ARM support - which architecture variants?
Hugo Vincent
I can't find in the documentation which ARM architecture types V8
supports. Does it support older ARM9 devices (I'm specifically
interested in an ARMv5te architecture, ARM926ej-s device) or only
newer ARMv7 (Cortex-A8 etc)? I can see that it is (supposed to) build
on ARMv5te, but do all the JIT features work or is it running in a
byte code interpreter fallback or something? Can I expect good
performance?
Thanks,
Hugo
Rodolph Perfetta
--
v8-users mailing list
v8-u...@googlegroups.com
http://groups.google.com/group/v8-users
Søren Gjesse
Hugo Vincent
anyone have any benchmarks? I'm hesitant to invest time in using V8
for my project if it's going to get substantially slower soon. Is
there any estimated time frame for when nocrankshaft will be
deprecated?
Thanks,
Hugo
On Feb 23, 9:14 pm, Søren Gjesse <sgje...@chromium.org> wrote:
> Just a follow-up note regarding the new optimizing compiler (crankshaft).
> This will be enabled by default for ARM quite soon, and the existing
> optimizing compiler will be removed at some point. For non ARMv7+VFP devices
> this means that the base JIT (non-optimizing/full-compiler) will be used. To
> measure the different compilers on a ARMv7+VFP device use following options:
>
> --nocrankshaft (current optimizing JIT - the current default)
> --crankshaft (new optimizing JIT - the soon to be default)
> --always-full-compiler (base/non-optimizing compiler)
>
> Going forward using --crankshaft on a non ARMv7+VFP device will have no
> effect and execution will fallback to --always-full-compiler.
>
> Regards,
> Søren
>
> On Wed, Feb 23, 2011 at 18:33, Rodolph Perfetta
>
>
>
>
>
>
> > V8 can run on ARMv4 devices (non T though).
>
> > There is no interpreter in V8 so you will be using the JIT every time,
> > perfromance should be good (keep in mind CPU like 926-ej-s do not have L2
> > cache and this is going to have a visible impact). There is a new JIT
> > infrastructure being developed (crankshaft) which features an optimising JIT
> > and this will only be for ARMv7+VFP devices.
>
> > HTH,
> > Rodolph.
>
Søren Gjesse
Arun M
Hi
Is Crankshaft optimizing compiler enabled for ARMv7-A + NEON devices
which does not have VFPv3 FPU?
Thanks & Regards
Arun
Alexandre Rames
Søren Gjesse
Víctor M. Jáquez L.
> Hello Arun,
>
> Crankshaft is now enabled by default on ARM processors supporting VFPv3.
> v8 does not use NEON (it is not worth using, at least currently), but I
> believe no processors with this configuration (NEON without VFPv3) exist
> anyway. So if the CPU feature detection work correctly, v8 should assume
> that you have VFPv3 if it detects NEON, and thus use use Crankshaft.
>
> If I remember correctly there has been some issues with incorrect feature
> detection. A simple run of the v8 benchmarks with the latest bleeding_edge
> and the frequency of your CPU should be enough to determine if crankshaft is
> enabled.
The CPU features detection is done parsing the /proc/cpuinfo file:
https://code.google.com/p/v8/source/browse/branches/bleeding_edge/src/platform-linux.cc#111
So, if your cpu information is not exposed correctly the detection will fail.
I'd never seen an ARM with NEON but without VFPv3.
vmjl
>
> I hope this helps.
>
> Regards,
>
> Alexandre
>
>
> On Sat, Apr 2, 2011 at 7:40 AM, Arun M <arunm....@gmail.com> wrote:
>
> >
> >
> > Hi
> >
> > Is Crankshaft optimizing compiler enabled for ARMv7-A + NEON devices
> > which does not have VFPv3 FPU?
> >
> > Thanks & Regards
> > Arun
> >
> >
> > On Mar 9, 6:02 pm, S�ren Gjesse <sgje...@chromium.org> wrote:
> > > For ARM crankshaft is now the default. This change is in the repository
> > > starting from V8 version 3.2. To use the previous optimizing compiler
> > > --nocrankshaft will have to be used. When crankshaft for ARM has been
> > fully
> > > stabilized the previous optimizing compiler will be removed from the
> > > repository and running with --nocrankshaft will no longer be possible.
