Compute Prefilter.

samples/00_util/gpu_resource_manager.h

@@ -130,9 +130,9 @@ struct GpuResourceManager
     void Release(TextureHandle handle);                   // Release and Destroy
     void Release(Texture *texture, TextureHandle handle); // Release and Return
 
-    StorageTextureHandle CommitStorageTexture(StorageTexture *storageTexture); // What they said ^
+    StorageTextureHandle CommitStorageTexture(StorageTexture *storageTexture); // Commit to GPU and take Ownership
-    void Release(StorageTextureHandle handle);                                 //
+    void Release(StorageTextureHandle handle);                                 // Release and Destroy
-    void Release(StorageTexture *texture, StorageTextureHandle handle);        //
+    void Release(StorageTexture *texture, StorageTextureHandle handle);        // Release and Return
 
     void Update(); // Update all the descriptors required.
 

samples/03_model_render/CMakeLists.txt

@@ -23,6 +23,7 @@ add_shader(model_render shader/eqrect_to_cube.cs.hlsl)
 add_shader(model_render shader/background.vs.hlsl)
 add_shader(model_render shader/background.ps.hlsl)
 add_shader(model_render shader/diffuse_irradiance.cs.hlsl)
+add_shader(model_render shader/prefilter.cs.hlsl)
 
 target_link_libraries(model_render PRIVATE aster_core)
 target_link_libraries(model_render PRIVATE util_helper)

samples/03_model_render/asset_loader.cpp

@@ -12,12 +12,12 @@
 
 #include "asset_loader.h"
 
+#include "EASTL/fixed_vector.h"
 #include "buffer.h"
 #include "device.h"
 #include "gpu_resource_manager.h"
 #include "helpers.h"
 #include "image.h"
-#include "EASTL/fixed_vector.h"
 
 #include <EASTL/hash_map.h>
 #include <glm/gtc/type_ptr.hpp>
@@ -191,10 +191,19 @@ AssetLoader::LoadHdrImage(Texture *texture, cstr path, cstr name) const
 }
 
 void
-GenerateMipMaps(vk::CommandBuffer commandBuffer, Texture *texture, vk::ImageLayout initialLayout, vk::ImageLayout finalLayout)
+GenerateMipMaps(vk::CommandBuffer commandBuffer, Texture *texture, vk::ImageLayout initialLayout,
+                vk::ImageLayout finalLayout, vk::PipelineStageFlags2 prevStage, vk::PipelineStageFlags2 finalStage)
 {
+#if !defined(ASTER_NDEBUG)
+    vk::DebugUtilsLabelEXT label = {
+        .pLabelName = "Generate Mipmap",
+        .color = std::array{0.9f, 0.9f, 0.9f, 1.0f},
+    };
+    commandBuffer.beginDebugUtilsLabelEXT(&label);
+#endif
+
     vk::ImageMemoryBarrier2 imageStartBarrier = {
-        .srcStageMask = vk::PipelineStageFlagBits2::eAllCommands,
+        .srcStageMask = prevStage,
         .srcAccessMask = vk::AccessFlagBits2::eNone,
         .dstStageMask = vk::PipelineStageFlagBits2::eTransfer,
         .dstAccessMask = vk::AccessFlagBits2::eTransferRead,
@@ -346,6 +355,9 @@ GenerateMipMaps(vk::CommandBuffer commandBuffer, Texture *texture, vk::ImageLayo
     }
 
     commandBuffer.pipelineBarrier2(&imageReadyDependency);
+#if !defined(ASTER_NDEBUG)
+    commandBuffer.endDebugUtilsLabelEXT();
+#endif
 }
 
 TextureHandle

samples/03_model_render/asset_loader.h

@@ -120,4 +120,6 @@ struct AssetLoader
 };
 
 void GenerateMipMaps(vk::CommandBuffer commandBuffer, Texture *texture, vk::ImageLayout initialLayout,
-                     vk::ImageLayout finalLayout);
+                     vk::ImageLayout finalLayout,
+                     vk::PipelineStageFlags2 prevStage = vk::PipelineStageFlagBits2::eAllCommands,
+                     vk::PipelineStageFlags2 finalStage = vk::PipelineStageFlagBits2::eAllCommands);

samples/03_model_render/ibl_helpers.cpp

@@ -16,29 +16,51 @@
 
 constexpr cstr EQUIRECT_TO_CUBE_SHADER_FILE = "shader/eqrect_to_cube.cs.hlsl.spv";
 constexpr cstr DIFFUSE_IRRADIANCE_SHADER_FILE = "shader/diffuse_irradiance.cs.hlsl.spv";
+constexpr cstr PREFILTER_SHADER_FILE = "shader/prefilter.cs.hlsl.spv";
 
