Brcb

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	x
Atlas inference product AI Core	√
Atlas inference product Vector Core	x
Atlas training product	x

Function Usage

Extracts eight elements from a given input tensor each time and fills them in eight data blocks (32 bytes) in the result tensor. Each element corresponds to a data block.

Prototype

template <typename T>
__aicore__ inline void Brcb(const LocalTensor<T>& dst, const LocalTensor<T>& src0, const uint8_t repeatTime, const BrcbRepeatParams& repeatParams)

Parameters

**Table 1** Parameters in the template
Parameter	Description
T	Operand data type. For the Atlas 350 Accelerator Card, the supported data types are int8_t, uint8_t, int16_t, uint16_t, int32_t, uint32_t, half, float, bfloat16_t, uint64_t, and int64_t. For the Atlas A3 training product/Atlas A3 inference product, the supported data types are int16_t, uint16_t, int32_t, uint32_t, half, bfloat16_t, and float. For the Atlas A2 training product/Atlas A2 inference product, the supported data types are int16_t, uint16_t, int32_t, uint32_t, half, bfloat16_t, and float. For the Atlas inference product AI Core, the supported data types are int16_t, uint16_t, int32_t, uint32_t, half, and float.

**Table 2** Parameters
Parameter	Input/Output	Meaning
dst	Output	Destination operand. The type is LocalTensor, and TPosition can be VECIN, VECCALC, or VECOUT. The start address of LocalTensor must be 32-byte aligned.
src0	Input	Source operand. The type is LocalTensor, and TPosition can be VECIN, VECCALC, or VECOUT. The start address of LocalTensor must be 32-byte aligned. Its data type must match that of dst. Eight elements are read from src0 in each iteration. Therefore, the total number of elements in src0 must be no less than repeatTime multiplied by 8.
repeatTime	Input	Number of instruction repeats. Eight data blocks are collected in each repeat. The value range is [0, 255]
repeatParams	Input	Instruction iteration parameters. The type is BrcbRepeatParams. For details about the definition, see ${INSTALL_DIR}/include/ascendc/basic_api/interface/kernel_struct_brcb.h. Replace *${INSTALL_DIR}* with the CANN installation path. dstBlkStride and dstRepStride can be set by users. For details about the parameters, see Table 3.

**Table 3** Parameters in the **BrcbRepeatParams** structure
Parameter	Meaning
dstBlkStride	Address stride of the vector destination operand between different data blocks in a single repeat. Note: When dstBlkStride is set to 0, the value 1 is used by default.
dstRepStride	Address stride of the vector destination operand for the same data block between adjacent repeats
blockNumber	Reserved parameter. This parameter is reserved for future functions. You can use the default value.
src0BlkStride
src1BlkStride
src0RepStride
src1RepStride
repeatStrideMode
strideSizeMode

Returns

None

Restrictions

For details about the operand address alignment requirements, see General Address Alignment Restrictions.

Overlapping memory between src0 and dst is not supported.
For the Atlas inference product AI Core, reserve an 8 KB Unified Buffer to serve as a temporary data storage area for the API.

Examples

// repeatTime = 4, 128 elements one repeat, 512 elements total
// The data type of srcLocal is half, and the data type of dstLocal is half.
// dstBlkStride = 1, no gap between blocks in one repeat
// dstRepStride = 8, no gap between repeats 
AscendC::Brcb(dstLocal, srcLocal, 4, {1,8});

Result example:

Input (srcLocal):
[1 2 3 ... 16]
Initial value of output data (dstLocal):
[0. 0. 0. 0. 0. 0. ... 0.]
Output (dstLocal) after Brcb computation:
[1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 ... 15 15 15 15 15 15 15 15 15 15 15 15 15 15 15 15 16 16 16 16 16 16 16 16 16 16 16 16 16 16 16 16]

Parent topic: Data Padding