BF1CVTL, BF2CVTL

Multi-vector 8-bit floating-point convert to deinterleaved BFloat16

This instruction converts each 8-bit floating-point element of the source vector to BFloat16 while downscaling the value, and places the two-way deinterleaved results in the corresponding 16-bit elements of the destination vectors. BF1CVTL scales the values by 2-UInt(FPMR.LSCALE[5:0]). BF2CVTL scales the values by 2-UInt(FPMR.LSCALE2[5:0]). The 8-bit floating-point encoding format for BF1CVTL is selected by FPMR.F8S1. The 8-bit floating-point encoding format for BF2CVTL is selected by FPMR.F8S2.

This instruction is unpredicated.

It has encodings from 2 classes: BF1CVTL and BF2CVTL

BF1CVTL
(FEAT_SME2 && FEAT_FP8)

313029282726252423222120191817161514131211109876543210
1100000101100110111000ZnZd1
opcL

Encoding

BF1CVTL { <Zd1>.H-<Zd2>.H }, <Zn>.B

Decode for this encoding

if !IsFeatureImplemented(FEAT_SME2) || !IsFeatureImplemented(FEAT_FP8) then EndOfDecode(Decode_UNDEF); end; let n : integer = UInt(Zn); let d : integer = UInt(Zd:: '0'); let issrc2 : boolean = FALSE;

BF2CVTL
(FEAT_SME2 && FEAT_FP8)

313029282726252423222120191817161514131211109876543210
1100000111100110111000ZnZd1
opcL

Encoding

BF2CVTL { <Zd1>.H-<Zd2>.H }, <Zn>.B

Decode for this encoding

if !IsFeatureImplemented(FEAT_SME2) || !IsFeatureImplemented(FEAT_FP8) then EndOfDecode(Decode_UNDEF); end; let n : integer = UInt(Zn); let d : integer = UInt(Zd:: '0'); let issrc2 : boolean = TRUE;

Assembler Symbols

<Zd1>

Is the name of the first scalable vector register of the destination multi-vector group, encoded as "Zd" times 2.

<Zd2>

Is the name of the second scalable vector register of the destination multi-vector group, encoded as "Zd" times 2 plus 1.

<Zn>

Is the name of the source scalable vector register, encoded in the "Zn" field.

Operation

CheckFPMREnabled(); CheckStreamingSVEEnabled(); let VL : integer{} = CurrentVL(); let pairs : integer = VL DIV 16; let operand : bits(VL) = Z{}(n); var result1 : bits(VL); var result2 : bits(VL); for p = 0 to pairs-1 do let element1 : bits(8) = operand[(2*p + 0)*:8]; let element2 : bits(8) = operand[(2*p + 1)*:8]; result1[p*:16] = FP8ConvertBF(element1, issrc2, FPCR(), FPMR()); result2[p*:16] = FP8ConvertBF(element2, issrc2, FPCR(), FPMR()); end; Z{VL}(d+0) = result1; Z{VL}(d+1) = result2;


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.