From 6e4134459bdb1e537731a16123f75c8d2ec66c5a Mon Sep 17 00:00:00 2001 From: Eugene Leviant <eleviant@accesssoftek.com> Date: Sun, 1 Jul 2018 11:02:07 +0000 Subject: [PATCH] [Evaluator] Improve evaluation of call instruction Recommit of r335324 after buildbot failure fix llvm-svn: 336059 --- .../include/llvm/Transforms/Utils/Evaluator.h | 13 +++ llvm/lib/Transforms/Utils/Evaluator.cpp | 69 +++++++++++-- .../GlobalOpt/evaluate-call-errors.ll | 99 +++++++++++++++++++ .../Transforms/GlobalOpt/evaluate-call.ll | 85 ++++++++++++++++ .../GlobalOpt/evaluate-constfold-call.ll | 55 +++++++++++ 5 files changed, 314 insertions(+), 7 deletions(-) create mode 100644 llvm/test/Transforms/GlobalOpt/evaluate-call-errors.ll create mode 100644 llvm/test/Transforms/GlobalOpt/evaluate-call.ll create mode 100644 llvm/test/Transforms/GlobalOpt/evaluate-constfold-call.ll diff --git a/llvm/include/llvm/Transforms/Utils/Evaluator.h b/llvm/include/llvm/Transforms/Utils/Evaluator.h index 0e987b93177a..9908ae6fd393 100644 --- a/llvm/include/llvm/Transforms/Utils/Evaluator.h +++ b/llvm/include/llvm/Transforms/Utils/Evaluator.h @@ -18,6 +18,7 @@ #include "llvm/ADT/SmallPtrSet.h" #include "llvm/ADT/SmallVector.h" #include "llvm/IR/BasicBlock.h" +#include "llvm/IR/CallSite.h" #include "llvm/IR/GlobalVariable.h" #include "llvm/IR/Value.h" #include "llvm/Support/Casting.h" @@ -73,6 +74,18 @@ public: ValueStack.back()[V] = C; } + /// Given call site return callee and list of its formal arguments + Function *getCalleeWithFormalArgs(CallSite &CS, + SmallVector<Constant *, 8> &Formals); + + /// Given call site and callee returns list of callee formal argument + /// values converting them when necessary + bool getFormalParams(CallSite &CS, Function *F, + SmallVector<Constant *, 8> &Formals); + + /// Casts call result to a type of bitcast call expression + Constant *castCallResultIfNeeded(Value *CallExpr, Constant *RV); + const DenseMap<Constant*, Constant*> &getMutatedMemory() const { return MutatedMemory; } diff --git a/llvm/lib/Transforms/Utils/Evaluator.cpp b/llvm/lib/Transforms/Utils/Evaluator.cpp index 9440ae3ef2af..264bc21d604a 100644 --- a/llvm/lib/Transforms/Utils/Evaluator.cpp +++ b/llvm/lib/Transforms/Utils/Evaluator.cpp @@ -217,6 +217,60 @@ Constant *Evaluator::ComputeLoadResult(Constant *P) { return nullptr; // don't know how to evaluate. } +Function * +Evaluator::getCalleeWithFormalArgs(CallSite &CS, + SmallVector<Constant *, 8> &Formals) { + auto *V = CS.getCalledValue(); + if (auto *Fn = dyn_cast<Function>(getVal(V))) + return getFormalParams(CS, Fn, Formals) ? Fn : nullptr; + + auto *CE = dyn_cast<ConstantExpr>(V); + if (!CE || CE->getOpcode() != Instruction::BitCast || + !getFormalParams(CS, cast<Function>(CE->getOperand(0)), Formals)) + return nullptr; + + return dyn_cast<Function>( + ConstantFoldLoadThroughBitcast(CE, CE->getOperand(0)->getType(), DL)); +} + +bool Evaluator::getFormalParams(CallSite &CS, Function *F, + SmallVector<Constant *, 8> &Formals) { + auto *FTy = F->getFunctionType(); + if (FTy->getNumParams() > CS.getNumArgOperands()) { + LLVM_DEBUG(dbgs() << "Too few arguments for function.