Adding upstream version 0.6.0.

Signed-off-by: Daniel Baumann <daniel@debian.org>

regressions/ck_pr/benchmark/Makefile

@@ -0,0 +1,31 @@
+.PHONY: clean
+
+all: ck_pr_cas_64 ck_pr_fas_64 ck_pr_cas_64_2 ck_pr_add_64 ck_pr_faa_64 ck_pr_neg_64 fp
+
+fp: fp.c
+	$(CC) $(CFLAGS) -o fp fp.c
+
+ck_pr_cas_64_2: ck_pr_cas_64_2.c
+	$(CC) $(CFLAGS) -o ck_pr_cas_64_2 ck_pr_cas_64_2.c -lm
+
+ck_pr_cas_64: ck_pr_cas_64.c
+	$(CC) $(CFLAGS) -o ck_pr_cas_64 ck_pr_cas_64.c -lm
+
+ck_pr_fas_64: ck_pr_fas_64.c
+	$(CC) $(CFLAGS) -o ck_pr_fas_64 ck_pr_fas_64.c -lm
+
+ck_pr_add_64: ck_pr_add_64.c
+	$(CC) $(CFLAGS) -o ck_pr_add_64 ck_pr_add_64.c -lm
+
+ck_pr_faa_64: ck_pr_faa_64.c
+	$(CC) $(CFLAGS) -o ck_pr_faa_64 ck_pr_faa_64.c -lm
+
+ck_pr_neg_64: ck_pr_neg_64.c
+	$(CC) $(CFLAGS) -o ck_pr_neg_64 ck_pr_neg_64.c -lm
+
+clean:
+	rm -rf ck_pr_cas_64 ck_pr_fas_64 ck_pr_cas_64_2 ck_pr_add_64 \
+	       ck_pr_faa_64 ck_pr_neg_64 *.dSYM *.exe
+
+include ../../../build/regressions.build
+CFLAGS+=$(PTHREAD_CFLAGS) -D_GNU_SOURCE

regressions/ck_pr/benchmark/benchmark.h

@@ -0,0 +1,130 @@
+#include <errno.h>
+#include <inttypes.h>
+#include <pthread.h>
+#include <math.h>
+#include <stdint.h>
+#include <stdio.h>
+#include <stdlib.h>
+#include <string.h>
+#include <strings.h>
+#include <unistd.h>
+#include <sys/time.h>
+
+#include <ck_pr.h>
+
+#include "../../common.h"
+
+/* 8! = 40320, evenly divide 1 .. 8 processor workload. */
+#define WORKLOAD (40320 * 2056)
+
+struct block {
+	unsigned int tid;
+};
+
+static struct affinity a;
+static unsigned int ready;
+static uint64_t *count;
+static uint64_t nthr;
+
+static uint64_t object[2] CK_CC_CACHELINE;
+
+static void *
+fairness(void *null)
+{
+	struct block *context = null;
+	unsigned int i = context->tid;
+
+        if (aff_iterate(&a)) {
+                perror("ERROR: Could not affine thread");
+                exit(EXIT_FAILURE);
+        }
+
+	while (ck_pr_load_uint(&ready) == 0);
+	while (ck_pr_load_uint(&ready)) {
+		ATOMIC;
+		ATOMIC;
+		ATOMIC;
+		ATOMIC;
+		ck_pr_store_64(count + i, count[i] + 1);
+	}
+
+	return (NULL);
+}
+
+int
+main(int argc, char *argv[])
+{
+	uint64_t v, d;
+	unsigned int i;
+	pthread_t *threads;
+	struct block *context;
+
+	if (argc != 3) {
+		ck_error("Usage: " ATOMIC_STRING " <number of threads> <affinity delta>\n");
+		exit(EXIT_FAILURE);
+	}
+
+	nthr = atoi(argv[1]);
+	if (nthr <= 0) {
+		ck_error("ERROR: Number of threads must be greater than 0\n");
+		exit(EXIT_FAILURE);
+	}
+
+	threads = malloc(sizeof(pthread_t) * nthr);
+	if (threads == NULL) {
+		ck_error("ERROR: Could not allocate thread structures\n");
+		exit(EXIT_FAILURE);
+	}
+
+	context = malloc(sizeof(struct block) * nthr);
+	if (context == NULL) {
+		ck_error("ERROR: Could not allocate thread contexts\n");
+		exit(EXIT_FAILURE);
+	}
+
+	a.delta = atoi(argv[2]);
+	a.request = 0;
+
+	count = malloc(sizeof(uint64_t) * nthr);
+	if (count == NULL) {
+		ck_error("ERROR: Could not create acquisition buffer\n");
+		exit(EXIT_FAILURE);
+	}
+	memset(count, 0, sizeof(uint64_t) * nthr);
+
+	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);
+			exit(EXIT_FAILURE);
+		}
+	}
+	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]);
+		v += count[i];
+	}
+
+	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] - v) * (count[i] - v);
+
+	printf("# average     : %15" PRIu64 "\n", v);
+	printf("# deviation   : %.2f (%.2f%%)\n\n", sqrt(d / nthr), (sqrt(d / nthr) / v) * 100.00);
+
+	return (0);
+}
+

