[llvm] r287178 - Introduce GlobalSplit pass.

Davide Italiano via llvm-commits llvm-commits at lists.llvm.org
Wed Nov 16 16:13:10 PST 2016


On Wed, Nov 16, 2016 at 3:40 PM, Peter Collingbourne via llvm-commits
<llvm-commits at lists.llvm.org> wrote:
> Author: pcc
> Date: Wed Nov 16 17:40:26 2016
> New Revision: 287178
>
> URL: http://llvm.org/viewvc/llvm-project?rev=287178&view=rev
> Log:
> Introduce GlobalSplit pass.
>
> This pass splits globals into elements using inrange annotations on
> getelementptr indices.
>

Sorry, I missed this review. As this is hooked in the default LTO
pipeline, I think it makes sense to provide a port of this pass to the
new pass manager.
While the port should be fairly straightforward, it's not very high priority.
We (internally) periodically try to run the LTO pipeline with the new
PM, so if you won't get to it, I'll try to take a look and send a
review the next time I'll experiment.

Thanks,

> Differential Revision: https://reviews.llvm.org/D22295
>
> Added:
>     llvm/trunk/lib/Transforms/IPO/GlobalSplit.cpp
>     llvm/trunk/test/Transforms/GlobalSplit/
>     llvm/trunk/test/Transforms/GlobalSplit/basic.ll
>     llvm/trunk/test/Transforms/GlobalSplit/non-beneficial.ll
>     llvm/trunk/test/Transforms/GlobalSplit/nonlocal.ll
> Modified:
>     llvm/trunk/include/llvm/InitializePasses.h
>     llvm/trunk/include/llvm/Transforms/IPO.h
>     llvm/trunk/lib/Transforms/IPO/CMakeLists.txt
>     llvm/trunk/lib/Transforms/IPO/IPO.cpp
>     llvm/trunk/lib/Transforms/IPO/PassManagerBuilder.cpp
>
> Modified: llvm/trunk/include/llvm/InitializePasses.h
> URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/include/llvm/InitializePasses.h?rev=287178&r1=287177&r2=287178&view=diff
> ==============================================================================
> --- llvm/trunk/include/llvm/InitializePasses.h (original)
> +++ llvm/trunk/include/llvm/InitializePasses.h Wed Nov 16 17:40:26 2016
> @@ -145,6 +145,7 @@ void initializeGVNLegacyPassPass(PassReg
>  void initializeGlobalDCELegacyPassPass(PassRegistry&);
>  void initializeGlobalMergePass(PassRegistry&);
>  void initializeGlobalOptLegacyPassPass(PassRegistry&);
> +void initializeGlobalSplitPass(PassRegistry&);
>  void initializeGlobalsAAWrapperPassPass(PassRegistry&);
>  void initializeGuardWideningLegacyPassPass(PassRegistry&);
>  void initializeIPCPPass(PassRegistry&);
>
> Modified: llvm/trunk/include/llvm/Transforms/IPO.h
> URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/include/llvm/Transforms/IPO.h?rev=287178&r1=287177&r2=287178&view=diff
> ==============================================================================
> --- llvm/trunk/include/llvm/Transforms/IPO.h (original)
> +++ llvm/trunk/include/llvm/Transforms/IPO.h Wed Nov 16 17:40:26 2016
> @@ -225,6 +225,10 @@ ModulePass *createCrossDSOCFIPass();
>  /// metadata.
>  ModulePass *createWholeProgramDevirtPass();
>
> +/// This pass splits globals into pieces for the benefit of whole-program
> +/// devirtualization and control-flow integrity.
> +ModulePass *createGlobalSplitPass();
> +
>  //===----------------------------------------------------------------------===//
>  // SampleProfilePass - Loads sample profile data from disk and generates
>  // IR metadata to reflect the profile.
