Bxdf fixes cook torrance #930

devshgraphicsprogramming · 2025-09-16T14:03:10Z

Description

Continues #899 , #916 and #919

Testing

TODO list:

…eturn quants

include/nbl/builtin/hlsl/bxdf/common.hlsl

include/nbl/builtin/hlsl/math/functions.hlsl

devshgraphicsprogramming · 2025-09-16T14:11:56Z

include/nbl/builtin/hlsl/cpp_compat/vector.hlsl

+#ifndef __HLSL_VERSION
+namespace std
+{
+template<typename T, uint16_t N>
+struct hash<nbl::hlsl::vector<T,N> >
+{
+    size_t operator()(const nbl::hlsl::vector<T,N>& v) const noexcept
+    {
+        size_t seed = 0;
+        NBL_UNROLL for (uint16_t i = 0; i < N; i++)
+            nbl::core::hash_combine(seed, v[i]);
+        return seed;
+    }
+};
+
+template<typename T>
+struct hash<nbl::hlsl::vector<T,1> >
+{
+    size_t operator()(const nbl::hlsl::vector<T,1>& v) const noexcept
+    {
+        std::hash<T> hasher;
+        return hasher(v.x);
+    }
+};
+}
+#endif


are you sure having this specialization globally is a good idea ?

What do you use it for ?

Also used in the bucket bxdf test, for hash map buckets

then define it in that example, nowhere else

I moved this over from the test due to this comment
Devsh-Graphics-Programming/Nabla-Examples-and-Tests#165 (comment)

ok but don't make a specialization of std::hash thats binding for everyone (different struct and name), esp that hlsl::vector is a typedef of glm::vec[1,2,3,4] and we're bound to introduce conflicts there with our future integrations

Removed it and used glm's std::hash instead. Dunno why I missed that initially

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

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

devshgraphicsprogramming · 2025-10-03T13:41:55Z

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

+        struct reflect_wrapper
+        {
+            vector3_type operator()() NBL_CONST_MEMBER_FUNC
+            {
+                return r(VdotH);
+            }
+            bxdf::Reflect<scalar_type> r;
+            scalar_type VdotH;
+        };
+        reflect_wrapper r;
+        r.r = bxdf::Reflect<scalar_type>::create(localV, localH);
+        r.VdotH = cache.getVdotH();


write a comment that we're doing all these gymnastics to have reflect use an externally computed VdotH such that it doesn't compute it itself again

devshgraphicsprogramming · 2025-10-03T13:47:27Z

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

+        reflect_wrapper r;
+        r.r = bxdf::Reflect<scalar_type>::create(localV, localH);
+        r.VdotH = cache.getVdotH();
+        ray_dir_info_type localL = localV_raydir.template reflect<reflect_wrapper>(r);


the whole point of the reflect_wrapper was to avoid turning the reflection of V around H into a three step process:

transforming fat V Ray-Dir into tangent space

reflecting V around H to get fat L

transforming tangent space fat L into original space of fat V

Note that we can do reflections and refractions in any space we like, so instead of bringing V to tangent and L back out of it, one can just bring H to fat V's space, and do the refraction using non-local V and H because cache.VdotH remains invariant under rotations.

(this is why we get our reflection wrapper to override the computation of VdotH, its not something compiler CSE can spot being identical)

This is not only 1 transformation instead of 2, but also a muuch cheaper transformation (always a 3x3 matrix mul with H a vector, not a transformation of a direction vector and its derivatives or covariance matrices).

I know that creating a sample from worldspace L requires 3 dot products with TBN which is same as creating a sample from tangentspace L (TBN dot products already given) and transforming it into worldspace

HOWEVER:

in isotropic BRDF usage, the TdotL and BdotL may be thrown away

when derivatives and covariance matrices come, taking worldspace ray-dir info and doint 3 dot products will be cheaper

Also the NdotL computation could be overwritten by 2.f*cache.getVdotH()*localH.z - localV.z if you want (the quantitiy you check to see if the sample has invalid path).

devshgraphicsprogramming · 2025-10-03T13:50:15Z

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

+        const vector3_type localH = ndf.generateH(upperHemisphereV, u.xy);
+        const vector3_type H = hlsl::mul(interaction.getFromTangentSpace(), localH);
+
+        const scalar_type VdotH = hlsl::dot(V.getDirection(), H);


compute this as dot(localV,localH) before computing H, it will shorten the instruction dependency cycle

~~also as soon as you know VdotH you can check its sign's consistency with NdotV, and reject the sample if it hits a microfacet from a different side than the surface~~

