1. Attribute¶

1.1. GlobalDefs¶

This section describes Global enum and defines of TOPS runtime API.

enum topsDeviceAttribute_t¶

Values:

enumerator topsDeviceAttributeClockRate¶: Peak clock frequency in kilohertz.

enumerator topsDeviceAttributeComputeCapabilityMajor¶: Major compute capability version number.

enumerator topsDeviceAttributeComputeCapabilityMinor¶: Minor compute capability version number.

enumerator topsDeviceAttributeMaxBlockDimX¶: Max block size in width.

enumerator topsDeviceAttributeMaxBlockDimY¶: Max block size in height.

enumerator topsDeviceAttributeMaxBlockDimZ¶: Max block size in depth.

enumerator topsDeviceAttributeMaxGridDimX¶: Max grid size in width.

enumerator topsDeviceAttributeMaxGridDimY¶: Max grid size in height.

enumerator topsDeviceAttributeMaxGridDimZ¶: Max grid size in depth.

enumerator topsDeviceAttributeMaxThreadsPerBlock¶: Maximum number of threads per block.

enumerator topsDeviceAttributeMultiprocessorCount¶: Number of multiprocessors on the device.

enumerator topsDeviceAttributeMaxSharedMemoryPerBlock¶: Maximum shared memory available per block in bytes.

enumerator topsDeviceAttributeEnflameSpecificBegin¶

enumerator topsDeviceAttributeEnflameMcChannel¶: Memory Control Channel count.

enum topsComputeMode¶

Values:

enumerator topsComputeModeDefault¶

enumerator topsComputeModeExclusive¶

enumerator topsComputeModeProhibited¶

enumerator topsComputeModeExclusiveProcess¶

enum topsMemoryAdvise¶

Values:

enumerator topsMemAdviseSetReadMostly¶

be written to

Data will mostly be read and only occasionally

enumerator topsMemAdviseUnsetReadMostly¶: Undo the effect of topsMemAdviseSetReadMostly.

enumerator topsMemAdviseSetPreferredLocation¶

the specified device

Set the preferred location for the data as

enumerator topsMemAdviseUnsetPreferredLocation¶: Clear the preferred location for the data.

enumerator topsMemAdviseSetAccessedBy¶

so prevent page faults as much as possible

Data will be accessed by the specified device,

enumerator topsMemAdviseUnsetAccessedBy¶

for the specified device

Let TOPS to decide on the page faulting policy

enumerator topsMemAdviseSetCoarseGrain¶: The default memory model is fine-grain. That allows coherent operations between host and device, while executing kernels. The coarse-grain can be used for data that only needs to be coherent at dispatch boundaries for better performance

enumerator topsMemAdviseUnsetCoarseGrain¶: Restores cache coherency policy back to fine-grain.

enum topsMemRangeCoherencyMode¶

Values:

enumerator topsMemRangeCoherencyModeFineGrain¶

done coherently from all devices

Updates to memory with this attribute can be

enumerator topsMemRangeCoherencyModeCoarseGrain¶

performed by a single device at a time

Writes to memory with this attribute can be

enumerator topsMemRangeCoherencyModeIndeterminate¶: Memory region queried contains subregions with both topsMemRangeCoherencyModeFineGrain and topsMemRangeCoherencyModeCoarseGrain attributes

enum topsMemRangeAttribute¶

Values:

enumerator topsMemRangeAttributeReadMostly¶

only occasionally be written to

Whether the range will mostly be read and

enumerator topsMemRangeAttributePreferredLocation¶: The preferred location of the range.

enumerator topsMemRangeAttributeAccessedBy¶

set for the specified device

Memory range has topsMemAdviseSetAccessedBy

enumerator topsMemRangeAttributeLastPrefetchLocation¶

prefetched

The last location to where the range was

enumerator topsMemRangeAttributeCoherencyMode¶

topsMemRangeCoherencyMode for the range

Returns coherency mode

enum topsMemPoolAttr¶

Values:

enumerator topsMemPoolReuseFollowEventDependencies¶: (value type = int) Allow topsMemAllocAsync to use memory asynchronously freed in another streams as long as a stream ordering dependency of the allocating stream on the free action exists. tops events and null stream interactions can create the required stream ordered dependencies. (default enabled)

enumerator topsMemPoolReuseAllowOpportunistic¶: (value type = int) Allow reuse of already completed frees when there is no dependency between the free and allocation. (default enabled)

enumerator topsMemPoolReuseAllowInternalDependencies¶: (value type = int) Allow topsMemAllocAsync to insert new stream dependencies in order to establish the stream ordering required to reuse a piece of memory released by cuFreeAsync (default enabled).

