Skip to content

Commit d2df54e

Browse files
teresajohnsonteresajohnson
committedAug 2, 2019
[ThinLTO] Implement index-based WPD
This patch adds support to the WholeProgramDevirt pass to perform index-based WPD, which is invoked from ThinLTO during the thin link. The ThinLTO backend (WPD import phase) behaves the same regardless of whether the WPD decisions were made with the index-based or (the existing) IR-based analysis. Depends on D54815. Reviewers: pcc Subscribers: mehdi_amini, inglorion, eraman, steven_wu, dexonsmith, arphaman, dang, llvm-commits Differential Revision: https://reviews.llvm.org/D55153 llvm-svn: 367679
·
llvmorg-21-initllvmorg-10.0.0-rc1
1 parent 4cfd015 commit d2df54e

File tree

9 files changed

+805
-15
lines changed

9 files changed

+805
-15
lines changed
 

‎llvm/include/llvm/IR/ModuleSummaryIndex.h

+8
Original file line numberDiff line numberDiff line change
@@ -632,6 +632,8 @@ class FunctionSummary : public GlobalValueSummary {
632632
/// Return the list of <CalleeValueInfo, CalleeInfo> pairs.
633633
ArrayRef<EdgeTy> calls() const { return CallGraphEdgeList; }
634634

635+
void addCall(EdgeTy E) { CallGraphEdgeList.push_back(E); }
636+
635637
/// Returns the list of type identifiers used by this function in
636638
/// llvm.type.test intrinsics other than by an llvm.assume intrinsic,
637639
/// represented as GUIDs.
@@ -1293,6 +1295,12 @@ class ModuleSummaryIndex {
12931295
return nullptr;
12941296
}
12951297

1298+
TypeIdSummary *getTypeIdSummary(StringRef TypeId) {
1299+
return const_cast<TypeIdSummary *>(
1300+
static_cast<const ModuleSummaryIndex *>(this)->getTypeIdSummary(
1301+
TypeId));
1302+
}
1303+
12961304
const std::map<std::string, TypeIdCompatibleVtableInfo> &
12971305
typeIdCompatibleVtableMap() const {
12981306
return TypeIdCompatibleVtableMap;

‎llvm/include/llvm/Transforms/IPO/WholeProgramDevirt.h

+26
Original file line numberDiff line numberDiff line change
@@ -16,8 +16,10 @@
1616

1717
#include "llvm/IR/Module.h"
1818
#include "llvm/IR/PassManager.h"
19+
#include "llvm/Transforms/IPO/FunctionImport.h"
1920
#include <cassert>
2021
#include <cstdint>
22+
#include <set>
2123
#include <utility>
2224
#include <vector>
2325

@@ -28,6 +30,7 @@ template <typename T> class MutableArrayRef;
2830
class Function;
2931
class GlobalVariable;
3032
class ModuleSummaryIndex;
33+
struct ValueInfo;
3134

3235
namespace wholeprogramdevirt {
3336

@@ -228,6 +231,29 @@ struct WholeProgramDevirtPass : public PassInfoMixin<WholeProgramDevirtPass> {
228231
PreservedAnalyses run(Module &M, ModuleAnalysisManager &);
229232
};
230233

234+
struct VTableSlotSummary {
235+
StringRef TypeID;
236+
uint64_t ByteOffset;
237+
};
238+
239+
/// Perform index-based whole program devirtualization on the \p Summary
240+
/// index. Any devirtualized targets used by a type test in another module
241+
/// are added to the \p ExportedGUIDs set. For any local devirtualized targets
242+
/// only used within the defining module, the information necessary for
243+
/// locating the corresponding WPD resolution is recorded for the ValueInfo
244+
/// in case it is exported by cross module importing (in which case the
245+
/// devirtualized target name will need adjustment).
246+
void runWholeProgramDevirtOnIndex(
247+
ModuleSummaryIndex &Summary, std::set<GlobalValue::GUID> &ExportedGUIDs,
248+
std::map<ValueInfo, std::vector<VTableSlotSummary>> &LocalWPDTargetsMap);
249+
250+
/// Call after cross-module importing to update the recorded single impl
251+
/// devirt target names for any locals that were exported.
252+
void updateIndexWPDForExports(
253+
ModuleSummaryIndex &Summary,
254+
StringMap<FunctionImporter::ExportSetTy> &ExportLists,
255+
std::map<ValueInfo, std::vector<VTableSlotSummary>> &LocalWPDTargetsMap);
256+
231257
} // end namespace llvm
232258

233259
#endif // LLVM_TRANSFORMS_IPO_WHOLEPROGRAMDEVIRT_H

‎llvm/lib/LTO/LTO.cpp

+15-1
Original file line numberDiff line numberDiff line change
@@ -44,6 +44,7 @@
4444
#include "llvm/Target/TargetOptions.h"
4545
#include "llvm/Transforms/IPO.h"
4646
#include "llvm/Transforms/IPO/PassManagerBuilder.h"
47+
#include "llvm/Transforms/IPO/WholeProgramDevirt.h"
4748
#include "llvm/Transforms/Utils/FunctionImportUtils.h"
4849
#include "llvm/Transforms/Utils/SplitModule.h"
4950

@@ -1274,15 +1275,28 @@ Error LTO::runThinLTO(AddStreamFn AddStream, NativeObjectCache Cache,
12741275
if (DumpThinCGSCCs)
12751276
ThinLTO.CombinedIndex.dumpSCCs(outs());
12761277

