Note: This is an archived version of the Blender Developer Wiki (archived 2024). The current developer documentation is available on developer.blender.org/docs.

User:Lukasstockner97/Weekly Reports

2023

Week 30 to 37 (2023-07-25 to 2023-09-18)

  • Was at SIGGRAPH and then on vacation
  • Principled v2:
    • Updated and merged Principled Sheen rework PR (PR #109949)
    • Created, updated and merged Principled layering rework PR (PR #110864)
    • Created, updated and merged Principled Subsurface rework PR (PR #110989)
    • Created, updated and merged Principled Clearcoat rework PR (PR #110993)
    • Created, updated and merged Principled Emission rework PR (PR #111155)
    • Created OSL closure layering bug fix PR (PR #112213)
    • Created Principled Metallic improvement PR (PR #112551)
  • Fixes for OSL vs. SVM differences:
  • Various fixes:
    • Build error after ShaderData change (b19011e2db)
    • Tests error after Sheen change (4bd8f24b81)
    • Unused argument in Microfacet BSDF code (825cc14e74)
    • Wasteful reflection normal calculation when not using Coat (e6296acdba)
    • Unused Normal input on Hair BSDF (5939810b3c)
    • Wasteful Closure Extra allocation for Chiang Hair BSDF (1b92284f86)
  • Reviews:
    • Reviewed (and made minor changes to) Microfacet Hair BSDF PR (PR #105600)
    • Reviewed Multiscatter GGX default PR (PR #111267)
    • Reviewed Microfacet Fresnel calculation refactor PR (PR #112158)
    • Reviewed Principled BSDF node panel PR (PR #112314)
  • Various Pull requests:

Week 28 and 29 (2023-07-11 to 2023-07-24)

  • Merged Sheen BSDF (PR #108869)
  • Created and merged cleanup to shrink ShaderData (PR #110366)
  • Rebased Principled Sheen rework (PR #109949)
  • Merged OSL Clearcoat cleanup (PR #109951)
  • Merged Transmission Roughness removal (PR #109950)
  • Fixed Fresnel handling in MNEE caustics code (8639bf013c)
  • Continued to work on upcoming Principled changes (Specular changes, Subsurface changes)
  • Implemented BSDF layering for GPU OSL and SVN
  • University exam today, so no PR for the last two points yet

Week 25 to 27 (2023-06-20 to 2023-07-10)

  • Reworked the Sheen BSDF PR to no longer be WIP. Now it implements the LTC-based model, with versioning code etc., so it's in a state where it could be merged.
  • Created a WIP PR for replacing the Sheen component of the Principled BSDF, is waiting for the first Sheen PR to be merged.
  • Created a PR for removing the transmission roughness option from the Principled BSDF. Later found an issue with the MNEE caustics code, still needs a fix before merging.
  • Created a PR for simplifying the OSL implementation of the clearcoat component.
  • Pushed a branch to my repo that combines all the PRs above, plus some commits that still need to be finished and split up into PRs:
    • Replacing the Principled Diffuse component with a regular lambertian Diffuse component
    • Removing the Principled Diffuse code, since it's now unused
    • Implementing layering support in OSL
      • This one still needs some work to support it on GPUs as well, the current code is CPU-only
    • Reworking the Principled BSDF to use layer stacking to enforce energy conservation
    • Using precomputed tables for accurately estimating specular layer albedo for correct layering
      • We should eventually have tables for all layerable closures, but for Principled this one and Clearcoat are the two that matter
  • Started working on the Principled BSDF Specular input rework (change Specular Tint to color input, change Specular to non-physical multiplier)
  • Started working on Principled BSDF Subsurface rework (removing SSS Color, changing Subsurface input to scale, adding entry refraction)
  • Reviewed Sharp Distribution cleanup PR

Week 24 (2023-06-13 to 2023-06-19)

  • Merged the #108211 (Glass BSDF glossy/transmission channel split) fix from last week.
  • Fixed #105555 (a longstanding bug related to baking indirect light).
  • Created and merged PR #108955 to fix an unreported bug that caused some direct light (especially from the background) leaking into indirect-only baking.
  • Created PR #108996 to change how the sun lamp (aka distant lamp) in Cycles works, in order to fix #108957. With this change, the brightness of the sun lamp is constant across its full extent, while previous versions would have a darker center at high angles.
  • Updated the Sheen BSDF PR #108869 to also have OSL support. Test builds are available here.

