IES updates #965

AnastaZIuk · 2025-12-12T19:13:08Z

No description provided.

small change to make sure `ditt` doesn't autodelete

…shaders

…ubresource bitflags to not trigger asserts, update examples_tests submodule

…experimental, no real validation), update examples_tests submodule

…arguments in json payload, CI should pass now

…ental)

…ate examples_tests

…ODO on how to improve it even more

… with GPU converter, encode to EF_A2R10G10B10_UNORM_PACK32, update examples_tests submodule

devshgraphicsprogramming · 2026-01-14T13:52:18Z

include/nbl/builtin/hlsl/ies/sampler.hlsl

+            return abs(phi);
+        case symmetry_t::NO_LATERAL_SYMMET: //! plot onto whole (in degress) [0, 360] range
+        {
+            NBL_CONSTEXPR float32_t M_TWICE_PI = numbers::pi<float32_t> *2.f;


same as https://github.com/Devsh-Graphics-Programming/Nabla/pull/965/files#r2690508812

devshgraphicsprogramming · 2026-01-14T13:53:53Z

include/nbl/builtin/hlsl/ies/sampler.hlsl

+            return (phi < 0.f) ? (phi + M_TWICE_PI) : phi;
+        }
+        }
+        return 69.f;


devshgraphicsprogramming · 2026-01-14T13:54:02Z

include/nbl/builtin/hlsl/ies/sampler.hlsl

+        }
+        }


indent the cases

devshgraphicsprogramming · 2026-01-14T13:55:58Z

include/nbl/builtin/hlsl/ies/sampler.hlsl

+        // TODO: DXC seems to have a bug and cannot use symmetry_t directly with == operator https://godbolt.devsh.eu/z/P9Kc5x
+        const ProfileProperties::LuminairePlanesSymmetry symmetry = accessor.getProperties().getSymmetry();


This

s.sample(); // error: translating binary operator '==' unimplemented

works now

devshgraphicsprogramming · 2026-01-14T13:57:16Z

include/nbl/builtin/hlsl/ies/sampler.hlsl

+        const float32_t vAngle = degrees(polar.theta);
+        const float32_t hAngle = degrees(wrapPhi(polar.phi, symmetry));


we have asserts in HLSL, you can assert the polar coordinates are valid (more than 0 and less than pi and 2pi respectively)

devshgraphicsprogramming · 2026-01-14T13:59:04Z

include/nbl/builtin/hlsl/ies/sampler.hlsl

+    struct impl_t
+    {
+        static uint32_t getVUB(NBL_CONST_REF_ARG(accessor_t) accessor, const float32_t angle)
+        {
+            for (uint32_t i = 0u; i < accessor.vAnglesCount(); ++i)
+                if (accessor.vAngle(i) > angle)
+                    return i;
+            return accessor.vAnglesCount();
+        }
+
+        static uint32_t getHUB(NBL_CONST_REF_ARG(accessor_t) accessor, const float32_t angle)
+        {
+            for (uint32_t i = 0u; i < accessor.hAnglesCount(); ++i)
+                if (accessor.hAngle(i) > angle)
+                    return i;
+            return accessor.hAnglesCount();
+        }
+    };


you can just prefix the methods with __ as is our convention

devshgraphicsprogramming · 2026-01-14T13:59:48Z

include/nbl/builtin/hlsl/ies/sampler.hlsl

+            for (uint32_t i = 0u; i < accessor.vAnglesCount(); ++i)
+                if (accessor.vAngle(i) > angle)
+                    return i;
+            return accessor.vAnglesCount();
+        }
+
+        static uint32_t getHUB(NBL_CONST_REF_ARG(accessor_t) accessor, const float32_t angle)
+        {
+            for (uint32_t i = 0u; i < accessor.hAnglesCount(); ++i)
+                if (accessor.hAngle(i) > angle)
+                    return i;
+            return accessor.hAnglesCount();


can you please use our upper_bound algorithm ?

