SQDMLSL, SQDMLSL2 (by element)

Signed saturating Doubling Multiply-Subtract Long (by element). This instruction multiplies each vector element in the lower or upper half of the first source SIMD&FP register by the specified vector element of the second source SIMD&FP register, doubles the results, and subtracts the final results from the vector elements of the destination SIMD&FP register. The destination vector elements are twice as long as the elements that are multiplied. All the values in this instruction are signed integer values.

If overflow occurs with any of the results, those results are saturated. If saturation occurs, the cumulative saturation bit FPSR.QC is set.

The SQDMLSL instruction extracts vector elements from the lower half of the first source register. The SQDMLSL2 instruction extracts vector elements from the upper half of the first source register.

Depending on the settings in the CPACR_EL1, CPTR_EL2, and CPTR_EL3 registers, and the current Security state and Exception level, an attempt to execute the instruction might be trapped.

It has encodings from 2 classes: Scalar and Vector

Scalar

313029282726252423222120191817161514131211109876543210
01011111sizeLMRm0111H0RnRd
Uo2

SQDMLSL <Va><d>, <Vb><n>, <Vm>.<Ts>[<index>]

constant integer idxdsize = 64 << UInt(H); integer index; bit Rmhi; case size of when '01' index = UInt(H:L:M); Rmhi = '0'; when '10' index = UInt(H:L); Rmhi = M; otherwise UNDEFINED; integer d = UInt(Rd); integer n = UInt(Rn); integer m = UInt(Rmhi:Rm); constant integer esize = 8 << UInt(size); constant integer datasize = esize; integer elements = 1; integer part = 0; boolean sub_op = (o2 == '1');

Vector

313029282726252423222120191817161514131211109876543210
0Q001111sizeLMRm0111H0RnRd
Uo2

SQDMLSL{2} <Vd>.<Ta>, <Vn>.<Tb>, <Vm>.<Ts>[<index>]

constant integer idxdsize = 64 << UInt(H); integer index; bit Rmhi; case size of when '01' index = UInt(H:L:M); Rmhi = '0'; when '10' index = UInt(H:L); Rmhi = M; otherwise UNDEFINED; integer d = UInt(Rd); integer n = UInt(Rn); integer m = UInt(Rmhi:Rm); constant integer esize = 8 << UInt(size); constant integer datasize = 64; integer part = UInt(Q); integer elements = datasize DIV esize; boolean sub_op = (o2 == '1');

Assembler Symbols

<Va>

Is the destination width specifier, encoded in size:

size <Va>
00 RESERVED
01 S
10 D
11 RESERVED
<d>

Is the number of the SIMD&FP destination register, encoded in the "Rd" field.

<Vb>

Is the source width specifier, encoded in size:

size <Vb>
00 RESERVED
01 H
10 S
11 RESERVED
<n>

Is the number of the first SIMD&FP source register, encoded in the "Rn" field.

<Vm>

Is the name of the second SIMD&FP source register, encoded in size:M:Rm:

size <Vm>
00 RESERVED
01 UInt('0':Rm)
10 UInt(M:Rm)
11 RESERVED
Restricted to V0-V15 when element size <Ts> is H.
<Ts>

Is an element size specifier, encoded in size:

size <Ts>
00 RESERVED
01 H
10 S
11 RESERVED
<index>

Is the element index, encoded in size:H:L:M:

size <index>
00 RESERVED
01 UInt(H:L:M)
10 UInt(H:L)
11 RESERVED
2

Is the second and upper half specifier. If present it causes the operation to be performed on the upper 64 bits of the registers holding the narrower elements, and is encoded in Q:

Q 2
0 [absent]
1 [present]
<Vd>

Is the name of the SIMD&FP destination register, encoded in the "Rd" field.

<Ta>

Is an arrangement specifier, encoded in size:

size <Ta>
00 RESERVED
01 4S
10 2D
11 RESERVED
<Vn>

Is the name of the first SIMD&FP source register, encoded in the "Rn" field.

<Tb>

Is an arrangement specifier, encoded in size:Q:

size Q <Tb>
00 x RESERVED
01 0 4H
01 1 8H
10 0 2S
10 1 4S
11 x RESERVED

Operation

CheckFPAdvSIMDEnabled64(); bits(datasize) operand1 = Vpart[n, part, datasize]; bits(idxdsize) operand2 = V[m, idxdsize]; bits(2*datasize) operand3 = V[d, 2*datasize]; bits(2*datasize) result; integer element1; integer element2; bits(2*esize) product; integer accum; boolean sat1; boolean sat2; element2 = SInt(Elem[operand2, index, esize]); for e = 0 to elements-1 element1 = SInt(Elem[operand1, e, esize]); (product, sat1) = SignedSatQ(2 * element1 * element2, 2 * esize); if sub_op then accum = SInt(Elem[operand3, e, 2*esize]) - SInt(product); else accum = SInt(Elem[operand3, e, 2*esize]) + SInt(product); (Elem[result, e, 2*esize], sat2) = SignedSatQ(accum, 2 * esize); if sat1 || sat2 then FPSR.QC = '1'; V[d, 2*datasize] = result;


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.