Convert

High-level convert

Converts the value from one data type to another data type. The source and destination operands can be vectors of 8/16/32 bit signed and unsigned integers, 32 bit float ,16 bit bfloat.

Scalar Convert

Represents CONVERT instruction in scalar slot.

Vector Convert

Represents CONVERT instruction in vector slot, in which the result vector element is of the same size as the source one.

Upconvert

Represents CONVERT instruction in vector slot in which result vector element is wider then the source one (up-convert).

Down Convert

Represents CONVERT instruction in vector slot in which result vector element is shorter then the source one (down-convert).

Convert using Single Lane

Represents CONVERT instruction in vector slot that make move 1xFP32 to 1xBF16/F16 (lane 0 only), which means the 64 values will be assigned into the 128 elements vector in every other lane. E.g. 0,2,4,6 … 126.

Convert using All Lanes

Represents CONVERT instruction in vector slot with ALL_LANES switch, converts all lanes and vector element size of source and destination is different.

Scalar Convert INT32

Represents scalar CONVERT_INT32 instruction.

Vector Convert INT32

Represents vector CONVERT_INT32 instruction.

Scalar Convert UINT32

Represents scalar CONVERT_UINT32 instruction.

Vector Convert UINT32

Represents vector CONVERT_UINT32 instruction.

Scalar Convert INT16

Represents scalar CONVERT_INT16 instruction.

Vector Convert INT16

Represents vector CONVERT_INT16 instruction.

Scalar Convert UINT16

Represents scalar CONVERT_UINT16 instruction.

Vector Convert UINT16

Represents vector CONVERT_UINT16 instruction.