mikros/rust
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity
d06ca0ffaf
rust/tests/codegen/function-arguments.rs

Ignoring revisions in .git-blame-ignore-revs. Click here to bypass and see the normal blame view.

291 lines
7.7 KiB
Rust
Raw Normal View History

Don't compute optimized PointerKind for unoptimized builds This saves us both the Freeze/Unpin queries, and avoids placing noalias attributes, which have a compile-time impact on LLVM even in optnone builds (due to always_inline functions).
2021-03-19 16:49:51 -05:00
// compile-flags: -O -C no-prepopulate-passes
Revamp codegen tests to check IR quality instead of quantity The current codegen tests only compare IR line counts between similar rust and C programs, the latter getting compiled with clang. That looked like a good idea back then, but actually things like lifetime intrinsics mean that less IR isn't always better, so the metric isn't really helpful. Instead, we can start doing tests that check specific aspects of the generated IR, like attributes or metadata. To do that, we can use LLVM's FileCheck tool which has a number of useful features for such tests. To start off, I created some tests for a few things that were recently added and/or broken.
2015-05-24 11:07:52 -05:00
Avoid loading the whole gdb debug scripts section. This is so LLVM isn't forced to load every byte of it. Also sets the alignment of the load. Adds a test for the debug script section.
2015-09-15 16:22:16 -05:00
#![crate_type = "lib"]
Add codegen test
2023-02-10 17:54:05 -06:00
#![feature(dyn_star)]
Revamp codegen tests to check IR quality instead of quantity The current codegen tests only compare IR line counts between similar rust and C programs, the latter getting compiled with clang. That looked like a good idea back then, but actually things like lifetime intrinsics mean that less IR isn't always better, so the metric isn't really helpful. Instead, we can start doing tests that check specific aspects of the generated IR, like attributes or metadata. To do that, we can use LLVM's FileCheck tool which has a number of useful features for such tests. To start off, I created some tests for a few things that were recently added and/or broken.
2015-05-24 11:07:52 -05:00
test that MaybeUninit<bool> is not noundef
2022-02-06 20:09:21 -06:00
use std::mem::MaybeUninit;
Apply noundef attribute to all scalar types which do not permit raw init Beyond `&`/`&mut`/`Box`, this covers `char`, discriminants, `NonZero*`, etc. All such types currently cause a Miri error if left uninitialized, and an `invalid_value` lint in cases like `mem::uninitialized::<char>()` Note that this _does not_ change whether or not it is UB for `u64` (or other integer types with no invalid values) to be undef.
2022-02-12 00:38:24 -06:00
use std::num::NonZeroU64;
do not mark interior mutable shared refs as dereferenceable
2022-06-20 22:51:15 -05:00
use std::marker::PhantomPinned;
Add more codegen tests
2022-12-30 14:11:30 -06:00
use std::ptr::NonNull;
test that MaybeUninit<bool> is not noundef
2022-02-06 20:09:21 -06:00
Revamp codegen tests to check IR quality instead of quantity The current codegen tests only compare IR line counts between similar rust and C programs, the latter getting compiled with clang. That looked like a good idea back then, but actually things like lifetime intrinsics mean that less IR isn't always better, so the metric isn't really helpful. Instead, we can start doing tests that check specific aspects of the generated IR, like attributes or metadata. To do that, we can use LLVM's FileCheck tool which has a number of useful features for such tests. To start off, I created some tests for a few things that were recently added and/or broken.
2015-05-24 11:07:52 -05:00
pub struct S {
rustc_trans: compute better align/dereferenceable attributes from pointees.
2017-10-03 02:45:07 -05:00
_field: [i32; 8],
Revamp codegen tests to check IR quality instead of quantity The current codegen tests only compare IR line counts between similar rust and C programs, the latter getting compiled with clang. That looked like a good idea back then, but actually things like lifetime intrinsics mean that less IR isn't always better, so the metric isn't really helpful. Instead, we can start doing tests that check specific aspects of the generated IR, like attributes or metadata. To do that, we can use LLVM's FileCheck tool which has a number of useful features for such tests. To start off, I created some tests for a few things that were recently added and/or broken.
2015-05-24 11:07:52 -05:00
}
pub struct UnsafeInner {
_field: std::cell::UnsafeCell<i16>,
}
do not mark interior mutable shared refs as dereferenceable
2022-06-20 22:51:15 -05:00
pub struct NotUnpin {
_field: i32,
_marker: PhantomPinned,
}
Apply noundef attribute to all scalar types which do not permit raw init Beyond `&`/`&mut`/`Box`, this covers `char`, discriminants, `NonZero*`, etc. All such types currently cause a Miri error if left uninitialized, and an `invalid_value` lint in cases like `mem::uninitialized::<char>()` Note that this _does not_ change whether or not it is UB for `u64` (or other integer types with no invalid values) to be undef.
2022-02-12 00:38:24 -06:00
pub enum MyBool {
True,
False,
}
Apply noundef attribute to &T, &mut T, Box<T>, bool This doesn't handle `char` because it's a bit awkward to distinguish it from u32 at this point in codegen. Note that for some types (like `&Struct` and `&mut Struct`), we already apply `dereferenceable`, which implies `noundef`, so the IR does not change.
2022-02-05 00:00:37 -06:00
// CHECK: noundef zeroext i1 @boolean(i1 noundef zeroext %x)
Revamp codegen tests to check IR quality instead of quantity The current codegen tests only compare IR line counts between similar rust and C programs, the latter getting compiled with clang. That looked like a good idea back then, but actually things like lifetime intrinsics mean that less IR isn't always better, so the metric isn't really helpful. Instead, we can start doing tests that check specific aspects of the generated IR, like attributes or metadata. To do that, we can use LLVM's FileCheck tool which has a number of useful features for such tests. To start off, I created some tests for a few things that were recently added and/or broken.
2015-05-24 11:07:52 -05:00
#[no_mangle]
pub fn boolean(x: bool) -> bool {
x
}
test that MaybeUninit<bool> is not noundef
2022-02-06 20:09:21 -06:00
// CHECK: i8 @maybeuninit_boolean(i8 %x)
#[no_mangle]
pub fn maybeuninit_boolean(x: MaybeUninit<bool>) -> MaybeUninit<bool> {
x
}
Apply noundef attribute to all scalar types which do not permit raw init Beyond `&`/`&mut`/`Box`, this covers `char`, discriminants, `NonZero*`, etc. All such types currently cause a Miri error if left uninitialized, and an `invalid_value` lint in cases like `mem::uninitialized::<char>()` Note that this _does not_ change whether or not it is UB for `u64` (or other integer types with no invalid values) to be undef.
