Vector Saturating Rounding Doubling Multiply Accumulate 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 accumulates the most significant half of the final results with 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 ) .
31 | 30 | 29 | 28 | 27 | 26 | 25 | 24 | 23 | 22 | 21 | 20 | 19 | 18 | 17 | 16 | 15 | 14 | 13 | 12 | 11 | 10 | 9 | 8 | 7 | 6 | 5 | 4 | 3 | 2 | 1 | 0 |
1 | 1 | 1 | 1 | 0 | 0 | 1 | 1 | 0 | D | size | Vn | Vd | 1 | 0 | 1 | 1 | N | Q | M | 1 | Vm |
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 = TRUE; 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;
31 | 30 | 29 | 28 | 27 | 26 | 25 | 24 | 23 | 22 | 21 | 20 | 19 | 18 | 17 | 16 | 15 | 14 | 13 | 12 | 11 | 10 | 9 | 8 | 7 | 6 | 5 | 4 | 3 | 2 | 1 | 0 |
1 | 1 | 1 | 1 | 0 | 0 | 1 | Q | 1 | D | != 11 | Vn | Vd | 1 | 1 | 1 | 0 | N | 1 | M | 0 | Vm | ||||||||||
size |
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 = TRUE; 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);
15 | 14 | 13 | 12 | 11 | 10 | 9 | 8 | 7 | 6 | 5 | 4 | 3 | 2 | 1 | 0 | 15 | 14 | 13 | 12 | 11 | 10 | 9 | 8 | 7 | 6 | 5 | 4 | 3 | 2 | 1 | 0 |
1 | 1 | 1 | 1 | 1 | 1 | 1 | 1 | 0 | D | size | Vn | Vd | 1 | 0 | 1 | 1 | N | Q | M | 1 | Vm |
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 = TRUE; 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;
If InITBlock(), then one of the following behaviors must occur:
15 | 14 | 13 | 12 | 11 | 10 | 9 | 8 | 7 | 6 | 5 | 4 | 3 | 2 | 1 | 0 | 15 | 14 | 13 | 12 | 11 | 10 | 9 | 8 | 7 | 6 | 5 | 4 | 3 | 2 | 1 | 0 |
1 | 1 | 1 | Q | 1 | 1 | 1 | 1 | 1 | D | != 11 | Vn | Vd | 1 | 1 | 1 | 0 | N | 1 | M | 0 | Vm | ||||||||||
size |
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 = TRUE; 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);
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.
<q> |
<dt> |
Is the data type for the elements of the operands,
encoded in
|
<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> |
Is the 64-bit name of the second SIMD&FP source register, encoded in the "M:Vm" field. |
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 rdmlah = RShr(op3 + 2*(op1*op2), esize, round); (result, sat) = SignedSatQ(rdmlah, 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.