[llvm] r209650 - ARM: teach AAPCS-VFP to deal with Cortex-M4.

Tim Northover tnorthover at apple.com
Mon Jun 2 09:05:48 PDT 2014


Hi Oliver,

> On 2 Jun 2014, at 12:20, Oliver Stannard <oliver.stannard at arm.com> wrote:
> Thanks for working on this, however there are still a few problems with
> targeting FPv4-SP and AAPCS-VFP:
> * Arguments of type double need to be aligned when passed in registers, as
> well as on the stack. For example, a function which takes a float and a
> double should pass the double in s2 and s3 (which overlap d1), not s1 and
> s2.
> * This patch only handles function arguments, double return values are
> still (incorrectly) passed in integer registers.

That doesn’t entirely surprise me. You should probably file a bug about it; I’ve not got any big Cortex-M4 AAPCS-validation plans at the moment (this was mostly to stop the compiler-rt build failing) so any time I get to spend on it is likely to be interrupted and spotty.

Cheers.

Tim.



More information about the llvm-commits mailing list