separate out use_fma, and add Th24 explainer images

arm/build/config/compiler/BUILD.gn

@@ -624,7 +624,7 @@ config("compiler") {
       }
     }
 
-    if (use_avx2 || use_avx512 || current_cpu == "arm64" || is_mac) {
+    if (use_avx2 || use_avx512 || use_fma || current_cpu == "arm64" || is_mac) {
       if (is_win) {
         cflags += [ "/clang:-ffp-contract=fast" ]
       } else {
@@ -1284,14 +1284,19 @@ config("compiler_cpu_abi") {
           "-mavx",
         ]
       }
+      if (use_fma) {
+        cflags += [
+          "-mfma",
+          "-ffp-contract=fast",
+          #"-ffast-math",
+        ]
+      }
       if (use_avx2) {
         cflags += [
           "-mavx2",
-          "-mfma",
           "-mf16c",
           "-mlzcnt",
           "-mbmi2",
-          "-ffp-contract=fast",
           "-mtune=haswell",
         ]
       }
@@ -1335,14 +1340,18 @@ config("compiler_cpu_abi") {
           "-mavx",
         ]
       }
+      if (use_fma) {
+        ldflags += [
+          "-mfma",
+          "-Wl,-mllvm,-fp-contract=fast",
+        ]
+      }
       if (use_avx2) {
         ldflags += [
           "-mavx2",
-          "-mfma",
           "-mf16c",
           "-mlzcnt",
           "-mbmi2",
-          "-Wl,-mllvm,-fp-contract=fast",
           "-Wl,-mllvm,-march=haswell",
         ]
       }
@@ -2596,14 +2605,18 @@ if (is_win) {
       "/clang:-mavx",
     ]
   }
+  if (use_fma) {
+    common_optimize_on_cflags += [
+      "/clang:-mfma",
+      "/clang:-ffp-contract=fast",
+    ]
+  }
   if (use_avx2) {
     common_optimize_on_cflags += [
       "/clang:-mavx2",
-      "/clang:-mfma",
       "/clang:-mf16c",
       "/clang:-mlzcnt",
       "/clang:-mbmi2",
-      "/clang:-ffp-contract=fast",
     ]
   }
   if (use_avx512) {
@@ -2689,14 +2702,18 @@ if (is_win) {
       "-mavx",
     ]
   }
+  if (use_fma) {
+    common_optimize_on_cflags += [
+      "-mfma",
+      "-ffp-contract=fast",
+    ]
+  }
   if (use_avx2) {
     common_optimize_on_cflags += [
       "-mavx2",
-      "-mfma",
       "-mf16c",
       "-mlzcnt",
       "-mbmi2",
-      "-ffp-contract=fast",
     ]
   }
   if (use_avx512) {
@@ -2743,14 +2760,18 @@ if (is_win) {
       "-mavx",
     ]
   }
+  if (use_fma) {
+    common_optimize_on_ldflags += [
+      "-mfma",
+      "-Wl,-mllvm,-fp-contract=fast",
+    ]
+  }
   if (use_avx2) {
     common_optimize_on_ldflags += [
       "-mavx2",
-      "-mfma",
       "-mf16c",
       "-mlzcnt",
       "-mbmi2",
-      "-Wl,-mllvm,-fp-contract=fast",
       "-Wl,-mllvm,-march=haswell",
     ]
   }