>
> Modified: llvm/trunk/lib/Transforms/IPO/CMakeLists.txt
> URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Transforms/IPO/CMakeLists.txt?rev=287178&r1=287177&r2=287178&view=diff
> ==============================================================================
> --- llvm/trunk/lib/Transforms/IPO/CMakeLists.txt (original)
> +++ llvm/trunk/lib/Transforms/IPO/CMakeLists.txt Wed Nov 16 17:40:26 2016
> @@ -12,6 +12,7 @@ add_llvm_library(LLVMipo
>    FunctionImport.cpp
>    GlobalDCE.cpp
>    GlobalOpt.cpp
> +  GlobalSplit.cpp
>    IPConstantPropagation.cpp
>    IPO.cpp
>    InferFunctionAttrs.cpp
>
> Added: llvm/trunk/lib/Transforms/IPO/GlobalSplit.cpp
> URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Transforms/IPO/GlobalSplit.cpp?rev=287178&view=auto
> ==============================================================================
> --- llvm/trunk/lib/Transforms/IPO/GlobalSplit.cpp (added)
> +++ llvm/trunk/lib/Transforms/IPO/GlobalSplit.cpp Wed Nov 16 17:40:26 2016
> @@ -0,0 +1,164 @@
> +//===- GlobalSplit.cpp - global variable splitter -------------------------===//
> +//
> +//                     The LLVM Compiler Infrastructure
> +//
> +// This file is distributed under the University of Illinois Open Source
> +// License. See LICENSE.TXT for details.
> +//
> +//===----------------------------------------------------------------------===//
> +//
> +// This pass uses inrange annotations on GEP indices to split globals where
> +// beneficial. Clang currently attaches these annotations to references to
> +// virtual table globals under the Itanium ABI for the benefit of the
> +// whole-program virtual call optimization and control flow integrity passes.
> +//
> +//===----------------------------------------------------------------------===//
> +
> +#include "llvm/Transforms/IPO.h"
> +#include "llvm/ADT/StringExtras.h"
> +#include "llvm/IR/Constants.h"
> +#include "llvm/IR/GlobalVariable.h"
> +#include "llvm/IR/Intrinsics.h"
> +#include "llvm/IR/Module.h"
> +#include "llvm/IR/Operator.h"
> +#include "llvm/Pass.h"
> +
> +#include <set>
> +
> +using namespace llvm;
> +
> +namespace {
> +
> +bool splitGlobal(GlobalVariable &GV) {
> +  // If the address of the global is taken outside of the module, we cannot
> +  // apply this transformation.
> +  if (!GV.hasLocalLinkage())
> +    return false;
> +
> +  // We currently only know how to split ConstantStructs.
> +  auto *Init = dyn_cast_or_null<ConstantStruct>(GV.getInitializer());
> +  if (!Init)
> +    return false;
> +
> +  // Verify that each user of the global is an inrange getelementptr constant.
> +  // From this it follows that any loads from or stores to that global must use
> +  // a pointer derived from an inrange getelementptr constant, which is
> +  // sufficient to allow us to apply the splitting transform.
> +  for (User *U : GV.users()) {
> +    if (!isa<Constant>(U))
> +      return false;
> +
> +    auto *GEP = dyn_cast<GEPOperator>(U);
> +    if (!GEP || !GEP->getInRangeIndex() || *GEP->getInRangeIndex() != 1 ||
> +        !isa<ConstantInt>(GEP->getOperand(1)) ||
> +        !cast<ConstantInt>(GEP->getOperand(1))->isZero() ||
> +        !isa<ConstantInt>(GEP->getOperand(2)))
> +      return false;
> +  }
> +
> +  SmallVector<MDNode *, 2> Types;
> +  GV.getMetadata(LLVMContext::MD_type, Types);
> +
> +  const DataLayout &DL = GV.getParent()->getDataLayout();
> +  const StructLayout *SL = DL.getStructLayout(Init->getType());
> +
> +  IntegerType *Int32Ty = Type::getInt32Ty(GV.getContext());
> +
> +  std::vector<GlobalVariable *> SplitGlobals(Init->getNumOperands());
> +  for (unsigned I = 0; I != Init->getNumOperands(); ++I) {
> +    // Build a global representing this split piece.
> +    auto *SplitGV =
> +        new GlobalVariable(*GV.getParent(), Init->getOperand(I)->getType(),
> +                           GV.isConstant(), GlobalValue::PrivateLinkage,
> +                           Init->getOperand(I), GV.getName() + "." + utostr(I));
> +    SplitGlobals[I] = SplitGV;
> +
> +    unsigned SplitBegin = SL->getElementOffset(I);
> +    unsigned SplitEnd = (I == Init->getNumOperands() - 1)
> +                            ? SL->getSizeInBytes()
> +                            : SL->getElementOffset(I + 1);
> +
> +    // Rebuild type metadata, adjusting by the split offset.
> +    // FIXME: See if we can use DW_OP_piece to preserve debug metadata here.
