vello/piet-gpu/shader/gen/tile_alloc.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;
};

struct SceneBuf
{
    uint scene[1];
};

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

static inline __attribute__((always_inline))
bool check_deps(thread const uint& dep_stage, device Memory& v_53)
{
    uint _60 = atomic_fetch_or_explicit((device atomic_uint*)&v_53.mem_error, 0u, memory_order_relaxed);
    return (_60 & dep_stage) == 0u;
}

static inline __attribute__((always_inline))
float4 load_draw_bbox(thread const uint& draw_ix, device Memory& v_53, const device ConfigBuf& v_148)
{
    uint base = (v_148.conf.draw_bbox_alloc.offset >> uint(2)) + (4u * draw_ix);
    float x0 = as_type<float>(v_53.memory[base]);
    float y0 = as_type<float>(v_53.memory[base + 1u]);
    float x1 = as_type<float>(v_53.memory[base + 2u]);
    float y1 = as_type<float>(v_53.memory[base + 3u]);
    float4 bbox = float4(x0, y0, x1, y1);
    return bbox;
}

static inline __attribute__((always_inline))
uint malloc_stage(thread const uint& size, thread const uint& mem_size, thread const uint& stage, device Memory& v_53)
{
    uint _70 = atomic_fetch_add_explicit((device atomic_uint*)&v_53.mem_offset, size, memory_order_relaxed);
    uint offset = _70;
    if ((offset + size) > mem_size)
    {
        uint _80 = atomic_fetch_or_explicit((device atomic_uint*)&v_53.mem_error, stage, memory_order_relaxed);
        offset = 0u;
    }
    return offset;
}

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

static inline __attribute__((always_inline))
void write_mem(thread const Alloc& alloc, thread const uint& offset, thread const uint& val, device Memory& v_53)
{
    Alloc param = alloc;
    uint param_1 = offset;
    if (!touch_mem(param, param_1))
    {
        return;
    }
    v_53.memory[offset] = val;
}

static inline __attribute__((always_inline))
void Path_write(thread const Alloc& a, thread const PathRef& ref, thread const Path& s, device Memory& v_53)
{
    uint ix = ref.offset >> uint(2);
    Alloc param = a;
    uint param_1 = ix + 0u;
    uint param_2 = s.bbox.x | (s.bbox.y << uint(16));
    write_mem(param, param_1, param_2, v_53);
    Alloc param_3 = a;
    uint param_4 = ix + 1u;
    uint param_5 = s.bbox.z | (s.bbox.w << uint(16));
    write_mem(param_3, param_4, param_5, v_53);
    Alloc param_6 = a;
    uint param_7 = ix + 2u;
    uint param_8 = s.tiles.offset;
    write_mem(param_6, param_7, param_8, v_53);
}

kernel void main0(device Memory& v_53 [[buffer(0)]], const device ConfigBuf& v_148 [[buffer(1)]], const device SceneBuf& _232 [[buffer(2)]], uint3 gl_LocalInvocationID [[thread_position_in_threadgroup]], uint3 gl_GlobalInvocationID [[thread_position_in_grid]])
{
    threadgroup uint sh_tile_count[256];
    threadgroup uint sh_tile_offset;
    uint param = 1u;
    bool _192 = check_deps(param, v_53);
    if (!_192)
    {
        return;
    }
    uint th_ix = gl_LocalInvocationID.x;
    uint element_ix = gl_GlobalInvocationID.x;
    PathRef path_ref = PathRef{ v_148.conf.tile_alloc.offset + (element_ix * 12u) };
    uint drawtag_base = v_148.conf.drawtag_offset >> uint(2);
    uint drawtag = 0u;
    if (element_ix < v_148.conf.n_elements)
    {
        drawtag = _232.scene[drawtag_base + element_ix];
    }
    int x0 = 0;
    int y0 = 0;
    int x1 = 0;
    int y1 = 0;
    if ((drawtag != 0u) && (drawtag != 37u))
    {
        uint param_1 = element_ix;
        float4 bbox = load_draw_bbox(param_1, v_53, v_148);
        x0 = int(floor(bbox.x * 0.0625));
        y0 = int(floor(bbox.y * 0.0625));
        x1 = int(ceil(bbox.z * 0.0625));
        y1 = int(ceil(bbox.w * 0.0625));
    }
    x0 = clamp(x0, 0, int(v_148.conf.width_in_tiles));
    y0 = clamp(y0, 0, int(v_148.conf.height_in_tiles));
    x1 = clamp(x1, 0, int(v_148.conf.width_in_tiles));
    y1 = clamp(y1, 0, int(v_148.conf.height_in_tiles));
    Path path;
    path.bbox = uint4(uint(x0), uint(y0), uint(x1), uint(y1));
    uint tile_count = uint((x1 - x0) * (y1 - y0));
    sh_tile_count[th_ix] = tile_count;
    uint total_tile_count = tile_count;
    for (uint i = 0u; i < 8u; i++)
    {
        threadgroup_barrier(mem_flags::mem_threadgroup);
        if (th_ix >= (1u << i))
        {
            total_tile_count += sh_tile_count[th_ix - (1u << i)];
        }
        threadgroup_barrier(mem_flags::mem_threadgroup);
        sh_tile_count[th_ix] = total_tile_count;
    }
    if (th_ix == 255u)
    {
        uint param_2 = total_tile_count * 8u;
        uint param_3 = v_148.conf.mem_size;
        uint param_4 = 2u;
        uint _370 = malloc_stage(param_2, param_3, param_4, v_53);
        sh_tile_offset = _370;
    }
    threadgroup_barrier(mem_flags::mem_threadgroup);
    uint offset_start = sh_tile_offset;
    if (offset_start == 0u)
    {
        return;
    }
    if (element_ix < v_148.conf.n_elements)
    {
        uint _387;
        if (th_ix > 0u)
        {
            _387 = sh_tile_count[th_ix - 1u];
        }
        else
        {
            _387 = 0u;
        }
        uint tile_subix = _387;
        path.tiles = TileRef{ offset_start + (8u * tile_subix) };
        Alloc param_5;
        param_5.offset = v_148.conf.tile_alloc.offset;
        PathRef param_6 = path_ref;
        Path param_7 = path;
        Path_write(param_5, param_6, param_7, v_53);
    }
    uint total_count = sh_tile_count[255] * 2u;
    uint start_ix = offset_start >> uint(2);
    for (uint i_1 = th_ix; i_1 < total_count; i_1 += 256u)
    {
        v_53.memory[start_ix + i_1] = 0u;
    }
}