1
0
mirror of https://github.com/RPCS3/llvm-mirror.git synced 2024-10-20 03:23:01 +02:00
llvm-mirror/lib/Transforms/Utils/LowerSwitch.cpp

519 lines
18 KiB
C++
Raw Normal View History

//===- LowerSwitch.cpp - Eliminate Switch instructions --------------------===//
//
// The LLVM Compiler Infrastructure
//
// This file is distributed under the University of Illinois Open Source
// License. See LICENSE.TXT for details.
//
//===----------------------------------------------------------------------===//
//
// The LowerSwitch transformation rewrites switch instructions with a sequence
// of branches, which allows targets to get away with not implementing the
// switch instruction until it is convenient.
//
//===----------------------------------------------------------------------===//
#include "llvm/Transforms/Scalar.h"
LowerSwitch: track bounding range for the condition tree. When LowerSwitch transforms a switch instruction into a tree of ifs it is actually performing a binary search into the various case ranges, to see if the current value falls into one cases range of values. So, if we have a program with something like this: switch (a) { case 0: do0(); break; case 1: do1(); break; case 2: do2(); break; default: break; } the code produced is something like this: if (a < 1) { if (a == 0) { do0(); } } else { if (a < 2) { if (a == 1) { do1(); } } else { if (a == 2) { do2(); } } } This code is inefficient because the check (a == 1) to execute do1() is not needed. The reason is that because we already checked that (a >= 1) initially by checking that also (a < 2) we basically already inferred that (a == 1) without the need of an extra basic block spawned to check if actually (a == 1). The patch addresses this problem by keeping track of already checked bounds in the LowerSwitch algorithm, so that when the time arrives to produce a Leaf Block that checks the equality with the case value / range the algorithm can decide if that block is really needed depending on the already checked bounds . For example, the above with "a = 1" would work like this: the bounds start as LB: NONE , UB: NONE as (a < 1) is emitted the bounds for the else path become LB: 1 UB: NONE. This happens because by failing the test (a < 1) we know that the value "a" cannot be smaller than 1 if we enter the else branch. After the emitting the check (a < 2) the bounds in the if branch become LB: 1 UB: 1. This is because by checking that "a" is smaller than 2 then the upper bound becomes 2 - 1 = 1. When it is time to emit the leaf block for "case 1:" we notice that 1 can be squeezed exactly in between the LB and UB, which means that if we arrived to that block there is no need to emit a block that checks if (a == 1). Patch by: Marcello Maggioni <hayarms@gmail.com> llvm-svn: 211038
2014-06-16 18:55:20 +02:00
#include "llvm/Transforms/Utils/BasicBlockUtils.h"
#include "llvm/ADT/STLExtras.h"
#include "llvm/IR/Constants.h"
#include "llvm/IR/Function.h"
#include "llvm/IR/Instructions.h"
#include "llvm/IR/LLVMContext.h"
LowerSwitch: track bounding range for the condition tree. When LowerSwitch transforms a switch instruction into a tree of ifs it is actually performing a binary search into the various case ranges, to see if the current value falls into one cases range of values. So, if we have a program with something like this: switch (a) { case 0: do0(); break; case 1: do1(); break; case 2: do2(); break; default: break; } the code produced is something like this: if (a < 1) { if (a == 0) { do0(); } } else { if (a < 2) { if (a == 1) { do1(); } } else { if (a == 2) { do2(); } } } This code is inefficient because the check (a == 1) to execute do1() is not needed. The reason is that because we already checked that (a >= 1) initially by checking that also (a < 2) we basically already inferred that (a == 1) without the need of an extra basic block spawned to check if actually (a == 1). The patch addresses this problem by keeping track of already checked bounds in the LowerSwitch algorithm, so that when the time arrives to produce a Leaf Block that checks the equality with the case value / range the algorithm can decide if that block is really needed depending on the already checked bounds . For example, the above with "a = 1" would work like this: the bounds start as LB: NONE , UB: NONE as (a < 1) is emitted the bounds for the else path become LB: 1 UB: NONE. This happens because by failing the test (a < 1) we know that the value "a" cannot be smaller than 1 if we enter the else branch. After the emitting the check (a < 2) the bounds in the if branch become LB: 1 UB: 1. This is because by checking that "a" is smaller than 2 then the upper bound becomes 2 - 1 = 1. When it is time to emit the leaf block for "case 1:" we notice that 1 can be squeezed exactly in between the LB and UB, which means that if we arrived to that block there is no need to emit a block that checks if (a == 1). Patch by: Marcello Maggioni <hayarms@gmail.com> llvm-svn: 211038
2014-06-16 18:55:20 +02:00
#include "llvm/IR/CFG.h"
#include "llvm/Pass.h"
#include "llvm/Support/Compiler.h"
#include "llvm/Support/Debug.h"
#include "llvm/Support/raw_ostream.h"
#include "llvm/Transforms/Utils/UnifyFunctionExitNodes.h"
#include <algorithm>
using namespace llvm;
[Modules] Make Support/Debug.h modular. This requires it to not change behavior based on other files defining DEBUG_TYPE, which means it cannot define DEBUG_TYPE at all. This is actually better IMO as it forces folks to define relevant DEBUG_TYPEs for their files. However, it requires all files that currently use DEBUG(...) to define a DEBUG_TYPE if they don't already. I've updated all such files in LLVM and will do the same for other upstream projects. This still leaves one important change in how LLVM uses the DEBUG_TYPE macro going forward: we need to only define the macro *after* header files have been #include-ed. Previously, this wasn't possible because Debug.h required the macro to be pre-defined. This commit removes that. By defining DEBUG_TYPE after the includes two things are fixed: - Header files that need to provide a DEBUG_TYPE for some inline code can do so by defining the macro before their inline code and undef-ing it afterward so the macro does not escape. - We no longer have rampant ODR violations due to including headers with different DEBUG_TYPE definitions. This may be mostly an academic violation today, but with modules these types of violations are easy to check for and potentially very relevant. Where necessary to suppor headers with DEBUG_TYPE, I have moved the definitions below the includes in this commit. I plan to move the rest of the DEBUG_TYPE macros in LLVM in subsequent commits; this one is big enough. The comments in Debug.h, which were hilariously out of date already, have been updated to reflect the recommended practice going forward. llvm-svn: 206822
2014-04-22 00:55:11 +02:00
#define DEBUG_TYPE "lower-switch"
namespace {
struct IntRange {
int64_t Low, High;
};
// Return true iff R is covered by Ranges.
static bool IsInRanges(const IntRange &R,
const std::vector<IntRange> &Ranges) {
// Note: Ranges must be sorted, non-overlapping and non-adjacent.
