From febb71f2c14ca60e1ed0e59f2c36c123ba51a391 Mon Sep 17 00:00:00 2001
From: Patrick Wuttke <mewin@mewin.de>
Date: Tue, 1 Oct 2024 16:56:28 +0200
Subject: [PATCH] Added simple post-processing.

---
 IDEAS.md                                      |   4 +-
 assets/shaders/glsl/post_process.comp         |  19 ++
 private/sdl_gpu_test/5_input/app.cpp          |   6 +-
 private/sdl_gpu_test/6_ui/app.cpp             |   6 +-
 private/sdl_gpu_test/7_3d_scene/app.cpp       |  16 +-
 private/sdl_gpu_test/7_3d_scene/app.hpp       |   2 -
 private/sdl_gpu_test/8_post_process/app.cpp   | 263 ++++++++++++++++++
 private/sdl_gpu_test/8_post_process/app.hpp   |  52 ++++
 private/sdl_gpu_test/SModule                  |   5 +-
 private/sdl_gpu_test/application.cpp          |  32 +++
 private/sdl_gpu_test/application.hpp          |   7 +
 private/sdl_gpu_test/main.cpp                 |   5 +-
 .../post_process/post_processor.cpp           |  47 ++++
 .../post_process/post_processor.hpp           |  35 +++
 private/sdl_gpu_test/scene/scene_renderer.cpp |  15 +-
 private/sdl_gpu_test/scene/scene_renderer.hpp |   9 +-
 private/sdl_gpu_test/sdlpp/gpu.hpp            | 186 +++++++++++++
 17 files changed, 672 insertions(+), 37 deletions(-)
 create mode 100644 assets/shaders/glsl/post_process.comp
 create mode 100644 private/sdl_gpu_test/8_post_process/app.cpp
 create mode 100644 private/sdl_gpu_test/8_post_process/app.hpp
 create mode 100644 private/sdl_gpu_test/post_process/post_processor.cpp
 create mode 100644 private/sdl_gpu_test/post_process/post_processor.hpp

diff --git a/IDEAS.md b/IDEAS.md
index f768237..a629ada 100644
--- a/IDEAS.md
+++ b/IDEAS.md
@@ -1,9 +1,9 @@
 more ideas on what to try out
 
-- [ ] post processing
+- [x] post processing
 - [x] UI rendering
 - [ ] DirectX
-- [ ] compute shaders
+- [x] compute shaders
 - [ ] mini game
 - [ ] ImGui?
 - [ ] extending SDL_gpu?
diff --git a/assets/shaders/glsl/post_process.comp b/assets/shaders/glsl/post_process.comp
new file mode 100644
index 0000000..1f105cb
--- /dev/null
+++ b/assets/shaders/glsl/post_process.comp
@@ -0,0 +1,19 @@
+#version 460
+
+layout(local_size_x = 16, local_size_y = 16, local_size_x = 1) in;
+
+layout(set = 0, binding = 0, rgba8) uniform readonly image2D u_sourceImage;
+layout(set = 1, binding = 0, rgba8) uniform writeonly image2D u_targetImage;
+
+float luma(vec4 color)
+{
+    return dot(color.rgb, vec3(0.299, 0.587, 0.114));
+}
+
+void main()
+{
+    const ivec2 uv = ivec2(gl_GlobalInvocationID.xy);
+    const vec4 pixel = imageLoad(u_sourceImage, uv);
+    const float lum = luma(pixel);
+    imageStore(u_targetImage, uv, vec4(lum, lum, lum, 1.0));
+}
diff --git a/private/sdl_gpu_test/5_input/app.cpp b/private/sdl_gpu_test/5_input/app.cpp
index 8d93a60..6e82aca 100644
--- a/private/sdl_gpu_test/5_input/app.cpp
+++ b/private/sdl_gpu_test/5_input/app.cpp
@@ -121,11 +121,7 @@ void InputApp::init(const AppInitArgs& args)
     mSampler.create(mDevice, {});
 
     // open gamepad
-    const std::vector<SDL_JoystickID> gamepads = sdlpp::getGamepads();
-    if (!gamepads.empty())
-    {
-        mGamepad.open(gamepads[0]);
-    }
+    mGamepad = openFirstGamepad();
 }
 
 void InputApp::update(const AppUpdateArgs& args)
diff --git a/private/sdl_gpu_test/6_ui/app.cpp b/private/sdl_gpu_test/6_ui/app.cpp
index 23098a0..a3a3848 100644
--- a/private/sdl_gpu_test/6_ui/app.cpp
+++ b/private/sdl_gpu_test/6_ui/app.cpp
@@ -64,11 +64,7 @@ void UIApp::init(const AppInitArgs& args)
     mSampler.create(mDevice, {});
 
     // open gamepad
-    const std::vector<SDL_JoystickID> gamepads = sdlpp::getGamepads();
-    if (!gamepads.empty())
-    {
-        mGamepad.open(gamepads[0]);
-    }
+    mGamepad = openFirstGamepad();
 
     // init the UI
     mWidgetTree.init({.renderer = &mUIRenderer});
diff --git a/private/sdl_gpu_test/7_3d_scene/app.cpp b/private/sdl_gpu_test/7_3d_scene/app.cpp
index 7139269..59cf3a4 100644
--- a/private/sdl_gpu_test/7_3d_scene/app.cpp
+++ b/private/sdl_gpu_test/7_3d_scene/app.cpp
@@ -43,10 +43,7 @@ void ThreeDSceneApp::init(const AppInitArgs& args)
 
