BFMLAL (multiple and single vector)

Multi-vector BFloat16 floating-point multiply-add long by vector

This BFloat16 floating-point multiply-add long instruction widens all 16-bit BFloat16 elements in the one, two, or four first source vectors and the second source vector to single-precision format, then multiplies the corresponding elements and destructively adds these values without intermediate rounding to the overlapping 32-bit single-precision elements of the ZA double-vector groups.

The double-vector group within all of, each half of, or each quarter of the ZA array is selected by the sum of the vector select register and offset range, modulo all, half, or quarter the number of ZA array vectors.

The vector group symbol, VGx2 or VGx4, indicates that the ZA operand consists of two or four ZA double-vector groups respectively. The vector group symbol is preferred for disassembly, but optional in assembler source code.

This instruction follows SME ZA-targeting floating-point behaviors.

This instruction is unpredicated.

It has encodings from 3 classes: One ZA double-vector , Two ZA double-vectors and Four ZA double-vectors

One ZA double-vector
(FEAT_SME2)

313029282726252423222120191817161514131211109876543210
110000010010Zm0Rv011Zn10off3
S

BFMLAL ZA.S[<Wv>, <offs1>:<offs2>], <Zn>.H, <Zm>.H

if !IsFeatureImplemented(FEAT_SME2) then UNDEFINED; constant integer v = UInt('010':Rv); constant integer n = UInt(Zn); constant integer m = UInt('0':Zm); constant integer offset = UInt(off3:'0'); constant integer nreg = 1;

Two ZA double-vectors
(FEAT_SME2)

313029282726252423222120191817161514131211109876543210
110000010010Zm0Rv010Zn100off2
S

BFMLAL ZA.S[<Wv>, <offs1>:<offs2>{, VGx2}], { <Zn1>.H-<Zn2>.H }, <Zm>.H

if !IsFeatureImplemented(FEAT_SME2) then UNDEFINED; constant integer v = UInt('010':Rv); constant integer n = UInt(Zn); constant integer m = UInt('0':Zm); constant integer offset = UInt(off2:'0'); constant integer nreg = 2;

Four ZA double-vectors
(FEAT_SME2)

313029282726252423222120191817161514131211109876543210
110000010011Zm0Rv010Zn100off2
S

BFMLAL ZA.S[<Wv>, <offs1>:<offs2>{, VGx4}], { <Zn1>.H-<Zn4>.H }, <Zm>.H

if !IsFeatureImplemented(FEAT_SME2) then UNDEFINED; constant integer v = UInt('010':Rv); constant integer n = UInt(Zn); constant integer m = UInt('0':Zm); constant integer offset = UInt(off2:'0'); constant integer nreg = 4;

Assembler Symbols

<Wv>

Is the 32-bit name of the vector select register W8-W11, encoded in the "Rv" field.

<offs1>

For the one ZA double-vector variant: is the first vector select offset, encoded as "off3" field times 2.

For the four ZA double-vectors and two ZA double-vectors variant: is the first vector select offset, encoded as "off2" field times 2.

<offs2>

For the one ZA double-vector variant: is the second vector select offset, encoded as "off3" field times 2 plus 1.

For the four ZA double-vectors and two ZA double-vectors variant: is the second vector select offset, encoded as "off2" field times 2 plus 1.

<Zn>

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

<Zn1>

Is the name of the first scalable vector register of the first source multi-vector group, encoded as "Zn".

<Zn4>

Is the name of the fourth scalable vector register of the first source multi-vector group, encoded as "Zn" plus 3 modulo 32.

<Zn2>

Is the name of the second scalable vector register of the first source multi-vector group, encoded as "Zn" plus 1 modulo 32.

<Zm>

Is the name of the second source scalable vector register Z0-Z15, encoded in the "Zm" field.

Operation

CheckStreamingSVEAndZAEnabled(); constant integer VL = CurrentVL; constant integer elements = VL DIV 32; constant integer vectors = VL DIV 8; constant integer vstride = vectors DIV nreg; constant bits(32) vbase = X[v, 32]; integer vec = (UInt(vbase) + offset) MOD vstride; bits(VL) result; vec = vec - (vec MOD 2); for r = 0 to nreg-1 constant bits(VL) op1 = Z[(n+r) MOD 32, VL]; constant bits(VL) op2 = Z[m, VL]; for i = 0 to 1 constant bits(VL) op3 = ZAvector[vec + i, VL]; for e = 0 to elements-1 constant bits(16) elem1 = Elem[op1, 2 * e + i, 16]; constant bits(16) elem2 = Elem[op2, 2 * e + i, 16]; constant bits(32) elem3 = Elem[op3, e, 32]; Elem[result, e, 32] = BFMulAddH_ZA(elem3, elem1, elem2, FPCR); ZAvector[vec + i, VL] = result; vec = vec + vstride;


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.