Memory Operation Grouping for Graphics Processing Consistency

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

In graphics processing architectures, memory fences are used to maintain sequential consistency between work items and work groups, but they cause threads to stall, leading to processing 'bubbles' that negatively impact performance.

Innovation Solution

Implementing a group identifier to track the completion of memory operations, allowing threads to bypass stalls and reduce processing bubbles by using a completion interface that monitors hardware counters to track memory operation completions and issue fence completion events.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If memory fences are used to maintain sequential consistency between work items and work groups, then reliability is improved, but productivity deteriorates due to thread stalls and processing bubbles

Engineering Contradiction:

Improvesequential consistencyVSAvoidprocessing throughput

Core Design Contradiction:

ReliabilityVSProductivity

Solution Approach 1:

The patent segments the memory coherence tracking by introducing group identifiers (color values) that associate memory operations with specific work groups. This allows the system to track completion status separately for different work groups, enabling fine-grained synchronization that avoids global stalls while maintaining sequential consistency within each group.

Inventive Principle:

Principle #1Segmentation

Solution Approach 2:

The patent implements a feedback mechanism where completion status of memory operations is tracked and monitored. When all memory operations for a work group are complete, the system generates a completion event that signals the work group can proceed. This feedback-driven approach eliminates unnecessary stalls by only waiting when actually needed, thereby maintaining reliability while improving productivity.

Inventive Principle:

Principle #23Feedback

2Measurement precision

If memory fences are used to ensure sequential consistency, then measurement precision of memory operation ordering is improved, but loss of time increases due to thread stalling

Engineering Contradiction:

Improvememory operation orderingVSAvoidthread stall time

Core Design Contradiction:

Measurement precisionVSLoss of time

Solution Approach 1:

The patent performs preliminary actions by attaching group identifiers to memory operations before they are executed. This pre-tagging allows the completion tracking system to identify and monitor specific work group operations without requiring threads to stall and check status repeatedly. The ordering precision is maintained through this preliminary identification, while time is saved by eliminating stall-based synchronization.

Inventive Principle:

Principle #10Preliminary action

Solution Approach 2:

The patent replaces the mechanical stalling mechanism with an event-driven completion notification system. Instead of threads physically stalling and waiting at fence instructions, the system uses software-based completion events triggered when memory operations finish. This substitution maintains precise ordering tracking while eliminating the time loss associated with mechanical thread stalls.

Inventive Principle:

Principle #28Mechanics substitution (Replace mechanical system)

Data Source

PatentUS11195497B2Handling surface level coherency without reliance on fencing

Publication Date: 2021.12.07 INTEL CORP

AI summary

Systems, apparatuses and methods may provide for technology that detects a memory fence in a thread, adds a group identifier to one or more memory operations in the thread that follow the memory fence, and sends the one or more memory operations and the group identifier to a memory structure. In one example, the group identifier is used to track completion of the one or more memory operations.