     // open gamepad
     const std::vector<SDL_JoystickID> gamepads = sdlpp::getGamepads();
-    if (!gamepads.empty())
-    {
-        mGamepad.open(gamepads[0]);
-    }
+    mGamepad = openFirstGamepad();
 
     // init the UI
     mWidgetTree.init({.renderer = &mUIRenderer});
@@ -57,21 +54,10 @@ void ThreeDSceneApp::init(const AppInitArgs& args)
         .posY = 100,
         .textHeight = 20
     });
-    mButton = mWidgetTree.getRootWidget().emplaceChild<Button>({
-        .text = "Click Me!",
-        .posX = 100,
-        .posY = 500
-    });
 
     // init the scene
     mScene.init({.renderer = &mSceneRenderer});
     loadScene("scenes/test_scene.yml", mScene);
-    // mScene.getRootNode().emplaceChild<MeshNode>({.mesh = "meshes/cube.obj", .texture = "meshes/cube.png"});
-
-    mButton->clicked.connect([&]()
-    {
-        mButton->setText("Thanks!");
-    });
 
     mWindow.setRelativeMouseMode(true);
 }
diff --git a/private/sdl_gpu_test/7_3d_scene/app.hpp b/private/sdl_gpu_test/7_3d_scene/app.hpp
index 4919ee0..5fab271 100644
--- a/private/sdl_gpu_test/7_3d_scene/app.hpp
+++ b/private/sdl_gpu_test/7_3d_scene/app.hpp
@@ -5,7 +5,6 @@
 #define SDL_GPU_TEST_PRIVATE_SDL_GPU_TEST_7_3D_SCENE_APP_HPP_INCLUDED 1
 
 #include "../application.hpp"
-#include "../gui/button.hpp"
 #include "../gui/label.hpp"
 #include "../gui/widget.hpp"
 #include "../gui/ui_renderer.hpp"
@@ -27,7 +26,6 @@ private:
     Scene mScene;
 
     Label* mLabel = nullptr;
-    Button* mButton = nullptr;
 