1278+
std::set<GlobalValue::GUID> ExportedGUIDs;
1279+
1280+
// Perform index-based WPD. This will return immediately if there are
1281+
// no index entries in the typeIdMetadata map (e.g. if we are instead
1282+
// performing IR-based WPD in hybrid regular/thin LTO mode).
1283+
std::map<ValueInfo, std::vector<VTableSlotSummary>> LocalWPDTargetsMap;
1284+
runWholeProgramDevirtOnIndex(ThinLTO.CombinedIndex, ExportedGUIDs,
1285+
LocalWPDTargetsMap);
1286+
12771287
if (Conf.OptLevel > 0)
12781288
ComputeCrossModuleImport(ThinLTO.CombinedIndex, ModuleToDefinedGVSummaries,
12791289
ImportLists, ExportLists);
12801290

1291+
// Update local devirtualized targets that were exported by cross-module
1292+
// importing
1293+
updateIndexWPDForExports(ThinLTO.CombinedIndex, ExportLists,
1294+
LocalWPDTargetsMap);
1295+
12811296
// Figure out which symbols need to be internalized. This also needs to happen
12821297
// at -O0 because summary-based DCE is implemented using internalization, and
12831298
// we must apply DCE consistently with the full LTO module in order to avoid
12841299
// undefined references during the final link.
1285-
std::set<GlobalValue::GUID> ExportedGUIDs;
12861300
for (auto &Res : GlobalResolutions) {
12871301
// If the symbol does not have external references or it is not prevailing,
12881302
// then not need to mark it as exported from a ThinLTO partition.

‎llvm/lib/Transforms/IPO/WholeProgramDevirt.cpp

+298-5
Large diffs are not rendered by default.

‎llvm/test/ThinLTO/X86/Inputs/devirt2.ll

+59
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,59 @@
1+
target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128"
2+
target triple = "x86_64-grtev4-linux-gnu"
3+
4+
%struct.A = type { i32 (...)** }
5+
%struct.B = type { %struct.A }
6+
%struct.C = type { %struct.A }
7+
%struct.D = type { i32 (...)** }
8+
%struct.E = type { i32 (...)** }
9+
10+
@_ZTV1B = constant { [4 x i8*] } { [4 x i8*] [i8* null, i8* undef, i8* bitcast (i32 (%struct.B*, i32)* @_ZN1B1fEi to i8*), i8* bitcast (i32 (%struct.A*, i32)* @_ZN1A1nEi to i8*)] }, !type !0, !type !1
11+
@_ZTV1C = constant { [4 x i8*] } { [4 x i8*] [i8* null, i8* undef, i8* bitcast (i32 (%struct.C*, i32)* @_ZN1C1fEi to i8*), i8* bitcast (i32 (%struct.A*, i32)* @_ZN1A1nEi to i8*)] }, !type !0, !type !2
12+
@_ZTV1D = linkonce_odr constant { [3 x i8*] } { [3 x i8*] [i8* null, i8* undef, i8* bitcast (i32 (%struct.D*, i32)* @_ZN1D1mEi to i8*)] }, !type !3
13+
@_ZTV1E = constant { [3 x i8*] } { [3 x i8*] [i8* null, i8* undef, i8* bitcast (i32 (%struct.E*, i32)* @_ZN1E1mEi to i8*)] }, !type !4
14+
15+
define i32 @_ZN1B1fEi(%struct.B* %this, i32 %a) #0 {
16+
ret i32 0;
17+
}
18+
19+
define internal i32 @_ZN1A1nEi(%struct.A* %this, i32 %a) #0 {
20+
ret i32 0;
21+
}
22+
23+
define i32 @_ZN1C1fEi(%struct.C* %this, i32 %a) #0 {
24+
ret i32 0;
25+
}
26+
27+
define linkonce_odr i32 @_ZN1D1mEi(%struct.D* %this, i32 %a) #0 {
28+
ret i32 0;
29+
}
30+
31+
define internal i32 @_ZN1E1mEi(%struct.E* %this, i32 %a) #0 {
32+
ret i32 0;
33+
}
34+
35+
define i32 @test2(%struct.E* %obj, i32 %a) {
36+
entry:
37+
%0 = bitcast %struct.E* %obj to i8***
38+
%vtable2 = load i8**, i8*** %0
39+
%1 = bitcast i8** %vtable2 to i8*
40+
%p2 = call i1 @llvm.type.test(i8* %1, metadata !"_ZTS1E")
41+
call void @llvm.assume(i1 %p2)
42+
43+
%2 = bitcast i8** %vtable2 to i32 (%struct.E*, i32)**
44+
%fptr33 = load i32 (%struct.E*, i32)*, i32 (%struct.E*, i32)** %2, align 8
45+
46+
%call4 = tail call i32 %fptr33(%struct.E* nonnull %obj, i32 %a)
47+
ret i32 %call4
48+
}
49+
50+
attributes #0 = { noinline optnone }
51+
52+
declare i1 @llvm.type.test(i8*, metadata)
53+
declare void @llvm.assume(i1)
54+
55+
!0 = !{i64 16, !"_ZTS1A"}
56+
!1 = !{i64 16, !"_ZTS1B"}
57+
!2 = !{i64 16, !"_ZTS1C"}
58+
!3 = !{i64 16, !"_ZTS1D"}
59+
!4 = !{i64 16, !"_ZTS1E"}

