[Android] Use ELF TLS for Android API level 29+

[brad]

Summary

Use ELF TLS by default instead of emulated TLS for Android build targets when the API level is >= 29.

Android Q features full TLS support. See release notes here and implementation details here.

(This is an updated https://reviews.llvm.org/D63904)

[danalbert]

+1, but I'd like @rprichard to be the one to actually sign off since he'll be the one to know if there are any caveats.

[rprichard]

nit: In the description/commit message, the "when the API level is > 29" should be "when the API level is >= 29".

I think it's OK, but I'd like confirmation that it works with LTO?

One issue I noticed earlier was that the compiler flag (-f[no-]emulated-tls) doesn't work with LTO:

• The flag is ignored when the compiler is only outputting LLVM IR, and
• The flag is also ignored at link-time when the IR is converted to emutls-or-efltls machine code.

This link-time LTO setting can be changed using -Wl,-plugin-opt,-emulated-tls={0,1}.

It's not clear to me from the patch whether the new default applies when -flto is used to compile. (I suspect that it does, but a confirmation might be useful?)

Also: This is technically an ABI break, but it only applies when a thread-local variable is accessed across an ABI boundary, which seems unlikely/rare. It seems like the change is worth it, in any case.

[enh]

In D147849#4256504, @rprichard wrote:

nit: In the description/commit message, the "when the API level is > 29" should be "when the API level is >= 29".

I think it's OK, but I'd like confirmation that it works with LTO?

One issue I noticed earlier was that the compiler flag (-f[no-]emulated-tls) doesn't work with LTO:

• The flag is ignored when the compiler is only outputting LLVM IR, and
• The flag is also ignored at link-time when the IR is converted to emutls-or-efltls machine code.

This link-time LTO setting can be changed using -Wl,-plugin-opt,-emulated-tls={0,1}.

does https://reviews.llvm.org/D147834 fix this?

It's not clear to me from the patch whether the new default applies when -flto is used to compile. (I suspect that it does, but a confirmation might be useful?)

Also: This is technically an ABI break, but it only applies when a thread-local variable is accessed across an ABI boundary, which seems unlikely/rare. It seems like the change is worth it, in any case.

[rprichard]

In D147849#4256641, @enh wrote:

In D147849#4256504, @rprichard wrote:

One issue I noticed earlier was that the compiler flag (-f[no-]emulated-tls) doesn't work with LTO:

• The flag is ignored when the compiler is only outputting LLVM IR, and
• The flag is also ignored at link-time when the IR is converted to emutls-or-efltls machine code.

This link-time LTO setting can be changed using -Wl,-plugin-opt,-emulated-tls={0,1}.

does https://reviews.llvm.org/D147834 fix this?

Thanks, I hadn't seen that before. Yes, that should make -f[no-]emulated-tls work with LTO.

[brad]

I'm not in a rush to commit this, it would be nice if at least one other person could Ok this.

[danalbert]

The rest of us are just echoing rprichard since he's the expert, but done :)

[enh]

(yeah, thanks so much for doing this!)

llvm/include/llvm/TargetParser/Triple.h

public:
 
   /// Tests whether the target supports comdat
   bool supportsCOMDAT() const {
     return !(isOSBinFormatMachO() || isOSBinFormatXCOFF() ||
              isOSBinFormatDXContainer());
   }
 
   /// Tests whether the target uses emulated TLS as default.
+  ///
+  /// Note: Android API level 29 (10) introduced ELF TLS.
   bool hasDefaultEmulatedTLS() const {
-    return isAndroid() || isOSOpenBSD() || isWindowsCygwinEnvironment() ||
-           isOHOSFamily();
+    return (isAndroid() && isAndroidVersionLT(29)) || isOSOpenBSD() ||
+           isWindowsCygwinEnvironment() || isOHOSFamily();
   }
 
   /// Tests whether the target uses -data-sections as default.
   bool hasDefaultDataSections() const {
     return isOSBinFormatXCOFF() || isWasm();
   }
 
   /// Tests if the environment supports dllimport/export annotations.

llvm/test/CodeGen/AArch64/emutls_generic.ll

 ; RUN:     | FileCheck -check-prefix=ARM_64 %s
 ; RUN: llc < %s -mtriple=aarch64-linux-android -relocation-model=pic -O3 \
 ; RUN:     | FileCheck -check-prefix=ARM_64 %s
 ; RUN: llc < %s -mtriple=aarch64-linux-android -O3 \
 ; RUN:     | FileCheck -check-prefix=ARM_64 %s
 ; aarch64-windows-gnu needs explicit -emulated-tls
 ; RUN: llc < %s -mtriple=aarch64-apple-darwin -O3 \
 ; RUN:     | FileCheck -check-prefix=NoEMU %s
+; RUN: llc < %s -mtriple=aarch64-linux-android29 -O3 \
+; RUN:     | FileCheck -check-prefix=NoEMU %s
 
 ; NoEMU-NOT: __emutls
 
 ; Make sure that TLS symbols are emitted in expected order.
 
