cc : Jacky he experienced some performance issues with Canvas

On Thu, Nov 28, 2013 at 12:43 PM, Hwang, Dongseong <
dongseong.hw...@intel.com> wrote:

> FW to mailing list
>
> ________________________________
> From: Hwang, Dongseong
> Sent: Thursday, November 28, 2013 12:51 PM
> To: Kondapally, Kalyan; Poussa, Sakari; Syrjala, Ilkka; Pozdnyakov,
> Mikhail; Christiansen, Kenneth R; Nikkanen, Kimmo; Saarinen, Jani;
> Balestrieri, Francesco; Menard, Alexis
> Subject: RE: Enable Accelerated Canvas 2D
>
> I've investigated A2DC in both Android and Tizen.
>
> A2DC is already enabled in Android.
>
> When it comes to Tizen, performace is improved by 2~3 times!!!
> Now, Android and Tizen has almost same performance.
>
> I'll apply it to Crosswalk soon.
>
>
>
> In technical detail,
> Current Tizen Crosswalk uses threaded compositing + software canvas.
>
> I measure four combinations.
> The result perfomance: threaded compositing + A2DC > A2DC >>>>> sw >
> threaded compositing + sw
>
> 1. http://www.craftymind.com/factory/guimark3/vector/GM3_JS_Vector.html
> sw: 17.99
> threaded comp + sw: 14.91
> A2DC : 37.52
> threaded comp + A2DC: 44.82
> 2. http://www.craftymind.com/factory/guimark3/bitmap/GM3_JS_Bitmap.html
> sw: 20.51
> threaded comp + sw: 17.16
> A2DC : 49.98
> threaded comp + A2DC: 54.58
> 3. http://www.craftymind.com/factory/guimark3/compute/GM3_JS_Compute.html
> sw: 20.98
> threded comp + sw: 19.36
> A2DC : 35.47
> threaded comp + A2DC: 39.18
>
> Let's analysis the result,
> sw is faster than threaded compositing + sw, because threaded compositing
> + sw needs to copy bitmap one more time.
> A2DC is pretty faster than software canvas, because of GPU.
> threaded comp + A2DC is faster than A2DC, because 1) both need to copy
> bitmap the same time. 2) replaying in compositor thread reduces Blink
> thread workload.
>
> more detail: https://crosswalk-project.org/jira/browse/XWALK-73
>
> - DS
>
> ________________________________
> From: Kondapally, Kalyan
> Sent: Wednesday, November 27, 2013 5:29 PM
> To: Hwang, Dongseong; Poussa, Sakari; Syrjala, Ilkka; Pozdnyakov, Mikhail;
> Christiansen, Kenneth R; Nikkanen, Kimmo; Saarinen, Jani; Balestrieri,
> Francesco; Menard, Alexis
> Subject: RE: Enable Accelerated Canvas 2D
>
> >> It would be interesting to see the performance improvements with this
> enabled.
>
> I mean benchmarking Canvas2D performance with current stack vs accelerated
> Canvas 2D support
>
> Br,
> Kalyan
>
>
> From: Kondapally, Kalyan
> Sent: Wednesday, November 27, 2013 4:19 PM
> To: Hwang, Dongseong; Poussa, Sakari; Syrjala, Ilkka; Pozdnyakov, Mikhail;
> Christiansen, Kenneth R; Nikkanen, Kimmo; Saarinen, Jani; Balestrieri,
> Francesco; Menard, Alexis
> Subject: RE: Enable Accelerated Canvas 2D
>
> Right, Canvas2D doesn’t support restoring a lost context. Do you know if
> the details have been moved to some Bug or where the specification is being
> finalized ? (I couldn’t find any reference to either).  It would be
> interesting to see the performance improvements with this enabled.
>
>
> Reg blacklisting of Intel gpu’s:
>
> This list can be found in gpu/config/software_rendering_list.json.  I
> don’t know  reasons behind the blacklisting though.
>
> br,
> Kalyan
>
> From: Hwang, Dongseong
> Sent: Wednesday, November 27, 2013 3:01 PM
> To: Kondapally, Kalyan; Poussa, Sakari; Syrjala, Ilkka; Pozdnyakov,
> Mikhail; Christiansen, Kenneth R; Nikkanen, Kimmo; Saarinen, Jani;
> Balestrieri, Francesco; Menard, Alexis
> Subject: RE: Enable Accelerated Canvas 2D
>
> According to
> https://groups.google.com/a/chromium.org/forum/?fromgroups#!topic/graphics-dev/CQJXpXxO6dk
> , they disable accelerated 2d canvas by default, because when gpu process
