VQRDMLSH

Vector Saturating Rounding Doubling Multiply Subtract Returning High Half. This instruction multiplies the vector elements of the first source SIMD&FP register with either the corresponding vector elements of the second source SIMD&FP register or the value of a vector element of the second source SIMD&FP register, without saturating the multiply results, doubles the results, and subtracts the most significant half of the final results from the vector elements of the destination SIMD&FP register. The results are rounded.

If any of the results overflow, they are saturated. The cumulative saturation bit, FPSCR.QC, is set if saturation occurs. For details see Pseudocode details of saturation.

Depending on settings in the CPACR, NSACR, and HCPTR registers, and the Security state and PE mode in which the instruction is executed, an attempt to execute the instruction might be undefined, or trapped to Hyp mode. For more information see Enabling Advanced SIMD and floating-point support.

It has encodings from the following instruction sets: A32 ( A1 and A2 ) and T32 ( T1 and T2 ) .

A1
(FEAT_RDM)

313029282726252423222120191817161514131211109876543210
111100110DsizeVnVd1100NQM1Vm

64-bit SIMD vector (Q == 0)

VQRDMLSH{<q>}.<dt> <Dd>, <Dn>, <Dm>

128-bit SIMD vector (Q == 1)

VQRDMLSH{<q>}.<dt> <Qd>, <Qn>, <Qm>

if !HaveQRDMLAHExt() then UNDEFINED; if Q == '1' && (Vd<0> == '1' || Vn<0> == '1' || Vm<0> == '1') then UNDEFINED; if size == '00' || size == '11' then UNDEFINED; add = FALSE; scalar_form = FALSE; esize = 8 << UInt(size); elements = 64 DIV esize; d = UInt(D:Vd); n = UInt(N:Vn); m = UInt(M:Vm); regs = if Q == '0' then 1 else 2; integer index = integer UNKNOWN;

A2
(FEAT_RDM)

313029282726252423222120191817161514131211109876543210
1111001Q1D!= 11VnVd1111N1M0Vm
size

64-bit SIMD vector (Q == 0)

VQRDMLSH{<q>}.<dt> <Dd>, <Dn>, <Dm[x]>

128-bit SIMD vector (Q == 1)

VQRDMLSH{<q>}.<dt> <Qd>, <Qn>, <Dm[x]>

if !HaveQRDMLAHExt() then UNDEFINED; if size == '11' then SEE "Related encodings"; if size == '00' then UNDEFINED; if Q == '1' && (Vd<0> == '1' || Vn<0> == '1') then UNDEFINED; add = FALSE; scalar_form = TRUE; d = UInt(D:Vd); n = UInt(N:Vn); regs = if Q == '0' then 1 else 2; integer esize; integer elements; integer m; integer index; if size == '01' then esize = 16; elements = 4; m = UInt(Vm<2:0>); index = UInt(M:Vm<3>); if size == '10' then esize = 32; elements = 2; m = UInt(Vm); index = UInt(M);

T1
(FEAT_RDM)

15141312111098765432101514131211109876543210
111111110DsizeVnVd1100NQM1Vm

64-bit SIMD vector (Q == 0)

VQRDMLSH{<q>}.<dt> <Dd>, <Dn>, <Dm>

128-bit SIMD vector (Q == 1)

VQRDMLSH{<q>}.<dt> <Qd>, <Qn>, <Qm>

if !HaveQRDMLAHExt() then UNDEFINED; if InITBlock() then UNPREDICTABLE; if Q == '1' && (Vd<0> == '1' || Vn<0> == '1' || Vm<0> == '1') then UNDEFINED; if size == '00' || size == '11' then UNDEFINED; add = FALSE; scalar_form = FALSE; esize = 8 << UInt(size); elements = 64 DIV esize; d = UInt(D:Vd); n = UInt(N:Vn); m = UInt(M:Vm); regs = if Q == '0' then 1 else 2; integer index = integer UNKNOWN;

CONSTRAINED UNPREDICTABLE behavior

If InITBlock(), then one of the following behaviors must occur:

T2
(FEAT_RDM)

15141312111098765432101514131211109876543210
111Q11111D!= 11VnVd1111N1M0Vm
size

64-bit SIMD vector (Q == 0)

VQRDMLSH{<q>}.<dt> <Dd>, <Dn>, <Dm[x]>

128-bit SIMD vector (Q == 1)

VQRDMLSH{<q>}.<dt> <Qd>, <Qn>, <Dm[x]>

if !HaveQRDMLAHExt() then UNDEFINED; if InITBlock() then UNPREDICTABLE; if size == '11' then SEE "Related encodings"; if size == '00' then UNDEFINED; if Q == '1' && (Vd<0> == '1' || Vn<0> == '1') then UNDEFINED; add = FALSE; scalar_form = TRUE; d = UInt(D:Vd); n = UInt(N:Vn); regs = if Q == '0' then 1 else 2; integer esize; integer m; integer index; integer elements; if size == '01' then esize = 16; elements = 4; m = UInt(Vm<2:0>); index = UInt(M:Vm<3>); if size == '10' then esize = 32; elements = 2; m = UInt(Vm); index = UInt(M);

CONSTRAINED UNPREDICTABLE behavior

If InITBlock(), then one of the following behaviors must occur:

Related encodings: See Advanced SIMD data-processing for the T32 instruction set, or Advanced SIMD data-processing for the A32 instruction set.

Assembler Symbols

<q>

See Standard assembler syntax fields.

<dt>

Is the data type for the elements of the operands, encoded in size:

size <dt>
01 S16
10 S32
<Qd>

Is the 128-bit name of the SIMD&FP register holding the accumulate vector, encoded in the "D:Vd" field as <Qd>*2.

<Qn>

Is the 128-bit name of the first SIMD&FP source register, encoded in the "N:Vn" field as <Qn>*2.

<Qm>

Is the 128-bit name of the second SIMD&FP source register, encoded in the "M:Vm" field as <Qm>*2.

<Dd>

Is the 64-bit name of the SIMD&FP register holding the accumulate vector, encoded in the "D:Vd" field.

<Dn>

Is the 64-bit name of the first SIMD&FP source register, encoded in the "N:Vn" field.

<Dm[x]>

Is the 64-bit name of the second SIMD&FP source register holding the scalar. If <dt> is S16, Dm is restricted to D0-D7. Dm is encoded in "Vm<2:0>", and x is encoded in "M:Vm<3>". If <dt> is S32, Dm is restricted to D0-D15. Dm is encoded in "Vm", and x is encoded in "M".

<Dm>

Is the 64-bit name of the second SIMD&FP source register, encoded in the "M:Vm" field.

Operation

EncodingSpecificOperations(); CheckAdvSIMDEnabled(); integer op2; boolean round = TRUE; if scalar_form then op2 = SInt(Elem[D[m],index,esize]); for r = 0 to regs-1 for e = 0 to elements-1 op1 = SInt(Elem[D[n+r],e,esize]); op3 = SInt(Elem[D[d+r],e,esize]) << esize; if !scalar_form then op2 = SInt(Elem[D[m+r],e,esize]); integer rdmlsh = RShr(op3 - 2*(op1*op2), esize, round); (result, sat) = SignedSatQ(rdmlsh, esize); Elem[D[d+r],e,esize] = result; if sat then FPSCR.QC = '1';


Internal version only: isa v01_31, pseudocode v2023-06_rel, sve v2023-06_rel ; Build timestamp: 2023-07-04T18:06

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