Function tvm::codegen::intrin::DispatchFastErf

Function tvm::codegen::intrin::DispatchFastErf#

Function Documentation#

PrimExpr tvm::codegen::intrin::DispatchFastErf(const PrimExpr &e)#

Lower a unary float16/float32 erf call to a rational approximation. The expansion clamps the input to [-4, 4] and evaluates numerator and denominator polynomials using the input element width; other types are rejected.