Module numeric 

Backend-neutral numeric boundary conversions. Backend-neutral numeric boundary conversions.

Concrete GPU backends cross the same host/API boundaries: host sizes become API u64s, high-resolution timers become telemetry u64s, and device timestamp deltas arrive as rounded floating-point nanoseconds. This module is the single policy for those lossy or fallible conversions; backend crates add only the backend label that makes the diagnostic actionable.

Structs

BackendNumericPolicy

Backend-bound numeric conversion policy.

Constants

BASIS_POINTS_DENOMINATOR

Integer basis-point denominator: 10_000 bps = 100%.

Functions

align_up_u64

Align value upward to alignment, after applying min_value.

align_up_usize

Align value upward to alignment, after applying min_value.

checked_ceil_div_u64

Compute ceil(value / divisor) in u64, returning None for zero divisors or arithmetic overflow.

checked_compose_basis_points_u64

Compose two basis-point multipliers as (left * right) / 10_000, returning None rather than saturating when the composed value cannot fit u64.

checked_dim_product_u32

Multiply three u32 dimensions into a u32 without wraparound.

checked_dim_product_u64

Multiply three u32 dimensions into a u64 without wraparound.

compose_basis_points_u32

Compose two basis-point multipliers as (left * right) / 10_000, with widened arithmetic and loud u32 overflow pinning.

elapsed_nanos_u64

Convert elapsed wall-clock time to telemetry nanoseconds.

finite_f64_ratio_basis_points_round

Convert a finite floating-point ratio into rounded basis points, with separate policies for invalid numerators and denominators.

finite_f64_ratio_basis_points_trunc

Convert a finite floating-point ratio into floor basis points, with separate policies for invalid numerators and denominators.

finite_f64_to_u32_round

Convert a finite non-negative floating-point telemetry value to u32 after rounding to the nearest integer, with loud saturation on invalid input.

finite_f64_to_u32_trunc

Convert a finite non-negative floating-point telemetry value to u32 by truncating toward zero, with loud saturation on invalid or oversized input.

finite_f64_unit_basis_points_trunc

Convert a finite scalar where 1.0 == 10_000 bps into floor basis points.

ratio_basis_points_u64

Compute part / whole as floor basis points with explicit zero-denominator policy and saturating telemetry overflow.

ratio_basis_points_u64_wide

Compute part / whole as floor basis points in a u64 telemetry domain with explicit zero-denominator policy and loud overflow pinning.

ratio_parts_per_million_u64

Compute part / whole as floor parts-per-million with explicit zero-denominator policy and loud u32 overflow pinning.

rounded_f64_to_u64

Round a finite floating-point nanosecond value into telemetry storage.

scale_u64_by_basis_points_floor_min

Apply a basis-point multiplier to a u64 with floor rounding and an output lower bound.

scale_u64_by_basis_points_round_clamped

Apply a basis-point multiplier to a u64 with nearest-integer rounding, optional high clamp, and explicit zero-scale policy.

u128_to_u64

Convert a wide counter to telemetry u64.

usize_to_u64

Convert a host usize to a backend/API u64.

weighted_u64_by_basis_points_u128

Weight a u64 cost by basis points into a widened exact u128 domain.