Week 22 and 23 (2023-05-29 to 2023-06-12)

  • Created WIP patch for new sheen models (#108869). One of those will be used for Principled v2, but for now I'm adding both as options to the Velvet BSDF to get feedback on which one is better.
  • Fixed #108560 by adding support for setting light groups for instanced objects (meshes and lamps) through the collection instance. If the original object already has a light group assigned, it continues to use it for its instances.
  • Created a branch to fix #108211. This changes the diffuse/glossy/transmission split logic to be able to have both glossy and transmission contributions from a single closure.
  • Fixed asserts in the Principled BSDF and Velvet BSDF test cases.
  • Merged the MultiGGX removal.

Week 20 and 21 (2023-05-16 to 2023-05-28)

In week 20, I implemented two missing parts of the MultiGGX removal patch: On the one hand, I had forgotten OSL support. Turns out that the OSL microfacet handling could be simplified a lot in general, so in addition to making it work the patch now also removes 10 or so legacy closure types from the OSL interface. On the other hand, the albedo passes (Diffuse/Glossy/Transmission Color) had not worked correctly. The fix is not as elegant as I'd like, but it works and it's the best option to minimize memory requirements that I can see. The patch is ready to be merged now.

Between week 21 and 22, I'm on vacation, so I didn't merge it yet. I did work on the sheen BSDFs before I left, and will post the patch (and merge the MultiGGX patch) in a few days.

Week 19 (2023-05-08 to 2023-05-15)

Finally finished the MultiGGX removal and albedo scaling patch after spending waay too much time getting lost in trying to find an accurate solution for the multibounce Fresnel for the generalized Schlick model. Finally decided that it makes no sense to try and account for the tint factors, which simplifies the whole thing a lot. This was the main thing blocking Principled v2.

Week 18 (2023-05-01 to 2023-05-07)

  • Fixed #72011 (Quantization artifacts when baking bevel normals to 8-bit image)
  • Fixed #107577 (Crash when using denoising with Ashikhmin-Shirley BSDF)
  • Looked into #107580 (Report about the Glossy Color change in 3.6)
  • Updated #104445 (Merging Glossy and Anisotropic BSDF)

Week 4 and 5 (2023-01-23 to 2023-02-05)

  • Fixed T103888 (Side-by-side stereoscopic renders in 3.3.1 ignore color management)
  • Fixed T104041 (Incorrect diffuse color pass for Sheen)
  • Looked into T103876 (AO node is not supported for direct lighting transparency)
  • Looked into T103795 (Emissive meshes and shadowcatcher)
  • Started submitting pieces of the Principled v2 work:
    • Created and merged D17099 (Switching to non-separable shadowing-masking term for Microfacet BSDFs)
    • Created and merged D17100 (Using a faster and exact GGX VNDF sampling algorithm)
    • Created and merged D17101 (Reorganizing Fresnel handling in the Microfacet BSDFs)
    • Created D17123 (Merge Anisotropic BSDF node into Glossy BSDF node)
    • Created D17149 (Use per-microfacet Fresnel term for Glass closures)
    • Created D17211 (Reorganizing Fresnel handling further, implement OSL MaterialX closures)
    • Working on splitting out the next step: Removing the MultiGGX code
  • Created and merged D17107 (Cycles tests comparison helper option)
  • Committed rB329eeacc66d13c9007e22f200304ebe902737abb (Small Cycles closure cleanup)

2022

Week 48 to 3 (2022-11-28 to 2023-01-22)

