2019-02-05 20:05:53 +08:00
|
|
|
// RUN: %clang_cc1 -std=c++14 -triple i386-unknown-unknown %s -emit-llvm -o - | FileCheck %s
|
2011-06-25 07:30:52 +08:00
|
|
|
|
|
|
|
|
// rdar://problem/9246208
|
|
|
|
|
|
|
|
|
|
// Basic test.
|
|
|
|
|
namespace test0 {
|
|
|
|
|
struct A {
|
|
|
|
|
A();
|
|
|
|
|
int x;
|
|
|
|
|
};
|
|
|
|
|
|
|
|
|
|
typedef A elt;
|
|
|
|
|
|
2020-12-31 16:27:11 +08:00
|
|
|
// CHECK: define{{.*}} [[A:%.*]]* @_ZN5test04testEs(i16 signext
|
2011-06-25 07:30:52 +08:00
|
|
|
// CHECK: [[N:%.*]] = sext i16 {{%.*}} to i32
|
2011-07-10 01:41:47 +08:00
|
|
|
// CHECK-NEXT: [[T0:%.*]] = call { i32, i1 } @llvm.umul.with.overflow.i32(i32 [[N]], i32 4)
|
|
|
|
|
// CHECK-NEXT: [[T1:%.*]] = extractvalue { i32, i1 } [[T0]], 1
|
|
|
|
|
// CHECK-NEXT: [[T2:%.*]] = extractvalue { i32, i1 } [[T0]], 0
|
2011-06-25 07:30:52 +08:00
|
|
|
// CHECK-NEXT: [[T3:%.*]] = select i1 [[T1]], i32 -1, i32 [[T2]]
|
[clang] Annotating C++'s `operator new` with more attributes
Summary:
Right now we annotate C++'s `operator new` with `noalias` attribute,
which very much is healthy for optimizations.
However as per [[ http://eel.is/c++draft/basic.stc.dynamic.allocation | `[basic.stc.dynamic.allocation]` ]],
there are more promises on global `operator new`, namely:
* non-`std::nothrow_t` `operator new` *never* returns `nullptr`
* If `std::align_val_t align` parameter is taken, the pointer will also be `align`-aligned
* ~~global `operator new`-returned pointer is `__STDCPP_DEFAULT_NEW_ALIGNMENT__`-aligned ~~ It's more caveated than that.
Supplying this information may not cause immediate landslide effects
on any specific benchmarks, but it for sure will be healthy for optimizer
in the sense that the IR will better reflect the guarantees provided in the source code.
The caveat is `-fno-assume-sane-operator-new`, which currently prevents emitting `noalias`
attribute, and is automatically passed by Sanitizers ([[ https://bugs.llvm.org/show_bug.cgi?id=16386 | PR16386 ]]) - should it also cover these attributes?
The problem is that the flag is back-end-specific, as seen in `test/Modules/explicit-build-flags.cpp`.
But while it is okay to add `noalias` metadata in backend, we really should be adding at least
the alignment metadata to the AST, since that allows us to perform sema checks on it.
Reviewers: erichkeane, rjmccall, jdoerfert, eugenis, rsmith
Reviewed By: rsmith
Subscribers: xbolva00, jrtc27, atanasyan, nlopes, cfe-commits
Tags: #llvm, #clang
Differential Revision: https://reviews.llvm.org/D73380
2020-02-26 06:37:17 +08:00
|
|
|
// CHECK-NEXT: call noalias nonnull i8* @_Znaj(i32 [[T3]])
|
2011-07-13 14:10:41 +08:00
|
|
|
// CHECK: getelementptr inbounds {{.*}}, i32 [[N]]
|
2011-06-25 07:30:52 +08:00
|
|
|
elt *test(short s) {
|
|
|
|
|
return new elt[s];
|
|
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
// test0 with a nested array.
