BFloat16 outer product, accumulating
This instruction works with a 16-bit element ZA tile.
This instruction generates an outer product of the first source vector and the second source vector. The first source is SVLH×1 vector and the second source is 1×SVLH vector.
Each source vector is independently predicated by a corresponding governing predicate. When either source vector element is Inactive the corresponding destination tile element remains unmodified.
The resulting outer product, SVLH×SVLH, 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 SME2 ZA-targeting non-widening BFloat16 numerical behaviors.
ID_AA64SMFR0_EL1.B16B16 indicates whether this instruction is implemented.
| 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 |
| 1 | 0 | 0 | 0 | 0 | 0 | 0 | 1 | 1 | 0 | 1 | Zm | Pm | Pn | Zn | 0 | 1 | 0 | 0 | ZAda | ||||||||||||
| S | |||||||||||||||||||||||||||||||
if !IsFeatureImplemented(FEAT_SME_B16B16) then EndOfDecode(Decode_UNDEF); end; let a : integer = UInt(Pn); let b : integer = UInt(Pm); let n : integer = UInt(Zn); let m : integer = UInt(Zm); let da : integer = UInt(ZAda);
| <ZAda> |
Is the name of the ZA tile ZA0-ZA1, encoded in the "ZAda" field. |
| <Pn> |
Is the name of the first governing scalable predicate register P0-P7, encoded in the "Pn" field. |
| <Pm> |
Is the name of the second governing scalable predicate register P0-P7, encoded in the "Pm" field. |
| <Zn> |
Is the name of the first source scalable vector register, encoded in the "Zn" field. |
| <Zm> |
Is the name of the second source scalable vector register, encoded in the "Zm" field. |
CheckStreamingSVEAndZAEnabled(); let VL : integer{} = CurrentVL(); let PL : integer{} = VL DIV 8; let dim : integer{} = VL DIV 16; let mask1 : bits(PL) = P{}(a); let mask2 : bits(PL) = P{}(b); let op1 : bits(VL) = Z{}(n); let op2 : bits(VL) = Z{}(m); let op3 : bits(dim*dim*16) = ZAtile{}(da, 16); var result : bits(dim*dim*16); for row = 0 to dim-1 do for col = 0 to dim-1 do let elem2 : bits(16) = op2[col*:16]; let elem3 : bits(16) = op3[(row*dim+col)*:16]; if ActivePredicateElement{PL}(mask1, row, 16) && ActivePredicateElement{PL}(mask2, col, 16) then let elem1 : bits(16) = op1[row*:16]; result[(row*dim+col)*:16] = BFMulAdd_ZA(elem3, elem1, elem2, FPCR()); else result[(row*dim+col)*:16] = elem3; end; end; end; ZAtile{dim*dim*16}(da, 16) = result;
2026-03_rel 2026-03-26 20:48:11
Copyright © 2010-2026 Arm Limited or its affiliates. All rights reserved. This document is Non-Confidential.