Currently, the documentation says:
Note that all current Nvidia devices return 32 for this variable, and all current AMD devices return 64.
(https://github.com/RadeonOpenCompute/ROCm_Documentation/blob/d54ddbd43dcc434211c55451445093e4c6a5bb07/Programming_Guides/Kernel_language.rst#warpsize)
However, this is not the case. On gfx1032 (AMD RDNA 2), the warpSize value in the kernel is (correctly) 32, so "all current AMD devices return 64" is not true.