Intel® Advisor User Guide

Download PDF
ID 766448
Date 3/22/2024
Public

A newer version of this document is available.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

Document Table of Contents
Document Table of Contents x
Intel® Advisor User Guide
Intel® Advisor User Guide x
Introduction Install and Launch Intel® Advisor Set Up Project Analyze Vectorization Perspective Analyze CPU Roofline Model Threading Designs Model Offloading to a GPU Analyze GPU Roofline Design and Analyze Flow Graphs Minimize Analysis Overhead Analyze MPI Applications Manage Results Command Line Interface Troubleshooting Reference Appendix
Introduction x
Design and Optimization Methodology Tutorials and Samples Get Help and Support
Install and Launch Intel® Advisor x
Install Intel® Advisor Set Up Environment Variables Set Up System to Analyze GPU Kernels Set Up Environment to Offload SYCL, OpenMP* target, and OpenCL™ Applications to CPU Launch Intel® Advisor GUI Navigation Quick Start
Set Up Project x
Configure Target Application Build Target Application Create Project
Configure Target Application x
Limit the Number of Threads Used by Parallel Frameworks Choose a Small, Representative Data Set
Create Project x
Configure Project Configure Binary/Symbol Search Directories Configure Source Search Directory Binary/Symbol Search and Source Search Locations
Analyze Vectorization Perspective x
Run Vectorization and Code Insights Perspective from GUI Run Vectorization and Code Insights Perspective from Command Line Explore Vectorization and Code Insights Results
Run Vectorization and Code Insights Perspective from GUI x
Vectorization Accuracy Presets Customize Vectorization and Code Insights Perspective
Run Vectorization and Code Insights Perspective from Command Line x
Vectorization Accuracy Levels in Command Line
Explore Vectorization and Code Insights Results x
Vectorization Report Overview Examine Not-Vectorized and Under-Vectorized Loops Analyze Loop Call Count Investigate Memory Usage and Traffic Find Data Dependencies
Analyze CPU Roofline x
Run CPU / Memory Roofline Insights Perspective from GUI Run CPU / Memory Roofline Insights Perspective from Command Line Explore CPU/Memory Roofline Results
Run CPU / Memory Roofline Insights Perspective from GUI x
CPU Roofline Accuracy Presets Customize CPU / Memory Roofline Insights Perspective
Run CPU / Memory Roofline Insights Perspective from Command Line x
CPU Roofline Accuracy Levels in Command Line
Explore CPU/Memory Roofline Results x
CPU Roofline Report Overview Examine Bottlenecks on CPU Roofline Chart Examine Relationships Between Memory Levels Compare CPU Roofline Results
Model Threading Designs x
Run Threading Perspective from GUI Run Threading Perspective from Command Line Annotate Code for Deeper Analysis Explore Threading Results Model Threading Parallelism Check for Dependencies Issues Add Parallelism to Your Program
Run Threading Perspective from GUI x
Customize Threading Perspective
Run Threading Perspective from Command Line x
Threading Accuracy Levels in Command Line
Annotate Code for Deeper Analysis x
Annotate Code to Model Parallelism Annotations Annotation Report
Annotate Code to Model Parallelism x
Before Annotating Code for Deeper Analysis Use Amdahl's Law and Measure the Program Task Organization and Annotations Annotate Parallel Sites and Tasks Task Patterns Choose the Tasks Use Partially Parallel Programs with Intel® Advisor
Task Patterns x
Multiple Parallel Sites Data and Task Parallelism Mix and Match Tasks
Choose the Tasks x
Task Interactions and Suitability How Big Should a Task Be?
Annotations x
Annotation Types Annotation Definitions Files Add Annotations into Your Source Code Tips for Annotation Use with C/C++ Programs
Annotation Types x
Annotation Types Summary Annotation General Characteristics Site and Task Annotations for Simple Loops With One Task Site and Task Annotations for Parallel Sites with Multiple Tasks Lock Annotations Pause Collection and Resume Collection Annotations Special-purpose Annotations
Annotation Definitions Files x
Reference the Annotations Definitions Directory Include the Annotations Header File in C/C++ Sources
Add Annotations into Your Source Code x
Copy Annotations and Build Settings Using the Annotation Assistant Pane Insert Annotations in a Text Editor
Tips for Annotation Use with C/C++ Programs x
Control the Expansion of advisor-annotate.h Handle Compilation Issues that Appear After Adding advisor-annotate.h
Annotation Report x
Annotation Report, Clear Description of Storage Row Annotation Report, Disable Observations in Region Row Annotation Report, Pause Collection Row Annotation Report, Inductive Expression Row Annotation Report, Lock Row Annotation Report, Observe Uses Row Annotation Report, Reduction Row Annotation Report, Re-enable Observations at End of Region Row Annotation Report, Resume Collection Row Annotation Report, Site Row Annotation Report, Task Row Annotation Report, User Memory Allocator Use Row Annotation Report, User Memory Deallocator Use Row
Model Threading Parallelism x
Suitability Report Overview Choose Modeling Parameters in the Suitability Report Fix Annotation-related Errors Detected by the Suitability Tool Advanced Modeling Options Reduce Parallel Overhead, Lock Contention, and Enable Chunking
Reduce Parallel Overhead, Lock Contention, and Enable Chunking x
Reduce Site Overhead Reduce Task Overhead Reduce Lock Overhead Reduce Lock Contention Enable Task Chunking
Check for Dependencies Issues x
Code Locations Pane Filter Pane (Dependencies Report) Problems and Messages Pane Dependencies Source Window
Dependencies Source Window x
Code Locations Pane (Dependencies Source Window) Focus Code Location Pane Focus Code Location Call Stack Pane Related Code Locations Pane Related Code Location Call Stack Pane Relationship Diagram Pane
Add Parallelism to Your Program x
Before You Add Parallelism: Choose a Parallel Framework Add the Parallel Framework to Your Build Environment Annotation Report Replace Annotations with Intel® oneAPI Threading Building Blocks (oneTBB) Code Replace Annotations with OpenMP* Code Next Steps for the Parallel Program
Before You Add Parallelism: Choose a Parallel Framework x
Parallel Frameworks Intel® oneAPI Threading Building Blocks (oneTBB) OpenMP* Microsoft Task Parallel Library* (TPL) Other Parallel Frameworks
Add the Parallel Framework to Your Build Environment x
