Cairo performance insight in stroke/fill/paint

Question

This is a more general question on how to write code for Cairo with performance in mind.

What gives better performance:

1. Make your lines/area's/... and than stroke/fill/paint
2. Make 1 item eg a box stroke/fill/paint and do the next box till they are all done.

Or is there no significant difference? Or maybe only for paint but not stroke/fill? Does it depend on the number of drawn items?

A real world example (high level code), the calculations abc are simple:

save cr
set_source_color cr rgb1
loop 100 times
   do calculations abc
   draw box
end loop
fill cr
restore cr

set_source_color cr rgb2
loop 100 times
   do calculations abc
   draw arc
end loop
fill cr
restore cr

set_source_color cr rgb3
loop 100 times
   do calculations abc
   draw rectangle
end loop
fill cr
restore cr

or

loop 100 times
   do calculations abc

   set_source_color cr rgb1
   draw box
   fill cr

   set_source_color cr rgb2
   draw arc
   fill cr

   set_source_color cr rgb3
   draw rectangle
   fill cr
end loop

Of course this only works if all the boxes ect have the same color. Otherwise option 2 has to be used.

This is meant as a more general question, the given example is just an illustration to make my example scenario clear. Do not hesitate to answer in depth/technical.

Answer 1

When you draw on a window, most drawing operations happen on GPU and i guess your calculations on CPU. Each processing unit has, dependent on the operation, throughput and processing time. To avoid waiting much for synchronization, you should mix your calculations and drawings to let the units work in parallel.

In case of an image surface as target, everything would happen on CPU, but internal multithreading might result in a parallel rendering pipeline as well.

I recommend to use your first approach.