Skip to content

Commit 51e459a

Browse files
nikicnikic
committed
Revert "[ConstantFold] Remove non-trivial gep-of-gep fold (#93823)"
This reverts commit e1cc9e4. This causes some non-trivial text size increases in unoptimized builds for Bullet. Revert while I investigate.
1 parent b86a9c5 commit 51e459a

9 files changed

+120
-71
lines changed

‎clang/test/CodeGenCXX/2011-12-19-init-list-ctor.cpp

Lines changed: 2 additions & 2 deletions
Original file line numberDiff line numberDiff line change
@@ -22,5 +22,5 @@ struct S {
2222
// CHECK: call void @_ZN1AC1EPKc(ptr {{[^,]*}} getelementptr inbounds (%struct.S, ptr @arr, i32 0, i32 1), ptr noundef @.str)
2323
// CHECK: store i32 1, ptr getelementptr inbounds (%struct.S, ptr @arr, i64 1)
2424
// CHECK: call void @_ZN1AC1EPKc(ptr {{[^,]*}} getelementptr inbounds (%struct.S, ptr @arr, i64 1, i32 1), ptr noundef @.str.1)
25-
// CHECK: store i32 2, ptr getelementptr inbounds (%struct.S, ptr getelementptr inbounds (%struct.S, ptr @arr, i64 1), i64 1)
26-
// CHECK: call void @_ZN1AC1EPKc(ptr {{[^,]*}} getelementptr inbounds (%struct.S, ptr getelementptr inbounds (%struct.S, ptr @arr, i64 1), i64 1, i32 1), ptr noundef @.str.2)
25+
// CHECK: store i32 2, ptr getelementptr inbounds (%struct.S, ptr @arr, i64 2)
26+
// CHECK: call void @_ZN1AC1EPKc(ptr {{[^,]*}} getelementptr inbounds (%struct.S, ptr @arr, i64 2, i32 1), ptr noundef @.str.2)

‎clang/test/CodeGenCXX/cxx0x-initializer-stdinitializerlist-pr12086.cpp

Lines changed: 4 additions & 4 deletions
Original file line numberDiff line numberDiff line change
@@ -83,8 +83,8 @@ std::initializer_list<std::initializer_list<int>> nested = {
8383
// CHECK-DYNAMIC-BL: store i32 5, ptr @_ZGR6nested2_
8484
// CHECK-DYNAMIC-BL: store i32 {{.*}}, ptr getelementptr inbounds (i32, ptr @_ZGR6nested2_, i64 1)
8585
// CHECK-DYNAMIC-BL: store ptr @_ZGR6nested2_,
86-
// CHECK-DYNAMIC-BL: ptr getelementptr inbounds ({{.*}}, ptr getelementptr inbounds ({{.*}}, ptr @_ZGR6nested_, i64 1), i64 1), align 8
87-
// CHECK-DYNAMIC-BL: store i64 2, ptr getelementptr inbounds ({{.*}}, ptr getelementptr inbounds ({{.*}}, ptr @_ZGR6nested_, i64 1), i64 1, i32 1), align 8
86+
// CHECK-DYNAMIC-BL: ptr getelementptr inbounds ({{.*}}, ptr @_ZGR6nested_, i64 2), align 8
87+
// CHECK-DYNAMIC-BL: store i64 2, ptr getelementptr inbounds ({{.*}}, ptr @_ZGR6nested_, i64 2, i32 1), align 8
8888
// CHECK-DYNAMIC-BL: store ptr @_ZGR6nested_,
8989
// CHECK-DYNAMIC-BL: ptr @nested, align 8
9090
// CHECK-DYNAMIC-BL: store i64 3, ptr getelementptr inbounds ({{.*}}, ptr @nested, i32 0, i32 1), align 8
@@ -123,9 +123,9 @@ std::initializer_list<std::initializer_list<int>> nested = {
123123
// CHECK-DYNAMIC-BE: store i32 5, ptr @_ZGR6nested2_
124124
// CHECK-DYNAMIC-BE: store i32 {{.*}}, ptr getelementptr inbounds (i32, ptr @_ZGR6nested2_, i64 1)
125125
// CHECK-DYNAMIC-BE: store ptr @_ZGR6nested2_,
126-
// CHECK-DYNAMIC-BE: ptr getelementptr inbounds ({{.*}}, ptr getelementptr inbounds ({{.*}}, ptr @_ZGR6nested_, i64 1), i64 1), align 8
126+
// CHECK-DYNAMIC-BE: ptr getelementptr inbounds ({{.*}}, ptr @_ZGR6nested_, i64 2), align 8
127127
// CHECK-DYNAMIC-BE: store ptr getelementptr inbounds ([2 x i32], ptr @_ZGR6nested2_, i64 0, i64 2),
128-
// CHECK-DYNAMIC-BE: ptr getelementptr inbounds ({{.*}}, ptr getelementptr inbounds ({{.*}}, ptr @_ZGR6nested_, i64 1), i64 1, i32 1), align 8
128+
// CHECK-DYNAMIC-BE: ptr getelementptr inbounds ({{.*}}, ptr @_ZGR6nested_, i64 2, i32 1), align 8
129129
// CHECK-DYNAMIC-BE: store ptr @_ZGR6nested_,
130130
// CHECK-DYNAMIC-BE: ptr @nested, align 8
131131
// CHECK-DYNAMIC-BE: store ptr getelementptr inbounds ([3 x {{.*}}], ptr @_ZGR6nested_, i64 0, i64 3),