Enable Intel® oneAPI Threading Building Blocks (oneTBB) in your Build Environment Define the TBBROOT Environment Variable Enable C++11 Lambda Expression Support with Intel® oneAPI Threading Building Blocks (oneTBB) Enable OpenMP* in your Build Environment
Annotation Report x
Annotation Report Overview Locate Annotations with the Annotation Report
Replace Annotations with Intel® oneAPI Threading Building Blocks (oneTBB) Code x
Intel® oneAPI Threading Building Blocks (oneTBB) Mutexes Intel® oneAPI Threading Building Blocks (oneTBB) Simple Mutex - Example Test the Intel® oneAPI Threading Building Blocks (oneTBB) Synchronization Code Parallelize Functions - Intel® oneAPI Threading Building Blocks (oneTBB) Tasks Parallelize Data - Intel® oneAPI Threading Building Blocks (oneTBB) Counted Loops Parallelize Data - Intel® oneAPI Threading Building Blocks (oneTBB) Loops with Complex Iteration Control
Replace Annotations with OpenMP* Code x
Add OpenMP Code to Synchronize the Shared Resources OpenMP Critical Sections Basic OpenMP Atomic Operations Advanced OpenMP Atomic Operations OpenMP Reduction Operations OpenMP Locks Test the OpenMP Synchronization Code Parallelize Functions - OpenMP Tasks Parallelize Data - OpenMP Counted Loops Parallelize Data - OpenMP Loops with Complex Iteration Control
Next Steps for the Parallel Program x
Use Intel® Inspector and Intel® VTune™ Profiler Debug Parallel Programs
Model Offloading to a GPU x
Run Offload Modeling Perspective from GUI Run Offload Modeling Perspective from Command Line Prerequisites Optional: Generate pre-Configured Command Lines Method 1. Use Collection Presets Method 2. Use per-Analysis Collection Method 3. Use Python* Scripts Next Steps See Also Explore Offload Modeling Results Advanced Modeling Configuration
Run Offload Modeling Perspective from GUI x
Offload Modeling Accuracy Presets Customize Offload Modeling Perspective
Run Offload Modeling Perspective from Command Line x
Prerequisites Optional: Generate pre-Configured Command Lines Method 1. Use Collection Presets Method 2. Use per-Analysis Collection Method 3. Use Python* Scripts Next Steps See Also Offload Modeling Accuracy Levels in Command Line Run GPU-to-GPU Performance Modeling from Command Line
Explore Offload Modeling Results x
Offload Modeling Report Overview Examine Regions Recommended for Offloading Examine Data Transfers for Modeled Regions Check for Dependency Issues Explore Performance Gain from GPU-to-GPU Modeling Investigate Non-Offloaded Code Regions
Advanced Modeling Configuration x
Model Application Performance on a Custom Target GPU Device Check How Assumed Dependencies Affect Modeling Manage Invocation Taxes Enforce Offloading for Specific Loops
Analyze GPU Roofline x
Test Topic for Embedded Help Run GPU Roofline Insights Perspective from GUI Run GPU Roofline Insights Perspective from Command Line Explore GPU Roofline Results
Run GPU Roofline Insights Perspective from GUI x
GPU Roofline Accuracy Presets Customize GPU Roofline Insights Perspective
Run GPU Roofline Insights Perspective from Command Line x
GPU Roofline Accuracy Levels in Command Line
Explore GPU Roofline Results x
Examine GPU Roofline Summary Examine Bottlenecks on GPU Roofline Chart Examine Kernel Details Compare GPU Roofline Results
Design and Analyze Flow Graphs x
Where to Find the Flow Graph Analyzer Launching the Flow Graph Analyzer Flow Graph Analyzer GUI Overview Flow Graph Analyzer Workflows Designer Workflow Generating C++ Stubs Preferences Scalability Analysis Collecting Traces from Applications Nested Parallelism in Flow Graph Analyzer Analyzer Workflow Experimental Support for OpenMP* Applications Sample Trace Files Additional Resources
Flow Graph Analyzer GUI Overview x
Menus Toolbars Tabs Main Canvas Charts
Designer Workflow x
Adding Nodes, Edges, and Ports Modifying Node Properties Viewing Edge Properties Validating a Graph Saving a Graph to a File
Scalability Analysis x
Activating the Graph Scalability Analysis Prerequisites Running the Scalability Analysis Exploring the Parallelism in a Concurrent Node Showing Non-Parallel Nature of a Serial Node Explore Parallelism Provided by the Topology of a Graph Understanding Analysis Color Codes
Scalability Analysis Prerequisites x
Setting Concurrency Specification Setting Data Count Setting Node Weight
Collecting Traces from Applications x
Building an Application for Trace Collection Collecting Trace Files
Building an Application for Trace Collection x
Building an Application on Windows* OS Building an Application on Linux* OS Building an Application on macOS*
Collecting Trace Files x
Collect Traces In the Flow Graph Analyzer GUI Collect Traces Outside the Flow Graph Analyzer GUI
Collect Traces Outside the Flow Graph Analyzer GUI x
Collecting Trace Files with fgtrun Script Collecting Trace Files without fgtrun Script
Analyzer Workflow x
Find Time Regions of Low Concurrency and Their Cause Finding a Critical Path Finding Tasks with Small Durations Reduce Scheduler Overhead using Lightweight Policy Identifying Tasks that Operate on Common Input Support for SYCL
Support for SYCL x
Collect SYCL Application Traces Examine a SYCL Application Graph Find Issues Using Static Rule-check Engine
Examine a SYCL Application Graph x
Hotspot View View Performance Inefficiencies of Data-parallel Constructs
Find Issues Using Static Rule-check Engine x
Issue: Const Reference to a Host Pointer Used to Initialize a Buffer Issue: Host Pointer Accessor Used in a Loop Issue: Data Parallel Construct Inefficiency
Experimental Support for OpenMP* Applications x
Collecting Traces for OpenMP* Applications OpenMP* Constructs in the Per-Thread Task View OpenMP* Constructs in the Graph Canvas
Sample Trace Files x
code_generation Samples performance_analysis Samples
Minimize Analysis Overhead x
Collection Controls to Minimize Analysis Overhead Loop Markup to Minimize Analysis Overhead Filtering to Minimize Analysis Overhead Execution Speed/Duration/Scope Properties to Minimize Analysis Overhead Miscellaneous Techniques to Minimize Analysis Overhead
Analyze MPI Applications x
Model MPI Application Performance on GPU Control Collection with an MPI_Pcontrol Function
Manage Results x
Open a Result Rename an Existing Result Delete a Result Save Results to a Custom Location Work with Standalone HTML Reports Create a Read-only Result Snapshot Create a Result Snapshot Dialog Box
Command Line Interface x
advisor Command Line Interface Reference Offload Modeling Command Line Reference Generate Pre-configured Command Lines
