Namespace JYPPX.CudaSharp
Classes
- CudaArray
Wraps a CUDA array allocation and its copy helpers. 封装 CUDA array 分配对象及其复制辅助方法。
- CudaDevice
Provides current-device helpers over the CUDA runtime. 提供 CUDA runtime 当前设备相关 helper。
- CudaDeviceGraphMemoryInfo
Captures graph-memory allocator counters for one CUDA device. 捕获单个 CUDA 设备的 graph memory 分配器计数。
- CudaDeviceGraphMemorySummary
Summarizes copied CUDA device graph-memory counters without exposing native pointers. 汇总已复制的 CUDA device graph-memory 计数，不暴露原生指针。
- CudaDeviceInfo
Describes one CUDA device as reported by the bridge. 描述桥接层返回的单个 CUDA 设备信息。
- CudaDeviceProperties
Aggregates deployment-relevant CUDA device properties and optional runtime attributes. 聚合模型部署相关的 CUDA 设备属性以及可选 runtime attribute。
- CudaDeviceScope
Restores the previous CUDA current device when disposed. 在释放时恢复之前的 CUDA 当前设备。
- CudaDeviceSelectionRequirements
Minimum CUDA device properties used by
cudaChooseDevice.cudaChooseDevice使用的最小 CUDA 设备约束。
- CudaDriver
High-level optional CUDA Driver capability entry point. 可选 CUDA Driver 能力的高层入口。
- CudaDriverCapability
Copied optional CUDA Driver capability snapshot. 可选 CUDA Driver 能力的复制型快照。
- CudaDriverKernelLaunch
Owns one asynchronous CUDA Driver kernel launch and its participating owners. 拥有一次异步 CUDA Driver kernel launch 及其参与的所有 owner。
- CudaDriverModule
Owns a CUDA Driver module loaded from copied code bytes. 拥有从复制型 code bytes 加载的 CUDA Driver module。
- CudaEnvironmentProbe
Probes the current CUDA bridge state and available devices. 探测当前 CUDA bridge 状态和可用设备。
- CudaEnvironmentSnapshot
High-level environment snapshot for CUDA bridge diagnostics. 用于 CUDA bridge 诊断的高层环境快照。
- CudaEvent
Managed wrapper around a CUDA event handle. CUDA event 句柄的托管封装。
- CudaException
Thrown when the CUDA bridge reports a non-success status. 当 CUDA bridge 返回非成功状态时引发。
- CudaGraph
Managed wrapper around a captured CUDA graph. 已捕获 CUDA graph 的托管封装。
- CudaGraphAdjacentNodeSnapshot
Captures copied diagnostics for an adjacent CUDA graph node. 捕获相邻 CUDA graph node 的复制型诊断信息。
- CudaGraphChildSnapshot
Captures copied topology metadata for an embedded CUDA child graph. 捕获 CUDA embedded child graph 的复制型拓扑元数据。
- CudaGraphConditionalHandle
Owns a CUDA conditional graph handle associated with one CudaGraph. 拥有一个与指定 CudaGraph 关联的 CUDA conditional graph handle。
- CudaGraphConditionalNode
Owns the bridge metadata for a CUDA conditional node and its CUDA-owned body graphs. 拥有 CUDA conditional node 及其 CUDA-owned body graph 的 bridge metadata。
- CudaGraphDiagnosticSnapshot
Captures a bounded copied diagnostic snapshot for a CUDA graph. 捕获 CUDA graph 的有界复制型诊断快照。
- CudaGraphDiagnosticSummary
Summarizes copied CUDA graph diagnostics without exposing native graph or node pointers. 汇总已复制的 CUDA graph 诊断信息，不暴露原生 graph 或 node 指针。
- CudaGraphEdgeSnapshot
Captures copied diagnostics for a CUDA graph dependency edge. 捕获 CUDA graph 依赖边的复制型诊断信息。
- CudaGraphExec
Managed wrapper around an executable CUDA graph. 可执行 CUDA graph 的托管封装。
- CudaGraphExecDiagnosticSnapshot
Captures copied executable-state diagnostics for multiple CUDA graph nodes. 捕获多个 CUDA graph node 的复制型 executable 状态诊断。
- CudaGraphExecDiagnosticSummary
Summarizes copied CUDA graph executable diagnostics without exposing native pointers. 汇总已复制的 CUDA graph executable 诊断信息，不暴露原生指针。
