-
Notifications
You must be signed in to change notification settings - Fork 349
userspace: add a performance test #10563
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
base: main
Are you sure you want to change the base?
Changes from all commits
File filter
Filter by extension
Conversations
Jump to
Diff view
Diff view
There are no files selected for viewing
| Original file line number | Diff line number | Diff line change |
|---|---|---|
| @@ -0,0 +1,172 @@ | ||
| // SPDX-License-Identifier: BSD-3-Clause | ||
| /* Copyright(c) 2026 Intel Corporation. */ | ||
|
|
||
| /* Test kernel vs. user-space performance. */ | ||
|
|
||
| #include <sof/boot_test.h> | ||
| #include <rtos/alloc.h> | ||
|
|
||
| #include <zephyr/kernel.h> | ||
| #include <zephyr/ztest.h> | ||
| #include <zephyr/logging/log.h> | ||
|
|
||
| LOG_MODULE_DECLARE(sof_boot_test, LOG_LEVEL_DBG); | ||
|
|
||
| static int load_add(void) | ||
| { | ||
| #define N_ADD (1000 * 1000 * 100) | ||
| unsigned long r = 0; | ||
|
|
||
| for (unsigned int i = 0; i < N_ADD; i++) | ||
| r += i; | ||
| #define N_DIV 10000 | ||
| for (unsigned int i = 1; i <= N_DIV; i++) | ||
| r = r / (i % 10 + 1) * (i % 10 + 3); | ||
| return (int)r; | ||
| } | ||
|
|
||
| #ifdef __XCC__ | ||
| #include <xtensa/tie/xt_hifi4.h> | ||
|
|
||
| /* Compute dot product of two vectors using HiFi4 SIMD instructions */ | ||
| static int32_t dot_product_hifi4(const int16_t *a, const int16_t *b, int length) | ||
| { | ||
| ae_int64 acc = AE_ZERO64(); /* 1. Initialize accumulator to zero */ | ||
| ae_int16x4 *pa = (ae_int16x4 *)a; /* Pointer to vector a */ | ||
| ae_int16x4 *pb = (ae_int16x4 *)b; /* Pointer to vector b */ | ||
|
|
||
| for (int i = 0; i < length / 4; i++) { | ||
| ae_int16x4 va, vb; | ||
|
|
||
| AE_L16X4_IP(va, pa, 8); /* 2. Load 4x 16-bit values from a */ | ||
| AE_L16X4_IP(vb, pb, 8); /* 3. Load 4x 16-bit values from b */ | ||
| AE_MULAAAAQ16(acc, va, vb); /* 4. Multiply-accumulate (4 MACs in parallel) */ | ||
| } | ||
|
|
||
| return AE_TRUNCA32F64S(acc, 0); /* 5. Convert 64-bit result to 32-bit */ | ||
| } | ||
|
|
||
| #define VECTOR_LENGTH 100 | ||
| static int load_hifi4(void) | ||
| { | ||
| uint16_t a[VECTOR_LENGTH], b[VECTOR_LENGTH]; | ||
| int ret = 0; | ||
|
|
||
| for (unsigned int j = 0; j < 1000; j++) { | ||
| for (unsigned int i = 0; i < VECTOR_LENGTH; i++) { | ||
| a[i] = i * 3 - 47 * j; | ||
| b[i] = 411 * j - i * 5; | ||
| } | ||
|
|
||
| ret += dot_product_hifi4(a, b, VECTOR_LENGTH); | ||
| } | ||
| return ret; | ||
| } | ||
| #endif /* __XCC__ */ | ||
|
|
||
| typedef int (*load_fn_t)(void); | ||
|
|
||
| load_fn_t load_fn[] = { | ||
| load_add, | ||
| #ifdef __XCC__ | ||
| load_hifi4, | ||
| #endif | ||
| }; | ||
|
|
||
| static unsigned int test_perf(load_fn_t fn, struct k_event *event, | ||
| struct k_sem *sem) | ||
| { | ||
| uint64_t start = k_uptime_ticks(); | ||
|
|
||
| k_event_set(event, (uint32_t)fn); | ||
|
|
||
| int ret = k_sem_take(sem, K_MSEC(200)); | ||
|
|
||
| zassert_ok(ret); | ||
|
|
||
| uint64_t end = k_uptime_ticks(); | ||
|
|
||
| return (unsigned int)(end - start); | ||
| } | ||
|
|
||
| static void thread_fn(void *p1, void *p2, void *p3) | ||
| { | ||
| struct k_event *event = p1; | ||
| struct k_sem *sem = p2; | ||
| bool first = true; | ||
|
|
||
| for (;;) { | ||
| load_fn_t fn = (load_fn_t)k_event_wait(event, 0xffffffff, !first, K_FOREVER); | ||
|
|
||
| first = false; | ||
| LOG_INF("fn %p ret %d", (void *)fn, fn()); | ||
|
|
||
| k_sem_give(sem); | ||
| } | ||
| } | ||
|
|
||
| #define STACK_SIZE 4096 | ||
|
|
||
| ZTEST(sof_boot, test_perf) | ||
| { | ||
| /* Synchronization objects allocated on original uncached heap */ | ||
| struct k_event *u_event = k_object_alloc(K_OBJ_EVENT); | ||
| struct k_event *k_event = k_object_alloc(K_OBJ_EVENT); | ||
