1//===- SplitModule.cpp - Split a module into partitions -------------------===//
2//
3// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
4// See https://llvm.org/LICENSE.txt for license information.
5// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
6//
7//===----------------------------------------------------------------------===//
8//
9// This file defines the function llvm::SplitModule, which splits a module
10// into multiple linkable partitions. It can be used to implement parallel code
11// generation for link-time optimization.
12//
13//===----------------------------------------------------------------------===//
14
15#include "llvm/Transforms/Utils/SplitModule.h"
16#include "llvm/ADT/DenseMap.h"
17#include "llvm/ADT/EquivalenceClasses.h"
18#include "llvm/ADT/SmallPtrSet.h"
19#include "llvm/ADT/SmallVector.h"
20#include "llvm/ADT/StringRef.h"
21#include "llvm/IR/Comdat.h"
22#include "llvm/IR/Constant.h"
23#include "llvm/IR/Constants.h"
24#include "llvm/IR/Function.h"
25#include "llvm/IR/GlobalAlias.h"
26#include "llvm/IR/GlobalObject.h"
27#include "llvm/IR/GlobalValue.h"
28#include "llvm/IR/GlobalVariable.h"
29#include "llvm/IR/Instruction.h"
30#include "llvm/IR/Module.h"
31#include "llvm/IR/User.h"
32#include "llvm/IR/Value.h"
33#include "llvm/Support/Casting.h"
34#include "llvm/Support/Debug.h"
35#include "llvm/Support/ErrorHandling.h"
36#include "llvm/Support/MD5.h"
37#include "llvm/Support/raw_ostream.h"
38#include "llvm/Transforms/Utils/Cloning.h"
39#include "llvm/Transforms/Utils/ValueMapper.h"
40#include <algorithm>
41#include <cassert>
42#include <iterator>
43#include <memory>
44#include <queue>
45#include <utility>
46#include <vector>
47
48using namespace llvm;
49
50#define DEBUG_TYPE "split-module"
51
52namespace {
53
54using ClusterMapType = EquivalenceClasses<const GlobalValue *>;
55using ComdatMembersType = DenseMap<const Comdat *, const GlobalValue *>;
56using ClusterIDMapType = DenseMap<const GlobalValue *, unsigned>;
57
58bool compareClusters(const std::pair<unsigned, unsigned> &A,
59 const std::pair<unsigned, unsigned> &B) {
60 if (A.second || B.second)
61 return A.second > B.second;
62 return A.first > B.first;
63}
64
65using BalancingQueueType =
66 std::priority_queue<std::pair<unsigned, unsigned>,
67 std::vector<std::pair<unsigned, unsigned>>,
68 decltype(compareClusters) *>;
69
70} // end anonymous namespace
71
72static void addNonConstUser(ClusterMapType &GVtoClusterMap,
73 const GlobalValue *GV, const User *U) {
74 assert((!isa<Constant>(U) || isa<GlobalValue>(U)) && "Bad user");
75
76 if (const Instruction *I = dyn_cast<Instruction>(Val: U)) {
77 const GlobalValue *F = I->getParent()->getParent();
78 GVtoClusterMap.unionSets(V1: GV, V2: F);
79 } else if (const GlobalValue *GVU = dyn_cast<GlobalValue>(Val: U)) {
80 GVtoClusterMap.unionSets(V1: GV, V2: GVU);
81 } else {
82 llvm_unreachable("Underimplemented use case");
83 }
84}
85
86// Adds all GlobalValue users of V to the same cluster as GV.
87static void addAllGlobalValueUsers(ClusterMapType &GVtoClusterMap,
88 const GlobalValue *GV, const Value *V) {
89 for (const auto *U : V->users()) {
90 SmallVector<const User *, 4> Worklist;
91 Worklist.push_back(Elt: U);
92 while (!Worklist.empty()) {
93 const User *UU = Worklist.pop_back_val();
94 // For each constant that is not a GV (a pure const) recurse.
95 if (isa<Constant>(Val: UU) && !isa<GlobalValue>(Val: UU)) {
96 Worklist.append(in_start: UU->user_begin(), in_end: UU->user_end());
97 continue;
98 }
99 addNonConstUser(GVtoClusterMap, GV, U: UU);
100 }
101 }
102}
103
104static const GlobalObject *getGVPartitioningRoot(const GlobalValue *GV) {
105 const GlobalObject *GO = GV->getAliaseeObject();
106 if (const auto *GI = dyn_cast_or_null<GlobalIFunc>(Val: GO))
107 GO = GI->getResolverFunction();
108 return GO;
109}
110
111// Find partitions for module in the way that no locals need to be
112// globalized.
