[llvm-bugs] [Bug 48965] New: Iteration over arrays until end pointer (C++ vector iterator, Rust iterators) doesn't preserve number of iterations for the optimizer
via llvm-bugs
llvm-bugs at lists.llvm.org
Sat Jan 30 05:00:27 PST 2021
https://bugs.llvm.org/show_bug.cgi?id=48965
Bug ID: 48965
Summary: Iteration over arrays until end pointer (C++ vector
iterator, Rust iterators) doesn't preserve number of
iterations for the optimizer
Product: new-bugs
Version: 11.0
Hardware: PC
OS: Linux
Status: NEW
Severity: enhancement
Priority: P
Component: new bugs
Assignee: unassignedbugs at nondot.org
Reporter: slomo at coaxion.net
CC: htmldeveloper at gmail.com, llvm-bugs at lists.llvm.org
This was originally reported here:
https://github.com/rust-lang/rust/issues/75935
In the end it boils down to the following minimal testcases in C/C++/Rust.
Godbolt links: https://godbolt.org/z/s6reoz and
https://rust.godbolt.org/z/a7f6rj
All of them have in common that the iteration happens not by having a counter
but instead iterating until the end pointer. When additionally counting the
iterations, the maximum (or here: exact) number of iterations does not seem to
be available during optimizations. As such the two assertions are not detected
as dead code and not optimized away. In this specific minimal case, the whole
function should've been possible to optimize away.
The testcase is extremely contrived but it also happens in real code (the Rust
issue contains something closer to my original code) and causes missed
optimization opportunities, and is especially problematic with Rust because of
automatic bounds checks for array indexing that can be impossible to optimize
away because of this and then preventing other optimizations (like
auto-vectorization or loop unrolling) to kick in.
# C
void foo(const uint32_t *y, size_t y_len) {
const uint32_t *y_end = y + y_len;
size_t c = 0;
for (const uint32_t *y_iter = y; y_iter != y_end; y_iter++, c++) {
assert(c < y_len);
}
assert(c == y_len);
}
# C++
void foo(const std::vector<uint32_t>& y) {
size_t c = 0;
for (auto y_iter = y.cbegin(); y_iter != y.cend(); y_iter++, c++) {
assert(c < y.size());
}
assert(c == y.size());
}
# Rust
pub fn foo(y: &[u32]) {
let mut x = 0;
for (c, _y) in y.iter().enumerate() {
assert!(c < y.len());
x = c;
}
assert!(x == y.len());
}
--
You are receiving this mail because:
You are on the CC list for the bug.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-bugs/attachments/20210130/d394233a/attachment.html>
More information about the llvm-bugs
mailing list