Iridescent fresnel, bxdfs #918

keptsecret · 2025-08-27T09:58:12Z

No description provided.

include/nbl/builtin/hlsl/bxdf/fresnel.hlsl

include/nbl/builtin/hlsl/bxdf/reflection/iridescent.hlsl

include/nbl/builtin/hlsl/colorspace.hlsl

include/nbl/builtin/hlsl/colorspace/decodeCIEXYZ.hlsl

include/nbl/builtin/hlsl/bxdf/fresnel.hlsl

…obal conversion matrices

include/nbl/builtin/hlsl/bxdf/fresnel.hlsl

include/nbl/builtin/hlsl/colorspace.hlsl

include/nbl/builtin/hlsl/bxdf/transmission/iridescent.hlsl

devshgraphicsprogramming · 2025-11-03T13:38:35Z

In this PR also resolve all left over comments from #930

…unc for that

devshgraphicsprogramming · 2025-11-12T11:35:02Z

include/nbl/builtin/hlsl/bxdf/base/cook_torrance_base.hlsl

+    static scalar_type __getScaledReflectance(NBL_CONST_REF_ARG(fresnel_type) orientedFresnel, NBL_CONST_REF_ARG(Interaction) interaction, scalar_type clampedVdotH)
+    {
+        spectral_type throughputWeights = interaction.getLuminosityContributionHint();
+        return hlsl::dot<spectral_type>(impl::__implicit_promote<spectral_type, typename fresnel_type::vector_type>::__call(orientedFresnel(clampedVdotH)), throughputWeights);


there shouldn't be an impl::__implicit_promote here at all, the spectral_type and fresnel_type::vector_type should match 100%

include/nbl/builtin/hlsl/bxdf/base/cook_torrance_base.hlsl

devshgraphicsprogramming · 2025-11-12T11:41:54Z

include/nbl/builtin/hlsl/bxdf/ndf/beckmann.hlsl

+            dmq.microfacetMeasure = scalar_type(0.0);
+            dmq.projectedLightMeasure = scalar_type(0.0);


both should be set to bit_cast<scale_type>(numeric_limits<scale_type>::inf) not 0s

devshgraphicsprogramming · 2025-11-12T11:42:02Z

include/nbl/builtin/hlsl/bxdf/ndf/ggx.hlsl

+            dmq.microfacetMeasure = scalar_type(0.0);
+            dmq.projectedLightMeasure = scalar_type(0.0);


both should be set to bit_cast<scale_type>(numeric_limits<scale_type>::inf) not 0s

devshgraphicsprogramming · 2025-11-12T12:16:00Z

include/nbl/builtin/hlsl/bxdf/fresnel.hlsl

-    // TODO: will probably merge with __call at some point
-    static void __polarized(const T orientedEta, const T orientedEtak, const T cosTheta, NBL_REF_ARG(T) Rp, NBL_REF_ARG(T) Rs)
+    static void __polarized(const T orientedEta, const T etaLen2, const T clampedCosTheta, NBL_REF_ARG(T) Rp, NBL_REF_ARG(T) Rs)


cool but this method/function should not be part of Conductor or Dielectric class, can make it free floating in fresnel namespace if you want

include/nbl/builtin/hlsl/bxdf/fresnel.hlsl

include/nbl/builtin/hlsl/math/angle_adding.hlsl

Also flesh out future optimizations for faster angle adding

devshgraphicsprogramming · 2025-11-12T13:07:31Z

include/nbl/builtin/hlsl/bxdf/fresnel.hlsl

        vector_type R12p, R23p, R12s, R23s;
        const vector_type scale = scalar_type(1.0)/eta12;
-        const vector_type cosTheta2_2 = hlsl::promote<vector_type>(1.0) - hlsl::promote<vector_type>(1-cosTheta_1*cosTheta_1) * scale * scale;
+        const vector_type cosTheta2_2 = hlsl::promote<vector_type>(1.0) - hlsl::promote<vector_type>(1.0-cosTheta_1*cosTheta_1) * scale * scale;


this 1.0 should be wrapped in scalar_type(1.0) before - with cos theta

you can declare a const scalar_type unity = 1.0; somwhere in the function to save yourself a headache

devshgraphicsprogramming · 2025-11-12T13:14:48Z

include/nbl/builtin/hlsl/bxdf/fresnel.hlsl


        // Optical Path Difference
-        const vector_type D = hlsl::promote<vector_type>(2.0 * Dinc) * thinFilmIor * cosTheta_2;
+        const vector_type D = hlsl::promote<vector_type>(2.0 * params.getDinc()) * params.getThinFilmIor() * cosTheta_2;


hlsl::promote<vector_type>(2.0 * params.getDinc()) * params.getThinFilmIor() is a single vector which could be precomputed instead of getThinFilmIor()

devshgraphicsprogramming · 2025-11-12T13:19:55Z

include/nbl/builtin/hlsl/bxdf/fresnel.hlsl

        R12s = hlsl::mix(R12s, hlsl::promote<vector_type>(1.0), cosTheta2_2 <= hlsl::promote<vector_type>(0.0));
        R12p = hlsl::mix(R12p, hlsl::promote<vector_type>(1.0), cosTheta2_2 <= hlsl::promote<vector_type>(0.0));

