cancel
Showing results for 
Search instead for 
Did you mean: 

Archives Discussions

bpurnomo
Staff

Stream KernelAnalyzer 1.4 with OpenCL support is now available!

A static analysis tool for OpenCL, Brook+ and IL.

We are pleased to announce the release of Stream KernelAnalyzer v1.4, a static analysis tool for stream kernels.  This tool can be used to analyze OpenCL kernels without requiring any GPU in the machine.

What's New in Version 1.4

  • Support for OpenCL™ (requires ATI Stream SDK v2.01).
  • Support for Catalyst™ driver 9.9-9.12.
  • Support for ATI Radeon™ HD 5870 graphics cards.
  • Support for ATI Radeon™ HD 5770 graphics cards.
  • Fixed issue where analysis of shader with complex flow control would fail.
  • Updated statistic names and descriptions.
  • Fetch instructions statistic now includes both texture and vertex fetch usage.
  • Compiler Output window now includes the OpenCL™ compiler build log (for warning and error messages).
  • Double-clicking on an OpenCL™ compilation error or warning will navigate to the reported line within the Source Code window.
  • Registered file extensions now supported automatically in file dialogs.
  • Object Code target now includes x86 assembly code for an OpenCL kernel.

Get it at http://developer.amd.com/GPU/SKA/Pages/default.aspx.

Please post your feedback here.

 

0 Likes
3 Replies
Hellblau
Adept I

Originally posted by: bpurnomo This tool can be used to analyze OpenCL kernels without requiring any GPU in the machine.


 

When running the tool on a system with a NVidia card compiling an OpenCL kernel results in the following output:

OpenCL Compile Error: clCreateContextFromType failed (CL_INVALID_VALUE).

This is with the NVidia System32/OpenCL.dll removed prior to installing the ATI Stream SDK.

Switching to my onboard ATI X1200 fixes the problem and then the tool runs fine.

0 Likes
kb1vc
Journeyman III

Are there any plans for tools that run under Linux? 

 

 

0 Likes
rotor
Journeyman III

Thanks for the Kernel Analyzer it is helpful but I still want to address several point it should have on next version:

-Specify the best and/or maximum workgroup size for the kernel targeted to specific device (i.e. 5870, 5770...)

-A better help document. It is really hard to know the meaning of the output of the tool (i.e. ALU:Fetch, min, max...) from the help document. I really don't know what does the ALU:Fetch mean. And also what the different between "filtering points"? What is the filtering point?

Rt

0 Likes