Summary: At the exit of the loop, the compiler uses a register to remember and accumulate the number of threads that have already exited. When all active threads exit the loop, this register is used to restore the exec mask, and the execution continues for the post loop code. When there is a "continue" in the loop, the compiler made a mistake to reset the register to 0 when the "continue" backedge is taken. This will result in some threads not executing the post loop code as they are supposed to. This patch fixed the issue. Reviewers: nhaehnle, arsenm Differential Revision: https://reviews.llvm.org/D59312 llvm-svn: 356298
366 lines
11 KiB
C++
366 lines
11 KiB
C++
//===- SIAnnotateControlFlow.cpp ------------------------------------------===//
|
|
//
|
|
// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
|
|
// See https://llvm.org/LICENSE.txt for license information.
|
|
// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
|
|
//
|
|
//===----------------------------------------------------------------------===//
|
|
//
|
|
/// \file
|
|
/// Annotates the control flow with hardware specific intrinsics.
|
|
//
|
|
//===----------------------------------------------------------------------===//
|
|
|
|
#include "AMDGPU.h"
|
|
#include "llvm/ADT/DepthFirstIterator.h"
|
|
#include "llvm/ADT/STLExtras.h"
|
|
#include "llvm/ADT/SmallVector.h"
|
|
#include "llvm/Analysis/LegacyDivergenceAnalysis.h"
|
|
#include "llvm/Analysis/LoopInfo.h"
|
|
#include "llvm/Transforms/Utils/Local.h"
|
|
#include "llvm/IR/BasicBlock.h"
|
|
#include "llvm/IR/CFG.h"
|
|
#include "llvm/IR/Constant.h"
|
|
#include "llvm/IR/Constants.h"
|
|
#include "llvm/IR/DerivedTypes.h"
|
|
#include "llvm/IR/Dominators.h"
|
|
#include "llvm/IR/Function.h"
|
|
#include "llvm/IR/Instruction.h"
|
|
#include "llvm/IR/Instructions.h"
|
|
#include "llvm/IR/Intrinsics.h"
|
|
#include "llvm/IR/Module.h"
|
|
#include "llvm/IR/Type.h"
|
|
#include "llvm/IR/ValueHandle.h"
|
|
#include "llvm/Pass.h"
|
|
#include "llvm/Support/Casting.h"
|
|
#include "llvm/Support/Debug.h"
|
|
#include "llvm/Support/ErrorHandling.h"
|
|
#include "llvm/Support/raw_ostream.h"
|
|
#include "llvm/Transforms/Utils/BasicBlockUtils.h"
|
|
#include <cassert>
|
|
#include <utility>
|
|
|
|
using namespace llvm;
|
|
|
|
#define DEBUG_TYPE "si-annotate-control-flow"
|
|
|
|
namespace {
|
|
|
|
// Complex types used in this pass
|
|
using StackEntry = std::pair<BasicBlock *, Value *>;
|
|
using StackVector = SmallVector<StackEntry, 16>;
|
|
|
|
class SIAnnotateControlFlow : public FunctionPass {
|
|
LegacyDivergenceAnalysis *DA;
|
|
|
|
Type *Boolean;
|
|
Type *Void;
|
|
Type *Int64;
|
|
Type *ReturnStruct;
|
|
|
|
ConstantInt *BoolTrue;
|
|
ConstantInt *BoolFalse;
|
|
UndefValue *BoolUndef;
|
|
Constant *Int64Zero;
|
|
|
|
Function *If;
|
|
Function *Else;
|
|
Function *IfBreak;
|
|
Function *Loop;
|
|
Function *EndCf;
|
|
|
|
DominatorTree *DT;
|
|
StackVector Stack;
|
|
|
|
LoopInfo *LI;
|
|
|
|
bool isUniform(BranchInst *T);
|
|
|
|
bool isTopOfStack(BasicBlock *BB);
|
|
|
|
Value *popSaved();
|
|
|
|
void push(BasicBlock *BB, Value *Saved);