-eastl::tuple<TextureHandle, TextureHandle>
+eastl::tuple<TextureHandle, TextureHandle, TextureHandle>
 CreateCubeFromHdrEnv(AssetLoader *assetLoader, vk::Queue computeQueue, const u32 cubeSide, TextureHandle hdrEnv,
                      const cstr name)
 {
     GpuResourceManager *resMan = assetLoader->m_ResourceManager;
     const Device *pDevice = resMan->m_Device;
 
-    StackString<64> skyboxName = "Skybox: ";
-    StackString<64> diffuseIrradianceName = "DiffIrr: ";
-    skyboxName += name ? name : "";
-    diffuseIrradianceName += name ? name : "";
-
     StorageTextureCube skybox;
-    skybox.Init(pDevice, cubeSide, vk::Format::eR16G16B16A16Sfloat, true, false, skyboxName.c_str());
+    skybox.Init(pDevice, cubeSide, vk::Format::eR16G16B16A16Sfloat, true, true, "Skybox");
     TextureHandle skyboxHandle = resMan->CommitTexture(&skybox);
     StorageTextureHandle skyboxStorageHandle = resMan->CommitStorageTexture(&skybox);
 
     StorageTextureCube diffuseIrradiance;
-    diffuseIrradiance.Init(pDevice, 64, vk::Format::eR16G16B16A16Sfloat, true, false, diffuseIrradianceName.c_str());
+    diffuseIrradiance.Init(pDevice, 64, vk::Format::eR16G16B16A16Sfloat, true, false, "Diffuse Irradiance");
     TextureHandle diffuseIrradianceHandle = resMan->CommitTexture(&diffuseIrradiance);
     StorageTextureHandle diffuseIrradianceStorageHandle = resMan->CommitStorageTexture(&diffuseIrradiance);
 
+    StorageTextureCube prefilterCube;
+    prefilterCube.Init(pDevice, cubeSide, vk::Format::eR16G16B16A16Sfloat, true, true, "Prefilter");
+    TextureHandle prefilterHandle = resMan->CommitTexture(&prefilterCube); // This stores the original view for us.
+    constexpr u32 prefilterMipCountMax = 6;
+    eastl::array<StorageTextureHandle, prefilterMipCountMax> prefilterStorageHandles;
+    // All non-owning copies.
+    for (u32 mipLevel = 0; auto &tex : prefilterStorageHandles)
+    {
+        vk::ImageViewCreateInfo imageViewCreateInfo = {
+            .image = prefilterCube.m_Image,
+            .viewType = vk::ImageViewType::eCube,
+            .format = vk::Format::eR16G16B16A16Sfloat,
+            .components = vk::ComponentMapping{},
+            .subresourceRange =
+                {
+                    .aspectMask = vk::ImageAspectFlagBits::eColor,
+                    .baseMipLevel = mipLevel++,
+                    .levelCount = 1,
+                    .baseArrayLayer = 0,
+                    .layerCount = 6,
+                },
+        };
+        AbortIfFailed(pDevice->m_Device.createImageView(&imageViewCreateInfo, nullptr, &prefilterCube.m_View));
+        tex = resMan->CommitStorageTexture(&prefilterCube);
+    }
+
 #pragma region Dependencies and Copies
     vk::ImageSubresourceRange cubeSubresRange = {
         .aspectMask = vk::ImageAspectFlagBits::eColor,
@@ -80,7 +102,7 @@ CreateCubeFromHdrEnv(AssetLoader *assetLoader, vk::Queue computeQueue, const u32
         .image = skybox.m_Image,
         .subresourceRange = cubeSubresRange,
     };
-    vk::DependencyInfo skyboxReadToWriteDependency = {
+    vk::DependencyInfo skyboxWriteToReadDependency = {
         .imageMemoryBarrierCount = 1,
         .pImageMemoryBarriers = &skyboxWriteToReadBarrier,
     };
@@ -116,13 +138,22 @@ CreateCubeFromHdrEnv(AssetLoader *assetLoader, vk::Queue computeQueue, const u32
         StorageTextureHandle m_OutputTexture;
         u32 m_CubeSide;
     };
+    struct PrefilterPushConstants
+    {
+        TextureHandle m_SkyboxHandle;
+        StorageTextureHandle m_OutputTexture;
+        u32 m_CubeSide;
+        f32 m_Roughness;
+        u32 m_EnvSide;
+    };
 
