18089780. SCALABLE ACCELERATION OF REENTRANT COMPUTE OPERATIONS simplified abstract (XILINX, INC.)
Contents
SCALABLE ACCELERATION OF REENTRANT COMPUTE OPERATIONS
Organization Name
Inventor(s)
Rajeev Patwari of San Jose CA (US)
Elliott Delaye of San Jose CA (US)
Xiao Teng of Cupertino CA (US)
Ephrem Wu of San Mateo CA (US)
SCALABLE ACCELERATION OF REENTRANT COMPUTE OPERATIONS - A simplified explanation of the abstract
This abstract first appeared for US patent application 18089780 titled 'SCALABLE ACCELERATION OF REENTRANT COMPUTE OPERATIONS
The abstract describes techniques for parallel processing using multiple processing elements and a controller for data with dependencies. The processing elements are assigned different chunks of data for processing and use tokens to inform the controller when they are done. The controller gathers partial results, determines an intermediate value, and distributes it back to the processing elements for final results.
- Parallel processing using multiple processing elements and a controller
- Processing elements assigned different data chunks for processing
- Tokens used to inform the controller when processing is complete
- Controller gathers partial results, determines intermediate value, and distributes it back to processing elements
- Processing elements re-process data chunks with intermediate value for final results
Potential Applications: - Data analysis - Image processing - Machine learning algorithms
Problems Solved: - Efficient processing of large datasets - Handling data dependencies - Improving processing speed
Benefits: - Faster data processing - Improved efficiency - Scalability for large datasets
Commercial Applications: - Big data analytics - Artificial intelligence systems - Scientific research
Questions about Parallel Processing: 1. How does parallel processing improve data analysis efficiency? Parallel processing allows multiple processing elements to work simultaneously on different data chunks, speeding up the overall processing time.
2. What are the advantages of using a controller in parallel processing? A controller helps manage data dependencies, gather partial results, and distribute intermediate values, improving the overall efficiency of the processing system.
Original Abstract Submitted
Examples herein describe techniques for performing parallel processing using a plurality of processing elements (PEs) and a controller for data that has data dependencies. For example, a calculation may require an entire row or column to be summed, or to determine its mean. The PEs can be assigned different chunks of a data set (e.g., a tensor set, a column, or a row) for processing. The PEs can use one or more tokens to inform the controller when they are done with partial processing of their data chunks. The controller can then gather the partial results and determine an intermediate value for the data set. The controller can then distribute this intermediate value to the PEs which then re-process their respective data chunks using the intermediate value to generate final results.