- CudaGraphExecNodeStateSnapshot
Captures copied executable-state diagnostics for a CUDA graph node. 捕获 CUDA graph executable 中某个 node 的复制型执行状态诊断。
- CudaGraphExecUpdateSnapshot
Captures pointer-free metadata from a CUDA executable-graph update. 捕获 CUDA executable graph 更新产生的不含指针的元数据。
- CudaGraphExternalSemaphoreNodeSnapshot
Captures copied count and presence metadata from an external-semaphore graph node. 捕获 external-semaphore graph 节点的复制型 count 和存在性元数据。
- CudaGraphHostNodeParametersSnapshot
Captures callback-presence metadata without exposing a CUDA host callback or user-data pointer. 捕获 callback 存在性元数据，不暴露 CUDA host callback 或 user-data pointer。
- CudaGraphKernelNodeParametersSnapshot
Captures pointer-free scalar parameters from a CUDA graph kernel node. 捕获 CUDA graph kernel 节点中不含指针的标量参数。
- CudaGraphMemoryAllocation
Owns pointer-free bridge metadata for a CUDA graph memory-allocation node. 拥有 CUDA graph memory-allocation node 的无指针 bridge 元数据。
- CudaGraphMemoryAllocationNodeSnapshot
Captures copied scalar metadata from a CUDA graph memory-allocation node. 捕获 CUDA graph memory-allocation 节点的复制型标量元数据。
- CudaGraphMemoryFreeNodeSnapshot
Captures whether a CUDA graph memory-free node has a configured device pointer without exposing it. 捕获 CUDA graph memory-free 节点是否配置 device pointer，但不暴露该 pointer。
- CudaGraphNodeSnapshot
Captures copied diagnostics for a CUDA graph node entry. 捕获 CUDA graph node 条目的复制型诊断信息。
- CudaGraphNodeTopologySnapshot
Captures copied topology counters for a CUDA graph node. 捕获 CUDA graph node 的复制型拓扑计数。
- CudaGraphTopologySnapshot
Captures a copied, ownership-neutral topology summary for a CUDA graph. 捕获 CUDA graph 的复制型、无所有权转移拓扑摘要。
- CudaIpcExportToken
Immutable managed copy of an opaque CUDA IPC export token. CUDA IPC opaque 导出 token 的不可变托管副本。
- CudaIpcMemoryExportDescriptor
Immutable transport descriptor for an exported CUDA IPC memory allocation. 导出 CUDA IPC memory allocation 的不可变传输 descriptor。
- CudaKernelArgument
Stores a copied scalar or a managed CUDA memory owner without exposing native pointers. 保存复制型标量或托管 CUDA memory owner，不暴露 native pointer。
- CudaKernelLaunch
Owns one asynchronous named-kernel launch and retains every participating CUDA owner until disposal. 拥有一次异步 named-kernel launch，并在释放前租用所有参与的 CUDA owner。
- CudaKernelLibrary
Owns a CUDA runtime kernel library and exposes only copied metadata. 拥有 CUDA runtime kernel library，并且只公开复制型元数据。
- CudaLogSnapshot
Captures a copied UTF-8 CUDA runtime log dump. 捕获复制出的 UTF-8 CUDA runtime 日志。
- CudaManagedMemory
Managed wrapper around a CUDA unified-memory allocation. CUDA unified memory 分配的托管封装。
- CudaManagedMemoryBatch
Provides owner-safe CUDA 13 managed-memory batch operations. 提供 owner-safe 的 CUDA 13 managed memory 批量操作。
- CudaMemory
Managed wrapper around a device memory allocation. CUDA 设备内存分配的托管封装。
- CudaMemoryInfo
Reports CUDA device memory availability for the current device. 报告当前 CUDA 设备的显存可用性。
- CudaMemoryPressureSnapshot
Captures CUDA memory pressure and device capability signals for deployment diagnostics. 捕获 CUDA 内存压力和设备能力信号，用于模型部署诊断。
- CudaMemoryRangeDiagnosticSummary