 #pragma region Pipeline Creation etc
 
     vk::PushConstantRange pcr = {
         .stageFlags = vk::ShaderStageFlagBits::eCompute,
         .offset = 0,
-        .size = eastl::max(sizeof(SkyboxPushConstants), sizeof(DiffuseIrradiancePushConstants)),
+        .size = eastl::max(sizeof(SkyboxPushConstants),
+                           eastl::max(sizeof(DiffuseIrradiancePushConstants), sizeof(PrefilterPushConstants))),
     };
 
     vk::PipelineLayout pipelineLayout;
@@ -136,6 +167,7 @@ CreateCubeFromHdrEnv(AssetLoader *assetLoader, vk::Queue computeQueue, const u32
 
     const auto eqRectToCubeShader = CreateShader(pDevice, EQUIRECT_TO_CUBE_SHADER_FILE);
     const auto diffuseRadianceShader = CreateShader(pDevice, DIFFUSE_IRRADIANCE_SHADER_FILE);
+    const auto prefilterShader = CreateShader(pDevice, PREFILTER_SHADER_FILE);
     eastl::array computePipelineCreateInfo = {
         vk::ComputePipelineCreateInfo{
             .stage =
@@ -155,16 +187,25 @@ CreateCubeFromHdrEnv(AssetLoader *assetLoader, vk::Queue computeQueue, const u32
                 },
             .layout = pipelineLayout,
         },
+        vk::ComputePipelineCreateInfo{
+            .stage =
+                {
+                    .stage = vk::ShaderStageFlagBits::eCompute,
+                    .module = prefilterShader,
+                    .pName = "main",
+                },
+            .layout = pipelineLayout,
+        },
     };
 
-    eastl::array<vk::Pipeline, 2> pipelines;
+    eastl::array<vk::Pipeline, computePipelineCreateInfo.size()> pipelines;
-    static_assert(pipelines.size() == computePipelineCreateInfo.size());
     AbortIfFailed(pDevice->m_Device.createComputePipelines(pDevice->m_PipelineCache, computePipelineCreateInfo.size(),
                                                            computePipelineCreateInfo.data(), nullptr,
                                                            pipelines.data()));
 
     vk::Pipeline eqRectToCubePipeline = pipelines[0];
     vk::Pipeline diffuseIrradiancePipeline = pipelines[1];
+    vk::Pipeline prefilterPipeline = pipelines[2];
 
     for (auto &createInfos : computePipelineCreateInfo)
     {
@@ -183,6 +224,10 @@ CreateCubeFromHdrEnv(AssetLoader *assetLoader, vk::Queue computeQueue, const u32
         .m_OutputTexture = diffuseIrradianceStorageHandle,
         .m_CubeSide = diffuseIrradiance.m_Extent.width,
     };
+    PrefilterPushConstants prefilterPushConstants = {
+        .m_SkyboxHandle = skyboxHandle,
+        .m_EnvSide = cubeSide,
+    };
 
     resMan->Update();
 
@@ -206,14 +251,31 @@ CreateCubeFromHdrEnv(AssetLoader *assetLoader, vk::Queue computeQueue, const u32
     cmd.bindPipeline(vk::PipelineBindPoint::eCompute, eqRectToCubePipeline);
     cmd.pushConstants(pipelineLayout, vk::ShaderStageFlagBits::eCompute, 0, sizeof skyboxPushConstant,
                       &skyboxPushConstant);
-    cmd.dispatch(cubeSide / 16, cubeSide / 16, 6);
+    cmd.dispatch(skybox.m_Extent.width / 16, skybox.m_Extent.height / 16, 6);
 