advisor Command Line Interface Reference x
advisor Command Action Reference advisor Command Option Reference
advisor Command Action Reference x
collect command create-project help import-dir mark-up-loops report snapshot version workflow
advisor Command Option Reference x
accuracy append app-working-dir assume-dependencies assume-hide-taxes assume-ndim-dependency assume-single-data-transfer auto-finalize batching benchmarks-sync bottom-up cache-binaries cache-binaries-mode cache-config cache-simulation cache-sources cachesim cachesim-associativity cachesim-cacheline-size cachesim-mode cachesim-sampling-factor cachesim-sets check-profitability clear config count-logical-instructions count-memory-instructions count-memory-objects-accesses count-mov-instructions count-send-latency cpu-scale-factor csv-delimiter custom-config data-limit data-reuse-analysis data-transfer data-transfer-histogram data-transfer-page-size data-type delete-tripcounts disable-fp64-math-optimization display-callstack dry-run duration dynamic enable-cache-simulation enable-data-transfer-analysis enable-task-chunking enforce-baseline-decomposition enforce-fallback enforce-offloads estimate-max-speedup evaluate-min-speedup exclude-files executable-of-interest exp-dir filter filter-by-scope filter-reductions flop force-32bit-arithmetics force-64bit-arithmetics format gpu gpu-carm gpu-kernel-of-interest gpu-sampling-interval hide-data-transfer-tax ignore ignore-app-mismatch ignore-checksums instance-of-interest integrated interval limit loop-call-count-limit loop-filter-threshold loops mark-up mark-up-list memory-level memory-operation-type mix mkl-user-mode model-baseline-gpu model-children model-extended-math model-system-calls module-filter module-filter-mode mpi-rank mrte-mode ndim-depth-limit option-file overlap-taxes pack profile-gpu profile-intel-perf-libs profile-jit profile-python profile-stripped-binaries project-dir quiet recalculate-time record-mem-allocations record-stack-frame reduce-lock-contention reduce-lock-overhead reduce-site-overhead reduce-task-overhead refinalize-survey remove report-output report-template result-dir resume-after return-app-exitcode search-dir search-n-dim select set-dependency set-parallel set-parameter show-all-columns show-all-rows show-functions show-loops show-not-executed show-report small-node-filter sort-asc sort-desc spill-analysis stack-access-granularity stack-stitching stack-unwind-limit stacks stackwalk-mode start-paused static-instruction-mix strategy support-multi-isa-binaries target-device target-gpu target-pid target-process target-system threading-model threads top-down trace-mode trace-mpi track-memory-objects track-stack-accesses track-stack-variables trip-counts verbose with-stack
Offload Modeling Command Line Reference x
run_oa.py Options collect.py Options analyze.py Options
Troubleshooting x
Error Message: Application Sets Its Own Handler for Signal Error Message: Cannot Collect GPU Hardware Metrics for the Selected GPU Adapter Error Message: Memory Model Cache Hierarchy Incompatible Error Message: No Annotations Found Error Message: No Data Is Collected Error Message: Stack Size Is Too Small Error Message: Undefined Linker References to dlopen or dlsym Problem: Broken Call Tree Problem: Code Region is not Marked Up Problem: Debug Information Not Available Problem: No Data Problem: Source Not Available Problem: Stack in the Top-Down Tree Window Is Incorrect Problem: Survey Tool does not Display Survey Report Problem: Unexpected C/C++ Compilation Errors After Adding Annotations Problem: Unexpected Unmatched Annotations in the Dependencies Report Warning: Analysis of Debug Build Warning: Analysis of Release Build
Reference x
Data Reference Dependencies Problem and Message Types Recommendation Reference User Interface Reference
Data Reference x
CPU Metrics Accelerator Metrics
Dependencies Problem and Message Types x
Dangling Lock Data Communication Data Communication, Child Task Inconsistent Lock Use Lock Hierarchy Violation Memory Reuse Memory Reuse, Child Task Memory Watch Missing End Site Missing End Task Missing Start Site Missing Start Task No Tasks in Parallel Site One Task Instance in Parallel Site Orphaned Task Parallel Site Information Thread Information Unhandled Application Exception
Recommendation Reference x
Vectorization Recommendations for C++ Vectorization Recommendations for Fortran
User Interface Reference x
Dialog Box: Corresponding Command Line Dialog Box: Create a Project Dialog Box: Create a Result Snapshot Dialog Box: Options - Assembly Editor Tab Dialog Box: Options - General Dialog Box: Options - Result Location Dialog Box: Project Properties - Analysis Target Dialog Box: Project Properties - Binary/Symbol Search Dialog Box: Project Properties - Source Search Pane: Advanced View Pane: Analysis Workflow Pane: Roofline Chart Pane: GPU Roofline Chart Project Navigator Pane Toolbar: Intel Advisor Annotation Report Window: Dependencies Source Window: GPU Roofline Regions Window: GPU Roofline Insights Summary Window: Memory Access Patterns Source Window: Offload Modeling Summary Window: Offload Modeling Report - Accelerated Regions Window: Perspective Selector Window: Refinement Reports Window: Suitability Report Window: Suitability Source Window: Survey Report Window: Survey Source Window: Threading Summary Window: Vectorization Summary
Window: Refinement Reports x
Tab: Dependencies Report Tab: Memory Access Patterns Report
Appendix x
Data Sharing Problems Notational Conventions Key Concepts Related Information
Data Sharing Problems x
Data Sharing Problem Types Problem Solving Strategies
Data Sharing Problem Types x
Incidental Sharing Independent Updates
Problem Solving Strategies x
Eliminate Incidental Sharing Synchronize Independent Updates Difficult Problems: Choosing a Different Set of Tasks Fix Problems in Code Used by Multiple Parallel Sites Memory That is Accessed Through a Pointer
Eliminate Incidental Sharing x
Examine the Task's Static and Dynamic Extent Verify Whether Incidental Sharing Exists Create the Private Memory Location Pointer Dereferences
Synchronize Independent Updates x
Synchronization Explicit Locking Assign Locks to Transactions Pitfalls from Using Synchronization
Key Concepts x
Glossary Parallelism
Parallelism x
Parallel Processing Terminology Add Parallelism Common Issues When Adding Parallelism Parallel Programming Implementations
Intel® Advisor User Guide