2022-02-12 00:38:24 -06:00
// CHECK: noundef zeroext i1 @enum_bool(i1 noundef zeroext %x)
#[no_mangle]
pub fn enum_bool(x: MyBool) -> MyBool {
x
}
// CHECK: i8 @maybeuninit_enum_bool(i8 %x)
#[no_mangle]
pub fn maybeuninit_enum_bool(x: MaybeUninit<MyBool>) -> MaybeUninit<MyBool> {
x
}
// CHECK: noundef i32 @char(i32 noundef %x)
#[no_mangle]
pub fn char(x: char) -> char {
x
}
// CHECK: i32 @maybeuninit_char(i32 %x)
#[no_mangle]
pub fn maybeuninit_char(x: MaybeUninit<char>) -> MaybeUninit<char> {
x
}
Put `noundef` on all scalars that don't allow uninit Previously, it was only put on scalars with range validity invariants like bool, was uninit was obviously invalid for those. Since then, we have normatively declared all uninit primitives to be undefined behavior and can therefore put `noundef` on them. The remaining concern was the `mem::uninitialized` function, which cause quite a lot of UB in the older parts of the ecosystem. This function now doesn't return uninit values anymore, making users of it safe from this change. The only real sources of UB where people could encounter uninit primitives are `MaybeUninit::uninit().assume_init()`, which has always be clear in the docs about being UB and from heap allocations (like reading from the spare capacity of a vec. This is hopefully rare enough to not break anything.
2022-12-30 08:55:05 -06:00
// CHECK: noundef i64 @int(i64 noundef %x)
Apply noundef attribute to all scalar types which do not permit raw init Beyond `&`/`&mut`/`Box`, this covers `char`, discriminants, `NonZero*`, etc. All such types currently cause a Miri error if left uninitialized, and an `invalid_value` lint in cases like `mem::uninitialized::<char>()` Note that this _does not_ change whether or not it is UB for `u64` (or other integer types with no invalid values) to be undef.
2022-02-12 00:38:24 -06:00
#[no_mangle]
pub fn int(x: u64) -> u64 {
x
}
// CHECK: noundef i64 @nonzero_int(i64 noundef %x)
#[no_mangle]
pub fn nonzero_int(x: NonZeroU64) -> NonZeroU64 {
x
}
Put `noundef` on all scalars that don't allow uninit Previously, it was only put on scalars with range validity invariants like bool, was uninit was obviously invalid for those. Since then, we have normatively declared all uninit primitives to be undefined behavior and can therefore put `noundef` on them. The remaining concern was the `mem::uninitialized` function, which cause quite a lot of UB in the older parts of the ecosystem. This function now doesn't return uninit values anymore, making users of it safe from this change. The only real sources of UB where people could encounter uninit primitives are `MaybeUninit::uninit().assume_init()`, which has always be clear in the docs about being UB and from heap allocations (like reading from the spare capacity of a vec. This is hopefully rare enough to not break anything.
2022-12-30 08:55:05 -06:00
// CHECK: noundef i64 @option_nonzero_int(i64 noundef %x)
Apply noundef attribute to all scalar types which do not permit raw init Beyond `&`/`&mut`/`Box`, this covers `char`, discriminants, `NonZero*`, etc. All such types currently cause a Miri error if left uninitialized, and an `invalid_value` lint in cases like `mem::uninitialized::<char>()` Note that this _does not_ change whether or not it is UB for `u64` (or other integer types with no invalid values) to be undef.
2022-02-12 00:38:24 -06:00
#[no_mangle]
pub fn option_nonzero_int(x: Option<NonZeroU64>) -> Option<NonZeroU64> {
x
}
CHECK only for opaque ptr
2023-07-27 16:44:13 -05:00
// CHECK: @readonly_borrow(ptr noalias noundef readonly align 4 dereferenceable(4) %_1)
Revamp codegen tests to check IR quality instead of quantity The current codegen tests only compare IR line counts between similar rust and C programs, the latter getting compiled with clang. That looked like a good idea back then, but actually things like lifetime intrinsics mean that less IR isn't always better, so the metric isn't really helpful. Instead, we can start doing tests that check specific aspects of the generated IR, like attributes or metadata. To do that, we can use LLVM's FileCheck tool which has a number of useful features for such tests. To start off, I created some tests for a few things that were recently added and/or broken.
2015-05-24 11:07:52 -05:00
// FIXME #25759 This should also have `nocapture`
#[no_mangle]
pub fn readonly_borrow(_: &i32) {
}
CHECK only for opaque ptr
2023-07-27 16:44:13 -05:00
// CHECK: noundef align 4 dereferenceable(4) ptr @readonly_borrow_ret()
make &mut !Unpin not dereferenceable See https://github.com/rust-lang/unsafe-code-guidelines/issues/381 for discussion.
2022-12-27 05:46:08 -06:00
#[no_mangle]
pub fn readonly_borrow_ret() -> &'static i32 {
loop {}
}
CHECK only for opaque ptr
2023-07-27 16:44:13 -05:00
// CHECK: @static_borrow(ptr noalias noundef readonly align 4 dereferenceable(4) %_1)
Revamp codegen tests to check IR quality instead of quantity The current codegen tests only compare IR line counts between similar rust and C programs, the latter getting compiled with clang. That looked like a good idea back then, but actually things like lifetime intrinsics mean that less IR isn't always better, so the metric isn't really helpful. Instead, we can start doing tests that check specific aspects of the generated IR, like attributes or metadata. To do that, we can use LLVM's FileCheck tool which has a number of useful features for such tests. To start off, I created some tests for a few things that were recently added and/or broken.
2015-05-24 11:07:52 -05:00
// static borrow may be captured
#[no_mangle]
pub fn static_borrow(_: &'static i32) {
}
CHECK only for opaque ptr
2023-07-27 16:44:13 -05:00
// CHECK: @named_borrow(ptr noalias noundef readonly align 4 dereferenceable(4) %_1)
Revamp codegen tests to check IR quality instead of quantity The current codegen tests only compare IR line counts between similar rust and C programs, the latter getting compiled with clang. That looked like a good idea back then, but actually things like lifetime intrinsics mean that less IR isn't always better, so the metric isn't really helpful. Instead, we can start doing tests that check specific aspects of the generated IR, like attributes or metadata. To do that, we can use LLVM's FileCheck tool which has a number of useful features for such tests. To start off, I created some tests for a few things that were recently added and/or broken.
2015-05-24 11:07:52 -05:00
// borrow with named lifetime may be captured
#[no_mangle]
pub fn named_borrow<'r>(_: &'r i32) {
}
CHECK only for opaque ptr
2023-07-27 16:44:13 -05:00
// CHECK: @unsafe_borrow(ptr noundef nonnull align 2 %_1)
Revamp codegen tests to check IR quality instead of quantity The current codegen tests only compare IR line counts between similar rust and C programs, the latter getting compiled with clang. That looked like a good idea back then, but actually things like lifetime intrinsics mean that less IR isn't always better, so the metric isn't really helpful. Instead, we can start doing tests that check specific aspects of the generated IR, like attributes or metadata. To do that, we can use LLVM's FileCheck tool which has a number of useful features for such tests. To start off, I created some tests for a few things that were recently added and/or broken.
2015-05-24 11:07:52 -05:00
// unsafe interior means this isn't actually readonly and there may be aliases ...
