Thank you for the feedback. It is great to hear that you now get the performance benefit of V8 Crankshaft on ARM. We are working on a better way of detecting VFPv3 to correctly enable Crankshaft.
Regards, Søren On Wed, Apr 27, 2011 at 11:23, Arun M <[email protected]> wrote: > Hi > > 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 <[email protected]> > wrote: > > Which platform are you running on? Do you know which CPU? (maybe send > your > > /proc/cpuinfo) > > > > Rodolph. > > > > On 13 April 2011 06:24, Arun M <[email protected]> wrote: > > > > > > > > > > > > > > > > > Hi > > > > > 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 > > > > > On Apr 7, 4:22 pm, Søren Gjesse <[email protected]> wrote: > > > > Arun, > > > > > > 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 > > > > > > On Thu, Apr 7, 2011 at 03:25, Arun M <[email protected]> wrote: > > > > > Hi > > > > > > > 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 <[email protected]> > > > > > wrote: > > > > > > Arun, > > > > > > > > Have you tried running Chrome with the options Soren mentioned? > The > > > > > result > > > > > > should tell you if crankshaft is enabled or not. > > > > > > > > Rodolph. > > > > > > > > 2011/4/6 Víctor M. Jáquez L. <[email protected]> > > > > > > > > > On Wed, Apr 06, 2011 at 12:00:27AM -0700, Arun M wrote: > > > > > > > > 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, Víctor M. Jáquez L. <[email protected]> > > > 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 < > > > [email protected]> > > > > > > > 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 < > [email protected]> > > > > > 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 < > > > > > > > [email protected]> > > > > > > > > > > > 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 < > > > [email protected]> > > > > > > > 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 > > > > > > > > > > > > > > > > On Wed, Feb 23, 2011 at 18:33, Rodolph Perfetta > > > > > > > > > > > > > > <[email protected]>wrote: > > > > > > > > > > > > > > > > > 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. > > > > > > > > > > > > > > > > > On 23 February 2011 17:12, Hugo Vincent < > > > > > > > [email protected]> > > > > > > > > > > > > > wrote: > > > > > > > > > > > > > > > > >> Hi, > > > > > > > > > > > > > > > > >> 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 > > > > > > > > > > > > > > > > >> -- > > > > > > > > > > > > > > >> 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 > > > > > > > > > -- > > > > ... > > > > read more » > > -- > 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
