Uploaded image for project: 'SAFe Program'
  1. SAFe Program
  2. SP-1770

Improve PSS CXFT/FPGA Algorithm Handling of FFTs that are other than 8 Million Points

Details

    • Feature
    • Could have
    • PI11, PI12
    • COM PSS SW
    • None
    • Data Processing
    • Hide

      Make algorithm more useable by making it  able to handle FFTs of varying size 

      Show
      Make algorithm more useable by making it  able to handle FFTs of varying size 
    • Hide

      Given – the Cheetah CXFT/Altera algorithm is requested
      When - the CXFT unit tests are executed with datasets of powers of 2 in the range 20 to 23 or truncated/padded to the nearest of these powers of 2
      Then — the unit tests shall pass indicating the correct results which will be demonstrated. 

      Show
      Given – the Cheetah CXFT/Altera algorithm is requested When  - the CXFT unit tests are executed with datasets of powers of 2 in the range 20 to 23 or truncated/padded to the nearest of these powers of 2 Then — the unit tests shall pass indicating the correct results which will be demonstrated. 
    • 1
    • 1
    • 1
    • 1
    • Team_PSS
    • Sprint 4
    • Hide

      See Child Issues for MR links:
      AT4-518
      This issue created the code to compute real to complex FFTs of sizes 128 to 32M for odd powers of 2 i.e 2^n where n is an odd number. The code is also capable of complex to complex transforms for even power of 2 from 64 to 16M data size.

      AT4-519
      Unit tests with different FFT sizes were executed. cuda code is also working fine. An argument for data size to be passed to the unit tests is now included instead of a static variable. Unit tests for the following data sizes with odd powers of 2 were successful: 128 (2^7), 2K (2^11), 8M (2^23).

      AT4-520
      Both Rabbit & Cheetah codes for flexible FFT sizes were merged to the dev/master branch.

      AT4-544:
      Created a Panda branch and uploaded hello_world.aocx binary image for emulator 17.1.2 and 20.2 SDK. Now unit tests load correct binary based on compilation environment.

      If a user compiles Panda with a SDK other than 17.1 & 20.2, a warning message will be displayed.

      Added a GetHelloAocx() function to generalize this driver specific image loading. This function obtains driver version based on FPGA device passed and returns a Boost path of the image.

      A summary document:

      https://docs.google.com/document/d/1AUn_KRpKc2lkHfzdf_q2uORM_a_93ypgLrdb3ic8FcY/edit?usp=sharing

      Show
      See Child Issues for MR links: AT4-518 This issue created the code to compute real to complex FFTs of sizes 128 to 32M for odd powers of 2 i.e 2^n where n is an odd number. The code is also capable of complex to complex transforms for even power of 2 from 64 to 16M data size. AT4-519 Unit tests with different FFT sizes were executed. cuda code is also working fine. An argument for data size to be passed to the unit tests is now included instead of a static variable. Unit tests for the following data sizes with odd powers of 2 were successful: 128 (2^7), 2K (2^11), 8M (2^23). AT4-520 Both Rabbit & Cheetah codes for flexible FFT sizes were merged to the dev/master branch. AT4-544: Created a Panda branch and uploaded hello_world.aocx binary image for emulator 17.1.2 and 20.2 SDK. Now unit tests load correct binary based on compilation environment. If a user compiles Panda with a SDK other than 17.1 & 20.2, a warning message will be displayed. Added a GetHelloAocx() function to generalize this driver specific image loading. This function obtains driver version based on FPGA device passed and returns a Boost path of the image. A summary document: https://docs.google.com/document/d/1AUn_KRpKc2lkHfzdf_q2uORM_a_93ypgLrdb3ic8FcY/edit?usp=sharing
    • 12.5
    • Stories Completed, Outcomes Reviewed, NFRS met, Demonstrated, Satisfies Acceptance Criteria, Accepted by FO
    • PI22 - UNCOVERED

    Description

      Current CXFT FPGA 8M point code does not correctly perform for FFTs of sizes < 8M points. The aim of this Feature is to improve the CXFT algorithm in order to make it capable of handling FFTs other than 8M points.

       

      Attachments

        Issue Links

          Structure

            Activity

              People

                A.Noutsos Noutsos, Aristeidis
                m.pearson Pearson, Mike [X] (Inactive)
                Votes:
                0 Vote for this issue
                Watchers:
                2 Start watching this issue

                Feature Progress

                  Story Point Burn-up: (100.00%)

                  Feature Estimate: 1.0

                  IssuesStory Points
                  To Do00.0
                  In Progress   00.0
                  Complete417.0
                  Total417.0

                  Dates

                    Created:
                    Updated:
                    Resolved:

                    Structure Helper Panel