SetBias

Applicability

Product	Supported
Atlas 350 Accelerator Card	√
Atlas A3 training product / Atlas A3 inference product	√
Atlas A2 training product / Atlas A2 inference product	√
Atlas 200I/500 A2 inference product	√
Atlas inference product AI Core	√
Atlas inference product Vector Core	x
Atlas training product	x

Function Usage

Sets the bias for matrix multiplication.

Prototype

      
           __aicore__ inline void SetBias(const GlobalTensor<BiasT>& biasGlobal)

      
           __aicore__ inline void SetBias(const LocalTensor<BiasT>& inputBias)

Parameters

**Table 1** Parameters
Parameter	Input/Output	Description
biasGlobal	Input	Bias. The type is GlobalTensor. For the Atlas 350 Accelerator Card, the supported data types are half, bfloat16_t, float, and int32_t. The data type of the bias can be set to int32_t only when the data type of matrices A and B is int8_t. For the Atlas A3 training product / Atlas A3 inference product , the supported data types are half, float, and int32_t. The data type of the bias can be set to int32_t only when the data type of matrices A and B is int8_t. For the Atlas inference product AI Core, the supported data types are half, float, and int32_t. The data type of the bias can be set to int32_t only when the data type of matrices A and B is int8_t. For the Atlas A2 training product / Atlas A2 inference product , the supported data types are half, float, and int32_t. The data type of the bias can be set to int32_t only when the data type of matrices A and B is int8_t. For the Atlas 200I/500 A2 inference product , the supported data types are half, float, and int32_t. The data type of the bias can be set to int32_t only when the data type of matrices A and B is int8_t. For details about the data type combinations supported by matrices A and B and the bias, see Combinations of Matmul input and output data types. For details about the data type combinations supported by matrices A and B and the bias in the MxMatmul scenario, see MatmulTypeWithScale parameters.
inputBias	Input	Bias. The type is LocalTensor, and TPosition can be TSCM or VECOUT. For the Atlas 350 Accelerator Card, the supported data types are half, bfloat16_t, float, and int32_t. The data type of the bias can be set to int32_t only when the data type of matrices A and B is int8_t. For the Atlas A3 training product / Atlas A3 inference product , the supported data types are half, float, and int32_t. The data type of the bias can be set to int32_t only when the data type of matrices A and B is int8_t. For the Atlas inference product AI Core, the supported data types are half, float, and int32_t. For the Atlas A2 training product / Atlas A2 inference product , the supported data types are half, float, and int32_t. The data type of the bias can be set to int32_t only when the data type of matrices A and B is int8_t. For the Atlas 200I/500 A2 inference product , the supported data types are half, float, and int32_t. The data type of the bias can be set to int32_t only when the data type of matrices A and B is int8_t. For details about the data type combinations supported by matrices A and B and the bias, see Combinations of Matmul input and output data types. For details about the data type combinations supported by matrices A and B and the bias in the MxMatmul scenario, see MatmulTypeWithScale parameters.

Returns

None

Restrictions

In Matmul tiling computation, the isBias parameter in the TCubeTiling structure must be set to 1 (that is, the bias is enabled) before this API is called to set the bias.
Ensure that the size of the input bias address space is greater than or equal to singleN.
For the Atlas 350 Accelerator Card, if the logical memory location of the bias is TSCM and the data type is float or int32_t, the LocalTensor space of the bias must be 64-byte aligned.

Example

      
           REGIST_MATMUL_OBJ(&pipe, GetSysWorkSpacePtr(), mm, &tiling);
mm.SetTensorA(gm_a);
mm.SetTensorB(gm_b);
if (tiling.isBias) {
    mm.SetBias(gmBias);  // Set the bias.
}
mm.IterateAll(gm_c);
mm.End();

Parent topic: Matmul Kernel APIs