Reserved APIs

GetTid()

Obtains the current thread ID.

GetSafeStr(T name)

Safely obtains a string pointer. The supported types are const char*, char*, and std::string.

GetOpInfo(const std::string& str)

Obtains the operator information string.

GetOpInfo(const char* str)

Obtains the operator information string. Null pointer protection is supported.

IsContextType<T>()

Checks whether the template type T is Context.

GetOpInfo(T context)

Obtains the operator type and name from TilingContext or InferShapeContext.

DLOG_DEBUG

DEBUG log constant. The value is 0.

DLOG_INFO

INFO log constant. The value is 1.

DLOG_WARN

WARN log constant. The value is 2.

DLOG_ERROR

ERROR log constant. The value is 3.

OP_MODULE_ID

Operator module ID constant. The value is 63.

CheckLogLevel(int32_t moduleId, int32_t logLevel)

Log level check function, which checks whether the log level of a specified module is enabled. If 1 is returned, the log level is enabled.

DlogRecord(int32_t moduleId, int32_t level, const char* fmt, ...)

Logging function, which records logs by module ID and log level.

OP_LOGE_LIBOPAPI_REPORT(opName, fmt, ...)

Reports bottom-layer logs and records ERROR logs.

OP_LOGE_WITHOUT_REPORT(opName, ...)

Records logs only and does not report errors.

OpLogSub(moduleId, level, opInfo, fmt, ...)

Logging macro for a specified module and level.

OpLogErrSub(moduleId, level, opInfo, fmt, ...)

Error logging macro for a specified module and level.

D_OP_LOGI(opName, fmt, ...)

INFO log debugging macro.

D_OP_LOGW(opName, fmt, ...)

WARN log debugging macro.

D_OP_LOGE(opName, fmt, ...)

ERROR log debugging macro.

D_OP_LOGD(opName, fmt, ...)

DEBUG log debugging macro.

unlikely(x)

Compiler branch prediction optimization macro, which marks x as a low-probability branch.

likely(x)

Compiler branch prediction optimization macro, which marks x as a high-probability branch.

ToString(ge::DataType type)

Obtains the string corresponding to the data type.

ToString(ge::Format format)

Obtains the string corresponding to the format.

ToString(const gert::Shape& shape)

Obtains the string corresponding to the shape.

ToString(const std::vector<const gert::Shape*>& v)

Obtains the string corresponding to the shape pointer list.

ViewErrorCode

Error code enumeration, which is used to report errors in scenarios such as infershape.

LastPow2(uint64_t n)

Obtains the maximum power of 2 that does not exceed n.

MurmurHash(const void *src, uint32_t srcLen, uint32_t seed)

Murmur hash algorithm, which is used to calculate the hash value of data.

SplitResult

Structure of the average result, including splitCount, splitFactor, and splitTailFactor.

SplitIntoEqualByParts(int64_t splitLen, int32_t parts, SplitResult& splitResult)

Splits splitLen evenly into parts and returns the size of each part.

SplitIntoEqualByFactor(int64_t splitLen, int32_t factor, SplitResult& splitResult)

Splits splitLen evenly based on the specified factor and returns the number of parts.

GetValueToInteger<T1, T2>(const gert::Tensor *constTensor, T2 &value)

Reads integer data from a const Tensor.

GetConstInt<ContextType, T>(const ContextType *context, int64_t inputIdx, T &value)

Obtains the constant integer value of a specified input from InferShapeContext or TilingContext.

GetValueToShape<T>(const gert::Tensor *constTensor, gert::Shape &shape)

Reads an integer array from a const Tensor to Shape.

GetConstIntToShape<ContextType>(const ContextType *context, int64_t inputIdx, gert::Shape &shape)

Obtains the constant integer of a specified input from InferShapeContext or TilingContext and writes it to Shape.

IsRegbaseSocVersion(const gert::TilingParseContext* context)

Checks whether the current platform is a Regbase SoC version.

IsRegbaseSocVersion(const gert::TilingContext* context)

Checks whether the current platform is a Regbase SoC version.

EnsureNotScalar(const gert::Shape& inShape)

Ensure that the shape is not a scalar. If it is a scalar, a one-dimensional shape is returned.

AiCoreParams

AiCore parameter structure, including ubSize, numBlocks, aicNum, l1Size, l0aSize, l0bSize, and l0cSize.

CompileInfoCommon

Common structure of compilation information, including platform information such as aivNum, aicNum, ubSize, and l1Size.