Run Offload Modeling Perspective from Command Line

Intel® Advisor provides several methods to run the Offload Modeling perspective from command line. Use one of the following:

  • Method 1. Run Offload Modeling with a command line collection presets. Use this method if you want to use basic Intel Advisor analysis and modeling functionality, especially for the first-run analysis. This simple method allows you to run multiple analyses with a single command and control the modeling accuracy.
  • Method 2. Run Offload Modeling analyses separately. Use this method if you want to analyze an MPI application or need more advanced analysis customization. This method allows you to select what performance data you want to collect for your application and configure each analysis separately.
  • Method 3. Run Offload Modeling with Python* scripts. Use this method if you need more analysis customization. This method is moderately flexible and allows you to customize data collection and performance modeling steps.
TIP:
See Intel Advisor cheat sheet for quick reference on command line interface.

After you run the Offload Modeling with any method above, you can view the results in Intel Advisor graphical user interface (GUI), command line interface (CLI), or an interactive HTML report. For example, the interactive HTML report is similar to the following:

Prerequisites

  1. Set Intel Advisor environment variables with an automated script.

    The script enables the advisor command line interface (CLI), advisor-python command line tool, and the APM environment variable, which points to the directory with Offload Modeling scripts and simplifies their use.

  2. For SYCL, OpenMP* target, OpenCL™ applications: Set Intel Advisor environment variables to offload temporarily your application to a CPU for the analysis.
    NOTE:
    You are recommended to run the GPU-to-GPU performance modeling to analyze SYCL, OpenMP target, and OpenCL application because it provides more accurate estimations.

