Taylor, MichaelMuralitharan, Sripathi2021-08-262021-08-262021-08-262021Muralitharan_washington_0250O_23009.pdfhttp://hdl.handle.net/1773/47276Thesis (Master's)--University of Washington, 2021Recent developments in architecture research warrant the need for efficient host cores to interact and manage multiple accelerators in a system-on-chip design. Existing designs suffer from low configurability, rely on non-standard or proprietary tooling, or require sophisticated mechanisms to interact with accelerators. BlackParrot [1] is a 64-bit Linux-capable, open-source multicore processor that aims to break these barriers to become an accelerator host processor used in state-of-the-art SoCs. One of the critical requirements to achieve this goal is modifying the core quickly to suit the application's needs. The work presented in this thesis discusses ways to optimize BlackParrot for integration with diverse architectures. While BlackParrot offers sufficient configurability to host standalone accelerators in its memory system, it did not have the tools to integrate with larger system designs. The creation of the BlackParrot unicore and the standard cache interface between the L1 cache and its controller, presented in this thesis, allows BlackParrot to integrate with designs with minimal modification. Its parameterizable cache, along with a multi-cycle fill strategy, has enabled the creation of a tiny core that can find uses in systems with physical limitations. The utility of these modifications, among others, was validated by integrating BlackParrot with HammerBlade and OpenPiton.application/pdfen-USCC BYAcceleratorASICBlackParrotFPGAIntegrationComputer engineeringElectrical engineeringThesis