TilingBaseClass(gert::TilingContext* context)

Constructor of the Tiling base class, which receives TilingContext.

DoTiling()

Entry of the Tiling execution framework, which calls each Tiling phase in sequence.

Reset(gert::TilingContext* context)

Resets the context member of TilingBaseClass.

IsCapable()

Pure virtual function, which determines whether the current class supports the tiling of the operator.

GetPlatformInfo()

Pure virtual function, which obtains platform information, such as the number of cores and the size of UB/L1/L0C resources.

GetShapeAttrsInfo()

Pure virtual function, which obtains the INPUT/OUTPUT/ATTR information.

DoOpTiling()

Pure virtual function, which calculates the tiling data.

DoLibApiTiling()

Pure virtual function, which calculates the tiling data of high-level APIs.

GetTilingKey()

Pure virtual function, which calculates the tiling key.

GetWorkspaceSize()

Pure virtual function, which calculates the workspace size.

PostTiling()

Pure virtual function, which saves the tiling data.

DumpTilingInfo()

Prints tiling data information.

DefaultTilingInfoDump()

Default implementation for printing tiling data.

CalcTschBlockDim(uint32_t sliceNum, uint32_t aicCoreNum, uint32_t aivCoreNum)

Calculates the BlockDim in the tsch scenario.

GetShapeDebugStr<T>(const T& shape)

Obtains the debugging string of the shape.

GetTensorDebugStr(const gert::StorageShape* shape, const gert::CompileTimeTensorDesc* tensor)

Obtains the debugging string of a tensor, including the dtype, shape, and format information.

GetTilingContextDebugStr()

Obtains the debugging strings of all input and output tensors in TilingContext.

GetTilingDataDebugStr()

Obtains the debugging string of TilingData.

bfloat16

bfloat16 data type structure, which supports conversion and arithmetic operations with types such as float and double.

fp16_t

fp16 half-precision floating-point data type structure, which supports conversion and arithmetic operations with types such as float and int.

castTraitB162B32

CastTrait constant configuration from B16 to B32.

castTraitB322B16

CastTrait constant configuration from B32 to B16.

castTraitB322Int32

CastTrait constant configuration from B32 to int32.

castTraitB322Int16

CastTrait constant configuration from B32 to int16.

castTraitB162Int8

CastTrait constant configuration from B16 to int8.

LoadOneTensorForDtypeT<T>(__local_mem__ T *input, MicroAPI::RegTensor<float> &dst, MicroAPI::MaskReg &preg, uint32_t offset)

Loads an aligned tensor from the UB to the float32 register.

LoadTwoTensorForDtypeT<T>(__local_mem__ T *src1, __local_mem__ T *src2, MicroAPI::RegTensor<float> &dst1, MicroAPI::RegTensor<float> &dst2, MicroAPI::MaskReg &dst1Preg, MicroAPI::MaskReg &dst2Preg, uint32_t src1Offset, uint32_t src2Offset)

Loads two aligned tensors from the UB to the float32 register.

StoreOneTensorForDtypeT<T>(__local_mem__ T *output, MicroAPI::RegTensor<float> &src, MicroAPI::MaskReg &preg, uint32_t offset)

Stores the float32 register data to the UB. The output can be of the bfloat16, float16, float32, int32, int16, int8, or uint8 type.

LoadUnAlignOneTensor<T>(__local_mem__ T *&input, MicroAPI::RegTensor<float> &dst, MicroAPI::UnalignReg &uSrc, MicroAPI::MaskReg &preg, uint32_t postUpdateStride)

Loads an unaligned tensor from the UB to the float32 register.

StoreUnAlignOneTensor<T>(__local_mem__ T *&output, MicroAPI::RegTensor<float> &src, MicroAPI::UnalignReg &uValue, MicroAPI::MaskReg &preg, uint32_t postUpdateStride)

Stores the float32 register data to the UB in an unaligned manner.

CeilDiv<T>(T a, T b)

Returns the quotient rounded up to the nearest integer on the kernel side.

FloorDiv<T>(T a, T b)

Returns the quotient rounded down to the nearest integer on the kernel side.

CeilAlign<T>(T a, T b)

Returns the value rounded up on the kernel side.

FloorAlign<T>(T a, T b)

Returns the value rounded down on the kernel side.

GetUbBlockSize()

Obtains the size of the UB block (32 bytes) on the kernel side.

GetVRegSize()

Obtains the size of the vector register on the kernel side.