build,win: enable x64 PGO

PR-URL: #62761
Refs: #61964
Reviewed-By: Xuguang Mei <meixuguang@gmail.com>
Reviewed-By: Luigi Pinca <luigipinca@gmail.com>
Reviewed-By: Tobias Nießen <tniessen@tnie.de>

Author: StefanStojanovic

common.gypi

@@ -12,6 +12,7 @@
     'msvs_multi_core_compile': '0',   # we do enable multicore compiles, but not using the V8 way
     'enable_pgo_generate%': '0',
     'enable_pgo_use%': '0',
+    'clang_profile_lib%': '',
     'python%': 'python',
 
     'node_shared%': 'false',
@@ -243,6 +244,65 @@
               },],
             ],
           },],
+          ['OS=="win"', {
+            'conditions': [
+              ['enable_lto=="true"', {
+                'msvs_settings': {
+                  'VCCLCompilerTool': {
+                    'AdditionalOptions': ['-flto=full'],
+                  },
+                  'VCLibrarianTool': {
+                    'AdditionalOptions': ['-flto=full'],
+                  },
+                  'VCLinkerTool': {
+                    'AdditionalOptions': ['-flto=full'],
+                  },
+                },
+              },],
+              ['enable_thin_lto=="true"', {
+                'msvs_settings': {
+                  'VCCLCompilerTool': {
+                    'AdditionalOptions': ['-flto=thin'],
+                  },
+                  'VCLibrarianTool': {
+                    'AdditionalOptions': ['-flto=thin'],
+                  },
+                  'VCLinkerTool': {
+                    'AdditionalOptions': ['-flto=thin'],
+                  },
+                },
+              },],
+            ],
+            'target_conditions': [
+              ['_toolset=="target"', {
+                'conditions': [
+                  ['enable_pgo_generate=="true"', {
+                    'msvs_settings': {
+                      'VCCLCompilerTool': {
+                        'AdditionalOptions': ['-fprofile-generate'],
+                      },
+                      'VCLinkerTool': {
+                        'AdditionalOptions': [
+                          '/NODEFAULTLIB:clang_rt.profile.lib',
+                          '"<(clang_profile_lib)"',
+                        ],
+                      },
+                    },
+                  },],
+                  ['enable_pgo_use=="true"', {
+                    'msvs_settings': {
+                      'VCCLCompilerTool': {
+                        'AdditionalOptions': ['-fprofile-use=$(SolutionDir)node.profdata'],
+                      },
+                      'VCLinkerTool': {
+                        'AdditionalOptions': ['-fprofile-use=$(SolutionDir)node.profdata'],
+                      },
+                    },
+                  },],
+                ],
+              },],
+            ],
+          },],
           ['OS == "android"', {
             'cflags': [ '-fPIC', '-I<(android_ndk_path)/sources/android/cpufeatures' ],
             'ldflags': [ '-fPIC' ]

configure.py

@@ -202,14 +202,14 @@
     dest="enable_pgo_generate",
     default=None,
     help="Enable profiling with pgo of a binary. This feature is only available "
-         "on linux with gcc and g++ 5.4.1 or newer.")
+         "on linux with gcc and g++ 5.4.1 or newer and on windows.")
 
 parser.add_argument("--enable-pgo-use",
     action="store_true",
     dest="enable_pgo_use",
     default=None,
     help="Enable use of the profile generated with --enable-pgo-generate. This "
-         "feature is only available on linux with gcc and g++ 5.4.1 or newer.")
+         "feature is only available on linux with gcc and g++ 5.4.1 or newer and on windows.")
 
 parser.add_argument("--enable-lto",
     action="store_true",
@@ -218,6 +218,13 @@
     help="Enable compiling with lto of a binary. This feature is only available "
          "with gcc 5.4.1+ or clang 3.9.1+.")
 
+parser.add_argument("--enable-thin-lto",
+    action="store_true",
+    dest="enable_thin_lto",
+    default=None,
+    help="Enable compiling with thin lto of a binary. This feature is only available "
+         "on windows.")
+
 parser.add_argument("--link-module",
     action="append",
     dest="linked_module",
@@ -925,7 +932,8 @@
     action='store_true',
     dest='with_ltcg',
     default=None,
-    help='Use Link Time Code Generation. This feature is only available on Windows.')
+    help='Use Thin LTO scoped to node.exe and libnode only. '
+         'This feature is only available on Windows.')
 
 parser.add_argument('--write-snapshot-as-array-literals',
     action='store_true',
@@ -1920,9 +1928,9 @@ def configure_node(o):
   else:
     o['variables']['node_enable_v8_vtunejit'] = 'false'
 
-  if flavor != 'linux' and (options.enable_pgo_generate or options.enable_pgo_use):
+  if (flavor != 'linux' and flavor != 'win') and (options.enable_pgo_generate or options.enable_pgo_use):
     raise Exception(
-      'The pgo option is supported only on linux.')
+      'The pgo option is supported only on linux and windows.')
 
   if flavor == 'linux':
     if options.enable_pgo_generate or options.enable_pgo_use:
@@ -1933,21 +1941,55 @@ def configure_node(o):
           'The options --enable-pgo-generate and --enable-pgo-use '
           f'are supported for gcc and gxx {version_checked_str} or newer only.')
 
