tests: add atomic_add-bench

With this microbenchmark we can measure the overhead of emulating atomic
instructions with a configurable degree of contention.

The benchmark spawns $n threads, each performing $o atomic ops (additions)
in a loop. Each atomic operation is performed on a different cache line
(assuming lines are 64b long) that is randomly selected from a range [0, $r).

[ Note: each $foo corresponds to a -foo flag ]

Signed-off-by: Emilio G. Cota <cota@braap.org>
Signed-off-by: Richard Henderson <rth@twiddle.net>
Message-Id: <1467054136-10430-20-git-send-email-cota@braap.org>

3 files changed, 167 insertions, 1 deletions

diff --git a/tests/.gitignore b/tests/.gitignore
index 4aec8bc5fa..64e050e859 100644
--- a/tests/.gitignore
+++ b/tests/.gitignore
@@ -1,3 +1,4 @@
+atomic_add-bench
 check-qdict
 check-qfloat
 check-qint
diff --git a/tests/Makefile.include b/tests/Makefile.include
index cd058efc14..22656eaf26 100644
--- a/tests/Makefile.include
+++ b/tests/Makefile.include
@@ -460,7 +460,8 @@ test-obj-y = tests/check-qint.o tests/check-qstring.o tests/check-qdict.o \
 	tests/test-opts-visitor.o tests/test-qmp-event.o \
 	tests/rcutorture.o tests/test-rcu-list.o \
 	tests/test-qdist.o \
-	tests/test-qht.o tests/qht-bench.o tests/test-qht-par.o
+	tests/test-qht.o tests/qht-bench.o tests/test-qht-par.o \
+	tests/atomic_add-bench.o
 
 $(test-obj-y): QEMU_INCLUDES += -Itests
 QEMU_CFLAGS += -I$(SRC_PATH)/tests
@@ -507,6 +508,7 @@ tests/test-qht$(EXESUF): tests/test-qht.o $(test-util-obj-y)
 tests/test-qht-par$(EXESUF): tests/test-qht-par.o tests/qht-bench$(EXESUF) $(test-util-obj-y)
 tests/qht-bench$(EXESUF): tests/qht-bench.o $(test-util-obj-y)
 tests/test-bufferiszero$(EXESUF): tests/test-bufferiszero.o $(test-util-obj-y)
+tests/atomic_add-bench$(EXESUF): tests/atomic_add-bench.o $(test-util-obj-y)
 
 tests/test-qdev-global-props$(EXESUF): tests/test-qdev-global-props.o \
 	hw/core/qdev.o hw/core/qdev-properties.o hw/core/hotplug.o\
diff --git a/tests/atomic_add-bench.c b/tests/atomic_add-bench.c
new file mode 100644
index 0000000000..caa1e8e689
--- /dev/null
+++ b/tests/atomic_add-bench.c
@@ -0,0 +1,163 @@
+#include "qemu/osdep.h"
+#include "qemu/thread.h"
+#include "qemu/host-utils.h"
+#include "qemu/processor.h"
+
+struct thread_info {
+    uint64_t r;
+} QEMU_ALIGNED(64);
+
+struct count {
+    unsigned long val;
+} QEMU_ALIGNED(64);
+
+static QemuThread *threads;
+static struct thread_info *th_info;
+static unsigned int n_threads = 1;
+static unsigned int n_ready_threads;
+static struct count *counts;
+static unsigned int duration = 1;
+static unsigned int range = 1024;
+static bool test_start;
+static bool test_stop;
+
+static const char commands_string[] =
+    " -n = number of threads\n"
+    " -d = duration in seconds\n"
+    " -r = range (will be rounded up to pow2)";
+
+static void usage_complete(char *argv[])
+{
+    fprintf(stderr, "Usage: %s [options]\n", argv[0]);
+    fprintf(stderr, "options:\n%s\n", commands_string);
+}
+
+/*
+ * From: https://en.wikipedia.org/wiki/Xorshift
+ * This is faster than rand_r(), and gives us a wider range (RAND_MAX is only
+ * guaranteed to be >= INT_MAX).
+ */
+static uint64_t xorshift64star(uint64_t x)
+{
+    x ^= x >> 12; /* a */
+    x ^= x << 25; /* b */
+    x ^= x >> 27; /* c */
+    return x * UINT64_C(2685821657736338717);
+}
+
+static void *thread_func(void *arg)
+{
+    struct thread_info *info = arg;
+
+    atomic_inc(&n_ready_threads);
+    while (!atomic_read(&test_start)) {
+        cpu_relax();
+    }
+
+    while (!atomic_read(&test_stop)) {
+        unsigned int index;
+
+        info->r = xorshift64star(info->r);
+        index = info->r & (range - 1);
+        atomic_inc(&counts[index].val);
+    }
+    return NULL;
+}
+
+static void run_test(void)
+{
+    unsigned int remaining;
+    unsigned int i;
+
+    while (atomic_read(&n_ready_threads) != n_threads) {
+        cpu_relax();
+    }
+    atomic_set(&test_start, true);
+    do {
+        remaining = sleep(duration);
+    } while (remaining);
+    atomic_set(&test_stop, true);
+
+    for (i = 0; i < n_threads; i++) {
+        qemu_thread_join(&threads[i]);
+    }
+}
+
+static void create_threads(void)
+{
+    unsigned int i;
+
+    threads = g_new(QemuThread, n_threads);
+    th_info = g_new(struct thread_info, n_threads);
+    counts = qemu_memalign(64, sizeof(*counts) * range);
+    memset(counts, 0, sizeof(*counts) * range);
+
+    for (i = 0; i < n_threads; i++) {
+        struct thread_info *info = &th_info[i];
+
+        info->r = (i + 1) ^ time(NULL);
+        qemu_thread_create(&threads[i], NULL, thread_func, info,
+                           QEMU_THREAD_JOINABLE);
+    }
+}
+
+static void pr_params(void)
+{
+    printf("Parameters:\n");
+    printf(" # of threads:      %u\n", n_threads);
+    printf(" duration:          %u\n", duration);
+    printf(" ops' range:        %u\n", range);
+}
+
+static void pr_stats(void)
+{
+    unsigned long long val = 0;
+    unsigned int i;
+    double tx;
+
+    for (i = 0; i < range; i++) {
+        val += counts[i].val;
+    }
+    tx = val / duration / 1e6;
+
+    printf("Results:\n");
+    printf("Duration:            %u s\n", duration);
+    printf(" Throughput:         %.2f Mops/s\n", tx);
+    printf(" Throughput/thread:  %.2f Mops/s/thread\n", tx / n_threads);
+}
+
+static void parse_args(int argc, char *argv[])
+{
+    int c;
+
+    for (;;) {
+        c = getopt(argc, argv, "hd:n:r:");
+        if (c < 0) {
+            break;
+        }
+        switch (c) {
+        case 'h':
+            usage_complete(argv);
+            exit(0);
+        case 'd':
+            duration = atoi(optarg);
+            break;
+        case 'n':
+            n_threads = atoi(optarg);
+            break;
+        case 'r':
+            range = pow2ceil(atoi(optarg));
+            break;
+        }
+    }
+}
+
+int main(int argc, char *argv[])
+{
+    parse_args(argc, argv);
+    pr_params();
+    create_threads();
+    run_test();
+    pr_stats();
+    return 0;
+}