|
|
|
|
|
namespace test1 {
|
|
|
|
|
struct A {
|
|
|
|
|
A();
|
|
|
|
|
int x;
|
|
|
|
|
};
|
|
|
|
|
|
|
|
|
|
typedef A elt[100];
|
|
|
|
|
|
2020-12-31 16:27:11 +08:00
|
|
|
// CHECK: define{{.*}} [100 x [[A:%.*]]]* @_ZN5test14testEs(i16 signext
|
2011-06-25 07:30:52 +08:00
|
|
|
// CHECK: [[N:%.*]] = sext i16 {{%.*}} to i32
|
2011-07-10 01:41:47 +08:00
|
|
|
// CHECK-NEXT: [[T0:%.*]] = call { i32, i1 } @llvm.umul.with.overflow.i32(i32 [[N]], i32 400)
|
|
|
|
|
// CHECK-NEXT: [[T1:%.*]] = extractvalue { i32, i1 } [[T0]], 1
|
|
|
|
|
// CHECK-NEXT: [[T2:%.*]] = extractvalue { i32, i1 } [[T0]], 0
|
2011-06-25 07:30:52 +08:00
|
|
|
// CHECK-NEXT: [[T3:%.*]] = mul i32 [[N]], 100
|
|
|
|
|
// CHECK-NEXT: [[T4:%.*]] = select i1 [[T1]], i32 -1, i32 [[T2]]
|
[clang] Annotating C++'s `operator new` with more attributes
Summary:
Right now we annotate C++'s `operator new` with `noalias` attribute,
which very much is healthy for optimizations.
However as per [[ http://eel.is/c++draft/basic.stc.dynamic.allocation | `[basic.stc.dynamic.allocation]` ]],
there are more promises on global `operator new`, namely:
* non-`std::nothrow_t` `operator new` *never* returns `nullptr`
* If `std::align_val_t align` parameter is taken, the pointer will also be `align`-aligned
* ~~global `operator new`-returned pointer is `__STDCPP_DEFAULT_NEW_ALIGNMENT__`-aligned ~~ It's more caveated than that.
Supplying this information may not cause immediate landslide effects
on any specific benchmarks, but it for sure will be healthy for optimizer
in the sense that the IR will better reflect the guarantees provided in the source code.
The caveat is `-fno-assume-sane-operator-new`, which currently prevents emitting `noalias`
attribute, and is automatically passed by Sanitizers ([[ https://bugs.llvm.org/show_bug.cgi?id=16386 | PR16386 ]]) - should it also cover these attributes?
The problem is that the flag is back-end-specific, as seen in `test/Modules/explicit-build-flags.cpp`.
But while it is okay to add `noalias` metadata in backend, we really should be adding at least
the alignment metadata to the AST, since that allows us to perform sema checks on it.
Reviewers: erichkeane, rjmccall, jdoerfert, eugenis, rsmith
Reviewed By: rsmith
Subscribers: xbolva00, jrtc27, atanasyan, nlopes, cfe-commits
Tags: #llvm, #clang
Differential Revision: https://reviews.llvm.org/D73380
2020-02-26 06:37:17 +08:00
|
|
|
// CHECK-NEXT: call noalias nonnull i8* @_Znaj(i32 [[T4]])
|
2011-07-13 14:10:41 +08:00
|
|
|
// CHECK: getelementptr inbounds {{.*}}, i32 [[T3]]
|
2011-06-25 07:30:52 +08:00
|
|
|
elt *test(short s) {
|
|
|
|
|
return new elt[s];
|
|
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
// test1 with an array cookie.
