CK Tile Index

CK Tile documentation structure:

• Introduction and Motivation - Why Tile Distribution Matters
  • Overview
  • The GPU Memory Problem
  • The Tile Distribution Solution
  • The Coordinate Mapping Insight
  • What’s Coming Next
  • Summary
  • Next Steps
• CK Tile buffer view
  • Address Space Usage Patterns
  • Get Operations
  • Update Operations
  • Atomic Operations
• Tensor Views - Multi-Dimensional Structure
  • Overview
  • TensorView Architecture
  • The Foundation: BufferView and TensorDescriptor
  • C++ Implementation
  • Memory Layouts and Strides
  • Advanced Operations
  • Performance Considerations
  • TensorView vs BufferView
  • Integration with Tile Distribution
  • Summary
  • Next Steps
• Tile Distribution - The Core API
  • Overview
  • Complete Tile Distribution System Overview
  • Coordinate System Architecture
  • What is Tile Distribution?
  • Problem Space Mapping
  • Creating a TileDistribution
  • Hierarchical Decomposition
  • Advanced Example: Matrix Multiplication Distribution
  • Work Distribution Pattern
  • Memory Access Patterns
  • Transformation Pipeline
  • Performance Comparison
  • Summary
  • Next Steps
• Coordinate Systems - The Mathematical Foundation
  • Overview
  • The Five Coordinate Spaces
  • Thread Identification
  • Logical Work Organization
  • Physical Tensor Coordinates
  • The Core Transformation: P + Y → X
  • Replication and Cooperation
  • Memory Linearization
  • Complete Pipeline Example
  • Real-World Example: Matrix Multiplication
  • Performance Implications
  • Summary
  • Next Steps
• Terminology Reference - Key Concepts and Definitions
  • Overview
  • Core Concepts
  • Coordinate Spaces
  • Dimension Types
  • Transformations
  • Operations
  • Memory Concepts
  • Distribution Components
  • GPU Hardware Terms
  • Template Parameters
  • Optimization Terms
  • Common Patterns
  • Performance Metrics
  • Usage Examples
  • Related Documentation
• Tensor Adaptors - Chaining Transformations
  • Overview
  • TensorAdaptor Basics
  • Transpose Adaptor: Dimension Reordering
  • Single-Stage Adaptors: Custom Transform Chains
  • Chaining Adaptors: Building Complex Transformations
  • Transform Addition: Extending Existing Adaptors
  • Advanced Patterns
  • Common Transform Chains
  • Key Concepts Summary
  • Key C++ Patterns in Composable Kernel
  • Next Steps
• Individual Transform Operations
  • What Are Transforms?
  • MergeTransform
  • UnmergeTransform
  • EmbedTransform
  • ReplicateTransform
  • OffsetTransform
  • PassThroughTransform - Identity
  • PadTransform
  • Additional Transform Types
  • Summary
  • Next Steps
• Tensor Descriptors - Complete Tensor Specifications
  • Overview
  • Creating Basic Tensor Layouts
  • The Pipeline Concept
  • Advanced Layouts: Step-by-Step Transformation
  • 5D to 3D Block Transformation
  • Common Descriptor Patterns
  • Key Concepts Summary
  • Performance Considerations
  • Next Steps
• Tile Window - Data Access Gateway
  • Overview
  • TileWindow Architecture
  • What is a TileWindow?
  • LoadStoreTraits - The Access Pattern Engine
  • Space-Filling Curves for Memory Access
  • TileWindow Data Flow
  • Creating and Using TileWindow
  • The Load Operation Deep Dive
  • Load Operation Architecture
  • Memory Access Patterns
  • Window Movement and Updates
  • Store Operations with Vectorization
  • Complete Load-Compute-Store Pipeline
  • Performance Characteristics
  • Best Practices
  • Summary
  • Next Steps
• LoadStoreTraits - Memory Access Optimization Engine
  • Overview
  • Key Concepts
  • C++ Implementation
  • Vectorization Selection Algorithm
  • Memory Access Patterns
  • Performance Analysis
  • Integration with Space-Filling Curves
  • Advanced Optimizations
  • Best Practices
  • Summary
  • Next Steps
• Space-Filling Curves - Optimal Memory Traversal
  • Overview
  • Key Concepts
  • C++ Implementation
  • Basic Usage Examples
  • Advanced Patterns
  • Performance Analysis
  • Handling Edge Cases
  • Integration with CK Tile
  • Best Practices
  • Summary
  • Next Steps
• Static Distributed Tensor
  • Overview
  • Thread-Local Storage Model
  • Memory Layout and Access Patterns
  • Element Access and Indexing
  • Thread Coordination and Synchronization
  • Practical Usage Patterns
  • Performance Considerations
  • Integration with CK Tile Ecosystem
  • Summary
• Convolution Implementation with CK Tile
  • Overview
  • Understanding Sliding Windows
  • Naive Convolution Implementation
  • Window Extraction with Tensor Descriptors
  • Im2col Transformation
  • Optimized Convolution Kernel
  • Multi-Channel Convolution
  • Performance Optimizations
  • Performance Characteristics
  • Summary
• Advanced Coordinate Movement
  • Overview
  • TensorCoordinate: Descriptor-Aware Navigation
  • TensorAdaptorCoordinate: Transform-Aware Tracking
  • Efficient Coordinate Movement
  • Advanced Movement Patterns
  • Performance Considerations
  • Integration with CK Tile Components
• Load Datat Share Index Swapping
  • Overview
  • Coordinate Transformation Pipeline
  • C++ Implementation
  • Practical Usage in GEMM
  • Bank Conflict Analysis
  • Performance Benefits
  • Advanced Configurations
  • Integration with Tile Distribution
  • Summary
• Memory Swizzling with Morton Ordering
  • Overview
  • Understanding Morton Ordering
  • Stage 1: Tiling with UnmergeTransform
  • Stage 2: Morton Ordering with MergeTransform
  • Memory Access Pattern Analysis
  • GPU Kernel Implementation
  • Bank Conflict Reduction
  • Practical Applications
  • Summary
• Tensor Coordinates
  • Overview
  • MultiIndex Implementation
  • Creating and Using MultiIndex
  • MultiIndex in Coordinate Flow
  • Common Usage Patterns
  • Advanced MultiIndex Operations
  • Integration with Tensor Operations
  • Performance Considerations
  • Summary
• Sweep Tile
  • Overview
  • The Complete GPU Workflow
  • Basic Sweep Implementation
  • Memory Efficiency Pattern
  • Practical Sweep Patterns
  • GEMM Sweep Pattern
  • Advanced Sweep Patterns
  • Performance Characteristics
  • Integration with CK Tile Components
  • Summary
• Encoding Internals
  • Overview
  • Encoding Structure
  • Parameter Breakdown
  • Transformation Pipeline
  • Y to D Linearization
  • Practical Examples
  • Performance Implications
  • Summary
• Thread Mapping - Connecting to Hardware
  • Thread Identification and Partition Indices
  • Thread-to-Data Mapping
  • Thread Cooperation Patterns
  • Memory Access Patterns
  • Practical Thread Mapping Example
  • Key Takeaways
• CK Tile Hardware Documentation
  • Overview
  • Documentation Structure
  • Key Hardware Considerations
  • Performance Guidelines
  • Next Steps