|
|
|
|
bool isElse(PHINode *Phi);
|
|
|
|
void eraseIfUnused(PHINode *Phi);
|
|
|
|
void openIf(BranchInst *Term);
|
|
|
|
void insertElse(BranchInst *Term);
|
|
|
|
Value *
|
|
handleLoopCondition(Value *Cond, PHINode *Broken, llvm::Loop *L,
|
|
BranchInst *Term);
|
|
|
|
void handleLoop(BranchInst *Term);
|
|
|
|
void closeControlFlow(BasicBlock *BB);
|
|
|
|
public:
|
|
static char ID;
|
|
|
|
SIAnnotateControlFlow() : FunctionPass(ID) {}
|
|
|
|
bool doInitialization(Module &M) override;
|
|
|
|
bool runOnFunction(Function &F) override;
|
|
|
|
StringRef getPassName() const override { return "SI annotate control flow"; }
|
|
|
|
void getAnalysisUsage(AnalysisUsage &AU) const override {
|
|
AU.addRequired<LoopInfoWrapperPass>();
|
|
AU.addRequired<DominatorTreeWrapperPass>();
|
|
AU.addRequired<LegacyDivergenceAnalysis>();
|
|
AU.addPreserved<DominatorTreeWrapperPass>();
|
|
FunctionPass::getAnalysisUsage(AU);
|
|
}
|
|
};
|
|
|
|
} // end anonymous namespace
|
|
|
|
INITIALIZE_PASS_BEGIN(SIAnnotateControlFlow, DEBUG_TYPE,
|
|
"Annotate SI Control Flow", false, false)
|
|
INITIALIZE_PASS_DEPENDENCY(DominatorTreeWrapperPass)
|
|
INITIALIZE_PASS_DEPENDENCY(LegacyDivergenceAnalysis)
|
|
INITIALIZE_PASS_END(SIAnnotateControlFlow, DEBUG_TYPE,
|
|
"Annotate SI Control Flow", false, false)
|
|
|
|
char SIAnnotateControlFlow::ID = 0;
|
|
|
|
/// Initialize all the types and constants used in the pass
|
|
bool SIAnnotateControlFlow::doInitialization(Module &M) {
|
|
LLVMContext &Context = M.getContext();
|
|
|
|
Void = Type::getVoidTy(Context);
|
|
Boolean = Type::getInt1Ty(Context);
|
|
Int64 = Type::getInt64Ty(Context);
|
|
ReturnStruct = StructType::get(Boolean, Int64);
|
|
|
|
BoolTrue = ConstantInt::getTrue(Context);
|
|
BoolFalse = ConstantInt::getFalse(Context);
|
|
BoolUndef = UndefValue::get(Boolean);
|
|
Int64Zero = ConstantInt::get(Int64, 0);
|
|
|
|
If = Intrinsic::getDeclaration(&M, Intrinsic::amdgcn_if);
|
|
Else = Intrinsic::getDeclaration(&M, Intrinsic::amdgcn_else);
|
|
IfBreak = Intrinsic::getDeclaration(&M, Intrinsic::amdgcn_if_break);
|
|
Loop = Intrinsic::getDeclaration(&M, Intrinsic::amdgcn_loop);
|
|
EndCf = Intrinsic::getDeclaration(&M, Intrinsic::amdgcn_end_cf);
|
|
return false;
|
|
}
|
|
|
|
/// Is the branch condition uniform or did the StructurizeCFG pass
|
|
/// consider it as such?
|
|
bool SIAnnotateControlFlow::isUniform(BranchInst *T) {
|
|
return DA->isUniform(T) ||
|
|
T->getMetadata("structurizecfg.uniform") != nullptr;
|
|
}
|
|
|
|
/// Is BB the last block saved on the stack ?
|
|
bool SIAnnotateControlFlow::isTopOfStack(BasicBlock *BB) {
|
|
return !Stack.empty() && Stack.back().first == BB;
|
|
}
|
|
|
|
/// Pop the last saved value from the control flow stack
|
|
Value *SIAnnotateControlFlow::popSaved() {
|
|
return Stack.pop_back_val().second;
|
|
}
|
|
|
|
/// Push a BB and saved value to the control flow stack
|
|
void SIAnnotateControlFlow::push(BasicBlock *BB, Value *Saved) {
|
|
Stack.push_back(std::make_pair(BB, Saved));
|
|
}
|
|
|
|
/// Can the condition represented by this PHI node treated like
|
|
/// an "Else" block?
