Reserve more RAM per Gold process.

build/toolchain/get_concurrent_links.py

 # Copyright 2014 The Chromium Authors. All rights reserved.
 # Use of this source code is governed by a BSD-style license that can be
 # found in the LICENSE file.
 
 # This script computs the number of concurrent links we want to run in the build
 # as a function of machine spec. It's based on GetDefaultConcurrentLinks in GYP.
 
 import optparse
 import os
 import re
@@ skipping 37 matching lines @@
         for line in meminfo:
           match = memtotal_re.match(line)
           if not match:
             continue
           mem_total_gb = float(match.group(1)) / (2 ** 20)
           # Allow 8Gb per link on Linux because Gold is quite memory hungry
           mem_per_link_gb = 8
           if is_lto:
             mem_total_gb -= 10 # Reserve
             # For LTO builds the RAM requirements are even higher
-            mem_per_link_gb = 16
+            mem_per_link_gb = 24
           return int(max(1, mem_total_gb / mem_per_link_gb))
     return 1
   elif sys.platform == 'darwin':
     try:
       avail_bytes = int(subprocess.check_output(['sysctl', '-n', 'hw.memsize']))
       # A static library debug build of Chromium's unit_tests takes ~2.7GB, so
       # 4GB per ld process allows for some more bloat.
       return max(1, avail_bytes / (4 * (2 ** 30)))  # total / 4GB
     except Exception:
       return 1
   else:
     # TODO(scottmg): Implement this for other platforms.
     return 1
 
 def main():
   parser = optparse.OptionParser()
   parser.add_option('--lto', action="store_true", default=False,
                     help='This is an LTO build with higher memory requirements')
   parser.disable_interspersed_args()
   options, args = parser.parse_args()
 
   print _GetDefaultConcurrentLinks(is_lto=options.lto)
   return 0
 
 if __name__ == '__main__':
   sys.exit(main())