You can assert it, VNDF sampling should guarantee this

devshgraphicsprogramming · 2025-10-03T13:53:28Z

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

+        fresnel_type _f = impl::getOrientedFresnel<fresnel_type, IsBSDF>::__call(fresnel, interaction.getNdotV());
+        fresnel::OrientedEtaRcps<monochrome_type> rcpEta = _f.getOrientedEtaRcps();
+
+        ray_dir_info_type V = interaction.getV();
+        const vector3_type localV = interaction.getTangentSpaceV();
+        const vector3_type upperHemisphereV = ieee754::flipSignIfRHSNegative<vector3_type>(localV, hlsl::promote<vector3_type>(interaction.getNdotV()));


assign interaction.getNdotV() to a const variable, or better use localV.z in its stead

devshgraphicsprogramming · 2025-10-03T13:53:51Z

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

+        fresnel_type _f = impl::getOrientedFresnel<fresnel_type, IsBSDF>::__call(fresnel, interaction.getNdotV());
+        fresnel::OrientedEtaRcps<monochrome_type> rcpEta = _f.getOrientedEtaRcps();
+
+        ray_dir_info_type V = interaction.getV();


you don't need this till much later

devshgraphicsprogramming · 2025-10-03T14:01:15Z

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

+        Refract<scalar_type> r = Refract<scalar_type>::create(V.getDirection(), H);
+        bxdf::ReflectRefract<scalar_type> rr;
+        rr.refract = r;
+        ray_dir_info_type L = V.reflectRefract(rr, transmitted, rcpEta.value[0]);


use a wrapper same as reflect to override operator()(const bool doRefract, const scalar_type rcpOrientedEta) into actually calling operator()(const scalar_type NdotTorR, const scalar_type rcpOrientedEta) which the first parameter coming from cache.LdotH (N.B. the NdotI internal computation will CSE with your VdotH computation)

devshgraphicsprogramming · 2025-10-03T14:01:42Z

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

+        if ((ComputeMicrofacetNormal<scalar_type>::isTransmissionPath(VdotH, LdotH) != transmitted) || (LdotH * hlsl::dot(_N, Ldir) < scalar_type(0.0)))
+            L.makeInvalid(); // should check if sample direction is invalid
+        else
+            cache = anisocache_type::create(VdotH, Ldir, H, T, B, _N, transmitted);


use this instead

Nabla/include/nbl/builtin/hlsl/bxdf/common.hlsl

Lines 725 to 730 in 804014f

static this_t create(

const vector3_type tangentSpaceV,

const vector3_type tangentSpaceH,

const bool transmitted,

NBL_CONST_REF_ARG(fresnel::OrientedEtaRcps<monochrome_type>) rcpOrientedEta

)

devshgraphicsprogramming · 2025-10-03T14:11:18Z

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

+        const vector3_type Ldir = L.getDirection();
+        const scalar_type LdotH = hlsl::dot(Ldir, H);
+        if ((ComputeMicrofacetNormal<scalar_type>::isTransmissionPath(VdotH, LdotH) != transmitted) || (LdotH * hlsl::dot(_N, Ldir) < scalar_type(0.0)))
+            L.makeInvalid(); // should check if sample direction is invalid


you actually want to make the isotropic part of the cache first (before L is computed) cause it computes VdotH and LdotH by itself without knowing the full L

continues https://github.com/Devsh-Graphics-Programming/Nabla/pull/930/files?diff=split&w=1#r2356451287

The ~~LdotH and~~ (this doesn't need to be checked, see next comment) NdotL sign consistency could be checked right after computing transmitted

You're also checking the wrong thing, ComputeMicrofacetNormal<scalar_type>::isTransmissionPath(VdotH, LdotH) != transmitted will always be true, why?

VNDF sampling guarantees that VdotH will have a sign consistent with NdotV, and transmitted controls the sign of LdotH relative to VdotH by construction (you either reflect and end up with the same sign, or you refract and you get the opposite sign)

So what you should really be checking is isTransmissionPath(NdotV,NdotL) != transmitted and the NdotL should be computed in a friendly way similarly to this https://github.com/Devsh-Graphics-Programming/Nabla/pull/930/files/4faecc38efe93237e48b795402451366eeb1c5ba..804014f28e7469036673169266a013f42e018f60?diff=unified&w=1#r2401939155

all the above means that the (LdotH * hlsl::dot(_N, Ldir) < scalar_type(0.0)) check is redundant because when you know that

isTransmissionPath(NdotV,VdotH) === false // VNDF

and

isTransmissionPath(VdotH,LdotH) === transmitted // VNDF + construction from ReflectRefract

for sure, then

(LdotH * hlsl::dot(_N, Ldir) < scalar_type(0.0)) === (isTransmissionPath(NdotV,NdotL) != transmitted)

what you should assert() though after you pass the isTransmissionPath(NdotV,NdotL) != transmitted check, is that