> > There
> > > is no specific date to when this will happen but most likely it will be
> > > within a month or two. The removal of the previous optimizing compiler
> > will
> > > happen for all supported platforms simultaneously,
> > >
> > > The previous optimizing compiler can of cause still be found in previous
> > > versions of V8.
> > >
> > > Regards,
> > > S�ren
> > > > > S�ren
Víctor M. Jáquez L.
> The Crankshaft optimizing compiler requires VFPv3. Without VFPv3 the
> non-optimizing will be used for all code. V8 cannot make use of NEON.
Why v8 cannot make use of NEON? Because of an specific reason?
vmjl
>
> Regards,
> S�ren
>
> On Sat, Apr 2, 2011 at 08:40, Arun M <arunm....@gmail.com> wrote:
>
> >
> >
> > Hi
> >
> > Is Crankshaft optimizing compiler enabled for ARMv7-A + NEON devices
> > which does not have VFPv3 FPU?
> >
> > Thanks & Regards
> > Arun
> >
> >
> > On Mar 9, 6:02 pm, S�ren Gjesse <sgje...@chromium.org> wrote:
> > > For ARM crankshaft is now the default. This change is in the repository
> > > starting from V8 version 3.2. To use the previous optimizing compiler
> > > --nocrankshaft will have to be used. When crankshaft for ARM has been
> > fully
> > > stabilized the previous optimizing compiler will be removed from the
> > > repository and running with --nocrankshaft will no longer be possible.
> > There
> > > is no specific date to when this will happen but most likely it will be
> > > within a month or two. The removal of the previous optimizing compiler
> > will
> > > happen for all supported platforms simultaneously,
> > >
> > > The previous optimizing compiler can of cause still be found in previous
> > > versions of V8.
> > >
> > > Regards,
> > > S�ren
> > > > > S�ren
Mads Sig Ager
we have not found any of the NEON instructions useful to speed up
computations in V8. We don't have anything against NEON. So far we
have been able to do better without.
Cheers, -- Mads
On Mon, Apr 4, 2011 at 11:20 AM, Víctor M. Jáquez L. <vja...@igalia.com> wrote:
> On Mon, Apr 04, 2011 at 10:46:29AM +0200, Søren Gjesse wrote:
>> The Crankshaft optimizing compiler requires VFPv3. Without VFPv3 the
>> non-optimizing will be used for all code. V8 cannot make use of NEON.
>
> Why v8 cannot make use of NEON? Because of an specific reason?
>
> vmjl
>
>>
>> Regards,
>> Søren
>>
>> On Sat, Apr 2, 2011 at 08:40, Arun M <arunm....@gmail.com> wrote:
>>
>> >
>> >
>> > Hi
>> >
>> > Is Crankshaft optimizing compiler enabled for ARMv7-A + NEON devices
>> > which does not have VFPv3 FPU?
>> >
>> > Thanks & Regards
>> > Arun
>> >
>> >
>> > On Mar 9, 6:02 pm, Søren Gjesse <sgje...@chromium.org> wrote:
>> > > For ARM crankshaft is now the default. This change is in the repository
>> > > starting from V8 version 3.2. To use the previous optimizing compiler
>> > > --nocrankshaft will have to be used. When crankshaft for ARM has been
>> > fully
>> > > stabilized the previous optimizing compiler will be removed from the
>> > > repository and running with --nocrankshaft will no longer be possible.
>> > There
>> > > is no specific date to when this will happen but most likely it will be
>> > > within a month or two. The removal of the previous optimizing compiler
>> > will
>> > > happen for all supported platforms simultaneously,
>> > >
>> > > The previous optimizing compiler can of cause still be found in previous
>> > > versions of V8.
>> > >
>> > > Regards,
>> > > Søren
>> > > > > Søren
Arun M
I am running the v8 benchmark suite from Chromium browser.
(http://v8.googlecode.com/svn/data/benchmarks/v6/run.html)
Currently the browser version is 12.0.703.0.
But I still have not got any improvement on ARM platform for the V8
benchmark suite.