|
|
|
|
|
namespace test2 {
|
|
|
|
|
struct A {
|
|
|
|
|
A();
|
|
|
|
|
~A();
|
|
|
|
|
int x;
|
|
|
|
|
};
|
|
|
|
|
|
|
|
|
|
typedef A elt[100];
|
|
|
|
|
|
2020-12-31 16:27:11 +08:00
|
|
|
// CHECK: define{{.*}} [100 x [[A:%.*]]]* @_ZN5test24testEs(i16 signext
|
2011-06-25 07:30:52 +08:00
|
|
|
// CHECK: [[N:%.*]] = sext i16 {{%.*}} to i32
|
2011-07-10 01:41:47 +08:00
|
|
|
// CHECK-NEXT: [[T0:%.*]] = call { i32, i1 } @llvm.umul.with.overflow.i32(i32 [[N]], i32 400)
|
|
|
|
|
// CHECK-NEXT: [[T1:%.*]] = extractvalue { i32, i1 } [[T0]], 1
|
|
|
|
|
// CHECK-NEXT: [[T2:%.*]] = extractvalue { i32, i1 } [[T0]], 0
|
2011-06-25 07:30:52 +08:00
|
|
|
// CHECK-NEXT: [[T3:%.*]] = mul i32 [[N]], 100
|
2011-07-10 01:41:47 +08:00
|
|
|
// CHECK-NEXT: [[T4:%.*]] = call { i32, i1 } @llvm.uadd.with.overflow.i32(i32 [[T2]], i32 4)
|
|
|
|
|
// CHECK-NEXT: [[T5:%.*]] = extractvalue { i32, i1 } [[T4]], 1
|
2011-06-25 07:30:52 +08:00
|
|
|
// CHECK-NEXT: [[T6:%.*]] = or i1 [[T1]], [[T5]]
|
2011-07-10 01:41:47 +08:00
|
|
|
// CHECK-NEXT: [[T7:%.*]] = extractvalue { i32, i1 } [[T4]], 0
|
2011-06-25 07:30:52 +08:00
|
|
|
// CHECK-NEXT: [[T8:%.*]] = select i1 [[T6]], i32 -1, i32 [[T7]]
|
[clang] Annotating C++'s `operator new` with more attributes
Summary:
Right now we annotate C++'s `operator new` with `noalias` attribute,
which very much is healthy for optimizations.
However as per [[ http://eel.is/c++draft/basic.stc.dynamic.allocation | `[basic.stc.dynamic.allocation]` ]],
there are more promises on global `operator new`, namely:
* non-`std::nothrow_t` `operator new` *never* returns `nullptr`
* If `std::align_val_t align` parameter is taken, the pointer will also be `align`-aligned
* ~~global `operator new`-returned pointer is `__STDCPP_DEFAULT_NEW_ALIGNMENT__`-aligned ~~ It's more caveated than that.
Supplying this information may not cause immediate landslide effects
on any specific benchmarks, but it for sure will be healthy for optimizer
in the sense that the IR will better reflect the guarantees provided in the source code.
The caveat is `-fno-assume-sane-operator-new`, which currently prevents emitting `noalias`
attribute, and is automatically passed by Sanitizers ([[ https://bugs.llvm.org/show_bug.cgi?id=16386 | PR16386 ]]) - should it also cover these attributes?
The problem is that the flag is back-end-specific, as seen in `test/Modules/explicit-build-flags.cpp`.
But while it is okay to add `noalias` metadata in backend, we really should be adding at least
the alignment metadata to the AST, since that allows us to perform sema checks on it.
Reviewers: erichkeane, rjmccall, jdoerfert, eugenis, rsmith
Reviewed By: rsmith
Subscribers: xbolva00, jrtc27, atanasyan, nlopes, cfe-commits
Tags: #llvm, #clang
Differential Revision: https://reviews.llvm.org/D73380
2020-02-26 06:37:17 +08:00
|
|
|
// CHECK-NEXT: call noalias nonnull i8* @_Znaj(i32 [[T8]])
|
2011-07-13 14:10:41 +08:00
|
|
|
// CHECK: getelementptr inbounds {{.*}}, i32 [[T3]]
|
2011-06-25 07:30:52 +08:00
|
|
|
elt *test(short s) {
|
|
|
|
|
return new elt[s];
|
|
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
// test0 with a 1-byte element.
