[object Object][object Object][object Object]undefined
[object Object]

Computes the difference element-wise, then performs ReLU computation (compares the result with 0 and takes the larger value), and provides the ability to convert the data type of the final result (s162s8, f322f16, f162s8). The computation formula is as follows:

dsti=Relu(src0isrc1i)dst_i = Relu(src0_i -src1_i) [object Object]
  • First-n data computation

    [object Object]
  • High-dimensional tiling computation

    [object Object]
  • Synchronous computation

    [object Object]
[object Object][object Object]undefined
[object Object]

None

[object Object]

PIPE_V

[object Object]
  • The starting addresses of dst, src0, and src1 must be 32-byte aligned.
  • For operand address overlap constraints, refer to .
[object Object]
[object Object]
[object Object]