#ifndef GPU_OFFSET_H
#define GPU_OFFSET_H

#ifdef _WIN32
  #ifdef GPU_OFFSET_EXPORTS
    #define GPU_OFFSET_API __declspec(dllexport)
  #else
    #define GPU_OFFSET_API __declspec(dllimport)
  #endif
#else
  #define GPU_OFFSET_API __attribute__((visibility("default")))
#endif

#ifdef __cplusplus
extern "C" {
#endif

#include <stddef.h>
#include <stdint.h>

/**
 * @brief Status and Error Codes
 */
#define GPU_OFFSET_SUCCESS           0
#define GPU_OFFSET_ERR_GENERIC      -1
#define GPU_OFFSET_ERR_NO_CUDA      -2
#define GPU_OFFSET_ERR_INVALID_ARG  -3

/**
 * @brief Compute Backend Modes
 */
typedef enum {
    GPU_OFFSET_BACKEND_AUTO = 0, // Auto: use CUDA if available, fallback to CPU
    GPU_OFFSET_BACKEND_CUDA = 1, // Require CUDA: returns GPU_OFFSET_ERR_NO_CUDA if unavailable
    GPU_OFFSET_BACKEND_CPU  = 2  // Compatibility mode: force multi-threaded CPU execution
} GpuOffsetBackend;

/**
 * @brief 2D Point structure (floating-point millimeters)
 */
typedef struct {
    float x;
    float y;
} GpuPoint2D;

/**
 * @brief 2D Polygon Contour Loop
 */
typedef struct {
    GpuPoint2D* points;
    size_t num_points;
    int is_hole; // 0 = exterior boundary, 1 = interior hole
} GpuContour;

/**
 * @brief Sliced Layer containing one or more planar polygon loops
 */
typedef struct {
    GpuContour* contours;
    size_t num_contours;
    float z_height;
} GpuSliceLayer;

/**
 * @brief Resulting Offset Contours for a single offset perimeter
 */
typedef struct {
    GpuContour* contours;
    size_t num_contours;
    float offset_distance_mm;
} GpuOffsetPerimeter;

/**
 * @brief Infill region mask and boundary contours for a single sliced layer
 */
typedef struct {
    GpuContour* contours;      // Infill boundary contours (at -perimeters_count * layer_width)
    size_t num_contours;
    uint8_t* mask;             // 2D binary raster mask (width * height uint8), or NULL if not exported
    int width;
    int height;
    float offset_distance_mm;  // Offset distance for infill (perimeters_count * layer_width)
} GpuInfillRegion;

/**
 * @brief All offset perimeters and infill mask for a single sliced layer
 */
typedef struct {
    GpuOffsetPerimeter* perimeters; // Array of perimeters (length = num_perimeters)
    size_t num_perimeters;         // perimeters_count
    GpuInfillRegion infill;        // Infill boundary contours & 2D mask
    float z_height;
} GpuLayerResult;

/**
 * @brief Configuration parameters for Offsetting & Slicing Engine
 */
typedef struct {
    float pitch_mm;            // Pixel raster pitch (default: 0.10 mm)
    float pad_mm;              // Bounding box padding margin in mm (default: 1.5 mm)
    int perimeters_count;      // Number of perimeter passes (e.g. 2, default: 2)
    float layer_width;         // Line/extrusion width in mm (e.g. 0.45 mm, default: 0.45 mm)
    const float* offset_distances_mm; // Optional custom offset distances (if NULL, auto-generated from perimeters_count & layer_width)
    size_t num_offsets;        // Total offsets (auto-set if offset_distances_mm is NULL)
    int use_gpu_rasterizer;    // 1 = GPU-accelerated rasterization (when CUDA enabled), 0 = CPU multi-threaded
    float first_layer_height;  // First layer height in mm (default: 0.0f -> uses layer_height)
    int export_infill_mask;    // 1 = export binary 2D infill mask into GpuLayerResult (default: 1)
    float simplify_tolerance_mm; // RDP simplification tolerance in mm (default: 0.02 mm, 0 to disable)
    int subpixel_lerp;         // 1 = Continuous SDF sub-pixel LERP vectorization (default: 1), 0 = discrete bitpack
    int backend;               // GpuOffsetBackend: 0 = AUTO, 1 = CUDA, 2 = CPU (default: 0)
} GpuOffsetConfig;

/**
 * @brief Microsecond-accurate timing breakdown
 */
typedef struct {
    double t_mesh_slice_ms;
    double t_raster_ms;
    double t_h2d_ms;
    double t_gpu_jfa_ms;
    double t_gpu_thresh_ms;
    double t_d2h_ms;
    double t_vectorize_ms;
    double t_total_pipeline_ms;
} GpuOffsetStats;

/**
 * @brief Returns 1 if a functional CUDA-capable device is detected, 0 otherwise.
 */
GPU_OFFSET_API int gpu_offset_has_cuda(void);

/**
 * @brief Set global default backend (0 = AUTO, 1 = CUDA, 2 = CPU).
 */
GPU_OFFSET_API int gpu_offset_set_backend(int backend);

/**
 * @brief Get global default backend (0 = AUTO, 1 = CUDA, 2 = CPU).
 */
GPU_OFFSET_API int gpu_offset_get_backend(void);

/**
 * @brief Initialize GPU device and allocate CUDA stream / warm up kernels.
 */
GPU_OFFSET_API int gpu_offset_init(int device_id);

/**
 * @brief Core Distance Field & Offsetting on raw 2D/3D binary masks (CUDA or CPU fallback).
 */
GPU_OFFSET_API int gpu_offset_compute_masks(
    const uint8_t* h_masks,
    int width,
    int height,
    int num_layers,
    float pitch_mm,
    const float* offset_distances,
    int num_offsets,
    uint8_t* h_out_masks,
    GpuOffsetStats* stats
);

/**
 * @brief 3D Triangle structure for direct mesh slicing (vertices in millimeters)
 */
typedef struct {
    float v0[3];
    float v1[3];
    float v2[3];
} GpuTriangle3D;

/**
 * @brief End-to-end polygon offsetting: Polygons -> Raster -> Distance Field -> Vector Contours & Infill
 */
GPU_OFFSET_API int gpu_offset_polygon_layers(
    const GpuSliceLayer* in_layers,
    size_t num_layers,
    float min_x, float min_y, float max_x, float max_y,
    const GpuOffsetConfig* config,
    GpuLayerResult** out_results,
    GpuOffsetStats* stats
);

/**
 * @brief End-to-End Mesh Slicing & Perimeter Offsetting:
 *        3D Triangles -> Mesh Slicer -> Distance Field -> Vector Contours & Infill.
 */
GPU_OFFSET_API int gpu_offset_slice_mesh_and_offset(
    const GpuTriangle3D* triangles,
    size_t num_triangles,
    float min_x, float min_y, float min_z,
    float max_x, float max_y, float max_z,
    float layer_height,
    float first_layer_height,
    const GpuOffsetConfig* config,
    GpuLayerResult** out_results,
    size_t* out_num_layers,
    GpuOffsetStats* stats
);

/**
 * @brief Free memory allocated for GpuLayerResult
 */
GPU_OFFSET_API void gpu_offset_free_results(GpuLayerResult* results, size_t num_layers);

/**
 * @brief Release GPU context and cached buffers
 */
GPU_OFFSET_API void gpu_offset_cleanup();

#ifdef __cplusplus
}
#endif

#endif // GPU_OFFSET_H