Cyclone V HPS bus - FPGA-to-SDRAM

[BrianSune_Froum]

Dear Intel, and all,

Brian here and there is a very puzzling documenting on:

https://www.intel.com/content/www/us/en/docs/programmable/683360/18-0/fpga-to-hps-sdram-access.html

Based on the 256 bit and # of allowed master to interact with the controller.

Why 256bit support 4 but 32bit only 1 at the same time?

Meanwhile, on the qsys design the behavior is completely reversed.

To achieve the maximum throughput I assume the 256bit is formed by the DDR3 quad DDR 32bit = 4*2*32 = 256.
Then the data either used band or page hit method to burst out.

So the 256 can split into 4x64 2x128  or 1x256 W/R calls.

Please FAE or internal staff confirm the above info is wrongly labeled?

[ShoH_Altera]

Hi @BrianSune_Froum ,

Thanks for you posting.

First, Let me confirm how FPGA-to-SDRAM bridge can be configured.

"There is a maximum of six command ports, four 64-bit read data port and four 64-bit

write data port."

These three types of ports can be combined to form multiple Avalon-MM/AXI interfaces. 

The table 5 to which you refer shows which ports and how many are required for each interface.

For example : You can configure

  1x  64bit AXI  (2:CommandPorts, 1:ReadPort, 1:WritePort)

  1x 128bit Avalon-MM Write Only (1:CommandPort, 0:ReadPort, 2:WritePorts)

  1x 128bit Avalon-MM Read Only (1:CommandPort, 2:ReadPorts, 0:WritePort)

  1x  64bit Avalon-MM Bidirectional (1xCommandPort, 1xReadPort, 1xWritePort)

The above configuration utilizes 4:CommandPorts, 4:ReadPorts, 4:WritePorts. (2:CommandPorts are un-used.)

Is there something strange or wrongly labeled?

Regards,

[BrianSune_Froum]

@ShoH_Altera 

With all due respect you had misinterpret my question.

The question is not above how to confirm.

It is above AXI3 settings the table from the document links are possibility wrongly labelled.

Perhaps you are try it using Quartus.

For AXI3 256 bit width bus how many interface can you used?

128 bit and 64 bit.

The question is still opened.

[BrianSune_Froum]

Dear Intel, and all,

Additional info:

Encourage employee @ShoH_Altera read document before laying down info that is not align with Intel pro

prosing?

https://www.intel.com/content/www/us/en/docs/programmable/683360/18-0/fpga-to-hps-sdram-access.html

[ShoH_Altera]

@BrianSune_Froum ,

Thank you for your responses and clarifications.

To understand how to interpret the numbers in the table, could you please read this section?

"12.4.4. FPGA-to-HPS SDRAM Interface" (Cyclone V HPS Technical Reference Manual)

https://www.intel.com/content/www/us/en/docs/programmable/683126/21-2/fpga-to-hps-sdram-interface-36440.html

The FPGA‑to‑HPS SDRAM bridge has 6 command ports, 4 read ports and 4 write ports.  In the FPGA‑to‑HPS SDRAM bridge, an Avalon-MM/AXI bus interface is formed by combining these ports according to the table.

: How many 256bit AXI interfaces can be generated in the FPGA‑to‑HPS SDRAM bridge?

With referring to the table, a 256bit AXI interface uses 2 command ports, 4 read ports and 4 write ports. 

Since the FPGA‑to‑HPS SDRAM interface has 6 command ports, 4 read ports and 4 write ports, you can generate ONE 256bit AXI interface.

Hope this clears up your doubts.

Regards,

[BrianSune_Froum]

@ShoH_Altera 

No you still misunderstand what i am trying to point out.

I do clearly known the HPS can only create "1" 256 bit AXI3 R/W interface based on the fixed HPS interface.

Either from the HPS Qsys or the link I enclosed from first place.

However, based on the link, the table mentioned "possible port utilization".

Why the AXI3 64bit only support 1 set?

With the above statement AXI3 allow what configurations?

What I am trying to pointed out is this information is labeled wrongly and very puzzling from first place.

[ShoH_Altera]

@BrianSune_Froum 

OK... I still don't quite understand your point. Please let me ask you some more questions...

Why do you think "the AXI3 64bit only support 1 set "?

Based on the link, the table mentioned : 64bit AXI uses "2 Command Ports, 1 Read Ports, 1 Write Ports".
The FPGA-HPS SDRAM interface has six command ports, four read ports, and four write ports, and taking the above into account, we can create three 64-bit AXI interfaces. In this case, six(=2x3) command ports, three(=1x3) read ports, and three3(=1x3) write ports are utilized.  (In this case, no more bus interfaces can be created, since all 6 command ports are used.)

This can be also confirmed with HPS Qsys as below:

Could you please elaborate a bit more on what exactly you mean by "this information is mislabeled"?

Thanks,

[BrianSune_Froum]

@ShoH_Altera 

Question 1: why it is 1 set:

Rely: based on this site the table labeled "possible port utilization" only shown 1 set of bus rather than maximum # of co-exist bus on the same settings.

https://www.intel.com/content/www/us/en/docs/programmable/683360/18-0/fpga-to-hps-sdram-access.html

The title "possible port utilization"  I am not sure this is what the table have completely shown.

i.e. only 1 set of 64bit AXI3 etc.

This is why I mentioned you miss my point on the ticket.

What those examples you shown are computed rather than follow the table info.

Which also does not mentioned via equations nor co-exist bus examples.

This is why this is so puzzling and simply using Qsys alone only shows two 256 AXI3 returns error.

Then in order to know what caused this error also confused from the site that only 1 set of AXI3 64bit possible port utilization from the table.