Compiler Register Allocation for Multi-Level Clustered Files

Solution Overview

Problem

Current techniques for register allocation in parallel computing systems with multiple-level register files often result in higher access energy and time due to storing values primarily used by one cluster of functional units in high-level register files, which can miss opportunities for optimization in terms of reducing access time and energy consumption.

Innovation Solution

A method is introduced where a compiler assigns instructions to processing clusters, determines the owner cluster for virtual registers, and configures instructions to access global virtual registers, allowing frequently accessed registers to be allocated to local register files, thereby reducing access time and energy.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Adaptability or versatility

If all values utilized by multiple clusters of functional units are stored in high-level register files, then values needed by multiple clusters are accessible to those clusters, but access time and energy consumption increase

Engineering Contradiction:

Improveaccessibility to multiple clustersVSAvoidaccess time

Core Design Contradiction:

Adaptability or versatilityVSLoss of time

Solution Approach 1:

The patent applies local quality by differentiating register allocation based on access patterns. Frequently accessed values are allocated to local register files for fast access, while less frequently accessed values are placed in high-level register files. This creates non-uniform quality across different register locations, optimizing for both speed and accessibility.

Inventive Principle:

Principle #3Local quality

Solution Approach 2:

The patent implements dynamic register allocation where the compiler analyzes access patterns and dynamically determines optimal register placement. The system adapts to different access scenarios by adjusting which values reside in local versus high-level register files, rather than using a static allocation strategy.

Inventive Principle:

Principle #15Dynamics

2Adaptability or versatility

If all values utilized by multiple clusters of functional units are stored in high-level register files, then values needed by multiple clusters are accessible to those clusters, but energy consumption increases

Engineering Contradiction:

Improveaccessibility to multiple clustersVSAvoidaccess energy

Core Design Contradiction:

Adaptability or versatilityVSUse of energy by stationary object

Solution Approach 1:

The patent applies local quality by differentiating register allocation based on access patterns. Frequently accessed values are allocated to local register files for fast access, while less frequently accessed values are placed in high-level register files. This creates non-uniform quality across different register locations, optimizing for both speed and accessibility.

Inventive Principle:

Principle #3Local quality

Solution Approach 2:

The patent implements dynamic register allocation where the compiler analyzes access patterns and dynamically determines optimal register placement. The system adapts to different access scenarios by adjusting which values reside in local versus high-level register files, rather than using a static allocation strategy.

Inventive Principle:

Principle #15Dynamics

3Adaptability or versatility

If values used primarily by a single cluster are stored in high-level register files, then multiple clusters can access these values, but opportunities for optimization in access time and energy are missed

Engineering Contradiction:

Improveaccessibility to multiple clustersVSAvoidaccess efficiency

Core Design Contradiction:

Adaptability or versatilityVSProductivity

Solution Approach 1:

The patent applies local quality by differentiating register allocation based on access patterns. Frequently accessed values are allocated to local register files for fast access, while less frequently accessed values are placed in high-level register files. This creates non-uniform quality across different register locations, optimizing for both speed and accessibility.

Inventive Principle:

Principle #3Local quality

Solution Approach 2:

The patent changes the parameter of register allocation from a uniform strategy to a selective strategy based on access frequency. The compiler modifies allocation parameters dynamically, placing values in different register levels according to their access patterns, thereby optimizing overall system efficiency.

Inventive Principle:

Principle #35Parameter changes

Data Source

PatentUS9229717B2Register allocation for clustered multi-level register files

Publication Date: 2016.01.05 NVIDIA CORP

AI summary

A method for allocating registers within a processing unit. A compiler assigns a plurality of instructions to a plurality of processing clusters. Each instruction is configured to access a first virtual register within a live range. The compiler determines which processing cluster in the plurality of processing clusters is an owner cluster for the first virtual register within the live range. The compiler configures a first instruction included in the plurality of instructions to access a first global virtual register.