enumerator topsMemPoolAttrReleaseThreshold¶: (value type = uint64_t) Amount of reserved memory in bytes to hold onto before trying to release memory back to the OS. When more than the release threshold bytes of memory are held by the memory pool, the allocator will try to release memory back to the OS on the next call to stream, event or context synchronize. (default 0)

enumerator topsMemPoolAttrReservedMemCurrent¶: (value type = uint64_t) Amount of backing memory currently allocated for the mempool.

enumerator topsMemPoolAttrReservedMemHigh¶: (value type = uint64_t) High watermark of backing memory allocated for the mempool since the last time it was reset. High watermark can only be reset to zero.

enumerator topsMemPoolAttrUsedMemCurrent¶: (value type = uint64_t) Amount of memory from the pool that is currently in use by the application.

enumerator topsMemPoolAttrUsedMemHigh¶: (value type = uint64_t) High watermark of the amount of memory from the pool that was in use by the application since the last time it was reset. High watermark can only be reset to zero.

enum topsMemLocationType¶

Values:

enumerator topsMemLocationTypeInvalid¶

enumerator topsMemLocationTypeDevice¶: Device location, thus it’s TOPS device ID.

enum topsMemAccessFlags¶

Values:

enumerator topsMemAccessFlagsProtNone¶: Default, make the address range not accessible.

enumerator topsMemAccessFlagsProtRead¶: Set the address range read accessible.

enumerator topsMemAccessFlagsProtReadWrite¶: Set the address range read-write accessible.

enum topsMemAllocationType¶

Values:

enumerator topsMemAllocationTypeInvalid¶

enumerator topsMemAllocationTypePinned¶: This allocation type is ‘pinned’, i.e. cannot migrate from its current location while the application is actively using it

enumerator topsMemAllocationTypeMax¶

enum topsMemAllocationHandleType¶

Values:

enumerator topsMemHandleTypeNone¶: Does not allow any export mechanism.

enumerator topsMemHandleTypePosixFileDescriptor¶: Allows a file descriptor for exporting. Permitted only on POSIX systems.

enumerator topsMemHandleTypeWin32¶: Allows a Win32 NT handle for exporting. (HANDLE)

enumerator topsMemHandleTypeWin32Kmt¶: Allows a Win32 KMT handle for exporting. (D3DKMT_HANDLE)

enum topsJitOption¶

Values:

enumerator topsJitOptionMaxRegisters¶

enumerator topsJitOptionThreadsPerBlock¶

enumerator topsJitOptionWallTime¶

enumerator topsJitOptionInfoLogBuffer¶

enumerator topsJitOptionInfoLogBufferSizeBytes¶

enumerator topsJitOptionErrorLogBuffer¶

enumerator topsJitOptionErrorLogBufferSizeBytes¶

enumerator topsJitOptionOptimizationLevel¶

enumerator topsJitOptionTargetFromContext¶

enumerator topsJitOptionTarget¶

enumerator topsJitOptionFallbackStrategy¶

enumerator topsJitOptionGenerateDebugInfo¶

enumerator topsJitOptionLogVerbose¶

enumerator topsJitOptionGenerateLineInfo¶

enumerator topsJitOptionCacheMode¶

enumerator topsJitOptionSm3xOpt¶

enumerator topsJitOptionFastCompile¶

enumerator topsJitOptionNumOptions¶

enum topsFuncAttribute¶

Warning

these hints and controls are ignored.

Values:

enumerator topsFuncAttributeMaxDynamicSharedMemorySize¶

enumerator topsFuncAttributePreferredSharedMemoryCarveout¶

enumerator topsFuncAttributeMax¶

enum topsFuncCache_t¶

Warning

these hints and controls are ignored.