  • Looked into T72011 (artifacts when baking Bevel normals in Cycles)
    • Created and merged D16727 to fix one of the two issues with Bevel baking
  • Fixed T103067 (Workbench render crash in 3.4)
  • Looked into and closed T103255 (Unexpected behavior when combining Color Attributes with White Noise Texture)
  • Fixed T98951 (Incorrect Depth Pass rendering for Shadow Catcher Objects)
  • Looked into T103327, was already fixed
  • Fixed T103507 (World background contributing to wrong lightgroup)
  • Fixed T103403 (Lightgroups receiving light on shadowcatcher objects)
  • Fixed T103408 (Cycles viewport deadlock)
  • Looked into and closed T103491 (Black areas in Refraction BSDF)
  • Fixed T94698 (Incorrect light passes with volumetric scattering)
  • Fixed T95244 (AO pass broken for shadow catcher objects)
  • Fixed T94752 (Incorrect stereoscopic panoramas)
  • Fixed T89037 (Several Cycles issues with objects with negative scale)
  • Fixed T88849 (Slightly incorrect handling of motion blur shutter curve)
  • Fixed T103786 (Missing viewport update for shadowcatcher self-shadowing)
  • Looked into T102761 (New particles aren't rendered when Vector pass is enabled)
  • Fixed T102920 (Incorrect handling of muted nodes in Cycles)
  • Started creating additional Cycles tests to cover some of the bugs above
  • Looked into updating the Node socket Panel patch, still need to discuss data layout further
  • Started splitting some of the parts of Principled v2 out into separate patches can be reviewed, first up are Microfacet changes and removal of MultiGGX code
  • Working on finishing several changes to Principled v2 so they can be pushed (thin sheet mode, metallic thin film, new sheen model, artistic controls)
  • Christmas break :)

Week 43 to 47 (2022-10-24 to 2022-11-27)

Week 41 and 42 (2022-10-09 to 2022-10-23)

  • Investigated T101270 (Random Object Info not working on instanced curves)
  • Investigated T101263 (Vector pass not being saved correctly when using the File Output node)
  • Fixed T99450 (Animated Holdout not updating on frame change)
  • Fixed T98672 (Noise texture shows incorrect behaviour for large scale values)
  • Fixed T99151 (Regression: cycles not updated when off object types in "View object Types" menu)
  • Investigated T92028 (Cycles: not updating for homogeneous to heterogeneous volume change in viewport render), was fixed already
  • Investigated T92416 (First render of barbershop appears different)
  • Investigated T93187 (Problem mixing Hair shader using color ramp and Hair Info Intercept value), was fixed already
  • Fixed T93382 (Blender still generates subsurface render passes)
    • Ended up being a overall change to the way Blender handles render passes, see D16295
  • Fixed T94136 (Cycles: No Hair Shadows with Transparent BSDF)
  • Fixed T101969 (Crash when using large texture with nonstandard colorspace)
    • Also created D16317 to speed up the image loading here
  • Fixed T101329 (Regression: EXR 'JPG Preview' doesn't use color space anymore)

Week 40 (2022-10-02 to 2022-10-08)

Finally back and able to spend time on Blender properly again!

  • Merged latest master changes (e.g. Spectral data type preparation, ray differentials change) into Principled v2 branch
  • Unified old and new Principled clearcoat BSDFs
  • Started working on more general Fresnel control for Principled v2 (to allow e.g. compatibility with glTF and Spec/Gloss workflow)
  • Fixed T101185 (Crash in the new Mikktspace code for meshes consisting only of invalid triangles)
  • Fixed T101651 (Cycles crash when device initialization fails)
  • Fixed Cycles side of T101702 (Crash when using a boolean input in an OSL script)

Week 31 to 33 (2022-07-25 to 2022-08-21)

  • Vacation! I was away for most of the time, but I did manage to get some stuff in.
  • Created D15589 (Mikktspace C++ port for more performance)
    • Most of the time I had was spent on the initial port as well as tons of cleanup and optimization
    • Also created D15636 for moving the relevant BKE files to C++
  • Looked into bugs T99450 and T99565
  • Continued working on Principled v2
    • Not a lot though, most of the time went into Mikktspace

Week 29 and 30 (2022-07-11 to 2022-07-24)

