Use --icf=all instead of --icf=safe

With --icf=safe, functions whose address is taken aren't folded.
Folding could change the results of function pointer comparisons.

However, on x86_64 with pie, --icf=safe disables almost all folding,
for technical reasons. Since using --icf=all will save 7 MB on x86_64
and all tests seem to pass with it, switch to --icf=all. If we end up
having a function where we absolutely need equality, we can pass
--keep-unique,<symbol-name> for that symbol.

And if this turns out to be some maintenance burden, we can switch
back to --icf=safe.

Change the flag on all platforms where we use gold, not just x86_64, for
consistency so that if we end up having problems with this, we notice
it faster.

BUG=76280,492177

Review URL: https://codereview.chromium.org/1153143004

Cr-Commit-Position: refs/heads/master@{#331474}

2 files changed, 3 insertions, 3 deletions

diff --git a/build/common.gypi b/build/common.gypi
index 9c30c4e..227e82f 100644
--- a/build/common.gypi
+++ b/build/common.gypi
@@ -665,7 +665,7 @@
       'use_lto_o2%': 0,
 
       # Allowed level of identical code folding in the gold linker.
-      'gold_icf_level%': 'safe',
+      'gold_icf_level%': 'all',
 
       # Libxkbcommon usage.
       'use_xkbcommon%': 0,
diff --git a/build/config/compiler/BUILD.gn b/build/config/compiler/BUILD.gn
index ff20eb6..23515a9 100644
--- a/build/config/compiler/BUILD.gn
+++ b/build/config/compiler/BUILD.gn
@@ -423,7 +423,7 @@ config("compiler") {
       "-fuse-ld=gold",
 
       # TODO(brettw) common.gypi has this only for target toolset.
-      "-Wl,--icf=safe",
+      "-Wl,--icf=all",
 
       # Experimentation found that using four linking threads
       # saved ~20% of link time.
@@ -520,7 +520,7 @@ config("compiler") {
     if (current_cpu == "arm") {
       ldflags += [
         # Enable identical code folding to reduce size.
-        "-Wl,--icf=safe",
+        "-Wl,--icf=all",
       ]
     }