[clang][llvm][aarch64][win] Add a clang flag and module attribute for import call optimization, and remove LLVM flag (#122831)

Switches import call optimization from being enabled by an LLVM flag to
instead using a module attribute, and creates a new Clang flag that will
set that attribute. This addresses the concern raised in the original
PR:
<llvm/llvm-project#121516 (comment)>

This change also only creates the Called Global info if the module
attribute is present, addressing this concern:
<llvm/llvm-project#122762 (review)>

Author: dpaoliello

clang/include/clang/Basic/CodeGenOptions.def

@@ -465,6 +465,10 @@ ENUM_CODEGENOPT(ZeroCallUsedRegs, llvm::ZeroCallUsedRegs::ZeroCallUsedRegsKind,
 /// non-deleting destructors. (No effect on Microsoft ABI.)
 CODEGENOPT(CtorDtorReturnThis, 1, 0)
 
+/// Enables emitting Import Call sections on supported targets that can be used
+/// by the Windows kernel to enable import call optimization.
+CODEGENOPT(ImportCallOptimization, 1, 0)
+
 /// FIXME: Make DebugOptions its own top-level .def file.
 #include "DebugOptions.def"
 

clang/include/clang/Driver/Options.td

@@ -7587,6 +7587,11 @@ def fexperimental_assignment_tracking_EQ : Joined<["-"], "fexperimental-assignme
 def enable_tlsdesc : Flag<["-"], "enable-tlsdesc">,
   MarshallingInfoFlag<CodeGenOpts<"EnableTLSDESC">>;
 
+def import_call_optimization : Flag<["-"], "import-call-optimization">,
+    HelpText<"Emit Import Call sections on supported targets that can be used "
+             "by the Windows kernel to enable import call optimization">,
+    MarshallingInfoFlag<CodeGenOpts<"ImportCallOptimization">>;
+
 } // let Visibility = [CC1Option]
 
 //===----------------------------------------------------------------------===//

clang/lib/CodeGen/CodeGenModule.cpp

@@ -1293,6 +1293,11 @@ void CodeGenModule::Release() {
   if (LangOpts.EHAsynch)
     getModule().addModuleFlag(llvm::Module::Warning, "eh-asynch", 1);
 
+  // Emit Import Call section.
+  if (CodeGenOpts.ImportCallOptimization)
+    getModule().addModuleFlag(llvm::Module::Warning, "import-call-optimization",
+                              1);
+
   // Indicate whether this Module was compiled with -fopenmp
   if (getLangOpts().OpenMP && !getLangOpts().OpenMPSimd)
     getModule().addModuleFlag(llvm::Module::Max, "openmp", LangOpts.OpenMP);

clang/test/CodeGen/import-call-optimization.c

@@ -0,0 +1,5 @@
+// RUN: %clang_cc1 -import-call-optimization -emit-llvm %s -o - | FileCheck %s
+
+void f(void) {}
+
+// CHECK: !"import-call-optimization", i32 1}

llvm/lib/Target/AArch64/AArch64AsmPrinter.cpp

@@ -78,11 +78,6 @@ static cl::opt<PtrauthCheckMode> PtrauthAuthChecks(
     cl::desc("Check pointer authentication auth/resign failures"),
     cl::init(Default));
 
-static cl::opt<bool> EnableImportCallOptimization(
-    "aarch64-win-import-call-optimization", cl::Hidden,
-    cl::desc("Enable import call optimization for AArch64 Windows"),
-    cl::init(false));
-
 #define DEBUG_TYPE "asm-printer"
 
 namespace {
@@ -95,6 +90,7 @@ class AArch64AsmPrinter : public AsmPrinter {
 #ifndef NDEBUG
   unsigned InstsEmitted;
 #endif
+  bool EnableImportCallOptimization = false;
   DenseMap<MCSection *, std::vector<std::pair<MCSymbol *, MCSymbol *>>>
       SectionToImportedFunctionCalls;
 
