ARM GCC machine options for RTEMS 6

[Chris Johns]

On 4/7/20 5:43 pm, Sebastian Huber wrote:
> On 04/07/2020 01:03, Chris Johns wrote:
>> On 3/7/20 5:06 pm, Sebastian Huber wrote:
>>> Hello,
>>> someone from ARM changed the machine options used to build the RTEMS multilibs
>>> some time ago (I think GCC 8):
>> Was a reason given?
> ARM thinks the new options are easier to use.

Hmm.

>>> arm-rtems6-gcc -print-multi-lib -march=armv4t
>>> .;
>>> thumb;@mthumb
>>> armv5te+fp/hard;@march=armv5te+fp at mfloat-abi=hard
>>> thumb/armv6-m;@mthumb at march=armv6-m
>>> thumb/armv7-a;@mthumb at march=armv7-a
>>> thumb/armv7-r;@mthumb at march=armv7-r
>>> thumb/cortex-m3;@mthumb at mcpu=cortex-m3
>>> thumb/cortex-m4+nofp;@mthumb at mcpu=cortex-m4+nofp
>>> thumb/armv7-a+simd/hard;@mthumb at march=armv7-a+simd at mfloat-abi=hard
>>> thumb/armv7-r+fp/hard;@mthumb at march=armv7-r+fp at mfloat-abi=hard
>>> thumb/cortex-m4/hard;@mthumb at mcpu=cortex-m4 at mfloat-abi=hard
>>> thumb/cortex-m7/hard;@mthumb at mcpu=cortex-m7 at mfloat-abi=hard
>>> eb/thumb/armv7-r;@mbig-endian at mthumb@march=armv7-r
>>> eb/thumb/armv7-r+fp/hard;@mbig-endian at mthumb@march=armv7-r+fp at mfloat-abi=hard
>>>
>>> It seems that our current machine options still map to the right multilib:
>>>
>>> arm-rtems6-gcc -print-multi-directory -march=armv7-a -mthumb -mfpu=neon
>>> -mfloat-abi=hard -mtune=cortex-a9
>>> thumb/armv7-a+simd/hard
>>>
>>> arm-rtems6-gcc -print-multi-directory -mthumb -mcpu=cortex-m4 -mfpu=fpv4-sp-d16
>>> -mfloat-abi=hard
>>> thumb/cortex-m4/hard
>>>
>>> arm-rtems6-gcc -print-multi-directory -march=armv7-r -mthumb -mbig-endian
>>> eb/thumb/armv7-r
>>>
>>> arm-rtems6-gcc -print-multi-directory -march=armv7-r -mthumb -mbig-endian
>>> -mfpu=vfpv3-d16 -mfloat-abi=hard
>>> eb/thumb/armv7-r+fp/hard
>>>
>>> Should we change the options to match the ones given in the multilib definition
>>> or should we keep the current options? If we change the options, then we can no
>>> longer build with GCC 7.
>> Would we need to build gcc 7 for RTEMS 6?
> 
> No, but for a git bisect it would be nice to use the RTEMS 5 compiler for a
> while for the master branch.

Sure.

> Since the new build system will be a hard break, maybe we should keep the
> options as is in the old build system and start with the new options in the new
> build system.

This seems reasonable.

Chris