‎llvm/test/ThinLTO/X86/devirt.ll

+47-9
Original file line numberDiff line numberDiff line change
@@ -20,10 +20,10 @@
2020
; and that we generate summary information needed for index-based WPD.
2121
; RUN: llvm-dis -o - %t2.o | FileCheck %s --check-prefix=NOENABLESPLITFLAG
2222
; NOENABLESPLITFLAG-DAG: !{i32 1, !"EnableSplitLTOUnit", i32 0}
23-
; NOENABLESPLITFLAG-DAG: [[An:\^[0-9]+]] = gv: (name: "_ZN1A1nEi")
24-
; NOENABLESPLITFLAG-DAG: [[Bf:\^[0-9]+]] = gv: (name: "_ZN1B1fEi")
25-
; NOENABLESPLITFLAG-DAG: [[Cf:\^[0-9]+]] = gv: (name: "_ZN1C1fEi")
26-
; NOENABLESPLITFLAG-DAG: [[Dm:\^[0-9]+]] = gv: (name: "_ZN1D1mEi")
23+
; NOENABLESPLITFLAG-DAG: [[An:\^[0-9]+]] = gv: (name: "_ZN1A1nEi"
24+
; NOENABLESPLITFLAG-DAG: [[Bf:\^[0-9]+]] = gv: (name: "_ZN1B1fEi"
25+
; NOENABLESPLITFLAG-DAG: [[Cf:\^[0-9]+]] = gv: (name: "_ZN1C1fEi"
26+
; NOENABLESPLITFLAG-DAG: [[Dm:\^[0-9]+]] = gv: (name: "_ZN1D1mEi"
2727
; NOENABLESPLITFLAG-DAG: [[B:\^[0-9]+]] = gv: (name: "_ZTV1B", {{.*}} vTableFuncs: ((virtFunc: [[Bf]], offset: 16), (virtFunc: [[An]], offset: 24)), refs: ([[Bf]], [[An]])
2828
; NOENABLESPLITFLAG-DAG: [[C:\^[0-9]+]] = gv: (name: "_ZTV1C", {{.*}} vTableFuncs: ((virtFunc: [[Cf]], offset: 16), (virtFunc: [[An]], offset: 24)), refs: ([[An]], [[Cf]])
2929
; NOENABLESPLITFLAG-DAG: [[D:\^[0-9]+]] = gv: (name: "_ZTV1D", {{.*}} vTableFuncs: ((virtFunc: [[Dm]], offset: 16)), refs: ([[Dm]])
@@ -33,7 +33,31 @@
3333
; Type Id on _ZTV1D should have been promoted
3434
; NOENABLESPLITFLAG-DAG: typeidCompatibleVTable: (name: "1${{.*}}", summary: ((offset: 16, [[D]])))
3535

36-
; TODO: Test index-based WPD one %t2.o once implemented.
36+
; Legacy PM, Index based WPD
37+
; RUN: llvm-lto2 run %t2.o -save-temps -pass-remarks=. \
38+
; RUN: -o %t3 \
39+
; RUN: -r=%t2.o,test,px \
40+
; RUN: -r=%t2.o,_ZN1A1nEi,p \
41+
; RUN: -r=%t2.o,_ZN1B1fEi,p \
42+
; RUN: -r=%t2.o,_ZN1C1fEi,p \
43+
; RUN: -r=%t2.o,_ZN1D1mEi,p \
44+
; RUN: -r=%t2.o,_ZTV1B,px \
45+
; RUN: -r=%t2.o,_ZTV1C,px \
46+
; RUN: -r=%t2.o,_ZTV1D,px 2>&1 | FileCheck %s --check-prefix=REMARK
47+
; RUN: llvm-dis %t3.1.4.opt.bc -o - | FileCheck %s --check-prefix=CHECK-IR
48+
49+
; New PM, Index based WPD
50+
; RUN: llvm-lto2 run %t2.o -save-temps -use-new-pm -pass-remarks=. \
51+
; RUN: -o %t3 \
52+
; RUN: -r=%t2.o,test,px \
53+
; RUN: -r=%t2.o,_ZN1A1nEi,p \
54+
; RUN: -r=%t2.o,_ZN1B1fEi,p \
55+
; RUN: -r=%t2.o,_ZN1C1fEi,p \
56+
; RUN: -r=%t2.o,_ZN1D1mEi,p \
57+
; RUN: -r=%t2.o,_ZTV1B,px \
58+
; RUN: -r=%t2.o,_ZTV1C,px \
59+
; RUN: -r=%t2.o,_ZTV1D,px 2>&1 | FileCheck %s --check-prefix=REMARK
60+
; RUN: llvm-dis %t3.1.4.opt.bc -o - | FileCheck %s --check-prefix=CHECK-IR
3761

3862
; Legacy PM
3963
; FIXME: Fix machine verifier issues and remove -verify-machineinstrs=0. PR39436.
@@ -138,10 +162,24 @@ entry:
138162
declare i1 @llvm.type.test(i8*, metadata)
139163
declare void @llvm.assume(i1)
140164

