[IR] make -stack-alignment= into a module attr

Similar to D102742, specifying the stack alignment via CodegenOpts means that this flag gets dropped during LTO, unless the command line is re-specified as a plugin opt. Instead, encode this information as a module level attribute so that we don't have to expose this llvm internal flag when linking the Linux kernel with LTO. Looks like external dependencies might need a fix: * llvm-hs/llvm-hs#345 * halide/Halide#6079 Link: ClangBuiltLinux/linux#1377 Reviewed By: tejohnson Differential Revision: https://reviews.llvm.org/D103048
steffenlarsen · Jun 8, 2021 · 433c8d9 · 433c8d9
1 parent c5d56fe
commit 433c8d9
Show file tree

Hide file tree

Showing 23 changed files with 189 additions and 120 deletions.
diff --git a/clang/lib/CodeGen/BackendUtil.cpp b/clang/lib/CodeGen/BackendUtil.cpp
@@ -530,7 +530,6 @@ static bool initTargetOptions(DiagnosticsEngine &Diags,
   Options.NoNaNsFPMath = LangOpts.NoHonorNaNs;
   Options.NoZerosInBSS = CodeGenOpts.NoZeroInitializedInBSS;
   Options.UnsafeFPMath = LangOpts.UnsafeFPMath;
-  Options.StackAlignmentOverride = CodeGenOpts.StackAlignment;
 
   Options.BBSections =
       llvm::StringSwitch<llvm::BasicBlockSection>(CodeGenOpts.BBSections)

diff --git a/clang/lib/CodeGen/CodeGenModule.cpp b/clang/lib/CodeGen/CodeGenModule.cpp
@@ -785,6 +785,8 @@ void CodeGenModule::Release() {
   if (getCodeGenOpts().StackProtectorGuardOffset != INT_MAX)
     getModule().setStackProtectorGuardOffset(
         getCodeGenOpts().StackProtectorGuardOffset);
+  if (getCodeGenOpts().StackAlignment)
+    getModule().setOverrideStackAlignment(getCodeGenOpts().StackAlignment);
 
   getTargetCodeGenInfo().emitTargetMetadata(*this, MangledDeclNames);
 

diff --git a/clang/test/CodeGen/stackrealign-main.c b/clang/test/CodeGen/stackrealign-main.c
@@ -1,4 +1,5 @@
 // RUN: %clang_cc1 -triple i386-unknown-unknown -emit-llvm -o - -mstack-alignment=64 %s | FileCheck %s
+// RUN: %clang_cc1 -triple i386-unknown-unknown -emit-llvm -o - %s | FileCheck %s --check-prefix=DEFAULT
 
 // CHECK-LABEL: define{{.*}} void @other()
 // CHECK: [[OTHER:#[0-9]+]]
@@ -17,3 +18,5 @@ int main(int argc, char **argv) {
 // CHECK-NOT: "stackrealign"
 // CHECK: }
 // CHECK: attributes [[MAIN]] = { noinline nounwind optnone {{.*}}"stackrealign"{{.*}} }
+// CHECK: !{i32 1, !"override-stack-alignment", i32 64}
+// DEFAULT-NOT: "override-stack-alignment"
diff --git a/llvm/include/llvm/IR/Module.h b/llvm/include/llvm/IR/Module.h
@@ -909,6 +909,10 @@ class Module {
   int getStackProtectorGuardOffset() const;
   void setStackProtectorGuardOffset(int Offset);
 
+  /// Get/set the stack alignment overridden from the default.
+  unsigned getOverrideStackAlignment() const;
+  void setOverrideStackAlignment(unsigned Align);
+
   /// @name Utility functions for querying and setting the build SDK version
   /// @{
 

diff --git a/llvm/include/llvm/Target/TargetOptions.h b/llvm/include/llvm/Target/TargetOptions.h
@@ -201,9 +201,6 @@ namespace llvm {
     /// as their parent function, etc.), using an alternate ABI if necessary.
     unsigned GuaranteedTailCallOpt : 1;
 
