Closed chhzh123 closed 2 years ago
Current MLIR has two issues that prevent us from supporting PowOp in a clean way:
math
math.powf
llvm
powi
The solution is to cast those input operands from integers to floating points and then cast them back from floating points to integers after the calculation.
Current MLIR has two issues that prevent us from supporting PowOp in a clean way:
math
dialect does not inherently support the power of integers. There is only amath.powf
operation. See the discussion here.llvm
dialect has apowi
operation but it does not provide Python binding.