Vector Execution Unit 512-Bit Broadcast Replication

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Existing instruction execution units in vector processors lack the ability to broadcast data values at granularities beyond 256 bits, limiting their replication patterns and operational complexity.

Innovation Solution

The introduction of advanced VBROADCAST instructions that support a 512-bit result width with write masking at the granularity of individual data elements, allowing for more complex replication patterns through a write masking layer that can mask at 32-bit or 64-bit granularity.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Adaptability or versatility

If existing instruction execution units are used, then the processor can execute standard vector operations, but the data broadcast width is limited to 256 bits which restricts operational complexity

Engineering Contradiction:

Improvedata replication pattern complexityVSAvoidexecution unit structure

Core Design Contradiction:

Adaptability or versatilityVSDevice complexity

Solution Approach 1:

The execution unit is segmented into distinct functional components: a replication unit that handles the complex 512-bit broadcast operation, and a write masking layer that operates at 32-bit or 64-bit granularity. This segmentation allows the replication unit to focus on high-level data distribution while the masking layer handles fine-grained control, resolving the contradiction between enhanced adaptability and device complexity.

Inventive Principle:

Principle #1Segmentation

Solution Approach 2:

A write masking layer is introduced as an intermediary component between the replication unit and the destination registers. This masking layer operates at 32-bit or 64-bit granularity to control which elements of the 512-bit broadcast result are actually written to memory or registers. The intermediary masking layer enables complex replication patterns without requiring the entire execution unit to be redesigned for fine-grained control.

Inventive Principle:

Principle #24Intermediary (Mediator)

2Productivity

If 512-bit broadcast operation is implemented, then larger data sets can be processed in parallel, but the instruction execution unit requires additional complexity

Engineering Contradiction:

Improvedata processing throughputVSAvoidexecution unit structure

Core Design Contradiction:

ProductivityVSDevice complexity

Solution Approach 1:

The patent merges the 512-bit replication capability with the existing write masking infrastructure. The replication unit broadcasts data at 512-bit width to improve productivity, while the existing write masking layers (operating at 32-bit or 64-bit granularity) are combined with this replication unit to control the actual write operations. This merging allows high-throughput data processing without proportionally increasing overall device complexity.

Inventive Principle:

Principle #5Merging (Combining)

Solution Approach 2:

The execution unit is extended into a new dimension of 512-bit width for the replication operation, while the write masking continues to operate at traditional 32-bit or 64-bit granularities. This dimensional extension allows larger data sets to be processed in parallel (improving productivity) while the masking infrastructure maintains its established complexity characteristics.

Inventive Principle:

Principle #17Another dimension (Dimensionality change)

3Manufacturing precision

If write masking at 32-bit or 64-bit granularity is added to 512-bit broadcast, then precise control over replication is achieved, but the instruction complexity increases

Engineering Contradiction:

Improvedata element control precisionVSAvoidmasking layer structure

Core Design Contradiction:

Manufacturing precisionVSDevice complexity

Solution Approach 1:

The write masking layer applies different masking granularities (32-bit or 64-bit) to different segments of the 512-bit broadcast result. This local quality approach allows precise control over specific data elements or groups of elements without requiring the entire masking infrastructure to operate at the finest granularity, thereby achieving manufacturing precision while managing masking layer complexity.

Inventive Principle:

Principle #3Local quality

Data Source

PatentUS9336000B2Instruction execution unit that broadcasts data values at different levels of granularity

Publication Date: 2016.05.10 INTEL CORP

AI summary

An apparatus is described that includes an execution unit to execute a first instruction and a second instruction. The execution unit includes input register space to store a first data structure to be replicated when executing the first instruction and to store a second data structure to be replicated when executing the second instruction. The first and second data structures are both packed data structures. Data values of the first packed data structure are twice as large as data values of the second packed data structure. The first data structure is four times as large as the second data structure. The execution unit also includes replication logic circuitry to replicate the first data structure when executing the first instruction to create a first replication data structure, and, to replicate the second data structure when executing the second instruction to create a second replication data structure.