|
|
|
|
|
namespace test4 {
|
|
|
|
|
struct A {
|
|
|
|
|
A();
|
|
|
|
|
};
|
|
|
|
|
|
|
|
|
|
typedef A elt;
|
|
|
|
|
|
2020-12-31 16:27:11 +08:00
|
|
|
// CHECK: define{{.*}} [[A:%.*]]* @_ZN5test44testEs(i16 signext
|
2011-06-25 07:30:52 +08:00
|
|
|
// CHECK: [[N:%.*]] = sext i16 {{%.*}} to i32
|
[clang] Annotating C++'s `operator new` with more attributes
Summary:
Right now we annotate C++'s `operator new` with `noalias` attribute,
which very much is healthy for optimizations.
However as per [[ http://eel.is/c++draft/basic.stc.dynamic.allocation | `[basic.stc.dynamic.allocation]` ]],
there are more promises on global `operator new`, namely:
* non-`std::nothrow_t` `operator new` *never* returns `nullptr`
* If `std::align_val_t align` parameter is taken, the pointer will also be `align`-aligned
* ~~global `operator new`-returned pointer is `__STDCPP_DEFAULT_NEW_ALIGNMENT__`-aligned ~~ It's more caveated than that.
Supplying this information may not cause immediate landslide effects
on any specific benchmarks, but it for sure will be healthy for optimizer
in the sense that the IR will better reflect the guarantees provided in the source code.
The caveat is `-fno-assume-sane-operator-new`, which currently prevents emitting `noalias`
attribute, and is automatically passed by Sanitizers ([[ https://bugs.llvm.org/show_bug.cgi?id=16386 | PR16386 ]]) - should it also cover these attributes?
The problem is that the flag is back-end-specific, as seen in `test/Modules/explicit-build-flags.cpp`.
But while it is okay to add `noalias` metadata in backend, we really should be adding at least
the alignment metadata to the AST, since that allows us to perform sema checks on it.
Reviewers: erichkeane, rjmccall, jdoerfert, eugenis, rsmith
Reviewed By: rsmith
Subscribers: xbolva00, jrtc27, atanasyan, nlopes, cfe-commits
Tags: #llvm, #clang
Differential Revision: https://reviews.llvm.org/D73380
2020-02-26 06:37:17 +08:00
|
|
|
// CHECK-NEXT: call noalias nonnull i8* @_Znaj(i32 [[N]])
|
2011-07-13 14:10:41 +08:00
|
|
|
// CHECK: getelementptr inbounds {{.*}}, i32 [[N]]
|
2011-06-25 07:30:52 +08:00
|
|
|
elt *test(short s) {
|
|
|
|
|
return new elt[s];
|
|
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
// test4 with no sext required.
|
|
|
|
|
namespace test5 {
|
|
|
|
|
struct A {
|
|
|
|
|
A();
|
|
|
|
|
};
|
|
|
|
|
|
|
|
|
|
typedef A elt;
|
|
|
|
|
|
2020-12-31 16:27:11 +08:00
|
|
|
// CHECK: define{{.*}} [[A:%.*]]* @_ZN5test54testEi(i32
|
2015-02-28 05:19:58 +08:00
|
|
|
// CHECK: [[N:%.*]] = load i32, i32*
|
[clang] Annotating C++'s `operator new` with more attributes
Summary:
Right now we annotate C++'s `operator new` with `noalias` attribute,
which very much is healthy for optimizations.
However as per [[ http://eel.is/c++draft/basic.stc.dynamic.allocation | `[basic.stc.dynamic.allocation]` ]],
there are more promises on global `operator new`, namely:
* non-`std::nothrow_t` `operator new` *never* returns `nullptr`
* If `std::align_val_t align` parameter is taken, the pointer will also be `align`-aligned
* ~~global `operator new`-returned pointer is `__STDCPP_DEFAULT_NEW_ALIGNMENT__`-aligned ~~ It's more caveated than that.
