Offload Modeling Report - Accelerated Regions
Review the summary offload characteristics for your application to decide if the application is recommended for offloading. The pane highlights total speedup, number of loops and functions offloaded, and a fraction of code accelerated.
Select code regions to show in the report based on offload type:
Review the detailed information about your application performance measured on a host platform and its performance modeled on a target platform. For details about metrics reported, see Accelerator Metrics.
Depending on a perspective configuration, you might see different metrics reported and some metrics might be not accurate. Refer to the following topics for interpretation details:
Switch between the
Data Transfer Estimationstab and
Data Transfer Estimationstab, examine details about estimated data transfers in a loop and memory objects tracked and review data transfer recommendations. Select a loop in the
CPU+GPUtable to see the data transfer estimations for it. Click a recommendation to expand it and view hints for offloading your code to another accelerator, code samples, and links to useful resources.
For details about how to read this pane, see Examine Data Transfers for Modeled Regions.
You need to enable data transfer analysis before running the perspective to see metrics in this pane,
Detailstab, view performance metrics for a selected function/loop.
Examine the source code and offload details for each source line. Select a loop in the
CPU+GPUtable to focus on the corresponding part of the source code.
Examine offload details about a loop. Select a loop in the
CPU+GPUtable to see the offload speed-up and status for this loop. If the loop is not recommended for offloading, this pane reports the reason why it is not recommended. for details, see Investigate Not-Offloaded Code Regions.
This pane is collapsed by default. Click an arrow icon on the top of it to expand the pane.