@@ -344,6 +340,9 @@ void AArch64AsmPrinter::emitStartOfAsmFile(Module &M) {
     OutStreamer->emitSymbolAttribute(S, MCSA_Global);
     OutStreamer->emitAssignment(
         S, MCConstantExpr::create(Feat00Value, MMI->getContext()));
+
+    if (M.getModuleFlag("import-call-optimization"))
+      EnableImportCallOptimization = true;
   }
 
   if (!TT.isOSBinFormatELF())
@@ -3172,8 +3171,7 @@ void AArch64AsmPrinter::emitInstruction(const MachineInstr *MI) {
 
 void AArch64AsmPrinter::recordIfImportCall(
     const llvm::MachineInstr *BranchInst) {
-  if (!EnableImportCallOptimization ||
-      !TM.getTargetTriple().isOSBinFormatCOFF())
+  if (!EnableImportCallOptimization)
     return;
 
   auto [GV, OpFlags] = BranchInst->getMF()->tryGetCalledGlobal(BranchInst);

llvm/lib/Target/AArch64/AArch64ISelLowering.cpp

@@ -9573,7 +9573,8 @@ AArch64TargetLowering::LowerCall(CallLoweringInfo &CLI,
 
     DAG.addNoMergeSiteInfo(Ret.getNode(), CLI.NoMerge);
     DAG.addCallSiteInfo(Ret.getNode(), std::move(CSInfo));
-    if (CalledGlobal)
+    if (CalledGlobal &&
+        MF.getFunction().getParent()->getModuleFlag("import-call-optimization"))
       DAG.addCalledGlobal(Ret.getNode(), CalledGlobal, OpFlags);
     return Ret;
   }
@@ -9586,7 +9587,8 @@ AArch64TargetLowering::LowerCall(CallLoweringInfo &CLI,
   DAG.addNoMergeSiteInfo(Chain.getNode(), CLI.NoMerge);
   InGlue = Chain.getValue(1);
   DAG.addCallSiteInfo(Chain.getNode(), std::move(CSInfo));
-  if (CalledGlobal)
+  if (CalledGlobal &&
+      MF.getFunction().getParent()->getModuleFlag("import-call-optimization"))
     DAG.addCalledGlobal(Chain.getNode(), CalledGlobal, OpFlags);
 
   uint64_t CalleePopBytes =

llvm/test/CodeGen/AArch64/win-import-call-optimization-nocalls.ll

@@ -1,4 +1,4 @@
-; RUN: llc -mtriple=aarch64-pc-windows-msvc -aarch64-win-import-call-optimization < %s | FileCheck %s
+; RUN: llc -mtriple=aarch64-pc-windows-msvc < %s | FileCheck %s
 
 define dso_local void @normal_call() local_unnamed_addr {
 entry:
@@ -16,3 +16,6 @@ declare void @a() local_unnamed_addr
 ; CHECK-LABEL  .section   .impcall,"yi"
 ; CHECK-NEXT   .asciz  "Imp_Call_V1"
 ; CHECK-NOT    .secnum
+
+!llvm.module.flags = !{!0}
+!0 = !{i32 1, !"import-call-optimization", i32 1}

llvm/test/CodeGen/AArch64/win-import-call-optimization.ll

@@ -1,48 +1,48 @@
-; RUN: llc -mtriple=aarch64-pc-windows-msvc -aarch64-win-import-call-optimization < %s | FileCheck %s --check-prefix=CHECK-ENABLED
-; RUN: llc -mtriple=aarch64-pc-windows-msvc < %s | FileCheck %s --check-prefix=CHECK-DISABLED
-
-; CHECK-DISABLED-NOT: .section        .impcall
+; RUN: llc -mtriple=aarch64-pc-windows-msvc < %s | FileCheck %s --check-prefix=CHECK
 
