1 / 17

Mapping the FFT Algorithm to the IBM Cell Processor

Mapping the FFT Algorithm to the IBM Cell Processor. Andy Polidore Advisors: Brendan Burns, Joseph Czechowski. Motivation. MRI Imaging Fast Fourier Transformations Efficient algorithm for computing a Discrete Fourier Transform DFT converts time-domain to frequency-domain

Download Presentation

Mapping the FFT Algorithm to the IBM Cell Processor

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. Mapping the FFT Algorithm to the IBM Cell Processor Andy Polidore Advisors: Brendan Burns, Joseph Czechowski

  2. Motivation • MRI Imaging • Fast Fourier Transformations • Efficient algorithm for computing a Discrete Fourier Transform • DFT converts time-domain to frequency-domain • 2D FFT: Perform a 1D FFT on each row of an image and then perform a 1D FFT on each resulting column • The Cell • Nine cores • 1 Power Processing Unit (PPU) • 8 Synergistic Processing Units (SPU)

  3. Strategy • Cell comes with 2d routine • Needs to be called twice • First call organizes the data in contiguous column form • Striping • Limited SPU memory • Quad Buffering

  4. PPU SPU 0 Input Buffer Input DMA In FFT Output Buffer FFT out DMA Out PPU SPU 0 Input Buffer Input DMA In FFT Output Buffer FFT out DMA Out

  5. SPU 0 PPU SPU 1 SPU 7 Input Buffer DMA In Input FFT Output Buffer FFT out

  6. PPU SPU 0 SPU 1 Input Buffer SPU 2 Input DMA In FFT Output Buffer FFT out DMA Out PPU Sync Point SPU 0 SPU 1 Input Buffer SPU 2 Input DMA In FFT Output Buffer FFT out DMA Out

  7. Quad buffering • Why it is required? • Space problems • Maximizing processing power • Buffers • IN to handle incoming data • FFTin and FFTout to process the data • OUT stores the data ready to be DMA’ed back to main memory

  8. Buffering A B C D 0 FILL ------- ------- ------- 1 2 3 4 5 6

  9. Buffering A B C D 0 FILL ------- ------- ------- 1 FFTOUT FILL ------- FFTIN 2 3 4 5 6

  10. Buffering A B C D 0 FILL ------- ------- ------- 1 FFTOUT FILL ------- FFTIN FFTOUT 2 OUT FFTIN FILL 3 4 5 6

  11. Buffering A B C D 0 FILL ------- ------- ------- 1 FFTOUT FILL ------- FFTIN FFTOUT 2 OUT FFTIN FILL 3 OUT FILL FFTOUT FFTIN 4 5 6

  12. Buffering A B C D 0 FILL ------- ------- ------- 1 FFTIN FFTOUT FILL ------- 2 FFTOUT OUT FFTIN FILL 3 OUT FILL FFTOUT FFTIN 4 FILL FFTIN OUT FFTOUT 5 FFTIN FFTOUT FILL OUT 6 FFTOUT OUT FFTIN FILL

  13. Striping Main Memory SPU 0 SPU 1 SPU 2 SPU 3 SPU 4 SPU 5 SPU 6 SPU 7

  14. Challenges • Simulator • Testing is slow • Alignment • Compiler • C coding • Working with bytes • Parallel processing • Data movement • Debugging

  15. Knowledge Gained • Mastering Linux • C make files, linking, etc • Data movement strategies • Multi-core processing • Debugging!

  16. Results and Conclusions • Success? • Future Work • Arbitrary size input

  17. Questions?

More Related