Some Adreno performance numbers in the sample database are incorrect

[Oletus]

At least some Adreno performance numbers in database.json seem to be inversely proportional to the actual performance. Adreno 225 from 2012 for example has a performance score of 2757, similar to mid-range desktop GPUs from a few years ago.

[gkjohnson]

Some of the GPUs are missing passmark data and in that scenario we try to use a score it does have and linearly interpolate between the passmark scores of two GPUs that also contain the secondary benchmark score. It's reasonable in some cases but could yield some odd results. This is the function that happens in.

I just took a look at the Qualcomm Adreno 225 case (which is only present in the notebook check site) and here's what's happening:

• The Adreno GPU has only a "3dMarkIceStorm" score.
• The "Quadro K5100M" and "Intel HD 500" GPUs are the only other two GPUs that have both a passmark score and "3dMarkIceStorm", but the data is pretty odd. Here's the score list:

	3dMarkIceStorm	Passmark
Adreno 225	4969	-
Quadro K5100M	2197	3130
Intel HD 500	22807	359

So as you can see the "Intel HD 500" is scored (a lot) higher than "Quadro K5100M" when comparing the 3dMarkIceStorm scores but lower with Passmark. Linearly interpolating using the values gives us a derived score of "2757". You can see in this case that the source data seems a little wonky if only be because the apparent reported capabilities of either GPU seems reversed depending on which score you're looking at. You could do something like mark "3dMarkIceStorm" as an unreliable data source and not give any result for the Adreno 225 or change the interpolation strategy to use multiple GPUs and weight the scores, reporting no result if we don't enough source data to properly derive a score. Right now it's really simple and just looks for the next lowest and highest scores and interpolates.