Tensor Dimension API for Fused GPU Image Processing Kernels
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Performing image processing operations on graphics processing units (GPUs) requires significant human expertise and inefficient use of resources, leading to suboptimal memory and computing resource utilization.
Innovation Solution
An application programming interface (API) is used to generate software kernels that perform image processing operations, including convolution and tensor operations, by converting tensors into matrices for efficient matrix multiplication without storing operands, and modifying tensor dimensions to combine convolution and reduction operations.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If custom graphics pipelines are built manually for image processing operations, then processing capability is achieved, but human expertise requirement increases and resource efficiency decreases
Solution Approach 1:
The system enables self-service by allowing the GPU to automatically generate and execute software kernels for image processing operations without requiring manual pipeline configuration. The CUDA program executes autonomously, converting tensors to matrices and performing convolutions through automated kernel generation, eliminating the need for human expertise in pipeline construction while maintaining high processing efficiency.
2Speed
If tensors are converted to matrices for matrix multiplication, then computing efficiency improves, but memory usage increases due to operand storage requirements
Solution Approach 1:
The invention extracts and eliminates the unnecessary operand storage step from the traditional matrix multiplication process. Instead of storing complete matrix operands in memory before computation, the system directly processes tensor data through automated kernel generation that performs conversions and multiplications in-place, removing the memory burden of storing intermediate matrix representations while maintaining computing speed.
Solution Approach 2:
The system changes the dimensional representation approach by working directly with tensors in their native multi-dimensional form rather than converting to traditional two-dimensional matrices for storage. This dimensionality approach allows the GPU to perform efficient computations on tensor data structures without requiring additional memory allocation for matrix conversion, achieving both speed and memory efficiency.
3Productivity
If convolution and reduction operations are performed separately, then operational clarity is maintained, but processing time increases
Solution Approach 1:
The invention merges convolution and reduction operations into a single integrated software kernel that executes atomically on the GPU. The unified kernel performs both tensor convolution and dimension reduction simultaneously without intermediate storage or separate execution steps, eliminating the time loss associated with sequential operation execution while maintaining the logical clarity of distinct computational stages through structured code organization.
Data Source
AI summary
Apparatuses, systems, and techniques to perform image processing. In at least one embodiment, an application programming interface is performed to cause dimensions of a tensor to be modified.