Development Tools ::.
OpenUH
OpenUH is a compiler based upon the open source Open64 compiler suite for C, C++, and Fortran 90 and the IA-64 Linux ABI and API standards. This infrastructure provides state-of-the-art analyses and transformations along with a high level of robustness. It also includes a feedback mechanism for improving compiler optimizations. The OpenUH feedback mechanism was extended to display information associated with one or more program runs, including the call paths taken at run time. The Dragon tool extends OpenUH by displaying high level program data structures in graphical and text format in response to user requests; it relates the information to the original source code lines visually. OpenUH is developed and maintained by the High Performance Computing and Tools Group from the University of Houston.KOJAK
KOJAK is an automatic performance evaluation system for MPI, OpenMP, and hybrid applications written in C/C++ or Fortran. It helps application developers detect sources of inefficient program behavior. KOJAK is able to identify the reasons for low performance on a very high abstraction level (e.g. "process was waiting for a message that was sent too late"). It generates event traces from running applications and automatically searches them offline for execution patterns indicating inefficient performance behavior. KOJAK is a collaborative effort between the Central Institute for Applied Mathematics at Forschungszentrum Julich and the Innovative Computing Lab at the University of Tennessee.PerfSuite
PerfSuite is a performance analysis infrastructure developed and deployed at NCSA since 2002, where it is the foundation for automatic performance data collection from all MPI-based user jobs that execute on NCSA Linux clusters. It is available for download in source form from NCSA and SourceForge. PerfSuite provides low overhead, low impact assessments of the effectiveness of application transformations. It eliminates most of the overhead imposed by performance monitoring software and enables "before" and "after" comparisons to be performed with minimal effort. PerfSuite is thread-aware, allowing its use with unmodified OpenMP applications executed via POSIX threads.High-End Applications::.
The following applications, developed by the HPCFD group from Virginia Tech, motivate, evaluate and demonstrate our efforts:
- Additive Schwarz Preconditioned Conjugate Gradient (ASPCG) solves a linear system of equations generated by a Laplace equation in Cartesian coordinates. It was developed to evaluate cache performance and memory bandwidth of different architectures. The code is available in a serial form, MPI, OpenMP and MPI-OpenMP.
