Merge branch 'test/fix_flaky_sched_time_test_v5.2' into 'release/v5.2'

test(freertos): Fixed flaky scheduling time test (v5.2)

See merge request espressif/esp-idf!32849
This commit is contained in:
Marius Vikhammer 2024-08-15 10:13:24 +08:00
commit a610a0e249

View File

@ -16,12 +16,26 @@
#include "unity.h"
#include "test_utils.h"
#define NUMBER_OF_ITERATIONS 10
#define NUMBER_OF_ITERATIONS 1023
static int compare_uint32(const void *a, const void *b)
{
return (*(uint32_t *)a - * (uint32_t *)b);
}
static uint32_t calculate_median(uint32_t *values, int size)
{
/* Sort the array of values containing the scheduling time of each iteration */
qsort(values, size, sizeof(uint32_t), compare_uint32);
/* Return the median value. (size is an odd number) */
return values[size / 2];
}
typedef struct {
SemaphoreHandle_t end_sema;
uint32_t before_sched;
uint32_t cycles_to_sched;
uint32_t cycles_to_sched[NUMBER_OF_ITERATIONS];
TaskHandle_t t1_handle;
} test_context_t;
@ -38,20 +52,18 @@ static void test_task_1(void *arg) {
static void test_task_2(void *arg) {
test_context_t *context = (test_context_t *)arg;
uint64_t accumulator = 0;
vTaskPrioritySet(NULL, CONFIG_UNITY_FREERTOS_PRIORITY + 1);
vTaskPrioritySet(context->t1_handle, CONFIG_UNITY_FREERTOS_PRIORITY + 1);
vPortYield();
for(int i = 0; i < NUMBER_OF_ITERATIONS; i++) {
accumulator += (esp_cpu_get_cycle_count() - context->before_sched);
for (int i = 0; i < NUMBER_OF_ITERATIONS; i++) {
context->cycles_to_sched[i] = esp_cpu_get_cycle_count() - context->before_sched;
vPortYield();
}
context->cycles_to_sched = accumulator / NUMBER_OF_ITERATIONS;
vTaskDelete(context->t1_handle);
xSemaphoreGive(context->end_sema);
vTaskDelete(context->t1_handle);
vTaskDelete(NULL);
}
@ -72,5 +84,11 @@ TEST_CASE("scheduling time test", "[freertos]")
BaseType_t result = xSemaphoreTake(context.end_sema, portMAX_DELAY);
TEST_ASSERT_EQUAL_HEX32(pdTRUE, result);
TEST_PERFORMANCE_LESS_THAN(SCHEDULING_TIME , "%"PRIu32" cycles" ,context.cycles_to_sched);
/* We calculate the median instead of the average to avoid outliers which may occur due to cache speed variabilities */
uint32_t median_cycles = calculate_median(context.cycles_to_sched, NUMBER_OF_ITERATIONS);
TEST_PERFORMANCE_LESS_THAN(SCHEDULING_TIME, "%"PRIu32" cycles", median_cycles);
/* Cleanup */
vSemaphoreDelete(context.end_sema);
}