-    cmd.pipelineBarrier2(&skyboxReadToWriteDependency);
+    GenerateMipMaps(cmd, &skybox, vk::ImageLayout::eGeneral, vk::ImageLayout::eGeneral,
+                    vk::PipelineStageFlagBits2::eComputeShader, vk::PipelineStageFlagBits2::eComputeShader);
 
     cmd.bindPipeline(vk::PipelineBindPoint::eCompute, diffuseIrradiancePipeline);
     cmd.pushConstants(pipelineLayout, vk::ShaderStageFlagBits::eCompute, 0, sizeof skyboxPushConstant,
                       &diffuseIrradiancePushConstants);
-    cmd.dispatch(cubeSide / 16, cubeSide / 16, 6);
+    cmd.dispatch(diffuseIrradiance.m_Extent.width / 16, diffuseIrradiance.m_Extent.width / 16, 6);
+
+    cmd.bindPipeline(vk::PipelineBindPoint::eCompute, prefilterPipeline);
+    u32 mipSize = prefilterCube.m_Extent.width;
+    for (u32 mipCount = 0; auto &tex : prefilterStorageHandles)
+    {
+        prefilterPushConstants.m_OutputTexture = tex;
+        prefilterPushConstants.m_CubeSide = mipSize;
+        prefilterPushConstants.m_Roughness = Cast<f32>(mipCount) / Cast<f32>(prefilterMipCountMax);
+        cmd.pushConstants(pipelineLayout, vk::ShaderStageFlagBits::eCompute, 0, sizeof prefilterPushConstants,
+                          &prefilterPushConstants);
+        u32 groupCount = eastl::max(mipSize / 16u, 1u);
+        cmd.dispatch(groupCount, groupCount, 6);
+
+        ++mipCount;
+        mipSize = mipSize >> 1;
+    }
 
     cmd.pipelineBarrier2(&skyboxToSampleDependency);
 
@@ -242,11 +304,17 @@ CreateCubeFromHdrEnv(AssetLoader *assetLoader, vk::Queue computeQueue, const u32
     skybox = {};
     resMan->Release(skyboxStorageHandle);
     resMan->Release(diffuseIrradianceStorageHandle);
+    for (auto &texHandles : prefilterStorageHandles)
+    {
+        StorageTextureCube st;
+        resMan->Release(&st, texHandles);
+        pDevice->m_Device.destroy(st.m_View, nullptr);
+    }
     for (auto &pipeline : pipelines)
     {
         pDevice->m_Device.destroy(pipeline, nullptr);
     }
     pDevice->m_Device.destroy(pipelineLayout, nullptr);
 
-    return {skyboxHandle, diffuseIrradianceHandle};
+    return {skyboxHandle, diffuseIrradianceHandle, prefilterHandle};
 }

samples/03_model_render/ibl_helpers.h

@@ -15,6 +15,6 @@ struct Texture;
 struct TextureCube;
 struct AssetLoader;
 
-eastl::tuple<TextureHandle, TextureHandle>
+eastl::tuple<TextureHandle, TextureHandle, TextureHandle>
 CreateCubeFromHdrEnv(AssetLoader *assetLoader, vk::Queue computeQueue, u32 cubeSide, TextureHandle hdrEnv,
                                    cstr name = nullptr);

samples/03_model_render/model_render.cpp

@@ -180,7 +180,7 @@ main(int, char **)
     assetLoader.LoadHdrImage(&environment, BACKDROP_FILE);
     auto envHandle = resourceManager.CommitTexture(&environment);
 
-    auto [texCube, diffuseIrr] = CreateCubeFromHdrEnv(&assetLoader, graphicsQueue, 512, envHandle, "Cube Env");
+    auto [texCube, diffuseIrr, prefilter] = CreateCubeFromHdrEnv(&assetLoader, graphicsQueue, 512, envHandle, "Cube Env");
 
     resourceManager.Release(envHandle);
 
