Monthly R&D Status Report Date: February 15, 1997 Title: "SYSTEM-LEVEL DESIGN METHODOLOGY FOR EMBEDDED SIGNAL PROCESSORS" Contract Number: F33615-93-C-1317 Principal Investigator: Edward A. Lee Organization: University of California at Berkeley 1. Tasks Performed We are on schedule for the 0.7 release of Ptolemy, and have focused the remaining effort in this project on maximizing the quality of this release. 2. Significant Accomplishments 2.1 Ptolemy Miniconference At least 35 people have signed up to attend the Ptolemy miniconference on March 10. See http://ptolemy.eecs.berkeley.edu/ for details. 2.2 Architecture Trade Tool from Sanders Released The Lockheed/Sanders Architecture Trade Tool was released on January 29, 1997. Lockheed/Sanders has developed a proof of concept architectural trade capability using Ptolemy's Discrete Event domain. It is meant as a first cut at a usable capability, and is provided for demonstration purposes only. There are many simplifications made in order to get the tool working for a few example cases. It being provided to members of the Ptolemy community interested in this type of architectural trade capability work. A custom graphical front-end to Ptolemy has been developed that allows a user to sketch a target architecture in one window and quickly map the stars in a SDF graph in another window to the processors in the architecture. Extensions to the DE domain have been implemented to allow a performance-level model of the architecture to be simulated. These extensions create a DE domain model representing the mapping of the algorithm to the architecture and use the Ptolemy kernel to simulate the performance. The product of the simulation is a Gantt chart showing the execution of stars over time as well as a thermometer display of estimates of certain other system level metrics (weight, size, power, reliability, etc.). This capability has been developed as a front-end architectural trade tool for the Sanders RASSP Program, under contract to the Naval Research Laboratory. The capability is available in the form of tar file, and requires Ptolemy 0.6. For further information or to download the tool, look under the "News in the Ptolemy Project" section of the Ptolemy home page at: http://ptolemy.eecs.berkeley.edu 2.3 Synchronous/Reactive Modeling After much internal discussion about demonstration applications for the synchronous/reactive (SR) domain, we settled on computer music as providing a suitable representative of systems that combine intensive signal processing with intensive control. In view of this, Stephen Edwards and Mary Stewart purchased a Midi keyboard to interface to the serial port of an Ultrasparc. The Midi keyboard is capable of providing highly complex and time-sensitive control signals, thus representing a demanding system environment. Stephen is using the Synchronous Reactive domain and SDF (synchronous dataflow) to create a synthesizer using the Ultrasparc to generate sound, and the keyboard to trigger events. The synthesizer itself will consist of a few modules decoding the Midi stream and distributing them to a few tone generator modules, described using SDF subsystems. The goal is not the best sound synthesis, but to demonstrate the utility of the SR domain in controlling such a real-time system, where both the control and signal processing are reasonably complex. 2.4 Infrastructure Christopher Hylands and Mary Stewart have installed three new ultrasparc2s (200Mhz Ultrasparc, 128mb memory, 24 bit color, fast disk). This should improve our build times considerably and speed up the 0.7 release. Mary Stewart upgraded our HP machine to HPUX10.20 in preparation for the Ptolemy0.7 port. Our hope is that some of the DCE threading bugs that caused problems with Tom Park's Process Network domain have been fixed in the new release. 2.5 Improvements to Ptolemy Tom Lane, from Structured Software Systems, has provided a revised type resolution system for Ptolemy. This is the first major change in the Ptolemy kernel in quite a while, and represents a big improvement. We have worked with Tom to fully test the system and correct problems it created in other parts of the system. These changes will be part of Ptolemy 0.7. Tom Lane also provided a fix to the management of scripted runs so that a scripted run can be interrupted. Mike Williamson tuned some key kernel classes in Ptolemy for improved speed. Uwe Trautwein of TU-Ilmenau in Germany has contributed a suite of Ptolemy blocks and demos for adaptive beamforming. Christopher Hylands, Praveen Murthy, and Mike Williamson have all tracked and repaired memory access errors in various parts of Ptolemy. Alberto Ferrari has developed an interface between Matrixx and Ptolemy. It is unclear whether this interface will be included in the 0.7 release. Luis Gutierrez, Mike Williamson and Christopher Hylands have updated the CG56 domain to use the latest version of 56x software tools from Motorola. 2.6 Improvements to Tycho Christopher Hylands added a capability to the Tcl and Itcl editors to evaluate the method body containing the cursor. He also modified the HTML viewer to handle .gif, .ppm and .xbm files and directory listings. A number of other improvements have been made to Tycho as well, but few of them bear directly on the 0.7 release of Ptolemy. 3. Publications [1] S. Bhattacharyya, S. Sriram & E. A. Lee, "Optimizing Synchronization in Multiprocessor DSP Systems," , accepted as a regular paper in Transactions of Signal Processing. [2] S. S. Bhattacharyya, P. K. Murthy, and E. A. Lee, "Optimized Software Synthesis for Synchronous Dataflow," to be presented at the International Conference on Application Specific Systems, Architectures, and Processors, July, 1997, invited paper. [3] S. S. Bhattacharyya, P. K. Murthy, and E. A. Lee, "APGAN and RPMC: Complimentary Heuristics for Translating DSP Block Diagrams into Efficient Software Implementations," Journal of Design Automation for Embedded Systems, Vol. 2, No. 1, pp. 33-60, January, 1997.