Add test for fctiwz

[Tilka]

Removed "FAIL".

[neobrain]

I support @magumagu's suggestion of just looping through all f32 inputs. Any chances of changing your tests to do this?

[hrydgard]

To compare all of them, you either need to generate the 16GB of data (2^32 * sizeof(float)) on the wii and copy them over and compare on the PC, or vice versa, right? Or am I missing something? Can't just add them as expected results in that table.

[neobrain]

You can specify an algorithm that generates all expected results though (i.e. copy-pasting a working Interpreter implementation to the hwtest).

[hrydgard]

That would have to be a known correct soft-fp-implementation, otherwise you're just testing PowerPC behaviour against PowerPC behaviour, no? And how to you test that implementation?

[delroth]

IMO hwtests would make way more sense if they didn't run anything on HW but simply sent back some values that could be compared to another implementation (for example Dolphin).

On Fri, Jun 13, 2014 at 11:04 PM, Henrik Rydgård notifications@github.com wrote:

That would have to be a known correct soft-fp-implementation, otherwise you're just testing PowerPC behaviour against PowerPC behaviour, no? And how to you test that implementation?

— Reply to this email directly or view it on GitHub https://github.com/dolphin-emu/hwtests/pull/4#issuecomment-46059570.

Pierre "delroth" Bourdon delroth@gmail.com Software Engineer @ Zürich, Switzerland http://code.delroth.net/

[neobrain]

@hrydgard Indeed it would need to be that. I still think it should ideally be that way, but I didn't take what you pointed out into consideration first and hence did not see that testing all valid floating point inputs is more effort than I thought.

[magumagu]

It has to be a soft-fp implementation, but it doesn't have to be known correct... you can easily verify it on hardware for any given set of values. (That's what I did for frsqrte.)

[hrydgard]

@magumagu ah, that's right.

My point was really just what neobrain just said, that it's unfair to hold up this pull request with such difficult demands :)

[shuffle2]

To extend what @delroth said, it would be great if hwtests was actually just a "test harness", allowing the real test cases to be sent to a static instance of the test runner executing on the real hw. Then results and new inputs could be passed back and forth. The main difference is focusing on making the real hw a dumb slave, instead of requiring any manual intervention.

I have a wii which has been sitting around for some years, I would be fine with having it running as a test server if such a test runner is made. (Or I'm sure anyone could buy a used unit very cheaply and do the same thing).

On Fri, Jun 13, 2014 at 2:17 PM, Henrik Rydgård notifications@github.com wrote:

@magumagu https://github.com/magumagu ah, that's right.

My point was really just what neobrain just said, that it's unfair to hold up this pull request with such difficult demands :)

— Reply to this email directly or view it on GitHub https://github.com/dolphin-emu/hwtests/pull/4#issuecomment-46060892.

[neobrain]

@shuffle2 I suppose such an "interactive mode" might be a nice idea, however ultimately implementing it in a good way sounds like a lot of work, and I don't know how much effort it would be to keep around some "default" tests (would probably need to store the input values in some script). Ultimately I'm open for such a feature if done properly, however I personally don't see enough need to do the work myself.

[neobrain]

@delroth I'm not sure what exactly it is that you're proposing, but either way it sounds like it's a similar case like @shuffle2's suggestion - maybe a nice thing in theory, but ultimately hwtests suits my personal requirements perfectly fine currently and hence I won't actively work towards a different design in the near future.