• Locked Locked
  • Replies 2 replies
  • Subscribers 136 subscribers
  • Views 3653 views
  • Users 0 members are here
Options
Related
How was your experience today?
This discussion has been locked.
You can no longer post new replies to this discussion. If you have a question you can start a new discussion

Renderscript on Nexus 10

Bastian Zuehlke
Note: This was originally posted on 23rd January 2013 at http://forums.arm.com

Hi,

i am very much interested in GPGPU on mobile devices.
According to Google Android 4.2 on Nexus 10 is supporting Renderscript running on GPU.


Over the past weeks we have implemented a benchmark with quite a number of filters running on .rs , .fs, using rs intrinsic and on OpenGL|ES.
We got best results (on Nexus 10) from either Intrinsic or OpenGL|ES. Pure .rs seems not to run on GPU and .fs is very limited.

The Renderscript documentation is kind of weak when it comes to Renderscript, in particular Filterscript.

Is  there a kind of "How to write Renderscript for Mali T-604" document available ?
Any further details would be very much appreciated.

BR Bastian
Parents
  • Note: This was originally posted on 20th February 2013 at http://forums.arm.com

    Hi Chris,

    as stated in http://developer.android.com/about/versions/jelly-bean.html .


    GPU Compute
    [color=#222222][font=Roboto, sans-serif][size=4]Renderscript Compute is the first computation platform ported to run directly on a mobile device GPU. It now automatically takes advantage of GPU computation resources whenver possible to improve performance. With GPU integration, even the most complex computations for graphics or image processing can execute with dramatically improved performance.[/size][/font][/color]

    [color=#222222][font=Roboto, sans-serif][size=4]Any app using Renderscript on a supported device can benefit immediately from this GPU integration, without recompiling. The Nexus 10 tablet is the first device to support this integration.[/size][/font][/color]

    As I stated we are currently implementing several algorithms (CV based) by using OGLES (GLSL), RS, FS and Intrinsic Functions like [font=Monaco][size=2]ScriptIntrinsicConvolve3x3.[/size][/font]

    Please tell me, in your understanding are the following Filterscripts actually be performed on Mali T-604 GPU ?

    If not, what do we need to do ?

    [font=Courier][size=2]uchar4 __attribute__((kernel)) root(const uchar4 in)[/size][/font]

    [font=Courier][size=2]{ [/size][/font]

    [font=Courier][size=2] return in;[/size][/font]

    [font=Courier][size=2]}[/size][/font]



    or

    [font=Courier][size=2]uchar4 __attribute__((kernel)) root(uint32_t x, uint32_t y) [/size][/font]

    [font=Courier][size=2]{ [/size][/font]

    [font=Courier][size=2] float4 res;[/size][/font]

    [font=Courier][size=2] res.x = x * (1.0f/512.0f);[/size][/font]

    [font=Courier][size=2] res.y = y * (1.0f/512.0f);[/size][/font]

    [font=Courier][size=2] res.z = 0.0f;[/size][/font]

    [font=Courier][size=2] res.w = 1.0f;[/size][/font]

    [font=Courier][size=2] [/size][/font]

    [font=Courier][size=2] return rsPackColorTo8888( res );      [/size][/font]

    [font=Courier][size=2]}[/size][/font]

    [font=Courier][size=2]
    [/size][/font]

    [font=Courier][size=2]
    [/size][/font]






Reply
  • Note: This was originally posted on 20th February 2013 at http://forums.arm.com

    Hi Chris,

    as stated in http://developer.android.com/about/versions/jelly-bean.html .


    GPU Compute
    [color=#222222][font=Roboto, sans-serif][size=4]Renderscript Compute is the first computation platform ported to run directly on a mobile device GPU. It now automatically takes advantage of GPU computation resources whenver possible to improve performance. With GPU integration, even the most complex computations for graphics or image processing can execute with dramatically improved performance.[/size][/font][/color]

    [color=#222222][font=Roboto, sans-serif][size=4]Any app using Renderscript on a supported device can benefit immediately from this GPU integration, without recompiling. The Nexus 10 tablet is the first device to support this integration.[/size][/font][/color]

    As I stated we are currently implementing several algorithms (CV based) by using OGLES (GLSL), RS, FS and Intrinsic Functions like [font=Monaco][size=2]ScriptIntrinsicConvolve3x3.[/size][/font]

    Please tell me, in your understanding are the following Filterscripts actually be performed on Mali T-604 GPU ?

    If not, what do we need to do ?

    [font=Courier][size=2]uchar4 __attribute__((kernel)) root(const uchar4 in)[/size][/font]

    [font=Courier][size=2]{ [/size][/font]

    [font=Courier][size=2] return in;[/size][/font]

    [font=Courier][size=2]}[/size][/font]



    or

    [font=Courier][size=2]uchar4 __attribute__((kernel)) root(uint32_t x, uint32_t y) [/size][/font]

    [font=Courier][size=2]{ [/size][/font]

    [font=Courier][size=2] float4 res;[/size][/font]

    [font=Courier][size=2] res.x = x * (1.0f/512.0f);[/size][/font]

    [font=Courier][size=2] res.y = y * (1.0f/512.0f);[/size][/font]

    [font=Courier][size=2] res.z = 0.0f;[/size][/font]

    [font=Courier][size=2] res.w = 1.0f;[/size][/font]

    [font=Courier][size=2] [/size][/font]

    [font=Courier][size=2] return rsPackColorTo8888( res );      [/size][/font]

    [font=Courier][size=2]}[/size][/font]

    [font=Courier][size=2]
    [/size][/font]

    [font=Courier][size=2]
    [/size][/font]






Children
No data