\n"); + return false; + } + + auto ArgI = CS.arg_begin(); + for (auto ParI = FTy->param_begin(), ParE = FTy->param_end(); ParI != ParE; + ++ParI) { + auto *ArgC = ConstantFoldLoadThroughBitcast(getVal(*ArgI), *ParI, DL); + if (!ArgC) { + LLVM_DEBUG(dbgs() << "Can not convert function argument.\n"); + return false; + } + Formals.push_back(ArgC); + ++ArgI; + } + return true; +} + +/// If call expression contains bitcast then we may need to cast +/// evaluated return value to a type of the call expression. +Constant *Evaluator::castCallResultIfNeeded(Value *CallExpr, Constant *RV) { + ConstantExpr *CE = dyn_cast<ConstantExpr>(CallExpr); + if (!RV || !CE || CE->getOpcode() != Instruction::BitCast) + return RV; + + if (auto *FT = + dyn_cast<FunctionType>(CE->getType()->getPointerElementType())) { + RV = ConstantFoldLoadThroughBitcast(RV, FT->getReturnType(), DL); + if (!RV) + LLVM_DEBUG(dbgs() << "Failed to fold bitcast call expr\n"); + } + return RV; +} + /// Evaluate all instructions in block BB, returning true if successful, false /// if we can't evaluate it. NewBB returns the next BB that control flows into, /// or null upon return. @@ -465,20 +519,19 @@ bool Evaluator::EvaluateBlock(BasicBlock::iterator CurInst, } // Resolve function pointers. - Function *Callee = dyn_cast<Function>(getVal(CS.getCalledValue())); + SmallVector<Constant *, 8> Formals; + Function *Callee = getCalleeWithFormalArgs(CS, Formals); if (!Callee || Callee->isInterposable()) { LLVM_DEBUG(dbgs() << "Can not resolve function pointer.\n"); return false; // Cannot resolve. } - SmallVector<Constant*, 8> Formals; - for (User::op_iterator i = CS.arg_begin(), e = CS.arg_end(); i != e; ++i) - Formals.push_back(getVal(*i)); - if (Callee->isDeclaration()) { // If this is a function we can constant fold, do it. if (Constant *C = ConstantFoldCall(CS, Callee, Formals, TLI)) { - InstResult = C; + InstResult = castCallResultIfNeeded(CS.getCalledValue(), C); + if (!InstResult) + return false; LLVM_DEBUG(dbgs() << "Constant folded function call. Result: " << *InstResult << "\n"); } else { @@ -499,7 +552,9 @@ bool Evaluator::EvaluateBlock(BasicBlock::iterator CurInst, return false; } ValueStack.pop_back(); - InstResult = RetVal; + InstResult = castCallResultIfNeeded(CS.getCalledValue(), RetVal); + if (RetVal && !InstResult) + return false; if (InstResult) { LLVM_DEBUG(dbgs() << "Successfully evaluated function. Result: " diff --git a/llvm/test/Transforms/GlobalOpt/evaluate-call-errors.ll b/llvm/test/Transforms/GlobalOpt/evaluate-call-errors.ll new file mode 100644 index 000000000000..88f7cbd8df1b --- /dev/null +++ b/llvm/test/Transforms/GlobalOpt/evaluate-call-errors.ll @@ -0,0 +1,99 @@ +; Checks for few bitcasted call evaluation errors + +; REQUIRES: asserts +; RUN: opt -globalopt -instcombine -S -debug-only=evaluator %s -o %t 2>&1 | FileCheck %s + +; CHECK: Failed to fold bitcast call expr +; CHECK: Can not convert function argument + +target datalayout = "e-m:o-i64:64-f80:128-n8:16:32:64-S128" +target