Welcome, Unregistered.

If this is your first visit, be sure to check out the FAQ by clicking the link above. You may have to register before you can post: click the register link above to proceed. To start viewing messages, select the forum that you want to visit from the selection below.

Reply
Old 03-May-2010, 05:59   #1
cho
Member
 
Join Date: Feb 2002
Posts: 410
Default ATI Stream SDK v2.1 include image object support

http://developer.amd.com/gpu/ATIStre...fault.aspx#one

What’s New in v2.1
  • Support for openSUSE™ 11.2 and Red Hat® Enterprise Linux® 5.4.
  • Support for OpenCL™ / OpenGL® interoperability.
  • Support for OpenCL™ byte addressable stores3,4.
  • Support for OpenCL™ images3.
  • Extension: Support for double-precision floating point basic arithmetic in OpenCL™ C kernels.
  • Extension: Support for AMD media operations in OpenCL™.
  • Extension: Support for device fission in OpenCL™ 4.
  • Extension: Support for device attribute queries in OpenCL™.
  • Preview Feature: Support forbinary OpenCL™ kernels.
  • Additional OpenCL™ samples:
    • HistogramAtomics
    • MatrixMulDouble (under cpp_cl)
    • MatrixMulImage
    • SimpleGL
    • SimpleImage
    • SobelFilterImage (under cpp_cl)
    • URNGNoiseGL
  • Stream KernelAnalyzer 1.5 installer now bundled with the ATI Stream SDK v2.1.
  • Various OpenCL™ compiler and runtime fixes and enhancements (see developer release notes for more details).
  • Support for new hardware:
    • ATI Radeon™ HD 5830 GPU
    • ATI Radeon™ HD 5450 GPU
    • ATI FirePro™ V8800 GPU
    • ATI FirePro™ V7800 GPU
    • ATI FirePro™ V5800 GPU
    • ATI FirePro™ V4800 GPU
    • ATI FirePro™ V3800 GPU
    • ATI Mobility Radeon™ HD 5800 Series GPUs
    • ATI Mobility Radeon™ HD 5700 Series GPUs
    • ATI Mobility Radeon™ HD 5600 Series GPUs
    • ATI Mobility Radeon™ HD 5400 Series GPUs
    • ATI FirePro M7820™ GPU
    • ATI FirePro M5800™ GPU
cho is offline   Reply With Quote
Old 03-May-2010, 07:14   #2
rpg.314
Senior Member
 
Join Date: Jul 2008
Location: /
Posts: 4,218
Send a message via Skype™ to rpg.314
Default

What is this device fission? Can't find a thing on Khronos API registry.
rpg.314 is offline   Reply With Quote
Old 03-May-2010, 08:50   #3
Dade
Member
 
Join Date: Dec 2009
Posts: 182
Default

Quote:
Originally Posted by cho View Post
  • Extension: Support for device fission in OpenCL™ 4.
OpenCL 4 ? May be a typo and they refer to OpenGL 4 ?

I was wondering too what device fission is
Dade is offline   Reply With Quote
Old 03-May-2010, 09:30   #4
denev2004
Member
 
Join Date: Apr 2010
Location: China
Posts: 143
Send a message via MSN to denev2004 Send a message via Skype™ to denev2004
Default

While AMD is giving its new support to Red Hat, NVIDIA is trying to stop Linux Driver Service = =
And why they put of DP?That's bad....
denev2004 is offline   Reply With Quote
Old 03-May-2010, 10:02   #5
rpg.314
Senior Member
 
Join Date: Jul 2008
Location: /
Posts: 4,218
Send a message via Skype™ to rpg.314
Default

Quote:
Originally Posted by Dade View Post
OpenCL 4 ? May be a typo and they refer to OpenGL 4 ?
No, it's a reference number in the original material.
Quote:

I was wondering too what device fission is
Split a multicore-CPU into 2 or more opencl cpu devices?

Why would you need such a thing?
rpg.314 is offline   Reply With Quote
Old 03-May-2010, 10:10   #6
Dade
Member
 
Join Date: Dec 2009
Posts: 182
Default

Quote:
Originally Posted by rpg.314 View Post
Split a multicore-CPU into 2 or more opencl cpu devices?

Why would you need such a thing?
I use AMD OpenCL CPU device only as debugging/developing tool (native code is always faster). This would be a very handy feature to emulate the presence of multiple OpenCL devices.

It would be a quite nice feature to have for testing/debugging.
Dade is offline   Reply With Quote
Old 03-May-2010, 10:50   #7
Jawed
Regular
 
Join Date: Oct 2004
Location: London
Posts: 9,948
Send a message via Skype™ to Jawed
Default

Blimey, that's more like v3 than v2.1. EDIT: hmm, the DP support is still "preview".

