Intel® Advisor User Guide

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A newer version of this document is available.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

Offload Modeling Accuracy Levels in Command Line

For each perspective, Intel® Advisor has several levels of collection accuracy. Each accuracy level is a set of analyses and properties that control what data is collected and the level of collection details. The higher accuracy value you choose, the higher runtime overhead is added.

You can generate commands for a desired accuracy level from the Intel Advisor GUI. See Generate Command Lines from GUI for details.

NOTE:
There is a variety of techniques available to minimize data collection, result size, and execution overhead. Check Minimize Analysis Overhead.

CPU-to-GPU Modeling

For the CPU-to-GPU modeling, the following accuracy levels are available:

Comparison / Accuracy Level

Low

Medium

High

Overhead

5 - 10x

15 - 50x

50 - 80x

Goal

Model performance of an application that is mostly compute bound and does not have dependencies

Model application performance considering memory traffic for all cache and memory levels

Model application performance with all potential limitations for offload candidates

Analyses

Survey + Characterization (Trip Counts and FLOP) + Performance Modeling with no assumed dependencies

Survey + Characterization (Trip Counts and FLOP with cache simulation for the selected target device, callstacks, and light data transfer simulation) + Performance Modeling with no assumed dependencies

Survey + Characterization (Trip Counts and FLOP with cache simulation for the selected target device, callstacks, and medium data transfer simulation) + Dependencies + Performance Modeling with assumed dependencies

Result

Basic Offload Modeling report that shows potential speedup and performance metrics estimated on a target considering memory traffic from execution units to L1 cache only. The result might be inaccurate for memory-bound applications.

Offload Modeling report extended with data transfers estimated between host and device platforms considering memory traffic for all cache and memory levels

Offload Modeling report with detailed data transfer estimations and automated check for loop-carried dependencies for more accurate search for the most profitable regions to offload

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.

NOTE:
Families of Intel® Xe graphics products starting with Intel® Arc™ Alchemist (formerly DG2) and newer generations feature GPU architecture terminology that shifts from legacy terms. For more information on the terminology changes and to understand their mapping with legacy content, see GPU Architecture Terminology for Intel® Xe Graphics.

Low Accuracy

To model application performance for with low accuracy for a default target device, run the following command:

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

This command runs the following analyses one by one:

  1. Survey analysis:
    advisor --collect=survey --auto-finalize --static-instruction-mix --project-dir=./advi_results -- ./myApplication 
  2. Characterization analysis to collect trip count and FLOP data
    advisor --collect=tripcounts --flop --auto-finalize --target-device=xehpg_512xve --project-dir=./advi_results -- ./myApplication
  3. Performance modeling:
    advisor --collect=projection --no-assume-dependencies --config=xehpg_512xve --project-dir=./advi_results

Medium Accuracy

This accuracy is set by default. To model application performance with medium accuracy for a default target device, run the following command:

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

This command runs the following analyses one by one:

  1. Survey analysis:
    advisor --collect=survey --auto-finalize --static-instruction-mix --project-dir=./advi_results -- ./myApplication
  2. Characterization analysis to collect trip count and FLOP data
    advisor --collect=tripcounts --flop --stacks --auto-finalize --cache-simulation=single --data-transfer=light --target-device=xehpg_512xve --project-dir=./advi_results -- ./myApplication
  3. Performance modeling:
    advisor --collect=projection --no-assume-dependencies --config=xehpg_512xve --project-dir=./advi_results

High Accuracy

To model application performance with high accuracy for a default target device, run the following command:

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

This command runs the following analyses one by one:

  1. Survey analysis:
    advisor --collect=survey --auto-finalize --static-instruction-mix --project-dir=./advi_results -- ./myApplication
  2. Characterization analysis to collect trip count and FLOP data
    advisor --collect=tripcounts --flop --stacks --auto-finalize --cache-simulation=single --target-device=xehpg_512xve --data-transfer=medium --project-dir=./advi_results -- ./myApplication
  3. Dependencies analysis:
    advisor --collect=dependencies --filter-reductions --loop-call-count-limit=16 --select markup=gpu_generic --project-dir=./advi_results -- ./myApplication
  4. Performance modeling:
    advisor --collect=projection --config=xehpg_512xve --project-dir=./advi_results

See Check How Dependencies Affect Modeling for a recommended strategy to check for loop-carried dependencies.

GPU-to-GPU Modeling

For the GPU-to-GPU modeling, the following accuracy levels are available:

Comparison / Accuracy Level

Low

Medium

High

Overhead

5 - 10x

15 - 50x

15 - 50x

Goal

Model performance of an application that is mostly compute bound

Model application performance considering memory traffic for all cache and memory levels

Model application performance with all potential limitations for offload candidates

Analyses

Survey + Characterization (Trip Counts and FLOP) + Performance Modeling

Survey + Characterization (Trip Counts and FLOP with light data transfer simulation) + Performance Modeling

Survey + Characterization (Trip Counts and FLOP with medium data transfer simulation) + Performance Modeling

Result

Basic Offload Modeling report that shows potential speedup and performance metrics estimated on a target considering memory traffic from execution units to L1 cache only. The result might be inaccurate for memory-bound applications.

Offload Modeling report extended with data transfers estimated between host and device platforms

Offload Modeling report with detailed data transfer estimations for more accurate search for the most profitable regions to offload

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.

Low Accuracy

To model application performance for with low accuracy for a default target device, run the following command:

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

This command runs the following analyses one by one:

  1. Survey analysis:
    advisor --collect=survey --auto-finalize --static-instruction-mix --profile-gpu --project-dir=./advi_results -- ./myApplication 
  2. Characterization analysis to collect trip count and FLOP data
    advisor --collect=tripcounts --flop --auto-finalize --target-device=xehpg_512xve --profile-gpu --project-dir=./advi_results -- ./myApplication
  3. Performance modeling:
    advisor --collect=projection --no-assume-dependencies --config=xehpg_512xve --profile-gpu --project-dir=./advi_results

Medium Accuracy

This accuracy is set by default. To model application performance with medium accuracy for a default target device, run the following command:

advisor --collect=offload --gpu --project-dir=./advi_results -- ./myApplication

This command runs the following analyses one by one:

  1. Survey analysis:
    advisor --collect=survey --auto-finalize --static-instruction-mix --profile-gpu --project-dir=./advi_results -- ./myApplication
  2. Characterization analysis to collect trip count and FLOP data
    advisor --collect=tripcounts --flop --auto-finalize --data-transfer=light --target-device=xehpg_512xve --profile-gpu --project-dir=./advi_results -- ./myApplication
  3. Performance modeling:
    advisor --collect=projection --no-assume-dependencies --config=xehpg_512xve --profile-gpu --project-dir=./advi_results

High Accuracy

To model application performance with high accuracy for a default target device, run the following command:

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

This command runs the following analyses one by one:

  1. Survey analysis:
    advisor --collect=survey --auto-finalize --static-instruction-mix --profile-gpu --project-dir=./advi_results -- ./myApplication
  2. Characterization analysis to collect trip count and FLOP data
    advisor --collect=tripcounts --flop --auto-finalize --target-device=xehpg_512xve --profile-gpu --data-transfer=medium --project-dir=./advi_results -- ./myApplication
  3. Performance modeling:
    advisor --collect=projection --config=xehpg_512xve --profile-gpu --project-dir=./advi_results

After you run the perspective, you can view the results in the Intel Advisor GUI, in CLI, or an interactive HTML report.

See Also