devshgraphicsprogramming · 2026-01-14T14:00:37Z

include/nbl/builtin/hlsl/ies/sampler.hlsl

+        return (uint32_t)hlsl::max((int64_t)impl_t::getVUB(accessor, angle) - 1ll, 0ll);
+    }
+
+    static uint32_t getHLB(NBL_CONST_REF_ARG(accessor_t) accessor, const float32_t angle)
+    {
+        return (uint32_t)hlsl::max((int64_t)impl_t::getHUB(accessor, angle) - 1ll, 0ll);
+    }
+
+    static uint32_t getVUB(NBL_CONST_REF_ARG(accessor_t) accessor, const float32_t angle)
+    {
+        return (uint32_t)hlsl::min((int64_t)impl_t::getVUB(accessor, angle), (int64_t)(accessor.vAnglesCount() - 1u));
+    }
+
+    static uint32_t getHUB(NBL_CONST_REF_ARG(accessor_t) accessor, const float32_t angle)
+    {
+        return (uint32_t)hlsl::min((int64_t)impl_t::getHUB(accessor, angle), (int64_t)(accessor.hAnglesCount() - 1u));


please don't use uint64_t, use the return type

devshgraphicsprogramming · 2026-01-14T14:07:25Z

src/nbl/asset/interchange/CIESProfileLoader.cpp

+        logger.log("%s: Invalid IES signature for \"%s\" file!", system::ILogger::ELL_ERROR, __FUNCTION__, fName);
+    }
+    else
+        logger.log("%s: Failed to read \"%s\" file!", system::ILogger::ELL_ERROR, __FUNCTION__, fName);


the're not errors, they' DEBUG or INFO, remember `isALoadableFileFormat can be called to query files if a particular loader is getting used

devshgraphicsprogramming · 2026-01-14T14:10:51Z

src/nbl/asset/utils/CIESProfile.cpp


-    if (height > CDC_MAX_TEXTURE_HEIGHT)
-        height = CDC_MAX_TEXTURE_HEIGHT;
+    // TODO: If no symmetry (no folding in half and abuse of mirror sampler) make dimensions odd-sized so middle texel taps the south pole


do it right now, and leave a TODO to undo it when implementing the exploitation of symmetries

devshgraphicsprogramming · 2026-01-14T14:22:31Z

include/nbl/builtin/hlsl/ies/texture.hlsl

+// TODO(?): should be in nbl::hlsl::ies (or in the Texutre struct) but I get
+// error GA3909C62: class template specialization of 'member_count' not in a namespace enclosing 'bda'
+// which I don't want to deal with rn to not (eventually) break stuff


its extremely annoying but you need to do it like this

namespace nbl { namespace hlsl { namespace ies { struct IESTextureInfo; } } } NBL_HLSL_DEFINE_STRUCT((::nbl::hlsl::ies::IESTextureInfo), ((inv, float32_t2)) ((flatten, float32_t)) ((maxValueRecip, float32_t)) ((flattenTarget, float32_t)) ((domainLo, float32_t)) ((domainHi, float32_t)) ((fullDomainFlatten, uint16_t)) // bool );

devshgraphicsprogramming · 2026-01-14T14:23:10Z

include/nbl/builtin/hlsl/ies/texture.hlsl

+
+struct IESTextureInfo;
+NBL_HLSL_DEFINE_STRUCT((IESTextureInfo),
+	((inv, float32_t2))


give it a better name than just inv, lastTexelRcp or something

devshgraphicsprogramming · 2026-01-14T14:24:10Z

include/nbl/builtin/hlsl/ies/texture.hlsl

+	((flatten, float32_t))
+	((maxValueRecip, float32_t))
+	((flattenTarget, float32_t))
+	((domainLo, float32_t))
+	((domainHi, float32_t))
+	((fullDomainFlatten, uint16_t)) // bool


you don't need these members anymore because flattening is no longer needed (except for maxValueRecip)