Fission might allow treating the cores of a CPU separately. Older versions of the SDK take over all of a CPU's cores I believe. With fission the programmer can presumably keep a host core (or more) to himself, unaffected by OpenCL device allocation.

Also, presumably, this means that task parallelism across CPU cores can be more finely-grained.

Jawed
__________________
Can it play WoW?
Jawed is offline   Reply With Quote
Old 03-May-2010, 11:26   #8
pcchen
Moderator
 
Join Date: Feb 2002
Location: Taiwan
Posts: 2,467
Default

Quote:
Originally Posted by Jawed View Post
Blimey, that's more like v3 than v2.1. EDIT: hmm, the DP support is still "preview".
I'm not sure about this preview thing. They have an extension for DP support which has reduced functionalities than the official Khronos extension. Are they going to continue using this or somehow transfer to the official extension in the future?
pcchen is offline   Reply With Quote
Old 03-May-2010, 11:42   #9
Jawed
Regular
 
Join Date: Oct 2004
Location: London
Posts: 9,948
Send a message via Skype™ to Jawed
Default

I presume they can't expose an official extension until the extension is entirely supported, hence "preview": a subset of the official extension's features.
__________________
Can it play WoW?
Jawed is offline   Reply With Quote
Old 03-May-2010, 12:13   #10
Dade
Member
 
Join Date: Dec 2009
Posts: 182
Default

Quote:
Originally Posted by Jawed View Post
Fission might allow treating the cores of a CPU separately. Older versions of the SDK take over all of a CPU's cores I believe. With fission the programmer can presumably keep a host core (or more) to himself, unaffected by OpenCL device allocation.
It was possible to set the amount of threads used by OpenCL CPU device in previous SDK by defining the CPU_MAX_COMPUTE_UNITS env. variable.

May be it is now possible to control this behaviour with an OpenCL extension.
Dade is offline   Reply With Quote
Old 03-May-2010, 13:49   #11
Arnold Beckenbauer
Senior Member
 
Join Date: Oct 2006
Location: Germany
Posts: 1,004
Default

Great: No image support for HD4000 series. Well done!
__________________
Hail Brothers and Sisters! Coranon Silaria, Ozoo Mahoke
Eta Kooram Nah Smech!

Find Chuck Norris.
Arnold Beckenbauer is offline   Reply With Quote
Old 04-May-2010, 07:57   #12
fellix
Senior Member
 
Join Date: Dec 2004
Location: Varna, Bulgaria
Posts: 2,989
Send a message via Skype™ to fellix
Default

Quote:
Originally Posted by Jawed View Post
Blimey, that's more like v3 than v2.1. EDIT: hmm, the DP support is still "preview".

Fission might allow treating the cores of a CPU separately. Older versions of the SDK take over all of a CPU's cores I believe. With fission the programmer can presumably keep a host core (or more) to himself, unaffected by OpenCL device allocation.

Also, presumably, this means that task parallelism across CPU cores can be more finely-grained.
Quote:
OpenCL extension support for device fission in OpenCL, enabling developers to sub-divide an OpenCL device and allowing multiple work kernels to be assigned to that device.
__________________
Apple: China -- Brutal leadership done right.
Google: United States -- Somewhat democratic.
Microsoft: Russia -- Big and bloated.
Linux: EU -- Diverse and broke.
fellix is offline   Reply With Quote
Old 04-May-2010, 09:15   #13
Dade
Member
 
Join Date: Dec 2009
Posts: 182
Default

From ATI SDK cl_ext.h:

Code:
typedef CL_API_ENTRY cl_int (CL_API_CALL * clCreateSubDevicesEXT_fn)(
    cl_device_id     /* in_device */,
    const cl_device_partition_property_ext * /* partition_properties */,
    cl_uint          /* num_entries */,
    cl_device_id *   /* out_devices */,
    cl_uint *        /* num_devices */);
It looks like you can split an OpenCL in multiple instances.

This define was interesting:

/* cl_affinity_domain_ext */
#define CL_AFFINITY_DOMAIN_NUMA_EXT 0x1
#define CL_AFFINITY_DOMAIN_L4_CACHE_EXT 0x2
#define CL_AFFINITY_DOMAIN_L3_CACHE_EXT 0x3
#define CL_AFFINITY_DOMAIN_L2_CACHE_EXT 0x4
#define CL_AFFINITY_DOMAIN_L1_CACHE_EXT 0x5
#define CL_AFFINITY_DOMAIN_NEXT_FISSIONABLE_EXT 0x6

Are they using OpenCL on some big CPU/NUMA cluster ?
Dade is offline   Reply With Quote
Old 04-May-2010, 10:23   #14
fellix
Senior Member
 
Join Date: Dec 2004
Location: Varna, Bulgaria
Posts: 2,989
Send a message via Skype™ to fellix
Default

