LUTI4

Lookup table read with 4-bit indices

This instruction copies indexed 8-bit or 16-bit elements from the low 128 or 256 bits of the table vector, or from the low 128 bits of the two table vectors to the destination vector using packed 4-bit indices from a segment of the source vector. A segment corresponds to a portion of the source vector that is consumed in order to fill the destination vector. The segment is selected by the vector segment index. This instruction is unpredicated.

It has encodings from 3 classes: Byte, single register table , Halfword, two register table and Halfword, single register table

Byte, single register table
(FEAT_LUT)

313029282726252423222120191817161514131211109876543210
01000101i111Zm101001ZnZd

LUTI4 <Zd>.B, { <Zn>.B }, <Zm>[<index>]

if (!HaveSVE2() && !HaveSME2()) || !IsFeatureImplemented(FEAT_LUT) then UNDEFINED; integer isize = 4; constant integer esize = 8; integer ntblr = 1; integer m = UInt(Zm); integer n = UInt(Zn); integer d = UInt(Zd); integer part = UInt(i1);

Halfword, two register table
(FEAT_LUT)

313029282726252423222120191817161514131211109876543210
01000101i21Zm101101ZnZd

LUTI4 <Zd>.H, { <Zn1>.H, <Zn2>.H }, <Zm>[<index>]

if (!HaveSVE2() && !HaveSME2()) || !IsFeatureImplemented(FEAT_LUT) then UNDEFINED; integer isize = 4; constant integer esize = 16; integer ntblr = 2; integer m = UInt(Zm); integer n = UInt(Zn); integer d = UInt(Zd); integer part = UInt(i2);

Halfword, single register table
(FEAT_LUT)

313029282726252423222120191817161514131211109876543210
01000101i21Zm101111ZnZd

LUTI4 <Zd>.H, { <Zn>.H }, <Zm>[<index>]

if (!HaveSVE2() && !HaveSME2()) || !IsFeatureImplemented(FEAT_LUT) then UNDEFINED; if MaxImplementedAnyVL() < 256 then UNDEFINED; integer isize = 4; constant integer esize = 16; integer ntblr = 1; integer m = UInt(Zm); integer n = UInt(Zn); integer d = UInt(Zd); integer part = UInt(i2);

Assembler Symbols

<Zd>

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

<Zn>

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

<Zn1>

Is the name of the first table vector register, encoded as "Zn".

<Zn2>

Is the name of the second table vector register, encoded as "Zn" plus 1 modulo 32.

<Zm>

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

<index>

For the "halfword, single register table" and "halfword, two register table" variant: is the vector segment index, in the range 0 to 3, encoded in the "i2" field.

For the byte, single register table variant: is the vector segment index, in the range 0 to 1, encoded in the "i1" field.

Operation

CheckSVEEnabled(); constant integer VL = CurrentVL; if ntblr == 1 && esize == 16 && VL < 256 then UNDEFINED; constant integer elements = VL DIV esize; constant integer tablesize = if ntblr == 1 && esize == 16 then 256 else 128; constant integer eltspertable = tablesize DIV esize; constant integer ibase = elements * part; bits(VL) indices = Z[m, VL]; bits(VL) table1 = Z[n+0, VL]; bits(VL) table2 = if ntblr == 2 then Z[(n+1) MOD 32, VL] else Zeros(VL); bits(VL) result; bits(esize) res; for e = 0 to elements-1 integer index = UInt(Elem[indices, ibase + e, isize]); if index < eltspertable then res = Elem[table1, index, esize]; else assert ntblr == 2; res = Elem[table2, index - eltspertable, esize]; Elem[result, e, esize] = res; Z[d, VL] = result;

Operational information

If FEAT_SVE2 is implemented or FEAT_SME is implemented, then if PSTATE.DIT is 1:


Internal version only: aarchmrs v2023-12_rel, pseudocode v2023-12_rel, sve v2023-12_rel ; Build timestamp: 2023-12-15T16:46

Copyright © 2010-2023 Arm Limited or its affiliates. All rights reserved. This document is Non-Confidential.