// Find the first range whose High field is >= R.High,
// then check if the Low field is <= R.Low. If so, we
// have a Range that covers R.
auto I = std::lower_bound(
Ranges.begin(), Ranges.end(), R,
[](const IntRange &A, const IntRange &B) { return A.High < B.High; });
return I != Ranges.end() && I->Low <= R.Low;
}
/// LowerSwitch Pass - Replace all SwitchInst instructions with chained branch
/// instructions.
class LowerSwitch : public FunctionPass {
public:
2007-05-06 15:37:16 +02:00
static char ID; // Pass identification, replacement for typeid
LowerSwitch() : FunctionPass(ID) {
initializeLowerSwitchPass(*PassRegistry::getPassRegistry());
}
bool runOnFunction(Function &F) override;
void getAnalysisUsage(AnalysisUsage &AU) const override {
// This is a cluster of orthogonal Transforms
AU.addPreserved<UnifyFunctionExitNodes>();
AU.addPreservedID(LowerInvokePassID);
}
struct CaseRange {
ConstantInt* Low;
ConstantInt* High;
BasicBlock* BB;
CaseRange(ConstantInt *low, ConstantInt *high, BasicBlock *bb)
: Low(low), High(high), BB(bb) {}
};
LowerSwitch: track bounding range for the condition tree. When LowerSwitch transforms a switch instruction into a tree of ifs it is actually performing a binary search into the various case ranges, to see if the current value falls into one cases range of values. So, if we have a program with something like this: switch (a) { case 0: do0(); break; case 1: do1(); break; case 2: do2(); break; default: break; } the code produced is something like this: if (a < 1) { if (a == 0) { do0(); } } else { if (a < 2) { if (a == 1) { do1(); } } else { if (a == 2) { do2(); } } } This code is inefficient because the check (a == 1) to execute do1() is not needed. The reason is that because we already checked that (a >= 1) initially by checking that also (a < 2) we basically already inferred that (a == 1) without the need of an extra basic block spawned to check if actually (a == 1). The patch addresses this problem by keeping track of already checked bounds in the LowerSwitch algorithm, so that when the time arrives to produce a Leaf Block that checks the equality with the case value / range the algorithm can decide if that block is really needed depending on the already checked bounds . For example, the above with "a = 1" would work like this: the bounds start as LB: NONE , UB: NONE as (a < 1) is emitted the bounds for the else path become LB: 1 UB: NONE. This happens because by failing the test (a < 1) we know that the value "a" cannot be smaller than 1 if we enter the else branch. After the emitting the check (a < 2) the bounds in the if branch become LB: 1 UB: 1. This is because by checking that "a" is smaller than 2 then the upper bound becomes 2 - 1 = 1. When it is time to emit the leaf block for "case 1:" we notice that 1 can be squeezed exactly in between the LB and UB, which means that if we arrived to that block there is no need to emit a block that checks if (a == 1). Patch by: Marcello Maggioni <hayarms@gmail.com> llvm-svn: 211038
2014-06-16 18:55:20 +02:00
typedef std::vector<CaseRange> CaseVector;
typedef std::vector<CaseRange>::iterator CaseItr;
private:
void processSwitchInst(SwitchInst *SI);
LowerSwitch: track bounding range for the condition tree. When LowerSwitch transforms a switch instruction into a tree of ifs it is actually performing a binary search into the various case ranges, to see if the current value falls into one cases range of values. So, if we have a program with something like this: switch (a) { case 0: do0(); break; case 1: do1(); break; case 2: do2(); break; default: break; } the code produced is something like this: if (a < 1) { if (a == 0) { do0(); } } else { if (a < 2) { if (a == 1) { do1(); } } else { if (a == 2) { do2(); } } } This code is inefficient because the check (a == 1) to execute do1() is not needed. The reason is that because we already checked that (a >= 1) initially by checking that also (a < 2) we basically already inferred that (a == 1) without the need of an extra basic block spawned to check if actually (a == 1). The patch addresses this problem by keeping track of already checked bounds in the LowerSwitch algorithm, so that when the time arrives to produce a Leaf Block that checks the equality with the case value / range the algorithm can decide if that block is really needed depending on the already checked bounds . For example, the above with "a = 1" would work like this: the bounds start as LB: NONE , UB: NONE as (a < 1) is emitted the bounds for the else path become LB: 1 UB: NONE. This happens because by failing the test (a < 1) we know that the value "a" cannot be smaller than 1 if we enter the else branch. After the emitting the check (a < 2) the bounds in the if branch become LB: 1 UB: 1. This is because by checking that "a" is smaller than 2 then the upper bound becomes 2 - 1 = 1. When it is time to emit the leaf block for "case 1:" we notice that 1 can be squeezed exactly in between the LB and UB, which means that if we arrived to that block there is no need to emit a block that checks if (a == 1). Patch by: Marcello Maggioni <hayarms@gmail.com> llvm-svn: 211038
2014-06-16 18:55:20 +02:00
BasicBlock *switchConvert(CaseItr Begin, CaseItr End,
ConstantInt *LowerBound, ConstantInt *UpperBound,
Value *Val, BasicBlock *Predecessor,
BasicBlock *OrigBlock, BasicBlock *Default,
const std::vector<IntRange> &UnreachableRanges);
LowerSwitch: track bounding range for the condition tree. When LowerSwitch transforms a switch instruction into a tree of ifs it is actually performing a binary search into the various case ranges, to see if the current value falls into one cases range of values. So, if we have a program with something like this: switch (a) { case 0: do0(); break; case 1: do1(); break; case 2: do2(); break; default: break; } the code produced is something like this: if (a < 1) { if (a == 0) { do0(); } } else { if (a < 2) { if (a == 1) { do1(); } } else { if (a == 2) { do2(); } } } This code is inefficient because the check (a == 1) to execute do1() is not needed. The reason is that because we already checked that (a >= 1) initially by checking that also (a < 2) we basically already inferred that (a == 1) without the need of an extra basic block spawned to check if actually (a == 1). The patch addresses this problem by keeping track of already checked bounds in the LowerSwitch algorithm, so that when the time arrives to produce a Leaf Block that checks the equality with the case value / range the algorithm can decide if that block is really needed depending on the already checked bounds . For example, the above with "a = 1" would work like this: the bounds start as LB: NONE , UB: NONE as (a < 1) is emitted the bounds for the else path become LB: 1 UB: NONE. This happens because by failing the test (a < 1) we know that the value "a" cannot be smaller than 1 if we enter the else branch. After the emitting the check (a < 2) the bounds in the if branch become LB: 1 UB: 1. This is because by checking that "a" is smaller than 2 then the upper bound becomes 2 - 1 = 1. When it is time to emit the leaf block for "case 1:" we notice that 1 can be squeezed exactly in between the LB and UB, which means that if we arrived to that block there is no need to emit a block that checks if (a == 1). Patch by: Marcello Maggioni <hayarms@gmail.com> llvm-svn: 211038
2014-06-16 18:55:20 +02:00
BasicBlock *newLeafBlock(CaseRange &Leaf, Value *Val, BasicBlock *OrigBlock,
BasicBlock *Default);
unsigned Clusterify(CaseVector &Cases, SwitchInst *SI);
};
/// The comparison function for sorting the switch case values in the vector.
