Performance

Gate B is equivalent-n loops: Buraaq --release versus C++ -O2, clang 22. The same n is in both sources. Unfavorable numbers stay published.

Results

Bench	n	C++ `-O2`	Buraaq	Ratio
nested_loop	1e4²	0.010s	0.002s	0.16×
integer_sum	1e8	0.015s	<0.001s	0.00×
float_saxpy	1e7	0.004s	0.004s	0.97×
numerical_loop	1e8	0.083s	0.084s	1.01×
fib_iter	1e8	0.020s	0.021s	1.02×

Four benches beat or match C++. integer_sum is an Orbit fold of wrapping sum = sum * 3 + i (same n=1e8; n=10 checksum 44281). Worst published: fib_iter 1.02×. n was not reduced.

What we measure

Timed region is only the loop. Setup and print sit outside the clock.
n is kept alive (volatile in C++, buraaq_keepalive_i32 in Buraaq) so the optimizer cannot delete the work.
Buraaq signed add / sub emit LLVM nsw, the same flag clang uses for C++ signed math.
for i in 1..=n parses when n is an identifier (the end of a range is not a struct literal).

How to write a hot loop

Write the obvious while or for. Measure --release before you clever it. --release-fast is -O3 plus thin LTO when you care about the last percent.

fn main() {
    n = 100000000
    mut sum = 0
    mut i = 1
    while i <= n {
        sum = sum * 3 + i
        i = i + 1
    }
    println(sum)
}

Do not drop bounds checks or reach for unsafe to “win” a bench. Gate B was taken in safe Buraaq.

Next: electronics if the loop sits on a board.