ARMC6: Use float-abi=hard

[kjbracey]

Description

For binary compatibility with ARMC5, use the hardfp ABI variant whenever
we have FP hardware - this is ARMC5's default behaviour, which we do not
override.

Cortex-M4F was already using hard; this brings M7F and M33F into line.

Pull request type

[X] Fix
[ ] Refactor
[ ] Target update
[ ] Functionality change
[ ] Docs update
[ ] Test update
[ ] Breaking change

[kjbracey]

See issue #9153 for more discussion. I believe this is a minimal change bringing ARMC6 into line with ARMC5 - but it would be good if someone can cross-check that I've understood what ARMC5 is doing correctly.

As to what the general direction should be, that's maybe a separate issue. We currently have 2 toolchains using hardfp, 1 using softfp...

[deepikabhavnani]

Had similar discussion while setting FPU for Cortex-M33, and as per @theotherjimmy we use softfp.
I am not sure if it was for nanostack only or something else. Will have to wait for this till @theotherjimmy is back from vacation (2019)

[kjbracey]

Yeah, I saw that discussion, but not any reasoning. I can also imagine either changing M4F to softfp, or changing just M7F to hardfp, depending on what the plan is. It'll have to wait.

[SenRamakri]

I think this is a good change, because it optimizes the function call overhead when we pass many floating point args.
But note that this will not be backward compatible with any libraries currently compiled with softfp. I hope this doesn't break any targets if using any such libraries(with softfp).

[SenRamakri]

@cmonr - Although I approved please wait for @theotherjimmy also to confirm.

[kjbracey]

But note that this will not be backward compatible with any libraries currently compiled with softfp. I hope this doesn't break any targets if using any such libraries(with softfp).

Actually, I believe this avoids binary compatibility problems - this makes the float ABI the same as we're using with ARMC5 (it defaults to hard if FP is present, and we don't select anything manually, afaict).

So only an issue if you've got precompiled ARMC6 binaries somewhere that you're only using with ARMC6.

[theotherjimmy]

@kjbracey-arm @deepikabhavnani That was mostly GCC I was talking about for softfp. I agree with @kjbracey-arm This change is for the best. Just a second on that approval.

[cmonr]

CI started

[mbed-ci]

Test run: FAILED

Summary: 1 of 11 test jobs failed
Build number : 1
Build artifacts

Failed test jobs:

• jenkins-ci/mbed-os-ci_greentea-test

[0xc0170]

Test restarted