qemu-devel
[Top][All Lists]
Advanced

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [Qemu-devel] [PATCH v2 09/10] target-arm: optimize neon vld/vst ops


From: Juha.Riihimaki
Subject: Re: [Qemu-devel] [PATCH v2 09/10] target-arm: optimize neon vld/vst ops
Date: Thu, 29 Oct 2009 14:45:49 +0100

On Oct 26, 2009, at 23:05, ext Aurelien Jarno wrote:

> On Mon, Oct 26, 2009 at 10:11:07AM +0100, Laurent Desnogues wrote:
>> On Mon, Oct 26, 2009 at 8:46 AM,  <address@hidden> wrote:
>>>
>>> On Oct 25, 2009, at 16:01, ext Laurent Desnogues wrote:
>>>
>>>> I don't really like the idea of having tcg_qemu_ld/st not factored
>>>> in some place, as it makes memory access tracing extensions
>>>> more intrusive.
>>>>
>>>> This brings us back to the problem having function freeing tmps.
>>>> In that case, you could perhaps create a gen_st16_dont_free
>>>> function as a temporary workaround?
>>>
>>> I could, but it is getting ugly =/ How about if I create another  
>>> patch
>>> that moves the temporary variable (de)allocation outside the  
>>> gen_ldx/
>>> gen_stx functions and then refactor this patch on top of that?
>>
>> I'd personally like this, but I guess you'd better wait for some  
>> inputs
>> from other QEMU dev's before doing it.
>
> Looks like the best option to me.

Alrighty then, I did the patch against the latest git and it's rather  
large... but seems to have broken nothing at least in my testing. The  
patch will remove all implicit tcg temp variable allocation and  
deallocation in target-arm/translate.c and make it the responsibility  
of the calling function. At the same time I also removed the new_tmp  
and dead_tmp functions completely because I see no point in only  
tracking some of the 32bit temp variables instead of everything.  
Personally I think the patch makes reading and understanding (and why  
not also writing) the file much easier. I do also have a version that  
has a compile time option of substituting the tcg temp variable alloc/ 
dealloc function calls with inline functions that track the usage but  
this is not included with the patch.

I'll send the patch shortly, should be nice bed-time reading I  
guess ;) Please comment when you have free time to read it through...


Cheers,
Juha




reply via email to

[Prev in Thread] Current Thread [Next in Thread]