GetSubBlockIdx

Applicability

Function Usage

Obtains the ID of the current AIV. In the separated architecture, an AI Core consists of Cube Cores (AICs) and Vector Cores (AIVs) in a fixed ratio of 1:N. The IDs of the N AIVs are 0, 1, ..., and N-1, respectively.

The difference between Matmul::GetSubBlockIdx() and the basic API AscendC::GetSubBlockIdx() is that Matmul::GetSubBlockIdx() obtains the ID of the current AIV in the current AI Core group, while AscendC::GetSubBlockIdx() obtains the logical ID of the AIV in all AI Core groups. For example, there are 10 AI Core groups, and the ratio of AICs to AIVs is 1:2. There are 20 AIVs in total. When Matmul::GetSubBlockIdx() is called, the obtained IDs of the 20 AIVs are 0, 1, 0, 1, 0, 1, ..., 0, and 1 in sequence. When AscendC::GetSubBlockIdx() is called, the obtained IDs of the 20 AIVs are 0, 1, 2, 3, 4, 5, ..., 18, and 19 in sequence.

Prototype

1

__aicore__  inline uint8_t GetSubBlockIdx()

Parameters

None

Returns

ID of the current AIV.

Restrictions

• This API can be used only in the separated architecture. Otherwise, a random value is returned.
• In the separated architecture, the AIV ID is automatically initialized and assigned within the REGIST_MATMUL_OBJ() API. Therefore, you need to call this API after calling REGIST_MATMUL_OBJ() to obtain the correct ID.
• If SetSubBlockIdx() is called in the operator program, the GetSubBlockIdx() API will return the ID set by the SetSubBlockIdx API.

Example

1
2
3
4
5
6
7
8

typedef AscendC::MatmulType<AscendC::TPosition::GM, CubeFormat::ND, half> aType; 
typedef AscendC::MatmulType<AscendC::TPosition::GM, CubeFormat::ND, half> bType; 
typedef AscendC::MatmulType<AscendC::TPosition::GM, CubeFormat::ND, float> cType; 
typedef AscendC::MatmulType<AscendC::TPosition::GM, CubeFormat::ND, float> biasType; 

AscendC::Matmul<aType, bType, cType, biasType, CFG_NORM> mm;
REGIST_MATMUL_OBJ(&pipe, GetSysWorkSpacePtr(), mm, &tiling);
mm.GetSubBlockIdx(); // Obtain the subcore ID.