Values:

enumerator topsFuncCachePreferNone¶: no preference for shared memory or L1 (default)

enumerator topsFuncCachePreferShared¶: prefer larger shared memory and smaller L1 cache

enumerator topsFuncCachePreferL1¶: prefer larger L1 cache and smaller shared memory

enumerator topsFuncCachePreferEqual¶: prefer equal size L1 cache and shared memory

enum topsSharedMemConfig¶

Warning

these hints and controls are ignored.

Values:

enumerator topsSharedMemBankSizeDefault¶: The compiler selects a device-specific value for the banking.

enumerator topsSharedMemBankSizeFourByte¶

when adjacent threads access data 4 bytes apart.

Shared mem is banked at 4-bytes intervals and performs best

enumerator topsSharedMemBankSizeEightByte¶

when adjacent threads access data 4 bytes apart.

Shared mem is banked at 8-byte intervals and performs best

enum topsExternalMemoryHandleType_enum¶

Values:

enumerator topsExternalMemoryHandleTypeOpaqueFd¶

enumerator topsExternalMemoryHandleTypeOpaqueWin32¶

enumerator topsExternalMemoryHandleTypeOpaqueWin32Kmt¶

enumerator topsExternalMemoryHandleTypeD3D12Heap¶

enumerator topsExternalMemoryHandleTypeD3D12Resource¶

enumerator topsExternalMemoryHandleTypeD3D11Resource¶

enumerator topsExternalMemoryHandleTypeD3D11ResourceKmt¶

enum topsExternalSemaphoreHandleType_enum¶

Values:

enumerator topsExternalSemaphoreHandleTypeOpaqueFd¶

enumerator topsExternalSemaphoreHandleTypeOpaqueWin32¶

enumerator topsExternalSemaphoreHandleTypeOpaqueWin32Kmt¶

enumerator topsExternalSemaphoreHandleTypeD3D12Fence¶

enum topsGLDeviceList¶

Values:

enumerator topsGLDeviceListAll¶: All tops devices used by current OpenGL context.

enumerator topsGLDeviceListCurrentFrame¶

frame

Tops devices used by current OpenGL context in current

enumerator topsGLDeviceListNextFrame¶

frame.

Tops devices used by current OpenGL context in next

enum __TOPS_NODISCARD topsError_t topsError_t

typedef enum topsDeviceAttribute_t topsDeviceAttribute_t

typedef enum topsMemoryAdvise topsMemoryAdvise

typedef enum topsMemRangeCoherencyMode topsMemRangeCoherencyMode

typedef enum topsMemRangeAttribute topsMemRangeAttribute

typedef enum topsMemPoolAttr topsMemPoolAttr

typedef enum topsMemLocationType topsMemLocationType

typedef struct topsMemLocation topsMemLocation¶

Specifies a memory location.

To specify a gpu, set type = topsMemLocationTypeDevice and set id = the gpu’s device ID

typedef enum topsMemAccessFlags topsMemAccessFlags

typedef struct topsMemAccessDesc topsMemAccessDesc¶: Memory access descriptor

typedef enum topsMemAllocationType topsMemAllocationType

typedef enum topsMemAllocationHandleType topsMemAllocationHandleType

typedef struct topsMemPoolProps topsMemPoolProps¶: Specifies the properties of allocations made from the pool.

typedef struct topsMemPoolPtrExportData topsMemPoolPtrExportData¶: Opaque data structure for exporting a pool allocation

typedef enum topsJitOption topsJitOption

typedef enum topsFuncAttribute topsFuncAttribute: Warning

these hints and controls are ignored.

typedef enum topsFuncCache_t topsFuncCache_t: Warning

these hints and controls are ignored.

typedef enum topsSharedMemConfig topsSharedMemConfig: Warning

these hints and controls are ignored.