#[no_mangle]
pub fn unsafe_borrow(_: &UnsafeInner) {
}
CHECK only for opaque ptr
2023-07-27 16:44:13 -05:00
// CHECK: @mutable_unsafe_borrow(ptr noalias noundef align 2 dereferenceable(2) %_1)
Revamp codegen tests to check IR quality instead of quantity The current codegen tests only compare IR line counts between similar rust and C programs, the latter getting compiled with clang. That looked like a good idea back then, but actually things like lifetime intrinsics mean that less IR isn't always better, so the metric isn't really helpful. Instead, we can start doing tests that check specific aspects of the generated IR, like attributes or metadata. To do that, we can use LLVM's FileCheck tool which has a number of useful features for such tests. To start off, I created some tests for a few things that were recently added and/or broken.
2015-05-24 11:07:52 -05:00
// ... unless this is a mutable borrow, those never alias
#[no_mangle]
pub fn mutable_unsafe_borrow(_: &mut UnsafeInner) {
}
CHECK only for opaque ptr
2023-07-27 16:44:13 -05:00
// CHECK: @mutable_borrow(ptr noalias noundef align 4 dereferenceable(4) %_1)
Revamp codegen tests to check IR quality instead of quantity The current codegen tests only compare IR line counts between similar rust and C programs, the latter getting compiled with clang. That looked like a good idea back then, but actually things like lifetime intrinsics mean that less IR isn't always better, so the metric isn't really helpful. Instead, we can start doing tests that check specific aspects of the generated IR, like attributes or metadata. To do that, we can use LLVM's FileCheck tool which has a number of useful features for such tests. To start off, I created some tests for a few things that were recently added and/or broken.
2015-05-24 11:07:52 -05:00
// FIXME #25759 This should also have `nocapture`
#[no_mangle]
pub fn mutable_borrow(_: &mut i32) {
}
CHECK only for opaque ptr
2023-07-27 16:44:13 -05:00
// CHECK: noundef align 4 dereferenceable(4) ptr @mutable_borrow_ret()
make &mut !Unpin not dereferenceable See https://github.com/rust-lang/unsafe-code-guidelines/issues/381 for discussion.
2022-12-27 05:46:08 -06:00
#[no_mangle]
pub fn mutable_borrow_ret() -> &'static mut i32 {
loop {}
}
do not mark interior mutable shared refs as dereferenceable
2022-06-20 22:51:15 -05:00
#[no_mangle]
CHECK only for opaque ptr
2023-07-27 16:44:13 -05:00
// CHECK: @mutable_notunpin_borrow(ptr noundef nonnull align 4 %_1)
do not mark interior mutable shared refs as dereferenceable
2022-06-20 22:51:15 -05:00
// This one is *not* `noalias` because it might be self-referential.
make &mut !Unpin not dereferenceable See https://github.com/rust-lang/unsafe-code-guidelines/issues/381 for discussion.
2022-12-27 05:46:08 -06:00
// It is also not `dereferenceable` due to
// <https://github.com/rust-lang/unsafe-code-guidelines/issues/381>.
do not mark interior mutable shared refs as dereferenceable
2022-06-20 22:51:15 -05:00
pub fn mutable_notunpin_borrow(_: &mut NotUnpin) {
}
CHECK only for opaque ptr
2023-07-27 16:44:13 -05:00
// CHECK: @notunpin_borrow(ptr noalias noundef readonly align 4 dereferenceable(4) %_1)
do not mark interior mutable shared refs as dereferenceable
2022-06-20 22:51:15 -05:00
// But `&NotUnpin` behaves perfectly normal.
#[no_mangle]
pub fn notunpin_borrow(_: &NotUnpin) {
}
CHECK only for opaque ptr
2023-07-27 16:44:13 -05:00
// CHECK: @indirect_struct(ptr noalias nocapture noundef readonly align 4 dereferenceable(32) %_1)
Revamp codegen tests to check IR quality instead of quantity The current codegen tests only compare IR line counts between similar rust and C programs, the latter getting compiled with clang. That looked like a good idea back then, but actually things like lifetime intrinsics mean that less IR isn't always better, so the metric isn't really helpful. Instead, we can start doing tests that check specific aspects of the generated IR, like attributes or metadata. To do that, we can use LLVM's FileCheck tool which has a number of useful features for such tests. To start off, I created some tests for a few things that were recently added and/or broken.
2015-05-24 11:07:52 -05:00
#[no_mangle]
pub fn indirect_struct(_: S) {
}
CHECK only for opaque ptr
2023-07-27 16:44:13 -05:00
// CHECK: @borrowed_struct(ptr noalias noundef readonly align 4 dereferenceable(32) %_1)
Revamp codegen tests to check IR quality instead of quantity The current codegen tests only compare IR line counts between similar rust and C programs, the latter getting compiled with clang. That looked like a good idea back then, but actually things like lifetime intrinsics mean that less IR isn't always better, so the metric isn't really helpful. Instead, we can start doing tests that check specific aspects of the generated IR, like attributes or metadata. To do that, we can use LLVM's FileCheck tool which has a number of useful features for such tests. To start off, I created some tests for a few things that were recently added and/or broken.
2015-05-24 11:07:52 -05:00
// FIXME #25759 This should also have `nocapture`
#[no_mangle]
pub fn borrowed_struct(_: &S) {
}
CHECK only for opaque ptr
2023-07-27 16:44:13 -05:00
// CHECK: @option_borrow(ptr noalias noundef readonly align 4 dereferenceable_or_null(4) %x)
Add more codegen tests
2022-12-30 14:11:30 -06:00
#[no_mangle]
pub fn option_borrow(x: Option<&i32>) {
}
CHECK only for opaque ptr
2023-07-27 16:44:13 -05:00
// CHECK: @option_borrow_mut(ptr noalias noundef align 4 dereferenceable_or_null(4) %x)
Add more codegen tests
2022-12-30 14:11:30 -06:00
#[no_mangle]
pub fn option_borrow_mut(x: Option<&mut i32>) {
}
CHECK only for opaque ptr
2023-07-27 16:44:13 -05:00
// CHECK: @raw_struct(ptr noundef %_1)
Apply noundef attribute to &T, &mut T, Box<T>, bool This doesn't handle `char` because it's a bit awkward to distinguish it from u32 at this point in codegen. Note that for some types (like `&Struct` and `&mut Struct`), we already apply `dereferenceable`, which implies `noundef`, so the IR does not change.
2022-02-05 00:00:37 -06:00
#[no_mangle]
pub fn raw_struct(_: *const S) {
}
CHECK only for opaque ptr
2023-07-27 16:44:13 -05:00
// CHECK: @raw_option_nonnull_struct(ptr noundef %_1)
Add more codegen tests
2022-12-30 14:11:30 -06:00
#[no_mangle]
pub fn raw_option_nonnull_struct(_: Option<NonNull<S>>) {
}
remove the 'dereferenceable' attribute from Box
2019-11-22 15:04:22 -06:00
// `Box` can get deallocated during execution of the function, so it should
// not get `dereferenceable`.
