[llvm-commits] [llvm] r162301 - in /llvm/trunk: foo lib/CodeGen/RegAllocGreedy.cpp

[David Blaikie]

On Tue, Aug 21, 2012 at 7:28 PM, Craig Topper <craig.topper at gmail.com> wrote:
> Something like this?

That looks about like what I'd half-done this afternoon (sorry, didn't
mean for you (or I) to duplicate effort - just been away from a
computer since 3:30).

> Do I need to check if the MachineFunction has a valid
> Function the way MachineBasicBlock checks and returns "(null)"? None of the
> callers I replaced seemed to be checking for a valid result from
> getFunction().

I just threw in an assert in the MachineFunction::getName. You could
do that & this looks generally fine to commit to me (under the "post
commit review for obvious changes") but you can wait for Jakob to
check it if you prefer.

(I'll check in any outstanding changes (if there are any) from my
similar work once you check this in)

- David

>
>
> On Tue, Aug 21, 2012 at 5:06 PM, Craig Topper <craig.topper at gmail.com>
> wrote:
>>
>> I'll work on this tonight.
>>
>>
>> On Tue, Aug 21, 2012 at 1:25 PM, Jakob Stoklund Olesen <stoklund at 2pi.dk>
>> wrote:
>>>
>>>
>>> On Aug 21, 2012, at 1:14 PM, Craig Topper <craig.topper at gmail.com> wrote:
>>>
>>> > I think the other register allocators have the same cast and should be
>>> > fixed too.
>>>
>>> Even better, add MachineFunction::getName(), and stop including
>>> "llvm/Function.h" in every backend pass.
>>>
>>> See MachineBasicBlock::getName().
>>>
>>> /jakob
>>>
>>
>>
>>
>> --
>> ~Craig
>
>
>
>
> --
> ~Craig