+        R23s = hlsl::mix(R23s, hlsl::promote<vector_type>(0.0), cosTheta2_2 <= hlsl::promote<vector_type>(0.0));
+        R23p = hlsl::mix(R23p, hlsl::promote<vector_type>(0.0), cosTheta2_2 <= hlsl::promote<vector_type>(0.0));


you can do const vector<bool,> tir = cosTheta2_2 <= hlsl::promote<vector_type>(0.0);

btw since you compute it as cosTheta_2 = hlsl::sqrt(hlsl::max(cosTheta2_2, hlsl::promote<vector_type>(0.0))); you can actually compute this condition as cosTheta2_2<=0.0

which in turns is hlsl::promote<vector_type>(1.0-cosTheta_1*cosTheta_1) * scale * scale >= 1.0

and in turn is hlsl::promote<>(1.0-cosTheta_1*cosTheta_1) >= eta12

So you can know the TIR waaay in advance

devshgraphicsprogramming · 2025-11-12T13:20:29Z

include/nbl/builtin/hlsl/bxdf/fresnel.hlsl

+template<typename T, bool SupportsTransmission NBL_PRIMARY_REQUIRES(concepts::FloatingPointLikeVectorial<T>)    
+struct iridescent_base
+{
+    using scalar_type = typename vector_traits<T>::scalar_type;
+    using vector_type = T;
+
+    scalar_type getDinc() NBL_CONST_MEMBER_FUNC { return Dinc; }
+    vector_type getThinFilmIor() NBL_CONST_MEMBER_FUNC { return thinFilmIor; }
+    vector_type getEta12() NBL_CONST_MEMBER_FUNC { return eta12; }
+    vector_type getEta23() NBL_CONST_MEMBER_FUNC { return eta23; }
+    vector_type getEtak23() NBL_CONST_MEMBER_FUNC
+    {
+        NBL_IF_CONSTEXPR(SupportsTransmission)
+            return hlsl::promote<vector_type>(0.0);
+        else
+            return etak23;
+    }

    scalar_type Dinc;       // thickness of thin film in nanometers, rec. 100-25000nm
    vector_type thinFilmIor;


still you are storking etak23 for dielectric iridescent.

don't take SupportsTransmission have etak23 and getEtak23() in your iridescent_base, define them in your Iridescent specializations and actually make them inherit instead of compose

Dinc and thinFilmIoR can be combined into a single member and getter

if you want more semantically clear initialization make creation parameter structs for Iridescent::create

devshgraphicsprogramming · 2025-11-12T13:21:05Z

include/nbl/builtin/hlsl/bxdf/fresnel.hlsl

-    static this_t create(scalar_type Dinc, vector_type ior1, vector_type ior2, vector_type ior3, vector_type iork3)
-    {
-        this_t retval;
-        retval.helper.Dinc = Dinc;
-        retval.helper.thinFilmIor = ior2;
-        retval.helper.eta12 = ior2/ior1;
-        retval.helper.eta23 = ior3/ior2;
-        retval.helper.etak23 = iork3/ior2;
-        return retval;
-    }


where's our create from a creation parameters struct ?

#918 (comment)

devshgraphicsprogramming · 2025-11-12T13:27:25Z

include/nbl/builtin/hlsl/bxdf/fresnel.hlsl

        // Check for total internal reflection
        R12s = hlsl::mix(R12s, hlsl::promote<vector_type>(1.0), cosTheta2_2 <= hlsl::promote<vector_type>(0.0));
        R12p = hlsl::mix(R12p, hlsl::promote<vector_type>(1.0), cosTheta2_2 <= hlsl::promote<vector_type>(0.0));

+        R23s = hlsl::mix(R23s, hlsl::promote<vector_type>(0.0), cosTheta2_2 <= hlsl::promote<vector_type>(0.0));
+        R23p = hlsl::mix(R23p, hlsl::promote<vector_type>(0.0), cosTheta2_2 <= hlsl::promote<vector_type>(0.0));
+
        // Compute the transmission coefficients
        vector_type T121p = hlsl::promote<vector_type>(1.0) - R12p;
        vector_type T121s = hlsl::promote<vector_type>(1.0) - R12s;


actually it would make sense to compute a vector<bool,> tir then do if (all(tir)) path that skips a lot of the calculations (since T121p, T121s, R23s, R23p and cosTheta2 are all zero then)

