FTMOPA (non-widening)

Floating-point sparse outer product, accumulating

This instruction generates floating-point outer product by multiplying the 1-in-2 selected elements from the dense sub-matrices in the two first source vectors with the corresponding elements of the compressed sparse sub-matrix in the second source vector and accumulates the results to the destination ZA tile elements.

In case of the half-precision variant, the outer product is generated by multiplying the selected 1-in-2 half-precision value from each overlapping 16-bit containers of the two SVL_H×1 sub-matrices in the first source vectors by the half-precision value from the corresponding 16-bit container of the compressed 1×SVL_H sub-matrix in the second source vector. In case of the single-precision variant, the outer product is generated by multiplying the selected 1-in-2 single-precision value from each overlapping 32-bit containers of the two SVL_S×1 sub-matrices in the first source vectors by the single-precision value from the corresponding 32-bit container of the compressed 1×SVL_S sub-matrix in the second source vector.

The 1-in-2 floating-point value in the first source vectors is selected by 2-bit controls in the indexed segment of the control vector register. If the control bit corresponding to an element in the first source vectors is 0, the element is discarded and does not contribute to the result. If both bits of the 2-bit control corresponding to the elements of the first source vectors are 1, only the element corresponding to the least significant bit is selected.

The instruction multiplies the selected elements of sub-matrices of floating-point values held in the first source vectors by the corresponding elements of sub-matrix of floating-point values in the second source vector. The resulting outer product, SVL_H×SVL_H in case of the half-precision variant or SVL_S×SVL_S in case of the single-precision variant, is then destructively added to the destination tile. This is equivalent to performing a single multiply-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 2 classes: Half-precision and Single-precision

Half-precision
(FEAT_SME_TMOP && FEAT_SME_F16F16)

Decode for this encoding

if !IsFeatureImplemented(FEAT_SME_TMOP) || !IsFeatureImplemented(FEAT_SME_F16F16) then EndOfDecode(Decode_UNDEF); constant integer n = UInt(Zn:'0'); constant integer m = UInt(Zm); constant integer k = UInt('1':K:'1':Zk); constant integer index = UInt(i2); constant integer da = UInt(ZAda); constant integer esize = 16;

Single-precision
(FEAT_SME_TMOP)

Decode for this encoding

if !IsFeatureImplemented(FEAT_SME_TMOP) then EndOfDecode(Decode_UNDEF); constant integer n = UInt(Zn:'0'); constant integer m = UInt(Zm); constant integer k = UInt('1':K:'1':Zk); constant integer index = UInt(i2); constant integer da = UInt(ZAda); constant integer esize = 32;