Let's make rint() work correctly once and for all.

[divVerent]

This new rint() now returns always matches output sign to input.

It should match C23's round() perfectly, but is explicit about the handling of negative zeroes in input and output and still keeps the original Quake logic for safety.

[divVerent]

@uis246 Comparison for test values with all deltas from C's rint function highlighted:

input	rintf	roundf	rint_q1	rint_dpold	rint_dpnew
-3e+09	-3e+09	-3e+09	! -2.14748365e+09 !	-3e+09	-3e+09
-2.14748365e+09	-2.14748365e+09	-2.14748365e+09	-2.14748365e+09	-2.14748365e+09	-2.14748365e+09
-16777218	-16777218	-16777218	-16777218	-16777218	-16777218
-16777216	-16777216	-16777216	-16777216	-16777216	-16777216
-16777215	-16777215	-16777215	-16777215	-16777215	-16777215
-16777214	-16777214	-16777214	-16777214	-16777214	-16777214
-8388609	-8388609	-8388609	-8388609	-8388609	-8388609
-8388608	-8388608	-8388608	-8388608	-8388608	-8388608
-8388607.5	-8388608	-8388608	-8388608	-8388608	-8388608
-8388607	-8388607	-8388607	-8388607	-8388607	-8388607
-1.6	-2	-2	-2	-2	-2
-1.5	-2	-2	-2	-2	-2
-1.4	-1	-1	-1	-1	-1
-1	-1	-1	-1	-1	-1
-0.6	-1	-1	-1	-1	-1
-0.5	-0	! -1 !	! -1 !	! -1 !	! -1 !
-0.4	-0	-0	! 0 !	-0	-0
-0	-0	-0	! 0 !	-0	-0
0	0	0	0	! -0 !	0
0.4	0	0	0	0	0
0.5	0	! 1 !	! 1 !	! 1 !	! 1 !
0.6	1	1	1	1	1
1	1	1	1	1	1
1.4	1	1	1	1	1
1.5	2	2	2	2	2
1.6	2	2	2	2	2
8388607	8388607	8388607	8388607	8388607	8388607
8388607.5	8388608	8388608	8388608	8388608	8388608
8388608	8388608	8388608	8388608	8388608	8388608
8388609	8388609	8388609	8388609	8388609	8388609
16777214	16777214	16777214	16777214	16777214	16777214
16777215	16777215	16777215	16777215	16777215	16777215
16777216	16777216	16777216	16777216	16777216	16777216
16777218	16777218	16777218	16777218	16777218	16777218
2.14748365e+09	2.14748365e+09	2.14748365e+09	! -2.14748365e+09 !	2.14748365e+09	2.14748365e+09
3e+09	3e+09	3e+09	! -2.14748365e+09 !	3e+09	3e+09

It actually matches roundf exactly on x86, but is explicit about things not cleared up in the standard. It in particular fixes the negative zero DP outputs when fed exactly 0. Let's consider this the best possible compromise between Quake and ISO C.

https://godbolt.org/z/z9a59ev96

[divVerent]

BTW, DP's change to rint came here:

commit 757a7dcbb67fcbeac0275a072dd984fd15a96b93
Author: havoc <havoc@d7cf8633-e32d-0410-b094-e92efae38249>
Date:   Sat Aug 26 07:02:49 2006 +0000

    CSQC fixes (less broken, still not spec compliant)
    CSQC can now link the same entity multiple times (each time modified differently) as per the CSQC spec (this is mainly useful for rendering multi-model players based on one csqc entity, like Quake3's cgame does)
    changed DrawQ functions to automatically set up 2D rendering if previous rendering was of a 3D view (necessary change for proper CSQC support)
    changed QuakeC rint builtin implementation to handle very large values (outside of int range) and possibly run a little bit faster, note that it still rounds toward the nearest integer, away from zero (as intended)
    refactored gun bob, damage kick fade, and stair smoothing code to eliminate a 'multiple fade steps per frame' bug
    renamed the misnamed VM_ftoi function to VM_ftoe, in accordance with the menu QC builtin ftoe
    
    
    git-svn-id: svn://svn.icculus.org/twilight/trunk/darkplaces@6568 d7cf8633-e32d-0410-b094-e92efae38249

So the goal of DP's change in 2006 was specifically to avoid the undefined behavior (which on x86 usually returns -2.14748365e+09, and on ARM usually saturates).

[Cloudwalk9]

I'm averse to changing the behavior of this to anything other than stock Quake, which obviously DP's version is not stock apparently, for compatibility reasons, because it involves floating point, and to give an example, last time we had 64 bit QCVM, there were so many codepaths with downcasting bugs that I had to turn it off until they were fixed in normal builds, and it caused so many subtle bugs in map logic.

But if this doesn't break entity logic, timing, etc, and is more correct, sure. Do buttons return to their original position when pushed? Do elevators randomly kill you or just not work? Do doors close? Do monsters seem to act weird(er than they already do in DP)?

[divVerent]

Let's put it behind a default-on prvm_ganeplayfix? Could even also add a setting -1 that replicates original Quake brokenness?Am 25.08.2025 19:38 schrieb Cloudwalk9 ***@***.***>:Cloudwalk9 left a comment (DarkPlacesEngine/DarkPlaces#282) I'm averse to changing the behavior of this to anything other than stock Quake, which obviously DP's version is not stock apparently, for compatibility reasons, because it involves floating point, and last time we had 64 bit QCVM, there were so many codepaths with downcasting bugs that I had to turn it off for now and it caused so many subtle bugs in map logic. But if this doesn't break entity logic, timing, etc, and is more correct, sure. Do buttons return to their original position when pushed? Do elevators randomly kill you or just not work? Do doors close? Do monsters seem to act weird(er than they already do in DP)? —Reply to this email directly, view it on GitHub, or unsubscribe.You are receiving this because you authored the thread.Message ID: ***@***.***>

[divVerent]

Put it behind a gameplayfix, that defaults to 1, and on 0 replicates the old broken Quake behavior.

[divVerent]

Any news here?