Module image 

SIMD-accelerated image operations.

Platform Support

• ARM64: NEON (vsetq_lane_u32 gather + vst1q_u32 store)
• x86_64: SSE2 (_mm_set_epi32 gather + _mm_storeu_si128 store)
• Other: Scalar u32-at-a-time pixel copy

Algorithms

Nearest-neighbor downsample: Processes 4 output pixels per SIMD iteration. Pre-computes x-coordinate mapping table (shared across rows) to avoid per-pixel integer division. Used for thumbhash generation (100x100 max).

Functions

nearest_neighbor_downsample_rgba

Nearest-neighbor RGBA downsample with SIMD acceleration.