/// WARNING: Case ranges should be disjoint!
struct CaseCmp {
bool operator () (const LowerSwitch::CaseRange& C1,
const LowerSwitch::CaseRange& C2) {
const ConstantInt* CI1 = cast<const ConstantInt>(C1.Low);
const ConstantInt* CI2 = cast<const ConstantInt>(C2.High);
return CI1->getValue().slt(CI2->getValue());
}
};
}
char LowerSwitch::ID = 0;
INITIALIZE_PASS(LowerSwitch, "lowerswitch",
"Lower SwitchInst's to branches", false, false)
// Publicly exposed interface to pass...
char &llvm::LowerSwitchID = LowerSwitch::ID;
// createLowerSwitchPass - Interface to this file...
FunctionPass *llvm::createLowerSwitchPass() {
return new LowerSwitch();
}
bool LowerSwitch::runOnFunction(Function &F) {
bool Changed = false;
for (Function::iterator I = F.begin(), E = F.end(); I != E; ) {
BasicBlock *Cur = I++; // Advance over block so we don't traverse new blocks
if (SwitchInst *SI = dyn_cast<SwitchInst>(Cur->getTerminator())) {
Changed = true;
processSwitchInst(SI);
}
}
return Changed;
}
// operator<< - Used for debugging purposes.
//
static raw_ostream& operator<<(raw_ostream &O,
const LowerSwitch::CaseVector &C)
LLVM_ATTRIBUTE_USED;
static raw_ostream& operator<<(raw_ostream &O,
const LowerSwitch::CaseVector &C) {
O << "[";
for (LowerSwitch::CaseVector::const_iterator B = C.begin(),
E = C.end(); B != E; ) {
O << *B->Low << " -" << *B->High;
if (++B != E) O << ", ";
}
return O << "]";
}
// \brief Update the first occurrence of the "switch statement" BB in the PHI
// node with the "new" BB. The other occurrences will:
//
// 1) Be updated by subsequent calls to this function. Switch statements may
// have more than one outcoming edge into the same BB if they all have the same
// value. When the switch statement is converted these incoming edges are now
// coming from multiple BBs.
// 2) Removed if subsequent incoming values now share the same case, i.e.,
// multiple outcome edges are condensed into one. This is necessary to keep the
// number of phi values equal to the number of branches to SuccBB.
static void fixPhis(BasicBlock *SuccBB, BasicBlock *OrigBB, BasicBlock *NewBB,
unsigned NumMergedCases) {
for (BasicBlock::iterator I = SuccBB->begin(), IE = SuccBB->getFirstNonPHI();
I != IE; ++I) {
PHINode *PN = cast<PHINode>(I);
// Only update the first occurence.
unsigned Idx = 0, E = PN->getNumIncomingValues();
unsigned LocalNumMergedCases = NumMergedCases;
for (; Idx != E; ++Idx) {
if (PN->getIncomingBlock(Idx) == OrigBB) {
PN->setIncomingBlock(Idx, NewBB);
break;
}
}
// Remove additional occurences coming from condensed cases and keep the
// number of incoming values equal to the number of branches to SuccBB.
for (++Idx; LocalNumMergedCases > 0 && Idx < E; ++Idx)
if (PN->getIncomingBlock(Idx) == OrigBB) {
PN->removeIncomingValue(Idx);
LocalNumMergedCases--;
}
}
}
// switchConvert - Convert the switch statement into a binary lookup of
// the case values. The function recursively builds this tree.
LowerSwitch: track bounding range for the condition tree. When LowerSwitch transforms a switch instruction into a tree of ifs it is actually performing a binary search into the various case ranges, to see if the current value falls into one cases range of values. So, if we have a program with something like this: switch (a) { case 0: do0(); break; case 1: do1(); break; case 2: do2(); break; default: break; } the code produced is something like this: if (a < 1) { if (a == 0) { do0(); } } else { if (a < 2) { if (a == 1) { do1(); } } else { if (a == 2) { do2(); } } } This code is inefficient because the check (a == 1) to execute do1() is not needed. The reason is that because we already checked that (a >= 1) initially by checking that also (a < 2) we basically already inferred that (a == 1) without the need of an extra basic block spawned to check if actually (a == 1). The patch addresses this problem by keeping track of already checked bounds in the LowerSwitch algorithm, so that when the time arrives to produce a Leaf Block that checks the equality with the case value / range the algorithm can decide if that block is really needed depending on the already checked bounds . For example, the above with "a = 1" would work like this: the bounds start as LB: NONE , UB: NONE as (a < 1) is emitted the bounds for the else path become LB: 1 UB: NONE. This happens because by failing the test (a < 1) we know that the value "a" cannot be smaller than 1 if we enter the else branch. After the emitting the check (a < 2) the bounds in the if branch become LB: 1 UB: 1. This is because by checking that "a" is smaller than 2 then the upper bound becomes 2 - 1 = 1. When it is time to emit the leaf block for "case 1:" we notice that 1 can be squeezed exactly in between the LB and UB, which means that if we arrived to that block there is no need to emit a block that checks if (a == 1). Patch by: Marcello Maggioni <hayarms@gmail.com> llvm-svn: 211038
2014-06-16 18:55:20 +02:00
// LowerBound and UpperBound are used to keep track of the bounds for Val
// that have already been checked by a block emitted by one of the previous
// calls to switchConvert in the call stack.
