GPU Surface State Access via 64-bit Virtual Addresses
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current graphics processing units (GPUs) face performance penalties due to the need to flush the pipeline when changing heap-based state settings outside of command list boundaries, particularly in multi-threaded applications, leading to inefficiencies in accessing surface and sampler states.
Innovation Solution
Implementing a memory access technique using 64-bit virtual addresses to access surface and sampler states, eliminating the reliance on heap-based state programming and thereby avoiding pipeline flushes.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If heap-based state programming is used for surface and sampler state, then state changes can be made, but pipeline flush is required which causes significant performance penalty
Solution Approach 1:
The patent introduces a virtual address space as an intermediary layer between the CPU and GPU heap-based state programming. This virtual address space acts as a mediator that translates CPU memory accesses into GPU-friendly state changes without requiring pipeline flushes, thereby resolving the contradiction between state change capability and pipeline throughput.
Solution Approach 2:
The patent replaces the mechanical pipeline flushing mechanism with a virtual memory addressing system. Instead of relying on the GPU pipeline to flush and resync after heap-based state changes, the system uses virtual addresses to directly access surface and sampler states, eliminating the need for pipeline flush operations and maintaining continuous throughput.
2Ease of operation
If explicit state parameters are programmed into hard coded data structures, then state settings can be configured, but changes require pipeline drain which introduces performance penalty
Solution Approach 1:
The patent transforms the static hard-coded data structure approach into a dynamic virtual address space system. Instead of configuring states through rigid, fixed-location data structures that require pipeline draining for changes, the system uses dynamic virtual addresses that can be directly modified without time-consuming pipeline drain operations, thereby improving ease of operation while reducing time loss.
3Extent of automation
If multi-threaded applications access surface and sampler states, then parallel processing is enabled, but pipeline flush is required for state changes causing inefficiency
Solution Approach 1:
The patent creates a universal virtual address space that serves multiple threads simultaneously. This unified addressing mechanism allows different threads to access and modify surface and sampler states independently through virtual addresses, enabling multi-threaded processing without requiring pipeline flushes for each thread's state changes, thereby maintaining high productivity across all threads.
Data Source
AI summary
Embodiments described herein provide a technique to enable access to entries in a surface state or sampler state using 64-bit virtual addresses. One embodiment provides a graphics core that includes memory access circuitry configured to facilitate access to the memory by functional units of the graphics core. The memory access circuitry is configured to receive a message to access an entry in a surface state or a sampler state associated with a parallel processing operation. The message specifies a base address for a surface state entry or sampler state entry. The circuitry can add the base address and the offset to determine a 64-bit virtual address for the entry in the surface state entry or the sampler state and submit a memory access request to the memory to access the entry of the surface state or sampler state.