@@ -373,6 +373,7 @@ main(int, char **)
     bool lockToScreen = true;
     bool showDiffuse = false;
     bool useDiffuse = true;
+    bool showPrefilter = false;
     i32 height = Cast<i32>(internalResolution.height);
     f32 camPitch = glm::degrees(cameraController.m_Pitch);
     f32 camYaw = glm::degrees(cameraController.m_Yaw);
@@ -451,6 +452,7 @@ main(int, char **)
         gui::Text("IBL");
         gui::Checkbox("Show DiffIrr", &showDiffuse);
         gui::Checkbox("Use DiffIrr", &useDiffuse);
+        gui::Checkbox("Show Prefilter", &showPrefilter);
         gui::Separator();
         gui::Checkbox("Rotate", &rotating);
         gui::PopItemWidth();
@@ -553,17 +555,25 @@ main(int, char **)
         cmd.pushConstants(pipeline.m_Layout, vk::ShaderStageFlagBits::eAll, pcbOffset, sizeof model.m_Handles,
                           &model.m_Handles);
         pcbOffset += sizeof model.m_Handles;
+
         if (showDiffuse)
         {
             cmd.pushConstants(pipeline.m_Layout, vk::ShaderStageFlagBits::eAll, pcbOffset, sizeof diffuseIrr,
                               &diffuseIrr);
             pcbOffset += sizeof diffuseIrr;
         }
+        else if (showPrefilter)
+        {
+            cmd.pushConstants(pipeline.m_Layout, vk::ShaderStageFlagBits::eAll, pcbOffset, sizeof prefilter,
+                              &prefilter);
+            pcbOffset += sizeof prefilter;
+        }
         else
         {
             cmd.pushConstants(pipeline.m_Layout, vk::ShaderStageFlagBits::eAll, pcbOffset, sizeof texCube, &texCube);
             pcbOffset += sizeof texCube;
         }
+
         if (useDiffuse)
         {
             cmd.pushConstants(pipeline.m_Layout, vk::ShaderStageFlagBits::eAll, pcbOffset, sizeof diffuseIrr,
@@ -656,6 +666,7 @@ main(int, char **)
 
     resourceManager.Release(texCube);
     resourceManager.Release(diffuseIrr);
+    resourceManager.Release(prefilter);
 
     pipelineCacheData = device.DumpPipelineCache();
     ERROR_IF(!WriteFileBytes(PIPELINE_CACHE_FILE, pipelineCacheData), "Pipeline Cache incorrectly written");

samples/03_model_render/shader/diffuse_irradiance.cs.hlsl

@@ -26,14 +26,14 @@ void main(uint3 GlobalInvocationID : SV_DispatchThreadID)
 {
     float3 Forward, Up, Right;
     
-    Forward = GetLocalDir(pcb.CubeSide, GlobalInvocationID);
+    Forward = GetCubeDir(GlobalInvocationID, pcb.CubeSide);
     Up = float3(0.0f, 1.0f, 0.0f); // 0.01f offset to 
     Right = normalize(cross(Up, Forward));
     Up = normalize(cross(Forward, Right));
     
     float3 Irradiance = 0.0f.xxx;
     float3 IrrDirr = 0.0f.xxx;
-    float SampleStep = 0.025f;
+    float SampleStep = 0.005f;
     float SampleCount = 0.0f;
   
     for (float Azimuth = 0.0f; Azimuth < TAU; Azimuth += SampleStep)
@@ -42,6 +42,7 @@ void main(uint3 GlobalInvocationID : SV_DispatchThreadID)
         {
             float3 DirectionTanSpace = float3(sin(Zenith) * cos(Azimuth), sin(Zenith) * sin(Azimuth), cos(Zenith));
             float3 DirectionWorld = DirectionTanSpace.x * Right + DirectionTanSpace.y * Up + DirectionTanSpace.z * Forward;
+            
             Irradiance += TextureCubes[pcb.SkyboxHandle].SampleLevel(ImmutableSamplers[pcb.SkyboxHandle], DirectionWorld, 0).xyz * (cos(Zenith) * sin(Zenith));
             SampleCount++;
         }

samples/03_model_render/shader/eqrect_to_cube.cs.hlsl

@@ -19,21 +19,10 @@ float2 SampleSphericalMap(float3 v)
     return UV;
 }
 
-/*
-| Axis | Layer | Up |
-|:----:|:-----:|:--:|
-|  +x  |   0   | -y |
-|  -x  |   1   | -y |
-|  +y  |   2   | +z |
-|  -y  |   3   | -z |
-|  -z  |   4   | -y |
-|  +z  |   5   | -y |
-*/
-
 [numthreads(16, 16, 1)]
 void main(uint3 GlobalInvocationID : SV_DispatchThreadID)
 {
-    float3 LocalDir = GetLocalDir(pcb.CubeSide, GlobalInvocationID);
+    float3 LocalDir = GetCubeDir(GlobalInvocationID, pcb.CubeSide);
     
     float2 UV = SampleSphericalMap(LocalDir);
     StorageTextureArrays[pcb.OutputTextureHandle][GlobalInvocationID.xyz] = Textures[pcb.HdrEnvHandle].SampleLevel(ImmutableSamplers[pcb.HdrEnvHandle], UV, 0);

samples/03_model_render/shader/ibl_common.hlsli

@@ -1,22 +1,39 @@
 #include "bindless_structs.hlsli"
 
