vapoursynth-bm3dhip 2.16__tar.gz

vapoursynth_bm3dhip-2.16/CMakeLists.txt

@@ -0,0 +1,8 @@
+cmake_minimum_required(VERSION 3.20)
+project(BM3DCUDA_Package)
+
+# Define the include path relative to this wrapper
+set(VAPOURSYNTH_INCLUDE_DIRECTORY "${CMAKE_CURRENT_SOURCE_DIR}/vapoursynth/include" CACHE PATH "Path to VapourSynth headers")
+
+# Add the submodule
+add_subdirectory(VapourSynth-BM3DCUDA)

vapoursynth_bm3dhip-2.16/DEVELOPMENT.md

@@ -0,0 +1,23 @@
+# Links & Reminder Notes
+
+## Compatibility Matrix
+
+https://rocm.docs.amd.com/en/latest/compatibility/compatibility-matrix.html#gpu-compatibility
+
+## How to install via package manager
+
+https://rocm.docs.amd.com/projects/install-on-linux/en/latest/install/install-methods/package-manager-index.html
+
+Currently targetting RHEL 8 -> glic 2.28
+
+## GPU_TARGETS
+
+https://rocm.docs.amd.com/projects/install-on-linux/en/docs-6.4.2/reference/system-requirements.html#supported-gpus
+https://rocm.docs.amd.com/projects/install-on-linux/en/latest/reference/system-requirements.html#supported-gpus
+
+## HIP Versionning
+
+- Linux has individual version available on package managers.
+- Windows has an installer AMD-Software-PRO-Edition released each quarter of the year. https://www.amd.com/en/developer/resources/rocm-hub/hip-sdk.html
+  - 25.Q3-WinSvr2022 -> ROCm 6.4.2
+  - 26.Q1-Win11 -> ROCm 7.1.1

vapoursynth_bm3dhip-2.16/PKG-INFO

@@ -0,0 +1,189 @@
+Metadata-Version: 2.4
+Name: vapoursynth-bm3dhip
+Version: 2.16
+Summary: BM3D denoising filter for VapourSynth, implemented in HIP
+Author: WolframRhodium
+Maintainer-Email: =?utf-8?b?VmFyZMOr?= <ichunjo.le.terrible@gmail.com>
+License-Expression: GPL-2.0-or-later
+License-File: VapourSynth-BM3DCUDA/LICENSE
+Project-URL: Source Code, https://github.com/WolframRhodium/VapourSynth-BM3DCUDA
+Project-URL: Bug Tracker, https://github.com/WolframRhodium/VapourSynth-BM3DCUDA/issues
+Requires-Python: >=3.12
+Requires-Dist: vapoursynth>=74
+Description-Content-Type: text/markdown
+
+# VapourSynth-BM3DHIP
+
+This package contains the HIP implementation of the [VapourSynth-BM3DCUDA](https://github.com/WolframRhodium/VapourSynth-BM3DCUDA) filter.
+
+## Installation
+
+```bash
+pip install vapoursynth-bm3dhip
+```
+
+---
+
+Detailed parameter information from the parent project follows.
+
+---
+# VapourSynth-BM3DCUDA
+
+Copyright© 2021 WolframRhodium
+
+BM3D denoising filter for VapourSynth, implemented in CUDA.
+
+## Description
+
+- Please check [VapourSynth-BM3D](https://github.com/HomeOfVapourSynthEvolution/VapourSynth-BM3D).
+
+- The `_rtc` version compiles GPU code at runtime, which might runs faster than standard version at the cost of a slight overhead.
+
+- The `cpu` version is implemented in AVX and AVX2 intrinsics, serves as a reference implementation on CPU. However, _bitwise identical_ outputs are not guaranteed across CPU and CUDA implementations.
+
+## Requirements
+
+- CPU with AVX support.
+
+- CUDA-enabled GPU(s) of [compute capability](https://developer.nvidia.com/cuda-gpus) 5.0 or higher (Maxwell+).
+
+- GPU driver 450 or newer.
+
+The minimum requirement on compute capability is 3.5, which requires manual compilation (specifying nvcc flag `-gencode arch=compute_35,code=sm_35`).
+
+The `cpu` version does not require any external libraries but requires AVX2 support on CPU in addition.
