[Tau-announcements] New versions of TAU, PDT, and LiveDVD released

Sameer Shende sameer at cs.uoregon.edu
Sat Nov 10 18:15:05 PST 2012


       We are pleased to announce the release of TAU v2.22, PDT v3.18.1, and the HPC Linux LiveDVD:


The following new features have been added since TAU v2.21.4 released on September 18, 2012

1. Introducing TAUdb

TAU's next generation performance database technology, TAUdb, is now included in this release. It replaces the PerfDMF (Performance Data Management Framework) and maintains compatibility with the old databases in our tools ParaProf and PerfExplorer. The hierarchy for managing the data has changed to an arbitrary level depth hierarchy of views using the performance metadata. (Seehttp://www.cs.uoregon.edu/Research/tau/docs/newguide/bk02ch02.html)  The update includes performance improvements over the previous schema with respect to time and space. The associated analysis tools have been updated to the new schema.

New C API for TAUdb:
A new C programming interface was developed for TAUdb, with support for querying an existing database. The support for inserting, updating and deleting data through the C API is included and will be improved in the future releases.

2. New Platforms

a. IBM Power 7 Linux (DARPA HPCS PERCS)
TAU now supports the new IBM XL compiler suite (including the new IBM XLUPC compiler) on this platform. Support for compiler and source-based instrumentation, as well as linker level instrumentation is available.

b. Cray Cascades prototype
TAU now supports the Intel Sandybridge based Cray systems. TAU has been tested on the Cascades "emulator" (a proof-of-concept cluster system). Unlike the other XE6 systems, we do not need the -arch=craycnl on this system. To configure TAU on this system using PrgEnv-Intel and MPI, please use:
./configure -mpi -mpiinc=<dir> -mpilib=<dir> -pdt=<dir> -c++=mpicxx -cc=mpicc -f90=mpif90; make install
and use TAU's compiler scripts (tau_cxx.sh, tau_cc.sh, tau_f90.sh...) as usual.

c. Cray XK7 and CUDA support
TAU improves support for CUDA device to device memory copies and the Cray CCE OpenACC compiler in this release. This release also supports the PGI OpenACC compiler.

d. Intel(R) Xeon Phil(TM) co-processor systems (MIC).
This release adds support for both native and offloading modes of execution on the Intel Xeon Phi systems. In the offload mode, where part of the application executes on the host and part of it executes on the MIC co-processor, please configure TAU twice using -arch=x86_64 and -arch=mic_linux. Export the TAU_OPTIONS variable to include -optMICOffload and link the application using the TAU compiler wrapper scripts. While running the program, please set environment variables TAU_MIC_OFFLOAD=1 and MIC_PROXY_FS_ROOT=`pwd` to generate host profiles as profile.0.0.* and the co-processor profiles as profile.0.<pid>.*. For the native mode of execution, simply using the mic_linux architecture is sufficient to compile with the TAU wrapper scripts and execute the code.

e. Fujitsu FX10 (K Computer)
This release updates ParaProf's 3D topology display to reveal the internal torus structure of the hardware. This is similar to TAU's support for topology views with the IBM BlueGene/P, BlueGene/Q, and the Cray XE6/XK7 systems. TAU and PDT configuration on Fujitsu has been improved.

3. Support for GPI
TAU supports GPI, a PGAS-API from Fraunhofer ITWM (www.gpi-site.com). You may confiugure TAU with the -gpi=<dir> option and use the TAU scripts. It is important to use tau.conf in the executables current working directory to pass TAU runtime environment variables to the remote tasks. TAU provides a wrapper interposition library that tracks the one-sided communication. You may use TAU options such as TAU_COMM_MATRIX=1 or TAU_SAMPLING=1 to generate a communication matrix and to use event based sampling for fine-grained statement-level profiling, respectively.

