This is the mail archive of the
gcc@gcc.gnu.org
mailing list for the GCC project.
Re: xscale-elf-gcc
- From: Richard Earnshaw <rearnsha at arm dot com>
- To: Arvind Krishnaswamy <arvind at CS dot Arizona dot EDU>
- Cc: gcc at gcc dot gnu dot org, Richard dot Earnshaw at arm dot com
- Date: Wed, 08 May 2002 11:21:10 +0100
- Subject: Re: xscale-elf-gcc
- Organization: ARM Ltd.
- Reply-to: Richard dot Earnshaw at arm dot com
> Hi,
> When I use the march=armv4t option with the xscale-elf-gcc compiler,
> what pipeline does it target? Does it use armv4t compliant instructions
> but still target the xscale pipeline? The reason I ask is because I
> notice that the compiler chooses to predicate branches even when the sequences
> are long (branches would incur 1 stall but long sequences of predicated code
> would result in no-ops).
>
The honest answer is that I'm not sure without completely
reverse-engineering the relevant part of the code. However, I suspect
that the if-conversion pass is sometimes getting too enthusiastic -- I've
occasionally seen sequences way longer than the 4+4 that I would normally
expect.
> When I specify mcpu=strongarm and march=armv4 the compiler complains about
> a conflict. Isn't the strongarm compliant with armv4?
Use -mcpu on its own, or -march=<arch> and -mtune=<processor> together.
>
> Another question: which is the first arm processor to support thumb instructions?
ARM7TDMI (architecture armv4t).
R.