Floating-point exponential accelerator
The FEXPA instruction computes 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 × (2bits<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 × (2bits<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 × (2bits<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_SME2p2 is implemented.
For a double-precision input value x in the range 70,368,744,177,655 <= x < 70,368,744,179,711, the operation performed by this instruction is equivalent to computing 2x-70,368,744,178,687. For a single-precision input value x in the range 131,073 <= x < 131,327, the operation performed by this instruction is equivalent to computing 2x-131,199. For a half-precision input value x in the range 33 <= x < 63, the operation performed by this instruction is equivalent to computing 2x-47.
Variants: FEAT_SVE || FEAT_SSVE_FEXPA (FEAT_SVE || FEAT_SSVE_FEXPA)
| 31 | 30 | 29 | 28 | 27 | 26 | 25 | 24 | 23 | 22 | 21 | 20 | 19 | 18 | 17 | 16 | 15 | 14 | 13 | 12 | 11 | 10 | 9 | 8 | 7 | 6 | 5 | 4 | 3 | 2 | 1 | 0 |
| 0 | 0 | 0 | 0 | 0 | 1 | 0 | 0 | 1 | 0 | 0 | 0 | 0 | 0 | 1 | 0 | 1 | 1 | 1 | 0 | ||||||||||||
| size | opc | Zn | Zd | ||||||||||||||||||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
if !IsFeatureImplemented(FEAT_SVE) && !IsFeatureImplemented(FEAT_SSVE_FEXPA) then
EndOfDecode(Decode_UNDEF);
if size == '00' then EndOfDecode(Decode_UNDEF);
constant integer esize = 8 << UInt(size);
constant integer n = UInt(Zn);
constant integer d = UInt(Zd);if IsFeatureImplemented(FEAT_SSVE_FEXPA) then
CheckSVEEnabled();
else
CheckNonStreamingSVEEnabled();
constant integer VL = CurrentVL;
constant integer elements = VL DIV esize;
constant bits(VL) operand = Z[n, VL];
bits(VL) result;
for e = 0 to elements-1
constant bits(esize) element = Elem[operand, e, esize];
Elem[result, e, esize] = FPExpA(element);
Z[d, VL] = result;