Data Display and Visualization

TotalView for HPC provides comprehensive and flexible tools for developers to explore large and complex data sets. The TotalView for HPC data window supports browsing through complex structures and arrays. Powerful slicing and filtering helps developers manage arrays of thousands or even millions of elements. Key features include:

• – Data watchpoints provide answers to questions about how data is changing.
• – Built-in graphical visualization displays a quick view of complex numerical data.
• – Type transformations, especially C++ View, help display data in a meaningful way.

Tools for Multi-threaded and

Parallel Applications

TotalView for HPC is designed to debug multi-process, multi-threaded programs, with a rich feature set to support fine-grained control over individual or multiple threads and processes. This level of control makes it possible to quickly resolve problems like deadlocks or race conditions in a complex program that spawns thousands of processes and threads across a broad network of servers.

When a program creates processes and threads, TotalView for HPC can automatically bring them under its control, whether they are local or remote. If the processes are already running, TotalView for HPC can acquire them as well, avoiding the need to run multiple debuggers.

TotalView for HPC places a server on each remote processor at the time of launch, then communicates with the main TotalView process. This debugging architecture gives users a central location from which they can manage and examine all aspects of their program.

Batch and Automated Debugging

Users can set up unattended batch debugging sessions using the TotalView for HPC scripting tool, TVScript. First, the user defines a series of events that may occur within the target program. TVScript loads the program under its control, sets breakpoints as necessary, and runs the program. At each program stop, TVScript logs the data for review when the job has completed.

Users can also execute message passing interface (MPI) programs using TVScript. While batch debugging of large-scale MPI applications through TVScript has long been a powerful tool, TVScript is fully certified to handle 1024 process jobs, 2048 threads per process, or more than two million running operations.

Debugging on a Remote Host

Using the Remote Display Client, users can easily set up and operate a TotalView debug session that is running on another system. A licensed copy of TotalView must be installed on the remote machine; however, the customer does not need an additional license to run the client.

The client also supports submission of jobs to batch queuing systems, such as Altair PBS Professional® and Load Leveler.

Using the TotalView Server, users can debug programs located on remote machines. Debugging a remote process is similar to debugging a native process, although performance depends on the load on the remote host and network latency. TotalView runs and accesses the process tvdsr on the remote machine.

CUDA Debugger

All host debugging features are supported, except ReplyEngine

The TotalView for HPC CUDA debugger is an integrated debugging tool capable of simultaneously debugging CUDA code running on the host system and the NVIDIA® GPU. CUDA support is an extension to the standard version of TotalView, and is capable of debugging 64-bit CUDA programs. Debugging 32-bit CUDA programs is currently not supported.

Supported features include:

• – Debug a CUDA application running directly on GPU hardware
• – Set breakpoints, pause execution, and single step in GPU code
• – View GPU variables in PX registers, and in local, parameter, global, or shared memory
• – Access runtime variables, such as threadldx, blockldx, and blockDim
• – Debug multiple GPU devices per process
• – Support for the CUDA MemoryChecker
• – Debug remote, distributed and clustered systems

Memory Debugging

TotalView for HPC includes a fully integrated version of the MemoryScape product for debugging memory issues. MemoryScape can monitor how the program uses malloc() and free() and related functions.

MemoryScape is also available as a standalone product.

Reverse Debugging

ReplayEngine records all program activities as it executes within TotalView for HPC. After recording information, users can move forward and backward by function, line or instruction. They can examine data and state in the past, as if it were a live process. Using ReplayEngine eliminates the cycle of starting and restarting common in debugging and greatly helps in finding hard-to-produce bugs.

Operating System

Compatibility

Perforce TotalView for HPC is operational on:

• – Red Hat® Enterprise Linux 6, and 7
• – Red Hat Fedora 23, 24 and 25
• – Ubuntu 14.04,15.04, 15.10, and, 16.04
• – Open SuSE 12 and 13
• – SuSE Linux Enterprise Server 11 and 12

Supported Compilers

Perforce TotalView for HPC supports the following compilers for Linux x86-64 platforms:

• – Intel Parallel Studio XE 15-17
• – GNU 4.3-6.2
• – PGI Workstation 11.2-16.1
• – Oracle Studio 12
• – C and C++: LLVM Clang 3.3 – 3.5
• – Fortran: Absoft Pro 16
• – Fortran: Lahey Fortran 6.2 - 8

Supported Message Passing Interface (MPI) Products

Perforce TotalView for HPC supports the following MPI products for Linux x86-64 platforms:

• – Argonne MPICH and MPICH2
• – GNU SLURM
• – Intel MPI
• – Open MPI
• – OSU MVPICH and MVAPICH2
• – HPE Performance Software – Message Passing Interface

Hardware Compatibility

Perforce TotalView for HPC is supported on the following Gen9 and Gen10 HPE systems:

• – HPE SGI 8600 system
• – HPE ProLiant servers, including the DL server line

For more information, please contact your Hewlett Packard Enterprise sales representative.

Support

Support for Perforce TotalView for HPC is direct-to-vendor. Information about support processes will be provided at the point of providing a license key to Perforce.