141-
declare i32 @_ZN1B1fEi(%struct.B* %this, i32 %a)
142-
declare i32 @_ZN1A1nEi(%struct.A* %this, i32 %a)
143-
declare i32 @_ZN1C1fEi(%struct.C* %this, i32 %a)
144-
declare i32 @_ZN1D1mEi(%struct.D* %this, i32 %a)
165+
define i32 @_ZN1B1fEi(%struct.B* %this, i32 %a) #0 {
166+
ret i32 0;
167+
}
168+
169+
define i32 @_ZN1A1nEi(%struct.A* %this, i32 %a) #0 {
170+
ret i32 0;
171+
}
172+
173+
define i32 @_ZN1C1fEi(%struct.C* %this, i32 %a) #0 {
174+
ret i32 0;
175+
}
176+
177+
define i32 @_ZN1D1mEi(%struct.D* %this, i32 %a) #0 {
178+
ret i32 0;
179+
}
180+
181+
; Make sure we don't inline or otherwise optimize out the direct calls.
182+
attributes #0 = { noinline optnone }
145183

146184
!0 = !{i64 16, !"_ZTS1A"}
147185
!1 = !{i64 16, !"_ZTS1B"}

‎llvm/test/ThinLTO/X86/devirt2.ll

+278
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,278 @@
1+
; REQUIRES: x86-registered-target
2+
3+
; Test devirtualization requiring promotion of local targets.
4+
5+
; Generate split module with summary for hybrid Thin/Regular LTO WPD.
6+
; RUN: opt -thinlto-bc -thinlto-split-lto-unit -o %t1.o %s
7+
; RUN: opt -thinlto-bc -thinlto-split-lto-unit -o %t2.o %p/Inputs/devirt3.ll
8+
9+
; Check that we have module flag showing splitting enabled, and that we don't
10+
; generate summary information needed for index-based WPD.
11+
; RUN: llvm-modextract -b -n=0 %t2.o -o %t2.o.0
12+
; RUN: llvm-dis -o - %t2.o.0 | FileCheck %s --check-prefix=ENABLESPLITFLAG --implicit-check-not=vTableFuncs --implicit-check-not=typeidCompatibleVTable
13+
; RUN: llvm-modextract -b -n=1 %t2.o -o %t2.o.1
14+
; RUN: llvm-dis -o - %t2.o.1 | FileCheck %s --check-prefix=ENABLESPLITFLAG --implicit-check-not=vTableFuncs --implicit-check-not=typeidCompatibleVTable
15+
; ENABLESPLITFLAG: !{i32 1, !"EnableSplitLTOUnit", i32 1}
16+
17+
; Generate unsplit module with summary for ThinLTO index-based WPD.
18+
; RUN: opt -thinlto-bc -o %t3.o %s
19+
; RUN: opt -thinlto-bc -o %t4.o %p/Inputs/devirt3.ll
20+
21+
; Check that we don't have module flag when splitting not enabled for ThinLTO,
22+
; and that we generate summary information needed for index-based WPD.
23+
; RUN: llvm-dis -o - %t4.o | FileCheck %s --check-prefix=NOENABLESPLITFLAG
24+
; NOENABLESPLITFLAG-DAG: !{i32 1, !"EnableSplitLTOUnit", i32 0}
25+
; NOENABLESPLITFLAG-DAG: [[An:\^[0-9]+]] = gv: (name: "_ZN1A1nEi"
26+
; NOENABLESPLITFLAG-DAG: [[Bf:\^[0-9]+]] = gv: (name: "_ZN1B1fEi"
27+
; NOENABLESPLITFLAG-DAG: [[Cf:\^[0-9]+]] = gv: (name: "_ZN1C1fEi"
28+
; NOENABLESPLITFLAG-DAG: [[Dm:\^[0-9]+]] = gv: (name: "_ZN1D1mEi"
29+
; NOENABLESPLITFLAG-DAG: [[B:\^[0-9]+]] = gv: (name: "_ZTV1B", {{.*}} vTableFuncs: ((virtFunc: [[Bf]], offset: 16), (virtFunc: [[An]], offset: 24)), refs: ([[Bf]], [[An]])
30+
; NOENABLESPLITFLAG-DAG: [[C:\^[0-9]+]] = gv: (name: "_ZTV1C", {{.*}} vTableFuncs: ((virtFunc: [[Cf]], offset: 16), (virtFunc: [[An]], offset: 24)), refs: ([[An]], [[Cf]])
31+
; NOENABLESPLITFLAG-DAG: [[D:\^[0-9]+]] = gv: (name: "_ZTV1D", {{.*}} vTableFuncs: ((virtFunc: [[Dm]], offset: 16)), refs: ([[Dm]])
32+
; NOENABLESPLITFLAG-DAG: typeidCompatibleVTable: (name: "_ZTS1A", summary: ((offset: 16, [[B]]), (offset: 16, [[C]])))
33+
; NOENABLESPLITFLAG-DAG: typeidCompatibleVTable: (name: "_ZTS1B", summary: ((offset: 16, [[B]])))
