Design Con 2015
Breaking News
News & Analysis

ARM Unveils New Architecture for Automotive

10/23/2013 07:55 AM EDT
17 comments
NO RATINGS
2 saves
< Previous Page 2 / 2
More Related Links
View Comments: Newest First | Oldest First | Threaded View
Page 1 / 2   >   >>
Wilco1
User Rank
CEO
Re: Automotive Hypervisors for Safe Add-ons
Wilco1   10/24/2013 8:50:01 AM
NO RATINGS
Well I'd be surprised if those PowerPCs had an interrupt latency of ~10 cycles. Cortex-R cores are used in hard realtime tasks like flying the heads of a HD. Typically when you have caches latencies go up to hundreds of cycles. This is due to having many outstanding cache misses and writebacks in flight - you can't just drop those on an interrupt as you end up in an inconsistent state...

All new cores like Cortex-A7, A12 and A15 have division. Cortex-A8/A9 don't have divide as division was introduced well after they were finished. I agree it's not ideal but then again around 10 years ago, when I talked to an ARM hardware guy and asked how fast he could implement division on something like ARM11, his answer was "about 40 cycles". I then said "well my software division easily beats that with an average time below 30 cycles, so don't even bother". It seems that after the introduction of Cortex-M3 with its amazingly fast division hardware designers made division fast enough on higher-end cores as well!

 

Bastian.Schick
User Rank
Rookie
Re: Automotive Hypervisors for Safe Add-ons
Bastian.Schick   10/24/2013 8:25:42 AM
NO RATINGS
The issue with running an RTOS on an -A core is that interrupt latency is much higher on fast OoO cores with large caches. So you lose the R in RTOS. The -R cores have special features to keep interrupt latency low (although to get the lowest interrupt latency you have to use an -M core).

Actually, I'd say no. There are a lot of CPUs arround with cache which are used in realtime enviroments (like all the automotive PowerPCs).

All Cortex-A cores since A15 do have integer divide. The latest QC and Apple cores do have divide as well.

The thing is, Cortex-R4 has it, but e.g. ZYNQ Cortex-A9 does not have it as it is "optional".

Encoding T1 ARMv7-R, ARMv7VE, otherwise OPTIONAL in ARMv7-A
SDIV<c> <Rd>, <Rn>, <Rm>


and


In an ARMv7-A implementation that does not include the Virtualization Extensions, the implementation of SDIV
and UDIV in both instruction sets is OPTIONAL, but the architecture permits an ARMv7-A implementation to not
implement SDIV and UDIV.


So what? I really wonder, what goes on in the brains of architecture developers (and as it is optional: implementers) :-(

Wilco1
User Rank
CEO
Re: Automotive Hypervisors for Safe Add-ons
Wilco1   10/24/2013 8:25:16 AM
NO RATINGS
It's very simple: v8-R supports both VMSA and PMSA, so it has an MPU and MMU. Otherwise you wouldn't be able to run standard Linux.

Wilco1
User Rank
CEO
Re: Automotive Hypervisors for Safe Add-ons
Wilco1   10/24/2013 8:03:05 AM
NO RATINGS
The R variant of the architecture has always been something of a middle ground, with some -M features and some -A features. That's just what the RTOS people seem to need.

If you want 32 registers then you do want 64-bit but use 32-bit pointers to avoid the overhead of 64-bit pointers. For x64 there is already a Linux ABI with 32-bit pointers, so I bet there will be one for ARM64 as well.

The issue with running an RTOS on an -A core is that interrupt latency is much higher on fast OoO cores with large caches. So you lose the R in RTOS. The -R cores have special features to keep interrupt latency low (although to get the lowest interrupt latency you have to use an -M core).

All Cortex-A cores since A15 do have integer divide. The latest QC and Apple cores do have divide as well.

 

 

Denis.Giri
User Rank
Manager
Re: Automotive Hypervisors for Safe Add-ons
Denis.Giri   10/24/2013 3:57:00 AM
NO RATINGS
Honestly, I'd like to see more information on how they deal with the virtual adressing issues of the richOS...

