[PATCH] ARM attributes and user intent

[Richard Earnshaw]

On Wed, 2009-01-21 at 11:07 +0000, Andrew Stubbs wrote:

> The patch adds a test case for every possible -march and -mfpu to ensure 
> that the attributes are set correctly, and a number of test cases to 
> ensure that the precedence rules above are obeyed. I have also updated 
> the expected results for all affected existing tests.
> 
> OK?
> 

This looks good.  However, if code is compiled/assembled without
interworking, then it seams a bit odd that the other instruction set is
still permitted.

Thus if the user compiles code with -mcpu=arm7tdmi -mno-thumb-interwork,
should the code be marked arm-only (ie permitted thumb use was none)?

Sadly, although gas has -mthumb-interwork, it doesn't have the inverse
(-mno-thumb-interwork).  That's a bit unfortunate given that the new ABI
mandates interworking by default.

There are also technical difficulties for some thumb non-interworking
code that still uses the ARM instruction set internally (but in a way
that is not visible to the user).  This is done, for example, in the
support code for soft-float in gcc.

R.