Intel® Advisor User Guide

ID 766448
Date 3/31/2023
Public

A newer version of this document is available.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

Document Table of Contents

Examine Bottlenecks on CPU Roofline Chart

Accuracy Level

Low

Enabled Analyses

Survey + FLOP (Characterization)

Result Interpretation

The farther a dot is from the topmost roofs, the more room for improvement there is. In accordance with Amdahl's Law, optimizing the loops that take the largest portion of the program's total run time will lead to greater speedups than optimizing the loops that take a smaller portion of the run time.

NOTE:
This topic describes data as it is shown in the CPU Roofline report in the Intel Advisor GUI. You can also view the result in an HTML report, but data arrangement and panes may vary.
  • By dot size and color, identify loops that take most of total program time and/or located very low in the chart. For example:
    • Small, green dots take up relatively little time, so are likely not worth optimizing.
    • Large, red dots take up the most time, so the best candidates for optimization are the large, red dots with a large amount of space between them and the topmost roofs.

    NOTE:
    You can switch between coloring the dots by execution time and coloring the dots by type (scalar or vectorized) in the roof view menu on the right.

  • Depending on the dots position, identify what the loops are bounded by. Intel® Advisor marks the roofline zones on the chart to help you identify what roofs bound the loop:
    • Loop is bounded by memory roofs.
    • Loop is bounded by compute roofs.
    • Loop is bounded by both memory and compute roofs.
  • In the Recommendations tab, scroll down to the Roofline Guidance section that provides you hints on next optimization steps for a selected loop/function.

The roofs above a dot represent the restrictions preventing it from achieving a higher performance, although the roofs below can contribute somewhat. Each roof represents the maximum performance achievable without taking advantage of a particular optimization, which is associated with the next roof up. Depending on a dot position, you can try the following optimizations.

NOTE:
For more precise optimization recommendations, see the Roofline Guidance in Code Analytics and Roofline Conclusions in Recommendations tabs.

Dot Position

Reason

To Optimize

Below a memory roof (DRAM Bandwidth, L1 Bandwidth, so on)

The loop/function uses memory inefficiently.

Run a Memory Access Patterns analysis for this loop.

  • If MAP analysis suggests cache optimization, make any appropriate optimizations.
  • If cache optimization is impossible, try reworking the algorithm to have a higher AI.

Below Vector Add Peak

The loop/function under-utilizes available instruction sets.

Check Traits column in the Survey report to see if FMAs are used.

  • If FMA is not used, try altering your code or compiler flags to induce FMA usage.

Just above Scalar Add Peak

The loop/function is undervectorized.

Check vectorization efficiency and performance issues in the Survey. Follow the recommendations to improve it if it's low.

Below Scalar Add Peak

The loop/function is scalar.

Check the Survey report to see if the loop vectorized. If not, try to get it to vectorize if possible. This may involve running Dependencies to see if it's safe to force it.

In the following Roofline chart representation, loops A and G (large red dots), and to a lesser extent B (yellow dot far below the roofs), are the best candidates for optimization. Loops C, D, and E (small green dots) and H (yellow dot) are poor candidates because they do not have much room to improve or are too small to have significant impact on performance.


Some algorithms are incapable of breaking certain roofs. For instance, if Loop A in the example above cannot be vectorized due to dependencies, it cannot break the Scalar Add Peak.

TIP:
If you cannot break a memory roof, try to rework your algorithm for higher arithmetic intensity. This will move you to the right and give you more room to increase performance before hitting the memory bandwidth roof. This would be the appropriate approach to optimizing loop F in the example, as well as loop G if its cache usage cannot be improved.

Analyze Specific Loops

Select a dot on the chart, open the Code Analytics tab to view detailed information about the selected loop:

  • Refer to Loop Information pane to examine total time, self time, instruction sets used, and instruction mix for the selected loop. Intel Advisor provides:
    • Static instruction mix data that is based on static assembly code analysis within a call stack. Use static instruction mix to examine instruction sets in the inner-most functions/loops.
    • Dynamic instruction mix that is based on dynamic assembly code analysis. This metric represents the total count of instructions executed by your function/loop. Use dynamic instruction mix to examine instruction sets in the outer loops and in complex loop-nests.

    Intel Advisor automatically determines the data type used in operations. View the classes of instructions grouped by categories in instruction mix:

    Category