Summarizes copied CUDA managed-memory range diagnostics without exposing native pointers. 汇总已复制的 CUDA managed-memory range 诊断信息，不暴露原生指针。
- CudaMipmappedArray
Managed wrapper around a CUDA mipmapped array allocation. CUDA mipmapped array 分配的托管封装。
- CudaOwnedMemoryPool
Owns a CUDA memory pool created through the bridge and releases it during disposal. 拥有通过桥接层创建的 CUDA 内存池，并在释放时销毁它。
- CudaPinnedMemory
Managed wrapper around CUDA pinned host memory. CUDA pinned host memory 的托管封装。
- CudaPitchedMemory
Managed wrapper around a CUDA pitched device allocation for 2D tensor/image-like data. 面向二维张量或图像类数据的 CUDA pitched 设备内存托管封装。
- CudaPrimaryExecutionContext
Owns a bridge wrapper for a CUDA device primary execution context. 拥有 CUDA 设备主执行上下文的 bridge 包装器。
- CudaRegisteredHostMemory
Registers a managed byte buffer as CUDA pinned host memory for asynchronous transfers. 将托管字节缓冲区注册为 CUDA pinned host memory 以支持异步传输。
- CudaRtcArtifact
Copied immutable NVRTC artifact with reproducibility metadata. 带可复现元数据的不可变 NVRTC 工件副本。
- CudaRtcCapability
Copied optional-NVRTC capability snapshot. 可选 NVRTC 的复制型能力快照。
- CudaRtcCompilationResult
Immutable NVRTC compilation result retaining failure logs and copied success artifacts. 保留失败日志与成功工件副本的不可变 NVRTC 编译结果。
- CudaRtcCompileOptions
Immutable validated NVRTC option snapshot. 不可变且已验证的 NVRTC option 快照。
- CudaRtcCompiler
High-level entry point for optional CUDA runtime compilation. 可选 CUDA runtime compilation 的高层入口。
- CudaRtcHeader
Immutable virtual header supplied to NVRTC. 提供给 NVRTC 的不可变虚拟 header。
- CudaRtcLoweredName
Copied mapping from a source name expression to its lowered name. source name expression 到 lowered name 的复制型映射。
- CudaRtcProgram
Owns a reusable native NVRTC program boundary without exposing the native handle. 拥有可复用 native NVRTC program 边界且不暴露 native handle。
- CudaRtcProgramSource
Immutable NVRTC source, virtual headers, and name expressions. 不可变的 NVRTC 源码、虚拟 header 与 name expression。
- CudaRuntimeInfo
Runtime-level CUDA bridge information. CUDA bridge 的 runtime 层信息。
- CudaRuntimeLogs
Provides copied access to CUDA runtime logs without exposing callbacks or native buffers. 提供 CUDA runtime 日志的复制型访问，不暴露 callback 或 native buffer。
- CudaSharpInfo
Repository-level metadata for the managed CUDA assembly. 托管 CUDA 程序集的仓库级元数据。
- CudaStream
Managed wrapper around a CUDA stream handle. CUDA stream 句柄的托管封装。
- CudaStreamCaptureToGraphSession
Owns the lifetime boundary for capture into an existing CUDA graph. 管理写入已有 CUDA graph 的 capture 生命周期边界。
- CudaSurfaceObject
Owns a CUDA surface object backed by a managed CUDA array owner. 拥有由托管 CUDA array owner 支撑的 CUDA surface object。
- CudaTextureObject
Owns a CUDA texture object backed by a managed CUDA array owner. 拥有由托管 CUDA array owner 支撑的 CUDA texture object。
Structs
- CudaArrayExtent
Represents the width, height, and depth of a CUDA array extent. 表示 CUDA array 的宽度、高度和深度范围。
- CudaArrayInfo
Provides immutable metadata about a CUDA array. 提供 CUDA array 的不可变元数据。
- CudaArrayMemoryRequirements
Describes device-memory requirements for a CUDA array. 描述 CUDA array 的设备内存需求。
- CudaArraySparseProperties
Describes sparse-tile metadata for a CUDA array. 描述 CUDA array 的稀疏 tile 元数据。
- CudaChannelFormatDescriptor
Describes the channel bit layout for a CUDA array. 描述 CUDA array 的通道位布局。
