HHostingByteSoftware
Ctrl K
ComunidadGuías…

docs/BUILDING.md

Rama main · réplica de lectura del repositorio original.

Archivo original ↗Preguntar a KiBot
# Building Meetily from Source

This guide explains source builds on each supported platform. Start with the build notes below, then use the platform instructions that match your machine.

## Build Notes

Install the committed frontend dependencies before building:

```bash
npm install -g pnpm@9.15.9
cd frontend
pnpm install --frozen-lockfile
```

Frozen installation keeps the lockfile and installed dependency set aligned. When intentionally changing dependencies, update and commit `pnpm-lock.yaml`.

- **Linux:** Meetily is built from source; choose acceleration for the environment where you build.
- **Windows packages:** Distribution builds use Vulkan-enabled Whisper and require an AVX2-capable x64 CPU. AVX-512 is not required.
- **CUDA:** NVIDIA CUDA support requires a compatible source build and CUDA toolchain; the standard Windows installer does not select it automatically.

<details>
<summary>Linux</summary>

## 🐧 Building on Linux

This guide helps you build Meetily on Linux with **automatic GPU acceleration**. The build system detects your hardware and configures the best performance automatically.

---

### 🚀 Quick Start (Recommended for Beginners)

If you're new to building on Linux, start here. These simple commands work for most users:

#### 1. Install Basic Dependencies

```bash
# Ubuntu/Debian
sudo apt update
sudo apt install build-essential cmake git

# Fedora/RHEL
sudo dnf install gcc-c++ cmake git

# Arch Linux
sudo pacman -S base-devel cmake git
```

#### 2. Build and Run

```bash
# Development mode (with hot reload)
./dev-gpu.sh

# Production build
./build-gpu.sh
```

**That's it!** The scripts automatically detect your GPU and configure acceleration.

### What Happens Automatically?

- ✅ **NVIDIA GPU** → CUDA acceleration (if toolkit installed)
- ✅ **AMD GPU** → ROCm acceleration (if ROCm installed)
- ✅ **No GPU** → Optimized CPU mode (still works great!)

> 💡 **Tip:** If you have an NVIDIA or AMD GPU but want better performance, jump to the [GPU Setup](#-gpu-setup-guides-intermediate) section below.

---

### 🧠 Understanding Auto-Detection

The build scripts (`dev-gpu.sh` and `build-gpu.sh`) orchestrate the entire build process. Here's how they work:

1.  **Detect location:** Find `package.json` (works from project root or `frontend/`)
2.  **Auto-detect GPU:** Run `scripts/auto-detect-gpu.js` (or use `TAURI_GPU_FEATURE` if set)
3.  **Build Sidecar:** Build `llama-helper` with the detected feature (debug or release)
4.  **Copy Binary:** Copy the built sidecar to `src-tauri/binaries` with the target triple
5.  **Run Tauri:** Call `npm run tauri:dev` or `tauri:build` with the feature flag passed via env var

#### Detection Priority

| Priority | Hardware        | What It Checks                                               | Result                  |
| -------- | --------------- | ------------------------------------------------------------ | ----------------------- |
| 1️⃣       | **NVIDIA CUDA** | `nvidia-smi` exists + (`CUDA_PATH` or `nvcc` found)          | `--features cuda`       |
| 2️⃣       | **AMD ROCm**    | `rocm-smi` exists + (`ROCM_PATH` or `hipcc` found)           | `--features hipblas`    |
| 3️⃣       | **Vulkan**      | `vulkaninfo` exists + `VULKAN_SDK` + `BLAS_INCLUDE_DIRS` set | `--features vulkan`     |
| 4️⃣       | **OpenBLAS**    | `BLAS_INCLUDE_DIRS` set                                      | `--features openblas`   |
| 5️⃣       | **CPU-only**    | None of the above                                            | (no features, pure CPU) |

