Regal: OpenGL over OpenGLES2 emulation - some tech. info and porting progress

@Daytona675x
Tested new version of ogles2.libary and with reverted back regal's changes: it works ! Thanks a bunch !:)

Quote:

Probably for more accurate test, i can just skip first frame from fps calcualtion ? Or, i may run test just for some longest time, if it will make any difference.


Quote:

Yeah, seems so, sadly. But at least if we can at first emulate ogl1/2/3/4 properly, later may come some optimisation, even small bits can help.

But, i assume that its only like that for such a small test case , for big ones it should make a differences in positive side ?

I just can't imagine that regal will be slower than minigl :)) Its kind of make no sense then.. Only as some "slow emulation for legacy OGL" :))

Do you have maybe by hands some example, which we can put on mgl, mgl-reloaded and regal to see _real_ difference, i mean maximum which we can expect. I mean if it just so small example give us bad feeling about perfomance, maybe we need some real example

EDIT: maybe we could speedup regal just by adding to compiler some more optimisation.. Currently i only use "-fvisibility=hidden -O2 -fomit-frame-pointer" as extra CFLAGS when building.

@Hans
Quote:
Awesome, thanks!

Quote:
Hope you can give it maximum priority in your list, then we can throw at regal some _big_ test cases, and there where interesting stuff will starts :) As well as arrays support is by far the last big missing bit from NOVA's shaders (?) which will help not only Regal , but in whole.

I ready to test any alpha, beta, when and how you wish, of course :)

Quote:
Yes, in such simple tests that should do. If your test has several "stages" of such simple tests (e.g. first a rotating colored cube, then some texture effect, etc.) you should not measure the first frame of each sub-test.
IIRC we had a similar effect here once with some SDL-speed-test.

Quote:
As soon as MiniGL spends more time in its software TCL than Regal/MGLReloaded need to eventually update the internal VBOs / DBOs / whatever-bookkeeping, MiniGL will become slower. One can asume from the previous test that this turning-point X is reached significantly earlier for MGLReloaded than for Regal.
However, at some other point Y, if the workload is high enough so that the time spent by the GPU outweights the time spent in the respective library significantly, I expect Regal and MGLReloaded to give more or less similar results. More or less Because then the quality of the shaders will become interesting and looking at Regal's shaders MGLReloaded will certainly beat it
I have no idea about concrete values for X or Y though (also system-dependent, of course, just look at your and my results).

Quote:
As being said, it's no surprise that MGL is faster if the TCL workload is low. However, the gap certainly shouldn't be thaaaaat large. But then again: Regal's primary aim is not performance and looks like nobody cared about that aspect.

Quote:
First of all you could massively increase the vertex-count here. Just put a nice for-loop around the glVertex-calls so that it creates some thousands of vertices. Other than that you could also make the TCL calculations more complicated. Most simple: move the objects to the edges of the viewport so that only half of them is visible. This will force MGL to modfiy / add triangles. And if smooth colored / textured shading is enabled it will also force it to create new color / tex-coord values.
You could also add lighting / more lights. However, then MGLReloaded's results will likely become useless because I didn't implement all features accurately yet, so I'd not recommend that for now if you want to compare all three variants. (oh, I forgot: won't work for now anyway because Regal uses arrays when doing lighting).

Quote:
May well be that ogles2.library will require some adjustments for array support too.

@all

Great progress and nice read, congrats