BasicBlock *
LowerSwitch::switchConvert(CaseItr Begin, CaseItr End, ConstantInt *LowerBound,
ConstantInt *UpperBound, Value *Val,
BasicBlock *Predecessor, BasicBlock *OrigBlock,
BasicBlock *Default,
const std::vector<IntRange> &UnreachableRanges) {
unsigned Size = End - Begin;
LowerSwitch: track bounding range for the condition tree. When LowerSwitch transforms a switch instruction into a tree of ifs it is actually performing a binary search into the various case ranges, to see if the current value falls into one cases range of values. So, if we have a program with something like this: switch (a) { case 0: do0(); break; case 1: do1(); break; case 2: do2(); break; default: break; } the code produced is something like this: if (a < 1) { if (a == 0) { do0(); } } else { if (a < 2) { if (a == 1) { do1(); } } else { if (a == 2) { do2(); } } } This code is inefficient because the check (a == 1) to execute do1() is not needed. The reason is that because we already checked that (a >= 1) initially by checking that also (a < 2) we basically already inferred that (a == 1) without the need of an extra basic block spawned to check if actually (a == 1). The patch addresses this problem by keeping track of already checked bounds in the LowerSwitch algorithm, so that when the time arrives to produce a Leaf Block that checks the equality with the case value / range the algorithm can decide if that block is really needed depending on the already checked bounds . For example, the above with "a = 1" would work like this: the bounds start as LB: NONE , UB: NONE as (a < 1) is emitted the bounds for the else path become LB: 1 UB: NONE. This happens because by failing the test (a < 1) we know that the value "a" cannot be smaller than 1 if we enter the else branch. After the emitting the check (a < 2) the bounds in the if branch become LB: 1 UB: 1. This is because by checking that "a" is smaller than 2 then the upper bound becomes 2 - 1 = 1. When it is time to emit the leaf block for "case 1:" we notice that 1 can be squeezed exactly in between the LB and UB, which means that if we arrived to that block there is no need to emit a block that checks if (a == 1). Patch by: Marcello Maggioni <hayarms@gmail.com> llvm-svn: 211038
2014-06-16 18:55:20 +02:00
if (Size == 1) {
// Check if the Case Range is perfectly squeezed in between
// already checked Upper and Lower bounds. If it is then we can avoid
// emitting the code that checks if the value actually falls in the range
// because the bounds already tell us so.
if (Begin->Low == LowerBound && Begin->High == UpperBound) {
unsigned NumMergedCases = 0;
if (LowerBound && UpperBound)
NumMergedCases =
UpperBound->getSExtValue() - LowerBound->getSExtValue();
fixPhis(Begin->BB, OrigBlock, Predecessor, NumMergedCases);
LowerSwitch: track bounding range for the condition tree. When LowerSwitch transforms a switch instruction into a tree of ifs it is actually performing a binary search into the various case ranges, to see if the current value falls into one cases range of values. So, if we have a program with something like this: switch (a) { case 0: do0(); break; case 1: do1(); break; case 2: do2(); break; default: break; } the code produced is something like this: if (a < 1) { if (a == 0) { do0(); } } else { if (a < 2) { if (a == 1) { do1(); } } else { if (a == 2) { do2(); } } } This code is inefficient because the check (a == 1) to execute do1() is not needed. The reason is that because we already checked that (a >= 1) initially by checking that also (a < 2) we basically already inferred that (a == 1) without the need of an extra basic block spawned to check if actually (a == 1). The patch addresses this problem by keeping track of already checked bounds in the LowerSwitch algorithm, so that when the time arrives to produce a Leaf Block that checks the equality with the case value / range the algorithm can decide if that block is really needed depending on the already checked bounds . For example, the above with "a = 1" would work like this: the bounds start as LB: NONE , UB: NONE as (a < 1) is emitted the bounds for the else path become LB: 1 UB: NONE. This happens because by failing the test (a < 1) we know that the value "a" cannot be smaller than 1 if we enter the else branch. After the emitting the check (a < 2) the bounds in the if branch become LB: 1 UB: 1. This is because by checking that "a" is smaller than 2 then the upper bound becomes 2 - 1 = 1. When it is time to emit the leaf block for "case 1:" we notice that 1 can be squeezed exactly in between the LB and UB, which means that if we arrived to that block there is no need to emit a block that checks if (a == 1). Patch by: Marcello Maggioni <hayarms@gmail.com> llvm-svn: 211038
2014-06-16 18:55:20 +02:00
return Begin->BB;
}
return newLeafBlock(*Begin, Val, OrigBlock, Default);
LowerSwitch: track bounding range for the condition tree. When LowerSwitch transforms a switch instruction into a tree of ifs it is actually performing a binary search into the various case ranges, to see if the current value falls into one cases range of values. So, if we have a program with something like this: switch (a) { case 0: do0(); break; case 1: do1(); break; case 2: do2(); break; default: break; } the code produced is something like this: if (a < 1) { if (a == 0) { do0(); } } else { if (a < 2) { if (a == 1) { do1(); } } else { if (a == 2) { do2(); } } } This code is inefficient because the check (a == 1) to execute do1() is not needed. The reason is that because we already checked that (a >= 1) initially by checking that also (a < 2) we basically already inferred that (a == 1) without the need of an extra basic block spawned to check if actually (a == 1). The patch addresses this problem by keeping track of already checked bounds in the LowerSwitch algorithm, so that when the time arrives to produce a Leaf Block that checks the equality with the case value / range the algorithm can decide if that block is really needed depending on the already checked bounds . For example, the above with "a = 1" would work like this: the bounds start as LB: NONE , UB: NONE as (a < 1) is emitted the bounds for the else path become LB: 1 UB: NONE. This happens because by failing the test (a < 1) we know that the value "a" cannot be smaller than 1 if we enter the else branch. After the emitting the check (a < 2) the bounds in the if branch become LB: 1 UB: 1. This is because by checking that "a" is smaller than 2 then the upper bound becomes 2 - 1 = 1. When it is time to emit the leaf block for "case 1:" we notice that 1 can be squeezed exactly in between the LB and UB, which means that if we arrived to that block there is no need to emit a block that checks if (a == 1). Patch by: Marcello Maggioni <hayarms@gmail.com> llvm-svn: 211038
2014-06-16 18:55:20 +02:00
}
unsigned Mid = Size / 2;
std::vector<CaseRange> LHS(Begin, Begin + Mid);
DEBUG(dbgs() << "LHS: " << LHS << "\n");
std::vector<CaseRange> RHS(Begin + Mid, End);
DEBUG(dbgs() << "RHS: " << RHS << "\n");