|
|
bool SIAnnotateControlFlow::isElse(PHINode *Phi) {
|
|
BasicBlock *IDom = DT->getNode(Phi->getParent())->getIDom()->getBlock();
|
|
for (unsigned i = 0, e = Phi->getNumIncomingValues(); i != e; ++i) {
|
|
if (Phi->getIncomingBlock(i) == IDom) {
|
|
|
|
if (Phi->getIncomingValue(i) != BoolTrue)
|
|
return false;
|
|
|
|
} else {
|
|
if (Phi->getIncomingValue(i) != BoolFalse)
|
|
return false;
|
|
|
|
}
|
|
}
|
|
return true;
|
|
}
|
|
|
|
// Erase "Phi" if it is not used any more
|
|
void SIAnnotateControlFlow::eraseIfUnused(PHINode *Phi) {
|
|
if (RecursivelyDeleteDeadPHINode(Phi)) {
|
|
LLVM_DEBUG(dbgs() << "Erased unused condition phi\n");
|
|
}
|
|
}
|
|
|
|
/// Open a new "If" block
|
|
void SIAnnotateControlFlow::openIf(BranchInst *Term) {
|
|
if (isUniform(Term))
|
|
return;
|
|
|
|
Value *Ret = CallInst::Create(If, Term->getCondition(), "", Term);
|
|
Term->setCondition(ExtractValueInst::Create(Ret, 0, "", Term));
|
|
push(Term->getSuccessor(1), ExtractValueInst::Create(Ret, 1, "", Term));
|
|
}
|
|
|
|
/// Close the last "If" block and open a new "Else" block
|
|
void SIAnnotateControlFlow::insertElse(BranchInst *Term) {
|
|
if (isUniform(Term)) {
|
|
return;
|
|
}
|
|
Value *Ret = CallInst::Create(Else, popSaved(), "", Term);
|
|
Term->setCondition(ExtractValueInst::Create(Ret, 0, "", Term));
|
|
push(Term->getSuccessor(1), ExtractValueInst::Create(Ret, 1, "", Term));
|
|
}
|
|
|
|
/// Recursively handle the condition leading to a loop
|
|
Value *SIAnnotateControlFlow::handleLoopCondition(
|
|
Value *Cond, PHINode *Broken, llvm::Loop *L, BranchInst *Term) {
|
|
if (Instruction *Inst = dyn_cast<Instruction>(Cond)) {
|
|
BasicBlock *Parent = Inst->getParent();
|
|
Instruction *Insert;
|
|
if (L->contains(Inst)) {
|
|
Insert = Parent->getTerminator();
|
|
} else {
|
|
Insert = L->getHeader()->getFirstNonPHIOrDbgOrLifetime();
|
|
}
|
|
|
|
Value *Args[] = { Cond, Broken };
|
|
return CallInst::Create(IfBreak, Args, "", Insert);
|
|
}
|
|
|
|
// Insert IfBreak in the loop header TERM for constant COND other than true.
|
|
if (isa<Constant>(Cond)) {
|
|
Instruction *Insert = Cond == BoolTrue ?
|
|
Term : L->getHeader()->getTerminator();
|
|
|
|
Value *Args[] = { Cond, Broken };
|
|
return CallInst::Create(IfBreak, Args, "", Insert);
|
|
}
|
|
|
|
llvm_unreachable("Unhandled loop condition!");
|
|
}
|
|
|
|
/// Handle a back edge (loop)
|
|
void SIAnnotateControlFlow::handleLoop(BranchInst *Term) {
|
|
if (isUniform(Term))
|
|
return;
|
|
|
|
BasicBlock *BB = Term->getParent();
|
|
llvm::Loop *L = LI->getLoopFor(BB);
|
|
if (!L)
|
|
return;
|
|
|
|
BasicBlock *Target = Term->getSuccessor(1);
|
|
PHINode *Broken = PHINode::Create(Int64, 0, "phi.broken", &Target->front());
|
|
|
|
Value *Cond = Term->getCondition();
|
|
Term->setCondition(BoolTrue);
|
|
Value *Arg = handleLoopCondition(Cond, Broken, L, Term);
|
|
|
|
for (BasicBlock *Pred : predecessors(Target)) {
|
|
Value *PHIValue = Int64Zero;
|
|
if (Pred == BB) // Remember the value of the previous iteration.
