Configured posit backend runtime #
The posit planner selects from representation-independent exact-dyadic backends, two-limb
model-valued backends, and direct packed-storage kernels that read UInt64 words or UInt64
pairs.
Packed carriers pair a code with a proof that it lies below the format modulus. Their constructors
use the arithmetic kernels' range theorems; these proof arguments are erased during compilation.
Semantic refinement of the configured entry points is proved in
FloatLib.Floats.Formats.Posit.Configured.Backend.Proof.
Representation-independent exact-dyadic kernels #
Exact-dyadic addition backend.
Finite inputs are aligned, added, and rounded entirely as signed binary integers.
Instances For
Exact-dyadic subtraction backend.
Instances For
Exact-dyadic multiplication backend.
Instances For
Width-generic quotient-prefix division backend.
Instances For
Rational-free square-root backend.
The kernel is DirectDyadicSquareRoot: one destination-width integer root prefix, with the exact
square remainder jammed into the sticky bit before the single rounding step.
Instances For
Exact-dyadic fused multiply-add backend with one final rounding step.
Instances For
Representation-independent two-limb kernels #
Two-limb-rounded addition for formats of at most 128 bits.
Instances For
Two-limb-rounded subtraction for formats of at most 128 bits.
Instances For
Two-limb-rounded multiplication for formats of at most 128 bits.
Instances For
Two-limb-rounded fused multiply-add for formats of at most 128 bits.
Instances For
Direct built-in packed-storage kernels #
These are the native kernels: addWordsCodeFlatValid, subWordsCodeFlatValid,
fmaWordsCodeFlatValid, PackedProduct.mulWordsCodeValid, PackedQuotient.divWordsCodeValid,
and PackedSquareRoot.sqrtWordCodeValid operate on UInt64 words and consume the eligibility
witness.
Native-word addition that reads built-in packed storage directly.
Unlike dyadicAdd, this entry point does not reconstruct Model operands. Its NativeCode
capability is available only for built-in UInt8/UInt16/UInt32/UInt64 plans.
Instances For
Direct built-in packed-storage subtraction.
Instances For
Direct built-in packed-storage multiplication.
Instances For
Direct built-in packed-storage division.
Instances For
Direct built-in packed-storage square root.
Instances For
Direct built-in packed-storage fused multiply-add.
Instances For
Direct two-limb packed-storage kernels #
Two-limb addition that reads and returns .pair storage directly.
Instances For
Direct two-limb packed-storage subtraction.
Instances For
Direct two-limb packed-storage multiplication.
Instances For
Width-generic quotient-prefix division with direct .pair operand decoding.
This is only a storage adapter: it removes the pair-to-model round trip while retaining the same
DirectDyadicQuotient.round kernel used by arbitrary-width Posits.
Instances For
Direct two-limb packed-storage square root.
Instances For
Direct two-limb packed-storage fused multiply-add.