[Libre-soc-bugs] [Bug 432] mulld pseudocode overflow calculation incorrect

[bugzilla-daemon at libre-soc.org]

https://bugs.libre-soc.org/show_bug.cgi?id=432

--- Comment #4 from Jacob Lifshay <programmerjake at gmail.com> ---
sec 3.3.9.1 of powerisa v3.1:
If OE=1 then OV and OV32 are set to 1 if the product can-
not be represented in 64 bits.

so, it doesn't explicitly say u64 or i64

however, I did include test cases in power-instruction-analyzer which wouldn't
overflow according to the buggy pseudocode but are still marked as overflowing
in the mulld instruction model:
-1i64 * -0x8000_0000_0000_0000i64
2i64 * -0x8000_0000_0000_0000i64

instruction model code:
https://salsa.debian.org/Kazan-team/power-instruction-analyzer/-/blob/fa9a4d450614adcdda9ec08dca40423492dcdbbd/src/instr_models.rs#L341

pub fn mulldo(inputs: InstructionInput) -> InstructionResult {
    let ra = inputs.ra as i64;
    let rb = inputs.rb as i64;
    let result = ra.wrapping_mul(rb) as u64;
    let overflow = ra.checked_mul(rb).is_none();
    InstructionResult {
        rt: Some(result),
        overflow: Some(OverflowFlags::from_overflow(overflow)),
        ..InstructionResult::default()
    }
}

-- 
You are receiving this mail because:
You are on the CC list for the bug.