#### Common Scenarios

| Your System               | Auto-Detection Result       | Why                          |
| ------------------------- | --------------------------- | ---------------------------- |
| Clean Linux install       | CPU-only                    | No GPU SDK detected          |
| NVIDIA GPU + drivers only | CPU-only                    | CUDA toolkit not installed   |
| NVIDIA GPU + CUDA toolkit | **CUDA acceleration** ✅    | Full detection successful    |
| AMD GPU + ROCm            | **HIPBlas acceleration** ✅ | Full detection successful    |
| Vulkan drivers only       | CPU-only                    | Vulkan SDK + env vars needed |
| Vulkan SDK configured     | **Vulkan acceleration** ✅  | All requirements met         |

> 💡 **Key Insight:** Having GPU drivers alone isn't enough. You need the **development SDK** (CUDA toolkit, ROCm, or Vulkan SDK) for acceleration.

---

### 🔧 GPU Setup Guides (Intermediate)

Want better performance? Follow these guides to enable GPU acceleration.

#### 🟢 NVIDIA CUDA Setup

**Prerequisites:** NVIDIA GPU with compute capability 5.0+ (check: `nvidia-smi --query-gpu=compute_cap --format=csv`)

##### Step 1: Install CUDA Toolkit

```bash
# Ubuntu/Debian (CUDA 12.x)
sudo apt install nvidia-driver-550 nvidia-cuda-toolkit

# Verify installation
nvidia-smi          # Shows GPU info
nvcc --version      # Shows CUDA version
```

##### Step 2: Build with CUDA

```bash
# Set your GPU's compute capability
# Example: RTX 3080 = 8.6 → use "86"
# Example: GTX 1080 = 6.1 → use "61"

CMAKE_CUDA_ARCHITECTURES=75 \
CMAKE_CUDA_STANDARD=17 \
CMAKE_POSITION_INDEPENDENT_CODE=ON \
./build-gpu.sh
```

> 💡 **Finding Your Compute Capability:**
>
> ```bash
> nvidia-smi --query-gpu=compute_cap --format=csv
> ```
>
> Convert `7.5` → `75`, `8.6` → `86`, etc.

**Why these flags?**

- `CMAKE_CUDA_ARCHITECTURES`: Optimizes for your specific GPU
- `CMAKE_CUDA_STANDARD=17`: Ensures C++17 compatibility
- `CMAKE_POSITION_INDEPENDENT_CODE=ON`: Fixes linking issues on modern systems

---

#### 🔵 Vulkan Setup (Cross-Platform Fallback)

Vulkan works on NVIDIA, AMD, and Intel GPUs. Good choice if CUDA/ROCm don't work.

##### Step 1: Install Vulkan SDK and BLAS

```bash
# Ubuntu/Debian
sudo apt install vulkan-sdk libopenblas-dev

# Fedora
sudo dnf install vulkan-devel openblas-devel

# Arch Linux
sudo pacman -S vulkan-devel openblas
```

##### Step 2: Configure Environment

```bash
# Add to ~/.bashrc or ~/.zshrc
export VULKAN_SDK=/usr
export BLAS_INCLUDE_DIRS=/usr/include/x86_64-linux-gnu

# Apply changes
source ~/.bashrc
```

##### Step 3: Build

```bash
./build-gpu.sh
```

The script will automatically detect Vulkan and build with `--features vulkan`.

---

#### 🔴 AMD ROCm Setup (AMD GPUs Only)

**Prerequisites:** AMD GPU with ROCm support (RX 5000+, Radeon VII, etc.)

```bash
# Ubuntu/Debian
# Add ROCm repository (see https://rocm.docs.amd.com for latest)
sudo apt install rocm-smi hipcc

# Set environment
export ROCM_PATH=/opt/rocm

# Verify
rocm-smi            # Shows GPU info
hipcc --version     # Shows ROCm version

# Build
./build-gpu.sh
```