Also it would make sense to have a notTIR instead of tir boolean (and if (!any(notTIR)) )

because you can make a const vector_type killFactor = vector_type(notTIR); // 0 when TIR, 1 otherwise

and then you can compute R23p, R23s, T121p and T121s as originalMixLHS*killFactor

then obviously you compute R12 from 1.0-T121

devshgraphicsprogramming · 2025-11-12T13:29:13Z

include/nbl/builtin/hlsl/bxdf/fresnel.hlsl

-        const vector_type cosTheta2_2 = hlsl::promote<vector_type>(1.0) - hlsl::promote<vector_type>(1-cosTheta_1*cosTheta_1) * scale * scale;
+        const vector_type cosTheta2_2 = hlsl::promote<vector_type>(1.0) - hlsl::promote<vector_type>(1.0-cosTheta_1*cosTheta_1) * scale * scale;

-        cosTheta_2 = hlsl::sqrt(cosTheta2_2);
+        cosTheta_2 = hlsl::sqrt(hlsl::max(cosTheta2_2, hlsl::promote<vector_type>(0.0)));


a little code-style correctness note, after you clamp cosTheta_2 to not have imaginary numbers due to sqrt you loose the property that cosTheta2_2 == cosTheta_2*cosTheta_2 so you should "hide" cosTheta_2_2 in its own {} scope so its not accessible after the cosTheta_2 assignment

devshgraphicsprogramming · 2025-11-12T13:33:30Z

include/nbl/builtin/hlsl/bxdf/fresnel.hlsl

+    template<typename Params>
+    static T __call(NBL_CONST_REF_ARG(Params) params, const scalar_type clampedCosTheta)
    {
        const vector_type wavelengths = vector_type(colorspace::scRGB::wavelength_R, colorspace::scRGB::wavelength_G, colorspace::scRGB::wavelength_B);


suppy colorspace from iridescent_base last template param (you can default to scSRGB)

include/nbl/builtin/hlsl/bxdf/base/cook_torrance_base.hlsl

devshgraphicsprogramming · 2025-11-12T13:51:29Z

include/nbl/builtin/hlsl/bxdf/base/cook_torrance_base.hlsl

+                quo = hlsl::mix(reflectance / scaled_reflectance,
+                        (hlsl::promote<spectral_type>(1.0) - reflectance) / (scalar_type(1.0) - scaled_reflectance), cache.isTransmission()) * G2_over_G1;


see #918 (comment)

you can simplify thismix to mix(reflectance,hlsl::promote<fresnel_type::vector_type>(1.0)-reflectance,cache.isTransmission())/scaled_reflectance

keptsecret added 10 commits August 26, 2025 11:50

initial iridescent fresnel, only single channel ior

4b3224a

iridescent fresnel does rgb IOR, use in brdf

9c71906

some bug fixes

0f51306

replace loop for vector operations in fresnel

f205d4e

added iridescent btdf

f6daa6f

added unit tests

09402ea

merge bxdfs, fix conflicts

a77c337

latest example

aa0f6e8

moved colorspace transform mats into struct functions

d2cb193

some more colorspace utility

33fc955

devshgraphicsprogramming reviewed Aug 28, 2025

View reviewed changes

include/nbl/builtin/hlsl/bxdf/fresnel.hlsl Outdated Show resolved Hide resolved

devshgraphicsprogramming reviewed Aug 28, 2025

View reviewed changes

include/nbl/builtin/hlsl/bxdf/reflection/iridescent.hlsl Outdated Show resolved Hide resolved

devshgraphicsprogramming reviewed Aug 28, 2025

View reviewed changes

include/nbl/builtin/hlsl/colorspace.hlsl Outdated Show resolved Hide resolved

devshgraphicsprogramming reviewed Aug 28, 2025

View reviewed changes

include/nbl/builtin/hlsl/colorspace.hlsl Outdated Show resolved Hide resolved

devshgraphicsprogramming reviewed Aug 28, 2025

View reviewed changes

include/nbl/builtin/hlsl/colorspace/decodeCIEXYZ.hlsl Show resolved Hide resolved

devshgraphicsprogramming reviewed Aug 28, 2025

View reviewed changes

include/nbl/builtin/hlsl/bxdf/fresnel.hlsl Outdated Show resolved Hide resolved

Base automatically changed from hlsl_bxdfs to master August 29, 2025 13:04

merge latest bxdf_fixes, fix conflicts

0f6c7a2

keptsecret changed the base branch from master to bxdf_fixes_cook_torrance November 3, 2025 04:05

keptsecret added 3 commits November 3, 2025 11:13

restore encode/decode matrices

d8b6773

colorspace specific dominant wavelengths (still missing ACES), use gl…

65239d9

…obal conversion matrices

update iridescent fresnel to match fresnel concept

9a8b77e

Base automatically changed from bxdf_fixes_cook_torrance to master November 3, 2025 13:24

devshgraphicsprogramming reviewed Nov 3, 2025

View reviewed changes