     sdlpp::Gamepad mGamepad;
     Uint32 mLastSwapchainWidth = 1280;
diff --git a/private/sdl_gpu_test/8_post_process/app.cpp b/private/sdl_gpu_test/8_post_process/app.cpp
new file mode 100644
index 0000000..eed19e0
--- /dev/null
+++ b/private/sdl_gpu_test/8_post_process/app.cpp
@@ -0,0 +1,263 @@
+
+#include "./app.hpp"
+
+#include <numbers>
+
+#include <glm/mat4x4.hpp>
+#include <glm/vec2.hpp>
+#include <glm/vec4.hpp>
+#include <glm/gtx/norm.hpp>
+
+#include "../scene/mesh.hpp"
+#include "../scene/loader.hpp"
+#include "../sdlpp/keyboard.hpp"
+#include "../util/bitmap.hpp"
+#include "../util/mesh.hpp"
+#include "../util/spdlog_wrapper.hpp"
+
+namespace sdl_gpu_test
+{
+namespace
+{
+inline constexpr float PITCH_MIN = -0.49f * std::numbers::pi_v<float>;
+inline constexpr float PITCH_MAX = 0.49f * std::numbers::pi_v<float>;
+inline constexpr Uint16 AXIS_DEADZONE = 5000;
+inline constexpr sdlpp::GPUTextureFormat COLOR_BUFFER_FORMAT = sdlpp::GPUTextureFormat::R32G32B32A32_FLOAT;
+inline constexpr sdlpp::GPUTextureFormat POSTPROCESS_BUFFER_FORMAT = sdlpp::GPUTextureFormat::R8G8B8A8_UNORM;
+}
+
+void PostProcessApp::init(const AppInitArgs& args)
+{
+    Application::init(args);
+
+    // create color buffer
+    mColorBuffer.create(mDevice, {
+        .format = sdlpp::GPUTextureFormat::R32G32B32A32_FLOAT,
+        .usage = sdlpp::GPUTextureUsageFlags{.colorTarget = true, .computeStorageRead = true},
+        .width = 1280,
+        .height = 720
+    });
+
+    // create depth buffer
+    mDepthBuffer.create(mDevice, {
+        .format = sdlpp::GPUTextureFormat::D16_UNORM,
+        .usage = sdlpp::GPUTextureUsageFlags{.depthStencilTarget = true},
+        .width = 1280,
+        .height = 720
+    });
+
+    // and the target for the compute pass (SDL_gpu doesn't support using the swapchain in compute shaders)
+    mPostProcessBuffer.create(mDevice, {
+        .format = POSTPROCESS_BUFFER_FORMAT,
+        .usage = sdlpp::GPUTextureUsageFlags{.computeStorageWrite = true},
+        .width = 1280,
+        .height = 720
+    });
+
+    // init UI renderer
+    mUIRenderer.init(*this);
+
+    // init scene renderer
+    mSceneRenderer.init(*this, {
+        .colorBufferFormat = COLOR_BUFFER_FORMAT
+    });
+
+    // and the post processor
+    mPostProcessor.init(*this);
+
+    // open gamepad
+    mGamepad = openFirstGamepad();
+
+    // init the UI
+    mWidgetTree.init({.renderer = &mUIRenderer});
+    mLabel = mWidgetTree.getRootWidget().emplaceChild<Label>({
+        .text = "Test-Text!\nSecond $f00line$fff?",
+        .color = glm::vec4(0.f, 1.f, 0.f, 1.f),
+        .posX = 100,
+        .posY = 100,
+        .textHeight = 20
+    });
+
+    // init the scene
+    mScene.init({.renderer = &mSceneRenderer});
+    loadScene("scenes/test_scene.yml", mScene);
+
+    mWindow.setRelativeMouseMode(true);
+}
+
+void PostProcessApp::update(const AppUpdateArgs& args)
+{
+    Application::update(args);
+
+    processInput(args);
+    mLabel->setText(std::format("Pitch: {:.1f}\nYaw: {:.1f}", mPitch, mYaw));
+    mWidgetTree.revalidateWidgets();
+
+    // begin rendering
+    sdlpp::GPUCommandBuffer cmdBuffer = mDevice.acquireCommandBuffer();
+    Uint32 swapchainWidth = 0, swapchainHeight = 0;
+    const sdlpp::GPUTexture swapchainTexture = cmdBuffer.acquireSwapchainTexture(mWindow, swapchainWidth, swapchainHeight);
+
+    if (swapchainWidth != mLastSwapchainWidth || swapchainHeight != mLastSwapchainHeight)
+    {
+        mColorBuffer.destroy();
+        mColorBuffer.create(mDevice, {
+            .format = COLOR_BUFFER_FORMAT,
+            .usage = sdlpp::GPUTextureUsageFlags{.colorTarget = true, .computeStorageRead = true},
+            .width = swapchainWidth,
+            .height = swapchainHeight
+        });
+        mDepthBuffer.destroy();
+        mDepthBuffer.create(mDevice, {
+            .format = sdlpp::GPUTextureFormat::D16_UNORM,
+            .usage = sdlpp::GPUTextureUsageFlags{.depthStencilTarget = true},
+            .width = swapchainWidth,
+            .height = swapchainHeight
+        });
+        mPostProcessBuffer.destroy();
+        mPostProcessBuffer.create(mDevice, {
+            .format = POSTPROCESS_BUFFER_FORMAT,
+            .usage = sdlpp::GPUTextureUsageFlags{.computeStorageWrite = true},
+            .width = swapchainWidth,
+            .height = swapchainHeight
+        });
+        mLastSwapchainWidth = swapchainWidth;
+        mLastSwapchainHeight = swapchainHeight;
+    }
+
+    // render the scene
+    mSceneRenderer.render({
+        .camera = {
+            .position = mPosition,
+            .pitch = mPitch,
+            .yaw = mYaw
+        },
+        .cmdBuffer = &cmdBuffer,
+        .targetTexture = &mColorBuffer,
+        .depthTexture = &mDepthBuffer,
+        .targetTextureWidth = swapchainWidth,
+        .targetTextureHeight = swapchainHeight
+    });
+
+    // do the post processing
+    mPostProcessor.execute({
+        .cmdBuffer = &cmdBuffer,
+        .sourceTexture = &mColorBuffer,
+        .targetTexture = &mPostProcessBuffer,
+        .textureWidth = swapchainWidth,
+        .textureHeight = swapchainHeight
+    });
+
+    // copy the result to the swapchain image
+    sdlpp::GPUCopyPass copyPass = cmdBuffer.beginCopyPass();
+    copyPass.copyTextureToTexture(
+        /* source = */ {.texture = mPostProcessBuffer},
+        /* destination = */ {.texture = swapchainTexture},
+        /* width = */ swapchainWidth,