+
+## Parameters
+
+```python3
+{bm3dcuda, bm3dcuda_rtc, bm3dcpu}.BM3D(clip clip[, clip ref=None, float[] sigma=3.0, int[] block_step=8, int[] bm_range=9, int radius=0, int[] ps_num=2, int[] ps_range=4, bint chroma=False, int device_id=0, bool fast=True, int extractor_exp=0])
+```
+
+- clip:
+
+    The input clip. Must be of 32 bit float format. Each plane is denoised separately if `chroma` is set to `False`. Data of unprocessed planes is undefined. Frame properties of the output clip are copied from it.
+
+- ref:
+
+    The reference clip. Must be of the same format, width, height, number of frames as `clip`.
+
+    Used in block-matching and as the reference in empirical Wiener filtering, i.e. `bm3d.Final` / `bm3d.VFinal`:
+
+    ```python3
+    basic = core.{bm3dcpu, bm3dcuda, bm3dcuda_rtc}.BM3D(src, radius=0)
+    final = core.{bm3d...}.BM3D(src, ref=basic, radius=0)
+
+    vbasic = core.{bm3d...}.BM3D(src, radius=radius_nonzero).bm3d.VAggregate(radius=radius_nonzero)
+    vfinal = core.{bm3d...}.BM3D(src, ref=vbasic, radius=r).bm3d.VAggregate(radius=r)
+    
+    # alternatively, using the v2 interface
+    basic_or_vbasic = core.{bm3dcpu, bm3dcuda, bm3dcuda_rtc}.BM3Dv2(src, radius=r)
+    final_or_vfinal = core.{bm3d...}.BM3Dv2(src, ref=basic_or_vbasic, radius=r)
+    ```
+
+    corresponds to the followings (ignoring color space handling and other differences in implementation), respectively
+
+    ```python3
+    basic = core.bm3d.Basic(clip)
+    final = core.bm3d.Final(basic, ref=basic)
+
+    vbasic = core.bm3d.VBasic(src, radius=r).bm3d.VAggregate(radius=r, sample=1)
+    vfinal = core.bm3d.VFinal(src, ref=vbasic, radius=r).bm3d.VAggregate(radius=r)
+    ```
+
+- sigma:
+    The strength of denoising for each plane.
+
+    The strength is similar (but not strictly equal) as `VapourSynth-BM3D` due to differences in implementation. (coefficient normalization is not implemented, for example)
+
+    Default `[3,3,3]`.
+
+- block_step, bm_range, radius, ps_num, ps_range:
+
+    Same as those in `VapourSynth-BM3D`.
+
+    If `chroma` is set to `True`, only the first value is in effect.
+
+    Otherwise an array of values may be specified for each plane (except `radius`).
+    
+    **Note**: It is generally not recommended to take a large value of `ps_num` as current implementations do not take duplicate block-matching candidates into account during temporary searching, which may leads to regression in denoising quality. This issue is not present in `VapourSynth-BM3D`.
+
+    **Note2**: Lowering the value of "block_step" will be useful in reducing blocking artifacts at the cost of slower processing.
+
+- chroma:
+
+    CBM3D algorithm. `clip` must be of `YUV444PS` format.
+
+    Y channel is used in block-matching of chroma channels.
+
+    Default `False`.
+
+- device_id:
+
+    Set GPU to be used.
+
+    Default `0`.
+
+- fast:
+
+    Multi-threaded copy between CPU and GPU at the expense of 4x memory consumption.
+
+    Default `True`.
+
+- extractor_exp:
+
+    Used for deterministic (bitwise) output. This parameter is not present in the `cpu` version since the implementation always produces deterministic output.
+
+    [Pre-rounding](https://ieeexplore.ieee.org/document/6545904) is employed for associative floating-point summation.
+
+    The value should be a positive integer not less than 3, and may need to be higher depending on the source video and filter parameters.
+
+    Default `0`. (non-determinism)
+
+## Notes
+
+- `bm3d.VAggregate` should be called after temporal filtering, as in `VapourSynth-BM3D`. Alternatively, you may use the `BM3Dv2()` interface for both spatial and temporal denoising in one step.
+
+- The `_rtc` version has three additional experimental parameters:
+
+    - bm_error_s: (string)
+
+        Specify cost for block similarity measurement.