  • Updated D13522 (Nishita Sky support in Eevee), should be ready for merging now
  • Merged D14740 (Render: Update lightgroup membership in objects and world if lightgroup is renamed)
  • Looked into T98113 (Black dots in Shadow Catcher)
  • Fixed T98367 (Light group passes do not work when shadow catcher is used)
  • Continued working on Principled v2
    • Looked through user feedback, especially regarding Fresnel term
    • Spent more time looking into how others are doing Fresnel controls to ensure broad compatibility
    • Replaced Generalized Schlick Fresnel for the metallic component with Adobe's F82 model
      • Provides more realistic look (darkening at near-grazing angles, full reflection at grazing angles)
      • Compatible with Principled v1 at default (edge color white)
    • Created script for fitting the F82 metallic model to real-world metal measurements
      • Provided values and example renders for several metals in T99447
  • Created D15485, the sampling patch from the last weeks

Week 27 and 28 (2022-06-28 to 2022-07-10)

  • Continued working on Principled v2
    • Published first version of the branch (https://developer.blender.org/diffusion/B/history/principled-v2/)
    • Created T99447, a task for tracking development progress
    • Fixed several bugs (e.g. incorrect glass reflection/transmission pick, black glass at zero roughness)
    • Started working on Iridescence component
    • Looked closer into what others in the field are doing for compatibility
  • Went to EGSR 2022, saw many great papers and talked to many great people
  • Revisited the sampling work from week 22/23, WIP version of the patch will go up soon
    • Significantly reduces noise in some cases
      • The improved stratification helps when individual dimensions are causing the noise (e.g. time dimension for strong motion blur, lens dimensions for DoF, light picking dimension for direct lighting)
      • When the noise comes from the overall light transport (e.g. indirect lighting), the effect is not really noticeable
    • Some details still need to be worked out (e.g. jittering distance controls)

Week 26 (2022-06-20 to 2022-06-27)

  • Continued working on Principled v2
    • Fixed incorrect clearcoat tint logic
      • Spent too much time trying to get an analytical and later numerical approximation to the average exit bounce extinction, ended up just leaving a comment for now
    • Replaced GTR1 distribution for clearcoat with GGX
    • Implemented proper energy conservation for diffuse/SSS and clearcoat layers
  • Started reviewing D15286 (path guiding)
  • Started reviewing current state of D14664) (GSoC multi light sampling)
  • Started reading EGSR 2022 papers

Week 25 (2022-06-13 to 2022-06-19)

  • Continued working on Principled v2
    • Basic components are all working now
    • Cleanup is still needed, but probably good enough to publish as a branch
    • Spent a lot of time on trying to avoid or shrink albedo LUTs, unfortunately with mixed results

Week 24 (2022-06-06 to 2022-06-12)

Week 22 and 23 (2022-05-23 to 2022-06-05)

  • Worked on "Principled v2"
    • Refactored SVM implementation of Principled node to clean up input passing
    • Split SVM implementation of old Principled BSDF into components (diffuse+SSS, specular, glass etc.)
    • Started implementing components of new Principled BSDF
  • Spent some time looking into Cycles sampling improvements
    • Better hash function for reordering samples, preserving 2D stratification for PMJ during reordering, proper hash-based Owen scrambling, padding Sobol sequence to avoid higher dimension vectors...
    • Got some promising results but needs further investigation
    • Will also look into screen-space blue noise
  • Ported Mikktspace tangent code to C++
    • Allows to simplify code considerably
    • Currently ~30% faster
    • New code opens up parallelization opportunities
    • Many additional small optimizations possible (e.g. SIMDify vector operations, use saved memory to cache normals, ...)
    • Should also look into parallelizing tangent computation across meshes in Cycles since this step does not make use of the Blender API

Week 21 (2022-05-16 to 2022-05-22)

  • Spent the entire week on-site in Amsterdam
  • Cleaned up the Microfacet code further to improve maintainability
  • Replaced Multi-scattering GGX with a albedo-scaling factor on top of the regular GGX closures
  • Worked on design for "Principled v2" (placeholder name) and started implementation
  • Still need to create a design task and branch

Week 20 (2022-05-09 to 2022-05-15)

  • Not a lot happened this week, I spent more time on university so I can spend the entire next week on Blender while I'm in Amsterdam.
  • Mostly researched options for the Principled BSDF work
    • The entire thing might be growing from a few small fixes into a Principled v2 at this point
    • Goal would be to "catch up" to improvements in the physically-based shading space over the last few years
    • Tricky part will be balancing flexibility and power vs. ease of use (avoid having 50 inputs...)

