// The coarse rasterizer stage of the pipeline.
//
// As input we have the ordered partitions of paths from the binning phase and
// the annotated tile list of segments and backdrop per path.
//
// Each workgroup operating on one bin by stream compacting
// the elements corresponding to the bin.
//
// As output we have an ordered command stream per tile. Every tile from a path (backdrop + segment list) will be encoded.

#version 450
#extension GL_GOOGLE_include_directive : enable

#include "setup.h"

layout(local_size_x = N_TILE, local_size_y = 1) in;

layout(set = 0, binding = 0) buffer AnnotatedBuf {
    uint[] annotated;
};

layout(set = 0, binding = 1) buffer BinsBuf {
    uint[] bins;
};

layout(set = 0, binding = 2) buffer TileBuf {
    uint[] tile;
};

layout(set = 0, binding = 3) buffer AllocBuf {
    uint n_elements;
    uint alloc;
};

layout(set = 0, binding = 4) buffer PtclBuf {
    uint[] ptcl;
};

#include "annotated.h"
#include "bins.h"
#include "tile.h"
#include "ptcl.h"

#define LG_N_PART_READ (7 + LG_WG_FACTOR)
#define N_PART_READ (1 << LG_N_PART_READ)

shared uint sh_elements[N_TILE];
shared float sh_right_edge[N_TILE];

// Number of elements in the partition; prefix sum.
shared uint sh_part_count[N_PART_READ];
shared uint sh_part_elements[N_PART_READ];

shared uint sh_bitmaps[N_SLICE][N_TILE];

shared uint sh_tile_count[N_TILE];
// The width of the tile rect for the element, intersected with this bin
shared uint sh_tile_width[N_TILE];
shared uint sh_tile_x0[N_TILE];
shared uint sh_tile_y0[N_TILE];

// These are set up so base + tile_y * stride + tile_x points to a Tile.
shared uint sh_tile_base[N_TILE];
shared uint sh_tile_stride[N_TILE];

// Perhaps cmd_limit should be a global? This is a style question.
void alloc_cmd(inout CmdRef cmd_ref, inout uint cmd_limit) {
    if (cmd_ref.offset > cmd_limit) {
        uint new_cmd = atomicAdd(alloc, PTCL_INITIAL_ALLOC);
        CmdJump jump = CmdJump(new_cmd);
        Cmd_Jump_write(cmd_ref, jump);
        cmd_ref = CmdRef(new_cmd);
        cmd_limit = new_cmd + PTCL_INITIAL_ALLOC - 2 * Cmd_size;
    }
}

void main() {
    // Could use either linear or 2d layouts for both dispatch and
    // invocations within the workgroup. We'll use variables to abstract.
    uint bin_ix = N_TILE_X * gl_WorkGroupID.y + gl_WorkGroupID.x;
    uint partition_ix = 0;
    uint n_partitions = (n_elements + N_TILE - 1) / N_TILE;
    uint th_ix = gl_LocalInvocationID.x;

    // Coordinates of top left of bin, in tiles.
    uint bin_tile_x = N_TILE_X * gl_WorkGroupID.x;
    uint bin_tile_y = N_TILE_Y * gl_WorkGroupID.y;
    uint tile_x = gl_LocalInvocationID.x % N_TILE_X;
    uint tile_y = gl_LocalInvocationID.x / N_TILE_X;
    uint this_tile_ix = (bin_tile_y + tile_y) * WIDTH_IN_TILES + bin_tile_x + tile_x;
    CmdRef cmd_ref = CmdRef(this_tile_ix * PTCL_INITIAL_ALLOC);
    uint cmd_limit = cmd_ref.offset + PTCL_INITIAL_ALLOC - 2 * Cmd_size;

    // I'm sure we can figure out how to do this with at least one fewer register...
    // Items up to rd_ix have been read from sh_elements
    uint rd_ix = 0;
    // Items up to wr_ix have been written into sh_elements
    uint wr_ix = 0;
    // Items between part_start_ix and ready_ix are ready to be transferred from sh_part_elements
    uint part_start_ix = 0;
    uint ready_ix = 0;
    while (true) {
        for (uint i = 0; i < N_SLICE; i++) {
            sh_bitmaps[i][th_ix] = 0;
        }