+
+        Currently implemented costs: 
+        `SSD` (Sum of Squared Differences), 
+        `SAD` (Sum of Absolute Differences), 
+        `ZSSD` (Zero-mean SSD), 
+        `ZSAD` (Zero-mean SAD), 
+        `SSD/NORM`.
+
+        Default `SSD`.
+
+    - transform_2d_s/transform_1d_s: (string)
+
+        Specify type of transform.
+
+        Currently implemented transforms: 
+        `DCT` (Discrete Cosine Transform), 
+        `Haar` (Haar Transform), 
+        `WHT` (Walsh–Hadamard Transform), 
+        `Bior1.5` (transform based on a bi-orthogonal spline wavelet).
+
+        Default `DCT`.
+
+    These features are not implemented in the standard version due to performance and binary size concerns.
+
+## Statistics
+
+GPU memory consumptions:
+
+`(ref ? 4 : 3) * (chroma ? 3 : 1) * (fast ? 4 : 1) * (2 * radius + 1) * size_of_a_single_frame`
+
+## Compilation
+- The CMake configuration of `BM3DCUDA_RTC` links to NVRTC static library by default, which requires CUDA 11.5 or later.
+
+```bash
+cmake -S . -B build -D CMAKE_BUILD_TYPE=Release -D CMAKE_CUDA_FLAGS="--threads 0 --use_fast_math -Wno-deprecated-gpu-targets" -D CMAKE_CUDA_ARCHITECTURES="50;61-real;75-real;86"
+
+cmake --build build --config Release
+```

vapoursynth_bm3dhip-2.16/README.md

@@ -0,0 +1,15 @@
+# VapourSynth-BM3DHIP
+
+This package contains the HIP implementation of the [VapourSynth-BM3DCUDA](https://github.com/WolframRhodium/VapourSynth-BM3DCUDA) filter.
+
+## Installation
+
+```bash
+pip install vapoursynth-bm3dhip
+```
+
+---
+
+Detailed parameter information from the parent project follows.
+
+---

vapoursynth_bm3dhip-2.16/VapourSynth-BM3DCUDA/.github/workflows/linux-arm64.yml

@@ -0,0 +1,63 @@
+name: Build (Linux, ARM64)
+
+on:
+  push:
+    paths:
+      - 'CMakeLists.txt'
+      - 'source/*'
+      - 'cpu_source/*'
+      - 'hip_source/*'
+      - 'rtc_source/*'
+      - 'sycl_source/*'
+      - '.github/workflows/linux-arm64.yml'
+  workflow_dispatch:
+
+jobs:
+  build-linux:
+    runs-on: ubuntu-24.04-arm
+    steps:
+    - name: Checkout repo
+      uses: actions/checkout@v4
+      with:
+        fetch-depth: 0
+
+    - name: Setup Ninja
+      run: pip install ninja
+
+    - name: Download VapourSynth headers
+      run: |
+        wget -q -O vs.zip https://github.com/vapoursynth/vapoursynth/archive/refs/tags/R57.zip
+        unzip -q vs.zip
+        mv vapoursynth*/ vapoursynth
+
+    - name: Setup CUDA
+      run: |
+        wget https://developer.download.nvidia.com/compute/cuda/repos/ubuntu2404/sbsa/cuda-keyring_1.1-1_all.deb
+        sudo dpkg -i cuda-keyring_1.1-1_all.deb
+        sudo apt-get update
+        sudo apt-get install -y cuda-nvcc-12-8 cuda-cudart-dev-12-8 cuda-nvrtc-dev-12-8
+        echo "PATH=/usr/local/cuda/bin${PATH:+:${PATH}}" >> $GITHUB_ENV
+        echo "CUDA_PATH=/usr/local/cuda" >> $GITHUB_ENV
+        echo "LD_LIBRARY_PATH=/usr/local/cuda/lib${LD_LIBRARY_PATH:+:${LD_LIBRARY_PATH}}" >> $GITHUB_ENV
+
+    - name: Configure (CUDA)
+      run: cmake -S . -B build -G Ninja -LA
+        -D ENABLE_CPU=OFF
+        -D USE_NVRTC_STATIC=ON
+        -D VAPOURSYNTH_INCLUDE_DIRECTORY="`pwd`/vapoursynth/include"