Which version of Chromium will include the crankshaft improvement for
V8 on ARM?
Thanks & Regards
Arun
On Apr 4, 7:52 pm, Mads Sig Ager <a...@chromium.org> wrote:
> Hi Victor,
>
> we have not found any of the NEON instructions useful to speed up
> computations in V8. We don't have anything against NEON. So far we
> have been able to do better without.
>
> Cheers, -- Mads
>
>
>
>
>
>
>
>
> > On Mon, Apr 04, 2011 at 10:46:29AM +0200, Søren Gjesse wrote:
> >> The Crankshaft optimizing compiler requires VFPv3. Without VFPv3 the
> >> non-optimizing will be used for all code. V8 cannot make use of NEON.
>
> > Why v8 cannot make use of NEON? Because of an specific reason?
>
> > vmjl
>
> >> Regards,
> >> Søren
>
Jérémy Lal
> Hi
>
> I am running the v8 benchmark suite from Chromium browser.
> (http://v8.googlecode.com/svn/data/benchmarks/v6/run.html)
> Currently the browser version is 12.0.703.0.
>
> But I still have not got any improvement on ARM platform for the V8
> benchmark suite.
>
> Which version of Chromium will include the crankshaft improvement for
> V8 on ARM?
Check
http://src.chromium.org/viewvc/chrome/releases/
http://src.chromium.org/viewvc/chrome/releases/12.0.725.0/DEPS?revision=80328&view=markup
has :
'src/v8':
'http://v8.googlecode.com/svn/trunk@7427',
and in : http://code.google.com/p/v8/source/list?num=25&start=7446
you can see that it's version 3.2.6.
Same reading for 12.0.703.0 gives v8 3.2.1.1
Maybe chromium builds v8 with --nocrankshaft ?
Along with :
>>>>>> For ARM crankshaft is now the default. This change is in the repository
>>>>>> starting from V8 version 3.2.
>
> Thanks & Regards
> Arun
>
> On Apr 4, 7:52 pm, Mads Sig Ager <a...@chromium.org> wrote:
>> Hi Victor,
>>
>> we have not found any of the NEON instructions useful to speed up
>> computations in V8. We don't have anything against NEON. So far we
>> have been able to do better without.
>>
>> Cheers, -- Mads
>>
>> On Mon, Apr 4, 2011 at 11:20 AM, V锟絚tor M. J锟絨uez L. <vjaq...@igalia.com> wrote:
>>
>>
>>
>>
>>
>>
>>
>>> On Mon, Apr 04, 2011 at 10:46:29AM +0200, S锟絩en Gjesse wrote:
>>>> The Crankshaft optimizing compiler requires VFPv3. Without VFPv3 the
>>>> non-optimizing will be used for all code. V8 cannot make use of NEON.
>>
>>> Why v8 cannot make use of NEON? Because of an specific reason?
>>
>>> vmjl
>>
>>>> Regards,
>>>> S锟絩en
>>
>>>> On Sat, Apr 2, 2011 at 08:40, Arun M <arunm.chr...@gmail.com> wrote:
>>
>>>>> Hi
>>
>>>>> Is Crankshaft optimizing compiler enabled for ARMv7-A + NEON devices
>>>>> which does not have VFPv3 FPU?
>>
>>>>> Thanks & Regards
>>>>> Arun
>>
>>>>> On Mar 9, 6:02 pm, S锟絩en Gjesse <sgje...@chromium.org> wrote:
>>>>>> For ARM crankshaft is now the default. This change is in the repository
>>>>>> starting from V8 version 3.2. To use the previous optimizing compiler
>>>>>> --nocrankshaft will have to be used. When crankshaft for ARM has been
>>>>> fully
>>>>>> stabilized the previous optimizing compiler will be removed from the
>>>>>> repository and running with --nocrankshaft will no longer be possible.
>>>>> There
>>>>>> is no specific date to when this will happen but most likely it will be
>>>>>> within a month or two. The removal of the previous optimizing compiler
>>>>> will
>>>>>> happen for all supported platforms simultaneously,
>>
>>>>>> The previous optimizing compiler can of cause still be found in previous
>>>>>> versions of V8.
