Download File

## New Features * Hand light now has screen space ray traced shadows. * A new Hand Light shader pack options screen has been added to control this feature.  ## Improvements * Hand light is now handled separately for each hand. * Hand light sources are now positioned more closely to actual hand positions. * Significant improvements have been made that depend on the **Subgroup Operations** Compatibility option. * Lights are now de-duplicated using subgroup operations before being added to the global light list, significantly reducing the capacity required to avoid overflow and the time spent on global de-duplication. * Work group bounding box computations are now accelerated using subgroup operations. * Atomically pushing to buffers is now accelerated using subgroup operations. * SSBO and shared memory loads are now explicitly made subgroup uniform where possible, using subgroup operations. * Volumetric light and SMAA neighborhood blending are now significantly faster thanks to optimizations made to work group neighborhood handling. * Chunks are now culled in uniform control flow. * Light list creation is now slightly faster when 16-bit integer types are unsupported. * Sky rendering is now slightly faster, by skipping a work group barrier in tiles with no block lit geometry. * Gbuffer data is now ordered more optimally following the [RDNA Performance Guide](https://gpuopen.com/learn/rdna-performance-guide/). ## Other Changes * Amethyst Shards are now emissive. * Ambient occlusion is now more visible. *More improvements to ambient occlusion are planned.* * Hand light brightness now matches block light by default. * Work group sizes have been retuned and optimized for modern AMD Radeon hardware. **This could result in worse performance on some other hardware. Please report any significant issues.** * Subgroup operations are now assumed by default to be supported on all Radeon hardware. * Immutable constants are now assumed by default to be supported on all Radeon hardware on Linux. * A driver bug workaround for the work group size of the global light list de-duplication pass has been disabled on Mesa graphics drivers for Radeon. This could improve performance slightly in some situations. ## Fixed Issues * Light list sorting fails when multiple lights are at the exact same distance from the player. * Unlit geometry is rendered with incorrect depth guarantees, causing undefined behavior. * Fallback functions collide with disabled built-in 16-bit functions on AMD graphics drivers on Windows (driver bug). [GitHub Release](https://github.com/Luracasmus/renderpearl/releases/tag/v2.7.0)