pto.mad_bias

pto.mad_bias is part of the Cube MAD Ops.

Summary

Bias-init cube matrix multiply: dst[m, n] = sum_k(lhs[m, k] * rhs[k, n]) + bias[n].

Mechanism

Like pto.mad, but seeds the accumulator with a per-N bias vector instead of zero. Useful as the first MAD in a K-tiled sequence where the bias is known up front; subsequent partial sums can accumulate via pto.mad_acc.

Syntax

pto.mad_bias %lhs, %rhs, %dst, %bias, %m, %n, %k
  unit_flag(check_only | check_and_set)?
  disable_gemv?
  (sat | nosat)?
  tf32_mode(round_even | round_away)?
  n_dir?
  : !pto.ptr<A, l0a>, !pto.ptr<B, l0b>, !pto.ptr<C, l0c>, !pto.ptr<C, bt>, i64, i64, i64

Inputs

Parameter Type Description
%lhs, %rhs, %dst, %m, %n, %k — Same as pto.mad
%bias !pto.ptr<C, bt> Bias vector in BT, interpreted as N values broadcast across M

See MAD Common Clauses for the optional clauses.

Expected Outputs

Result Type Description
None — Writes the produced M x N tile to L0C with bias-init seed.

Side Effects

Engages the CUBE pipe, reads %bias from BT, writes to L0C. The caller is responsible for staging %bias into BT via pto.mte_l1_bt prior to this op.

Constraints

Constraints

  • %bias must be in bt address space.
  • %bias element type must match %dst element type.
  • Only N bias values are consumed; %bias is not an M x N matrix.
  • Other constraints match pto.mad.

Examples

pto.mad_bias %l0a, %l0b, %l0c, %bt, %c16_i64, %c16_i64, %c32_i64
  : !pto.ptr<f16, l0a>, !pto.ptr<f16, l0b>, !pto.ptr<f32, l0c>, !pto.ptr<f32, bt>, i64, i64, i64