HIP interface

[purepani]

I've successfully compiled the project to HIP*! Addresses #131.
*Currently only compiles to amd, haven't tested it(am currently running one of the demos which is taking a while), and no way to switch between nvidia backend and amd backend for compilation atm, and only on the Python backend right now, so definitely not ready to be merged.

It does compile successfully though!
How long are the demos supposed to take on a reasonably powerful GPU(say a 3080)? I've been leaving demos/d16_2Dtomography.py running on my 7800 XT for ~10-15 minutes now with my GPU pinned at 100%(so I assume it's doing something 😄)

[purepani]

Note: all cpp and cu files were autogenerated using hipify, aside from some small fixes needed to get it to work.
One of those fixes will technically change the numbers slightly: for __frcp_rd and __fsqrt_rd, the corresponding HIP api doesn't exist, so they were switched to __frcp_rn and __fsqrt_rn(both of them are switched from round down to round to nearest even mode).

[purepani]

I left it on overnight and it still didn't finish
Hmm

[AnderBiguri]

@purepani sorry, I am away and with no access to computers now. Thanks for this PR! but indeed, the demos should be quite short, at they should tell you how long they will run for (the algorithms) within a minute or so.

[purepani]

Yeah that demo said it estimated 9 hours(and running it overnight was definitely longer than that too).

Anyway, I might not have time to get back to this PR in particular, but someone else can pick this up as a starting point to finish this.
Or maybe i'll procrastinate on other stuff and try and fix this since it's frustrating me that i got it to compile but not work correctly haha

[AnderBiguri]

That 9h is definetly something wrong, demos should finish fairly quick....