ComputeMicrofacetNormal::isValidMicrofacet(transmitted,cache.VdotL,localH.z,orientedEta)==true

devshgraphicsprogramming · 2025-10-03T14:27:07Z

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

+        else
+            cache = anisocache_type::create(VdotH, Ldir, H, T, B, _N, transmitted);
+
+        return sample_type::create(L, T, B, _N);


same as for the BRDF generation, NdotL can be overwriten by whatever you had to compute to check that L is in the correct hemisphere as per transmitted and the sign of VdotN

devshgraphicsprogramming

GH UI screwing with me again

devshgraphicsprogramming · 2025-10-03T14:53:59Z

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

+    template<typename C=bool_constant<!IsAnisotropic>, typename D=bool_constant<!IsBSDF> >
+    enable_if_t<C::value && !IsAnisotropic && D::value && !IsBSDF, sample_type> generate(NBL_CONST_REF_ARG(isotropic_interaction_type) interaction, const vector2_type u, NBL_REF_ARG(isocache_type) cache)
+    {
+        anisocache_type aniso_cache;
+        sample_type s = generate(anisotropic_interaction_type::create(interaction), u, aniso_cache);
+        cache = aniso_cache.iso_cache;
+        return s;
+    }
+    template<typename C=bool_constant<!IsAnisotropic>, typename D=bool_constant<IsBSDF> >
+    enable_if_t<C::value && !IsAnisotropic && D::value && IsBSDF, sample_type> generate(NBL_CONST_REF_ARG(isotropic_interaction_type) interaction, const vector3_type u, NBL_REF_ARG(isocache_type) cache)
+    {
+        anisocache_type aniso_cache;
+        sample_type s = generate(anisotropic_interaction_type::create(interaction), u, aniso_cache);
+        cache = aniso_cache.iso_cache;
+        return s;
+    }


you could have just had one templated on V used to type u instead of D

devshgraphicsprogramming · 2025-10-03T14:58:59Z

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

+        if (IsBSDF || (_sample.getNdotL() > numeric_limits<scalar_type>::min && interaction.getNdotV() > numeric_limits<scalar_type>::min))
+        {
+            scalar_type _pdf = __pdf<isotropic_interaction_type, isocache_type>(_sample, interaction, cache);
+            return hlsl::mix(scalar_type(0.0), _pdf, _pdf < bit_cast<scalar_type>(numeric_limits<scalar_type>::infinity));
+        }
+        else
+            return scalar_type(0.0);
+    }
+    scalar_type pdf(NBL_CONST_REF_ARG(sample_type) _sample, NBL_CONST_REF_ARG(anisotropic_interaction_type) interaction, NBL_CONST_REF_ARG(anisocache_type) cache)
+    {
+        if (IsBSDF || (_sample.getNdotL() > numeric_limits<scalar_type>::min && interaction.getNdotV() > numeric_limits<scalar_type>::min))


checks are not complete for a BSDF (same problem as eval), not checking for TIR invalid microfacet or incosistent transmission with microfacets.

I'd really abstract the whole if (IsBSDF || (_sample.getNdotL() > numeric_limits<scalar_type>::min && interaction.getNdotV() > numeric_limits<scalar_type>::min)) into a pseduo-private method (prefixed with __) to use for eval and pdf

see https://github.com/Devsh-Graphics-Programming/Nabla/pull/930/files#r2401699310

devshgraphicsprogramming · 2025-10-03T15:01:36Z

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

+        const bool notTIR = impl::check_TIR_helper<fresnel_type, IsBSDF>::template __call<MicrofacetCache>(_f, cache);
+        if (notTIR)


quotient_and_pdf is meant to be only called on samples we generated, so I'd turn these checks into asserts (also different thing needs to be asserted, see https://github.com/Devsh-Graphics-Programming/Nabla/pull/930/files#r2401699310 )

devshgraphicsprogramming · 2025-10-03T15:02:58Z

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

+            NBL_IF_CONSTEXPR(IsBSDF)
+                quo = hlsl::promote<spectral_type>(G2_over_G1);
+            else
+                quo = _f(cache.getVdotH()) * G2_over_G1;


