Performance SDL1 vs SDL2

On ARM (Asus Tinker board) I have a Atari emulator called hatari that provides the ability to use either SDL1 or SDL2.
The performance on SDL2 is so bad, that the emulator is unusable, while the SDL1 version is working perfectly.
I have the same emulator running on Windows, OSX and Linux on x86_64, and on those platforms SDL2 works better than the SDL1 version, so I’m pretty sure the issue is SDL and not the emulator itself.

The Asus Tinker board uses a mali GPU, OS is Tinker OS (Debian variant) and the CPU is Cortex 12. Am I missing something in SDL2 that should somehow make it work better? I’m very capable of compiling and installing SDL myself, so I’m not “bound” to use the provides SDL2 packages in TinkerOS.
It seems like the only “driver” for SDL2 is x11 (dynamic),even if configure reports these ones: dummy x11(dynamic) opengl opengl_es1 opengl_es2 vulkan.
Am I doing anything wrong or is this to be expected?