I skimmed through some of the related docs about host1x and it seems that the Tegra isn't implementing a VGA with fixed-function 2D acceleration but is a standard general-purpose GPU. The host1x driver seems to be designed to use DMA to push data into buffers managed on the GPU device via normal kernel DRM. Presumably acceleration is had by sending the appropriate programs to the GPU and having them run on the buffers sent to produce an output image. This is using the "3D" engine to provide 2D acceleration as this is a subset of its full functionality, as you pointed out, and what is meant when someone notes that GPUs don't have 2D-only modules anymore.