CPU utilization #208
CPU utilization #208
Conversation
| // Can ccheck by creating vApplicationTickHook and comparing # tick and | ||
| // # idle cycles << todo | ||
|
|
||
| if (xPortGetCoreID() == 0) { |
There was a problem hiding this comment.
I think it'd be cleaner to make the variable an array and index into it with the output of the function to increment the variable. Just make sure it always succeeds or else you need to add error checking.
| // For ISR to handle (?) | ||
| BaseType_t xHigherPriorityTaskWoken = pdFALSE; | ||
|
|
||
| // Which core? (assumption 1 cycle == 1 tick) |
There was a problem hiding this comment.
The FreeRTOS tick is triggered by some kind of SysTick timer (in our case I believe it's TIMER0.
| #define TIMER_DIVIDER 80 // Divide Base by 80 | ||
|
|
||
|
|
||
| // Timer group 0, timer 0 configuration |
There was a problem hiding this comment.
See how these comments are redundant? Even the last field wouldn't need a comment if you use a descriptive constant name.
|
|
||
| #include <stdio.h> | ||
|
|
||
| #define TIMER_DIVIDER 80 // Divide Base by 80 |
There was a problem hiding this comment.
Comment is self-evident, would you mind #defineing constants for an expression as to how you arrived to this value?
|
|
||
| // Let run for 5 sec | ||
| // Interrupts can be called here to run other tasks | ||
| // What if interrupt task runs for more than 5 secs? |
| // Calculate total ticks/cycles in 1 sec | ||
| uint64_t ticks_elapsed = end_time - start_time; | ||
|
|
||
| float core0_utilization = 100.0 |
There was a problem hiding this comment.
I think you should use ticks to calculate this over cpu cycles since that's much harder to keep track of. There is a FreeRTOS config #define that sets the SysTick frequency in Hz. ESP-IDF is quirky and sets this in the sdkconfig and I think it generates the config for their custom fork automatically. You can either look at the file manually or run the menuconfig target to see this.
| printf("Core 1 CPU Utilization: %.2f\n", core1_utilization); | ||
|
|
||
| // Reset idle cycles | ||
| core0_idle_cycles = 0; |
There was a problem hiding this comment.
What's the reason for discarding these? Ideally we keep a global running count and a sliding window which we would average so that we have current stats and stats for the duration of the power cycle.
| .intr_type = TIMER_INTR_DISABLE, // No timer interrupts | ||
| .counter_dir = TIMER_COUNT_UP, // Count up | ||
| .auto_reload = TIMER_AUTORELOAD_EN, // Auto-reload | ||
| .divider = TIMER_DIVIDER // Set divider for 1ms (assumption that base |
There was a problem hiding this comment.
If you do expand this into a proper #define you wouldn't need to make this assumption! FreeRTOS should have a config #define that has the CPU frequency.
| uint64_t ticks_elapsed = end_time - start_time; | ||
|
|
||
| float core0_utilization = 100.0 | ||
| - ((float) core0_idle_cycles / ticks_elapsed) * 100.0; |
There was a problem hiding this comment.
Let's also try to keep the CPU utilization similar to how Linux reports it (as a float between 0 and 1 per core).
Added method to measure and print cpu utilization