FEXPA

Floating-point exponential accelerator

This instruction calculates an exponentiation acceleration operation on each floating-point element in the source vector, where the result sign is zero, the result exponent field is copied from a set of significant bits of the input fraction, and the result fraction is inserted from a lookup table indexed by the least-significant input fraction bits, and returns each result in the corresponding element of the destination vector.

This instruction is fully defined by its bit-manipulation semantics, does not generate floating-point exceptions, and does not guarantee NaN propagation.

For double-precision variants, the result element exponent is copied from the source element bits<16:6>, and the result fraction is set based on the source element to the rounded value of 2⁵² × (2^bits<5:0>/64 - 1).

For single-precision variants, the result element exponent is copied from the source element bits<13:6>, and the result fraction is set based on the source element to the rounded value of 2²³ × (2^bits<5:0>/64 - 1).

For half-precision variants, the result element exponent is copied from the source element bits<9:5>, and the result fraction is set based on the source element to the rounded value of 2¹⁰ × (2^bits<4:0>/32 - 1).

This instruction is unpredicated.

This instruction is illegal when executed in Streaming SVE mode, unless FEAT_SME_FA64 is implemented and enabled, or FEAT_SSVE_FEXPA is implemented.

SVE
(FEAT_SVE || FEAT_SSVE_FEXPA)

Decode for this encoding

if !IsFeatureImplemented(FEAT_SVE) && !IsFeatureImplemented(FEAT_SSVE_FEXPA) then EndOfDecode(Decode_UNDEF); end; if size == '00' then EndOfDecode(Decode_UNDEF); end; let esize : integer{} = 8 << UInt(size); let n : integer = UInt(Zn); let d : integer = UInt(Zd);