[AIX][PGO] Enable linux style PGO on AIX

[w2yehia]

Summary

This patch switches the PGO implementation on AIX from using the runtime registration-based section tracking to the __start_SECNAME/__stop_SECNAME based.
In order to enable the recognition of __start_SECNAME/__stop_SECNAME symbols in the AIX linker, the -bdbg:namedsects:ss needs to be used.

[jsji]

compiler-rt/lib/profile/InstrProfilingPlatformLinux.c

11

What is the major reason that we want to share the file with Linux? Would it be better to have AIX on its own file?

[MaskRay]

Forking InstrProfilingPlatformLinux.c looks good to me. The misnomer *Linux.c file works with a number of ELF operating systems. Developers know that one feature working on one OS will likely work on others.

[w2yehia]

compiler-rt/lib/profile/InstrProfilingPlatformLinux.c

11

The main reason is the non-trivial amount of code duplication we will have if we split the AIX into its own file.

[jsji]

LGTM. Thanks.

[davidxl]

compiler-rt/lib/profile/InstrProfilingPlatformLinux.c

235

Why is this needed? For objects compiled with older compiler?

[w2yehia]

compiler-rt/lib/profile/InstrProfilingPlatformLinux.c

235

For objects compiled with older compiler?

Yes. On AIX, we would like to allow old object files to be linkable (and function correctly) with the new compiler. On AIX, old object files really mean object files compiled with LLVM 13.0 or newer.

[w2yehia]

Add dummy variables to allow linking in general against the profile-rt on AIX in the default linking model (-bcdtos:all).

[w2yehia]

@davidxl @MaskRay @jsji FYI, please review the addition on line 242 of compiler-rt/lib/profile/InstrProfilingPlatformLinux.c

[davidxl]

lgtm

[MaskRay]

compiler-rt/lib/profile/InstrProfilingPlatformLinux.c

253

This has a -Wzero-length-array warning in -pedantic mode.

llvm/lib/Transforms/Instrumentation/InstrProfiling.cpp

860

You may place AIX before others for a relatively alphabetical order.

[w2yehia]

Address one comment.

[w2yehia]

compiler-rt/lib/profile/InstrProfilingPlatformLinux.c

253

sorry but what's the correct way to resolve this on all platforms?
I see -Wno-pedantic is used when building this file if(FUCHSIA OR UNIX). Should I add -Wno-pedantic if such flag is supported by the build compiler in general?

[MaskRay]

compiler-rt/lib/profile/InstrProfilingPlatformLinux.c

253

Many people don't build AIX so this isn't a problem for them (most build bots).

If you want suppress the warning for the AIX specific code, you may use

#pragma GCC diagnostic ignored "-Wpedantic"

This works with both GCC and Clang. I do not know how IBM XL/C behaves.

[w2yehia]

compiler-rt/lib/profile/InstrProfilingPlatformLinux.c

253

Right, I forgot this is guarded by the _AIX macro.
Since we already build with -Wno-pedantic on AIX, I don't need to suppress anything, right?
Here's the code in cmake that adds it. AIX is UNIX.

compiler-rt/lib/profile/CMakeLists.txt:
     89 if(FUCHSIA OR UNIX)
     90  set(EXTRA_FLAGS
     91      -fPIC
     92      -Wno-pedantic)
     93 endif()

[MaskRay]

compiler-rt/lib/profile/InstrProfilingPlatformLinux.c

253

If you will not add -Wpedantic in the future, landing just the current form LGTM.

clang/lib/Driver/ToolChains/AIX.cpp

if (Args.hasFlag(options::OPT_fprofile_arcs, options::OPT_fno_profile_arcs,
        Args.hasFlag(options::OPT_fprofile_instr_generate_EQ,
                     options::OPT_fno_profile_instr_generate, false) ||
        Args.hasFlag(options::OPT_fcs_profile_generate,
                     options::OPT_fno_profile_generate, false) ||
        Args.hasFlag(options::OPT_fcs_profile_generate_EQ,
                     options::OPT_fno_profile_generate, false) ||
        Args.hasArg(options::OPT_fcreate_profile) ||
        Args.hasArg(options::OPT_coverage))
-    CmdArgs.push_back("-bdbg:namedsects");
+    CmdArgs.push_back("-bdbg:namedsects:ss");
 