typedef struct dim3 dim3¶

typedef struct topsLaunchParams_t topsLaunchParams¶

typedef enum topsExternalMemoryHandleType_enum topsExternalMemoryHandleType¶

typedef struct topsExternalMemoryHandleDesc_st topsExternalMemoryHandleDesc¶

typedef struct topsExternalMemoryBufferDesc_st topsExternalMemoryBufferDesc¶

typedef void *topsExternalMemory_t¶

typedef enum topsExternalSemaphoreHandleType_enum topsExternalSemaphoreHandleType¶

typedef struct topsExternalSemaphoreHandleDesc_st topsExternalSemaphoreHandleDesc¶

typedef void *topsExternalSemaphore_t¶

typedef struct topsExternalSemaphoreSignalParams_st topsExternalSemaphoreSignalParams¶

typedef struct topsExternalSemaphoreWaitParams_st topsExternalSemaphoreWaitParams¶: External semaphore wait parameters, compatible with driver type

typedef enum topsGLDeviceList topsGLDeviceList

typedef struct _topsGraphicsResource topsGraphicsResource¶

typedef topsGraphicsResource *topsGraphicsResource_t¶

__TOPS_NODISCARD¶

topsStreamDefault¶: Default stream creation flags. These are used with topsStreamCreate().

topsStreamNonBlocking¶: Stream does not implicitly synchronize with null stream.

topsStreamCallbackBlocking¶

topsStreamNonMerging¶

topsEventDefault¶: Default flags.

topsEventBlockingSync¶: Waiting will yield CPU. Power-friendly and usage-friendly but may increase latency.

topsEventDisableTiming¶: Disable event’s capability to record timing information. May improve performance.

topsEventInterprocess¶: Event can support IPC. Warning: It is not supported in TOPS.

topsEventRecordOnce¶: Event can only support record once. NOTE: tops extern

topsEventStrongOrder¶: Event can only support strong order. NOTE: tops extern

topsEventDisableQueryAndSync¶: Disable event query/sync operation.

topsEventReleaseToDevice¶: Use a device-scope release when recording this event. This flag is useful to obtain more precise timings of commands between events.

topsEventReleaseToSystem¶: Use a system-scope release when recording this event. This flag is useful to make non-coherent host memory visible to the host.

topsHostMallocDefault¶: Default pinned memory allocation on the host.

topsHostMallocPortable¶: Memory is considered allocated by all contexts.

topsHostMallocMapped¶: Map the allocation into the address space for the current device. The device pointer can be obtained with topsHostGetDevicePointer.

topsHostMallocWriteCombined¶: Allocates the memory as write-combined. On some system configurations, write-combined allocation may be transferred faster across the PCI Express bus, however, could have low read efficiency by most CPUs. It’s a good option for data transfer from host to device via mapped pinned memory.

topsHostMallocNumaUser¶: Host memory allocation will follow numa policy set by user.

topsHostMallocCoherent¶: Allocate coherent memory. Overrides TOPS_COHERENT_HOST_ALLOC for specific allocation.

topsHostMallocNonCoherent¶: Allocate non-coherent memory. Overrides TOPS_COHERENT_HOST_ALLOC for specific allocation.

topsMemAttachGlobal¶: Memory can be accessed by any stream on any device

topsMemAttachHost¶: Memory cannot be accessed by any stream on any device.

topsMemAttachSingle¶: Memory can only be accessed by a single stream on the associated device.

topsDeviceMallocDefault¶

topsDeviceMallocFinegrained¶: Memory is allocated in fine grained region of device.

topsMallocSignalMemory¶: Memory represents a HSA signal.

topsMallocTopDown¶: Memory allocation from top to down.

topsMallocForbidMergeMove¶: Memory allocation with forbidden to do fragment.

topsMallocPreferHighSpeedMem¶: Memory allocation prefer to be allocated on high speed mem, if it can not be satisfied, then allocate on global device memory.

topsMallocHostAccessable¶: Memory allocation host accessable.

topsHostRegisterDefault¶: Memory is Mapped and Portable.

topsHostRegisterPortable¶: Memory is considered registered by all contexts.

topsHostRegisterMapped¶: Map the allocation into the address space for the current device. The device pointer can be obtained with topsHostGetDevicePointer.

topsHostRegisterIoMemory¶: Not supported.

topsExtHostRegisterCoarseGrained¶: Coarse Grained host memory lock.

topsDeviceScheduleAuto¶: Automatically select between Spin and Yield.