34+
; NOENABLESPLITFLAG-DAG: typeidCompatibleVTable: (name: "_ZTS1C", summary: ((offset: 16, [[C]])))
35+
; NOENABLESPLITFLAG-DAG: typeidCompatibleVTable: (name: "_ZTS1D", summary: ((offset: 16, [[D]])))
36+
37+
; Legacy PM, Index based WPD
38+
; RUN: llvm-lto2 run %t3.o %t4.o -save-temps -pass-remarks=. \
39+
; RUN: -wholeprogramdevirt-print-index-based \
40+
; RUN: -o %t5 \
41+
; RUN: -r=%t3.o,test,px \
42+
; RUN: -r=%t3.o,_ZTV1B, \
43+
; RUN: -r=%t3.o,_ZTV1C, \
44+
; RUN: -r=%t3.o,_ZTV1D, \
45+
; RUN: -r=%t3.o,_ZN1D1mEi, \
46+
; RUN: -r=%t3.o,test2, \
47+
; RUN: -r=%t4.o,_ZN1B1fEi,p \
48+
; RUN: -r=%t4.o,_ZN1C1fEi,p \
49+
; RUN: -r=%t4.o,_ZN1D1mEi,p \
50+
; RUN: -r=%t4.o,test2,px \
51+
; RUN: -r=%t4.o,_ZTV1B,px \
52+
; RUN: -r=%t4.o,_ZTV1C,px \
53+
; RUN: -r=%t4.o,_ZTV1D,px \
54+
; RUN: -r=%t4.o,_ZTV1E,px 2>&1 | FileCheck %s --check-prefix=REMARK --check-prefix=PRINT
55+
; RUN: llvm-dis %t5.1.4.opt.bc -o - | FileCheck %s --check-prefix=CHECK-IR1
56+
; RUN: llvm-dis %t5.2.4.opt.bc -o - | FileCheck %s --check-prefix=CHECK-IR2
57+
; RUN: nm %t5.1 | FileCheck %s --check-prefix=NM-INDEX1
58+
; RUN: nm %t5.2 | FileCheck %s --check-prefix=NM-INDEX2
59+
60+
; New PM, Index based WPD
61+
; RUN: llvm-lto2 run %t3.o %t4.o -save-temps -use-new-pm -pass-remarks=. \
62+
; RUN: -wholeprogramdevirt-print-index-based \
63+
; RUN: -o %t5 \
64+
; RUN: -r=%t3.o,test,px \
65+
; RUN: -r=%t3.o,_ZTV1B, \
66+
; RUN: -r=%t3.o,_ZTV1C, \
67+
; RUN: -r=%t3.o,_ZTV1D, \
68+
; RUN: -r=%t3.o,_ZN1D1mEi, \
69+
; RUN: -r=%t3.o,test2, \
70+
; RUN: -r=%t4.o,_ZN1B1fEi,p \
71+
; RUN: -r=%t4.o,_ZN1C1fEi,p \
72+
; RUN: -r=%t4.o,_ZN1D1mEi,p \
73+
; RUN: -r=%t4.o,test2,px \
74+
; RUN: -r=%t4.o,_ZTV1B,px \
75+
; RUN: -r=%t4.o,_ZTV1C,px \
76+
; RUN: -r=%t4.o,_ZTV1D,px \
77+
; RUN: -r=%t4.o,_ZTV1E,px 2>&1 | FileCheck %s --check-prefix=REMARK --check-prefix=PRINT
78+
; RUN: llvm-dis %t5.1.4.opt.bc -o - | FileCheck %s --check-prefix=CHECK-IR1
79+
; RUN: llvm-dis %t5.2.4.opt.bc -o - | FileCheck %s --check-prefix=CHECK-IR2
80+
; RUN: nm %t5.1 | FileCheck %s --check-prefix=NM-INDEX1
81+
; RUN: nm %t5.2 | FileCheck %s --check-prefix=NM-INDEX2
82+
83+
; NM-INDEX1-DAG: U _ZN1A1nEi.llvm.
84+
; NM-INDEX1-DAG: U _ZN1E1mEi.llvm.
85+
; NM-INDEX1-DAG: U _ZN1D1mEi
86+
87+
; NM-INDEX2-DAG: T _ZN1A1nEi.llvm.
88+
; NM-INDEX2-DAG: T _ZN1E1mEi.llvm.
89+
; NM-INDEX2-DAG: W _ZN1D1mEi
90+
; NM-INDEX2-DAG: t _ZN1B1fEi
91+
; NM-INDEX2-DAG: t _ZN1C1fEi
92+
93+
; Index based WPD, distributed backends
94+
; RUN: llvm-lto2 run %t3.o %t4.o -save-temps -use-new-pm \
95+
; RUN: -thinlto-distributed-indexes -wholeprogramdevirt-print-index-based \
96+
; RUN: -o %t5 \
97+
; RUN: -r=%t3.o,test,px \
98+
; RUN: -r=%t3.o,_ZTV1B, \
99+
; RUN: -r=%t3.o,_ZTV1C, \
100+
; RUN: -r=%t3.o,_ZTV1D, \
101+
; RUN: -r=%t3.o,_ZN1D1mEi, \
102+
; RUN: -r=%t3.o,test2, \
103+
; RUN: -r=%t4.o,_ZN1B1fEi,p \
104+
; RUN: -r=%t4.o,_ZN1C1fEi,p \
105+
; RUN: -r=%t4.o,_ZN1D1mEi,p \
106+
; RUN: -r=%t4.o,test2,px \
107+
; RUN: -r=%t4.o,_ZTV1B,px \
108+
; RUN: -r=%t4.o,_ZTV1C,px \
109+
; RUN: -r=%t4.o,_ZTV1D,px \
110+
; RUN: -r=%t4.o,_ZTV1E,px 2>&1 | FileCheck %s --check-prefix=PRINT
111+
112+
; PRINT-DAG: Devirtualized call to {{.*}} (_ZN1A1nEi)
113+
; PRINT-DAG: Devirtualized call to {{.*}} (_ZN1E1mEi)