|
|
||
| zassert_not_null(u_event); | ||
| zassert_not_null(k_event); | ||
|
|
||
| k_event_init(u_event); | ||
| k_event_init(k_event); | ||
|
|
||
| struct k_sem *sem = k_object_alloc(K_OBJ_SEM); | ||
|
|
||
| zassert_not_null(sem); | ||
| k_sem_init(sem, 0, 1); | ||
|
|
||
| /* Allocate kernel stack and thread and start it */ | ||
| struct k_thread *k_thread = k_object_alloc(K_OBJ_THREAD); | ||
|
|
||
| zassert_not_null(k_thread); | ||
| /* Important: Xtensa thread initialization code checks certain fields for 0 */ | ||
| memset(&k_thread->arch, 0, sizeof(k_thread->arch)); | ||
|
|
||
| k_thread_stack_t *k_stack = k_thread_stack_alloc(STACK_SIZE, 0); | ||
|
|
||
| zassert_not_null(k_stack); | ||
|
|
||
| struct k_thread *pk_thread = k_thread_create(k_thread, k_stack, STACK_SIZE, thread_fn, | ||
| k_event, sem, NULL, 0, 0, K_FOREVER); | ||
|
|
||
| k_thread_start(pk_thread); | ||
|
|
||
| /* Allocate userspace stack and thread and start it */ | ||
| struct k_thread *u_thread = k_object_alloc(K_OBJ_THREAD); | ||
|
|
||
| zassert_not_null(u_thread); | ||
| memset(&u_thread->arch, 0, sizeof(u_thread->arch)); | ||
|
|
||
| k_thread_stack_t *u_stack = k_thread_stack_alloc(STACK_SIZE, K_USER); | ||
|
|
||
| zassert_not_null(u_stack); | ||
|
|
||
| struct k_thread *pu_thread = k_thread_create(u_thread, u_stack, STACK_SIZE, thread_fn, | ||
| u_event, sem, NULL, 0, K_USER, K_FOREVER); | ||
|
|
||
| zassert_not_null(pu_thread); | ||
| k_thread_access_grant(pu_thread, u_event, sem); | ||
| k_thread_start(pu_thread); | ||
|
|
||
| for (unsigned int i = 0; i < ARRAY_SIZE(load_fn); i++) { | ||
|
Contributor
There was a problem hiding this comment. Choose a reason for hiding this commentThe reason will be displayed to describe this comment to others. Learn more. Did you see any meaningful difference between kernel space and user space in respect to the two load types (hifi4 vs regular code)? I would imagine that a given load type is more or less constant for both user thread and kernel thread.
Collaborator
Author
There was a problem hiding this comment. Choose a reason for hiding this commentThe reason will be displayed to describe this comment to others. Learn more.
@wjablon1 no, after we've fixed the uncached stack bug we didn't measure any difference any more. You certainly will have a difference in cumulative CPU load when executing a significant number of system calls, but that overhead will depend strongly exactly on those syscalls, so it's hard to measure a "typical" overhead |
||
| LOG_INF("user: fn %p took %u", load_fn[i], test_perf(load_fn[i], u_event, sem)); | ||
| LOG_INF("kernel: fn %p took %u", load_fn[i], test_perf(load_fn[i], k_event, sem)); | ||
| } | ||
|
|
||
| k_thread_abort(pu_thread); | ||
| k_thread_stack_free(u_stack); | ||
| k_thread_abort(pk_thread); | ||
| k_thread_stack_free(k_stack); | ||
| k_object_free(sem); | ||
| k_object_free(u_event); | ||
| k_object_free(k_event); | ||
| } | ||
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
That is some serious event overuse to pass a value here 😆. You might also try
k_poll_signal_raise(), it can pass a value too. The only catch is that we don't have enabledk_poll, so it won't help. I guess this is just a quick draft for tests, right?There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
@softwarecki well, I do think it would be good to have it merged. "overuse" - well, maybe indeed this isn't how events are intended to be used, but I'd call it a "creative use" ;-)
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
This would be the perfect place to use RTIO! Just kidding 😄 I'd suggest using something actually meant for passing data, maybe a FIFO or a queue? https://docs.zephyrproject.org/latest/kernel/services/index.html#data-passing