113// Try to balance pack those partitions into N files since this roughly equals
114// thread balancing for the backend codegen step.
115static void findPartitions(Module &M, ClusterIDMapType &ClusterIDMap,
116 unsigned N) {
117 // At this point module should have the proper mix of globals and locals.
118 // As we attempt to partition this module, we must not change any
119 // locals to globals.
120 LLVM_DEBUG(dbgs() << "Partition module with (" << M.size()
121 << ") functions\n");
122 ClusterMapType GVtoClusterMap;
123 ComdatMembersType ComdatMembers;
124
125 auto recordGVSet = [&GVtoClusterMap, &ComdatMembers](GlobalValue &GV) {
126 if (GV.isDeclaration())
127 return;
128
129 if (!GV.hasName())
130 GV.setName("__llvmsplit_unnamed");
131
132 // Comdat groups must not be partitioned. For comdat groups that contain
133 // locals, record all their members here so we can keep them together.
134 // Comdat groups that only contain external globals are already handled by
135 // the MD5-based partitioning.
136 if (const Comdat *C = GV.getComdat()) {
137 auto &Member = ComdatMembers[C];
138 if (Member)
139 GVtoClusterMap.unionSets(V1: Member, V2: &GV);
140 else
141 Member = &GV;
142 }
143
144 // Aliases should not be separated from their aliasees and ifuncs should
145 // not be separated from their resolvers regardless of linkage.
146 if (const GlobalObject *Root = getGVPartitioningRoot(GV: &GV))
147 if (&GV != Root)
148 GVtoClusterMap.unionSets(V1: &GV, V2: Root);
149
150 if (const Function *F = dyn_cast<Function>(Val: &GV)) {
151 for (const BasicBlock &BB : *F) {
152 BlockAddress *BA = BlockAddress::lookup(BB: &BB);
153 if (!BA || !BA->isConstantUsed())
154 continue;
155 addAllGlobalValueUsers(GVtoClusterMap, GV: F, V: BA);
156 }
157 }
158
159 if (GV.hasLocalLinkage())
160 addAllGlobalValueUsers(GVtoClusterMap, GV: &GV, V: &GV);
161 };
162
163 llvm::for_each(Range: M.functions(), F: recordGVSet);
164 llvm::for_each(Range: M.globals(), F: recordGVSet);
165 llvm::for_each(Range: M.aliases(), F: recordGVSet);
166
167 // Assigned all GVs to merged clusters while balancing number of objects in
168 // each.
169 BalancingQueueType BalancingQueue(compareClusters);
170 // Pre-populate priority queue with N slot blanks.
171 for (unsigned i = 0; i < N; ++i)
172 BalancingQueue.push(x: std::make_pair(x&: i, y: 0));
173
174 using SortType = std::pair<unsigned, ClusterMapType::iterator>;
175
176 SmallVector<SortType, 64> Sets;
177 SmallPtrSet<const GlobalValue *, 32> Visited;
178
179 // To guarantee determinism, we have to sort SCC according to size.
180 // When size is the same, use leader's name.
181 for (ClusterMapType::iterator I = GVtoClusterMap.begin(),
182 E = GVtoClusterMap.end();
183 I != E; ++I)
184 if (I->isLeader())
185 Sets.push_back(
186 Elt: std::make_pair(x: std::distance(first: GVtoClusterMap.member_begin(I),
187 last: GVtoClusterMap.member_end()),
188 y&: I));
189
190 llvm::sort(C&: Sets, Comp: [](const SortType &a, const SortType &b) {
191 if (a.first == b.first)
192 return a.second->getData()->getName() > b.second->getData()->getName();
193 else
194 return a.first > b.first;
195 });
196
197 for (auto &I : Sets) {
198 unsigned CurrentClusterID = BalancingQueue.top().first;
199 unsigned CurrentClusterSize = BalancingQueue.top().second;
200 BalancingQueue.pop();
201
202 LLVM_DEBUG(dbgs() << "Root[" << CurrentClusterID << "] cluster_size("
203 << I.first << ") ----> " << I.second->getData()->getName()
204 << "\n");
205
206 for (ClusterMapType::member_iterator MI =
207 GVtoClusterMap.findLeader(I: I.second);
208 MI != GVtoClusterMap.member_end(); ++MI) {
209 if (!Visited.insert(Ptr: *MI).second)
210 continue;
211 LLVM_DEBUG(dbgs() << "----> " << (*MI)->getName()
212 << ((*MI)->hasLocalLinkage() ? " l " : " e ") << "\n");
213 Visited.insert(Ptr: *MI);
214 ClusterIDMap[*MI] = CurrentClusterID;
215 CurrentClusterSize++;
216 }
217 // Add this set size to the number of entries in this cluster.