> down there is no way to recover it.
>
> It means that turning on A2DC in Android is exception case. They turn on
> it blindly. It makes sense because Android can not recover GPU process
> because Android always run GPU process with --single-process or
> --in-process-gpu. So just turn on A2DC and then if GPU has some issues,
> just crash chrome android.
>
> btw, what's mean about "on Certain Intel chipsets"? could you share about
> it? it there certain code about Intel GPU or do you have interesting
> experience related to Intel GPU?
>
> - DS
> ________________________________
> From: Kondapally, Kalyan
> Sent: Wednesday, November 27, 2013 2:15 PM
> To: Hwang, Dongseong; Poussa, Sakari; Syrjala, Ilkka; Pozdnyakov, Mikhail;
> Christiansen, Kenneth R; Nikkanen, Kimmo; Saarinen, Jani; Balestrieri,
> Francesco; Menard, Alexis
> Subject: Enable Accelerated Canvas 2D
> Changed the subject of email.
>
> HI DS,
>
> K, thanks for the information. Did we already identify the reasons as to
> why Accelerated 2D Canvas is blacklisted on Linux and on Certain Intel
> chipsets. This is would be very valuable information to evaluate if we need
> any support from the core drivers team (Mesa and SGX). Do we have a list
> somewhere ?
>
> Br,
> Kalyan
>
> From: Hwang, Dongseong
> Sent: Wednesday, November 27, 2013 11:39 AM
> To: Poussa, Sakari; Kondapally, Kalyan; Syrjala, Ilkka; Pozdnyakov,
> Mikhail; Christiansen, Kenneth R; Nikkanen, Kimmo; Saarinen, Jani;
> Balestrieri, Francesco; Menard, Alexis
> Subject: RE: Visualization and Web update
>
> Thank you for your opinion. I file accelerated 2d canvas in
> https://crosswalk-project.org/jira/browse/XWALK-531
>
> Xwalk-73 is originated from
> https://github.com/crosswalk-project/crosswalk/issues/806 . Unfotunately,
> all old github issues perhaps were removed.
> In the old bug, we understood texture uploading affects performance badly,
> and accelerated 2d canvas can be solution.
>
> Android as well as Linux needs to enable Accelerated 2d canvas. I'll do it
> soon in XWALK-531.
>
> - DS
> ________________________________
> From: Poussa, Sakari
> Sent: Wednesday, November 27, 2013 9:26 AM
> To: Kondapally, Kalyan; Syrjala, Ilkka; Pozdnyakov, Mikhail; Christiansen,
> Kenneth R; Nikkanen, Kimmo; Saarinen, Jani; Balestrieri, Francesco; Menard,
> Alexis; Hwang, Dongseong
> Subject: Re: Visualization and Web update
> +DS: Any comments ?
>
> From: <Kondapally>, Kalyan <kalyan.kondapa...@intel.com<mailto:
> kalyan.kondapa...@intel.com>>
> Date: Wednesday, November 27, 2013 at 7:21
> To: Sakari Poussa <sakari.pou...@intel.com<mailto:sakari.pou...@intel.com>>,
> "Syrjala, Ilkka" <ilkka.syrj...@intel.com<mailto:ilkka.syrj...@intel.com>>,
> "Pozdnyakov, Mikhail" <mikhail.pozdnya...@intel.com<mailto:
> mikhail.pozdnya...@intel.com>>, "Christiansen, Kenneth R" <
> kenneth.r.christian...@intel.com<mailto:kenneth.r.christian...@intel.com>>,
> "Nikkanen, Kimmo" <kimmo.nikka...@intel.com<mailto:
> kimmo.nikka...@intel.com>>, "Saarinen, Jani" <jani.saari...@intel.com
> <mailto:jani.saari...@intel.com>>, "Balestrieri, Francesco" <
> francesco.balestri...@intel.com<mailto:francesco.balestri...@intel.com>>,
> "Menard, Alexis" <alexis.men...@intel.com<mailto:alexis.men...@intel.com>>
> Subject: RE: Visualization and Web update
>
> Regarding Xwalk-73:
>
> I see that Accelerated Canvas2D is disabled on all Linux platforms and in
> some Intel chipsets specifically. I think the best solution for this, is to
> identify those issues and fix them.
>
> Back-up solutions:
> 1)Support for Zero copy would help here.
> Jani, Francesco:  Do you already have the internal Tizen Jira task related
> to SGX drivers study from Tapani. It might be good to add it to the list of