   // Specify linker output file.
   assert((Output.isFilename() || Output.isNothing()) && "Invalid output.");
   if (Output.isFilename()) {
     CmdArgs.push_back("-o");
     CmdArgs.push_back(Output.getFilename());
   }
 

clang/test/Driver/aix-ld.c

 // RUN:        --sysroot %S/Inputs/aix_ppc_tree \
 // RUN:        --unwindlib=libunwind \
 // RUN:   | FileCheck --check-prefix=CHECK-PGO-NON-LTO %s
 // CHECK-PGO-NON-LTO-NOT: warning:
 // CHECK-PGO-NON-LTO:     "-cc1" "-triple" "powerpc-ibm-aix7.3.0.0"
 // CHECK-PGO-NON-LTO:     "-resource-dir" "[[RESOURCE_DIR:[^"]+]]"
 // CHECK-PGO-NON-LTO:     "-isysroot" "[[SYSROOT:[^"]+]]"
 // CHECK-PGO-NON-LTO:     "{{.*}}ld{{(.exe)?}}"
-// CHECK-PGO-NON-LTO:     "-bdbg:namedsects"
+// CHECK-PGO-NON-LTO:     "-bdbg:namedsects:ss"
 // CHECK-PGO-NON-LTO:     "-b32"
 // CHECK-PGO-NON-LTO:     "-bpT:0x10000000" "-bpD:0x20000000"
 // CHECK-PGO-NON-LTO:     "[[SYSROOT]]/usr/lib{{/|\\\\}}crt0.o"
 // CHECK-PGO-NON-LTO-NOT: "-lc++"
 // CHECK-PGO-NON-LTO-NOT: "-lc++abi"
 // CHECK-PGO-NON-LTO:     "[[RESOURCE_DIR]]{{/|\\\\}}lib{{/|\\\\}}aix{{/|\\\\}}libclang_rt.builtins-powerpc.a"
 // CHECK-PGO-NON-LTO-NOT: "--as-needed"
 // CHECK-PGO-NON-LTO-NOT: "-lunwind"

 // RUN:        --sysroot %S/Inputs/aix_ppc_tree \
 // RUN:        --unwindlib=libunwind \
 // RUN:   | FileCheck --check-prefix=CHECK-PGO-LTO %s
 // CHECK-PGO-LTO-NOT: warning:
 // CHECK-PGO-LTO:     "-cc1" "-triple" "powerpc-ibm-aix7.2.5.3"
 // CHECK-PGO-LTO:     "-resource-dir" "[[RESOURCE_DIR:[^"]+]]"
 // CHECK-PGO-LTO:     "-isysroot" "[[SYSROOT:[^"]+]]"
 // CHECK-PGO-LTO:     "{{.*}}ld{{(.exe)?}}"
-// CHECK-PGO-LTO:     "-bdbg:namedsects"
+// CHECK-PGO-LTO:     "-bdbg:namedsects:ss"
 // CHECK-PGO-LTO:     "-b32"
 // CHECK-PGO-LTO:     "-bpT:0x10000000" "-bpD:0x20000000"
 // CHECK-PGO-LTO:     "[[SYSROOT]]/usr/lib{{/|\\\\}}crt0.o"
 // CHECK-PGO-LTO-NOT: "-lc++"
 // CHECK-PGO-LTO-NOT: "-lc++abi"
 // CHECK-PGO-LTO:     "[[RESOURCE_DIR]]{{/|\\\\}}lib{{/|\\\\}}aix{{/|\\\\}}libclang_rt.builtins-powerpc.a"
 // CHECK-PGO-LTO-NOT: "--as-needed"
 // CHECK-PGO-LTO-NOT: "-lunwind"
 // CHECK-PGO-LTO-NOT: "--no-as-needed"
 // CHECK-PGO-LTO-NOT: "-lm"
 // CHECK-PGO-LTO:     "-lc"

compiler-rt/lib/profile/InstrProfilingPlatformLinux.c

 /*===- InstrProfilingPlatformLinux.c - Profile data Linux platform ------===*\
 |*
 |* Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
 |* See https://llvm.org/LICENSE.txt for license information.
 |* SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
 |*
 \*===----------------------------------------------------------------------===*/
 
