Dynamic Register Reconfiguration via Fix-up Micro-ops

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

The narrow-to-wide dependency in processor register files leads to increased power consumption and performance bottlenecks due to fractured architectural registers being split across multiple physical registers, complicating instruction processing.

Innovation Solution

A method and apparatus for reconfiguring a register file using a rename table to record fracture information about source registers, generating fix-up micro-ops to merge smaller size registers into larger ones, and stalling micro-ops until the fracture is fixed, thereby addressing the fracture and reducing power consumption and performance bottlenecks.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Adaptability or versatility

If the architectural register is fractured and split across multiple physical registers to handle narrow-to-wide dependency, then the register mapping flexibility is improved, but the power consumption and processing complexity increase

Engineering Contradiction:
Improveregister mapping flexibilityVSAvoidpower consumption
Core Design Contradiction:
Adaptability or versatilityVSUse of energy by moving object

Solution Approach 1:

The system performs preliminary actions by detecting register fractures during the rename table lookup process and proactively generating fix-up micro-ops before the actual instruction execution. This allows the fracture detection and repair mechanisms to be prepared in advance, reducing the overhead during critical execution phases and thereby lowering overall power consumption while maintaining register mapping flexibility.

Inventive Principle:
Principle #10Preliminary action

Solution Approach 2:

The patent introduces fix-up micro-ops as intermediary operations that mediate between the fractured physical registers and the architectural register abstraction. These intermediary micro-ops consolidate data from multiple physical registers into a single architectural register view, reducing the number of active read ports and decreasing power consumption while preserving the adaptability of the register mapping system.

Inventive Principle:
Principle #24Intermediary (Mediator)

2Adaptability or versatility

If multiple physical registers are used to represent a single architectural register, then the register renaming capability is improved, but the number of read ports and processing complexity increase

Engineering Contradiction:
Improveregister renaming capabilityVSAvoidnumber of read ports
Core Design Contradiction:
Adaptability or versatilityVSDevice complexity

Solution Approach 1:

The patent merges multiple physical register reads into a single architectural register operation by generating fix-up micro-ops that consolidate data from multiple sources. This merging approach reduces the number of simultaneous read ports required, as the fix-up operations can be executed sequentially or in batches, thereby reducing device complexity while maintaining register renaming capability.

Inventive Principle:
Principle #5Merging (Combining)

Solution Approach 2:

The system performs preliminary consolidation of register data during the rename phase by detecting fractures and pre-generating fix-up micro-ops. This preliminary action allows the actual instruction execution to proceed with fewer active read ports, as the fracture repair work is already prepared or can be executed in advance, thereby reducing processing complexity.

Inventive Principle:
Principle #10Preliminary action

3Ease of operation

If register fracture detection and fix-up micro-ops are implemented, then the narrow-to-wide dependency handling is improved, but the instruction processing time increases

Engineering Contradiction:
Improvenarrow-to-wide dependency handlingVSAvoidinstruction processing time
Core Design Contradiction:
Ease of operationVSLoss of time

Solution Approach 1:

The system performs fracture detection and fix-up micro-op generation during the rename table lookup phase, which occurs before instruction execution. This preliminary action allows the actual instruction to execute more quickly, as the fracture repair work is already prepared or can be executed in advance, thereby reducing overall instruction processing time while improving dependency handling.

Inventive Principle:
Principle #10Preliminary action

Solution Approach 2:

The patent implements skip logic where, if no fracture is detected during rename table lookup, the system skips the entire fix-up micro-op generation process and proceeds directly to instruction execution. This skipping mechanism minimizes processing time for non-fractured registers while maintaining the ability to handle fractured registers when necessary, thereby improving overall instruction processing efficiency.

Inventive Principle:
Principle #21Skipping (Rushing through)

Data Source

PatentUS9588770B2Dynamic rename based register reconfiguration of a vector register file
Publication Date: 2017.03.07 SAMSUNG ELECTRONICS CO LTD
  • US9588770B2 patent drawing
  • US9588770B2 patent drawing
  • US9588770B2 patent drawing

AI summary

Reconfiguring a register file using a rename table having a plurality of fields that indicate fracture information about a source register of an instruction for instructions which have narrow to wide dependencies.