‎clang/test/CodeGenCXX/cxx0x-initializer-stdinitializerlist.cpp

Lines changed: 2 additions & 2 deletions
Original file line numberDiff line numberDiff line change
@@ -370,8 +370,8 @@ namespace partly_constant {
370370
//
371371
// Third init list.
372372
// CHECK-NOT: @[[PARTLY_CONSTANT_THIRD]],
373-
// CHECK: store ptr {{.*}}@[[PARTLY_CONSTANT_THIRD]]{{.*}}, ptr getelementptr inbounds ({{.*}}, ptr getelementptr inbounds ({{.*}}, ptr {{.*}}@[[PARTLY_CONSTANT_INNER]]{{.*}}, i64 1), i64 1)
374-
// CHECK: store i64 4, ptr getelementptr inbounds ({{.*}}, ptr getelementptr inbounds ({{.*}}, ptr {{.*}}@[[PARTLY_CONSTANT_INNER]]{{.*}}, i64 1), i64 1, i32 1)
373+
// CHECK: store ptr {{.*}}@[[PARTLY_CONSTANT_THIRD]]{{.*}}, ptr getelementptr inbounds ({{.*}}, ptr {{.*}}@[[PARTLY_CONSTANT_INNER]]{{.*}}, i64 2)
374+
// CHECK: store i64 4, ptr getelementptr inbounds ({{.*}}, ptr {{.*}}@[[PARTLY_CONSTANT_INNER]]{{.*}}, i64 2, i32 1)
375375
// CHECK-NOT: @[[PARTLY_CONSTANT_THIRD]],
376376
//
377377
// Outer init list.

‎clang/test/OpenMP/threadprivate_codegen.cpp

Lines changed: 48 additions & 48 deletions
Large diffs are not rendered by default.

‎llvm/lib/IR/ConstantFold.cpp

Lines changed: 52 additions & 4 deletions
Original file line numberDiff line numberDiff line change
@@ -1432,16 +1432,64 @@ static Constant *foldGEPOfGEP(GEPOperator *GEP, Type *PointeeTy, bool InBounds,
14321432
if (GEP->getInRange())
14331433
return nullptr;
14341434

1435-
// Only handle simple case with leading zero index. We cannot perform an
1436-
// actual addition as we don't know the correct index type size to use.
14371435
Constant *Idx0 = cast<Constant>(Idxs[0]);
1438-
if (!Idx0->isNullValue())
1436+
if (Idx0->isNullValue()) {
1437+
// Handle the simple case of a zero index.
1438+
SmallVector<Value*, 16> NewIndices;
1439+
NewIndices.reserve(Idxs.size() + GEP->getNumIndices());
1440+
NewIndices.append(GEP->idx_begin(), GEP->idx_end());
1441+
NewIndices.append(Idxs.begin() + 1, Idxs.end());
1442+
return ConstantExpr::getGetElementPtr(
1443+
GEP->getSourceElementType(), cast<Constant>(GEP->getPointerOperand()),
1444+
NewIndices, InBounds && GEP->isInBounds());
1445+
}
1446+
1447+
gep_type_iterator LastI = gep_type_end(GEP);
1448+
for (gep_type_iterator I = gep_type_begin(GEP), E = gep_type_end(GEP);
1449+
I != E; ++I)
1450+
LastI = I;
1451+
1452+
// We can't combine GEPs if the last index is a struct type.
1453+
if (!LastI.isSequential())
1454+
return nullptr;
1455+
// We could perform the transform with non-constant index, but prefer leaving
1456+
// it as GEP of GEP rather than GEP of add for now.
1457+
ConstantInt *CI = dyn_cast<ConstantInt>(Idx0);
1458+
if (!CI)
1459+
return nullptr;
1460+
1461+
// TODO: This code may be extended to handle vectors as well.
1462+
auto *LastIdx = cast<Constant>(GEP->getOperand(GEP->getNumOperands()-1));
1463+
Type *LastIdxTy = LastIdx->getType();
1464+
if (LastIdxTy->isVectorTy())
14391465
return nullptr;
14401466