+        -D CMAKE_BUILD_TYPE=Release
+        -D CMAKE_CXX_FLAGS="-Wall -ffast-math"
+        -D CMAKE_CUDA_FLAGS="--threads 0 --use_fast_math --resource-usage -Wno-deprecated-gpu-targets"
+        -D CMAKE_CUDA_ARCHITECTURES="50;61-real;70;75-real;86-real;89-real;90-real;100-real;120-real"
+
+    - name: Build (CUDA)
+      run: cmake --build build --verbose
+
+    - name: Install (CUDA)
+      run: cmake --install build --prefix artifact
+
+    - name: Upload
+      uses: actions/upload-artifact@v4
+      with:
+        name: VapourSynth-BM3DCUDA-Linux-ARM64
+        path: artifact

vapoursynth_bm3dhip-2.16/VapourSynth-BM3DCUDA/.github/workflows/linux.yml

@@ -0,0 +1,140 @@
+name: Build (Linux)
+
+on:
+  push:
+    paths:
+      - 'CMakeLists.txt'
+      - 'source/*'
+      - 'cpu_source/*'
+      - 'hip_source/*'
+      - 'rtc_source/*'
+      - 'sycl_source/*'
+      - '.github/workflows/linux.yml'
+  workflow_dispatch:
+
+jobs:
+  build-linux:
+    runs-on: ubuntu-24.04
+    steps:
+    - name: Checkout repo
+      uses: actions/checkout@v4
+      with:
+        fetch-depth: 0
+
+    - name: Setup Ninja
+      run: pip install ninja
+
+    - name: Download VapourSynth headers
+      run: |
+        wget -q -O vs.zip https://github.com/vapoursynth/vapoursynth/archive/refs/tags/R57.zip
+        unzip -q vs.zip
+        mv vapoursynth*/ vapoursynth
+
+    - name: Setup CUDA
+      run: |
+        wget https://developer.download.nvidia.com/compute/cuda/repos/ubuntu2404/x86_64/cuda-keyring_1.1-1_all.deb
+        sudo dpkg -i cuda-keyring_1.1-1_all.deb
+        sudo apt-get update
+        sudo apt-get install -y cuda-nvcc-12-8 cuda-cudart-dev-12-8 cuda-nvrtc-dev-12-8
+        echo "PATH=/usr/local/cuda/bin${PATH:+:${PATH}}" >> $GITHUB_ENV
+        echo "CUDA_PATH=/usr/local/cuda" >> $GITHUB_ENV
+        echo "LD_LIBRARY_PATH=/usr/local/cuda/lib${LD_LIBRARY_PATH:+:${LD_LIBRARY_PATH}}" >> $GITHUB_ENV
+
+    - name: Configure (CUDA, CPU)
+      run: cmake -S . -B build -G Ninja -LA
+        -D USE_NVRTC_STATIC=ON
+        -D VAPOURSYNTH_INCLUDE_DIRECTORY="`pwd`/vapoursynth/include"
+        -D CMAKE_BUILD_TYPE=Release
+        -D CMAKE_CXX_FLAGS="-Wall -ffast-math -march=x86-64-v3"
+        -D CMAKE_CUDA_FLAGS="--threads 0 --use_fast_math --resource-usage -Wno-deprecated-gpu-targets"
+        -D CMAKE_CUDA_ARCHITECTURES="50;61-real;70;75-real;86-real;89-real;90-real;100-real;120-real"
+
+    - name: Build (CUDA, CPU)
+      run: cmake --build build --verbose
+
+    - name: Install (CUDA, CPU)
+      run: cmake --install build --prefix artifact
+
+    - name: Setup HIP
+      run: |
+        wget https://repo.radeon.com/rocm/rocm.gpg.key -O - | gpg --dearmor | sudo tee /etc/apt/keyrings/rocm.gpg > /dev/null
+        echo "deb [arch=amd64 signed-by=/etc/apt/keyrings/rocm.gpg] https://repo.radeon.com/rocm/apt/6.0 jammy main" | sudo tee --append /etc/apt/sources.list.d/rocm.list
+        echo -e 'Package: *\nPin: release o=repo.radeon.com\nPin-Priority: 600' | sudo tee /etc/apt/preferences.d/rocm-pin-600
+        sudo apt update
+        sudo apt install hip-runtime-amd rocm-device-libs