LowerSwitch: track bounding range for the condition tree. When LowerSwitch transforms a switch instruction into a tree of ifs it is actually performing a binary search into the various case ranges, to see if the current value falls into one cases range of values. So, if we have a program with something like this: switch (a) { case 0: do0(); break; case 1: do1(); break; case 2: do2(); break; default: break; } the code produced is something like this: if (a < 1) { if (a == 0) { do0(); } } else { if (a < 2) { if (a == 1) { do1(); } } else { if (a == 2) { do2(); } } } This code is inefficient because the check (a == 1) to execute do1() is not needed. The reason is that because we already checked that (a >= 1) initially by checking that also (a < 2) we basically already inferred that (a == 1) without the need of an extra basic block spawned to check if actually (a == 1). The patch addresses this problem by keeping track of already checked bounds in the LowerSwitch algorithm, so that when the time arrives to produce a Leaf Block that checks the equality with the case value / range the algorithm can decide if that block is really needed depending on the already checked bounds . For example, the above with "a = 1" would work like this: the bounds start as LB: NONE , UB: NONE as (a < 1) is emitted the bounds for the else path become LB: 1 UB: NONE. This happens because by failing the test (a < 1) we know that the value "a" cannot be smaller than 1 if we enter the else branch. After the emitting the check (a < 2) the bounds in the if branch become LB: 1 UB: 1. This is because by checking that "a" is smaller than 2 then the upper bound becomes 2 - 1 = 1. When it is time to emit the leaf block for "case 1:" we notice that 1 can be squeezed exactly in between the LB and UB, which means that if we arrived to that block there is no need to emit a block that checks if (a == 1). Patch by: Marcello Maggioni <hayarms@gmail.com> llvm-svn: 211038
2014-06-16 18:55:20 +02:00
CaseRange &Pivot = *(Begin + Mid);
DEBUG(dbgs() << "Pivot ==> "
<< Pivot.Low->getValue()
<< " -" << Pivot.High->getValue() << "\n");
LowerSwitch: track bounding range for the condition tree. When LowerSwitch transforms a switch instruction into a tree of ifs it is actually performing a binary search into the various case ranges, to see if the current value falls into one cases range of values. So, if we have a program with something like this: switch (a) { case 0: do0(); break; case 1: do1(); break; case 2: do2(); break; default: break; } the code produced is something like this: if (a < 1) { if (a == 0) { do0(); } } else { if (a < 2) { if (a == 1) { do1(); } } else { if (a == 2) { do2(); } } } This code is inefficient because the check (a == 1) to execute do1() is not needed. The reason is that because we already checked that (a >= 1) initially by checking that also (a < 2) we basically already inferred that (a == 1) without the need of an extra basic block spawned to check if actually (a == 1). The patch addresses this problem by keeping track of already checked bounds in the LowerSwitch algorithm, so that when the time arrives to produce a Leaf Block that checks the equality with the case value / range the algorithm can decide if that block is really needed depending on the already checked bounds . For example, the above with "a = 1" would work like this: the bounds start as LB: NONE , UB: NONE as (a < 1) is emitted the bounds for the else path become LB: 1 UB: NONE. This happens because by failing the test (a < 1) we know that the value "a" cannot be smaller than 1 if we enter the else branch. After the emitting the check (a < 2) the bounds in the if branch become LB: 1 UB: 1. This is because by checking that "a" is smaller than 2 then the upper bound becomes 2 - 1 = 1. When it is time to emit the leaf block for "case 1:" we notice that 1 can be squeezed exactly in between the LB and UB, which means that if we arrived to that block there is no need to emit a block that checks if (a == 1). Patch by: Marcello Maggioni <hayarms@gmail.com> llvm-svn: 211038
2014-06-16 18:55:20 +02:00
// NewLowerBound here should never be the integer minimal value.
// This is because it is computed from a case range that is never
// the smallest, so there is always a case range that has at least
// a smaller value.
ConstantInt *NewLowerBound = Pivot.Low;
// Because NewLowerBound is never the smallest representable integer
// it is safe here to subtract one.
ConstantInt *NewUpperBound = ConstantInt::get(NewLowerBound->getContext(),
NewLowerBound->getValue() - 1);
if (!UnreachableRanges.empty()) {
// Check if the gap between LHS's highest and NewLowerBound is unreachable.
int64_t GapLow = LHS.back().High->getSExtValue() + 1;
int64_t GapHigh = NewLowerBound->getSExtValue() - 1;
IntRange Gap = { GapLow, GapHigh };
if (GapHigh >= GapLow && IsInRanges(Gap, UnreachableRanges))
NewUpperBound = LHS.back().High;
LowerSwitch: track bounding range for the condition tree. When LowerSwitch transforms a switch instruction into a tree of ifs it is actually performing a binary search into the various case ranges, to see if the current value falls into one cases range of values. So, if we have a program with something like this: switch (a) { case 0: do0(); break; case 1: do1(); break; case 2: do2(); break; default: break; } the code produced is something like this: if (a < 1) { if (a == 0) { do0(); } } else { if (a < 2) { if (a == 1) { do1(); } } else { if (a == 2) { do2(); } } } This code is inefficient because the check (a == 1) to execute do1() is not needed. The reason is that because we already checked that (a >= 1) initially by checking that also (a < 2) we basically already inferred that (a == 1) without the need of an extra basic block spawned to check if actually (a == 1). The patch addresses this problem by keeping track of already checked bounds in the LowerSwitch algorithm, so that when the time arrives to produce a Leaf Block that checks the equality with the case value / range the algorithm can decide if that block is really needed depending on the already checked bounds . For example, the above with "a = 1" would work like this: the bounds start as LB: NONE , UB: NONE as (a < 1) is emitted the bounds for the else path become LB: 1 UB: NONE. This happens because by failing the test (a < 1) we know that the value "a" cannot be smaller than 1 if we enter the else branch. After the emitting the check (a < 2) the bounds in the if branch become LB: 1 UB: 1. This is because by checking that "a" is smaller than 2 then the upper bound becomes 2 - 1 = 1. When it is time to emit the leaf block for "case 1:" we notice that 1 can be squeezed exactly in between the LB and UB, which means that if we arrived to that block there is no need to emit a block that checks if (a == 1). Patch by: Marcello Maggioni <hayarms@gmail.com> llvm-svn: 211038
2014-06-16 18:55:20 +02:00
}
