[llvm-commits] [patch][arm] Fix eabi calling convention when a 64 bit value shadows r3

[Sandeep Patel]

On Fri, Aug 6, 2010 at 3:18 PM, Rafael Espindola <espindola at google.com> wrote:
> On 6 August 2010 10:55, Sandeep Patel <deeppatel1987 at gmail.com> wrote:
>> This is PR4058, right? arguments-nosplit-i64.ll seems to be testing
>> the same thing. If the problem is that R3 isn't marked as used and can
>> get backfilled by mistake, shouldn't we expand the test to confirm
>> that?
>
> I don't think that is exactly the same problem. What I have seen was
>
> * R3 is not marked as "used"
> * ARM backend thinks it has to save it to the stack because of vaarg
> * Offset computation correctly ignores it
> * Offsets are wrong
>
> That is why in the test I check that there is a load from [sp]
> immediately followed by a ret. Without this patch we would get the
> same load followed by a stack adjustment and then a ret.
>
> Note that this is only an issue with vaarg. A new test is probably appropriate.

Sounds good then.

deep