Data Transform Accelerator Metadata Pipeline Configuration
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Data transform operations in decode direction often suffer from high latency and low throughput due to the lack of efficient metadata utilization across pipelines in data transform accelerators.
Innovation Solution
The method involves obtaining and generating metadata during the encode process, which is then used to configure a second pipeline for efficient decoding, reducing latency and increasing throughput by leveraging metadata generated in the encode direction to facilitate decode direction operations.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of time
If data transform operations are performed in decode direction without utilizing metadata from encode process, then the system can operate independently in each direction, but latency increases and throughput decreases
Solution Approach 1:
The patent applies preliminary action by generating and storing metadata during the encode process before decode operations are needed. This metadata is prepared in advance and stored in memory, so when decode operations occur, the pipeline can be quickly configured using this pre-existing metadata without performing complex computations during decoding, thereby reducing latency.
Solution Approach 2:
The patent uses metadata as an intermediary between the encode and decode processes. This metadata acts as a bridge that carries configuration information from the encode direction to the decode direction, allowing the decode pipeline to be efficiently configured without directly reprocessing the original input data, thus reducing computational overhead and latency.
2Productivity
If data transform operations are performed in decode direction without efficient metadata utilization, then the pipeline configuration can be simple, but throughput decreases
Solution Approach 1:
The patent applies preliminary action by pre-generating metadata during the encode process that contains all necessary configuration information for the decode pipeline. This metadata is stored and reused during decode operations, enabling the pipeline to be quickly configured and executed without repeated complex computations, thereby increasing throughput.
Solution Approach 2:
The patent uses copying by creating a metadata representation of the encode pipeline configuration and using this copied information to configure the decode pipeline. Instead of reprocessing the original data or recreating the transformation logic, the system copies the essential configuration parameters into metadata and reuses them, significantly improving throughput while reducing computational complexity.
Data Source
AI summary
A method includes obtaining, by a data transform accelerator, first metadata. The method also includes configuring a first pipeline in the data transform accelerator using the first metadata. The method further includes obtaining input data to be transformed by the data transform accelerator. The method also includes generating encoded data and second metadata in the first pipeline using the input data and the first metadata. The encoded data and the second metadata may be stored together for later operations, such as decode operations. The method further includes configuring a second pipeline in the data transform accelerator using the second metadata.