6 Replies Latest reply on Jan 10, 2010 12:32 AM by genaganna

    profiling tools

    mohit2710

      Hi,

      can anyone tell me from where can I download the cache grind profiling tool for ubuntu9.04...in order to check the performance of my opencl code... my output of matrixmultiplication is giving very irregular results...

      any kind of help is appreciated..

        • profiling tools
          genaganna

          Mohit2710,

                I am not sure what tool you are looking. Presently No tool available to profile OpenCL kernel in linux.  Hand tuning is the only way at this stage. 

          Please see the Preformance and Optimization notes doc at http://developer.amd.com/gpu/ATIStreamSDK/pages/Documentation.aspx

          for more details.

          Use your own timer code or OpenCL profiling API's to calculate performance calculation.

           

            • profiling tools
              mohit2710

              Hi,

              I want to find out why i am getting poor results..

              for this purpose i want to see my cache performance for the matrix multiplication program....

              what can be the other reasons for poor results....??

              I am using ubuntu 9.04, and amd profiling tool only supports visual studio...

                • profiling tools
                  genaganna

                  Mohit2710,

                           which cache performance you are talking about?.  Are you running your code on CPU or GPU?  Even profiler won't say anything about cache.

                    • profiling tools
                      ryta1203

                      It would be nice if it would though.

                      • profiling tools
                        mohit2710

                        I am running my code on CPU Intel Core2Duo T6400 @ 2.00 GHz with 2.00 MB cache size.

                        My code does not give proper results for higher block sizes and what could be he possible reasons for for this, I can think of limited cache size as a reason. Can you suggest any other reason.

                        Also about the tools that can be used to check this.

                          • profiling tools
                            genaganna

                             

                            Originally posted by: mohit2710 I am running my code on CPU Intel Core2Duo T6400 @ 2.00 GHz with 2.00 MB cache size.

                             

                            My code does not give proper results for higher block sizes and what could be he possible reasons for for this, I can think of limited cache size as a reason. Can you suggest any other reason.

                             

                            Also about the tools that can be used to check this.

                             

                            Mohit2710,

                                           Could you please post your code?  You have to use AMD codeanalyst or Intel VTune to profile your code on CPU.