At 18:45 15/09/00 +0200, Xavier DEBREUIL wrote:
> Consider the following structure :
> struct foo{
> u8 buf[5];
> u8 frame[0];
> } ;
(snipped)
sizeof(foo) == 8 with arm-linux-gcc, or 5 with (x86?) gcc.
You are seeing that the minimum structure alignment is 4 for the ARM
compiler, but 1 for the x86 gcc.
This is due to an historical optimisation choice, made for the original
commercial ARM compiler. Somebody thought they could compile faster code
for ARM by making this restriction. For example, you can use a word
aligned memcpy for "foo" on ARM, but you have to use the generic version on
x86.
I believe arm-gcc maintainers chose to do the same for object-level
compatibility.
Unfortunately changing this breaks compatibility, So if you change it for
your own code, you may not be able to use some parts of glibc anymore. It
could fail in bad ways. (Unless you recompile everything on your system
yourself.)
ARM has changed this in our tool-chain, and we took a backwards
compatibility hit when we did so.
So "arm-gcc" is not compatible with the ARM ADS tool-chain, as a result.
Phil - are there any plans to change this in a future gcc?
Efforts are still underway to document this as a standard. Currently this
is not part of either the ARM ATPCS or the ARM ELF specs, so it's not
surprising there are compatibility problems.
There's a similar problem with 4 byte alignment of global variables.
> Is there any option for the arm-linux-gcc to give the same result as the
gcc ?
"-mstructure-size-boundary=8" (for 1 byte = 8 bits alignment)
Be very careful with this, because it will break compatibility with some
libraries, and other object files.
Cheers,
Michael Nonweiler.
(speaking for myself)
_______________________________________________
http://lists.arm.linux.org.uk/mailman/listinfo/linux-arm
