Memory Copy Instruction Parallel Data Transfer

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Current techniques for copying data between memory locations, such as from DRAM to local or shared memory, face limitations including bottlenecks and wasted power due to limited outstanding data transfers and inefficient data processing.

Innovation Solution

A memory copy instruction is implemented, allowing data to be copied between memory locations using a processor or thread, supporting operands for source and destination addresses, size, and synchronization to avoid race conditions, enabling parallel execution and efficient data transfer.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Productivity

If data is copied from external memory to register file then transferred to local scratchpad memory, then data transfer can be performed, but the number of outstanding data transfers is limited causing bottlenecks and wasted power

Engineering Contradiction:
Improvedata transfer efficiencyVSAvoidpower consumption
Core Design Contradiction:
ProductivityVSLoss of energy

Solution Approach 1:

The memory copy instruction is divided into multiple parallel copy operations that can execute simultaneously. Each copy operation transfers data from external memory to local memory independently, increasing the number of outstanding transfers and eliminating bottlenecks while reducing total power consumption through parallel processing

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent introduces a new dimension to data transfer by implementing multiple copy operations that proceed in parallel across different memory locations. This transforms sequential transfers into simultaneous operations, effectively increasing throughput without proportionally increasing power consumption

Inventive Principle:
Principle #17Another dimension (Dimensionality change)

2Productivity

If data copying operations are performed sequentially, then simplicity is maintained, but processing time increases and system performance decreases

Engineering Contradiction:
Improvesystem performanceVSAvoiddata copying time
Core Design Contradiction:
ProductivityVSLoss of time

Solution Approach 1:

Multiple data copy operations are prepared and initiated in advance before the actual data processing begins. This allows data to be transferred to local memory ahead of time, so when processing starts, the data is already ready and can be processed immediately without sequential waiting delays

Inventive Principle:
Principle #10Preliminary action

Solution Approach 2:

The memory copy instruction enables continuous data transfer operations to proceed simultaneously rather than sequentially. Multiple copy operations continue in parallel without interruption, maintaining useful action throughout the data transfer process and eliminating idle waiting time between sequential operations

Inventive Principle:
Principle #20Continuity of useful action

Data Source

PatentUS9164690B2System, method, and computer program product for copying data between memory locations
Publication Date: 2015.10.20 NVIDIA CORP
  • US9164690B2 patent drawing
  • US9164690B2 patent drawing
  • US9164690B2 patent drawing

AI summary

A system, method, and computer program product are provided for copying data between memory locations. In use, a memory copy instruction is implemented. Additionally, data is copied from a first memory location to a second memory location, utilizing the memory copy instruction.