@woosh/meep-engine 3.60.3 → 3.60.5

package/src/shade/renderer/postprocess/ssr/prefilter/NOTES.md

@@ -1,39 +0,0 @@
-# `ssr/prefilter/` — what is live in here and what is not
-
-This directory is a port of AMD's FFX DNSR reflections denoiser. **It is not the denoiser the SSR
-pipeline runs.** Two files in it are, and the other ten are not, which is worth stating because a
-directory that the live path imports from reads as a dependency of the live path.
-
-## Live
-
-`ssr_spatial_denoise_shader.js`, one directory up, is the denoiser `SSR.graph_pass` builds. It is an
-à-trous filter, not FFX's, and it reaches in here for exactly two chunks:
-
-- `chunk_ffx_dnsr_reflections_get_edge_stopping_depth_weight`
-- `chunk_ffx_dnsr_reflections_get_edge_stopping_normal_weight`
-
-Those two are the edge-stopping weights, which are shared between the two denoisers because the
-weight function is the same question either way.
-
-## Not live
-
-Everything else:
-
-- `shader_ssr_prefilter.js` — the compute pass
-- `chunk_ffx_dnsr_reflections_prefilter.js`, `chunk_ffx_dnsr_reflections_resolve.js`
-- `chunk_ffx_dnsr_reflections_get_radiance_weight.js`
-- `NEIGHBOURHOOD_SAMPLE.js`, `chunk_load_neighbourhood_sample.js`
-- the four LDS tiling chunks: `chunk_ffx_dnsr_group_shared_memory`,
-  `chunk_initialize_group_shared_memory`, `chunk_load_from_group_shared_memory`,
-  `chunk_store_in_group_shared_memory`
-
-A grep for any of those symbols outside this directory returns nothing. They are kept because
-`@woosh/meep-engine` publishes raw `src`, so an unreferenced export is still public surface — not
-because anything schedules them.
-
-## If the FFX path is ever revived
-
-It has drifted against the live pipeline in one way that a reader will not notice from inside this
-directory: the à-trous path grew `chunk_ssr_metadata`, and the variance/metadata contract the
-prefilter expects predates it. Reconcile those before wiring the pass in, rather than after.
-

package/src/shade/renderer/postprocess/taa/NOTES.md

@@ -1,11 +0,0 @@
-References:
-
-* ["High Quality Temporal Supersampling" by Brian Karis](https://de45xmedrsdbp.cloudfront.net/Resources/files/TemporalAA_small-59732822.pdf)
-* https://github.com/GameTechDev/TAA/blob/main/MiniEngine/Core/Shaders/TAAResolve.hlsl
-* https://github.com/TheRealMJP/MSAAFilter/tree/master
-* https://www.elopezr.com/temporal-aa-and-the-quest-for-the-holy-trail/
-* https://github.com/google/filament/blob/edece8f3dc0223e445cbbc768fb3e0fa2facd8f1/filament/src/materials/antiAliasing/taa.mat#L97
-
-Check:
-
-* ["Temporally Reliable Motion Vectors for Real-time Ray Tracing" by Zheng Zeng  et al](https://sites.cs.ucsb.edu/~lingqi/publications/paper_trmv.pdf)

package/src/shade/renderer/postprocess/upscale-dither/NOTES.md

@@ -1,78 +0,0 @@
-# Dithered Bilateral Upscale
-
-Depth-aware upscale for low-resolution screen-space effects (volumetric
-lighting / fog, GTAO, soft shadows, etc.). Reconstructs a full-resolution
-image by weighting low-res neighbours by depth similarity, with a per-pixel
-noise-driven tap rotation so the reconstruction error becomes high-frequency
-dither instead of coherent banding.
-
-## Algorithm
-
-Per full-resolution pixel:
-
-1. Linearise the full-res device depth to view-space Z via
-   `chunk_get_view_space_depth`.
-2. Draw a per-pixel blue-noise angle from `HILBERT_R2_INDEX_TEXTURE` using
-   `chunk_spatio_temporal_noise_r2_64`.
-3. Sample 4 low-res taps arranged on a unit square rotated by that angle
-   around the full-res pixel's position expressed in low-res texel space.
-4. Weight each tap by `exp(-|z_tap - z_full| / depth_phi)`
-   (`chunk_depth_edge_stopping_weight`, reused from the SVGF denoiser).
-5. Return the weighted average. If the accumulated weight falls below a
-   small epsilon (every tap failed the depth test), fall back to the tap
-   with the smallest depth delta.
-
-## Inputs the caller must provide
-
-- **Low-res color** - whatever the upscaler is reconstructing
-  (HDR `rgba16float` for volumetrics, `rg16float` moments for GTAO, etc.).
-- **Low-res depth** - view-space (linear) Z at the low-res sample positions.
-  Ideally produced alongside the low-res effect: a raymarching volumetric
-  shader already knows its sample depth, and a dedicated min/max depth
-  downsample shader works for effects that do not. Storing linear Z
-  (not device Z) avoids re-linearising on every tap.
-- **Full-res depth** - the renderer's regular device-space G-buffer depth.
-
-## Expected output format
-
-The default `shader_upscale_dither` writes `rgba16float`. For non-HDR targets
-construct a fresh `ImageShader` from `fragment_shader_upscale_dither` with
-the format you need.
-
-## `depth_phi` tuning
-
-`depth_phi` is the bilateral tolerance in view-space meters. Typical values:
-
-| Effect                  | `depth_phi` |
-|-------------------------|-------------|
-| Volumetric fog (soft)   | 1.0 - 5.0   |
-| GTAO / contact shadows  | 0.1 - 0.5   |
-
-Smaller = stricter depth matching, sharper silhouettes, more fallback hits.
-Larger = softer, risks light leaks across depth edges.
-
-## What this pass does **not** do
-
-- No normal-aware weighting. The original RDR2 variant is depth-only;
-  add normal edge-stopping (see `chunk_normal_edge_stopping_weight`) if a
-  future effect needs it and you have a matching low-res normal buffer.
-- No TAA / temporal accumulation. Pair with the existing TAA pass if the
-  single-frame dither is too visible.
-- No downsample helper. Callers either render the effect at low-res
-  natively (which produces the matching depth for free) or run a dedicated
-  depth downsample pass first.
-
-## References
-
-- Fabian Bauer, *Creating the Atmospheric World of Red Dead Redemption 2:
-  A Complete and Integrated Solution*, SIGGRAPH 2019 Advances in Real-Time
-  Rendering. https://advances.realtimerendering.com/s2019/
-- Angelo Pesce (c0de517e), *Low-resolution effects with depth-aware
-  upsampling*.
-  https://c0de517e.blogspot.com/2016/02/downsampled-effects-with-depth-aware.html
-- pixelmager, depth-aware upsampling gist (summarises the RDR2 variant and
-  its three iterations).
-  https://gist.github.com/pixelmager/a4364ea18305ed5ca707d89ddc5f8743
-- Bart Wronski, *Volumetric Fog* (SIGGRAPH 2014) - earlier prior art on
-  the same family of bilateral-upsample techniques.
-  https://bartwronski.com/wp-content/uploads/2014/08/bwronski_volumetric_fog_siggraph2014.pdf

package/src/shade/renderer/postprocess/vrs/NOTES.md

@@ -1,4 +0,0 @@
-Variable-Rate Shading
-
-*
-see https://github.com/GPUOpen-LibrariesAndSDKs/FidelityFX-SDK/blob/main/sdk/include/FidelityFX/gpu/vrs/ffx_variable_shading.h