Intrinsics and overflow

[oli-obk]

fixes #151

[solson]

Hmm, but I think we are supposed to always mask the shift width...

Compare the Debug and Release mode MIR and LLVM IR here: https://is.gd/LglRp7

Debug mode uses CheckedShl and an assert MIR statement on the overflow result, while Release uses a plain Shl. Both do a shift width mask in the LLVM IR, but only Debug mode can panic (from the assert).

So I think we need to leave the code for Shl alone and change something for CheckedShl?

[solson]

Oh, I guess this code handles both Shl and CheckedShl. Maybe we just need to adjust the cases where the overflow result is true to make sure the assert is triggered.

[oli-obk]

In debug mode, the mir shows a concrete assert. In the release mode, we still forward and the masking is done by the forwarded to function. The masking was entirely redundant

[solson]

Ohhh. In release mode we still use the overflowing_shl intrinsic function and that does the masking?

[solson]

Huh, TIL the masking for overflowing_shl is done in std code, not via an intrinsic: https://doc.rust-lang.org/stable/src/core/num/mod.rs.html#1092-1094