LowerSwitch: track bounding range for the condition tree. When LowerSwitch transforms a switch instruction into a tree of ifs it is actually performing a binary search into the various case ranges, to see if the current value falls into one cases range of values. So, if we have a program with something like this: switch (a) { case 0: do0(); break; case 1: do1(); break; case 2: do2(); break; default: break; } the code produced is something like this: if (a < 1) { if (a == 0) { do0(); } } else { if (a < 2) { if (a == 1) { do1(); } } else { if (a == 2) { do2(); } } } This code is inefficient because the check (a == 1) to execute do1() is not needed. The reason is that because we already checked that (a >= 1) initially by checking that also (a < 2) we basically already inferred that (a == 1) without the need of an extra basic block spawned to check if actually (a == 1). The patch addresses this problem by keeping track of already checked bounds in the LowerSwitch algorithm, so that when the time arrives to produce a Leaf Block that checks the equality with the case value / range the algorithm can decide if that block is really needed depending on the already checked bounds . For example, the above with "a = 1" would work like this: the bounds start as LB: NONE , UB: NONE as (a < 1) is emitted the bounds for the else path become LB: 1 UB: NONE. This happens because by failing the test (a < 1) we know that the value "a" cannot be smaller than 1 if we enter the else branch. After the emitting the check (a < 2) the bounds in the if branch become LB: 1 UB: 1. This is because by checking that "a" is smaller than 2 then the upper bound becomes 2 - 1 = 1. When it is time to emit the leaf block for "case 1:" we notice that 1 can be squeezed exactly in between the LB and UB, which means that if we arrived to that block there is no need to emit a block that checks if (a == 1). Patch by: Marcello Maggioni <hayarms@gmail.com> llvm-svn: 211038
2014-06-16 18:55:20 +02:00
DEBUG(dbgs() << "LHS Bounds ==> ";
if (LowerBound) {
dbgs() << LowerBound->getSExtValue();
LowerSwitch: track bounding range for the condition tree. When LowerSwitch transforms a switch instruction into a tree of ifs it is actually performing a binary search into the various case ranges, to see if the current value falls into one cases range of values. So, if we have a program with something like this: switch (a) { case 0: do0(); break; case 1: do1(); break; case 2: do2(); break; default: break; } the code produced is something like this: if (a < 1) { if (a == 0) { do0(); } } else { if (a < 2) { if (a == 1) { do1(); } } else { if (a == 2) { do2(); } } } This code is inefficient because the check (a == 1) to execute do1() is not needed. The reason is that because we already checked that (a >= 1) initially by checking that also (a < 2) we basically already inferred that (a == 1) without the need of an extra basic block spawned to check if actually (a == 1). The patch addresses this problem by keeping track of already checked bounds in the LowerSwitch algorithm, so that when the time arrives to produce a Leaf Block that checks the equality with the case value / range the algorithm can decide if that block is really needed depending on the already checked bounds . For example, the above with "a = 1" would work like this: the bounds start as LB: NONE , UB: NONE as (a < 1) is emitted the bounds for the else path become LB: 1 UB: NONE. This happens because by failing the test (a < 1) we know that the value "a" cannot be smaller than 1 if we enter the else branch. After the emitting the check (a < 2) the bounds in the if branch become LB: 1 UB: 1. This is because by checking that "a" is smaller than 2 then the upper bound becomes 2 - 1 = 1. When it is time to emit the leaf block for "case 1:" we notice that 1 can be squeezed exactly in between the LB and UB, which means that if we arrived to that block there is no need to emit a block that checks if (a == 1). Patch by: Marcello Maggioni <hayarms@gmail.com> llvm-svn: 211038
2014-06-16 18:55:20 +02:00
} else {
dbgs() << "NONE";
}
dbgs() << " - " << NewUpperBound->getSExtValue() << "\n";
dbgs() << "RHS Bounds ==> ";
dbgs() << NewLowerBound->getSExtValue() << " - ";
if (UpperBound) {
dbgs() << UpperBound->getSExtValue() << "\n";
LowerSwitch: track bounding range for the condition tree. When LowerSwitch transforms a switch instruction into a tree of ifs it is actually performing a binary search into the various case ranges, to see if the current value falls into one cases range of values. So, if we have a program with something like this: switch (a) { case 0: do0(); break; case 1: do1(); break; case 2: do2(); break; default: break; } the code produced is something like this: if (a < 1) { if (a == 0) { do0(); } } else { if (a < 2) { if (a == 1) { do1(); } } else { if (a == 2) { do2(); } } } This code is inefficient because the check (a == 1) to execute do1() is not needed. The reason is that because we already checked that (a >= 1) initially by checking that also (a < 2) we basically already inferred that (a == 1) without the need of an extra basic block spawned to check if actually (a == 1). The patch addresses this problem by keeping track of already checked bounds in the LowerSwitch algorithm, so that when the time arrives to produce a Leaf Block that checks the equality with the case value / range the algorithm can decide if that block is really needed depending on the already checked bounds . For example, the above with "a = 1" would work like this: the bounds start as LB: NONE , UB: NONE as (a < 1) is emitted the bounds for the else path become LB: 1 UB: NONE. This happens because by failing the test (a < 1) we know that the value "a" cannot be smaller than 1 if we enter the else branch. After the emitting the check (a < 2) the bounds in the if branch become LB: 1 UB: 1. This is because by checking that "a" is smaller than 2 then the upper bound becomes 2 - 1 = 1. When it is time to emit the leaf block for "case 1:" we notice that 1 can be squeezed exactly in between the LB and UB, which means that if we arrived to that block there is no need to emit a block that checks if (a == 1). Patch by: Marcello Maggioni <hayarms@gmail.com> llvm-svn: 211038
2014-06-16 18:55:20 +02:00
} else {
dbgs() << "NONE\n";
});
// Create a new node that checks if the value is < pivot. Go to the
// left branch if it is and right branch if not.
Function* F = OrigBlock->getParent();
BasicBlock* NewNode = BasicBlock::Create(Val->getContext(), "NodeBlock");
ICmpInst* Comp = new ICmpInst(ICmpInst::ICMP_SLT,
Val, Pivot.Low, "Pivot");
BasicBlock *LBranch = switchConvert(LHS.begin(), LHS.end(), LowerBound,
NewUpperBound, Val, NewNode, OrigBlock,
Default, UnreachableRanges);
BasicBlock *RBranch = switchConvert(RHS.begin(), RHS.end(), NewLowerBound,
UpperBound, Val, NewNode, OrigBlock,
Default, UnreachableRanges);
Function::iterator FI = OrigBlock;
F->getBasicBlockList().insert(++FI, NewNode);
NewNode->getInstList().push_back(Comp);
BranchInst::Create(LBranch, RBranch, Comp, NewNode);
return NewNode;
}
// newLeafBlock - Create a new leaf block for the binary lookup tree. It
// checks if the switch's value == the case's value. If not, then it
// jumps to the default branch. At this point in the tree, the value
// can't be another valid case value, so the jump to the "default" branch
// is warranted.
//
BasicBlock* LowerSwitch::newLeafBlock(CaseRange& Leaf, Value* Val,
BasicBlock* OrigBlock,
BasicBlock* Default)
{
Function* F = OrigBlock->getParent();
BasicBlock* NewLeaf = BasicBlock::Create(Val->getContext(), "LeafBlock");
Function::iterator FI = OrigBlock;
F->getBasicBlockList().insert(++FI, NewLeaf);
// Emit comparison
ICmpInst* Comp = nullptr;
if (Leaf.Low == Leaf.High) {
// Make the seteq instruction...