>>
>>>>>> Regards,
>>>>>> S锟絩en
>>
>>>>>> On Tue, Mar 8, 2011 at 20:05, Hugo Vincent <hugo.vinc...@gmail.com>
>>>>> wrote:
>>>>>>> How much slower is full-compiler than nocrankshaft on ARM926ej-s -
>>>>>>> anyone have any benchmarks? I'm hesitant to invest time in using V8
>>>>>>> for my project if it's going to get substantially slower soon. Is
>>>>>>> there any estimated time frame for when nocrankshaft will be
>>>>>>> deprecated?
>>
>>>>>>> Thanks,
>>>>>>> Hugo
>>
>>>>>>> On Feb 23, 9:14 pm, S锟絩en Gjesse <sgje...@chromium.org> wrote:
>>>>>>>> Just a follow-up note regarding the new optimizing compiler
>>>>> (crankshaft).
>>>>>>>> This will be enabled by default for ARM quite soon, and the existing
>>>>>>>> optimizing compiler will be removed at some point. For non ARMv7+VFP
>>>>>>> devices
>>>>>>>> this means that the base JIT (non-optimizing/full-compiler) will be
>>>>> used.
>>>>>>> To
>>>>>>>> measure the different compilers on a ARMv7+VFP device use following
>>>>>>> options:
>>
>>>>>>>> --nocrankshaft (current optimizing JIT - the current default)
>>>>>>>> --crankshaft (new optimizing JIT - the soon to be default)
>>>>>>>> --always-full-compiler (base/non-optimizing compiler)
>>
>>>>>>>> Going forward using --crankshaft on a non ARMv7+VFP device will have
>>>>> no
>>>>>>>> effect and execution will fallback to --always-full-compiler.
>>
>>>>>>>> Regards,
>>>>>>>> S锟絩en
Søren Gjesse
--js-flags="--nocrankshaft"
Arun M
Will this method of detecting VFPv3 work in chromium browser when
sandbox is enabled?
Will browser sandbox allow to fopen "/proc/cpuinfo" file?
Arun
On Apr 4, 6:19 pm, Víctor M. Jáquez L. <vjaq...@igalia.com> wrote:
> On Mon, Apr 04, 2011 at 09:45:52AM +0100, Alexandre Rames wrote:
> > Hello Arun,
>
> > Crankshaft is now enabled by default on ARM processors supporting VFPv3.
> > v8 does not use NEON (it is not worth using, at least currently), but I
> > believe no processors with this configuration (NEON without VFPv3) exist
> > anyway. So if the CPU feature detection work correctly, v8 should assume
> > that you have VFPv3 if it detects NEON, and thus use use Crankshaft.
>
> > If I remember correctly there has been some issues with incorrect feature
> > detection. A simple run of the v8 benchmarks with the latest bleeding_edge
> > and the frequency of your CPU should be enough to determine if crankshaft is
> > enabled.
>
> The CPU features detection is done parsing the /proc/cpuinfo file:
>
> So, if your cpu information is not exposed correctly the detection will fail.
>
> I'd never seen an ARM with NEON but without VFPv3.
>
> vmjl
>
>
>
>
>
>
>
>
>
> > I hope this helps.
>
> > Regards,
>
> > Alexandre
>
>
> > > Hi
>
> > > Is Crankshaft optimizing compiler enabled for ARMv7-A + NEON devices
> > > which does not have VFPv3 FPU?
>
> > > Thanks & Regards
> > > Arun
>
> > > > For ARM crankshaft is now the default. This change is in the repository
> > > > starting from V8 version 3.2. To use the previous optimizing compiler
> > > > --nocrankshaft will have to be used. When crankshaft for ARM has been
> > > fully
> > > > stabilized the previous optimizing compiler will be removed from the
> > > > repository and running with --nocrankshaft will no longer be possible.
> > > There
> > > > is no specific date to when this will happen but most likely it will be
> > > > within a month or two. The removal of the previous optimizing compiler
> > > will
> > > > happen for all supported platforms simultaneously,
>
> > > > The previous optimizing compiler can of cause still be found in previous
> > > > versions of V8.
