How to config DMA2D in stm32mp13 linux

[inman]

Hi Dear Sir,

I am using linux6.6.48 (with st patch)based on stm32mp135d,I trying to setup DMA2D but cannot find correct dts setting. and I can see the dma2d driver is supported in this kernel. but how to config dma2d in dts ?

please help. thank you very much.

[inman]

Hi suppose this should be in the software topic. but it should be ok here for the DMA2D in the MP1.

[PatrickF · Accepted solution]

Hi,

There is no DMA2D inside STM32MP13x family.
https://www.st.com/en/microcontrollers-microprocessors/stm32mp135d.html

For which purposes are you willing to use DMA2D ?
if you are meaning DMA, please have a look to https://wiki.st.com/stm32mpu

Regards

[inman]

I want to use the DMA2D convernt RGB565 to RGB888 at camera application. is there any good solution for this? my applcation is using OPENCV for camera usage, but the opencv does not accept RGB565 image, I can use NEON convert RGB565 but it is not good. 

[PatrickF]

Hi @inman 

I'm not expert on this.
I could guess the RGB565 to RGB888 conversion for image analysis could be done mostly by moving/copying bits (no math).
Did you investigates the possible camera sensor settings + DCMIPP settings together with adequate wiring connection to DCMIPP to get RGB888 directly (might impact frame buffer size and frame rate) ?

If you are willing to get more performances, STM32MP25 family could offer more features around this (as far as I understand, its DCMIPP could get sensor RGB565 and output as RGB888 to processing pipe)

Regards.

[inman]

Hi dear sir,

is there pin compatiable chip i can use the dma2d or gpu to do image convesion? i have pcb out the board with stm32mp135d, it is hard to change other chip for re-design.

[inman]

I use DVP to connect external camera, so the Directly RGB888 cannot be got from camera.

[PatrickF]

Hi @inman 

unfortunately, there is no devices in the STM32MP13 family with DMA2D.
As it is for image processing, where quality is less important, you could look a simple conversion by bit shifting and MSB replication to form the RGB888 word, using Cortex-A7 for this could be efficient (image size should remain small).

Regards,