Optional: Generate pre-Configured Command Lines

With the Intel Advisor, you can generate pre-configured command lines for your application and hardware. Use this feature if you want to:

  • Analyze an MPI application
  • Customize pre-set Offload Modeling commands

Offload Modeling perspective consists of multiple analysis steps executed for the same application and project. You can configure each step from scratch or use pre-configured command lines that do not require you to provide the paths to project directory and an application executable manually.

Option 1. Generate pre-configured command lines with --collect=offload and the --dry-run option. The option generates:

  • Commands for the Intel Advisor CLI collection workflow
  • Commands that correspond to a specified accuracy level
  • Commands not configured to analyze an MPI application. You need to manually adjust the commands for MPI.

Note: In the commands below, make sure to replace the myApplication with your application executable path and name before executing a command. If your application requires additional command line options, add them after the executable name.

The workflow includes the following steps:

  1. Generate the command using the --collect=offload action with the --dry-run option. Specify accuracy level and paths to your project directory and application executable.

    For example, to generate the low-accuracy commands for the myApplication application, run the following command:

    • On Linux* OS: 
      advisor --collect=offload --accuracy=low --dry-run --project-dir=./advi_results -- ./myApplication
    • On Windows* OS: 
      advisor --collect=offload --accuracy=low --dry-run --project-dir=.\advi_results -- .\myApplication.exe

    You should see a list of commands for each analysis step to get the Offload Modeling result with the specified accuracy level (for the commands above, it is low).

  2. If you analyze an MPI application: Copy the generated commands to your preferred text editor and modify each command to use an MPI tool. For details about the syntax, see Analyze MPI Applications.
  3. Run the generated commands one by one from a command prompt or a terminal.

Option 2. If you have an Intel Advisor graphical user interface (GUI) available on your system and you want to analyze an MPI application from command line, you can generate the pre-configured command lines from GUI.

The GUI generates:

  • Commands for the Intel Advisor CLI collection workflow
  • Commands for a selected accuracy level if you want to run a pre-defined accuracy level or commands for a custom project configuration if you want to enable/disable additional analysis options
  • Command configured for MPI application with Intel® MPI Library. You do not need to manually modify the commands for the MPI application syntax.

For detailed instructions, see Generate Pre-configured Command Lines.

Method 1. Use Collection Presets

For the Offload Modeling perspective, Intel Advisor has a special collection mode --collect=offload that allows you to run several analyses using only oneIntel Advisor CLI command. When you run the collection, it sequentially runs data collection and performance modeling steps. The specific analyses and options depend on the accuracy level you specify for the collection.

Note: In the commands below, make sure to replace the myApplication with your application executable path and name before executing a command. If your application requires additional command line options, add them after the executable name.

For example, to run the Offload Modeling perspective with the default (medium) accuracy level:

  • On Linux* OS: 
    advisor --collect=offload --project-dir=./advi_results -- ./myApplication
  • On Windows* OS: 
    advisor --collect=offload --project-dir=.\advi_results -- .\myApplication.exe

The collection progress and commands for each analysis executed will be printed to a terminal or a command prompt. By default, the performance is modeled for the Intel® Arc™ graphics code-named Alchemist (xehpg_512xve configuration). When the collection is finished, you will see the result summary.

Analysis Details

To change the analyses to run and their option, you can specify a different accuracy level with the --accuracy=<level> option. The default accuracy level is medium.

The following accuracy levels are available:

  • low accuracy includes Survey, Characterization with Trip Counts and FLOP collections, and Performance Modeling analyses.
  • medium (default) accuracy includes Survey, Characterization with Trip Counts and FLOP collections, cache and data transfer simulation, and Performance Modeling analyses.
  • high accuracy includes Survey, Characterization with Trip Counts and FLOP collections, cache, data transfer, and memory object attribution simulation, and Performance Modeling analyses. For CPU applications, also includes the Dependencies analysis.

    For CPU applications, this accuracy level adds a high collection overhead because it includes the Dependencies analysis. This analysis is not required if your application is highly parallelized or vectorized on a CPU or if you know that key hotspots in your application do not have loop-carried dependencies. Otherwise, to learn how dependencies might affect your application performance on a GPU, see How Assumed Dependencies Affect Modeling?>.

For example, to run the low accuracy level: 

advisor --collect=offload --accuracy=low --project-dir=./advi_results -- ./myApplication

To run the high accuracy level: 

advisor --collect=offload --accuracy=high --project-dir=./advi_results -- ./myApplication

If you want to see the commands that are executed at each accuracy level, you can run the collection with the --dry-run option. The commands will be printed to a terminal or a command prompt.

For details about each accuracy level, see Offload Modeling Accuracy Levels in Command Line.

Customize Collection

You can also specify additional options if you want to run the Offload Modeling with custom configuration. This collection accepts most options of the Performance Modeling analysis (--collect=projection) and some options of the Survey, Trip Counts, and Dependencies analyses that can be useful for the Offload Modeling.

IMPORTANT:
Make sure to specify the additional options after the --accuracy option to make sure they take precedence over the accuracy level configurations.