-    /// StackAlignmentOverride - Override default stack alignment for target.
-    unsigned StackAlignmentOverride = 0;
-
     /// StackSymbolOrdering - When true, this will allow CodeGen to order
     /// the local stack symbols (for code size, code locality, or any other
     /// heuristics). When false, the local symbols are left in whatever order

diff --git a/llvm/lib/CodeGen/CommandFlags.cpp b/llvm/lib/CodeGen/CommandFlags.cpp
@@ -69,7 +69,6 @@ CGOPT(bool, DontPlaceZerosInBSS)
 CGOPT(bool, EnableGuaranteedTailCallOpt)
 CGOPT(bool, DisableTailCalls)
 CGOPT(bool, StackSymbolOrdering)
-CGOPT(unsigned, OverrideStackAlignment)
 CGOPT(bool, StackRealign)
 CGOPT(std::string, TrapFuncName)
 CGOPT(bool, UseCtors)
@@ -305,11 +304,6 @@ codegen::RegisterCodeGenFlags::RegisterCodeGenFlags() {
       cl::init(true));
   CGBINDOPT(StackSymbolOrdering);
 
-  static cl::opt<unsigned> OverrideStackAlignment(
-      "stack-alignment", cl::desc("Override default stack alignment"),
-      cl::init(0));
-  CGBINDOPT(OverrideStackAlignment);
-
   static cl::opt<bool> StackRealign(
       "stackrealign",
       cl::desc("Force align the stack to the minimum alignment"),
@@ -508,7 +502,6 @@ codegen::InitTargetOptionsFromCodeGenFlags(const Triple &TheTriple) {
   Options.EnableAIXExtendedAltivecABI = getEnableAIXExtendedAltivecABI();
   Options.NoZerosInBSS = getDontPlaceZerosInBSS();
   Options.GuaranteedTailCallOpt = getEnableGuaranteedTailCallOpt();
-  Options.StackAlignmentOverride = getOverrideStackAlignment();
   Options.StackSymbolOrdering = getStackSymbolOrdering();
   Options.UseInitArray = !getUseCtors();
   Options.RelaxELFRelocations = getRelaxELFRelocations();

diff --git a/llvm/lib/IR/Module.cpp b/llvm/lib/IR/Module.cpp
@@ -721,6 +721,17 @@ void Module::setStackProtectorGuardOffset(int Offset) {
   addModuleFlag(ModFlagBehavior::Error, "stack-protector-guard-offset", Offset);
 }
 
+unsigned Module::getOverrideStackAlignment() const {
+  Metadata *MD = getModuleFlag("override-stack-alignment");
+  if (auto *CI = mdconst::dyn_extract_or_null<ConstantInt>(MD))
+    return CI->getZExtValue();
+  return 0;
+}
+
+void Module::setOverrideStackAlignment(unsigned Align) {
+  addModuleFlag(ModFlagBehavior::Error, "override-stack-alignment", Align);
+}
+
 void Module::setSDKVersion(const VersionTuple &V) {
   SmallVector<unsigned, 3> Entries;
   Entries.push_back(V.getMajor());

diff --git a/llvm/lib/Target/X86/X86TargetMachine.cpp b/llvm/lib/Target/X86/X86TargetMachine.cpp
@@ -314,8 +314,8 @@ X86TargetMachine::getSubtargetImpl(const Function &F) const {
     resetTargetOptions(F);
     I = std::make_unique<X86Subtarget>(
         TargetTriple, CPU, TuneCPU, FS, *this,
-        MaybeAlign(Options.StackAlignmentOverride), PreferVectorWidthOverride,
-        RequiredVectorWidth);
+        MaybeAlign(F.getParent()->getOverrideStackAlignment()),
+        PreferVectorWidthOverride, RequiredVectorWidth);
   }
   return I.get();
 }

diff --git a/llvm/test/CodeGen/Generic/ForceStackAlign.ll b/llvm/test/CodeGen/Generic/ForceStackAlign.ll
@@ -1,7 +1,7 @@
 ; Check that stack alignment can be forced. Individual targets should test their
 ; specific implementation details.
 