devshgraphicsprogramming · 2026-01-14T14:25:19Z

include/nbl/builtin/hlsl/ies/texture.hlsl

+    static inline SInfo createInfo(NBL_CONST_REF_ARG(accessor_t) accessor, NBL_CONST_REF_ARG(uint32_t2) size, float32_t flatten, bool fullDomainFlatten)
+    {
+        SInfo retval;
+		const ProfileProperties props = accessor.getProperties();


store the info in the SProceduralTexture

devshgraphicsprogramming · 2026-01-14T14:25:27Z

include/nbl/builtin/hlsl/ies/texture.hlsl

+{
+
+template<typename Accessor NBL_FUNC_REQUIRES(concepts::IsIESAccessor<Accessor>)
+struct Texture


call it SProceduralTexture

and remove the template

devshgraphicsprogramming · 2026-01-14T14:27:14Z

include/nbl/builtin/hlsl/ies/texture.hlsl

+    static inline SInfo createInfo(NBL_CONST_REF_ARG(accessor_t) accessor, NBL_CONST_REF_ARG(uint32_t2) size, float32_t flatten, bool fullDomainFlatten)
+    {
+        SInfo retval;


make this into a static inline SProceduralTexture create(const float32_t minTheta, const float32_t maxTheta, const uint32_t2 resolution=Default...)

actually just static inline SProceduralTexture create(const float32_t maxCandelaValue, const uint32_t2 resolution=Default...) because flattening is no longer necessary so no need to know about the domain sizes

you don't actually need to template the whole class on the accessor, just operator()

devshgraphicsprogramming · 2026-01-14T14:27:37Z

include/nbl/builtin/hlsl/ies/profile.hlsl

+    //! max 16K resolution
+    NBL_CONSTEXPR_STATIC_INLINE uint32_t CDC_MAX_TEXTURE_WIDTH = 15360u;
+    NBL_CONSTEXPR_STATIC_INLINE uint32_t CDC_MAX_TEXTURE_HEIGHT = 8640u;
+
+    // TODO: This constraint is hack because the mitsuba loader and its material compiler use Virtual Texturing, and there's some bug with IES not sampling sub 128x128 mip levels
+    // don't want to spend time to fix this since we'll be using descriptor indexing for the next iteration
+    NBL_CONSTEXPR_STATIC_INLINE uint32_t CDC_MIN_TEXTURE_WIDTH = 128u;
+    NBL_CONSTEXPR_STATIC_INLINE uint32_t CDC_MIN_TEXTURE_HEIGHT = 128u;
+
+    NBL_CONSTEXPR_STATIC_INLINE uint32_t CDC_DEFAULT_TEXTURE_WIDTH = 1024u;
+    NBL_CONSTEXPR_STATIC_INLINE uint32_t CDC_DEFAULT_TEXTURE_HEIGHT = 1024u;


all the texture stuff should go to SProceduralTexture

devshgraphicsprogramming · 2026-01-14T14:30:51Z

include/nbl/builtin/hlsl/ies/texture.hlsl

+        return retval;
+    }
+
+	static inline float32_t eval(NBL_CONST_REF_ARG(accessor_t) accessor, NBL_CONST_REF_ARG(SInfo) info, NBL_CONST_REF_ARG(float32_t2) uv)


make this method into template operator(), its this function that needs to be templated on the accessor

also hold SInfo as a member`

actually turn this into __evalNDC and use octahedral_t::ndcToDir

devshgraphicsprogramming · 2026-01-14T14:32:01Z

include/nbl/builtin/hlsl/ies/texture.hlsl

+        const float32_t3 dir = octahedral_t::uvToDir(uv);
+        const polar_t polar = polar_t::createFromCartesian(dir);
+
+		sampler_t sampler;