Consider the following action options:

Option

Description

--accuracy=<level>

Set an accuracy level for a collection preset. Available accuracy levels:

  • low
  • medium (default)
  • high

For details, see Offload Modeling Accuracy Levels in Command Line.

--config

Select a target GPU configuration to model performance for. For example, xehpg_512xve (default), gen12_dg1, or gen9_gt3.

See conig for a full list of possible values and mapping to device names.

--gpu

Analyze a SYCL, OpenCL™, or OpenMP* target application on a graphics processing unit (GPU) device. This option automatically adds all related options to each analysis included in the preset.

If you use this option, the high accuracy does not include the Dependencies analysis.

For details about this workflow, see Run GPU-to-GPU Performance Modeling from Command Line.

--data-reuse-analysis

Analyze potential data reuse between code regions. This option automatically adds all related options to each analysis included in the preset.

--enforce-fallback

Emulate data distribution over stacks if stacks collection is disabled. This option automatically adds all related options to each analysis included in the preset.

For details about other available options, see collect.

Method 2. Use per-Analysis Collection

You can collect data and model performance for your application by running each Offload Modeling analysis in a separate command using Intel Advisor CLI. This option allows you to:

  • Control what analyses you want to run to profile your application and what data you want to collect
  • Modify behavior of each analysis you run with an extensive set of options
  • Remodel application performance without re-collecting performance data. This can save time if you want to see how the performance of your application might change with different modeling parameters using the same performance data as baseline
  • Profile and model performance of an MPI application

Consider the following workflow example. Using this example, you can run the Survey, Trip Counts, and FLOP analyses to profile an application and the Performance Modeling to model its performance on a selected target device.

Note: In the commands below, make sure to replace the myApplication with your application executable path and name before executing a command. If your application requires additional command line options, add them after the executable name.

On Linux OS:

  1. Run the Survey analysis. 
    advisor --collect=survey --static-instruction-mix --project-dir=./advi_results -- ./myApplication
  2. Run the Trip Counts and FLOP analyses with data transfer simulation for the default Intel® Arc™ graphics code-named Alchemist (xehpg_512xve configuration). 
    advisor --collect=tripcounts --flop --stacks --cache-simulation=single --target-device=xehpg_512xve --data-transfer=light --project-dir=./advi_results -- ./myApplication
  3. Optional: Run the Dependencies analysis to check for loop-carried dependencies. 
    advisor -collect=dependencies --loop-call-count-limit=16 --select markup=gpu_generic --filter-reductions --project-dir=./advi_results -- ./myApplication

    The Dependencies analysis adds a high collection overhead. You can skip it if your application is highly parallelized or vectorized on a CPU or if you know that key hotspots in your application do not have loop-carried dependencies.. If you are not sure, see Check How Assumed Dependencies Affect Modeling to learn how dependencies affect your application performance on a GPU.

  4. Run the Performance Modeling analysis to model application performance on the default Intel® Arc™ graphics code-named Alchemist(xehpg_512xve configuration). 
    advisor --collect=projection --project-dir=./advi_results

    You will see the result summary printed to the command prompt.

    Tip: If you already have an analysis result saved as a snapshot or a result for an MPI rank, you can use the exp-dir option instead of project-dir to model performance for the result.

On Windows OS:

  1. Run the Survey analysis. 
    advisor --collect=survey --static-instruction-mix --project-dir=.\advi_results -- .\myApplication.exe
  2. Run the Trip Counts and FLOP analyses with cache simulation for the default Intel® Arc™ graphics code-named Alchemist (xehpg_512xve configuration). 
    advisor --collect=tripcounts --flop --stacks --cache-simulation=single --target-device=xehpg_512xve --data-transfer=light --project-dir=.\advi_results -- .\myApplication.exe
  3. Optional: Run the Dependencies analysis to check for loop-carried dependencies. 
    advisor -collect=dependencies --loop-call-count-limit=16 --select markup=gpu_generic --filter-reductions --project-dir=.\advi_results -- myApplication.exe

    The Dependencies analysis adds a high collection overhead. You can skip it if your application is highly parallelized or vectorized on a CPU or if you know that key hotspots in your application do not have loop-carried dependencies.. If you are not sure, see Check How Assumed Dependencies Affect Modeling to learn how dependencies affect your application performance on a GPU.

  4. Run the Performance Modeling analysis to model application performance on the default Intel® Arc™ graphics code-named Alchemist (xehpg_512xve configuration). 
    advisor --collect=projection --project-dir=.\advi_results

    You will see the result summary printed to the command prompt.

    Tip: If you already have a collected analysis result saved as a snapshot or result for an MPI rank, you can use the exp-dir option instead of project-dir to model performance for the result.

For more useful options, see the Analysis Details section below.

Analysis Details

The Offload Modeling workflow includes the following analyses:

  1. Survey to collect initial performance data.
  2. Characterization with trip counts and FLOP to collect performance details.
  3. Dependencies (optional) to identify loop-carried dependencies that might limit offloading.
  4. Performance Modeling to model performance on a selected target device.

Each analysis has a set of additional options that modify its behavior and collect additional performance data. The more analyses you run and option you use, the higher the modeling accuracy.

Consider the following options:

Survey Options

To run the Survey analysis, use the following command line action: --collect=survey.

Recommended action options:

Options

Description

--static-instruction-mix

Collect static instruction mix data. This option is recommended for the Offload Modeling perspective.

