Contents

• 1. Introduction
  • 1.1. What is CUDA-GDB?
  • 1.2. Supported Features
  • 1.3. About This Document
• 2. Release Notes
• 3. Getting Started
  • 3.1. Setting Up the Debugger Environment
    • 3.1.1. Temporary Directory
    • 3.1.2. Using the CUDA-GDB debugger on Jetson and Drive Tegra devices
  • 3.2. Compiling the Application
    • 3.2.1. Debug Compilation
    • 3.2.2. Compilation With Linenumber Information
    • 3.2.3. Compilation With PTX Debug Option
    • 3.2.4. Compiling For Specific GPU architectures
    • 3.2.5. Compiling with tensor memory checks
  • 3.3. Using the Debugger
    • 3.3.1. Multi-GPU Debugging
    • 3.3.2. Remote Debugging
    • 3.3.3. Multiple Debuggers
    • 3.3.4. Attaching/Detaching
• 4. CUDA-GDB Extensions
  • 4.1. Command Naming Convention
  • 4.2. Getting Help
  • 4.3. Initialization File
  • 4.4. GUI Integration
  • 4.5. GPU core dump support
• 5. Kernel Focus
  • 5.1. Software Coordinates vs. Hardware Coordinates
  • 5.2. Current Focus
  • 5.3. Switching Focus
• 6. Program Execution
  • 6.1. Interrupting the Application
  • 6.2. Single Stepping
• 7. Breakpoints and Watchpoints
  • 7.1. Symbolic Breakpoints
  • 7.2. Line Breakpoints
  • 7.3. Address Breakpoints
  • 7.4. Kernel Entry Breakpoints
  • 7.5. Conditional Breakpoints
  • 7.6. Watchpoints
• 8. Inspecting Program State
  • 8.1. Memory and Variables
  • 8.2. Variable Storage and Accessibility
  • 8.3. Info CUDA Commands
    • 8.3.1. info cuda devices
    • 8.3.2. info cuda sms
    • 8.3.3. info cuda warps
    • 8.3.4. info cuda lanes
    • 8.3.5. info cuda kernels
    • 8.3.6. info cuda barriers
    • 8.3.7. info cuda blocks
    • 8.3.8. info cuda clusters
    • 8.3.9. info cuda threads
    • 8.3.10. info cuda launch trace
    • 8.3.11. info cuda launch children
    • 8.3.12. info cuda contexts
    • 8.3.13. info cuda managed
  • 8.4. Disassembly
  • 8.5. Registers
  • 8.6. Const banks
• 9. CUDA Debugger Variables
  • 9.1. Physical Coordinates
  • 9.2. Logical Coordinates
  • 9.3. Clusters
  • 9.4. Lane Masks
  • 9.5. Shared‑Memory Usage
  • 9.6. Error Tracking
  • 9.7. Convenience Variables
• 10. Event Notifications
  • 10.1. Context Events
  • 10.2. Kernel Events
• 11. Automatic Error Checking
  • 11.1. Checking API Errors
  • 11.2. GPU Error Reporting
  • 11.3. Autostep
• 12. Walk-Through Examples
  • 12.1. Example: bitreverse
    • 12.1.1. Walking through the Code
  • 12.2. Example: autostep
    • 12.2.1. Debugging with Autosteps
  • 12.3. Example: MPI CUDA Application
• 13. Tips and Tricks
  • 13.1. set cuda break_on_launch
  • 13.2. set cuda launch_blocking
  • 13.3. set cuda notify
  • 13.4. set cuda ptx_cache
  • 13.5. set cuda single_stepping_optimizations
  • 13.6. set cuda thread_selection
  • 13.7. set cuda value_extrapolation
  • 13.8. set cuda driver_logs
  • 13.9. set cuda driver_log_level
  • 13.10. Debugging Docker Containers
  • 13.11. Thread Block Clusters
  • 13.12. Debugging OptiX/RTCore applications
  • 13.13. Debugging on Windows Subsystem for Linux
  • 13.14. Accessing unmanaged device memory from host thread
• 14. Supported Platforms
• 15. Common Issues on Supported Operating Systems
• 16. Known Issues
• 17. Notices
  • 17.1. Notice
  • 17.2. OpenCL
  • 17.3. Trademarks