src/build/config/compiler/BUILD.gn

@@ -624,7 +624,7 @@ config("compiler") {
       }
     }
 
-    if (use_avx2 || use_avx512 || current_cpu == "arm64" || is_mac) {
+    if (use_avx2 || use_avx512 || use_fma || current_cpu == "arm64" || is_mac) {
       if (is_win) {
         cflags += [ "/clang:-ffp-contract=fast" ]
       } else {
@@ -1284,14 +1284,19 @@ config("compiler_cpu_abi") {
           "-mavx",
         ]
       }
+      if (use_fma) {
+        cflags += [
+          "-mfma",
+          "-ffp-contract=fast",
+          #"-ffast-math",
+        ]
+      }
       if (use_avx2) {
         cflags += [
           "-mavx2",
-          "-mfma",
           "-mf16c",
           "-mlzcnt",
           "-mbmi2",
-          "-ffp-contract=fast",
           "-mtune=haswell",
         ]
       }
@@ -1335,14 +1340,18 @@ config("compiler_cpu_abi") {
           "-mavx",
         ]
       }
+      if (use_fma) {
+        ldflags += [
+          "-mfma",
+          "-Wl,-mllvm,-fp-contract=fast",
+        ]
+      }
       if (use_avx2) {
         ldflags += [
           "-mavx2",
-          "-mfma",
           "-mf16c",
           "-mlzcnt",
           "-mbmi2",
-          "-Wl,-mllvm,-fp-contract=fast",
           "-Wl,-mllvm,-march=haswell",
         ]
       }
@@ -2596,14 +2605,18 @@ if (is_win) {
       "/clang:-mavx",
     ]
   }
+  if (use_fma) {
+    common_optimize_on_cflags += [
+      "/clang:-mfma",
+      "/clang:-ffp-contract=fast",
+    ]
+  }
   if (use_avx2) {
     common_optimize_on_cflags += [
       "/clang:-mavx2",
-      "/clang:-mfma",
       "/clang:-mf16c",
       "/clang:-mlzcnt",
       "/clang:-mbmi2",
-      "/clang:-ffp-contract=fast",
     ]
   }
   if (use_avx512) {
@@ -2689,14 +2702,18 @@ if (is_win) {
       "-mavx",
     ]
   }
+  if (use_fma) {
+    common_optimize_on_cflags += [
+      "-mfma",
+      "-ffp-contract=fast",
+    ]
+  }
   if (use_avx2) {
     common_optimize_on_cflags += [
       "-mavx2",
-      "-mfma",
       "-mf16c",
       "-mlzcnt",
       "-mbmi2",
-      "-ffp-contract=fast",
     ]
   }
   if (use_avx512) {
@@ -2743,14 +2760,18 @@ if (is_win) {
       "-mavx",
     ]
   }
+  if (use_fma) {
+    common_optimize_on_ldflags += [
+      "-mfma",
+      "-Wl,-mllvm,-fp-contract=fast",
+    ]
+  }
   if (use_avx2) {
     common_optimize_on_ldflags += [
       "-mavx2",
-      "-mfma",
       "-mf16c",
       "-mlzcnt",
       "-mbmi2",
-      "-Wl,-mllvm,-fp-contract=fast",
       "-Wl,-mllvm,-march=haswell",
     ]
   }

