I use the following command to compile the Vivado libraries:

$(GHDL_PATH)/lib/ghdl/vendors/compile-xilinx-vivado.sh --unisim --vhdl2008 --source $(VIVADO_PATH)/data/vhdl/src --output $(VENDOR_LIB_RELDIR)/$(SIMULATOR_NAME)

When using the debug flag -d, I noticed that GHDL compiles each component individually, i.e., GHDL is called each time a new file is compiled. I think this creates a big overhead during the compilation of the libraries.

With Questa or Modelsim, we can do the same but performance (the time it takes to compile everything) is severely affected.

A much better approach is to list all the components in a file and provide this file to a single call of the vcom command, e.g.:

vcom (...) -f <file_with_all_lib_components_listed>.cmf

This approach is really really much faster than calling vcom for each file.

I do use a Questa version that is not supported by Vivado so I have to compile the libraries manually (i.e., not using Vivado) and I can see the big different with both approaches.

In our runners setup, we always compile the Vivado libraries before running simulation. As this takes less than 1 minute with Questa, the time cost is very minimal so we do it all the time. But with GHDL, this is taking from 5 to 10 minutes to finish, depending on the machine specs.

1. Does GHDL support such a feature where we can compile all files in one go (or alternatively we could use xargs)?

2. Would that require a lot of effort to refactor compile-xilinx-vivado.sh to call GHDL per library instead of per component?

For your reference, using this approach in NVC, took the compilation time from few minutes to few seconds:
nickg/nvc#1098