+
+    - name: Configure (HIP)
+      run: cmake -S . -B build_hip -G Ninja -LA
+        -D CMAKE_BUILD_TYPE=Release
+        -D ENABLE_CPU=OFF -D ENABLE_CUDA=OFF -D ENABLE_HIP=ON
+        -D CMAKE_CXX_COMPILER=/opt/rocm/bin/amdclang++
+        -D CMAKE_CXX_FLAGS="-Wall -ffast-math -munsafe-fp-atomics -Rpass-analysis=kernel-resource-usage"
+        -D VAPOURSYNTH_INCLUDE_DIRECTORY="`pwd`/vapoursynth/include"
+        -D HIP_DIR=/opt/rocm/lib/cmake/hip
+        -D AMDDeviceLibs_DIR=/opt/rocm/lib/cmake/AMDDeviceLibs
+        -D amd_comgr_DIR=/opt/rocm/lib/cmake/amd_comgr
+        -D hsa-runtime64_DIR=/opt/rocm/lib/cmake/hsa-runtime64
+        -D GPU_TARGETS="gfx1010;gfx1011;gfx1012;gfx1030;gfx1031;gfx1032;gfx1033;gfx1034;gfx1035;gfx1036;gfx1100;gfx1101;gfx1102;gfx1103"
+
+    - name: Build (HIP)
+      run: cmake --build build_hip --verbose
+
+    - name: Install (HIP)
+      run: cmake --install build_hip --prefix artifact
+
+    - name: Setup SYCL
+      if: false
+      run: |
+        wget -O- https://apt.repos.intel.com/intel-gpg-keys/GPG-PUB-KEY-INTEL-SW-PRODUCTS.PUB \
+          | gpg --dearmor \
+          | sudo tee /usr/share/keyrings/oneapi-archive-keyring.gpg > /dev/null
+
+        echo "deb [signed-by=/usr/share/keyrings/oneapi-archive-keyring.gpg] https://apt.repos.intel.com/oneapi all main" \
+          | sudo tee /etc/apt/sources.list.d/oneAPI.list
+
+        sudo apt-get update
+        sudo apt-get install -y intel-oneapi-dpcpp-cpp-2024.0
+
+    - name: Setup ocloc
+      if: false
+      run: |
+        wget -qO - https://repositories.intel.com/gpu/intel-graphics.key \
+          | sudo gpg --dearmor --output /usr/share/keyrings/intel-graphics.gpg
+
+        echo "deb [arch=amd64 signed-by=/usr/share/keyrings/intel-graphics.gpg] https://repositories.intel.com/gpu/ubuntu jammy client" \
+          | sudo tee /etc/apt/sources.list.d/intel-gpu-jammy.list
+
+        sudo apt-get update
+        sudo apt-get install -y intel-ocloc
+        
+        ocloc compile --help
+
+    - name: Configure (SYCL)
+      if: false
+      run: |
+        source /opt/intel/oneapi/compiler/latest/env/vars.sh
+
+        cmake -S . -B build_sycl -G Ninja -LA \
+          -D CMAKE_BUILD_TYPE=Release \
+          -D ENABLE_CPU=OFF -D ENABLE_CUDA=OFF -D ENABLE_HIP=OFF -D ENABLE_SYCL=ON \
+          -D VAPOURSYNTH_INCLUDE_DIRECTORY="`pwd`/vapoursynth/include" \
+          -D CMAKE_CXX_COMPILER=icpx \
+          -D CMAKE_CXX_FLAGS="-Wall -ffast-math -march=x86-64-v3" \
+          -D CMAKE_SHARED_LINKER_FLAGS="-fsycl -fsycl-targets=spir64_gen -Xsycl-target-backend=spir64_gen \"-device gen12lp,xe-hpg,xe-lpg,xe-lpgplus\""
+
+    - name: Build (SYCL)
+      if: false
+      run: |
+        source /opt/intel/oneapi/compiler/latest/env/vars.sh
+
+        cmake --build build_sycl --verbose
+
+    - name: Install (SYCL)
+      if: false
+      run: cmake --install build_sycl --prefix artifact
+
+    - name: Upload
+      uses: actions/upload-artifact@v4
+      with:
+        name: VapourSynth-BM3DCUDA-Linux
+        path: artifact

vapoursynth_bm3dhip-2.16/VapourSynth-BM3DCUDA/.github/workflows/windows.yml

