Skip to content

Latest commit

 

History

History
133 lines (96 loc) · 4.3 KB

docker.md

File metadata and controls

133 lines (96 loc) · 4.3 KB
Raw

Docker Files

Setting up Docker and Running a chipStar Sample

  1. Install Docker:

  2. Pull the chipStar 'latest' image:

    docker pull pveleskopglc/chipstar:latest

  3. Run the container:

    docker run -it pveleskopglc/chipstar:latest /bin/bash

    This command starts an interactive session without binding any GPUs to the container - a couple of OpenCL CPU runtimes are setup already.

  4. Execute a matrix multiply sample:

    CHIP_DEVICE_TYPE=cpu CHIP_BE=opencl ~/chipStar/build/samples/0_MatrixMultiply/MatrixMultiply 
Device name 13th Gen Intel(R) Core(TM) i9-13900K
Running 1 iterations 
hipLaunchKernel 0 time taken: 185.903
hipLaunchKernel BEST TIME: 185.903
GPU real time taken(ms): 193.007
matrixMultiplyCPUReference time taken(ms): 2861.23
Verification PASSED!

This process will set up Docker, pull the latest chipStar image, start a container with the necessary environment, and run a sample application to verify the setup.

Running Docker Image with GPU

To run the Docker image with GPU support, follow these steps:

  1. Run the Docker container with GPU support:

    docker run -it --device /dev/dri pveleskopglc/chipstar:latest /bin/bash

  2. Unload PoCL module to expose Intel GPUs (if available):

    module unload pocl/5.0-llvm-15

  3. Run a GPU-accelerated sample:

    CHIP_DEVICE_TYPE=gpu CHIP_BE=opencl ~/chipStar/build/samples/0_MatrixMultiply/MatrixMultiply

This process allows you to run the chipStar Docker image with GPU support, enabling GPU-accelerated computations within the container.

DockerfileBase Overview

  • Base: Ubuntu latest
  • User: 'chipStarUser' with sudo, video, render group access
  • Core tools: gcc, g++, cmake, python3, git, OpenCL dev environment
  • LLVM/Clang: Customizable version (default 15)
  • Lmod: For environment module management
  • POCL: Portable OpenCL implementation
  • Intel OneAPI: Via Miniconda, includes MKL, TBB, DPC++
  • Level Zero API: For low-level device control

Purpose: Comprehensive chipStar development environment, supporting both open-source and Intel proprietary frameworks.

DockerfileCPPLinter Overview

This layer builds upon the base image and adds:

  • Python virtual environment setup
  • C++ linting tools installation

Key components:

  • Python3 venv: For isolated Python environment
  • clang-tools (version 0.13.0): Provides static analysis and linting capabilities for C++
  • cpp-linter (version 1.10.0): A tool for linting C++ code

Purpose: Enhances the development environment with code quality tools specifically for C++ projects, enabling better code analysis and consistency checks.

DockerfileFull Overview

This layer builds upon the base image and adds:

  • Multiple LLVM/Clang versions (16, 17, 18)
  • POCL (Portable Computing Language) for each LLVM version

Key components:

  • LLVM/Clang versions 16, 17, and 18:
    • Built from source
    • Installed in /apps/llvm/${LLVM_VERSION}
    • Environment modules created for each version
  • POCL 4.0:
    • Built for each LLVM version
    • Installed in /apps/pocl/4.0-llvm-${LLVM_VERSION}
    • Environment modules created for each version

Build process for each LLVM/POCL pair:

  1. Configure and build LLVM
  2. Install LLVM and create its environment module
  3. Clone POCL repository
  4. Configure and build POCL
  5. Install POCL and create its environment module

Purpose: Provides a comprehensive development environment with multiple LLVM toolchains and corresponding POCL installations, allowing for flexible OpenCL development and testing across different LLVM versions, suitable for CI use.

DockerfileLatest Overview

This layer builds upon the base image and adds:

  • Additional Python packages
  • LLVM/Clang 15 environment setup
  • Vim common tools
  • chipStar build and installation

Key components:

  • PyYAML: Python package for YAML parsing
  • LLVM/Clang 15: Loaded as a module
  • Vim common: Includes 'xxd' utility
  • chipStar:
    • Cloned from GitHub
    • Built with CMake (Release mode, HIPBLAS enabled)
    • Installed system-wide
    • Test suite built (for OpenCL backend, CPU device type)

Purpose: Provides a complete environment for chipStar development and testing.