 #if defined(__linux__) || defined(__FreeBSD__) || defined(__Fuchsia__) || \
-    (defined(__sun__) && defined(__svr4__)) || defined(__NetBSD__)
+    (defined(__sun__) && defined(__svr4__)) || defined(__NetBSD__) || \
+    defined(_AIX)
 
+#if !defined(_AIX)
 #include <elf.h>
 #include <link.h>
+#endif
 #include <stdlib.h>
 #include <string.h>
 
 #include "InstrProfiling.h"
 #include "InstrProfilingInternal.h"
 
 #if defined(__FreeBSD__) && !defined(ElfW)
 /*

  * Fallback implementation for targets that don't support the GNU
  * extensions NT_GNU_BUILD_ID and __ehdr_start.
  */
 COMPILER_RT_VISIBILITY int __llvm_write_binary_ids(ProfDataWriter *Writer) {
   return 0;
 }
 #endif
 
+#if defined(_AIX)
+// Empty stubs to allow linking object files using the registration-based scheme
+COMPILER_RT_VISIBILITY
+void __llvm_profile_register_function(void *Data_) {}
+
+COMPILER_RT_VISIBILITY
+void __llvm_profile_register_names_function(void *NamesStart,
+                                            uint64_t NamesSize) {}
+
+// The __start_SECNAME and __stop_SECNAME symbols (for SECNAME \in
+// {"__llvm_prf_cnts", "__llvm_prf_data", "__llvm_prf_name", "__llvm_prf_vnds"})
+// are always live when linking on AIX, regardless if the .o's being linked
+// reference symbols from the profile library (for example when no files were
+// compiled with -fprofile-generate). That's because these symbols are kept
+// alive through references in constructor functions that are always live in the
+// default linking model on AIX (-bcdtors:all). The __start_SECNAME and
+// __stop_SECNAME symbols are only resolved by the linker when the SECNAME
+// section exists. So for the scenario where the user objects have no such
+// section (i.e. when they are compiled with -fno-profile-generate), we always
+// define these zero length variables in each of the above 4 sections.
+COMPILER_RT_VISIBILITY int dummy_cnts[0] COMPILER_RT_SECTION(
+    COMPILER_RT_SEG INSTR_PROF_CNTS_SECT_NAME);
+COMPILER_RT_VISIBILITY int dummy_data[0] COMPILER_RT_SECTION(
+    COMPILER_RT_SEG INSTR_PROF_DATA_SECT_NAME);
+COMPILER_RT_VISIBILITY const int dummy_name[0] COMPILER_RT_SECTION(
+    COMPILER_RT_SEG INSTR_PROF_NAME_SECT_NAME);
+COMPILER_RT_VISIBILITY int dummy_vnds[0] COMPILER_RT_SECTION(
+    COMPILER_RT_SEG INSTR_PROF_VNODES_SECT_NAME);
+
+// Create a fake reference to avoid GC'ing of the dummy variables by the linker.
+// Ideally, we create a ".ref" of each variable inside the function
+// __llvm_profile_begin_counters(), but there's no source level construct
+// that allows us to generate that.
+__attribute__((destructor)) void keep() {
+  int volatile use = &dummy_cnts < &dummy_data && &dummy_name < &dummy_vnds;
+  (void)use;
+}
+#endif
+
 #endif

compiler-rt/lib/profile/InstrProfilingPlatformOther.c

 /*===- InstrProfilingPlatformOther.c - Profile data default platform ------===*\
 |*
 |* Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
 |* See https://llvm.org/LICENSE.txt for license information.
 |* SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
 |*
 \*===----------------------------------------------------------------------===*/
 
 #if !defined(__APPLE__) && !defined(__linux__) && !defined(__FreeBSD__) &&     \
     !defined(__Fuchsia__) && !(defined(__sun__) && defined(__svr4__)) &&       \
-    !defined(__NetBSD__) && !defined(_WIN32)
+    !defined(__NetBSD__) && !defined(_WIN32) && !defined(_AIX)
 
