Specifically, I have used only 2 simple functions rand() and rank (4500x2 array) and I get no GPU acceleration or usage at all. First, I generated a 4500x1 array of rand() numbers and then used another column to rank those random numbers. OpenCL works correctly, as I have downloaded the test spreadsheet “BuildingDesign” and get the proper results with significant GPU usage during test.
I do not know such a list, but:
There once was a much better structured forum on LibO which died (I suppose from missing support by TDF) a while ago. Applaudably there was user @charlie.it who took care of the preservation of at least a fixed state of the contents. There you find the thread http://www.imaccanici.org/en.libreofficeforum.org/node/9119.html which goes back to @oweng, the most valuable contributor ever to that forum and to this one, too. Probably you find still relevant information there.
(Editing:)
Just made a basic test with 1000000 (one million) contiguous cells of one column calculating RAND(). It was about 4 times the speed with OpenCL enabled as compared with not enabled. (My GPU is an older Nvidia onboard type. It may lack capabilities of modern GPU.)
Thanks. LibreO is very slow on my dual Xeon (like 50x) due to single thread and non optimal code (probably poor memory efficiency) and so 4x speed up is not enough, and risk of functions not being accelerated may make it not worth the investment. Only way I can see GPU acceleration worthwhile in LibreO calc is if it could support multiple video cards at the same time. This is probably why the development stopped. I used rand() array nested within rank, so likely rank is the constraint here.