Documentation recommendation for Driver API

[s]Att nVidia:

Just a suggestion regarding documentation of the Driver API… in addition to the other suggestions I’ve made regarding documentation of the alignment of the formal parameter list…

It would be nice to see documentation covering the fact that using textures referencing linear memory (eg: cuTexRefSetAddress) in conjunction with a memory address allocated by cuMemAllocPitch - do not require pitch compensation when sampling the texture via tex1Dfetch - as it appears to be handled auto-magically. (that is, the texture index doesn’t need it’s row-offset multipled by the pitch - just the ‘width’, if anything)

I’m guessing either the TU handles it transparently, or when gmem is copied to the TU (if that’s how the hardware works, not sure if memory is copied - or if the TU simply references gmem) that the TU doesn’t require the alignment (thus discards the buffer).

In any case, things like this ‘really’ do need to be documented… yet another ~5 hours wasted over such a small/trivial issue that simply wasn’t documented. This all too familiar case is really becoming tiresome.

Neither the Reference Manual nor the Programming Guide refer to this technicality at all - I’d personally expect it to be documented in ‘both’.[/s]

Edit: It would appear I was quite wrong and simply jumped to conclusions that ‘seemed to fit’ my problem… (5 hours of frustration over a single line of code does that…)

You can scrap this - it would appear everything works as documented… except for my own code (which was of course the problem all along).

I was passing in the unaligned size of the memory into cuTexRefSetAddress - and thus it only let me access that much of the texture…

It’s times like these I wish I could delete my own posts. :">