+        /* height = */ swapchainHeight
+    );
+    copyPass.end();
+
+    // render the "UI"
+    mUIRenderer.render({
+        .cmdBuffer = &cmdBuffer,
+        .targetTexture = &swapchainTexture,
+        .targetTextureWidth = swapchainWidth,
+        .targetTextureHeight = swapchainHeight
+    });
+
+    // finalize
+    cmdBuffer.submit();
+}
+
+void PostProcessApp::handleKeyboardEvent(const sdlpp::KeyboardEvent& event)
+{
+    switch (event.key)
+    {
+        case SDLK_Q:
+            if (!event.down)
+            {
+                mRunning = false;
+            }
+            break;
+        default: break;
+    }
+}
+
+void PostProcessApp::handleMouseMotionEvent(const sdlpp::MouseMotionEvent& event)
+{
+    mWidgetTree.notifyMouseMoved(event);
+
+    mYaw -= 0.002f * event.xrel;
+    mPitch -= 0.002f * event.yrel;
+}
+
+void PostProcessApp::handleMouseButtonEvent(const sdlpp::MouseButtonEvent& event)
+{
+    mWidgetTree.notifyMouseButton(event);
+}
+
+void PostProcessApp::processInput(const AppUpdateArgs& args)
+{
+    glm::vec2 movement = {0.f, 0.f};
+    const std::span<const SDL_bool> keyboardState = sdlpp::getKeyboardState();
+    if (mGamepad)
+    {
+        Sint16 axisValue = mGamepad.getAxis(sdlpp::GamepadAxis::RIGHTX);
+        if (std::abs(axisValue) > AXIS_DEADZONE)
+        {
+            mYaw -= 0.0001f * args.tickSeconds * static_cast<float>(axisValue);
+        }
+
+        axisValue = mGamepad.getAxis(sdlpp::GamepadAxis::RIGHTY);
+        if (std::abs(axisValue) > AXIS_DEADZONE)
+        {
+            mPitch -= 0.0001f * args.tickSeconds * static_cast<float>(axisValue);
+        }
+
+        axisValue = mGamepad.getAxis(sdlpp::GamepadAxis::LEFTY);
+        if (std::abs(axisValue) > AXIS_DEADZONE)
+        {
+            movement.y -= static_cast<float>(axisValue) / static_cast<float>(std::numeric_limits<Sint16>::max());
+        }
+
+        axisValue = mGamepad.getAxis(sdlpp::GamepadAxis::LEFTX);
+        if (std::abs(axisValue) > AXIS_DEADZONE)
+        {
+            movement.x += static_cast<float>(axisValue) / static_cast<float>(std::numeric_limits<Sint16>::max());
+        }
+    }
+
+    if (keyboardState[SDL_SCANCODE_W])
+    {
+        movement.y += 1.f;
+    }
+    if (keyboardState[SDL_SCANCODE_S])
+    {
+        movement.y -= 1.f;
+    }
+    if (keyboardState[SDL_SCANCODE_A])
+    {
+        movement.x -= 1.f;
+    }
+    if (keyboardState[SDL_SCANCODE_D])
+    {
+        movement.x += 1.f;
+    }
+
+    if (glm::length2(movement) > 1.f)
+    {
+        movement = glm::normalize(movement);
+    }
+
+    mPosition.x -= 2.f * args.tickSeconds * std::sin(mYaw) * movement.y;
+    mPosition.z -= 2.f * args.tickSeconds * std::cos(mYaw) * movement.y;
+
+    mPosition.x += 2.f * args.tickSeconds * std::cos(mYaw) * movement.x;
+    mPosition.z -= 2.f * args.tickSeconds * std::sin(mYaw) * movement.x;
+
+    while (mYaw >= 2.f * std::numbers::pi) { mYaw -= 2.f * std::numbers::pi_v<float>; }
+    while (mYaw < 0.f) { mYaw += 2.f * std::numbers::pi_v<float>; }
+    mPitch = std::clamp(mPitch, PITCH_MIN, PITCH_MAX);
+}
+}
diff --git a/private/sdl_gpu_test/8_post_process/app.hpp b/private/sdl_gpu_test/8_post_process/app.hpp
new file mode 100644
index 0000000..37c03a1
--- /dev/null
+++ b/private/sdl_gpu_test/8_post_process/app.hpp
@@ -0,0 +1,52 @@
+
+#pragma once
+
+#if !defined(SDL_GPU_TEST_PRIVATE_SDL_GPU_TEST_8_POST_PROCESS_APP_HPP_INCLUDED)
+#define SDL_GPU_TEST_PRIVATE_SDL_GPU_TEST_8_POST_PROCESS_APP_HPP_INCLUDED 1
+
+#include "../application.hpp"
+#include "../gui/label.hpp"
+#include "../gui/widget.hpp"
+#include "../gui/ui_renderer.hpp"
+#include "../post_process/post_processor.hpp"
+#include "../scene/scene.hpp"
+#include "../scene/scene_renderer.hpp"
+#include "../sdlpp/gamepad.hpp"
+
+namespace sdl_gpu_test
+{
+class PostProcessApp : public Application
+{
+private:
+    sdlpp::GPUTexture mColorBuffer;
+    sdlpp::GPUTexture mDepthBuffer;
+    sdlpp::GPUTexture mPostProcessBuffer;
+
+    UIRenderer mUIRenderer;
+    WidgetTree mWidgetTree;
+
+    SceneRenderer mSceneRenderer;
+    Scene mScene;
+
+    PostProcessor mPostProcessor;
+
+    Label* mLabel = nullptr;
+
+    sdlpp::Gamepad mGamepad;
+    Uint32 mLastSwapchainWidth = 1280;
+    Uint32 mLastSwapchainHeight = 720;
+    glm::vec3 mPosition = {5.f, 0.f, -7.f};
+    float mYaw = 1.57f;
+    float mPitch = 0.f;
+public:
+    void init(const AppInitArgs& args) override;
+    void update(const AppUpdateArgs& args) override;
+    void handleKeyboardEvent(const sdlpp::KeyboardEvent& event) override;
+    void handleMouseMotionEvent(const sdlpp::MouseMotionEvent& event) override;
+    void handleMouseButtonEvent(const sdlpp::MouseButtonEvent&) override;
+private:
+    void processInput(const AppUpdateArgs& args);
+};
+} // namespace sdl_gpu_test
+
+#endif // !defined(SDL_GPU_TEST_PRIVATE_SDL_GPU_TEST_8_POST_PROCESS_APP_HPP_INCLUDED)
diff --git a/private/sdl_gpu_test/SModule b/private/sdl_gpu_test/SModule
index ddddeb3..ccab8dc 100644
--- a/private/sdl_gpu_test/SModule
+++ b/private/sdl_gpu_test/SModule
@@ -9,6 +9,7 @@ src_files = Split("""
     gui/label.cpp
     gui/ui_renderer.cpp
     gui/widget.cpp
+    post_process/post_processor.cpp
     scene/loader.cpp
     scene/mesh.cpp
     scene/scene.cpp
@@ -26,10 +27,12 @@ src_files = Split("""
     5_input/app.cpp
     6_ui/app.cpp
     7_3d_scene/app.cpp
+    8_post_process/app.cpp
 """)
 
