Added new profiling scripts

tools/bench.rb

@@ -1,23 +1,70 @@
 # coding: utf-8
 
-# a quick script for profiling performance with perftools.
-#
-# USAGE
-#
-#     ruby tools/bench.rb
-#     evince bench.pdf
-
-$:.unshift "../lib"
-require 'pdf-reader'
-require 'perftools'
-
-PerfTools::CpuProfiler.start("/tmp/restart_profile") do
-  PDF::Reader.open("restart.pdf") do |reader|
+# a script for measuring text extraction performance
+
+# TO BENCHMARK: ruby tools/bench.rb <runs>
+# TO PROFILE:   ruby tools/bench.rb perftools
+#         OR:   ruby-prof tools/bench.rb <runs>
+# FOR OBJECT ALLOCATION STATS: ruby tools/bench.rb memprof
+# TO COUNT GC RUNS: ruby tools/bench.rb gc
+
+$project_root = File.expand_path(File.join(File.dirname(__FILE__), ".."))
+require 'rubygems' # for Ruby 1.8
+$:.unshift "#{$project_root}/lib"
+require 'pdf/reader'
+
+# Extract all the text from a large PDF
+
+def extract_text
+  PDF::Reader.open("#{$project_root}/spec/data/no_text_spaces.pdf") do |reader|
     reader.pages.each do |page|
       page.text
     end
   end
 end
 
-`pprof.rb --text /tmp/restart_profile > bench.txt`
-`pprof.rb --pdf  /tmp/restart_profile > bench.pdf`
+case ARGV[0]
+when "memprof"
+  # Measure object allocation with memprof
+  require 'memprof'
+  GC.disable
+  Memprof.track { extract_text }
+
+when "perftools"
+  # Profile with perftools.rb
+  # (The best thing about perftools.rb is that it shows you time spent on
+  #   garbage collection)
+  require 'perftools'
+  PerfTools::CpuProfiler.start("/tmp/perftools_data") do
+    extract_text
+  end
+  `pprof.rb --text /tmp/perftools_data > #{$project_root}/tools/profiles/perftools.txt`
+  `pprof.rb --pdf /tmp/perftools_data > #{$project_root}/tools/profiles/perftools.pdf`
+
+when "gc"
+  before = GC.count
+  extract_text
+  puts "GC ran #{GC.count - before} times"  
+
+else
+  # Benchmark
+  # Average the results over multiple runs
+  # Throw out the best and worst results, and average what remains
+  # With 10 runs, the results seem to fluctuate by as much as 6-7%
+  # I'd like that to be 1-2%, but that requires a VERY high number of runs
+
+  runs  = (ARGV[0] || 10).to_i
+  times = []
+
+  runs.times do
+    start = Time.new
+    extract_text
+    times << (Time.new - start)
+    sleep(0.1) # results seem more consistent this way
+  end
+
+  times.sort!
+  times = times.drop(runs / 5).take(runs - (runs * 2 / 3))
+  average = times.reduce(0,&:+).to_f / times.size
+  puts "#{"%0.3f" % average} seconds"
+end

tools/profile.rb

@@ -0,0 +1,20 @@
+# Driver to run a bunch of profiling scripts in parallel,
+#   leaving all the results in tools/profiles
+# Assumes "ruby" is Ruby 1.9, and "ruby1.8" is Ruby 1.8.7
+# Also assumes that all needed gems are installed
+# This script itself should be run under Ruby 1.9
+
+require 'fileutils'
+
+project_root = File.expand_path(File.join(File.dirname(__FILE__), ".."))
+dir = "#{project_root}/tools/profiles"
+FileUtils.mkdir(dir) unless File.exist?(dir)
+
+pids = []
+pids << fork { `ruby-prof #{project_root}/tools/bench.rb 1 --file=#{dir}/rubyprof.txt` }
+pids << fork { `ruby-prof #{project_root}/tools/bench.rb 1 --file=#{dir}/rubyprof-graph.htm --printer=graph_html` }
+pids << fork { `ruby-prof #{project_root}/tools/bench.rb 1 --file=#{dir}/rubyprof-stack.htm --printer=call_stack` }
+pids << fork { `ruby1.8 #{project_root}/tools/bench.rb memprof > #{dir}/memprof.txt` }
+pids << fork { `ruby #{project_root}/tools/bench.rb perftools` }
+
+pids.each { |pid| Process.wait(pid) }