- CudaDevResourceSnapshot
Pointer-free copy of the observable fields in a CUDA device resource. CUDA 设备资源可观察字段的无指针复制快照。
- CudaDim3
Defines a non-zero CUDA grid or block extent. 定义非零 CUDA grid 或 block 范围。
- CudaGraphEdge
Describes a dependency edge between two CUDA graph nodes. 描述两个 CUDA graph node 之间的一条依赖边。
- CudaGraphEdgeData
Describes the CUDA dependency behavior attached to a graph edge. 描述 CUDA graph 边上的依赖行为。
- CudaGraphEdgeWithData
Describes a CUDA graph dependency edge together with CUDA edge data. 描述带 CUDA edge data 的 CUDA graph 依赖边。
- CudaGraphKernelNodeAttributeValue
Describes a CUDA graph kernel node attribute using a copied scalar descriptor. 使用复制出的标量 descriptor 描述 CUDA graph kernel node attribute。
- CudaGraphMemcpyNodeParameters
Describes a CUDA graph memcpy node using copied diagnostic values. 使用复制出的诊断值描述 CUDA graph memcpy 节点。
- CudaGraphMemsetNodeParameters
Describes a CUDA graph memset node using copied scalar parameters. 使用复制出的标量参数描述 CUDA graph memset 节点。
- CudaGraphNode
Identifies a CUDA graph node owned by a CudaGraph. 标识一个由 CudaGraph 拥有的 CUDA graph node。
- CudaGraphNodeDependency
Describes a graph node reached through a dependency edge and its edge data. 描述通过依赖边关联的 graph node 及其 edge data。
- CudaGraphNodeParamsDescriptor
Describes copied, pointer-free CUDA graph node parameters. 描述复制出的、无指针暴露的 CUDA graph node 参数。
- CudaKernelLaunchConfiguration
Describes an owner-bound CUDA kernel launch. 描述 owner-bound CUDA kernel launch。
- CudaKernelLibraryInventorySnapshot
Describes a copied CUDA kernel-library inventory without exposing borrowed kernel handles. 描述不暴露 borrowed kernel handle 的 CUDA kernel-library 复制型清单。
- CudaLogCursor
Identifies a position in CUDA's runtime log buffer. 标识 CUDA runtime 日志缓冲区中的位置。
- CudaManagedMemoryPrefetchRange
Describes an owner-bound CUDA managed-memory range and its prefetch destination. 描述与 owner 绑定的 CUDA managed memory 范围及其预取目标设备。
- CudaManagedMemoryRange
Describes an owner-bound subrange of a CUDA managed-memory allocation. 描述 CUDA managed memory 分配中与 owner 绑定的子范围。
- CudaMemoryLocation
Represents a validated, pointer-free CUDA managed-memory location. 表示经过校验且不包含指针的 CUDA managed memory 位置。
- CudaMemoryPool
Identifies a CUDA device memory pool without exposing the raw native handle to ordinary users. 标识一个 CUDA 设备内存池，同时避免向普通用户暴露原生句柄。
- CudaMemoryRangeAttributeValue
Managed snapshot of one CUDA memory range scalar attribute. 一个 CUDA memory range 标量属性的托管快照。
- CudaPointerAttributes
Provides diagnostic CUDA pointer metadata without exposing a raw pointer as the main user API. 提供 CUDA 指针诊断元数据，同时避免把裸指针作为主要用户 API 暴露。
- CudaResourceDescriptorSnapshot
Contains a pointer-free copied CUDA resource descriptor. 包含无指针的 CUDA resource descriptor 副本。
- CudaStreamCaptureInfo
Captures CUDA stream graph-capture metadata. 表示 CUDA stream 的 graph capture 元数据。
- CudaStreamCaptureScalarInfo
Captures the scalar metadata returned by CUDA's per-thread stream capture query. 表示 CUDA per-thread stream capture 查询返回的标量元数据。
- CudaStreamPriorityRange
CUDA stream priority range for the current device. 当前设备的 CUDA stream 优先级范围。
- CudaTextureDescriptor
Defines pointer-free CUDA texture sampling parameters. 定义无指针的 CUDA texture 采样参数。
- CudaTextureResourceViewSnapshot
