|
| 1 | +# frozen_string_literal: true |
| 2 | + |
| 3 | +require 'open3' |
| 4 | + |
| 5 | +DEFAULT_NWORKERS = 6 |
| 6 | +PROGRAM_NAME = 'main.rb' |
| 7 | + |
| 8 | +Measurement = Struct.new(:min, :max, :sum, :count) |
| 9 | + |
| 10 | +# usage: `ruby main.rb <file> <threads>` |
| 11 | +# single worker: `ruby main.rb <file> <threads> worker <index>` |
| 12 | + |
| 13 | +def main |
| 14 | + mode = ARGV[2] |
| 15 | + |
| 16 | + if mode == 'worker' |
| 17 | + run_worker |
| 18 | + else |
| 19 | + results = calculate_in_processes |
| 20 | + Dir['worker_data_*.bin'].each { File.delete(_1) } |
| 21 | + merge(results) |
| 22 | + end |
| 23 | +end |
| 24 | + |
| 25 | +def result_file_path(process_index) |
| 26 | + "worker_data_#{process_index}.bin" |
| 27 | +end |
| 28 | + |
| 29 | +def calculate_in_processes |
| 30 | + file = ARGV[0] |
| 31 | + nworkers = ARGV[1]&.to_i || DEFAULT_NWORKERS |
| 32 | + threads = (0...nworkers).map do |i| |
| 33 | + Thread.new do |
| 34 | + _, stderr_str, status = Open3.capture3("ruby --yjit #{PROGRAM_NAME} #{file} #{nworkers} worker #{i}") |
| 35 | + unless status.success? |
| 36 | + puts stderr_str |
| 37 | + # TODO: kill running processes when one is not successful |
| 38 | + exit |
| 39 | + end |
| 40 | + result = |
| 41 | + File.open(result_file_path(i), 'rb') do |inp| |
| 42 | + Marshal.load(inp) |
| 43 | + end |
| 44 | + File.delete(result_file_path(i)) |
| 45 | + result |
| 46 | + end |
| 47 | + end |
| 48 | + threads.map(&:value) |
| 49 | +end |
| 50 | + |
| 51 | +def merge(results) |
| 52 | + result, *other = results |
| 53 | + |
| 54 | + |
| 55 | + result.merge!(*other) do |_, old_value, new_value| |
| 56 | + Measurement.new( |
| 57 | + [old_value.min, new_value.min].min, |
| 58 | + [old_value.max, new_value.max].max, |
| 59 | + old_value.sum + new_value.sum, |
| 60 | + old_value.count + new_value.count |
| 61 | + ) |
| 62 | + end |
| 63 | + |
| 64 | + sorted = result.sort.map do |name, stats| |
| 65 | + min = stats.min.round(1) |
| 66 | + mean = (stats.sum / stats.count + 0.0000001).round(1) |
| 67 | + max = stats.max.round(1) |
| 68 | + "#{name}=#{min}/#{mean}/#{max}" |
| 69 | + end |
| 70 | + |
| 71 | + print '{' |
| 72 | + print sorted.join(', ') |
| 73 | + puts '}' |
| 74 | +end |
| 75 | + |
| 76 | +def run_worker |
| 77 | + file = ARGV[0] |
| 78 | + nworkers = ARGV[1].to_i |
| 79 | + index = ARGV[3].to_i |
| 80 | + |
| 81 | + chunk_size = File.size(file) / nworkers |
| 82 | + offset = index * chunk_size |
| 83 | + |
| 84 | + result = calculate_data(file, offset, chunk_size, index) |
| 85 | + File.open(result_file_path(index), 'wb') do |out| |
| 86 | + Marshal.dump(result, out) |
| 87 | + end |
| 88 | +end |
| 89 | + |
| 90 | +def calculate_data(file, offset, chunk_size, index) |
| 91 | + skip_line = index != 0 |
| 92 | + buffer = IO::Buffer.map(File.open(file), nil, 0, IO::Buffer::READONLY) |
| 93 | + data = {} |
| 94 | + |
| 95 | + name = nil |
| 96 | + line_start_offset = 0 |
| 97 | + number_start_offset = 0 |
| 98 | + |
| 99 | + if skip_line |
| 100 | + buffer.each(:U8, offset) do |byte_offset, byte| |
| 101 | + next unless byte == 10 |
| 102 | + |
| 103 | + offset = offset + (byte_offset - offset) + 1 |
| 104 | + line_start_offset = offset |
| 105 | + break |
| 106 | + end |
| 107 | + end |
| 108 | + |
| 109 | + buffer.each(:U8, offset) do |byte_offset, byte| |
| 110 | + if byte == 59 # ';' |
| 111 | + name = buffer.get_string(line_start_offset, byte_offset - line_start_offset) |
| 112 | + number_start_offset = byte_offset + 1 |
| 113 | + elsif byte == 10 # '\n' |
| 114 | + value = buffer.get_string(number_start_offset, byte_offset - number_start_offset).to_f |
| 115 | + measurement = data[name] |
| 116 | + measurement ||= data[name] = Measurement.new(value, value, 0, 0) |
| 117 | + measurement.min = value if value < measurement.min |
| 118 | + measurement.max = value if value > measurement.max |
| 119 | + measurement.count += 1 |
| 120 | + measurement.sum += value |
| 121 | + |
| 122 | + line_start_offset = byte_offset + 1 |
| 123 | + |
| 124 | + break if byte_offset >= offset + chunk_size |
| 125 | + end |
| 126 | + end |
| 127 | + |
| 128 | + data |
| 129 | +end |
| 130 | + |
| 131 | +main |
0 commit comments