@@ -0,0 +1,234 @@
+name: Build (Windows)
+
+on:
+  push:
+    paths:
+      - 'CMakeLists.txt'
+      - 'source/*'
+      - 'cpu_source/*'
+      - 'rtc_source/*'
+      - 'hip_source/*'
+      - 'sycl_source/*'
+      - '.github/workflows/windows.yml'
+  workflow_dispatch:
+    inputs:
+      tag:
+        description: 'which tag to upload to'
+        default: ''
+
+jobs:
+  build-windows:
+    runs-on: windows-2022
+
+    defaults:
+      run:
+        shell: cmd
+
+    steps:
+    - name: Checkout repo
+      uses: actions/checkout@v4
+      with:
+        fetch-depth: 1
+
+    - name: Setup MSVC
+      uses: ilammy/msvc-dev-cmd@v1
+
+    - name: Setup Ninja
+      run: pip install ninja
+
+    - name: Download VapourSynth headers
+      run: |
+        curl -s -o vs.zip -L https://github.com/vapoursynth/vapoursynth/archive/refs/tags/R57.zip
+        unzip -q vs.zip
+        mv vapoursynth-*/ vapoursynth/
+
+    - name: Cache CUDA
+      id: cache-cuda
+      uses: actions/cache@v4
+      with:
+        path: C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA
+        key: ${{ runner.os }}-cuda-13.0.1
+
+    - name: Setup CUDA
+      if: steps.cache-cuda.outputs.cache-hit != 'true'
+      run: |
+        curl -s -o cuda_installer.exe -L https://developer.download.nvidia.com/compute/cuda/13.0.1/network_installers/cuda_13.0.1_windows_network.exe
+        cuda_installer.exe -s nvcc_13.0 cudart_13.0 nvrtc_dev_13.0 crt_13.0 nvptxcompiler_13.0 nvvm_13.0
+
+    - name: Configure (CUDA)
+      shell: bash
+      run: cmake -S . -B build -G Ninja -LA
+        -D CMAKE_BUILD_TYPE=Release
+        -D ENABLE_CUDA=ON
+        -D ENABLE_CPU=OFF
+        -D USE_NVRTC_STATIC=ON
+        -D VAPOURSYNTH_INCLUDE_DIRECTORY="$(pwd)\vapoursynth\include"
+        -D CMAKE_CXX_FLAGS="/fp:fast /arch:AVX /EHsc"
+        -D CMAKE_CUDA_FLAGS="--threads 0 --use_fast_math --resource-usage -Wno-deprecated-gpu-targets"
+        -D CMAKE_CUDA_ARCHITECTURES="75;86-real;89-real;120-real"
+        -D CMAKE_MSVC_RUNTIME_LIBRARY=MultiThreaded
+      env:
+        CUDA_PATH: C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v13.0
+
+    - name: Build (CUDA)
+      run: cmake --build build --verbose
+
+    - name: Install (CUDA)
+      run: cmake --install build --prefix install
+
+    - name: Configure, build, install (CPU)
+      shell: bash
+      run: for arch in haswell skylake icelake-client alderlake znver1 znver2 znver3 znver4 znver5; do
+          cmake -S . -B build_cpu -G Ninja -LA
+            -D CMAKE_BUILD_TYPE=Release
+            -D ENABLE_CUDA=OFF
+            -D ENABLE_CPU=ON
+            -D VAPOURSYNTH_INCLUDE_DIRECTORY="$(pwd)\vapoursynth\include"
+            -D CMAKE_CXX_COMPILER="clang++.exe"
+            -D CMAKE_CXX_FLAGS="-mtune=${arch} -ffast-math"
+            -D CMAKE_MSVC_RUNTIME_LIBRARY=MultiThreaded
+          && cmake --build build_cpu --verbose
+          && cmake --install build_cpu --prefix temp
+          && mv temp/bin/bm3dcpu.dll install/bin/bm3dcpu-${arch}.dll -v;
+        done
+
+    - name: Cache HIP
+      id: cache-hip
+      uses: actions/cache@v4
+      with:
