#ifdef COMPILING_HALIDE_RUNTIME
#include "HalideRuntime.h"
#define HALIDE_BUFFER_HELPER_ATTRS WEAK_INLINE
#else
#define HALIDE_BUFFER_HELPER_ATTRS inline
#endif

// Structs are annoying to deal with from within Halide Stmts. These
// utility functions are for dealing with halide_buffer_t in that
// context. They are not intended for use outside of Halide code, and
// not exposed in HalideRuntime.h. The symbols are private to the
// module and should be inlined and then stripped. This blob of code
// also gets copy-pasted into C outputs.

extern "C" {

HALIDE_BUFFER_HELPER_ATTRS
int _halide_buffer_get_dimensions(const halide_buffer_t *buf) {
    return buf->dimensions;
}

HALIDE_BUFFER_HELPER_ATTRS
uint8_t *_halide_buffer_get_host(const halide_buffer_t *buf) {
    return buf->host;
}

HALIDE_BUFFER_HELPER_ATTRS
uint64_t _halide_buffer_get_device(const halide_buffer_t *buf) {
    return buf->device;
}

HALIDE_BUFFER_HELPER_ATTRS
const struct halide_device_interface_t *_halide_buffer_get_device_interface(const halide_buffer_t *buf) {
    return buf->device_interface;
}

HALIDE_BUFFER_HELPER_ATTRS
int _halide_buffer_get_min(const halide_buffer_t *buf, int d) {
    return buf->dim[d].min;
}

HALIDE_BUFFER_HELPER_ATTRS
int _halide_buffer_get_max(const halide_buffer_t *buf, int d) {
    return buf->dim[d].min + buf->dim[d].extent - 1;
}

HALIDE_BUFFER_HELPER_ATTRS
int _halide_buffer_get_extent(const halide_buffer_t *buf, int d) {
    return buf->dim[d].extent;
}

HALIDE_BUFFER_HELPER_ATTRS
int _halide_buffer_get_stride(const halide_buffer_t *buf, int d) {
    return buf->dim[d].stride;
}

HALIDE_BUFFER_HELPER_ATTRS
int _halide_buffer_set_host_dirty(halide_buffer_t *buf, bool val) {
    buf->set_host_dirty(val);
    return 0;
}

HALIDE_BUFFER_HELPER_ATTRS
int _halide_buffer_set_device_dirty(halide_buffer_t *buf, bool val) {
    buf->set_device_dirty(val);
    return 0;
}

HALIDE_BUFFER_HELPER_ATTRS
bool _halide_buffer_get_host_dirty(const halide_buffer_t *buf) {
    return buf->host_dirty();
}

HALIDE_BUFFER_HELPER_ATTRS
bool _halide_buffer_get_device_dirty(const halide_buffer_t *buf) {
    return buf->device_dirty();
}

HALIDE_BUFFER_HELPER_ATTRS
halide_dimension_t *_halide_buffer_get_shape(halide_buffer_t *buf) {
    return buf->dim;
}

HALIDE_BUFFER_HELPER_ATTRS
bool _halide_buffer_is_bounds_query(const halide_buffer_t *buf) {
    return buf->host == nullptr && buf->device == 0;
}

HALIDE_BUFFER_HELPER_ATTRS
uint32_t _halide_buffer_get_type(const halide_buffer_t *buf) {
    return buf->type.as_u32();
}

HALIDE_BUFFER_HELPER_ATTRS
halide_buffer_t *_halide_buffer_init(halide_buffer_t *dst,
                                     halide_dimension_t *dst_shape,
                                     void *host,
                                     uint64_t device,
                                     const halide_device_interface_t *device_interface,
                                     int type_code, int type_bits,
                                     int dimensions,
                                     halide_dimension_t *shape,
                                     uint64_t flags) {
    dst->host = (uint8_t *)host;
    dst->device = device;
    dst->device_interface = device_interface;
    dst->type.code = (halide_type_code_t)type_code;
    dst->type.bits = (uint8_t)type_bits;
    dst->type.lanes = 1;
    dst->dimensions = dimensions;
    dst->dim = dst_shape;
    if (shape != dst->dim) {
        for (int i = 0; i < dimensions; i++) {
            dst->dim[i] = shape[i];
        }
    }
    dst->flags = flags;
    return dst;
}

