Should "Fallback" `mul` operations be implemented as truncating multiplication?

[ajakubowicz-canva]

While implementing WASMmul_u8x16 and mul_i8x16 I added the following two tests (which test parity between WASM and Fallback:

test_wasm_simd_parity! {
    fn mul_u8x16() {
        |s| -> [u8; 16] {
            let a = u8x16::from_slice(s, &[0,0,1,1,2,4,8,16,32,64,128,255,255,255,255,255]);
            let b = u8x16::from_slice(s, &[0,0,0, 1,1,2,4,8,16,32,64,128,255,255,255,0]);
            (a * b).into()
        }
    }
}

test_wasm_simd_parity! {
    fn mul_i8x16() {
        |s| -> [i8; 16] {
            let a = i8x16::from_slice(s, &[0,0,127,-0,2,4,8,16,32,64,127,127,127,127,127,127]);
            let b = i8x16::from_slice(s, &[0,0,-128,0,2,4,-8,-16,-32,-64,127,0,1,-1,0,1]);
            (a * b).into()
        }
    }
}

These failed with a panic message: attempt to multiply with overflow from fallback.

How should the Fallback implementation handle those test cases? I've implemented the WASM operations as truncated multiplication. I believe the neon implementation also uses truncating multiplication as Neon's mul_u8x16 operation uses the vmulq_u8 intrinsic.

[LaurenzV]

I think it's fine to document that overflows are basically unspecified behavior (i.e. anything can happen depending on the architecture), and add explicit instructions for wrapping/truncating arithmetic operations.

[raphlinus]

[Edit: Comment was in response to a now deleted comment]

You are correct that it is inappropriate.

In this case, the use of a different language is a major problem. While C++ does not yet have checked arithmetic (C23 does, so it will likely be added later), Rust does. Similarly, the use of exceptions is unacceptable for the problem domain of fearless_simd.

From this, it is clear that you do not understand the problem we are trying to solve, and your suggestion is not helpful. You are banned from this repo and from Linebender projects. Further interaction will be reported. I am sorry to be so blunt, but you did not get the message from an earlier attempt to be more gentle.

Again, I wish you the best in your open source journey, in repos other than the ones I moderate.