-float3 GetLocalDir(float Side, float3 GlobalInvocationID)
+/*
-{
+ The Goal is simply to convert from a (0,0,0) to (CubeSide, CubeSide, FaceCount) Invocation ID space to a
-    float HalfSide = Side / 2;
+ (-1,-1,-1) to (1, 1, 1) space.
-    float AxisSign = 2.0f * (0.5f - (GlobalInvocationID.z % 2));
 
-    float3 LocalDir;
+| Axis | Layer | Up |
-    if (GlobalInvocationID.z < 2)
+|:----:|:-----:|:--:|
+|  +x  |   0   | -y |
+|  -x  |   1   | -y |
+|  +y  |   2   | +z |
+|  -y  |   3   | -z |
+|  -z  |   4   | -y |
+|  +z  |   5   | -y |
+*/
+float3 GetCubeDir(uint3 GlobalInvocationID, float SideLength)
+{
+    float2 FaceUV = float2(GlobalInvocationID.xy) / SideLength; // (0, SideLength) -> (0, 1)
+    FaceUV = 2.0f * FaceUV - 1.0f; // (0, 1) -> (-1, 1)
+    
+    switch (GlobalInvocationID.z)
     {
-        LocalDir = float3(AxisSign * HalfSide, HalfSide - GlobalInvocationID.y, (HalfSide - GlobalInvocationID.x) * AxisSign);
+        case 0:
+            return normalize(float3(1.0f, -FaceUV.y, -FaceUV.x)); // Face +X; x = 1, y = -v, z = -u
+        case 1:
+            return normalize(float3(-1.0f, -FaceUV.y, FaceUV.x)); // Face -X; x = -1, y = -v, z = u
+        case 2:
+            return normalize(float3(FaceUV.x, 1.0f, FaceUV.y)); // Face +Y; x = u, y = 1, z = v
+        case 3:
+            return normalize(float3(FaceUV.x, -1.0f, -FaceUV.y)); // Face -Y; x=u, y=-1, z=-v
+        case 4:
+            return normalize(float3(FaceUV.x, -FaceUV.y, 1.0f)); // Face +Z; x=u,y=-v, z=1
+        case 5:
+            return normalize(float3(-FaceUV.x, -FaceUV.y, -1.0f)); // Face -Z; x=u,y=-v, z=-1
+        default:
+            // Never reach here.
+            return 0.0f.xxx;
     }
-    else if (GlobalInvocationID.z < 4)
-    {
-        LocalDir = float3(GlobalInvocationID.x - HalfSide, AxisSign * HalfSide, AxisSign * (GlobalInvocationID.y - HalfSide));
-    }
-    else
-    {
-        LocalDir = float3((GlobalInvocationID.x - HalfSide) * AxisSign, HalfSide - GlobalInvocationID.y, HalfSide * AxisSign);
-    }
-    return normalize(LocalDir);
 }