Supplying this information may not cause immediate landslide effects
on any specific benchmarks, but it for sure will be healthy for optimizer
in the sense that the IR will better reflect the guarantees provided in the source code.
The caveat is `-fno-assume-sane-operator-new`, which currently prevents emitting `noalias`
attribute, and is automatically passed by Sanitizers ([[ https://bugs.llvm.org/show_bug.cgi?id=16386 | PR16386 ]]) - should it also cover these attributes?
The problem is that the flag is back-end-specific, as seen in `test/Modules/explicit-build-flags.cpp`.
But while it is okay to add `noalias` metadata in backend, we really should be adding at least
the alignment metadata to the AST, since that allows us to perform sema checks on it.
Reviewers: erichkeane, rjmccall, jdoerfert, eugenis, rsmith
Reviewed By: rsmith
Subscribers: xbolva00, jrtc27, atanasyan, nlopes, cfe-commits
Tags: #llvm, #clang
Differential Revision: https://reviews.llvm.org/D73380
2020-02-26 06:37:17 +08:00
|
|
|
// CHECK-NEXT: call noalias nonnull i8* @_Znaj(i32 [[N]])
|
2011-07-13 14:10:41 +08:00
|
|
|
// CHECK: getelementptr inbounds {{.*}}, i32 [[N]]
|
2011-06-25 07:30:52 +08:00
|
|
|
elt *test(int s) {
|
|
|
|
|
return new elt[s];
|
|
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
// test0 with an unsigned size.
|
|
|
|
|
namespace test6 {
|
|
|
|
|
struct A {
|
|
|
|
|
A();
|
|
|
|
|
int x;
|
|
|
|
|
};
|
|
|
|
|
|
|
|
|
|
typedef A elt;
|
|
|
|
|
|
2020-12-31 16:27:11 +08:00
|
|
|
// CHECK: define{{.*}} [[A:%.*]]* @_ZN5test64testEt(i16 zeroext
|
2011-06-25 07:30:52 +08:00
|
|
|
// CHECK: [[N:%.*]] = zext i16 {{%.*}} to i32
|
2011-07-10 01:41:47 +08:00
|
|
|
// CHECK-NEXT: [[T0:%.*]] = call { i32, i1 } @llvm.umul.with.overflow.i32(i32 [[N]], i32 4)
|
|
|
|
|
// CHECK-NEXT: [[T1:%.*]] = extractvalue { i32, i1 } [[T0]], 1
|
|
|
|
|
// CHECK-NEXT: [[T2:%.*]] = extractvalue { i32, i1 } [[T0]], 0
|
2011-06-25 07:30:52 +08:00
|
|
|
// CHECK-NEXT: [[T3:%.*]] = select i1 [[T1]], i32 -1, i32 [[T2]]
|
[clang] Annotating C++'s `operator new` with more attributes
Summary:
Right now we annotate C++'s `operator new` with `noalias` attribute,
which very much is healthy for optimizations.
However as per [[ http://eel.is/c++draft/basic.stc.dynamic.allocation | `[basic.stc.dynamic.allocation]` ]],
there are more promises on global `operator new`, namely:
* non-`std::nothrow_t` `operator new` *never* returns `nullptr`
* If `std::align_val_t align` parameter is taken, the pointer will also be `align`-aligned
* ~~global `operator new`-returned pointer is `__STDCPP_DEFAULT_NEW_ALIGNMENT__`-aligned ~~ It's more caveated than that.
Supplying this information may not cause immediate landslide effects
on any specific benchmarks, but it for sure will be healthy for optimizer
in the sense that the IR will better reflect the guarantees provided in the source code.
The caveat is `-fno-assume-sane-operator-new`, which currently prevents emitting `noalias`
attribute, and is automatically passed by Sanitizers ([[ https://bugs.llvm.org/show_bug.cgi?id=16386 | PR16386 ]]) - should it also cover these attributes?
The problem is that the flag is back-end-specific, as seen in `test/Modules/explicit-build-flags.cpp`.
