[script]Pitch broken for REAL now.

[Graf Zahl]

Exactly. If this comes up again, it's a handy tool to do a quick check.

[Graf Zahl]

And here's some output I created on an iPad:
(No, that's not ZDoom running, I just copied the test stuff into an App I am currently working on)

Code: Select all

Angle = -900.00000: xs_RoundToInt = 80000000, unsigned cast = 00000000, signed cast = 80000000
Angle = -855.00000: xs_RoundToInt = a0000000, unsigned cast = 00000000, signed cast = 80000000
Angle = -810.00000: xs_RoundToInt = c0000000, unsigned cast = 00000000, signed cast = 80000000
Angle = -765.00000: xs_RoundToInt = e0000000, unsigned cast = 00000000, signed cast = 80000000
Angle = -720.00000: xs_RoundToInt = 00000000, unsigned cast = 00000000, signed cast = 80000000
Angle = -675.00000: xs_RoundToInt = 20000000, unsigned cast = 00000000, signed cast = 80000000
Angle = -630.00000: xs_RoundToInt = 40000000, unsigned cast = 00000000, signed cast = 80000000
Angle = -585.00000: xs_RoundToInt = 60000000, unsigned cast = 00000000, signed cast = 80000000
Angle = -540.00000: xs_RoundToInt = 80000000, unsigned cast = 00000000, signed cast = 80000000
Angle = -495.00000: xs_RoundToInt = a0000000, unsigned cast = 00000000, signed cast = 80000000
Angle = -450.00000: xs_RoundToInt = c0000000, unsigned cast = 00000000, signed cast = 80000000
Angle = -405.00000: xs_RoundToInt = e0000000, unsigned cast = 00000000, signed cast = 80000000
Angle = -360.00000: xs_RoundToInt = 00000000, unsigned cast = 00000000, signed cast = 80000000
Angle = -315.00000: xs_RoundToInt = 20000000, unsigned cast = 00000000, signed cast = 80000000
Angle = -270.00000: xs_RoundToInt = 40000000, unsigned cast = 00000000, signed cast = 80000000
Angle = -225.00000: xs_RoundToInt = 60000000, unsigned cast = 00000000, signed cast = 80000000
Angle = -180.00000: xs_RoundToInt = 80000000, unsigned cast = 00000000, signed cast = 80000000
Angle = -135.00000: xs_RoundToInt = a0000000, unsigned cast = 00000000, signed cast = a0000000
Angle = -90.00000: xs_RoundToInt = c0000000, unsigned cast = 00000000, signed cast = c0000000
Angle = -45.00000: xs_RoundToInt = e0000000, unsigned cast = 00000000, signed cast = e0000000
Angle = 0.00000: xs_RoundToInt = 00000000, unsigned cast = 00000000, signed cast = 00000000
Angle = 45.00000: xs_RoundToInt = 20000000, unsigned cast = 20000000, signed cast = 20000000
Angle = 90.00000: xs_RoundToInt = 40000000, unsigned cast = 40000000, signed cast = 40000000
Angle = 135.00000: xs_RoundToInt = 60000000, unsigned cast = 60000000, signed cast = 60000000
Angle = 180.00000: xs_RoundToInt = 80000000, unsigned cast = 80000000, signed cast = 7fffffff
Angle = 225.00000: xs_RoundToInt = a0000000, unsigned cast = a0000000, signed cast = 7fffffff
Angle = 270.00000: xs_RoundToInt = c0000000, unsigned cast = c0000000, signed cast = 7fffffff
Angle = 315.00000: xs_RoundToInt = e0000000, unsigned cast = e0000000, signed cast = 7fffffff
Angle = 360.00000: xs_RoundToInt = 00000000, unsigned cast = ffffffff, signed cast = 7fffffff
Angle = 405.00000: xs_RoundToInt = 20000000, unsigned cast = ffffffff, signed cast = 7fffffff
Angle = 450.00000: xs_RoundToInt = 40000000, unsigned cast = ffffffff, signed cast = 7fffffff
Angle = 495.00000: xs_RoundToInt = 60000000, unsigned cast = ffffffff, signed cast = 7fffffff
Angle = 540.00000: xs_RoundToInt = 80000000, unsigned cast = ffffffff, signed cast = 7fffffff
Angle = 585.00000: xs_RoundToInt = a0000000, unsigned cast = ffffffff, signed cast = 7fffffff
Angle = 630.00000: xs_RoundToInt = c0000000, unsigned cast = ffffffff, signed cast = 7fffffff
Angle = 675.00000: xs_RoundToInt = e0000000, unsigned cast = ffffffff, signed cast = 7fffffff
Angle = 720.00000: xs_RoundToInt = 00000000, unsigned cast = ffffffff, signed cast = 7fffffff
Angle = 765.00000: xs_RoundToInt = 20000000, unsigned cast = ffffffff, signed cast = 7fffffff
Angle = 810.00000: xs_RoundToInt = 40000000, unsigned cast = ffffffff, signed cast = 7fffffff
Angle = 855.00000: xs_RoundToInt = 60000000, unsigned cast = ffffffff, signed cast = 7fffffff

And yes, xs_RoundToInt is the only method that works here, too!

[Graf Zahl]

... which begs the question: If we want reproducable results for float -> int conversions in general because of netgame consistency, I think there's a lot more code that needs to be refactored.
But that's something someone else will have to do...

[Major Cooke]

Just a heads up, this means GZDoom won't compile because bam2rad is missing for now.

[Edward-san]

... which begs the question: If we want reproducable results for float -> int conversions in general because of netgame consistency, I think there's a lot more code that needs to be refactored.
But that's something someone else will have to do...

I believe the problem is more related to the use of intermediate operations with one or more operands in float, then convert the result to int. See what I had to do to fix this bug.

[Graf Zahl]

Just a heads up, this means GZDoom won't compile because bam2rad is missing for now.

I'll get there.

[Graf Zahl]

... which begs the question: If we want reproducable results for float -> int conversions in general because of netgame consistency, I think there's a lot more code that needs to be refactored.
But that's something someone else will have to do...

I believe the problem is more related to the use of intermediate operations with one or more operands in float, then convert the result to int. See what I had to do to fix this bug.

Yes, but if some overflow gets incorrectly truncated, effects can be similar. Who knows what's really in there...