aaronmurniadi/ofxrawhost v0.3.5 on GitHub

What's Changed

Performance enhancements:

ICC profile caching — lcms2 profiles and CMS transforms are now cached per color space instead of rebuilt on every color conversion (per-frame in toDisplayRGBA8, per-export in writeImage)
Hardware half-float conversion — halfToFloat uses F16C intrinsics (_cvtsh_ss) on supported CPUs instead of the software normalization loop
Link-time optimization — LTO is now enabled by default for Release builds via check_ipo_supported
Native architecture optimization — optional -march=native via OFX_NATIVE_ARCH=1 ./build.sh (enables AVX, F16C, etc.)
Multithreaded preview resize — makePreview uses stbir2's split API to parallelize downscaling across CPU cores
Pooled render buffers — renderChain reuses thread-local Image buffers with O(1) swap instead of allocating a new buffer per plugin node per render
Eliminated per-request allocations — clipGetImage reuses a pooled PropSet from the Clip struct instead of new/delete per call
Faster property lookups — PropSet switched from std::map to std::unordered_map for O(1) property access
Faster pixel conversion — replaced std::lround with inline float multiplication + clamp in toTopDown8; flipRows uses std::swap_ranges to avoid a temporary buffer allocation

Full Changelog: v0.3.11...v0.3.5