1 //===- DivergenceAnalysis.cpp --------- Divergence Analysis Implementation -==//
3 // The LLVM Compiler Infrastructure
5 // This file is distributed under the University of Illinois Open Source
6 // License. See LICENSE.TXT for details.
8 //===----------------------------------------------------------------------===//
10 // This file implements divergence analysis which determines whether a branch
11 // in a GPU program is divergent.It can help branch optimizations such as jump
12 // threading and loop unswitching to make better decisions.
14 // GPU programs typically use the SIMD execution model, where multiple threads
15 // in the same execution group have to execute in lock-step. Therefore, if the
16 // code contains divergent branches (i.e., threads in a group do not agree on
17 // which path of the branch to take), the group of threads has to execute all
18 // the paths from that branch with different subsets of threads enabled until
19 // they converge at the immediately post-dominating BB of the paths.
21 // Due to this execution model, some optimizations such as jump
22 // threading and loop unswitching can be unfortunately harmful when performed on
23 // divergent branches. Therefore, an analysis that computes which branches in a
24 // GPU program are divergent can help the compiler to selectively run these
27 // This file defines divergence analysis which computes a conservative but
28 // non-trivial approximation of all divergent branches in a GPU program. It
29 // partially implements the approach described in
31 // Divergence Analysis
32 // Sampaio, Souza, Collange, Pereira
35 // The divergence analysis identifies the sources of divergence (e.g., special
36 // variables that hold the thread ID), and recursively marks variables that are
37 // data or sync dependent on a source of divergence as divergent.
39 // While data dependency is a well-known concept, the notion of sync dependency
40 // is worth more explanation. Sync dependence characterizes the control flow
41 // aspect of the propagation of branch divergence. For example,
43 // %cond = icmp slt i32 %tid, 10
44 // br i1 %cond, label %then, label %else
50 // %a = phi i32 [ 0, %then ], [ 1, %else ]
52 // Suppose %tid holds the thread ID. Although %a is not data dependent on %tid
53 // because %tid is not on its use-def chains, %a is sync dependent on %tid
54 // because the branch "br i1 %cond" depends on %tid and affects which value %a
57 // The current implementation has the following limitations:
58 // 1. intra-procedural. It conservatively considers the arguments of a
59 // non-kernel-entry function and the return value of a function call as
61 // 2. memory as black box. It conservatively considers values loaded from
62 // generic or local address as divergent. This can be improved by leveraging
65 //===----------------------------------------------------------------------===//
67 #include "llvm/Analysis/DivergenceAnalysis.h"
68 #include "llvm/Analysis/Passes.h"
69 #include "llvm/Analysis/PostDominators.h"
70 #include "llvm/Analysis/TargetTransformInfo.h"
71 #include "llvm/IR/Dominators.h"
72 #include "llvm/IR/InstIterator.h"
73 #include "llvm/IR/Instructions.h"
74 #include "llvm/IR/IntrinsicInst.h"
75 #include "llvm/IR/Value.h"
76 #include "llvm/Support/CommandLine.h"
77 #include "llvm/Support/Debug.h"
78 #include "llvm/Support/raw_ostream.h"
79 #include "llvm/Transforms/Scalar.h"
85 class DivergencePropagator {
87 DivergencePropagator(Function &F, TargetTransformInfo &TTI, DominatorTree &DT,
88 PostDominatorTree &PDT, DenseSet<const Value *> &DV)
89 : F(F), TTI(TTI), DT(DT), PDT(PDT), DV(DV) {}
90 void populateWithSourcesOfDivergence();
94 // A helper function that explores data dependents of V.
95 void exploreDataDependency(Value *V);
96 // A helper function that explores sync dependents of TI.
97 void exploreSyncDependency(TerminatorInst *TI);
98 // Computes the influence region from Start to End. This region includes all
99 // basic blocks on any path from Start to End.