>
> > > > Regards,
> > > > On Tue, Mar 8, 2011 at 20:05, Hugo Vincent <hugo.vinc...@gmail.com>
> > > wrote:
> > > > > How much slower is full-compiler than nocrankshaft on ARM926ej-s -
> > > > > anyone have any benchmarks? I'm hesitant to invest time in using V8
> > > > > for my project if it's going to get substantially slower soon. Is
> > > > > there any estimated time frame for when nocrankshaft will be
> > > > > deprecated?
>
> > > > > Thanks,
> > > > > Hugo
>
> > > > > > Just a follow-up note regarding the new optimizing compiler
> > > (crankshaft).
> > > > > > This will be enabled by default for ARM quite soon, and the existing
> > > > > > optimizing compiler will be removed at some point. For non ARMv7+VFP
> > > > > devices
> > > > > > this means that the base JIT (non-optimizing/full-compiler) will be
> > > used.
> > > > > To
> > > > > > measure the different compilers on a ARMv7+VFP device use following
> > > > > options:
>
> > > > > > --nocrankshaft (current optimizing JIT - the current default)
> > > > > > --crankshaft (new optimizing JIT - the soon to be default)
> > > > > > --always-full-compiler (base/non-optimizing compiler)
>
> > > > > > Going forward using --crankshaft on a non ARMv7+VFP device will have
> > > no
> > > > > > effect and execution will fallback to --always-full-compiler.
>
> > > > > > Regards,
Víctor M. Jáquez L.
> Hi Victor
>
> Will this method of detecting VFPv3 work in chromium browser when
> sandbox is enabled?
> Will browser sandbox allow to fopen "/proc/cpuinfo" file?
Sorry Arun, I don't know. I'm not familiar with chrome in general.
vmjl
>
> Arun
>
> On Apr 4, 6:19�pm, V�ctor M. J�quez L. <vjaq...@igalia.com> wrote:
> > On Mon, Apr 04, 2011 at 09:45:52AM +0100, Alexandre Rames wrote:
> > > Hello Arun,
> >
> > > Crankshaft is now enabled by default on ARM processors supporting VFPv3..
> > > v8 does not use NEON (it is not worth using, at least currently), but I
> > > believe no processors with this configuration (NEON without VFPv3) exist
> > > anyway. So if the CPU feature detection work correctly, v8 should assume
> > > that you have VFPv3 if it detects NEON, and thus use use Crankshaft.
> >
> > > If I remember correctly there has been some issues with incorrect feature
> > > detection. A simple run of the v8 benchmarks with the latest bleeding_edge
> > > and the frequency of your CPU should be enough to determine if crankshaft is
> > > enabled.
> >
> > The CPU features detection is done parsing the /proc/cpuinfo file:
> >
> > https://code.google.com/p/v8/source/browse/branches/bleeding_edge/src...
> >
> > So, if your cpu information is not exposed correctly the detection will fail.
> >
> > I'd never seen an ARM with NEON but without VFPv3.
> >
> > vmjl
> >
> >
> >
> >
> >
> >
> >
> >
> >
> > > I hope this helps.
> >
> > > Regards,
> >
> > > Alexandre
> >
> > > On Sat, Apr 2, 2011 at 7:40 AM, Arun M <arunm.chr...@gmail.com> wrote:
> >
> > > > Hi
> >
> > > > Is Crankshaft optimizing compiler enabled for ARMv7-A + NEON devices
> > > > which does not have VFPv3 FPU?
> >
> > > > Thanks & Regards
> > > > Arun
> >
> > > > On Mar 9, 6:02 pm, S�ren Gjesse <sgje...@chromium.org> wrote:
> > > > > For ARM crankshaft is now the default. This change is in the repository
> > > > > starting from V8 version 3.2. To use the previous optimizing compiler
> > > > > --nocrankshaft will have to be used. When crankshaft for ARM has been
> > > > fully
> > > > > stabilized the previous optimizing compiler will be removed from the
> > > > > repository and running with --nocrankshaft will no longer be possible.
> > > > There
> > > > > is no specific date to when this will happen but most likely it will be
> > > > > within a month or two. The removal of the previous optimizing compiler
> > > > will
> > > > > happen for all supported platforms simultaneously,
> >
> > > > > The previous optimizing compiler can of cause still be found in previous
> > > > > versions of V8.