-; RUN: llc < %s -stackrealign -stack-alignment=32 | FileCheck %s
+; RUN: llc < %s -stackrealign | FileCheck %s
 ; CHECK-LABEL: @f
 ; CHECK-LABEL: @g
 
@@ -25,3 +25,6 @@ if.then:
 }
 
 declare void @llvm.memset.p0i8.i32(i8*, i8, i32, i1) nounwind
+
+!llvm.module.flags = !{!0}
+!0 = !{i32 2, !"override-stack-alignment", i32 32}
diff --git a/llvm/test/CodeGen/X86/base-pointer-and-cmpxchg.ll b/llvm/test/CodeGen/X86/base-pointer-and-cmpxchg.ll
@@ -1,7 +1,7 @@
-; RUN: llc -mtriple=x86_64-apple-macosx -mattr=+cx16 -x86-use-base-pointer=true -stackrealign -stack-alignment=32  %s -o - | FileCheck --check-prefix=CHECK --check-prefix=USE_BASE --check-prefix=USE_BASE_64 %s
-; RUN: llc -mtriple=x86_64-apple-macosx -mattr=+cx16 -x86-use-base-pointer=false -stackrealign -stack-alignment=32  %s -o - | FileCheck --check-prefix=CHECK --check-prefix=DONT_USE_BASE %s
-; RUN: llc -mtriple=x86_64-linux-gnux32 -mattr=+cx16 -x86-use-base-pointer=true -stackrealign -stack-alignment=32  %s -o - | FileCheck --check-prefix=CHECK --check-prefix=USE_BASE --check-prefix=USE_BASE_32 %s
-; RUN: llc -mtriple=x86_64-linux-gnux32 -mattr=+cx16 -x86-use-base-pointer=false -stackrealign -stack-alignment=32  %s -o - | FileCheck --check-prefix=CHECK --check-prefix=DONT_USE_BASE %s
+; RUN: llc -mtriple=x86_64-apple-macosx -mattr=+cx16 -x86-use-base-pointer=true -stackrealign %s -o - | FileCheck --check-prefix=CHECK --check-prefix=USE_BASE --check-prefix=USE_BASE_64 %s
+; RUN: llc -mtriple=x86_64-apple-macosx -mattr=+cx16 -x86-use-base-pointer=false -stackrealign %s -o - | FileCheck --check-prefix=CHECK --check-prefix=DONT_USE_BASE %s
+; RUN: llc -mtriple=x86_64-linux-gnux32 -mattr=+cx16 -x86-use-base-pointer=true -stackrealign %s -o - | FileCheck --check-prefix=CHECK --check-prefix=USE_BASE --check-prefix=USE_BASE_32 %s
+; RUN: llc -mtriple=x86_64-linux-gnux32 -mattr=+cx16 -x86-use-base-pointer=false -stackrealign %s -o - | FileCheck --check-prefix=CHECK --check-prefix=DONT_USE_BASE %s
 
