CROP Atomic Transaction Processing via Blend Unit Datapath

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Existing graphics processing units (GPUs) require significant area on the chip for dedicated hardware units to perform atomic transactions, limiting throughput.

Innovation Solution

Configuring the color raster operations unit (CROP) to process both color blending operations and atomic transactions, eliminating the need for dedicated hardware units for atomic transactions by using existing hardware units within the CROP.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Productivity

If dedicated hardware units are included to execute atomic transactions, then the throughput of atomic transactions is improved, but the area consumed on the GPU chip increases

Engineering Contradiction:

Improvethroughput of atomic transactionsVSAvoidarea on GPU chip

Core Design Contradiction:

ProductivityVSArea of stationary object

Solution Approach 1:

The patent applies multi-functionality by configuring the color raster operations unit (CROP) to perform both its traditional color blending operations and atomic transactions. The CROP is enhanced with an atomic transaction engine that allows it to execute atomic operations on memory locations while maintaining its primary function. This eliminates the need for separate dedicated hardware units for atomic transactions, thereby improving throughput without proportionally increasing chip area.

Inventive Principle:

Principle #6Universality (Multi-functionality)

2Productivity

If multiple dedicated hardware units are included for atomic transactions, then the throughput of atomic transactions is improved, but the device complexity increases

Engineering Contradiction:

Improvethroughput of atomic transactionsVSAvoidnumber of dedicated hardware units

Core Design Contradiction:

ProductivityVSDevice complexity

Solution Approach 1:

The patent merges the atomic transaction processing capability into the existing CROP structure. Instead of adding multiple independent dedicated hardware units, the atomic transaction engine is integrated within the CROP, sharing resources such as the blending logic, memory interfaces, and control units. This consolidation improves atomic transaction throughput while avoiding the complexity increase that would result from multiple separate hardware units.

Inventive Principle:

Principle #5Merging (Combining)

Data Source

PatentUS8624916B2Processing global atomic operations using the bending unit datapath

Publication Date: 2014.01.07 NVIDIA CORP

AI summary

One embodiment of the invention sets forth a CROP configured to perform both color raster operations and atomic transactions. Upon receiving an atomic transaction, the distribution unit within the CROP transmits a read request to the L2 cache for retrieving the destination operand. The distribution unit also transmits the source operands and the operation code to the latency buffer for storage until the destination operand is retrieved from the L2 cache. The processing pipeline transmits the operation code, the source and destination operands and an atomic flag to the blend unit for processing. The blend unit performs the atomic transaction on the source and destination operands based on the operation code and returns the result of the atomic transaction to the processing pipeline for storage in the internal cache. The processing pipeline writes the result of the atomic transaction to the L2 cache for storage at the memory location associated with the atomic transaction.