Function tvm::codegen::intrin::DispatchNumericallyStableTanh

Contents

Function tvm::codegen::intrin::DispatchNumericallyStableTanh#

Defined in File intrin_rule.h

Function Documentation#

PrimExpr tvm::codegen::intrin::DispatchNumericallyStableTanh(const PrimExpr &e)#: Lower a unary floating-point tanh call using sign-dependent exponential formulas. The expansion avoids the large-positive-input overflow of the symmetric formula; accuracy follows the floating-point arithmetic and exponential lowering in use.