MiniDxNN v0.3.0

Author: shoikeda

.gitignore

@@ -50,8 +50,8 @@ build/*
 !/build/.gitkeep
 
 # Runtime
-unittest/runtime/*
-!/unittest/runtime/.gitkeep
+third_party/runtime/*
+!/third_party/runtime/.gitkeep
 
 # CMake
 CMakeLists.txt.user

CMakeLists.txt

@@ -15,7 +15,7 @@ include(${CMAKE_CURRENT_LIST_DIR}/cmake/project.cmake)
 include(${CMAKE_CURRENT_LIST_DIR}/cmake/utility.cmake)
 
 set(project_desc "A minimal DirectX-based neural network library")
-project(MiniDXNN DESCRIPTION ${project_desc} VERSION 0.2.0 LANGUAGES CXX)
+project(MiniDXNN DESCRIPTION ${project_desc} VERSION 0.3.0 LANGUAGES CXX)
 
   # Configure project options (build flags, sanitizers, etc.)
   setProjectOptions()

NOTICE.md

@@ -15,3 +15,12 @@ and contains third-party components under various open source licenses.
 * [GoogleTest](https://github.com/google/googletest)
   - **used for:** To introduce testing framework
   - **license:** BSD-3-Clause license
+* [stb_image / stb_image_write](https://github.com/nothings/stb)
+  - **used for:** PNG image loading and writing
+  - **license:** MIT License / Public Domain
+
+## Resources
+
+* [tokyo-skyline.png](resources/image/tokyo-skyline.png)
+  - **source:** [Tokyo Skyline at Night with Tokyo Tower — Pexels](https://www.pexels.com/photo/tokyo-skyline-at-night-with-tokyo-tower-31558042/)
+  - **license:** [Pexels License](https://www.pexels.com/license/) — Free for personal and commercial use, no attribution required

README.md

@@ -1,21 +1,21 @@
-# MiniDXNN — MLP Inference & Training on DirectX 12 with Cooperative Vector
+# MiniDXNN — MLP Inference & Training on DirectX 12 with LinAlg Matrix
 
 ![CMake build on Windows](../../actions/workflows/cmake.yaml/badge.svg)
 
 <p align="center"><img src="./docs/minidxnn-logo.png" width="200"></p>
 
-An implementation of **MLP** (Multi-Layer Perceptron) inference and training using DirectX 12 [Cooperative Vector][coop-vec-spec]. This library demonstrates GPU-accelerated neural network inference and training with cutting-edge shader features.
+An implementation of **MLP** (Multi-Layer Perceptron) inference and training using DirectX 12 [LinAlg Matrix][linalg-spec]. This library demonstrates GPU-accelerated neural network inference and training with cutting-edge shader features.
 
-- 🚀 **High Performance**: GPU-accelerated inference and training using [Cooperative Vector][coop-vec-overview]
+- 🚀 **High Performance**: GPU-accelerated inference and training using [LinAlg Matrix][linalg-overview]
 - 🔧 **Flexible Architecture**: Configurable layers, activations, and data types
 - 🎯 **Single-header HLSL**: Easy to integrate into any DX12 project
 
 ## Requirements
 
 - **OS**: Windows 11 with [Developer Mode][win-dev-mode] enabled
-- **GPU**: Supports Shader Model 6.9 and Cooperative Vector in D3D12 (AMD Radeon™ RX 9000 Series GPUs or equivalent NVIDIA)
+- **GPU**: Supports Shader Model 6.10 and LinAlg Matrix in D3D12 (AMD Radeon™ RX 9000 Series GPUs or equivalent NVIDIA)
 - **Build**: CMake ≥ 3.21, Visual Studio 2022 (C++20), Windows SDK
-- **DX12 Runtime**: [Agility SDK 1.717.1-preview][dx12-agility-sdk-download], [DXC v1.8.2505.1][dx12-dxc-download]
+- **DX12 Runtime**: [Agility SDK 1.720-preview][dx12-agility-sdk-download], [DXC v1.10.2605.2][dx12-dxc-download]
 - **Python**: Python 3.8+ with PyTorch (optional, for example python training)
 
 ## Getting Started
@@ -39,11 +39,13 @@ Example binaries are output to `build/example/Release/`. Run them from `build/ex
 
 ## DX12 Setup
 
-⚠️ **Important**: As of early 2026, Cooperative Vector requires experimental feature support.
+⚠️ **Important**: As of early 2026, LinAlg Matrix requires experimental feature support.
 
