Hlsl path tracer by devshgraphicsprogramming · Pull Request #224 · Devsh-Graphics-Programming/Nabla-Examples-and-Tests

devshgraphicsprogramming · 2025-11-03T16:17:03Z

@keptsecret after you merge master again, you'll probably have the UI mess up and show changed from the merge commit, so close and reopen again

TODO

4 unresolved conversations from RWMC #218, check if done

31_HLSLPathTracer/app_resources/hlsl/render.comp.hlsl

31_HLSLPathTracer/app_resources/hlsl/example_common.hlsl

31_HLSLPathTracer/main.cpp

…ate so can make lights static const array members

…(12 bitfield) with invalid id

devshgraphicsprogramming · 2026-02-25T07:31:51Z

31_HLSLPathTracer/app_resources/hlsl/next_event_estimator.hlsl

+        if (hlsl::isinf(pdf))
+        {
+            retval.quotient_pdf = quotient_pdf_type::create(hlsl::promote<spectral_type>(0.0), 0.0);


@keptsecret you need a very good comment to justify this, because normally thats a point light

So write something about PDF being inf when SolidAngle is 0, so the quotient of finite area emission would be 0 due to division by int

also comment that the "PDF" getting returned is really a thing for a MIS weight

devshgraphicsprogramming · 2026-02-25T07:34:34Z

31_HLSLPathTracer/app_resources/hlsl/scene_rectangle_light.hlsl

+        ray_dir_info_t V;
+        V.setDirection(-ray.direction);
+        interaction_type interaction = interaction_type::create(V, N);
+        interaction.luminosityContributionHint = hlsl::normalize(colorspace::scRGBtoXYZ[1] * ray.getPayloadThroughput());


this is a good default, just rmemeber that our Unidirectional PT will overwrite it anyway

devshgraphicsprogramming · 2026-02-25T07:35:03Z

31_HLSLPathTracer/app_resources/hlsl/scene_base.hlsl

+using spectral_t = vector<float, 3>;
+using bxdfnode_type = BxDFNode<spectral_t>;
+
+static const bxdfnode_type bxdfs[SceneBase::SCENE_BXDF_COUNT] = {
+    bxdfnode_type::create(MaterialType::DIFFUSE, false, float2(0,0), spectral_t(0.8,0.8,0.8)),
+    bxdfnode_type::create(MaterialType::DIFFUSE, false, float2(0,0), spectral_t(0.8,0.4,0.4)),
+    bxdfnode_type::create(MaterialType::DIFFUSE, false, float2(0,0), spectral_t(0.4,0.8,0.4)),
+    bxdfnode_type::create(MaterialType::CONDUCTOR, false, float2(0,0), spectral_t(1.02,1.02,1.3), spectral_t(1.0,1.0,2.0)),
+    bxdfnode_type::create(MaterialType::CONDUCTOR, false, float2(0,0), spectral_t(1.02,1.3,1.02), spectral_t(1.0,2.0,1.0)),
+    bxdfnode_type::create(MaterialType::CONDUCTOR, false, float2(0.15,0.15), spectral_t(1.02,1.3,1.02), spectral_t(1.0,2.0,1.0)),
+    bxdfnode_type::create(MaterialType::DIELECTRIC, false, float2(0.0625,0.0625), spectral_t(1,1,1), spectral_t(1.4,1.45,1.5)),
+    bxdfnode_type::create(MaterialType::IRIDESCENT_CONDUCTOR, false, 0.0, 505.0, spectral_t(1.39,1.39,1.39), spectral_t(1.2,1.2,1.2), spectral_t(0.5,0.5,0.5)),
+    bxdfnode_type::create(MaterialType::IRIDESCENT_DIELECTRIC, false, 0.0, 400.0, spectral_t(1.7,1.7,1.7), spectral_t(1.0,1.0,1.0), spectral_t(0,0,0)),
+    bxdfnode_type::create(MaterialType::EMISSIVE, LightEminence)
+};


erm, shouldn't the table of bxdfs be in the material system and not the scene ?