But while it is okay to add `noalias` metadata in backend, we really should be adding at least
the alignment metadata to the AST, since that allows us to perform sema checks on it.
Reviewers: erichkeane, rjmccall, jdoerfert, eugenis, rsmith
Reviewed By: rsmith
Subscribers: xbolva00, jrtc27, atanasyan, nlopes, cfe-commits
Tags: #llvm, #clang
Differential Revision: https://reviews.llvm.org/D73380
2020-02-26 06:37:17 +08:00
|
|
|
// CHECK-NEXT: call noalias nonnull i8* @_Znaj(i32 [[T3]])
|
2011-07-13 14:10:41 +08:00
|
|
|
// CHECK: getelementptr inbounds {{.*}}, i32 [[N]]
|
2011-06-25 07:30:52 +08:00
|
|
|
elt *test(unsigned short s) {
|
|
|
|
|
return new elt[s];
|
|
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
// test1 with an unsigned size.
|
|
|
|
|
namespace test7 {
|
|
|
|
|
struct A {
|
|
|
|
|
A();
|
|
|
|
|
int x;
|
|
|
|
|
};
|
|
|
|
|
|
|
|
|
|
typedef A elt[100];
|
|
|
|
|
|
2020-12-31 16:27:11 +08:00
|
|
|
// CHECK: define{{.*}} [100 x [[A:%.*]]]* @_ZN5test74testEt(i16 zeroext
|
2011-06-25 07:30:52 +08:00
|
|
|
// CHECK: [[N:%.*]] = zext i16 {{%.*}} to i32
|
2011-07-10 01:41:47 +08:00
|
|
|
// CHECK-NEXT: [[T0:%.*]] = call { i32, i1 } @llvm.umul.with.overflow.i32(i32 [[N]], i32 400)
|
|
|
|
|
// CHECK-NEXT: [[T1:%.*]] = extractvalue { i32, i1 } [[T0]], 1
|
|
|
|
|
// CHECK-NEXT: [[T2:%.*]] = extractvalue { i32, i1 } [[T0]], 0
|
2011-06-25 07:30:52 +08:00
|
|
|
// CHECK-NEXT: [[T3:%.*]] = mul i32 [[N]], 100
|
|
|
|
|
// CHECK-NEXT: [[T4:%.*]] = select i1 [[T1]], i32 -1, i32 [[T2]]
|
[clang] Annotating C++'s `operator new` with more attributes
Summary:
Right now we annotate C++'s `operator new` with `noalias` attribute,
which very much is healthy for optimizations.
However as per [[ http://eel.is/c++draft/basic.stc.dynamic.allocation | `[basic.stc.dynamic.allocation]` ]],
there are more promises on global `operator new`, namely:
* non-`std::nothrow_t` `operator new` *never* returns `nullptr`
* If `std::align_val_t align` parameter is taken, the pointer will also be `align`-aligned
* ~~global `operator new`-returned pointer is `__STDCPP_DEFAULT_NEW_ALIGNMENT__`-aligned ~~ It's more caveated than that.
Supplying this information may not cause immediate landslide effects
on any specific benchmarks, but it for sure will be healthy for optimizer
in the sense that the IR will better reflect the guarantees provided in the source code.
The caveat is `-fno-assume-sane-operator-new`, which currently prevents emitting `noalias`
attribute, and is automatically passed by Sanitizers ([[ https://bugs.llvm.org/show_bug.cgi?id=16386 | PR16386 ]]) - should it also cover these attributes?
The problem is that the flag is back-end-specific, as seen in `test/Modules/explicit-build-flags.cpp`.
But while it is okay to add `noalias` metadata in backend, we really should be adding at least
the alignment metadata to the AST, since that allows us to perform sema checks on it.
