Crate oximedia_simd 

Hand-written assembly SIMD kernels for OxiMedia

This crate provides highly optimized assembly implementations of critical performance paths in the OxiMedia video codec, including:

• DCT (Discrete Cosine Transform) in various sizes
• Interpolation kernels (bilinear, bicubic, 8-tap)
• SAD (Sum of Absolute Differences) for motion estimation

All assembly is wrapped in safe Rust APIs with proper alignment checks, buffer validation, and runtime CPU feature detection.

Modules

accumulator

Running statistics accumulator for audio/video sample streams.

alpha_premul

Alpha pre-multiplication and un-pre-multiplication helpers.

audio_ops

SIMD-optimized audio buffer operations.

avx512

AVX-512 SIMD implementations for high-throughput media processing.

bitwise_ops

Bitwise SIMD-style operations for multimedia processing.

blend

Alpha blending and compositing operations.

blend_simd

SIMD-optimized blending operations

color_convert_simd

BT.709 RGB ↔ YUV colour conversion using scalar SIMD-friendly code.

color_space

Color space conversion operations optimized for SIMD-friendly access patterns.

convolution

SIMD convolution operations

deblock_filter

SIMD-accelerated deblocking filter for codec post-processing.

dispatch

Runtime-dispatched SIMD wrappers for core media processing operations.

dot_product

Dot product operations for SIMD-accelerated signal processing.

entropy_coding

SIMD helpers for bit packing / unpacking used in entropy coding.

filter

SIMD-friendly image filtering operations.

fixed_point

Fixed-point arithmetic helpers for SIMD-friendly computations.

gather_scatter

SIMD-style gather and scatter memory operations.

histogram

Histogram computation for pixel data analysis.

interleave

Audio channel interleaving and de-interleaving.

lookup_table

Lookup table (LUT) primitives for fast pixel transformations.

math_ops

SIMD-optimized mathematical operations on f32 slices.

matrix

SIMD-friendly 4×4 matrix operations.

min_max

SIMD-optimized minimum and maximum operations for pixel/sample data.

motion_search

SIMD-accelerated motion search using diamond and hexagonal search patterns.

neon

ARM NEON SIMD backend for OxiMedia.

pack_unpack

Pixel packing and unpacking utilities for SIMD pipelines.

pixel_ops

SIMD-optimized pixel / image operations.

portable

Portable SIMD-friendly operations using scalar code structured for auto-vectorization by the compiler.

prefix_sum

Parallel prefix sum (scan) operations for multimedia processing.

psnr

Peak Signal-to-Noise Ratio (PSNR) computation kernel.

reduce

Reduction operations (sum, min, max, mean, variance, normalize) on f32 slices.

resize

SIMD-accelerated image scaling (resize) with bilinear and Lanczos filters.

satd

Sum of Absolute Transformed Differences (SATD) kernel.

saturate

Saturating arithmetic primitives for media pixel pipelines.

simd_bench

Runtime benchmark utilities for oximedia-simd.

ssim

Structural Similarity Index Measure (SSIM) kernel with scalar and SIMD-accelerated implementations.

swizzle

SIMD lane swizzle, permute, and shuffle operations.

threshold

SIMD threshold operations

transpose

Block transpose primitives for matrix and image data.

vector_math

Vector math types: Vec2, Vec3, Vec4.

yuv_ops

YUV colour-space conversion helpers.

Structs

CpuFeatures

CPU features detected at runtime.

Enums

BlockSize

Block sizes for SAD operations

DctSize

DCT transform sizes

InterpolationFilter

Interpolation filter types

SimdError

Error types for SIMD operations

Functions

detect_cpu_features

Detect CPU features at runtime

forward_dct

Perform forward DCT transform

has_neon

Returns true when the executing CPU provides NEON (always true on aarch64).

interpolate

Perform interpolation for motion compensation

inverse_dct

Perform inverse DCT transform

is_aligned

Check if a pointer is properly aligned for SIMD operations

sad

Calculate Sum of Absolute Differences (SAD)

validate_avx2_alignment

Validate buffer alignment for AVX2 (32-byte alignment)

validate_avx512_alignment

Validate buffer alignment for AVX-512 (64-byte alignment)

validate_neon_alignment

Validate buffer alignment for NEON (16-byte alignment)

Type Aliases

Result