art049/image-rs - CodSpeed

image-rs

Blog Docs Changelog

Performance History

No performance history available yetOnce you have some commits, you will be able to see the performance history of your primary branch.

Latest Results

by

art049

3 days ago

by

art049

3 days ago

Use platform-aware fast rounding for FloatNearest::to_u8/to_u16 Introduce fast_round_f32 that delegates to hardware rounding (roundss/frintn) on SSE 4.1 and aarch64, and falls back to the mantissa snapping trick ((x + 2^23) - 2^23) elsewhere to avoid the costly libm roundf call.

perf/faster-blur

3 days ago

Use platform-aware fast rounding for FloatNearest::to_u8/to_u16 Introduce fast_round_f32 that delegates to hardware rounding (roundss/frintn) on SSE 4.1 and aarch64, and falls back to the mantissa snapping trick ((x + 2^23) - 2^23) elsewhere to avoid the costly libm roundf call.

perf/faster-blur

3 days ago

Replace integer division with reciprocal multiplication in u8 blur Precompute ceil(2^32 / kernel_size) once per pass, then use a multiply-shift to normalize accumulators instead of integer division.

perf/faster-blur

3 days ago

by

art049

3 days ago

by

art049

3 days ago

Use u32 integer accumulators for u8 fast blur The box blur hot path used f32 accumulators for all pixel types. For u8 images (the dominant case), this meant every pixel went through to_f32/from_f32 conversions and software roundf. Replace with u32 integer arithmetic for u8 pixels, dispatched at compile time by pixel size and channel count (1-4). ~1.64x wall-clock speedup on all fast blur benchmarks. wip

perf/faster-blur

3 days ago

Active Branches

Make fast and gaussian blur faster

last run

3 days ago

×5.6

Add CodSpeed continuous performance benchmarking

last run

4 days ago

N/A

© 2026 CodSpeed Technology

Home Terms Privacy Docs