LDNT1B (scalar plus immediate, consecutive registers)

Contiguous load non-temporal of bytes to multiple consecutive vectors (immediate index)

Contiguous load non-temporal of bytes to elements of two or four consecutive vector registers from the memory address generated by a 64-bit scalar base and immediate index which is multiplied by the vector's in-memory size, irrespective of predication, and added to the base address.

Inactive elements will not cause a read from Device memory or signal a fault, and are set to zero in the destination vector.

A non-temporal load is a hint to the system that this data is unlikely to be referenced again soon.

It has encodings from 2 classes: Two registers and Four registers

Two registers
(FEAT_SVE2p1)

313029282726252423222120191817161514131211109876543210
101000000100imm4000PNgRnZt1
msz<1>msz<0>N

LDNT1B { <Zt1>.B-<Zt2>.B }, <PNg>/Z, [<Xn|SP>{, #<imm>, MUL VL}]

if !IsFeatureImplemented(FEAT_SME2) && !IsFeatureImplemented(FEAT_SVE2p1) then UNDEFINED; constant integer n = UInt(Rn); constant integer g = UInt('1':PNg); constant integer nreg = 2; constant integer t = UInt(Zt:'0'); constant integer esize = 8; constant integer offset = SInt(imm4);

Four registers
(FEAT_SVE2p1)

313029282726252423222120191817161514131211109876543210
101000000100imm4100PNgRnZt01
msz<1>msz<0>N

LDNT1B { <Zt1>.B-<Zt4>.B }, <PNg>/Z, [<Xn|SP>{, #<imm>, MUL VL}]

if !IsFeatureImplemented(FEAT_SME2) && !IsFeatureImplemented(FEAT_SVE2p1) then UNDEFINED; constant integer n = UInt(Rn); constant integer g = UInt('1':PNg); constant integer nreg = 4; constant integer t = UInt(Zt:'00'); constant integer esize = 8; constant integer offset = SInt(imm4);

Assembler Symbols

<Zt1>

For the two registers variant: is the name of the first scalable vector register to be transferred, encoded as "Zt" times 2.

For the four registers variant: is the name of the first scalable vector register to be transferred, encoded as "Zt" times 4.

<Zt4>

Is the name of the fourth scalable vector register to be transferred, encoded as "Zt" times 4 plus 3.

<Zt2>

Is the name of the second scalable vector register to be transferred, encoded as "Zt" times 2 plus 1.

<PNg>

Is the name of the governing scalable predicate register PN8-PN15, with predicate-as-counter encoding, encoded in the "PNg" field.

<Xn|SP>

Is the 64-bit name of the general-purpose base register or stack pointer, encoded in the "Rn" field.

<imm>

For the two registers variant: is the optional signed immediate vector offset, a multiple of 2 in the range -16 to 14, defaulting to 0, encoded in the "imm4" field.

For the four registers variant: is the optional signed immediate vector offset, a multiple of 4 in the range -32 to 28, defaulting to 0, encoded in the "imm4" field.

Operation

if IsFeatureImplemented(FEAT_SVE2p1) then CheckSVEEnabled(); else CheckStreamingSVEEnabled(); constant integer VL = CurrentVL; constant integer PL = VL DIV 8; constant integer elements = VL DIV esize; constant integer mbytes = esize DIV 8; bits(64) base; bits(64) addr; constant bits(PL) pred = P[g, PL]; constant bits(PL * nreg) mask = CounterToPredicate(pred<15:0>, PL * nreg); array [0..3] of bits(VL) values; constant boolean contiguous = TRUE; constant boolean nontemporal = TRUE; constant integer transfer = t; constant boolean tagchecked = n != 31; constant AccessDescriptor accdesc = CreateAccDescSVE(MemOp_LOAD, nontemporal, contiguous, tagchecked); if !AnyActiveElement(mask, esize) then if n == 31 && ConstrainUnpredictableBool(Unpredictable_CHECKSPNONEACTIVE) then CheckSPAlignment(); else if n == 31 then CheckSPAlignment(); base = if n == 31 then SP[] else X[n, 64]; addr = AddressAdd(base, offset * nreg * elements * mbytes, accdesc); for r = 0 to nreg-1 for e = 0 to elements-1 if ActivePredicateElement(mask, r * elements + e, esize) then Elem[values[r], e, esize] = Mem[addr, mbytes, accdesc]; else Elem[values[r], e, esize] = Zeros(esize); addr = AddressIncrement(addr, mbytes, accdesc); for r = 0 to nreg-1 Z[transfer+r, VL] = values[r];

Operational information

If PSTATE.DIT is 1, the timing of this instruction is insensitive to the value of the data being loaded or stored when its governing predicate register contains the same value for each execution.


Internal version only: aarchmrs v2024-03_relA, pseudocode v2024-03_rel, sve v2024-03_rel ; Build timestamp: 2024-03-26T09:45

Copyright © 2010-2024 Arm Limited or its affiliates. All rights reserved. This document is Non-Confidential.