This discussion has been locked.

You can no longer post new replies to this discussion. If you have a question you can start a new discussion

Issue regarding running code on RVDS with cortex A8

Note: This was originally posted on 29th March 2010 at http://forums.arm.com

Hi all,

The issue is i'm trying to execute a code,written in NEON instruction set.For example,any code,say a simple for loop calculation.
When i'm step debugging,i found that code gets into infinite loop after going into __fp_init function(This function is called by __main() before program main).
these are my compiler settings:
--arm --debug --cpu=Cortex-A8 --fpu=VFPv3 --fpmode=fast -O1 --no_inline

If i change my --fpu setting to --softfpu than whenever code reaches NEON instruction,it again goes into infinite loop.

One info i found on ARM website is that go for --fpu=VFPv3D32 which is implied NEON registers.

see the link:

[url="http://infocenter.arm.com/help/index.jsp?topic=/com.arm.doc.dai0133c/index.html"]http://infocenter.arm.com/help/index.jsp?t...133c/index.html[/url]

Sadly,this option is not provided in my RVDS 4.0.only close to this option i'm getting is --fpu=VFPv3D16.

Evaluation copy should not be a problem as it includes all option of RVDS 4.0 professional.

Other answer i found is go for vfp initilization code(Application Code 133).Currently,i'm looking into it.let's see.
Even a simple hello world program is not executing with--cpu=Cortex-A8 --fpu=VFPv3,

i hope to find a solution soon.
any help on this would be appreciated.

Regards
Suvir

Offline suvir bhargav over 9 years ago

Note: This was originally posted on 30th March 2010 at http://forums.arm.com

Thanks scott for your such a quick reply.

Your example initialization link was easy to understand and now my code runs perfectly with co-processor enabled.
i compiled the code with cpu=cortex A8 and -fpu=auto(default) and -fpu=VFPv3.its working fine with these options.

Thanks & Regards
Suvir
Cancel
Up 0 Down

Cancel
Offline Scott Douglass over 9 years ago

Note: This was originally posted on 29th March 2010 at http://forums.arm.com

Both VFP and NEON instructions require coprocessors 10 & 11 to be enabled in the CAR and the EN bit to be set in the FPEXC register -- otherwise they will cause an undefined instruction exception. You don't say what your target is (harware? model?), but if the undefined exception is not caught/handled then it could wind up trying to run your image again (i.e. an infinite loop).

When building for --fpu=vfp* (which is the default with --cpu=Cortex-A8) the runtime library in __fp_init() will initialize the FPEXC register, but the OS or VFP support code is responsible for enabling coprocessors 10 & 11 the CAR.

For an example of initializing all of this "by hand" see this example code [url="http://infocenter.arm.com/help/topic/com.arm.doc.dui0205i/BGBEJBCF.html"]http://infocenter.arm.com/help/topic/com.arm.doc.dui0205i/BGBEJBCF.html[/url]
Cancel
Up 0 Down

Cancel