-1. Install a [Cooperative Vector supported driver][coop-vec-driver]
+1. Install a [LinAlg Matrix supported driver][linalg-driver]
 2. Enable [Experimental Shader Model][dx-experimental-shader-model] with [D3D12EnableExperimentalFeatures][dx-enable-experimental-features] **before** creating the device
-3. Compile shaders with **Shader Model 6.9**
+3. Compile shaders with **Shader Model 6.10**
+
+For a detailed walkthrough — including feature checks, weight matrix conversion (`GetLinearAlgebraMatrixConversionDestinationInfo` / `ConvertLinearAlgebraMatrix`), bias alignment, and full sample code — see the **[LinAlg Matrix MLP Guide](docs/linalg_matrix_mlp.md)**.
 
 ## HLSL Usage
 
@@ -147,23 +149,26 @@ MiniDXNN/
 | **Operations** | Forward pass (inference), backward pass (training with gradient accumulation) |
 | **Activations** | Identity, Sigmoid, ReLU, Leaky ReLU (custom activations supported — e.g. Tanh) |
 | **Data type** | float16 (`DATA_TYPE_FLOAT16`) — currently the only tested type |
-| **Matrix layout** | Row-major (`MATRIX_LAYOUT_ROW_MAJOR`) — currently the only tested layout |
+| **Matrix layout** | Row-major, Column-major, Mul-optimal, Outer-product-optimal |
 
 ## Examples
 
 | # | Name | Description |
 |---|------|-------------|
 | 01 | [Texture Inference](./example/01_texture_inference) | Load a pre-trained MLP binary and reconstruct a texture on the GPU |
 | 02 | [Texture Training](./example/02_texture_training) | Train an MLP on-GPU to learn a 2D texture pattern, then reconstruct it |
+| 03 | [Texture Compression with Input Encoding](./example/03_texture_compression_with_input_encoding) | Train with positional/grid input encoding for higher-quality texture compression |
 
 See [example/README.md](./example/README.md) for step-by-step instructions.
 
 ## Documentation
 
+- [LinAlg Matrix MLP Guide](docs/linalg_matrix_mlp.md) — Step-by-step setup, weight conversion, bias alignment, and shader compilation
 - [HLSL API Reference](docs/mlp_hlsl.md) — `mlp.hlsl` types, functions, and memory layout
 - [Example Guide](example/README.md) — building and running the examples
-- [Cooperative Vector Spec][coop-vec-spec] — HLSL specification
-- [D3D12 Cooperative Vector Blog][coop-vec-overview] — overview and getting started
+- [LinAlg Matrix Spec][linalg-spec] — HLSL specification
+- [D3D12 LinAlg Matrix Overview][linalg-overview] — runtime feature support and getting started
+- [LinAlg Examples][linalg-examples] — official example code
 
 ## License
 
@@ -177,16 +182,22 @@ Copyright (c) 2026 Advanced Micro Devices, Inc. All rights reserved.
 - [gfx](https://github.com/gboisse/gfx) — MIT
 - [CLI11](https://github.com/CLIUtils/CLI11) — BSD-3-Clause
 - [GoogleTest](https://github.com/google/googletest) — BSD-3-Clause
+- [stb_image](https://github.com/nothings/stb) — MIT/Public Domain
 
 See [NOTICE.md](NOTICE.md) for details.
 
+### CMake dependency downloads
+
+When building with GPU support (without `MINIDXNN_CPP_FALLBACK_ONLY`), CMake auto-downloads dependencies to `third_party/gfx_dep/gfx/third_party/`.
+
 ---
 
