Stream-Based Address Compression for Higher Memory Bandwidth
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current address compression methods, such as dynamic base register caching and stride address compression, do not achieve a high enough compression ratio due to their processing unit being a request, leading to inefficient bandwidth usage and limited system memory capacity, especially for 32-bit CPUs.
Innovation Solution
The method determines a compression algorithm based on address features for operation request messages with the same stream number, compressing addresses within a time window, and constructs a message packet with both compressed and uncompressed information parts to improve the address compression ratio.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If complete address is transferred for each request without using locality features, then address information is fully transmitted, but redundant information transfer occurs, wasting physical bandwidth and transmission power consumption
Solution Approach 1:
The patent merges multiple operation request messages with the same stream number within a time window into a single compressed message. By grouping requests that share common address features (such as same high-order address bits or address patterns), the system transmits shared information once and only transmits differences for each individual request, thereby reducing redundant transmissions and power consumption while maintaining complete address information availability.
Solution Approach 2:
The patent segments the address into multiple parts (e.g., high-order part and low-order part, or base address and offset). The high-order part or common address features are extracted and transmitted once for a group of requests, while only the varying low-order parts or differences are transmitted for each individual request. This segmentation enables selective transmission of necessary information, reducing overall transmission volume and energy consumption.
2Reliability
If complete address is transferred for each request without using locality features, then address information is fully transmitted, but physical bandwidth is wasted
Solution Approach 1:
Multiple operation request messages sharing the same stream number and address features are merged into a single compressed message. The common address information is transmitted once, and only the varying portions are transmitted for each request, significantly reducing the total volume of data transmitted over the physical bandwidth while ensuring complete address information is available for execution.
Solution Approach 2:
The address is segmented into common parts and variable parts. The common parts (such as high-order address bits or base addresses) are extracted and transmitted once for a group of requests, while only the variable parts (such as low-order address bits or offsets) are transmitted for each individual request. This segmentation reduces the total transmission volume and optimizes bandwidth utilization.
3Quantity of substance
If 32-bit address transfer is used for 32-bit CPU, then address transfer capacity is sufficient, but system memory capacity is limited to 4G
Solution Approach 1:
The patent introduces a new dimension for address representation by using compressed format that includes not only the address bits but also compression algorithm identification and parameter information. This allows the system to represent addresses beyond the traditional 32-bit limit by encoding addresses in a more efficient format that can accommodate larger address spaces while maintaining compatibility with 32-bit CPU architecture through intelligent compression and decompression.
Solution Approach 2:
The patent changes the parameter representation by using variable-length encoding and compression algorithms. Instead of fixed 32-bit addresses, the system uses compressed address representations that can efficiently encode both small and large address spaces. The compression ratio and encoding format are dynamically adjusted based on address patterns, enabling the system to support memory capacities beyond 4G while maintaining efficient address transfer for 32-bit CPUs.
4Productivity
If dynamic base register caching or stride address compression is used, then address compression is achieved, but compression ratio is not high due to request-level processing
Solution Approach 1:
The patent merges multiple operation request messages with the same stream number within a time window into a single compressed message. By processing a group of requests together rather than individually, the system can identify and compress common address features across multiple requests, achieving a much higher compression ratio. The merged compression allows shared address information to be encoded once and reused for all requests in the group, maximizing compression efficiency.
Solution Approach 2:
The patent performs preliminary analysis of address features across multiple requests within a time window before compression. By预先 identifying common patterns, shared high-order address bits, or repetitive address structures across the group of requests, the system can optimize the compression algorithm selection and parameter configuration to achieve maximum compression ratio. This preliminary group-level analysis enables more effective compression than request-level processing.
Data Source
Figure 1
Figure 2~3
Figure 4
AI summary
Embodiments of the present invention provide an address compression method, an address decompression method, a compressor, and a decompressor, which can improve an address compression ratio. The address compression method includes: after a compressor receives multiple operation request messages that are sent by a first processor, determining, according to an address feature formed by address information carried in all operation request messages that have a same stream number, a compression algorithm corresponding to the operation request messages that have a same stream number; and then compressing, according to the determined compression algorithm, addresses carried in the operation request messages that have a same stream number. The present invention is applicable to the computer field.