[InstrProfiling] Use llvm.compiler.used if applicable for Mach-O

[MaskRay]

Summary

Similar to D97585.

D25456 used S_ATTR_LIVE_SUPPORT to ensure the data variable will be retained
or discarded as a unit with the counter variable, so llvm.compiler.used is
sufficient. It allows ld to dead strip unneeded profc and profd variables.

[MaskRay]

-Wl,-dead_strip

Tested on aarch64 macOS.

[vsk]

Thanks for tackling this.

compiler-rt/test/profile/Darwin/coverage-linkage.cpp

10

The behavior of linker optimizations with LTO enabled may be a bit different. Could you check this case (test/profile/instrprof-darwin-dead-strip.c has an example)?

[MaskRay]

I don't have access to an x86_64 macOS.

If I remove REQUIRES: lto from test/profile/instrprof-darwin-dead-strip.c, it fails on aarch64 macOS before this patch.

But this patch doesn't change the section contents:

Contents of (__DATA,__llvm_prf_names) section
0000000100018030 6f660008 616d016f 69 6e

Contents of (__DATA,__llvm_prf_cnts) section
0000000100014000        00000000 00000000

[vsk]

x86 access shouldn't be necessary. Here's a patch for the otool formatting issue: https://reviews.llvm.org/D108929 (feel free to fold this in to your own change, if you'd like to).

[MaskRay]

Does it look good? :)

[vsk]

Yes, thanks! The test from D108929 should cover the lto case.

compiler-rt/test/profile/Darwin/coverage-linkage.cpp

+/// Test instrumentation can handle various linkages.
+// RUN: %clang_profgen -fcoverage-mapping %s -o %t
+// RUN: env LLVM_PROFILE_FILE=%t.profraw %run %t
+// RUN: llvm-profdata show %t.profraw --all-functions | FileCheck %s
+
+// RUN: %clang_profgen -fcoverage-mapping -Wl,-dead_strip %s -o %t
+// RUN: env LLVM_PROFILE_FILE=%t.profraw %run %t
+// RUN: llvm-profdata show %t.profraw --all-functions | FileCheck %s
+
+// CHECK:      {{.*}}external{{.*}}:
+// CHECK-NEXT:    Hash:
+// CHECK-NEXT:    Counters: 1
+// CHECK-NEXT:    Function count: 1
+// CHECK:      {{.*}}weak{{.*}}:
+// CHECK-NEXT:    Hash:
+// CHECK-NEXT:    Counters: 1
+// CHECK-NEXT:    Function count: 1
+// CHECK:      main:
+// CHECK-NEXT:    Hash:
+// CHECK-NEXT:    Counters: 1
+// CHECK-NEXT:    Function count: 1
+// CHECK:      {{.*}}internal{{.*}}:
+// CHECK-NEXT:    Hash:
+// CHECK-NEXT:    Counters: 1
+// CHECK-NEXT:    Function count: 1
+// CHECK:      {{.*}}linkonce_odr{{.*}}:
+// CHECK-NEXT:    Hash:
+// CHECK-NEXT:    Counters: 1
+// CHECK-NEXT:    Function count: 1
+
+#include <stdio.h>
+
+void discarded0() {}
+__attribute__((weak)) void discarded1() {}
+
+void external() { puts("external"); }
+__attribute__((weak)) void weak() { puts("weak"); }
+static void internal() { puts("internal"); }
+__attribute__((noinline)) inline void linkonce_odr() { puts("linkonce_odr"); }
+
+int main() {
+  internal();
+  external();
+  weak();
+  linkonce_odr();
+}

compiler-rt/test/profile/Darwin/lit.local.cfg.py

+def getRoot(config):
+  if not config.parent:
+    return config
+  return getRoot(config.parent)
+
+root = getRoot(config)
+
+if root.host_os not in ['Darwin']:
+  config.unsupported = True

llvm/lib/Transforms/Instrumentation/InstrProfiling.cpp

bool InstrProfiling::emitRuntimeHook() {
   return true;
 }
 
 void InstrProfiling::emitUses() {
   // The metadata sections are parallel arrays. Optimizers (e.g.
   // GlobalOpt/ConstantMerge) may not discard associated sections as a unit, so
   // we conservatively retain all unconditionally in the compiler.
   //
-  // On ELF, the linker can guarantee the associated sections will be retained
-  // or discarded as a unit, so llvm.compiler.used is sufficient. Similarly on
-  // COFF, if prof data is not referenced by code we use one comdat and ensure
-  // this GC property as well. Otherwise, we have to conservatively make all of
-  // the sections retained by the linker.
-  if (TT.isOSBinFormatELF() ||
+  // On ELF and Mach-O, the linker can guarantee the associated sections will be
+  // retained or discarded as a unit, so llvm.compiler.used is sufficient.
+  // Similarly on COFF, if prof data is not referenced by code we use one comdat
+  // and ensure this GC property as well. Otherwise, we have to conservatively
+  // make all of the sections retained by the linker.
+  if (TT.isOSBinFormatELF() || TT.isOSBinFormatMachO() ||
       (TT.isOSBinFormatCOFF() && !profDataReferencedByCode(*M)))
     appendToCompilerUsed(*M, CompilerUsedVars);
   else
     appendToUsed(*M, CompilerUsedVars);
 
   // We do not add proper references from used metadata sections to NamesVar and
   // VNodesVar, so we have to be conservative and place them in llvm.used
   // regardless of the target,

llvm/test/Instrumentation/InstrProfiling/profiling.ll

define available_externally void @foo_extern() {
   call void @llvm.instrprof.increment(i8* getelementptr inbounds ([10 x i8], [10 x i8]* @__profn_foo_extern, i32 0, i32 0), i64 0, i32 1, i32 0)
   ret void
 }
 
 declare void @llvm.instrprof.increment(i8*, i64, i32, i32)
 
 ; ELF:         @llvm.compiler.used = appending global {{.*}} @__profd_foo {{.*}}
 ; ELF_GENERIC: @llvm.compiler.used = appending global {{.*}} @__llvm_profile_runtime {{.*}} @__profd_foo {{.*}}
-; MACHO:       @llvm.used = appending global {{.*}} @__llvm_profile_runtime_user {{.*}} @__profd_foo {{.*}}
+; MACHO:       @llvm.compiler.used = appending global {{.*}} @__llvm_profile_runtime_user {{.*}} @__profd_foo {{.*}}
 ; COFF:        @llvm.compiler.used = appending global {{.*}} @__llvm_profile_runtime_user {{.*}} @__profd_foo {{.*}}
 
 ; MACHO: define linkonce_odr hidden i32 @__llvm_profile_runtime_user() {{.*}} {
 ; MACHO:   %[[REG:.*]] = load i32, i32* @__llvm_profile_runtime
 ; MACHO:   ret i32 %[[REG]]
 ; MACHO: }
 ; COFF: define linkonce_odr hidden i32 @__llvm_profile_runtime_user() {{.*}} comdat {
 ; ELF-NOT: define linkonce_odr hidden i32 @__llvm_profile_runtime_user() {{.*}} {