        // parallel read of input partitions
        do {
            if (ready_ix == wr_ix && partition_ix < n_partitions) {
                part_start_ix = ready_ix;
                uint count = 0;
                if (th_ix < N_PART_READ && partition_ix + th_ix < n_partitions) {
                    uint in_ix = ((partition_ix + th_ix) * N_TILE + bin_ix) * 2;
                    count = bins[in_ix];
                    sh_part_elements[th_ix] = bins[in_ix + 1];
                }
                // prefix sum of counts
                for (uint i = 0; i < LG_N_PART_READ; i++) {
                    if (th_ix < N_PART_READ) {
                        sh_part_count[th_ix] = count;
                    }
                    barrier();
                    if (th_ix < N_PART_READ) {
                        if (th_ix >= (1 << i)) {
                            count += sh_part_count[th_ix - (1 << i)];
                        }
                    }
                    barrier();
                }
                if (th_ix < N_PART_READ) {
                    sh_part_count[th_ix] = part_start_ix + count;
                }
                barrier();
                ready_ix = sh_part_count[N_PART_READ - 1];
                partition_ix += N_PART_READ;
            }
            // use binary search to find element to read
            uint ix = rd_ix + th_ix;
            if (ix >= wr_ix && ix < ready_ix) {
                uint part_ix = 0;
                for (uint i = 0; i < LG_N_PART_READ; i++) {
                    uint probe = part_ix + ((N_PART_READ / 2) >> i);
                    if (ix >= sh_part_count[probe - 1]) {
                        part_ix = probe;
                    }
                }
                ix -= part_ix > 0 ? sh_part_count[part_ix - 1] : part_start_ix;
                BinInstanceRef inst_ref = BinInstanceRef(sh_part_elements[part_ix]);
                BinInstance inst = BinInstance_read(BinInstance_index(inst_ref, ix));
                sh_elements[th_ix] = inst.element_ix;
                sh_right_edge[th_ix] = inst.right_edge;
            }
            barrier();

            wr_ix = min(rd_ix + N_TILE, ready_ix);
        } while (wr_ix - rd_ix < N_TILE && (wr_ix < ready_ix || partition_ix < n_partitions));

        // We've done the merge and filled the buffer.

        // Read one element, compute coverage.
        uint tag = Annotated_Nop;
        uint element_ix;
        AnnotatedRef ref;
        float right_edge = 0.0;
        if (th_ix + rd_ix < wr_ix) {
            element_ix = sh_elements[th_ix];
            right_edge = sh_right_edge[th_ix];
            ref = AnnotatedRef(element_ix * Annotated_size);
            tag = Annotated_tag(ref);
        }

        // Bounding box of element in pixel coordinates.
        uint tile_count;
        switch (tag) {
        case Annotated_Fill:
        case Annotated_Stroke:
            // Because the only elements we're processing right now are
            // paths, we can just use the element index as the path index.
            // In future, when we're doing a bunch of stuff, the path index
            // should probably be stored in the annotated element.
            uint path_ix = element_ix;
            Path path = Path_read(PathRef(path_ix * Path_size));
            uint stride = path.bbox.z - path.bbox.x;
            sh_tile_stride[th_ix] = stride;
            int dx = int(path.bbox.x) - int(bin_tile_x);
            int dy = int(path.bbox.y) - int(bin_tile_y);
            int x0 = clamp(dx, 0, N_TILE_X);
            int y0 = clamp(dy, 0, N_TILE_Y);
            int x1 = clamp(int(path.bbox.z) - int(bin_tile_x), 0, N_TILE_X);
            int y1 = clamp(int(path.bbox.w) - int(bin_tile_y), 0, N_TILE_Y);
            sh_tile_width[th_ix] = uint(x1 - x0);
            sh_tile_x0[th_ix] = x0;
            sh_tile_y0[th_ix] = y0;
            tile_count = uint(x1 - x0) * uint(y1 - y0);
            // base relative to bin
            uint base = path.tiles.offset - uint(dy * stride + dx) * Tile_size;
            sh_tile_base[th_ix] = base;
            break;
        default:
            tile_count = 0;
            break;
        }

