Floating-point matrix multiply-accumulate
The floating-point matrix multiply-accumulate instruction supports single-precision and double-precision data types in a 2×2 matrix contained in segments of 128 or 256 bits, respectively. It multiplies the 2×2 matrix in each segment of the first source vector by the 2×2 matrix in the corresponding segment of the second source vector. The resulting 2×2 matrix product is then destructively added to the matrix accumulator held in the corresponding segment of the addend and destination vector. This is equivalent to performing a 2-way dot product per destination element. This instruction is unpredicated. The single-precision variant is vector length agnostic. The double-precision variant requires that the current vector length is at least 256 bits, and if the current vector length is not an integer multiple of 256 bits then the trailing bits are set to zero.
ID_AA64ZFR0_EL1.F32MM indicates whether the single-precision variant is implemented.
ID_AA64ZFR0_EL1.F64MM indicates whether the double-precision variant is implemented.
It has encodings from 2 classes: 32-bit element and 64-bit element
| 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 |
| 0 | 1 | 1 | 0 | 0 | 1 | 0 | 0 | 1 | 0 | 1 | Zm | 1 | 1 | 1 | 0 | 0 | 1 | Zn | Zda | ||||||||||||
if !HaveSVE() || !HaveSVEFP32MatMulExt() then UNDEFINED; integer esize = 32; integer n = UInt(Zn); integer m = UInt(Zm); integer da = UInt(Zda);
| 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 |
| 0 | 1 | 1 | 0 | 0 | 1 | 0 | 0 | 1 | 1 | 1 | Zm | 1 | 1 | 1 | 0 | 0 | 1 | Zn | Zda | ||||||||||||
if !HaveSVE() || !HaveSVEFP64MatMulExt() then UNDEFINED; integer esize = 64; integer n = UInt(Zn); integer m = UInt(Zm); integer da = UInt(Zda);
| <Zda> |
Is the name of the third source and destination scalable vector register, encoded in the "Zda" field. |
| <Zn> |
Is the name of the first source scalable vector register, encoded in the "Zn" field. |
| <Zm> |
Is the name of the second source scalable vector register, encoded in the "Zm" field. |
CheckNonStreamingSVEEnabled(); if VL < esize * 4 then UNDEFINED; integer segments = VL DIV (4 * esize); bits(VL) operand1 = Z[n]; bits(VL) operand2 = Z[m]; bits(VL) operand3 = Z[da]; bits(VL) result = Zeros(); bits(4*esize) op1, op2; bits(4*esize) res, addend; for s = 0 to segments-1 op1 = Elem[operand1, s, 4*esize]; op2 = Elem[operand2, s, 4*esize]; addend = Elem[operand3, s, 4*esize]; res = FPMatMulAdd(addend, op1, op2, esize, FPCR[]); Elem[result, s, 4*esize] = res; Z[da] = result;
This instruction might be immediately preceded in program order by a MOVPRFX instruction. The MOVPRFX instruction must conform to all of the following requirements, otherwise the behavior of the MOVPRFX and this instruction is unpredictable:
Internal version only: isa v33.11seprel, AdvSIMD v29.05, pseudocode v2021-09_rel, sve v2021-09_rc3d ; Build timestamp: 2021-10-06T11:41
Copyright © 2010-2021 Arm Limited or its affiliates. All rights reserved. This document is Non-Confidential.