1. Start Here
   1. Welcome
   2. Getting Started
   3. Examples
   4. FAQ
   5. Debugging
   6. Native Libraries and CUDA
   7. Mailing List
   8. Language Specification
   9. Delite

Debugging

Usage

Using the Delite profiler involves three steps. In the first step, the Delite application is instrumented with calls to the profiler. In the second step, profiling data is generated and output to a JavaScript file. In the third step, the profiling data is converted to a web-based visualization which allows exploring the profiling information interactively.

Step 1: Instrumentation

When staging the Delite application, enable instrumentation by adding the --profile flag.

Example (GDA):

delitec --profile ppl.apps.ml.gda.GDARunner

Step 2: Collecting Profiling Data

Run the Delite application with the --profile flag. This generates a JavaScript file (/profileData.js) with the collected profile data.

Example (GDA):

delite --profile ppl.apps.ml.gda.GDARunner benchmark/data/ml/gda/q1x.dat benchmark/data/ml/gda/q1y.dat

Step 3: Generating a Web-based Visualization

Open profile/grid.html in a Chrome browser (the tool has not been tested for other browsers). On the siderbar located at the left, you will find a section named 'Inputs' at the top. Upload the DEG file and profile data file (named profile/profileData.js in the app's working directory) using the buttons provided. The src_dir input is optional. If provided, it enables the tool to locate the source line corresponding to each kernel. Once the files have been uploaded correctly, the 'View' button will turn green, indicating that the input files have been uploaded. Click the 'View' button' to start generating the visualization. Since the tool is in its initial version, it may take some time.