# Device and platform utilities

Functions for detecting and configuring the compute platform and device.

## Device information

```python
import genesis as gs

gs.init(backend=gs.gpu)

# Get PyTorch device
device = gs.device
print(device)  # cuda:0, mps:0, cpu, etc.

# Get active backend
backend = gs.backend
print(backend)  # gs.cuda, gs.metal, gs.cpu, etc.
```

## Backend selection

### Automatic selection

```python
# GPU auto-selects based on platform
gs.init(backend=gs.gpu)
# Linux -> CUDA
# macOS -> Metal
```

### Manual selection

```python
# Force specific backend
gs.init(backend=gs.cuda)    # NVIDIA CUDA
gs.init(backend=gs.metal)   # Apple Metal
gs.init(backend=gs.cpu)     # CPU fallback
```

## Random seed

```python
import genesis as gs

# Set random seed for reproducibility
gs.init(seed=42)

# Or set later
gs.set_random_seed(42)
```

## Global variables

After `gs.init()`, these are available:

| Variable | Type | Description |
|----------|------|-------------|
| `gs.device` | torch.device | PyTorch device for tensors |
| `gs.backend` | gs.backend | Active compute backend |
| `gs.EPS` | float | Machine epsilon for the current precision |

## Type hints

```python
# Quadrants types
gs.qd_float  # Quadrants float type
gs.qd_int    # Quadrants int type
gs.qd_vec3   # Quadrants 3D vector
gs.qd_mat3   # Quadrants 3x3 matrix

# PyTorch types
gs.tc_float  # PyTorch float dtype
gs.tc_int    # PyTorch int dtype
```

## See also

- {doc}`constants`: Backend enums
- {doc}`tensor_utils`: Tensor operations