SIGSEGV 11 on ReleaseFast but not Debug

[thuiop]

Zig Version

0.13.0

Steps to Reproduce and Observed Behavior

Compile the following code with -OReleaseFast to see it crashing

const std = @import("std");

fn next_secret(secret: usize) usize {
    var temp: usize = ((secret * 64) ^ secret) % 16777216;
    temp = ((temp / 32) ^ temp) % 16777216;
    temp = ((temp * 2048) ^ temp) % 16777216;
    return temp;
}

fn run(input: [:0]const u8) usize {
    var it = std.mem.splitScalar(u8, input, "\n"[0]);

    var results: [1048576]usize = [_]usize{0} ** 1048576;
    var is_done: [1048576]bool = [_]bool{false} ** 1048576;
    while (it.next()) |row| {
        @memset(&is_done, false);
        var secret: usize = std.fmt.parseInt(usize, row, 10) catch unreachable;
        var prev_price: i5 = @intCast(secret % 10);
        var last_4: u20 = 0;
        var i: usize = 0;
        for (0..2000) |_| {
            secret = next_secret(secret);
            const price: i5 = @intCast(secret % 10);
            const diff: i5 = price - prev_price;
            last_4 = (last_4 << 5) + @as(u5, @bitCast(diff));
            if (i >= 4) {
                if (!is_done[last_4]) {
                    results[last_4] += secret % 10;
                    is_done[last_4] = true;
                }
            } else {
                i += 1;
            }
            prev_price = price;
        }
    }
    return std.mem.max(usize, &results);
}

pub fn main() !void {
    _ = run("1");
}

However, it does not crash when in Debug.

System: Nixos 24.05
Kernel: 6.6.41
Architecture: x86_64

Expected Behavior

I am really unsure of why it crashes; I stumbled upon it during the Advent of Code. What I investigated:

• Originally I allocated results and is_done dynamically, and it worked fine. It also works fine when only is_done is a literal.
• Removing the results[last_4] += secret % 10; line removes the crash. However, debug printing showed that last_4 stays within the bounds of results as I expected.
• The program seems to crash instantly; even a print at the very beginning does not appear.
• I dared not remove more code when submitting the issue just in case, but the crash still happens even when changing next_secret to a dummy function.
• For some reason, replacing all the usize with u32 also removed the crash.

Maybe there is a good reason for this, but I am not seeing it. I thought it might be something related to the size of the allocated array but this does not explain why it only happens in ReleaseFast mode.

Thanks in advance.

[alexrp]

Without having looked at the code in detail: That's... a lot of data to put on the stack. Could well be a stack overflow. I suggest you try to move it to global variables and see if the issue still reproduces.

[nektro]

yeah just the two arrays are using 9,437,184 bytes of stack

[thuiop]

Without having looked at the code in detail: That's... a lot of data to put on the stack. Could well be a stack overflow. I suggest you try to move it to global variables and see if the issue still reproduces.

Indeed, moving them to global variables seems to solve the issue, so I guess it is indeed the size of the array. However, why does it work in Debug mode?