sampler is a reserved keyword in HLSL, use _sampler

devshgraphicsprogramming · 2026-01-14T14:32:20Z

include/nbl/builtin/hlsl/ies/texture.hlsl

+        const polar_t polar = polar_t::createFromCartesian(dir);
+
+		sampler_t sampler;
+        const float32_t intensity = sampler.sample(accessor, polar);


sample is a reserved keyword, turn the method into operator()

not possible due to dxc overload resolution bug, we have __call

devshgraphicsprogramming · 2026-01-14T14:32:50Z

include/nbl/builtin/hlsl/ies/texture.hlsl

+		//! blend the IES texture with "flatten"
+        float32_t blendV = intensity * (1.f - info.flatten);
+
+        const bool inDomain = (info.domainLo <= polar.theta) && (polar.theta <= info.domainHi);
+
+        if ((info.fullDomainFlatten && inDomain) || intensity > 0.0f)
+            blendV += info.flattenTarget * info.flatten;


no more flattening needed

devshgraphicsprogramming · 2026-01-14T14:33:45Z

include/nbl/builtin/hlsl/ies/texture.hlsl

+	static inline float32_t eval(NBL_CONST_REF_ARG(accessor_t) accessor, NBL_CONST_REF_ARG(SInfo) info, NBL_CONST_REF_ARG(uint32_t2) position)
+    {
+        const float32_t2 uv = float32_t2(position) * info.inv;


ok so for this, you take a corner sampled unnorm position, and need to turn it into UV, so needs to have +0.5 and then mul by rcp of texsize

OR mul by rcp of texsize and add half-texel in norm coords

Alternatively you can do this

template<typename Accessor, typename Coordinate NBL_FUNC_REQUIRES(concepts::IsIESAccessor<Accessor,> && is_same_v<Coordinate ,float32_t2>) inline float32_t operator()(NBL_CONST_REG_ARG(Accessor) accessor, const Coordinate coord) NBL_CONST_METHOD { const float32_t2 ndc = // uv to NDC with half texel handling return __evalNDC(ndc); } template<typename Accessor, typename Coordinate NBL_FUNC_REQUIRES(concepts::IsIESAccessor<Accessor,> && is_same_v<Coordinate ,int32_t2>) inline float32_t operator()(NBL_CONST_REG_ARG(Accessor) accessor, const Coordinate coord) NBL_CONST_METHOD { const float32_t2 ndc = coord*lastTexelRcp*2.f-float32_t2(1,1); return __evalNDC(ndc); }

devshgraphicsprogramming · 2026-01-14T14:38:27Z

include/nbl/builtin/hlsl/ies/sampler.hlsl

+	using polar_t = math::Polar<float32_t>;
+	using octahedral_t = math::OctahedralTransform<float32_t>;
+
+    static value_t sample(NBL_CONST_REF_ARG(accessor_t) accessor, NBL_CONST_REF_ARG(math::Polar<float32_t>) polar)


turn sample(...) into template<typename CoordinateT> requires static value_t __call(...)

because sample is a reserved keyword and sampling is already kinda like a function call oeprator

…into ies # Conflicts: # examples_tests

AnastaZIuk · 2026-02-02T08:33:50Z

src/nbl/video/utilities/CAssetConverter.cpp

 							for (auto i=0; i<combinedCount; i++)
 							{
-								const auto& toBind = binItems[i];
+								const auto& toBind = binItemsIt[i];


It used binItems[i] so it always indexed from the start of the full list. Later batches rebound earlier items and could trigger validation errors. binItemsIt[i] is indexed from the current batch slice so each batch binds its own items

ok start here, wrong index was used for the current batch slice and later batches rebound earlier objects

