TorchLean API

FloatLib.Floats.Formats.Posit.Arithmetic.Limb.Packed.Fma.Runtime

Direct packed-pair posit fused multiply-add runtime #

The packed-pair fused multiply-add kernel returns posit codes for configured values stored in two native limbs. Range and semantic refinement proofs live in Fma.Proof.

@[noinline]

Decode three operands, form the exact product in four native limbs, add without intermediate rounding, and return the result code.

The product crosses to the shared arbitrary-precision sum only once. This avoids allocating two input dyadics and avoids generic big-integer multiplication.

Instances For