Add a benchmarker that stores sizes in a skia perf compatible file. (#22097)

This will be run by emscripten releases to start tracking size
benchmarks on our skia perf instance. I've intentially only added two
benchmarks to start for testing.

The output format was verified using perf-tool. More info on the format
at
https://skia.googlesource.com/buildbot/+/refs/heads/main/perf/FORMAT.md

---------

Co-authored-by: Alon Zakai <alonzakai@gmail.com>

Author: brendandahl

test/test_benchmark.py

@@ -10,6 +10,7 @@
 import sys
 import time
 import unittest
+import json
 import zlib
 from pathlib import Path
 from typing import List
@@ -56,6 +57,9 @@
 
 
 class Benchmarker():
+  # Whether to record statistics. Set by SizeBenchmarker.
+  record_stats = False
+
   # called when we init the object, which is during startup, even if we are
   # not running benchmarks
   def __init__(self, name):
@@ -112,14 +116,39 @@ def display(self, baseline=None):
 
     # size
 
-    size = sum(os.path.getsize(f) for f in self.get_output_files())
-    gzip_size = sum(len(zlib.compress(read_binary(f))) for f in self.get_output_files())
+    recorded_stats = []
+
+    def add_stat(name, size, gzip_size):
+      recorded_stats.append({
+        'value': name,
+        'measurement': size,
+      })
+      recorded_stats.append({
+        'value': name + ' (gzipped)',
+        'measurement': gzip_size,
+      })
+
+    total_size = 0
+    total_gzip_size = 0
 
-    print('        size: %8s, compressed: %8s' % (size, gzip_size), end=' ')
+    for file in self.get_output_files():
+      size = os.path.getsize(file)
+      gzip_size = len(zlib.compress(read_binary(file)))
+      if self.record_stats:
+        add_stat(utils.removeprefix(os.path.basename(file), 'size_'), size, gzip_size)
+      total_size += size
+      total_gzip_size += gzip_size
+
+    if self.record_stats:
+      add_stat('total', total_size, total_gzip_size)
+
+    print('        size: %8s, compressed: %8s' % (total_size, total_gzip_size), end=' ')
     if self.get_size_text():
       print('  (' + self.get_size_text() + ')', end=' ')
     print()
 
+    return recorded_stats
+
   def get_size_text(self):
     return ''
 
@@ -240,6 +269,21 @@ def get_output_files(self):
     return ret
 
 
+# This benchmarker will make a test benchmark build with Emscripten and record
+# the file output sizes in out/test/stats.json. The file format is specified at
+# https://skia.googlesource.com/buildbot/+/refs/heads/main/perf/FORMAT.md
+# Running the benchmark will be skipped.
+class SizeBenchmarker(EmscriptenBenchmarker):
+  record_stats = True
+
+  def __init__(self, name):
+    # do not set an engine, as we will not run the code
+    super().__init__(name, engine=None)
+
+  # we will not actually run the benchmarks
+  run = None
+
+
 CHEERP_BIN = '/opt/cheerp/bin/'
 
 
@@ -317,6 +361,7 @@ def get_output_files(self):
 named_benchmarkers = {
   'clang': NativeBenchmarker('clang', [CLANG_CC], [CLANG_CXX]),
   'gcc': NativeBenchmarker('gcc',   ['gcc', '-no-pie'],  ['g++', '-no-pie']),
+  'size': SizeBenchmarker('size'),
   'v8': EmscriptenBenchmarker('v8', aot_v8),
   'v8-lto': EmscriptenBenchmarker('v8-lto', aot_v8, ['-flto']),
   'v8-ctors': EmscriptenBenchmarker('v8-ctors', aot_v8, ['-sEVAL_CTORS']),
@@ -337,6 +382,7 @@ def get_output_files(self):
 
 class benchmark(common.RunnerCore):
   save_dir = True
+  stats = [] # type: ignore
 
   @classmethod
   def setUpClass(cls):
@@ -358,6 +404,17 @@ def setUpClass(cls):
     fingerprint.append('llvm: ' + config.LLVM_ROOT)
     print('Running Emscripten benchmarks... [ %s ]' % ' | '.join(fingerprint))
 
+  @classmethod
+  def tearDownClass(cls):
+    super().tearDownClass()
+    if cls.stats:
+      output = {
+        'version': 1,
+        'git_hash': '',
+        'results': cls.stats
+      }
+      utils.write_file('stats.json', json.dumps(output, indent=2) + '\n')
+
   # avoid depending on argument reception from the commandline
   def hardcode_arguments(self, code):
     if not code or 'int main()' in code:
@@ -397,11 +454,27 @@ def do_benchmark(self, name, src, expected_output='FAIL', args=None,
     for b in benchmarkers:
       if skip_native and isinstance(b, NativeBenchmarker):
         continue
+      if not b.run:
+        # If we won't run the benchmark, we don't need repetitions.
+        reps = 0
       baseline = b
       print('Running benchmarker: %s: %s' % (b.__class__.__name__, b.name))
       b.build(self, filename, args, shared_args, emcc_args, native_args, native_exec, lib_builder, has_output_parser=output_parser is not None)
       b.bench(args, output_parser, reps, expected_output)
-      b.display(baseline)
+      recorded_stats = b.display(baseline)
+      if recorded_stats:
+        self.add_stats(name, recorded_stats)
+
+  def add_stats(self, name, stats):
+    self.stats.append({
+      'key': {
+        'test': name,
+        'units': 'bytes'
+      },
+      'measurements': {
+        'stats': stats
+      }
+    })
 
   def test_primes(self, check=True):
     src = r'''