Controlling Burst Mode in software

--- Quote Start ---

I want the Nios processor to read my buffer in burst mode.

--- Quote End ---

As I mentioned in my previous response, the only way Nios can perform a burst transfer is through the data or code cache. If you enable the data cache in Nios and enable bursting for the cache and then if you access your component via a memory pointer, then Nios will read from your component using bursts. You need to specifically access the buffer through a memory pointer instead of using IORD or IOWR operations as IO operations will bypass the cache.

Tnx for ur help!!. So do I have to instantiate a DMA controller core in SOPC builder? another question : so u mean i instantiate a pointer like :"" int *p"" and assign the address of my component to the pointer ? and use something like "memcpy" to transfer the data?

Thank u very very much!

--- Quote Start ---

So do I have to instantiate a DMA controller core in SOPC builder?

--- Quote End ---

To move data directly between two slaves you can use a DMA controller. Technically, in this case it is the DMA controller doing the burst transfers, not the Nios processor. The Nios processor is only indirectly performing a burst by controlling the DMA controller.

--- Quote Start ---

so u mean i instantiate a pointer like :"" int *p"" and assign the address of my component to the pointer

--- Quote End ---

Yes, that is basically what I mean. Although, you don't actually have to use a pointer variable. Instead you can cast the base address to a pointer type which will generate more efficient code in some cases. As you suggested, you can then use things like memcpy or you can dereference the pointer directly.

If you are simply moving data around and the Nios processor does not actually need to manipulate the data at all, then using DMA will probably be the faster approach. When using DMA, the Nios processor can also continue to do other things in parallel while the DMA is working.

If you need to manipulate the data within Nios itself, relying only on the bursting of the data cache instead of DMA could be faster. Also, if performance isn't critical this is probably the easier method to implement.