The AI Engine scalar unit supports signed and unsigned integers in 8, 16, and 32bit widths, along with some singleprecision floatingpoint for specific operations.
The AI Engine vector unit supports integers and complex integers in 8, 16, and 32bit widths, along with real and complex singleprecision floatingpoint numbers. It also supports accumulator vector data types, with 48 and 80bit wide elements. Intrinsic functions such as absolute value, addition, subtraction, comparison, multiplication, and MAC operate using these vector data types. Vector data types are named using a convention that includes the number of elements, real or complex, vector type or accumulator type, and bit width as follows:
v{NumLanes}[c]{[u]intfloatacc}{SizeofElement}
Optional specifications include:

NumLanes
 Denotes the number of elements in the vector which can be 2, 4, 8, 16, 32, 64, or 128.

c
 Denotes complex data with real and imaginary parts packed together.

int
 denotes integer vector data values.

float
 Denotes single precision floating point values.Note: There are no accumulator registers for floatingpoint vectors.

acc
 Denotes accumulator vector data values.

u
 Denotes unsigned. Unsigned only exists for int8 vectors.

SizeofElement
 Denotes the size of the vector data type element.
 1024bit integer vector types are vectors of 8bit, 16bit, or 32bit vector elements. These vectors have 16, 32, 64, or 128 lanes.
 512bit integer vector types are vectors of 8bit, 16bit, 32bit, or 64bit vector elements. These vectors have 4, 8, 16, 32, or 64 lanes.
 256bit integer vector types are vectors of 8bit, 16bit, 32bit, 64bit, or 128bit vector elements. These vectors have 1, 2, 4, 8, 16, or 32 lanes.
 128bit integer vector types are vectors of 8bit, 16bit, or 32bit vector elements. These vectors have 2, 4, 8, or 16 lanes.
 Accumulator data types are vectors of 80bit or 48bit elements These vectors have 2, 4, 8, or 16 lanes.
The total datawidth of the vector datatypes can be 128bit, 256bit, 512bit, or 1024bit. The total datawidth of the accumulator datatypes can be 320/384bit or 640/768bit.
For example, v16int32 is a sixteen element vector of integers with 32 bits. Each element of the vector is referred to as a lane. Using the smallest bit width necessary can improve performance by making good use of registers.