Updating the makefile to have armv7=0 should do it, but I never tried.

Rodolph.

On 3 September 2012 12:50, Kevin Read <[email protected]> wrote:
> Hi Rodolph,
>
> I see, thanks. Do you know if this also applies to the android_arm target? 
> The Makefile.android states that v7 is active and neon off. Would I have to 
> change the Makefile to create an armv5 build?
>
> Cheers,
> Kevin
>
> On 03.09.2012, at 13:16, Rodolph Perfetta wrote:
>
>> Hi Kevin,
>>
>> The standalone V8 for ARM builds for armv7+vfpv3 by default using gyp.
>>
>> V8 has two compilers, one non optimising which compiles everything the
>> first time and an optimising compiler (crankshaft) for hotspots. To
>> use crankshaft on ARM you need armv7+vfpv3. The non optimising
>> compiler works on armv5 (and probably armv4). If performance is
>> important then crankshaft is a must.
>>
>> Note that V8 will check the CPU at runtime so even if statically
>> compiled for armv5 it may use cranckshaft when running.
>>
>> NEON is a SIMD engine and JS VM currently don't emit SIMD like code.
>>
>> Cheers,
>> Rodolph.
>>
>> On 2 September 2012 22:15, Paul Lind <[email protected]> wrote:
>>> Hi Kevin -
>>>
>>> Regarding your MIPS question, you are best off pulling from Google's release
>>> branches, which match up to major Chrome releases. Those will have received
>>> the most testing, on all platforms. The latest right now is 3.12, and
>>> Android Jelly Bean is using 3.9, if that matters to you at all.
>>>
>>> As you may or may not know, the MIPS port is developed by our outside team,
>>> and on any given day we may not have finished porting necessary to work with
>>> recent changes the Google team has made. Roughly once a week, they push to
>>> trunk, and we currently don't sync up with them on these pushes. So MIPS may
>>> be broken in trunk. We definitely do sync up on the release branches, and
>>> get everything thoroughly tested.
>>>
>>> We do keep the code at our own repo (https://github.com/paul99/v8m-rb)
>>> working and our bleeding_edge branch will match theirs (though possibly a
>>> day or two behind as port changes). So far there has not been demand for
>>> keeping 'trunk' in sync for MIPS. But again, our bleeding_edge and theirs
>>> can be broken at a given moment in time, so release branches are your best
>>> bet for product development.
>>>
>>> I don't know the answer to all your ARM questions, but I do know that Neon
>>> is not used in v8, however the VFP3 floating point is required for the
>>> optimizing (crankshaft) compiler.
>>>
>>> thanks,
>>>
>>> paul
>>>
>>>
>>>
>>> On Sep 2, 2012, at 10:55 AM, Kevin Read wrote:
>>>
>>> Hi all,
>>>
>>> I succeeded in using v8 via JNI and the ndk from within my android app. The
>>> idea here is to reuse existing JS code but with performance improvements. I
>>> have some questions regarding the build flags for the different
>>> architectures.
>>>
>>> I was under the impression that v8 builds for ARMv5 by default. Is that
>>> correct? I also read somewhere that when compiled for ARMv5 v8 would not use
>>> the JIT and so performance would suffer. In any case, how would I go about
>>> compiling for ARMv5 or ARMv7 explicitly?
>>> Would it make sense to compile v8 with softfp and/or neon? I'm working with
>>> fairly math intensive code.
>>>
>>> I would also like to investigate building for MIPS, so that I have all
>>> tablet hardware covered. Can I use trunk for that at the moment, or should I
>>> try to work with the code from https://github.com/paul99/v8m-rb/?
>>>
>>> Thanks for your input, and for this best of JS VMs :)
>>>
>>> Regards,
>>> Kevin
>>>
>>>
>>> --
>>> v8-users mailing list
>>> [email protected]
>>> http://groups.google.com/group/v8-users
>>>
>>>
>>> --
>>> v8-users mailing list
>>> [email protected]
>>> http://groups.google.com/group/v8-users
>>
>> --
>> v8-users mailing list
>> [email protected]
>> http://groups.google.com/group/v8-users
>
> --
> v8-users mailing list
> [email protected]
> http://groups.google.com/group/v8-users

-- 
v8-users mailing list
[email protected]
http://groups.google.com/group/v8-users

Reply via email to