Vector Register Complex Data Continuous Access
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing information processing systems face inefficiencies in performing complex vector operations due to the need for stride access when loading and storing real and imaginary parts of complex data, leading to slower processing speeds and the inability to execute division operations as vector operations.
Innovation Solution
The system generates instructions for continuous access and masked vector operations, allowing real and imaginary parts to be loaded and processed as elements of a vector register, enabling operations on different element numbers and accelerating memory access.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If real-part data and imaginary-part data are loaded into different vector registers using stride access, then complex number multiplication and division can be performed, but processing speed becomes slower
Solution Approach 1:
The patent merges real-part data and imaginary-part data into the same vector register by loading them contiguously from alternately arranged memory locations. This combining approach eliminates the need for separate vector registers, enabling continuous access to memory and significantly improving processing speed while maintaining the capability to perform complex number operations.
Solution Approach 2:
Instead of the conventional approach of separating real and imaginary parts into different registers, the patent inverts this approach by interleaving them within the same register. The load instruction generates alternating real-imaginary element sequences, and subsequent mask operations selectively process the required components, thereby achieving both speed and operational versatility.
2Ease of operation
If stride access is used to load real and imaginary parts alternately from memory, then data can be loaded into vector registers, but memory access speed decreases
Solution Approach 1:
The patent performs preliminary arrangement of real and imaginary parts in alternating sequence within the vector register during the load operation. This preliminary interleaving allows subsequent operations to directly access the required data elements without additional memory access, thereby improving memory access speed while maintaining ease of operation for complex number computations.
3Productivity
If specialized complex multiplication instructions are used, then multiplication can be performed efficiently, but other vector operations like division cannot be executed
Solution Approach 1:
The patent creates a universal vector operation framework that can handle addition, subtraction, multiplication, and division of complex numbers using the same basic mechanism. By loading real and imaginary parts alternately into the same vector register and using mask operations to selectively process elements, the system achieves multi-functionality without requiring specialized instructions for each operation type.
Data Source
AI summary
An information processing apparatus according to the present invention includes: a load instruction generating unit configured to generate an instruction to continuously access a memory in which a real part and an imaginary part composing complex data are alternately arranged, in accordance with arrangement of the real part and the imaginary part, and load the real part and the imaginary part as respective elements of a vector register; and an operation instruction generating unit configured to generate a vector operation instruction including an instruction to perform a vector operation of elements corresponding to element numbers different from each other between two vector registers and an instruction to perform a masked vector operation.