 shader_files = env.Glob("#assets/shaders/glsl/*.frag") \
-    + env.Glob("#assets/shaders/glsl/*.vert")
+             + env.Glob("#assets/shaders/glsl/*.vert") \
+             + env.Glob("#assets/shaders/glsl/*.comp")
 
 env.Append(CPPDEFINES = ['GLM_FORCE_DEPTH_ZERO_TO_ONE', 'GLM_FORCE_RADIANS', 'GLM_ENABLE_EXPERIMENTAL'])
 prog_app = env.UnityProgram(
diff --git a/private/sdl_gpu_test/application.cpp b/private/sdl_gpu_test/application.cpp
index f36d6b8..9e1fc1a 100644
--- a/private/sdl_gpu_test/application.cpp
+++ b/private/sdl_gpu_test/application.cpp
@@ -114,6 +114,16 @@ sdlpp::GPUShader Application::loadShader(const fs::path& path, sdlpp::GPUShaderC
     return shader;
 }
 
+sdlpp::GPUComputePipeline Application::loadComputePipeline(const fs::path& path, sdlpp::GPUComputePipelineCreateArgs args)
+{
+    mijin::TypelessBuffer source = getFileContentsBinary(path);
+    sdlpp::GPUComputePipeline pipeline;
+    args.code = {static_cast<const Uint8*>(source.data()), source.byteSize()};
+    pipeline.create(mDevice, args);
+    return pipeline;
+}
+
+
 sdlpp::GPUTexture Application::loadTexture(const fs::path& path, sdlpp::GPUTextureCreateArgs& inout_args)
 {
     const Bitmap bitmap = loadBitmap(mFileSystem.getPath(path));
@@ -159,4 +169,26 @@ void Application::uploadTextureData(const sdlpp::GPUTexture& texture, const Bitm
     copyPass.end();
     cmdBuffer.submit();
 }
+
+sdlpp::Gamepad Application::openFirstGamepad() const
+{
+    sdlpp::Gamepad gamepad;
+    const std::vector<SDL_JoystickID> gamepads = sdlpp::getGamepads();
+    if (!gamepads.empty())
+    {
+        for (SDL_JoystickID gamepadId : gamepads)
+        {
+            try
+            {
+                gamepad.open(gamepadId);
+                break;
+            }
+            catch(std::runtime_error& err)
+            {
+                spdlog::warn("Error opening gamepad: {}.", err.what());
+            }
+        }
+    }
+    return gamepad;
+}
 }
diff --git a/private/sdl_gpu_test/application.hpp b/private/sdl_gpu_test/application.hpp
index a9f304b..de62374 100644
--- a/private/sdl_gpu_test/application.hpp
+++ b/private/sdl_gpu_test/application.hpp
@@ -10,6 +10,7 @@
 #include <mijin/virtual_filesystem/stacked.hpp>
 
 #include "./sdlpp/event.hpp"
+#include "./sdlpp/gamepad.hpp"
 #include "./sdlpp/gpu.hpp"
 #include "./sdlpp/window.hpp"
 
@@ -74,6 +75,9 @@ public:
     [[nodiscard]]
     sdlpp::GPUShader loadShader(const fs::path& path, sdlpp::GPUShaderCreateArgs args);
 
+    [[nodiscard]]
+    sdlpp::GPUComputePipeline loadComputePipeline(const fs::path& path, sdlpp::GPUComputePipelineCreateArgs args);
+
     [[nodiscard]]
     sdlpp::GPUTexture loadTexture(const fs::path& path, sdlpp::GPUTextureCreateArgs& inout_args);
 
@@ -81,6 +85,9 @@ public:
 
     void uploadTextureData(const sdlpp::GPUTexture& texture, const struct Bitmap& bitmap) const;
 
+    [[nodiscard]]
+    sdlpp::Gamepad openFirstGamepad() const;
+
     template<typename TVertex>
     void uploadVertexData(const sdlpp::GPUBuffer& vertexBuffer, std::span<TVertex> vertices) const;
 };
diff --git a/private/sdl_gpu_test/main.cpp b/private/sdl_gpu_test/main.cpp
index b412cee..78ef930 100644
--- a/private/sdl_gpu_test/main.cpp
+++ b/private/sdl_gpu_test/main.cpp
@@ -10,7 +10,9 @@
 #include "./5_input/app.hpp"
 #include "./6_ui/app.hpp"
 #include "./7_3d_scene/app.hpp"
+#include "./8_post_process/app.hpp"
 #include "./util/spdlog_wrapper.hpp"
+#include "8_post_process/app.hpp"
 
 namespace
 {
@@ -40,7 +42,8 @@ const std::array APPS = {
     makeAppHelper<sdl_gpu_test::TexturedCubeApp>("textured_cube"),
     makeAppHelper<sdl_gpu_test::InputApp>("input"),
     makeAppHelper<sdl_gpu_test::UIApp>("ui"),
-    makeAppHelper<sdl_gpu_test::ThreeDSceneApp>("3d_scene")
+    makeAppHelper<sdl_gpu_test::ThreeDSceneApp>("3d_scene"),
+    makeAppHelper<sdl_gpu_test::PostProcessApp>("post_process")
 };
 }
 
