Internal raw-double compute signatures for SIMD batch kernels. More...

#include <cstddef>

Namespaces
namespace	srfm

namespace	srfm::simd

namespace	srfm::simd::detail

Macros
#define	SRFM_RESTRICT __restrict__

Functions
void	srfm::simd::detail::compute_beta_scalar (const double __restrict__ velocities, std::size_t n, double &running_max, double __restrict__ out) noexcept
	Scalar reference implementation of the beta batch kernel.

void	srfm::simd::detail::compute_beta_avx2 (const double __restrict__ velocities, std::size_t n, double &running_max, double __restrict__ out) noexcept
	AVX2 (256-bit, 4-wide) beta batch kernel.

void	srfm::simd::detail::compute_beta_avx512 (const double __restrict__ velocities, std::size_t n, double &running_max, double __restrict__ out) noexcept
	AVX-512F (512-bit, 8-wide) beta batch kernel.

void	srfm::simd::detail::compute_gamma_scalar (const double __restrict__ betas, std::size_t n, double __restrict__ out) noexcept
	Scalar reference implementation of the gamma batch kernel.

void	srfm::simd::detail::compute_gamma_avx2 (const double __restrict__ betas, std::size_t n, double __restrict__ out) noexcept
	AVX2 (256-bit, 4-wide) gamma batch kernel.

void	srfm::simd::detail::compute_gamma_avx512 (const double __restrict__ betas, std::size_t n, double __restrict__ out) noexcept
	AVX-512F (512-bit, 8-wide) gamma batch kernel.

Detailed Description

Internal raw-double compute signatures for SIMD batch kernels.

Module: src/simd/ (internal — do NOT include from public headers) Owner: AGT-08 — 2026-03-01

Responsibility

Declare the low-level SIMD and scalar kernel signatures that operate on raw double pointers. These are implementation details; callers should use the public API in include/srfm/simd/simd_dispatch.hpp instead.

Kernels

Beta batch: Computes beta_i = |velocities[i]| / running_max for all i in [0, n). running_max is updated in-place: at each chunk boundary it is set to max(running_max, max(|velocities[chunk]|)). This guarantees: • beta_i ∈ [0.0, BETA_MAX_SAFE] always • running_max is monotonically non-decreasing across calls

Gamma batch: Computes gamma_i = 1.0 / sqrt(1.0 − betas[i]²) for all i in [0, n). Input betas must be pre-validated (0 ≤ betas[i] ≤ BETA_MAX_SAFE). Clamping is applied before sqrt to prevent NaN.

Guarantees

• All functions are noexcept. • Tail elements (n % LANE != 0) are handled with a scalar fallback. • Output pointers must be writable for n doubles; they need not be aligned (all loads/stores use unaligned intrinsics). • Input and output pointers must not alias (SRFM_RESTRICT).

NOT Responsible For • Wrapping outputs into BetaVelocity / LorentzFactor (dispatch layer). • Selecting the correct kernel at runtime (simd_dispatch.cpp).

Definition in file simd_batch_detail.hpp.

Macro Definition Documentation

◆ SRFM_RESTRICT

#define SRFM_RESTRICT __restrict__

Definition at line 50 of file simd_batch_detail.hpp.

Namespaces