218 BalancingQueue.push(x: std::make_pair(x&: CurrentClusterID, y&: CurrentClusterSize));
219 }
220}
221
222static void externalize(GlobalValue *GV) {
223 if (GV->hasLocalLinkage()) {
224 GV->setLinkage(GlobalValue::ExternalLinkage);
225 GV->setVisibility(GlobalValue::HiddenVisibility);
226 }
227
228 // Unnamed entities must be named consistently between modules. setName will
229 // give a distinct name to each such entity.
230 if (!GV->hasName())
231 GV->setName("__llvmsplit_unnamed");
232}
233
234// Returns whether GV should be in partition (0-based) I of N.
235static bool isInPartition(const GlobalValue *GV, unsigned I, unsigned N) {
236 if (const GlobalObject *Root = getGVPartitioningRoot(GV))
237 GV = Root;
238
239 StringRef Name;
240 if (const Comdat *C = GV->getComdat())
241 Name = C->getName();
242 else
243 Name = GV->getName();
244
245 // Partition by MD5 hash. We only need a few bits for evenness as the number
246 // of partitions will generally be in the 1-2 figure range; the low 16 bits
247 // are enough.
248 MD5 H;
249 MD5::MD5Result R;
250 H.update(Str: Name);
251 H.final(Result&: R);
252 return (R[0] | (R[1] << 8)) % N == I;
253}
254
255void llvm::SplitModule(
256 Module &M, unsigned N,
257 function_ref<void(std::unique_ptr<Module> MPart)> ModuleCallback,
258 bool PreserveLocals, bool RoundRobin) {
259 if (!PreserveLocals) {
260 for (Function &F : M)
261 externalize(GV: &F);
262 for (GlobalVariable &GV : M.globals())
263 externalize(GV: &GV);
264 for (GlobalAlias &GA : M.aliases())
265 externalize(GV: &GA);
266 for (GlobalIFunc &GIF : M.ifuncs())
267 externalize(GV: &GIF);
268 }
269
270 // This performs splitting without a need for externalization, which might not
271 // always be possible.
272 ClusterIDMapType ClusterIDMap;
273 findPartitions(M, ClusterIDMap, N);
274
275 // Find functions not mapped to modules in ClusterIDMap and count functions
276 // per module. Map unmapped functions using round-robin so that they skip
277 // being distributed by isInPartition() based on function name hashes below.
278 // This provides better uniformity of distribution of functions to modules
279 // in some cases - for example when the number of functions equals to N.
280 if (RoundRobin) {
281 DenseMap<unsigned, unsigned> ModuleFunctionCount;
282 SmallVector<const GlobalValue *> UnmappedFunctions;
283 for (const auto &F : M.functions()) {
284 if (F.isDeclaration() ||
285 F.getLinkage() != GlobalValue::LinkageTypes::ExternalLinkage)
286 continue;
287 auto It = ClusterIDMap.find(Val: &F);
288 if (It == ClusterIDMap.end())
289 UnmappedFunctions.push_back(Elt: &F);
290 else
291 ++ModuleFunctionCount[It->second];
292 }
293 BalancingQueueType BalancingQueue(compareClusters);
294 for (unsigned I = 0; I < N; ++I) {
295 if (auto It = ModuleFunctionCount.find(Val: I);
296 It != ModuleFunctionCount.end())
297 BalancingQueue.push(x: *It);
298 else
299 BalancingQueue.push(x: {I, 0});
300 }
301 for (const auto *const F : UnmappedFunctions) {
302 const unsigned I = BalancingQueue.top().first;
303 const unsigned Count = BalancingQueue.top().second;
304 BalancingQueue.pop();
305 ClusterIDMap.insert(KV: {F, I});
306 BalancingQueue.push(x: {I, Count + 1});
307 }
308 }
309
310 // FIXME: We should be able to reuse M as the last partition instead of
311 // cloning it. Note that the callers at the moment expect the module to
312 // be preserved, so will need some adjustments as well.
313 for (unsigned I = 0; I < N; ++I) {
314 ValueToValueMapTy VMap;
315 std::unique_ptr<Module> MPart(
316 CloneModule(M, VMap, ShouldCloneDefinition: [&](const GlobalValue *GV) {
317 if (auto It = ClusterIDMap.find(Val: GV); It != ClusterIDMap.end())
318 return It->second == I;
319 else
320 return isInPartition(GV, I, N);
321 }));
322 if (I != 0)
323 MPart->setModuleInlineAsm("");
324 ModuleCallback(std::move(MPart));
325 }
326}
327