Comp = new ICmpInst(*NewLeaf, ICmpInst::ICMP_EQ, Val,
Leaf.Low, "SwitchLeaf");
} else {
// Make range comparison
if (Leaf.Low->isMinValue(true /*isSigned*/)) {
// Val >= Min && Val <= Hi --> Val <= Hi
Comp = new ICmpInst(*NewLeaf, ICmpInst::ICMP_SLE, Val, Leaf.High,
"SwitchLeaf");
} else if (Leaf.Low->isZero()) {
// Val >= 0 && Val <= Hi --> Val <=u Hi
Comp = new ICmpInst(*NewLeaf, ICmpInst::ICMP_ULE, Val, Leaf.High,
"SwitchLeaf");
} else {
// Emit V-Lo <=u Hi-Lo
Constant* NegLo = ConstantExpr::getNeg(Leaf.Low);
Instruction* Add = BinaryOperator::CreateAdd(Val, NegLo,
Val->getName()+".off",
NewLeaf);
Constant *UpperBound = ConstantExpr::getAdd(NegLo, Leaf.High);
Comp = new ICmpInst(*NewLeaf, ICmpInst::ICMP_ULE, Add, UpperBound,
"SwitchLeaf");
}
}
// Make the conditional branch...
BasicBlock* Succ = Leaf.BB;
BranchInst::Create(Succ, Default, Comp, NewLeaf);
// If there were any PHI nodes in this successor, rewrite one entry
// from OrigBlock to come from NewLeaf.
for (BasicBlock::iterator I = Succ->begin(); isa<PHINode>(I); ++I) {
PHINode* PN = cast<PHINode>(I);
// Remove all but one incoming entries from the cluster
uint64_t Range = Leaf.High->getSExtValue() -
Leaf.Low->getSExtValue();
for (uint64_t j = 0; j < Range; ++j) {
PN->removeIncomingValue(OrigBlock);
}
int BlockIdx = PN->getBasicBlockIndex(OrigBlock);
assert(BlockIdx != -1 && "Switch didn't go to this successor??");
PN->setIncomingBlock((unsigned)BlockIdx, NewLeaf);
}
return NewLeaf;
}
// Clusterify - Transform simple list of Cases into list of CaseRange's
unsigned LowerSwitch::Clusterify(CaseVector& Cases, SwitchInst *SI) {
unsigned numCmps = 0;
// Start with "simple" cases
for (SwitchInst::CaseIt i = SI->case_begin(), e = SI->case_end(); i != e; ++i)
Cases.push_back(CaseRange(i.getCaseValue(), i.getCaseValue(),
i.getCaseSuccessor()));
std::sort(Cases.begin(), Cases.end(), CaseCmp());
// Merge case into clusters
if (Cases.size()>=2)
for (CaseItr I = Cases.begin(), J = std::next(Cases.begin());
J != Cases.end();) {
int64_t nextValue = J->Low->getSExtValue();
int64_t currentValue = I->High->getSExtValue();
BasicBlock* nextBB = J->BB;
BasicBlock* currentBB = I->BB;
// If the two neighboring cases go to the same destination, merge them
// into a single case.
if ((nextValue-currentValue==1) && (currentBB == nextBB)) {
I->High = J->High;
J = Cases.erase(J);
} else {
I = J++;
}
}
for (CaseItr I=Cases.begin(), E=Cases.end(); I!=E; ++I, ++numCmps) {
if (I->Low != I->High)
// A range counts double, since it requires two compares.
++numCmps;
}
return numCmps;
}
// processSwitchInst - Replace the specified switch instruction with a sequence
// of chained if-then insts in a balanced binary search.
//
void LowerSwitch::processSwitchInst(SwitchInst *SI) {
BasicBlock *CurBlock = SI->getParent();
BasicBlock *OrigBlock = CurBlock;
Function *F = CurBlock->getParent();
Value *Val = SI->getCondition(); // The value we are switching on...
BasicBlock* Default = SI->getDefaultDest();
// If there is only the default destination, just branch.
SwitchInst refactoring. The purpose of refactoring is to hide operand roles from SwitchInst user (programmer). If you want to play with operands directly, probably you will need lower level methods than SwitchInst ones (TerminatorInst or may be User). After this patch we can reorganize SwitchInst operands and successors as we want. What was done: 1. Changed semantics of index inside the getCaseValue method: getCaseValue(0) means "get first case", not a condition. Use getCondition() if you want to resolve the condition. I propose don't mix SwitchInst case indexing with low level indexing (TI successors indexing, User's operands indexing), since it may be dangerous. 2. By the same reason findCaseValue(ConstantInt*) returns actual number of case value. 0 means first case, not default. If there is no case with given value, ErrorIndex will returned. 3. Added getCaseSuccessor method. I propose to avoid usage of TerminatorInst::getSuccessor if you want to resolve case successor BB. Use getCaseSuccessor instead, since internal SwitchInst organization of operands/successors is hidden and may be changed in any moment. 4. Added resolveSuccessorIndex and resolveCaseIndex. The main purpose of these methods is to see how case successors are really mapped in TerminatorInst. 4.1 "resolveSuccessorIndex" was created if you need to level down from SwitchInst to TerminatorInst. It returns TerminatorInst's successor index for given case successor. 4.2 "resolveCaseIndex" converts low level successors index to case index that curresponds to the given successor. Note: There are also related compatability fix patches for dragonegg, klee, llvm-gcc-4.0, llvm-gcc-4.2, safecode, clang. llvm-svn: 149481
2012-02-01 08:49:51 +01:00
if (!SI->getNumCases()) {
BranchInst::Create(Default, CurBlock);
SI->eraseFromParent();
return;
}
// Prepare cases vector.
CaseVector Cases;
unsigned numCmps = Clusterify(Cases, SI);
DEBUG(dbgs() << "Clusterify finished. Total clusters: " << Cases.size()
<< ". Total compares: " << numCmps << "\n");
DEBUG(dbgs() << "Cases: " << Cases << "\n");
(void)numCmps;
ConstantInt *LowerBound = nullptr;
ConstantInt *UpperBound = nullptr;
std::vector<IntRange> UnreachableRanges;
if (isa<UnreachableInst>(Default->getFirstNonPHIOrDbg())) {
// Make the bounds tightly fitted around the case value range, becase we
// know that the value passed to the switch must be exactly one of the case
// values.
assert(!Cases.empty());
LowerBound = Cases.front().Low;
UpperBound = Cases.back().High;
DenseMap<BasicBlock *, unsigned> Popularity;
unsigned MaxPop = 0;
BasicBlock *PopSucc = nullptr;
IntRange R = { INT64_MIN, INT64_MAX };
UnreachableRanges.push_back(R);
for (const auto &I : Cases) {
int64_t Low = I.Low->getSExtValue();
int64_t High = I.High->getSExtValue();
IntRange &LastRange = UnreachableRanges.back();
if (LastRange.Low == Low) {
// There is nothing left of the previous range.