src/build/config/compiler/pgo/BUILD.gn.bak

@@ -0,0 +1,190 @@
+# Copyright 2024 The Chromium Authors, RobRich999, and Alex313031
+# Use of this source code is governed by a BSD-style license that can be
+# found in the LICENSE file.
+
+import("//build/config/chromeos/ui_mode.gni")
+import("//build/config/clang/clang.gni")
+import("//build/config/compiler_opt.gni")
+import("//build/config/compiler/compiler.gni")
+import("//build/config/compiler/pgo/pgo.gni")
+import("//build/config/features.gni")
+import("//build/toolchain/toolchain.gni")
+
+# Configuration that enables PGO instrumentation.
+config("pgo_instrumentation_flags") {
+  visibility = [ ":default_pgo_flags" ]
+
+  # Only add flags when chrome_pgo_phase == 1, so that variables we would use
+  # are not required to be defined when we're not actually using PGO.
+  if (chrome_pgo_phase == 1 && is_clang && !is_nacl && is_a_target_toolchain) {
+    cflags = [ "-fprofile-generate" ]
+    if (temporal_pgo_profile) {
+      cflags += [
+        "-mllvm",
+        "-pgo-temporal-instrumentation",
+      ]
+    }
+    if (!is_win) {
+      # Windows directly calls link.exe instead of the compiler driver when
+      # linking, and embeds the path to the profile runtime library as
+      # dependent library into each object file.
+      ldflags = [ "-fprofile-generate" ]
+    }
+  }
+}
+
+# Configuration that enables optimization using profile data.
+config("pgo_optimization_flags") {
+  visibility = [ ":default_pgo_flags" ]
+
+  # Only add flags when chrome_pgo_phase == 2, so that variables we would use
+  # are not required to be defined when we're not actually using PGO.
+  if (chrome_pgo_phase == 2 && is_clang && !is_nacl && is_a_target_toolchain) {
+    _pgo_target = ""
+
+    # There are txt files used by //tools/update_pgo_profiles.py to decide which
+    # profiles to use, adding them as inputs so that analyzer recognizes the
+    # dependencies.
+    inputs = []
+
+    if (is_win) {
+      if (target_cpu == "arm64") {
+        _pgo_target = "win-arm64"
+      } else if (target_cpu == "x64") {
+        _pgo_target = "win64"
+      } else {
+        _pgo_target = "win32"
+      }
+    } else if (is_mac) {
+      if (target_cpu == "arm64") {
+        _pgo_target = "mac-arm"
+      } else {
+        _pgo_target = "mac"
+      }
+    } else if (is_linux) {
+      _pgo_target = "linux"
+    } else if (is_chromeos_lacros) {
+      if (target_cpu == "arm") {
+        # We don't have the arm device to train arm pgo data. So Lacros arm
+        # would use arm64 profile.
+        _pgo_target = "lacros-arm64"
+      } else if (target_cpu == "arm64") {
+        _pgo_target = "lacros-arm64"
+      } else {
+        _pgo_target = "lacros64"
+      }
+    } else if (is_android) {
+      # Use |current_cpu| and not |target_cpu|; for Android we may built both.
+      if (current_cpu == "arm64") {
+        _pgo_target = "android-arm64"
+      } else {
+        _pgo_target = "android-arm32"
+      }
+    } else if (is_fuchsia) {
+      if (target_cpu == "arm64") {
+        _pgo_target = "mac-arm"
+      } else {
+        _pgo_target = "mac"
+      }
+    } else if (is_ios && use_blink) {
+      if (target_cpu == "arm64") {
+        _pgo_target = "mac-arm"
+      } else {
+        _pgo_target = "mac"
+      }
+    }
+
+    if (_pgo_target == "win-arm64") {
+      inputs = [ "//chrome/build/win-arm64.pgo.txt" ]
+    } else if (_pgo_target == "win64") {
+      inputs = [ "//chrome/build/win64.pgo.txt" ]
+    } else if (_pgo_target == "win32") {
+      inputs = [ "//chrome/build/win32.pgo.txt" ]
+    } else if (_pgo_target == "mac-arm") {
+      inputs = [ "//chrome/build/mac-arm.pgo.txt" ]
+    } else if (_pgo_target == "mac") {
+      inputs = [ "//chrome/build/mac.pgo.txt" ]
+    } else if (_pgo_target == "linux") {
+      inputs = [ "//chrome/build/linux.pgo.txt" ]
+    } else if (_pgo_target == "lacros64") {
+      inputs = [ "//chrome/build/lacros64.pgo.txt" ]
+    } else if (_pgo_target == "lacros-arm") {
+      inputs = [ "//chrome/build/lacros-arm.pgo.txt" ]
+    } else if (_pgo_target == "lacros-arm64") {
+      inputs = [ "//chrome/build/lacros-arm64.pgo.txt" ]
+    } else if (_pgo_target == "android-arm32") {
+      inputs = [ "//chrome/build/android-arm32.pgo.txt" ]
+    } else if (_pgo_target == "android-arm64") {
+      inputs = [ "//chrome/build/android-arm64.pgo.txt" ]
+    }
+
+    if (_pgo_target != "" && pgo_data_path == "") {
+      pgo_data_path = exec_script("//tools/update_pgo_profiles.py",
+                                  [
+                                    "--target",
+                                    _pgo_target,
+                                    "get_profile_path",
+                                  ],
+                                  "value")
+    }
+    assert(pgo_data_path != "",
+           "Please set pgo_data_path to point at the profile data")
+    cflags = [
+      "-fprofile-use=" + rebase_path(pgo_data_path, root_build_dir),
+
+      # It's possible to have some profile data legitimately missing,
+      # and at least some profile data always ends up being considered
+      # out of date, so make sure we don't error for those cases.
+      "-Wno-profile-instr-unprofiled",
+      "-Wno-profile-instr-out-of-date",
+
+      # Some hashing conflict results in a lot of warning like this when doing
+      # a PGO build:
+      #   warning: foo.cc: Function control flow change detected (hash mismatch)
+      #   [-Wbackend-plugin]
+      # See https://crbug.com/978401
+      "-Wno-backend-plugin",
+    ]
+
+    # Enable basic block layout based on the extended TSP problem. This aims to
+    # improve icache utilization and reduce the binary size.
+    if (use_thin_lto) {
+      if (current_cpu == "x64" || current_cpu == "x86") {
+        if (is_win) {
+          ldflags = [ "-mllvm:-enable-ext-tsp-block-placement=1" ]
+        } else if (is_linux) {
+          ldflags = [
+            "-Wl,-mllvm,-enable-ext-tsp-block-placement=1",
+            "-Wl,-mllvm,-enable-split-machine-functions"
+          ]
+        } else {
+          ldflags = [ "-Wl,-mllvm,-enable-ext-tsp-block-placement=1" ]
+        }
+      } else {
+        if (is_win) {
+          ldflags = [ "-mllvm:-enable-ext-tsp-block-placement=1" ]
+        } else {
+          ldflags = [ "-Wl,-mllvm,-enable-ext-tsp-block-placement=1" ]
+        }
+      }
+    } else {
+      cflags += [
+        "-mllvm",
+        "-enable-ext-tsp-block-placement=1",
+      ]
+    }
+  }
+}
+
+# Applies flags necessary when profile-guided optimization is used.
+# Flags are only added if PGO is enabled, so that this config is safe to
+# include by default.
+config("default_pgo_flags") {
+  if (chrome_pgo_phase == 0) {
+    # Nothing. This config should be a no-op when chrome_pgo_phase == 0.
+  } else if (chrome_pgo_phase == 1) {
+    configs = [ ":pgo_instrumentation_flags" ]
+  } else if (chrome_pgo_phase == 2) {
+    configs = [ ":pgo_optimization_flags" ]
+  }
+}