VMSA usually assumes a MMU, but ARMv8-R apparently only has a MPU (PMSA instead of VMSA)

Bastian.Schick
User Rank
Rookie
Re: Automotive Hypervisors for Safe Add-ons
Bastian.Schick   10/24/2013 1:51:00 AM
NO RATINGS
Ok, reading this document shows really some improvements over v7R.
But it again makes me wonder why ARM differs -A and -R but now they mix -A and -R with the v8-R only that there is no 64bit support (which I do not miss) and no support for the new ISA with 32 registers (which I miss).
Granted, todays feature rich (overloaded?) OSes will not run on a R or M CPU, but on the other hand, an RTOS very well runs on an A one.

I really wonder: Do the ARM ISA developers talk to (RT)OS designers asking what they think an ISA needs? I mean, how come an Cortex-A CPU does not have integer divide? (Ok, now going off-topic...)

So lets wait for the first dual-core v8-R (no more Cortex I guess) CPU.



junko.yoshida
User Rank
Blogger
Re: Licensees?
junko.yoshida   10/24/2013 12:27:03 AM
NO RATINGS
From what I understand, Renesas does use ARM cores more often in multimedia applications -- SoCs for in-vehicle infotainment , or Advanced Driver Assistant Systems.

rick merritt
User Rank
Author
Re: Licensees?
rick merritt   10/23/2013 9:34:17 PM
NO RATINGS
1 saves
Good point, Sanjaac. Renesas is playing both sides of the fence although I believe its propritary MCU architecture is one of the most widely used.

sanjaac
User Rank
Rookie
Re: Licensees?
sanjaac   10/23/2013 1:46:19 PM
NO RATINGS
Rick,

Maybe you are not well informed, which would be really surprising, but Renesas has long been an ARM licensee in the Automotive -namely Infotaiment segment- market, with R-Car family (http://am.renesas.com/applications/automotive/cis/cis_highend/index.jsp). A quick excerpt, excluding SuperH-only devices:
  • R-Car E1: ARM® Cortex™-A9 533 MHz, 1330 DMIPS
  • R-Car M1A/S: ARM® Cortex™-A9 800 MHz + SH-4A core 800 MHz (M1A), 3760 DMIPS
  • R-Car H1: ARM® Cortex™-A9 1 GHz x 4 + SH-4A core 11760DMIPS
  • R-Car M2: ARM® Cortex™-A15 × 2 + SH-4A core over 12000DMIPS
  • R-Car H2: ARM® Cortex™-A15 × 4 + Cortex™-A7 × 4 + SH-4A core over 25,000 DMIPS

A few more details of the R-Car H2: http://am.renesas.com/press/news/2013/news20130325.jsp

So, no really clear where is the split you mention.

Or is it maybe, because some MCU vendors - Renesas for instance- also have automotive devices, more dedicated for Real Time tasks, like the RH850 (http://am.renesas.com/products/mpumcu/rh850/index.jsp?campaign=gn_prod) or V850 (http://am.renesas.com/products/mpumcu/v850/sub/automotive.jsp) families, among others, and now with ARM's announcement we see a marriage of RealTime and Non-RealTime tasks on a same device?

But in any case, it is not an ARM or not-ARM split, at least not in Renesas case, neither in Industrial nor in Automotive fields.

Cheers.

Wilco1
User Rank
CEO
Re: Automotive Hypervisors for Safe Add-ons
Wilco1   10/23/2013 1:10:14 PM
NO RATINGS
This is the v8 version of the v7-R architecture. So it is indeed compatible with existing cores like Cortex-R4 but adds new features as well, such as the Hypervisor and support for running Linux. Note v8-R is not 64-bit.

See http://arm.com/files/pdf/ARMv8R__Architecture_Oct13.pdf for all the details.

Page 1 / 2   >   >>
Radio
LATEST ARCHIVED BROADCAST
EE Times Senior Technical Editor Martin Rowe will interview EMC engineer Kenneth Wyatt.
Like Us on Facebook

Datasheets.com Parts Search

185 million searchable parts
(please enter a part number or hit search to begin)
EE Times on Twitter
EE Times Twitter Feed
Flash Poll