I'm using the serial library for the first time. What are the advantages/disadvantages of using IRQ and DMA? How should FIFO queue and buffer sizes be chosen?
I'm using the serial library for the first time. What are the advantages/disadvantages of using IRQ and DMA? How should FIFO queue and buffer sizes be chosen?
Hi,
What are the advantages/disadvantages of using IRQ and DMA?
How should FIFO queue and buffer sizes be chosen?
Please see the answer in this post.
-Amanda H.
What is the advantage of using EasyDMA? I understand that direct memory access means memory copying is not required. But nrf_serial_read copies the data anyway. So is there any advantage of using EasyDMA? I assume the RX buffer could be read directly instead of using nrf_serial_read (but what function should then be called to mark the data as processed?), but if the size is 1 then there is no efficiency improvements really, the data would need copying anyway and function calls would be a greater overhead.
In terms of buffer sizes – I understand that the buffer size is the minimum number of bytes that can be sent or received in one shot, from your serial library docs: "They are the smallest slices of data that can be transfered in a single UART driver - legacy layer driver request". So in terms of TX, it's the minimum data length I would want to send? And RX is the minimum data I expect to receive? So for most protocols, I imagine this would be problematic if set to anything other than 1, as the data may be corrupted or missing bytes, or may not fit neatly into fixed size buffers. Unless you can dynamically set the expected rx size?
FIFO sizes – I assume these are queues of pointers to data passed into nrf_serial_write, I'm not sure what the queue is for when reading...where is the data stored for the queue of pointers waiting to be processed?
These could do with better explanation in the docs.
What is the advantage of using EasyDMA? I understand that direct memory access means memory copying is not required. But nrf_serial_read copies the data anyway. So is there any advantage of using EasyDMA? I assume the RX buffer could be read directly instead of using nrf_serial_read (but what function should then be called to mark the data as processed?), but if the size is 1 then there is no efficiency improvements really, the data would need copying anyway and function calls would be a greater overhead.
In terms of buffer sizes – I understand that the buffer size is the minimum number of bytes that can be sent or received in one shot, from your serial library docs: "They are the smallest slices of data that can be transfered in a single UART driver - legacy layer driver request". So in terms of TX, it's the minimum data length I would want to send? And RX is the minimum data I expect to receive? So for most protocols, I imagine this would be problematic if set to anything other than 1, as the data may be corrupted or missing bytes, or may not fit neatly into fixed size buffers. Unless you can dynamically set the expected rx size?
FIFO sizes – I assume these are queues of pointers to data passed into nrf_serial_write, I'm not sure what the queue is for when reading...where is the data stored for the queue of pointers waiting to be processed?
These could do with better explanation in the docs.
Hi,
nrbrook said:is there any advantage of using EasyDMA?
EasyDMA data can transfer from/to RAM directly.
The Serial port library allows to create and handle serial port instances. In the Serial Port Library Example project code, you could see the Serial port library connect with UART. Serial Port Library Example with two UARTEs shows how to use the Serial port library with two UARTEs. The DMA operation is implemented by the UARTE driver.
nrbrook said:So in terms of TX, it's the minimum data length I would want to send? And RX is the minimum data I expect to receive?
Yes, they are the smallest transfer data slice used by nrf_drv_uart_tx and nrf_drv_uart_rx.
nrbrook said:where is the data stored for the queue of pointers waiting to be processed?
In your Serial Port Library Example with two UARTEs example you copy each byte out one at a time. I suppose as you are using the timeout parameter the DMA improves performance slightly by writing directly to the pointer provided. However my application is a bluetooth application and so I will not be using the timeout parameter as I need serial communication to be asynchronous. In this scenario, is using IRQ effectively the same as using DMA? In both, I get an interrupt and then copy out a single byte, so I don't understand how DMA can improve performance.
Hi Nick,
nrbrook said:However my application is a bluetooth application
What is your application? The UART/Serial Port Emulation over BLE example might help.
-Amanda H.
The serial comm is not over BLE, it is with another micro, using the MIN protocol. That example just receives one character at once as your other examples do.
Hi Nick,
nrbrook said:In this scenario, is using IRQ effectively the same as using DMA?
Yes, it is essentially the same under this situation.
-Amanda H.