> >
> > > > > Regards,
> > > > > S�ren
> >
> > > > > On Tue, Mar 8, 2011 at 20:05, Hugo Vincent <hugo.vinc...@gmail.com>
> > > > wrote:
> > > > > > How much slower is full-compiler than nocrankshaft on ARM926ej-s -
> > > > > > anyone have any benchmarks? I'm hesitant to invest time in using V8
> > > > > > for my project if it's going to get substantially slower soon. Is
> > > > > > there any estimated time frame for when nocrankshaft will be
> > > > > > deprecated?
> >
> > > > > > Thanks,
> > > > > > Hugo
> >
> > > > > > On Feb 23, 9:14 pm, S�ren Gjesse <sgje...@chromium.org> wrote:
> > > > > > > Just a follow-up note regarding the new optimizing compiler
> > > > (crankshaft).
> > > > > > > This will be enabled by default for ARM quite soon, and the existing
> > > > > > > optimizing compiler will be removed at some point. For non ARMv7+VFP
> > > > > > devices
> > > > > > > this means that the base JIT (non-optimizing/full-compiler) will be
> > > > used.
> > > > > > To
> > > > > > > measure the different compilers on a ARMv7+VFP device use following
> > > > > > options:
> >
> > > > > > > � --nocrankshaft (current optimizing JIT - the current default)
> > > > > > > � --crankshaft (new optimizing JIT - the soon to be default)
> > > > > > > � --always-full-compiler (base/non-optimizing compiler)
> >
> > > > > > > Going forward using --crankshaft on a non ARMv7+VFP device will have
> > > > no
> > > > > > > effect and execution will fallback to --always-full-compiler.
> >
> > > > > > > Regards,
> > > > > > > S�ren
Rodolph Perfetta
On Wed, Apr 06, 2011 at 12:00:27AM -0700, Arun M wrote:Sorry Arun, I don't know. I'm not familiar with chrome in general.
> Hi Victor
>
> Will this method of detecting VFPv3 work in chromium browser when
> sandbox is enabled?
> Will browser sandbox allow to fopen "/proc/cpuinfo" file?
vmjl
>
> Arun
> > > > > S�ren
> > > > > > > S�ren
Arun M
I ran with --js-flags="--crankshaft" and --js-flags="--nocrankshaft"
options.
But the V8 scores are nearly same.
V8::DisableCrankshaft() function is called from cpu-arm.cc.
I think Chromium is not able to parse /proc/cpuinfo file and get VFPv3
information.
Arun
On Apr 6, 6:31 pm, Rodolph Perfetta <rodolph.perfe...@gmail.com>
wrote:
>
> Have you tried running Chrome with the options Soren mentioned? The result
> should tell you if crankshaft is enabled or not.
>
> Rodolph.
>
Søren Gjesse
These are interesting experiments, and it would be great to get ot the bottom of this.
I assume that /proc/cpuinfo contains the strings "ARMv7" and "vfpv3", as that is the exact strings which are checked for. You could also try to run chrome with the --no-sandbox option to see whether that makes a diffrence. V8 has a number of options which can be used to trace the optimizations performed by crankshaft, try --js-flags="--trace-opt --trace-deopt --trace-osr --trace-bailout" for the full monty.
If you know that VFPv3 is available you could try to hack the CPU detection in OS:::ArmCpuHasFeature (platform-linux.cc) always return true for VFP3.
Regards,
Søren
Arun M
I tried all options.
When i disable sandbox or hack the CPU detection
in OS:::ArmCpuHasFeature and run v8 benchmark,
the browser tab crashes.
I am trying to find the issue.
Regards
Arun
Rodolph Perfetta
Arun M
I updated the chromium browser version to 12.0.742.0.
V8 score has almost doubled after the hack in OS:::ArmCpuHasFeature.
Thanks & Regards
Arun
On Apr 13, 4:46 pm, Rodolph Perfetta <rodolph.perfe...@gmail.com>
wrote:
> /proc/cpuinfo)
>
> Rodolph.
>
>
> read more »