GPU Texture Wave Allocation Under GPR and Wave Slot Limits

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

The constraint of limited memory, registers, and wave slots in graphics processing units (GPUs) hinders efficient texture mapping and general-purpose computing, leading to increased latency and costs when additional memory is used to mitigate latency.

Innovation Solution

Efficient allocation of general purpose registers (GPR) and physical wave slots by deferring their use for texture operations, utilizing virtual wave slots and texture buffers to optimize resource utilization, allowing continuous processing without initial allocation, and converting to physical slots when available.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Loss of time

If additional memory is used to mitigate latency in texture mapping, then latency is reduced, but device cost increases

Engineering Contradiction:
ImprovelatencyVSAvoiddevice cost
Core Design Contradiction:
Loss of timeVSDevice complexity

Solution Approach 1:

The system performs preliminary actions by pre-allocating virtual wave slots and deferring physical wave slot allocation until needed. This allows the GPU to prepare for future texture mapping operations without immediately consuming additional physical memory resources, thereby reducing latency through better resource management rather than simply adding more memory

Inventive Principle:
Principle #10Preliminary action

Solution Approach 2:

Virtual wave slots act as an intermediary between the need for wave slot allocation and the physical wave slots available in hardware. This intermediary layer allows the system to manage wave slots more efficiently, deferring physical allocation until necessary and reducing the need for expensive additional memory to handle latency issues

Inventive Principle:
Principle #24Intermediary (Mediator)

2Productivity

If more wave slots are allocated for texture operations, then throughput is improved, but device complexity increases

Engineering Contradiction:
ImprovethroughputVSAvoiddevice complexity
Core Design Contradiction:
ProductivityVSDevice complexity

Solution Approach 1:

The system segments wave slot allocation into virtual and physical layers. Virtual wave slots can be allocated freely to represent different texture mapping operations, while physical wave slots are allocated only when needed based on actual hardware availability. This segmentation allows the system to improve throughput by managing more wave slots virtually while keeping physical device complexity manageable

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The wave slot allocation system is made dynamic through deferred allocation and conversion mechanisms. The system can adaptively convert virtual wave slots to physical wave slots based on real-time availability and needs, allowing throughput to be optimized without fixed physical allocation constraints that would increase device complexity

Inventive Principle:
Principle #15Dynamics

3Ease of operation

If general purpose registers are allocated upfront, then resource availability is improved, but resource utilization efficiency decreases

Engineering Contradiction:
Improveresource availabilityVSAvoidresource utilization efficiency
Core Design Contradiction:
Ease of operationVSProductivity

Solution Approach 1:

The system performs preliminary allocation of virtual wave slots and register resources, but defers the actual consumption of physical resources until needed. This allows the system to guarantee resource availability for upcoming operations while improving utilization efficiency by avoiding premature allocation of physical GPRs and wave slots that would reduce the pool of available resources for other operations

Inventive Principle:
Principle #10Preliminary action

Data Source

PatentEP3948791B1General purpose register and wave slot allocation in graphics processing
Publication Date: 2025.12.03 QUALCOMM INC
  • EP3948791B1 patent drawingFigure 1
  • EP3948791B1 patent drawingFigure 2
  • EP3948791B1 patent drawingFigure 3

AI summary

Example techniques are described for generating graphics content by obtaining texture operation instructions corresponding to a texture operation, in response to determining at least one of insufficient general purpose register space is available for the texture operation or insufficient wave slots are available for the texture operation, generating an indication that the texture operation corresponds to a deferred wave, executing the texture operation, sending, to a texture processor, initial texture sample instructions corresponding to the texture operation that was executed, and receiving texture mapped data corresponding to the initial texture sample instructions.