UnreachableRanges.pop_back();
} else {
// Terminate the previous range.
assert(Low > LastRange.Low);
LastRange.High = Low - 1;
}
if (High != INT64_MAX) {
IntRange R = { High + 1, INT64_MAX };
UnreachableRanges.push_back(R);
}
// Count popularity.
int64_t N = High - Low + 1;
unsigned &Pop = Popularity[I.BB];
if ((Pop += N) > MaxPop) {
MaxPop = Pop;
PopSucc = I.BB;
}
}
#ifndef NDEBUG
/* UnreachableRanges should be sorted and the ranges non-adjacent. */
for (auto I = UnreachableRanges.begin(), E = UnreachableRanges.end();
I != E; ++I) {
assert(I->Low <= I->High);
auto Next = I + 1;
if (Next != E) {
assert(Next->Low > I->High);
}
}
#endif
// Use the most popular block as the new default, reducing the number of
// cases.
assert(MaxPop > 0 && PopSucc);
Default = PopSucc;
for (CaseItr I = Cases.begin(); I != Cases.end();) {
if (I->BB == PopSucc)
I = Cases.erase(I);
else
++I;
}
// If there are no cases left, just branch.
if (Cases.empty()) {
BranchInst::Create(Default, CurBlock);
SI->eraseFromParent();
return;
}
}
// Create a new, empty default block so that the new hierarchy of
// if-then statements go to this and the PHI nodes are happy.
BasicBlock *NewDefault = BasicBlock::Create(SI->getContext(), "NewDefault");
F->getBasicBlockList().insert(Default, NewDefault);
BranchInst::Create(Default, NewDefault);
// If there is an entry in any PHI nodes for the default edge, make sure
// to update them as well.
for (BasicBlock::iterator I = Default->begin(); isa<PHINode>(I); ++I) {
PHINode *PN = cast<PHINode>(I);
int BlockIdx = PN->getBasicBlockIndex(OrigBlock);
assert(BlockIdx != -1 && "Switch didn't go to this successor??");
PN->setIncomingBlock((unsigned)BlockIdx, NewDefault);
}
LowerSwitch: track bounding range for the condition tree. When LowerSwitch transforms a switch instruction into a tree of ifs it is actually performing a binary search into the various case ranges, to see if the current value falls into one cases range of values. So, if we have a program with something like this: switch (a) { case 0: do0(); break; case 1: do1(); break; case 2: do2(); break; default: break; } the code produced is something like this: if (a < 1) { if (a == 0) { do0(); } } else { if (a < 2) { if (a == 1) { do1(); } } else { if (a == 2) { do2(); } } } This code is inefficient because the check (a == 1) to execute do1() is not needed. The reason is that because we already checked that (a >= 1) initially by checking that also (a < 2) we basically already inferred that (a == 1) without the need of an extra basic block spawned to check if actually (a == 1). The patch addresses this problem by keeping track of already checked bounds in the LowerSwitch algorithm, so that when the time arrives to produce a Leaf Block that checks the equality with the case value / range the algorithm can decide if that block is really needed depending on the already checked bounds . For example, the above with "a = 1" would work like this: the bounds start as LB: NONE , UB: NONE as (a < 1) is emitted the bounds for the else path become LB: 1 UB: NONE. This happens because by failing the test (a < 1) we know that the value "a" cannot be smaller than 1 if we enter the else branch. After the emitting the check (a < 2) the bounds in the if branch become LB: 1 UB: 1. This is because by checking that "a" is smaller than 2 then the upper bound becomes 2 - 1 = 1. When it is time to emit the leaf block for "case 1:" we notice that 1 can be squeezed exactly in between the LB and UB, which means that if we arrived to that block there is no need to emit a block that checks if (a == 1). Patch by: Marcello Maggioni <hayarms@gmail.com> llvm-svn: 211038
2014-06-16 18:55:20 +02:00
BasicBlock *SwitchBlock =
switchConvert(Cases.begin(), Cases.end(), LowerBound, UpperBound, Val,
OrigBlock, OrigBlock, NewDefault, UnreachableRanges);
// Branch to our shiny new if-then stuff...
BranchInst::Create(SwitchBlock, OrigBlock);
// We are now done with the switch instruction, delete it.
BasicBlock *OldDefault = SI->getDefaultDest();
CurBlock->getInstList().erase(SI);
LowerSwitch: track bounding range for the condition tree. When LowerSwitch transforms a switch instruction into a tree of ifs it is actually performing a binary search into the various case ranges, to see if the current value falls into one cases range of values. So, if we have a program with something like this: switch (a) { case 0: do0(); break; case 1: do1(); break; case 2: do2(); break; default: break; } the code produced is something like this: if (a < 1) { if (a == 0) { do0(); } } else { if (a < 2) { if (a == 1) { do1(); } } else { if (a == 2) { do2(); } } } This code is inefficient because the check (a == 1) to execute do1() is not needed. The reason is that because we already checked that (a >= 1) initially by checking that also (a < 2) we basically already inferred that (a == 1) without the need of an extra basic block spawned to check if actually (a == 1). The patch addresses this problem by keeping track of already checked bounds in the LowerSwitch algorithm, so that when the time arrives to produce a Leaf Block that checks the equality with the case value / range the algorithm can decide if that block is really needed depending on the already checked bounds . For example, the above with "a = 1" would work like this: the bounds start as LB: NONE , UB: NONE as (a < 1) is emitted the bounds for the else path become LB: 1 UB: NONE. This happens because by failing the test (a < 1) we know that the value "a" cannot be smaller than 1 if we enter the else branch. After the emitting the check (a < 2) the bounds in the if branch become LB: 1 UB: 1. This is because by checking that "a" is smaller than 2 then the upper bound becomes 2 - 1 = 1. When it is time to emit the leaf block for "case 1:" we notice that 1 can be squeezed exactly in between the LB and UB, which means that if we arrived to that block there is no need to emit a block that checks if (a == 1). Patch by: Marcello Maggioni <hayarms@gmail.com> llvm-svn: 211038
2014-06-16 18:55:20 +02:00
// If the Default block has no more predecessors just remove it.
if (pred_begin(OldDefault) == pred_end(OldDefault))
DeleteDeadBlock(OldDefault);
}