Week 19 (2022-05-02 to 2022-05-08)

  • Fixed T97169, then realized that a commit with the same fix (for a different bug) had already been committed a few days ago and I hadn't updated my checkout :(
  • Looked into some other open bugs (T96606, T96719, and a bunch of others related to Multiscattering GGX)
  • Created D14889 (GGX Microfacet refactor)
    • Not much changes from the user perspective, just slightly brighter reflections at grazing angles due to a more accurate shadowing-masking function
    • Also basically no impact on performance or noise, but makes the code easier to understand and maintain
    • Finally produces expected results for approximate multiscattering precomputation (see Week 17)
  • Got initial version of the two approximate multiscattering approaches working, will compare further
    • Still needs glass version and color handling to properly replace the current multiscattering code
    • Will create test build when it's ready for user feedback

Week 18 (2022-04-25 to 2022-05-01)

  • Reviewed D14754 (LRU cache for dynamic Cycles kernels)
  • Fixed T96576 (Light leaks with Multiscatter GGX and normal maps)
  • Created D14815 to fix T95644 (Object Attributes not updating when rendering on GPU)
  • Started to fix T96718, currently tracked it down to an uninitialized stack read in the SVM code of a shader
  • Looked into the GGX issue mentioned last week a bit more, nothing solid yet. Might be related to different versions of the Smith shadowing term?
  • More planning regarding Principled BSDF improvements, will write proper description soon.

Week 17 (2022-04-18 to 2022-04-24)

  • Merged D14675 (Speeing up tangent space computation)
  • Created D14735, fixes T97386 (Nodes not draggable when clicking on labels since the tooltip change)
  • Created D14740 (Updating lightgroup membership of objects and world when renaming lightgroups)
  • Started to work on Principled BSDF improvements:
    • Created precomputation code for the approximate multiscattering code, noticed discrepancy between Cycles' GGX and other renderers at grazing angles. Need to investigate further.
    • Implemented improved GGX VNDF importance sampling code, doesn't make much difference for noise or render time but is simpler.

Week 16 (2022-04-11 to 2022-04-17)

Week 15 (2022-04-04 to 2022-04-10)

  • Reviewed D14395 (UDIM packing), ready for merge now.
  • Updated T78008 (Light group improvement collection)
  • Merged D14540 (Light Group UI improvement: Adding light groups from object/world properties)
  • Fixed T97159 (Color AOV passes missing their alpha pass since Blender 3.0)
  • Created D14596 (Light group improvement: Operators for adding all used or removing all unused lightgroups)
  • Started working on showing lightgroups in viewport rendering
  • Updated D9967 (Node socket tooltip support), much cleaner implementation now.

Week 14 (2022-03-28 to 2022-04-03)

  • Reviewed, polished and merged D12871 (Cycles Light Groups)
  • Reviewed and merged D14527 (Fix for Light Groups on lamps)
  • Fixed T96978 (Emitters themselves not showing up in Light Groups)
  • Reviewed D14395 (UDIM packing)
  • Created D14540 (Light Group UI improvement: Adding light groups from object/world properties)

Week 13 (2022-03-21 to 2022-03-27)

  • Continued to catch up on code changes (e.g. C++ification of the core)
  • Updated D13522 (Nishita Sky support in Eevee) to address review comments.
  • Updated D9967 (Tooltip support for node sockets) to address review comments
    • Uses a different approach now to reduce hackiness and also supports tooltips in the Material Properties tab
    • Also spent way too much time digging through the UI code trying yet another approach that went nowhere...
  • Reviewed D14327 (Preparation for UDIM packing)
  • Reviewed D14395 (UDIM packing)
  • Started reviewing D12871 (Cycles Light Groups)

Week 12 (2022-03-14 to 2022-03-20)

  • Went through the onboarding process
  • Set up my local development setup properly again
  • Caught up on the development process changes and the Cycles-X architecture (not fully happy with my understanding yet, will continue, probably by fixing bugs)
  • Updated D13522 (Nishita Sky support in Eevee) and prepared compatibility with rewrite branch
  • Updated D9967 (Tooltip support for node sockets) and integrated the existing work on Geometry Node introspection. The actual tooltip content and showing them in the Material Properties tab is still TODO.