新手求助：如何使用Intel Parallel Studio XE对C++代码做性能分析？

Hey there! I totally get where you're coming from—Yourkit's Eclipse integration is super straightforward, so jumping to Intel Parallel Studio XE can feel a bit overwhelming at first. Let's break this down step by step to get you up and running.

First: Fixing the psxevars.sh Issue #

When you run source psxevars.sh, "partial output" might be normal, but we need to confirm the environment variables are actually loading correctly:

• Double-check the script path: This file lives in your Parallel Studio XE installation's bin directory. For example, on Linux it's usually /opt/intel/psxe/bin/psxevars.sh. If you're running it from a different folder, use the full absolute path to avoid issues.
• Specify your architecture: Try running it with the target architecture explicitly:

  source /opt/intel/psxe/bin/psxevars.sh intel64  # For 64-bit systems
  # Or for 32-bit: source /opt/intel/psxe/bin/psxevars.sh ia32
  

• Verify the environment: After sourcing, run echo $INTEL_PSXEDIR—if it outputs your Parallel Studio installation path, the variables loaded successfully. If not, double-check your installation directory (student versions sometimes install to ~/intel/psxe instead of /opt/intel).

Integrating with Eclipse (Like Yourkit) #

Intel Parallel Studio XE has official Eclipse plugins that make analysis just as seamless as Yourkit once set up:

1. Install the Eclipse plugins:

   • Open Eclipse, go to Help > Install New Software
   • Click Add, then Local, and navigate to your Parallel Studio installation's eclipse_plugins folder (e.g., /opt/intel/psxe/eclipse_plugins)
   • Select the plugins you need: Intel VTune Amplifier (for performance profiling), Intel Inspector (for memory/thread debugging), and Intel C++ Compiler Integration (optional but recommended for better optimization insights)
   • Follow the installation prompts and restart Eclipse

2. Set up your C++ project for analysis:

   • Make sure your project is compiled with debug info: Go to Project Properties > C/C++ Build > Settings, add -g to the compiler flags (under Intel C++ Compiler > Debugging or GCC C++ Compiler > Debugging if you're using GCC)
   • For better profiling results, avoid aggressive optimizations like -O3 while you're learning—stick to -O0 or -O1 first

Running Performance Analysis (Similar to Yourkit's Listening) #

Let's focus on VTune Amplifier (the equivalent of Yourkit for C++ performance profiling):

• Open your project in Eclipse, build it to generate an executable
• Click the VTune Amplifier icon in the toolbar (looks like a speedometer)
• Select New Analysis, then pick a profiling type:
  • Hotspots: The most common starting point—finds functions that take the most CPU time, just like Yourkit's CPU profiling
  • Memory Access: For analyzing cache misses and memory bottlenecks
  • Threading: If you're working with parallel code, this helps identify synchronization issues
• In the analysis configuration window:
  • Select your project's executable file under Application
  • Add any command-line arguments your program needs under Arguments
  • Click Start—VTune will run your program and collect profiling data
• Once the run finishes, you'll get a detailed report with:
  • Top CPU-consuming functions
  • Call stacks to trace where time is spent
  • Optimization suggestions specific to Intel hardware

Quick Tips for New Users #

• Use Intel's compiler: Switching to icc/icpc (via Project Properties > C/C++ Build > Tool Chain Editor) will give you more accurate profiling data and tailored optimization tips from VTune
• Start simple: Don't dive into advanced profiling types first—master Hotspots analysis before moving to memory or threading tools
• Check the Eclipse perspective: After installing the plugins, switch to the Intel VTune Amplifier perspective (Window > Perspective > Open Perspective > Other) for a dedicated profiling workspace

内容的提问来源于stack exchange，提问作者Felipe