Repository navigation
step_by() on new ranges is less optimized than the old ranges. #157754
Description
I compiled the following code:
#[unsafe(no_mangle)]
fn a(n: u32, callback: fn(u32)) {
let range = 0..n;
for x in range.step_by(2) {
callback(x);
}
}
#[unsafe(no_mangle)]
fn b(n: u32, callback: fn(u32)) {
let range = std::range::Range { start: 0, end: n }.into_iter();
for x in range.step_by(2) {
callback(x);
}
}Ideally, both functions should compile to the same or similar assembly. Instead, the inner loop of b is clearly worse than that of a:
Assembly code
a:
push rbp
push r14
push rbx
mov ebp, edi
shr ebp
cmp edi, ebp
je .LBB0_3
mov rbx, rsi
sub ebp, edi
xor edi, edi
.LBB0_2:
lea r14d, [rdi + 2]
call rbx
mov edi, r14d
inc ebp
jne .LBB0_2
.LBB0_3:
pop rbx
pop r14
pop rbp
ret
b:
test edi, edi
je .LBB1_5
push rbp
push r14
push rbx
mov rbx, rsi
mov ebp, edi
mov r14d, 2
.LBB1_2:
lea edi, [r14 - 2]
call rbx
test r14d, r14d
je .LBB1_4
cmp r14d, ebp
lea eax, [r14 + 2]
mov r14d, eax
jb .LBB1_2
.LBB1_4:
pop rbx
pop r14
pop rbp
.LBB1_5:
retI believe that this is because step_by has a specialized implementation for the old ranges, but not for the new ranges. See https://github.com/rust-lang/rust/blob/485ec3fbcc12fa14ef6596dabb125ad710499c9e/library/core/src/iter/adapters/step_by.rs
Meta
Reproducible on godbolt with version
rustc 1.98.0-nightly (485ec3fbc 2026-06-10)
binary: rustc
commit-hash: 485ec3fbcc12fa14ef6596dabb125ad710499c9e
commit-date: 2026-06-10
host: x86_64-unknown-linux-gnu
release: 1.98.0-nightly
LLVM version: 22.1.6
Internal compiler ID: nightly
Activity
Rollup merge of rust-lang#159518 - Lfan-ke:fix/step-by-range-iter-spe…
Rollup merge of rust-lang#159518 - Lfan-ke:fix/step-by-range-iter-spe…
Rollup merge of rust-lang#159518 - Lfan-ke:fix/step-by-range-iter-spe…
Rollup merge of rust-lang#159518 - Lfan-ke:fix/step-by-range-iter-spe…
Unrolled build for #159518
Rollup merge of #159518 - Lfan-ke:fix/step-by-range-iter-specializati…