diff --git a/private/sdl_gpu_test/post_process/post_processor.cpp b/private/sdl_gpu_test/post_process/post_processor.cpp
new file mode 100644
index 0000000..40e34a0
--- /dev/null
+++ b/private/sdl_gpu_test/post_process/post_processor.cpp
@@ -0,0 +1,47 @@
+
+#include "./post_processor.hpp"
+
+namespace sdl_gpu_test
+{
+namespace
+{
+inline constexpr Uint32 LOCAL_SIZE_X = 16;
+inline constexpr Uint32 LOCAL_SIZE_Y = 16;
+}
+
+void PostProcessor::init(Application& application)
+{
+    mApplication = &application;
+
+    createPipeline();
+}
+
+void PostProcessor::createPipeline()
+{
+    mPipeline = mApplication->loadComputePipeline("shaders/glsl/post_process.comp.spv", {
+        .format = sdlpp::GPUShaderFormat::SPIRV,
+        .numReadonlyStorageTextures = 1,
+        .numWriteonlyStorageTextures = 1,
+        .threadcountX = LOCAL_SIZE_X,
+        .threadcountY = LOCAL_SIZE_Y
+    });
+}
+
+void PostProcessor::execute(const PostProcessorExecuteArgs& args)
+{
+    const std::array textueBindings = {
+        sdlpp::GPUStorageTextureWriteOnlyBinding{
+            .texture = *args.targetTexture,
+            .cycle = true
+        }
+    };
+    sdlpp::GPUComputePass computePass = args.cmdBuffer->beginComputePass({.storageTextureBindings = textueBindings});
+    computePass.bindComputePipeline(mPipeline);
+    computePass.bindStorageTexture(0, *args.sourceTexture);
+    computePass.dispatch(
+        /* groupCountX = */ (args.textureWidth + LOCAL_SIZE_X - 1) / LOCAL_SIZE_X,
+        /* groupCountY = */ (args.textureHeight + LOCAL_SIZE_Y - 1) / LOCAL_SIZE_Y
+    );
+    computePass.end();
+}
+}
diff --git a/private/sdl_gpu_test/post_process/post_processor.hpp b/private/sdl_gpu_test/post_process/post_processor.hpp
new file mode 100644
index 0000000..55aa45e
--- /dev/null
+++ b/private/sdl_gpu_test/post_process/post_processor.hpp
@@ -0,0 +1,35 @@
+
+#pragma once
+
+#if !defined(SDL_GPU_TEST_PRIVATE_SDL_GPU_TEST_POST_PROCESS_POST_PROCESSOR_HPP_INCLUDED)
+#define SDL_GPU_TEST_PRIVATE_SDL_GPU_TEST_POST_PROCESS_POST_PROCESSOR_HPP_INCLUDED 1
+
+#include "../application.hpp"
+#include "../sdlpp/gpu.hpp"
+
+namespace sdl_gpu_test
+{
+struct PostProcessorExecuteArgs
+{
+    const sdlpp::GPUCommandBuffer* cmdBuffer;
+    const sdlpp::GPUTexture* sourceTexture;
+    const sdlpp::GPUTexture* targetTexture;
+    unsigned textureWidth;
+    unsigned textureHeight;
+};
+
+class PostProcessor
+{
+private:
+    Application* mApplication = nullptr;
+    sdlpp::GPUComputePipeline mPipeline;
+public:
+    void init(Application& application);
+
+    void execute(const PostProcessorExecuteArgs& args);
+private:
+    void createPipeline();
+};
+} // namespace sdl_gpu_test
+
+#endif // !defined(SDL_GPU_TEST_PRIVATE_SDL_GPU_TEST_POST_PROCESS_POST_PROCESSOR_HPP_INCLUDED)
diff --git a/private/sdl_gpu_test/scene/scene_renderer.cpp b/private/sdl_gpu_test/scene/scene_renderer.cpp
index fd538a1..a23bb9b 100644
--- a/private/sdl_gpu_test/scene/scene_renderer.cpp
+++ b/private/sdl_gpu_test/scene/scene_renderer.cpp
@@ -24,11 +24,17 @@ struct PerModelParameters
     glm::mat4 modelToWorld;
 };
 }
-void SceneRenderer::init(Application& application)
+void SceneRenderer::init(Application& application, SceneRendererInitArgs args)
 {
     mApplication = &application;
 
-    createPipeline();
+    if (args.colorBufferFormat == sdlpp::GPUTextureFormat::INVALID)
+    {
+        // default to render directly to the swapchain
+        args.colorBufferFormat = mApplication->getDevice().getSwapchainTextureFormat(mApplication->getWindow());
+    }
+
+    createPipeline(args);
     mSampler.create(mApplication->getDevice(), {});
 }
 
@@ -100,6 +106,7 @@ void SceneRenderer::render(const SceneRendererRenderArgs& args)
         .texture = *args.targetTexture,
         .clearColor = {.r = 0.f, .g = 0.f, .b = 0.f, .a = 1.f},
         .loadOp = sdlpp::GPULoadOp::CLEAR,
+        .cycle = true
     }};
     sdlpp::GPURenderPass renderPass = args.cmdBuffer->beginRenderPass({
         .colorTargetInfos = colorTargets,
@@ -193,7 +200,7 @@ bool SceneRenderer::updateRenderListEntry(mesh_id_t meshId, const RenderListUpda
     return true;
 }
 
-void SceneRenderer::createPipeline()
+void SceneRenderer::createPipeline(const SceneRendererInitArgs& args)
 {
     // create shaders
     const sdlpp::GPUShader vertexShader = mApplication->loadShader("shaders/glsl/scene_renderer.vert.spv", {
@@ -209,7 +216,7 @@ void SceneRenderer::createPipeline()
     });
     std::array colorTargetsDescs = {
         sdlpp::GPUColorTargetDescription{
-            .format = mApplication->getDevice().getSwapchainTextureFormat(mApplication->getWindow()),
+            .format = args.colorBufferFormat,
             .blendState = {
                 .enableBlend = true,
                 .srcColorBlendfactor = sdlpp::GPUBlendFactor::SRC_ALPHA,
diff --git a/private/sdl_gpu_test/scene/scene_renderer.hpp b/private/sdl_gpu_test/scene/scene_renderer.hpp
index 8b6ef0a..dad3e1c 100644
--- a/private/sdl_gpu_test/scene/scene_renderer.hpp
+++ b/private/sdl_gpu_test/scene/scene_renderer.hpp
@@ -37,6 +37,11 @@ struct CameraOptions
     float zFar = 100.f;
 };
 