14411467
SmallVector<Value*, 16> NewIndices;
14421468
NewIndices.reserve(Idxs.size() + GEP->getNumIndices());
1443-
NewIndices.append(GEP->idx_begin(), GEP->idx_end());
1469+
NewIndices.append(GEP->idx_begin(), GEP->idx_end() - 1);
1470+
1471+
// Add the last index of the source with the first index of the new GEP.
1472+
// Make sure to handle the case when they are actually different types.
1473+
if (LastIdxTy != Idx0->getType()) {
1474+
unsigned CommonExtendedWidth =
1475+
std::max(LastIdxTy->getIntegerBitWidth(),
1476+
Idx0->getType()->getIntegerBitWidth());
1477+
CommonExtendedWidth = std::max(CommonExtendedWidth, 64U);
1478+
1479+
Type *CommonTy =
1480+
Type::getIntNTy(LastIdxTy->getContext(), CommonExtendedWidth);
1481+
if (Idx0->getType() != CommonTy)
1482+
Idx0 = ConstantFoldCastInstruction(Instruction::SExt, Idx0, CommonTy);
1483+
if (LastIdx->getType() != CommonTy)
1484+
LastIdx =
1485+
ConstantFoldCastInstruction(Instruction::SExt, LastIdx, CommonTy);
1486+
if (!Idx0 || !LastIdx)
1487+
return nullptr;
1488+
}
1489+
1490+
NewIndices.push_back(ConstantExpr::get(Instruction::Add, Idx0, LastIdx));
14441491
NewIndices.append(Idxs.begin() + 1, Idxs.end());
1492+
14451493
return ConstantExpr::getGetElementPtr(
14461494
GEP->getSourceElementType(), cast<Constant>(GEP->getPointerOperand()),
14471495
NewIndices, InBounds && GEP->isInBounds());

‎llvm/test/Other/constant-fold-gep.ll

Lines changed: 1 addition & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -104,7 +104,7 @@
104104

105105
; Fold GEP of a GEP. Very simple cases are folded without targetdata.
106106

107-
; PLAIN: @Y = global ptr getelementptr inbounds ([3 x { i32, i32 }], ptr getelementptr inbounds ([3 x { i32, i32 }], ptr @ext, i64 1), i64 1)
107+
; PLAIN: @Y = global ptr getelementptr inbounds ([3 x { i32, i32 }], ptr @ext, i64 2)
108108
; PLAIN: @Z = global ptr getelementptr inbounds (i32, ptr getelementptr inbounds ([3 x { i32, i32 }], ptr @ext, i64 0, i64 1, i32 0), i64 1)
109109
; OPT: @Y = local_unnamed_addr global ptr getelementptr inbounds (i8, ptr @ext, i64 48)
110110
; OPT: @Z = local_unnamed_addr global ptr getelementptr inbounds (i8, ptr @ext, i64 12)

‎llvm/test/Transforms/InstCombine/gepgep.ll

Lines changed: 1 addition & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -10,7 +10,7 @@ declare void @use(ptr)
1010

1111
define void @f() {
1212
; CHECK-LABEL: define void @f() {
13-
; CHECK-NEXT: call void @use(ptr getelementptr (i8, ptr getelementptr (i8, ptr @buffer, i64 add (i64 sub (i64 0, i64 ptrtoint (ptr @buffer to i64)), i64 63)), i64 64))
13+
; CHECK-NEXT: call void @use(ptr getelementptr (i8, ptr @buffer, i64 add (i64 sub (i64 0, i64 ptrtoint (ptr @buffer to i64)), i64 127)))
1414
; CHECK-NEXT: ret void
1515
;
1616
call void @use(ptr getelementptr (i8, ptr getelementptr (i8, ptr @buffer, i64 add (i64 sub (i64 0, i64 ptrtoint (ptr @buffer to i64)), i64 63)), i64 64))

‎llvm/test/Transforms/InstCombine/getelementptr.ll

Lines changed: 2 additions & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -1684,9 +1684,10 @@ if.else:
16841684

16851685
@g = external global i8
16861686

1687+
; FIXME: This is a miscompile
16871688
define ptr @constexpr_gep_of_gep_with_narrow_type() {
16881689
; CHECK-LABEL: @constexpr_gep_of_gep_with_narrow_type(
1689-
; CHECK-NEXT: ret ptr getelementptr (i8, ptr @g, i64 254)
1690+
; CHECK-NEXT: ret ptr getelementptr (i8, ptr @g, i64 -2)
16901691
;
16911692
ret ptr getelementptr (i8, ptr getelementptr (i8, ptr @g, i8 127), i8 127)
16921693
}

