[AArch64][ELF] Prefer to lower MC_GlobalAddress operands to .Lfoo$local

Similar to X86 D73230 & 46788a21f9152be3950e57dc526454655682bdd4

With this change, we can set dso_local in clang's -fpic -fno-semantic-interposition mode,
for default visibility external linkage non-ifunc-non-COMDAT definitions.

For such dso_local definitions, variable access/taking the address of a
function/calling a function will go through a local alias to avoid GOT/PLT.

Note: the 'S' inline assembly constraint refers to an absolute symbolic address
or a label reference (D46745).

Differential Revision: https://reviews.llvm.org/D101872

lib/CodeGen/AsmPrinter/AsmPrinterInlineAsm.cpp

@@ -589,7 +589,7 @@ void AsmPrinter::PrintSpecial(const MachineInstr *MI, raw_ostream &OS,
 
 void AsmPrinter::PrintSymbolOperand(const MachineOperand &MO, raw_ostream &OS) {
   assert(MO.isGlobal() && "caller should check MO.isGlobal");
-  getSymbol(MO.getGlobal())->print(OS, MAI);
+  getSymbolPreferLocal(*MO.getGlobal())->print(OS, MAI);
   printOffset(MO.getOffset(), OS);
 }
 

lib/Target/AArch64/AArch64MCInstLower.cpp

@@ -39,7 +39,7 @@ AArch64MCInstLower::GetGlobalAddressSymbol(const MachineOperand &MO) const {
   unsigned TargetFlags = MO.getTargetFlags();
   const Triple &TheTriple = Printer.TM.getTargetTriple();
   if (!TheTriple.isOSBinFormatCOFF())
-    return Printer.getSymbol(GV);
+    return Printer.getSymbolPreferLocal(*GV);
 
   assert(TheTriple.isOSWindows() &&
          "Windows is the only supported COFF target");

test/CodeGen/AArch64/basic-pic.ll

@@ -22,33 +22,3 @@ define i32* @get_globalvaraddr() {
 
   ret i32* @var
 }
-
-@hiddenvar = hidden global i32 0
-
-define i32 @get_hiddenvar() {
-; CHECK-LABEL: get_hiddenvar:
-
-  %val = load i32, i32* @hiddenvar
-; CHECK: adrp x[[HI:[0-9]+]], hiddenvar
-; CHECK: ldr w0, [x[[HI]], :lo12:hiddenvar]
-
-  ret i32 %val
-}
-
-define i32* @get_hiddenvaraddr() {
-; CHECK-LABEL: get_hiddenvaraddr:
-
-  %val = load i32, i32* @hiddenvar
-; CHECK: adrp [[HI:x[0-9]+]], hiddenvar
-; CHECK: add x0, [[HI]], :lo12:hiddenvar
-
-  ret i32* @hiddenvar
-}
-
-define void()* @get_func() {
-; CHECK-LABEL: get_func:
-
-  ret void()* bitcast(void()*()* @get_func to void()*)
-; CHECK: adrp x[[GOTHI:[0-9]+]], :got:get_func
-; CHECK: ldr x0, [x[[GOTHI]], :got_lo12:get_func]
-}

test/CodeGen/AArch64/elf-globals-static.ll

@@ -78,16 +78,6 @@ define i64* @test_addr() {
 ; CHECK-FAST: add x0, [[HIREG]], :lo12:var64
 }
 
-@hiddenvar = hidden global i32 0, align 4
-@protectedvar = protected global i32 0, align 4
-
-define i32 @test_vis() {
-  %lhs = load i32, i32* @hiddenvar, align 4
-  %rhs = load i32, i32* @protectedvar, align 4
-  %ret = add i32 %lhs, %rhs
-  ret i32 %ret
-}
-
 @var_default = external dso_local global [2 x i32]
 