 @external_x = external thread_local global i32, align 8
 @external_y = thread_local global i8 7, align 2
 @internal_y = internal thread_local global i64 9, align 16

llvm/test/CodeGen/X86/emutls-pic.ll

 ; RUN: llc < %s -emulated-tls -mtriple=i386-linux-gnu -relocation-model=pic | FileCheck -check-prefix=X86 %s
 ; RUN: llc < %s -emulated-tls -mtriple=x86_64-linux-gnu -relocation-model=pic | FileCheck -check-prefix=X64 %s
 ; RUN: llc < %s -emulated-tls -mtriple=i386-linux-android -relocation-model=pic | FileCheck -check-prefix=X86 %s
 ; RUN: llc < %s -emulated-tls -mtriple=x86_64-linux-android -relocation-model=pic | FileCheck -check-prefix=X64 %s
 
 ; RUN: llc < %s -mtriple=i386-linux-gnu -relocation-model=pic | FileCheck -check-prefix=NoEMU %s
 ; RUN: llc < %s -mtriple=x86_64-linux-gnu -relocation-model=pic | FileCheck -check-prefix=NoEMU %s
+; RUN: llc < %s -mtriple=x86_64-linux-android29 -relocation-model=pic | FileCheck -check-prefix=NoEMU %s
 ; RUN: llc < %s -mtriple=i386-linux-android -relocation-model=pic | FileCheck -check-prefix=X86 %s
 ; RUN: llc < %s -mtriple=x86_64-linux-android -relocation-model=pic | FileCheck -check-prefix=X64 %s
 ; RUN: llc < %s -mtriple=i386-linux-ohos -relocation-model=pic | FileCheck -check-prefix=X86 %s
 ; RUN: llc < %s -mtriple=x86_64-linux-ohos -relocation-model=pic | FileCheck -check-prefix=X64 %s
 
 ; NoEMU-NOT: __emutls
 
 ; Use my_emutls_get_address like __emutls_get_address.

llvm/test/CodeGen/X86/emutls-pie.ll

 ; RUN: llc < %s -emulated-tls -mcpu=generic -mtriple=i386-linux-gnu -relocation-model=pic \
 ; RUN:   | FileCheck -check-prefix=X86 %s
 ; RUN: llc < %s -emulated-tls -mcpu=generic -mtriple=x86_64-linux-gnu -relocation-model=pic \
 ; RUN:   | FileCheck -check-prefix=X64 %s
 ; RUN: llc < %s -emulated-tls -mcpu=generic -mtriple=i386-linux-android -relocation-model=pic \
 ; RUN:   | FileCheck -check-prefix=X86 %s
 ; RUN: llc < %s -emulated-tls -mcpu=generic -mtriple=x86_64-linux-android -relocation-model=pic \
 ; RUN:   | FileCheck -check-prefix=X64 %s
 
 ; RUN: llc < %s -mcpu=generic -mtriple=i386-linux-gnu -relocation-model=pic \
 ; RUN:   | FileCheck -check-prefix=NoEMU %s
 ; RUN: llc < %s -mcpu=generic -mtriple=x86_64-linux-gnu -relocation-model=pic \
 ; RUN:   | FileCheck -check-prefix=NoEMU %s
+; RUN: llc < %s -mcpu=generic -mtriple=i386-linux-android29 -relocation-model=pic \
+; RUN:   | FileCheck -check-prefix=NoEMU %s
 ; RUN: llc < %s -mcpu=generic -mtriple=i386-linux-android -relocation-model=pic \
 ; RUN:   | FileCheck -check-prefix=X86 %s
 ; RUN: llc < %s -mcpu=generic -mtriple=x86_64-linux-android -relocation-model=pic \
 ; RUN:   | FileCheck -check-prefix=X64 %s
 
 ; NoEMU-NOT: __emutls
 
 ; Use my_emutls_get_address like __emutls_get_address.

llvm/test/CodeGen/X86/emutls_generic.ll

 ; RUN: llc < %s -mtriple=i686-linux-android -relocation-model=pic \
 ; RUN:     | FileCheck -check-prefix=X86_32 %s
 ; RUN: llc < %s -mtriple=i686-linux-android -relocation-model=pic \
 ; RUN:     | FileCheck -check-prefix=X86_32 %s
 ; RUN: llc < %s -mtriple=x86_64-linux-android -relocation-model=pic \
 ; RUN:     | FileCheck -check-prefix=X86_64 %s
 ; RUN: llc < %s -mtriple=i386-linux-gnu -relocation-model=pic \
 ; RUN:     | FileCheck -check-prefix=NoEMU %s
+; RUN: llc < %s -mtriple=i686-linux-android29 -relocation-model=pic \
+; RUN:     | FileCheck -check-prefix=NoEMU %s
 
 ; NoEMU-NOT: __emutls
 
 ; Make sure that TLS symbols are emitted in expected order.
 
 @external_x = external thread_local global i32, align 8
 @external_y = thread_local global i8 7, align 2
 @internal_y = internal thread_local global i64 9, align 16

llvm/test/CodeGen/X86/fast-isel-emutls.ll

 ; RUN: llc < %s -emulated-tls -relocation-model=pic -mtriple=i686-unknown-linux-gnu -fast-isel | FileCheck %s
 ; RUN: llc < %s -relocation-model=pic -mtriple=i686-unknown-linux-gnu -fast-isel \
 ; RUN: | FileCheck -check-prefix=NoEMU %s
+; RUN: llc < %s -relocation-model=pic -mtriple=i686-linux-android29 -fast-isel \
+; RUN: | FileCheck -check-prefix=NoEMU %s
 ; PR3654
 
 ; NoEMU-NOT: __emutls
 
 @v = thread_local global i32 0
 define i32 @f() nounwind {
 entry:
           %t = load i32, ptr @v