 ; This function uses dynamic allocated stack to force the use
 ; of a frame pointer.
@@ -49,3 +49,5 @@ tail call void asm sideeffect "nop", "~{rax},~{rcx},~{rdx},~{rsi},~{rdi},~{rbp},
   store i32 %n, i32* %idx
   ret i1 %res
 }
+!llvm.module.flags = !{!0}
+!0 = !{i32 2, !"override-stack-alignment", i32 32}
diff --git a/llvm/test/CodeGen/X86/base-pointer-and-mwaitx.ll b/llvm/test/CodeGen/X86/base-pointer-and-mwaitx.ll
@@ -1,7 +1,11 @@
-; RUN: llc -mtriple=x86_64-pc-linux-gnu -mattr=+mwaitx -x86-use-base-pointer=true -stackrealign -stack-alignment=32  %s -o - | FileCheck --check-prefix=CHECK --check-prefix=USE_BASE_64 %s
-; RUN: llc -mtriple=x86_64-pc-linux-gnux32 -mattr=+mwaitx -x86-use-base-pointer=true -stackrealign -stack-alignment=32  %s -o - | FileCheck --check-prefix=CHECK --check-prefix=USE_BASE_32 %s
-; RUN: llc -mtriple=x86_64-pc-linux-gnu -mattr=+mwaitx -x86-use-base-pointer=true  %s -o - | FileCheck --check-prefix=CHECK --check-prefix=NO_BASE_64 %s
-; RUN: llc -mtriple=x86_64-pc-linux-gnux32 -mattr=+mwaitx -x86-use-base-pointer=true  %s -o - | FileCheck --check-prefix=CHECK --check-prefix=NO_BASE_32 %s
+; RUN: split-file %s %t
+; RUN: cat %t/main.ll %t/_align32.ll > %t/align32.ll
+; RUN: llc -mtriple=x86_64-pc-linux-gnu -mattr=+mwaitx -x86-use-base-pointer=true -stackrealign %t/align32.ll -o - | FileCheck --check-prefix=CHECK --check-prefix=USE_BASE_64 %s
+; RUN: llc -mtriple=x86_64-pc-linux-gnux32 -mattr=+mwaitx -x86-use-base-pointer=true -stackrealign %t/align32.ll -o - | FileCheck --check-prefix=CHECK --check-prefix=USE_BASE_32 %s
+; RUN: llc -mtriple=x86_64-pc-linux-gnu -mattr=+mwaitx -x86-use-base-pointer=true %t/main.ll -o - | FileCheck --check-prefix=CHECK --check-prefix=NO_BASE_64 %s
+; RUN: llc -mtriple=x86_64-pc-linux-gnux32 -mattr=+mwaitx -x86-use-base-pointer=true %t/main.ll -o - | FileCheck --check-prefix=CHECK --check-prefix=NO_BASE_32 %s
+
+;--- main.ll
 
 ; This test checks that we save and restore the base pointer (ebx or rbx) in the
 ; presence of the mwaitx intrinsic which requires to use ebx for one of its
@@ -208,3 +212,6 @@ if.end:
 ; NO_BASE_32-NEXT: {{.+$}}
 
 declare void @llvm.x86.mwaitx(i32, i32, i32) nounwind
+;--- _align32.ll
+!llvm.module.flags = !{!0}
+!0 = !{i32 2, !"override-stack-alignment", i32 32}
diff --git a/llvm/test/CodeGen/X86/dynamic-allocas-VLAs-stack-align.ll b/llvm/test/CodeGen/X86/dynamic-allocas-VLAs-stack-align.ll
@@ -0,0 +1,46 @@
+; RUN: llc < %s -stack-symbol-ordering=0 -mcpu=generic -stackrealign -mattr=+avx -mtriple=x86_64-apple-darwin10 | FileCheck %s
+; rdar://11496434
+declare void @t1_helper(i32*)
+declare void @t3_helper(i32*, i32*)
+
+; Test when forcing stack alignment
+define i32 @t8() nounwind uwtable {
+entry:
+  %a = alloca i32, align 4
+  call void @t1_helper(i32* %a) nounwind
+  %0 = load i32, i32* %a, align 4
+  %add = add nsw i32 %0, 13
+  ret i32 %add
+
+; CHECK: _t8
+; CHECK:      movq %rsp, %rbp
+; CHECK:      andq $-32, %rsp
+; CHECK-NEXT: subq $32, %rsp
+; CHECK:      movq %rbp, %rsp
+; CHECK:      popq %rbp
+}
+
+; VLAs
+define i32 @t9(i64 %sz) nounwind uwtable {
+entry:
+  %a = alloca i32, align 4
+  %vla = alloca i32, i64 %sz, align 16
+  call void @t3_helper(i32* %a, i32* %vla) nounwind
+  %0 = load i32, i32* %a, align 4
+  %add = add nsw i32 %0, 13
+  ret i32 %add
+
+; CHECK: _t9
+; CHECK: pushq %rbp
+; CHECK: movq %rsp, %rbp
+; CHECK: pushq %rbx
+; CHECK: andq $-32, %rsp
+; CHECK: subq $32, %rsp
+; CHECK: movq %rsp, %rbx
+
+; CHECK: leaq -8(%rbp), %rsp
+; CHECK: popq %rbx
+; CHECK: popq %rbp
+}
+!llvm.module.flags = !{!0}
+!0 = !{i32 2, !"override-stack-alignment", i32 32}
diff --git a/llvm/test/CodeGen/X86/dynamic-allocas-VLAs.ll b/llvm/test/CodeGen/X86/dynamic-allocas-VLAs.ll
@@ -1,5 +1,4 @@
 ; RUN: llc < %s -stack-symbol-ordering=0 -mcpu=generic -mattr=+avx -mtriple=x86_64-apple-darwin10 | FileCheck %s
-; RUN: llc < %s -stack-symbol-ordering=0 -mcpu=generic -stackrealign -stack-alignment=32 -mattr=+avx -mtriple=x86_64-apple-darwin10 | FileCheck %s -check-prefix=FORCE-ALIGN
 ; rdar://11496434
 