CHECK only for opaque ptr
2023-07-27 16:44:13 -05:00
// CHECK: noundef nonnull align 4 ptr @_box(ptr noalias noundef nonnull align 4 %x)
Revamp codegen tests to check IR quality instead of quantity The current codegen tests only compare IR line counts between similar rust and C programs, the latter getting compiled with clang. That looked like a good idea back then, but actually things like lifetime intrinsics mean that less IR isn't always better, so the metric isn't really helpful. Instead, we can start doing tests that check specific aspects of the generated IR, like attributes or metadata. To do that, we can use LLVM's FileCheck tool which has a number of useful features for such tests. To start off, I created some tests for a few things that were recently added and/or broken.
2015-05-24 11:07:52 -05:00
#[no_mangle]
pub fn _box(x: Box<i32>) -> Box<i32> {
x
}
CHECK only for opaque ptr
2023-07-27 16:44:13 -05:00
// CHECK: noundef nonnull align 4 ptr @notunpin_box(ptr noundef nonnull align 4 %x)
also do not add noalias on not-Unpin Box
2023-01-02 07:09:01 -06:00
#[no_mangle]
pub fn notunpin_box(x: Box<NotUnpin>) -> Box<NotUnpin> {
x
}
CHECK only for opaque ptr
2023-07-27 16:44:13 -05:00
// CHECK: @struct_return(ptr noalias nocapture noundef sret(%S) align 4 dereferenceable(32){{( %_0)?}})
Revamp codegen tests to check IR quality instead of quantity The current codegen tests only compare IR line counts between similar rust and C programs, the latter getting compiled with clang. That looked like a good idea back then, but actually things like lifetime intrinsics mean that less IR isn't always better, so the metric isn't really helpful. Instead, we can start doing tests that check specific aspects of the generated IR, like attributes or metadata. To do that, we can use LLVM's FileCheck tool which has a number of useful features for such tests. To start off, I created some tests for a few things that were recently added and/or broken.
2015-05-24 11:07:52 -05:00
#[no_mangle]
pub fn struct_return() -> S {
S {
rustc_trans: compute better align/dereferenceable attributes from pointees.
2017-10-03 02:45:07 -05:00
_field: [0, 0, 0, 0, 0, 0, 0, 0]
Revamp codegen tests to check IR quality instead of quantity The current codegen tests only compare IR line counts between similar rust and C programs, the latter getting compiled with clang. That looked like a good idea back then, but actually things like lifetime intrinsics mean that less IR isn't always better, so the metric isn't really helpful. Instead, we can start doing tests that check specific aspects of the generated IR, like attributes or metadata. To do that, we can use LLVM's FileCheck tool which has a number of useful features for such tests. To start off, I created some tests for a few things that were recently added and/or broken.
2015-05-24 11:07:52 -05:00
}
}
Pass fat pointers in two immediate arguments This has a number of advantages compared to creating a copy in memory and passing a pointer. The obvious one is that we don't have to put the data into memory but can keep it in registers. Since we're currently passing a pointer anyway (instead of using e.g. a known offset on the stack, which is what the `byval` attribute would achieve), we only use a single additional register for each fat pointer, but save at least two pointers worth of stack in exchange (sometimes more because more than one copy gets eliminated). On archs that pass arguments on the stack, we save a pointer worth of stack even without considering the omitted copies. Additionally, LLVM can optimize the code a lot better, to a large degree due to the fact that lots of copies are gone or can be optimized away. Additionally, we can now emit attributes like nonnull on the data and/or vtable pointers contained in the fat pointer, potentially allowing for even more optimizations. This results in LLVM passes being about 3-7% faster (depending on the crate), and the resulting code is also a few percent smaller, for example: text data filename 5671479 3941461 before/librustc-d8ace771.so 5447663 3905745 after/librustc-d8ace771.so 1944425 2394024 before/libstd-d8ace771.so 1896769 2387610 after/libstd-d8ace771.so I had to remove a call in the backtrace-debuginfo test, because LLVM can now merge the tails of some blocks when optimizations are turned on, which can't correctly preserve line info. Fixes #22924 Cc #22891 (at least for fat pointers the code is good now)
2015-06-18 16:57:40 -05:00
// Hack to get the correct size for the length part in slices
Put `noundef` on all scalars that don't allow uninit Previously, it was only put on scalars with range validity invariants like bool, was uninit was obviously invalid for those. Since then, we have normatively declared all uninit primitives to be undefined behavior and can therefore put `noundef` on them. The remaining concern was the `mem::uninitialized` function, which cause quite a lot of UB in the older parts of the ecosystem. This function now doesn't return uninit values anymore, making users of it safe from this change. The only real sources of UB where people could encounter uninit primitives are `MaybeUninit::uninit().assume_init()`, which has always be clear in the docs about being UB and from heap allocations (like reading from the spare capacity of a vec. This is hopefully rare enough to not break anything.
2022-12-30 08:55:05 -06:00
// CHECK: @helper([[USIZE:i[0-9]+]] noundef %_1)
Pass fat pointers in two immediate arguments This has a number of advantages compared to creating a copy in memory and passing a pointer. The obvious one is that we don't have to put the data into memory but can keep it in registers. Since we're currently passing a pointer anyway (instead of using e.g. a known offset on the stack, which is what the `byval` attribute would achieve), we only use a single additional register for each fat pointer, but save at least two pointers worth of stack in exchange (sometimes more because more than one copy gets eliminated). On archs that pass arguments on the stack, we save a pointer worth of stack even without considering the omitted copies. Additionally, LLVM can optimize the code a lot better, to a large degree due to the fact that lots of copies are gone or can be optimized away. Additionally, we can now emit attributes like nonnull on the data and/or vtable pointers contained in the fat pointer, potentially allowing for even more optimizations. This results in LLVM passes being about 3-7% faster (depending on the crate), and the resulting code is also a few percent smaller, for example: text data filename 5671479 3941461 before/librustc-d8ace771.so 5447663 3905745 after/librustc-d8ace771.so 1944425 2394024 before/libstd-d8ace771.so 1896769 2387610 after/libstd-d8ace771.so I had to remove a call in the backtrace-debuginfo test, because LLVM can now merge the tails of some blocks when optimizations are turned on, which can't correctly preserve line info. Fixes #22924 Cc #22891 (at least for fat pointers the code is good now)
2015-06-18 16:57:40 -05:00
#[no_mangle]
Update codegen tests.
