FMOP4A (widening, 2-way, FP16 to FP32)

Half-precision quarter-tile sum of outer products to single-precision, accumulating

This instruction generates four independent quarter-tile half-precision 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 32-bit element ZA tile.

Each of the quarter-tile sums of outer products is generated by multiplying the SVL_S÷2 × 2 sub-matrix of half-precision values held in the half-vectors of the first source vectors by the 2 × SVL_S÷2 sub-matrix of half-precision values held in the half-vectors of the second source vectors. Each 32-bit container of the first source vectors holds 2 elements of each row of a SVL_S÷2 × 2 sub-matrix. Similarly, each 32-bit container of the second source vectors holds 2 elements of each column of a 2 × SVL_S÷2 sub-matrix.

The instruction widens the sub-matrices of half-precision values held in the first source vectors to single-precision values and multiplies them by the corresponding widened sub-matrices of half-precision values in the second source vectors to single-precision values. The resulting quarter-tile SVL_S÷2 × SVL_S÷2 single-precision sums of outer products are then destructively added to the single-precision destination tile. This is equivalent to performing a 2-way dot product and accumulate to each of the destination tile elements.

This instruction follows SME ZA-targeting floating-point behaviors.

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)

Decode for this encoding

if !IsFeatureImplemented(FEAT_SME_MOP4) 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); let sub_op : boolean = FALSE;

Single vectors
(FEAT_SME_MOP4)

Decode for this encoding

if !IsFeatureImplemented(FEAT_SME_MOP4) 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); let sub_op : boolean = FALSE;

Multiple and single vectors
(FEAT_SME_MOP4)

Decode for this encoding

if !IsFeatureImplemented(FEAT_SME_MOP4) 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); let sub_op : boolean = FALSE;

Multiple vectors
(FEAT_SME_MOP4)

Decode for this encoding

if !IsFeatureImplemented(FEAT_SME_MOP4) 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); let sub_op : boolean = FALSE;