> +    for (MDNode *Type : Types) {
> +      uint64_t ByteOffset = cast<ConstantInt>(
> +              cast<ConstantAsMetadata>(Type->getOperand(0))->getValue())
> +              ->getZExtValue();
> +      if (ByteOffset < SplitBegin || ByteOffset >= SplitEnd)
> +        continue;
> +      SplitGV->addMetadata(
> +          LLVMContext::MD_type,
> +          *MDNode::get(GV.getContext(),
> +                       {ConstantAsMetadata::get(
> +                            ConstantInt::get(Int32Ty, ByteOffset - SplitBegin)),
> +                        Type->getOperand(1)}));
> +    }
> +  }
> +
> +  for (User *U : GV.users()) {
> +    auto *GEP = cast<GEPOperator>(U);
> +    unsigned I = cast<ConstantInt>(GEP->getOperand(2))->getZExtValue();
> +    if (I >= SplitGlobals.size())
> +      continue;
> +
> +    SmallVector<Value *, 4> Ops;
> +    Ops.push_back(ConstantInt::get(Int32Ty, 0));
> +    for (unsigned I = 3; I != GEP->getNumOperands(); ++I)
> +      Ops.push_back(GEP->getOperand(I));
> +
> +    auto *NewGEP = ConstantExpr::getGetElementPtr(
> +        SplitGlobals[I]->getInitializer()->getType(), SplitGlobals[I], Ops,
> +        GEP->isInBounds());
> +    GEP->replaceAllUsesWith(NewGEP);
> +  }
> +
> +  // Finally, remove the original global. Any remaining uses refer to invalid
> +  // elements of the global, so replace with undef.
> +  if (!GV.use_empty())
> +    GV.replaceAllUsesWith(UndefValue::get(GV.getType()));
> +  GV.eraseFromParent();
> +  return true;
> +}
> +
> +bool splitGlobals(Module &M) {
> +  // First, see if the module uses either of the llvm.type.test or
> +  // llvm.type.checked.load intrinsics, which indicates that splitting globals
> +  // may be beneficial.
> +  Function *TypeTestFunc =
> +      M.getFunction(Intrinsic::getName(Intrinsic::type_test));
> +  Function *TypeCheckedLoadFunc =
> +      M.getFunction(Intrinsic::getName(Intrinsic::type_checked_load));
> +  if ((!TypeTestFunc || TypeTestFunc->use_empty()) &&
> +      (!TypeCheckedLoadFunc || TypeCheckedLoadFunc->use_empty()))
> +    return false;
> +
> +  bool Changed = false;
> +  for (auto I = M.global_begin(); I != M.global_end();) {
> +    GlobalVariable &GV = *I;
> +    ++I;
> +    Changed |= splitGlobal(GV);
> +  }
> +  return Changed;
> +}
> +
> +struct GlobalSplit : public ModulePass {
> +  static char ID;
> +  GlobalSplit() : ModulePass(ID) {
> +    initializeGlobalSplitPass(*PassRegistry::getPassRegistry());
> +  }
> +  bool runOnModule(Module &M) {
> +    if (skipModule(M))
> +      return false;
> +
> +    return splitGlobals(M);
> +  }
> +};
> +
> +}
> +
> +INITIALIZE_PASS(GlobalSplit, "globalsplit", "Global splitter", false, false)
> +char GlobalSplit::ID = 0;
> +
> +ModulePass *llvm::createGlobalSplitPass() {
> +  return new GlobalSplit;
> +}
>
> Modified: llvm/trunk/lib/Transforms/IPO/IPO.cpp
> URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Transforms/IPO/IPO.cpp?rev=287178&r1=287177&r2=287178&view=diff
> ==============================================================================
> --- llvm/trunk/lib/Transforms/IPO/IPO.cpp (original)
> +++ llvm/trunk/lib/Transforms/IPO/IPO.cpp Wed Nov 16 17:40:26 2016
> @@ -32,6 +32,7 @@ void llvm::initializeIPO(PassRegistry &R
>    initializeForceFunctionAttrsLegacyPassPass(Registry);
>    initializeGlobalDCELegacyPassPass(Registry);
>    initializeGlobalOptLegacyPassPass(Registry);
> +  initializeGlobalSplitPass(Registry);
>    initializeIPCPPass(Registry);
>    initializeAlwaysInlinerLegacyPassPass(Registry);
>    initializeSimpleInlinerPass(Registry);
>
> Modified: llvm/trunk/lib/Transforms/IPO/PassManagerBuilder.cpp
> URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Transforms/IPO/PassManagerBuilder.cpp?rev=287178&r1=287177&r2=287178&view=diff
> ==============================================================================
> --- llvm/trunk/lib/Transforms/IPO/PassManagerBuilder.cpp (original)
> +++ llvm/trunk/lib/Transforms/IPO/PassManagerBuilder.cpp Wed Nov 16 17:40:26 2016
> @@ -692,6 +692,11 @@ void PassManagerBuilder::addLTOOptimizat
>    PM.add(createPostOrderFunctionAttrsLegacyPass());
>    PM.add(createReversePostOrderFunctionAttrsPass());
>
> +  // Split globals using inrange annotations on GEP indices. This can help
> +  // improve the quality of generated code when virtual constant propagation or
> +  // control flow integrity are enabled.
