vello/tests/shader/prefix_scan.comp

// SPDX-License-Identifier: Apache-2.0 OR MIT OR Unlicense

// A scan for a tree reduction prefix scan (either root or not, by ifdef).

#version 450

#define N_ROWS 8
#define LG_WG_SIZE 9
#define WG_SIZE (1 << LG_WG_SIZE)
#define PARTITION_SIZE (WG_SIZE * N_ROWS)

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

struct Monoid {
    uint element;
};

layout(set = 0, binding = 0) buffer DataBuf {
    Monoid[] data;
};

#ifndef ROOT
layout(set = 0, binding = 1) readonly buffer ParentBuf {
    Monoid[] parent;
};
#endif

shared Monoid sh_scratch[WG_SIZE];

Monoid combine_monoid(Monoid a, Monoid b) {
    return Monoid(a.element + b.element);
}

void main() {
    Monoid local[N_ROWS];

    uint ix = gl_GlobalInvocationID.x * N_ROWS;

    // TODO: gate buffer read
    local[0] = data[ix];
    for (uint i = 1; i < N_ROWS; i++) {
        local[i] = combine_monoid(local[i - 1], data[ix + i]);
    }
    Monoid agg = local[N_ROWS - 1];
    sh_scratch[gl_LocalInvocationID.x] = agg;
    for (uint i = 0; i < LG_WG_SIZE; i++) {
        barrier();
        if (gl_LocalInvocationID.x >= (1u << i)) {
            Monoid other = sh_scratch[gl_LocalInvocationID.x - (1u << i)];
            agg = combine_monoid(other, agg);
        }
        barrier();
        sh_scratch[gl_LocalInvocationID.x] = agg;
    }
    
    barrier();
    // This could be a semigroup instead of a monoid if we reworked the
    // conditional logic, but that might impact performance.
    Monoid row = Monoid(0);
#ifdef ROOT
    if (gl_LocalInvocationID.x > 0) {
        row = sh_scratch[gl_LocalInvocationID.x - 1];
    }
#else
    if (gl_WorkGroupID.x > 0) {
        row = parent[gl_WorkGroupID.x - 1];
    }
    if (gl_LocalInvocationID.x > 0) {
        row = combine_monoid(row, sh_scratch[gl_LocalInvocationID.x - 1]);
    }
#endif
    for (uint i = 0; i < N_ROWS; i++) {
        Monoid m = combine_monoid(row, local[i]);
        // TODO: gate buffer write
        data[ix + i] = m;
    }
}