 define dso_local void @normal_call() local_unnamed_addr section "nc_sect" {
 entry:
   call void @a()
   call void @a()
   ret void
 }
-; CHECK-ENABLED-LABEL:  normal_call:
-; CHECK-ENABLED:        adrp    [[ADRPREG:x[0-9]+]], __imp_a
-; CHECK-ENABLED-NEXT:   ldr     [[LDRREG:x[0-9]+]], [[[ADRPREG]], :lo12:__imp_a]
-; CHECK-ENABLED-NEXT:   .Limpcall0:
-; CHECK-ENABLED-NEXT:   blr     [[LDRREG]]
-; CHECK-ENABLED-NEXT:   .Limpcall1:
-; CHECK-ENABLED-NEXT:   blr     [[LDRREG]]
+; CHECK-LABEL:  normal_call:
+; CHECK:        adrp    [[ADRPREG:x[0-9]+]], __imp_a
+; CHECK-NEXT:   ldr     [[LDRREG:x[0-9]+]], [[[ADRPREG]], :lo12:__imp_a]
+; CHECK-NEXT:   .Limpcall0:
+; CHECK-NEXT:   blr     [[LDRREG]]
+; CHECK-NEXT:   .Limpcall1:
+; CHECK-NEXT:   blr     [[LDRREG]]
 
 define dso_local void @tail_call() local_unnamed_addr section "tc_sect" {
 entry:
   tail call void @b()
   ret void
 }
-; CHECK-ENABLED-LABEL:  tail_call:
-; CHECK-ENABLED:        adrp    [[ADRPREG:x[0-9]+]], __imp_b
-; CHECK-ENABLED-NEXT:   ldr     [[LDRREG:x[0-9]+]], [[[ADRPREG]], :lo12:__imp_b]
-; CHECK-ENABLED-NEXT:   .Limpcall2:
-; CHECK-ENABLED-NEXT:   br      [[LDRREG]]
+; CHECK-LABEL:  tail_call:
+; CHECK:        adrp    [[ADRPREG:x[0-9]+]], __imp_b
+; CHECK-NEXT:   ldr     [[LDRREG:x[0-9]+]], [[[ADRPREG]], :lo12:__imp_b]
+; CHECK-NEXT:   .Limpcall2:
+; CHECK-NEXT:   br      [[LDRREG]]
 
 declare dllimport void @a() local_unnamed_addr
 declare dllimport void @b() local_unnamed_addr
 
-; CHECK-ENABLED-LABEL  .section   .impcall,"yi"
-; CHECK-ENABLED-NEXT   .asciz  "Imp_Call_V1"
-; CHECK-ENABLED-NEXT   .word   32
-; CHECK-ENABLED-NEXT   .secnum nc_sect
-; CHECK-ENABLED-NEXT   .word   19
-; CHECK-ENABLED-NEXT   .secoffset      .Limpcall0
-; CHECK-ENABLED-NEXT   .symidx __imp_a
-; CHECK-ENABLED-NEXT   .word   19
-; CHECK-ENABLED-NEXT   .secoffset      .Limpcall1
-; CHECK-ENABLED-NEXT   .symidx __imp_a
-; CHECK-ENABLED-NEXT   .word   20
-; CHECK-ENABLED-NEXT   .secnum tc_sect
-; CHECK-ENABLED-NEXT   .word   19
-; CHECK-ENABLED-NEXT   .secoffset      .Limpcall2
-; CHECK-ENABLED-NEXT   .symidx __imp_b
+; CHECK-LABEL  .section   .impcall,"yi"
+; CHECK-NEXT   .asciz  "Imp_Call_V1"
+; CHECK-NEXT   .word   32
+; CHECK-NEXT   .secnum nc_sect
+; CHECK-NEXT   .word   19
+; CHECK-NEXT   .secoffset      .Limpcall0
+; CHECK-NEXT   .symidx __imp_a
+; CHECK-NEXT   .word   19
+; CHECK-NEXT   .secoffset      .Limpcall1
+; CHECK-NEXT   .symidx __imp_a
+; CHECK-NEXT   .word   20
+; CHECK-NEXT   .secnum tc_sect
+; CHECK-NEXT   .word   19
+; CHECK-NEXT   .secoffset      .Limpcall2
+; CHECK-NEXT   .symidx __imp_b
+
+!llvm.module.flags = !{!0}
+!0 = !{i32 1, !"import-call-optimization", i32 1}