--profile-gpu

Analyze a SYCL, OpenCL, or OpenMP target application on a GPU device.

If you use this option, skip the Dependencies analysis.

For details about this workflow, see Run GPU-to-GPU Performance Modeling from Command Line.

Characterization Options

To run the Characterization analysis, use the following command line action: --collect=tripcounts.

Recommended action options:

Options

Description

--flop

Collect data about floating-point and integer operations, memory traffic, and mask utilization metrics for AVX-512 platforms.

--stacks

Enable advanced collection of call stack data.

--cache-simulation=<mode>

Simulate cache behavior for a target device. Available modes:

  • off - disable cache simulation.
  • single - simulate cache behavior only for the selected target device. Make sure to use with the --target-device=<target> option.
  • multi - simulate cache behavior for all available target devices to enable performance remodeling without running the Characterization analysis.

--target-device=<target>

Specify a target graphics processing unit (GPU) to model cache for. For example, xehpg_512xve (default), gen12_dg1, or gen9_gt3. See target-device for a full list of possible values and mapping to device names.

Use with the --cache-simulation=single option.

IMPORTANT:
Make sure to specify the same target device as for the --collect=projection --config=<config>.

--data-transfer=<mode>

Enable modeling data transfers between host and target devices. The following modes are available:

  • Use off (default) to disable data transfer modeling.
  • Use light to model only data transfers.
  • Use medium to model data transfers, attributes memory objects, and tracks accesses to stack memory.
  • Use full to model data transfers, attributes memory objects, tracks accesses to stack memory, and enables data reuse analysis as well.

--profile-gpu

Analyze a SYCL, OpenCL, or OpenMP target application on a GPU device.

If you use this option, skip the Dependencies analysis.

For details about this workflow, see Run GPU-to-GPU Performance Modeling from Command Line.

Dependencies Options

The Dependencies analysis is optional because it adds a high overhead and is mostly necessary if you have scalar loops/functions in your application or if you do not know about loop-carried dependencies in key hotspots. For details about when you need to run the Dependencies analysis, see Check How Assumed Dependencies Affect Modeling.

To run the Dependencies analysis, use the following command line action: --collect=dependencies.

Recommended action options:

Options

Description

--select=<string>

Select loops to run the analysis for.

For the Offload Modeling, the recommended value is --select markup=gpu_generic, which selects only loops/functions profitable for offloading to a target device to reduce the analysis overhead.

For more information about markup options, see Loop Markup to Minimize Analysis Overhead.

NOTE:
The generic markup strategy is recommended if you want to run the Dependencies analysis for an application that does not use SYCL, C++/Fortran with OpenMP target, or OpenCL.

--loop-call-count-limit=<num>

Set the maximum number of call instances to analyze assuming similar runtime properties over different call instances.

The recommended value is 16.

--filter-reductions

Mark all potential reductions with a specific diagnostic.

Performance Modeling Options

To run the Performance Modeling analysis, use the following command line action: --collect=projection.

Recommended action options:

Options

Description

--exp-dir=<path>

Specify a path to an unpacked result snapshot or an MPI rank result to model performance. Use this option instead of project-dir if you already have an analysis result ready.

--config=<config>

Select a target GPU configuration to model performance for. For example, xehpg_512xve (default), gen12_dg1, or gen9_gt3.

IMPORTANT:
Make sure to specify the same target device as for the --collect=tripcounts --target-device=<target>.

For details about configuration files, see config.

--no-assume-dependencies

Assume that a loop does not have dependencies if a loop dependency type is unknown.

Use this option if your application contains parallel and/or vectorized loops and you did not run the Dependencies analysis.

--data-reuse-analysis

Analyze potential data reuse between code regions when offloaded to a target GPU.

IMPORTANT:
Make sure to use --data-transfer=full with --collect=tripcounts for this option to work correctly.

--assume-hide-taxes

Assume that an invocation tax is paid only for the first time a kernel is launched.

--set-parameter

Specify a single-line configuration parameter to modify in a format "<group>.<parameter>=<new-value>". For example, "min_required_speed_up=0".

For details about the option, see set-parameter. For details about some of the possible modifications, see Advanced Modeling Configuration.

--profile-gpu

Analyze a SYCL, OpenCL, or OpenMP target application on a GPU device.

If you use this option, skip the Dependencies analysis.

For details about this workflow, see Run GPU-to-GPU Performance Modeling from Command Line.

See advisor Command Option Reference for more options.

Method 3. Use Python* Scripts

Intel Advisor has three scripts that use the Intel Advisor Python* API to run the Offload Modeling. You can run the scripts with the advisor-python command line tool or with your local Python 3.6 or 3.7.

The scripts vary in functionality and run different sets of Intel Advisor analyses. Depending on what you want to run, use one or several of the following scripts:

  • run_oa.py is the simplest script with limited modification flexibility. Use this script to run the collection and modeling steps with a single command. This script is the equivalent of the Intel Advisor command line collection preset.
  • collect.py is a moderately flexible script that runs only the collection step.
  • analyze.py is a moderately flexible script that runs only the performance modeling step.
NOTE:
The scripts do not support the analysis of MPI applications. For an MPI application, use the per-analysis collection with the Intel Advisor CLI.

