tests: benchmarks: move pthread_pressure to benchmarks/posix
The pthread_pressure test was not a typical test per se. It was a benchmark in search of the proper home. Let's move it to the correct place in the Zephyr tree, add a doc, and provide some reporting. Currently, k_threads out-perform pthreads by almost a factor of 2. The theoretical maximum performance of pthreads would be at parity of k_threads, since pthreads are a wrapper around kernel threads. It would be great to reduce the gap. Signed-off-by: Chris Friedt <cfriedt@tenstorrent.com>
This commit is contained in:
Chris Friedt
Benjamin Cabé
parent
7497b8bbbe
commit
3f60489fae
10 changed files with 327 additions and 268 deletions
|
|
@ -1,10 +0,0 @@
|
|||
# Copyright (c) 2023, Meta
|
||||
#
|
||||
# SPDX-License-Identifier: Apache-2.0
|
||||
|
||||
cmake_minimum_required(VERSION 3.20.0)
|
||||
find_package(Zephyr REQUIRED HINTS $ENV{ZEPHYR_BASE})
|
||||
project(pthread_pressure)
|
||||
|
||||
FILE(GLOB app_sources src/*.c)
|
||||
target_sources(app PRIVATE ${app_sources})
|
||||
|
|
@ -1,51 +0,0 @@
|
|||
# Copyright (c) 2023, Meta
|
||||
#
|
||||
# SPDX-License-Identifier: Apache-2.0
|
||||
|
||||
source "Kconfig.zephyr"
|
||||
|
||||
config TEST_NUM_CPUS
|
||||
int "Number of CPUs to use in parallel"
|
||||
range 1 MP_MAX_NUM_CPUS
|
||||
default MP_MAX_NUM_CPUS
|
||||
help
|
||||
The number of parallel threads to run during the test.
|
||||
|
||||
config TEST_DURATION_S
|
||||
int "Number of seconds to run the test"
|
||||
default 5
|
||||
help
|
||||
Duration for the test, in seconds.
|
||||
|
||||
config TEST_DELAY_US
|
||||
int "Microseconds to delay between pthread join and create"
|
||||
default 0
|
||||
help
|
||||
If there is a race condition, a value of zero here should
|
||||
cause a crash.
|
||||
|
||||
config TEST_STACK_SIZE
|
||||
int "Size of each thread stack in this test"
|
||||
default 1024 if 64BIT
|
||||
default 1024 if SPARC
|
||||
default 512
|
||||
help
|
||||
The minimal stack size required to run a minimal thread.
|
||||
|
||||
config TEST_KTHREADS
|
||||
bool "Test k_threads"
|
||||
default y
|
||||
help
|
||||
Run tests for k_threads
|
||||
|
||||
config TEST_PTHREADS
|
||||
bool "Test pthreads"
|
||||
default y
|
||||
help
|
||||
Run tests for pthreads
|
||||
|
||||
config TEST_EXTRA_ASSERTIONS
|
||||
bool "Add extra assertions into the hot path"
|
||||
default y
|
||||
help
|
||||
This can be disabled for benchmarking.
|
||||
|
|
@ -1,11 +0,0 @@
|
|||
CONFIG_ZTEST=y
|
||||
CONFIG_POSIX_API=y
|
||||
CONFIG_POSIX_PRIORITY_SCHEDULING=y
|
||||
|
||||
## Note: for benchmarking purposes, uncomment the Kconfig below
|
||||
# CONFIG_TEST_DURATION_S=60
|
||||
# CONFIG_TEST_EXTRA_ASSERTIONS=n
|
||||
# CONFIG_ASSERT=n
|
||||
|
||||
## Optionally, uncomment this to only test pthreads:
|
||||
# CONFIG_TEST_KTHREADS=n
|
||||
|
|
@ -1,230 +0,0 @@
|
|||
/*
|
||||
* Copyright (c) 2023, Meta
|
||||
*
|
||||
* SPDX-License-Identifier: Apache-2.0
|
||||
*/
|
||||
|
||||
#include <pthread.h>
|
||||
|
||||
#include <zephyr/sys/util.h>
|
||||
#include <zephyr/ztest.h>
|
||||
|
||||
#define STACK_SIZE K_THREAD_STACK_LEN(CONFIG_TEST_STACK_SIZE)
|
||||
|
||||
/* update interval for printing stats */
|
||||
#if CONFIG_TEST_DURATION_S >= 60
|
||||
#define UPDATE_INTERVAL_S 10
|
||||
#elif CONFIG_TEST_DURATION_S >= 30
|
||||
#define UPDATE_INTERVAL_S 5
|
||||
#else
|
||||
#define UPDATE_INTERVAL_S 1
|
||||
#endif
|
||||
|
||||