-    if options.enable_pgo_generate and options.enable_pgo_use:
-      raise Exception(
-        'Only one of the --enable-pgo-generate or --enable-pgo-use options '
-        'can be specified at a time. You would like to use '
-        '--enable-pgo-generate first, profile node, and then recompile '
-        'with --enable-pgo-use')
+  if options.enable_pgo_generate and options.enable_pgo_use:
+    raise Exception(
+      'Only one of the --enable-pgo-generate or --enable-pgo-use options '
+      'can be specified at a time. You would like to use '
+      '--enable-pgo-generate first, profile node, and then recompile '
+      'with --enable-pgo-use')
 
   o['variables']['enable_pgo_generate'] = b(options.enable_pgo_generate)
   o['variables']['enable_pgo_use']      = b(options.enable_pgo_use)
 
-  if flavor == 'win' and (options.enable_lto):
+  if flavor == 'win' and (options.enable_pgo_generate or options.enable_pgo_use):
+    lib_suffix = 'aarch64' if target_arch == 'arm64' else 'x86_64'
+    lib_name = f'clang_rt.profile-{lib_suffix}.lib'
+    msvc_dir = target_arch  # 'x64' or 'arm64'
+
+    vc_tools_dir = os.environ.get('VCToolsInstallDir', '')
+    if vc_tools_dir:
+      clang_profile_lib = os.path.join(vc_tools_dir, 'lib', msvc_dir, lib_name)
+      if os.path.isfile(clang_profile_lib):
+        o['variables']['clang_profile_lib'] = clang_profile_lib
+      else:
+        raise Exception(
+          f'PGO profile runtime library not found at {clang_profile_lib}. '
+          'Ensure the ClangCL toolset is installed.')
+    else:
+      raise Exception(
+        'VCToolsInstallDir not set. Run from a Visual Studio command prompt.')
+
+  if flavor != 'win' and options.enable_thin_lto:
     raise Exception(
-      'Use Link Time Code Generation instead.')
+      'Use --enable-lto instead.')
+
+  # LTO mutual exclusion
+  if flavor == 'win':
+    lto_options = []
+    if options.enable_lto:
+      lto_options.append('--enable-lto')
+    if options.enable_thin_lto:
+      lto_options.append('--enable-thin-lto')
+    if options.with_ltcg:
+      lto_options.append('--with-ltcg')
+    if len(lto_options) > 1:
+      raise Exception(
+        f'Only one LTO option can be specified at a time: {", ".join(lto_options)}. '
+        'Use --enable-lto for Full LTO (global), '
+        '--enable-thin-lto for Thin LTO (global), '
+        'or --with-ltcg for Thin LTO (scoped to node.exe and libnode).')
 
-  if options.enable_lto:
+  if options.enable_lto and flavor != 'win':
     gcc_version_checked = (5, 4, 1)
     clang_version_checked = (3, 9, 1)
     if not gcc_version_ge(gcc_version_checked) and not clang_version_ge(clang_version_checked):
@@ -1958,6 +2000,7 @@ def configure_node(o):
         f'or clang {clang_version_checked_str}+ only.')
 
   o['variables']['enable_lto'] = b(options.enable_lto)
+  o['variables']['enable_thin_lto'] = b(options.enable_thin_lto)
 
   if options.node_use_large_pages or options.node_use_large_pages_script_lld:
     warn('''The `--use-largepages` and `--use-largepages-script-lld` options

deps/openssl/openssl-cli.gypi

@@ -25,5 +25,15 @@
     ['enable_lto=="true"', {
       'ldflags': [ '-fno-lto' ],
     }],
+    ['node_with_ltcg=="true" or enable_lto=="true" or enable_thin_lto=="true"', {
+      'msvs_settings': {
+        'VCCLCompilerTool': {
+          'AdditionalOptions': ['-fno-lto'],
+        },
+        'VCLinkerTool': {
+          'AdditionalOptions': ['-fno-lto'],
+        },
+      },
+    }],
   ],
 }

deps/openssl/openssl.gyp

@@ -77,6 +77,16 @@
         ['enable_lto=="true"', {
           'ldflags': [ '-fno-lto' ],
         }],
+        ['node_with_ltcg=="true" or enable_lto=="true" or enable_thin_lto=="true"', {
+          'msvs_settings': {
+            'VCCLCompilerTool': {
+              'AdditionalOptions': ['-fno-lto'],
+            },
+            'VCLinkerTool': {
+              'AdditionalOptions': ['-fno-lto'],
+            },
+          },
+        }],
      ]
     }, {
       # openssl-fipsmodule target

