Intel® Advisor Help
To plot a Roofline chart, the Intel® Advisor does the following:
Intel® Advisor calculates compute operations (FLOP and INTOP) as a weighted sum of the following groups of instructions: BASIC COMPUTE, FMA, BIT, DIV, POW, MATH.
Intel Advisor automatically determines data type in the collected operations using the dst register.
Set Intel Advisor environment variables with an automated script to enable the advisor command line interface (CLI).
There are two methods to run the CPU Roofline. Use one of the following:
Info: 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.
Method 1. Run the Shortcut Command
To collect data for a CPU Roofline chart with a shortcut, run the following command:
advisor --collect=roofline --project-dir=./advi_results -– ./myApplication
This command collects data for a basic CPU Roofline chart based on the Cache-Aware Roofline model. You can add other option to the command to collect more data. See Analysis Details below for more options.
Method 2. Run the Analyses Separately
Use this method if you want to analyze an MPI application.
advisor --collect=survey --project-dir=./advi_results -- ./myApplication
advisor --collect=tripcounts --flops --project-dir=./advi_results -- ./myApplication
These commands collect data for a basic CPU Roofline chart based on the Cache-Aware Roofline model. You can add other option to the command to collect more data. See Analysis Details below for more options.
You can view the results in the Intel Advisor graphical user interface (GUI), or generate an interactive HTML report. See View the Results below for details.
Analysis Details
The CPU / Memory Roofline Insights workflow includes the following analyses:
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 more useful data about your application you get.
Consider the following options:
Roofline Options
To run the Roofline analysis, use the following command line action: --collect=roofline.
Recommended action options:
Options |
Description |
---|---|
--stacks |
Enable advanced collection of call stack data. Use this option to get a CPU Roofline with callstacks. |
--enable-cache-simulation |
Model CPU cache behavior on your target application. Use this option to get a Memory-level CPU Roofline that shows data for all memory levels. |
--cache-config=<config> |
Set the cache hierarchy to collect modeling data for CPU cache behavior. Use with enable-cache-simulation. The value should follow the template: [<num_of_caches>]:[<num_of_ways_caches_connected> ]:[<cache_size>]:[<cacheline_size>] for each of three cache levels separated with a /. |
--cachesim-associativity=<num> |
Set the cache associativity for modeling CPU cache behavior: 1 | 2 | 4 | 8 (default) | 16. Use with enable-cache-simulation. |
--cachesim-mode=<mode> |
Set the focus for modeling CPU cache behavior: cache-misses | footprint | utilization. Use with enable-cache-simulation. |
See advisor Command Option Reference for more options.
Memory Access Patterns Options
The Memory Access Patterns analysis is optional because it adds a high overhead. This analysis does not add more information to the CPU Roofline chart. The results are added to the Refinement report, which you can view from GUI or from CLI. Use it to understand the Memory-Level Roofline chart better and get more detailed optimization recommendations.
To run the Memory Access Patterns analysis, use the following command line action: --collect=map.
Recommended action options:
Options |
Description |
---|---|
--select=<string> |
Select loops for the analysis by loop IDs, source locations, or criteria such as scalar, has-issue, or markup=<markup-mode>. This option is required. See select for more selection options. |
--enable-cache-simulation |
Model CPU cache behavior on your target application. |
--cachesim-cacheline-size=<num> |
Set the cache line size (in bytes) for modeling CPU cache behavior: 4 | 8 | 16 | 32 | 64 (default) | 128 | 256 | 512 | 1024 | 2048 | 4096 | 8192 | 16384 | 32768 | 65536. Use with enable-cache-simulation. |
--cachesim-sets=<num> |
Set the cache set size (in bytes) for modeling CPU cache behavior: 256 | 512 | 1024 | 2048 | 4096 (default) | 8192. Use with enable-cache-simulation. |
See advisor Command Option Reference for more options.
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. This analysis does not add more information to the CPU Roofline chart. The results are added to the Refinement report, which you can view from GUI or from CLI. Use it to get more detailed optimization recommendations.
To run the Dependencies analysis, use the following command line action: --collect=dependencies.
Recommended action options:
Options |
Description |
---|---|
--select=<string> |
Select loops for the analysis by loop IDs, source locations, criteria such as scalar, has-issue, or markup=<markup-mode>. This option is required. See select for more selection options. |
--filter-reductions |
Mark all potential reductions with a specific diagnostic. |
See advisor Command Option Reference for more options.
Intel Advisor provides several ways to work with the CPU / Memory Roofline Insights results.
View Results in GUI
When you run Intel Advisor CLI, a project is created automatically in the directory specified with --project-dir. All the collected results and analysis configurations are stored in the .advixeproj project, which you can view in the Intel Advisor.
To open the project in GUI, run the following command:
advisor-gui <project-dir>
You will see the CPU Roofline report that includes:
Roofline chart that plots an application's achieved performance and arithmetic intensity against the CPU maximum achievable performance
Additional information about your application in the Advanced View pane under the chart, including source code, detailed code analytics for trip counts and FLOP/INTOP data, optimization recommendations, and compiler diagnostics
Select a dot on the Roofline chart to see details for the selected loop in all tabs of the Advanced View pane
View an Interactive HTML Report
Intel Advisor enables you to export an interactive HTML report for the CPU Roofline chart, which you can open in your preferred browser and share.
When you open the report, you see the CPU Roofline chart with the selected configuration. In this report, you can:
Expand the Performance Metrics Summary drop-down to view the summary performance characteristics for your application.
Double-click a dot on the chart to see a roof ruler that point to exact roofs that bound the dot.
Hover over a dot to see a detailed tooltip with performance metrics.
If you have a Memory-level Roofline report, you can also:
Select memory levels to show dots for from the filter drop-down list on the chart.
Double-click a dot on the chart to expand it for other memory levels and see roof rulers.
For details on exporting HTML reports, see Work with Standalone HTML Reports.
Save a Read-only Snapshot
A snapshot is a read-only copy of a project result, which you can view at any time using the Intel Advisor GUI. To save an active project result as a read-only snapshot:
advisor --snapshot --project-dir=<project-dir> [--cache-sources] [--cache-binaries] -- <snapshot-path>
where:
To open the result snapshot in the Intel Advisor GUI, you can run the following command:
advisor-gui <snapshot-path>
You can visually compare the saved snapshot against the current active result or other snapshot results.
Continue to Examine Bottlenecks on CPU Roofline Chart if you generated the Cache-Aware CPU Roofline.
Continue to Examine Relationships Between Memory Levels if you generated the Memory-level CPU Roofline with or without call stacks.
These sections are GUI-focused, but you can still use them to understand the results. For details about the metrics reported, see CPU Metrics.