 ; no VLAs or dynamic alignment
@@ -184,44 +183,3 @@ declare i8* @llvm.stacksave() nounwind
 declare void @bar(i32, i32*, %struct.struct_t* byval(%struct.struct_t) align 8)
 
 declare void @llvm.stackrestore(i8*) nounwind
-
-
-; Test when forcing stack alignment
-define i32 @t8() nounwind uwtable {
-entry:
-  %a = alloca i32, align 4
-  call void @t1_helper(i32* %a) nounwind
-  %0 = load i32, i32* %a, align 4
-  %add = add nsw i32 %0, 13
-  ret i32 %add
-
-; FORCE-ALIGN: _t8
-; FORCE-ALIGN:      movq %rsp, %rbp
-; FORCE-ALIGN:      andq $-32, %rsp
-; FORCE-ALIGN-NEXT: subq $32, %rsp
-; FORCE-ALIGN:      movq %rbp, %rsp
-; FORCE-ALIGN:      popq %rbp
-}
-
-; VLAs
-define i32 @t9(i64 %sz) nounwind uwtable {
-entry:
-  %a = alloca i32, align 4
-  %vla = alloca i32, i64 %sz, align 16
-  call void @t3_helper(i32* %a, i32* %vla) nounwind
-  %0 = load i32, i32* %a, align 4
-  %add = add nsw i32 %0, 13
-  ret i32 %add
-
-; FORCE-ALIGN: _t9
-; FORCE-ALIGN: pushq %rbp
-; FORCE-ALIGN: movq %rsp, %rbp
-; FORCE-ALIGN: pushq %rbx
-; FORCE-ALIGN: andq $-32, %rsp
-; FORCE-ALIGN: subq $32, %rsp
-; FORCE-ALIGN: movq %rsp, %rbx
-
-; FORCE-ALIGN: leaq -8(%rbp), %rsp
-; FORCE-ALIGN: popq %rbx
-; FORCE-ALIGN: popq %rbp
-}
diff --git a/llvm/test/CodeGen/X86/force-align-stack-alloca.ll b/llvm/test/CodeGen/X86/force-align-stack-alloca.ll
@@ -3,7 +3,7 @@
 ; arbitrarily force alignment up to 32-bytes for i386 hoping that this will
 ; exceed any ABI provisions.
 ;
-; RUN: llc < %s -mcpu=generic -stackrealign -stack-alignment=32 | FileCheck %s
+; RUN: llc < %s -mcpu=generic -stackrealign | FileCheck %s
 
 target datalayout = "e-p:32:32:32-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:32:64-f32:32:32-f64:32:64-v64:64:64-v128:128:128-a0:0:64-f80:32:32-n8:16:32-S128"
 target triple = "i386-unknown-linux-gnu"
@@ -74,3 +74,6 @@ if.then:
 }
 
 declare void @llvm.memset.p0i8.i32(i8*, i8, i32, i1) nounwind
+
+!llvm.module.flags = !{!0}
+!0 = !{i32 2, !"override-stack-alignment", i32 32}
diff --git a/llvm/test/CodeGen/X86/hipe-cc.ll b/llvm/test/CodeGen/X86/hipe-cc.ll
@@ -1,4 +1,4 @@
-; RUN: llc < %s -stack-symbol-ordering=0 -tailcallopt -code-model=medium -stack-alignment=4 -mtriple=i686-linux-gnu -mcpu=pentium | FileCheck %s
+; RUN: llc < %s -stack-symbol-ordering=0 -tailcallopt -code-model=medium -mtriple=i686-linux-gnu -mcpu=pentium | FileCheck %s
 