Reviewers: erichkeane, rjmccall, jdoerfert, eugenis, rsmith
Reviewed By: rsmith
Subscribers: xbolva00, jrtc27, atanasyan, nlopes, cfe-commits
Tags: #llvm, #clang
Differential Revision: https://reviews.llvm.org/D73380
2020-02-26 06:37:17 +08:00
|
|
|
// CHECK-NEXT: call noalias nonnull i8* @_Znaj(i32 [[T4]])
|
2011-07-13 14:10:41 +08:00
|
|
|
// CHECK: getelementptr inbounds {{.*}}, i32 [[T3]]
|
2011-06-25 07:30:52 +08:00
|
|
|
elt *test(unsigned short s) {
|
|
|
|
|
return new elt[s];
|
|
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
// test0 with a signed type larger than size_t.
|
|
|
|
|
namespace test8 {
|
|
|
|
|
struct A {
|
|
|
|
|
A();
|
|
|
|
|
int x;
|
|
|
|
|
};
|
|
|
|
|
|
|
|
|
|
typedef A elt;
|
|
|
|
|
|
2020-12-31 16:27:11 +08:00
|
|
|
// CHECK: define{{.*}} [[A:%.*]]* @_ZN5test84testEx(i64
|
2015-02-28 05:19:58 +08:00
|
|
|
// CHECK: [[N:%.*]] = load i64, i64*
|
2011-06-25 07:30:52 +08:00
|
|
|
// CHECK-NEXT: [[T1:%.*]] = trunc i64 [[N]] to i32
|
2011-07-10 01:41:47 +08:00
|
|
|
// CHECK-NEXT: [[T2:%.*]] = call { i32, i1 } @llvm.umul.with.overflow.i32(i32 [[T1]], i32 4)
|
|
|
|
|
// CHECK-NEXT: [[T3:%.*]] = extractvalue { i32, i1 } [[T2]], 1
|
|
|
|
|
// CHECK-NEXT: [[T5:%.*]] = extractvalue { i32, i1 } [[T2]], 0
|
2017-12-09 20:09:54 +08:00
|
|
|
// CHECK-NEXT: [[T6:%.*]] = select i1 [[T3]], i32 -1, i32 [[T5]]
|
[clang] Annotating C++'s `operator new` with more attributes
Summary:
Right now we annotate C++'s `operator new` with `noalias` attribute,
which very much is healthy for optimizations.
However as per [[ http://eel.is/c++draft/basic.stc.dynamic.allocation | `[basic.stc.dynamic.allocation]` ]],
there are more promises on global `operator new`, namely:
* non-`std::nothrow_t` `operator new` *never* returns `nullptr`
* If `std::align_val_t align` parameter is taken, the pointer will also be `align`-aligned
* ~~global `operator new`-returned pointer is `__STDCPP_DEFAULT_NEW_ALIGNMENT__`-aligned ~~ It's more caveated than that.
Supplying this information may not cause immediate landslide effects
on any specific benchmarks, but it for sure will be healthy for optimizer
in the sense that the IR will better reflect the guarantees provided in the source code.
The caveat is `-fno-assume-sane-operator-new`, which currently prevents emitting `noalias`
attribute, and is automatically passed by Sanitizers ([[ https://bugs.llvm.org/show_bug.cgi?id=16386 | PR16386 ]]) - should it also cover these attributes?
The problem is that the flag is back-end-specific, as seen in `test/Modules/explicit-build-flags.cpp`.
But while it is okay to add `noalias` metadata in backend, we really should be adding at least
the alignment metadata to the AST, since that allows us to perform sema checks on it.
