[Interpreter] Add Ignition profile visualization tool.

tools/ignition_perf_report.py

Index: tools/ignition_perf_report.py
diff --git a/tools/ignition_perf_report.py b/tools/ignition_perf_report.py
new file mode 100755
index 0000000000000000000000000000000000000000..7679084ddb49e3febbd275371bd3c3e87ad315ee
--- /dev/null
+++ b/tools/ignition_perf_report.py
@@ -0,0 +1,140 @@
+#! /usr/bin/python2
+#
+# Copyright 2016 the V8 project authors. All rights reserved.
+# Use of this source code is governed by a BSD-style license that can be
+# found in the LICENSE file.
+#
+
+import argparse
+import collections
+import re
+import subprocess
+import sys
+
+
+DESCRIPTION = """
+Processes a perf.data sample file and reports the hottest Ignition bytecodes,
+or write an input file for flamegraph.pl.
+"""
+
+
+EPILOGUE = """

[rmcilroy, 2016/03/11 11:26:49]

HELP_EPILOGUE. Also make these all private by adding two leading underscores.

[Stefano Sanfilippo, 2016/03/11 16:33:51]

Done.

[rmcilroy, 2016/03/14 12:26:52]

You never made the fields private.

[Stefano Sanfilippo, 2016/03/14 13:14:20]

Wooops, sorry about that. Done for the help strings. About the regex, we have
test cases for those, so they need to be exportable. How should we handle this
case?

[rmcilroy, 2016/03/14 14:48:00]

This is fine, thanks.

+examples:
+  # Get a flamegraph for Ignition bytecode handlers on Octane benchmark,
+  # without considering time spent compiling JS code.
+  #
+  $ tools/run-perf.sh out/x64.release/d8 --ignition octane/run.js
+  $ tools/ignition_perf_report.py --flamegraph --hide-compile -o out.collapsed
+  $ flamegraph.pl --colors js out.collapsed > out.svg
+
+  # See the hottest bytecodes on Octane benchmark, by number of samples.
+  #
+  $ tools/run-perf.sh out/x64.release/d8 --ignition octane/run.js
+  $ tools/ignition_perf_report.py
+"""
+
+
+COMPILER_SYMBOLS_RE = re.compile(r"Builtin:Compile(?:Lazy|OptimizedConcurrent)"
+                                  "$|LazyCompile:|v8::internal::Compile")
+
+
+# Function name, strip parameters
+SYMBOL_NAME_RE = re.compile(r"((?:\(anonymous namespace\)|[^(])+)")
+
+
+def yield_collapsed_callchains(perf_stream, hide_compile_time=False):

[rmcilroy, 2016/03/11 11:26:48]

Probably better named collapsed_callchains_generator.

Also please add comments to the to of the function stating what it does and what
the return value is - also for count_callchains and count_handler_samples)

[Stefano Sanfilippo, 2016/03/11 16:33:51]

Done.

+  current_chain = []
+  keep_parsing_chain = True
+  for line in perf_stream:
+    if line[0] == "#":

[rmcilroy, 2016/03/11 11:26:49]

Add some comments (e.g., # Skip comments, # Empty line signals a new call chain,
etc.)

[Stefano Sanfilippo, 2016/03/11 16:33:51]

Done.

+      continue
+    line = line.strip()
+    if not line:
+      keep_parsing_chain = True
+      current_chain = []
+      continue
+    if not keep_parsing_chain:
+      continue
+    symbol = SYMBOL_NAME_RE.match(line.split(" ", 1)[1]).group(1)

[rmcilroy, 2016/03/11 11:26:49]

nit newline above

[Stefano Sanfilippo, 2016/03/11 16:33:51]

Done.

+    current_chain.append(symbol)
+    if hide_compile_time and COMPILER_SYMBOLS_RE.match(symbol):
+      keep_parsing_chain = False
+    elif symbol.startswith("BytecodeHandler:"):
+      keep_parsing_chain = False
+      yield current_chain
+
+
+def count_callchains(callchains):

[rmcilroy, 2016/03/11 11:26:49]

get_callchain_sample_counts ? 

[Stefano Sanfilippo, 2016/03/11 16:33:51]

I think calculate_* better conveys the idea that there's some computation going
on, get_* seems more a plain getter than anything.

+  chain_counters = collections.defaultdict(int)
+  for callchain in callchains:
+    key = ";".join(reversed(callchain))
+    chain_counters[key] += 1
+  return chain_counters.items()
+
+
+def count_handler_samples(callchains):

[rmcilroy, 2016/03/11 11:26:49]

get_bytecode_handler_sample_counts ?

[Stefano Sanfilippo, 2016/03/11 16:33:51]

Same as above.

+  handler_counters = collections.defaultdict(int)
+  for callchain in callchains:
+    # Strip the "BytecodeHandler:" prefix
+    handler = callchain[-1].split(":", 1)[1]
+    handler_counters[handler] += 1
+  # Sort by decreasing number of samples
+  return sorted(handler_counters.items(),

[rmcilroy, 2016/03/11 11:26:49]

Maybe just do the sort in main below, to keep these two functions similar in
their API. 

[Stefano Sanfilippo, 2016/03/11 16:33:51]

Done.

+                key=lambda entry: entry[1], reverse=True)
+
+
+def parse_command_line():
+  command_line_parser = argparse.ArgumentParser(
+    formatter_class=argparse.RawDescriptionHelpFormatter,
+    description=DESCRIPTION,
+    epilog=EPILOGUE)
+
+  command_line_parser.add_argument(
+    "perf_filename",
+    help="perf sample file to process (default: perf.data)",
+    nargs="?",
+    default="perf.data",
+    metavar="<perf filename>")
+  command_line_parser.add_argument(
+    "--flamegraph", "-f",
+    help="output an input file for flamegraph.pl, not a report",
+    action="store_true",
+    dest="output_flamegraph")
+  command_line_parser.add_argument(
+    "--hide-compile", "-c",
+    help="do not count samples inside compiler routines",
+    action="store_true",
+    dest="hide_compile_time")
+  command_line_parser.add_argument(
+    "--output", "-o",
+    help="output file name (stdout if omitted)",
+    type=argparse.FileType('wt'),
+    default=sys.stdout,
+    metavar="<output filename>",
+    dest="output_stream")
+
+  return command_line_parser.parse_args()
+
+
+def main():
+  program_options = parse_command_line()
+  output_stream = program_options.output_stream
+  perf = subprocess.Popen(["perf", "script", "-f", "ip,sym",
+                           "-i", program_options.perf_filename],
+                          stdout=subprocess.PIPE)
+  callchains = yield_collapsed_callchains(perf.stdout,
+                                          program_options.hide_compile_time)
+  if program_options.output_flamegraph:
+    for callchain, count in count_callchains(callchains):
+      output_stream.write("{} {}\n".format(callchain, count))

[rmcilroy, 2016/03/11 11:26:49]

nit - move to a seperate function (for clarity) and the same for the code to
output hottest bytecodes.

[Stefano Sanfilippo, 2016/03/11 16:33:51]

Done.

+  else:
+    handler_counters = count_handler_samples(callchains)
+    samples_num = sum(counter for _, counter in handler_counters)
+    for bytecode_name, count in handler_counters:
+      output_stream.write(
+        "{}\t{}\t{:.3f}%\n".format(bytecode_name, count,
+                                   100. * count / samples_num))
+
+
+if __name__ == "__main__":
+  main()