4. Support for Binary Rewriting using PEBIL
This release of TAU and PDT 3.18.1 provides a new tool (tau_pebil_rewrite) to rewrite binaries with TAU instrumentation code. This is a new beta/experimental release. It supports outer-loop level instrumentation and selective instrumentation of routines based on exclude/include lists using TAU's format for selective instrumentation. It may be used as follows:
tau_pebil_rewrite -T <tags> -f select.tau a.out a.inst

% tau_pebil_rewrite -T icpc,mpi,pdt -f select.tau a.out a.inst
% cat select.tau
loops routine="#"

Only dynamic executables are supported at this time. Support for compilers and runtimes will improve over time. The TAU team would like to thank the PMaC labs at SDSC, UCSD for their support of the TAU project.

5. Improved Support for Binary Rewriting using MAQAO
PDT release 3.18.1 includes an updated MAQAO binary rewriting tool. TAU provides tau_rewrite, a tool that uses MAQAO internally.
% tau_rewrite -T <tags> a.out a.inst
We would like to thank the Intel Exascale Labs, UVSQ for their support of the TAU project.

6. Improved Support for UPC
TAU supports BUPC 2.16.0, IBM XLUPC, Cray CCE, GNU UPC, and has improvements for tracking the time spent in UPC runtime layer and the compiler scripts.

7. Reduced Measurement Overhead
TAU's runtime overhead has been reduced considerably if the user sets the TAU_LITE=1 runtime measurement option for simple flat profiling. Throttling is not available in this mode.

8. ParaProf and PerfExplorer Enhancements
TAU's configuration scripts now support the 3D windows better. Communication matrix display, 3D topology window are all improved. TAUdb support is now available in ParaProf and PerfExplorer.

9. Improved Support for OpenMP Instrumentation
This release of TAU features the latest Opari2 v1.0.6 for source-instrumentation. While this does not cover the full OpenMP 3.0 standard, we are making progress in OpenMP instrumentation.

TAU_SUMMARY=1 environment variable can generate profile output for node 0, node 1, and compute the max,mean,min,total,std. deviation across all nodes. This can reduce the profile output considerably and provide a glimpse of the performance data without generating data from each rank. This option now supports multi-threaded executions.

PDT 3.18.1 Release
As mentioned above, PDT provides an updated MAQAO, and PEBIL binary rewriters. It also has improvements in C++ headers, and configuration for Fujitsu FX10, Ubuntu Linux, IBM BlueGene/Q, and Mac OS X Mountain Lion.

HPC Linux
TAU, PDT, and VI-HPS (www.vi-hps.org) and ACTS (acts.nersc.gov) tools are featured on the HPC Linux LiveDVD and virtual images available for download from:


There is also a pre-release version ready for deploying cloud images and VMs for Amazon EC2, Windows Azure, KVM, VMWare, VirtualBox, and Xen.

Our team will be at SC'12. Please stop by the PGAS Booth (#2137) at SC to pick up the LiveDVDs or talk with one of our team members. Also, please join us at:

a. J. Vetter, A. Malony, P. Roth, K. Spafford, J. Meredith, "Scalable Heterogeneous Computing on GPU Clusters," Tutorial, Sunday, Nov 11, 2012, 8:30am - 5pm. Room 355-E.

b. B. Tibbits, G. Watson, J. Alameda, J. Overbey, "Developing and Tuning Parallel Scientific Applications in Eclipse," Tutorial, Monday, Nov 12, 2012, 8:30am - 5pm. Room 251-A.

c. Extreme Scale Performance Tools, Workshop, Friday, Nov. 16, 2012, 8:30am - 12:30pm. Room 155-F.


d. S. Shende, Talk, "Performance Evaluation using the TAU Performance System,"
Wednesday, Nov. 14, 3:15pm,
Oakleaf/Kashiwa Alliance Exhibitor Booth (#1943). U. Tokyo.

e. S. Shende, Talk, "TAU Performance System", Wednesday, Nov. 14, 4:10pm,
Intel Booth (#2601-2617),

  Please let us know if we may assist you with our tools in any way.
  - Sameer
 (for tau-team@ cs.uoregon.edu)

-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://nic.uoregon.edu/pipermail/tau-announcements/attachments/20121110/154c787f/attachment.html 

More information about the Tau-announcements mailing list