New XMOS chips??!

Technical discussions around xCORE processors (e.g. xcore-200 & xcore.ai).
User avatar
Folknology
XCore Legend
Posts: 1274
Joined: Thu Dec 10, 2009 10:20 pm
Contact:

Post by Folknology »

On the xCORE-200 eXplorerKIT port map the links look very different compared to current port maps, any chance of explaining those, are the Xmos links themselves different in nature?

I also notice there is no MSEL on the sys header for boot selection

regards
Al


User avatar
XMatt
XCore Addict
Posts: 147
Joined: Tue Feb 23, 2010 6:55 pm

Post by XMatt »

- The links are fully compatible with the XS1-L/U/A series of devices.
- MSEL is not needed on xCORE-200, JTAG boot mode is selected via an internal register in the top level JTAG TAP.
User avatar
Folknology
XCore Legend
Posts: 1274
Joined: Thu Dec 10, 2009 10:20 pm
Contact:

Post by Folknology »

Thanks Xmatt

Jtag boot feature is nice , a few less pins to worry about.. It's good to hear that the links are L1 backward compatible, looking at the port map they appear somewhat spread out compared to how they are on the L series, what I cannot see is the chip pinout, are they closely grouped from a pin point of view for each set A,D,E,H.

Does the boot actually use QSPI to improve performance as well as overlays?

Are there any architectural changes or improvements to the way ports operate, do they have the same 4ma/8ma drive capability?

I notice there are 3 VDDIO supplies VDDIOL,R and T, I am guessing the first 2 are Left and Right tiles but T is a new 2.5v requirement, what is this used for?

*Ok found the preliminary datasheet now can prob answer these already

regards
Al
User avatar
DavidW
Active Member
Posts: 51
Joined: Mon Dec 16, 2013 3:23 pm

Post by DavidW »

I'll post up a proper thread soon but in the meantime, please enjoy the disembodied head of Mark Lippett (XMOS COO) discussing the xcore-200 technology:

https://youtu.be/_0R5VS4SugU
User avatar
XMatt
XCore Addict
Posts: 147
Joined: Tue Feb 23, 2010 6:55 pm

Post by XMatt »

Folknology wrote:Thanks Xmatt

Does the boot actually use QSPI to improve performance as well as overlays?
Yes QSPI boot, the embedded flash parts have QSPI flash internally.
User avatar
infiniteimprobability
XCore Legend
Posts: 1126
Joined: Thu May 27, 2010 10:08 am
Contact:

Post by infiniteimprobability »

new 2.5v requirement, what is this used for?
RGMII - Gig Ethernet!
bearcat
Respected Member
Posts: 283
Joined: Fri Mar 19, 2010 4:49 am

Post by bearcat »

My likes:
External power again.
No 1.8V rail for the USB part.
TQ parts.

My concerns:
To use more than 2 tiles, must use 128+ pin parts. Rules me out. 128 pin part is too big.

I noticed the speeds as listed:
XL208Speed.PNG
XL208Speed.PNG (10.93 KiB) Viewed 12346 times
XL208Speed.PNG
XL208Speed.PNG (10.93 KiB) Viewed 12346 times
So you can no longer get 125Mhz per core with only 4 active cores (500 MHz grade), as you can with the older parts?

Any details on the new instructions?

More options are better. Keep up the great designs...
srinie
XCore Addict
Posts: 158
Joined: Thu Mar 20, 2014 8:04 am

Post by srinie »

bearcat wrote: So you can no longer get 125Mhz per core with only 4 active cores (500 MHz grade), as you can with the older parts?


Any details on the new instructions?
Yes, the maximum MIPS per core is 100

Please keep an eye on the documentation updates - they are to be made available soon.
User avatar
infiniteimprobability
XCore Legend
Posts: 1126
Joined: Thu May 27, 2010 10:08 am
Contact:

Post by infiniteimprobability »

Yes, the maximum MIPS per core is 100
This is true (due to a 5 stage pipeline,up from 4) if you restrict the compiler/assembler to the XS1 instruction set.

Enabling dual issue allows 16b (short) instructions to be dual issued (this has to be explicitly done in the stream and mode change can be done at function call granularity). One lane can execute resource (I/O) or arithmetic and the other lane can do memory or arithmetic.

So at a maximum, you can chew through twice as many instructions making your 100MIPS actually 200MIPs. Of course getting to the 2x number is hard to achieve but expect 1.3-1.5x for optimised loops. The compiler is managing 1.2x average for general C code already. There will be a similar code size penalty, but with 4x the SRAM per tile and the ability to only compile certain files with dual issue, this can be mitigated.

There are a few other instructions which will boost speed too - especially load/store double (64b). This will really help on memcpy, dsp and push/pops.

As srinie says - more details coming shortly!
User avatar
Folknology
XCore Legend
Posts: 1274
Joined: Thu Dec 10, 2009 10:20 pm
Contact:

Post by Folknology »

On the new XE216-512-TQ128 chip is it possible to use USB or RMII on tile 1 and still be able to use XL3 link in 2 bit mode, likewise for XL4?
Post Reply