Reduce the use of porting::getTimeMs() when rendering frames

[x2048]

Optimization is based on this graph from hotspot:

1. Remove excessive calls to TimeTaker in renderMap() and renderMapShadow()
2. Centralize clock time per frame in ClientEnvironment and avoid calling porting::getTimeMs() from shader constant setters.

To do

This PR is Ready for Review.

How to test

1. Start any game
2. FPS should be higher, and rendering should be correct.

[rubenwardy]

Do you get a higher FPS with this? Interested to know if there's a visible difference

[Desour]

Tested using quicktune: Desour@3f9c3c2
I've seen no difference on my machine.

[x2048]

@Desour updateFrameTime() is the optimized method, can you try to quicktune on master?

The difference for me is from 37 to 47 FPS in a similar scene in the same location in a RelWithDebInfo build:

[SmallJoker]

This would be the best location to remind contributors of the purpose of this function.

Suggested change

	m_frame_time = porting::getTimeMs();
	// Cache the time each frame to reduce time-consuming getTimeMs() calls for (semi-)debug builds.
	m_frame_time = porting::getTimeMs();

[sfan5]

It's also more correct to always use the same timestamp during rendering of a specific frame.

[x2048]

It's an added bonus :)

[Desour]

@Desour updateFrameTime() is the optimized method, can you try to quicktune on master?

I'm not sure what you mean with "quicktune on master".
I've used minetest's quicktune thing to toggle the optimisation on and off in-game (i.e. make getFrameTime call porting::getTimeMs or just return the cached value).

I've used a Release build, not RelWithDebInfo. Idk if that makes a difference.

[sfan5]

I'm all for reducing useless calls but isn't clock_gettime supposed to be very fast? It goes via vdso on Linux.

[x2048]

Clean release builds show even bigger difference between the branches:

@sfan5 If my math is right, we were calling getTimeMs() 200K times a second (without shadow mapping) at 33 FPS. Apparently, clock_gettime is not that fast, at least on my Ubuntu kernel.

[sfan5]

Works but the improvement is tiny on my machine.