Nabla/src/nbl/video/utilities/CAssetConverter.cpp

Line 2483 in 4c0da88

const auto& toBind = binItemsIt[i];

batch binding happens in finalize

Nabla/src/nbl/video/utilities/CAssetConverter.cpp

Line 2346 in 4c0da88

void finalize()

then finalize is triggered from convert

Nabla/src/nbl/video/utilities/CAssetConverter.cpp

Line 3799 in 4c0da88

deferredAllocator.finalize();

convert_impl entry

Nabla/src/nbl/video/utilities/CAssetConverter.cpp

Line 4088 in 4c0da88

ISemaphore::future_t<IQueue::RESULT> CAssetConverter::convert_impl(SReserveResult& reservations, SConvertParams& params)

now let four GPU buffers land in the same bin. One allocation for all 4 fails, allocator succeeds for 2 then 2. The second batch rebinds buffers 0–1 instead of 2–3, leaving 2–3 unbound

wow, thanks for fixing this!

devshgraphicsprogramming and others added 30 commits September 13, 2025 20:52

Update CONTRIBUTING.md

199b75a

small change to make sure `ditt` doesn't autodelete

Get IES cpps to compile

36d9943

cap kind args

e312993

fix a bug after my NSC rules update

25c4bdc

update examples_tests submodule, we will first make CI pass with IES …

3d18b58

…shaders

update examples_tests submodule

27de67f

update examples_tests submodule

0110f98

ah IES was not added to asset manager, also correct some aspectMask s…

9ed5aee

…ubresource bitflags to not trigger asserts, update examples_tests submodule

update examples_tests submodule

35931bd

update examples_tests submodule

a892678

update examples_tests submodule

5e236e6

update examples_tests submodule

7b27168

update examples_tests submodule

77e9e57

update ImGUI extension to allow hijacking precompiled SPIRV shaders (…

699e841

…experimental, no real validation), update examples_tests submodule

update examples_tests submodule

bd6cc4d

a few minor updates to NBL_CREATE_NSC_COMPILE_RULES to skip optional …

9381491

…arguments in json payload, CI should pass now

CP_UTF8 for dxc source buffer's encoding just like code page (experim…

ae0b160

…ental)

update examples_tests submodule

d45e3a7

fix IES Profile average emission and domain calculations

5ecf50e

Update CI references

07feaff

-Wno-local-type-template-args for NBL_CREATE_NSC_COMPILE_RULES, upd…

e8ea119

…ate examples_tests

make the IES octahedral maps corner sampled

1e9138e

improve how the resolution of an IES is calculated but also leave a T…

be62e3a

…ODO on how to improve it even more

Update CI references

0c03e2f

update examples_tests submodule

80de6aa

update examples_tests submodule

e05478d

add createGrid to geometry creator

8d0c237

remove debug code I forgot about

dbd2805

add positions to geometry creator's grid for the polygon to be usable…

054038f

… with GPU converter, encode to EF_A2R10G10B10_UNORM_PACK32, update examples_tests submodule

update examples_tests submodule

f64a8f3

devshgraphicsprogramming reviewed Jan 14, 2026

View reviewed changes

AnastaZIuk added 4 commits January 25, 2026 17:33

Merge master

08a83fd

adjust to comments

925023e

Merge branch 'master' of github.com:Devsh-Graphics-Programming/Nabla …

4fdfb15

…into ies # Conflicts: # examples_tests

adjust to comments

4c0da88

AnastaZIuk commented Feb 2, 2026

View reviewed changes

AnastaZIuk added 2 commits February 2, 2026 12:25

pull master, resolve conflicts (+ adjust)

9b3891b

update examples_tests submodule pointer

e3a57df

		// TODO: DXC seems to have a bug and cannot use symmetry_t directly with == operator https://godbolt.devsh.eu/z/P9Kc5x
		const ProfileProperties::LuminairePlanesSymmetry symmetry = accessor.getProperties().getSymmetry();

		const float32_t vAngle = degrees(polar.theta);
		const float32_t hAngle = degrees(wrapPhi(polar.phi, symmetry));

IES updates #965

Are you sure you want to change the base?

IES updates #965

Conversation

AnastaZIuk commented Dec 12, 2025

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!

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!

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!

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 Jan 14, 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!

devshgraphicsprogramming Jan 14, 2026 •

edited

Loading