# GPU Procedural Phase 4: Texture Composition

**Status:** ✅ Complete

## Implementation

Multi-input composite shaders with configurable sampler support.

### API

```cpp
enum class SamplerType {
  LinearClamp, LinearRepeat, NearestClamp, NearestRepeat
};

void create_gpu_composite_texture(
    const std::string& name,
    const std::string& shader_func,
    const char* shader_code,
    const void* uniform_data,
    size_t uniform_size,
    int width, int height,
    const std::vector<std::string>& input_names,
    SamplerType sampler = SamplerType::LinearClamp);
```

### Shaders

**gen_blend.wgsl** - Blend two textures with lerp factor:
- Bindings: output (0), uniform (1), input_a (2), input_b (3), sampler (4)
- Uniform: `{u32 width, height; f32 blend_factor, _pad0}`

**gen_mask.wgsl** - Multiply textures (masking):
- Bindings: output (0), uniform (1), input_a (2), input_b (3), sampler (4)
- Uniform: `{u32 width, height}`

### Usage

```cpp
extern const char* gen_blend_compute_wgsl;

struct { uint32_t width, height; float blend_factor, _pad0; } uni = {256, 256, 0.5f, 0.0f};

tex_mgr.create_gpu_composite_texture(
    "blended", "gen_blend", gen_blend_compute_wgsl,
    &uni, sizeof(uni), 256, 256,
    {"noise_a", "noise_b"},
    SamplerType::LinearClamp);
```

### Features

- **Dynamic bind groups:** N input textures + 1 sampler
- **Lazy sampler creation:** Map-based cache, 4 preset types
- **Multi-stage composition:** Composite of composites supported
- **Guarded with `#if !defined(STRIP_GPU_COMPOSITE)`**

### Size Impact

- Code: ~460 lines added
- Compressed: ~830 bytes (2 shaders + dispatch logic)

### Tests

`test_gpu_composite.cc`:
- Blend two noise textures
- Mask noise with grid
- Multi-stage composite (composite of composites)

All 35 tests passing.