/* 32 threads is mainly a limitation of find_lsb_set() */
|
||||
#define NUM_THREADS MIN(32, MIN(CONFIG_TEST_NUM_CPUS, CONFIG_POSIX_THREAD_THREADS_MAX))
|
||||
|
||||
typedef int (*create_fn)(int i);
|
||||
typedef int (*join_fn)(int i);
|
||||
|
||||
static void *setup(void);
|
||||
static void before(void *fixture);
|
||||
|
||||
/* bitmask of available threads */
|
||||
static bool alive[NUM_THREADS];
|
||||
|
||||
/* array of thread stacks */
|
||||
static K_THREAD_STACK_ARRAY_DEFINE(thread_stacks, NUM_THREADS, STACK_SIZE);
|
||||
|
||||
static struct k_thread k_threads[NUM_THREADS];
|
||||
static uint64_t counters[NUM_THREADS];
|
||||
static uint64_t prev_counters[NUM_THREADS];
|
||||
|
||||
static void print_stats(uint64_t now, uint64_t end)
|
||||
{
|
||||
printk("now (ms): %llu end (ms): %llu\n", now, end);
|
||||
for (int i = 0; i < NUM_THREADS; ++i) {
|
||||
printk("Thread %d created and joined %llu times (%llu joins/s)\n", i, counters[i],
|
||||
(counters[i] - prev_counters[i]) / UPDATE_INTERVAL_S);
|
||||
prev_counters[i] = counters[i];
|
||||
}
|
||||
}
|
||||
|
||||
static void test_create_join_common(const char *tag, create_fn create, join_fn join)
|
||||
{
|
||||
int i;
|
||||
int ret;
|
||||
uint64_t now_ms = k_uptime_get();
|
||||
const uint64_t end_ms = now_ms + MSEC_PER_SEC * CONFIG_TEST_DURATION_S;
|
||||
uint64_t update_ms = now_ms + MSEC_PER_SEC * UPDATE_INTERVAL_S;
|
||||
|
||||
printk("BOARD: %s\n", CONFIG_BOARD);
|
||||
printk("CONFIG_SMP: %s\n", IS_ENABLED(CONFIG_SMP) ? "y" : "n");
|
||||
printk("NUM_THREADS: %u\n", NUM_THREADS);
|
||||
printk("TEST_NUM_CPUS: %u\n", CONFIG_TEST_NUM_CPUS);
|
||||
printk("TEST_DURATION_S: %u\n", CONFIG_TEST_DURATION_S);
|
||||
printk("TEST_DELAY_US: %u\n", CONFIG_TEST_DELAY_US);
|
||||
|
||||
for (i = 0; i < NUM_THREADS; ++i) {
|
||||
/* spawn thread i */
|
||||
prev_counters[i] = 0;
|
||||
ret = create(i);
|
||||
if (IS_ENABLED(CONFIG_TEST_EXTRA_ASSERTIONS)) {
|
||||
zassert_ok(ret, "%s_create(%d)[%zu] failed: %d", tag, i, counters[i], ret);
|
||||
}
|
||||
}
|
||||
|
||||
do {
|
||||
if (!IS_ENABLED(CONFIG_SMP)) {
|
||||
/* allow the test thread to be swapped-out */
|
||||
k_yield();
|
||||
}
|
||||
|
||||
for (i = 0; i < NUM_THREADS; ++i) {
|
||||
if (alive[i]) {
|
||||
ret = join(i);
|
||||
if (IS_ENABLED(CONFIG_TEST_EXTRA_ASSERTIONS)) {
|
||||
zassert_ok(ret, "%s_join(%d)[%zu] failed: %d", tag, i,
|
||||
counters[i], ret);
|
||||
}
|
||||
alive[i] = false;
|
||||
|
||||
/* update counter i after each (create,join) pair */
|
||||
++counters[i];
|
||||
|
||||
if (IS_ENABLED(CONFIG_TEST_DELAY_US)) {
|
||||
/* success with 0 delay means we are ~raceless */
|
||||
k_busy_wait(CONFIG_TEST_DELAY_US);
|
||||
}
|
||||
|
||||
/* re-spawn thread i */
|
||||
ret = create(i);
|
||||
if (IS_ENABLED(CONFIG_TEST_EXTRA_ASSERTIONS)) {
|
||||
zassert_ok(ret, "%s_create(%d)[%zu] failed: %d", tag, i,
|
||||
counters[i], ret);
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
/* are we there yet? */
|
||||
now_ms = k_uptime_get();
|
||||
|
||||
/* dump some stats periodically */
|
||||
if (now_ms > update_ms) {
|
||||
update_ms += MSEC_PER_SEC * UPDATE_INTERVAL_S;
|
||||
|
||||
/* at this point, we should have seen many context switches */
|
||||
for (i = 0; i < NUM_THREADS; ++i) {
|
||||
if (IS_ENABLED(CONFIG_TEST_EXTRA_ASSERTIONS)) {
|
||||
zassert_true(counters[i] > 0, "%s %d was never scheduled",
|
||||
tag, i);
|
||||
}
|
||||
}
|
||||
|
||||
print_stats(now_ms, end_ms);
