FCVT

Floating-point convert (predicated)

This instruction converts the size and precision of each active floating-point element of the source vector, and places the results in the corresponding elements of the destination vector. Inactive elements in the destination vector register remain unmodified or are set to zero, depending on whether merging or zeroing predication is selected.

Since the input and result types have a different size the smaller type is held unpacked in the least significant bits of elements of the larger size. When the input is the smaller type the upper bits of each source element are ignored. When the result is the smaller type the results are zero-extended to fill each destination element.

It has encodings from 12 classes: Half-precision to single-precision, merging , Half-precision to single-precision, zeroing , Half-precision to double-precision, merging , Half-precision to double-precision, zeroing , Single-precision to half-precision, merging , Single-precision to half-precision, zeroing , Single-precision to double-precision, merging , Single-precision to double-precision, zeroing , Double-precision to half-precision, merging , Double-precision to half-precision, zeroing , Double-precision to single-precision, merging and Double-precision to single-precision, zeroing

Half-precision to single-precision, merging
(FEAT_SVE || FEAT_SME)

313029282726252423222120191817161514131211109876543210
0110010110001001101PgZnZd
opcopc2

Encoding

FCVT <Zd>.S, <Pg>/M, <Zn>.H

Decode for this encoding

if !IsFeatureImplemented(FEAT_SVE) && !IsFeatureImplemented(FEAT_SME) then EndOfDecode(Decode_UNDEF); end; let esize : integer{} = 32; let g : integer = UInt(Pg); let n : integer = UInt(Zn); let d : integer = UInt(Zd); let s_esize : integer{} = 16; let d_esize : integer{} = 32; let merging : boolean = TRUE;

Half-precision to single-precision, zeroing
(FEAT_SVE2p2 || FEAT_SME2p2)

313029282726252423222120191817161514131211109876543210
0110010010011010101PgZnZd
opcopc2

Encoding

FCVT <Zd>.S, <Pg>/Z, <Zn>.H

Decode for this encoding

if !IsFeatureImplemented(FEAT_SVE2p2) && !IsFeatureImplemented(FEAT_SME2p2) then EndOfDecode(Decode_UNDEF); end; let esize : integer{} = 32; let g : integer = UInt(Pg); let n : integer = UInt(Zn); let d : integer = UInt(Zd); let s_esize : integer{} = 16; let d_esize : integer{} = 32; let merging : boolean = FALSE;

Half-precision to double-precision, merging
(FEAT_SVE || FEAT_SME)

313029282726252423222120191817161514131211109876543210
0110010111001001101PgZnZd
opcopc2

Encoding

FCVT <Zd>.D, <Pg>/M, <Zn>.H

Decode for this encoding

if !IsFeatureImplemented(FEAT_SVE) && !IsFeatureImplemented(FEAT_SME) then EndOfDecode(Decode_UNDEF); end; let esize : integer{} = 64; let g : integer = UInt(Pg); let n : integer = UInt(Zn); let d : integer = UInt(Zd); let s_esize : integer{} = 16; let d_esize : integer{} = 64; let merging : boolean = TRUE;

Half-precision to double-precision, zeroing
(FEAT_SVE2p2 || FEAT_SME2p2)

313029282726252423222120191817161514131211109876543210
0110010011011010101PgZnZd
opcopc2

Encoding

FCVT <Zd>.D, <Pg>/Z, <Zn>.H

Decode for this encoding

if !IsFeatureImplemented(FEAT_SVE2p2) && !IsFeatureImplemented(FEAT_SME2p2) then EndOfDecode(Decode_UNDEF); end; let esize : integer{} = 64; let g : integer = UInt(Pg); let n : integer = UInt(Zn); let d : integer = UInt(Zd); let s_esize : integer{} = 16; let d_esize : integer{} = 64; let merging : boolean = FALSE;

Single-precision to half-precision, merging
(FEAT_SVE || FEAT_SME)

313029282726252423222120191817161514131211109876543210
0110010110001000101PgZnZd
opcopc2

Encoding

FCVT <Zd>.H, <Pg>/M, <Zn>.S

Decode for this encoding

if !IsFeatureImplemented(FEAT_SVE) && !IsFeatureImplemented(FEAT_SME) then EndOfDecode(Decode_UNDEF); end; let esize : integer{} = 32; let g : integer = UInt(Pg); let n : integer = UInt(Zn); let d : integer = UInt(Zd); let s_esize : integer{} = 32; let d_esize : integer{} = 16; let merging : boolean = TRUE;

Single-precision to half-precision, zeroing
(FEAT_SVE2p2 || FEAT_SME2p2)

313029282726252423222120191817161514131211109876543210
0110010010011010100PgZnZd
opcopc2

Encoding

FCVT <Zd>.H, <Pg>/Z, <Zn>.S

Decode for this encoding

if !IsFeatureImplemented(FEAT_SVE2p2) && !IsFeatureImplemented(FEAT_SME2p2) then EndOfDecode(Decode_UNDEF); end; let esize : integer{} = 32; let g : integer = UInt(Pg); let n : integer = UInt(Zn); let d : integer = UInt(Zd); let s_esize : integer{} = 32; let d_esize : integer{} = 16; let merging : boolean = FALSE;

Single-precision to double-precision, merging
(FEAT_SVE || FEAT_SME)

313029282726252423222120191817161514131211109876543210
0110010111001011101PgZnZd
opcopc2