100 void computeInfluenceRegion(BasicBlock *Start, BasicBlock *End,
101 DenseSet<BasicBlock *> &InfluenceRegion);
102 // Finds all users of I that are outside the influence region, and add these
103 // users to Worklist.
104 void findUsersOutsideInfluenceRegion(
105 Instruction &I, const DenseSet<BasicBlock *> &InfluenceRegion);
108 TargetTransformInfo &TTI;
110 PostDominatorTree &PDT;
111 std::vector<Value *> Worklist; // Stack for DFS.
112 DenseSet<const Value *> &DV; // Stores all divergent values.
115 void DivergencePropagator::populateWithSourcesOfDivergence() {
118 for (auto &I : instructions(F)) {
119 if (TTI.isSourceOfDivergence(&I)) {
120 Worklist.push_back(&I);
124 for (auto &Arg : F.args()) {
125 if (TTI.isSourceOfDivergence(&Arg)) {
126 Worklist.push_back(&Arg);
132 void DivergencePropagator::exploreSyncDependency(TerminatorInst *TI) {
133 // Propagation rule 1: if branch TI is divergent, all PHINodes in TI's
134 // immediate post dominator are divergent. This rule handles if-then-else
135 // patterns. For example,
141 // a = phi(a1, a2); // sync dependent on (tid < 5)
142 BasicBlock *ThisBB = TI->getParent();
143 BasicBlock *IPostDom = PDT.getNode(ThisBB)->getIDom()->getBlock();
144 if (IPostDom == nullptr)
147 for (auto I = IPostDom->begin(); isa<PHINode>(I); ++I) {
148 // A PHINode is uniform if it returns the same value no matter which path is
150 if (!cast<PHINode>(I)->hasConstantValue() && DV.insert(I).second)
151 Worklist.push_back(I);
154 // Propagation rule 2: if a value defined in a loop is used outside, the user
155 // is sync dependent on the condition of the loop exits that dominate the
156 // user. For example,
161 // if (foo(i)) ... // uniform
162 // } while (i < tid);
163 // if (bar(i)) ... // divergent
165 // A program may contain unstructured loops. Therefore, we cannot leverage
166 // LoopInfo, which only recognizes natural loops.
168 // The algorithm used here handles both natural and unstructured loops. Given
169 // a branch TI, we first compute its influence region, the union of all simple
170 // paths from TI to its immediate post dominator (IPostDom). Then, we search
171 // for all the values defined in the influence region but used outside. All
172 // these users are sync dependent on TI.
173 DenseSet<BasicBlock *> InfluenceRegion;
174 computeInfluenceRegion(ThisBB, IPostDom, InfluenceRegion);
175 // An insight that can speed up the search process is that all the in-region
176 // values that are used outside must dominate TI. Therefore, instead of
177 // searching every basic blocks in the influence region, we search all the
178 // dominators of TI until it is outside the influence region.
179 BasicBlock *InfluencedBB = ThisBB;
180 while (InfluenceRegion.count(InfluencedBB)) {
181 for (auto &I : *InfluencedBB)
182 findUsersOutsideInfluenceRegion(I, InfluenceRegion);
183 DomTreeNode *IDomNode = DT.getNode(InfluencedBB)->getIDom();
184 if (IDomNode == nullptr)
186 InfluencedBB = IDomNode->getBlock();
190 void DivergencePropagator::findUsersOutsideInfluenceRegion(
191 Instruction &I, const DenseSet<BasicBlock *> &InfluenceRegion) {
192 for (User *U : I.users()) {
193 Instruction *UserInst = cast<Instruction>(U);
194 if (!InfluenceRegion.count(UserInst->getParent())) {
195 if (DV.insert(UserInst).second)
196 Worklist.push_back(UserInst);
201 void DivergencePropagator::computeInfluenceRegion(
202 BasicBlock *Start, BasicBlock *End,
203 DenseSet<BasicBlock *> &InfluenceRegion) {
204 assert(PDT.properlyDominates(End, Start) &&
205 "End does not properly dominate Start");
206 std::vector<BasicBlock *> InfluenceStack;
207 InfluenceStack.push_back(Start);
208 InfluenceRegion.insert(Start);
209 while (!InfluenceStack.empty()) {
210 BasicBlock *BB = InfluenceStack.back();
211 InfluenceStack.pop_back();
212 for (BasicBlock *Succ : successors(BB)) {
213 if (End != Succ && InfluenceRegion.insert(Succ).second)
214 InfluenceStack.push_back(Succ);
219 void DivergencePropagator::exploreDataDependency(Value *V) {
220 // Follow def-use chains of V.