|
||||
}
|
||||
Z_SPIN_DELAY(100);
|
||||
} while (end_ms > now_ms);
|
||||
|
||||
print_stats(now_ms, end_ms);
|
||||
}
|
||||
|
||||
/*
|
||||
* Wrappers for k_threads
|
||||
*/
|
||||
|
||||
static void k_thread_fun(void *arg1, void *arg2, void *arg3)
|
||||
{
|
||||
int i = POINTER_TO_INT(arg1);
|
||||
|
||||
alive[i] = true;
|
||||
}
|
||||
|
||||
static int k_thread_create_wrapper(int i)
|
||||
{
|
||||
k_thread_create(&k_threads[i], thread_stacks[i], STACK_SIZE, k_thread_fun,
|
||||
INT_TO_POINTER(i), NULL, NULL, K_HIGHEST_APPLICATION_THREAD_PRIO, 0,
|
||||
K_NO_WAIT);
|
||||
|
||||
return 0;
|
||||
}
|
||||
|
||||
static int k_thread_join_wrapper(int i)
|
||||
{
|
||||
return k_thread_join(&k_threads[i], K_FOREVER);
|
||||
}
|
||||
|
||||
ZTEST(pthread_pressure, test_k_thread_create_join)
|
||||
{
|
||||
if (IS_ENABLED(CONFIG_TEST_KTHREADS)) {
|
||||
test_create_join_common("k_thread", k_thread_create_wrapper, k_thread_join_wrapper);
|
||||
} else {
|
||||
ztest_test_skip();
|
||||
}
|
||||
}
|
||||
|
||||
/*
|
||||
* Wrappers for pthreads
|
||||
*/
|
||||
|
||||
static pthread_t pthreads[NUM_THREADS];
|
||||
static pthread_attr_t pthread_attrs[NUM_THREADS];
|
||||
|
||||
static void *pthread_fun(void *arg)
|
||||
{
|
||||
k_thread_fun(arg, NULL, NULL);
|
||||
return NULL;
|
||||
}
|
||||
|
||||
static int pthread_create_wrapper(int i)
|
||||
{
|
||||
return pthread_create(&pthreads[i], &pthread_attrs[i], pthread_fun, INT_TO_POINTER(i));
|
||||
}
|
||||
|
||||
static int pthread_join_wrapper(int i)
|
||||
{
|
||||
return pthread_join(pthreads[i], NULL);
|
||||
}
|
||||
|
||||
ZTEST(pthread_pressure, test_pthread_create_join)
|
||||
{
|
||||
if (IS_ENABLED(CONFIG_TEST_PTHREADS)) {
|
||||
test_create_join_common("pthread", pthread_create_wrapper, pthread_join_wrapper);
|
||||
} else {
|
||||
ztest_test_skip();
|
||||
}
|
||||
}
|
||||
|
||||
/*
|
||||
* Test suite / fixture
|
||||
*/
|
||||
|
||||
ZTEST_SUITE(pthread_pressure, NULL, setup, before, NULL, NULL);
|
||||
|
||||
static void *setup(void)
|
||||
{
|
||||
if (IS_ENABLED(CONFIG_TEST_PTHREADS)) {
|
||||
const struct sched_param param = {
|
||||
.sched_priority = sched_get_priority_max(SCHED_FIFO),
|
||||
};
|
||||
|
||||
/* setup pthread stacks */
|
||||
for (int i = 0; i < NUM_THREADS; ++i) {
|
||||
zassert_ok(pthread_attr_init(&pthread_attrs[i]));
|
||||
zassert_ok(pthread_attr_setstack(&pthread_attrs[i], thread_stacks[i],
|
||||
STACK_SIZE));
|
||||
zassert_ok(pthread_attr_setschedpolicy(&pthread_attrs[i], SCHED_FIFO));
|
||||
zassert_ok(pthread_attr_setschedparam(&pthread_attrs[i], ¶m));
|
||||
}
|
||||
}
|
||||
|
||||
return NULL;
|
||||
}
|
||||
|
||||
static void before(void *fixture)
|
||||
{
|
||||
ARG_UNUSED(before);
|
||||
|
||||
for (int i = 0; i < NUM_THREADS; ++i) {
|
||||
counters[i] = 0;
|
||||
}
|
||||
}
|
||||
|
|
@ -1,16 +0,0 @@
|
|||
common:
|
||||
tags: posix
|
||||
min_ram: 64
|
||||
arch_exclude:
|
||||
- posix
|
||||
integration_platforms:
|
||||
- qemu_cortex_a53/qemu_cortex_a53/smp
|
||||
- qemu_riscv64/qemu_virt_riscv64/smp
|
||||
- qemu_riscv32/qemu_virt_riscv32/smp
|
||||
- qemu_x86_64
|
||||
tests:
|
||||
portability.posix.pthread_pressure:
|
||||
extra_configs:
|
||||
# Enabled for GitHub CI to reduce host scheduling noise while running
|
||||
# several concurrent Qemu processes each under stressful SMP load.
|
||||
- CONFIG_PTHREAD_CREATE_BARRIER=y
|
||||
Loading…
Add table
Add a link
Reference in a new issue