async inclusive_scan gist

test.cpp

    std::vector<double> output(input.size());
    const std::size_t tile_size = (input.size() + tile_count - 1) / tile_count;

    std::vector<double> partials(tile_count + 1);
    partials[0] = init;

    return ex::just_on(sched, partials) |
        ex::bulk(tile_count,
            [&](std::size_t i, std::vector<double>& partials) {
                auto start = i * tile_size;
                auto end = std::min(input.size(), (i + 1) * tile_size);
                partials[i + 1] =
                    *--std::inclusive_scan(std::begin(input) + start,
                        std::begin(input) + end, std::begin(output) + start);
            }) |
        ex::transform([](std::vector<double>& partials) {
            std::inclusive_scan(
                std::begin(partials), std::end(partials), std::begin(partials));
            return std::move(partials);
        }) |
        ex::bulk(tile_count,
            [&](std::size_t i, std::vector<double>& partials) {
                auto start = i * tile_size;
                auto end = std::min(input.size(), (i + 1) * tile_size);
                std::for_each(output.begin() + start,
                    output.begin() + end,
                    [&](double& e) { e = partials[i] + e; });
            }) |
        ex::transform([&](std::vector<double>&) { return output; }) |
        ex::sync_wait();