2017-10-30 12:18:00 -05:00
pub fn helper(_: usize) {
Pass fat pointers in two immediate arguments This has a number of advantages compared to creating a copy in memory and passing a pointer. The obvious one is that we don't have to put the data into memory but can keep it in registers. Since we're currently passing a pointer anyway (instead of using e.g. a known offset on the stack, which is what the `byval` attribute would achieve), we only use a single additional register for each fat pointer, but save at least two pointers worth of stack in exchange (sometimes more because more than one copy gets eliminated). On archs that pass arguments on the stack, we save a pointer worth of stack even without considering the omitted copies. Additionally, LLVM can optimize the code a lot better, to a large degree due to the fact that lots of copies are gone or can be optimized away. Additionally, we can now emit attributes like nonnull on the data and/or vtable pointers contained in the fat pointer, potentially allowing for even more optimizations. This results in LLVM passes being about 3-7% faster (depending on the crate), and the resulting code is also a few percent smaller, for example: text data filename 5671479 3941461 before/librustc-d8ace771.so 5447663 3905745 after/librustc-d8ace771.so 1944425 2394024 before/libstd-d8ace771.so 1896769 2387610 after/libstd-d8ace771.so I had to remove a call in the backtrace-debuginfo test, because LLVM can now merge the tails of some blocks when optimizations are turned on, which can't correctly preserve line info. Fixes #22924 Cc #22891 (at least for fat pointers the code is good now)
2015-06-18 16:57:40 -05:00
}
CHECK only for opaque ptr
2023-07-27 16:44:13 -05:00
// CHECK: @slice(ptr noalias noundef nonnull readonly align 1 %_1.0, [[USIZE]] noundef %_1.1)
Pass fat pointers in two immediate arguments This has a number of advantages compared to creating a copy in memory and passing a pointer. The obvious one is that we don't have to put the data into memory but can keep it in registers. Since we're currently passing a pointer anyway (instead of using e.g. a known offset on the stack, which is what the `byval` attribute would achieve), we only use a single additional register for each fat pointer, but save at least two pointers worth of stack in exchange (sometimes more because more than one copy gets eliminated). On archs that pass arguments on the stack, we save a pointer worth of stack even without considering the omitted copies. Additionally, LLVM can optimize the code a lot better, to a large degree due to the fact that lots of copies are gone or can be optimized away. Additionally, we can now emit attributes like nonnull on the data and/or vtable pointers contained in the fat pointer, potentially allowing for even more optimizations. This results in LLVM passes being about 3-7% faster (depending on the crate), and the resulting code is also a few percent smaller, for example: text data filename 5671479 3941461 before/librustc-d8ace771.so 5447663 3905745 after/librustc-d8ace771.so 1944425 2394024 before/libstd-d8ace771.so 1896769 2387610 after/libstd-d8ace771.so I had to remove a call in the backtrace-debuginfo test, because LLVM can now merge the tails of some blocks when optimizations are turned on, which can't correctly preserve line info. Fixes #22924 Cc #22891 (at least for fat pointers the code is good now)
2015-06-18 16:57:40 -05:00
// FIXME #25759 This should also have `nocapture`
#[no_mangle]
Update codegen tests.
2017-10-30 12:18:00 -05:00
pub fn slice(_: &[u8]) {
Pass fat pointers in two immediate arguments This has a number of advantages compared to creating a copy in memory and passing a pointer. The obvious one is that we don't have to put the data into memory but can keep it in registers. Since we're currently passing a pointer anyway (instead of using e.g. a known offset on the stack, which is what the `byval` attribute would achieve), we only use a single additional register for each fat pointer, but save at least two pointers worth of stack in exchange (sometimes more because more than one copy gets eliminated). On archs that pass arguments on the stack, we save a pointer worth of stack even without considering the omitted copies. Additionally, LLVM can optimize the code a lot better, to a large degree due to the fact that lots of copies are gone or can be optimized away. Additionally, we can now emit attributes like nonnull on the data and/or vtable pointers contained in the fat pointer, potentially allowing for even more optimizations. This results in LLVM passes being about 3-7% faster (depending on the crate), and the resulting code is also a few percent smaller, for example: text data filename 5671479 3941461 before/librustc-d8ace771.so 5447663 3905745 after/librustc-d8ace771.so 1944425 2394024 before/libstd-d8ace771.so 1896769 2387610 after/libstd-d8ace771.so I had to remove a call in the backtrace-debuginfo test, because LLVM can now merge the tails of some blocks when optimizations are turned on, which can't correctly preserve line info. Fixes #22924 Cc #22891 (at least for fat pointers the code is good now)
2015-06-18 16:57:40 -05:00
}
CHECK only for opaque ptr
2023-07-27 16:44:13 -05:00
// CHECK: @mutable_slice(ptr noalias noundef nonnull align 1 %_1.0, [[USIZE]] noundef %_1.1)
Pass fat pointers in two immediate arguments This has a number of advantages compared to creating a copy in memory and passing a pointer. The obvious one is that we don't have to put the data into memory but can keep it in registers. Since we're currently passing a pointer anyway (instead of using e.g. a known offset on the stack, which is what the `byval` attribute would achieve), we only use a single additional register for each fat pointer, but save at least two pointers worth of stack in exchange (sometimes more because more than one copy gets eliminated). On archs that pass arguments on the stack, we save a pointer worth of stack even without considering the omitted copies. Additionally, LLVM can optimize the code a lot better, to a large degree due to the fact that lots of copies are gone or can be optimized away. Additionally, we can now emit attributes like nonnull on the data and/or vtable pointers contained in the fat pointer, potentially allowing for even more optimizations. This results in LLVM passes being about 3-7% faster (depending on the crate), and the resulting code is also a few percent smaller, for example: text data filename 5671479 3941461 before/librustc-d8ace771.so 5447663 3905745 after/librustc-d8ace771.so 1944425 2394024 before/libstd-d8ace771.so 1896769 2387610 after/libstd-d8ace771.so I had to remove a call in the backtrace-debuginfo test, because LLVM can now merge the tails of some blocks when optimizations are turned on, which can't correctly preserve line info. Fixes #22924 Cc #22891 (at least for fat pointers the code is good now)
2015-06-18 16:57:40 -05:00
// FIXME #25759 This should also have `nocapture`
#[no_mangle]
Update codegen tests.