Contains a copied CUDA texture resource-view descriptor. 包含复制型 CUDA texture resource-view descriptor。
Enums
- CudaArrayCreationFlags
Flags that control CUDA array creation behavior. 控制 CUDA array 创建行为的标志。
- CudaArraySparseFlags
Flags that describe sparse CUDA array behavior. 描述 sparse CUDA array 行为的标志。
- CudaAtomicCapability
Capability bitmask returned for a CUDA atomic operation. CUDA atomic operation 返回的能力位掩码。
- CudaAtomicOperation
CUDA atomic operations accepted by CUDA 13 host/P2P atomic capability queries. CUDA 13 host/P2P atomic 能力查询支持的 atomic operation。
- CudaChannelFormatKind
Channel-format kinds supported by CUDA arrays and textures. CUDA array 与 texture 支持的通道格式类型。
- CudaClusterSchedulingPolicyPreference
Describes CUDA cluster scheduling policy preference values. 描述 CUDA cluster scheduling policy preference 值。
- CudaDevResourceType
Identifies the CUDA device-resource union variant returned by CUDA 13. 标识 CUDA 13 返回的设备资源 union 变体。
- CudaDeviceAttribute
CUDA device attributes exposed through the runtime API. 通过 CUDA runtime API 暴露的设备属性。
- CudaDeviceLimit
Identifies CUDA runtime device limits that affect device-side execution behavior. 标识影响设备端执行行为的 CUDA runtime device limit。
- CudaDeviceP2PAttribute
CUDA peer-to-peer attribute identifiers. CUDA 点对点访问属性标识。
- CudaDeviceRuntimeFlags
CUDA runtime device scheduling and mapped-host-memory flags. CUDA runtime 设备调度与 mapped host memory 标志。
- CudaEventCreationFlags
Flags that control CUDA event creation behavior. 控制 CUDA event 创建行为的标志。
- CudaEventRecordFlags
Flags used when recording an event into a stream. 向 stream 记录 event 时使用的标志。
- CudaFunctionCachePreference
Selects the preferred shared-memory/L1 cache split for subsequent CUDA launches. 选择后续 CUDA launch 使用的 shared memory / L1 cache 倾向。
- CudaGpuDirectRdmaWritesScope
Selects the visibility scope for GPU Direct RDMA write flushing. 选择 GPU Direct RDMA 写入刷新的可见性范围。
- CudaGpuDirectRdmaWritesTarget
Selects the target for GPU Direct RDMA write flushing. 选择 GPU Direct RDMA 写入刷新的目标。
- CudaGraphConditionalHandleFlags
Controls initialization of a CUDA conditional value at graph launch. 控制 graph 启动时 CUDA 条件值的初始化。
- CudaGraphConditionalNodeType
Identifies the CUDA conditional graph node body policy. 标识 CUDA conditional graph 节点的 body 策略。
- CudaGraphDebugDotFlags
Flags that control which additional information CUDA writes to graph debug DOT output. 控制 CUDA graph debug DOT 输出附加信息的标志。
- CudaGraphDependencyType
Identifies the CUDA graph dependency type encoded in graph edge data. 标识 CUDA graph edge data 中的依赖类型。
- CudaGraphExecUpdateResult
Describes the result of a CUDA executable-graph update attempt. 描述 CUDA executable graph 更新尝试的结果。
- CudaGraphKernelNodeAttribute
Identifies CUDA graph kernel node attributes supported by the safe scalar bridge. 标识安全标量 bridge 支持的 CUDA graph kernel node attribute。
- CudaGraphMemoryAttribute
Identifies CUDA graph-memory attributes exposed by
cudaDeviceGetGraphMemAttribute. 标识cudaDeviceGetGraphMemAttribute暴露的 CUDA graph memory 属性。
- CudaGraphNodeParamsDescriptorKind
Identifies the copied, pointer-free CUDA graph node parameter descriptor kind. 标识复制型、无指针 CUDA graph node 参数 descriptor 类型。
- CudaGraphNodeType
Identifies CUDA graph node kinds reported by the CUDA runtime. 标识 CUDA runtime 报告的 CUDA graph node 类型。
