1

shr(x,n) and lshr(x,n) have always handled shift values outside the 0…31 range in a peculiar way: negative values of n are treated like n&31, while values ≥ 32 always return 0 (or 0xffff.ffff for a signed shift).

But now the new >>> operator diverges from lshr by treating n ≥ 32 like n&31, so we get:

 ``` 1 >> 32 = 0 shr(1,32) = 0 1 >>> 32 = 1 lshr(1,32) = 0 ```

edit: same with << which no longer behaves like shl()

P#75744 2020-05-02 13:43 ( Edited 2020-05-02 13:47)

:: Felice

Oof, that's not good. I hope it's just a bug.

@zep
Same thing happens if the shift value is a variable, so it's not just an opcode encoding issue with immediate values.

P#75766 2020-05-03 01:39 ( Edited 2020-05-03 01:41)
:: zep

Thanks @samhocevar, that was a timely catch -- fixed for 0.2.0f

I think it's worth the low risk of breakage to do something better for negative shift values too. To follow suit with the n >= 32 behaviour I suppose shl(x, -n) should give the same result as lshr(x, n) and vice versa, as Lua 5.3 does. It's not the best for porting / transpiling to non-Lua languages, but oh well.

P#75772 2020-05-03 10:25
:: Felice
1

I also like the idea of negative shifts Just Working™, as it makes a lot of stuff work without needing any ugly branching.

P#75775 2020-05-03 18:03
:: zep

extra note: negative shift values round down, same as rotl, rotr. so when n < 0:

 ```shl(x, n) == lshr(x, -(n\1)) shr(x, n) == lshl(x, -(n\1)) lshr(x, n) == lshl(x, -(n\1)) ```
P#75796 2020-05-04 12:35
:: Felice

@zep

That makes sense. Intuitively I'd expect the fractional bits to be ignored, effectively the same as rounding down.

P#75856 2020-05-05 16:46

@zep I don’t think I understand the new 0.2.0f logic wrt. overflows:

 ```?7 << 256 0 ?7 << -256 7 ?7 >> 256 0 ?7 >> -256 0 ?7 >>> 256 7 ?7 >>> -256 0 ```

(also they can now crash/freeze as reported in https://www.lexaloffle.com/bbs/?tid=37768)

P#75898 2020-05-05 22:42
:: zep

WHAT

ok, 0.2.0f doesn't exist. It's correct in 0.2.0g which is up now.

P#75956 2020-05-06 12:27