Design resource usage and H/W 8b/10b encoder status

19 views
Skip to first unread message

Mike Field

unread,
Oct 14, 2015, 3:27:56 AM10/14/15
to foss-displayport
I've got hardware 8B/10B encoding sort of working - it looks right in simulation but doesn't work on the hardware. However it works enough that I can measure the impact of moving the 8b/10b into the transceiver, and the answer is about 170 slices per pair of channels - as two channels can share the same lookup ROM)

It is also interesting that a config for four channels is only 30% bigger than that required for a single channel. A four lane H/W assisted interface is smaller than a single lane using my soft encoder.

Mike

Single lane enabled, Spartan 6 LXT
 Ch | 8b/10b |  Regs | LUTs | Slices | BRAM
----+--------+-------+------+--------+------
  1 |  SW    |   649 |  966 |    366 | 2
  1 |  HW    |   606 |  822 |    290 | 2

All four lanes enabled, Spartan 6 LXT
 Ch | 8b/10b |  Regs | LUTs | Slices | BRAM
----+--------+-------+------+--------+------
  1 |  SW    |   890 | 1437 |    654 | 2
  1 |  HW    |   646 |  924 |    326 | 2
  4 |  SW    |  1062 | 1600 |    697 | 2
  4 |  HW    |   818 | 1086 |    380 | 2

Tim Ansell

unread,
Oct 14, 2015, 7:01:59 AM10/14/15
to Mike Field, foss-displayport
For that 50% reduction in the number of slices, it seems like it is worth figuring it out the hardware path?

From a learning exercise, it is definitely good to have the software version too :-)

Tim 'mithro' Ansell



--
You received this message because you are subscribed to the Google Groups "foss-displayport" group.
To unsubscribe from this group and stop receiving emails from it, send an email to foss-displaypo...@googlegroups.com.
To post to this group, send email to foss-dis...@googlegroups.com.
To view this discussion on the web, visit https://groups.google.com/d/msgid/foss-displayport/2bc23d97-bc9c-493e-8f36-ed743c0021ec%40googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

Mike Field

unread,
Oct 27, 2015, 2:49:45 AM10/27/15
to foss-displayport
This has been actioned, and now both Artix-7 and Spartan 6 have examples that support h/w 8b/10b coding.

Mike

Tim Ansell

unread,
Oct 27, 2015, 2:50:46 AM10/27/15
to hamster, foss-displayport

Thanks for all your hard work on this Mike!

Tim 'mithro' Ansell

On 27 Oct 2015 5:49 p.m., "Mike Field" <ham...@snap.net.nz> wrote:
This has been actioned, and now both Artix-7 and Spartan 6 have examples that support h/w 8b/10b coding.

Mike

--
You received this message because you are subscribed to the Google Groups "foss-displayport" group.
To unsubscribe from this group and stop receiving emails from it, send an email to foss-displaypo...@googlegroups.com.

Tim Ansell

unread,
Oct 28, 2015, 8:49:26 AM10/28/15
to hamster, foss-displayport
Out of interest how does the Spartan-6 design compare to the Artix-7 design in "resource usage"?

Tim

Mike Field

unread,
Oct 30, 2015, 5:33:12 AM10/30/15
to foss-displayport
With H/W 8b/10b = 320 slices, two of RAMB8BWER blocks

With soft  8b/10b = 562 slices, two of RAMB8BWER blocks

So almost half the size 40% smaller....

Mike

Tim Ansell

unread,
Nov 2, 2015, 12:34:33 AM11/2/15
to hamster, foss-displayport

I was more wondering Spartan-6 verse Artix-7.

How much were you able to share much between the two implementations? I assume pretty much everything apart from the GTP configuration bits should be common?

Do you have time to do a short write up on the differences / similarities between the high speed transceivers on both architectures from implementing the DisplayPort core point of view?

Tim 'mithro' Ansell

--
You received this message because you are subscribed to the Google Groups "foss-displayport" group.
To unsubscribe from this group and stop receiving emails from it, send an email to foss-displaypo...@googlegroups.com.
Reply all
Reply to author
Forward
0 new messages