221 for (User *U : V->users()) {
222 Instruction *UserInst = cast<Instruction>(U);
223 if (DV.insert(UserInst).second)
224 Worklist.push_back(UserInst);
228 void DivergencePropagator::propagate() {
229 // Traverse the dependency graph using DFS.
230 while (!Worklist.empty()) {
231 Value *V = Worklist.back();
233 if (TerminatorInst *TI = dyn_cast<TerminatorInst>(V)) {
234 // Terminators with less than two successors won't introduce sync
235 // dependency. Ignore them.
236 if (TI->getNumSuccessors() > 1)
237 exploreSyncDependency(TI);
239 exploreDataDependency(V);
243 } /// end namespace anonymous
245 // Register this pass.
246 char DivergenceAnalysis::ID = 0;
247 INITIALIZE_PASS_BEGIN(DivergenceAnalysis, "divergence", "Divergence Analysis",
249 INITIALIZE_PASS_DEPENDENCY(DominatorTreeWrapperPass)
250 INITIALIZE_PASS_DEPENDENCY(PostDominatorTree)
251 INITIALIZE_PASS_END(DivergenceAnalysis, "divergence", "Divergence Analysis",
254 FunctionPass *llvm::createDivergenceAnalysisPass() {
255 return new DivergenceAnalysis();
258 void DivergenceAnalysis::getAnalysisUsage(AnalysisUsage &AU) const {
259 AU.addRequired<DominatorTreeWrapperPass>();
260 AU.addRequired<PostDominatorTree>();
261 AU.setPreservesAll();
264 bool DivergenceAnalysis::runOnFunction(Function &F) {
265 auto *TTIWP = getAnalysisIfAvailable<TargetTransformInfoWrapperPass>();
266 if (TTIWP == nullptr)
269 TargetTransformInfo &TTI = TTIWP->getTTI(F);
270 // Fast path: if the target does not have branch divergence, we do not mark
271 // any branch as divergent.
272 if (!TTI.hasBranchDivergence())
275 DivergentValues.clear();
276 DivergencePropagator DP(F, TTI,
277 getAnalysis<DominatorTreeWrapperPass>().getDomTree(),
278 getAnalysis<PostDominatorTree>(), DivergentValues);
279 DP.populateWithSourcesOfDivergence();
284 void DivergenceAnalysis::print(raw_ostream &OS, const Module *) const {
285 if (DivergentValues.empty())
287 const Value *FirstDivergentValue = *DivergentValues.begin();
289 if (const Argument *Arg = dyn_cast<Argument>(FirstDivergentValue)) {
290 F = Arg->getParent();
291 } else if (const Instruction *I =
292 dyn_cast<Instruction>(FirstDivergentValue)) {
293 F = I->getParent()->getParent();
295 llvm_unreachable("Only arguments and instructions can be divergent");
298 // Dumps all divergent values in F, arguments and then instructions.
299 for (auto &Arg : F->args()) {
300 if (DivergentValues.count(&Arg))
301 OS << "DIVERGENT: " << Arg << "\n";
303 // Iterate instructions using instructions() to ensure a deterministic order.
304 for (auto &I : instructions(F)) {
305 if (DivergentValues.count(&I))
306 OS << "DIVERGENT:" << I << "\n";