node.gyp

@@ -730,35 +730,34 @@
             'Ws2_32.lib',
           ],
         }],
+        # Thin LTO for node_main.cc and linker (scoped to node_exe)
         ['node_with_ltcg=="true"', {
           'msvs_settings': {
             'VCCLCompilerTool': {
-              'WholeProgramOptimization': 'true'   # /GL, whole program optimization, needed for LTCG
+              'AdditionalOptions': ['-flto=thin'],
             },
-            'VCLibrarianTool': {
-              'AdditionalOptions': [
-                '/LTCG:INCREMENTAL',               # link time code generation
-              ],
+            'VCLinkerTool': {
+              'AdditionalOptions': ['-flto=thin'],
             },
+          },
+        }],
+        # Whole-program optimization: either Thin LTO or PGO
+        ['node_with_ltcg=="true" or enable_lto=="true" or enable_thin_lto=="true" or enable_pgo_generate=="true" or enable_pgo_use=="true"', {
+          'msvs_settings': {
             'VCLinkerTool': {
               'OptimizeReferences': 2,             # /OPT:REF
               'EnableCOMDATFolding': 2,            # /OPT:ICF
               'LinkIncremental': 1,                # disable incremental linking
-              'AdditionalOptions': [
-                '/LTCG:INCREMENTAL',               # incremental link-time code generation
-              ],
-            }
-          }
+            },
+          },
         }, {
+          # No whole-program optimization
           'msvs_settings': {
-            'VCCLCompilerTool': {
-              'WholeProgramOptimization': 'false'
-            },
             'VCLinkerTool': {
-              'LinkIncremental': 2                 # enable incremental linking
+              'LinkIncremental': 2,                # enable incremental linking
             },
           },
-         }],
+        }],
          ['node_use_node_snapshot=="true"', {
           'dependencies': [
             'node_mksnapshot',
@@ -1139,6 +1138,17 @@
         [ 'debug_nghttp2==1', {
           'defines': [ 'NODE_DEBUG_NGHTTP2=1' ]
         }],
+        # Thin LTO for node sources (scoped to libnode, not global)
+        ['node_with_ltcg=="true"', {
+          'msvs_settings': {
+            'VCCLCompilerTool': {
+              'AdditionalOptions': ['-flto=thin'],
+            },
+            'VCLibrarianTool': {
+              'AdditionalOptions': ['-flto=thin'],
+            },
+          },
+        }],
       ],
       'actions': [
         {
@@ -1439,6 +1449,16 @@
         ['enable_lto=="true"', {
           'ldflags': [ '-fno-lto' ],
         }],
+        ['node_with_ltcg=="true" or enable_lto=="true" or enable_thin_lto=="true"', {
+          'msvs_settings': {
+            'VCCLCompilerTool': {
+              'AdditionalOptions': ['-fno-lto'],
+            },
+            'VCLinkerTool': {
+              'AdditionalOptions': ['-fno-lto'],
+            },
+          },
+        }],
       ],
     }, # cctest
 
@@ -1503,6 +1523,16 @@
         ['enable_lto=="true"', {
           'ldflags': [ '-fno-lto' ],
         }],
+        ['node_with_ltcg=="true" or enable_lto=="true" or enable_thin_lto=="true"', {
+          'msvs_settings': {
+            'VCCLCompilerTool': {
+              'AdditionalOptions': ['-fno-lto'],
+            },
+            'VCLinkerTool': {
+              'AdditionalOptions': ['-fno-lto'],
+            },
+          },
+        }],
       ],
     }, # embedtest
 
@@ -1580,6 +1610,16 @@
         ['enable_lto=="true"', {
           'ldflags': [ '-fno-lto' ],
         }],
+        ['node_with_ltcg=="true" or enable_lto=="true" or enable_thin_lto=="true"', {
+          'msvs_settings': {
+            'VCCLCompilerTool': {
+              'AdditionalOptions': ['-fno-lto'],
+            },
+            'VCLinkerTool': {
+              'AdditionalOptions': ['-fno-lto'],
+            },
+          },
+        }],
       ]
     }, # overlapped-checker
     {
@@ -1706,6 +1746,16 @@
         ['enable_lto=="true"', {
           'ldflags': [ '-fno-lto' ],
         }],
+        ['node_with_ltcg=="true" or enable_lto=="true" or enable_thin_lto=="true"', {
+          'msvs_settings': {
+            'VCCLCompilerTool': {
+              'AdditionalOptions': ['-fno-lto'],
+            },
+            'VCLinkerTool': {
+              'AdditionalOptions': ['-fno-lto'],
+            },
+          },
+        }],
       ],
     }, # node_mksnapshot
   ], # end targets