In article <2025Nov10.074620@mips.complang.tuwien.ac.at>,
Anton Ertl wrote:
>dxf writes:
>>I can't think of that many instances where I've used signed division.
>
>When Gforth switched from symmetric to floored for / MOD, etc., there
>were no complaints that I remember.
>
>This lack of complaints indicates that division with negative operands
>either does not occur, or that it occurs, but the users wanted floored
>division and have not used the earlier symmetric /, MOD etc., but have
>used FM/MOD instead.
>
>In the latter case, the users might have started using /, MOD etc. for
>such cases in the meantime, and switching back might lead to
>complaints. We are not going to switch back to find that out.
>
>There have been regular discussions of floored vs. symmetric division,
>and the Forth-83 standardization committee found it important enough
>to specify that /, MOD etc. are floored, so division with negative
>operands is at least on the mind of people.
>
>Forth-94 specified /, MOD etc. as implementation-defined, which has
>led to the regular discussions mentioned above.
>
>Bernd Paysan, who switched Gforth to floored, used to work on
>signal-processing equipment for audio applications. These signals
>tend to be signed numbers (for positive and negative voltages), and
>when you scale them with */, rounding towards 0 leads to flatness when
>the signal passes through 0 that was not there in the original signal.
>Floored division avoids that; rounded division may avoid that, too,
>but the round-towards-even tie-breaker is probably a problem.
>
>>Are
>>there use cases for floored? Sure. But if the argument be 'a positive mod
>>is best' then floored doesn't guarantee that.
>
>If the divisor is positive, the result of floored MOD is >=0.
>
>If you have negative divisors and want results that are >=0, there is
>also something called "Euclidean division"
> where the remainder
>satisfies 0<=remainder<|divisor|.
>
>As long as the divisor is positive, the result of Euclidean division
>and floored division are the same, and the remainders are the same,
>too.
>
>If you look at
>,
>you find that there are some programming languages that specify
>Euclidean modulo, sometimes just that and sometimes in addition to
>symmetric ("truncated" in the table) or rounded; there seems to be no
>language that has both Euclidean and floored, which may indicate that
>the cases where they differ (i.e., negative divisors) occur too rarely
>to provide both. In any case, the variants and the page itself
>indicate that the topic is not just on the minds of the Forth
>community.
The more mindless, the more problems. I think that FM/MOD and SM/REM
are the right decision for Forth. I don't like it that a simple program
ported from ciforth versus gforth or vice versa stops working because
of floored/symmetric.
And then discover that the problem lies in a standard word
that is implementation defined and overlooked.
I agree with the theoretical underpinning of gforth's decision
and ciforth will be compatible in the upcoming version 6.#.#.
MOD that is compatible with FM/MOD.
I have a modulo arithmetic screen +m -m *m /m (modulo is _m).
Negative signed or unsigned modulo is all but nonsensical
for 64 bit.
So my modulo arithmetic is based on a positive modulo,
allowing efficient implementation using SM/REM that is probably
a low level word. (In riscv ciforth it is.)
>
>- anton
Groetjes Albert
--
The Chinese government is satisfied with its military superiority over USA.
The next 5 year plan has as primary goal to advance life expectancy
over 80 years, like Western Europe.
--- SoupGate-Win32 v1.05
* Origin: you cannot sedate... all the things you hate (1:229/2)
|
