Module simd 

SIMD vectorization module for high-performance batch operations.

Provides optimized implementations of common mathematical operations for processing 8 values in parallel, matching Tier 2 batch size.

Strategy

Uses Rust’s simd via the wide crate for f64x2-based vectorization. 8 f64 values = 4 f64x2 vectors processed in parallel. Hardware: SSE2 minimum, optimizes for AVX on modern x86_64.

Performance

• True SIMD: Actual vector hardware instructions via wide crate
• f64x2 provides good vectorization across all architectures
• Expected 5-15% speedup from ILP and vectorization
• Trigonometric operations: primary bottleneck, now vectorized
• GPU code already vectorizing color mapping

Functions

simd_abs_8

Vectorized absolute value using vector SIMD

simd_acos_8

Vectorized inverse cosine for 8 f64 values

simd_add_8

Vectorized element-wise addition using vector SIMD

simd_asin_8

Vectorized inverse sine for 8 f64 values

simd_asinh_scale_8

Vectorized asinh scale transformation (handles positive and negative data)

simd_atan2_8

Vectorized inverse tangent (atan2) for 8 point pairs

simd_batch_scale_8

Batch process 8 raw values through scaling operation with caching

simd_batch_scale_16

Batch scale 16 values with validity masking (for 16-pixel rendering)

simd_clamp_8

Vectorized clamp operation using vector SIMD

simd_colormap_sample_8

Vectorized colormap LUT lookup (fast palette sampling)

simd_cos_8

Vectorized cosine for 8 f64 values

simd_cross_8

Vectorized 3D cross product

simd_dot3_8

Vectorized 3D dot product

simd_gamma_correct_8

Vectorized gamma correction for 8 values

simd_linear_scale_8

Vectorized linear scaling (normalization to [0, 1] range)

simd_ln_8

Vectorized natural logarithm

simd_log_scale_8

Vectorized log scale transformation (for positive data)

simd_madd_8

Vectorized fused multiply-add: result = a * b + c using vector SIMD

simd_matvec3_8

Vectorized 3x3 matrix-vector multiplication (8 vectors)

simd_mul_8

Vectorized element-wise multiplication using vector SIMD

simd_normalize_vec3_8

Vectorized 3D vector normalization

simd_plancklog_scale_8

Vectorized PlanckLog scale transformation

simd_pow_8

Vectorized power function (y = x^exp)

simd_recip_8

Vectorized reciprocal (1/x) using vector SIMD

simd_sin_8

Vectorized sine for 8 f64 values

simd_sin_cos_8

Vectorized sine and cosine simultaneously (more efficient than separate calls)

simd_sin_cos_16

Vectorized sin_cos for 16 f64 values

simd_sph_to_vec_8

Vectorized spherical to Cartesian conversion (8 theta-phi pairs)

simd_sqrt_8

Vectorized square root using vector SIMD

simd_symlog_scale_8

Vectorized symlog scale transformation (supports negative values)

simd_to_pixel_values

Convert SIMD linear scale results to PixelValue enum array

simd_to_pixel_values_16

Convert 16 SIMD scaling results to PixelValue array

simd_vec_to_sph_8

Vectorized Cartesian to spherical conversion (8 vectors)