114+
; PRINT-DAG: Devirtualized call to {{.*}} (_ZN1D1mEi)
115+
116+
; Legacy PM
117+
; RUN: llvm-lto2 run %t1.o %t2.o -save-temps -pass-remarks=. \
118+
; RUN: -o %t5 \
119+
; RUN: -r=%t1.o,test,px \
120+
; RUN: -r=%t1.o,_ZTV1B, \
121+
; RUN: -r=%t1.o,_ZTV1C, \
122+
; RUN: -r=%t1.o,_ZTV1D, \
123+
; RUN: -r=%t1.o,_ZTV1D, \
124+
; RUN: -r=%t1.o,_ZN1D1mEi, \
125+
; RUN: -r=%t1.o,_ZN1D1mEi, \
126+
; RUN: -r=%t1.o,test2, \
127+
; RUN: -r=%t2.o,_ZN1A1nEi,p \
128+
; RUN: -r=%t2.o,_ZN1B1fEi,p \
129+
; RUN: -r=%t2.o,_ZN1C1fEi,p \
130+
; RUN: -r=%t2.o,_ZN1D1mEi,p \
131+
; RUN: -r=%t2.o,_ZN1E1mEi,p \
132+
; RUN: -r=%t2.o,_ZTV1B, \
133+
; RUN: -r=%t2.o,_ZTV1C, \
134+
; RUN: -r=%t2.o,_ZTV1D, \
135+
; RUN: -r=%t2.o,_ZTV1E, \
136+
; RUN: -r=%t2.o,test2,px \
137+
; RUN: -r=%t2.o,_ZN1A1nEi, \
138+
; RUN: -r=%t2.o,_ZN1B1fEi, \
139+
; RUN: -r=%t2.o,_ZN1C1fEi, \
140+
; RUN: -r=%t2.o,_ZN1D1mEi, \
141+
; RUN: -r=%t2.o,_ZN1E1mEi, \
142+
; RUN: -r=%t2.o,_ZTV1B,px \
143+
; RUN: -r=%t2.o,_ZTV1C,px \
144+
; RUN: -r=%t2.o,_ZTV1D,px \
145+
; RUN: -r=%t2.o,_ZTV1E,px 2>&1 | FileCheck %s --check-prefix=REMARK
146+
; RUN: llvm-dis %t5.1.4.opt.bc -o - | FileCheck %s --check-prefix=CHECK-IR1
147+
; RUN: llvm-dis %t5.2.4.opt.bc -o - | FileCheck %s --check-prefix=CHECK-IR2
148+
; RUN: nm %t5.1 | FileCheck %s --check-prefix=NM-HYBRID1
149+
; RUN: nm %t5.2 | FileCheck %s --check-prefix=NM-HYBRID2
150+
151+
; New PM
152+
; RUN: llvm-lto2 run %t1.o %t2.o -save-temps -use-new-pm -pass-remarks=. \
153+
; RUN: -o %t5 \
154+
; RUN: -r=%t1.o,test,px \
155+
; RUN: -r=%t1.o,_ZTV1B, \
156+
; RUN: -r=%t1.o,_ZTV1C, \
157+
; RUN: -r=%t1.o,_ZTV1D, \
158+
; RUN: -r=%t1.o,_ZTV1D, \
159+
; RUN: -r=%t1.o,_ZN1D1mEi, \
160+
; RUN: -r=%t1.o,_ZN1D1mEi, \
161+
; RUN: -r=%t1.o,test2, \
162+
; RUN: -r=%t2.o,_ZN1A1nEi,p \
163+
; RUN: -r=%t2.o,_ZN1B1fEi,p \
164+
; RUN: -r=%t2.o,_ZN1C1fEi,p \
165+
; RUN: -r=%t2.o,_ZN1D1mEi,p \
166+
; RUN: -r=%t2.o,_ZN1E1mEi,p \
167+
; RUN: -r=%t2.o,_ZTV1B, \
168+
; RUN: -r=%t2.o,_ZTV1C, \
169+
; RUN: -r=%t2.o,_ZTV1D, \
170+
; RUN: -r=%t2.o,_ZTV1E, \
171+
; RUN: -r=%t2.o,test2,px \
172+
; RUN: -r=%t2.o,_ZN1A1nEi, \
173+
; RUN: -r=%t2.o,_ZN1B1fEi, \
174+
; RUN: -r=%t2.o,_ZN1C1fEi, \
175+
; RUN: -r=%t2.o,_ZN1D1mEi, \
176+
; RUN: -r=%t2.o,_ZN1E1mEi, \
177+
; RUN: -r=%t2.o,_ZTV1B,px \
178+
; RUN: -r=%t2.o,_ZTV1C,px \
179+
; RUN: -r=%t2.o,_ZTV1D,px \
180+
; RUN: -r=%t2.o,_ZTV1E,px 2>&1 | FileCheck %s --check-prefix=REMARK
181+
; RUN: llvm-dis %t5.1.4.opt.bc -o - | FileCheck %s --check-prefix=CHECK-IR1
182+
; RUN: llvm-dis %t5.2.4.opt.bc -o - | FileCheck %s --check-prefix=CHECK-IR2
183+
; RUN: nm %t5.1 | FileCheck %s --check-prefix=NM-HYBRID1
184+
; RUN: nm %t5.2 | FileCheck %s --check-prefix=NM-HYBRID2
185+
186+
; NM-HYBRID1-DAG: U _ZN1A1nEi$
187+
; NM-HYBRID1-DAG: U _ZN1E1mEi$
188+
; NM-HYBRID1-DAG: U _ZN1D1mEi
189+
190+
; NM-HYBRID2-DAG: T _ZN1A1nEi$
191+
; NM-HYBRID2-DAG: T _ZN1E1mEi$
192+
; NM-HYBRID2-DAG: W _ZN1D1mEi
193+
; NM-HYBRID2-DAG: T _ZN1B1fEi
194+
; NM-HYBRID2-DAG: T _ZN1C1fEi
195+
196+
; REMARK-DAG: single-impl: devirtualized a call to _ZN1A1nEi
197+
; REMARK-DAG: single-impl: devirtualized a call to _ZN1D1mEi
198+
; We should devirt call to _ZN1E1mEi once in importing module and once
199+
; in original (exporting) module.