topsDeviceScheduleSpin¶: Dedicate a CPU core to spin-wait. Provides lowest latency, but burns a CPU core and may consume more power.

topsDeviceScheduleYield¶: Yield the CPU to the operating system when waiting. May increase latency, but lowers power and is friendlier to other threads in the system.

topsDeviceScheduleBlockingSync¶

topsDeviceScheduleMask¶

topsDeviceMapHost¶

topsDeviceLmemResizeToMax¶

topsArrayDefault¶: Default TOPS array allocation flag.

topsArrayLayered¶

topsArraySurfaceLoadStore¶

topsArrayCubemap¶

topsArrayTextureGather¶

topsOccupancyDefault¶

topsCooperativeLaunchMultiDeviceNoPreSync¶

topsCooperativeLaunchMultiDeviceNoPostSync¶

topsCpuDeviceId¶

topsInvalidDeviceId¶

topsExtAnyOrderLaunch¶: AnyOrderLaunch of kernels.

topsStreamWaitValueGte¶

topsStreamWaitValueEq¶

topsStreamWaitValueAnd¶

topsStreamWaitValueNor¶

topsStreamPerThread¶: Implicit stream per application thread.

struct itopsStream_t¶

struct itopsEvent_t¶

struct itopsExecutable_t¶

struct itopsResource_t¶

struct itopsCtx_t¶

struct itopsTensor_t¶

struct topsMemLocation

#include <tops_runtime_api.h>

Specifies a memory location.

To specify a gpu, set type = topsMemLocationTypeDevice and set id = the gpu’s device ID

struct topsMemAccessDesc: #include <tops_runtime_api.h>

Memory access descriptor

struct topsMemPoolProps: #include <tops_runtime_api.h>

Specifies the properties of allocations made from the pool.

struct topsMemPoolPtrExportData: #include <tops_runtime_api.h>

Opaque data structure for exporting a pool allocation

union uint3¶

#include <tops_runtime_api.h>

Struct for data in 3D

Public Members

uint32_t x¶

uint32_t y¶

uint32_t z¶

struct uint3::[anonymous] [anonymous]¶

uint32_t data[3]¶

struct dim3

struct topsLaunchParams_t¶

struct topsExternalMemoryHandleDesc_st¶

struct topsExternalMemoryBufferDesc_st¶

struct topsExternalSemaphoreHandleDesc_st¶

struct topsExternalSemaphoreSignalParams_st¶

struct topsExternalSemaphoreWaitParams_st¶: #include <tops_runtime_api.h>

External semaphore wait parameters, compatible with driver type

1.2. GlobalDefsExt¶

This section describes Ext Global enum and defines of TOPS runtime API.

enum topsResLocMode¶

Values:

enumerator TOPS_RES_LOC__SAME¶

enumerator TOPS_RES_LOC__DIST¶

enum topsHbmPolicy¶

Values:

enumerator TOPS_HBM_POLICY__BEST_FIT¶

enumerator TOPS_HBM_POLICY__AFFINITY_ONLY¶

enumerator TOPS_HBM_POLICY__NONE¶

enum topsResBundlMode¶

Values:

enumerator TOPS_RES_BUNDLE_MODE__UNKNOWN¶

enumerator TOPS_RES_BUNDLE_MODE__DEVICE¶

enumerator TOPS_RES_BUNDLE_MODE__CLUSTER¶

enumerator TOPS_RES_BUNDLE_MODE__VG¶

enumerator TOPS_RES_BUNDLE_MODE__PG¶

enumerator TOPS_RES_BUNDLE_MODE__USER¶

enum topsExecutableInfoType¶

topsExecutableInfoType types

Values:

enumerator topsExecutableInfoInputCount¶

enumerator topsExecutableInfoOutputCount¶

enumerator topsExecutableInfoInputSizeList¶

enumerator topsExecutableInfoOutputSizeList¶

enumerator topsExecutableInfoInputRank¶

enumerator topsExecutableInfoInputDimsList¶

enumerator topsExecutableInfoOutputRank¶

enumerator topsExecutableInfoOutputDimsList¶

enumerator topsExecutableInfoInputDataTypeList¶

enumerator topsExecutableInfoOutputDataTypeList¶

enumerator topsExecutableInfoTensorTableSize¶

enumerator topsExecutableInfoInputMinDimsList¶

enumerator topsExecutableInfoInputMaxDimsList¶

enumerator topsExecutableInfoOutputMaxDimsList¶

enumerator topsExecutableInfoBlockCount¶

enumerator topsExecutableInfoInputBankList¶

enumerator topsExecutableInfoOutputBankList¶

enumerator topsExecutableInfoThreadNumberPerBlock¶

enum topsResourceBundleInfoType¶

topsResourceBundleInfoType types

Values:

enumerator topsResourceBundleProcessorCount¶

enum class topsReduceOpType¶

Reduce Operation define.

Values:

enumerator REDUCE_SUM¶

enumerator REDUCE_PROD¶

enumerator REDUCE_MAX¶

enumerator REDUCE_MIN¶

enumerator REDUCE_UNKNOW¶

enum class topsReduceDataType¶

Reduce Data type define.

Values:

enumerator INT8¶

enumerator CHAR¶

enumerator UINT8¶

enumerator INT32¶

enumerator INT¶

enumerator UINT32¶

enumerator INT64¶

enumerator UINT64¶

enumerator FLOAT16¶

enumerator HALF¶

enumerator FLOAT32¶

enumerator FLOAT¶

enumerator LFLOAT64¶

enumerator DOUBLE¶

enumerator BFLOAT16¶

enumerator UNKNOW¶

enum topsExtExecutableSectionHeaderType¶

topsExecutable section header types

Values:

enumerator SHT_NULL_¶

enumerator SHT_HEADER¶

enumerator SHT_RESOURCE¶

enumerator SHT_INPUT¶

enumerator SHT_OUT¶

enumerator SHT_CONSTANT¶

enumerator SHT_SIPCODE¶

enumerator SHT_SIP_FUNC_PARAM¶

enumerator SHT_SIP_FUNC_PARAM_UPDATE¶

enumerator SHT_STREAM¶

enumerator SHT_PACKET¶

enumerator SHT_PACKET_UPDATE¶

enumerator SHT_CLUSTER¶

enumerator SHT_JITBINARY¶

enumerator SHT_CPU_FUNC_PARAM¶

enumerator SHT_CPU_FUNC_DATA¶

enumerator SHT_CPU_FUNCTION¶

enumerator SHT_PROFILE¶

enumerator SHT_HOST_CONST¶

enumerator SHT_TASK¶

enumerator SHT_SIP_CALLBACK¶

enumerator SHT_TARGET_RESOURCE¶

enumerator SHT_TENSOR_TABLE¶

enumerator SHT_KERNEL_ASSERT_INFO¶

enumerator SHT_RAND_STATE¶

enumerator SHT_HOST_PROGRAM¶

enumerator SHT_RODATA¶

enumerator SHT_MANAGED_CONSTANT¶

enumerator SHT_USER5¶

enumerator SHT_USER6¶

enumerator SHT_USER7¶

enumerator SHT_USER8¶

enumerator SHT_USER9¶

enumerator SHT_LAST_TYPE¶

typedef struct itopsExecutable_t *topsExecutable_t¶

typedef struct itopsResource_t *topsResource_t¶

typedef struct topsResourceBundle topsResourceBundle_t¶

typedef struct topsResourceRequest topsResourceRequest_t¶: topsResourceRequest descriptor

typedef enum topsExecutableInfoType topsExecutableInfoType_t¶: topsExecutableInfoType types

typedef enum topsResourceBundleInfoType topsResourceBundleInfoType_t¶: topsResourceBundleInfoType types

typedef struct topsKernelDescriptor topsKernelDescriptor_t¶

typedef struct topsExtLaunchParams topsExtLaunchParams_t¶: topsExtLaunchParams descriptor

typedef struct topsExtExecutableSectionInfo topsExtExecutableSectionInfo_t¶: topsExecutable section information

typedef enum topsExtExecutableSectionHeaderType topsExtExecutableSectionHeaderType_t¶: topsExecutable section header types

typedef struct topsResourceRequestV2_st topsResourceRequestV2_t¶

topsResourceRequestV2 descriptor