Namespace DotCompute.Abstractions.Types
Classes
- CoalescingIssue
Represents a specific memory coalescing issue identified during analysis.
- MemoryAccessInfo
Represents information about a memory access pattern in CUDA kernels. Used to analyze coalescing efficiency and identify optimization opportunities.
- Optimized
Represents an optimized kernel implementation strategy.
- PerformanceTrend
Represents a comprehensive performance trend analysis for kernels and operations. Combines trend detection, statistical analysis, and performance recommendations to provide actionable insights into performance patterns over time.
- StructureOfArrays
Structure of Arrays (SoA) utilities for optimal SIMD performance. Converts Array of Structures (AoS) to Structure of Arrays (SoA) for better vectorization.
- ThroughputMetrics
Throughput performance metrics for pipeline execution. Measures data processing rates and operational throughput.
- TileAnalysis
Represents the analysis of tile efficiency for shared memory optimization.
- TransferTypeStats
Statistics for a specific type of memory transfer.
Structs
- Dim2
Represents a 2-dimensional configuration for CUDA operations.
- Dim3
Represents a three-dimensional structure for defining grid and block dimensions in GPU computing.
- EventId
Unique identifier for CUDA events.
- MemoryAccessMetrics
Comprehensive metrics for memory access operations including performance counters, bandwidth utilization, and access patterns
- Point3D
Simple 3D point structure for demonstration.
Enums
- AccessOrder
Defines the memory access order patterns for 2D data structures.
- BottleneckType
Defines types of performance bottlenecks that can occur in computational workloads. These bottleneck types help identify the primary limiting factors in system performance and guide optimization strategies for improving overall throughput.
- CacheConfig
Defines cache configuration preferences for CUDA kernels. Controls the allocation of on-chip memory between L1 cache and shared memory.
- DataType
Defines data types supported by tensor core operations.
- ErrorCategory
Categories of CUDA errors for handling strategies.
- ErrorSeverity
Defines error severity levels.
- ExecutionStatus
Represents the execution status of a pipeline or stage.
- ExecutionStrategyType
Defines the type of execution strategy for kernel and pipeline execution.
- ExportFormat
Export formats for validation reports.
- FloatingPointMode
Specifies the floating-point precision and behavior mode.
- IssueSeverity
Defines the severity levels for memory coalescing issues.
- IssueType
Defines the types of memory coalescing issues that can be identified.
- KernelType
Enumeration of kernel types for cross-platform kernel categorization and optimization.
- ManagedMemoryOptions
Flags for CUDA managed memory allocation and behavior.
- MemoryAccessPattern
Defines memory access patterns for kernel optimization. Understanding access patterns enables better memory coalescing and cache utilization.
- MemoryOptimizationLevel
Specifies the memory optimization level for kernel execution.
- MemoryResidence
Defines where unified memory is currently resident.
- MemoryTransferType
Defines the types of memory transfers in CUDA.
- OptimizationHint
Provides hints for kernel launch optimization based on workload characteristics.
- OptimizationLevel
Defines optimization levels for kernel compilation. These levels correspond to standard compiler optimization flags and provide a balance between compilation time and runtime performance.
- SharedMemoryCarveout
Shared memory carveout preferences for Ada generation GPUs.
- StreamCreationFlags
Flags for CUDA stream creation and behavior.
- StreamPriority
Defines priority levels for CUDA streams.
- SynchronizationMode
Synchronization modes for parallel stage execution.
- TensorCoreArchitecture
Defines NVIDIA GPU architectures with tensor core support.
- TrendDirection
Defines the possible directions for performance trends observed over time. Indicates whether a performance metric is getting better, worse, or staying consistent.
- ValidationStatus
Status of a validation test.
- WarpSchedulingMode
Defines warp scheduling modes for CUDA kernel execution. Controls how warps are scheduled on streaming multiprocessors.
- WorkStatus
Defines the status of a work item in the execution pipeline. Used for tracking work items through the execution lifecycle.
- WorkloadType
Defines the type of workload being processed