Getting started

ReProspect spans three complementary forms of analysis of CUDA applications: API tracing, kernel profiling, and binary analysis. The following notebooks introduce how to use ReProspect for each of these forms of analysis.

• Getting started with API tracing
  • About this page
  • Source code
  • Compilation
  • Running API tracing
  • CUDA API Trace default report
  • Querying CUDA API tracing data
  • Querying CUDA API tracing data by nested NVTX range
  • Assertions on CUDA API tracing data
  • Outlook
• Getting started with kernel profiling
  • About this page
  • Source code
  • Compilation
  • Measured throughputs
  • Kernel profiling metrics
  • Running kernel profiling
  • Kernel profiling results
  • Interpreting kernel profiling results
  • Assertions on kernel profiling results
  • Going further: insight through binary analysis
  • Outlook
• Getting started with binary analysis
  • About this page
  • Source code
  • Compilation
  • Extracting and disassembling CUDA binary code
  • Decoding SASS instructions
  • Decomposing SASS instructions
  • Matching SASS instructions across target architectures
  • Outlook