Customized LibSci (including BLAS, LAPACK, and ScaLAPACK), our iterative refinement toolkit, and LibSci_ACC (accelerated BLAS, and LAPACK) are designed to take full advantage of the underlying hardware, optimizing for both intra-node and inter-node performance on all HPE Cray Supercomputing systems. Using auto-tuning and adaptation, the libraries choose the known best algorithms at runtime. They also feature simplified interfaces into complex software (no source code changes required to access optimized algorithms) and integrate with the compiling environment for better productivity.

NetCDF and HDF5 I/O libraries are built with the supported compiling environments and included in the suite for convenience.

Debugging tools

The HPE Cray Supercomputing Programming Environment Software offers traditional debuggers combined with new innovative techniques. Together, these technologies allow users to address debugging problems at a broader range and scale than conventional techniques. This means that programmers can spend less time debugging and more time creating Included are:

• – Comparative debugger: This market-unique tool helps programmers uncover issue by running two applications side by side. If the values of the selected data structures diverge, the user is notified that an error may exist. This capability is useful for locating errors that are introduced when applications are modified through code, compiler, or library changes, and for application porting between architectures (for example between CPUs and GPUs) or programming models.

• – GDB for HPC is based on the popular GDB command-line debugger used to debug applications compiled with Fortran, C, and C++ compilers with enhancements to provide a GDB debugging experience for applications that run at scale across many nodes. The tool enables users to run a traditional scalable debugging session—either by launching an application or by attaching it to an already-running application. A GDB for HPC debug session retrieves debug information from thousands of processes and presents merged backtraces and data, removing vast amounts of duplicate information.

• – Valgrind for HPC: Parallel memory analysis tool based on Valgrind debugger used for applications compiled with Fortran, C, and C++ compilers—it aggregates common errors into a single output record for easier analysis of potential memory problems within applications that run at scale.

• – Sanitizers for HPC: Sanitizers for HPC enables the users to easily detect memory and thread errors for easier analysis and debugging. It collects and analyses LLVM sanitizer output at scale, manages job launch through system’s workload manager and processes Sanitizers tool error reports and aggregates them for easier analysis. Sanitizers for HPC supports Clang AddressSanitizer, LeakSanitizer, ThreadSanitizer and Sanitizer libraries included with both the Cray CCE and the GNU GCC compilers. It also supports AMD’s GPU Sanitizerlibrary and NVIDIA’s CUDA-Memcheck tool.

• – Stack Trace Analysis Tool (STAT): Helps developers identify if an application is hung or still making progress when running. Generates a merged backtrace for applications so users can get a better insight into application behavior at a function level.

• – Tool for abnormal termination processing (ATP): When an application crashes, the tool detects a signal and generates a merged backtrace resulting in a minimal core file set so that programmers do not have to plough through many core files when debugging the application.