FMLALL (multiple and single vector)

Multi-vector 8-bit floating-point multiply-add long-long by vector to single-precision

This 8-bit floating-point multiply-add long long instruction widens all 8-bit floating-point elements in the one, two, or four first source vectors and the second source vector to single-precision format and multiplies the corresponding elements. The intermediate products are scaled by 2-UInt(FPMR.LSCALE) before being destructively added without intermediate rounding to the overlapping 32-bit single-precision elements of the ZA quad-vector groups.

The lowest of the four consecutive vector numbers forming the quad-vector group within all of, each half of, or each quarter of the ZA array are selected by the sum of the vector select register and immediate offset, 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 quad-vector groups respectively. The vector group symbol is preferred for disassembly, but optional in assembler source code.

The 8-bit floating-point encoding format for the elements of the first source vector and the second source vector is selected by FPMR.F8S1 and FPMR.F8S2 respectively.

This instruction is unpredicated.

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

One ZA quad-vector
(FEAT_SME_F8F32)

313029282726252423222120191817161514131211109876543210
110000010011Zm0Rv001Zn000off2

FMLALL ZA.S[<Wv>, <offs1>:<offs4>], <Zn>.B, <Zm>.B

if !IsFeatureImplemented(FEAT_SME_F8F32) then UNDEFINED; integer v = UInt('010':Rv); integer n = UInt(Zn); integer m = UInt('0':Zm); integer offset = UInt(off2:'00'); constant integer nreg = 1;

Two ZA quad-vectors
(FEAT_SME_F8F32)

313029282726252423222120191817161514131211109876543210
110000010010Zm0Rv000Zn0001o1

FMLALL ZA.S[<Wv>, <offs1>:<offs4>{, VGx2}], { <Zn1>.B-<Zn2>.B }, <Zm>.B

if !IsFeatureImplemented(FEAT_SME_F8F32) then UNDEFINED; integer v = UInt('010':Rv); integer n = UInt(Zn); integer m = UInt('0':Zm); integer offset = UInt(o1:'00'); constant integer nreg = 2;

Four ZA quad-vectors
(FEAT_SME_F8F32)

313029282726252423222120191817161514131211109876543210
110000010011Zm0Rv000Zn0001o1

FMLALL ZA.S[<Wv>, <offs1>:<offs4>{, VGx4}], { <Zn1>.B-<Zn4>.B }, <Zm>.B

if !IsFeatureImplemented(FEAT_SME_F8F32) then UNDEFINED; integer v = UInt('010':Rv); integer n = UInt(Zn); integer m = UInt('0':Zm); integer offset = UInt(o1:'00'); 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 quad-vector variant: is the first vector select offset, encoded as "off2" field times 4.

For the four ZA quad-vectors and two ZA quad-vectors variant: is the first vector select offset, encoded as "o1" field times 4.

<offs4>

For the one ZA quad-vector variant: is the fourth vector select offset, encoded as "off2" field times 4 plus 3.

For the four ZA quad-vectors and two ZA quad-vectors variant: is the fourth vector select offset, encoded as "o1" field times 4 plus 3.

<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

CheckFPMREnabled(); CheckStreamingSVEAndZAEnabled(); constant integer VL = CurrentVL; constant integer elements = VL DIV 32; integer vectors = VL DIV 8; integer vstride = vectors DIV nreg; bits(32) vbase = X[v, 32]; integer vec = (UInt(vbase) + offset) MOD vstride; bits(VL) result; vec = vec - (vec MOD 4); for r = 0 to nreg-1 bits(VL) operand1 = Z[(n+r) MOD 32, VL]; bits(VL) operand2 = Z[m, VL]; for i = 0 to 3 bits(VL) operand3 = ZAvector[vec + i, VL]; for e = 0 to elements-1 bits(8) element1 = Elem[operand1, 4 * e + i, 8]; bits(8) element2 = Elem[operand2, 4 * e + i, 8]; bits(32) element3 = Elem[operand3, e, 32]; Elem[result, e, 32] = FP8MulAddFP(element3, element1, element2, FPCR, FPMR); ZAvector[vec + i, VL] = result; vec = vec + vstride;


Internal version only: aarchmrs v2023-12_rel, pseudocode v2023-12_rel, sve v2023-12_rel ; Build timestamp: 2023-12-15T16:46

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