AMD has demoed an internal build of Powder Toy, running parallel across 24 CPU cores in a 4S system while ago, using x86 OCL run-time.
__________________
Apple: China -- Brutal leadership done right.
Google: United States -- Somewhat democratic.
Microsoft: Russia -- Big and bloated.
Linux: EU -- Diverse and broke.
fellix is offline   Reply With Quote
Old 04-May-2010, 11:04   #15
Jawed
Regular
 
Join Date: Oct 2004
Location: London
Posts: 9,948
Send a message via Skype™ to Jawed
Default

Note the original post links the wrong documents, e.g. this is the correct version (1.0d, not 1.0c as described on the SDK Documentation page) of the OpenCL Programming Guide:

http://developer.amd.com/gpu/ATIStre...ming_Guide.pdf

Documentation page:

http://developer.amd.com/gpu/ATIStre...mentation.aspx

Table A.1 indicates that Fission is not available on any GPU.

That section of the document also describes the media extensions, including the pack/unpack/SAD stuff.

Jawed
__________________
Can it play WoW?
Jawed is offline   Reply With Quote
Old 04-May-2010, 11:16   #16
rpg.314
Senior Member
 
Join Date: Jul 2008
Location: /
Posts: 4,218
Send a message via Skype™ to rpg.314
Default

Where have all the extensions gone? All of them aren't here.
rpg.314 is offline   Reply With Quote
Old 08-May-2010, 10:49   #17
Lightman
Senior Member
 
Join Date: Jun 2008
Location: Torquay, UK
Posts: 1,096
Default

Quote:
Originally Posted by Dade View Post
From ATI SDK cl_ext.h:

Code:
typedef CL_API_ENTRY cl_int (CL_API_CALL * clCreateSubDevicesEXT_fn)(
    cl_device_id     /* in_device */,
    const cl_device_partition_property_ext * /* partition_properties */,
    cl_uint          /* num_entries */,
    cl_device_id *   /* out_devices */,
    cl_uint *        /* num_devices */);
It looks like you can split an OpenCL in multiple instances.

This define was interesting:

/* cl_affinity_domain_ext */
#define CL_AFFINITY_DOMAIN_NUMA_EXT 0x1
#define CL_AFFINITY_DOMAIN_L4_CACHE_EXT 0x2
#define CL_AFFINITY_DOMAIN_L3_CACHE_EXT 0x3
#define CL_AFFINITY_DOMAIN_L2_CACHE_EXT 0x4
#define CL_AFFINITY_DOMAIN_L1_CACHE_EXT 0x5
#define CL_AFFINITY_DOMAIN_NEXT_FISSIONABLE_EXT 0x6

Are they using OpenCL on some big CPU/NUMA cluster ?
Looks like AMD is preparing for FUSION CPU/APU launch!
Lightman is offline   Reply With Quote
Old 14-May-2010, 14:02   #18
MfA
Regular
 
Join Date: Feb 2002
Posts: 5,520
Send a message via ICQ to MfA
Default

With those binary OpenCL kernels would it be possible to hack inline assembly now? (Do subroutine calls in OpenCL kernels get compiled to actual CALLs?)
MfA is offline   Reply With Quote
Old 14-May-2010, 14:58   #19
Jawed
Regular
 
Join Date: Oct 2004
Location: London
Posts: 9,948
Send a message via Skype™ to Jawed
Default

Quote:
Originally Posted by MfA View Post
(Do subroutine calls in OpenCL kernels get compiled to actual CALLs?)
Usually subroutines are in-lined.

The ISA has CALL (since R600) with a stack (32 level) and supports jump tables (since Evergreen) of up to 256 elements. Never seen these in ISA code I've looked at, though I haven't tried to generate the latter.
__________________
Can it play WoW?
Jawed is offline   Reply With Quote
Old 14-May-2010, 17:07   #20
MfA
Regular
 
Join Date: Feb 2002
Posts: 5,520
Send a message via ICQ to MfA
Default

Shame, would make patching a lot easier.
MfA is offline   Reply With Quote
Old 14-May-2010, 17:17   #21
Florin
Merrily dodgy
 
Join Date: Aug 2003
Location: The colonies
Posts: 1,432
Default

Quote:
Originally Posted by Lightman View Post
Looks like AMD is preparing for FUSION CPU/APU launch!
The Future is Fission.

whehe
__________________
"A man generally has two reasons for doing a thing. One that sounds good, and a real one." - J.P. Morgan
Florin is offline   Reply With Quote

Reply

Thread Tools
Display Modes

Posting Rules
You may not post new threads
You may not post replies
You may not post attachments
You may not edit your posts

BB code is On
Smilies are On
[IMG] code is On
HTML code is Off

Forum Jump


All times are GMT +1. The time now is 21:44.


Powered by vBulletin® Version 3.8.6
Copyright ©2000 - 2014, Jelsoft Enterprises Ltd.