+struct SceneRendererInitArgs
+{
+    sdlpp::GPUTextureFormat colorBufferFormat = sdlpp::GPUTextureFormat::INVALID;
+};
+
 struct SceneRendererRenderArgs
 {
     CameraOptions camera;
@@ -78,7 +83,7 @@ private:
     sdlpp::GPUGraphicsPipeline mPipeline;
     sdlpp::GPUSampler mSampler;
 public:
-    void init(Application& application);
+    void init(Application& application, SceneRendererInitArgs args = {});
 
     [[nodiscard]]
     std::shared_ptr<SceneMesh> getMesh(const std::string& resourcePath);
@@ -93,7 +98,7 @@ public:
 
     void render(const SceneRendererRenderArgs& args);
 private:
-    void createPipeline();
+    void createPipeline(const SceneRendererInitArgs& args);
 };
 } // namespace sdl_gpu_test
 
diff --git a/private/sdl_gpu_test/sdlpp/gpu.hpp b/private/sdl_gpu_test/sdlpp/gpu.hpp
index 3e2c467..96b3562 100644
--- a/private/sdl_gpu_test/sdlpp/gpu.hpp
+++ b/private/sdl_gpu_test/sdlpp/gpu.hpp
@@ -475,6 +475,26 @@ struct GPUDepthStencilTargetInfo
 static_assert(sizeof(GPUDepthStencilTargetInfo) == sizeof(SDL_GPUDepthStencilTargetInfo)
               && alignof(GPUDepthStencilTargetInfo) == alignof(SDL_GPUDepthStencilTargetInfo));
 
+struct GPUStorageTextureWriteOnlyBinding
+{
+    SDL_GPUTexture* texture = nullptr;
+    Uint32 mipLevel = 0;
+    Uint32 layer = 0;
+    bool cycle = false;
+    std::array<Uint8, 3> padding_;
+};
+static_assert(sizeof(GPUStorageTextureWriteOnlyBinding) == sizeof(SDL_GPUStorageTextureWriteOnlyBinding)
+              && alignof(GPUStorageTextureWriteOnlyBinding) == alignof(SDL_GPUStorageTextureWriteOnlyBinding));
+
+struct GPUStorageBufferWriteOnlyBinding
+{
+    SDL_GPUBuffer* buffer = nullptr;
+    bool cycle = false;
+    std::array<Uint8, 3> padding;
+};
+static_assert(sizeof(GPUStorageBufferWriteOnlyBinding) == sizeof(SDL_GPUStorageBufferWriteOnlyBinding)
+              && alignof(GPUStorageBufferWriteOnlyBinding) == alignof(SDL_GPUStorageBufferWriteOnlyBinding));
+
 struct GPUTransferBufferLocation
 {
     SDL_GPUTransferBuffer* transferBuffer;
@@ -513,6 +533,17 @@ struct GPUTextureRegion
 };
 static_assert(sizeof(GPUTextureRegion) == sizeof(SDL_GPUTextureRegion));
 
+struct GPUTextureLocation
+{
+    SDL_GPUTexture* texture = nullptr;
+    Uint32 mipLevel = 0;
+    Uint32 layer = 0;
+    Uint32 x = 0;
+    Uint32 y = 0;
+    Uint32 z = 0;
+};
+static_assert(sizeof(GPUTextureLocation) == sizeof(SDL_GPUTextureLocation));
+
 using GPUBufferBinding = SDL_GPUBufferBinding;
 
 using GPUTextureSamplerBinding = SDL_GPUTextureSamplerBinding;
@@ -598,6 +629,60 @@ public:
     friend class GPUCommandBuffer;
 };
 
+class GPUComputePass : public Base<SDL_GPUComputePass, GPUComputePass>
+{
+public:
+    GPUComputePass() noexcept = default;
+    GPUComputePass(const GPUComputePass&) = delete;
+    GPUComputePass(GPUComputePass&& other) noexcept : Base(std::move(other)) {}
+
+    GPUComputePass& operator=(const GPUComputePass&) = delete;
+    GPUComputePass& operator=(GPUComputePass&& other) noexcept
+    {
+        Base::operator=(std::move(other));
+        return *this;
+    }
+    auto operator<=>(const GPUComputePass& other) const noexcept = default;
+
+    void end() noexcept
+    {
+        SDL_EndGPUComputePass(mHandle);
+        mHandle = nullptr;
+    }
+
+    void destroy() noexcept
+    {
+        MIJIN_ASSERT(mHandle == nullptr, "Computepass has not been ended.");
+    }
+
+    void bindComputePipeline(SDL_GPUComputePipeline* computePipeline) const noexcept
+    {
+        SDL_BindGPUComputePipeline(mHandle, computePipeline);
+    }
+
+    void bindStorageTextures(Uint32 slotIndex, std::span<SDL_GPUTexture* const> storageTextures) const noexcept
+    {
+        SDL_BindGPUComputeStorageTextures(
+            /* compute_pass     = */ mHandle,
+            /* first_slot       = */ slotIndex,
+            /* storage_textures = */ storageTextures.data(),
+            /* num_bindings     = */ static_cast<Uint32>(storageTextures.size())
+        );
+    }
+
+    void bindStorageTexture(Uint32 slotIndex, SDL_GPUTexture* storageTexture) const noexcept
+    {
+        bindStorageTextures(slotIndex, {&storageTexture, 1});
+    }
+
+    void dispatch(Uint32 groupCountX, Uint32 groupCountY = 1, Uint32 groupCountZ = 1) const noexcept
+    {
+        SDL_DispatchGPUCompute(mHandle, groupCountX, groupCountY, groupCountZ);
+    }
+
+    friend class GPUCommandBuffer;
+};
+
 class GPUCopyPass : public Base<SDL_GPUCopyPass, GPUCopyPass>
 {
 public:
@@ -644,6 +729,20 @@ public:
         );
     }
 
