
+			}
+		}
+		ck_swlock_read_unlock(&lock);
+	}
+
+	return (NULL);
+}
+
+static void
+swlock_test(pthread_t *threads, void *(*f)(void *), const char *test)
+{
+	int i;
+
+	fprintf(stderr, "Creating threads (%s)...", test);
+	for (i = 0; i < nthr; i++) {
+		if (pthread_create(&threads[i], NULL, f, &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 <number of threads> <affinity delta>\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_recursive, "recursive");
+	return 0;
+}
+