devshgraphicsprogramming · 2026-02-25T07:35:12Z

31_HLSLPathTracer/app_resources/hlsl/scene_base.hlsl

+
+    NBL_CONSTEXPR_STATIC_INLINE uint32_t SCENE_SPHERE_COUNT = 10u;
+    NBL_CONSTEXPR_STATIC_INLINE uint32_t SCENE_LIGHT_COUNT = 1u;
+    NBL_CONSTEXPR_STATIC_INLINE uint32_t SCENE_BXDF_COUNT = 10u;


material system ?

the BXDF count that is

devshgraphicsprogramming · 2026-02-25T07:39:41Z

31_HLSLPathTracer/app_resources/hlsl/material_system.hlsl

+                ray_dir_info_type L;
+                L.makeInvalid();
+                return sample_type::create(L, hlsl::promote<vector3_type>(0.0));
+            }
+        }
+
+        ray_dir_info_type L;
+        L.makeInvalid();
+        return sample_type::create(L, hlsl::promote<vector3_type>(0.0));


put a break in the switch, or remove the unreachable code after the switch

devshgraphicsprogramming · 2026-02-25T07:40:03Z

31_HLSLPathTracer/app_resources/hlsl/material_system.hlsl

+                return iridescentDielectricBxDF.generate(interaction, u, _cache);
+            }
+            break;


if you return from a case label, you don't need to break, Clang actually warns you about that in C++

devshgraphicsprogramming · 2026-02-25T07:41:35Z

31_HLSLPathTracer/app_resources/hlsl/material_system.hlsl

+    DiffuseBxDF diffuseBxDF;
+    ConductorBxDF conductorBxDF;
+    DielectricBxDF dielectricBxDF;
+    IridescentConductorBxDF iridescentConductorBxDF;
+    IridescentDielectricBxDF iridescentDielectricBxDF;


compiler and perf will hate you because these are live variables.

I think I have a solution for this

see Devsh-Graphics-Programming/Nabla#991 (comment)

Basically make the "cache" more generic, and instead of it just being an aniso cache, also slap the BxDF in there so it doesn't need to be live state here.

Obviously this means some very light weight serialization to emulate a union

I wish we could have KHR_shader_untyped_pointers that work on any storage class, then unions would be easy
KhronosGroup/Vulkan-Docs#2684

…DF + add config for that, refactor stuff around ray and nee methods changes in concept

devshgraphicsprogramming · 2026-02-25T13:51:39Z

31_HLSLPathTracer/app_resources/hlsl/next_event_estimator.hlsl

+        }
+        else
+        {
+            retval.quotient_pdf = quotient_pdf_type::create(1.0, 0.0);


why when the sample is invalid or below the hemisphere are you returning 1 for the quotient?

devshgraphicsprogramming · 2026-02-25T13:52:23Z

31_HLSLPathTracer/app_resources/hlsl/next_event_estimator.hlsl

+    }
+
+    light_type lights[scene_type::SCENE_LIGHT_COUNT];
+    uint32_t lightCount;


make it a constant, not a variable, will speed us up a lot

basically remove it and use scene_type::SCENE_LIGHT_COUNT in its place

devshgraphicsprogramming · 2026-02-25T14:23:53Z

31_HLSLPathTracer/app_resources/hlsl/next_event_estimator.hlsl

+        if (hlsl::isinf(pdf))
+        {
+            retval.quotient_pdf = quotient_pdf_type::create(hlsl::promote<spectral_type>(0.0), 0.0);
+            return retval;
+        }
+
+        if (retval.sample_.getNdotL() > numeric_limits<scalar_type>::min && retval.sample_.isValid())
+        {


also you could join the failure cases if you write enough comments

if (pdf < Inf && NdotL > min && retval.sample_.isValid())

normally you'd have to do conditionalMaxOrAbs(NdotL,0.0f,isBSDF) > min because BSDFs should receive light from the backside, and its only in this scene that they can't because the scene has watertight geometry everywhere

P.S. do you even need to check if the sample is valid, I thought invalid samples produce NaN or 0

so I guess first check if the pdf is non-zero and finite (with comment about finiteness being due to area lights) and the NdotL (with a comment)

then create the sample and do the rest

keptsecret added 30 commits February 3, 2025 14:52

copied ex 30 to new ex 31

eab0f70

ignore events on imgui focus

8521123

initial files for pathtracer

b171724

intersection logic

af35393

changes to intersection logic

5a5fbfe

completed material system?