 ; Check the HiPE calling convention works (x86-32)
 
@@ -89,3 +89,6 @@ define cc 11 { i32, i32, i32 } @tailcaller(i32 %hp, i32 %p) nounwind {
 @clos = external dso_local constant i32
 declare cc 11 void @bar(i32, i32, i32, i32, i32)
 declare cc 11 { i32, i32, i32 } @tailcallee(i32, i32, i32, i32, i32, i32)
+
+!llvm.module.flags = !{!3}
+!3 = !{i32 2, !"override-stack-alignment", i32 4}
diff --git a/llvm/test/CodeGen/X86/hipe-cc64.ll b/llvm/test/CodeGen/X86/hipe-cc64.ll
@@ -1,4 +1,4 @@
-; RUN: llc < %s -stack-symbol-ordering=0 -tailcallopt -relocation-model=static -code-model=medium -stack-alignment=8 -mtriple=x86_64-linux-gnu -mcpu=opteron | FileCheck %s
+; RUN: llc < %s -stack-symbol-ordering=0 -tailcallopt -relocation-model=static -code-model=medium -mtriple=x86_64-linux-gnu -mcpu=opteron | FileCheck %s
 
 ; Check the HiPE calling convention works (x86-64)
 
@@ -100,3 +100,5 @@ define cc 11 { i64, i64, i64 } @tailcaller(i64 %hp, i64 %p) #0 {
 @clos = external constant i64
 declare cc 11 void @bar(i64, i64, i64, i64, i64, i64)
 declare cc 11 { i64, i64, i64 } @tailcallee(i64, i64, i64, i64, i64, i64, i64)
+!llvm.module.flags = !{!3}
+!3 = !{i32 2, !"override-stack-alignment", i32 8}
diff --git a/llvm/test/CodeGen/X86/movtopush-stack-align.ll b/llvm/test/CodeGen/X86/movtopush-stack-align.ll
@@ -0,0 +1,52 @@
+; RUN: llc < %s -mtriple=i686-windows -stackrealign | FileCheck %s
+
+declare void @good(i32 %a, i32 %b, i32 %c, i32 %d)
+declare void @oneparam(i32 %a)
+declare void @eightparams(i32 %a, i32 %b, i32 %c, i32 %d, i32 %e, i32 %f, i32 %g, i32 %h)
+
+; When there is no reserved call frame, check that additional alignment
+; is added when the pushes don't add up to the required alignment.
+; CHECK-LABEL: test5:
+; CHECK: subl    $16, %esp
+; CHECK-NEXT: pushl   $4
+; CHECK-NEXT: pushl   $3
+; CHECK-NEXT: pushl   $2
+; CHECK-NEXT: pushl   $1
+; CHECK-NEXT: call
+define void @test5(i32 %k) {
+entry:
+  %a = alloca i32, i32 %k
+  call void @good(i32 1, i32 2, i32 3, i32 4)
+  ret void
+}
+
+; When the alignment adds up, do the transformation
+; CHECK-LABEL: test5b:
+; CHECK: pushl   $8
+; CHECK-NEXT: pushl   $7
+; CHECK-NEXT: pushl   $6
+; CHECK-NEXT: pushl   $5
+; CHECK-NEXT: pushl   $4
+; CHECK-NEXT: pushl   $3
+; CHECK-NEXT: pushl   $2
+; CHECK-NEXT: pushl   $1
+; CHECK-NEXT: call
+define void @test5b() optsize {
+entry:
+  call void @eightparams(i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8)
+  ret void
+}
+
+; When having to compensate for the alignment isn't worth it,
+; don't use pushes.
+; CHECK-LABEL: test5c:
+; CHECK: movl $1, (%esp)
+; CHECK-NEXT: call
+define void @test5c() optsize {
+entry:
+  call void @oneparam(i32 1)
+  ret void
+}
+
+!llvm.module.flags = !{!0}
+!0 = !{i32 2, !"override-stack-alignment", i32 32}