samples/03_model_render/shader/prefilter.cs.hlsl

@@ -0,0 +1,106 @@
+#include "ibl_common.hlsli"
+
+struct Block
+{
+    uint SkyboxHandle;
+    uint OutputTextureHandle;
+    int CubeSide;
+    float Roughness;
+    uint EnvRes;
+};
+
+[[vk::push_constant]]
+Block Pcb;
+
+float RadicalInverse_VdC(uint Bits)
+{
+    Bits = (Bits << 16u) | (Bits >> 16u);
+    Bits = ((Bits & 0x55555555u) << 1u) | ((Bits & 0xAAAAAAAAu) >> 1u);
+    Bits = ((Bits & 0x33333333u) << 2u) | ((Bits & 0xCCCCCCCCu) >> 2u);
+    Bits = ((Bits & 0x0F0F0F0Fu) << 4u) | ((Bits & 0xF0F0F0F0u) >> 4u);
+    Bits = ((Bits & 0x00FF00FFu) << 8u) | ((Bits & 0xFF00FF00u) >> 8u);
+    return float(Bits) * 2.3283064365386963e-10; // / 0x100000000
+}
+
+float2 Hammersley(uint SampleIndex, uint SampleCount)
+{
+    return float2(float(SampleIndex) / float(SampleCount), RadicalInverse_VdC(SampleIndex));
+}
+
+float3 ImportanceSampleGGX(float2 Xi, float3 Normal, float Roughness)
+{
+    float A = Roughness * Roughness;
+	
+    float Phi = 2.0f * PI * Xi.x;
+    float CosTheta = sqrt((1.0f - Xi.y) / (1.0f + (A * A - 1.0f) * Xi.y));
+    float SinTheta = sqrt(1.0f - CosTheta * CosTheta);
+	
+	// from spherical coordinates to cartesian coordinates
+    float3 H;
+    H.x = cos(Phi) * SinTheta;
+    H.y = sin(Phi) * SinTheta;
+    H.z = CosTheta;
+	
+	// from tangent-space vector to world-space sample vector
+    float3 Up = abs(Normal.z) < 0.999f ? float3(0.0f, 0.0f, 1.0f) : float3(1.0f, 0.0f, 0.0f);
+    float3 Tangent = normalize(cross(Up, Normal));
+    float3 Bitangent = cross(Normal, Tangent);
+	
+    float3 SampleVec = Tangent * H.x + Bitangent * H.y + Normal * H.z;
+    return normalize(SampleVec);
+}
+
+float TrowbridgeReitzGGX(float NdotH, float Roughness)
+{
+    float Coeff = Roughness * Roughness;
+    float Alpha2 = Coeff * Coeff;
+    float NdotH2 = NdotH * NdotH;
+	
+    float Numerator = Alpha2;
+    float Denominator = NdotH2 * (Alpha2 - 1.0f) + 1.0f;
+    Denominator = PI * Denominator * Denominator;
+	
+    return Numerator / Denominator;
+}
+
+float GetSampleMipLevel(float NdotH, float HdotV, float SampleCount)
+{
+    float D = TrowbridgeReitzGGX(NdotH, Pcb.Roughness);
+    float Pdf = (D * NdotH / (4.0f * HdotV)) + 0.0001f;
+
+    float SurfAreaTexel = 4.0f * PI / (6.0f * Pcb.EnvRes * Pcb.EnvRes);
+    float SurfAreaSample = 1.0f / (SampleCount * Pdf + 0.0001f);
+
+    return Pcb.Roughness == 0.0f ? 0.0f : 0.5f * log2(SurfAreaSample / SurfAreaTexel);
+}
+
+[numthreads(16, 16, 1)]
+void main(uint3 GlobalInvocationID : SV_DispatchThreadID)
+{   
+    float3 Normal = GetCubeDir(GlobalInvocationID, Pcb.CubeSide);
+    float3 ViewDir = Normal;
+
+    const uint SAMPLE_COUNT = 2048u;
+    float TotalWeight = 0.0f;
+    float3 PrefilterColor = 0.0f.xxx;
+    for (uint i = 0u; i < SAMPLE_COUNT; ++i)
+    {
+        float2 Xi = Hammersley(i, SAMPLE_COUNT);
+        float3 Halfway = ImportanceSampleGGX(Xi, Normal, Pcb.Roughness);
+        float3 LightDir = normalize(2.0 * dot(ViewDir, Halfway) * Halfway - ViewDir);
+        
+        float NdotH = max(dot(Normal, Halfway), 0.0f);
+    
+        float MipLevel = GetSampleMipLevel(NdotH, NdotH /* N = V :: NdotH = HdotV */, SAMPLE_COUNT);
+        
+        float NdotL = max(dot(Normal, LightDir), 0.0);
+        if (NdotL > 0.0)
+        {
+            PrefilterColor += TextureCubes[Pcb.SkyboxHandle].SampleLevel(ImmutableSamplers[Pcb.SkyboxHandle], LightDir, MipLevel).rgb * NdotL;
+            TotalWeight += NdotL;
+        }
+    }
+    PrefilterColor = PrefilterColor / TotalWeight;
+
+    StorageTextureArrays[Pcb.OutputTextureHandle][GlobalInvocationID] = float4(PrefilterColor, 1.0f);
+}