cancel
Showing results for 
Show  only  | Search instead for 
Did you mean: 
drsys
Visitor
Visitor
612 Views
Registered: ‎05-17-2018

AXI DMA IP Core upgrade

Jump to solution

Dear all,

I have a working Zynq-7000 device running Linux designed and programmed by me. The design needs to be upgraded. The bottle neck of the design is an AXI DMA IP core, which can handle only 8 Mb of data per time. Are there any other IP cores which can handle more data per single burst on the market? Other solutions also welcome especially if they will not require much modifications in the software.

Thank you in advance!

0 Kudos
1 Solution

Accepted Solutions
dgisselq
Scholar
Scholar
552 Views
Registered: ‎05-21-2015

32-data bits on the downstream bus seems kind of narrow.  Is that the biggest path you have to memory?  Most DDR3 SDRAMs I know of have at least a 64-bit path to memory.

That said, I do know of an open source S2MM design you are welcome to try.  Unlike the datamover IP, this one achieves a full 100% bus throughput.  Sadly, whereas the data mover IP can handle unaligned transactions, this one can only handle aligned transactions, so you get tradeoffs both ways.

Dan

View solution in original post

0 Kudos
4 Replies
dgisselq
Scholar
Scholar
578 Views
Registered: ‎05-21-2015

Can you tell me which DMA core you are using?

Dan

0 Kudos
drsys
Visitor
Visitor
560 Views
Registered: ‎05-17-2018

Hello Den,

 

I am using LogiCORE IP AXI DMA v7.1 which comes with Vivado 2017.4. The documentation says that the width of S2MM_LENGTH is 26 bits (64 Mb) but Vivado does not allow me to set this parameter more than 23 bit (see attached screenshot). Is it 26 bit in a new versions of Vivado?

 

Misha.

vivado.png

0 Kudos
dgisselq
Scholar
Scholar
553 Views
Registered: ‎05-21-2015

32-data bits on the downstream bus seems kind of narrow.  Is that the biggest path you have to memory?  Most DDR3 SDRAMs I know of have at least a 64-bit path to memory.

That said, I do know of an open source S2MM design you are welcome to try.  Unlike the datamover IP, this one achieves a full 100% bus throughput.  Sadly, whereas the data mover IP can handle unaligned transactions, this one can only handle aligned transactions, so you get tradeoffs both ways.

Dan

View solution in original post

0 Kudos
drsys
Visitor
Visitor
539 Views
Registered: ‎05-17-2018

Hello Den,

 

Thank you very much! Performance of current DMA IP is enough for my demands but alightment of transfers requires care.

 

Misha.

0 Kudos