High Speed Memory array copy

[VolterPhase]

Hello,

I have implemented a high-speed USART @921600 baud rate with STM32f10x @72Mhz. I want to send 3000 Bytes to a slave MCU.

In slave MCU, I used USART DMA to read MSG[300]byte data collect the data, and copy them into a larger array(Data[3000]).

My problem is that when the first 300 bytes come from the USART and a DMA interrupt occurs, I don't have enough time to copy MSG[300] bytes to my larger array Data[3000] and if I use a memory-to-memory DMA, I am worried about the variable changing when the incoming the second-300byte packet changes the previous values of MSG variable.

what is you idea?

[MasterT]

Use two buffers, ping-pong style. Small 300 bytes, or if memory is available than two big

[Peter BENSCH]

Or use Half Transfers via XferHalfCpltCallback, which is similar to the proposal of @MasterT:

In addition to the DMA interrupt for a filled buffer, there is also one for a half-filled buffer. Your main process can then store the first half of the buffer in your data array, while the second half is simultaneously filled by the DMA - and later do the same again with the second half of the buffer.

[VolterPhase]

Hi all,

I am appreciative of your responses.

I followed your advice and implemented the suggested procedure. After each transfer-completed-interrupt, I switch the DMA buffers. However, I am experiencing data loss during the next transmission because the sender transmits all 3000 bytes continuously, and during the 300 bytes interrupts I miss some of them. This version is more polished and clear.

void DMA1_Channel3_IRQHandler(void){
			

			if(TCP_Buffer_Selection==1)
			{
				TCP_Buffer_Selection=2;
				DMA_Cmd(DMA1_Channel3, DISABLE);	
				DMA_DeInit(DMA1_Channel3);
			        DMA_InitStructure.DMA_MemoryBaseAddr = (uint32_t)Buffer2;
				DMA_Init(DMA1_Channel3, &DMA_InitStructure);
				DMA_ClearITPendingBit(DMA1_IT_TC3| DMA1_IT_TE3);
				DMA_Cmd(DMA1_Channel3, ENABLE);
			}
			else
			{
				TCP_Buffer_Selection=1;
				DMA_Cmd(DMA1_Channel3, DISABLE);	
				DMA_DeInit(DMA1_Channel3);
				DMA_InitStructure.DMA_MemoryBaseAddr = (uint32_t)Buffer1;
				DMA_Init(DMA1_Channel3, &DMA_InitStructure);
				New_DMA_Packet=2;
				DMA_ClearITPendingBit(DMA1_IT_TC3| DMA1_IT_TE3);
				DMA_Cmd(DMA1_Channel3, ENABLE);
			}

}

As I know, each time that I want to change the buffer I should configure the DMA and enable it again.

Did I make a mistake?

[Tesla DeLorean]

The fastest way is not to do it...

It will likely be cheaper to implement as a scatter-gather list for the DMA to process, and leave the data in-situ

[MasterT · Accepted solution]

There is an example of using USART with DMA:

https://github.com/STMicroelectronics/STM32CubeF1/tree/master/Projects/STM32F103RB-Nucleo/Examples/UART/UART_HyperTerminal_DMA/Src

[VolterPhase]

I appreciate all of your comments. Thank you.