200+
; REMARK-DAG: single-impl: devirtualized a call to _ZN1E1mEi
201+
; REMARK-DAG: single-impl: devirtualized a call to _ZN1E1mEi
202+
203+
target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128"
204+
target triple = "x86_64-grtev4-linux-gnu"
205+
206+
%struct.A = type { i32 (...)** }
207+
%struct.B = type { %struct.A }
208+
%struct.C = type { %struct.A }
209+
%struct.D = type { i32 (...)** }
210+
%struct.E = type { i32 (...)** }
211+
212+
@_ZTV1B = external constant [4 x i8*]
213+
@_ZTV1C = external constant [4 x i8*]
214+
;@_ZTV1D = external constant [3 x i8*]
215+
@_ZTV1D = linkonce_odr constant { [3 x i8*] } { [3 x i8*] [i8* null, i8* undef, i8* bitcast (i32 (%struct.D*, i32)* @_ZN1D1mEi to i8*)] }, !type !3
216+
217+
define linkonce_odr i32 @_ZN1D1mEi(%struct.D* %this, i32 %a) #0 {
218+
ret i32 0
219+
}
220+
221+
; CHECK-IR1-LABEL: define i32 @test
222+
define i32 @test(%struct.A* %obj, %struct.D* %obj2, %struct.E* %obj3, i32 %a) {
223+
entry:
224+
%0 = bitcast %struct.A* %obj to i8***
225+
%vtable = load i8**, i8*** %0
226+
%1 = bitcast i8** %vtable to i8*
227+
%p = call i1 @llvm.type.test(i8* %1, metadata !"_ZTS1A")
228+
call void @llvm.assume(i1 %p)
229+
%fptrptr = getelementptr i8*, i8** %vtable, i32 1
230+
%2 = bitcast i8** %fptrptr to i32 (%struct.A*, i32)**
231+
%fptr1 = load i32 (%struct.A*, i32)*, i32 (%struct.A*, i32)** %2, align 8
232+
233+
; Check that the call was devirtualized. Ignore extra character before
234+
; symbol name which would happen if it was promoted during module
235+
; splitting for hybrid WPD.
236+
; CHECK-IR1: %call = tail call i32 bitcast (void ()* @{{.*}}_ZN1A1nEi
237+
%call = tail call i32 %fptr1(%struct.A* nonnull %obj, i32 %a)
238+
239+
%3 = bitcast i8** %vtable to i32 (%struct.A*, i32)**
240+
%fptr22 = load i32 (%struct.A*, i32)*, i32 (%struct.A*, i32)** %3, align 8
241+
242+
; We still have to call it as virtual.
243+
; CHECK-IR1: %call3 = tail call i32 %fptr22
244+
%call3 = tail call i32 %fptr22(%struct.A* nonnull %obj, i32 %call)
245+
246+
%4 = bitcast %struct.D* %obj2 to i8***
247+
%vtable2 = load i8**, i8*** %4
248+
%5 = bitcast i8** %vtable2 to i8*
249+
%p2 = call i1 @llvm.type.test(i8* %5, metadata !"_ZTS1D")
250+
call void @llvm.assume(i1 %p2)
251+
252+
%6 = bitcast i8** %vtable2 to i32 (%struct.D*, i32)**
253+
%fptr33 = load i32 (%struct.D*, i32)*, i32 (%struct.D*, i32)** %6, align 8
254+
255+
; Check that the call was devirtualized.
256+
; CHECK-IR1: %call4 = tail call i32 @_ZN1D1mEi
257+
%call4 = tail call i32 %fptr33(%struct.D* nonnull %obj2, i32 %call3)
258+
259+
%call5 = tail call i32 @test2(%struct.E* nonnull %obj3, i32 %call4)
260+
ret i32 %call5
261+
}
262+
; CHECK-IR1-LABEL: ret i32
263+
; CHECK-IR1-LABEL: }
264+
265+
; CHECK-IR2: define i32 @test2
266+
; CHECK-IR2-NEXT: entry:
267+
; Check that the call was devirtualized. Ignore extra character before
268+
; symbol name which would happen if it was promoted during module
269+
; splitting for hybrid WPD.
270+
; CHECK-IR2-NEXT: %call4 = tail call i32 @{{.*}}_ZN1E1mEi
271+
272+
declare i1 @llvm.type.test(i8*, metadata)
273+
declare void @llvm.assume(i1)
274+
declare i32 @test2(%struct.E* %obj, i32 %a)
275+
276+
attributes #0 = { noinline optnone }
277+
278+
!3 = !{i64 16, !"_ZTS1D"}