> +  PM.add(createGlobalSplitPass());
> +
>    // Apply whole-program devirtualization and virtual constant propagation.
>    PM.add(createWholeProgramDevirtPass());
>
>
> Added: llvm/trunk/test/Transforms/GlobalSplit/basic.ll
> URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/Transforms/GlobalSplit/basic.ll?rev=287178&view=auto
> ==============================================================================
> --- llvm/trunk/test/Transforms/GlobalSplit/basic.ll (added)
> +++ llvm/trunk/test/Transforms/GlobalSplit/basic.ll Wed Nov 16 17:40:26 2016
> @@ -0,0 +1,56 @@
> +; RUN: opt -S -globalsplit %s | FileCheck %s
> +
> +target datalayout = "e-p:64:64"
> +target triple = "x86_64-unknown-linux-gnu"
> +
> +; CHECK: @vtt = constant [3 x i8*] [i8* bitcast ([2 x i8* ()*]* @global.0 to i8*), i8* bitcast (i8* ()** getelementptr inbounds ([2 x i8* ()*], [2 x i8* ()*]* @global.0, i32 0, i32 1) to i8*), i8* bitcast ([1 x i8* ()*]* @global.1 to i8*)]
> + at vtt = constant [3 x i8*] [
> +  i8* bitcast (i8* ()** getelementptr ({ [2 x i8* ()*], [1 x i8* ()*] }, { [2 x i8* ()*], [1 x i8* ()*] }* @global, i32 0, inrange i32 0, i32 0) to i8*),
> +  i8* bitcast (i8* ()** getelementptr ({ [2 x i8* ()*], [1 x i8* ()*] }, { [2 x i8* ()*], [1 x i8* ()*] }* @global, i32 0, inrange i32 0, i32 1) to i8*),
> +  i8* bitcast (i8* ()** getelementptr ({ [2 x i8* ()*], [1 x i8* ()*] }, { [2 x i8* ()*], [1 x i8* ()*] }* @global, i32 0, inrange i32 1, i32 0) to i8*)
> +]
> +
> +; CHECK-NOT: @global =
> +; CHECK: @global.0 = private constant [2 x i8* ()*] [i8* ()* @f1, i8* ()* @f2], !type [[T1:![0-9]+$]]
> +; CHECK: @global.1 = private constant [1 x i8* ()*] [i8* ()* @f3], !type [[T2:![0-9]+$]]
> +; CHECK-NOT: @global =
> + at global = internal constant { [2 x i8* ()*], [1 x i8* ()*] } {
> +  [2 x i8* ()*] [i8* ()* @f1, i8* ()* @f2],
> +  [1 x i8* ()*] [i8* ()* @f3]
> +}, !type !0, !type !1
> +
> +; CHECK: define i8* @f1()
> +define i8* @f1() {
> +  ; CHECK-NEXT: ret i8* bitcast ([2 x i8* ()*]* @global.0 to i8*)
> +  ret i8* bitcast (i8* ()** getelementptr ({ [2 x i8* ()*], [1 x i8* ()*] }, { [2 x i8* ()*], [1 x i8* ()*] }* @global, i32 0, inrange i32 0, i32 0) to i8*)
> +}
> +
> +; CHECK: define i8* @f2()
> +define i8* @f2() {
> +  ; CHECK-NEXT: ret i8* bitcast (i8* ()** getelementptr inbounds ([2 x i8* ()*], [2 x i8* ()*]* @global.0, i32 0, i32 1) to i8*)
> +  ret i8* bitcast (i8* ()** getelementptr ({ [2 x i8* ()*], [1 x i8* ()*] }, { [2 x i8* ()*], [1 x i8* ()*] }* @global, i32 0, inrange i32 0, i32 1) to i8*)
> +}
> +
> +; CHECK: define i8* @f3()
> +define i8* @f3() {
> +  ; CHECK-NEXT: ret i8* bitcast (i8* ()** getelementptr inbounds ([2 x i8* ()*], [2 x i8* ()*]* @global.0, i64 1, i32 0) to i8*)
> +  ret i8* bitcast (i8* ()** getelementptr ({ [2 x i8* ()*], [1 x i8* ()*] }, { [2 x i8* ()*], [1 x i8* ()*] }* @global, i32 0, inrange i32 0, i32 2) to i8*)
> +}
> +
> +; CHECK: define i8* @f4()
> +define i8* @f4() {
> +  ; CHECK-NEXT: ret i8* bitcast ([1 x i8* ()*]* @global.1 to i8*)
> +  ret i8* bitcast (i8* ()** getelementptr ({ [2 x i8* ()*], [1 x i8* ()*] }, { [2 x i8* ()*], [1 x i8* ()*] }* @global, i32 0, inrange i32 1, i32 0) to i8*)
> +}
> +
> +define void @foo() {
> +  %p = call i1 @llvm.type.test(i8* null, metadata !"")