‎llvm/test/Transforms/InstCombine/ptrtoint-nullgep.ll

Lines changed: 8 additions & 8 deletions
Original file line numberDiff line numberDiff line change
@@ -68,10 +68,10 @@ define void @constant_fold_ptrtoint_of_gep_of_nullgep() {
6868
; LLPARSER-NEXT: call void @use_i64(i64 ptrtoint (ptr addrspace(1) getelementptr inbounds (i8, ptr addrspace(1) null, i64 1234) to i64))
6969
; LLPARSER-NEXT: call void @use_i64(i64 ptrtoint (ptr addrspace(1) getelementptr (i8, ptr addrspace(1) null, i64 1234) to i64))
7070
; LLPARSER-NEXT: call void @use_i64(i64 ptrtoint (ptr addrspace(1) getelementptr (i8, ptr addrspace(1) null, i64 1234) to i64))
71-
; LLPARSER-NEXT: call void @use_i64(i64 ptrtoint (ptr addrspace(1) getelementptr inbounds (i8, ptr addrspace(1) getelementptr inbounds (i8, ptr addrspace(1) null, i64 -1), i64 1) to i64))
72-
; LLPARSER-NEXT: call void @use_i64(i64 ptrtoint (ptr addrspace(1) getelementptr (i8, ptr addrspace(1) getelementptr inbounds (i8, ptr addrspace(1) null, i64 -1), i64 1) to i64))
73-
; LLPARSER-NEXT: call void @use_i64(i64 ptrtoint (ptr addrspace(1) getelementptr inbounds (i8, ptr addrspace(1) getelementptr (i8, ptr addrspace(1) null, i64 -1), i64 1) to i64))
74-
; LLPARSER-NEXT: call void @use_i64(i64 ptrtoint (ptr addrspace(1) getelementptr (i8, ptr addrspace(1) getelementptr (i8, ptr addrspace(1) null, i64 -1), i64 1) to i64))
71+
; LLPARSER-NEXT: call void @use_i64(i64 0)
72+
; LLPARSER-NEXT: call void @use_i64(i64 0)
73+
; LLPARSER-NEXT: call void @use_i64(i64 0)
74+
; LLPARSER-NEXT: call void @use_i64(i64 0)
7575
; LLPARSER-NEXT: ret void
7676
;
7777
; INSTSIMPLIFY-LABEL: define {{[^@]+}}@constant_fold_ptrtoint_of_gep_of_nullgep() {
@@ -83,10 +83,10 @@ define void @constant_fold_ptrtoint_of_gep_of_nullgep() {
8383
; INSTSIMPLIFY-NEXT: call void @use_i64(i64 ptrtoint (ptr addrspace(1) getelementptr inbounds (i8, ptr addrspace(1) null, i64 1234) to i64))
8484
; INSTSIMPLIFY-NEXT: call void @use_i64(i64 ptrtoint (ptr addrspace(1) getelementptr (i8, ptr addrspace(1) null, i64 1234) to i64))
8585
; INSTSIMPLIFY-NEXT: call void @use_i64(i64 ptrtoint (ptr addrspace(1) getelementptr (i8, ptr addrspace(1) null, i64 1234) to i64))
86-
; INSTSIMPLIFY-NEXT: call void @use_i64(i64 ptrtoint (ptr addrspace(1) getelementptr inbounds (i8, ptr addrspace(1) getelementptr inbounds (i8, ptr addrspace(1) null, i64 -1), i64 1) to i64))
87-
; INSTSIMPLIFY-NEXT: call void @use_i64(i64 ptrtoint (ptr addrspace(1) getelementptr (i8, ptr addrspace(1) getelementptr inbounds (i8, ptr addrspace(1) null, i64 -1), i64 1) to i64))
88-
; INSTSIMPLIFY-NEXT: call void @use_i64(i64 ptrtoint (ptr addrspace(1) getelementptr inbounds (i8, ptr addrspace(1) getelementptr (i8, ptr addrspace(1) null, i64 -1), i64 1) to i64))
89-
; INSTSIMPLIFY-NEXT: call void @use_i64(i64 ptrtoint (ptr addrspace(1) getelementptr (i8, ptr addrspace(1) getelementptr (i8, ptr addrspace(1) null, i64 -1), i64 1) to i64))
86+
; INSTSIMPLIFY-NEXT: call void @use_i64(i64 0)
87+
; INSTSIMPLIFY-NEXT: call void @use_i64(i64 0)
88+
; INSTSIMPLIFY-NEXT: call void @use_i64(i64 0)
89+
; INSTSIMPLIFY-NEXT: call void @use_i64(i64 0)
9090
; INSTSIMPLIFY-NEXT: ret void
9191
;
9292
; INSTCOMBINE-LABEL: define {{[^@]+}}@constant_fold_ptrtoint_of_gep_of_nullgep() {

0 commit comments

Comments
 (0)