ATi: fragment programs expose wrong light position?

[tellaman]

this sounds extremely weird to me but, believe it or not, i'm getting light positions in world space on ati boards when using arb_fragment_program and asking for state.light[x].position

i always thought light vectors were transformed by the current mv matrix when calling a glLightfv command, and actually this is what i get on nv boards in the same test application

vertex programs are working fine instead (returns eye space position)

can anyone confirm this?

oh btw i am using the latest omega drivers, based on CATALYST 5.1 Driver Build 6.14.10.6505

thanks for any help you may provide

[Chuck0]

this is something i ran into a few weeks ago as well... but not only that state.light[x].position seems to be wrong within an arb_fp (and correct in an arb_vp) also the modelview matrix in a fragment program seems not to be correct...
i tested this by writing a vertex program that was passing one row of the modelview matrix as color and comparing the results to a fp that did the same... in the fp the modelviewmatrix seemed to be the inverse of the true modelviewmatrix.

[tellaman]

sheez, you are definitely right, i've been trying to output row[0] of texmatrix[0], and ati fps actually output column 0, while vps output row 0

is this the way it's meant to be?!?

on the other side, nv boards show coherent behaviour between vps and fps

[mogumbo]

This is one of multiple fragment program bugs that I reported to ATI months ago. If it isn't fixed by now I don't expect it to ever be fixed. That's why I have an NVidia card now

[tellaman]

* ouch *
that hurts

[mogumbo]

Yes, that may have been a bit harsh. I'll start a flame war if I'm not careful I guess I also switched because I got tired of waiting for GLSL support under Linux. I almost never use arb_vp and arb_fp anymore; GLSL is too much fun.

[tellaman]

so, is there any chance this could be fixed in the near future? how do i report this to ati driver team?

[jwatte]

devrel@ati.com

I've found that they listen better if you include a program that is easy to run to see the problem; ideally in source.

[tellaman]

thanks for your reply
i have already sent em an email saying i may provide some test app if needed
but i think they should definitely be aware of it judging by the gravity of the problem and by the number of people experiencing it