DrGPUM: A Memory Profiler for GPUs

DrGPUM is an advanced memory profiler that locates memory inefficiencies in GPU-accelerated applications. DrGPUM’s code is available on Github.

DrGPUM Basics

• Preface
  • HPCToolkit (Profiling Runtime)
  • Redshow
  • GPU Patch
  • Program Analyzer and Aggregator
• Install
  • GPU Patch
  • Dependencies
  • Redshow
  • HPCToolkit
  • Setup and Test
• Manual
  • Compile with Line Information
  • Profile Using DrGPUM
  • Profile Using HPCToolkit
    • First pass
    • Second pass
    • HPCToolkit separate pass
  • Control Knobs
    • Calling context view
    • Data flow view
    • Fine grain pattern views
  • Example
• FAQ
  • Profile Python applications
  • Accelerate data flow profiling
  • Accelerate value pattern profiling

DrGPUM Development

• Workflow
• Roadmap

DrGPUM Samples

• Unit Tests
  • interval_merge
  • op_graph_simple
  • op_pattern_simple
  • stress
  • vectorAdd
• Rodinia GPU Benchmark
  • backprop
  • bfs
  • cfd
  • hotspot
  • hotspot3D
  • huffman
  • lavaMD
  • pathfinder
  • srad
  • streamcluster
• QMCPACK
  • Introduction
  • Profiling
  • Optimization
• Castro
  • Introduction
  • Profiling
  • Optimization
• Deepwave
  • Introduction
  • Profiling
  • Optimization
• Darknet
  • Introduction
  • Profiling
  • Optimization
• PyTorch
  • Introduction
  • Profiling
  • Optimization
• NAMD
  • Introduction
  • Profiling
  • Optimization
• BarraCUDA
  • Introduction
  • Profiling
  • Optimizations

Indices and Tables

• Index

• Search Page