> items to be tracked by Cross-Walk on Tizen-Mobile.
>
> 2)Texture upload speed. Depending on the actual problem, this might show
> some immediate improvements.
>
> Reg 258:
>
> This is something we need Mobica team to have a look and determine why
> shader compilation fails.
>
> Br,
> Kalyan
>
> From: Poussa, Sakari
> Sent: Tuesday, November 26, 2013 2:27 PM
> To: Syrjala, Ilkka; Pozdnyakov, Mikhail; Kondapally, Kalyan; Christiansen,
> Kenneth R; Nikkanen, Kimmo; Saarinen, Jani; Balestrieri, Francesco; Menard,
> Alexis
> Subject: Re: Visualization and Web update
>
> https://crosswalk-project.org/jira/browse/XWALK-73
> https://crosswalk-project.org/jira/browse/XWALK-258
>
> Good candidates for co-ops between the teams.
>
> From: ilkka.syrj...@intel.com<mailto:ilkka.syrj...@intel.com>
> When: 14:00 - 15:00 November 26, 2013
> Subject: Visualization and Web update
> Location: Lync Meeting/Saana
>
>
> All,
>
> Let’s start a bi-weekly meetings to share information regarding what is
> happening in GFX and Web. The goal is to talk about the latest news and
> efforts from individual contributors to programs, share news, identify what
> other companies might be contributing in the future and find synergies
> between efforts.
>
> Please feel free to send this invitation forwards to those who might be
> interested to join in the talks.
>
> Br,
> Ilkka
>
>
> .........................................................................................................................................
> -->Join Lync Meeting<https://meet.intel.com/ilkka.syrjala/2HL9KB0B>
>
> Join by phone
> +1(916)356-2663 (or your local bridge access #) Choose bridge
> 5.<tel:+1(916)356-2663%20(or%20your%20local%20bridge%20access%20#)%20Choose%20bridge%205.>
> (Global)                English (United States)
> Find a local number<https://dial.intel.com>
>
> Conference ID: 725036066
>
> Forgot your dial-in PIN?<https://dial.intel.com>|Help<
> http://o15.officeredir.microsoft.com/r/rlidLync15?clid=1033&p1=5&p2=2009>
>
> [!OC([1033])!]
>
> .........................................................................................................................................
>
> Lync Bridge users: Click here<http://goto/lyncdisclaimer> if you don’t
> want your phone number to be displayed. Recordings are subject to the
> Audio/Video Recording Policy<
> http://itdoclib.intel.com/getdoc.asp?id=112152>.
> ---------------------------------------------------------------------
> Intel Finland Oy
> Registered Address: PL 281, 00181 Helsinki
> Business Identity Code: 0357606 - 4
> Domiciled in Helsinki
>
> This e-mail and any attachments may contain confidential material for
> the sole use of the intended recipient(s). Any review or distribution
> by others is strictly prohibited. If you are not the intended
> recipient, please contact the sender and delete all copies.
>
> _______________________________________________
> Crosswalk-dev mailing list
> Crosswalk-dev@lists.crosswalk-project.org
> https://lists.crosswalk-project.org/mailman/listinfo/crosswalk-dev
>



-- 
Alexis Menard
Software Engineer @
Intel Open Source Technology Center

Intel Semiconductores do Brasil Ltda.
Ave Dr. Chucri Zaidan, 940, Brooklin, 10 Andar
04583-904 São Paulo, SP
Brazil

This e-mail and any attachments may contain confidential material for the
sole use of the intended recipient(s). Any review or distribution by others
is strictly prohibited. If you are not the intended recipient, please
contact the sender and delete all copies.
_______________________________________________
Crosswalk-dev mailing list
Crosswalk-dev@lists.crosswalk-project.org
https://lists.crosswalk-project.org/mailman/listinfo/crosswalk-dev

Reply via email to