|
|
PHIValue = Arg;
|
|
// If the backedge from Pred to Target could be executed before the exit
|
|
// of the loop at BB, it should not reset or change "Broken", which keeps
|
|
// track of the number of threads exited the loop at BB.
|
|
else if (L->contains(Pred) && DT->dominates(Pred, BB))
|
|
PHIValue = Broken;
|
|
Broken->addIncoming(PHIValue, Pred);
|
|
}
|
|
|
|
Term->setCondition(CallInst::Create(Loop, Arg, "", Term));
|
|
|
|
push(Term->getSuccessor(0), Arg);
|
|
}
|
|
|
|
/// Close the last opened control flow
|
|
void SIAnnotateControlFlow::closeControlFlow(BasicBlock *BB) {
|
|
llvm::Loop *L = LI->getLoopFor(BB);
|
|
|
|
assert(Stack.back().first == BB);
|
|
|
|
if (L && L->getHeader() == BB) {
|
|
// We can't insert an EndCF call into a loop header, because it will
|
|
// get executed on every iteration of the loop, when it should be
|
|
// executed only once before the loop.
|
|
SmallVector <BasicBlock *, 8> Latches;
|
|
L->getLoopLatches(Latches);
|
|
|
|
SmallVector<BasicBlock *, 2> Preds;
|
|
for (BasicBlock *Pred : predecessors(BB)) {
|
|
if (!is_contained(Latches, Pred))
|
|
Preds.push_back(Pred);
|
|
}
|
|
|
|
BB = SplitBlockPredecessors(BB, Preds, "endcf.split", DT, LI, nullptr,
|
|
false);
|
|
}
|
|
|
|
Value *Exec = popSaved();
|
|
Instruction *FirstInsertionPt = &*BB->getFirstInsertionPt();
|
|
if (!isa<UndefValue>(Exec) && !isa<UnreachableInst>(FirstInsertionPt))
|
|
CallInst::Create(EndCf, Exec, "", FirstInsertionPt);
|
|
}
|
|
|
|
/// Annotate the control flow with intrinsics so the backend can
|
|
/// recognize if/then/else and loops.
|
|
bool SIAnnotateControlFlow::runOnFunction(Function &F) {
|
|
DT = &getAnalysis<DominatorTreeWrapperPass>().getDomTree();
|
|
LI = &getAnalysis<LoopInfoWrapperPass>().getLoopInfo();
|
|
DA = &getAnalysis<LegacyDivergenceAnalysis>();
|
|
|
|
for (df_iterator<BasicBlock *> I = df_begin(&F.getEntryBlock()),
|
|
E = df_end(&F.getEntryBlock()); I != E; ++I) {
|
|
BasicBlock *BB = *I;
|
|
BranchInst *Term = dyn_cast<BranchInst>(BB->getTerminator());
|
|
|
|
if (!Term || Term->isUnconditional()) {
|
|
if (isTopOfStack(BB))
|
|
closeControlFlow(BB);
|
|
|
|
continue;
|
|
}
|
|
|
|
if (I.nodeVisited(Term->getSuccessor(1))) {
|
|
if (isTopOfStack(BB))
|
|
closeControlFlow(BB);
|
|
|
|
handleLoop(Term);
|
|
continue;
|
|
}
|
|
|
|
if (isTopOfStack(BB)) {
|
|
PHINode *Phi = dyn_cast<PHINode>(Term->getCondition());
|
|
if (Phi && Phi->getParent() == BB && isElse(Phi)) {
|
|
insertElse(Term);
|
|
eraseIfUnused(Phi);
|
|
continue;
|
|
}
|
|
|
|
closeControlFlow(BB);
|
|
}
|
|
|
|
openIf(Term);
|
|
}
|
|
|
|
if (!Stack.empty()) {
|
|
// CFG was probably not structured.
|
|
report_fatal_error("failed to annotate CFG");
|
|
}
|
|
|
|
return true;
|
|
}
|
|
|
|
/// Create the annotation pass
|
|
FunctionPass *llvm::createSIAnnotateControlFlowPass() {
|
|
return new SIAnnotateControlFlow();
|
|
}
|