Main Content

Code Generation Reports

GPU Coder™ produces a code generation report that helps you to:

  • Debug code generation issues and verify that your MATLAB® code is suitable for code generation.

  • View generated CUDA® code.

  • Trace between MATLAB source code and generated CUDA code.

  • See how the code generator determines and propagates type information for variables and expressions in your MATLAB code.

  • Identify potential issues in the generated code.

  • Access additional reports available with Embedded Coder®.

Report Generation

When you enable report generation or when an error occurs, the code generator produces a code generation report. To control production and opening of a code generation report, use app settings, codegen options, or configuration object properties.

In the GPU Coder app:

  • To generate a report, set Always create a report to Yes.

  • If you want the app to open the report for you, set Automatically launch a report if one is generated to Yes.

At the command line, use codegen options:

  • To generate a report, use the -report option.

  • To generate and open a report, use the -launchreport option.

Alternatively, use the configuration object properties (coder.CodeConfig):

  • To generate a report, set GenerateReport to true.

  • If you want codegen to open the report for you, set LaunchReport to true.

Report Location

The code generation report is named report.mldatx. It is located in the html subfolder of the code generation output folder. If you have MATLAB R2018a or later, you can open the report.mldatx file by double-clicking it.

Errors and Warnings

View code generation error, warning, and information messages on the All Messages tab. To highlight the source code for an error or warning, click the message. It is a best practice to address the first message because subsequent errors and warnings can be related to the first message.

View compilation and linking errors and warnings on the Build Logs tab.

Files and Functions

The report lists MATLAB source functions and generated files. In the MATLAB Source pane, the Function List view organizes functions according to the containing file. To visualize functions according to the call structure, use the Call Tree view.

To view a function in the code pane of the report, click the function in the list. Clicking a function opens the file that contains the function. To edit the selected file in the MATLAB Editor, click Edit in MATLAB or click a line number in the code pane.

If you have Embedded Coder and generate the report with traceability enabled, to view the source code and generated code next to each other in the code pane, click Trace Code. You can interactively trace between the source code and the generated code. See Interactively Trace Between MATLAB Code and Generated C/C++ Code (Embedded Coder).

If you want to move the generated files for standalone code (library or executable) to another development environment, you can put them into a zip file by clicking Package Code.

Specialized Functions or Classes

When a function is called with different types of inputs or a class uses different types for its properties, the code generator produces specializations. In the MATLAB Source pane, numbered functions (or classes) indicate specializations. For example:

MATLAB Source

To view a MATLAB function in the code pane, click the function in the MATLAB Source pane. To see information about the type of a variable or expression, pause over the variable or expression.

In the code pane, syntax highlighting of MATLAB source code helps you to identify MATLAB syntax elements. Syntax highlighting also helps you to identify certain code generation attributes such as whether a function is extrinsic or whether an argument is constant.

CUDA Kernels

The green GPU marker next to mandelbrot_count function indicates that the generated code has both CPU and GPU sections. The green vertical bar indicates the lines of code that are mapped to the GPU. To see information about the type of a variable or expression and the name of the corresponding GPU Kernel Function, pause over the variable or expression. When you select highlighted code by clicking it, the code becomes blue and you can see the information even when you move your pointer away from the selection. The code remains selected until you press Esc or select different code.

Entry-point function in the code pane highlighting code sections mapped to GPU kernels

Extrinsic Functions

In the MATLAB code, the report identifies an extrinsic function with purple text. The information window indicates that the function is extrinsic.

Entry-point function in the code pane highlighting extrinsic functions

Constant Arguments

In the MATLAB code, orange text indicates a compile-time constant argument to an entry-point function or a specialized function. The information window includes the constant value.

Knowing the value of the constant arguments helps you to understand generated function signatures. It also helps you to see when code generation created function specializations for different constant argument values.

To export the value to a variable in the workspace, click .

Generated Code

To view a generated CUDA source or header file in the code pane, click the file in the Files tab on the Generated Code pane. The GPU Kernels tab on the Generated Code pane contains the list of CUDA kernels in the generated code. Click on the kernel name to navigate directly to the definition of the corresponding kernel in the generated code.

MATLAB Variables

The Variables tab provides information about the variables for the selected MATLAB function. To select a function, click the function in the MATLAB Source pane.

The variables table shows:

  • Class, size, and complexity

  • Properties of fixed-point types

This information helps you to debug errors, such as type mismatch errors, and to understand how the code generator propagates types and represents data in the generated code.

Visual Indicators on the Variables Tab

This table describes symbols, badges, and other indicators in the variables table.

Column in the Variables TableIndicatorDescription
NameexpanderVariable has elements or properties that you can see by clicking the expander.
Name{:}Heterogeneous cell array (all elements have the same properties)
Name{n}nth element of a heterogeneous cell array
Classv > nv is reused with a different class, size, and complexity. The number n identifies each unique reuse (a reuse with a unique set of properties). When you pause over a renamed variable, the report highlights only the instances of this variable that share the class, size, and complexity.
Size:nVariable-size dimension with an upper bound of n
Size:?Variable-size with no upper bound
SizeitalicsVariable-size array whose dimensions do not change size during execution
Classsparse prefixSparse array
Classcomplex prefixComplex number

Array Layout Indicators on the Variables Tab

This table describes the badges that indicate array layout in the variables table.

BadgeDescription

Row-major array layout.

Column-major array layout.

A mixture of row-major and column-major layouts.

See Row-Major and Column-Major Array Layouts.

Tracing Code

You can trace between MATLAB source code and generated CUDA code by using one of these methods:

  • Interactively visualize the mapping between the MATLAB code and the generated code. To access interactive tracing, in the report, click Trace Code. The Trace Code button is enabled only if you have Embedded Coder and you enabled code traceability when you generated code. See Interactively Trace Between MATLAB Code and Generated C/C++ Code (Embedded Coder).

  • Include source code as comments in the generated CUDA code. In a comment, the code generator produces a tag that helps you find the corresponding MATLAB source code. If you have Embedded Coder, the tag is a link to the source code. See Trace Between Generated CUDA Code and MATLAB Source Code.

Code Insights

The code generator can detect and report issues that can potentially occur in the generated code. View the messages on the Code Insights tab. The issues include:

  • Potential differences between the behavior of the generated code and the behavior of the MATLAB code. The report includes potential differences messages only if you enabled potential differences reporting. See Potential Differences Reporting.

  • GPU code generation diagnostics report that identifies issues during code generation and suggests potential solutions to maximize performance.

  • Potential row-major issues. See Code Design for Row-Major Array Layout.

Additional Reports

The Summary tab can have links to these additional reports:

Report Limitations

  • The report does not show full information for unrolled loops. It displays data types of one arbitrary iteration.

  • The report does not show information about dead code.

Related Topics