+        path: C:\Program Files\AMD\ROCM
+        key: ${{ runner.os }}-rocm-6.4.2
+
+    - name: Setup HIP
+      if: steps.cache-hip.outputs.cache-hit != 'true'
+      shell: pwsh
+      run: |
+        curl -s -o hip_installer.exe -L https://download.amd.com/developer/eula/rocm-hub/AMD-Software-PRO-Edition-25.Q3-WinSvr2022-For-HIP.exe
+        Start-Process hip_installer.exe -ArgumentList '-install' -NoNewWindow -Wait
+
+    - name: Configure (HIP)
+      shell: bash
+      run: cmake -S . -B build_hip -G Ninja -LA
+        -D CMAKE_BUILD_TYPE=Release
+        -D ENABLE_CUDA=OFF
+        -D ENABLE_CPU=OFF
+        -D ENABLE_HIP=ON
+        -D VAPOURSYNTH_INCLUDE_DIRECTORY="$(pwd)\vapoursynth\include"
+        -D CMAKE_CXX_COMPILER="C:/Program Files/AMD/ROCm/6.4/bin/clang++.exe"
+        -D CMAKE_CXX_FLAGS="-ffast-math -munsafe-fp-atomics -Rpass-analysis=kernel-resource-usage"
+        -D HIP_DIR="C:/Program Files/AMD/ROCm/6.4/lib/cmake/hip"
+        -D HIP_PLATFORM=amd
+        -D GPU_TARGETS="gfx1030;gfx1031;gfx1032;gfx1100;gfx1101;gfx1102;gfx1151;gfx1200;gfx1201"
+        -D CMAKE_MSVC_RUNTIME_LIBRARY=MultiThreaded
+
+    - name: Build (HIP)
+      run: cmake --build build_hip --verbose
+
+    - name: Install (HIP)
+      run: cmake --install build_hip --prefix install
+
+    - name: Cache SYCL
+      if: false
+      id: cache-sycl
+      uses: actions/cache@v4
+      with:
+        path: C:\Program Files (x86)\Intel\oneAPI
+        key: ${{ runner.os }}-dpcpp-2024.0
+
+    - name: Setup SYCL
+      if: false && steps.cache-sycl.outputs.cache-hit != 'true'
+      run: |
+        curl -J -o dpcpp_installer.exe https://registrationcenter-download.intel.com/akdlm/IRC_NAS/94e15cb5-4bcc-4fdd-91cf-0f819a54e42e/w_dpcpp-cpp-compiler_p_2024.0.2.28_offline.exe
+        dpcpp_installer -s -a -s --eula accept
+
+    - name: Setup ocloc (Gen12+)
+      if: false
+      shell: pwsh
+      run: |
+        $env:PATH = "C:\Program Files (x86)\Intel\oneAPI\compiler\latest\lib\ocloc;$env:PATH"
+        "PATH=$env:PATH" | Out-File -Append -FilePath $env:GITHUB_ENV
+
+    - name: Configure (SYCL, Gen12+)
+      if: false
+      run: |
+        ocloc compile --help
+
+        call "C:\Program Files (x86)\Intel\oneAPI\compiler\latest\env\vars.bat"
+      
+        cmake -S . -B build_sycl_gen12+ -G Ninja -LA ^
+          -D CMAKE_BUILD_TYPE=Release ^
+          -D ENABLE_CUDA=OFF ^
+          -D ENABLE_CPU=OFF ^
+          -D ENABLE_HIP=OFF ^
+          -D ENABLE_SYCL=ON ^
+          -D VAPOURSYNTH_INCLUDE_DIRECTORY="%cd%\vapoursynth\include" ^
+          -D CMAKE_CXX_COMPILER=icx ^
+          -D CMAKE_CXX_FLAGS="-ffast-math -mavx" ^
+          -D CMAKE_SHARED_LINKER_FLAGS="-fsycl -fsycl-targets=spir64_gen -Xsycl-target-backend=spir64_gen \"-device gen12lp,xe-hpg\""
+
+    - name: Build (SYCL, Gen12+)
+      if: false
+      run: |
+        call "C:\Program Files (x86)\Intel\oneAPI\compiler\latest\env\vars.bat"
+
+        cmake --build build_sycl_gen12+ --verbose
+
+    - name: Install (SYCL, Gen12+)
+      if: false
+      run: echo f | xcopy build_sycl_gen12+\sycl_source\bm3dsycl.dll install\bin\bm3dsycl_gen12+.dll /f
+
