Prediction Circuitry for Fetch Bandwidth Bottleneck
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The performance of data processing apparatuses is limited by the bandwidth of fetch and decode circuitry, which restricts the number of operations that can be processed each cycle, and increasing this bandwidth comes at a higher area and power cost, making it unfeasible for low-power applications.
Innovation Solution
Implementing prediction circuitry that uses a lookup in a prediction table to identify instructions that can be executed without being fetched, allowing the fetch circuitry to skip these instructions and enabling the processing circuitry to perform the operations directly, thereby improving performance without increasing fetch bandwidth.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If the bandwidth of fetch and decode circuitry is increased to improve performance, then the number of operations that can be processed each cycle increases, but the area and power cost increases
Solution Approach 1:
The prediction circuitry performs lookups in the prediction table in advance to identify instructions that can be executed without fetching. This preliminary identification allows the system to skip fetching certain instructions, effectively increasing the number of operations processed per cycle without requiring increased fetch bandwidth, thus resolving the contradiction between productivity and power consumption
Solution Approach 2:
The invention extracts and identifies specific instructions (execute-without-fetch instructions) that can be executed without being fetched from the instruction stream. By separating these instructions from the normal fetch path, the system improves processing throughput without increasing the bandwidth requirements of the fetch circuitry, thereby improving productivity without increasing power cost
2Productivity
If the bandwidth of fetch and decode circuitry is increased to improve performance, then the number of operations that can be processed each cycle increases, but the area cost increases
Solution Approach 1:
The prediction circuitry performs preliminary lookups in the prediction table to identify instructions that can be executed without fetching. This advance identification enables the system to increase operational throughput without requiring additional fetch bandwidth or larger circuitry area, thus resolving the contradiction between productivity and area cost
Solution Approach 2:
The invention extracts specific instructions that can be executed without fetching, removing them from the normal fetch path. This extraction allows the system to process more operations per cycle without increasing the area of the fetch and decode circuitry, as these extracted instructions bypass the bandwidth-constrained fetch path
Data Source
AI summary
An apparatus comprises fetch circuitry configured to fetch instructions for processing by processing circuitry, and prediction circuitry configured to identify instructions to be fetched by the fetch circuitry. The prediction circuitry is configured to perform a lookup in a prediction table based on an address of a first instruction, and in response to the lookup identifying that the prediction table includes a skip-fetch-instruction entry corresponding to the address of the first instruction, the skip-fetch-instruction entry specifying information identifying an operation represented by at least one execute-without-fetch instruction subsequent in program order to the first instruction, the prediction circuitry is configured to control the fetch circuitry to skip fetching of the at least one execute-without-fetch instruction, and control the processing circuitry to perform the operation represented by the at least one execute-without-fetch instruction despite fetching of the at least one execute-without-fetch instruction being skipped.