assert that the !(VdotH<0.f) for certainty

devshgraphicsprogramming · 2025-10-03T15:04:25Z

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

+
+        dg1_query_type dq = ndf.template createDG1Query<Interaction, MicrofacetCache>(interaction, cache);
+
+        fresnel_type _f = impl::getOrientedFresnel<fresnel_type, IsBSDF>::__call(fresnel, interaction.getNdotV());


you only need to reorient fresnels for BSDFs, see https://github.com/Devsh-Graphics-Programming/Nabla/pull/930/files#r2401699310

devshgraphicsprogramming · 2025-10-03T15:11:01Z

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

+    quotient_pdf_type __quotient_and_pdf(NBL_CONST_REF_ARG(sample_type) _sample, NBL_CONST_REF_ARG(Interaction) interaction, NBL_CONST_REF_ARG(MicrofacetCache) cache)
+    {
+        scalar_type _pdf = __pdf<Interaction, MicrofacetCache>(_sample, interaction, cache);
+        fresnel_type _f = impl::getOrientedFresnel<fresnel_type, IsBSDF>::__call(fresnel, interaction.getNdotV());


only BSDFs need to re-orient

devshgraphicsprogramming · 2025-10-03T15:15:41Z

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

+            scalar_type DG = D.projectedLightMeasure;
+            if (D.microfacetMeasure < bit_cast<scalar_type>(numeric_limits<scalar_type>::infinity))
+            {
+                g2g1_query_type gq = ndf.template createG2G1Query<sample_type, Interaction>(_sample, interaction);
+                DG *= ndf.template correlated<sample_type, Interaction>(gq, _sample, interaction);
+            }


hmm actually since Eval is not Quotient, it should return 0 whenever PDF would have been INF, and therefore whenever Eval is inf, you can just return 0

devshgraphicsprogramming · 2025-10-03T15:19:09Z

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

+            g2g1_query_type gq = ndf.template createG2G1Query<sample_type, Interaction>(_sample, interaction);
+            scalar_type G2_over_G1 = ndf.template G2_over_G1<sample_type, Interaction, MicrofacetCache>(gq, _sample, interaction, cache);


gq is not needed if the NDF is smooth, cause G2_over_G1 will approach 1

an NDF is smooth if PDF = INF

devshgraphicsprogramming · 2025-10-03T15:20:19Z

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

+        // set pdf=0 when quo=0 because we don't want to give high weight to sampling strategy that yields 0 contribution
+        _pdf = hlsl::mix(_pdf, scalar_type(0.0), hlsl::all(quo < hlsl::promote<spectral_type>(numeric_limits<scalar_type>::min)));


quotient will never be 0 after the _sample.valid() is checked (valid microfacetness can be asserted), its only when a sample is invalid that you need to return {0,0}

devshgraphicsprogramming · 2025-10-03T15:24:26Z

Please take care of this #916 (comment)

devshgraphicsprogramming

more GH UI stuff

devshgraphicsprogramming · 2025-10-03T15:31:58Z

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


    scalar_type G1_wo_numerator(scalar_type absNdotX, scalar_type NdotX2)
    {
        return scalar_type(1.0) / (absNdotX + devsh_part(NdotX2));


clamped not abs see #916 (comment)

don't even have this method #916 (comment)

devshgraphicsprogramming · 2025-10-03T15:41:42Z

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

+template<typename T>
+struct PhongExponent
+{
+    //conversion between alpha and Phong exponent, Walter et.al.
+    static T ToAlpha2(T _n)
+    {
+        return T(2.0) / (_n + T(2.0));
+    }
+    //+INF for a2==0.0
+    static T FromAlpha2(T a2)
+    {
+        return T(2.0) / a2 - T(2.0);
+    }


make a BeckmannBase class that your Common inherits from, cause right now this is bxdf::ndf::PhongExponent and people will think it can be used to translate phgon shininess into any NDF roughness