        // Prefix sum of sh_tile_count
        sh_tile_count[th_ix] = tile_count;
        for (uint i = 0; i < LG_N_TILE; i++) {
            barrier();
            if (th_ix >= (1 << i)) {
                tile_count += sh_tile_count[th_ix - (1 << i)];
            }
            barrier();
            sh_tile_count[th_ix] = tile_count;
        }
        barrier();
        uint total_tile_count = sh_tile_count[N_TILE - 1];
        for (uint ix = th_ix; ix < total_tile_count; ix += N_TILE) {
            // Binary search to find element
            uint el_ix = 0;
            for (uint i = 0; i < LG_N_TILE; i++) {
                uint probe = el_ix + ((N_TILE / 2) >> i);
                if (ix >= sh_tile_count[probe - 1]) {
                    el_ix = probe;
                }
            }
            uint seq_ix = ix - (el_ix > 0 ? sh_tile_count[el_ix - 1] : 0);
            uint width = sh_tile_width[el_ix];
            uint x = sh_tile_x0[el_ix] + seq_ix % width;
            uint y = sh_tile_y0[el_ix] + seq_ix / width;
            Tile tile = Tile_read(TileRef(sh_tile_base[el_ix] + (sh_tile_stride[el_ix] * y + x) * Tile_size));
            if (tile.tile.offset != 0 || tile.backdrop != 0) {
                uint el_slice = el_ix / 32;
                uint el_mask = 1 << (el_ix & 31);
                atomicOr(sh_bitmaps[el_slice][y * N_TILE_X + x], el_mask);
            }
        }

        barrier();

        // Output non-segment elements for this tile. The thread does a sequential walk
        // through the non-segment elements, and for segments, count and backdrop are
        // aggregated using bit counting.
        uint slice_ix = 0;
        uint bitmap = sh_bitmaps[0][th_ix];
        while (true) {
            if (bitmap == 0) {
                slice_ix++;
                if (slice_ix == N_SLICE) {
                    break;
                }
                bitmap = sh_bitmaps[slice_ix][th_ix];
                if (bitmap == 0) {
                    continue;
                }
            }
            uint element_ref_ix = slice_ix * 32 + findLSB(bitmap);
            uint element_ix = sh_elements[element_ref_ix];

            // Clear LSB
            bitmap &= bitmap - 1;

            // At this point, we read the element again from global memory.
            // If that turns out to be expensive, maybe we can pack it into
            // shared memory (or perhaps just the tag).
            ref = AnnotatedRef(element_ix * Annotated_size);
            tag = Annotated_tag(ref);

            switch (tag) {
            case Annotated_Fill:
                Tile tile = Tile_read(TileRef(sh_tile_base[element_ref_ix]
                    + (sh_tile_stride[element_ref_ix] * tile_y + tile_x) * Tile_size));
                AnnoFill fill = Annotated_Fill_read(ref);
                alloc_cmd(cmd_ref, cmd_limit);
                if (tile.tile.offset != 0) {
                    CmdFill cmd_fill;
                    cmd_fill.tile_ref = tile.tile.offset;
                    cmd_fill.backdrop = tile.backdrop;
                    cmd_fill.rgba_color = fill.rgba_color;
                    Cmd_Fill_write(cmd_ref, cmd_fill);
                } else {
                    Cmd_Solid_write(cmd_ref, CmdSolid(fill.rgba_color));
                }
                cmd_ref.offset += Cmd_size;
                break;
            case Annotated_Stroke:
                tile = Tile_read(TileRef(sh_tile_base[element_ref_ix]
                    + (sh_tile_stride[element_ref_ix] * tile_y + tile_x) * Tile_size));
                AnnoStroke stroke = Annotated_Stroke_read(ref);
                CmdStroke cmd_stroke;
                cmd_stroke.tile_ref = tile.tile.offset;
                cmd_stroke.half_width = 0.5 * stroke.linewidth;
                cmd_stroke.rgba_color = stroke.rgba_color;
                alloc_cmd(cmd_ref, cmd_limit);
                Cmd_Stroke_write(cmd_ref, cmd_stroke);
                cmd_ref.offset += Cmd_size;
                break;
            }
        }
        barrier();

        rd_ix += N_TILE;
        if (rd_ix >= ready_ix && partition_ix >= n_partitions) break;
    }
    Cmd_End_write(cmd_ref);
}