 #include <stdlib.h>
 #include <stdio.h>
 
 #include "InstrProfiling.h"
 #include "InstrProfilingInternal.h"
 
 static const __llvm_profile_data *DataFirst = NULL;

llvm/lib/Transforms/Instrumentation/InstrProfiling.cpp

static inline bool shouldRecordFunctionAddr(Function *F) {
   return F->hasAddressTaken() || F->hasLinkOnceLinkage();
 }
 
 static bool needsRuntimeRegistrationOfSectionRange(const Triple &TT) {
   // Don't do this for Darwin.  compiler-rt uses linker magic.
   if (TT.isOSDarwin())
     return false;
   // Use linker script magic to get data/cnts/name start/end.
-  if (TT.isOSLinux() || TT.isOSFreeBSD() || TT.isOSNetBSD() ||
+  if (TT.isOSAIX() || TT.isOSLinux() || TT.isOSFreeBSD() || TT.isOSNetBSD() ||
       TT.isOSSolaris() || TT.isOSFuchsia() || TT.isPS4() || TT.isOSWindows())
     return false;
 
   return true;
 }
 
 GlobalVariable *
 InstrProfiling::createRegionCounters(InstrProfInstBase *Inc, StringRef Name,
                                      GlobalValue::LinkageTypes Linkage) {
   uint64_t NumCounters = Inc->getNumCounters()->getZExtValue();

llvm/test/Instrumentation/InstrProfiling/platform.ll

 ;; symbols by their sections.
 
 ; MACHO-NOT: define internal void @__llvm_profile_register_functions
 ; LINUX-NOT: define internal void @__llvm_profile_register_functions
 ; FREEBSD-NOT: define internal void @__llvm_profile_register_functions
 ; SOLARIS-NOT: define internal void @__llvm_profile_register_functions
 ; PS4-NOT: define internal void @__llvm_profile_register_functions
 ; WINDOWS-NOT: define internal void @__llvm_profile_register_functions
-; AIX: define internal void @__llvm_profile_register_functions
+; AIX-NOT: define internal void @__llvm_profile_register_functions
 
 ;; PR38340: When dynamic registration is used, we had a bug where we'd register
 ;; something that's not a __profd_* variable.
 
 ; MACHO-NOT: define internal void @__llvm_profile_init
 ; LINUX-NOT: define internal void @__llvm_profile_init
 ; FREEBSD-NOT: define internal void @__llvm_profile_init
 ; SOLARIS-NOT: define internal void @__llvm_profile_init
 ; PS4-NOT: define internal void @__llvm_profile_init
 ; WINDOWS-NOT: define internal void @__llvm_profile_init
-; AIX: define internal void @__llvm_profile_init
+; AIX-NOT: define internal void @__llvm_profile_init

llvm/test/Instrumentation/InstrProfiling/profiling.ll

 ; ELF_GENERIC:      define internal void @__llvm_profile_register_functions() unnamed_addr {
 ; ELF_GENERIC-NEXT:   call void @__llvm_profile_register_function(i8* bitcast (i32* @__llvm_profile_runtime to i8*))
 ; ELF_GENERIC-NEXT:   call void @__llvm_profile_register_function(i8* bitcast ({ i64, i64, i64, i8*, i8*, i32, [{{.*}} x i16] }* @__profd_foo to i8*))
 ; ELF_GENERIC-NEXT:   call void @__llvm_profile_register_function(i8* bitcast ({ i64, i64, i64, i8*, i8*, i32, [{{.*}} x i16] }* @__profd_foo_weak to i8*))
 ; ELF_GENERIC:        call void @__llvm_profile_register_names_function(i8* getelementptr inbounds {{.*}} @__llvm_prf_nm
 ; ELF_GENERIC-NEXT:   ret void
 ; ELF_GENERIC-NEXT: }
 
-; XCOFF:      define internal void @__llvm_profile_register_functions() unnamed_addr {
-; XCOFF-NEXT:   call void @__llvm_profile_register_function(i8* bitcast ({ i64, i64, i64, i8*, i8*, i32, [{{.*}} x i16] }* @__profd_foo to i8*))
-; XCOFF-NEXT:   call void @__llvm_profile_register_function(i8* bitcast ({ i64, i64, i64, i8*, i8*, i32, [{{.*}} x i16] }* @__profd_foo_weak to i8*))
-; XCOFF:   call void @__llvm_profile_register_names_function(i8* getelementptr inbounds {{.*}} @__llvm_prf_nm
-; XCOFF-NEXT:   ret void
-; XCOFF-NEXT: }
+; XCOFF-NOT:  internal void @__llvm_profile_register_functions()