 define i32 @test_default_align() {

test/CodeGen/AArch64/elf-preemption.ll

@@ -0,0 +1,114 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
+; RUN: llc -mtriple=aarch64 -relocation-model=static < %s | FileCheck %s --check-prefixes=CHECK,STATIC
+; RUN: llc -mtriple=aarch64 -relocation-model=pic < %s | FileCheck %s --check-prefixes=CHECK,PIC
+
+@preemptable_var = dso_preemptable global i32 42
+define i32* @get_preemptable_var() nounwind {
+; CHECK-LABEL: get_preemptable_var:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    adrp x0, :got:preemptable_var
+; CHECK-NEXT:    ldr x0, [x0, :got_lo12:preemptable_var]
+; CHECK-NEXT:    ret
+  ret i32* @preemptable_var
+}
+
+@dsolocal_var = dso_local global i32 42
+define i32* @get_dsolocal_var() nounwind {
+; STATIC-LABEL: get_dsolocal_var:
+; STATIC:       // %bb.0:
+; STATIC-NEXT:    adrp x0, dsolocal_var
+; STATIC-NEXT:    add x0, x0, :lo12:dsolocal_var
+; STATIC-NEXT:    ret
+;
+; PIC-LABEL: get_dsolocal_var:
+; PIC:       // %bb.0:
+; PIC-NEXT:    adrp x0, .Ldsolocal_var$local
+; PIC-NEXT:    add x0, x0, :lo12:.Ldsolocal_var$local
+; PIC-NEXT:    ret
+  ret i32* @dsolocal_var
+}
+
+@weak_dsolocal_var = weak dso_local global i32 42
+define i32* @get_weak_dsolocal_var() nounwind {
+; CHECK-LABEL: get_weak_dsolocal_var:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    adrp x0, weak_dsolocal_var
+; CHECK-NEXT:    add x0, x0, :lo12:weak_dsolocal_var
+; CHECK-NEXT:    ret
+  ret i32* @weak_dsolocal_var
+}
+
+@hidden_var = hidden global i32 42
+define i32* @get_hidden_var() nounwind {
+; CHECK-LABEL: get_hidden_var:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    adrp x0, hidden_var
+; CHECK-NEXT:    add x0, x0, :lo12:hidden_var
+; CHECK-NEXT:    ret
+  ret i32* @hidden_var
+}
+
+@protected_var = protected global i32 42
+define i32* @get_protected_var() nounwind {
+; CHECK-LABEL: get_protected_var:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    adrp x0, protected_var
+; CHECK-NEXT:    add x0, x0, :lo12:protected_var
+; CHECK-NEXT:    ret
+  ret i32* @protected_var
+}
+
+define dso_preemptable void()* @preemptable_func() nounwind {
+; CHECK-LABEL: preemptable_func:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    adrp x0, :got:preemptable_func
+; CHECK-NEXT:    ldr x0, [x0, :got_lo12:preemptable_func]
+; CHECK-NEXT:    ret
+  ret void()* bitcast(void()*()* @preemptable_func to void()*)
+}
+
+define dso_local void()* @dsolocal_func() nounwind {
+; STATIC-LABEL: dsolocal_func:
+; STATIC:       // %bb.0:
+; STATIC-NEXT:    adrp x0, dsolocal_func
+; STATIC-NEXT:    add x0, x0, :lo12:dsolocal_func
+; STATIC-NEXT:    ret
+;
+; PIC-LABEL: dsolocal_func:
+; PIC:       .Ldsolocal_func$local:
+; PIC-NEXT:  // %bb.0:
+; PIC-NEXT:    adrp x0, .Ldsolocal_func$local
+; PIC-NEXT:    add x0, x0, :lo12:.Ldsolocal_func$local
+; PIC-NEXT:    ret
+  ret void()* bitcast(void()*()* @dsolocal_func to void()*)
+}
+
+define weak dso_local void()* @weak_dsolocal_func() nounwind {
+; CHECK-LABEL: weak_dsolocal_func:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    adrp x0, weak_dsolocal_func
+; CHECK-NEXT:    add x0, x0, :lo12:weak_dsolocal_func
+; CHECK-NEXT:    ret
+  ret void()* bitcast(void()*()* @weak_dsolocal_func to void()*)
+}
+
+;; bl .Ldsolocal_func$local either resolves to a constant at assembly time
+;; or produces a relocation which can potentially cause a veneer.
+define dso_local void @call_dsolocal_func() nounwind {
+; STATIC-LABEL: call_dsolocal_func:
+; STATIC:       // %bb.0:
+; STATIC-NEXT:    str x30, [sp, #-16]! // 8-byte Folded Spill
+; STATIC-NEXT:    bl dsolocal_func
+; STATIC-NEXT:    ldr x30, [sp], #16 // 8-byte Folded Reload
+; STATIC-NEXT:    ret
+;
+; PIC-LABEL: call_dsolocal_func:
+; PIC:       .Lcall_dsolocal_func$local:
+; PIC-NEXT:  // %bb.0:
+; PIC-NEXT:    str x30, [sp, #-16]! // 8-byte Folded Spill
+; PIC-NEXT:    bl .Ldsolocal_func$local
+; PIC-NEXT:    ldr x30, [sp], #16 // 8-byte Folded Reload
+; PIC-NEXT:    ret
+  call void()* @dsolocal_func()
+  ret void
+}

test/CodeGen/AArch64/semantic-interposition-asm.ll

@@ -0,0 +1,25 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
+; RUN: llc -mtriple=aarch64 -relocation-model=pic < %s | FileCheck %s
+
+;; Test that we use the local alias for dso_local globals in inline assembly.
+
+@gv0 = dso_local global i32 0
+@gv1 = dso_preemptable global i32 1
+
+define i32 @load() nounwind {
+; CHECK-LABEL: load:
+; CHECK:       // %bb.0: // %entry
+; CHECK-NEXT:    //APP
+; CHECK-NEXT:    adrp x0, .Lgv0$local
+; CHECK-NEXT:    ldr w0, [x0, :lo12:.Lgv0$local]
+; CHECK-NEXT:    adrp x8, gv1
+; CHECK-NEXT:    ldr w8, [x8, :lo12:gv1]
+; CHECK-NEXT:    add x0, x8, x0
+; CHECK-NEXT:    //NO_APP
+; CHECK-NEXT:    // kill: def $w0 killed $w0 killed $x0
+; CHECK-NEXT:    ret
+entry:
+  %0 = tail call i64 asm "adrp $0, $1\0Aldr ${0:w}, [$0, :lo12:$1]\0Aadrp x8, $2\0Aldr w8, [x8, :lo12:$2]\0Aadd $0,x8,$0", "=r,S,S,~{x8}"(i32* nonnull @gv0, i32* nonnull @gv1)
+  %conv = trunc i64 %0 to i32
+  ret i32 %conv
+}