Skip to content

Latest commit

 

History

History
118 lines (53 loc) · 4.08 KB

README_Codebase.md

File metadata and controls

118 lines (53 loc) · 4.08 KB

Codebase Overview:

  • .github/workflows: Define CI/CD implementation and actions

  • pre-commit-config.yaml: Part of CI/CD implementation used to check if you can commit your changes

  • .gitignore: Gives list of types of files the system should ignore when committing your changes

  • compilation: helps transform IR’s into ones that are compatible for Vortex

    • HostTranslation: transforming the Host IR

      • include: All header files for files in src folder

      • src: Source code for HostTranslation

        • vortex: support for Vortex architecture

          • GenerateHostStub.cpp:

          • InitializeDevice.cpp:

          • RemoveCudeBuiltin.cpp: Removing all cuda builtin functions (not compatible with CPU’s)

          • ReplaceConstantMemory.cpp: Converting CUDA constant memory declarations into CPU compatible ones

          • ReplaceCudaBuiltin.cpp: Replace all built-in functions compatible to CUDA with those compatible to CPU

          • ReplaceKernelArgs.cpp: Make calls to cudaLaunchKernel compatible with CPU or device

        • x86: Support for x86 architecture(files are similar to that in vortex folder)

    • KernelTranslation: trasnforming the Kernel IR

      • include: All header files for files in src folder

      • src: Source code for HostTranslation

        • vortex: support for Vortex architecture

          • generate_wrapper.cpp: generates a wrapper function around a kernel function

          • handle_sync.cpp: Finds barriers and split blocks using these barriers

          • init.cpp: Initializes block of components needed before execution for kernel translation

          • insert_sync.cpp: Insert barriers within kernel code to ensure coordination in execution

          • insert_warp_loop.cpp: Inserts warp-level loop and ensures handling at warp-level is correct

          • memory_hierarchy.cpp: Maps different memory hierarchy abstractions (shared, constant, global memory) to ones LLVM can use

          • performance.cpp: Displays metrics (LAM, FAM, CGAM, MAM) that can be used to analyze and optimize performance

          • tool.cpp: Provide helper functions that can be useful for working with the kernel

          • warp_func.cpp: Provide helper functions for warp-level operations

        • x86: Support for x86 architecture (files are similar to that in vortex folder)

    • HostTranslation.cpp: Combines functions in HostTranslation folder to transform the Host IR

    • KernelTranslation.cpp: Combines functions in KernelTranslation folder to transform the Kernel IR

  • data: data about some benchmarks in examples folder

  • examples: benchmarks used in testing

  • runtime: relates to runtime library used to compile and link files together

    • include: All header files for files in src folder

    • src: source code

      • vortex: support for Vortex architecture

        • host: dealing with the transformed IR for source code

          • cudaRuntimeImpl.cpp: managing and interfacing CUDA runtime operations

        • kernel: dealing with the transformed IR for kernel code

          • Makefile: Lists targets and rules for compiling and linking the kernel sources

          • cudaKernelImpl.cpp: Implements basic arithmetic functions Vortex can use in runtime compilation

          • cudaKernelImpl.o: Object file from cudaKernelImpl.cpp file

          • cudaKernelImpl_64.o: Object file for 64-bit architecture

      • x86: Support for x86 architecture

    • threadPool: managing thread pool

      • include: All header files for files in src folder

      • src: source code

        • api.cpp: Implements threads in threadpool

  • CuPBoP_env_setup.sh: Environment setup for CuPBoP

  • CuPBoP_env_setup_wo_PoCL.sh: Environment setup for CuPBoP if device can't use the Portable Computing Language

  • docs: folder containing all files for README file documentation