Signed by unsigned integer sum of outer products and subtract

The 8-bit integer variant works with a 32-bit element ZA tile.

The 16-bit integer variant works with a 64-bit element ZA tile.

The signed by unsigned integer sum of outer products and subtract instructions multiply the sub-matrix in the first source vector by the sub-matrix in the second source vector. In case of the 8-bit integer variant, the first source holds SVL_{S}×4 sub-matrix of signed 8-bit integer values, and the second source holds 4×SVL_{S} sub-matrix of unsigned 8-bit integer values. In case of the 16-bit integer variant, the first source holds SVL_{D}×4 sub-matrix of signed 16-bit integer values, and the second source holds 4×SVL_{D} sub-matrix of unsigned 16-bit integer values.

Each source vector is independently predicated by a corresponding governing predicate. When an 8-bit source element in case of 8-bit integer variant or a 16-bit source element in case of 16-bit integer variant is Inactive, it is treated as having the value 0.

The resulting SVL_{S}×SVL_{S} widened 32-bit integer or SVL_{D}×SVL_{D} widened 64-bit integer sum of outer products is then destructively subtracted from the 32-bit integer or 64-bit integer destination tile, respectively for 8-bit integer and 16-bit integer instruction variants. This is equivalent to performing a 4-way dot product and subtract from each of the destination tile elements.

In case of the 8-bit integer variant, each 32-bit container of the first source vector holds 4 consecutive column elements of each row of a SVL_{S}×4 sub-matrix, and each 32-bit container of the second source vector holds 4 consecutive row elements of each column of a 4×SVL_{S} sub-matrix. In case of the 16-bit integer variant, each 64-bit container of the first source vector holds 4 consecutive column elements of each row of a SVL_{D}×4 sub-matrix, and each 64-bit container of the second source vector holds 4 consecutive row elements of each column of a 4×SVL_{D} sub-matrix.

ID_AA64SMFR0_EL1.I16I64 indicates whether the 16-bit integer variant is implemented.

It has encodings from 2 classes: 32-bit and 64-bit

(FEAT_SME)

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 | 0 | 1 | 0 | 0 | 0 | 0 | 0 | 1 | 0 | 1 | Zm | Pm | Pn | Zn | 1 | 0 | 0 | ZAda | |||||||||||||

u0 | u1 | S |

if !IsFeatureImplemented(FEAT_SME) then UNDEFINED; constant integer esize = 32; constant integer a = UInt(Pn); constant integer b = UInt(Pm); constant integer n = UInt(Zn); constant integer m = UInt(Zm); constant integer da = UInt(ZAda); constant boolean op1_unsigned = FALSE; constant boolean op2_unsigned = TRUE;

(FEAT_SME_I16I64)

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 | 0 | 1 | 0 | 0 | 0 | 0 | 0 | 1 | 1 | 1 | Zm | Pm | Pn | Zn | 1 | 0 | ZAda | ||||||||||||||

u0 | u1 | S |

if !IsFeatureImplemented(FEAT_SME_I16I64) then UNDEFINED; constant integer esize = 64; constant integer a = UInt(Pn); constant integer b = UInt(Pm); constant integer n = UInt(Zn); constant integer m = UInt(Zm); constant integer da = UInt(ZAda); constant boolean op1_unsigned = FALSE; constant boolean op2_unsigned = TRUE;

<ZAda> |
For the 32-bit variant: is the name of the ZA tile ZA0-ZA3, encoded in the "ZAda" field. |

For the 64-bit variant: is the name of the ZA tile ZA0-ZA7, encoded in the "ZAda" field. |

<Pn> |
Is the name of the first governing scalable predicate register P0-P7, encoded in the "Pn" field. |

<Pm> |
Is the name of the second governing scalable predicate register P0-P7, encoded in the "Pm" 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. |

CheckStreamingSVEAndZAEnabled(); constant integer VL = CurrentVL; constant integer PL = VL DIV 8; constant integer dim = VL DIV esize; constant bits(PL) mask1 = P[a, PL]; constant bits(PL) mask2 = P[b, PL]; constant bits(VL) operand1 = Z[n, VL]; constant bits(VL) operand2 = Z[m, VL]; constant bits(dim*dim*esize) operand3 = ZAtile[da, esize, dim*dim*esize]; bits(dim*dim*esize) result; integer prod; for row = 0 to dim-1 for col = 0 to dim-1 bits(esize) sum = Elem[operand3, row*dim+col, esize]; for k = 0 to 3 if (ActivePredicateElement(mask1, 4*row + k, esize DIV 4) && ActivePredicateElement(mask2, 4*col + k, esize DIV 4)) then prod = (Int(Elem[operand1, 4*row + k, esize DIV 4], op1_unsigned) * Int(Elem[operand2, 4*col + k, esize DIV 4], op2_unsigned)); sum = sum - prod; Elem[result, row*dim+col, esize] = sum; ZAtile[da, esize, dim*dim*esize] = result;

If PSTATE.DIT is 1:

- The execution time of this instruction is independent of:
- The values of the data supplied in any of its operand registers when its governing predicate registers contain the same value for each execution.
- The values of the NZCV flags.

- The response of this instruction to asynchronous exceptions does not vary based on:
- The values of the data supplied in any of its operand registers when its governing predicate registers contain the same value for each execution.
- The values of the NZCV flags.

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.