+    void copyTextureToTexture(const GPUTextureLocation& source, const GPUTextureLocation& destination, Uint32 width,
+                              Uint32 height, Uint32 depth = 1, bool cycle = false) const noexcept
+    {
+        SDL_CopyGPUTextureToTexture(
+            /* copy_pass   = */ mHandle,
+            /* source      = */ std::bit_cast<const SDL_GPUTextureLocation*>(&source),
+            /* destination = */ std::bit_cast<const SDL_GPUTextureLocation*>(&destination),
+            /* w           = */ width,
+            /* h           = */ height,
+            /* d           = */ depth,
+            /* cycle       = */ cycle
+        );
+    }
+
     friend class GPUCommandBuffer;
 };
 
@@ -787,6 +886,12 @@ struct GPUBeginRenderPassArgs
     std::optional<GPUDepthStencilTargetInfo> depthStencilTargetInfo;
 };
 
+struct GPUBeginComputePassArgs
+{
+    std::span<const GPUStorageTextureWriteOnlyBinding> storageTextureBindings;
+    std::span<const GPUStorageBufferWriteOnlyBinding> storageBufferBindings;
+};
+
 class GPUCommandBuffer : public Base<SDL_GPUCommandBuffer, GPUCommandBuffer>
 {
 public:
@@ -826,6 +931,20 @@ public:
         return renderPass;
     }
 
+    [[nodiscard]]
+    GPUComputePass beginComputePass(const GPUBeginComputePassArgs& args) const noexcept
+    {
+        GPUComputePass computePass;
+        computePass.mHandle = SDL_BeginGPUComputePass(
+            /* command_buffer               = */ mHandle,
+            /* storage_texture_bindings     = */ std::bit_cast<const SDL_GPUStorageTextureWriteOnlyBinding*>(args.storageTextureBindings.data()),
+            /* num_storage_texture_bindings = */ static_cast<Uint32>(args.storageTextureBindings.size()),
+            /* storage_buffer_bindings      = */ std::bit_cast<const SDL_GPUStorageBufferWriteOnlyBinding*>(args.storageBufferBindings.data()),
+            /* num_storage_buffer_bindings  = */ static_cast<Uint32>(args.storageBufferBindings.size())
+        );
+        return computePass;
+    }
+
     [[nodiscard]]
     GPUCopyPass beginCopyPass() const noexcept
     {
@@ -1039,6 +1158,73 @@ public:
     }
 };
 
+struct GPUComputePipelineCreateArgs
+{
+    std::span<const Uint8> code;
+    std::string entrypoint = "main";
+    GPUShaderFormat format;
+    Uint32 numReadonlyStorageTextures = 0;
+    Uint32 numReadonlyStorageBuffers = 0;
+    Uint32 numWriteonlyStorageTextures = 0;
+    Uint32 numWriteonlyStorageBuffers = 0;
+    Uint32 numUniformBuffers = 0;
+    Uint32 threadcountX = 1;
+    Uint32 threadcountY = 1;
+    Uint32 threadcountZ = 1;
+};
+
+class GPUComputePipeline : public BaseWithDevice<SDL_GPUComputePipeline, GPUComputePipeline>
+{
+public:
+    GPUComputePipeline() noexcept = default;
+    GPUComputePipeline(const GPUComputePipeline&) = delete;
+    GPUComputePipeline(GPUComputePipeline&& other) noexcept : BaseWithDevice(std::move(other)) {}
+
+    GPUComputePipeline& operator=(const GPUComputePipeline&) = delete;
+    GPUComputePipeline& operator=(GPUComputePipeline&& other) noexcept
+    {
+        BaseWithDevice::operator=(std::move(other));
+        return *this;
+    }
+    auto operator<=>(const GPUComputePipeline& other) const noexcept = default;
+
+    void create(SDL_GPUDevice* device, const GPUComputePipelineCreateArgs& args)
+    {
+        MIJIN_ASSERT(mHandle == nullptr, "GPUComputePipeline has already been created.");
+        const SDL_GPUComputePipelineCreateInfo createInfo =
+        {
+            .code_size = args.code.size(),
+            .code = args.code.data(),
+            .entrypoint = args.entrypoint.c_str(),
+            .format = static_cast<SDL_GPUShaderFormat>(args.format),
+            .num_readonly_storage_textures = args.numReadonlyStorageTextures,
+            .num_readonly_storage_buffers = args.numReadonlyStorageBuffers,
+            .num_writeonly_storage_textures = args.numWriteonlyStorageTextures,
+            .num_writeonly_storage_buffers = args.numWriteonlyStorageBuffers,
+            .threadcount_x = args.threadcountX,
+            .threadcount_y = args.threadcountY,
+            .threadcount_z = args.threadcountZ,
+            .props = 0
+        };
+        mHandle = SDL_CreateGPUComputePipeline(device, &createInfo);
+        if (mHandle == nullptr)
+        {
+            throw SDLError();
+        }
+        mDevice = device;
+    }
+
+    void destroy() noexcept
+    {
+        if (mHandle != nullptr)
+        {
+            SDL_ReleaseGPUComputePipeline(mDevice, mHandle);
+            mDevice = nullptr;
+            mHandle = nullptr;
+        }
+    }
+};
+
 struct GPUShaderCreateArgs
 {
     std::span<const Uint8> code;