regressions/ck_pr/benchmark/ck_pr_add_64.c

@@ -0,0 +1,16 @@
+#include <ck_pr.h>
+
+#ifdef CK_F_PR_ADD_64
+#define ATOMIC ck_pr_add_64(object, 1)
+#define ATOMIC_STRING "ck_pr_add_64"
+#include "benchmark.h"
+#else
+#warning Did not find ADD_64 implementation.
+#include <stdlib.h>
+
+int
+main(void)
+{
+	exit(EXIT_FAILURE);
+}
+#endif

regressions/ck_pr/benchmark/ck_pr_cas_64.c

@@ -0,0 +1,16 @@
+#include <ck_pr.h>
+
+#ifdef CK_F_PR_CAS_64
+#define ATOMIC ck_pr_cas_64(object, 1, 1)
+#define ATOMIC_STRING "ck_pr_cas_64"
+#include "benchmark.h"
+#else
+#warning Did not find CAS_64 implementation.
+#include <stdlib.h>
+
+int
+main(void)
+{
+	exit(EXIT_FAILURE);
+}
+#endif

regressions/ck_pr/benchmark/ck_pr_cas_64_2.c

@@ -0,0 +1,17 @@
+#include <ck_pr.h>
+
+#ifdef CK_F_PR_CAS_64_2
+#define ATOMIC { uint64_t z[2] = {1, 2}; ck_pr_cas_64_2(object, z, z); }
+#define ATOMIC_STRING "ck_pr_cas_64_2"
+#include "benchmark.h"
+#else
+#include <stdio.h>
+#include <stdlib.h>
+
+int
+main(void)
+{
+	fprintf(stderr, "Unsupported.\n");
+	return 0;
+}
+#endif

regressions/ck_pr/benchmark/ck_pr_faa_64.c

@@ -0,0 +1,16 @@
+#include <ck_pr.h>
+
+#ifdef CK_F_PR_FAA_64
+#define ATOMIC ck_pr_faa_64(object, 1)
+#define ATOMIC_STRING "ck_pr_faa_64"
+#include "benchmark.h"
+#else
+#warning Did not find FAA_64 implementation.
+#include <stdlib.h>
+
+int
+main(void)
+{
+	exit(EXIT_FAILURE);
+}
+#endif

regressions/ck_pr/benchmark/ck_pr_fas_64.c

@@ -0,0 +1,17 @@
+#include <ck_pr.h>
+
+#ifdef CK_F_PR_FAS_64
+#define ATOMIC ck_pr_fas_64(object, 1)
+#define ATOMIC_STRING "ck_pr_fas_64"
+#include "benchmark.h"
+#else
+#warning Did not find FAS_64 implementation.
+#include <stdlib.h>
+
+int
+main(void)
+{
+
+	return 0;
+}
+#endif

regressions/ck_pr/benchmark/ck_pr_neg_64.c

@@ -0,0 +1,16 @@
+#include <ck_pr.h>
+
+#ifdef CK_F_PR_NEG_64
+#define ATOMIC ck_pr_neg_64(object)
+#define ATOMIC_STRING "ck_pr_neg_64"
+#include "benchmark.h"
+#else
+#warning Did not find NEG_64 implementation.
+#include <stdlib.h>
+
+int
+main(void)
+{
+	exit(EXIT_FAILURE);
+}
+#endif

regressions/ck_pr/benchmark/fp.c

@@ -0,0 +1,66 @@
+#include <stdio.h>
+#include <inttypes.h>
+#include <stdint.h>
+
+#include "../../common.h"
+
+#ifndef IR
+#define IR 3000000
+#endif /* IR */
+
+static int a CK_CC_CACHELINE;
+static int b CK_CC_CACHELINE;
+
+int
+main(void)
+{
+	uint64_t s, e;
+	unsigned int i;
+
+	s = rdtsc();
+	for (i = 0; i < IR; i++) {
+		ck_pr_load_int(&a);
+		ck_pr_fence_strict_load();
+		ck_pr_load_int(&b);
+	}
+	e = rdtsc();
+	printf("[A] fence_load: %" PRIu64 "\n", (e - s) / IR);
+
+	s = rdtsc();
+	for (i = 0; i < IR; i++) {
+		if (ck_pr_load_int(&a) == 0)
+			ck_pr_barrier();
+		ck_pr_fence_strict_lock();
+		ck_pr_load_int(&b);
+	}
+	e = rdtsc();
+	printf("[A] fence_lock: %" PRIu64 "\n", (e - s) / IR);
+
+	s = rdtsc();
+	for (i = 0; i < IR; i++) {
+		ck_pr_store_int(&a, 0);
+		ck_pr_fence_strict_store();
+		ck_pr_store_int(&b, 0);
+	}
+	e = rdtsc();
+	printf("[B] fence_store: %" PRIu64 "\n", (e - s) / IR);
+
+	s = rdtsc();
+	for (i = 0; i < IR; i++) {
+		ck_pr_store_int(&a, 0);
+		ck_pr_fence_strict_memory();
+		ck_pr_load_int(&b);
+	}
+	e = rdtsc();
+	printf("[C] fence_memory: %" PRIu64 "\n", (e - s) / IR);
+
+	s = rdtsc();
+	for (i = 0; i < IR; i++) {
+		ck_pr_store_int(&a, 0);
+		ck_pr_faa_int(&a, 0);
+		ck_pr_load_int(&b);
+	}
+	e = rdtsc();
+	printf("[C] atomic: %" PRIu64 "\n", (e - s) / IR);
+	return 0;
+}