keptsecret added 23 commits September 4, 2025 10:46

changes to dual measure quant, old stuff commented out for now

5218545

create query structs in beckmann ndf, removed query concepts, funcs r…

f7525af

…eturn quants

did the same for ggx ndf, some fixes to beckmann

c50db68

Merge branch 'master' into bxdf_fixes_cook_torrance

c49bedb

fixes to ndf concept

319c954

moved eval, quotient/pdf into cook torrance base

4eeacf1

moved generate H into ndfs, generate impl in cook torrance base

e7bc784

use new cook torrance base in microfacet bxdfs

72226bb

Merge branch 'master' into bxdf_fixes_cook_torrance

f34b348

numerous typo bug fixes

ad13044

fixes to ggx ndf

d8d2116

fixed conductor fresnel naming ior as eta

238b08e

minor fixes to beckmann ndf, removed obsolete comments

8bc707f

added luminosity contribution hint to cook torrance bsdfs

9fb28cf

added aniso overloads to isotropic bxdf methods

4cb8a0a

checks for invalid generate sample

392dc31

removed obsolete commented out stuff

02de86e

added cartesian-polar conversions from unit tests

4a7f532

Merge branch 'fix_rotation_mat' into bxdf_fixes_cook_torrance

80c4f67

Merge branch 'master' into bxdf_fixes_cook_torrance

6233bd1

vector hashes

ce5fbac

Merge branch 'fix_rotation_mat' into bxdf_fixes_cook_torrance

47e814b

Merge branch 'master' into bxdf_fixes_cook_torrance

c983975

devshgraphicsprogramming commented Sep 16, 2025

View reviewed changes

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

devshgraphicsprogramming commented Sep 16, 2025

View reviewed changes

include/nbl/builtin/hlsl/math/functions.hlsl Outdated Show resolved Hide resolved

devshgraphicsprogramming commented Sep 16, 2025

View reviewed changes

keptsecret added 2 commits September 17, 2025 15:26

fix fresnel usage in bxdf pdf

0f2ee0b

fix F calc in generate

b5f02e6

devshgraphicsprogramming commented Sep 17, 2025

View reviewed changes

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

devshgraphicsprogramming commented Sep 17, 2025

View reviewed changes

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

devshgraphicsprogramming commented Oct 3, 2025

View reviewed changes

keptsecret added 7 commits October 6, 2025 10:29

projectedSphere generate should take v by ref

260f7a3

removed vector std::hash because glm already implemented it

fd54ac4

Merge branch 'master' into bxdf_fixes_cook_torrance

5caf006

get orientedeta and rcp methods in fresnel + concept

7508b82

slight change to flipSignIfRHSNeg struct

2b4a9c1

added two-sided fresnel concept

d3fa872

removed cook torrance param structs/create funcs

5765e87

	static this_t create(
	const vector3_type tangentSpaceV,
	const vector3_type tangentSpaceH,
	const bool transmitted,
	NBL_CONST_REF_ARG(fresnel::OrientedEtaRcps<monochrome_type>) rcpOrientedEta
	)

		const bool notTIR = impl::check_TIR_helper<fresnel_type, IsBSDF>::template __call<MicrofacetCache>(_f, cache);
		if (notTIR)


		dg1_query_type dq = ndf.template createDG1Query<Interaction, MicrofacetCache>(interaction, cache);

		fresnel_type _f = impl::getOrientedFresnel<fresnel_type, IsBSDF>::__call(fresnel, interaction.getNdotV());

		g2g1_query_type gq = ndf.template createG2G1Query<sample_type, Interaction>(_sample, interaction);
		scalar_type G2_over_G1 = ndf.template G2_over_G1<sample_type, Interaction, MicrofacetCache>(gq, _sample, interaction, cache);

		// set pdf=0 when quo=0 because we don't want to give high weight to sampling strategy that yields 0 contribution
		_pdf = hlsl::mix(_pdf, scalar_type(0.0), hlsl::all(quo < hlsl::promote<spectral_type>(numeric_limits<scalar_type>::min)));

Bxdf fixes cook torrance #930

Are you sure you want to change the base?

Bxdf fixes cook torrance #930

Uh oh!

Conversation

devshgraphicsprogramming commented Sep 16, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Description

Testing

TODO list:

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!

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

devshgraphicsprogramming Oct 3, 2025 • 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!

devshgraphicsprogramming Oct 3, 2025 • 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!

devshgraphicsprogramming Oct 3, 2025 • 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!

devshgraphicsprogramming Oct 3, 2025 • 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!

devshgraphicsprogramming Oct 3, 2025 • 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!

devshgraphicsprogramming Oct 3, 2025 • 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!

devshgraphicsprogramming left a comment

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 commented Sep 16, 2025 •

edited

Loading

devshgraphicsprogramming Oct 3, 2025 •

edited

Loading

devshgraphicsprogramming Oct 3, 2025 •

edited

Loading

devshgraphicsprogramming Oct 3, 2025 •

edited

Loading

devshgraphicsprogramming Oct 3, 2025 •

edited

Loading

devshgraphicsprogramming Oct 3, 2025 •

edited

Loading

devshgraphicsprogramming Oct 3, 2025 •

edited

Loading

devshgraphicsprogramming Oct 3, 2025 •

edited

Loading