Reviewers: erichkeane, rjmccall, jdoerfert, eugenis, rsmith
Reviewed By: rsmith
Subscribers: xbolva00, jrtc27, atanasyan, nlopes, cfe-commits
Tags: #llvm, #clang
Differential Revision: https://reviews.llvm.org/D73380
2020-02-26 06:37:17 +08:00
|
|
|
// CHECK-NEXT: call noalias nonnull i8* @_Znaj(i32 [[T6]])
|
2011-07-13 14:10:41 +08:00
|
|
|
// CHECK: getelementptr inbounds {{.*}}, i32 [[T1]]
|
2011-06-25 07:30:52 +08:00
|
|
|
elt *test(long long s) {
|
|
|
|
|
return new elt[s];
|
|
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
// test8 with an unsigned type.
|
|
|
|
|
namespace test9 {
|
|
|
|
|
struct A {
|
|
|
|
|
A();
|
|
|
|
|
int x;
|
|
|
|
|
};
|
|
|
|
|
|
|
|
|
|
typedef A elt;
|
|
|
|
|
|
2020-12-31 16:27:11 +08:00
|
|
|
// CHECK: define{{.*}} [[A:%.*]]* @_ZN5test94testEy(i64
|
2015-02-28 05:19:58 +08:00
|
|
|
// CHECK: [[N:%.*]] = load i64, i64*
|
2011-06-25 07:30:52 +08:00
|
|
|
// CHECK-NEXT: [[T1:%.*]] = trunc i64 [[N]] to i32
|
2011-07-10 01:41:47 +08:00
|
|
|
// CHECK-NEXT: [[T2:%.*]] = call { i32, i1 } @llvm.umul.with.overflow.i32(i32 [[T1]], i32 4)
|
|
|
|
|
// CHECK-NEXT: [[T3:%.*]] = extractvalue { i32, i1 } [[T2]], 1
|
|
|
|
|
// CHECK-NEXT: [[T5:%.*]] = extractvalue { i32, i1 } [[T2]], 0
|
2017-12-09 20:09:54 +08:00
|
|
|
// CHECK-NEXT: [[T6:%.*]] = select i1 [[T3]], i32 -1, i32 [[T5]]
|
[clang] Annotating C++'s `operator new` with more attributes
Summary:
Right now we annotate C++'s `operator new` with `noalias` attribute,
which very much is healthy for optimizations.
However as per [[ http://eel.is/c++draft/basic.stc.dynamic.allocation | `[basic.stc.dynamic.allocation]` ]],
there are more promises on global `operator new`, namely:
* non-`std::nothrow_t` `operator new` *never* returns `nullptr`
* If `std::align_val_t align` parameter is taken, the pointer will also be `align`-aligned
* ~~global `operator new`-returned pointer is `__STDCPP_DEFAULT_NEW_ALIGNMENT__`-aligned ~~ It's more caveated than that.
Supplying this information may not cause immediate landslide effects
on any specific benchmarks, but it for sure will be healthy for optimizer
in the sense that the IR will better reflect the guarantees provided in the source code.
The caveat is `-fno-assume-sane-operator-new`, which currently prevents emitting `noalias`
attribute, and is automatically passed by Sanitizers ([[ https://bugs.llvm.org/show_bug.cgi?id=16386 | PR16386 ]]) - should it also cover these attributes?
The problem is that the flag is back-end-specific, as seen in `test/Modules/explicit-build-flags.cpp`.
But while it is okay to add `noalias` metadata in backend, we really should be adding at least
the alignment metadata to the AST, since that allows us to perform sema checks on it.
Reviewers: erichkeane, rjmccall, jdoerfert, eugenis, rsmith
Reviewed By: rsmith
Subscribers: xbolva00, jrtc27, atanasyan, nlopes, cfe-commits
Tags: #llvm, #clang
Differential Revision: https://reviews.llvm.org/D73380
2020-02-26 06:37:17 +08:00
|
|
|
// CHECK-NEXT: call noalias nonnull i8* @_Znaj(i32 [[T6]])
|
2011-07-13 14:10:41 +08:00
|
|
|
// CHECK: getelementptr inbounds {{.*}}, i32 [[T1]]
|
2011-06-25 07:30:52 +08:00
|
|
|
elt *test(unsigned long long s) {
|
|
|
|
|
return new elt[s];
|
|
|
|
|
}
|
|
|
|
|
}
|