85e67ad

sphere nee stuff

2c500b1

triangle sampling

e6a9916

rectangle sampling

6dc25eb

nee stuff

22bd6f9

scene representation, getmeasure for pt

3bb858b

moved scene rep out, some closest hit stuff

f5adbf6

sample light part of closest hit

159d153

fix bugs, reorganize traceRay

a7350db

sample bsdf in closest hit

8a4e0a9

set up path tracer render shader

72104b8

finish render shader

202c645

hlsl path tracer initial, bug fixes

2f77555

fix shader bugs

99aed47

Merge branch 'master' into hlsl_path_tracer

a1a8ec0

bug fixes #3

a802a97

fix include when embed resources off

eed47e7

fixed more bugs #4

6e26dae

Merge branch 'master' into hlsl_path_tracer

3827fd3

fix compile hlsl shader bug

da661c0

more bug fixes #5

f97757b

more bug fixes #6

8e759f2

Merge branch 'master' into hlsl_path_tracer

2dfadfe

refactor to use new frisvad

b1831d9

fix bugs again

cb5662a

put everything into the same descriptor set

6a19568

devshgraphicsprogramming commented Feb 24, 2026

View reviewed changes

31_HLSLPathTracer/app_resources/hlsl/render.comp.hlsl Outdated Show resolved Hide resolved

devshgraphicsprogramming commented Feb 24, 2026

View reviewed changes

31_HLSLPathTracer/app_resources/hlsl/example_common.hlsl Outdated Show resolved Hide resolved

devshgraphicsprogramming commented Feb 24, 2026

View reviewed changes

31_HLSLPathTracer/app_resources/hlsl/example_common.hlsl Outdated Show resolved Hide resolved

devshgraphicsprogramming commented Feb 24, 2026

View reviewed changes

31_HLSLPathTracer/main.cpp Outdated Show resolved Hide resolved

keptsecret added 6 commits February 24, 2026 15:08

split different scenes into different hlsl files, made them not templ…

3b85362

…ate so can make lights static const array members

make ray templated on payload

5e207c8

comments to clarify randgen parameters

b18c162

removed intersect mode because it's pointless, made object id 16 bit …

c861674

…(12 bitfield) with invalid id

refactor use splatting param create

2b4c44e

refactor scene to return interaction

a02c060

devshgraphicsprogramming commented Feb 25, 2026

View reviewed changes

refactor intersector, nee concept changes

171aa95

devshgraphicsprogramming commented Feb 25, 2026

View reviewed changes

keptsecret added 2 commits February 25, 2026 16:45

added custom interactions structs that stores isInteractionMaterialBS…

d9bea1f

…DF + add config for that, refactor stuff around ray and nee methods changes in concept

fix normalize throughputCIE, refactor changes to method names

07c43a6

devshgraphicsprogramming commented Feb 25, 2026

View reviewed changes

keptsecret added 6 commits February 26, 2026 12:32

account for if otherTechniqueHeuristic is 0

480e73a

refactor mat system takes in aniso types

666f911

refactor material system pdf method, ray payload shouldDoMIS method

8f5ccf6

refactor scene getIntersection method

66bface

refactor minor name changes

80b0801

refactor method for intersection data to get geometric normal

ebff013

Conversation

devshgraphicsprogramming commented Nov 3, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

TODO

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

devshgraphicsprogramming Feb 25, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

5 participants

devshgraphicsprogramming commented Nov 3, 2025 •

edited

Loading

devshgraphicsprogramming Feb 25, 2026 •

edited

Loading