You can run the Offload Modeling using different combinations of the scripts and/or the Intel Advisor CLI. For example:

  • Run run_oa.py to profile application and model its performance.
  • Run the collect.py to profile application and analyze.py to model its performance. Re-run analyze.py to remodel with a different configuration.
  • Run the Intel Advisor CLI to collect performance data and analyze.py to model performance. Re-run analyze.py to remodel with a different configuration.
  • Run run_oa.py to collect data and model performance for the first time and run analyze.py to remodel with a different configuration.

Consider the following examples of some typical scenarios with Python scripts.

Note: In the commands below, make sure to replace the myApplication with your application executable path and name before executing a command. If your application requires additional command line options, add them after the executable name.

Example 1. Run the run_oa.py script to profile an application and model its performance for the Intel® ™ graphics code-named Alchemist (xehpg_512xve configuration).

  • On Linux OS: 
    advisor-python $APM/run_oa.py ./advi_results --collect=basic --config=xehpg_512xve -- ./myApplication
  • On Windows OS: 
    advisor-python %APM%\run_oa.py .\advi_results --collect=basic --config=xehpg_512xve -- .\myApplication.exe

You will see the result summary printed to the command prompt.

For more useful options, see the Analysis Details section below.

Example 2. Run the collect.py to profile an application and run the analyze.py to model its performance for the Intel® Arc™ graphics code-named Alchemist (xehpg_512xve configuration).

  • On Linux OS:
    1. Collect performance data. 
      advisor-python $APM/collect.py ./advi_results --collect=basic --config=xehpg_512xve -- ./myApplication
    2. Model application performance. 
      advisor-python $APM/analyze.py ./advi_results --config=xehpg_512xve

    You will see the result summary printed to the command prompt.

  • On Windows OS:
    1. Collect performance data. 
      advisor-python %APM%\collect.py .\advi_results --collect=basic --config=xehpg_512xve -- .\myApplication.exe
    2. Model application performance. 
      advisor-python %APM%\analyze.py .\advi_results --config=xehpg_512xve

For more useful options, see the Analysis Details section below.

Analysis Details

Each script has a set of additional options that modify its behavior and collect additional performance data. The more analyses you run and options you use, the higher the modeling accuracy.

Collection Options

The following options are applicable to the run_oa.py and collect.py scripts.

Option

Description

--collect=<mode>

Specify data to collect for your application:

  • Use basic to run only Survey, Trip Counts and FLOP analyses, analyze data transfer between host and device memory, attribute memory objects to loops, and track accesses to stack memory. This value corresponds to the Medium accuracy.
  • Use refinement to run only Dependencies analysis. Do not analyze data transfers.
  • Use full (default) to run Survey, Trip Counts, FLOP, and Dependencies analyses, analyze data transfer between host and device memory and potential data reuse, attribute memory objects to loops, and track accesses to stack memory. This value corresponds to the High accuracy.

See Check How Assumed Dependencies Affect Modeling to learn when you need to collect dependency data.

--config=<config>

Select a target GPU configuration to model performance for. For example, xehpg_512xve (default), gen12_dg1, or gen9_gt3.

IMPORTANT:
For collect.py, make sure to specify the same value of the --config option for the analyze.py.

For details about configuration files, see config.

--markup=<markup-mode>

Select loops to collect Trip Counts and FLOP and/or Dependencies data for with a pre-defined markup algorithm. This option decreases collection overhead.

By default, it is set to generic to analyze all loops profitable for offloading.

--gpu

Analyze a SYCL, OpenCL, or OpenMP target application on a GPU device.

For details about this workflow, see Run GPU-to-GPU Performance Modeling from Command Line.

For a full list of available options, see:

Performance Modeling Options

The following options are applicable to the run_oa.py and analyze.py scripts.

Option

Description

--config=<config>

Select a target GPU configuration to model performance for. For example, xehpg_512xve (default), gen12_dg1, or gen9_gt3.

IMPORTANT:
For analyze.py, make sure to specify the same value of the --config option for the collect.py.

For details about configuration files, see config.

--assume-parallel

Assume that a loop does not have dependencies if there is no information about the loop dependency type and you did not run the Dependencies analysis.

--data-reuse-analysis

Analyze potential data reuse between code regions when offloaded to a target GPU.

IMPORTANT:
Make sure to use --collect=full when running the analyses with collect.py or use the --data-transfer=full when running the Trip Counts analysis with Intel Advisor CLI.

--gpu

Analyze a SYCL, OpenCL, or OpenMP target application on a GPU device.

For details about this workflow, see Run GPU-to-GPU Performance Modeling from Command Line.

For a full list of available options, see:

Next Steps

Continue to explore the Offload Modeling results with a preferred method. For details about the metrics reported, see Accelerator Metrics.

Parent topic: Model Offloading to a GPU

See Also

Model Offloading to a GPU Find high-impact opportunities to offload/run your code and identify potential performance bottlenecks on a target graphics processing unit (GPU) by running the Offload Modeling perspective.
Command Line Interface This reference section describes the Intel® Advisor command line interface (CLI) used to run the analysis.
Model MPI Application Performance on GPU You can model your MPI application performance on a target graphics processing unit (GPU) device to determine whether you can get a performance speedup from offloading the application to the GPU.
Level Two Title