- CudaHostRegistrationFlags
Flags that control registered host-memory behavior. 控制 registered host memory 行为的标志。
- CudaIpcExportTokenKind
Identifies the CUDA resource represented by a copied IPC export token. 标识复制型 CUDA IPC 导出 token 对应的资源类型。
- CudaKernelArgumentKind
Identifies a copied scalar or owner-bound device-memory kernel argument. 标识复制型标量或 owner-bound device-memory kernel 参数。
- CudaKernelAttribute
Selects a CUDA kernel attribute that cudaKernelSetAttributeForDevice permits callers to modify. 选择 cudaKernelSetAttributeForDevice 允许调用方修改的 CUDA kernel 属性。
- CudaKernelScalarType
Identifies the managed scalar layout copied into a CUDA kernel argument. 标识复制到 CUDA kernel 参数的托管标量布局。
- CudaManagedMemoryAttachmentFlags
Flags that control CUDA managed-memory attachment behavior. 控制 CUDA managed memory 附着行为的标志。
- CudaMemcpyKind
Describes the CUDA memcpy direction for copied diagnostics. 描述复制诊断信息中的 CUDA memcpy 方向。
- CudaMemoryAdvice
Memory-advice commands accepted by CUDA managed memory. CUDA managed memory 接受的内存建议命令。
- CudaMemoryLocationKind
Identifies the kind of a CUDA managed-memory location. 标识 CUDA managed memory 位置的种类。
- CudaMemoryLocationType
CUDA memory location types returned by CUDA 12.3+ range-location attributes. CUDA 12.3+ range location 属性返回的位置类型。
- CudaMemoryPointerType
Describes how CUDA classifies a memory pointer returned by the runtime. 描述 CUDA Runtime 对内存指针的分类方式。
- CudaMemoryPoolAccessFlags
Describes CUDA memory-pool access permissions for a device location. 描述某个 CUDA 设备位置对 memory pool 的访问权限。
- CudaMemoryPoolAttribute
Identifies CUDA memory-pool attributes used by deployment memory management. 标识部署内存管理中常用的 CUDA 内存池属性。
- CudaMemoryRangeAttribute
CUDA managed-memory range attributes that can be queried as scalar values. 可作为标量查询的 CUDA managed memory range 属性。
- CudaPinnedMemoryAllocationFlags
Flags that control pinned host-memory allocation behavior. 控制 pinned host memory 分配行为的标志。
- CudaResourceType
Identifies the CUDA resource kind copied from a descriptor. 标识从 descriptor 复制出的 CUDA 资源类型。
- CudaRtcArtifactKind
Identifies a copied runtime-compilation artifact. 标识复制型 runtime compilation 工件。
- CudaRtcResultCode
Stable NVRTC result codes returned as diagnostic data. 作为诊断数据返回的稳定 NVRTC 结果码。
- CudaSharedMemoryConfig
Selects CUDA shared-memory bank width for compatible devices. 选择兼容设备上的 CUDA shared-memory bank 宽度。
- CudaStreamCaptureDependencyMode
Controls whether stream-capture dependencies are added or replaced. 控制 stream capture dependencies 是追加还是替换。
- CudaStreamCaptureMode
Modes that control CUDA stream capture validation. 控制 CUDA stream capture 校验行为的模式。
- CudaStreamCaptureStatus
Describes the CUDA stream capture state used by CUDA Graph capture. 描述 CUDA Graph 捕获流程中的 CUDA stream 捕获状态。
- CudaStreamCreationFlags
Flags that control CUDA stream creation behavior. 控制 CUDA stream 创建行为的标志。
- CudaTextureAddressMode
Specifies CUDA texture addressing behavior. 指定 CUDA texture 地址处理行为。
- CudaTextureFilterMode
Specifies CUDA texture filtering. 指定 CUDA texture 过滤方式。
- CudaTextureReadMode
Specifies CUDA texture read conversion. 指定 CUDA texture 读取转换。
- CudaTextureResourceViewFormat
Identifies a CUDA texture resource-view format. 标识 CUDA texture resource-view 格式。