triple = "x86_64-apple-macosx10.12.0" + +%struct.S = type { i32 } +%struct.Q = type { i32 } +%struct.Foo = type { i32 } + +@_s = global %struct.S zeroinitializer, align 4 +@_q = global %struct.Q zeroinitializer, align 4 +@llvm.global_ctors = appending global [2 x { i32, void ()*, i8* }] [{ i32, void ()*, i8* } { i32 65535, void ()* @_GLOBAL__sub_I_main2.cpp, i8* null }, { i32, void ()*, i8* } { i32 65535, void ()* @_GLOBAL__sub_I_main3.cpp, i8* null }] + +define internal void @__cxx_global_var_init() section "__TEXT,__StaticInit,regular,pure_instructions" { + call void @_ZN1SC1Ev(%struct.S* @_s) + ret void +} + +define linkonce_odr void @_ZN1SC1Ev(%struct.S*) unnamed_addr align 2 { + %2 = alloca %struct.S*, align 8 + store %struct.S* %0, %struct.S** %2, align 8 + %3 = load %struct.S*, %struct.S** %2, align 8 + call void @_ZN1SC2Ev(%struct.S* %3) + ret void +} + +define internal void @__cxx_global_var_init.1() #0 section "__TEXT,__StaticInit,regular,pure_instructions" { + call void @_ZN1QC1Ev(%struct.Q* @_q) + ret void +} + +define linkonce_odr void @_ZN1QC1Ev(%struct.Q*) unnamed_addr align 2 { + %2 = alloca %struct.Q*, align 8 + store %struct.Q* %0, %struct.Q** %2, align 8 + %3 = load %struct.Q*, %struct.Q** %2, align 8 + call void @_ZN1QC2Ev(%struct.Q* %3) + ret void +} + +define i32 @main() { + %1 = alloca i32, align 4 + store i32 0, i32* %1, align 4 + ret i32 0 +} + +define linkonce_odr void @_ZN1SC2Ev(%struct.S*) unnamed_addr align 2 { + %2 = alloca %struct.S*, align 8 + %3 = alloca %struct.Foo, align 4 + store %struct.S* %0, %struct.S** %2, align 8 + %4 = load %struct.S*, %struct.S** %2, align 8 + %5 = getelementptr inbounds %struct.S, %struct.S* %4, i32 0, i32 0 + %6 = call i32 bitcast (%struct.Foo* ()* @_ZL3foov to i32 ()*)() + %7 = getelementptr inbounds %struct.Foo, %struct.Foo* %3, i32 0, i32 0 + store i32 %6, i32* %7, align 4 + %8 = getelementptr inbounds %struct.Foo, %struct.Foo* %3, i32 0, i32 0 + %9 = load i32, i32* %8, align 4 + store i32 %9, i32* %5, align 4 + ret void +} + +define internal %struct.Foo* @_ZL3foov() { + ret %struct.Foo* null +} + +define linkonce_odr void @_ZN1QC2Ev(%struct.Q*) unnamed_addr align 2 { + %2 = alloca %struct.Q*, align 8 + store %struct.Q* %0, %struct.Q** %2, align 8 + %3 = load %struct.Q*, %struct.Q** %2, align 8 + %4 = getelementptr inbounds %struct.Q, %struct.Q* %3, i32 0, i32 0 + %5 = call i32 bitcast (i32 (i32)* @_ZL3baz3Foo to i32 (%struct.Foo*)*)(%struct.Foo* null) + store i32 %5, i32* %4, align 4 + ret void +} + +define internal i32 @_ZL3baz3Foo(i32) { + %2 = alloca %struct.Foo, align 4 + %3 = getelementptr inbounds %struct.Foo, %struct.Foo* %2, i32 0, i32 0 + store i32 %0, i32* %3, align 4 + %4 = getelementptr inbounds %struct.Foo, %struct.Foo* %2, i32 0, i32 0 + %5 = load i32, i32* %4, align 4 + ret i32 %5 +} + +; Function Attrs: noinline ssp uwtable +define internal void @_GLOBAL__sub_I_main2.cpp() section "__TEXT,__StaticInit,regular,pure_instructions" { + call void @__cxx_global_var_init() + ret void +} + +define internal void @_GLOBAL__sub_I_main3.cpp() section "__TEXT,__StaticInit,regular,pure_instructions" { + call void @__cxx_global_var_init.1() + ret void +} diff --git a/llvm/test/Transforms/GlobalOpt/evaluate-call.ll b/llvm/test/Transforms/GlobalOpt/evaluate-call.ll new file mode 100644 index 000000000000..27211007c6e0 --- /dev/null +++ b/llvm/test/Transforms/GlobalOpt/evaluate-call.ll @@ -0,0 +1,85 @@ +; Checks if bitcasted call expression can be evaluated +; Given call expresion: +; %struct.Bar* bitcast (%struct.Foo* (%struct.Foo*)* @_ZL3fooP3Foo to %struct.Bar* (%struct.Bar*)*)(%struct.Bar* @gBar) +; We evaluate call to function @_ZL3fooP3Foo casting both parameter and return value +; Given call expression: +; void bitcast (void (%struct.Foo*)* @_ZL3bazP3Foo to void (%struct.Bar*)*)(%struct.Bar* @gBar) +; We evaluate call to function _ZL3bazP3Foo casting its parameter and check that evaluated value (nullptr) +; is handled correctly + +; RUN: opt -globalopt -instcombine -S %s -o - | FileCheck %s + +; CHECK: @gBar = local_unnamed_addr global %struct.Bar { i32 2 } +; CHECK-NEXT: @_s = local_unnamed_addr global %struct.S { i32 1 }, align 4 +; CHECK-NEXT: @llvm.global_ctors = appending global [0 x { i32, void ()*, i8* }] zeroinitializer + +; CHECK: define i32 @main() +; CHECK-NEXT: ret i32 0 + +target datalayout = "e-m:o-i64:64-f80:128-n8:16:32:64-S128" +target triple = "x86_64-apple-macosx10.12.0" + +%struct.Bar = type { i32 } +%struct.S = type { i32 } +%struct.Foo = type { i32 } + +@gBar = global %struct.Bar zeroinitializer, align 4 +@_s = global %struct.S zeroinitializer, align 4 +@llvm.global_ctors = appending global [1 x { i32, void ()*, i8* }] [{ i32, void ()*, i8* } { i32 65535, void ()* @_GLOBAL__sub_I_main.cpp, i8* null }] + +define internal void @__cxx_global_var_init() section "__TEXT,__StaticInit,regular,pure_instructions" { + call void @_ZN1SC1Ev(%struct.S* @_s) + ret void +} + +define linkonce_odr void @_ZN1SC1Ev(%struct.S*) unnamed_addr align 2 { + %2 = alloca %struct.S*, align 8 + store %struct.S* %0, %struct.S** %2, align 8 + %3 = load %struct.S*, %struct.S** %2, align 8 + call void @_ZN1SC2Ev(%struct.S* %3) + ret void +} + +define i32 @main() { + %1 = alloca i32, align 4 + store i32 0, i32* %1, align 4 + ret i32 0 +} + +define linkonce_odr void @_ZN1SC2Ev(%struct.S*) unnamed_addr align 2 { + %2 = alloca %struct.S*, align 8 + store %struct.S* %0, %struct.S** %2, align 8 + %3 = load %struct.S*, %struct.S** %2, align 8 + %4 = getelementptr inbounds %struct.S, %struct.S* %3, i32 0, i32 0 + %5 = call %struct.Bar* bitcast (%struct.Foo* (%struct.Foo*)* @_ZL3fooP3Foo to %struct.Bar* (%struct.Bar*)*)(%struct.Bar* @gBar) + %6 = getelementptr inbounds %struct.Bar, %struct.Bar* %5, i32 0, i32 0 + %7 = load i32, i32* %6, align 4 + store i32 %7, i32* %4, align 4 + call void bitcast (void (%struct.Foo*)* @_ZL3bazP3Foo to void (%struct.Bar*)*)(%struct.Bar* @gBar) + ret void +} + +define internal %struct.Foo* @_ZL3fooP3Foo(%struct.Foo*) { + %2 = alloca %struct.Foo*, align 8 + store %struct.Foo* %0, %struct.Foo** %2, align 8 + %3 = load %struct.Foo*, %struct.Foo** %2, align 