Open
Description
I tried this code:
https://godbolt.org/z/4KKGvWs8d
#[derive(Copy, Clone)]
pub enum Foo {
Foo1 = 1,
Foo2 = 2,
Foo4 = 4,
Foo8 = 8,
}
#[no_mangle]
pub fn load(foo: &Foo) -> u32 {
*foo as u32
}
I expected to see this happen: The metadata attatched to the load
should specify that the loaded value is either 1, 2, 4 or 8:
define noundef range(i32 1, 8) i32 @load(ptr noalias nocapture noundef readonly align 1 dereferenceable(1) %foo) unnamed_addr #0 {
%0 = load i8, ptr %foo, align 1, !range !2, !noundef !3
%_0 = zext nneg i8 %0 to i32
ret i32 %_0
}
!2 = !{
i8 1, i8 3, ; 1 or 2
i8 4, i8 5, ; 4
i8 8, i8 9, ; 8
}
Instead, this happened: the range metadata gives a still correct, but less accurate range of [1,9)
:
define noundef range(i32 1, 8) i32 @load(ptr noalias nocapture noundef readonly align 1 dereferenceable(1) %foo) unnamed_addr #0 {
%0 = load i8, ptr %foo, align 1, !range !2, !noundef !3
%_0 = zext nneg i8 %0 to i32
ret i32 %_0
}
!2 = !{i8 1, i8 9}
Similarly, the range
attribute for the return value of load
(and the range
attribute for a value of type Foo
if it is passed by value) is also less precise than it could be. But it appears from reading the LLVM LangRef that range
attributes can only specify a single range, rather than a union of disjoint ranges like llvm.range
metadata can.
Meta
rustc --version --verbose
:
rustc 1.85.0-nightly (5e1440ae5 2024-12-01)
binary: rustc
commit-hash: 5e1440ae514d98ddfcbf1607acb64d41e07ef616
commit-date: 2024-12-01
host: x86_64-unknown-linux-gnu
release: 1.85.0-nightly
LLVM version: 19.1.4
Backtrace
<backtrace>
Metadata
Metadata
Assignees
Labels
Area: Code generation parts specific to LLVM. Both correctness bugs and optimization-related issues.Area: Code generationCategory: An issue highlighting optimization opportunities or PRs implementing suchIssue: Problems and improvements with respect to performance of generated code.Relevant to the compiler team, which will review and decide on the PR/issue.