-[coop-vec-spec]: https://github.com/microsoft/hlsl-specs/blob/main/proposals/0029-cooperative-vector.md
-[coop-vec-overview]: https://devblogs.microsoft.com/directx/cooperative-vector/
-[coop-vec-driver]: https://devblogs.microsoft.com/directx/cooperative-vector/#get-running
+[linalg-spec]: https://github.com/microsoft/hlsl-specs/blob/main/proposals/0035-linalg-matrix.md
+[linalg-overview]: https://microsoft.github.io/DirectX-Specs/d3d/D3D12LinearAlgebraRuntimeFeatureSupport.html#tier-1-support
+[linalg-driver]: https://devblogs.microsoft.com/directx/shader-model-6-10-agilitysdk-720-preview/
+[linalg-examples]: https://github.com/llvm-beanz/linalg-examples
 [win-dev-mode]: https://learn.microsoft.com/en-us/windows/apps/get-started/enable-your-device-for-development
 [dx-enable-experimental-features]: https://learn.microsoft.com/en-us/windows/win32/api/d3d12/nf-d3d12-d3d12enableexperimentalfeatures
 [dx-experimental-shader-model]: https://devblogs.microsoft.com/directx/ser/#availability
-[dx12-agility-sdk-download]: https://www.nuget.org/api/v2/package/Microsoft.Direct3D.D3D12/1.717.1-preview
-[dx12-dxc-download]: https://github.com/microsoft/DirectXShaderCompiler/releases/download/v1.8.2505.1/dxc_2025_07_14.zip
+[dx12-agility-sdk-download]: https://devblogs.microsoft.com/directx/directx12agility/
+[dx12-dxc-download]: https://github.com/microsoft/DirectXShaderCompiler/releases/tag/v1.10.2605.2

cmake/compiler.cmake

@@ -37,7 +37,7 @@ function(setCxxCompileFlags target scope)
                  /options:strict      # Strict options checking
                  )
   set(msvc_linker_flags)
-  set(msvc_definitions)
+  set(msvc_definitions _CRT_SECURE_NO_WARNINGS)  # Suppress deprecated CRT warnings from third-party headers (stb_image)
 
   # GCC compiler flags
   set(gcc_flags)
@@ -144,6 +144,7 @@ function(setCxxWarningFlags target scope)
                        /wd4514   # unreferenced inline function removed (informational)
                        /wd4625   # copy constructor implicitly deleted (GoogleTest)
                        /wd4626   # copy assignment implicitly deleted (GoogleTest)
+                       /wd4702   # unreachable code (can be triggered by compile-time branching)
                        /wd4710   # function not inlined (compiler decision, informational)
                        /wd4711   # function selected for automatic inline expansion (informational)
                        /wd4820   # struct padding added (informational)

cmake/utility.cmake

@@ -38,11 +38,34 @@ endfunction(checkTarget)
 ##   dest_path: Destination path for the symlink
 ## Note: On Windows, developer mode should be enabled for symlink creation without admin privileges
 function(createSymlink source_path dest_path)
-  file(CREATE_LINK "${source_path}" "${dest_path}" 
-    RESULT link_result 
-    COPY_ON_ERROR 
+  file(CREATE_LINK "${source_path}" "${dest_path}"
+    RESULT link_result
+    COPY_ON_ERROR
     SYMBOLIC)
   if(link_result)
     message(FATAL_ERROR "Failed to create symlink from ${source_path} to ${dest_path}: ${link_result}")
   endif()
 endfunction(createSymlink)
+
+
+## Copy runtime DLL/SO overrides from third_party/runtime to the target's output directory
+## If third_party/runtime contains DLL or SO files, they overwrite the corresponding
+## files already copied by TARGET_RUNTIME_DLLS, allowing custom driver/compiler builds.
+## Args:
+##   target: The executable target whose output directory receives the overrides
+function(copyRuntimeOverrides target)
+  cmake_path(SET runtime_dir "${CMAKE_SOURCE_DIR}/third_party/runtime")
+  if(WIN32)
+    file(GLOB runtime_files "${runtime_dir}/*.dll")
+  else()
+    file(GLOB runtime_files "${runtime_dir}/*.so" "${runtime_dir}/*.so.*")
+  endif()
+  if(runtime_files)
+    foreach(runtime_file IN LISTS runtime_files)
+      cmake_path(GET runtime_file FILENAME filename)
+      add_custom_command(TARGET ${target} POST_BUILD
+        COMMAND ${CMAKE_COMMAND} -E copy "${runtime_file}" "$<TARGET_FILE_DIR:${target}>/${filename}"
+        COMMENT "Overriding ${filename} from third_party/runtime")
+    endforeach()
+  endif()
+endfunction(copyRuntimeOverrides)