Intel® Advisor User Guide

ID 766448
Date 12/16/2022
Public

A newer version of this document is available. Customers should click here to go to the newest version.

Document Table of Contents

profile-gpu

Analyze OpenCL™ and oneAPI Level Zero programs running on Intel® Graphics.

GUI Equivalent

Analysis Workflow > Baseline Device > GPU

Project Properties > Analysis Target > Performance Modeling > GPU

Syntax

--profile-gpu

--no-profile-gpu

Default

Off (no-profile-gpu)

Actions Modified

collect=survey

collect=tripcounts

collect=roofline

collect=projection

Usage

Prerequisite: Set up system environment to enable GPU kernel profiling.

Use this option to analyze a GPU-enabled application that uses SYCL, OpenMP* target, or OpenCL™ programming model.

  • For the GPU Roofline Insights, use this option to analyze code regions running on a CPU and code regions running on a GPU. This option may affect the performance of your application on the CPU side.
  • For the Offload Modeling, use this option to analyze only code regions running on a GPU. This is a preview feature.
    NOTE:
    Make sure to use this option with the Survey, Trip Counts, and Performance Modeling analyses.
NOTE:
GPU profiling is applicable only to Intel® Graphics.

Example

Run the Roofline analysis and enable GPU profiling to analyze OpenCL™ and oneAPI Level Zero programs running on Intel® Graphics.

advisor --collect=roofline --profile-gpu --project-dir=./advi_results -- ./myApplication
  1. Run the Survey analysis with the GPU kernel profiling enabled.
  2. Run the Trip Counts and FLOP analysis with the GPU kernel profiling enabled.
  3. Run the Performance Modeling with the GPU kernel profiling enabled.
advisor --collect=survey --static-instruction-mix --profile-gpu --project-dir=./advi_results -- ./myApplication
advisor --collect=tripcounts --flop --profile-gpu --project-dir=./advi_results -- ./myApplication
advisor --collect=projection --profile-gpu --project-dir=./advi_results -- ./myApplication

See Also