FMOP4A (widening, 2-way, FP8 to FP16)

8-bit floating-point quarter-tile sum of outer products to half-precision, accumulating

This instruction generates four independent quarter-tile 8-bit floating-point sums of outer products from the sub-matrices in the half-vectors of the one or two first and second source vectors and accumulates the results to the corresponding elements of a 16-bit element ZA tile.

Each of the quarter-tile sums of outer products is generated by multiplying the SVL_H÷2 × 2 sub-matrix of 8-bit floating-point values held in the half-vectors of the first source vectors by the 2 × SVL_H÷2 sub-matrix of 8-bit floating-point values held in the half-vectors of the second source vectors. Each 16-bit container of the first source vectors holds 2 elements of each row of a SVL_H÷2 × 2 sub-matrix. Similarly, each 16-bit container of the second source vectors holds 2 elements of each column of a 2 × SVL_H÷2 sub-matrix.

This instruction widens the sub-matrices of 8-bit floating-point values held in the first source vectors to half-precision values and multiplies them by the corresponding widened sub-matrices of 8-bit floating-point values in the second source vectors to half-precision values. The resulting quarter-tile SVL_H÷2 × SVL_H÷2 half-precision sums of outer products are scaled by 2^{-UInt(FPMR.LSCALE[3:0])}, before being destructively added to the half-precision destination tile. This is equivalent to performing a downscaled 2-way dot product and accumulate to each of the destination tile elements.

The 8-bit floating-point encoding format for the elements of the first source vector and the second source vector is selected by FPMR.F8S1 and FPMR.F8S2 respectively.

This instruction is unpredicated.

It has encodings from 4 classes: Single and multiple vectors , Single vectors , Multiple and single vectors and Multiple vectors

Single and multiple vectors
(FEAT_SME_MOP4 && FEAT_SME_F8F16)

Decode for this encoding

if !IsFeatureImplemented(FEAT_SME_MOP4) || !IsFeatureImplemented(FEAT_SME_F8F16) then EndOfDecode(Decode_UNDEF); end; let n : integer = UInt('0'::Zn::'0'); let m : integer = UInt('1'::Zm::'0'); let nreg : integer{} = 1; let mreg : integer = 2; let da : integer = UInt(ZAda);

Single vectors
(FEAT_SME_MOP4 && FEAT_SME_F8F16)

Decode for this encoding

if !IsFeatureImplemented(FEAT_SME_MOP4) || !IsFeatureImplemented(FEAT_SME_F8F16) then EndOfDecode(Decode_UNDEF); end; let n : integer = UInt('0'::Zn::'0'); let m : integer = UInt('1'::Zm::'0'); let nreg : integer{} = 1; let mreg : integer = 1; let da : integer = UInt(ZAda);

Multiple and single vectors
(FEAT_SME_MOP4 && FEAT_SME_F8F16)

Decode for this encoding

if !IsFeatureImplemented(FEAT_SME_MOP4) || !IsFeatureImplemented(FEAT_SME_F8F16) then EndOfDecode(Decode_UNDEF); end; let n : integer = UInt('0'::Zn::'0'); let m : integer = UInt('1'::Zm::'0'); let nreg : integer{} = 2; let mreg : integer = 1; let da : integer = UInt(ZAda);

Multiple vectors
(FEAT_SME_MOP4 && FEAT_SME_F8F16)

Decode for this encoding

if !IsFeatureImplemented(FEAT_SME_MOP4) || !IsFeatureImplemented(FEAT_SME_F8F16) then EndOfDecode(Decode_UNDEF); end; let n : integer = UInt('0'::Zn::'0'); let m : integer = UInt('1'::Zm::'0'); let nreg : integer{} = 2; let mreg : integer = 2; let da : integer = UInt(ZAda);