8 + %4 = getelementptr inbounds %struct.Foo, %struct.Foo* %3, i32 0, i32 0 + store i32 1, i32* %4, align 4 + %5 = load %struct.Foo*, %struct.Foo** %2, align 8 + ret %struct.Foo* %5 +} + +define internal void @_ZL3bazP3Foo(%struct.Foo*) { + %2 = alloca %struct.Foo*, align 8 + store %struct.Foo* %0, %struct.Foo** %2, align 8 + %3 = load %struct.Foo*, %struct.Foo** %2, align 8 + %4 = getelementptr inbounds %struct.Foo, %struct.Foo* %3, i32 0, i32 0 + store i32 2, i32* %4, align 4 + ret void +} + +; Function Attrs: noinline ssp uwtable +define internal void @_GLOBAL__sub_I_main.cpp() section "__TEXT,__StaticInit,regular,pure_instructions" { + call void @__cxx_global_var_init() + ret void +} diff --git a/llvm/test/Transforms/GlobalOpt/evaluate-constfold-call.ll b/llvm/test/Transforms/GlobalOpt/evaluate-constfold-call.ll new file mode 100644 index 000000000000..aeaa5e800c70 --- /dev/null +++ b/llvm/test/Transforms/GlobalOpt/evaluate-constfold-call.ll @@ -0,0 +1,55 @@ +; Check if we can evaluate a bitcasted call to a function which is constant folded. +; Evaluator folds call to fmodf, replacing it with constant value in case both operands +; are known at compile time. +; RUN: opt -globalopt -instcombine %s -S -o - | FileCheck %s + +; CHECK: @_q = dso_local local_unnamed_addr global %struct.Q { i32 1066527622 } +; CHECK: define dso_local i32 @main +; CHECK-NEXT: %[[V:.+]] = load i32, i32* getelementptr inbounds (%struct.Q, %struct.Q* @_q, i64 0, i32 0) +; CHECK-NEXT: ret i32 %[[V]] + +source_filename = "main.cpp" +target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128" +target triple = "x86_64-none-linux-gnu" + +%struct.Q = type { i32 } + +$_ZN1QC2Ev = comdat any + +@_q = dso_local global %struct.Q zeroinitializer, align 4 +@llvm.global_ctors = appending global [1 x { i32, void ()*, i8* }] [{ i32, void ()*, i8* } { i32 65535, void ()* @_GLOBAL__sub_I_main.cpp, i8* null }] + +define internal void @__cxx_global_var_init() section ".text.startup" { + call void @_ZN1QC2Ev(%struct.Q* @_q) + ret void +} + +define linkonce_odr dso_local void @_ZN1QC2Ev(%struct.Q*) unnamed_addr #1 comdat align 2 { + %2 = alloca %struct.Q*, align 8 + store %struct.Q* %0, %struct.Q** %2, align 8 + %3 = load %struct.Q*, %struct.Q** %2, align 8 + %4 = getelementptr inbounds %struct.Q, %struct.Q* %3, i32 0, i32 0 + %5 = call i32 bitcast (float (float, float)* @fmodf to i32 (float, float)*)(float 0x40091EB860000000, float 2.000000e+00) + store i32 %5, i32* %4, align 4 + ret void +} + +define dso_local i32 @main(i32, i8**) { + %3 = alloca i32, align 4 + %4 = alloca i32, align 4 + %5 = alloca i8**, align 8 + store i32 0, i32* %3, align 4 + store i32 %0, i32* %4, align 4 + store i8** %1, i8*** %5, align 8 + %6 = load i32, i32* getelementptr inbounds (%struct.Q, %struct.Q* @_q, i32 0, i32 0), align 4 + ret i32 %6 +} + +; Function Attrs: nounwind +declare dso_local float @fmodf(float, float) + +; Function Attrs: noinline uwtable +define internal void @_GLOBAL__sub_I_main.cpp() section ".text.startup" { + call void @__cxx_global_var_init() + ret void +}