RealView or Green Hills for ARM? Opinions?

I am getting ready to begin my first project using an ARM core.  We are 
porting embedded software from an existing product that uses a Motorola 
56311 DSP to a new design that uses an ARM 966E-S based ASIC.  I am in 
the process of choosing the compiler/toolset that we will use for this 
project.  I have narrowed my list down to the RealView Developer Suite 
from ARM and MULTI from Green Hills Software.

I have looked at RVDS briefly and I was not impressed.  The IDE seemed 
like an unfinised product.  If I choose RVDS I would probably dump the 
IDE and use CodeWright and makefiles.  The evaluation CD for MULI is on 
the way.  

I am most concerned about the effieciency of the compiler and the 
quality of the IDE.  ARM and Green Hills each say they have the most 
efficient compiler (what did I expect to hear).

Is there anyone out there who has experience with these tools that is 
willing to share some of what they have learned?
  

Thanks,

Mike

Mike Harvey <Mike.h@flashmail.com> wrote in message news:<Xns947E73FDBA7DBmikehflashmailcom@130.133.1.4>...
> I am getting ready to begin my first project using an ARM core.  We are 
> porting embedded software from an existing product that uses a Motorola 
> 56311 DSP to a new design that uses an ARM 966E-S based ASIC.  I am in 
> the process of choosing the compiler/toolset that we will use for this 
> project.  I have narrowed my list down to the RealView Developer Suite 
> from ARM and MULTI from Green Hills Software.
> 
> I have looked at RVDS briefly and I was not impressed.  The IDE seemed 
> like an unfinised product.  If I choose RVDS I would probably dump the 
> IDE and use CodeWright and makefiles.  The evaluation CD for MULI is on 
> the way.  
> 
> I am most concerned about the effieciency of the compiler and the 
> quality of the IDE.  ARM and Green Hills each say they have the most 
> efficient compiler (what did I expect to hear).
> 
> Is there anyone out there who has experience with these tools that is 
> willing to share some of what they have learned?
>   
> 
> Thanks,
> 
> Mike


Hi Mike,

depending on your estimated program size and planned usage of the
Thumb mode, I would like to add a third option to your compiler list.
The IAR compiler is VERY good with Thumb and can compete with both
Green Hills and ARM regarding speed. We used evaluation versions of
all three and found the most significant difference in the compactness
of the IAR Thumb code. If code size is of maximum concern for you, it
might be a good idea to check the IAR as well.

Cheers, Schwob

It's true, IAR EWARM 4.10A can beat both ARM and Greenhills compilers
for THUMB code. Sadly THUMB code is geting less common with all this
wide/cheap/integrated memory around!

For ARM cored ASICs I personally would go with ARM factory tools -as
if it does go wrong you want to be as near the core vendor as
possible.

But for LPC2106 and the new baby ARM7 generic microcontrollers (NO OS,
NO TCP/IP) due in 2004 I would back IAR every time.

Jason Moore
IAR UK

"Jason Moore" <jason_ceng@yahoo.co.uk> wrote in message
news:335ca3c9.0403021511.713bb8c0@posting.google.com...
> It's true, IAR EWARM 4.10A can beat both ARM and Greenhills compilers
> for THUMB code. Sadly THUMB code is geting less common with all this
> wide/cheap/integrated memory around!

Well... I'd have to disagree (in part) - I have had more than my fair share
rootling around in Thumb compiler-generated assembly code over the past
couple of years, I think Thumb isn't going to go away - the trick will be to
get the best performance/code size balance out of Thumb code ;) Thumb can
also give surprising benefits in a cached or memory prefetch system when you
consider you can potentially fit almost twice as much code in a cache line.

If you're that confident about the IAR Thumb code-density, I'll have to give
them a test drive - on high volume products, memory is always at a premium
:)

Peter.

"CodeSprite" <pmaloy@codesprite.com> wrote in message
news:40467261$0$4881$812600b3@news.nntpaccess.com...
> "Jason Moore" <jason_ceng@yahoo.co.uk> wrote in message
> news:335ca3c9.0403021511.713bb8c0@posting.google.com...
> > It's true, IAR EWARM 4.10A can beat both ARM and Greenhills compilers
> > for THUMB code.

If most of your ROM consists of C libraries, yes. The benchmark
comparison on iar.com mainly shows that different toolkits use different
library strategies (small and slow or large and fast - both strategies
have advantages and disadvantages). However for serious codesize
benchmarking I normally use applications 2 to 3 orders of magnitude
larger.

>>Sadly THUMB code is geting less common with all this
> > wide/cheap/integrated memory around!
>
> Well... I'd have to disagree (in part) - I have had more than my fair
share
> rootling around in Thumb compiler-generated assembly code over the past
> couple of years, I think Thumb isn't going to go away -

Yes, Thumb only has a performance advantage over ARM when used
with 16-bit wide instruction memory, but this doesn't mean Thumb
isn't useful when wider memory is used. Thumb effectively makes
your flash, I-cache or on-chip memory look 50% larger (at 66% of
ARM size), and that is a saving well worth having. Also the saving in
compulsory and capacity cache misses reduces the performance
difference between ARM and Thumb.

> the trick will be to
> get the best performance/code size balance out of Thumb code ;)

That trick is called Thumb-2: ~ Thumb size with ~ ARM performance.
Since Thumb-2 combines all of the nice features of ARM and Thumb,
which do you think will last longer? :-)

Wilco