Programmability Framework

Challenges in Current Multiprocessor Programming

Current programming approaches for CMPs use fixed processor allocation schemes. Synchronization is ad-hoc – varying from application to application. This makes the application programming task highly complex. Current approaches have the following disadvantages:

• Lack of scalability – static allocation does not allow integration of components or reallocation of resources

• Lack of portability – the current methodologies tie the application down to specific hardware configurations and inflexible application designs

• Heavy overheads – high thread switching costs make it impossible to write parallel applications with optimal utilization and performance

• Inefficient resource utilisation – static data and functional pipelines lead to sub-optimal usage of the parallel resources available

Codito Technologies has developed patent pending technology for developing complex parallel applications on multi-core chips. We work with leading Chip Manufacturers to deploy this solution across their developer and customer networks.

Key Features of Technology

• Parallel programs can be developed visually

• The technology solves synchronization issues related to multiple processors - a major problem area in parallel programming

• Heterogeneous stream applications can be integrated due to dynamic allocation, with run-time guarantees on performance

• Data flow based programming ensures re-use of application components

• Solution significantly simplifies design, development, testing and debugging cycles for CMPs

• Technology enables meeting the most demanding time to market requirements for chip manufacturers, OEMs and ODMs

Codito's technology offering for CMPs include:

• An engineering methodology for designing and developing highly parallel data flow based applications

• Mathematically proven techniques for deadlock free synchronization

• Optimized runtime executive and resource allocator embedded with each multiprocessor

• A chip-specific custom API and libraries for parallel programs

• Custom visual workbench integrated with existing application development Toolchain for the CMP and Codito's runtime

• Tools and techniques to aid application development and integration

• Optimised pre-fabricated application components integrated with the methodology and tools