2017-10-30 12:18:00 -05:00
pub fn mutable_slice(_: &mut [u8]) {
Pass fat pointers in two immediate arguments This has a number of advantages compared to creating a copy in memory and passing a pointer. The obvious one is that we don't have to put the data into memory but can keep it in registers. Since we're currently passing a pointer anyway (instead of using e.g. a known offset on the stack, which is what the `byval` attribute would achieve), we only use a single additional register for each fat pointer, but save at least two pointers worth of stack in exchange (sometimes more because more than one copy gets eliminated). On archs that pass arguments on the stack, we save a pointer worth of stack even without considering the omitted copies. Additionally, LLVM can optimize the code a lot better, to a large degree due to the fact that lots of copies are gone or can be optimized away. Additionally, we can now emit attributes like nonnull on the data and/or vtable pointers contained in the fat pointer, potentially allowing for even more optimizations. This results in LLVM passes being about 3-7% faster (depending on the crate), and the resulting code is also a few percent smaller, for example: text data filename 5671479 3941461 before/librustc-d8ace771.so 5447663 3905745 after/librustc-d8ace771.so 1944425 2394024 before/libstd-d8ace771.so 1896769 2387610 after/libstd-d8ace771.so I had to remove a call in the backtrace-debuginfo test, because LLVM can now merge the tails of some blocks when optimizations are turned on, which can't correctly preserve line info. Fixes #22924 Cc #22891 (at least for fat pointers the code is good now)
2015-06-18 16:57:40 -05:00
}
CHECK only for opaque ptr
2023-07-27 16:44:13 -05:00
// CHECK: @unsafe_slice(ptr noundef nonnull align 2 %_1.0, [[USIZE]] noundef %_1.1)
Pass fat pointers in two immediate arguments This has a number of advantages compared to creating a copy in memory and passing a pointer. The obvious one is that we don't have to put the data into memory but can keep it in registers. Since we're currently passing a pointer anyway (instead of using e.g. a known offset on the stack, which is what the `byval` attribute would achieve), we only use a single additional register for each fat pointer, but save at least two pointers worth of stack in exchange (sometimes more because more than one copy gets eliminated). On archs that pass arguments on the stack, we save a pointer worth of stack even without considering the omitted copies. Additionally, LLVM can optimize the code a lot better, to a large degree due to the fact that lots of copies are gone or can be optimized away. Additionally, we can now emit attributes like nonnull on the data and/or vtable pointers contained in the fat pointer, potentially allowing for even more optimizations. This results in LLVM passes being about 3-7% faster (depending on the crate), and the resulting code is also a few percent smaller, for example: text data filename 5671479 3941461 before/librustc-d8ace771.so 5447663 3905745 after/librustc-d8ace771.so 1944425 2394024 before/libstd-d8ace771.so 1896769 2387610 after/libstd-d8ace771.so I had to remove a call in the backtrace-debuginfo test, because LLVM can now merge the tails of some blocks when optimizations are turned on, which can't correctly preserve line info. Fixes #22924 Cc #22891 (at least for fat pointers the code is good now)
2015-06-18 16:57:40 -05:00
// unsafe interior means this isn't actually readonly and there may be aliases ...
#[no_mangle]
pub fn unsafe_slice(_: &[UnsafeInner]) {
}
CHECK only for opaque ptr
2023-07-27 16:44:13 -05:00
// CHECK: @raw_slice(ptr noundef %_1.0, [[USIZE]] noundef %_1.1)
Apply noundef attribute to &T, &mut T, Box<T>, bool This doesn't handle `char` because it's a bit awkward to distinguish it from u32 at this point in codegen. Note that for some types (like `&Struct` and `&mut Struct`), we already apply `dereferenceable`, which implies `noundef`, so the IR does not change.
2022-02-05 00:00:37 -06:00
#[no_mangle]
pub fn raw_slice(_: *const [u8]) {
}
CHECK only for opaque ptr
2023-07-27 16:44:13 -05:00
// CHECK: @str(ptr noalias noundef nonnull readonly align 1 %_1.0, [[USIZE]] noundef %_1.1)
Pass fat pointers in two immediate arguments This has a number of advantages compared to creating a copy in memory and passing a pointer. The obvious one is that we don't have to put the data into memory but can keep it in registers. Since we're currently passing a pointer anyway (instead of using e.g. a known offset on the stack, which is what the `byval` attribute would achieve), we only use a single additional register for each fat pointer, but save at least two pointers worth of stack in exchange (sometimes more because more than one copy gets eliminated). On archs that pass arguments on the stack, we save a pointer worth of stack even without considering the omitted copies. Additionally, LLVM can optimize the code a lot better, to a large degree due to the fact that lots of copies are gone or can be optimized away. Additionally, we can now emit attributes like nonnull on the data and/or vtable pointers contained in the fat pointer, potentially allowing for even more optimizations. This results in LLVM passes being about 3-7% faster (depending on the crate), and the resulting code is also a few percent smaller, for example: text data filename 5671479 3941461 before/librustc-d8ace771.so 5447663 3905745 after/librustc-d8ace771.so 1944425 2394024 before/libstd-d8ace771.so 1896769 2387610 after/libstd-d8ace771.so I had to remove a call in the backtrace-debuginfo test, because LLVM can now merge the tails of some blocks when optimizations are turned on, which can't correctly preserve line info. Fixes #22924 Cc #22891 (at least for fat pointers the code is good now)
2015-06-18 16:57:40 -05:00
// FIXME #25759 This should also have `nocapture`
#[no_mangle]
Update codegen tests.
2017-10-30 12:18:00 -05:00
pub fn str(_: &[u8]) {
Pass fat pointers in two immediate arguments This has a number of advantages compared to creating a copy in memory and passing a pointer. The obvious one is that we don't have to put the data into memory but can keep it in registers. Since we're currently passing a pointer anyway (instead of using e.g. a known offset on the stack, which is what the `byval` attribute would achieve), we only use a single additional register for each fat pointer, but save at least two pointers worth of stack in exchange (sometimes more because more than one copy gets eliminated). On archs that pass arguments on the stack, we save a pointer worth of stack even without considering the omitted copies. Additionally, LLVM can optimize the code a lot better, to a large degree due to the fact that lots of copies are gone or can be optimized away. Additionally, we can now emit attributes like nonnull on the data and/or vtable pointers contained in the fat pointer, potentially allowing for even more optimizations. This results in LLVM passes being about 3-7% faster (depending on the crate), and the resulting code is also a few percent smaller, for example: text data filename 5671479 3941461 before/librustc-d8ace771.so 5447663 3905745 after/librustc-d8ace771.so 1944425 2394024 before/libstd-d8ace771.so 1896769 2387610 after/libstd-d8ace771.so I had to remove a call in the backtrace-debuginfo test, because LLVM can now merge the tails of some blocks when optimizations are turned on, which can't correctly preserve line info. Fixes #22924 Cc #22891 (at least for fat pointers the code is good now)
2015-06-18 16:57:40 -05:00
}
CHECK only for opaque ptr
2023-07-27 16:44:13 -05:00
// CHECK: @trait_borrow(ptr noundef nonnull align 1 %_1.0, {{.+}} noalias noundef readonly align {{.*}} dereferenceable({{.*}}) %_1.1)
Pass fat pointers in two immediate arguments This has a number of advantages compared to creating a copy in memory and passing a pointer. The obvious one is that we don't have to put the data into memory but can keep it in registers. Since we're currently passing a pointer anyway (instead of using e.g. a known offset on the stack, which is what the `byval` attribute would achieve), we only use a single additional register for each fat pointer, but save at least two pointers worth of stack in exchange (sometimes more because more than one copy gets eliminated). On archs that pass arguments on the stack, we save a pointer worth of stack even without considering the omitted copies. Additionally, LLVM can optimize the code a lot better, to a large degree due to the fact that lots of copies are gone or can be optimized away. Additionally, we can now emit attributes like nonnull on the data and/or vtable pointers contained in the fat pointer, potentially allowing for even more optimizations. This results in LLVM passes being about 3-7% faster (depending on the crate), and the resulting code is also a few percent smaller, for example: text data filename 5671479 3941461 before/librustc-d8ace771.so 5447663 3905745 after/librustc-d8ace771.so 1944425 2394024 before/libstd-d8ace771.so 1896769 2387610 after/libstd-d8ace771.so I had to remove a call in the backtrace-debuginfo test, because LLVM can now merge the tails of some blocks when optimizations are turned on, which can't correctly preserve line info. Fixes #22924 Cc #22891 (at least for fat pointers the code is good now)
2015-06-18 16:57:40 -05:00
// FIXME #25759 This should also have `nocapture`
#[no_mangle]
Put `noundef` on all scalars that don't allow uninit Previously, it was only put on scalars with range validity invariants like bool, was uninit was obviously invalid for those. Since then, we have normatively declared all uninit primitives to be undefined behavior and can therefore put `noundef` on them. The remaining concern was the `mem::uninitialized` function, which cause quite a lot of UB in the older parts of the ecosystem. This function now doesn't return uninit values anymore, making users of it safe from this change. The only real sources of UB where people could encounter uninit primitives are `MaybeUninit::uninit().assume_init()`, which has always be clear in the docs about being UB and from heap allocations (like reading from the spare capacity of a vec. This is hopefully rare enough to not break anything.