HALIDE_BUFFER_HELPER_ATTRS
halide_buffer_t *_halide_buffer_init_from_buffer(halide_buffer_t *dst,
                                                 halide_dimension_t *dst_shape,
                                                 const halide_buffer_t *src) {
    dst->host = src->host;
    dst->device = src->device;
    dst->device_interface = src->device_interface;
    dst->type = src->type;
    dst->dimensions = src->dimensions;
    dst->dim = dst_shape;
    dst->flags = src->flags;
    for (int i = 0; i < dst->dimensions; i++) {
        dst->dim[i] = src->dim[i];
    }
    return dst;
}

HALIDE_BUFFER_HELPER_ATTRS
halide_buffer_t *_halide_buffer_crop(void *user_context,
                                     halide_buffer_t *dst,
                                     halide_dimension_t *dst_shape,
                                     const halide_buffer_t *src,
                                     const int *min, const int *extent) {
    *dst = *src;
    dst->dim = dst_shape;
    int64_t offset = 0;
    for (int i = 0; i < dst->dimensions; i++) {
        dst->dim[i] = src->dim[i];
        dst->dim[i].min = min[i];
        dst->dim[i].extent = extent[i];
        offset += (min[i] - src->dim[i].min) * (int64_t)src->dim[i].stride;
    }
    if (dst->host) {
        dst->host += offset * src->type.bytes();
    }
    dst->device_interface = nullptr;
    dst->device = 0;
    if (src->device_interface) {
        if (src->device_interface->device_crop(user_context, src, dst) != 0) {
            // This is uncommon: either a runtime error, or a backend that
            // doesn't replace the default definition of device_crop. But it
            // does happen, so let's return a nullptr here, and require the caller
            // to check the result.
            return nullptr;
        }
    }
    return dst;
}

// Called on return from an extern stage where the output buffer was a
// crop of some other larger buffer. This happens for extern stages
// with distinct store_at/compute_at levels. Each call to the stage
// only fills in part of the buffer.
HALIDE_BUFFER_HELPER_ATTRS
int _halide_buffer_retire_crop_after_extern_stage(void *user_context,
                                                  void *obj) {
    int result;
    halide_buffer_t **buffers = (halide_buffer_t **)obj;
    halide_buffer_t *crop = buffers[0];
    halide_buffer_t *parent = buffers[1];

    if (crop->device) {
        if (!parent->device) {
            // We have been given a device allocation by the extern
            // stage. It only represents the cropped region, so we
            // can't just give it to the parent.
            if (crop->device_dirty()) {
                result = crop->device_interface->copy_to_host(user_context, crop);
                if (result != 0) {
                    return result;
                }
            }
            result = crop->device_interface->device_free(user_context, crop);
            if (result != 0) {
                return result;
            }
        } else {
            // We are a crop of an existing device allocation.
            if (crop->device_dirty()) {
                parent->set_device_dirty();
            }
            result = crop->device_interface->device_release_crop(user_context, crop);
            if (result != 0) {
                return result;
            }
        }
    }
    if (crop->host_dirty()) {
        parent->set_host_dirty();
    }
    return 0;
}

HALIDE_BUFFER_HELPER_ATTRS
int _halide_buffer_retire_crops_after_extern_stage(void *user_context,
                                                   void *obj) {
    halide_buffer_t **buffers = (halide_buffer_t **)obj;
    while (*buffers) {
        _halide_buffer_retire_crop_after_extern_stage(user_context, buffers);
        buffers += 2;
    }
    return 0;
}

HALIDE_BUFFER_HELPER_ATTRS
halide_buffer_t *_halide_buffer_set_bounds(halide_buffer_t *buf,
                                           int dim, int min, int extent) {
    // This can be called with the result of _halide_buffer_crop(), which
    // can return nullptr if an error occurs -- so don't crash, just propagate
    // the nullptr result to our caller.
    if (buf != nullptr) {
        buf->dim[dim].min = min;
        buf->dim[dim].extent = extent;
    }
    return buf;
}
}

#undef HALIDE_BUFFER_HELPER_ATTRS