#pragma clang diagnostic ignored "-Wmissing-prototypes"

#include <metal_stdlib>
#include <simd/simd.h>

using namespace metal;

struct DrawMonoid
{
    uint path_ix;
    uint clip_ix;
    uint scene_offset;
    uint info_offset;
};

struct Alloc
{
    uint offset;
};

struct Config
{
    uint n_elements;
    uint n_pathseg;
    uint width_in_tiles;
    uint height_in_tiles;
    Alloc tile_alloc;
    Alloc bin_alloc;
    Alloc ptcl_alloc;
    Alloc pathseg_alloc;
    Alloc anno_alloc;
    Alloc trans_alloc;
    Alloc path_bbox_alloc;
    Alloc drawmonoid_alloc;
    Alloc clip_alloc;
    Alloc clip_bic_alloc;
    Alloc clip_stack_alloc;
    Alloc clip_bbox_alloc;
    Alloc draw_bbox_alloc;
    Alloc drawinfo_alloc;
    uint n_trans;
    uint n_path;
    uint n_clip;
    uint trans_offset;
    uint linewidth_offset;
    uint pathtag_offset;
    uint pathseg_offset;
    uint drawtag_offset;
    uint drawdata_offset;
};

struct ConfigBuf
{
    Config conf;
};

struct SceneBuf
{
    uint scene[1];
};

struct DrawMonoid_1
{
    uint path_ix;
    uint clip_ix;
    uint scene_offset;
    uint info_offset;
};

struct OutBuf
{
    DrawMonoid_1 outbuf[1];
};

struct Memory
{
    uint mem_offset;
    uint mem_error;
    uint memory[1];
};

constant uint3 gl_WorkGroupSize [[maybe_unused]] = uint3(256u, 1u, 1u);

static inline __attribute__((always_inline))
DrawMonoid map_tag(thread const uint& tag_word)
{
    uint has_path = uint(tag_word != 0u);
    return DrawMonoid{ has_path, tag_word & 1u, tag_word & 28u, (tag_word >> uint(4)) & 28u };
}

static inline __attribute__((always_inline))
DrawMonoid combine_draw_monoid(thread const DrawMonoid& a, thread const DrawMonoid& b)
{
    DrawMonoid c;
    c.path_ix = a.path_ix + b.path_ix;
    c.clip_ix = a.clip_ix + b.clip_ix;
    c.scene_offset = a.scene_offset + b.scene_offset;
    c.info_offset = a.info_offset + b.info_offset;
    return c;
}

kernel void main0(const device ConfigBuf& _86 [[buffer(1)]], const device SceneBuf& _96 [[buffer(2)]], device OutBuf& _187 [[buffer(3)]], uint3 gl_GlobalInvocationID [[thread_position_in_grid]], uint3 gl_LocalInvocationID [[thread_position_in_threadgroup]], uint3 gl_WorkGroupID [[threadgroup_position_in_grid]])
{
    threadgroup DrawMonoid sh_scratch[256];
    uint ix = gl_GlobalInvocationID.x * 8u;
    uint drawtag_base = _86.conf.drawtag_offset >> uint(2);
    uint tag_word = _96.scene[drawtag_base + ix];
    uint param = tag_word;
    DrawMonoid agg = map_tag(param);
    for (uint i = 1u; i < 8u; i++)
    {
        uint tag_word_1 = _96.scene[(drawtag_base + ix) + i];
        uint param_1 = tag_word_1;
        DrawMonoid param_2 = agg;
        DrawMonoid param_3 = map_tag(param_1);
        agg = combine_draw_monoid(param_2, param_3);
    }
    sh_scratch[gl_LocalInvocationID.x] = agg;
    for (uint i_1 = 0u; i_1 < 8u; i_1++)
    {
        threadgroup_barrier(mem_flags::mem_threadgroup);
        if ((gl_LocalInvocationID.x + (1u << i_1)) < 256u)
        {
            DrawMonoid other = sh_scratch[gl_LocalInvocationID.x + (1u << i_1)];
            DrawMonoid param_4 = agg;
            DrawMonoid param_5 = other;
            agg = combine_draw_monoid(param_4, param_5);
        }
        threadgroup_barrier(mem_flags::mem_threadgroup);
        sh_scratch[gl_LocalInvocationID.x] = agg;
    }
    if (gl_LocalInvocationID.x == 0u)
    {
        _187.outbuf[gl_WorkGroupID.x].path_ix = agg.path_ix;
        _187.outbuf[gl_WorkGroupID.x].clip_ix = agg.clip_ix;
        _187.outbuf[gl_WorkGroupID.x].scene_offset = agg.scene_offset;
        _187.outbuf[gl_WorkGroupID.x].info_offset = agg.info_offset;
    }
}