BFloat16 multiply-add by indexed element to single-precision (bottom)
This instruction widens the even-numbered BFloat16 elements in the first source vector and the indexed element from the corresponding 128-bit segment in the second source vector to single-precision format and then destructively multiplies and adds these values without intermediate rounding to the single-precision elements of the destination vector that overlap with the corresponding BFloat16 elements in the first source vector. This instruction is unpredicated.
ID_AA64ZFR0_EL1.BF16 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 |
| 0 | 1 | 1 | 0 | 0 | 1 | 0 | 0 | 1 | 1 | 1 | i3h | Zm | 0 | 1 | 0 | 0 | i3l | 0 | Zn | Zda | |||||||||||
| o2 | op | T | |||||||||||||||||||||||||||||
if ((!IsFeatureImplemented(FEAT_SVE) && !IsFeatureImplemented(FEAT_SME)) || !IsFeatureImplemented(FEAT_BF16)) then EndOfDecode(Decode_UNDEF); end; let n : integer = UInt(Zn); let m : integer = UInt(Zm); let da : integer = UInt(Zda); let index : integer = UInt(i3h::i3l); let op1_neg : boolean = FALSE;
| <Zda> |
Is the name of the third source and destination scalable vector register, encoded in the "Zda" 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 Z0-Z7, encoded in the "Zm" field. |
| <imm> |
Is the immediate index, in the range 0 to 7, encoded in the "i3h:i3l" fields. |
CheckSVEEnabled(); let VL : integer{} = CurrentVL(); let elements : integer = VL DIV 32; let eltspersegment : integer = 128 DIV 32; let op1 : bits(VL) = Z{}(n); let op2 : bits(VL) = Z{}(m); let op3 : bits(VL) = Z{}(da); var result : bits(VL); for e = 0 to elements-1 do let segmentbase : integer = e - (e MOD eltspersegment); let s : integer = 2 * segmentbase + index; let elem1 : bits(16) = (if op1_neg then BFNeg(op1[(2*e + 0)*:16]) else op1[(2*e + 0)*:16]); let elem2 : bits(16) = op2[s*:16]; let elem3 : bits(32) = op3[e*:32]; result[e*:32] = BFMulAddH(elem3, elem1, elem2, FPCR()); end; Z{VL}(da) = result;
This instruction might be immediately preceded in program order by a MOVPRFX instruction. The MOVPRFX must conform to all of the following requirements, otherwise the behavior of the MOVPRFX and this instruction is CONSTRAINED UNPREDICTABLE:
2025-12_rel 2025-12-10 16:41:45
Copyright © 2010-2025 Arm Limited or its affiliates. All rights reserved. This document is Non-Confidential.