Encoding

FCVT <Zd>.D, <Pg>/M, <Zn>.S

Decode for this encoding

if !IsFeatureImplemented(FEAT_SVE) && !IsFeatureImplemented(FEAT_SME) then EndOfDecode(Decode_UNDEF); end; let esize : integer{} = 64; let g : integer = UInt(Pg); let n : integer = UInt(Zn); let d : integer = UInt(Zd); let s_esize : integer{} = 32; let d_esize : integer{} = 64; let merging : boolean = TRUE;

Single-precision to double-precision, zeroing
(FEAT_SVE2p2 || FEAT_SME2p2)

313029282726252423222120191817161514131211109876543210
0110010011011010111PgZnZd
opcopc2

Encoding

FCVT <Zd>.D, <Pg>/Z, <Zn>.S

Decode for this encoding

if !IsFeatureImplemented(FEAT_SVE2p2) && !IsFeatureImplemented(FEAT_SME2p2) then EndOfDecode(Decode_UNDEF); end; let esize : integer{} = 64; let g : integer = UInt(Pg); let n : integer = UInt(Zn); let d : integer = UInt(Zd); let s_esize : integer{} = 32; let d_esize : integer{} = 64; let merging : boolean = FALSE;

Double-precision to half-precision, merging
(FEAT_SVE || FEAT_SME)

313029282726252423222120191817161514131211109876543210
0110010111001000101PgZnZd
opcopc2

Encoding

FCVT <Zd>.H, <Pg>/M, <Zn>.D

Decode for this encoding

if !IsFeatureImplemented(FEAT_SVE) && !IsFeatureImplemented(FEAT_SME) then EndOfDecode(Decode_UNDEF); end; let esize : integer{} = 64; let g : integer = UInt(Pg); let n : integer = UInt(Zn); let d : integer = UInt(Zd); let s_esize : integer{} = 64; let d_esize : integer{} = 16; let merging : boolean = TRUE;

Double-precision to half-precision, zeroing
(FEAT_SVE2p2 || FEAT_SME2p2)

313029282726252423222120191817161514131211109876543210
0110010011011010100PgZnZd
opcopc2

Encoding

FCVT <Zd>.H, <Pg>/Z, <Zn>.D

Decode for this encoding

if !IsFeatureImplemented(FEAT_SVE2p2) && !IsFeatureImplemented(FEAT_SME2p2) then EndOfDecode(Decode_UNDEF); end; let esize : integer{} = 64; let g : integer = UInt(Pg); let n : integer = UInt(Zn); let d : integer = UInt(Zd); let s_esize : integer{} = 64; let d_esize : integer{} = 16; let merging : boolean = FALSE;

Double-precision to single-precision, merging
(FEAT_SVE || FEAT_SME)

313029282726252423222120191817161514131211109876543210
0110010111001010101PgZnZd
opcopc2

Encoding

FCVT <Zd>.S, <Pg>/M, <Zn>.D

Decode for this encoding

if !IsFeatureImplemented(FEAT_SVE) && !IsFeatureImplemented(FEAT_SME) then EndOfDecode(Decode_UNDEF); end; let esize : integer{} = 64; let g : integer = UInt(Pg); let n : integer = UInt(Zn); let d : integer = UInt(Zd); let s_esize : integer{} = 64; let d_esize : integer{} = 32; let merging : boolean = TRUE;

Double-precision to single-precision, zeroing
(FEAT_SVE2p2 || FEAT_SME2p2)

313029282726252423222120191817161514131211109876543210
0110010011011010110PgZnZd
opcopc2

Encoding

FCVT <Zd>.S, <Pg>/Z, <Zn>.D

Decode for this encoding

if !IsFeatureImplemented(FEAT_SVE2p2) && !IsFeatureImplemented(FEAT_SME2p2) then EndOfDecode(Decode_UNDEF); end; let esize : integer{} = 64; let g : integer = UInt(Pg); let n : integer = UInt(Zn); let d : integer = UInt(Zd); let s_esize : integer{} = 64; let d_esize : integer{} = 32; let merging : boolean = FALSE;

Assembler Symbols

<Zd>

Is the name of the destination scalable vector register, encoded in the "Zd" field.

<Pg>

Is the name of the governing scalable predicate register P0-P7, encoded in the "Pg" field.

<Zn>

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

Operation

CheckSVEEnabled(); let VL : integer{} = CurrentVL(); let PL : integer{} = VL DIV 8; let elements : integer = VL DIV esize; let mask : bits(PL) = P{}(g); let operand : bits(VL) = if AnyActiveElement{PL}(mask, esize) then Z{VL}(n) else Zeros{VL}; var result : bits(VL) = if merging then Z{VL}(d) else Zeros{VL}; for e = 0 to elements-1 do if ActivePredicateElement{PL}(mask, e, esize) then let element : bits(esize) = operand[e*:esize]; let res : bits(d_esize) = FPConvertSVE{d_esize, s_esize}(element[s_esize-1:0], FPCR()); result[e*:esize] = ZeroExtend{esize}(res); end; end; Z{VL}(d) = result;

Operational information

For the "Half-precision to single-precision, merging" , "Half-precision to double-precision, merging" , "Single-precision to half-precision, merging" , "Single-precision to double-precision, merging" , "Double-precision to half-precision, merging" , "Double-precision to single-precision, merging" variants:

The merging variant of 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 the merging variant of this instruction is CONSTRAINED UNPREDICTABLE:


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.