Re: [PATCH 2/8] wined3d: Avoid potential division by 0 when computing fog scale.

[Henri Verbeet]

On 16 June 2015 at 22:45, Matteo Bruni <mbruni(a)codeweavers.com> wrote:
> get_fog_start_end(context, state, &start, &end);
> -    scale = 1.0f / (end - start);
> +    scale = end == start ? INFINITY : 1.0f / (end - start);

Do we really need this? It almost feels nicer to just pass "end -
start" to the shader and handle the division there, or perhaps even
just pass "start".

[PATCH 2/8] wined3d: Avoid potential division by 0 when computing fog scale.

[Stefan Dösinger]

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

Am 2015-06-17 um 11:36 schrieb Henri Verbeet:
> Do we really need this? It almost feels nicer to just pass "end - 
> start" to the shader and handle the division there, or perhaps even
> just pass "start".
> I like the idea of calculating this once per draw rather than once per
> fragment. I did notice a performance difference when I wrote
> ba61730daa6a614ef07128a69a0bbca819e716b4, although it was only visible
> in a test application I wrote for World of Tanks performance, not in a
> real world game.

I think Matteo's patch comment should be put inside the code to
explain why the explicit check and INFINITE are there.


-----BEGIN PGP SIGNATURE-----
Version: GnuPG v2
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=gosR
-----END PGP SIGNATURE-----

[PATCH 2/8] wined3d: Avoid potential division by 0 when computing fog scale.

[Matteo Bruni]

2015-06-17 11:57 GMT+02:00 Stefan Dösinger <stefandoesinger(a)gmail.com>:
> -----BEGIN PGP SIGNED MESSAGE-----
> Hash: SHA1
>
> Am 2015-06-17 um 11:36 schrieb Henri Verbeet:
> > Do we really need this? It almost feels nicer to just pass "end -
> > start" to the shader and handle the division there, or perhaps even
> > just pass "start".
> > I like the idea of calculating this once per draw rather than once per
> > fragment. I did notice a performance difference when I wrote
> > ba61730daa6a614ef07128a69a0bbca819e716b4, although it was only visible
> > in a test application I wrote for World of Tanks performance, not in a
> > real world game.

Yeah, that was the reasoning for passing scale instead of start to the
shader (actually I used start in my initial implementation of the fog
emulation).
> I think Matteo's patch comment should be put inside the code to
> explain why the explicit check and INFINITE are there.

That's a good idea.

I don't know of any specific application requiring this right now. I
guess I can defer resending the patch in any form until (and if) we
get any indication that this is a problem.

[PATCH 2/8] wined3d: Avoid potential division by 0 when computing fog scale.

[Stefan Dösinger]

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

Am 2015-06-17 um 15:35 schrieb Matteo Bruni:
> I don't know of any specific application requiring this right now. 
> I guess I can defer resending the patch in any form until (and if) 
> we get any indication that this is a problem.
> The game in question is called Risk of Rain. The bug applies to any
> game made with Game Maker from YoYo Games. Unfortunately that problem
> was never reported to us, it was reported to Nvidia and I got the
> information from them.

Note that there's a demo of the game that is not affected by the
problem. It seems an update to the game later introduced the issue.
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v2
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=+5+n
-----END PGP SIGNATURE-----