‎llvm/test/ThinLTO/X86/nodevirt-nonpromoted-typeid.ll

+66
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,66 @@
1+
; REQUIRES: x86-registered-target
2+
3+
; Test that index-only devirtualization handles and ignores any
4+
; type metadata that could not be summarized (because it was internal
5+
; and could not be promoted due to the fact that the module has
6+
; no external symbols and therefore could not be assigned a unique
7+
; identifier). In this case we should simply not get the type
8+
; metadata summary entries, and no promotion will occur.
9+
10+
; Generate unsplit module with summary for ThinLTO index-based WPD.
11+
; RUN: opt -thinlto-bc -thinlto-split-lto-unit=false -o %t2.o %s
12+
13+
; Check that we don't have module flag when splitting not enabled for ThinLTO,
14+
; and that we generate summary information needed for index-based WPD.
15+
; RUN: llvm-dis -o - %t2.o | FileCheck %s --check-prefix=DIS
16+
; DIS-NOT: typeIdInfo
17+
; DIS-NOT: typeidMetadata
18+
19+
; Legacy PM, Index based WPD
20+
; RUN: llvm-lto2 run %t2.o -save-temps -pass-remarks=. \
21+
; RUN: -o %t3 \
22+
; RUN: -r=%t2.o,test,plx \
23+
; RUN: -r=%t2.o,_ZN1D1mEi,
24+
; RUN: llvm-dis %t3.1.4.opt.bc -o - | FileCheck %s --check-prefix=CHECK-IR
25+
26+
; New PM, Index based WPD
27+
; RUN: llvm-lto2 run %t2.o -save-temps -use-new-pm -pass-remarks=. \
28+
; RUN: -o %t3 \
29+
; RUN: -r=%t2.o,test,plx \
30+
; RUN: -r=%t2.o,_ZN1D1mEi,
31+
; RUN: llvm-dis %t3.1.4.opt.bc -o - | FileCheck %s --check-prefix=CHECK-IR
32+
33+
target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128"
34+
target triple = "x86_64-grtev4-linux-gnu"
35+
36+
%struct.D = type { i32 (...)** }
37+
38+
@_ZTV1D = internal constant { [3 x i8*] } { [3 x i8*] [i8* null, i8* undef, i8* bitcast (i32 (%struct.D*, i32)* @_ZN1D1mEi to i8*)] }, !type !3
39+
40+
; CHECK-IR-LABEL: define weak_odr dso_local i32 @test
41+
define weak_odr i32 @test(%struct.D* %obj2, i32 %a) {
42+
entry:
43+
%0 = bitcast %struct.D* %obj2 to i8***
44+
%vtable2 = load i8**, i8*** %0
45+
%1 = bitcast i8** %vtable2 to i8*
46+
%p2 = call i1 @llvm.type.test(i8* %1, metadata !4)
47+
call void @llvm.assume(i1 %p2)
48+
49+
%2 = bitcast i8** %vtable2 to i32 (%struct.D*, i32)**
50+
%fptr33 = load i32 (%struct.D*, i32)*, i32 (%struct.D*, i32)** %2, align 8
51+
52+
; Check that the call was not devirtualized.
53+
; CHECK-IR: %call4 = tail call i32 %fptr33
54+
%call4 = tail call i32 %fptr33(%struct.D* nonnull %obj2, i32 0)
55+
ret i32 %call4
56+
}
57+
; CHECK-IR-LABEL: ret i32
58+
; CHECK-IR-LABEL: }
59+
60+
declare i1 @llvm.type.test(i8*, metadata)
61+
declare void @llvm.assume(i1)
62+
63+
declare i32 @_ZN1D1mEi(%struct.D* %this, i32 %a)
64+
65+
!3 = !{i64 16, !4}
66+
!4 = distinct !{}

‎llvm/tools/llvm-lto2/llvm-lto2.cpp

+8
Original file line numberDiff line numberDiff line change
@@ -291,6 +291,14 @@ static int run(int argc, char **argv) {
291291
std::vector<SymbolResolution> Res;
292292
for (const InputFile::Symbol &Sym : Input->symbols()) {
293293
auto I = CommandLineResolutions.find({F, Sym.getName()});
294+
// If it isn't found, look for "$", which would have been added
295+
// (followed by a hash) when the symbol was promoted during module
296+
// splitting if it was defined in one part and used in the other.
297+
// Try looking up the symbol name before the "$".
298+
if (I == CommandLineResolutions.end()) {
299+
auto SplitName = Sym.getName().rsplit("$");
300+
I = CommandLineResolutions.find({F, SplitName.first});
301+
}
294302
if (I == CommandLineResolutions.end()) {
295303
llvm::errs() << argv[0] << ": missing symbol resolution for " << F
296304
<< ',' << Sym.getName() << '\n';

0 commit comments

Comments
 (0)
Please sign in to comment.