---

### 🎯 Advanced Usage

#### Manual Feature Override

Want to force a specific acceleration method? Use the `TAURI_GPU_FEATURE` environment variable with the shell scripts:

```bash
# Force CUDA (ignore auto-detection)
TAURI_GPU_FEATURE=cuda ./dev-gpu.sh
TAURI_GPU_FEATURE=cuda ./build-gpu.sh

# Force Vulkan
TAURI_GPU_FEATURE=vulkan ./dev-gpu.sh
TAURI_GPU_FEATURE=vulkan ./build-gpu.sh

# Force ROCm (HIPBlas)
TAURI_GPU_FEATURE=hipblas ./dev-gpu.sh
TAURI_GPU_FEATURE=hipblas ./build-gpu.sh

# Force CPU-only (for testing)
TAURI_GPU_FEATURE="" ./dev-gpu.sh
TAURI_GPU_FEATURE="" ./build-gpu.sh

# Force OpenBLAS (CPU-optimized)
TAURI_GPU_FEATURE=openblas ./dev-gpu.sh
TAURI_GPU_FEATURE=openblas ./build-gpu.sh
```

#### Build Output Location

After successful build:

```
src-tauri/target/release/bundle/appimage/Meetily_<version>_amd64.AppImage
```

---

### 🧭 Troubleshooting

#### "CUDA toolkit not found"

- **Fix:** Install `nvidia-cuda-toolkit` or set `CUDA_PATH` environment variable
- **Check:** `nvcc --version` should work

#### "Vulkan detected but missing dependencies"

- **Fix:** Set both `VULKAN_SDK` and `BLAS_INCLUDE_DIRS` environment variables
- **Example:**
  ```bash
  export VULKAN_SDK=/usr
  export BLAS_INCLUDE_DIRS=/usr/include/x86_64-linux-gnu
  ```

#### "AppImage build stripping symbols"

- **Fix:** Already handled! `build-gpu.sh` sets `NO_STRIP=true` automatically
- **Why:** Prevents runtime errors from missing symbols

#### Build works but no GPU acceleration

- **Check detection:** Look at the build output for GPU detection messages
- **Verify:** `nvidia-smi` (NVIDIA) or `rocm-smi` (AMD) should work
- **Missing SDK:** Install the development toolkit, not just drivers

</details>

<details>
<summary>macOS</summary>

## 🍎 Building on macOS

On macOS, the build process is simplified as GPU acceleration (Metal) is enabled by default.

### 1. Install Dependencies

```bash
# Install Homebrew (if not already installed)
/bin/bash -c "$(curl -fsSL https://raw.githubusercontent.com/Homebrew/install/HEAD/install.sh)"

# Install required tools
brew install cmake node pnpm
```

### 2. Build and Run

```bash
# Development mode (with hot reload)
pnpm tauri:dev

# Production build
pnpm tauri:build
```

The application will be built with Metal GPU acceleration automatically.

</details>

<details>
<summary>Windows</summary>

## 🪟 Building on Windows

### 1. Install Dependencies

- **Node.js:** Download and install from [nodejs.org](https://nodejs.org/).
- **Rust:** Install from [rust-lang.org](https://www.rust-lang.org/tools/install).
- **pnpm:** Install version 9.15.9 with `npm install -g pnpm@9.15.9`.
- **Visual Studio Build Tools:** Install the "Desktop development with C++" workload from the Visual Studio Installer.
- **CMake:** Download and install from [cmake.org](https://cmake.org/download/).

### 2. Build and Run

```powershell
pnpm install --frozen-lockfile
```

```powershell
# Development mode (with hot reload)
pnpm tauri:dev

# Production build
pnpm tauri:build
```

By default, the application will be built with CPU-only processing. To enable GPU acceleration, see the [GPU Acceleration Guide](GPU_ACCELERATION.md).

### Windows Distribution Builds

The commands above create a local source build. Use the production Windows build workflow for an installer intended for other computers: it enables Vulkan. Rust targets `x86-64-v2`; native Whisper retains AVX2 with host-native specialization and AVX-512 disabled.

The distribution workflows (`build.yml`, `build-windows.yml`, and `build-devtest.yml`) set `CMAKE_PROJECT_INCLUDE` to `.github/force-portable-ggml.cmake`, which forces `GGML_NATIVE=OFF`, and use `RUSTFLAGS=-C target-cpu=x86-64-v2`. The hook configures Whisper's native C/C++ build; Rust flags do not. `WHISPER_NATIVE=OFF` and plain `GGML_*` variables are not replacements.

</details>