2022-12-30 08:55:05 -06:00
pub fn trait_borrow(_: &dyn Drop) {
Pass fat pointers in two immediate arguments This has a number of advantages compared to creating a copy in memory and passing a pointer. The obvious one is that we don't have to put the data into memory but can keep it in registers. Since we're currently passing a pointer anyway (instead of using e.g. a known offset on the stack, which is what the `byval` attribute would achieve), we only use a single additional register for each fat pointer, but save at least two pointers worth of stack in exchange (sometimes more because more than one copy gets eliminated). On archs that pass arguments on the stack, we save a pointer worth of stack even without considering the omitted copies. Additionally, LLVM can optimize the code a lot better, to a large degree due to the fact that lots of copies are gone or can be optimized away. Additionally, we can now emit attributes like nonnull on the data and/or vtable pointers contained in the fat pointer, potentially allowing for even more optimizations. This results in LLVM passes being about 3-7% faster (depending on the crate), and the resulting code is also a few percent smaller, for example: text data filename 5671479 3941461 before/librustc-d8ace771.so 5447663 3905745 after/librustc-d8ace771.so 1944425 2394024 before/libstd-d8ace771.so 1896769 2387610 after/libstd-d8ace771.so I had to remove a call in the backtrace-debuginfo test, because LLVM can now merge the tails of some blocks when optimizations are turned on, which can't correctly preserve line info. Fixes #22924 Cc #22891 (at least for fat pointers the code is good now)
2015-06-18 16:57:40 -05:00
}
CHECK only for opaque ptr
2023-07-27 16:44:13 -05:00
// CHECK: @option_trait_borrow(ptr noundef align 1 %x.0, ptr %x.1)
Add more codegen tests
2022-12-30 14:11:30 -06:00
#[no_mangle]
pub fn option_trait_borrow(x: Option<&dyn Drop>) {
}
CHECK only for opaque ptr
2023-07-27 16:44:13 -05:00
// CHECK: @option_trait_borrow_mut(ptr noundef align 1 %x.0, ptr %x.1)
Add more codegen tests
2022-12-30 14:11:30 -06:00
#[no_mangle]
pub fn option_trait_borrow_mut(x: Option<&mut dyn Drop>) {
}
CHECK only for opaque ptr
2023-07-27 16:44:13 -05:00
// CHECK: @trait_raw(ptr noundef %_1.0, {{.+}} noalias noundef readonly align {{.*}} dereferenceable({{.*}}) %_1.1)
Apply noundef attribute to &T, &mut T, Box<T>, bool This doesn't handle `char` because it's a bit awkward to distinguish it from u32 at this point in codegen. Note that for some types (like `&Struct` and `&mut Struct`), we already apply `dereferenceable`, which implies `noundef`, so the IR does not change.
2022-02-05 00:00:37 -06:00
#[no_mangle]
Put `noundef` on all scalars that don't allow uninit Previously, it was only put on scalars with range validity invariants like bool, was uninit was obviously invalid for those. Since then, we have normatively declared all uninit primitives to be undefined behavior and can therefore put `noundef` on them. The remaining concern was the `mem::uninitialized` function, which cause quite a lot of UB in the older parts of the ecosystem. This function now doesn't return uninit values anymore, making users of it safe from this change. The only real sources of UB where people could encounter uninit primitives are `MaybeUninit::uninit().assume_init()`, which has always be clear in the docs about being UB and from heap allocations (like reading from the spare capacity of a vec. This is hopefully rare enough to not break anything.
2022-12-30 08:55:05 -06:00
pub fn trait_raw(_: *const dyn Drop) {
Apply noundef attribute to &T, &mut T, Box<T>, bool This doesn't handle `char` because it's a bit awkward to distinguish it from u32 at this point in codegen. Note that for some types (like `&Struct` and `&mut Struct`), we already apply `dereferenceable`, which implies `noundef`, so the IR does not change.
2022-02-05 00:00:37 -06:00
}
CHECK only for opaque ptr
2023-07-27 16:44:13 -05:00
// CHECK: @trait_box(ptr noalias noundef nonnull align 1{{( %0)?}}, {{.+}} noalias noundef readonly align {{.*}} dereferenceable({{.*}}){{( %1)?}})
Pass fat pointers in two immediate arguments This has a number of advantages compared to creating a copy in memory and passing a pointer. The obvious one is that we don't have to put the data into memory but can keep it in registers. Since we're currently passing a pointer anyway (instead of using e.g. a known offset on the stack, which is what the `byval` attribute would achieve), we only use a single additional register for each fat pointer, but save at least two pointers worth of stack in exchange (sometimes more because more than one copy gets eliminated). On archs that pass arguments on the stack, we save a pointer worth of stack even without considering the omitted copies. Additionally, LLVM can optimize the code a lot better, to a large degree due to the fact that lots of copies are gone or can be optimized away. Additionally, we can now emit attributes like nonnull on the data and/or vtable pointers contained in the fat pointer, potentially allowing for even more optimizations. This results in LLVM passes being about 3-7% faster (depending on the crate), and the resulting code is also a few percent smaller, for example: text data filename 5671479 3941461 before/librustc-d8ace771.so 5447663 3905745 after/librustc-d8ace771.so 1944425 2394024 before/libstd-d8ace771.so 1896769 2387610 after/libstd-d8ace771.so I had to remove a call in the backtrace-debuginfo test, because LLVM can now merge the tails of some blocks when optimizations are turned on, which can't correctly preserve line info. Fixes #22924 Cc #22891 (at least for fat pointers the code is good now)
2015-06-18 16:57:40 -05:00
#[no_mangle]
also do not add noalias on not-Unpin Box
2023-01-02 07:09:01 -06:00
pub fn trait_box(_: Box<dyn Drop + Unpin>) {
Pass fat pointers in two immediate arguments This has a number of advantages compared to creating a copy in memory and passing a pointer. The obvious one is that we don't have to put the data into memory but can keep it in registers. Since we're currently passing a pointer anyway (instead of using e.g. a known offset on the stack, which is what the `byval` attribute would achieve), we only use a single additional register for each fat pointer, but save at least two pointers worth of stack in exchange (sometimes more because more than one copy gets eliminated). On archs that pass arguments on the stack, we save a pointer worth of stack even without considering the omitted copies. Additionally, LLVM can optimize the code a lot better, to a large degree due to the fact that lots of copies are gone or can be optimized away. Additionally, we can now emit attributes like nonnull on the data and/or vtable pointers contained in the fat pointer, potentially allowing for even more optimizations. This results in LLVM passes being about 3-7% faster (depending on the crate), and the resulting code is also a few percent smaller, for example: text data filename 5671479 3941461 before/librustc-d8ace771.so 5447663 3905745 after/librustc-d8ace771.so 1944425 2394024 before/libstd-d8ace771.so 1896769 2387610 after/libstd-d8ace771.so I had to remove a call in the backtrace-debuginfo test, because LLVM can now merge the tails of some blocks when optimizations are turned on, which can't correctly preserve line info. Fixes #22924 Cc #22891 (at least for fat pointers the code is good now)
