New Features for TotalView 8.9.1

Enhanced CUDA Support

Support for CUDA 3.2 on Linux x86 and Linux x86-64 has been added. The following features are included

Support for apps built with the CUDA 3.0, 3.1, or 3.2 SDK

Compatibility with CUDA 3.0, 3.1 or 3.2 drivers

Support for CUDA function calls on the stack (in addition to the inlined function support in previous versions)

Handles exceptions in CUDA code

Display of variables in GPU hardware registers

Support for host pinned memory regions

Support for CUDA contexts

New CLI commands for CUDA functionality

Array Statistics CLI Commands

You can now use a dprint -stats to programmatically gather the array statistics that were previously only available in the GUI. These can be directly printed to the TCL prompt or placed in a TCL associative array with the -data option.

Array Type Inheritance

In the multi-dimensional array display, changes to types made in the variable window are picked up prior to launching the array display, resulting in the most up-to-date information.

Enhanced Parallel Backtrace

We've improved the way we store stack trace data to improve the representation of recursive function calls in the parallel backtrace view. In addition, parallel backtrace data is now available through the CLI with the dcalltree command.