Single Pass Scan in CUDA

Project for the Programming Massively Parallel Hardware class from University of Copenhagen A.Y. 2023/2024

Usage

The program can be ran with Makefile using make. To run different experiments modify the Makefile with the desired parameters. The Makefile generate an executable called sp-scan that needs 4 arguments to run.

./sp-scan <benchmark> <array-length> <block-size> <kernel-version>

• <benchmark> set benchmark mode, next arguments wil not be considered.
• <array-length> size of the input array to test
• <block-size> size of one CUDA block
• <kernel-version> kernel version from kernels.cu file to run:
  • 1 -> Without loopback (spScanKernel)
  • 2 -> Single thread Loopback (spLookbackScanKernel)
  • 3 -> Warp Loopback (spWarpLookbackScanKernel)
  • 4 -> Optimized Warp Loopback (spWarpLookbackScanKernelOpt)

Authors

• Leonardo Stoppani (@lilf4p)
• Lukas Mikelionis (@lukas-mi)

References

• Merrill, Duane and Michael Garland. “Single-pass Parallel Prefix Scan with Decoupled Lookback.” (2016).Nvidia Link