vello/piet-gpu/shader/gen/backdrop_lg.msl

#pragma clang diagnostic ignored "-Wmissing-prototypes"
#pragma clang diagnostic ignored "-Wunused-variable"

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

using namespace metal;

struct Alloc
{
    uint offset;
};

struct PathRef
{
    uint offset;
};

struct TileRef
{
    uint offset;
};

struct Path
{
    uint4 bbox;
    TileRef tiles;
};

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

struct Alloc_1
{
    uint offset;
};

struct Config
{
    uint mem_size;
    uint n_elements;
    uint n_pathseg;
    uint width_in_tiles;
    uint height_in_tiles;
    Alloc_1 tile_alloc;
    Alloc_1 bin_alloc;
    Alloc_1 ptcl_alloc;
    Alloc_1 pathseg_alloc;
    Alloc_1 anno_alloc;
    Alloc_1 trans_alloc;
    Alloc_1 path_bbox_alloc;
    Alloc_1 drawmonoid_alloc;
    Alloc_1 clip_alloc;
    Alloc_1 clip_bic_alloc;
    Alloc_1 clip_stack_alloc;
    Alloc_1 clip_bbox_alloc;
    Alloc_1 draw_bbox_alloc;
    Alloc_1 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;
};

constant uint3 gl_WorkGroupSize [[maybe_unused]] = uint3(256u, 4u, 1u);

static inline __attribute__((always_inline))
bool check_deps(thread const uint& dep_stage, device Memory& v_59)
{
    uint _65 = atomic_fetch_or_explicit((device atomic_uint*)&v_59.mem_error, 0u, memory_order_relaxed);
    return (_65 & dep_stage) == 0u;
}

static inline __attribute__((always_inline))
bool touch_mem(thread const Alloc& alloc, thread const uint& offset)
{
    return true;
}

static inline __attribute__((always_inline))
uint read_mem(thread const Alloc& alloc, thread const uint& offset, device Memory& v_59)
{
    Alloc param = alloc;
    uint param_1 = offset;
    if (!touch_mem(param, param_1))
    {
        return 0u;
    }
    uint v = v_59.memory[offset];
    return v;
}

static inline __attribute__((always_inline))
Path Path_read(thread const Alloc& a, thread const PathRef& ref, device Memory& v_59)
{
    uint ix = ref.offset >> uint(2);
    Alloc param = a;
    uint param_1 = ix + 0u;
    uint raw0 = read_mem(param, param_1, v_59);
    Alloc param_2 = a;
    uint param_3 = ix + 1u;
    uint raw1 = read_mem(param_2, param_3, v_59);
    Alloc param_4 = a;
    uint param_5 = ix + 2u;
    uint raw2 = read_mem(param_4, param_5, v_59);
    Path s;
    s.bbox = uint4(raw0 & 65535u, raw0 >> uint(16), raw1 & 65535u, raw1 >> uint(16));
    s.tiles = TileRef{ raw2 };
    return s;
}

static inline __attribute__((always_inline))
Alloc new_alloc(thread const uint& offset, thread const uint& size, thread const bool& mem_ok)
{
    Alloc a;
    a.offset = offset;
    return a;
}

static inline __attribute__((always_inline))
void write_mem(thread const Alloc& alloc, thread const uint& offset, thread const uint& val, device Memory& v_59)
{
    Alloc param = alloc;
    uint param_1 = offset;
    if (!touch_mem(param, param_1))
    {
        return;
    }
    v_59.memory[offset] = val;
}

kernel void main0(device Memory& v_59 [[buffer(0)]], const device ConfigBuf& _181 [[buffer(1)]], uint gl_LocalInvocationIndex [[thread_index_in_threadgroup]], uint3 gl_GlobalInvocationID [[thread_position_in_grid]], uint3 gl_LocalInvocationID [[thread_position_in_threadgroup]])
{
    threadgroup uint sh_row_width[256];
    threadgroup Alloc sh_row_alloc[256];
    threadgroup uint sh_row_count[256];
    uint param = 7u;
    bool _154 = check_deps(param, v_59);
    if (!_154)
    {
        return;
    }
    uint th_ix = gl_LocalInvocationIndex;
    uint element_ix = gl_GlobalInvocationID.x;
    uint row_count = 0u;
    if (gl_LocalInvocationID.y == 0u)
    {
        if (element_ix < _181.conf.n_elements)
        {
            PathRef path_ref = PathRef{ _181.conf.tile_alloc.offset + (element_ix * 12u) };
            Alloc param_1;
            param_1.offset = _181.conf.tile_alloc.offset;
            PathRef param_2 = path_ref;
            Path path = Path_read(param_1, param_2, v_59);
            sh_row_width[th_ix] = path.bbox.z - path.bbox.x;
            row_count = path.bbox.w - path.bbox.y;
            bool _225 = row_count == 1u;
            bool _231;
            if (_225)
            {
                _231 = path.bbox.y > 0u;
            }
            else
            {
                _231 = _225;
            }
            if (_231)
            {
                row_count = 0u;
            }
            uint param_3 = path.tiles.offset;
            uint param_4 = ((path.bbox.z - path.bbox.x) * (path.bbox.w - path.bbox.y)) * 8u;
            bool param_5 = true;
            Alloc path_alloc = new_alloc(param_3, param_4, param_5);
            sh_row_alloc[th_ix] = path_alloc;
        }
        sh_row_count[th_ix] = row_count;
    }
    for (uint i = 0u; i < 8u; i++)
    {
        threadgroup_barrier(mem_flags::mem_threadgroup);
        bool _276 = gl_LocalInvocationID.y == 0u;
        bool _283;
        if (_276)
        {
            _283 = th_ix >= (1u << i);
        }
        else
        {
            _283 = _276;
        }
        if (_283)
        {
            row_count += sh_row_count[th_ix - (1u << i)];
        }
        threadgroup_barrier(mem_flags::mem_threadgroup);
        if (gl_LocalInvocationID.y == 0u)
        {
            sh_row_count[th_ix] = row_count;
        }
    }
    threadgroup_barrier(mem_flags::mem_threadgroup);
    uint total_rows = sh_row_count[255];
    uint _360;
    for (uint row = th_ix; row < total_rows; row += 1024u)
    {
        uint el_ix = 0u;
        for (uint i_1 = 0u; i_1 < 8u; i_1++)
        {
            uint probe = el_ix + (128u >> i_1);
            if (row >= sh_row_count[probe - 1u])
            {
                el_ix = probe;
            }
        }
        uint width = sh_row_width[el_ix];
        if (width > 0u)
        {
            Alloc tiles_alloc = sh_row_alloc[el_ix];
            if (el_ix > 0u)
            {
                _360 = sh_row_count[el_ix - 1u];
            }
            else
            {
                _360 = 0u;
            }
            uint seq_ix = row - _360;
            uint tile_el_ix = ((tiles_alloc.offset >> uint(2)) + 1u) + ((seq_ix * 2u) * width);
            Alloc param_6 = tiles_alloc;
            uint param_7 = tile_el_ix;
            uint sum = read_mem(param_6, param_7, v_59);
            for (uint x = 1u; x < width; x++)
            {
                tile_el_ix += 2u;
                Alloc param_8 = tiles_alloc;
                uint param_9 = tile_el_ix;
                sum += read_mem(param_8, param_9, v_59);
                Alloc param_10 = tiles_alloc;
                uint param_11 = tile_el_ix;
                uint param_12 = sum;
                write_mem(param_10, param_11, param_12, v_59);
            }
        }
    }
}