+    - name: Prepare for upload
+      run: |
+        mkdir artifact
+        copy install\bin\*.dll artifact
+
+    - name: Upload
+      uses: actions/upload-artifact@v4
+      with:
+        name: VapourSynth-BM3DCUDA-Windows
+        path: artifact
+
+    - name: Compress artifact for release
+      if: github.event_name == 'workflow_dispatch' && github.event.inputs.tag != ''
+      run: |
+        cd artifact
+
+        mkdir VapourSynth-BM3DCUDA-${{ github.event.inputs.tag }}
+        xcopy bm3dcuda.dll VapourSynth-BM3DCUDA-${{ github.event.inputs.tag }} /f
+        7z a -t7z -mx=9 ../VapourSynth-BM3DCUDA-${{ github.event.inputs.tag }}.7z VapourSynth-BM3DCUDA-${{ github.event.inputs.tag }}
+
+        mkdir VapourSynth-BM3DCUDA_RTC-${{ github.event.inputs.tag }}
+        xcopy bm3dcuda_rtc.dll VapourSynth-BM3DCUDA_RTC-${{ github.event.inputs.tag }} /f
+        7z a -t7z -mx=9 ../VapourSynth-BM3DCUDA_RTC-${{ github.event.inputs.tag }}.7z VapourSynth-BM3DCUDA_RTC-${{ github.event.inputs.tag }}
+
+        mkdir VapourSynth-BM3DCPU-${{ github.event.inputs.tag }}
+        xcopy bm3dcpu-*.dll VapourSynth-BM3DCPU-${{ github.event.inputs.tag }} /f
+        7z a -t7z -mx=9 ../VapourSynth-BM3DCPU-${{ github.event.inputs.tag }}.7z VapourSynth-BM3DCPU-${{ github.event.inputs.tag }}
+
+        mkdir VapourSynth-BM3DHIP-${{ github.event.inputs.tag }}
+        xcopy bm3dhip.dll VapourSynth-BM3DHIP-${{ github.event.inputs.tag }} /f
+        7z a -t7z -mx=9 ../VapourSynth-BM3DHIP-${{ github.event.inputs.tag }}.7z VapourSynth-BM3DHIP-${{ github.event.inputs.tag }}
+        
+        :: mkdir VapourSynth-BM3DSYCL-${{ github.event.inputs.tag }}
+        :: xcopy bm3dsycl_*.dll VapourSynth-BM3DSYCL-${{ github.event.inputs.tag }} /f
+        :: 7z a -t7z -mx=9 ../VapourSynth-BM3DSYCL-${{ github.event.inputs.tag }}.7z VapourSynth-BM3DSYCL-${{ github.event.inputs.tag }}
+        
+        :: mkdir SYCL-Runtime-${{ github.event.inputs.tag }}
+        :: echo f | xcopy "C:\Program Files (x86)\Intel\oneAPI\compiler\latest\windows\bin\sycl6.dll" SYCL-Runtime-${{ github.event.inputs.tag }} /f
+        :: echo f | xcopy "C:\Program Files (x86)\Intel\oneAPI\compiler\latest\windows\bin\pi_*.dll" SYCL-Runtime-${{ github.event.inputs.tag }} /f /i
+        :: echo f | xcopy "C:\Program Files (x86)\Intel\oneAPI\compiler\latest\windows\redist\intel64_win\compiler\libmmd.dll" SYCL-Runtime-${{ github.event.inputs.tag }} /f
+        :: 7z a -t7z -mx=9 ../SYCL-Runtime-${{ github.event.inputs.tag }}.7z SYCL-Runtime-${{ github.event.inputs.tag }}
+
+    - name: Release
+      uses: softprops/action-gh-release@v1
+      if: github.event_name == 'workflow_dispatch' && github.event.inputs.tag != ''
+      with:
+        tag_name: ${{ github.event.inputs.tag }}
+        files: |
+          VapourSynth-BM3DCUDA-${{ github.event.inputs.tag }}.7z
+          VapourSynth-BM3DCUDA_RTC-${{ github.event.inputs.tag }}.7z
+          VapourSynth-BM3DCPU-${{ github.event.inputs.tag }}.7z
+          VapourSynth-BM3DHIP-${{ github.event.inputs.tag }}.7z
+        fail_on_unmatched_files: true
+        generate_release_notes: false
+        prerelease: true