[VE] Target-specific bit size for sjljehprepare

[simoll]

Summary

This patch extends the TargetMachine to let targets specify the integer size used by the sjljehprepare pass. This is 64bit for the VE target and otherwise defaults to 32bit for all targets, which was hard-wired before.

[simoll]

To clarify: this patch is not intended to be committed as-is but gives an overview of the required changes. It's the a patch of a series for the VE backend and includes all the earlier patches.

[simoll]

Trimmed down to SjLj changes only.

[simoll]

• Rebased
• NFC (getArch() == .. -- > isVE()).
• Dug through layers of cosmetic commits to find potential reviewers for the SjLjEHPreparePass.

[simoll]

Rebased

[merge_guards_bot]

Unit tests: pass. 62043 tests passed, 0 failed and 783 were skipped.

clang-tidy: unknown.

clang-format: fail. Please format your changes with clang-format by running git-clang-format HEAD^ or applying this patch.

Build artifacts: diff.json, clang-format.patch, CMakeCache.txt, console-log.txt, test-results.xml

[arsenm]

llvm/lib/CodeGen/TargetPassConfig.cpp

698–699

I would expect this to be a target hook of some kind the pass checks, not a pass parameter controlled here. Checking a target triple here is probably not kosher

[simoll]

llvm/lib/CodeGen/TargetPassConfig.cpp

698–699

Ok. How about querying the SjLj data size from TTI (eg TTI::getSjLjDataSize())?

[arsenm]

llvm/lib/CodeGen/TargetPassConfig.cpp

698–699

I think this probably belongs in either the TargetMachine or TargetLowering. I don't think things required for correctness should go in TTI

[simoll]

• Rebased
• Moved data size query to TargetMachine

[merge_guards_bot]

Unit tests: pass. 62057 tests passed, 0 failed and 784 were skipped.

clang-tidy: unknown.

clang-format: fail. Please format your changes with clang-format by running git-clang-format HEAD^ or applying this patch.

Build artifacts: diff.json, clang-format.patch, CMakeCache.txt, console-log.txt, test-results.xml

[simoll]

• Rebased
• Style pass

[merge_guards_bot]

Unit tests: pass. 62057 tests passed, 0 failed and 784 were skipped.

clang-tidy: unknown.

clang-format: pass.

Build artifacts: diff.json, clang-format.patch, CMakeCache.txt, console-log.txt, test-results.xml

[arsenm]

Needs test

[simoll]

• test

[simoll]

• initialize SjljEHPrepare pass with CodeGen library.
• rebased
• added test (32bit on X86, 64bit on VE).

[simoll]

ping?

[arsenm]

llvm/test/CodeGen/VE/sjlj_except.ll

2–3

You should be able to test this with just opt running the pass? You certainly don't need to rely on -print-after

14

I don't know anything about SJLJ but I would guess it's possible to shrink this test more?

33

Weird indent?

[simoll]

Thanks for having a look :)

llvm/test/CodeGen/VE/sjlj_except.ll

2–3

I was hoping to do that. However, when the pass is created without a target machine (as happens in opt) - it defaults to the standard bit size. Since VE is the only target that actually uses a non-default value, i found no other way to test it than to go all the way to VE codegen.

14

Probably. I'll look into it.

[simoll]

• rebased
• simplified test

[simoll]

ping. Any more comments for this one?

[arsenm]

Maybe this should go in TLI instead, but it's not really important

llvm/include/llvm/CodeGen/Passes.h

/// MachineDominanaceFrontier - This pass is a machine dominators analysis pass.
 
   /// createWinEHPass - Prepares personality functions used by MSVC on Windows,
   /// in addition to the Itanium LSDA based personalities.
   FunctionPass *createWinEHPass(bool DemoteCatchSwitchPHIOnly = false);
 
   /// createSjLjEHPreparePass - This pass adapts exception handling code to use
   /// the GCC-style builtin setjmp/longjmp (sjlj) to handling EH control flow.
   ///
-  FunctionPass *createSjLjEHPreparePass();
+  FunctionPass *createSjLjEHPreparePass(const TargetMachine *TM);
 
   /// createWasmEHPass - This pass adapts exception handling code to use
   /// WebAssembly's exception handling scheme.
   FunctionPass *createWasmEHPass();
 
   /// LocalStackSlotAllocation - This pass assigns local frame indices to stack
   /// slots relative to one another and allocates base registers to access them
   /// when it is estimated by the target to be out of range of normal frame

llvm/include/llvm/Target/TargetMachine.h

public:
   /// Branch relaxation, which must happen after block placement, can
   /// on some targets (e.g. SystemZ) expose additional post-RA
   /// scheduling opportunities.
   virtual bool targetSchedulesPostRAScheduling() const { return false; };
 
   void getNameWithPrefix(SmallVectorImpl<char> &Name, const GlobalValue *GV,
                          Mangler &Mang, bool MayAlwaysUsePrivate = false) const;
   MCSymbol *getSymbol(const GlobalValue *GV) const;
+
+  /// The integer bit size to use for SjLj based exception handling.
+  static constexpr unsigned DefaultSjLjDataSize = 32;
+  virtual unsigned getSjLjDataSize() const { return DefaultSjLjDataSize; }
 };
 
 /// This class describes a target machine that is implemented with the LLVM
 /// target-independent code generator.
 ///
 class LLVMTargetMachine : public TargetMachine {
 protected: // Can only create subclasses.
   LLVMTargetMachine(const Target &T, StringRef DataLayoutString,

llvm/lib/CodeGen/CodeGen.cpp

void llvm::initializeCodeGen(PassRegistry &Registry) {
   initializeRegAllocFastPass(Registry);
   initializeRegUsageInfoCollectorPass(Registry);
   initializeRegUsageInfoPropagationPass(Registry);
   initializeRegisterCoalescerPass(Registry);
   initializeRenameIndependentSubregsPass(Registry);
   initializeSafeStackLegacyPassPass(Registry);
   initializeScalarizeMaskedMemIntrinPass(Registry);
   initializeShrinkWrapPass(Registry);
+  initializeSjLjEHPreparePass(Registry);
   initializeSlotIndexesPass(Registry);
   initializeStackColoringPass(Registry);
   initializeStackMapLivenessPass(Registry);
   initializeStackProtectorPass(Registry);
   initializeStackSlotColoringPass(Registry);
   initializeTailDuplicatePass(Registry);
   initializeTargetPassConfigPass(Registry);
   initializeTwoAddressInstructionPassPass(Registry);

llvm/lib/CodeGen/SjLjEHPrepare.cpp

 #include "llvm/IR/IRBuilder.h"
 #include "llvm/IR/Instructions.h"
 #include "llvm/IR/Intrinsics.h"
 #include "llvm/IR/Module.h"
 #include "llvm/InitializePasses.h"
 #include "llvm/Pass.h"
 #include "llvm/Support/Debug.h"
 #include "llvm/Support/raw_ostream.h"
+#include "llvm/Target/TargetMachine.h"
 #include "llvm/Transforms/Utils/Local.h"
 using namespace llvm;
 
 #define DEBUG_TYPE "sjljehprepare"
 
 STATISTIC(NumInvokes, "Number of invokes replaced");
 STATISTIC(NumSpilled, "Number of registers live across unwind edges");
 
 namespace {
 class SjLjEHPrepare : public FunctionPass {
+  IntegerType *DataTy;
   Type *doubleUnderDataTy;
   Type *doubleUnderJBufTy;
   Type *FunctionContextTy;
   FunctionCallee RegisterFn;
   FunctionCallee UnregisterFn;
   Function *BuiltinSetupDispatchFn;
   Function *FrameAddrFn;
   Function *StackAddrFn;
   Function *StackRestoreFn;
   Function *LSDAAddrFn;
   Function *CallSiteFn;
   Function *FuncCtxFn;
   AllocaInst *FuncCtx;
+  const TargetMachine *TM;
 
 public:
   static char ID; // Pass identification, replacement for typeid
-  explicit SjLjEHPrepare() : FunctionPass(ID) {}
+  explicit SjLjEHPrepare(const TargetMachine *TM = nullptr)
+      : FunctionPass(ID), TM(TM) {}
   bool doInitialization(Module &M) override;
   bool runOnFunction(Function &F) override;
 
   void getAnalysisUsage(AnalysisUsage &AU) const override {}
   StringRef getPassName() const override {
     return "SJLJ Exception Handling preparation";
   }
 
 private:
   bool setupEntryBlockAndCallSites(Function &F);
   void substituteLPadValues(LandingPadInst *LPI, Value *ExnVal, Value *SelVal);
   Value *setupFunctionContext(Function &F, ArrayRef<LandingPadInst *> LPads);
   void lowerIncomingArguments(Function &F);
   void lowerAcrossUnwindEdges(Function &F, ArrayRef<InvokeInst *> Invokes);
   void insertCallSiteStore(Instruction *I, int Number);
 };
 } // end anonymous namespace
 
 char SjLjEHPrepare::ID = 0;
 INITIALIZE_PASS(SjLjEHPrepare, DEBUG_TYPE, "Prepare SjLj exceptions",
                 false, false)
 
 // Public Interface To the SjLjEHPrepare pass.
-FunctionPass *llvm::createSjLjEHPreparePass() { return new SjLjEHPrepare(); }
+FunctionPass *llvm::createSjLjEHPreparePass(const TargetMachine *TM) {
+  return new SjLjEHPrepare(TM);
+}
+
 // doInitialization - Set up decalarations and types needed to process
 // exceptions.
 bool SjLjEHPrepare::doInitialization(Module &M) {
   // Build the function context structure.
   // builtin_setjmp uses a five word jbuf
   Type *VoidPtrTy = Type::getInt8PtrTy(M.getContext());
-  Type *Int32Ty = Type::getInt32Ty(M.getContext());
-  doubleUnderDataTy = ArrayType::get(Int32Ty, 4);
+  unsigned DataBits =
+      TM ? TM->getSjLjDataSize() : TargetMachine::DefaultSjLjDataSize;
+  DataTy = Type::getIntNTy(M.getContext(), DataBits);
+  doubleUnderDataTy = ArrayType::get(DataTy, 4);
   doubleUnderJBufTy = ArrayType::get(VoidPtrTy, 5);
   FunctionContextTy = StructType::get(VoidPtrTy,         // __prev
-                                      Int32Ty,           // call_site
+                                      DataTy,            // call_site
                                       doubleUnderDataTy, // __data
                                       VoidPtrTy,         // __personality
                                       VoidPtrTy,         // __lsda
                                       doubleUnderJBufTy  // __jbuf
-                                      );
 
   return true;
 }
 
 /// insertCallSiteStore - Insert a store of the call-site value to the
 /// function context
 void SjLjEHPrepare::insertCallSiteStore(Instruction *I, int Number) {
   IRBuilder<> Builder(I);
 
   // Get a reference to the call_site field.
   Type *Int32Ty = Type::getInt32Ty(I->getContext());
   Value *Zero = ConstantInt::get(Int32Ty, 0);
   Value *One = ConstantInt::get(Int32Ty, 1);
   Value *Idxs[2] = { Zero, One };
   Value *CallSite =
       Builder.CreateGEP(FunctionContextTy, FuncCtx, Idxs, "call_site");
 
   // Insert a store of the call-site number
-  ConstantInt *CallSiteNoC =
-      ConstantInt::get(Type::getInt32Ty(I->getContext()), Number);
+  ConstantInt *CallSiteNoC = ConstantInt::get(DataTy, Number);
   Builder.CreateStore(CallSiteNoC, CallSite, true /*volatile*/);
 }
 
 /// MarkBlocksLiveIn - Insert BB and all of its predecessors into LiveBBs until
 /// we reach blocks we've already seen.
 static void MarkBlocksLiveIn(BasicBlock *BB,
                              SmallPtrSetImpl<BasicBlock *> &LiveBBs) {
   if (!LiveBBs.insert(BB).second)
     return; // already been here.
 
   df_iterator_default_set<BasicBlock*> Visited;
 
   for (BasicBlock *B : inverse_depth_first_ext(BB, Visited))
     LiveBBs.insert(B);
-
 }
 
 /// substituteLPadValues - Substitute the values returned by the landingpad
 /// instruction with those returned by the personality function.
 void SjLjEHPrepare::substituteLPadValues(LandingPadInst *LPI, Value *ExnVal,
                                          Value *SelVal) {
   SmallVector<Value *, 8> UseWorkList(LPI->user_begin(), LPI->user_end());
   while (!UseWorkList.empty()) {
for (LandingPadInst *LPI : LPads) {
     IRBuilder<> Builder(LPI->getParent(),
                         LPI->getParent()->getFirstInsertionPt());
 
     // Reference the __data field.
     Value *FCData =
         Builder.CreateConstGEP2_32(FunctionContextTy, FuncCtx, 0, 2, "__data");
 
     // The exception values come back in context->__data[0].
-    Type *Int32Ty = Type::getInt32Ty(F.getContext());
     Value *ExceptionAddr = Builder.CreateConstGEP2_32(doubleUnderDataTy, FCData,
                                                       0, 0, "exception_gep");
-    Value *ExnVal = Builder.CreateLoad(Int32Ty, ExceptionAddr, true, "exn_val");
+    Value *ExnVal = Builder.CreateLoad(DataTy, ExceptionAddr, true, "exn_val");
     ExnVal = Builder.CreateIntToPtr(ExnVal, Builder.getInt8PtrTy());
 
     Value *SelectorAddr = Builder.CreateConstGEP2_32(doubleUnderDataTy, FCData,
                                                      0, 1, "exn_selector_gep");
     Value *SelVal =
-        Builder.CreateLoad(Int32Ty, SelectorAddr, true, "exn_selector_val");
+        Builder.CreateLoad(DataTy, SelectorAddr, true, "exn_selector_val");
+
+    // SelVal must be Int32Ty, so trunc it
+    SelVal = Builder.CreateTrunc(SelVal, Type::getInt32Ty(F.getContext()));
 
     substituteLPadValues(LPI, ExnVal, SelVal);
   }
 
   // Personality function
   IRBuilder<> Builder(EntryBB->getTerminator());
   Value *PersonalityFn = F.getPersonalityFn();
   Value *PersonalityFieldPtr = Builder.CreateConstGEP2_32(

llvm/lib/CodeGen/TargetPassConfig.cpp

void TargetPassConfig::addPassesToHandleExceptions() {
   switch (MCAI->getExceptionHandlingType()) {
   case ExceptionHandling::SjLj:
     // SjLj piggy-backs on dwarf for this bit. The cleanups done apply to both
     // Dwarf EH prepare needs to be run after SjLj prepare. Otherwise,
     // catch info can get misplaced when a selector ends up more than one block
     // removed from the parent invoke(s). This could happen when a landing
     // pad is shared by multiple invokes and is also a target of a normal
     // edge from elsewhere.
-    addPass(createSjLjEHPreparePass());
+    addPass(createSjLjEHPreparePass(TM));
     LLVM_FALLTHROUGH;
   case ExceptionHandling::DwarfCFI:
   case ExceptionHandling::ARM:
     addPass(createDwarfEHPass());
     break;
   case ExceptionHandling::WinEH:
     // We support using both GCC-style and MSVC-style exceptions on Windows, so
     // add both preparation passes. Each pass will only actually run if it
     // recognizes the personality function.

llvm/lib/Target/VE/VETargetMachine.h

public:
   TargetPassConfig *createPassConfig(PassManagerBase &PM) override;
   TargetLoweringObjectFile *getObjFileLowering() const override {
     return TLOF.get();
   }
 
   bool isMachineVerifierClean() const override { return false; }
 
   TargetTransformInfo getTargetTransformInfo(const Function &F) override;
+
+  unsigned getSjLjDataSize() const override { return 64; }
 };
 
 } // namespace llvm
 
 #endif

llvm/test/CodeGen/VE/sjlj_except.ll

+; RUN: llc  -mtriple=x86_64-unknown-unknown --exception-model=sjlj --print-after=sjljehprepare < %s |& FileCheck --check-prefix=CHECK-X86 %s
+; RUN: (llc  -mtriple=ve-unknown-unknown --exception-model=sjlj  --print-after=sjljehprepare < %s || true) |& FileCheck --check-prefix=CHECK-VE %s
+
+@SomeGlobal = external dso_local global i8
+
+define dso_local i32 @foo(i32 %arg) local_unnamed_addr personality i8* bitcast (i32 (...)* @__gxx_personality_sj0 to i8*) {
+; CHECK-VE: *** IR Dump After SJLJ Exception Handling preparation ***
+; CHECK-VE-NEXT: define dso_local i32 @foo(i32 %arg) local_unnamed_addr personality i8* bitcast (i32 (...)* @__gxx_personality_sj0 to i8*) {
+; CHECK-VE-NEXT: entry:
+; CHECK-VE-NEXT:   %fn_context = alloca { i8*, i64, [4 x i64], i8*, i8*, [5 x i8*] }, align 8
+; CHECK-VE-NEXT:   %arg.tmp = select i1 true, i32 %arg, i32 undef
+; CHECK-VE-NEXT:   %pers_fn_gep = getelementptr { i8*, i64, [4 x i64], i8*, i8*, [5 x i8*] }, { i8*, i64, [4 x i64], i8*, i8*, [5 x i8*] }* %fn_context, i32 0, i32 3
+; CHECK-X86: *** IR Dump After SJLJ Exception Handling preparation ***
+; CHECK-X86-NEXT: define dso_local i32 @foo(i32 %arg) local_unnamed_addr personality i8* bitcast (i32 (...)* @__gxx_personality_sj0 to i8*) {
+; CHECK-X86-NEXT: entry:
+; CHECK-X86-NEXT:   %fn_context = alloca { i8*, i32, [4 x i32], i8*, i8*, [5 x i8*] }, align 8
+; CHECK-X86-NEXT:   %arg.tmp = select i1 true, i32 %arg, i32 undef
+; CHECK-X86-NEXT:   %pers_fn_gep = getelementptr { i8*, i32, [4 x i32], i8*, i8*, [5 x i8*] }, { i8*, i32, [4 x i32], i8*, i8*, [5 x i8*] }* %fn_context, i32 0, i32 3
+entry:
+  invoke void @errorbar() to label %exit unwind label %handle
+
+handle:
+  %error = landingpad { i8*, i32 } catch i8* @SomeGlobal
+  ret i32 1
+
+exit:
+  ret i32 0
+}
+
+declare dso_local void @errorbar() local_unnamed_addr
+
+declare dso_local i32 @__gxx_personality_sj0(...)