> +  ret void
> +}
> +
> +declare i1 @llvm.type.test(i8*, metadata) nounwind readnone
> +
> +; CHECK: [[T1]] = !{i32 8, !"foo"}
> +; CHECK: [[T2]] = !{i32 0, !"bar"}
> +!0 = !{i32 8, !"foo"}
> +!1 = !{i32 16, !"bar"}
>
> Added: llvm/trunk/test/Transforms/GlobalSplit/non-beneficial.ll
> URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/Transforms/GlobalSplit/non-beneficial.ll?rev=287178&view=auto
> ==============================================================================
> --- llvm/trunk/test/Transforms/GlobalSplit/non-beneficial.ll (added)
> +++ llvm/trunk/test/Transforms/GlobalSplit/non-beneficial.ll Wed Nov 16 17:40:26 2016
> @@ -0,0 +1,24 @@
> +; RUN: opt -S -globalsplit %s | FileCheck %s
> +
> +target datalayout = "e-p:64:64"
> +target triple = "x86_64-unknown-linux-gnu"
> +
> +; CHECK: @global =
> + at global = internal constant { [2 x i8* ()*], [1 x i8* ()*] } {
> +  [2 x i8* ()*] [i8* ()* @f, i8* ()* @g],
> +  [1 x i8* ()*] [i8* ()* @h]
> +}
> +
> +define i8* @f() {
> +  ret i8* bitcast (i8* ()** getelementptr ({ [2 x i8* ()*], [1 x i8* ()*] }, { [2 x i8* ()*], [1 x i8* ()*] }* @global, i32 0, inrange i32 0, i32 0) to i8*)
> +}
> +
> +define i8* @g() {
> +  ret i8* null
> +}
> +
> +define i8* @h() {
> +  ret i8* null
> +}
> +
> +!0 = !{i32 16}
>
> Added: llvm/trunk/test/Transforms/GlobalSplit/nonlocal.ll
> URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/Transforms/GlobalSplit/nonlocal.ll?rev=287178&view=auto
> ==============================================================================
> --- llvm/trunk/test/Transforms/GlobalSplit/nonlocal.ll (added)
> +++ llvm/trunk/test/Transforms/GlobalSplit/nonlocal.ll Wed Nov 16 17:40:26 2016
> @@ -0,0 +1,29 @@
> +; RUN: opt -S -globalsplit %s | FileCheck %s
> +
> +target datalayout = "e-p:64:64"
> +target triple = "x86_64-unknown-linux-gnu"
> +
> +; CHECK: @global =
> + at global = constant { [2 x i8* ()*], [1 x i8* ()*] } {
> +  [2 x i8* ()*] [i8* ()* @f, i8* ()* @g],
> +  [1 x i8* ()*] [i8* ()* @h]
> +}
> +
> +define i8* @f() {
> +  ret i8* bitcast (i8* ()** getelementptr ({ [2 x i8* ()*], [1 x i8* ()*] }, { [2 x i8* ()*], [1 x i8* ()*] }* @global, i32 0, inrange i32 0, i32 0) to i8*)
> +}
> +
> +define i8* @g() {
> +  ret i8* null
> +}
> +
> +define i8* @h() {
> +  ret i8* null
> +}
> +
> +define void @foo() {
> +  %p = call i1 @llvm.type.test(i8* null, metadata !"")
> +  ret void
> +}
> +
> +declare i1 @llvm.type.test(i8*, metadata) nounwind readnone
>
>
> _______________________________________________
> llvm-commits mailing list
> llvm-commits at lists.llvm.org
> http://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-commits

-- 
Davide

"There are no solved problems; there are only problems that are more
or less solved" -- Henri Poincare


More information about the llvm-commits mailing list