2015-06-18 16:57:40 -05:00
}
CHECK only for opaque ptr
2023-07-27 16:44:13 -05:00
// CHECK: { ptr, ptr } @trait_option(ptr noalias noundef align 1 %x.0, ptr %x.1)
Allow niche-filling dataful variants to be represented as a ScalarPair
2018-03-26 09:26:03 -05:00
#[no_mangle]
also do not add noalias on not-Unpin Box
2023-01-02 07:09:01 -06:00
pub fn trait_option(x: Option<Box<dyn Drop + Unpin>>) -> Option<Box<dyn Drop + Unpin>> {
Allow niche-filling dataful variants to be represented as a ScalarPair
2018-03-26 09:26:03 -05:00
x
}
CHECK only for opaque ptr
2023-07-27 16:44:13 -05:00
// CHECK: { ptr, [[USIZE]] } @return_slice(ptr noalias noundef nonnull readonly align 2 %x.0, [[USIZE]] noundef %x.1)
Pass fat pointers in two immediate arguments This has a number of advantages compared to creating a copy in memory and passing a pointer. The obvious one is that we don't have to put the data into memory but can keep it in registers. Since we're currently passing a pointer anyway (instead of using e.g. a known offset on the stack, which is what the `byval` attribute would achieve), we only use a single additional register for each fat pointer, but save at least two pointers worth of stack in exchange (sometimes more because more than one copy gets eliminated). On archs that pass arguments on the stack, we save a pointer worth of stack even without considering the omitted copies. Additionally, LLVM can optimize the code a lot better, to a large degree due to the fact that lots of copies are gone or can be optimized away. Additionally, we can now emit attributes like nonnull on the data and/or vtable pointers contained in the fat pointer, potentially allowing for even more optimizations. This results in LLVM passes being about 3-7% faster (depending on the crate), and the resulting code is also a few percent smaller, for example: text data filename 5671479 3941461 before/librustc-d8ace771.so 5447663 3905745 after/librustc-d8ace771.so 1944425 2394024 before/libstd-d8ace771.so 1896769 2387610 after/libstd-d8ace771.so I had to remove a call in the backtrace-debuginfo test, because LLVM can now merge the tails of some blocks when optimizations are turned on, which can't correctly preserve line info. Fixes #22924 Cc #22891 (at least for fat pointers the code is good now)
2015-06-18 16:57:40 -05:00
#[no_mangle]
Update codegen tests.
2017-10-30 12:18:00 -05:00
pub fn return_slice(x: &[u16]) -> &[u16] {
Pass fat pointers in two immediate arguments This has a number of advantages compared to creating a copy in memory and passing a pointer. The obvious one is that we don't have to put the data into memory but can keep it in registers. Since we're currently passing a pointer anyway (instead of using e.g. a known offset on the stack, which is what the `byval` attribute would achieve), we only use a single additional register for each fat pointer, but save at least two pointers worth of stack in exchange (sometimes more because more than one copy gets eliminated). On archs that pass arguments on the stack, we save a pointer worth of stack even without considering the omitted copies. Additionally, LLVM can optimize the code a lot better, to a large degree due to the fact that lots of copies are gone or can be optimized away. Additionally, we can now emit attributes like nonnull on the data and/or vtable pointers contained in the fat pointer, potentially allowing for even more optimizations. This results in LLVM passes being about 3-7% faster (depending on the crate), and the resulting code is also a few percent smaller, for example: text data filename 5671479 3941461 before/librustc-d8ace771.so 5447663 3905745 after/librustc-d8ace771.so 1944425 2394024 before/libstd-d8ace771.so 1896769 2387610 after/libstd-d8ace771.so I had to remove a call in the backtrace-debuginfo test, because LLVM can now merge the tails of some blocks when optimizations are turned on, which can't correctly preserve line info. Fixes #22924 Cc #22891 (at least for fat pointers the code is good now)
2015-06-18 16:57:40 -05:00
x
}
Apply noundef attribute to all scalar types which do not permit raw init Beyond `&`/`&mut`/`Box`, this covers `char`, discriminants, `NonZero*`, etc. All such types currently cause a Miri error if left uninitialized, and an `invalid_value` lint in cases like `mem::uninitialized::<char>()` Note that this _does not_ change whether or not it is UB for `u64` (or other integer types with no invalid values) to be undef.
2022-02-12 00:38:24 -06:00
// CHECK: { i16, i16 } @enum_id_1(i16 noundef %x.0, i16 %x.1)
Use ScalarPair for tagged enums
2018-03-27 09:44:03 -05:00
#[no_mangle]
pub fn enum_id_1(x: Option<Result<u16, u16>>) -> Option<Result<u16, u16>> {
x
}
Apply noundef attribute to &T, &mut T, Box<T>, bool This doesn't handle `char` because it's a bit awkward to distinguish it from u32 at this point in codegen. Note that for some types (like `&Struct` and `&mut Struct`), we already apply `dereferenceable`, which implies `noundef`, so the IR does not change.
2022-02-05 00:00:37 -06:00
// CHECK: { i8, i8 } @enum_id_2(i1 noundef zeroext %x.0, i8 %x.1)
Use ScalarPair for tagged enums
2018-03-27 09:44:03 -05:00
#[no_mangle]
pub fn enum_id_2(x: Option<u8>) -> Option<u8> {
x
}
Add codegen test
2023-02-10 17:54:05 -06:00
CHECK only for opaque ptr
2023-07-27 16:44:13 -05:00
// CHECK: { ptr, {{.+}} } @dyn_star(ptr noundef %x.0, {{.+}} noalias noundef readonly align {{.*}} dereferenceable({{.*}}) %x.1)
Make dyn* have the same scalar pair ABI as corresponding fat pointer
2023-02-14 21:42:45 -06:00
// Expect an ABI something like `{ {}*, [3 x i64]* }`, but that's hard to match on generically,
// so do like the `trait_box` test and just match on `{{.+}}` for the vtable.
Add codegen test
2023-02-10 17:54:05 -06:00
#[no_mangle]
pub fn dyn_star(x: dyn* Drop) -> dyn* Drop {
x
}
Reference in New Issue Copy Permalink