[llvm-commits] [llvm] r72593 - in /llvm/trunk/lib/Target/ARM: ARM.td ARMInstrInfo.td ARMInstrThumb2.td ARMSubtarget.cpp ARMSubtarget.h

[Evan Cheng]

Sent from my iPhone

On May 30, 2009, at 12:23 PM, Evan Cheng <evan.cheng at apple.com> wrote:

>
>
> Sent from my iPhone
>
> On May 30, 2009, at 1:20 AM, Anton Korobeynikov
> <anton at korobeynikov.info> wrote:
>
>> Hello, Evan
>>
>>> I don't think we want to distinguish thumb2 from thumb1. Thumb2 is  
>>> an
>>> ISA extension, it should not be considered a separate ISA from
>>> thumb1.
>> It's not separate... As other ARM stuff Thumb2 implies thumb1...
>>
>>> I would think it makes more sense to have just two modes: arm vs
>>> thumb
>>> mode. When in thumb mode, if thumb2 ISA is available, then that
>>> implies these instructions are available.
>> I thought about some future stuff. Consider we'll have Thumb3,  
>> Thumb4,
>> Thumb-A20, whatever, each one implying the previous ones.... It seems
>> ugly to have bunch of booleans for this...
>
> You are thinking too far ahead. :)
>
>>
>> How's about having isThumb and ThumbMode instead? The first one will
>> catch ARM vs Thumb and second one - the actual thumbmode...
>
> That sounds fine. It's similar to the various sse levels.
>
>>
>>> This is not necessary. We don't want thumb2-foo-bar. We have -
>>> march=arm and -march=thumb, but no -march=thumb2.
>> Right, but:
>> 1. We need to distinguish v6 vs v6t2. Which differs only in presence
>> of thumb2 instructions. The latter should be available via
>> "thumbv6t2".
>
> Why?
>
>> 2. We might want to have v6t2/v7 arch but generate Thumb1
>> instructions only...
>
> The first part of the triplet should corespond llvm -march=arm / -
> march=thumb. What instructions are available should be controlled by -
> mattr=, no? If we really need to encode other information, e.g.
> forcing thumb1, they should be controlled by function attributes. What
> do you think?

I should clarify. I think triplet should be used to specify arch, i.e.  
arm / thumb. -mcpu and -mattr should be used to control what  
instructions are available to isel.

Evan
>
> Evan
>
>>
>> -- 
>> With best regards, Anton Korobeynikov
>> Faculty of Mathematics and Mechanics, Saint Petersburg State
>> University
>> _______________________________________________
>> llvm-commits mailing list
>> llvm-commits at cs.uiuc.edu
>> http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits
> _______________________________________________
> llvm-commits mailing list
> llvm-commits at cs.uiuc.edu
> http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits